diff options
author | hairball <xhairball@gmail.com> | 2014-02-08 03:21:02 +0000 |
---|---|---|
committer | Tim Angus <tim@ngus.net> | 2014-06-17 17:43:38 +0100 |
commit | 35064811c0ac104acddd7777e00bfd9e054c2db6 (patch) | |
tree | 4063389141542093ecc3de47e3999ed099520621 /src | |
parent | 1778f3fb8cabe7400011c84331018b0ebf6a44b1 (diff) |
Upgrade opus 1.0.2 -> 1.1
Diffstat (limited to 'src')
-rw-r--r-- | src/opus-1.0.2/celt/celt.c | 2906 | ||||
-rw-r--r-- | src/opus-1.0.2/celt/celt_lpc.c | 188 | ||||
-rw-r--r-- | src/opus-1.0.2/celt/cwrs.c | 645 | ||||
-rw-r--r-- | src/opus-1.0.2/src/opus_multistream.c | 1027 | ||||
-rw-r--r-- | src/opus-1.1/celt/_kiss_fft_guts.h (renamed from src/opus-1.0.2/celt/_kiss_fft_guts.h) | 8 | ||||
-rw-r--r-- | src/opus-1.1/celt/arch.h (renamed from src/opus-1.0.2/celt/arch.h) | 15 | ||||
-rwxr-xr-x | src/opus-1.1/celt/arm/arm2gnu.pl | 316 | ||||
-rw-r--r-- | src/opus-1.1/celt/arm/arm_celt_map.c | 49 | ||||
-rw-r--r-- | src/opus-1.1/celt/arm/armcpu.c | 174 | ||||
-rw-r--r-- | src/opus-1.1/celt/arm/armcpu.h | 71 | ||||
-rw-r--r-- | src/opus-1.1/celt/arm/armopts.s.in (renamed from src/opus-1.0.2/src/opus.c) | 28 | ||||
-rw-r--r-- | src/opus-1.1/celt/arm/celt_pitch_xcorr_arm.s | 545 | ||||
-rw-r--r-- | src/opus-1.1/celt/arm/fixed_armv4.h | 76 | ||||
-rw-r--r-- | src/opus-1.1/celt/arm/fixed_armv5e.h | 116 | ||||
-rw-r--r-- | src/opus-1.1/celt/arm/kiss_fft_armv4.h | 121 | ||||
-rw-r--r-- | src/opus-1.1/celt/arm/kiss_fft_armv5e.h | 118 | ||||
-rw-r--r-- | src/opus-1.1/celt/arm/pitch_arm.h | 57 | ||||
-rw-r--r-- | src/opus-1.1/celt/bands.c (renamed from src/opus-1.0.2/celt/bands.c) | 1204 | ||||
-rw-r--r-- | src/opus-1.1/celt/bands.h (renamed from src/opus-1.0.2/celt/bands.h) | 31 | ||||
-rw-r--r-- | src/opus-1.1/celt/celt.c | 223 | ||||
-rw-r--r-- | src/opus-1.1/celt/celt.h (renamed from src/opus-1.0.2/celt/celt.h) | 107 | ||||
-rw-r--r-- | src/opus-1.1/celt/celt_decoder.c | 1195 | ||||
-rw-r--r-- | src/opus-1.1/celt/celt_encoder.c | 2353 | ||||
-rw-r--r-- | src/opus-1.1/celt/celt_lpc.c | 309 | ||||
-rw-r--r-- | src/opus-1.1/celt/celt_lpc.h (renamed from src/opus-1.0.2/celt/celt_lpc.h) | 3 | ||||
-rw-r--r-- | src/opus-1.1/celt/cpu_support.h (renamed from src/opus-1.0.2/celt/pitch.h) | 40 | ||||
-rw-r--r-- | src/opus-1.1/celt/cwrs.c | 697 | ||||
-rw-r--r-- | src/opus-1.1/celt/cwrs.h (renamed from src/opus-1.0.2/celt/cwrs.h) | 0 | ||||
-rw-r--r-- | src/opus-1.1/celt/ecintrin.h (renamed from src/opus-1.0.2/celt/ecintrin.h) | 2 | ||||
-rw-r--r-- | src/opus-1.1/celt/entcode.c (renamed from src/opus-1.0.2/celt/entcode.c) | 0 | ||||
-rw-r--r-- | src/opus-1.1/celt/entcode.h (renamed from src/opus-1.0.2/celt/entcode.h) | 9 | ||||
-rw-r--r-- | src/opus-1.1/celt/entdec.c (renamed from src/opus-1.0.2/celt/entdec.c) | 2 | ||||
-rw-r--r-- | src/opus-1.1/celt/entdec.h (renamed from src/opus-1.0.2/celt/entdec.h) | 0 | ||||
-rw-r--r-- | src/opus-1.1/celt/entenc.c (renamed from src/opus-1.0.2/celt/entenc.c) | 0 | ||||
-rw-r--r-- | src/opus-1.1/celt/entenc.h (renamed from src/opus-1.0.2/celt/entenc.h) | 0 | ||||
-rw-r--r-- | src/opus-1.1/celt/fixed_debug.h (renamed from src/opus-1.0.2/celt/fixed_debug.h) | 68 | ||||
-rw-r--r-- | src/opus-1.1/celt/fixed_generic.h (renamed from src/opus-1.0.2/celt/fixed_generic.h) | 9 | ||||
-rw-r--r-- | src/opus-1.1/celt/float_cast.h (renamed from src/opus-1.0.2/celt/float_cast.h) | 4 | ||||
-rw-r--r-- | src/opus-1.1/celt/kiss_fft.c (renamed from src/opus-1.0.2/celt/kiss_fft.c) | 3 | ||||
-rw-r--r-- | src/opus-1.1/celt/kiss_fft.h (renamed from src/opus-1.0.2/celt/kiss_fft.h) | 0 | ||||
-rw-r--r-- | src/opus-1.1/celt/laplace.c (renamed from src/opus-1.0.2/celt/laplace.c) | 0 | ||||
-rw-r--r-- | src/opus-1.1/celt/laplace.h (renamed from src/opus-1.0.2/celt/laplace.h) | 0 | ||||
-rw-r--r-- | src/opus-1.1/celt/mathops.c (renamed from src/opus-1.0.2/celt/mathops.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/celt/mathops.h (renamed from src/opus-1.0.2/celt/mathops.h) | 83 | ||||
-rw-r--r-- | src/opus-1.1/celt/mdct.c (renamed from src/opus-1.0.2/celt/mdct.c) | 113 | ||||
-rw-r--r-- | src/opus-1.1/celt/mdct.h (renamed from src/opus-1.0.2/celt/mdct.h) | 0 | ||||
-rw-r--r-- | src/opus-1.1/celt/mfrngcod.h (renamed from src/opus-1.0.2/celt/mfrngcod.h) | 0 | ||||
-rw-r--r-- | src/opus-1.1/celt/modes.c (renamed from src/opus-1.0.2/celt/modes.c) | 8 | ||||
-rw-r--r-- | src/opus-1.1/celt/modes.h (renamed from src/opus-1.0.2/celt/modes.h) | 0 | ||||
-rw-r--r-- | src/opus-1.1/celt/opus_custom_demo.c (renamed from src/opus-1.0.2/celt/opus_custom_demo.c) | 0 | ||||
-rw-r--r-- | src/opus-1.1/celt/os_support.h (renamed from src/opus-1.0.2/celt/os_support.h) | 9 | ||||
-rw-r--r-- | src/opus-1.1/celt/pitch.c (renamed from src/opus-1.0.2/celt/pitch.c) | 197 | ||||
-rw-r--r-- | src/opus-1.1/celt/pitch.h | 173 | ||||
-rw-r--r-- | src/opus-1.1/celt/quant_bands.c (renamed from src/opus-1.0.2/celt/quant_bands.c) | 50 | ||||
-rw-r--r-- | src/opus-1.1/celt/quant_bands.h (renamed from src/opus-1.0.2/celt/quant_bands.h) | 8 | ||||
-rw-r--r-- | src/opus-1.1/celt/rate.c (renamed from src/opus-1.0.2/celt/rate.c) | 10 | ||||
-rw-r--r-- | src/opus-1.1/celt/rate.h (renamed from src/opus-1.0.2/celt/rate.h) | 8 | ||||
-rw-r--r-- | src/opus-1.1/celt/stack_alloc.h (renamed from src/opus-1.0.2/celt/stack_alloc.h) | 29 | ||||
-rw-r--r-- | src/opus-1.1/celt/static_modes_fixed.h (renamed from src/opus-1.0.2/celt/static_modes_fixed.h) | 0 | ||||
-rw-r--r-- | src/opus-1.1/celt/static_modes_float.h (renamed from src/opus-1.0.2/celt/static_modes_float.h) | 0 | ||||
-rw-r--r-- | src/opus-1.1/celt/vq.c (renamed from src/opus-1.0.2/celt/vq.c) | 0 | ||||
-rw-r--r-- | src/opus-1.1/celt/vq.h (renamed from src/opus-1.0.2/celt/vq.h) | 7 | ||||
-rw-r--r-- | src/opus-1.1/celt/x86/pitch_sse.h | 156 | ||||
-rw-r--r-- | src/opus-1.1/include/opus.h (renamed from src/opus-1.0.2/include/opus.h) | 79 | ||||
-rw-r--r-- | src/opus-1.1/include/opus_custom.h (renamed from src/opus-1.0.2/include/opus_custom.h) | 89 | ||||
-rw-r--r-- | src/opus-1.1/include/opus_defines.h (renamed from src/opus-1.0.2/include/opus_defines.h) | 79 | ||||
-rw-r--r-- | src/opus-1.1/include/opus_multistream.h (renamed from src/opus-1.0.2/include/opus_multistream.h) | 28 | ||||
-rw-r--r-- | src/opus-1.1/include/opus_types.h (renamed from src/opus-1.0.2/include/opus_types.h) | 0 | ||||
-rw-r--r-- | src/opus-1.1/silk/A2NLSF.c (renamed from src/opus-1.0.2/silk/A2NLSF.c) | 10 | ||||
-rw-r--r-- | src/opus-1.1/silk/API.h (renamed from src/opus-1.0.2/silk/API.h) | 5 | ||||
-rw-r--r-- | src/opus-1.1/silk/CNG.c (renamed from src/opus-1.0.2/silk/CNG.c) | 13 | ||||
-rw-r--r-- | src/opus-1.1/silk/HP_variable_cutoff.c (renamed from src/opus-1.0.2/silk/HP_variable_cutoff.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/Inlines.h (renamed from src/opus-1.0.2/silk/Inlines.h) | 16 | ||||
-rw-r--r-- | src/opus-1.1/silk/LPC_analysis_filter.c (renamed from src/opus-1.0.2/silk/LPC_analysis_filter.c) | 27 | ||||
-rw-r--r-- | src/opus-1.1/silk/LPC_inv_pred_gain.c (renamed from src/opus-1.0.2/silk/LPC_inv_pred_gain.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/LP_variable_cutoff.c (renamed from src/opus-1.0.2/silk/LP_variable_cutoff.c) | 6 | ||||
-rw-r--r-- | src/opus-1.1/silk/MacroCount.h (renamed from src/opus-1.0.2/silk/MacroCount.h) | 164 | ||||
-rw-r--r-- | src/opus-1.1/silk/MacroDebug.h (renamed from src/opus-1.0.2/silk/MacroDebug.h) | 114 | ||||
-rw-r--r-- | src/opus-1.1/silk/NLSF2A.c (renamed from src/opus-1.0.2/silk/NLSF2A.c) | 6 | ||||
-rw-r--r-- | src/opus-1.1/silk/NLSF_VQ.c (renamed from src/opus-1.0.2/silk/NLSF_VQ.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/NLSF_VQ_weights_laroia.c (renamed from src/opus-1.0.2/silk/NLSF_VQ_weights_laroia.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/NLSF_decode.c (renamed from src/opus-1.0.2/silk/NLSF_decode.c) | 6 | ||||
-rw-r--r-- | src/opus-1.1/silk/NLSF_del_dec_quant.c (renamed from src/opus-1.0.2/silk/NLSF_del_dec_quant.c) | 6 | ||||
-rw-r--r-- | src/opus-1.1/silk/NLSF_encode.c (renamed from src/opus-1.0.2/silk/NLSF_encode.c) | 20 | ||||
-rw-r--r-- | src/opus-1.1/silk/NLSF_stabilize.c (renamed from src/opus-1.0.2/silk/NLSF_stabilize.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/NLSF_unpack.c (renamed from src/opus-1.0.2/silk/NLSF_unpack.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/NSQ.c (renamed from src/opus-1.0.2/silk/NSQ.c) | 25 | ||||
-rw-r--r-- | src/opus-1.1/silk/NSQ_del_dec.c (renamed from src/opus-1.0.2/silk/NSQ_del_dec.c) | 40 | ||||
-rw-r--r-- | src/opus-1.1/silk/PLC.c (renamed from src/opus-1.0.2/silk/PLC.c) | 12 | ||||
-rw-r--r-- | src/opus-1.1/silk/PLC.h (renamed from src/opus-1.0.2/silk/PLC.h) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/SigProc_FIX.h (renamed from src/opus-1.0.2/silk/SigProc_FIX.h) | 49 | ||||
-rw-r--r-- | src/opus-1.1/silk/VAD.c (renamed from src/opus-1.0.2/silk/VAD.c) | 59 | ||||
-rw-r--r-- | src/opus-1.1/silk/VQ_WMat_EC.c (renamed from src/opus-1.0.2/silk/VQ_WMat_EC.c) | 15 | ||||
-rw-r--r-- | src/opus-1.1/silk/ana_filt_bank_1.c (renamed from src/opus-1.0.2/silk/ana_filt_bank_1.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/arm/SigProc_FIX_armv4.h | 47 | ||||
-rw-r--r-- | src/opus-1.1/silk/arm/SigProc_FIX_armv5e.h | 61 | ||||
-rw-r--r-- | src/opus-1.1/silk/arm/macros_armv4.h | 103 | ||||
-rw-r--r-- | src/opus-1.1/silk/arm/macros_armv5e.h | 213 | ||||
-rw-r--r-- | src/opus-1.1/silk/biquad_alt.c (renamed from src/opus-1.0.2/silk/biquad_alt.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/bwexpander.c (renamed from src/opus-1.0.2/silk/bwexpander.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/bwexpander_32.c (renamed from src/opus-1.0.2/silk/bwexpander_32.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/check_control_input.c (renamed from src/opus-1.0.2/silk/check_control_input.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/code_signs.c (renamed from src/opus-1.0.2/silk/code_signs.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/control.h (renamed from src/opus-1.0.2/silk/control.h) | 7 | ||||
-rw-r--r-- | src/opus-1.1/silk/control_SNR.c (renamed from src/opus-1.0.2/silk/control_SNR.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/control_audio_bandwidth.c (renamed from src/opus-1.0.2/silk/control_audio_bandwidth.c) | 7 | ||||
-rw-r--r-- | src/opus-1.1/silk/control_codec.c (renamed from src/opus-1.0.2/silk/control_codec.c) | 47 | ||||
-rw-r--r-- | src/opus-1.1/silk/debug.c (renamed from src/opus-1.0.2/silk/debug.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/debug.h (renamed from src/opus-1.0.2/silk/debug.h) | 9 | ||||
-rw-r--r-- | src/opus-1.1/silk/dec_API.c (renamed from src/opus-1.0.2/silk/dec_API.c) | 11 | ||||
-rw-r--r-- | src/opus-1.1/silk/decode_core.c (renamed from src/opus-1.0.2/silk/decode_core.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/decode_frame.c (renamed from src/opus-1.0.2/silk/decode_frame.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/decode_indices.c (renamed from src/opus-1.0.2/silk/decode_indices.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/decode_parameters.c (renamed from src/opus-1.0.2/silk/decode_parameters.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/decode_pitch.c (renamed from src/opus-1.0.2/silk/decode_pitch.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/decode_pulses.c (renamed from src/opus-1.0.2/silk/decode_pulses.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/decoder_set_fs.c (renamed from src/opus-1.0.2/silk/decoder_set_fs.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/define.h (renamed from src/opus-1.0.2/silk/define.h) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/enc_API.c (renamed from src/opus-1.0.2/silk/enc_API.c) | 50 | ||||
-rw-r--r-- | src/opus-1.1/silk/encode_indices.c (renamed from src/opus-1.0.2/silk/encode_indices.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/encode_pulses.c (renamed from src/opus-1.0.2/silk/encode_pulses.c) | 19 | ||||
-rw-r--r-- | src/opus-1.1/silk/errors.h (renamed from src/opus-1.0.2/silk/errors.h) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/LTP_analysis_filter_FIX.c (renamed from src/opus-1.0.2/silk/fixed/LTP_analysis_filter_FIX.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/LTP_scale_ctrl_FIX.c (renamed from src/opus-1.0.2/silk/fixed/LTP_scale_ctrl_FIX.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/apply_sine_window_FIX.c (renamed from src/opus-1.0.2/silk/fixed/apply_sine_window_FIX.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/autocorr_FIX.c (renamed from src/opus-1.0.2/silk/fixed/autocorr_FIX.c) | 42 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/burg_modified_FIX.c (renamed from src/opus-1.0.2/silk/fixed/burg_modified_FIX.c) | 24 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/corrMatrix_FIX.c (renamed from src/opus-1.0.2/silk/fixed/corrMatrix_FIX.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/encode_frame_FIX.c (renamed from src/opus-1.0.2/silk/fixed/encode_frame_FIX.c) | 47 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/find_LPC_FIX.c (renamed from src/opus-1.0.2/silk/fixed/find_LPC_FIX.c) | 16 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/find_LTP_FIX.c (renamed from src/opus-1.0.2/silk/fixed/find_LTP_FIX.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/find_pitch_lags_FIX.c (renamed from src/opus-1.0.2/silk/fixed/find_pitch_lags_FIX.c) | 34 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/find_pred_coefs_FIX.c (renamed from src/opus-1.0.2/silk/fixed/find_pred_coefs_FIX.c) | 21 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/k2a_FIX.c (renamed from src/opus-1.0.2/silk/fixed/k2a_FIX.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/k2a_Q16_FIX.c (renamed from src/opus-1.0.2/silk/fixed/k2a_Q16_FIX.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/main_FIX.h (renamed from src/opus-1.0.2/silk/fixed/main_FIX.h) | 13 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/noise_shape_analysis_FIX.c (renamed from src/opus-1.0.2/silk/fixed/noise_shape_analysis_FIX.c) | 19 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/pitch_analysis_core_FIX.c (renamed from src/opus-1.0.2/silk/fixed/pitch_analysis_core_FIX.c) | 435 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/prefilter_FIX.c (renamed from src/opus-1.0.2/silk/fixed/prefilter_FIX.c) | 17 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/process_gains_FIX.c (renamed from src/opus-1.0.2/silk/fixed/process_gains_FIX.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/regularize_correlations_FIX.c (renamed from src/opus-1.0.2/silk/fixed/regularize_correlations_FIX.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/residual_energy16_FIX.c (renamed from src/opus-1.0.2/silk/fixed/residual_energy16_FIX.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/residual_energy_FIX.c (renamed from src/opus-1.0.2/silk/fixed/residual_energy_FIX.c) | 12 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/schur64_FIX.c (renamed from src/opus-1.0.2/silk/fixed/schur64_FIX.c) | 21 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/schur_FIX.c (renamed from src/opus-1.0.2/silk/fixed/schur_FIX.c) | 20 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/solve_LS_FIX.c (renamed from src/opus-1.0.2/silk/fixed/solve_LS_FIX.c) | 26 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/structs_FIX.h (renamed from src/opus-1.0.2/silk/fixed/structs_FIX.h) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/vector_ops_FIX.c (renamed from src/opus-1.0.2/silk/fixed/vector_ops_FIX.c) | 35 | ||||
-rw-r--r-- | src/opus-1.1/silk/fixed/warped_autocorrelation_FIX.c (renamed from src/opus-1.0.2/silk/fixed/warped_autocorrelation_FIX.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/LPC_analysis_filter_FLP.c (renamed from src/opus-1.0.2/silk/float/LPC_analysis_filter_FLP.c) | 14 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/LPC_inv_pred_gain_FLP.c (renamed from src/opus-1.0.2/silk/float/LPC_inv_pred_gain_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/LTP_analysis_filter_FLP.c (renamed from src/opus-1.0.2/silk/float/LTP_analysis_filter_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/LTP_scale_ctrl_FLP.c (renamed from src/opus-1.0.2/silk/float/LTP_scale_ctrl_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/SigProc_FLP.h (renamed from src/opus-1.0.2/silk/float/SigProc_FLP.h) | 17 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/apply_sine_window_FLP.c (renamed from src/opus-1.0.2/silk/float/apply_sine_window_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/autocorrelation_FLP.c (renamed from src/opus-1.0.2/silk/float/autocorrelation_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/burg_modified_FLP.c (renamed from src/opus-1.0.2/silk/float/burg_modified_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/bwexpander_FLP.c (renamed from src/opus-1.0.2/silk/float/bwexpander_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/corrMatrix_FLP.c (renamed from src/opus-1.0.2/silk/float/corrMatrix_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/encode_frame_FLP.c (renamed from src/opus-1.0.2/silk/float/encode_frame_FLP.c) | 18 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/energy_FLP.c (renamed from src/opus-1.0.2/silk/float/energy_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/find_LPC_FLP.c (renamed from src/opus-1.0.2/silk/float/find_LPC_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/find_LTP_FLP.c (renamed from src/opus-1.0.2/silk/float/find_LTP_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/find_pitch_lags_FLP.c (renamed from src/opus-1.0.2/silk/float/find_pitch_lags_FLP.c) | 9 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/find_pred_coefs_FLP.c (renamed from src/opus-1.0.2/silk/float/find_pred_coefs_FLP.c) | 7 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/inner_product_FLP.c (renamed from src/opus-1.0.2/silk/float/inner_product_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/k2a_FLP.c (renamed from src/opus-1.0.2/silk/float/k2a_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/levinsondurbin_FLP.c (renamed from src/opus-1.0.2/silk/float/levinsondurbin_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/main_FLP.h (renamed from src/opus-1.0.2/silk/float/main_FLP.h) | 11 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/noise_shape_analysis_FLP.c (renamed from src/opus-1.0.2/silk/float/noise_shape_analysis_FLP.c) | 8 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/pitch_analysis_core_FLP.c (renamed from src/opus-1.0.2/silk/float/pitch_analysis_core_FLP.c) | 132 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/prefilter_FLP.c (renamed from src/opus-1.0.2/silk/float/prefilter_FLP.c) | 8 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/process_gains_FLP.c (renamed from src/opus-1.0.2/silk/float/process_gains_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/regularize_correlations_FLP.c (renamed from src/opus-1.0.2/silk/float/regularize_correlations_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/residual_energy_FLP.c (renamed from src/opus-1.0.2/silk/float/residual_energy_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/scale_copy_vector_FLP.c (renamed from src/opus-1.0.2/silk/float/scale_copy_vector_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/scale_vector_FLP.c (renamed from src/opus-1.0.2/silk/float/scale_vector_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/schur_FLP.c (renamed from src/opus-1.0.2/silk/float/schur_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/solve_LS_FLP.c (renamed from src/opus-1.0.2/silk/float/solve_LS_FLP.c) | 16 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/sort_FLP.c (renamed from src/opus-1.0.2/silk/float/sort_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/structs_FLP.h (renamed from src/opus-1.0.2/silk/float/structs_FLP.h) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/warped_autocorrelation_FLP.c (renamed from src/opus-1.0.2/silk/float/warped_autocorrelation_FLP.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/float/wrappers_FLP.c (renamed from src/opus-1.0.2/silk/float/wrappers_FLP.c) | 7 | ||||
-rw-r--r-- | src/opus-1.1/silk/gain_quant.c (renamed from src/opus-1.0.2/silk/gain_quant.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/init_decoder.c (renamed from src/opus-1.0.2/silk/init_decoder.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/init_encoder.c (renamed from src/opus-1.0.2/silk/init_encoder.c) | 10 | ||||
-rw-r--r-- | src/opus-1.1/silk/inner_prod_aligned.c (renamed from src/opus-1.0.2/silk/inner_prod_aligned.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/interpolate.c (renamed from src/opus-1.0.2/silk/interpolate.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/lin2log.c (renamed from src/opus-1.0.2/silk/lin2log.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/log2lin.c (renamed from src/opus-1.0.2/silk/log2lin.c) | 8 | ||||
-rw-r--r-- | src/opus-1.1/silk/macros.h (renamed from src/opus-1.0.2/silk/macros.h) | 72 | ||||
-rw-r--r-- | src/opus-1.1/silk/main.h (renamed from src/opus-1.0.2/silk/main.h) | 8 | ||||
-rw-r--r-- | src/opus-1.1/silk/pitch_est_defines.h (renamed from src/opus-1.0.2/silk/pitch_est_defines.h) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/pitch_est_tables.c (renamed from src/opus-1.0.2/silk/pitch_est_tables.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/process_NLSFs.c (renamed from src/opus-1.0.2/silk/process_NLSFs.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/quant_LTP_gains.c (renamed from src/opus-1.0.2/silk/quant_LTP_gains.c) | 25 | ||||
-rw-r--r-- | src/opus-1.1/silk/resampler.c (renamed from src/opus-1.0.2/silk/resampler.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/resampler_down2.c (renamed from src/opus-1.0.2/silk/resampler_down2.c) | 8 | ||||
-rw-r--r-- | src/opus-1.1/silk/resampler_down2_3.c (renamed from src/opus-1.0.2/silk/resampler_down2_3.c) | 11 | ||||
-rw-r--r-- | src/opus-1.1/silk/resampler_private.h (renamed from src/opus-1.0.2/silk/resampler_private.h) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/resampler_private_AR2.c (renamed from src/opus-1.0.2/silk/resampler_private_AR2.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/resampler_private_IIR_FIR.c (renamed from src/opus-1.0.2/silk/resampler_private_IIR_FIR.c) | 20 | ||||
-rw-r--r-- | src/opus-1.1/silk/resampler_private_down_FIR.c (renamed from src/opus-1.0.2/silk/resampler_private_down_FIR.c) | 17 | ||||
-rw-r--r-- | src/opus-1.1/silk/resampler_private_up2_HQ.c (renamed from src/opus-1.0.2/silk/resampler_private_up2_HQ.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/resampler_rom.c (renamed from src/opus-1.0.2/silk/resampler_rom.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/resampler_rom.h (renamed from src/opus-1.0.2/silk/resampler_rom.h) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/resampler_structs.h (renamed from src/opus-1.0.2/silk/resampler_structs.h) | 9 | ||||
-rw-r--r-- | src/opus-1.1/silk/shell_coder.c (renamed from src/opus-1.0.2/silk/shell_coder.c) | 10 | ||||
-rw-r--r-- | src/opus-1.1/silk/sigm_Q15.c (renamed from src/opus-1.0.2/silk/sigm_Q15.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/sort.c (renamed from src/opus-1.0.2/silk/sort.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/stereo_LR_to_MS.c (renamed from src/opus-1.0.2/silk/stereo_LR_to_MS.c) | 20 | ||||
-rw-r--r-- | src/opus-1.1/silk/stereo_MS_to_LR.c (renamed from src/opus-1.0.2/silk/stereo_MS_to_LR.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/stereo_decode_pred.c (renamed from src/opus-1.0.2/silk/stereo_decode_pred.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/stereo_encode_pred.c (renamed from src/opus-1.0.2/silk/stereo_encode_pred.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/stereo_find_predictor.c (renamed from src/opus-1.0.2/silk/stereo_find_predictor.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/stereo_quant_pred.c (renamed from src/opus-1.0.2/silk/stereo_quant_pred.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/structs.h (renamed from src/opus-1.0.2/silk/structs.h) | 7 | ||||
-rw-r--r-- | src/opus-1.1/silk/sum_sqr_shift.c (renamed from src/opus-1.0.2/silk/sum_sqr_shift.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/table_LSF_cos.c (renamed from src/opus-1.0.2/silk/table_LSF_cos.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/tables.h (renamed from src/opus-1.0.2/silk/tables.h) | 6 | ||||
-rw-r--r-- | src/opus-1.1/silk/tables_LTP.c (renamed from src/opus-1.0.2/silk/tables_LTP.c) | 28 | ||||
-rw-r--r-- | src/opus-1.1/silk/tables_NLSF_CB_NB_MB.c (renamed from src/opus-1.0.2/silk/tables_NLSF_CB_NB_MB.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/tables_NLSF_CB_WB.c (renamed from src/opus-1.0.2/silk/tables_NLSF_CB_WB.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/tables_gain.c (renamed from src/opus-1.0.2/silk/tables_gain.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/tables_other.c (renamed from src/opus-1.0.2/silk/tables_other.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/tables_pitch_lag.c (renamed from src/opus-1.0.2/silk/tables_pitch_lag.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/tables_pulses_per_block.c (renamed from src/opus-1.0.2/silk/tables_pulses_per_block.c) | 4 | ||||
-rw-r--r-- | src/opus-1.1/silk/tuning_parameters.h (renamed from src/opus-1.0.2/silk/tuning_parameters.h) | 9 | ||||
-rw-r--r-- | src/opus-1.1/silk/typedef.h (renamed from src/opus-1.0.2/silk/typedef.h) | 7 | ||||
-rw-r--r-- | src/opus-1.1/src/analysis.c | 645 | ||||
-rw-r--r-- | src/opus-1.1/src/analysis.h | 90 | ||||
-rw-r--r-- | src/opus-1.1/src/mlp.c | 140 | ||||
-rw-r--r-- | src/opus-1.1/src/mlp.h | 41 | ||||
-rw-r--r-- | src/opus-1.1/src/mlp_data.c | 105 | ||||
-rw-r--r-- | src/opus-1.1/src/opus.c | 329 | ||||
-rw-r--r-- | src/opus-1.1/src/opus_decoder.c (renamed from src/opus-1.0.2/src/opus_decoder.c) | 379 | ||||
-rw-r--r-- | src/opus-1.1/src/opus_encoder.c (renamed from src/opus-1.0.2/src/opus_encoder.c) | 1195 | ||||
-rw-r--r-- | src/opus-1.1/src/opus_multistream.c | 92 | ||||
-rw-r--r-- | src/opus-1.1/src/opus_multistream_decoder.c | 537 | ||||
-rw-r--r-- | src/opus-1.1/src/opus_multistream_encoder.c | 1174 | ||||
-rw-r--r-- | src/opus-1.1/src/opus_private.h (renamed from src/opus-1.0.2/src/opus_private.h) | 54 | ||||
-rw-r--r-- | src/opus-1.1/src/repacketizer.c (renamed from src/opus-1.0.2/src/repacketizer.c) | 197 | ||||
-rw-r--r-- | src/opus-1.1/src/tansig_table.h | 45 |
243 files changed, 14618 insertions, 7081 deletions
diff --git a/src/opus-1.0.2/celt/celt.c b/src/opus-1.0.2/celt/celt.c deleted file mode 100644 index 9bbe8524..00000000 --- a/src/opus-1.0.2/celt/celt.c +++ /dev/null @@ -1,2906 +0,0 @@ -/* Copyright (c) 2007-2008 CSIRO - Copyright (c) 2007-2010 Xiph.Org Foundation - Copyright (c) 2008 Gregory Maxwell - Written by Jean-Marc Valin and Gregory Maxwell */ -/* - Redistribution and use in source and binary forms, with or without - modification, are permitted provided that the following conditions - are met: - - - Redistributions of source code must retain the above copyright - notice, this list of conditions and the following disclaimer. - - - Redistributions in binary form must reproduce the above copyright - notice, this list of conditions and the following disclaimer in the - documentation and/or other materials provided with the distribution. - - THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER - OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, - EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, - PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR - PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF - LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING - NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS - SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -*/ - -#ifdef HAVE_CONFIG_H -#include "config.h" -#endif - -#define CELT_C - -#include "os_support.h" -#include "mdct.h" -#include <math.h> -#include "celt.h" -#include "pitch.h" -#include "bands.h" -#include "modes.h" -#include "entcode.h" -#include "quant_bands.h" -#include "rate.h" -#include "stack_alloc.h" -#include "mathops.h" -#include "float_cast.h" -#include <stdarg.h> -#include "celt_lpc.h" -#include "vq.h" - -#ifndef OPUS_VERSION -#define OPUS_VERSION "unknown" -#endif - -#ifdef CUSTOM_MODES -#define OPUS_CUSTOM_NOSTATIC -#else -#define OPUS_CUSTOM_NOSTATIC static inline -#endif - -static const unsigned char trim_icdf[11] = {126, 124, 119, 109, 87, 41, 19, 9, 4, 2, 0}; -/* Probs: NONE: 21.875%, LIGHT: 6.25%, NORMAL: 65.625%, AGGRESSIVE: 6.25% */ -static const unsigned char spread_icdf[4] = {25, 23, 2, 0}; - -static const unsigned char tapset_icdf[3]={2,1,0}; - -#ifdef CUSTOM_MODES -static const unsigned char toOpusTable[20] = { - 0xE0, 0xE8, 0xF0, 0xF8, - 0xC0, 0xC8, 0xD0, 0xD8, - 0xA0, 0xA8, 0xB0, 0xB8, - 0x00, 0x00, 0x00, 0x00, - 0x80, 0x88, 0x90, 0x98, -}; - -static const unsigned char fromOpusTable[16] = { - 0x80, 0x88, 0x90, 0x98, - 0x40, 0x48, 0x50, 0x58, - 0x20, 0x28, 0x30, 0x38, - 0x00, 0x08, 0x10, 0x18 -}; - -static inline int toOpus(unsigned char c) -{ - int ret=0; - if (c<0xA0) - ret = toOpusTable[c>>3]; - if (ret == 0) - return -1; - else - return ret|(c&0x7); -} - -static inline int fromOpus(unsigned char c) -{ - if (c<0x80) - return -1; - else - return fromOpusTable[(c>>3)-16] | (c&0x7); -} -#endif /* CUSTOM_MODES */ - -#define COMBFILTER_MAXPERIOD 1024 -#define COMBFILTER_MINPERIOD 15 - -static int resampling_factor(opus_int32 rate) -{ - int ret; - switch (rate) - { - case 48000: - ret = 1; - break; - case 24000: - ret = 2; - break; - case 16000: - ret = 3; - break; - case 12000: - ret = 4; - break; - case 8000: - ret = 6; - break; - default: -#ifndef CUSTOM_MODES - celt_assert(0); -#endif - ret = 0; - break; - } - return ret; -} - -/** Encoder state - @brief Encoder state - */ -struct OpusCustomEncoder { - const OpusCustomMode *mode; /**< Mode used by the encoder */ - int overlap; - int channels; - int stream_channels; - - int force_intra; - int clip; - int disable_pf; - int complexity; - int upsample; - int start, end; - - opus_int32 bitrate; - int vbr; - int signalling; - int constrained_vbr; /* If zero, VBR can do whatever it likes with the rate */ - int loss_rate; - int lsb_depth; - - /* Everything beyond this point gets cleared on a reset */ -#define ENCODER_RESET_START rng - - opus_uint32 rng; - int spread_decision; - opus_val32 delayedIntra; - int tonal_average; - int lastCodedBands; - int hf_average; - int tapset_decision; - - int prefilter_period; - opus_val16 prefilter_gain; - int prefilter_tapset; -#ifdef RESYNTH - int prefilter_period_old; - opus_val16 prefilter_gain_old; - int prefilter_tapset_old; -#endif - int consec_transient; - - opus_val32 preemph_memE[2]; - opus_val32 preemph_memD[2]; - - /* VBR-related parameters */ - opus_int32 vbr_reservoir; - opus_int32 vbr_drift; - opus_int32 vbr_offset; - opus_int32 vbr_count; - -#ifdef RESYNTH - celt_sig syn_mem[2][2*MAX_PERIOD]; -#endif - - celt_sig in_mem[1]; /* Size = channels*mode->overlap */ - /* celt_sig prefilter_mem[], Size = channels*COMBFILTER_MAXPERIOD */ - /* opus_val16 oldBandE[], Size = channels*mode->nbEBands */ - /* opus_val16 oldLogE[], Size = channels*mode->nbEBands */ - /* opus_val16 oldLogE2[], Size = channels*mode->nbEBands */ -#ifdef RESYNTH - /* opus_val16 overlap_mem[], Size = channels*overlap */ -#endif -}; - -int celt_encoder_get_size(int channels) -{ - CELTMode *mode = opus_custom_mode_create(48000, 960, NULL); - return opus_custom_encoder_get_size(mode, channels); -} - -OPUS_CUSTOM_NOSTATIC int opus_custom_encoder_get_size(const CELTMode *mode, int channels) -{ - int size = sizeof(struct CELTEncoder) - + (channels*mode->overlap-1)*sizeof(celt_sig) /* celt_sig in_mem[channels*mode->overlap]; */ - + channels*COMBFILTER_MAXPERIOD*sizeof(celt_sig) /* celt_sig prefilter_mem[channels*COMBFILTER_MAXPERIOD]; */ - + 3*channels*mode->nbEBands*sizeof(opus_val16); /* opus_val16 oldBandE[channels*mode->nbEBands]; */ - /* opus_val16 oldLogE[channels*mode->nbEBands]; */ - /* opus_val16 oldLogE2[channels*mode->nbEBands]; */ -#ifdef RESYNTH - size += channels*mode->overlap*sizeof(celt_sig); /* celt_sig overlap_mem[channels*mode->nbEBands]; */ -#endif - return size; -} - -#ifdef CUSTOM_MODES -CELTEncoder *opus_custom_encoder_create(const CELTMode *mode, int channels, int *error) -{ - int ret; - CELTEncoder *st = (CELTEncoder *)opus_alloc(opus_custom_encoder_get_size(mode, channels)); - /* init will handle the NULL case */ - ret = opus_custom_encoder_init(st, mode, channels); - if (ret != OPUS_OK) - { - opus_custom_encoder_destroy(st); - st = NULL; - } - if (error) - *error = ret; - return st; -} -#endif /* CUSTOM_MODES */ - -int celt_encoder_init(CELTEncoder *st, opus_int32 sampling_rate, int channels) -{ - int ret; - ret = opus_custom_encoder_init(st, opus_custom_mode_create(48000, 960, NULL), channels); - if (ret != OPUS_OK) - return ret; - st->upsample = resampling_factor(sampling_rate); - return OPUS_OK; -} - -OPUS_CUSTOM_NOSTATIC int opus_custom_encoder_init(CELTEncoder *st, const CELTMode *mode, int channels) -{ - if (channels < 0 || channels > 2) - return OPUS_BAD_ARG; - - if (st==NULL || mode==NULL) - return OPUS_ALLOC_FAIL; - - OPUS_CLEAR((char*)st, opus_custom_encoder_get_size(mode, channels)); - - st->mode = mode; - st->overlap = mode->overlap; - st->stream_channels = st->channels = channels; - - st->upsample = 1; - st->start = 0; - st->end = st->mode->effEBands; - st->signalling = 1; - - st->constrained_vbr = 1; - st->clip = 1; - - st->bitrate = OPUS_BITRATE_MAX; - st->vbr = 0; - st->force_intra = 0; - st->complexity = 5; - st->lsb_depth=24; - - opus_custom_encoder_ctl(st, OPUS_RESET_STATE); - - return OPUS_OK; -} - -#ifdef CUSTOM_MODES -void opus_custom_encoder_destroy(CELTEncoder *st) -{ - opus_free(st); -} -#endif /* CUSTOM_MODES */ - -static inline opus_val16 SIG2WORD16(celt_sig x) -{ -#ifdef FIXED_POINT - x = PSHR32(x, SIG_SHIFT); - x = MAX32(x, -32768); - x = MIN32(x, 32767); - return EXTRACT16(x); -#else - return (opus_val16)x; -#endif -} - -static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int C, - int overlap) -{ - int i; - VARDECL(opus_val16, tmp); - opus_val32 mem0=0,mem1=0; - int is_transient = 0; - int block; - int N; - VARDECL(opus_val16, bins); - SAVE_STACK; - ALLOC(tmp, len, opus_val16); - - block = overlap/2; - N=len/block; - ALLOC(bins, N, opus_val16); - if (C==1) - { - for (i=0;i<len;i++) - tmp[i] = SHR32(in[i],SIG_SHIFT); - } else { - for (i=0;i<len;i++) - tmp[i] = SHR32(ADD32(in[i],in[i+len]), SIG_SHIFT+1); - } - - /* High-pass filter: (1 - 2*z^-1 + z^-2) / (1 - z^-1 + .5*z^-2) */ - for (i=0;i<len;i++) - { - opus_val32 x,y; - x = tmp[i]; - y = ADD32(mem0, x); -#ifdef FIXED_POINT - mem0 = mem1 + y - SHL32(x,1); - mem1 = x - SHR32(y,1); -#else - mem0 = mem1 + y - 2*x; - mem1 = x - .5f*y; -#endif - tmp[i] = EXTRACT16(SHR32(y,2)); - } - /* First few samples are bad because we don't propagate the memory */ - for (i=0;i<12;i++) - tmp[i] = 0; - - for (i=0;i<N;i++) - { - int j; - opus_val16 max_abs=0; - for (j=0;j<block;j++) - max_abs = MAX16(max_abs, ABS16(tmp[i*block+j])); - bins[i] = max_abs; - } - for (i=0;i<N;i++) - { - int j; - int conseq=0; - opus_val16 t1, t2, t3; - - t1 = MULT16_16_Q15(QCONST16(.15f, 15), bins[i]); - t2 = MULT16_16_Q15(QCONST16(.4f, 15), bins[i]); - t3 = MULT16_16_Q15(QCONST16(.15f, 15), bins[i]); - for (j=0;j<i;j++) - { - if (bins[j] < t1) - conseq++; - if (bins[j] < t2) - conseq++; - else - conseq = 0; - } - if (conseq>=3) - is_transient=1; - conseq = 0; - for (j=i+1;j<N;j++) - { - if (bins[j] < t3) - conseq++; - else - conseq = 0; - } - if (conseq>=7) - is_transient=1; - } - RESTORE_STACK; -#ifdef FUZZING - is_transient = rand()&0x1; -#endif - return is_transient; -} - -/** Apply window and compute the MDCT for all sub-frames and - all channels in a frame */ -static void compute_mdcts(const CELTMode *mode, int shortBlocks, celt_sig * OPUS_RESTRICT in, celt_sig * OPUS_RESTRICT out, int C, int LM) -{ - if (C==1 && !shortBlocks) - { - const int overlap = OVERLAP(mode); - clt_mdct_forward(&mode->mdct, in, out, mode->window, overlap, mode->maxLM-LM, 1); - } else { - const int overlap = OVERLAP(mode); - int N = mode->shortMdctSize<<LM; - int B = 1; - int b, c; - if (shortBlocks) - { - N = mode->shortMdctSize; - B = shortBlocks; - } - c=0; do { - for (b=0;b<B;b++) - { - /* Interleaving the sub-frames while doing the MDCTs */ - clt_mdct_forward(&mode->mdct, in+c*(B*N+overlap)+b*N, &out[b+c*N*B], mode->window, overlap, shortBlocks ? mode->maxLM : mode->maxLM-LM, B); - } - } while (++c<C); - } -} - -/** Compute the IMDCT and apply window for all sub-frames and - all channels in a frame */ -static void compute_inv_mdcts(const CELTMode *mode, int shortBlocks, celt_sig *X, - celt_sig * OPUS_RESTRICT out_mem[], - celt_sig * OPUS_RESTRICT overlap_mem[], int C, int LM) -{ - int c; - const int N = mode->shortMdctSize<<LM; - const int overlap = OVERLAP(mode); - VARDECL(opus_val32, x); - SAVE_STACK; - - ALLOC(x, N+overlap, opus_val32); - c=0; do { - int j; - int b; - int N2 = N; - int B = 1; - - if (shortBlocks) - { - N2 = mode->shortMdctSize; - B = shortBlocks; - } - /* Prevents problems from the imdct doing the overlap-add */ - OPUS_CLEAR(x, overlap); - - for (b=0;b<B;b++) - { - /* IMDCT on the interleaved the sub-frames */ - clt_mdct_backward(&mode->mdct, &X[b+c*N2*B], x+N2*b, mode->window, overlap, shortBlocks ? mode->maxLM : mode->maxLM-LM, B); - } - - for (j=0;j<overlap;j++) - out_mem[c][j] = x[j] + overlap_mem[c][j]; - for (;j<N;j++) - out_mem[c][j] = x[j]; - for (j=0;j<overlap;j++) - overlap_mem[c][j] = x[N+j]; - } while (++c<C); - RESTORE_STACK; -} - -static void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, const opus_val16 *coef, celt_sig *mem) -{ - int c; - int count=0; - c=0; do { - int j; - celt_sig * OPUS_RESTRICT x; - opus_val16 * OPUS_RESTRICT y; - celt_sig m = mem[c]; - x =in[c]; - y = pcm+c; - for (j=0;j<N;j++) - { - celt_sig tmp = *x + m; - m = MULT16_32_Q15(coef[0], tmp) - - MULT16_32_Q15(coef[1], *x); - tmp = SHL32(MULT16_32_Q15(coef[3], tmp), 2); - x++; - /* Technically the store could be moved outside of the if because - the stores we don't want will just be overwritten */ - if (count==0) - *y = SCALEOUT(SIG2WORD16(tmp)); - if (++count==downsample) - { - y+=C; - count=0; - } - } - mem[c] = m; - } while (++c<C); -} - -static void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N, - opus_val16 g0, opus_val16 g1, int tapset0, int tapset1, - const opus_val16 *window, int overlap) -{ - int i; - /* printf ("%d %d %f %f\n", T0, T1, g0, g1); */ - opus_val16 g00, g01, g02, g10, g11, g12; - static const opus_val16 gains[3][3] = { - {QCONST16(0.3066406250f, 15), QCONST16(0.2170410156f, 15), QCONST16(0.1296386719f, 15)}, - {QCONST16(0.4638671875f, 15), QCONST16(0.2680664062f, 15), QCONST16(0.f, 15)}, - {QCONST16(0.7998046875f, 15), QCONST16(0.1000976562f, 15), QCONST16(0.f, 15)}}; - g00 = MULT16_16_Q15(g0, gains[tapset0][0]); - g01 = MULT16_16_Q15(g0, gains[tapset0][1]); - g02 = MULT16_16_Q15(g0, gains[tapset0][2]); - g10 = MULT16_16_Q15(g1, gains[tapset1][0]); - g11 = MULT16_16_Q15(g1, gains[tapset1][1]); - g12 = MULT16_16_Q15(g1, gains[tapset1][2]); - for (i=0;i<overlap;i++) - { - opus_val16 f; - f = MULT16_16_Q15(window[i],window[i]); - y[i] = x[i] - + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g00),x[i-T0]) - + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g01),x[i-T0-1]) - + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g01),x[i-T0+1]) - + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g02),x[i-T0-2]) - + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g02),x[i-T0+2]) - + MULT16_32_Q15(MULT16_16_Q15(f,g10),x[i-T1]) - + MULT16_32_Q15(MULT16_16_Q15(f,g11),x[i-T1-1]) - + MULT16_32_Q15(MULT16_16_Q15(f,g11),x[i-T1+1]) - + MULT16_32_Q15(MULT16_16_Q15(f,g12),x[i-T1-2]) - + MULT16_32_Q15(MULT16_16_Q15(f,g12),x[i-T1+2]); - - } - for (i=overlap;i<N;i++) - y[i] = x[i] - + MULT16_32_Q15(g10,x[i-T1]) - + MULT16_32_Q15(g11,x[i-T1-1]) - + MULT16_32_Q15(g11,x[i-T1+1]) - + MULT16_32_Q15(g12,x[i-T1-2]) - + MULT16_32_Q15(g12,x[i-T1+2]); -} - -static const signed char tf_select_table[4][8] = { - {0, -1, 0, -1, 0,-1, 0,-1}, - {0, -1, 0, -2, 1, 0, 1,-1}, - {0, -2, 0, -3, 2, 0, 1,-1}, - {0, -2, 0, -3, 3, 0, 1,-1}, -}; - -static opus_val32 l1_metric(const celt_norm *tmp, int N, int LM, int width) -{ - int i, j; - static const opus_val16 sqrtM_1[4] = {Q15ONE, QCONST16(.70710678f,15), QCONST16(0.5f,15), QCONST16(0.35355339f,15)}; - opus_val32 L1; - opus_val16 bias; - L1=0; - for (i=0;i<1<<LM;i++) - { - opus_val32 L2 = 0; - for (j=0;j<N>>LM;j++) - L2 = MAC16_16(L2, tmp[(j<<LM)+i], tmp[(j<<LM)+i]); - L1 += celt_sqrt(L2); - } - L1 = MULT16_32_Q15(sqrtM_1[LM], L1); - if (width==1) - bias = QCONST16(.12f,15)*LM; - else if (width==2) - bias = QCONST16(.05f,15)*LM; - else - bias = QCONST16(.02f,15)*LM; - L1 = MAC16_32_Q15(L1, bias, L1); - return L1; -} - -static int tf_analysis(const CELTMode *m, int len, int C, int isTransient, - int *tf_res, int nbCompressedBytes, celt_norm *X, int N0, int LM, - int start, int *tf_sum) -{ - int i; - VARDECL(int, metric); - int cost0; - int cost1; - VARDECL(int, path0); - VARDECL(int, path1); - VARDECL(celt_norm, tmp); - int lambda; - int tf_select=0; - SAVE_STACK; - - if (nbCompressedBytes<15*C || start!=0) - { - *tf_sum = 0; - for (i=0;i<len;i++) - tf_res[i] = isTransient; - return 0; - } - if (nbCompressedBytes<40) - lambda = 12; - else if (nbCompressedBytes<60) - lambda = 6; - else if (nbCompressedBytes<100) - lambda = 4; - else - lambda = 3; - - ALLOC(metric, len, int); - ALLOC(tmp, (m->eBands[len]-m->eBands[len-1])<<LM, celt_norm); - ALLOC(path0, len, int); - ALLOC(path1, len, int); - - *tf_sum = 0; - for (i=0;i<len;i++) - { - int j, k, N; - opus_val32 L1, best_L1; - int best_level=0; - N = (m->eBands[i+1]-m->eBands[i])<<LM; - for (j=0;j<N;j++) - tmp[j] = X[j+(m->eBands[i]<<LM)]; - /* Just add the right channel if we're in stereo */ - if (C==2) - for (j=0;j<N;j++) - tmp[j] = ADD16(SHR16(tmp[j], 1),SHR16(X[N0+j+(m->eBands[i]<<LM)], 1)); - L1 = l1_metric(tmp, N, isTransient ? LM : 0, N>>LM); - best_L1 = L1; - /*printf ("%f ", L1);*/ - for (k=0;k<LM;k++) - { - int B; - - if (isTransient) - B = (LM-k-1); - else - B = k+1; - - if (isTransient) - haar1(tmp, N>>(LM-k), 1<<(LM-k)); - else - haar1(tmp, N>>k, 1<<k); - - L1 = l1_metric(tmp, N, B, N>>LM); - - if (L1 < best_L1) - { - best_L1 = L1; - best_level = k+1; - } - } - /*printf ("%d ", isTransient ? LM-best_level : best_level);*/ - if (isTransient) - metric[i] = best_level; - else - metric[i] = -best_level; - *tf_sum += metric[i]; - } - /*printf("\n");*/ - /* NOTE: Future optimized implementations could detect extreme transients and set - tf_select = 1 but so far we have not found a reliable way of making this useful */ - tf_select = 0; - - cost0 = 0; - cost1 = isTransient ? 0 : lambda; - /* Viterbi forward pass */ - for (i=1;i<len;i++) - { - int curr0, curr1; - int from0, from1; - - from0 = cost0; - from1 = cost1 + lambda; - if (from0 < from1) - { - curr0 = from0; - path0[i]= 0; - } else { - curr0 = from1; - path0[i]= 1; - } - - from0 = cost0 + lambda; - from1 = cost1; - if (from0 < from1) - { - curr1 = from0; - path1[i]= 0; - } else { - curr1 = from1; - path1[i]= 1; - } - cost0 = curr0 + abs(metric[i]-tf_select_table[LM][4*isTransient+2*tf_select+0]); - cost1 = curr1 + abs(metric[i]-tf_select_table[LM][4*isTransient+2*tf_select+1]); - } - tf_res[len-1] = cost0 < cost1 ? 0 : 1; - /* Viterbi backward pass to check the decisions */ - for (i=len-2;i>=0;i--) - { - if (tf_res[i+1] == 1) - tf_res[i] = path1[i+1]; - else - tf_res[i] = path0[i+1]; - } - RESTORE_STACK; -#ifdef FUZZING - tf_select = rand()&0x1; - tf_res[0] = rand()&0x1; - for (i=1;i<len;i++) - tf_res[i] = tf_res[i-1] ^ ((rand()&0xF) == 0); -#endif - return tf_select; -} - -static void tf_encode(int start, int end, int isTransient, int *tf_res, int LM, int tf_select, ec_enc *enc) -{ - int curr, i; - int tf_select_rsv; - int tf_changed; - int logp; - opus_uint32 budget; - opus_uint32 tell; - budget = enc->storage*8; - tell = ec_tell(enc); - logp = isTransient ? 2 : 4; - /* Reserve space to code the tf_select decision. */ - tf_select_rsv = LM>0 && tell+logp+1 <= budget; - budget -= tf_select_rsv; - curr = tf_changed = 0; - for (i=start;i<end;i++) - { - if (tell+logp<=budget) - { - ec_enc_bit_logp(enc, tf_res[i] ^ curr, logp); - tell = ec_tell(enc); - curr = tf_res[i]; - tf_changed |= curr; - } - else - tf_res[i] = curr; - logp = isTransient ? 4 : 5; - } - /* Only code tf_select if it would actually make a difference. */ - if (tf_select_rsv && - tf_select_table[LM][4*isTransient+0+tf_changed]!= - tf_select_table[LM][4*isTransient+2+tf_changed]) - ec_enc_bit_logp(enc, tf_select, 1); - else - tf_select = 0; - for (i=start;i<end;i++) - tf_res[i] = tf_select_table[LM][4*isTransient+2*tf_select+tf_res[i]]; - /*printf("%d %d ", isTransient, tf_select); for(i=0;i<end;i++)printf("%d ", tf_res[i]);printf("\n");*/ -} - -static void tf_decode(int start, int end, int isTransient, int *tf_res, int LM, ec_dec *dec) -{ - int i, curr, tf_select; - int tf_select_rsv; - int tf_changed; - int logp; - opus_uint32 budget; - opus_uint32 tell; - - budget = dec->storage*8; - tell = ec_tell(dec); - logp = isTransient ? 2 : 4; - tf_select_rsv = LM>0 && tell+logp+1<=budget; - budget -= tf_select_rsv; - tf_changed = curr = 0; - for (i=start;i<end;i++) - { - if (tell+logp<=budget) - { - curr ^= ec_dec_bit_logp(dec, logp); - tell = ec_tell(dec); - tf_changed |= curr; - } - tf_res[i] = curr; - logp = isTransient ? 4 : 5; - } - tf_select = 0; - if (tf_select_rsv && - tf_select_table[LM][4*isTransient+0+tf_changed] != - tf_select_table[LM][4*isTransient+2+tf_changed]) - { - tf_select = ec_dec_bit_logp(dec, 1); - } - for (i=start;i<end;i++) - { - tf_res[i] = tf_select_table[LM][4*isTransient+2*tf_select+tf_res[i]]; - } -} - -static void init_caps(const CELTMode *m,int *cap,int LM,int C) -{ - int i; - for (i=0;i<m->nbEBands;i++) - { - int N; - N=(m->eBands[i+1]-m->eBands[i])<<LM; - cap[i] = (m->cache.caps[m->nbEBands*(2*LM+C-1)+i]+64)*C*N>>2; - } -} - -static int alloc_trim_analysis(const CELTMode *m, const celt_norm *X, - const opus_val16 *bandLogE, int end, int LM, int C, int N0) -{ - int i; - opus_val32 diff=0; - int c; - int trim_index = 5; - if (C==2) - { - opus_val16 sum = 0; /* Q10 */ - /* Compute inter-channel correlation for low frequencies */ - for (i=0;i<8;i++) - { - int j; - opus_val32 partial = 0; - for (j=m->eBands[i]<<LM;j<m->eBands[i+1]<<LM;j++) - partial = MAC16_16(partial, X[j], X[N0+j]); - sum = ADD16(sum, EXTRACT16(SHR32(partial, 18))); - } - sum = MULT16_16_Q15(QCONST16(1.f/8, 15), sum); - /*printf ("%f\n", sum);*/ - if (sum > QCONST16(.995f,10)) - trim_index-=4; - else if (sum > QCONST16(.92f,10)) - trim_index-=3; - else if (sum > QCONST16(.85f,10)) - trim_index-=2; - else if (sum > QCONST16(.8f,10)) - trim_index-=1; - } - - /* Estimate spectral tilt */ - c=0; do { - for (i=0;i<end-1;i++) - { - diff += bandLogE[i+c*m->nbEBands]*(opus_int32)(2+2*i-m->nbEBands); - } - } while (++c<C); - /* We divide by two here to avoid making the tilt larger for stereo as a - result of a bug in the loop above */ - diff /= 2*C*(end-1); - /*printf("%f\n", diff);*/ - if (diff > QCONST16(2.f, DB_SHIFT)) - trim_index--; - if (diff > QCONST16(8.f, DB_SHIFT)) - trim_index--; - if (diff < -QCONST16(4.f, DB_SHIFT)) - trim_index++; - if (diff < -QCONST16(10.f, DB_SHIFT)) - trim_index++; - - if (trim_index<0) - trim_index = 0; - if (trim_index>10) - trim_index = 10; -#ifdef FUZZING - trim_index = rand()%11; -#endif - return trim_index; -} - -static int stereo_analysis(const CELTMode *m, const celt_norm *X, - int LM, int N0) -{ - int i; - int thetas; - opus_val32 sumLR = EPSILON, sumMS = EPSILON; - - /* Use the L1 norm to model the entropy of the L/R signal vs the M/S signal */ - for (i=0;i<13;i++) - { - int j; - for (j=m->eBands[i]<<LM;j<m->eBands[i+1]<<LM;j++) - { - opus_val32 L, R, M, S; - /* We cast to 32-bit first because of the -32768 case */ - L = EXTEND32(X[j]); - R = EXTEND32(X[N0+j]); - M = ADD32(L, R); - S = SUB32(L, R); - sumLR = ADD32(sumLR, ADD32(ABS32(L), ABS32(R))); - sumMS = ADD32(sumMS, ADD32(ABS32(M), ABS32(S))); - } - } - sumMS = MULT16_32_Q15(QCONST16(0.707107f, 15), sumMS); - thetas = 13; - /* We don't need thetas for lower bands with LM<=1 */ - if (LM<=1) - thetas -= 8; - return MULT16_32_Q15((m->eBands[13]<<(LM+1))+thetas, sumMS) - > MULT16_32_Q15(m->eBands[13]<<(LM+1), sumLR); -} - -int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes, ec_enc *enc) -{ - int i, c, N; - opus_int32 bits; - ec_enc _enc; - VARDECL(celt_sig, in); - VARDECL(celt_sig, freq); - VARDECL(celt_norm, X); - VARDECL(celt_ener, bandE); - VARDECL(opus_val16, bandLogE); - VARDECL(int, fine_quant); - VARDECL(opus_val16, error); - VARDECL(int, pulses); - VARDECL(int, cap); - VARDECL(int, offsets); - VARDECL(int, fine_priority); - VARDECL(int, tf_res); - VARDECL(unsigned char, collapse_masks); - celt_sig *prefilter_mem; - opus_val16 *oldBandE, *oldLogE, *oldLogE2; - int shortBlocks=0; - int isTransient=0; - const int CC = st->channels; - const int C = st->stream_channels; - int LM, M; - int tf_select; - int nbFilledBytes, nbAvailableBytes; - int effEnd; - int codedBands; - int tf_sum; - int alloc_trim; - int pitch_index=COMBFILTER_MINPERIOD; - opus_val16 gain1 = 0; - int intensity=0; - int dual_stereo=0; - int effectiveBytes; - opus_val16 pf_threshold; - int dynalloc_logp; - opus_int32 vbr_rate; - opus_int32 total_bits; - opus_int32 total_boost; - opus_int32 balance; - opus_int32 tell; - int prefilter_tapset=0; - int pf_on; - int anti_collapse_rsv; - int anti_collapse_on=0; - int silence=0; - ALLOC_STACK; - - if (nbCompressedBytes<2 || pcm==NULL) - return OPUS_BAD_ARG; - - frame_size *= st->upsample; - for (LM=0;LM<=st->mode->maxLM;LM++) - if (st->mode->shortMdctSize<<LM==frame_size) - break; - if (LM>st->mode->maxLM) - return OPUS_BAD_ARG; - M=1<<LM; - N = M*st->mode->shortMdctSize; - - prefilter_mem = st->in_mem+CC*(st->overlap); - oldBandE = (opus_val16*)(st->in_mem+CC*(st->overlap+COMBFILTER_MAXPERIOD)); - oldLogE = oldBandE + CC*st->mode->nbEBands; - oldLogE2 = oldLogE + CC*st->mode->nbEBands; - - if (enc==NULL) - { - tell=1; - nbFilledBytes=0; - } else { - tell=ec_tell(enc); - nbFilledBytes=(tell+4)>>3; - } - -#ifdef CUSTOM_MODES - if (st->signalling && enc==NULL) - { - int tmp = (st->mode->effEBands-st->end)>>1; - st->end = IMAX(1, st->mode->effEBands-tmp); - compressed[0] = tmp<<5; - compressed[0] |= LM<<3; - compressed[0] |= (C==2)<<2; - /* Convert "standard mode" to Opus header */ - if (st->mode->Fs==48000 && st->mode->shortMdctSize==120) - { - int c0 = toOpus(compressed[0]); - if (c0<0) - return OPUS_BAD_ARG; - compressed[0] = c0; - } - compressed++; - nbCompressedBytes--; - } -#else - celt_assert(st->signalling==0); -#endif - - /* Can't produce more than 1275 output bytes */ - nbCompressedBytes = IMIN(nbCompressedBytes,1275); - nbAvailableBytes = nbCompressedBytes - nbFilledBytes; - - if (st->vbr && st->bitrate!=OPUS_BITRATE_MAX) - { - opus_int32 den=st->mode->Fs>>BITRES; - vbr_rate=(st->bitrate*frame_size+(den>>1))/den; -#ifdef CUSTOM_MODES - if (st->signalling) - vbr_rate -= 8<<BITRES; -#endif - effectiveBytes = vbr_rate>>(3+BITRES); - } else { - opus_int32 tmp; - vbr_rate = 0; - tmp = st->bitrate*frame_size; - if (tell>1) - tmp += tell; - if (st->bitrate!=OPUS_BITRATE_MAX) - nbCompressedBytes = IMAX(2, IMIN(nbCompressedBytes, - (tmp+4*st->mode->Fs)/(8*st->mode->Fs)-!!st->signalling)); - effectiveBytes = nbCompressedBytes; - } - - if (enc==NULL) - { - ec_enc_init(&_enc, compressed, nbCompressedBytes); - enc = &_enc; - } - - if (vbr_rate>0) - { - /* Computes the max bit-rate allowed in VBR mode to avoid violating the - target rate and buffering. - We must do this up front so that bust-prevention logic triggers - correctly if we don't have enough bits. */ - if (st->constrained_vbr) - { - opus_int32 vbr_bound; - opus_int32 max_allowed; - /* We could use any multiple of vbr_rate as bound (depending on the - delay). - This is clamped to ensure we use at least two bytes if the encoder - was entirely empty, but to allow 0 in hybrid mode. */ - vbr_bound = vbr_rate; - max_allowed = IMIN(IMAX(tell==1?2:0, - (vbr_rate+vbr_bound-st->vbr_reservoir)>>(BITRES+3)), - nbAvailableBytes); - if(max_allowed < nbAvailableBytes) - { - nbCompressedBytes = nbFilledBytes+max_allowed; - nbAvailableBytes = max_allowed; - ec_enc_shrink(enc, nbCompressedBytes); - } - } - } - total_bits = nbCompressedBytes*8; - - effEnd = st->end; - if (effEnd > st->mode->effEBands) - effEnd = st->mode->effEBands; - - ALLOC(in, CC*(N+st->overlap), celt_sig); - - /* Find pitch period and gain */ - { - VARDECL(celt_sig, _pre); - celt_sig *pre[2]; - SAVE_STACK; - ALLOC(_pre, CC*(N+COMBFILTER_MAXPERIOD), celt_sig); - - pre[0] = _pre; - pre[1] = _pre + (N+COMBFILTER_MAXPERIOD); - - silence = 1; - c=0; do { - int count = 0; - const opus_val16 * OPUS_RESTRICT pcmp = pcm+c; - celt_sig * OPUS_RESTRICT inp = in+c*(N+st->overlap)+st->overlap; - - for (i=0;i<N;i++) - { - celt_sig x, tmp; - - x = SCALEIN(*pcmp); -#ifndef FIXED_POINT - if (!(x==x)) - x = 0; - if (st->clip) - x = MAX32(-65536.f, MIN32(65536.f,x)); -#endif - if (++count==st->upsample) - { - count=0; - pcmp+=CC; - } else { - x = 0; - } - /* Apply pre-emphasis */ - tmp = MULT16_16(st->mode->preemph[2], x); - *inp = tmp + st->preemph_memE[c]; - st->preemph_memE[c] = MULT16_32_Q15(st->mode->preemph[1], *inp) - - MULT16_32_Q15(st->mode->preemph[0], tmp); - silence = silence && *inp == 0; - inp++; - } - OPUS_COPY(pre[c], prefilter_mem+c*COMBFILTER_MAXPERIOD, COMBFILTER_MAXPERIOD); - OPUS_COPY(pre[c]+COMBFILTER_MAXPERIOD, in+c*(N+st->overlap)+st->overlap, N); - } while (++c<CC); - -#ifdef FUZZING - if ((rand()&0x3F)==0) - silence = 1; -#endif - if (tell==1) - ec_enc_bit_logp(enc, silence, 15); - else - silence=0; - if (silence) - { - /*In VBR mode there is no need to send more than the minimum. */ - if (vbr_rate>0) - { - effectiveBytes=nbCompressedBytes=IMIN(nbCompressedBytes, nbFilledBytes+2); - total_bits=nbCompressedBytes*8; - nbAvailableBytes=2; - ec_enc_shrink(enc, nbCompressedBytes); - } - /* Pretend we've filled all the remaining bits with zeros - (that's what the initialiser did anyway) */ - tell = nbCompressedBytes*8; - enc->nbits_total+=tell-ec_tell(enc); - } - if (nbAvailableBytes>12*C && st->start==0 && !silence && !st->disable_pf && st->complexity >= 5) - { - VARDECL(opus_val16, pitch_buf); - ALLOC(pitch_buf, (COMBFILTER_MAXPERIOD+N)>>1, opus_val16); - - pitch_downsample(pre, pitch_buf, COMBFILTER_MAXPERIOD+N, CC); - pitch_search(pitch_buf+(COMBFILTER_MAXPERIOD>>1), pitch_buf, N, - COMBFILTER_MAXPERIOD-COMBFILTER_MINPERIOD, &pitch_index); - pitch_index = COMBFILTER_MAXPERIOD-pitch_index; - - gain1 = remove_doubling(pitch_buf, COMBFILTER_MAXPERIOD, COMBFILTER_MINPERIOD, - N, &pitch_index, st->prefilter_period, st->prefilter_gain); - if (pitch_index > COMBFILTER_MAXPERIOD-2) - pitch_index = COMBFILTER_MAXPERIOD-2; - gain1 = MULT16_16_Q15(QCONST16(.7f,15),gain1); - if (st->loss_rate>2) - gain1 = HALF32(gain1); - if (st->loss_rate>4) - gain1 = HALF32(gain1); - if (st->loss_rate>8) - gain1 = 0; - prefilter_tapset = st->tapset_decision; - } else { - gain1 = 0; - } - - /* Gain threshold for enabling the prefilter/postfilter */ - pf_threshold = QCONST16(.2f,15); - - /* Adjusting the threshold based on rate and continuity */ - if (abs(pitch_index-st->prefilter_period)*10>pitch_index) - pf_threshold += QCONST16(.2f,15); - if (nbAvailableBytes<25) - pf_threshold += QCONST16(.1f,15); - if (nbAvailableBytes<35) - pf_threshold += QCONST16(.1f,15); - if (st->prefilter_gain > QCONST16(.4f,15)) - pf_threshold -= QCONST16(.1f,15); - if (st->prefilter_gain > QCONST16(.55f,15)) - pf_threshold -= QCONST16(.1f,15); - - /* Hard threshold at 0.2 */ - pf_threshold = MAX16(pf_threshold, QCONST16(.2f,15)); - if (gain1<pf_threshold) - { - if(st->start==0 && tell+16<=total_bits) - ec_enc_bit_logp(enc, 0, 1); - gain1 = 0; - pf_on = 0; - } else { - /*This block is not gated by a total bits check only because - of the nbAvailableBytes check above.*/ - int qg; - int octave; - - if (ABS16(gain1-st->prefilter_gain)<QCONST16(.1f,15)) - gain1=st->prefilter_gain; - -#ifdef FIXED_POINT - qg = ((gain1+1536)>>10)/3-1; -#else - qg = (int)floor(.5f+gain1*32/3)-1; -#endif - qg = IMAX(0, IMIN(7, qg)); - ec_enc_bit_logp(enc, 1, 1); - pitch_index += 1; - octave = EC_ILOG(pitch_index)-5; - ec_enc_uint(enc, octave, 6); - ec_enc_bits(enc, pitch_index-(16<<octave), 4+octave); - pitch_index -= 1; - ec_enc_bits(enc, qg, 3); - if (ec_tell(enc)+2<=total_bits) - ec_enc_icdf(enc, prefilter_tapset, tapset_icdf, 2); - else - prefilter_tapset = 0; - gain1 = QCONST16(0.09375f,15)*(qg+1); - pf_on = 1; - } - /*printf("%d %f\n", pitch_index, gain1);*/ - - c=0; do { - int offset = st->mode->shortMdctSize-st->mode->overlap; - st->prefilter_period=IMAX(st->prefilter_period, COMBFILTER_MINPERIOD); - OPUS_COPY(in+c*(N+st->overlap), st->in_mem+c*(st->overlap), st->overlap); - if (offset) - comb_filter(in+c*(N+st->overlap)+st->overlap, pre[c]+COMBFILTER_MAXPERIOD, - st->prefilter_period, st->prefilter_period, offset, -st->prefilter_gain, -st->prefilter_gain, - st->prefilter_tapset, st->prefilter_tapset, NULL, 0); - - comb_filter(in+c*(N+st->overlap)+st->overlap+offset, pre[c]+COMBFILTER_MAXPERIOD+offset, - st->prefilter_period, pitch_index, N-offset, -st->prefilter_gain, -gain1, - st->prefilter_tapset, prefilter_tapset, st->mode->window, st->mode->overlap); - OPUS_COPY(st->in_mem+c*(st->overlap), in+c*(N+st->overlap)+N, st->overlap); - - if (N>COMBFILTER_MAXPERIOD) - { - OPUS_MOVE(prefilter_mem+c*COMBFILTER_MAXPERIOD, pre[c]+N, COMBFILTER_MAXPERIOD); - } else { - OPUS_MOVE(prefilter_mem+c*COMBFILTER_MAXPERIOD, prefilter_mem+c*COMBFILTER_MAXPERIOD+N, COMBFILTER_MAXPERIOD-N); - OPUS_MOVE(prefilter_mem+c*COMBFILTER_MAXPERIOD+COMBFILTER_MAXPERIOD-N, pre[c]+COMBFILTER_MAXPERIOD, N); - } - } while (++c<CC); - - RESTORE_STACK; - } - - isTransient = 0; - shortBlocks = 0; - if (LM>0 && ec_tell(enc)+3<=total_bits) - { - if (st->complexity > 1) - { - isTransient = transient_analysis(in, N+st->overlap, CC, - st->overlap); - if (isTransient) - shortBlocks = M; - } - ec_enc_bit_logp(enc, isTransient, 3); - } - - ALLOC(freq, CC*N, celt_sig); /**< Interleaved signal MDCTs */ - ALLOC(bandE,st->mode->nbEBands*CC, celt_ener); - ALLOC(bandLogE,st->mode->nbEBands*CC, opus_val16); - /* Compute MDCTs */ - compute_mdcts(st->mode, shortBlocks, in, freq, CC, LM); - - if (CC==2&&C==1) - { - for (i=0;i<N;i++) - freq[i] = ADD32(HALF32(freq[i]), HALF32(freq[N+i])); - } - if (st->upsample != 1) - { - c=0; do - { - int bound = N/st->upsample; - for (i=0;i<bound;i++) - freq[c*N+i] *= st->upsample; - for (;i<N;i++) - freq[c*N+i] = 0; - } while (++c<C); - } - ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ - - compute_band_energies(st->mode, freq, bandE, effEnd, C, M); - - amp2Log2(st->mode, effEnd, st->end, bandE, bandLogE, C); - - /* Band normalisation */ - normalise_bands(st->mode, freq, X, bandE, effEnd, C, M); - - ALLOC(tf_res, st->mode->nbEBands, int); - tf_select = tf_analysis(st->mode, effEnd, C, isTransient, tf_res, effectiveBytes, X, N, LM, st->start, &tf_sum); - for (i=effEnd;i<st->end;i++) - tf_res[i] = tf_res[effEnd-1]; - - ALLOC(error, C*st->mode->nbEBands, opus_val16); - quant_coarse_energy(st->mode, st->start, st->end, effEnd, bandLogE, - oldBandE, total_bits, error, enc, - C, LM, nbAvailableBytes, st->force_intra, - &st->delayedIntra, st->complexity >= 4, st->loss_rate); - - tf_encode(st->start, st->end, isTransient, tf_res, LM, tf_select, enc); - - if (ec_tell(enc)+4<=total_bits) - { - if (shortBlocks || st->complexity < 3 - || nbAvailableBytes < 10*C || st->start!=0) - { - if (st->complexity == 0) - st->spread_decision = SPREAD_NONE; - else - st->spread_decision = SPREAD_NORMAL; - } else { - st->spread_decision = spreading_decision(st->mode, X, - &st->tonal_average, st->spread_decision, &st->hf_average, - &st->tapset_decision, pf_on&&!shortBlocks, effEnd, C, M); - } - ec_enc_icdf(enc, st->spread_decision, spread_icdf, 5); - } - - ALLOC(cap, st->mode->nbEBands, int); - ALLOC(offsets, st->mode->nbEBands, int); - - init_caps(st->mode,cap,LM,C); - for (i=0;i<st->mode->nbEBands;i++) - offsets[i] = 0; - /* Dynamic allocation code */ - /* Make sure that dynamic allocation can't make us bust the budget */ - if (effectiveBytes > 50 && LM>=1) - { - int t1, t2; - if (LM <= 1) - { - t1 = 3; - t2 = 5; - } else { - t1 = 2; - t2 = 4; - } - for (i=st->start+1;i<st->end-1;i++) - { - opus_val32 d2; - d2 = 2*bandLogE[i]-bandLogE[i-1]-bandLogE[i+1]; - if (C==2) - d2 = HALF32(d2 + 2*bandLogE[i+st->mode->nbEBands]- - bandLogE[i-1+st->mode->nbEBands]-bandLogE[i+1+st->mode->nbEBands]); -#ifdef FUZZING - if((rand()&0xF)==0) - { - offsets[i] += 1; - if((rand()&0x3)==0) - offsets[i] += 1+(rand()&0x3); - } -#else - if (d2 > SHL16(t1,DB_SHIFT)) - offsets[i] += 1; - if (d2 > SHL16(t2,DB_SHIFT)) - offsets[i] += 1; -#endif - } - } - dynalloc_logp = 6; - total_bits<<=BITRES; - total_boost = 0; - tell = ec_tell_frac(enc); - for (i=st->start;i<st->end;i++) - { - int width, quanta; - int dynalloc_loop_logp; - int boost; - int j; - width = C*(st->mode->eBands[i+1]-st->mode->eBands[i])<<LM; - /* quanta is 6 bits, but no more than 1 bit/sample - and no less than 1/8 bit/sample */ - quanta = IMIN(width<<BITRES, IMAX(6<<BITRES, width)); - dynalloc_loop_logp = dynalloc_logp; - boost = 0; - for (j = 0; tell+(dynalloc_loop_logp<<BITRES) < total_bits-total_boost - && boost < cap[i]; j++) - { - int flag; - flag = j<offsets[i]; - ec_enc_bit_logp(enc, flag, dynalloc_loop_logp); - tell = ec_tell_frac(enc); - if (!flag) - break; - boost += quanta; - total_boost += quanta; - dynalloc_loop_logp = 1; - } - /* Making dynalloc more likely */ - if (j) - dynalloc_logp = IMAX(2, dynalloc_logp-1); - offsets[i] = boost; - } - alloc_trim = 5; - if (tell+(6<<BITRES) <= total_bits - total_boost) - { - alloc_trim = alloc_trim_analysis(st->mode, X, bandLogE, - st->end, LM, C, N); - ec_enc_icdf(enc, alloc_trim, trim_icdf, 7); - tell = ec_tell_frac(enc); - } - - /* Variable bitrate */ - if (vbr_rate>0) - { - opus_val16 alpha; - opus_int32 delta; - /* The target rate in 8th bits per frame */ - opus_int32 target; - opus_int32 min_allowed; - int lm_diff = st->mode->maxLM - LM; - - /* Don't attempt to use more than 510 kb/s, even for frames smaller than 20 ms. - The CELT allocator will just not be able to use more than that anyway. */ - nbCompressedBytes = IMIN(nbCompressedBytes,1275>>(3-LM)); - target = vbr_rate + (st->vbr_offset>>lm_diff) - ((40*C+20)<<BITRES); - - /* Shortblocks get a large boost in bitrate, but since they - are uncommon long blocks are not greatly affected */ - if (shortBlocks || tf_sum < -2*(st->end-st->start)) - target = 7*target/4; - else if (tf_sum < -(st->end-st->start)) - target = 3*target/2; - else if (M > 1) - target-=(target+14)/28; - - /* The current offset is removed from the target and the space used - so far is added*/ - target=target+tell; - - /* In VBR mode the frame size must not be reduced so much that it would - result in the encoder running out of bits. - The margin of 2 bytes ensures that none of the bust-prevention logic - in the decoder will have triggered so far. */ - min_allowed = ((tell+total_boost+(1<<(BITRES+3))-1)>>(BITRES+3)) + 2 - nbFilledBytes; - - nbAvailableBytes = (target+(1<<(BITRES+2)))>>(BITRES+3); - nbAvailableBytes = IMAX(min_allowed,nbAvailableBytes); - nbAvailableBytes = IMIN(nbCompressedBytes,nbAvailableBytes+nbFilledBytes) - nbFilledBytes; - - /* By how much did we "miss" the target on that frame */ - delta = target - vbr_rate; - - target=nbAvailableBytes<<(BITRES+3); - - /*If the frame is silent we don't adjust our drift, otherwise - the encoder will shoot to very high rates after hitting a - span of silence, but we do allow the bitres to refill. - This means that we'll undershoot our target in CVBR/VBR modes - on files with lots of silence. */ - if(silence) - { - nbAvailableBytes = 2; - target = 2*8<<BITRES; - delta = 0; - } - - if (st->vbr_count < 970) - { - st->vbr_count++; - alpha = celt_rcp(SHL32(EXTEND32(st->vbr_count+20),16)); - } else - alpha = QCONST16(.001f,15); - /* How many bits have we used in excess of what we're allowed */ - if (st->constrained_vbr) - st->vbr_reservoir += target - vbr_rate; - /*printf ("%d\n", st->vbr_reservoir);*/ - - /* Compute the offset we need to apply in order to reach the target */ - st->vbr_drift += (opus_int32)MULT16_32_Q15(alpha,(delta*(1<<lm_diff))-st->vbr_offset-st->vbr_drift); - st->vbr_offset = -st->vbr_drift; - /*printf ("%d\n", st->vbr_drift);*/ - - if (st->constrained_vbr && st->vbr_reservoir < 0) - { - /* We're under the min value -- increase rate */ - int adjust = (-st->vbr_reservoir)/(8<<BITRES); - /* Unless we're just coding silence */ - nbAvailableBytes += silence?0:adjust; - st->vbr_reservoir = 0; - /*printf ("+%d\n", adjust);*/ - } - nbCompressedBytes = IMIN(nbCompressedBytes,nbAvailableBytes+nbFilledBytes); - /* This moves the raw bits to take into account the new compressed size */ - ec_enc_shrink(enc, nbCompressedBytes); - } - if (C==2) - { - int effectiveRate; - - /* Always use MS for 2.5 ms frames until we can do a better analysis */ - if (LM!=0) - dual_stereo = stereo_analysis(st->mode, X, LM, N); - - /* Account for coarse energy */ - effectiveRate = (8*effectiveBytes - 80)>>LM; - - /* effectiveRate in kb/s */ - effectiveRate = 2*effectiveRate/5; - if (effectiveRate<35) - intensity = 8; - else if (effectiveRate<50) - intensity = 12; - else if (effectiveRate<68) - intensity = 16; - else if (effectiveRate<84) - intensity = 18; - else if (effectiveRate<102) - intensity = 19; - else if (effectiveRate<130) - intensity = 20; - else - intensity = 100; - intensity = IMIN(st->end,IMAX(st->start, intensity)); - } - - /* Bit allocation */ - ALLOC(fine_quant, st->mode->nbEBands, int); - ALLOC(pulses, st->mode->nbEBands, int); - ALLOC(fine_priority, st->mode->nbEBands, int); - - /* bits = packet size - where we are - safety*/ - bits = (((opus_int32)nbCompressedBytes*8)<<BITRES) - ec_tell_frac(enc) - 1; - anti_collapse_rsv = isTransient&&LM>=2&&bits>=((LM+2)<<BITRES) ? (1<<BITRES) : 0; - bits -= anti_collapse_rsv; - codedBands = compute_allocation(st->mode, st->start, st->end, offsets, cap, - alloc_trim, &intensity, &dual_stereo, bits, &balance, pulses, - fine_quant, fine_priority, C, LM, enc, 1, st->lastCodedBands); - st->lastCodedBands = codedBands; - - quant_fine_energy(st->mode, st->start, st->end, oldBandE, error, fine_quant, enc, C); - -#ifdef MEASURE_NORM_MSE - float X0[3000]; - float bandE0[60]; - c=0; do - for (i=0;i<N;i++) - X0[i+c*N] = X[i+c*N]; - while (++c<C); - for (i=0;i<C*st->mode->nbEBands;i++) - bandE0[i] = bandE[i]; -#endif - - /* Residual quantisation */ - ALLOC(collapse_masks, C*st->mode->nbEBands, unsigned char); - quant_all_bands(1, st->mode, st->start, st->end, X, C==2 ? X+N : NULL, collapse_masks, - bandE, pulses, shortBlocks, st->spread_decision, dual_stereo, intensity, tf_res, - nbCompressedBytes*(8<<BITRES)-anti_collapse_rsv, balance, enc, LM, codedBands, &st->rng); - - if (anti_collapse_rsv > 0) - { - anti_collapse_on = st->consec_transient<2; -#ifdef FUZZING - anti_collapse_on = rand()&0x1; -#endif - ec_enc_bits(enc, anti_collapse_on, 1); - } - quant_energy_finalise(st->mode, st->start, st->end, oldBandE, error, fine_quant, fine_priority, nbCompressedBytes*8-ec_tell(enc), enc, C); - - if (silence) - { - for (i=0;i<C*st->mode->nbEBands;i++) - oldBandE[i] = -QCONST16(28.f,DB_SHIFT); - } - -#ifdef RESYNTH - /* Re-synthesis of the coded audio if required */ - { - celt_sig *out_mem[2]; - celt_sig *overlap_mem[2]; - - log2Amp(st->mode, st->start, st->end, bandE, oldBandE, C); - if (silence) - { - for (i=0;i<C*st->mode->nbEBands;i++) - bandE[i] = 0; - } - -#ifdef MEASURE_NORM_MSE - measure_norm_mse(st->mode, X, X0, bandE, bandE0, M, N, C); -#endif - if (anti_collapse_on) - { - anti_collapse(st->mode, X, collapse_masks, LM, C, N, - st->start, st->end, oldBandE, oldLogE, oldLogE2, pulses, st->rng); - } - - /* Synthesis */ - denormalise_bands(st->mode, X, freq, bandE, effEnd, C, M); - - OPUS_MOVE(st->syn_mem[0], st->syn_mem[0]+N, MAX_PERIOD); - if (CC==2) - OPUS_MOVE(st->syn_mem[1], st->syn_mem[1]+N, MAX_PERIOD); - - c=0; do - for (i=0;i<M*st->mode->eBands[st->start];i++) - freq[c*N+i] = 0; - while (++c<C); - c=0; do - for (i=M*st->mode->eBands[st->end];i<N;i++) - freq[c*N+i] = 0; - while (++c<C); - - if (CC==2&&C==1) - { - for (i=0;i<N;i++) - freq[N+i] = freq[i]; - } - - out_mem[0] = st->syn_mem[0]+MAX_PERIOD; - if (CC==2) - out_mem[1] = st->syn_mem[1]+MAX_PERIOD; - - overlap_mem[0] = (celt_sig*)(oldLogE2 + CC*st->mode->nbEBands); - if (CC==2) - overlap_mem[1] = overlap_mem[0] + st->overlap; - - compute_inv_mdcts(st->mode, shortBlocks, freq, out_mem, overlap_mem, CC, LM); - - c=0; do { - st->prefilter_period=IMAX(st->prefilter_period, COMBFILTER_MINPERIOD); - st->prefilter_period_old=IMAX(st->prefilter_period_old, COMBFILTER_MINPERIOD); - comb_filter(out_mem[c], out_mem[c], st->prefilter_period_old, st->prefilter_period, st->mode->shortMdctSize, - st->prefilter_gain_old, st->prefilter_gain, st->prefilter_tapset_old, st->prefilter_tapset, - st->mode->window, st->overlap); - if (LM!=0) - comb_filter(out_mem[c]+st->mode->shortMdctSize, out_mem[c]+st->mode->shortMdctSize, st->prefilter_period, pitch_index, N-st->mode->shortMdctSize, - st->prefilter_gain, gain1, st->prefilter_tapset, prefilter_tapset, - st->mode->window, st->mode->overlap); - } while (++c<CC); - - deemphasis(out_mem, (opus_val16*)pcm, N, CC, st->upsample, st->mode->preemph, st->preemph_memD); - st->prefilter_period_old = st->prefilter_period; - st->prefilter_gain_old = st->prefilter_gain; - st->prefilter_tapset_old = st->prefilter_tapset; - } -#endif - - st->prefilter_period = pitch_index; - st->prefilter_gain = gain1; - st->prefilter_tapset = prefilter_tapset; -#ifdef RESYNTH - if (LM!=0) - { - st->prefilter_period_old = st->prefilter_period; - st->prefilter_gain_old = st->prefilter_gain; - st->prefilter_tapset_old = st->prefilter_tapset; - } -#endif - - if (CC==2&&C==1) { - for (i=0;i<st->mode->nbEBands;i++) - oldBandE[st->mode->nbEBands+i]=oldBandE[i]; - } - - if (!isTransient) - { - for (i=0;i<CC*st->mode->nbEBands;i++) - oldLogE2[i] = oldLogE[i]; - for (i=0;i<CC*st->mode->nbEBands;i++) - oldLogE[i] = oldBandE[i]; - } else { - for (i=0;i<CC*st->mode->nbEBands;i++) - oldLogE[i] = MIN16(oldLogE[i], oldBandE[i]); - } - /* In case start or end were to change */ - c=0; do - { - for (i=0;i<st->start;i++) - { - oldBandE[c*st->mode->nbEBands+i]=0; - oldLogE[c*st->mode->nbEBands+i]=oldLogE2[c*st->mode->nbEBands+i]=-QCONST16(28.f,DB_SHIFT); - } - for (i=st->end;i<st->mode->nbEBands;i++) - { - oldBandE[c*st->mode->nbEBands+i]=0; - oldLogE[c*st->mode->nbEBands+i]=oldLogE2[c*st->mode->nbEBands+i]=-QCONST16(28.f,DB_SHIFT); - } - } while (++c<CC); - - if (isTransient) - st->consec_transient++; - else - st->consec_transient=0; - st->rng = enc->rng; - - /* If there's any room left (can only happen for very high rates), - it's already filled with zeros */ - ec_enc_done(enc); - -#ifdef CUSTOM_MODES - if (st->signalling) - nbCompressedBytes++; -#endif - - RESTORE_STACK; - if (ec_get_error(enc)) - return OPUS_INTERNAL_ERROR; - else - return nbCompressedBytes; -} - - -#ifdef CUSTOM_MODES - -#ifdef FIXED_POINT -int opus_custom_encode(CELTEncoder * OPUS_RESTRICT st, const opus_int16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) -{ - return celt_encode_with_ec(st, pcm, frame_size, compressed, nbCompressedBytes, NULL); -} - -#ifndef DISABLE_FLOAT_API -int opus_custom_encode_float(CELTEncoder * OPUS_RESTRICT st, const float * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) -{ - int j, ret, C, N; - VARDECL(opus_int16, in); - ALLOC_STACK; - - if (pcm==NULL) - return OPUS_BAD_ARG; - - C = st->channels; - N = frame_size; - ALLOC(in, C*N, opus_int16); - - for (j=0;j<C*N;j++) - in[j] = FLOAT2INT16(pcm[j]); - - ret=celt_encode_with_ec(st,in,frame_size,compressed,nbCompressedBytes, NULL); -#ifdef RESYNTH - for (j=0;j<C*N;j++) - ((float*)pcm)[j]=in[j]*(1.f/32768.f); -#endif - RESTORE_STACK; - return ret; -} -#endif /* DISABLE_FLOAT_API */ -#else - -int opus_custom_encode(CELTEncoder * OPUS_RESTRICT st, const opus_int16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) -{ - int j, ret, C, N; - VARDECL(celt_sig, in); - ALLOC_STACK; - - if (pcm==NULL) - return OPUS_BAD_ARG; - - C=st->channels; - N=frame_size; - ALLOC(in, C*N, celt_sig); - for (j=0;j<C*N;j++) { - in[j] = SCALEOUT(pcm[j]); - } - - ret = celt_encode_with_ec(st,in,frame_size,compressed,nbCompressedBytes, NULL); -#ifdef RESYNTH - for (j=0;j<C*N;j++) - ((opus_int16*)pcm)[j] = FLOAT2INT16(in[j]); -#endif - RESTORE_STACK; - return ret; -} - -int opus_custom_encode_float(CELTEncoder * OPUS_RESTRICT st, const float * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) -{ - return celt_encode_with_ec(st, pcm, frame_size, compressed, nbCompressedBytes, NULL); -} - -#endif - -#endif /* CUSTOM_MODES */ - -int opus_custom_encoder_ctl(CELTEncoder * OPUS_RESTRICT st, int request, ...) -{ - va_list ap; - - va_start(ap, request); - switch (request) - { - case OPUS_SET_COMPLEXITY_REQUEST: - { - int value = va_arg(ap, opus_int32); - if (value<0 || value>10) - goto bad_arg; - st->complexity = value; - } - break; - case CELT_SET_START_BAND_REQUEST: - { - opus_int32 value = va_arg(ap, opus_int32); - if (value<0 || value>=st->mode->nbEBands) - goto bad_arg; - st->start = value; - } - break; - case CELT_SET_END_BAND_REQUEST: - { - opus_int32 value = va_arg(ap, opus_int32); - if (value<1 || value>st->mode->nbEBands) - goto bad_arg; - st->end = value; - } - break; - case CELT_SET_PREDICTION_REQUEST: - { - int value = va_arg(ap, opus_int32); - if (value<0 || value>2) - goto bad_arg; - st->disable_pf = value<=1; - st->force_intra = value==0; - } - break; - case OPUS_SET_PACKET_LOSS_PERC_REQUEST: - { - int value = va_arg(ap, opus_int32); - if (value<0 || value>100) - goto bad_arg; - st->loss_rate = value; - } - break; - case OPUS_SET_VBR_CONSTRAINT_REQUEST: - { - opus_int32 value = va_arg(ap, opus_int32); - st->constrained_vbr = value; - } - break; - case OPUS_SET_VBR_REQUEST: - { - opus_int32 value = va_arg(ap, opus_int32); - st->vbr = value; - } - break; - case OPUS_SET_BITRATE_REQUEST: - { - opus_int32 value = va_arg(ap, opus_int32); - if (value<=500 && value!=OPUS_BITRATE_MAX) - goto bad_arg; - value = IMIN(value, 260000*st->channels); - st->bitrate = value; - } - break; - case CELT_SET_CHANNELS_REQUEST: - { - opus_int32 value = va_arg(ap, opus_int32); - if (value<1 || value>2) - goto bad_arg; - st->stream_channels = value; - } - break; - case OPUS_SET_LSB_DEPTH_REQUEST: - { - opus_int32 value = va_arg(ap, opus_int32); - if (value<8 || value>24) - goto bad_arg; - st->lsb_depth=value; - } - break; - case OPUS_GET_LSB_DEPTH_REQUEST: - { - opus_int32 *value = va_arg(ap, opus_int32*); - *value=st->lsb_depth; - } - break; - case OPUS_RESET_STATE: - { - int i; - opus_val16 *oldBandE, *oldLogE, *oldLogE2; - oldBandE = (opus_val16*)(st->in_mem+st->channels*(st->overlap+COMBFILTER_MAXPERIOD)); - oldLogE = oldBandE + st->channels*st->mode->nbEBands; - oldLogE2 = oldLogE + st->channels*st->mode->nbEBands; - OPUS_CLEAR((char*)&st->ENCODER_RESET_START, - opus_custom_encoder_get_size(st->mode, st->channels)- - ((char*)&st->ENCODER_RESET_START - (char*)st)); - for (i=0;i<st->channels*st->mode->nbEBands;i++) - oldLogE[i]=oldLogE2[i]=-QCONST16(28.f,DB_SHIFT); - st->vbr_offset = 0; - st->delayedIntra = 1; - st->spread_decision = SPREAD_NORMAL; - st->tonal_average = 256; - st->hf_average = 0; - st->tapset_decision = 0; - } - break; -#ifdef CUSTOM_MODES - case CELT_SET_INPUT_CLIPPING_REQUEST: - { - opus_int32 value = va_arg(ap, opus_int32); - st->clip = value; - } - break; -#endif - case CELT_SET_SIGNALLING_REQUEST: - { - opus_int32 value = va_arg(ap, opus_int32); - st->signalling = value; - } - break; - case CELT_GET_MODE_REQUEST: - { - const CELTMode ** value = va_arg(ap, const CELTMode**); - if (value==0) - goto bad_arg; - *value=st->mode; - } - break; - case OPUS_GET_FINAL_RANGE_REQUEST: - { - opus_uint32 * value = va_arg(ap, opus_uint32 *); - if (value==0) - goto bad_arg; - *value=st->rng; - } - break; - default: - goto bad_request; - } - va_end(ap); - return OPUS_OK; -bad_arg: - va_end(ap); - return OPUS_BAD_ARG; -bad_request: - va_end(ap); - return OPUS_UNIMPLEMENTED; -} - -/**********************************************************************/ -/* */ -/* DECODER */ -/* */ -/**********************************************************************/ -#define DECODE_BUFFER_SIZE 2048 - -/** Decoder state - @brief Decoder state - */ -struct OpusCustomDecoder { - const OpusCustomMode *mode; - int overlap; - int channels; - int stream_channels; - - int downsample; - int start, end; - int signalling; - - /* Everything beyond this point gets cleared on a reset */ -#define DECODER_RESET_START rng - - opus_uint32 rng; - int error; - int last_pitch_index; - int loss_count; - int postfilter_period; - int postfilter_period_old; - opus_val16 postfilter_gain; - opus_val16 postfilter_gain_old; - int postfilter_tapset; - int postfilter_tapset_old; - - celt_sig preemph_memD[2]; - - celt_sig _decode_mem[1]; /* Size = channels*(DECODE_BUFFER_SIZE+mode->overlap) */ - /* opus_val16 lpc[], Size = channels*LPC_ORDER */ - /* opus_val16 oldEBands[], Size = 2*mode->nbEBands */ - /* opus_val16 oldLogE[], Size = 2*mode->nbEBands */ - /* opus_val16 oldLogE2[], Size = 2*mode->nbEBands */ - /* opus_val16 backgroundLogE[], Size = 2*mode->nbEBands */ -}; - -int celt_decoder_get_size(int channels) -{ - const CELTMode *mode = opus_custom_mode_create(48000, 960, NULL); - return opus_custom_decoder_get_size(mode, channels); -} - -OPUS_CUSTOM_NOSTATIC int opus_custom_decoder_get_size(const CELTMode *mode, int channels) -{ - int size = sizeof(struct CELTDecoder) - + (channels*(DECODE_BUFFER_SIZE+mode->overlap)-1)*sizeof(celt_sig) - + channels*LPC_ORDER*sizeof(opus_val16) - + 4*2*mode->nbEBands*sizeof(opus_val16); - return size; -} - -#ifdef CUSTOM_MODES -CELTDecoder *opus_custom_decoder_create(const CELTMode *mode, int channels, int *error) -{ - int ret; - CELTDecoder *st = (CELTDecoder *)opus_alloc(opus_custom_decoder_get_size(mode, channels)); - ret = opus_custom_decoder_init(st, mode, channels); - if (ret != OPUS_OK) - { - opus_custom_decoder_destroy(st); - st = NULL; - } - if (error) - *error = ret; - return st; -} -#endif /* CUSTOM_MODES */ - -int celt_decoder_init(CELTDecoder *st, opus_int32 sampling_rate, int channels) -{ - int ret; - ret = opus_custom_decoder_init(st, opus_custom_mode_create(48000, 960, NULL), channels); - if (ret != OPUS_OK) - return ret; - st->downsample = resampling_factor(sampling_rate); - if (st->downsample==0) - return OPUS_BAD_ARG; - else - return OPUS_OK; -} - -OPUS_CUSTOM_NOSTATIC int opus_custom_decoder_init(CELTDecoder *st, const CELTMode *mode, int channels) -{ - if (channels < 0 || channels > 2) - return OPUS_BAD_ARG; - - if (st==NULL) - return OPUS_ALLOC_FAIL; - - OPUS_CLEAR((char*)st, opus_custom_decoder_get_size(mode, channels)); - - st->mode = mode; - st->overlap = mode->overlap; - st->stream_channels = st->channels = channels; - - st->downsample = 1; - st->start = 0; - st->end = st->mode->effEBands; - st->signalling = 1; - - st->loss_count = 0; - - opus_custom_decoder_ctl(st, OPUS_RESET_STATE); - - return OPUS_OK; -} - -#ifdef CUSTOM_MODES -void opus_custom_decoder_destroy(CELTDecoder *st) -{ - opus_free(st); -} -#endif /* CUSTOM_MODES */ - -static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, opus_val16 * OPUS_RESTRICT pcm, int N, int LM) -{ - int c; - int pitch_index; - opus_val16 fade = Q15ONE; - int i, len; - const int C = st->channels; - int offset; - celt_sig *out_mem[2]; - celt_sig *decode_mem[2]; - celt_sig *overlap_mem[2]; - opus_val16 *lpc; - opus_val32 *out_syn[2]; - opus_val16 *oldBandE, *oldLogE, *oldLogE2, *backgroundLogE; - const OpusCustomMode *mode; - int nbEBands; - int overlap; - const opus_int16 *eBands; - SAVE_STACK; - - mode = st->mode; - nbEBands = mode->nbEBands; - overlap = mode->overlap; - eBands = mode->eBands; - - c=0; do { - decode_mem[c] = st->_decode_mem + c*(DECODE_BUFFER_SIZE+st->overlap); - out_mem[c] = decode_mem[c]+DECODE_BUFFER_SIZE-MAX_PERIOD; - overlap_mem[c] = decode_mem[c]+DECODE_BUFFER_SIZE; - } while (++c<C); - lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+st->overlap)*C); - oldBandE = lpc+C*LPC_ORDER; - oldLogE = oldBandE + 2*nbEBands; - oldLogE2 = oldLogE + 2*nbEBands; - backgroundLogE = oldLogE2 + 2*nbEBands; - - c=0; do { - out_syn[c] = out_mem[c]+MAX_PERIOD-N; - } while (++c<C); - - len = N+overlap; - - if (st->loss_count >= 5 || st->start!=0) - { - /* Noise-based PLC/CNG */ - VARDECL(celt_sig, freq); - VARDECL(celt_norm, X); - VARDECL(celt_ener, bandE); - opus_uint32 seed; - int effEnd; - - effEnd = st->end; - if (effEnd > mode->effEBands) - effEnd = mode->effEBands; - - ALLOC(freq, C*N, celt_sig); /**< Interleaved signal MDCTs */ - ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ - ALLOC(bandE, nbEBands*C, celt_ener); - - if (st->loss_count >= 5) - log2Amp(mode, st->start, st->end, bandE, backgroundLogE, C); - else { - /* Energy decay */ - opus_val16 decay = st->loss_count==0 ? QCONST16(1.5f, DB_SHIFT) : QCONST16(.5f, DB_SHIFT); - c=0; do - { - for (i=st->start;i<st->end;i++) - oldBandE[c*nbEBands+i] -= decay; - } while (++c<C); - log2Amp(mode, st->start, st->end, bandE, oldBandE, C); - } - seed = st->rng; - for (c=0;c<C;c++) - { - for (i=0;i<(st->mode->eBands[st->start]<<LM);i++) - X[c*N+i] = 0; - for (i=st->start;i<mode->effEBands;i++) - { - int j; - int boffs; - int blen; - boffs = N*c+(eBands[i]<<LM); - blen = (eBands[i+1]-eBands[i])<<LM; - for (j=0;j<blen;j++) - { - seed = celt_lcg_rand(seed); - X[boffs+j] = (celt_norm)((opus_int32)seed>>20); - } - renormalise_vector(X+boffs, blen, Q15ONE); - } - for (i=(st->mode->eBands[st->end]<<LM);i<N;i++) - X[c*N+i] = 0; - } - st->rng = seed; - - denormalise_bands(mode, X, freq, bandE, mode->effEBands, C, 1<<LM); - - c=0; do - for (i=0;i<st->mode->eBands[st->start]<<LM;i++) - freq[c*N+i] = 0; - while (++c<C); - c=0; do { - int bound = eBands[effEnd]<<LM; - if (st->downsample!=1) - bound = IMIN(bound, N/st->downsample); - for (i=bound;i<N;i++) - freq[c*N+i] = 0; - } while (++c<C); - c=0; do { - OPUS_MOVE(decode_mem[c], decode_mem[c]+N, DECODE_BUFFER_SIZE-N+overlap); - } while (++c<C); - compute_inv_mdcts(mode, 0, freq, out_syn, overlap_mem, C, LM); - } else { - /* Pitch-based PLC */ - VARDECL(opus_val32, etmp); - - if (st->loss_count == 0) - { - opus_val16 pitch_buf[DECODE_BUFFER_SIZE>>1]; - /* Corresponds to a min pitch of 67 Hz. It's possible to save CPU in this - search by using only part of the decode buffer */ - int poffset = 720; - pitch_downsample(decode_mem, pitch_buf, DECODE_BUFFER_SIZE, C); - /* Max pitch is 100 samples (480 Hz) */ - pitch_search(pitch_buf+((poffset)>>1), pitch_buf, DECODE_BUFFER_SIZE-poffset, - poffset-100, &pitch_index); - pitch_index = poffset-pitch_index; - st->last_pitch_index = pitch_index; - } else { - pitch_index = st->last_pitch_index; - fade = QCONST16(.8f,15); - } - - ALLOC(etmp, overlap, opus_val32); - c=0; do { - opus_val16 exc[MAX_PERIOD]; - opus_val32 ac[LPC_ORDER+1]; - opus_val16 decay; - opus_val16 attenuation; - opus_val32 S1=0; - opus_val16 mem[LPC_ORDER]={0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0}; - opus_val32 *e = out_syn[c]; - - - offset = MAX_PERIOD-pitch_index; - for (i=0;i<MAX_PERIOD;i++) - exc[i] = ROUND16(out_mem[c][i], SIG_SHIFT); - - /* Compute LPC coefficients for the last MAX_PERIOD samples before the loss so we can - work in the excitation-filter domain */ - if (st->loss_count == 0) - { - _celt_autocorr(exc, ac, mode->window, overlap, - LPC_ORDER, MAX_PERIOD); - - /* Noise floor -40 dB */ -#ifdef FIXED_POINT - ac[0] += SHR32(ac[0],13); -#else - ac[0] *= 1.0001f; -#endif - /* Lag windowing */ - for (i=1;i<=LPC_ORDER;i++) - { - /*ac[i] *= exp(-.5*(2*M_PI*.002*i)*(2*M_PI*.002*i));*/ -#ifdef FIXED_POINT - ac[i] -= MULT16_32_Q15(2*i*i, ac[i]); -#else - ac[i] -= ac[i]*(.008f*i)*(.008f*i); -#endif - } - - _celt_lpc(lpc+c*LPC_ORDER, ac, LPC_ORDER); - } - /* Samples just before the beginning of exc */ - for (i=0;i<LPC_ORDER;i++) - mem[i] = ROUND16(out_mem[c][-1-i], SIG_SHIFT); - /* Compute the excitation for MAX_PERIOD samples before the loss */ - celt_fir(exc, lpc+c*LPC_ORDER, exc, MAX_PERIOD, LPC_ORDER, mem); - - /* Check if the waveform is decaying (and if so how fast) - We do this to avoid adding energy when concealing in a segment - with decaying energy */ - { - opus_val32 E1=1, E2=1; - int period; - if (pitch_index <= MAX_PERIOD/2) - period = pitch_index; - else - period = MAX_PERIOD/2; - for (i=0;i<period;i++) - { - E1 += SHR32(MULT16_16(exc[MAX_PERIOD-period+i],exc[MAX_PERIOD-period+i]),8); - E2 += SHR32(MULT16_16(exc[MAX_PERIOD-2*period+i],exc[MAX_PERIOD-2*period+i]),8); - } - if (E1 > E2) - E1 = E2; - decay = celt_sqrt(frac_div32(SHR32(E1,1),E2)); - attenuation = decay; - } - - /* Move memory one frame to the left */ - OPUS_MOVE(decode_mem[c], decode_mem[c]+N, DECODE_BUFFER_SIZE-N+overlap); - - /* Extrapolate excitation with the right period, taking decay into account */ - for (i=0;i<len;i++) - { - opus_val16 tmp; - if (offset+i >= MAX_PERIOD) - { - offset -= pitch_index; - attenuation = MULT16_16_Q15(attenuation, decay); - } - e[i] = SHL32(EXTEND32(MULT16_16_Q15(attenuation, exc[offset+i])), SIG_SHIFT); - /* Compute the energy of the previously decoded signal whose - excitation we're copying */ - tmp = ROUND16(out_mem[c][-N+offset+i],SIG_SHIFT); - S1 += SHR32(MULT16_16(tmp,tmp),8); - } - - /* Copy the last decoded samples (prior to the overlap region) to - synthesis filter memory so we can have a continuous signal. */ - for (i=0;i<LPC_ORDER;i++) - mem[i] = ROUND16(out_mem[c][MAX_PERIOD-N-1-i], SIG_SHIFT); - /* Apply the fading if not the first loss */ - for (i=0;i<len;i++) - e[i] = MULT16_32_Q15(fade, e[i]); - /* Synthesis filter -- back in the signal domain */ - celt_iir(e, lpc+c*LPC_ORDER, e, len, LPC_ORDER, mem); - - /* Check if the synthesis energy is higher than expected, which can - happen with the signal changes during our window. If so, attenuate. */ - { - opus_val32 S2=0; - for (i=0;i<len;i++) - { - opus_val16 tmp = ROUND16(e[i],SIG_SHIFT); - S2 += SHR32(MULT16_16(tmp,tmp),8); - } - /* This checks for an "explosion" in the synthesis */ -#ifdef FIXED_POINT - if (!(S1 > SHR32(S2,2))) -#else - /* Float test is written this way to catch NaNs at the same time */ - if (!(S1 > 0.2f*S2)) -#endif - { - for (i=0;i<len;i++) - e[i] = 0; - } else if (S1 < S2) - { - opus_val16 ratio = celt_sqrt(frac_div32(SHR32(S1,1)+1,S2+1)); - for (i=0;i<overlap;i++) - { - opus_val16 tmp_g = Q15ONE - MULT16_16_Q15(mode->window[i], Q15ONE-ratio); - e[i] = MULT16_32_Q15(tmp_g, e[i]); - } - for (i=overlap;i<len;i++) - e[i] = MULT16_32_Q15(ratio, e[i]); - } - } - - /* Apply pre-filter to the MDCT overlap for the next frame because the - post-filter will be re-applied in the decoder after the MDCT overlap */ - comb_filter(etmp, out_mem[c]+MAX_PERIOD, st->postfilter_period, st->postfilter_period, st->overlap, - -st->postfilter_gain, -st->postfilter_gain, st->postfilter_tapset, st->postfilter_tapset, - NULL, 0); - - /* Simulate TDAC on the concealed audio so that it blends with the - MDCT of next frames. */ - for (i=0;i<overlap/2;i++) - { - opus_val32 tmp; - tmp = MULT16_32_Q15(mode->window[i], etmp[overlap-1-i]) + - MULT16_32_Q15(mode->window[overlap-i-1], etmp[i ]); - out_mem[c][MAX_PERIOD+i] = MULT16_32_Q15(mode->window[overlap-i-1], tmp); - out_mem[c][MAX_PERIOD+overlap-i-1] = MULT16_32_Q15(mode->window[i], tmp); - } - } while (++c<C); - } - - deemphasis(out_syn, pcm, N, C, st->downsample, mode->preemph, st->preemph_memD); - - st->loss_count++; - - RESTORE_STACK; -} - -int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_val16 * OPUS_RESTRICT pcm, int frame_size, ec_dec *dec) -{ - int c, i, N; - int spread_decision; - opus_int32 bits; - ec_dec _dec; - VARDECL(celt_sig, freq); - VARDECL(celt_norm, X); - VARDECL(celt_ener, bandE); - VARDECL(int, fine_quant); - VARDECL(int, pulses); - VARDECL(int, cap); - VARDECL(int, offsets); - VARDECL(int, fine_priority); - VARDECL(int, tf_res); - VARDECL(unsigned char, collapse_masks); - celt_sig *out_mem[2]; - celt_sig *decode_mem[2]; - celt_sig *overlap_mem[2]; - celt_sig *out_syn[2]; - opus_val16 *lpc; - opus_val16 *oldBandE, *oldLogE, *oldLogE2, *backgroundLogE; - - int shortBlocks; - int isTransient; - int intra_ener; - const int CC = st->channels; - int LM, M; - int effEnd; - int codedBands; - int alloc_trim; - int postfilter_pitch; - opus_val16 postfilter_gain; - int intensity=0; - int dual_stereo=0; - opus_int32 total_bits; - opus_int32 balance; - opus_int32 tell; - int dynalloc_logp; - int postfilter_tapset; - int anti_collapse_rsv; - int anti_collapse_on=0; - int silence; - int C = st->stream_channels; - ALLOC_STACK; - - frame_size *= st->downsample; - - c=0; do { - decode_mem[c] = st->_decode_mem + c*(DECODE_BUFFER_SIZE+st->overlap); - out_mem[c] = decode_mem[c]+DECODE_BUFFER_SIZE-MAX_PERIOD; - overlap_mem[c] = decode_mem[c]+DECODE_BUFFER_SIZE; - } while (++c<CC); - lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+st->overlap)*CC); - oldBandE = lpc+CC*LPC_ORDER; - oldLogE = oldBandE + 2*st->mode->nbEBands; - oldLogE2 = oldLogE + 2*st->mode->nbEBands; - backgroundLogE = oldLogE2 + 2*st->mode->nbEBands; - -#ifdef CUSTOM_MODES - if (st->signalling && data!=NULL) - { - int data0=data[0]; - /* Convert "standard mode" to Opus header */ - if (st->mode->Fs==48000 && st->mode->shortMdctSize==120) - { - data0 = fromOpus(data0); - if (data0<0) - return OPUS_INVALID_PACKET; - } - st->end = IMAX(1, st->mode->effEBands-2*(data0>>5)); - LM = (data0>>3)&0x3; - C = 1 + ((data0>>2)&0x1); - data++; - len--; - if (LM>st->mode->maxLM) - return OPUS_INVALID_PACKET; - if (frame_size < st->mode->shortMdctSize<<LM) - return OPUS_BUFFER_TOO_SMALL; - else - frame_size = st->mode->shortMdctSize<<LM; - } else { -#else - { -#endif - for (LM=0;LM<=st->mode->maxLM;LM++) - if (st->mode->shortMdctSize<<LM==frame_size) - break; - if (LM>st->mode->maxLM) - return OPUS_BAD_ARG; - } - M=1<<LM; - - if (len<0 || len>1275 || pcm==NULL) - return OPUS_BAD_ARG; - - N = M*st->mode->shortMdctSize; - - effEnd = st->end; - if (effEnd > st->mode->effEBands) - effEnd = st->mode->effEBands; - - if (data == NULL || len<=1) - { - celt_decode_lost(st, pcm, N, LM); - RESTORE_STACK; - return frame_size/st->downsample; - } - - ALLOC(freq, IMAX(CC,C)*N, celt_sig); /**< Interleaved signal MDCTs */ - ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ - ALLOC(bandE, st->mode->nbEBands*C, celt_ener); - c=0; do - for (i=0;i<M*st->mode->eBands[st->start];i++) - X[c*N+i] = 0; - while (++c<C); - c=0; do - for (i=M*st->mode->eBands[effEnd];i<N;i++) - X[c*N+i] = 0; - while (++c<C); - - if (dec == NULL) - { - ec_dec_init(&_dec,(unsigned char*)data,len); - dec = &_dec; - } - - if (C==1) - { - for (i=0;i<st->mode->nbEBands;i++) - oldBandE[i]=MAX16(oldBandE[i],oldBandE[st->mode->nbEBands+i]); - } - - total_bits = len*8; - tell = ec_tell(dec); - - if (tell >= total_bits) - silence = 1; - else if (tell==1) - silence = ec_dec_bit_logp(dec, 15); - else - silence = 0; - if (silence) - { - /* Pretend we've read all the remaining bits */ - tell = len*8; - dec->nbits_total+=tell-ec_tell(dec); - } - - postfilter_gain = 0; - postfilter_pitch = 0; - postfilter_tapset = 0; - if (st->start==0 && tell+16 <= total_bits) - { - if(ec_dec_bit_logp(dec, 1)) - { - int qg, octave; - octave = ec_dec_uint(dec, 6); - postfilter_pitch = (16<<octave)+ec_dec_bits(dec, 4+octave)-1; - qg = ec_dec_bits(dec, 3); - if (ec_tell(dec)+2<=total_bits) - postfilter_tapset = ec_dec_icdf(dec, tapset_icdf, 2); - postfilter_gain = QCONST16(.09375f,15)*(qg+1); - } - tell = ec_tell(dec); - } - - if (LM > 0 && tell+3 <= total_bits) - { - isTransient = ec_dec_bit_logp(dec, 3); - tell = ec_tell(dec); - } - else - isTransient = 0; - - if (isTransient) - shortBlocks = M; - else - shortBlocks = 0; - - /* Decode the global flags (first symbols in the stream) */ - intra_ener = tell+3<=total_bits ? ec_dec_bit_logp(dec, 3) : 0; - /* Get band energies */ - unquant_coarse_energy(st->mode, st->start, st->end, oldBandE, - intra_ener, dec, C, LM); - - ALLOC(tf_res, st->mode->nbEBands, int); - tf_decode(st->start, st->end, isTransient, tf_res, LM, dec); - - tell = ec_tell(dec); - spread_decision = SPREAD_NORMAL; - if (tell+4 <= total_bits) - spread_decision = ec_dec_icdf(dec, spread_icdf, 5); - - ALLOC(pulses, st->mode->nbEBands, int); - ALLOC(cap, st->mode->nbEBands, int); - ALLOC(offsets, st->mode->nbEBands, int); - ALLOC(fine_priority, st->mode->nbEBands, int); - - init_caps(st->mode,cap,LM,C); - - dynalloc_logp = 6; - total_bits<<=BITRES; - tell = ec_tell_frac(dec); - for (i=st->start;i<st->end;i++) - { - int width, quanta; - int dynalloc_loop_logp; - int boost; - width = C*(st->mode->eBands[i+1]-st->mode->eBands[i])<<LM; - /* quanta is 6 bits, but no more than 1 bit/sample - and no less than 1/8 bit/sample */ - quanta = IMIN(width<<BITRES, IMAX(6<<BITRES, width)); - dynalloc_loop_logp = dynalloc_logp; - boost = 0; - while (tell+(dynalloc_loop_logp<<BITRES) < total_bits && boost < cap[i]) - { - int flag; - flag = ec_dec_bit_logp(dec, dynalloc_loop_logp); - tell = ec_tell_frac(dec); - if (!flag) - break; - boost += quanta; - total_bits -= quanta; - dynalloc_loop_logp = 1; - } - offsets[i] = boost; - /* Making dynalloc more likely */ - if (boost>0) - dynalloc_logp = IMAX(2, dynalloc_logp-1); - } - - ALLOC(fine_quant, st->mode->nbEBands, int); - alloc_trim = tell+(6<<BITRES) <= total_bits ? - ec_dec_icdf(dec, trim_icdf, 7) : 5; - - bits = (((opus_int32)len*8)<<BITRES) - ec_tell_frac(dec) - 1; - anti_collapse_rsv = isTransient&&LM>=2&&bits>=((LM+2)<<BITRES) ? (1<<BITRES) : 0; - bits -= anti_collapse_rsv; - codedBands = compute_allocation(st->mode, st->start, st->end, offsets, cap, - alloc_trim, &intensity, &dual_stereo, bits, &balance, pulses, - fine_quant, fine_priority, C, LM, dec, 0, 0); - - unquant_fine_energy(st->mode, st->start, st->end, oldBandE, fine_quant, dec, C); - - /* Decode fixed codebook */ - ALLOC(collapse_masks, C*st->mode->nbEBands, unsigned char); - quant_all_bands(0, st->mode, st->start, st->end, X, C==2 ? X+N : NULL, collapse_masks, - NULL, pulses, shortBlocks, spread_decision, dual_stereo, intensity, tf_res, - len*(8<<BITRES)-anti_collapse_rsv, balance, dec, LM, codedBands, &st->rng); - - if (anti_collapse_rsv > 0) - { - anti_collapse_on = ec_dec_bits(dec, 1); - } - - unquant_energy_finalise(st->mode, st->start, st->end, oldBandE, - fine_quant, fine_priority, len*8-ec_tell(dec), dec, C); - - if (anti_collapse_on) - anti_collapse(st->mode, X, collapse_masks, LM, C, N, - st->start, st->end, oldBandE, oldLogE, oldLogE2, pulses, st->rng); - - log2Amp(st->mode, st->start, st->end, bandE, oldBandE, C); - - if (silence) - { - for (i=0;i<C*st->mode->nbEBands;i++) - { - bandE[i] = 0; - oldBandE[i] = -QCONST16(28.f,DB_SHIFT); - } - } - /* Synthesis */ - denormalise_bands(st->mode, X, freq, bandE, effEnd, C, M); - - OPUS_MOVE(decode_mem[0], decode_mem[0]+N, DECODE_BUFFER_SIZE-N); - if (CC==2) - OPUS_MOVE(decode_mem[1], decode_mem[1]+N, DECODE_BUFFER_SIZE-N); - - c=0; do - for (i=0;i<M*st->mode->eBands[st->start];i++) - freq[c*N+i] = 0; - while (++c<C); - c=0; do { - int bound = M*st->mode->eBands[effEnd]; - if (st->downsample!=1) - bound = IMIN(bound, N/st->downsample); - for (i=bound;i<N;i++) - freq[c*N+i] = 0; - } while (++c<C); - - out_syn[0] = out_mem[0]+MAX_PERIOD-N; - if (CC==2) - out_syn[1] = out_mem[1]+MAX_PERIOD-N; - - if (CC==2&&C==1) - { - for (i=0;i<N;i++) - freq[N+i] = freq[i]; - } - if (CC==1&&C==2) - { - for (i=0;i<N;i++) - freq[i] = HALF32(ADD32(freq[i],freq[N+i])); - } - - /* Compute inverse MDCTs */ - compute_inv_mdcts(st->mode, shortBlocks, freq, out_syn, overlap_mem, CC, LM); - - c=0; do { - st->postfilter_period=IMAX(st->postfilter_period, COMBFILTER_MINPERIOD); - st->postfilter_period_old=IMAX(st->postfilter_period_old, COMBFILTER_MINPERIOD); - comb_filter(out_syn[c], out_syn[c], st->postfilter_period_old, st->postfilter_period, st->mode->shortMdctSize, - st->postfilter_gain_old, st->postfilter_gain, st->postfilter_tapset_old, st->postfilter_tapset, - st->mode->window, st->overlap); - if (LM!=0) - comb_filter(out_syn[c]+st->mode->shortMdctSize, out_syn[c]+st->mode->shortMdctSize, st->postfilter_period, postfilter_pitch, N-st->mode->shortMdctSize, - st->postfilter_gain, postfilter_gain, st->postfilter_tapset, postfilter_tapset, - st->mode->window, st->mode->overlap); - - } while (++c<CC); - st->postfilter_period_old = st->postfilter_period; - st->postfilter_gain_old = st->postfilter_gain; - st->postfilter_tapset_old = st->postfilter_tapset; - st->postfilter_period = postfilter_pitch; - st->postfilter_gain = postfilter_gain; - st->postfilter_tapset = postfilter_tapset; - if (LM!=0) - { - st->postfilter_period_old = st->postfilter_period; - st->postfilter_gain_old = st->postfilter_gain; - st->postfilter_tapset_old = st->postfilter_tapset; - } - - if (C==1) { - for (i=0;i<st->mode->nbEBands;i++) - oldBandE[st->mode->nbEBands+i]=oldBandE[i]; - } - - /* In case start or end were to change */ - if (!isTransient) - { - for (i=0;i<2*st->mode->nbEBands;i++) - oldLogE2[i] = oldLogE[i]; - for (i=0;i<2*st->mode->nbEBands;i++) - oldLogE[i] = oldBandE[i]; - for (i=0;i<2*st->mode->nbEBands;i++) - backgroundLogE[i] = MIN16(backgroundLogE[i] + M*QCONST16(0.001f,DB_SHIFT), oldBandE[i]); - } else { - for (i=0;i<2*st->mode->nbEBands;i++) - oldLogE[i] = MIN16(oldLogE[i], oldBandE[i]); - } - c=0; do - { - for (i=0;i<st->start;i++) - { - oldBandE[c*st->mode->nbEBands+i]=0; - oldLogE[c*st->mode->nbEBands+i]=oldLogE2[c*st->mode->nbEBands+i]=-QCONST16(28.f,DB_SHIFT); - } - for (i=st->end;i<st->mode->nbEBands;i++) - { - oldBandE[c*st->mode->nbEBands+i]=0; - oldLogE[c*st->mode->nbEBands+i]=oldLogE2[c*st->mode->nbEBands+i]=-QCONST16(28.f,DB_SHIFT); - } - } while (++c<2); - st->rng = dec->rng; - - deemphasis(out_syn, pcm, N, CC, st->downsample, st->mode->preemph, st->preemph_memD); - st->loss_count = 0; - RESTORE_STACK; - if (ec_tell(dec) > 8*len) - return OPUS_INTERNAL_ERROR; - if(ec_get_error(dec)) - st->error = 1; - return frame_size/st->downsample; -} - - -#ifdef CUSTOM_MODES - -#ifdef FIXED_POINT -int opus_custom_decode(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_int16 * OPUS_RESTRICT pcm, int frame_size) -{ - return celt_decode_with_ec(st, data, len, pcm, frame_size, NULL); -} - -#ifndef DISABLE_FLOAT_API -int opus_custom_decode_float(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, float * OPUS_RESTRICT pcm, int frame_size) -{ - int j, ret, C, N; - VARDECL(opus_int16, out); - ALLOC_STACK; - - if (pcm==NULL) - return OPUS_BAD_ARG; - - C = st->channels; - N = frame_size; - - ALLOC(out, C*N, opus_int16); - ret=celt_decode_with_ec(st, data, len, out, frame_size, NULL); - if (ret>0) - for (j=0;j<C*ret;j++) - pcm[j]=out[j]*(1.f/32768.f); - - RESTORE_STACK; - return ret; -} -#endif /* DISABLE_FLOAT_API */ - -#else - -int opus_custom_decode_float(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, float * OPUS_RESTRICT pcm, int frame_size) -{ - return celt_decode_with_ec(st, data, len, pcm, frame_size, NULL); -} - -int opus_custom_decode(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_int16 * OPUS_RESTRICT pcm, int frame_size) -{ - int j, ret, C, N; - VARDECL(celt_sig, out); - ALLOC_STACK; - - if (pcm==NULL) - return OPUS_BAD_ARG; - - C = st->channels; - N = frame_size; - ALLOC(out, C*N, celt_sig); - - ret=celt_decode_with_ec(st, data, len, out, frame_size, NULL); - - if (ret>0) - for (j=0;j<C*ret;j++) - pcm[j] = FLOAT2INT16 (out[j]); - - RESTORE_STACK; - return ret; -} - -#endif -#endif /* CUSTOM_MODES */ - -int opus_custom_decoder_ctl(CELTDecoder * OPUS_RESTRICT st, int request, ...) -{ - va_list ap; - - va_start(ap, request); - switch (request) - { - case CELT_SET_START_BAND_REQUEST: - { - opus_int32 value = va_arg(ap, opus_int32); - if (value<0 || value>=st->mode->nbEBands) - goto bad_arg; - st->start = value; - } - break; - case CELT_SET_END_BAND_REQUEST: - { - opus_int32 value = va_arg(ap, opus_int32); - if (value<1 || value>st->mode->nbEBands) - goto bad_arg; - st->end = value; - } - break; - case CELT_SET_CHANNELS_REQUEST: - { - opus_int32 value = va_arg(ap, opus_int32); - if (value<1 || value>2) - goto bad_arg; - st->stream_channels = value; - } - break; - case CELT_GET_AND_CLEAR_ERROR_REQUEST: - { - opus_int32 *value = va_arg(ap, opus_int32*); - if (value==NULL) - goto bad_arg; - *value=st->error; - st->error = 0; - } - break; - case OPUS_GET_LOOKAHEAD_REQUEST: - { - opus_int32 *value = va_arg(ap, opus_int32*); - if (value==NULL) - goto bad_arg; - *value = st->overlap/st->downsample; - } - break; - case OPUS_RESET_STATE: - { - int i; - opus_val16 *lpc, *oldBandE, *oldLogE, *oldLogE2; - lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+st->overlap)*st->channels); - oldBandE = lpc+st->channels*LPC_ORDER; - oldLogE = oldBandE + 2*st->mode->nbEBands; - oldLogE2 = oldLogE + 2*st->mode->nbEBands; - OPUS_CLEAR((char*)&st->DECODER_RESET_START, - opus_custom_decoder_get_size(st->mode, st->channels)- - ((char*)&st->DECODER_RESET_START - (char*)st)); - for (i=0;i<2*st->mode->nbEBands;i++) - oldLogE[i]=oldLogE2[i]=-QCONST16(28.f,DB_SHIFT); - } - break; - case OPUS_GET_PITCH_REQUEST: - { - opus_int32 *value = va_arg(ap, opus_int32*); - if (value==NULL) - goto bad_arg; - *value = st->postfilter_period; - } - break; - case CELT_GET_MODE_REQUEST: - { - const CELTMode ** value = va_arg(ap, const CELTMode**); - if (value==0) - goto bad_arg; - *value=st->mode; - } - break; - case CELT_SET_SIGNALLING_REQUEST: - { - opus_int32 value = va_arg(ap, opus_int32); - st->signalling = value; - } - break; - case OPUS_GET_FINAL_RANGE_REQUEST: - { - opus_uint32 * value = va_arg(ap, opus_uint32 *); - if (value==0) - goto bad_arg; - *value=st->rng; - } - break; - default: - goto bad_request; - } - va_end(ap); - return OPUS_OK; -bad_arg: - va_end(ap); - return OPUS_BAD_ARG; -bad_request: - va_end(ap); - return OPUS_UNIMPLEMENTED; -} - - - -const char *opus_strerror(int error) -{ - static const char * const error_strings[8] = { - "success", - "invalid argument", - "buffer too small", - "internal error", - "corrupted stream", - "request not implemented", - "invalid state", - "memory allocation failed" - }; - if (error > 0 || error < -7) - return "unknown error"; - else - return error_strings[-error]; -} - -const char *opus_get_version_string(void) -{ - return "libopus " OPUS_VERSION -#ifdef FIXED_POINT - "-fixed" -#endif -#ifdef FUZZING - "-fuzzing" -#endif - ; -} diff --git a/src/opus-1.0.2/celt/celt_lpc.c b/src/opus-1.0.2/celt/celt_lpc.c deleted file mode 100644 index d2addbf2..00000000 --- a/src/opus-1.0.2/celt/celt_lpc.c +++ /dev/null @@ -1,188 +0,0 @@ -/* Copyright (c) 2009-2010 Xiph.Org Foundation - Written by Jean-Marc Valin */ -/* - Redistribution and use in source and binary forms, with or without - modification, are permitted provided that the following conditions - are met: - - - Redistributions of source code must retain the above copyright - notice, this list of conditions and the following disclaimer. - - - Redistributions in binary form must reproduce the above copyright - notice, this list of conditions and the following disclaimer in the - documentation and/or other materials provided with the distribution. - - THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER - OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, - EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, - PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR - PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF - LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING - NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS - SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -*/ - -#ifdef HAVE_CONFIG_H -#include "config.h" -#endif - -#include "celt_lpc.h" -#include "stack_alloc.h" -#include "mathops.h" - -void _celt_lpc( - opus_val16 *_lpc, /* out: [0...p-1] LPC coefficients */ -const opus_val32 *ac, /* in: [0...p] autocorrelation values */ -int p -) -{ - int i, j; - opus_val32 r; - opus_val32 error = ac[0]; -#ifdef FIXED_POINT - opus_val32 lpc[LPC_ORDER]; -#else - float *lpc = _lpc; -#endif - - for (i = 0; i < p; i++) - lpc[i] = 0; - if (ac[0] != 0) - { - for (i = 0; i < p; i++) { - /* Sum up this iteration's reflection coefficient */ - opus_val32 rr = 0; - for (j = 0; j < i; j++) - rr += MULT32_32_Q31(lpc[j],ac[i - j]); - rr += SHR32(ac[i + 1],3); - r = -frac_div32(SHL32(rr,3), error); - /* Update LPC coefficients and total error */ - lpc[i] = SHR32(r,3); - for (j = 0; j < (i+1)>>1; j++) - { - opus_val32 tmp1, tmp2; - tmp1 = lpc[j]; - tmp2 = lpc[i-1-j]; - lpc[j] = tmp1 + MULT32_32_Q31(r,tmp2); - lpc[i-1-j] = tmp2 + MULT32_32_Q31(r,tmp1); - } - - error = error - MULT32_32_Q31(MULT32_32_Q31(r,r),error); - /* Bail out once we get 30 dB gain */ -#ifdef FIXED_POINT - if (error<SHR32(ac[0],10)) - break; -#else - if (error<.001f*ac[0]) - break; -#endif - } - } -#ifdef FIXED_POINT - for (i=0;i<p;i++) - _lpc[i] = ROUND16(lpc[i],16); -#endif -} - -void celt_fir(const opus_val16 *x, - const opus_val16 *num, - opus_val16 *y, - int N, - int ord, - opus_val16 *mem) -{ - int i,j; - - for (i=0;i<N;i++) - { - opus_val32 sum = SHL32(EXTEND32(x[i]), SIG_SHIFT); - for (j=0;j<ord;j++) - { - sum += MULT16_16(num[j],mem[j]); - } - for (j=ord-1;j>=1;j--) - { - mem[j]=mem[j-1]; - } - mem[0] = x[i]; - y[i] = ROUND16(sum, SIG_SHIFT); - } -} - -void celt_iir(const opus_val32 *x, - const opus_val16 *den, - opus_val32 *y, - int N, - int ord, - opus_val16 *mem) -{ - int i,j; - for (i=0;i<N;i++) - { - opus_val32 sum = x[i]; - for (j=0;j<ord;j++) - { - sum -= MULT16_16(den[j],mem[j]); - } - for (j=ord-1;j>=1;j--) - { - mem[j]=mem[j-1]; - } - mem[0] = ROUND16(sum,SIG_SHIFT); - y[i] = sum; - } -} - -void _celt_autocorr( - const opus_val16 *x, /* in: [0...n-1] samples x */ - opus_val32 *ac, /* out: [0...lag-1] ac values */ - const opus_val16 *window, - int overlap, - int lag, - int n - ) -{ - opus_val32 d; - int i; - VARDECL(opus_val16, xx); - SAVE_STACK; - ALLOC(xx, n, opus_val16); - celt_assert(n>0); - celt_assert(overlap>=0); - for (i=0;i<n;i++) - xx[i] = x[i]; - for (i=0;i<overlap;i++) - { - xx[i] = MULT16_16_Q15(x[i],window[i]); - xx[n-i-1] = MULT16_16_Q15(x[n-i-1],window[i]); - } -#ifdef FIXED_POINT - { - opus_val32 ac0=0; - int shift; - for(i=0;i<n;i++) - ac0 += SHR32(MULT16_16(xx[i],xx[i]),9); - ac0 += 1+n; - - shift = celt_ilog2(ac0)-30+10; - shift = (shift+1)/2; - for(i=0;i<n;i++) - xx[i] = VSHR32(xx[i], shift); - } -#endif - while (lag>=0) - { - for (i = lag, d = 0; i < n; i++) - d += xx[i] * xx[i-lag]; - ac[lag] = d; - /*printf ("%f ", ac[lag]);*/ - lag--; - } - /*printf ("\n");*/ - ac[0] += 10; - - RESTORE_STACK; -} diff --git a/src/opus-1.0.2/celt/cwrs.c b/src/opus-1.0.2/celt/cwrs.c deleted file mode 100644 index 8edc919d..00000000 --- a/src/opus-1.0.2/celt/cwrs.c +++ /dev/null @@ -1,645 +0,0 @@ -/* Copyright (c) 2007-2008 CSIRO - Copyright (c) 2007-2009 Xiph.Org Foundation - Copyright (c) 2007-2009 Timothy B. Terriberry - Written by Timothy B. Terriberry and Jean-Marc Valin */ -/* - Redistribution and use in source and binary forms, with or without - modification, are permitted provided that the following conditions - are met: - - - Redistributions of source code must retain the above copyright - notice, this list of conditions and the following disclaimer. - - - Redistributions in binary form must reproduce the above copyright - notice, this list of conditions and the following disclaimer in the - documentation and/or other materials provided with the distribution. - - THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER - OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, - EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, - PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR - PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF - LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING - NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS - SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -*/ - -#ifdef HAVE_CONFIG_H -#include "config.h" -#endif - -#include "os_support.h" -#include "cwrs.h" -#include "mathops.h" -#include "arch.h" - -#ifdef CUSTOM_MODES - -/*Guaranteed to return a conservatively large estimate of the binary logarithm - with frac bits of fractional precision. - Tested for all possible 32-bit inputs with frac=4, where the maximum - overestimation is 0.06254243 bits.*/ -int log2_frac(opus_uint32 val, int frac) -{ - int l; - l=EC_ILOG(val); - if(val&(val-1)){ - /*This is (val>>l-16), but guaranteed to round up, even if adding a bias - before the shift would cause overflow (e.g., for 0xFFFFxxxx). - Doesn't work for val=0, but that case fails the test above.*/ - if(l>16)val=((val-1)>>(l-16))+1; - else val<<=16-l; - l=(l-1)<<frac; - /*Note that we always need one iteration, since the rounding up above means - that we might need to adjust the integer part of the logarithm.*/ - do{ - int b; - b=(int)(val>>16); - l+=b<<frac; - val=(val+b)>>b; - val=(val*val+0x7FFF)>>15; - } - while(frac-->0); - /*If val is not exactly 0x8000, then we have to round up the remainder.*/ - return l+(val>0x8000); - } - /*Exact powers of two require no rounding.*/ - else return (l-1)<<frac; -} -#endif - -#ifndef SMALL_FOOTPRINT - -#define MASK32 (0xFFFFFFFF) - -/*INV_TABLE[i] holds the multiplicative inverse of (2*i+1) mod 2**32.*/ -static const opus_uint32 INV_TABLE[53]={ - 0x00000001,0xAAAAAAAB,0xCCCCCCCD,0xB6DB6DB7, - 0x38E38E39,0xBA2E8BA3,0xC4EC4EC5,0xEEEEEEEF, - 0xF0F0F0F1,0x286BCA1B,0x3CF3CF3D,0xE9BD37A7, - 0xC28F5C29,0x684BDA13,0x4F72C235,0xBDEF7BDF, - 0x3E0F83E1,0x8AF8AF8B,0x914C1BAD,0x96F96F97, - 0xC18F9C19,0x2FA0BE83,0xA4FA4FA5,0x677D46CF, - 0x1A1F58D1,0xFAFAFAFB,0x8C13521D,0x586FB587, - 0xB823EE09,0xA08AD8F3,0xC10C9715,0xBEFBEFBF, - 0xC0FC0FC1,0x07A44C6B,0xA33F128D,0xE327A977, - 0xC7E3F1F9,0x962FC963,0x3F2B3885,0x613716AF, - 0x781948B1,0x2B2E43DB,0xFCFCFCFD,0x6FD0EB67, - 0xFA3F47E9,0xD2FD2FD3,0x3F4FD3F5,0xD4E25B9F, - 0x5F02A3A1,0xBF5A814B,0x7C32B16D,0xD3431B57, - 0xD8FD8FD9, -}; - -/*Computes (_a*_b-_c)/(2*_d+1) when the quotient is known to be exact. - _a, _b, _c, and _d may be arbitrary so long as the arbitrary precision result - fits in 32 bits, but currently the table for multiplicative inverses is only - valid for _d<=52.*/ -static inline opus_uint32 imusdiv32odd(opus_uint32 _a,opus_uint32 _b, - opus_uint32 _c,int _d){ - celt_assert(_d<=52); - return (_a*_b-_c)*INV_TABLE[_d]&MASK32; -} - -/*Computes (_a*_b-_c)/_d when the quotient is known to be exact. - _d does not actually have to be even, but imusdiv32odd will be faster when - it's odd, so you should use that instead. - _a and _d are assumed to be small (e.g., _a*_d fits in 32 bits; currently the - table for multiplicative inverses is only valid for _d<=54). - _b and _c may be arbitrary so long as the arbitrary precision reuslt fits in - 32 bits.*/ -static inline opus_uint32 imusdiv32even(opus_uint32 _a,opus_uint32 _b, - opus_uint32 _c,int _d){ - opus_uint32 inv; - int mask; - int shift; - int one; - celt_assert(_d>0); - celt_assert(_d<=54); - shift=EC_ILOG(_d^(_d-1)); - inv=INV_TABLE[(_d-1)>>shift]; - shift--; - one=1<<shift; - mask=one-1; - return (_a*(_b>>shift)-(_c>>shift)+ - ((_a*(_b&mask)+one-(_c&mask))>>shift)-1)*inv&MASK32; -} - -#endif /* SMALL_FOOTPRINT */ - -/*Although derived separately, the pulse vector coding scheme is equivalent to - a Pyramid Vector Quantizer \cite{Fis86}. - Some additional notes about an early version appear at - http://people.xiph.org/~tterribe/notes/cwrs.html, but the codebook ordering - and the definitions of some terms have evolved since that was written. - - The conversion from a pulse vector to an integer index (encoding) and back - (decoding) is governed by two related functions, V(N,K) and U(N,K). - - V(N,K) = the number of combinations, with replacement, of N items, taken K - at a time, when a sign bit is added to each item taken at least once (i.e., - the number of N-dimensional unit pulse vectors with K pulses). - One way to compute this is via - V(N,K) = K>0 ? sum(k=1...K,2**k*choose(N,k)*choose(K-1,k-1)) : 1, - where choose() is the binomial function. - A table of values for N<10 and K<10 looks like: - V[10][10] = { - {1, 0, 0, 0, 0, 0, 0, 0, 0, 0}, - {1, 2, 2, 2, 2, 2, 2, 2, 2, 2}, - {1, 4, 8, 12, 16, 20, 24, 28, 32, 36}, - {1, 6, 18, 38, 66, 102, 146, 198, 258, 326}, - {1, 8, 32, 88, 192, 360, 608, 952, 1408, 1992}, - {1, 10, 50, 170, 450, 1002, 1970, 3530, 5890, 9290}, - {1, 12, 72, 292, 912, 2364, 5336, 10836, 20256, 35436}, - {1, 14, 98, 462, 1666, 4942, 12642, 28814, 59906, 115598}, - {1, 16, 128, 688, 2816, 9424, 27008, 68464, 157184, 332688}, - {1, 18, 162, 978, 4482, 16722, 53154, 148626, 374274, 864146} - }; - - U(N,K) = the number of such combinations wherein N-1 objects are taken at - most K-1 at a time. - This is given by - U(N,K) = sum(k=0...K-1,V(N-1,k)) - = K>0 ? (V(N-1,K-1) + V(N,K-1))/2 : 0. - The latter expression also makes clear that U(N,K) is half the number of such - combinations wherein the first object is taken at least once. - Although it may not be clear from either of these definitions, U(N,K) is the - natural function to work with when enumerating the pulse vector codebooks, - not V(N,K). - U(N,K) is not well-defined for N=0, but with the extension - U(0,K) = K>0 ? 0 : 1, - the function becomes symmetric: U(N,K) = U(K,N), with a similar table: - U[10][10] = { - {1, 0, 0, 0, 0, 0, 0, 0, 0, 0}, - {0, 1, 1, 1, 1, 1, 1, 1, 1, 1}, - {0, 1, 3, 5, 7, 9, 11, 13, 15, 17}, - {0, 1, 5, 13, 25, 41, 61, 85, 113, 145}, - {0, 1, 7, 25, 63, 129, 231, 377, 575, 833}, - {0, 1, 9, 41, 129, 321, 681, 1289, 2241, 3649}, - {0, 1, 11, 61, 231, 681, 1683, 3653, 7183, 13073}, - {0, 1, 13, 85, 377, 1289, 3653, 8989, 19825, 40081}, - {0, 1, 15, 113, 575, 2241, 7183, 19825, 48639, 108545}, - {0, 1, 17, 145, 833, 3649, 13073, 40081, 108545, 265729} - }; - - With this extension, V(N,K) may be written in terms of U(N,K): - V(N,K) = U(N,K) + U(N,K+1) - for all N>=0, K>=0. - Thus U(N,K+1) represents the number of combinations where the first element - is positive or zero, and U(N,K) represents the number of combinations where - it is negative. - With a large enough table of U(N,K) values, we could write O(N) encoding - and O(min(N*log(K),N+K)) decoding routines, but such a table would be - prohibitively large for small embedded devices (K may be as large as 32767 - for small N, and N may be as large as 200). - - Both functions obey the same recurrence relation: - V(N,K) = V(N-1,K) + V(N,K-1) + V(N-1,K-1), - U(N,K) = U(N-1,K) + U(N,K-1) + U(N-1,K-1), - for all N>0, K>0, with different initial conditions at N=0 or K=0. - This allows us to construct a row of one of the tables above given the - previous row or the next row. - Thus we can derive O(NK) encoding and decoding routines with O(K) memory - using only addition and subtraction. - - When encoding, we build up from the U(2,K) row and work our way forwards. - When decoding, we need to start at the U(N,K) row and work our way backwards, - which requires a means of computing U(N,K). - U(N,K) may be computed from two previous values with the same N: - U(N,K) = ((2*N-1)*U(N,K-1) - U(N,K-2))/(K-1) + U(N,K-2) - for all N>1, and since U(N,K) is symmetric, a similar relation holds for two - previous values with the same K: - U(N,K>1) = ((2*K-1)*U(N-1,K) - U(N-2,K))/(N-1) + U(N-2,K) - for all K>1. - This allows us to construct an arbitrary row of the U(N,K) table by starting - with the first two values, which are constants. - This saves roughly 2/3 the work in our O(NK) decoding routine, but costs O(K) - multiplications. - Similar relations can be derived for V(N,K), but are not used here. - - For N>0 and K>0, U(N,K) and V(N,K) take on the form of an (N-1)-degree - polynomial for fixed N. - The first few are - U(1,K) = 1, - U(2,K) = 2*K-1, - U(3,K) = (2*K-2)*K+1, - U(4,K) = (((4*K-6)*K+8)*K-3)/3, - U(5,K) = ((((2*K-4)*K+10)*K-8)*K+3)/3, - and - V(1,K) = 2, - V(2,K) = 4*K, - V(3,K) = 4*K*K+2, - V(4,K) = 8*(K*K+2)*K/3, - V(5,K) = ((4*K*K+20)*K*K+6)/3, - for all K>0. - This allows us to derive O(N) encoding and O(N*log(K)) decoding routines for - small N (and indeed decoding is also O(N) for N<3). - - @ARTICLE{Fis86, - author="Thomas R. Fischer", - title="A Pyramid Vector Quantizer", - journal="IEEE Transactions on Information Theory", - volume="IT-32", - number=4, - pages="568--583", - month=Jul, - year=1986 - }*/ - -#ifndef SMALL_FOOTPRINT -/*Compute U(2,_k). - Note that this may be called with _k=32768 (maxK[2]+1).*/ -static inline unsigned ucwrs2(unsigned _k){ - celt_assert(_k>0); - return _k+(_k-1); -} - -/*Compute V(2,_k).*/ -static inline opus_uint32 ncwrs2(int _k){ - celt_assert(_k>0); - return 4*(opus_uint32)_k; -} - -/*Compute U(3,_k). - Note that this may be called with _k=32768 (maxK[3]+1).*/ -static inline opus_uint32 ucwrs3(unsigned _k){ - celt_assert(_k>0); - return (2*(opus_uint32)_k-2)*_k+1; -} - -/*Compute V(3,_k).*/ -static inline opus_uint32 ncwrs3(int _k){ - celt_assert(_k>0); - return 2*(2*(unsigned)_k*(opus_uint32)_k+1); -} - -/*Compute U(4,_k).*/ -static inline opus_uint32 ucwrs4(int _k){ - celt_assert(_k>0); - return imusdiv32odd(2*_k,(2*_k-3)*(opus_uint32)_k+4,3,1); -} - -/*Compute V(4,_k).*/ -static inline opus_uint32 ncwrs4(int _k){ - celt_assert(_k>0); - return ((_k*(opus_uint32)_k+2)*_k)/3<<3; -} - -#endif /* SMALL_FOOTPRINT */ - -/*Computes the next row/column of any recurrence that obeys the relation - u[i][j]=u[i-1][j]+u[i][j-1]+u[i-1][j-1]. - _ui0 is the base case for the new row/column.*/ -static inline void unext(opus_uint32 *_ui,unsigned _len,opus_uint32 _ui0){ - opus_uint32 ui1; - unsigned j; - /*This do-while will overrun the array if we don't have storage for at least - 2 values.*/ - j=1; do { - ui1=UADD32(UADD32(_ui[j],_ui[j-1]),_ui0); - _ui[j-1]=_ui0; - _ui0=ui1; - } while (++j<_len); - _ui[j-1]=_ui0; -} - -/*Computes the previous row/column of any recurrence that obeys the relation - u[i-1][j]=u[i][j]-u[i][j-1]-u[i-1][j-1]. - _ui0 is the base case for the new row/column.*/ -static inline void uprev(opus_uint32 *_ui,unsigned _n,opus_uint32 _ui0){ - opus_uint32 ui1; - unsigned j; - /*This do-while will overrun the array if we don't have storage for at least - 2 values.*/ - j=1; do { - ui1=USUB32(USUB32(_ui[j],_ui[j-1]),_ui0); - _ui[j-1]=_ui0; - _ui0=ui1; - } while (++j<_n); - _ui[j-1]=_ui0; -} - -/*Compute V(_n,_k), as well as U(_n,0..._k+1). - _u: On exit, _u[i] contains U(_n,i) for i in [0..._k+1].*/ -static opus_uint32 ncwrs_urow(unsigned _n,unsigned _k,opus_uint32 *_u){ - opus_uint32 um2; - unsigned len; - unsigned k; - len=_k+2; - /*We require storage at least 3 values (e.g., _k>0).*/ - celt_assert(len>=3); - _u[0]=0; - _u[1]=um2=1; -#ifndef SMALL_FOOTPRINT - /*_k>52 doesn't work in the false branch due to the limits of INV_TABLE, - but _k isn't tested here because k<=52 for n=7*/ - if(_n<=6) -#endif - { - /*If _n==0, _u[0] should be 1 and the rest should be 0.*/ - /*If _n==1, _u[i] should be 1 for i>1.*/ - celt_assert(_n>=2); - /*If _k==0, the following do-while loop will overflow the buffer.*/ - celt_assert(_k>0); - k=2; - do _u[k]=(k<<1)-1; - while(++k<len); - for(k=2;k<_n;k++)unext(_u+1,_k+1,1); - } -#ifndef SMALL_FOOTPRINT - else{ - opus_uint32 um1; - opus_uint32 n2m1; - _u[2]=n2m1=um1=(_n<<1)-1; - for(k=3;k<len;k++){ - /*U(N,K) = ((2*N-1)*U(N,K-1)-U(N,K-2))/(K-1) + U(N,K-2)*/ - _u[k]=um2=imusdiv32even(n2m1,um1,um2,k-1)+um2; - if(++k>=len)break; - _u[k]=um1=imusdiv32odd(n2m1,um2,um1,(k-1)>>1)+um1; - } - } -#endif /* SMALL_FOOTPRINT */ - return _u[_k]+_u[_k+1]; -} - -#ifndef SMALL_FOOTPRINT - -/*Returns the _i'th combination of _k elements (at most 32767) chosen from a - set of size 1 with associated sign bits. - _y: Returns the vector of pulses.*/ -static inline void cwrsi1(int _k,opus_uint32 _i,int *_y){ - int s; - s=-(int)_i; - _y[0]=(_k+s)^s; -} - -/*Returns the _i'th combination of _k elements (at most 32767) chosen from a - set of size 2 with associated sign bits. - _y: Returns the vector of pulses.*/ -static inline void cwrsi2(int _k,opus_uint32 _i,int *_y){ - opus_uint32 p; - int s; - int yj; - p=ucwrs2(_k+1U); - s=-(_i>=p); - _i-=p&s; - yj=_k; - _k=(_i+1)>>1; - p=_k?ucwrs2(_k):0; - _i-=p; - yj-=_k; - _y[0]=(yj+s)^s; - cwrsi1(_k,_i,_y+1); -} - -/*Returns the _i'th combination of _k elements (at most 32767) chosen from a - set of size 3 with associated sign bits. - _y: Returns the vector of pulses.*/ -static void cwrsi3(int _k,opus_uint32 _i,int *_y){ - opus_uint32 p; - int s; - int yj; - p=ucwrs3(_k+1U); - s=-(_i>=p); - _i-=p&s; - yj=_k; - /*Finds the maximum _k such that ucwrs3(_k)<=_i (tested for all - _i<2147418113=U(3,32768)).*/ - _k=_i>0?(isqrt32(2*_i-1)+1)>>1:0; - p=_k?ucwrs3(_k):0; - _i-=p; - yj-=_k; - _y[0]=(yj+s)^s; - cwrsi2(_k,_i,_y+1); -} - -/*Returns the _i'th combination of _k elements (at most 1172) chosen from a set - of size 4 with associated sign bits. - _y: Returns the vector of pulses.*/ -static void cwrsi4(int _k,opus_uint32 _i,int *_y){ - opus_uint32 p; - int s; - int yj; - int kl; - int kr; - p=ucwrs4(_k+1); - s=-(_i>=p); - _i-=p&s; - yj=_k; - /*We could solve a cubic for k here, but the form of the direct solution does - not lend itself well to exact integer arithmetic. - Instead we do a binary search on U(4,K).*/ - kl=0; - kr=_k; - for(;;){ - _k=(kl+kr)>>1; - p=_k?ucwrs4(_k):0; - if(p<_i){ - if(_k>=kr)break; - kl=_k+1; - } - else if(p>_i)kr=_k-1; - else break; - } - _i-=p; - yj-=_k; - _y[0]=(yj+s)^s; - cwrsi3(_k,_i,_y+1); -} - -#endif /* SMALL_FOOTPRINT */ - -/*Returns the _i'th combination of _k elements chosen from a set of size _n - with associated sign bits. - _y: Returns the vector of pulses. - _u: Must contain entries [0..._k+1] of row _n of U() on input. - Its contents will be destructively modified.*/ -static void cwrsi(int _n,int _k,opus_uint32 _i,int *_y,opus_uint32 *_u){ - int j; - celt_assert(_n>0); - j=0; - do{ - opus_uint32 p; - int s; - int yj; - p=_u[_k+1]; - s=-(_i>=p); - _i-=p&s; - yj=_k; - p=_u[_k]; - while(p>_i)p=_u[--_k]; - _i-=p; - yj-=_k; - _y[j]=(yj+s)^s; - uprev(_u,_k+2,0); - } - while(++j<_n); -} - -/*Returns the index of the given combination of K elements chosen from a set - of size 1 with associated sign bits. - _y: The vector of pulses, whose sum of absolute values is K. - _k: Returns K.*/ -static inline opus_uint32 icwrs1(const int *_y,int *_k){ - *_k=abs(_y[0]); - return _y[0]<0; -} - -#ifndef SMALL_FOOTPRINT - -/*Returns the index of the given combination of K elements chosen from a set - of size 2 with associated sign bits. - _y: The vector of pulses, whose sum of absolute values is K. - _k: Returns K.*/ -static inline opus_uint32 icwrs2(const int *_y,int *_k){ - opus_uint32 i; - int k; - i=icwrs1(_y+1,&k); - i+=k?ucwrs2(k):0; - k+=abs(_y[0]); - if(_y[0]<0)i+=ucwrs2(k+1U); - *_k=k; - return i; -} - -/*Returns the index of the given combination of K elements chosen from a set - of size 3 with associated sign bits. - _y: The vector of pulses, whose sum of absolute values is K. - _k: Returns K.*/ -static inline opus_uint32 icwrs3(const int *_y,int *_k){ - opus_uint32 i; - int k; - i=icwrs2(_y+1,&k); - i+=k?ucwrs3(k):0; - k+=abs(_y[0]); - if(_y[0]<0)i+=ucwrs3(k+1U); - *_k=k; - return i; -} - -/*Returns the index of the given combination of K elements chosen from a set - of size 4 with associated sign bits. - _y: The vector of pulses, whose sum of absolute values is K. - _k: Returns K.*/ -static inline opus_uint32 icwrs4(const int *_y,int *_k){ - opus_uint32 i; - int k; - i=icwrs3(_y+1,&k); - i+=k?ucwrs4(k):0; - k+=abs(_y[0]); - if(_y[0]<0)i+=ucwrs4(k+1); - *_k=k; - return i; -} - -#endif /* SMALL_FOOTPRINT */ - -/*Returns the index of the given combination of K elements chosen from a set - of size _n with associated sign bits. - _y: The vector of pulses, whose sum of absolute values must be _k. - _nc: Returns V(_n,_k).*/ -static inline opus_uint32 icwrs(int _n,int _k,opus_uint32 *_nc,const int *_y, - opus_uint32 *_u){ - opus_uint32 i; - int j; - int k; - /*We can't unroll the first two iterations of the loop unless _n>=2.*/ - celt_assert(_n>=2); - _u[0]=0; - for(k=1;k<=_k+1;k++)_u[k]=(k<<1)-1; - i=icwrs1(_y+_n-1,&k); - j=_n-2; - i+=_u[k]; - k+=abs(_y[j]); - if(_y[j]<0)i+=_u[k+1]; - while(j-->0){ - unext(_u,_k+2,0); - i+=_u[k]; - k+=abs(_y[j]); - if(_y[j]<0)i+=_u[k+1]; - } - *_nc=_u[k]+_u[k+1]; - return i; -} - -#ifdef CUSTOM_MODES -void get_required_bits(opus_int16 *_bits,int _n,int _maxk,int _frac){ - int k; - /*_maxk==0 => there's nothing to do.*/ - celt_assert(_maxk>0); - _bits[0]=0; - if (_n==1) - { - for (k=1;k<=_maxk;k++) - _bits[k] = 1<<_frac; - } - else { - VARDECL(opus_uint32,u); - SAVE_STACK; - ALLOC(u,_maxk+2U,opus_uint32); - ncwrs_urow(_n,_maxk,u); - for(k=1;k<=_maxk;k++) - _bits[k]=log2_frac(u[k]+u[k+1],_frac); - RESTORE_STACK; - } -} -#endif /* CUSTOM_MODES */ - -void encode_pulses(const int *_y,int _n,int _k,ec_enc *_enc){ - opus_uint32 i; - celt_assert(_k>0); -#ifndef SMALL_FOOTPRINT - switch(_n){ - case 2:{ - i=icwrs2(_y,&_k); - ec_enc_uint(_enc,i,ncwrs2(_k)); - }break; - case 3:{ - i=icwrs3(_y,&_k); - ec_enc_uint(_enc,i,ncwrs3(_k)); - }break; - case 4:{ - i=icwrs4(_y,&_k); - ec_enc_uint(_enc,i,ncwrs4(_k)); - }break; - default: - { -#endif - VARDECL(opus_uint32,u); - opus_uint32 nc; - SAVE_STACK; - ALLOC(u,_k+2U,opus_uint32); - i=icwrs(_n,_k,&nc,_y,u); - ec_enc_uint(_enc,i,nc); - RESTORE_STACK; -#ifndef SMALL_FOOTPRINT - } - break; - } -#endif -} - -void decode_pulses(int *_y,int _n,int _k,ec_dec *_dec) -{ - celt_assert(_k>0); -#ifndef SMALL_FOOTPRINT - switch(_n){ - case 2:cwrsi2(_k,ec_dec_uint(_dec,ncwrs2(_k)),_y);break; - case 3:cwrsi3(_k,ec_dec_uint(_dec,ncwrs3(_k)),_y);break; - case 4:cwrsi4(_k,ec_dec_uint(_dec,ncwrs4(_k)),_y);break; - default: - { -#endif - VARDECL(opus_uint32,u); - SAVE_STACK; - ALLOC(u,_k+2U,opus_uint32); - cwrsi(_n,_k,ec_dec_uint(_dec,ncwrs_urow(_n,_k,u)),_y,u); - RESTORE_STACK; -#ifndef SMALL_FOOTPRINT - } - break; - } -#endif -} diff --git a/src/opus-1.0.2/src/opus_multistream.c b/src/opus-1.0.2/src/opus_multistream.c deleted file mode 100644 index a7f25a52..00000000 --- a/src/opus-1.0.2/src/opus_multistream.c +++ /dev/null @@ -1,1027 +0,0 @@ -/* Copyright (c) 2011 Xiph.Org Foundation - Written by Jean-Marc Valin */ -/* - Redistribution and use in source and binary forms, with or without - modification, are permitted provided that the following conditions - are met: - - - Redistributions of source code must retain the above copyright - notice, this list of conditions and the following disclaimer. - - - Redistributions in binary form must reproduce the above copyright - notice, this list of conditions and the following disclaimer in the - documentation and/or other materials provided with the distribution. - - THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER - OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, - EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, - PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR - PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF - LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING - NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS - SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -*/ - -#ifdef HAVE_CONFIG_H -#include "config.h" -#endif - -#include "opus_multistream.h" -#include "opus.h" -#include "opus_private.h" -#include "stack_alloc.h" -#include <stdarg.h> -#include "float_cast.h" -#include "os_support.h" - -typedef struct ChannelLayout { - int nb_channels; - int nb_streams; - int nb_coupled_streams; - unsigned char mapping[256]; -} ChannelLayout; - -struct OpusMSEncoder { - ChannelLayout layout; - int bitrate; - /* Encoder states go here */ -}; - -struct OpusMSDecoder { - ChannelLayout layout; - /* Decoder states go here */ -}; - -#ifdef FIXED_POINT -#define opus_encode_native opus_encode -#else -#define opus_encode_native opus_encode_float -#endif - -static int validate_layout(const ChannelLayout *layout) -{ - int i, max_channel; - - max_channel = layout->nb_streams+layout->nb_coupled_streams; - if (max_channel>255) - return 0; - for (i=0;i<layout->nb_channels;i++) - { - if (layout->mapping[i] >= max_channel && layout->mapping[i] != 255) - return 0; - } - return 1; -} - - -static int get_left_channel(const ChannelLayout *layout, int stream_id, int prev) -{ - int i; - i = (prev<0) ? 0 : prev+1; - for (;i<layout->nb_channels;i++) - { - if (layout->mapping[i]==stream_id*2) - return i; - } - return -1; -} - -static int get_right_channel(const ChannelLayout *layout, int stream_id, int prev) -{ - int i; - i = (prev<0) ? 0 : prev+1; - for (;i<layout->nb_channels;i++) - { - if (layout->mapping[i]==stream_id*2+1) - return i; - } - return -1; -} - -static int get_mono_channel(const ChannelLayout *layout, int stream_id, int prev) -{ - int i; - i = (prev<0) ? 0 : prev+1; - for (;i<layout->nb_channels;i++) - { - if (layout->mapping[i]==stream_id+layout->nb_coupled_streams) - return i; - } - return -1; -} - -static int validate_encoder_layout(const ChannelLayout *layout) -{ - int s; - for (s=0;s<layout->nb_streams;s++) - { - if (s < layout->nb_coupled_streams) - { - if (get_left_channel(layout, s, -1)==-1) - return 0; - if (get_right_channel(layout, s, -1)==-1) - return 0; - } else { - if (get_mono_channel(layout, s, -1)==-1) - return 0; - } - } - return 1; -} - -opus_int32 opus_multistream_encoder_get_size(int nb_streams, int nb_coupled_streams) -{ - int coupled_size; - int mono_size; - - if(nb_streams<1||nb_coupled_streams>nb_streams||nb_coupled_streams<0)return 0; - coupled_size = opus_encoder_get_size(2); - mono_size = opus_encoder_get_size(1); - return align(sizeof(OpusMSEncoder)) - + nb_coupled_streams * align(coupled_size) - + (nb_streams-nb_coupled_streams) * align(mono_size); -} - - - -int opus_multistream_encoder_init( - OpusMSEncoder *st, - opus_int32 Fs, - int channels, - int streams, - int coupled_streams, - const unsigned char *mapping, - int application -) -{ - int coupled_size; - int mono_size; - int i, ret; - char *ptr; - - if ((channels>255) || (channels<1) || (coupled_streams>streams) || - (coupled_streams+streams>255) || (streams<1) || (coupled_streams<0)) - return OPUS_BAD_ARG; - - st->layout.nb_channels = channels; - st->layout.nb_streams = streams; - st->layout.nb_coupled_streams = coupled_streams; - - for (i=0;i<st->layout.nb_channels;i++) - st->layout.mapping[i] = mapping[i]; - if (!validate_layout(&st->layout) || !validate_encoder_layout(&st->layout)) - return OPUS_BAD_ARG; - ptr = (char*)st + align(sizeof(OpusMSEncoder)); - coupled_size = opus_encoder_get_size(2); - mono_size = opus_encoder_get_size(1); - - for (i=0;i<st->layout.nb_coupled_streams;i++) - { - ret = opus_encoder_init((OpusEncoder*)ptr, Fs, 2, application); - if(ret!=OPUS_OK)return ret; - ptr += align(coupled_size); - } - for (;i<st->layout.nb_streams;i++) - { - ret = opus_encoder_init((OpusEncoder*)ptr, Fs, 1, application); - if(ret!=OPUS_OK)return ret; - ptr += align(mono_size); - } - return OPUS_OK; -} - -OpusMSEncoder *opus_multistream_encoder_create( - opus_int32 Fs, - int channels, - int streams, - int coupled_streams, - const unsigned char *mapping, - int application, - int *error -) -{ - int ret; - OpusMSEncoder *st; - if ((channels>255) || (channels<1) || (coupled_streams>streams) || - (coupled_streams+streams>255) || (streams<1) || (coupled_streams<0)) - { - if (error) - *error = OPUS_BAD_ARG; - return NULL; - } - st = (OpusMSEncoder *)opus_alloc(opus_multistream_encoder_get_size(streams, coupled_streams)); - if (st==NULL) - { - if (error) - *error = OPUS_ALLOC_FAIL; - return NULL; - } - ret = opus_multistream_encoder_init(st, Fs, channels, streams, coupled_streams, mapping, application); - if (ret != OPUS_OK) - { - opus_free(st); - st = NULL; - } - if (error) - *error = ret; - return st; -} - -typedef void (*opus_copy_channel_in_func)( - opus_val16 *dst, - int dst_stride, - const void *src, - int src_stride, - int src_channel, - int frame_size -); - -/* Max size in case the encoder decides to return three frames */ -#define MS_FRAME_TMP (3*1275+7) -static int opus_multistream_encode_native -( - OpusMSEncoder *st, - opus_copy_channel_in_func copy_channel_in, - const void *pcm, - int frame_size, - unsigned char *data, - opus_int32 max_data_bytes -) -{ - opus_int32 Fs; - int coupled_size; - int mono_size; - int s; - char *ptr; - int tot_size; - VARDECL(opus_val16, buf); - unsigned char tmp_data[MS_FRAME_TMP]; - OpusRepacketizer rp; - ALLOC_STACK; - - ptr = (char*)st + align(sizeof(OpusMSEncoder)); - opus_encoder_ctl((OpusEncoder*)ptr, OPUS_GET_SAMPLE_RATE(&Fs)); - /* Validate frame_size before using it to allocate stack space. - This mirrors the checks in opus_encode[_float](). */ - if (400*frame_size != Fs && 200*frame_size != Fs && - 100*frame_size != Fs && 50*frame_size != Fs && - 25*frame_size != Fs && 50*frame_size != 3*Fs) - { - RESTORE_STACK; - return OPUS_BAD_ARG; - } - ALLOC(buf, 2*frame_size, opus_val16); - coupled_size = opus_encoder_get_size(2); - mono_size = opus_encoder_get_size(1); - - if (max_data_bytes < 4*st->layout.nb_streams-1) - { - RESTORE_STACK; - return OPUS_BUFFER_TOO_SMALL; - } - /* Counting ToC */ - tot_size = 0; - for (s=0;s<st->layout.nb_streams;s++) - { - OpusEncoder *enc; - int len; - int curr_max; - - opus_repacketizer_init(&rp); - enc = (OpusEncoder*)ptr; - if (s < st->layout.nb_coupled_streams) - { - int left, right; - left = get_left_channel(&st->layout, s, -1); - right = get_right_channel(&st->layout, s, -1); - (*copy_channel_in)(buf, 2, - pcm, st->layout.nb_channels, left, frame_size); - (*copy_channel_in)(buf+1, 2, - pcm, st->layout.nb_channels, right, frame_size); - ptr += align(coupled_size); - } else { - int chan = get_mono_channel(&st->layout, s, -1); - (*copy_channel_in)(buf, 1, - pcm, st->layout.nb_channels, chan, frame_size); - ptr += align(mono_size); - } - /* number of bytes left (+Toc) */ - curr_max = max_data_bytes - tot_size; - /* Reserve three bytes for the last stream and four for the others */ - curr_max -= IMAX(0,4*(st->layout.nb_streams-s-1)-1); - curr_max = IMIN(curr_max,MS_FRAME_TMP); - len = opus_encode_native(enc, buf, frame_size, tmp_data, curr_max); - if (len<0) - { - RESTORE_STACK; - return len; - } - /* We need to use the repacketizer to add the self-delimiting lengths - while taking into account the fact that the encoder can now return - more than one frame at a time (e.g. 60 ms CELT-only) */ - opus_repacketizer_cat(&rp, tmp_data, len); - len = opus_repacketizer_out_range_impl(&rp, 0, opus_repacketizer_get_nb_frames(&rp), data, max_data_bytes-tot_size, s != st->layout.nb_streams-1); - data += len; - tot_size += len; - } - RESTORE_STACK; - return tot_size; - -} - -#if !defined(DISABLE_FLOAT_API) -static void opus_copy_channel_in_float( - opus_val16 *dst, - int dst_stride, - const void *src, - int src_stride, - int src_channel, - int frame_size -) -{ - const float *float_src; - int i; - float_src = (const float *)src; - for (i=0;i<frame_size;i++) -#if defined(FIXED_POINT) - dst[i*dst_stride] = FLOAT2INT16(float_src[i*src_stride+src_channel]); -#else - dst[i*dst_stride] = float_src[i*src_stride+src_channel]; -#endif -} -#endif - -static void opus_copy_channel_in_short( - opus_val16 *dst, - int dst_stride, - const void *src, - int src_stride, - int src_channel, - int frame_size -) -{ - const opus_int16 *short_src; - int i; - short_src = (const opus_int16 *)src; - for (i=0;i<frame_size;i++) -#if defined(FIXED_POINT) - dst[i*dst_stride] = short_src[i*src_stride+src_channel]; -#else - dst[i*dst_stride] = (1/32768.f)*short_src[i*src_stride+src_channel]; -#endif -} - -#ifdef FIXED_POINT -int opus_multistream_encode( - OpusMSEncoder *st, - const opus_val16 *pcm, - int frame_size, - unsigned char *data, - opus_int32 max_data_bytes -) -{ - return opus_multistream_encode_native(st, opus_copy_channel_in_short, - pcm, frame_size, data, max_data_bytes); -} - -#ifndef DISABLE_FLOAT_API -int opus_multistream_encode_float( - OpusMSEncoder *st, - const float *pcm, - int frame_size, - unsigned char *data, - opus_int32 max_data_bytes -) -{ - return opus_multistream_encode_native(st, opus_copy_channel_in_float, - pcm, frame_size, data, max_data_bytes); -} -#endif - -#else - -int opus_multistream_encode_float -( - OpusMSEncoder *st, - const opus_val16 *pcm, - int frame_size, - unsigned char *data, - opus_int32 max_data_bytes -) -{ - return opus_multistream_encode_native(st, opus_copy_channel_in_float, - pcm, frame_size, data, max_data_bytes); -} - -int opus_multistream_encode( - OpusMSEncoder *st, - const opus_int16 *pcm, - int frame_size, - unsigned char *data, - opus_int32 max_data_bytes -) -{ - return opus_multistream_encode_native(st, opus_copy_channel_in_short, - pcm, frame_size, data, max_data_bytes); -} -#endif - -int opus_multistream_encoder_ctl(OpusMSEncoder *st, int request, ...) -{ - va_list ap; - int coupled_size, mono_size; - char *ptr; - int ret = OPUS_OK; - - va_start(ap, request); - - coupled_size = opus_encoder_get_size(2); - mono_size = opus_encoder_get_size(1); - ptr = (char*)st + align(sizeof(OpusMSEncoder)); - switch (request) - { - case OPUS_SET_BITRATE_REQUEST: - { - int chan, s; - opus_int32 value = va_arg(ap, opus_int32); - chan = st->layout.nb_streams + st->layout.nb_coupled_streams; - value /= chan; - for (s=0;s<st->layout.nb_streams;s++) - { - OpusEncoder *enc; - enc = (OpusEncoder*)ptr; - if (s < st->layout.nb_coupled_streams) - ptr += align(coupled_size); - else - ptr += align(mono_size); - opus_encoder_ctl(enc, request, value * (s < st->layout.nb_coupled_streams ? 2 : 1)); - } - } - break; - case OPUS_GET_BITRATE_REQUEST: - { - int s; - opus_int32 *value = va_arg(ap, opus_int32*); - *value = 0; - for (s=0;s<st->layout.nb_streams;s++) - { - opus_int32 rate; - OpusEncoder *enc; - enc = (OpusEncoder*)ptr; - if (s < st->layout.nb_coupled_streams) - ptr += align(coupled_size); - else - ptr += align(mono_size); - opus_encoder_ctl(enc, request, &rate); - *value += rate; - } - } - break; - case OPUS_GET_LSB_DEPTH_REQUEST: - case OPUS_GET_VBR_REQUEST: - case OPUS_GET_APPLICATION_REQUEST: - case OPUS_GET_BANDWIDTH_REQUEST: - case OPUS_GET_COMPLEXITY_REQUEST: - case OPUS_GET_PACKET_LOSS_PERC_REQUEST: - case OPUS_GET_DTX_REQUEST: - case OPUS_GET_VOICE_RATIO_REQUEST: - case OPUS_GET_VBR_CONSTRAINT_REQUEST: - case OPUS_GET_SIGNAL_REQUEST: - case OPUS_GET_LOOKAHEAD_REQUEST: - case OPUS_GET_SAMPLE_RATE_REQUEST: - case OPUS_GET_INBAND_FEC_REQUEST: - { - OpusEncoder *enc; - /* For int32* GET params, just query the first stream */ - opus_int32 *value = va_arg(ap, opus_int32*); - enc = (OpusEncoder*)ptr; - ret = opus_encoder_ctl(enc, request, value); - } - break; - case OPUS_GET_FINAL_RANGE_REQUEST: - { - int s; - opus_uint32 *value = va_arg(ap, opus_uint32*); - opus_uint32 tmp; - *value=0; - for (s=0;s<st->layout.nb_streams;s++) - { - OpusEncoder *enc; - enc = (OpusEncoder*)ptr; - if (s < st->layout.nb_coupled_streams) - ptr += align(coupled_size); - else - ptr += align(mono_size); - ret = opus_encoder_ctl(enc, request, &tmp); - if (ret != OPUS_OK) break; - *value ^= tmp; - } - } - break; - case OPUS_SET_LSB_DEPTH_REQUEST: - case OPUS_SET_COMPLEXITY_REQUEST: - case OPUS_SET_VBR_REQUEST: - case OPUS_SET_VBR_CONSTRAINT_REQUEST: - case OPUS_SET_BANDWIDTH_REQUEST: - case OPUS_SET_SIGNAL_REQUEST: - case OPUS_SET_APPLICATION_REQUEST: - case OPUS_SET_INBAND_FEC_REQUEST: - case OPUS_SET_PACKET_LOSS_PERC_REQUEST: - case OPUS_SET_DTX_REQUEST: - case OPUS_SET_FORCE_MODE_REQUEST: - { - int s; - /* This works for int32 params */ - opus_int32 value = va_arg(ap, opus_int32); - for (s=0;s<st->layout.nb_streams;s++) - { - OpusEncoder *enc; - - enc = (OpusEncoder*)ptr; - if (s < st->layout.nb_coupled_streams) - ptr += align(coupled_size); - else - ptr += align(mono_size); - ret = opus_encoder_ctl(enc, request, value); - if (ret != OPUS_OK) - break; - } - } - break; - case OPUS_MULTISTREAM_GET_ENCODER_STATE_REQUEST: - { - int s; - opus_int32 stream_id; - OpusEncoder **value; - stream_id = va_arg(ap, opus_int32); - if (stream_id<0 || stream_id >= st->layout.nb_streams) - ret = OPUS_BAD_ARG; - value = va_arg(ap, OpusEncoder**); - for (s=0;s<stream_id;s++) - { - if (s < st->layout.nb_coupled_streams) - ptr += align(coupled_size); - else - ptr += align(mono_size); - } - *value = (OpusEncoder*)ptr; - } - break; - default: - ret = OPUS_UNIMPLEMENTED; - break; - } - - va_end(ap); - return ret; -} - -void opus_multistream_encoder_destroy(OpusMSEncoder *st) -{ - opus_free(st); -} - - -/* DECODER */ - -opus_int32 opus_multistream_decoder_get_size(int nb_streams, int nb_coupled_streams) -{ - int coupled_size; - int mono_size; - - if(nb_streams<1||nb_coupled_streams>nb_streams||nb_coupled_streams<0)return 0; - coupled_size = opus_decoder_get_size(2); - mono_size = opus_decoder_get_size(1); - return align(sizeof(OpusMSDecoder)) - + nb_coupled_streams * align(coupled_size) - + (nb_streams-nb_coupled_streams) * align(mono_size); -} - -int opus_multistream_decoder_init( - OpusMSDecoder *st, - opus_int32 Fs, - int channels, - int streams, - int coupled_streams, - const unsigned char *mapping -) -{ - int coupled_size; - int mono_size; - int i, ret; - char *ptr; - - if ((channels>255) || (channels<1) || (coupled_streams>streams) || - (coupled_streams+streams>255) || (streams<1) || (coupled_streams<0)) - return OPUS_BAD_ARG; - - st->layout.nb_channels = channels; - st->layout.nb_streams = streams; - st->layout.nb_coupled_streams = coupled_streams; - - for (i=0;i<st->layout.nb_channels;i++) - st->layout.mapping[i] = mapping[i]; - if (!validate_layout(&st->layout)) - return OPUS_BAD_ARG; - - ptr = (char*)st + align(sizeof(OpusMSDecoder)); - coupled_size = opus_decoder_get_size(2); - mono_size = opus_decoder_get_size(1); - - for (i=0;i<st->layout.nb_coupled_streams;i++) - { - ret=opus_decoder_init((OpusDecoder*)ptr, Fs, 2); - if(ret!=OPUS_OK)return ret; - ptr += align(coupled_size); - } - for (;i<st->layout.nb_streams;i++) - { - ret=opus_decoder_init((OpusDecoder*)ptr, Fs, 1); - if(ret!=OPUS_OK)return ret; - ptr += align(mono_size); - } - return OPUS_OK; -} - - -OpusMSDecoder *opus_multistream_decoder_create( - opus_int32 Fs, - int channels, - int streams, - int coupled_streams, - const unsigned char *mapping, - int *error -) -{ - int ret; - OpusMSDecoder *st; - if ((channels>255) || (channels<1) || (coupled_streams>streams) || - (coupled_streams+streams>255) || (streams<1) || (coupled_streams<0)) - { - if (error) - *error = OPUS_BAD_ARG; - return NULL; - } - st = (OpusMSDecoder *)opus_alloc(opus_multistream_decoder_get_size(streams, coupled_streams)); - if (st==NULL) - { - if (error) - *error = OPUS_ALLOC_FAIL; - return NULL; - } - ret = opus_multistream_decoder_init(st, Fs, channels, streams, coupled_streams, mapping); - if (error) - *error = ret; - if (ret != OPUS_OK) - { - opus_free(st); - st = NULL; - } - return st; -} - -typedef void (*opus_copy_channel_out_func)( - void *dst, - int dst_stride, - int dst_channel, - const opus_val16 *src, - int src_stride, - int frame_size -); - -static int opus_multistream_decode_native( - OpusMSDecoder *st, - const unsigned char *data, - opus_int32 len, - void *pcm, - opus_copy_channel_out_func copy_channel_out, - int frame_size, - int decode_fec -) -{ - opus_int32 Fs; - int coupled_size; - int mono_size; - int s, c; - char *ptr; - int do_plc=0; - VARDECL(opus_val16, buf); - ALLOC_STACK; - - /* Limit frame_size to avoid excessive stack allocations. */ - opus_multistream_decoder_ctl(st, OPUS_GET_SAMPLE_RATE(&Fs)); - frame_size = IMIN(frame_size, Fs/25*3); - ALLOC(buf, 2*frame_size, opus_val16); - ptr = (char*)st + align(sizeof(OpusMSDecoder)); - coupled_size = opus_decoder_get_size(2); - mono_size = opus_decoder_get_size(1); - - if (len==0) - do_plc = 1; - if (len < 0) - return OPUS_BAD_ARG; - if (!do_plc && len < 2*st->layout.nb_streams-1) - return OPUS_INVALID_PACKET; - for (s=0;s<st->layout.nb_streams;s++) - { - OpusDecoder *dec; - int packet_offset, ret; - - dec = (OpusDecoder*)ptr; - ptr += (s < st->layout.nb_coupled_streams) ? align(coupled_size) : align(mono_size); - - if (!do_plc && len<=0) - { - RESTORE_STACK; - return OPUS_INVALID_PACKET; - } - packet_offset = 0; - ret = opus_decode_native(dec, data, len, buf, frame_size, decode_fec, s!=st->layout.nb_streams-1, &packet_offset); - data += packet_offset; - len -= packet_offset; - if (ret > frame_size) - { - RESTORE_STACK; - return OPUS_BUFFER_TOO_SMALL; - } - if (s>0 && ret != frame_size) - { - RESTORE_STACK; - return OPUS_INVALID_PACKET; - } - if (ret <= 0) - { - RESTORE_STACK; - return ret; - } - frame_size = ret; - if (s < st->layout.nb_coupled_streams) - { - int chan, prev; - prev = -1; - /* Copy "left" audio to the channel(s) where it belongs */ - while ( (chan = get_left_channel(&st->layout, s, prev)) != -1) - { - (*copy_channel_out)(pcm, st->layout.nb_channels, chan, - buf, 2, frame_size); - prev = chan; - } - prev = -1; - /* Copy "right" audio to the channel(s) where it belongs */ - while ( (chan = get_right_channel(&st->layout, s, prev)) != -1) - { - (*copy_channel_out)(pcm, st->layout.nb_channels, chan, - buf+1, 2, frame_size); - prev = chan; - } - } else { - int chan, prev; - prev = -1; - /* Copy audio to the channel(s) where it belongs */ - while ( (chan = get_mono_channel(&st->layout, s, prev)) != -1) - { - (*copy_channel_out)(pcm, st->layout.nb_channels, chan, - buf, 1, frame_size); - prev = chan; - } - } - } - /* Handle muted channels */ - for (c=0;c<st->layout.nb_channels;c++) - { - if (st->layout.mapping[c] == 255) - { - (*copy_channel_out)(pcm, st->layout.nb_channels, c, - NULL, 0, frame_size); - } - } - RESTORE_STACK; - return frame_size; -} - -#if !defined(DISABLE_FLOAT_API) -static void opus_copy_channel_out_float( - void *dst, - int dst_stride, - int dst_channel, - const opus_val16 *src, - int src_stride, - int frame_size -) -{ - float *float_dst; - int i; - float_dst = (float*)dst; - if (src != NULL) - { - for (i=0;i<frame_size;i++) -#if defined(FIXED_POINT) - float_dst[i*dst_stride+dst_channel] = (1/32768.f)*src[i*src_stride]; -#else - float_dst[i*dst_stride+dst_channel] = src[i*src_stride]; -#endif - } - else - { - for (i=0;i<frame_size;i++) - float_dst[i*dst_stride+dst_channel] = 0; - } -} -#endif - -static void opus_copy_channel_out_short( - void *dst, - int dst_stride, - int dst_channel, - const opus_val16 *src, - int src_stride, - int frame_size -) -{ - opus_int16 *short_dst; - int i; - short_dst = (opus_int16*)dst; - if (src != NULL) - { - for (i=0;i<frame_size;i++) -#if defined(FIXED_POINT) - short_dst[i*dst_stride+dst_channel] = src[i*src_stride]; -#else - short_dst[i*dst_stride+dst_channel] = FLOAT2INT16(src[i*src_stride]); -#endif - } - else - { - for (i=0;i<frame_size;i++) - short_dst[i*dst_stride+dst_channel] = 0; - } -} - - - -#ifdef FIXED_POINT -int opus_multistream_decode( - OpusMSDecoder *st, - const unsigned char *data, - opus_int32 len, - opus_int16 *pcm, - int frame_size, - int decode_fec -) -{ - return opus_multistream_decode_native(st, data, len, - pcm, opus_copy_channel_out_short, frame_size, decode_fec); -} - -#ifndef DISABLE_FLOAT_API -int opus_multistream_decode_float(OpusMSDecoder *st, const unsigned char *data, - opus_int32 len, float *pcm, int frame_size, int decode_fec) -{ - return opus_multistream_decode_native(st, data, len, - pcm, opus_copy_channel_out_float, frame_size, decode_fec); -} -#endif - -#else - -int opus_multistream_decode(OpusMSDecoder *st, const unsigned char *data, - opus_int32 len, opus_int16 *pcm, int frame_size, int decode_fec) -{ - return opus_multistream_decode_native(st, data, len, - pcm, opus_copy_channel_out_short, frame_size, decode_fec); -} - -int opus_multistream_decode_float( - OpusMSDecoder *st, - const unsigned char *data, - opus_int32 len, - float *pcm, - int frame_size, - int decode_fec -) -{ - return opus_multistream_decode_native(st, data, len, - pcm, opus_copy_channel_out_float, frame_size, decode_fec); -} -#endif - -int opus_multistream_decoder_ctl(OpusMSDecoder *st, int request, ...) -{ - va_list ap; - int coupled_size, mono_size; - char *ptr; - int ret = OPUS_OK; - - va_start(ap, request); - - coupled_size = opus_decoder_get_size(2); - mono_size = opus_decoder_get_size(1); - ptr = (char*)st + align(sizeof(OpusMSDecoder)); - switch (request) - { - case OPUS_GET_BANDWIDTH_REQUEST: - case OPUS_GET_SAMPLE_RATE_REQUEST: - case OPUS_GET_GAIN_REQUEST: - case OPUS_GET_LAST_PACKET_DURATION_REQUEST: - { - OpusDecoder *dec; - /* For int32* GET params, just query the first stream */ - opus_int32 *value = va_arg(ap, opus_int32*); - dec = (OpusDecoder*)ptr; - ret = opus_decoder_ctl(dec, request, value); - } - break; - case OPUS_GET_FINAL_RANGE_REQUEST: - { - int s; - opus_uint32 *value = va_arg(ap, opus_uint32*); - opus_uint32 tmp; - *value = 0; - for (s=0;s<st->layout.nb_streams;s++) - { - OpusDecoder *dec; - dec = (OpusDecoder*)ptr; - if (s < st->layout.nb_coupled_streams) - ptr += align(coupled_size); - else - ptr += align(mono_size); - ret = opus_decoder_ctl(dec, request, &tmp); - if (ret != OPUS_OK) break; - *value ^= tmp; - } - } - break; - case OPUS_RESET_STATE: - { - int s; - for (s=0;s<st->layout.nb_streams;s++) - { - OpusDecoder *dec; - - dec = (OpusDecoder*)ptr; - if (s < st->layout.nb_coupled_streams) - ptr += align(coupled_size); - else - ptr += align(mono_size); - ret = opus_decoder_ctl(dec, OPUS_RESET_STATE); - if (ret != OPUS_OK) - break; - } - } - break; - case OPUS_MULTISTREAM_GET_DECODER_STATE_REQUEST: - { - int s; - opus_int32 stream_id; - OpusDecoder **value; - stream_id = va_arg(ap, opus_int32); - if (stream_id<0 || stream_id >= st->layout.nb_streams) - ret = OPUS_BAD_ARG; - value = va_arg(ap, OpusDecoder**); - for (s=0;s<stream_id;s++) - { - if (s < st->layout.nb_coupled_streams) - ptr += align(coupled_size); - else - ptr += align(mono_size); - } - *value = (OpusDecoder*)ptr; - } - break; - case OPUS_SET_GAIN_REQUEST: - { - int s; - /* This works for int32 params */ - opus_int32 value = va_arg(ap, opus_int32); - for (s=0;s<st->layout.nb_streams;s++) - { - OpusDecoder *dec; - - dec = (OpusDecoder*)ptr; - if (s < st->layout.nb_coupled_streams) - ptr += align(coupled_size); - else - ptr += align(mono_size); - ret = opus_decoder_ctl(dec, request, value); - if (ret != OPUS_OK) - break; - } - } - break; - default: - ret = OPUS_UNIMPLEMENTED; - break; - } - - va_end(ap); - return ret; -} - - -void opus_multistream_decoder_destroy(OpusMSDecoder *st) -{ - opus_free(st); -} diff --git a/src/opus-1.0.2/celt/_kiss_fft_guts.h b/src/opus-1.1/celt/_kiss_fft_guts.h index 33e62c6b..aefe490e 100644 --- a/src/opus-1.0.2/celt/_kiss_fft_guts.h +++ b/src/opus-1.1/celt/_kiss_fft_guts.h @@ -94,6 +94,14 @@ do {(res).r = ADD32((res).r,(a).r); (res).i = SUB32((res).i,(a).i); \ }while(0) +#if defined(OPUS_ARM_INLINE_ASM) +#include "arm/kiss_fft_armv4.h" +#endif + +#if defined(OPUS_ARM_INLINE_EDSP) +#include "arm/kiss_fft_armv5e.h" +#endif + #else /* not FIXED_POINT*/ # define S_MUL(a,b) ( (a)*(b) ) diff --git a/src/opus-1.0.2/celt/arch.h b/src/opus-1.1/celt/arch.h index 03cda40f..3bbcd366 100644 --- a/src/opus-1.0.2/celt/arch.h +++ b/src/opus-1.1/celt/arch.h @@ -35,6 +35,7 @@ #define ARCH_H #include "opus_types.h" +#include "opus_defines.h" # if !defined(__GNUC_PREREQ) # if defined(__GNUC__)&&defined(__GNUC_MINOR__) @@ -54,7 +55,7 @@ #ifdef __GNUC__ __attribute__((noreturn)) #endif -static inline void _celt_fatal(const char *str, const char *file, int line) +static OPUS_INLINE void _celt_fatal(const char *str, const char *file, int line) { fprintf (stderr, "Fatal (internal) error in %s, line %d: %s\n", file, line, str); abort(); @@ -100,6 +101,7 @@ typedef opus_val32 celt_ener; #define DB_SHIFT 10 #define EPSILON 1 +#define VERY_SMALL 0 #define VERY_LARGE16 ((opus_val16)32767) #define Q15_ONE ((opus_val16)32767) @@ -112,10 +114,10 @@ typedef opus_val32 celt_ener; #include "fixed_generic.h" -#ifdef ARM5E_ASM -#include "fixed_arm5e.h" -#elif defined (ARM4_ASM) -#include "fixed_arm4.h" +#ifdef OPUS_ARM_INLINE_EDSP +#include "arm/fixed_armv5e.h" +#elif defined (OPUS_ARM_INLINE_ASM) +#include "arm/fixed_armv4.h" #elif defined (BFIN_ASM) #include "fixed_bfin.h" #elif defined (TI_C5X_ASM) @@ -140,6 +142,7 @@ typedef float celt_ener; #define NORM_SCALING 1.f #define EPSILON 1e-15f +#define VERY_SMALL 1e-30f #define VERY_LARGE16 1e15f #define Q15_ONE ((opus_val16)1.f) @@ -161,6 +164,7 @@ typedef float celt_ener; #define SHR(a,shift) (a) #define SHL(a,shift) (a) #define SATURATE(x,a) (x) +#define SATURATE16(x) (x) #define ROUND16(a,shift) (a) #define HALF16(x) (.5f*(x)) @@ -182,6 +186,7 @@ typedef float celt_ener; #define MAC16_32_Q15(c,a,b) ((c)+(a)*(b)) #define MULT16_16_Q11_32(a,b) ((a)*(b)) +#define MULT16_16_Q11(a,b) ((a)*(b)) #define MULT16_16_Q13(a,b) ((a)*(b)) #define MULT16_16_Q14(a,b) ((a)*(b)) #define MULT16_16_Q15(a,b) ((a)*(b)) diff --git a/src/opus-1.1/celt/arm/arm2gnu.pl b/src/opus-1.1/celt/arm/arm2gnu.pl new file mode 100755 index 00000000..eab42efa --- /dev/null +++ b/src/opus-1.1/celt/arm/arm2gnu.pl @@ -0,0 +1,316 @@ +#!/usr/bin/perl + +my $bigend; # little/big endian +my $nxstack; + +$nxstack = 0; + +eval 'exec /usr/local/bin/perl -S $0 ${1+"$@"}' + if $running_under_some_shell; + +while ($ARGV[0] =~ /^-/) { + $_ = shift; + last if /^--/; + if (/^-n/) { + $nflag++; + next; + } + die "I don't recognize this switch: $_\\n"; +} +$printit++ unless $nflag; + +$\ = "\n"; # automatically add newline on print +$n=0; + +$thumb = 0; # ARM mode by default, not Thumb. +@proc_stack = (); + +LINE: +while (<>) { + + # For ADRLs we need to add a new line after the substituted one. + $addPadding = 0; + + # First, we do not dare to touch *anything* inside double quotes, do we? + # Second, if you want a dollar character in the string, + # insert two of them -- that's how ARM C and assembler treat strings. + s/^([A-Za-z_]\w*)[ \t]+DCB[ \t]*\"/$1: .ascii \"/ && do { s/\$\$/\$/g; next }; + s/\bDCB\b[ \t]*\"/.ascii \"/ && do { s/\$\$/\$/g; next }; + s/^(\S+)\s+RN\s+(\S+)/$1 .req r$2/ && do { s/\$\$/\$/g; next }; + # If there's nothing on a line but a comment, don't try to apply any further + # substitutions (this is a cheap hack to avoid mucking up the license header) + s/^([ \t]*);/$1@/ && do { s/\$\$/\$/g; next }; + # If substituted -- leave immediately ! + + s/@/,:/; + s/;/@/; + while ( /@.*'/ ) { + s/(@.*)'/$1/g; + } + s/\{FALSE\}/0/g; + s/\{TRUE\}/1/g; + s/\{(\w\w\w\w+)\}/$1/g; + s/\bINCLUDE[ \t]*([^ \t\n]+)/.include \"$1\"/; + s/\bGET[ \t]*([^ \t\n]+)/.include \"${ my $x=$1; $x =~ s|\.s|-gnu.S|; \$x }\"/; + s/\bIMPORT\b/.extern/; + s/\bEXPORT\b/.global/; + s/^(\s+)\[/$1IF/; + s/^(\s+)\|/$1ELSE/; + s/^(\s+)\]/$1ENDIF/; + s/IF *:DEF:/ .ifdef/; + s/IF *:LNOT: *:DEF:/ .ifndef/; + s/ELSE/ .else/; + s/ENDIF/ .endif/; + + if( /\bIF\b/ ) { + s/\bIF\b/ .if/; + s/=/==/; + } + if ( $n == 2) { + s/\$/\\/g; + } + if ($n == 1) { + s/\$//g; + s/label//g; + $n = 2; + } + if ( /MACRO/ ) { + s/MACRO *\n/.macro/; + $n=1; + } + if ( /\bMEND\b/ ) { + s/\bMEND\b/.endm/; + $n=0; + } + + # ".rdata" doesn't work in 'as' version 2.13.2, as it is ".rodata" there. + # + if ( /\bAREA\b/ ) { + my $align; + $align = "2"; + if ( /ALIGN=(\d+)/ ) { + $align = $1; + } + if ( /CODE/ ) { + $nxstack = 1; + } + s/^(.+)CODE(.+)READONLY(.*)/ .text/; + s/^(.+)DATA(.+)READONLY(.*)/ .section .rdata/; + s/^(.+)\|\|\.data\|\|(.+)/ .data/; + s/^(.+)\|\|\.bss\|\|(.+)/ .bss/; + s/$/; .p2align $align/; + # Enable NEON instructions but don't produce a binary that requires + # ARMv7. RVCT does not have equivalent directives, so we just do this + # for all CODE areas. + if ( /.text/ ) { + # Separating .arch, .fpu, etc., by semicolons does not work (gas + # thinks the semicolon is part of the arch name, even when there's + # whitespace separating them). Sadly this means our line numbers + # won't match the original source file (we could use the .line + # directive, which is documented to be obsolete, but then gdb will + # show the wrong line in the translated source file). + s/$/; .arch armv7-a\n .fpu neon\n .object_arch armv4t/; + } + } + + s/\|\|\.constdata\$(\d+)\|\|/.L_CONST$1/; # ||.constdata$3|| + s/\|\|\.bss\$(\d+)\|\|/.L_BSS$1/; # ||.bss$2|| + s/\|\|\.data\$(\d+)\|\|/.L_DATA$1/; # ||.data$2|| + s/\|\|([a-zA-Z0-9_]+)\@([a-zA-Z0-9_]+)\|\|/@ $&/; + s/^(\s+)\%(\s)/ .space $1/; + + s/\|(.+)\.(\d+)\|/\.$1_$2/; # |L80.123| -> .L80_123 + s/\bCODE32\b/.code 32/ && do {$thumb = 0}; + s/\bCODE16\b/.code 16/ && do {$thumb = 1}; + if (/\bPROC\b/) + { + my $prefix; + my $proc; + /^([A-Za-z_\.]\w+)\b/; + $proc = $1; + $prefix = ""; + if ($proc) + { + $prefix = $prefix.sprintf("\t.type\t%s, %%function; ",$proc); + push(@proc_stack, $proc); + s/^[A-Za-z_\.]\w+/$&:/; + } + $prefix = $prefix."\t.thumb_func; " if ($thumb); + s/\bPROC\b/@ $&/; + $_ = $prefix.$_; + } + s/^(\s*)(S|Q|SH|U|UQ|UH)ASX\b/$1$2ADDSUBX/; + s/^(\s*)(S|Q|SH|U|UQ|UH)SAX\b/$1$2SUBADDX/; + if (/\bENDP\b/) + { + my $proc; + s/\bENDP\b/@ $&/; + $proc = pop(@proc_stack); + $_ = "\t.size $proc, .-$proc".$_ if ($proc); + } + s/\bSUBT\b/@ $&/; + s/\bDATA\b/@ $&/; # DATA directive is deprecated -- Asm guide, p.7-25 + s/\bKEEP\b/@ $&/; + s/\bEXPORTAS\b/@ $&/; + s/\|\|(.)+\bEQU\b/@ $&/; + s/\|\|([\w\$]+)\|\|/$1/; + s/\bENTRY\b/@ $&/; + s/\bASSERT\b/@ $&/; + s/\bGBLL\b/@ $&/; + s/\bGBLA\b/@ $&/; + s/^\W+OPT\b/@ $&/; + s/:OR:/|/g; + s/:SHL:/<</g; + s/:SHR:/>>/g; + s/:AND:/&/g; + s/:LAND:/&&/g; + s/CPSR/cpsr/; + s/SPSR/spsr/; + s/ALIGN$/.balign 4/; + s/ALIGN\s+([0-9x]+)$/.balign $1/; + s/psr_cxsf/psr_all/; + s/LTORG/.ltorg/; + s/^([A-Za-z_]\w*)[ \t]+EQU/ .set $1,/; + s/^([A-Za-z_]\w*)[ \t]+SETL/ .set $1,/; + s/^([A-Za-z_]\w*)[ \t]+SETA/ .set $1,/; + s/^([A-Za-z_]\w*)[ \t]+\*/ .set $1,/; + + # {PC} + 0xdeadfeed --> . + 0xdeadfeed + s/\{PC\} \+/ \. +/; + + # Single hex constant on the line ! + # + # >>> NOTE <<< + # Double-precision floats in gcc are always mixed-endian, which means + # bytes in two words are little-endian, but words are big-endian. + # So, 0x0000deadfeed0000 would be stored as 0x0000dead at low address + # and 0xfeed0000 at high address. + # + s/\bDCFD\b[ \t]+0x([a-fA-F0-9]{8})([a-fA-F0-9]{8})/.long 0x$1, 0x$2/; + # Only decimal constants on the line, no hex ! + s/\bDCFD\b[ \t]+([0-9\.\-]+)/.double $1/; + + # Single hex constant on the line ! +# s/\bDCFS\b[ \t]+0x([a-f0-9]{8})([a-f0-9]{8})/.long 0x$1, 0x$2/; + # Only decimal constants on the line, no hex ! +# s/\bDCFS\b[ \t]+([0-9\.\-]+)/.double $1/; + s/\bDCFS[ \t]+0x/.word 0x/; + s/\bDCFS\b/.float/; + + s/^([A-Za-z_]\w*)[ \t]+DCD/$1 .word/; + s/\bDCD\b/.word/; + s/^([A-Za-z_]\w*)[ \t]+DCW/$1 .short/; + s/\bDCW\b/.short/; + s/^([A-Za-z_]\w*)[ \t]+DCB/$1 .byte/; + s/\bDCB\b/.byte/; + s/^([A-Za-z_]\w*)[ \t]+\%/.comm $1,/; + s/^[A-Za-z_\.]\w+/$&:/; + s/^(\d+)/$1:/; + s/\%(\d+)/$1b_or_f/; + s/\%[Bb](\d+)/$1b/; + s/\%[Ff](\d+)/$1f/; + s/\%[Ff][Tt](\d+)/$1f/; + s/&([\dA-Fa-f]+)/0x$1/; + if ( /\b2_[01]+\b/ ) { + s/\b2_([01]+)\b/conv$1&&&&/g; + while ( /[01][01][01][01]&&&&/ ) { + s/0000&&&&/&&&&0/g; + s/0001&&&&/&&&&1/g; + s/0010&&&&/&&&&2/g; + s/0011&&&&/&&&&3/g; + s/0100&&&&/&&&&4/g; + s/0101&&&&/&&&&5/g; + s/0110&&&&/&&&&6/g; + s/0111&&&&/&&&&7/g; + s/1000&&&&/&&&&8/g; + s/1001&&&&/&&&&9/g; + s/1010&&&&/&&&&A/g; + s/1011&&&&/&&&&B/g; + s/1100&&&&/&&&&C/g; + s/1101&&&&/&&&&D/g; + s/1110&&&&/&&&&E/g; + s/1111&&&&/&&&&F/g; + } + s/000&&&&/&&&&0/g; + s/001&&&&/&&&&1/g; + s/010&&&&/&&&&2/g; + s/011&&&&/&&&&3/g; + s/100&&&&/&&&&4/g; + s/101&&&&/&&&&5/g; + s/110&&&&/&&&&6/g; + s/111&&&&/&&&&7/g; + s/00&&&&/&&&&0/g; + s/01&&&&/&&&&1/g; + s/10&&&&/&&&&2/g; + s/11&&&&/&&&&3/g; + s/0&&&&/&&&&0/g; + s/1&&&&/&&&&1/g; + s/conv&&&&/0x/g; + } + + if ( /commandline/) + { + if( /-bigend/) + { + $bigend=1; + } + } + + if ( /\bDCDU\b/ ) + { + my $cmd=$_; + my $value; + my $prefix; + my $w1; + my $w2; + my $w3; + my $w4; + + s/\s+DCDU\b/@ $&/; + + $cmd =~ /\bDCDU\b\s+0x(\d+)/; + $value = $1; + $value =~ /(\w\w)(\w\w)(\w\w)(\w\w)/; + $w1 = $1; + $w2 = $2; + $w3 = $3; + $w4 = $4; + + if( $bigend ne "") + { + # big endian + $prefix = "\t.byte\t0x".$w1.";". + "\t.byte\t0x".$w2.";". + "\t.byte\t0x".$w3.";". + "\t.byte\t0x".$w4."; "; + } + else + { + # little endian + $prefix = "\t.byte\t0x".$w4.";". + "\t.byte\t0x".$w3.";". + "\t.byte\t0x".$w2.";". + "\t.byte\t0x".$w1."; "; + } + $_=$prefix.$_; + } + + if ( /\badrl\b/i ) + { + s/\badrl\s+(\w+)\s*,\s*(\w+)/ldr $1,=$2/i; + $addPadding = 1; + } + s/\bEND\b/@ END/; +} continue { + printf ("%s", $_) if $printit; + if ($addPadding != 0) + { + printf (" mov r0,r0\n"); + $addPadding = 0; + } +} +#If we had a code section, mark that this object doesn't need an executable +# stack. +if ($nxstack) { + printf (" .section\t.note.GNU-stack,\"\",\%\%progbits\n"); +} diff --git a/src/opus-1.1/celt/arm/arm_celt_map.c b/src/opus-1.1/celt/arm/arm_celt_map.c new file mode 100644 index 00000000..547a84d1 --- /dev/null +++ b/src/opus-1.1/celt/arm/arm_celt_map.c @@ -0,0 +1,49 @@ +/* Copyright (c) 2010 Xiph.Org Foundation + * Copyright (c) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "pitch.h" + +#if defined(OPUS_HAVE_RTCD) + +# if defined(FIXED_POINT) +opus_val32 (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *, + const opus_val16 *, opus_val32 *, int , int) = { + celt_pitch_xcorr_c, /* ARMv4 */ + MAY_HAVE_EDSP(celt_pitch_xcorr), /* EDSP */ + MAY_HAVE_MEDIA(celt_pitch_xcorr), /* Media */ + MAY_HAVE_NEON(celt_pitch_xcorr) /* NEON */ +}; +# else +# error "Floating-point implementation is not supported by ARM asm yet." \ + "Reconfigure with --disable-rtcd or send patches." +# endif + +#endif diff --git a/src/opus-1.1/celt/arm/armcpu.c b/src/opus-1.1/celt/arm/armcpu.c new file mode 100644 index 00000000..17685258 --- /dev/null +++ b/src/opus-1.1/celt/arm/armcpu.c @@ -0,0 +1,174 @@ +/* Copyright (c) 2010 Xiph.Org Foundation + * Copyright (c) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/* Original code from libtheora modified to suit to Opus */ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#ifdef OPUS_HAVE_RTCD + +#include "armcpu.h" +#include "cpu_support.h" +#include "os_support.h" +#include "opus_types.h" + +#define OPUS_CPU_ARM_V4 (1) +#define OPUS_CPU_ARM_EDSP (1<<1) +#define OPUS_CPU_ARM_MEDIA (1<<2) +#define OPUS_CPU_ARM_NEON (1<<3) + +#if defined(_MSC_VER) +/*For GetExceptionCode() and EXCEPTION_ILLEGAL_INSTRUCTION.*/ +# define WIN32_LEAN_AND_MEAN +# define WIN32_EXTRA_LEAN +# include <windows.h> + +static OPUS_INLINE opus_uint32 opus_cpu_capabilities(void){ + opus_uint32 flags; + flags=0; + /* MSVC has no OPUS_INLINE __asm support for ARM, but it does let you __emit + * instructions via their assembled hex code. + * All of these instructions should be essentially nops. */ +# if defined(OPUS_ARM_MAY_HAVE_EDSP) + __try{ + /*PLD [r13]*/ + __emit(0xF5DDF000); + flags|=OPUS_CPU_ARM_EDSP; + } + __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){ + /*Ignore exception.*/ + } +# if defined(OPUS_ARM_MAY_HAVE_MEDIA) + __try{ + /*SHADD8 r3,r3,r3*/ + __emit(0xE6333F93); + flags|=OPUS_CPU_ARM_MEDIA; + } + __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){ + /*Ignore exception.*/ + } +# if defined(OPUS_ARM_MAY_HAVE_NEON) + __try{ + /*VORR q0,q0,q0*/ + __emit(0xF2200150); + flags|=OPUS_CPU_ARM_NEON; + } + __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){ + /*Ignore exception.*/ + } +# endif +# endif +# endif + return flags; +} + +#elif defined(__linux__) +/* Linux based */ +opus_uint32 opus_cpu_capabilities(void) +{ + opus_uint32 flags = 0; + FILE *cpuinfo; + + /* Reading /proc/self/auxv would be easier, but that doesn't work reliably on + * Android */ + cpuinfo = fopen("/proc/cpuinfo", "r"); + + if(cpuinfo != NULL) + { + /* 512 should be enough for anybody (it's even enough for all the flags that + * x86 has accumulated... so far). */ + char buf[512]; + + while(fgets(buf, 512, cpuinfo) != NULL) + { +# if defined(OPUS_ARM_MAY_HAVE_EDSP) || defined(OPUS_ARM_MAY_HAVE_NEON) + /* Search for edsp and neon flag */ + if(memcmp(buf, "Features", 8) == 0) + { + char *p; +# if defined(OPUS_ARM_MAY_HAVE_EDSP) + p = strstr(buf, " edsp"); + if(p != NULL && (p[5] == ' ' || p[5] == '\n')) + flags |= OPUS_CPU_ARM_EDSP; +# endif + +# if defined(OPUS_ARM_MAY_HAVE_NEON) + p = strstr(buf, " neon"); + if(p != NULL && (p[5] == ' ' || p[5] == '\n')) + flags |= OPUS_CPU_ARM_NEON; +# endif + } +# endif + +# if defined(OPUS_ARM_MAY_HAVE_MEDIA) + /* Search for media capabilities (>= ARMv6) */ + if(memcmp(buf, "CPU architecture:", 17) == 0) + { + int version; + version = atoi(buf+17); + + if(version >= 6) + flags |= OPUS_CPU_ARM_MEDIA; + } +# endif + } + + fclose(cpuinfo); + } + return flags; +} +#else +/* The feature registers which can tell us what the processor supports are + * accessible in priveleged modes only, so we can't have a general user-space + * detection method like on x86.*/ +# error "Configured to use ARM asm but no CPU detection method available for " \ + "your platform. Reconfigure with --disable-rtcd (or send patches)." +#endif + +int opus_select_arch(void) +{ + opus_uint32 flags = opus_cpu_capabilities(); + int arch = 0; + + if(!(flags & OPUS_CPU_ARM_EDSP)) + return arch; + arch++; + + if(!(flags & OPUS_CPU_ARM_MEDIA)) + return arch; + arch++; + + if(!(flags & OPUS_CPU_ARM_NEON)) + return arch; + arch++; + + return arch; +} + +#endif diff --git a/src/opus-1.1/celt/arm/armcpu.h b/src/opus-1.1/celt/arm/armcpu.h new file mode 100644 index 00000000..ac574460 --- /dev/null +++ b/src/opus-1.1/celt/arm/armcpu.h @@ -0,0 +1,71 @@ +/* Copyright (c) 2010 Xiph.Org Foundation + * Copyright (c) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(ARMCPU_H) +# define ARMCPU_H + +# if defined(OPUS_ARM_MAY_HAVE_EDSP) +# define MAY_HAVE_EDSP(name) name ## _edsp +# else +# define MAY_HAVE_EDSP(name) name ## _c +# endif + +# if defined(OPUS_ARM_MAY_HAVE_MEDIA) +# define MAY_HAVE_MEDIA(name) name ## _media +# else +# define MAY_HAVE_MEDIA(name) MAY_HAVE_EDSP(name) +# endif + +# if defined(OPUS_ARM_MAY_HAVE_NEON) +# define MAY_HAVE_NEON(name) name ## _neon +# else +# define MAY_HAVE_NEON(name) MAY_HAVE_MEDIA(name) +# endif + +# if defined(OPUS_ARM_PRESUME_EDSP) +# define PRESUME_EDSP(name) name ## _edsp +# else +# define PRESUME_EDSP(name) name ## _c +# endif + +# if defined(OPUS_ARM_PRESUME_MEDIA) +# define PRESUME_MEDIA(name) name ## _media +# else +# define PRESUME_MEDIA(name) PRESUME_EDSP(name) +# endif + +# if defined(OPUS_ARM_PRESUME_NEON) +# define PRESUME_NEON(name) name ## _neon +# else +# define PRESUME_NEON(name) PRESUME_MEDIA(name) +# endif + +# if defined(OPUS_HAVE_RTCD) +int opus_select_arch(void); +# endif + +#endif diff --git a/src/opus-1.0.2/src/opus.c b/src/opus-1.1/celt/arm/armopts.s.in index d6ae7bab..3d8aaf27 100644 --- a/src/opus-1.0.2/src/opus.c +++ b/src/opus-1.1/celt/arm/armopts.s.in @@ -1,5 +1,4 @@ -/* Copyright (c) 2011 Xiph.Org Foundation, Skype Limited - Written by Jean-Marc Valin and Koen Vos */ +/* Copyright (C) 2013 Mozilla Corporation */ /* Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions @@ -25,23 +24,14 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ -#ifdef HAVE_CONFIG_H -#include "config.h" -#endif +; Set the following to 1 if we have EDSP instructions +; (LDRD/STRD, etc., ARMv5E and later). +OPUS_ARM_MAY_HAVE_EDSP * @OPUS_ARM_MAY_HAVE_EDSP@ -#include "opus.h" -#include "opus_private.h" +; Set the following to 1 if we have ARMv6 media instructions. +OPUS_ARM_MAY_HAVE_MEDIA * @OPUS_ARM_MAY_HAVE_MEDIA@ -int encode_size(int size, unsigned char *data) -{ - if (size < 252) - { - data[0] = size; - return 1; - } else { - data[0] = 252+(size&0x3); - data[1] = (size-(int)data[0])>>2; - return 2; - } -} +; Set the following to 1 if we have NEON (some ARMv7) +OPUS_ARM_MAY_HAVE_NEON * @OPUS_ARM_MAY_HAVE_NEON@ +END diff --git a/src/opus-1.1/celt/arm/celt_pitch_xcorr_arm.s b/src/opus-1.1/celt/arm/celt_pitch_xcorr_arm.s new file mode 100644 index 00000000..09917b16 --- /dev/null +++ b/src/opus-1.1/celt/arm/celt_pitch_xcorr_arm.s @@ -0,0 +1,545 @@ +; Copyright (c) 2007-2008 CSIRO +; Copyright (c) 2007-2009 Xiph.Org Foundation +; Copyright (c) 2013 Parrot +; Written by Aurélien Zanelli +; +; Redistribution and use in source and binary forms, with or without +; modification, are permitted provided that the following conditions +; are met: +; +; - Redistributions of source code must retain the above copyright +; notice, this list of conditions and the following disclaimer. +; +; - Redistributions in binary form must reproduce the above copyright +; notice, this list of conditions and the following disclaimer in the +; documentation and/or other materials provided with the distribution. +; +; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +; ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +; LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +; A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER +; OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, +; EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +; PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +; PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF +; LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING +; NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +; SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + + AREA |.text|, CODE, READONLY + + GET celt/arm/armopts.s + +IF OPUS_ARM_MAY_HAVE_EDSP + EXPORT celt_pitch_xcorr_edsp +ENDIF + +IF OPUS_ARM_MAY_HAVE_NEON + EXPORT celt_pitch_xcorr_neon +ENDIF + +IF OPUS_ARM_MAY_HAVE_NEON + +; Compute sum[k]=sum(x[j]*y[j+k],j=0...len-1), k=0...3 +xcorr_kernel_neon PROC + ; input: + ; r3 = int len + ; r4 = opus_val16 *x + ; r5 = opus_val16 *y + ; q0 = opus_val32 sum[4] + ; output: + ; q0 = opus_val32 sum[4] + ; preserved: r0-r3, r6-r11, d2, q4-q7, q9-q15 + ; internal usage: + ; r12 = int j + ; d3 = y_3|y_2|y_1|y_0 + ; q2 = y_B|y_A|y_9|y_8|y_7|y_6|y_5|y_4 + ; q3 = x_7|x_6|x_5|x_4|x_3|x_2|x_1|x_0 + ; q8 = scratch + ; + ; Load y[0...3] + ; This requires len>0 to always be valid (which we assert in the C code). + VLD1.16 {d5}, [r5]! + SUBS r12, r3, #8 + BLE xcorr_kernel_neon_process4 +; Process 8 samples at a time. +; This loop loads one y value more than we actually need. Therefore we have to +; stop as soon as there are 8 or fewer samples left (instead of 7), to avoid +; reading past the end of the array. +xcorr_kernel_neon_process8 + ; This loop has 19 total instructions (10 cycles to issue, minimum), with + ; - 2 cycles of ARM insrtuctions, + ; - 10 cycles of load/store/byte permute instructions, and + ; - 9 cycles of data processing instructions. + ; On a Cortex A8, we dual-issue the maximum amount (9 cycles) between the + ; latter two categories, meaning the whole loop should run in 10 cycles per + ; iteration, barring cache misses. + ; + ; Load x[0...7] + VLD1.16 {d6, d7}, [r4]! + ; Unlike VMOV, VAND is a data processsing instruction (and doesn't get + ; assembled to VMOV, like VORR would), so it dual-issues with the prior VLD1. + VAND d3, d5, d5 + SUBS r12, r12, #8 + ; Load y[4...11] + VLD1.16 {d4, d5}, [r5]! + VMLAL.S16 q0, d3, d6[0] + VEXT.16 d16, d3, d4, #1 + VMLAL.S16 q0, d4, d7[0] + VEXT.16 d17, d4, d5, #1 + VMLAL.S16 q0, d16, d6[1] + VEXT.16 d16, d3, d4, #2 + VMLAL.S16 q0, d17, d7[1] + VEXT.16 d17, d4, d5, #2 + VMLAL.S16 q0, d16, d6[2] + VEXT.16 d16, d3, d4, #3 + VMLAL.S16 q0, d17, d7[2] + VEXT.16 d17, d4, d5, #3 + VMLAL.S16 q0, d16, d6[3] + VMLAL.S16 q0, d17, d7[3] + BGT xcorr_kernel_neon_process8 +; Process 4 samples here if we have > 4 left (still reading one extra y value). +xcorr_kernel_neon_process4 + ADDS r12, r12, #4 + BLE xcorr_kernel_neon_process2 + ; Load x[0...3] + VLD1.16 d6, [r4]! + ; Use VAND since it's a data processing instruction again. + VAND d4, d5, d5 + SUB r12, r12, #4 + ; Load y[4...7] + VLD1.16 d5, [r5]! + VMLAL.S16 q0, d4, d6[0] + VEXT.16 d16, d4, d5, #1 + VMLAL.S16 q0, d16, d6[1] + VEXT.16 d16, d4, d5, #2 + VMLAL.S16 q0, d16, d6[2] + VEXT.16 d16, d4, d5, #3 + VMLAL.S16 q0, d16, d6[3] +; Process 2 samples here if we have > 2 left (still reading one extra y value). +xcorr_kernel_neon_process2 + ADDS r12, r12, #2 + BLE xcorr_kernel_neon_process1 + ; Load x[0...1] + VLD2.16 {d6[],d7[]}, [r4]! + ; Use VAND since it's a data processing instruction again. + VAND d4, d5, d5 + SUB r12, r12, #2 + ; Load y[4...5] + VLD1.32 {d5[]}, [r5]! + VMLAL.S16 q0, d4, d6 + VEXT.16 d16, d4, d5, #1 + ; Replace bottom copy of {y5,y4} in d5 with {y3,y2} from d4, using VSRI + ; instead of VEXT, since it's a data-processing instruction. + VSRI.64 d5, d4, #32 + VMLAL.S16 q0, d16, d7 +; Process 1 sample using the extra y value we loaded above. +xcorr_kernel_neon_process1 + ; Load next *x + VLD1.16 {d6[]}, [r4]! + ADDS r12, r12, #1 + ; y[0...3] are left in d5 from prior iteration(s) (if any) + VMLAL.S16 q0, d5, d6 + MOVLE pc, lr +; Now process 1 last sample, not reading ahead. + ; Load last *y + VLD1.16 {d4[]}, [r5]! + VSRI.64 d4, d5, #16 + ; Load last *x + VLD1.16 {d6[]}, [r4]! + VMLAL.S16 q0, d4, d6 + MOV pc, lr + ENDP + +; opus_val32 celt_pitch_xcorr_neon(opus_val16 *_x, opus_val16 *_y, +; opus_val32 *xcorr, int len, int max_pitch) +celt_pitch_xcorr_neon PROC + ; input: + ; r0 = opus_val16 *_x + ; r1 = opus_val16 *_y + ; r2 = opus_val32 *xcorr + ; r3 = int len + ; output: + ; r0 = int maxcorr + ; internal usage: + ; r4 = opus_val16 *x (for xcorr_kernel_neon()) + ; r5 = opus_val16 *y (for xcorr_kernel_neon()) + ; r6 = int max_pitch + ; r12 = int j + ; q15 = int maxcorr[4] (q15 is not used by xcorr_kernel_neon()) + STMFD sp!, {r4-r6, lr} + LDR r6, [sp, #16] + VMOV.S32 q15, #1 + ; if (max_pitch < 4) goto celt_pitch_xcorr_neon_process4_done + SUBS r6, r6, #4 + BLT celt_pitch_xcorr_neon_process4_done +celt_pitch_xcorr_neon_process4 + ; xcorr_kernel_neon parameters: + ; r3 = len, r4 = _x, r5 = _y, q0 = {0, 0, 0, 0} + MOV r4, r0 + MOV r5, r1 + VEOR q0, q0, q0 + ; xcorr_kernel_neon only modifies r4, r5, r12, and q0...q3. + ; So we don't save/restore any other registers. + BL xcorr_kernel_neon + SUBS r6, r6, #4 + VST1.32 {q0}, [r2]! + ; _y += 4 + ADD r1, r1, #8 + VMAX.S32 q15, q15, q0 + ; if (max_pitch < 4) goto celt_pitch_xcorr_neon_process4_done + BGE celt_pitch_xcorr_neon_process4 +; We have less than 4 sums left to compute. +celt_pitch_xcorr_neon_process4_done + ADDS r6, r6, #4 + ; Reduce maxcorr to a single value + VMAX.S32 d30, d30, d31 + VPMAX.S32 d30, d30, d30 + ; if (max_pitch <= 0) goto celt_pitch_xcorr_neon_done + BLE celt_pitch_xcorr_neon_done +; Now compute each remaining sum one at a time. +celt_pitch_xcorr_neon_process_remaining + MOV r4, r0 + MOV r5, r1 + VMOV.I32 q0, #0 + SUBS r12, r3, #8 + BLT celt_pitch_xcorr_neon_process_remaining4 +; Sum terms 8 at a time. +celt_pitch_xcorr_neon_process_remaining_loop8 + ; Load x[0...7] + VLD1.16 {q1}, [r4]! + ; Load y[0...7] + VLD1.16 {q2}, [r5]! + SUBS r12, r12, #8 + VMLAL.S16 q0, d4, d2 + VMLAL.S16 q0, d5, d3 + BGE celt_pitch_xcorr_neon_process_remaining_loop8 +; Sum terms 4 at a time. +celt_pitch_xcorr_neon_process_remaining4 + ADDS r12, r12, #4 + BLT celt_pitch_xcorr_neon_process_remaining4_done + ; Load x[0...3] + VLD1.16 {d2}, [r4]! + ; Load y[0...3] + VLD1.16 {d3}, [r5]! + SUB r12, r12, #4 + VMLAL.S16 q0, d3, d2 +celt_pitch_xcorr_neon_process_remaining4_done + ; Reduce the sum to a single value. + VADD.S32 d0, d0, d1 + VPADDL.S32 d0, d0 + ADDS r12, r12, #4 + BLE celt_pitch_xcorr_neon_process_remaining_loop_done +; Sum terms 1 at a time. +celt_pitch_xcorr_neon_process_remaining_loop1 + VLD1.16 {d2[]}, [r4]! + VLD1.16 {d3[]}, [r5]! + SUBS r12, r12, #1 + VMLAL.S16 q0, d2, d3 + BGT celt_pitch_xcorr_neon_process_remaining_loop1 +celt_pitch_xcorr_neon_process_remaining_loop_done + VST1.32 {d0[0]}, [r2]! + VMAX.S32 d30, d30, d0 + SUBS r6, r6, #1 + ; _y++ + ADD r1, r1, #2 + ; if (--max_pitch > 0) goto celt_pitch_xcorr_neon_process_remaining + BGT celt_pitch_xcorr_neon_process_remaining +celt_pitch_xcorr_neon_done + VMOV.32 r0, d30[0] + LDMFD sp!, {r4-r6, pc} + ENDP + +ENDIF + +IF OPUS_ARM_MAY_HAVE_EDSP + +; This will get used on ARMv7 devices without NEON, so it has been optimized +; to take advantage of dual-issuing where possible. +xcorr_kernel_edsp PROC + ; input: + ; r3 = int len + ; r4 = opus_val16 *_x (must be 32-bit aligned) + ; r5 = opus_val16 *_y (must be 32-bit aligned) + ; r6...r9 = opus_val32 sum[4] + ; output: + ; r6...r9 = opus_val32 sum[4] + ; preserved: r0-r5 + ; internal usage + ; r2 = int j + ; r12,r14 = opus_val16 x[4] + ; r10,r11 = opus_val16 y[4] + STMFD sp!, {r2,r4,r5,lr} + LDR r10, [r5], #4 ; Load y[0...1] + SUBS r2, r3, #4 ; j = len-4 + LDR r11, [r5], #4 ; Load y[2...3] + BLE xcorr_kernel_edsp_process4_done + LDR r12, [r4], #4 ; Load x[0...1] + ; Stall +xcorr_kernel_edsp_process4 + ; The multiplies must issue from pipeline 0, and can't dual-issue with each + ; other. Every other instruction here dual-issues with a multiply, and is + ; thus "free". There should be no stalls in the body of the loop. + SMLABB r6, r12, r10, r6 ; sum[0] = MAC16_16(sum[0],x_0,y_0) + LDR r14, [r4], #4 ; Load x[2...3] + SMLABT r7, r12, r10, r7 ; sum[1] = MAC16_16(sum[1],x_0,y_1) + SUBS r2, r2, #4 ; j-=4 + SMLABB r8, r12, r11, r8 ; sum[2] = MAC16_16(sum[2],x_0,y_2) + SMLABT r9, r12, r11, r9 ; sum[3] = MAC16_16(sum[3],x_0,y_3) + SMLATT r6, r12, r10, r6 ; sum[0] = MAC16_16(sum[0],x_1,y_1) + LDR r10, [r5], #4 ; Load y[4...5] + SMLATB r7, r12, r11, r7 ; sum[1] = MAC16_16(sum[1],x_1,y_2) + SMLATT r8, r12, r11, r8 ; sum[2] = MAC16_16(sum[2],x_1,y_3) + SMLATB r9, r12, r10, r9 ; sum[3] = MAC16_16(sum[3],x_1,y_4) + LDRGT r12, [r4], #4 ; Load x[0...1] + SMLABB r6, r14, r11, r6 ; sum[0] = MAC16_16(sum[0],x_2,y_2) + SMLABT r7, r14, r11, r7 ; sum[1] = MAC16_16(sum[1],x_2,y_3) + SMLABB r8, r14, r10, r8 ; sum[2] = MAC16_16(sum[2],x_2,y_4) + SMLABT r9, r14, r10, r9 ; sum[3] = MAC16_16(sum[3],x_2,y_5) + SMLATT r6, r14, r11, r6 ; sum[0] = MAC16_16(sum[0],x_3,y_3) + LDR r11, [r5], #4 ; Load y[6...7] + SMLATB r7, r14, r10, r7 ; sum[1] = MAC16_16(sum[1],x_3,y_4) + SMLATT r8, r14, r10, r8 ; sum[2] = MAC16_16(sum[2],x_3,y_5) + SMLATB r9, r14, r11, r9 ; sum[3] = MAC16_16(sum[3],x_3,y_6) + BGT xcorr_kernel_edsp_process4 +xcorr_kernel_edsp_process4_done + ADDS r2, r2, #4 + BLE xcorr_kernel_edsp_done + LDRH r12, [r4], #2 ; r12 = *x++ + SUBS r2, r2, #1 ; j-- + ; Stall + SMLABB r6, r12, r10, r6 ; sum[0] = MAC16_16(sum[0],x,y_0) + LDRGTH r14, [r4], #2 ; r14 = *x++ + SMLABT r7, r12, r10, r7 ; sum[1] = MAC16_16(sum[1],x,y_1) + SMLABB r8, r12, r11, r8 ; sum[2] = MAC16_16(sum[2],x,y_2) + SMLABT r9, r12, r11, r9 ; sum[3] = MAC16_16(sum[3],x,y_3) + BLE xcorr_kernel_edsp_done + SMLABT r6, r14, r10, r6 ; sum[0] = MAC16_16(sum[0],x,y_1) + SUBS r2, r2, #1 ; j-- + SMLABB r7, r14, r11, r7 ; sum[1] = MAC16_16(sum[1],x,y_2) + LDRH r10, [r5], #2 ; r10 = y_4 = *y++ + SMLABT r8, r14, r11, r8 ; sum[2] = MAC16_16(sum[2],x,y_3) + LDRGTH r12, [r4], #2 ; r12 = *x++ + SMLABB r9, r14, r10, r9 ; sum[3] = MAC16_16(sum[3],x,y_4) + BLE xcorr_kernel_edsp_done + SMLABB r6, r12, r11, r6 ; sum[0] = MAC16_16(sum[0],tmp,y_2) + CMP r2, #1 ; j-- + SMLABT r7, r12, r11, r7 ; sum[1] = MAC16_16(sum[1],tmp,y_3) + LDRH r2, [r5], #2 ; r2 = y_5 = *y++ + SMLABB r8, r12, r10, r8 ; sum[2] = MAC16_16(sum[2],tmp,y_4) + LDRGTH r14, [r4] ; r14 = *x + SMLABB r9, r12, r2, r9 ; sum[3] = MAC16_16(sum[3],tmp,y_5) + BLE xcorr_kernel_edsp_done + SMLABT r6, r14, r11, r6 ; sum[0] = MAC16_16(sum[0],tmp,y_3) + LDRH r11, [r5] ; r11 = y_6 = *y + SMLABB r7, r14, r10, r7 ; sum[1] = MAC16_16(sum[1],tmp,y_4) + SMLABB r8, r14, r2, r8 ; sum[2] = MAC16_16(sum[2],tmp,y_5) + SMLABB r9, r14, r11, r9 ; sum[3] = MAC16_16(sum[3],tmp,y_6) +xcorr_kernel_edsp_done + LDMFD sp!, {r2,r4,r5,pc} + ENDP + +celt_pitch_xcorr_edsp PROC + ; input: + ; r0 = opus_val16 *_x (must be 32-bit aligned) + ; r1 = opus_val16 *_y (only needs to be 16-bit aligned) + ; r2 = opus_val32 *xcorr + ; r3 = int len + ; output: + ; r0 = maxcorr + ; internal usage + ; r4 = opus_val16 *x + ; r5 = opus_val16 *y + ; r6 = opus_val32 sum0 + ; r7 = opus_val32 sum1 + ; r8 = opus_val32 sum2 + ; r9 = opus_val32 sum3 + ; r1 = int max_pitch + ; r12 = int j + STMFD sp!, {r4-r11, lr} + MOV r5, r1 + LDR r1, [sp, #36] + MOV r4, r0 + TST r5, #3 + ; maxcorr = 1 + MOV r0, #1 + BEQ celt_pitch_xcorr_edsp_process1u_done +; Compute one sum at the start to make y 32-bit aligned. + SUBS r12, r3, #4 + ; r14 = sum = 0 + MOV r14, #0 + LDRH r8, [r5], #2 + BLE celt_pitch_xcorr_edsp_process1u_loop4_done + LDR r6, [r4], #4 + MOV r8, r8, LSL #16 +celt_pitch_xcorr_edsp_process1u_loop4 + LDR r9, [r5], #4 + SMLABT r14, r6, r8, r14 ; sum = MAC16_16(sum, x_0, y_0) + LDR r7, [r4], #4 + SMLATB r14, r6, r9, r14 ; sum = MAC16_16(sum, x_1, y_1) + LDR r8, [r5], #4 + SMLABT r14, r7, r9, r14 ; sum = MAC16_16(sum, x_2, y_2) + SUBS r12, r12, #4 ; j-=4 + SMLATB r14, r7, r8, r14 ; sum = MAC16_16(sum, x_3, y_3) + LDRGT r6, [r4], #4 + BGT celt_pitch_xcorr_edsp_process1u_loop4 + MOV r8, r8, LSR #16 +celt_pitch_xcorr_edsp_process1u_loop4_done + ADDS r12, r12, #4 +celt_pitch_xcorr_edsp_process1u_loop1 + LDRGEH r6, [r4], #2 + ; Stall + SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, *x, *y) + SUBGES r12, r12, #1 + LDRGTH r8, [r5], #2 + BGT celt_pitch_xcorr_edsp_process1u_loop1 + ; Restore _x + SUB r4, r4, r3, LSL #1 + ; Restore and advance _y + SUB r5, r5, r3, LSL #1 + ; maxcorr = max(maxcorr, sum) + CMP r0, r14 + ADD r5, r5, #2 + MOVLT r0, r14 + SUBS r1, r1, #1 + ; xcorr[i] = sum + STR r14, [r2], #4 + BLE celt_pitch_xcorr_edsp_done +celt_pitch_xcorr_edsp_process1u_done + ; if (max_pitch < 4) goto celt_pitch_xcorr_edsp_process2 + SUBS r1, r1, #4 + BLT celt_pitch_xcorr_edsp_process2 +celt_pitch_xcorr_edsp_process4 + ; xcorr_kernel_edsp parameters: + ; r3 = len, r4 = _x, r5 = _y, r6...r9 = sum[4] = {0, 0, 0, 0} + MOV r6, #0 + MOV r7, #0 + MOV r8, #0 + MOV r9, #0 + BL xcorr_kernel_edsp ; xcorr_kernel_edsp(_x, _y+i, xcorr+i, len) + ; maxcorr = max(maxcorr, sum0, sum1, sum2, sum3) + CMP r0, r6 + ; _y+=4 + ADD r5, r5, #8 + MOVLT r0, r6 + CMP r0, r7 + MOVLT r0, r7 + CMP r0, r8 + MOVLT r0, r8 + CMP r0, r9 + MOVLT r0, r9 + STMIA r2!, {r6-r9} + SUBS r1, r1, #4 + BGE celt_pitch_xcorr_edsp_process4 +celt_pitch_xcorr_edsp_process2 + ADDS r1, r1, #2 + BLT celt_pitch_xcorr_edsp_process1a + SUBS r12, r3, #4 + ; {r10, r11} = {sum0, sum1} = {0, 0} + MOV r10, #0 + MOV r11, #0 + LDR r8, [r5], #4 + BLE celt_pitch_xcorr_edsp_process2_loop_done + LDR r6, [r4], #4 + LDR r9, [r5], #4 +celt_pitch_xcorr_edsp_process2_loop4 + SMLABB r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_0) + LDR r7, [r4], #4 + SMLABT r11, r6, r8, r11 ; sum1 = MAC16_16(sum1, x_0, y_1) + SUBS r12, r12, #4 ; j-=4 + SMLATT r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_1, y_1) + LDR r8, [r5], #4 + SMLATB r11, r6, r9, r11 ; sum1 = MAC16_16(sum1, x_1, y_2) + LDRGT r6, [r4], #4 + SMLABB r10, r7, r9, r10 ; sum0 = MAC16_16(sum0, x_2, y_2) + SMLABT r11, r7, r9, r11 ; sum1 = MAC16_16(sum1, x_2, y_3) + SMLATT r10, r7, r9, r10 ; sum0 = MAC16_16(sum0, x_3, y_3) + LDRGT r9, [r5], #4 + SMLATB r11, r7, r8, r11 ; sum1 = MAC16_16(sum1, x_3, y_4) + BGT celt_pitch_xcorr_edsp_process2_loop4 +celt_pitch_xcorr_edsp_process2_loop_done + ADDS r12, r12, #2 + BLE celt_pitch_xcorr_edsp_process2_1 + LDR r6, [r4], #4 + ; Stall + SMLABB r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_0) + LDR r9, [r5], #4 + SMLABT r11, r6, r8, r11 ; sum1 = MAC16_16(sum1, x_0, y_1) + SUB r12, r12, #2 + SMLATT r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_1, y_1) + MOV r8, r9 + SMLATB r11, r6, r9, r11 ; sum1 = MAC16_16(sum1, x_1, y_2) +celt_pitch_xcorr_edsp_process2_1 + LDRH r6, [r4], #2 + ADDS r12, r12, #1 + ; Stall + SMLABB r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_0) + LDRGTH r7, [r4], #2 + SMLABT r11, r6, r8, r11 ; sum1 = MAC16_16(sum1, x_0, y_1) + BLE celt_pitch_xcorr_edsp_process2_done + LDRH r9, [r5], #2 + SMLABT r10, r7, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_1) + SMLABB r11, r7, r9, r11 ; sum1 = MAC16_16(sum1, x_0, y_2) +celt_pitch_xcorr_edsp_process2_done + ; Restore _x + SUB r4, r4, r3, LSL #1 + ; Restore and advance _y + SUB r5, r5, r3, LSL #1 + ; maxcorr = max(maxcorr, sum0) + CMP r0, r10 + ADD r5, r5, #2 + MOVLT r0, r10 + SUB r1, r1, #2 + ; maxcorr = max(maxcorr, sum1) + CMP r0, r11 + ; xcorr[i] = sum + STR r10, [r2], #4 + MOVLT r0, r11 + STR r11, [r2], #4 +celt_pitch_xcorr_edsp_process1a + ADDS r1, r1, #1 + BLT celt_pitch_xcorr_edsp_done + SUBS r12, r3, #4 + ; r14 = sum = 0 + MOV r14, #0 + BLT celt_pitch_xcorr_edsp_process1a_loop_done + LDR r6, [r4], #4 + LDR r8, [r5], #4 + LDR r7, [r4], #4 + LDR r9, [r5], #4 +celt_pitch_xcorr_edsp_process1a_loop4 + SMLABB r14, r6, r8, r14 ; sum = MAC16_16(sum, x_0, y_0) + SUBS r12, r12, #4 ; j-=4 + SMLATT r14, r6, r8, r14 ; sum = MAC16_16(sum, x_1, y_1) + LDRGE r6, [r4], #4 + SMLABB r14, r7, r9, r14 ; sum = MAC16_16(sum, x_2, y_2) + LDRGE r8, [r5], #4 + SMLATT r14, r7, r9, r14 ; sum = MAC16_16(sum, x_3, y_3) + LDRGE r7, [r4], #4 + LDRGE r9, [r5], #4 + BGE celt_pitch_xcorr_edsp_process1a_loop4 +celt_pitch_xcorr_edsp_process1a_loop_done + ADDS r12, r12, #2 + LDRGE r6, [r4], #4 + LDRGE r8, [r5], #4 + ; Stall + SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, x_0, y_0) + SUBGE r12, r12, #2 + SMLATTGE r14, r6, r8, r14 ; sum = MAC16_16(sum, x_1, y_1) + ADDS r12, r12, #1 + LDRGEH r6, [r4], #2 + LDRGEH r8, [r5], #2 + ; Stall + SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, *x, *y) + ; maxcorr = max(maxcorr, sum) + CMP r0, r14 + ; xcorr[i] = sum + STR r14, [r2], #4 + MOVLT r0, r14 +celt_pitch_xcorr_edsp_done + LDMFD sp!, {r4-r11, pc} + ENDP + +ENDIF + +END diff --git a/src/opus-1.1/celt/arm/fixed_armv4.h b/src/opus-1.1/celt/arm/fixed_armv4.h new file mode 100644 index 00000000..b690bc8c --- /dev/null +++ b/src/opus-1.1/celt/arm/fixed_armv4.h @@ -0,0 +1,76 @@ +/* Copyright (C) 2013 Xiph.Org Foundation and contributors */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_ARMv4_H +#define FIXED_ARMv4_H + +/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q16 +static OPUS_INLINE opus_val32 MULT16_32_Q16_armv4(opus_val16 a, opus_val32 b) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#MULT16_32_Q16\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(b),"r"(a<<16) + ); + return rd_hi; +} +#define MULT16_32_Q16(a, b) (MULT16_32_Q16_armv4(a, b)) + + +/** 16x32 multiplication, followed by a 15-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q15 +static OPUS_INLINE opus_val32 MULT16_32_Q15_armv4(opus_val16 a, opus_val32 b) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#MULT16_32_Q15\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(b), "r"(a<<16) + ); + /*We intentionally don't OR in the high bit of rd_lo for speed.*/ + return rd_hi<<1; +} +#define MULT16_32_Q15(a, b) (MULT16_32_Q15_armv4(a, b)) + + +/** 16x32 multiply, followed by a 15-bit shift right and 32-bit add. + b must fit in 31 bits. + Result fits in 32 bits. */ +#undef MAC16_32_Q15 +#define MAC16_32_Q15(c, a, b) ADD32(c, MULT16_32_Q15(a, b)) + + +/** 32x32 multiplication, followed by a 31-bit shift right. Results fits in 32 bits */ +#undef MULT32_32_Q31 +#define MULT32_32_Q31(a,b) (opus_val32)((((opus_int64)(a)) * ((opus_int64)(b)))>>31) + +#endif diff --git a/src/opus-1.1/celt/arm/fixed_armv5e.h b/src/opus-1.1/celt/arm/fixed_armv5e.h new file mode 100644 index 00000000..1194a7d3 --- /dev/null +++ b/src/opus-1.1/celt/arm/fixed_armv5e.h @@ -0,0 +1,116 @@ +/* Copyright (C) 2007-2009 Xiph.Org Foundation + Copyright (C) 2003-2008 Jean-Marc Valin + Copyright (C) 2007-2008 CSIRO + Copyright (C) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_ARMv5E_H +#define FIXED_ARMv5E_H + +#include "fixed_armv4.h" + +/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q16 +static OPUS_INLINE opus_val32 MULT16_32_Q16_armv5e(opus_val16 a, opus_val32 b) +{ + int res; + __asm__( + "#MULT16_32_Q16\n\t" + "smulwb %0, %1, %2\n\t" + : "=r"(res) + : "r"(b),"r"(a) + ); + return res; +} +#define MULT16_32_Q16(a, b) (MULT16_32_Q16_armv5e(a, b)) + + +/** 16x32 multiplication, followed by a 15-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q15 +static OPUS_INLINE opus_val32 MULT16_32_Q15_armv5e(opus_val16 a, opus_val32 b) +{ + int res; + __asm__( + "#MULT16_32_Q15\n\t" + "smulwb %0, %1, %2\n\t" + : "=r"(res) + : "r"(b), "r"(a) + ); + return res<<1; +} +#define MULT16_32_Q15(a, b) (MULT16_32_Q15_armv5e(a, b)) + + +/** 16x32 multiply, followed by a 15-bit shift right and 32-bit add. + b must fit in 31 bits. + Result fits in 32 bits. */ +#undef MAC16_32_Q15 +static OPUS_INLINE opus_val32 MAC16_32_Q15_armv5e(opus_val32 c, opus_val16 a, + opus_val32 b) +{ + int res; + __asm__( + "#MAC16_32_Q15\n\t" + "smlawb %0, %1, %2, %3;\n" + : "=r"(res) + : "r"(b<<1), "r"(a), "r"(c) + ); + return res; +} +#define MAC16_32_Q15(c, a, b) (MAC16_32_Q15_armv5e(c, a, b)) + +/** 16x16 multiply-add where the result fits in 32 bits */ +#undef MAC16_16 +static OPUS_INLINE opus_val32 MAC16_16_armv5e(opus_val32 c, opus_val16 a, + opus_val16 b) +{ + int res; + __asm__( + "#MAC16_16\n\t" + "smlabb %0, %1, %2, %3;\n" + : "=r"(res) + : "r"(a), "r"(b), "r"(c) + ); + return res; +} +#define MAC16_16(c, a, b) (MAC16_16_armv5e(c, a, b)) + +/** 16x16 multiplication where the result fits in 32 bits */ +#undef MULT16_16 +static OPUS_INLINE opus_val32 MULT16_16_armv5e(opus_val16 a, opus_val16 b) +{ + int res; + __asm__( + "#MULT16_16\n\t" + "smulbb %0, %1, %2;\n" + : "=r"(res) + : "r"(a), "r"(b) + ); + return res; +} +#define MULT16_16(a, b) (MULT16_16_armv5e(a, b)) + +#endif diff --git a/src/opus-1.1/celt/arm/kiss_fft_armv4.h b/src/opus-1.1/celt/arm/kiss_fft_armv4.h new file mode 100644 index 00000000..e4faad6f --- /dev/null +++ b/src/opus-1.1/celt/arm/kiss_fft_armv4.h @@ -0,0 +1,121 @@ +/*Copyright (c) 2013, Xiph.Org Foundation and contributors. + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_ARMv4_H +#define KISS_FFT_ARMv4_H + +#if !defined(KISS_FFT_GUTS_H) +#error "This file should only be included from _kiss_fft_guts.h" +#endif + +#ifdef FIXED_POINT + +#undef C_MUL +#define C_MUL(m,a,b) \ + do{ \ + int br__; \ + int bi__; \ + int tt__; \ + __asm__ __volatile__( \ + "#C_MUL\n\t" \ + "ldrsh %[br], [%[bp], #0]\n\t" \ + "ldm %[ap], {r0,r1}\n\t" \ + "ldrsh %[bi], [%[bp], #2]\n\t" \ + "smull %[tt], %[mi], r1, %[br]\n\t" \ + "smlal %[tt], %[mi], r0, %[bi]\n\t" \ + "rsb %[bi], %[bi], #0\n\t" \ + "smull %[br], %[mr], r0, %[br]\n\t" \ + "mov %[tt], %[tt], lsr #15\n\t" \ + "smlal %[br], %[mr], r1, %[bi]\n\t" \ + "orr %[mi], %[tt], %[mi], lsl #17\n\t" \ + "mov %[br], %[br], lsr #15\n\t" \ + "orr %[mr], %[br], %[mr], lsl #17\n\t" \ + : [mr]"=r"((m).r), [mi]"=r"((m).i), \ + [br]"=&r"(br__), [bi]"=r"(bi__), [tt]"=r"(tt__) \ + : [ap]"r"(&(a)), [bp]"r"(&(b)) \ + : "r0", "r1" \ + ); \ + } \ + while(0) + +#undef C_MUL4 +#define C_MUL4(m,a,b) \ + do{ \ + int br__; \ + int bi__; \ + int tt__; \ + __asm__ __volatile__( \ + "#C_MUL4\n\t" \ + "ldrsh %[br], [%[bp], #0]\n\t" \ + "ldm %[ap], {r0,r1}\n\t" \ + "ldrsh %[bi], [%[bp], #2]\n\t" \ + "smull %[tt], %[mi], r1, %[br]\n\t" \ + "smlal %[tt], %[mi], r0, %[bi]\n\t" \ + "rsb %[bi], %[bi], #0\n\t" \ + "smull %[br], %[mr], r0, %[br]\n\t" \ + "mov %[tt], %[tt], lsr #17\n\t" \ + "smlal %[br], %[mr], r1, %[bi]\n\t" \ + "orr %[mi], %[tt], %[mi], lsl #15\n\t" \ + "mov %[br], %[br], lsr #17\n\t" \ + "orr %[mr], %[br], %[mr], lsl #15\n\t" \ + : [mr]"=r"((m).r), [mi]"=r"((m).i), \ + [br]"=&r"(br__), [bi]"=r"(bi__), [tt]"=r"(tt__) \ + : [ap]"r"(&(a)), [bp]"r"(&(b)) \ + : "r0", "r1" \ + ); \ + } \ + while(0) + +#undef C_MULC +#define C_MULC(m,a,b) \ + do{ \ + int br__; \ + int bi__; \ + int tt__; \ + __asm__ __volatile__( \ + "#C_MULC\n\t" \ + "ldrsh %[br], [%[bp], #0]\n\t" \ + "ldm %[ap], {r0,r1}\n\t" \ + "ldrsh %[bi], [%[bp], #2]\n\t" \ + "smull %[tt], %[mr], r0, %[br]\n\t" \ + "smlal %[tt], %[mr], r1, %[bi]\n\t" \ + "rsb %[bi], %[bi], #0\n\t" \ + "smull %[br], %[mi], r1, %[br]\n\t" \ + "mov %[tt], %[tt], lsr #15\n\t" \ + "smlal %[br], %[mi], r0, %[bi]\n\t" \ + "orr %[mr], %[tt], %[mr], lsl #17\n\t" \ + "mov %[br], %[br], lsr #15\n\t" \ + "orr %[mi], %[br], %[mi], lsl #17\n\t" \ + : [mr]"=r"((m).r), [mi]"=r"((m).i), \ + [br]"=&r"(br__), [bi]"=r"(bi__), [tt]"=r"(tt__) \ + : [ap]"r"(&(a)), [bp]"r"(&(b)) \ + : "r0", "r1" \ + ); \ + } \ + while(0) + +#endif /* FIXED_POINT */ + +#endif /* KISS_FFT_ARMv4_H */ diff --git a/src/opus-1.1/celt/arm/kiss_fft_armv5e.h b/src/opus-1.1/celt/arm/kiss_fft_armv5e.h new file mode 100644 index 00000000..9eca183d --- /dev/null +++ b/src/opus-1.1/celt/arm/kiss_fft_armv5e.h @@ -0,0 +1,118 @@ +/*Copyright (c) 2013, Xiph.Org Foundation and contributors. + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_ARMv5E_H +#define KISS_FFT_ARMv5E_H + +#if !defined(KISS_FFT_GUTS_H) +#error "This file should only be included from _kiss_fft_guts.h" +#endif + +#ifdef FIXED_POINT + +#if defined(__thumb__)||defined(__thumb2__) +#define LDRD_CONS "Q" +#else +#define LDRD_CONS "Uq" +#endif + +#undef C_MUL +#define C_MUL(m,a,b) \ + do{ \ + int mr1__; \ + int mr2__; \ + int mi__; \ + long long aval__; \ + int bval__; \ + __asm__( \ + "#C_MUL\n\t" \ + "ldrd %[aval], %H[aval], %[ap]\n\t" \ + "ldr %[bval], %[bp]\n\t" \ + "smulwb %[mi], %H[aval], %[bval]\n\t" \ + "smulwb %[mr1], %[aval], %[bval]\n\t" \ + "smulwt %[mr2], %H[aval], %[bval]\n\t" \ + "smlawt %[mi], %[aval], %[bval], %[mi]\n\t" \ + : [mr1]"=r"(mr1__), [mr2]"=r"(mr2__), [mi]"=r"(mi__), \ + [aval]"=&r"(aval__), [bval]"=r"(bval__) \ + : [ap]LDRD_CONS(a), [bp]"m"(b) \ + ); \ + (m).r = SHL32(SUB32(mr1__, mr2__), 1); \ + (m).i = SHL32(mi__, 1); \ + } \ + while(0) + +#undef C_MUL4 +#define C_MUL4(m,a,b) \ + do{ \ + int mr1__; \ + int mr2__; \ + int mi__; \ + long long aval__; \ + int bval__; \ + __asm__( \ + "#C_MUL4\n\t" \ + "ldrd %[aval], %H[aval], %[ap]\n\t" \ + "ldr %[bval], %[bp]\n\t" \ + "smulwb %[mi], %H[aval], %[bval]\n\t" \ + "smulwb %[mr1], %[aval], %[bval]\n\t" \ + "smulwt %[mr2], %H[aval], %[bval]\n\t" \ + "smlawt %[mi], %[aval], %[bval], %[mi]\n\t" \ + : [mr1]"=r"(mr1__), [mr2]"=r"(mr2__), [mi]"=r"(mi__), \ + [aval]"=&r"(aval__), [bval]"=r"(bval__) \ + : [ap]LDRD_CONS(a), [bp]"m"(b) \ + ); \ + (m).r = SHR32(SUB32(mr1__, mr2__), 1); \ + (m).i = SHR32(mi__, 1); \ + } \ + while(0) + +#undef C_MULC +#define C_MULC(m,a,b) \ + do{ \ + int mr__; \ + int mi1__; \ + int mi2__; \ + long long aval__; \ + int bval__; \ + __asm__( \ + "#C_MULC\n\t" \ + "ldrd %[aval], %H[aval], %[ap]\n\t" \ + "ldr %[bval], %[bp]\n\t" \ + "smulwb %[mr], %[aval], %[bval]\n\t" \ + "smulwb %[mi1], %H[aval], %[bval]\n\t" \ + "smulwt %[mi2], %[aval], %[bval]\n\t" \ + "smlawt %[mr], %H[aval], %[bval], %[mr]\n\t" \ + : [mr]"=r"(mr__), [mi1]"=r"(mi1__), [mi2]"=r"(mi2__), \ + [aval]"=&r"(aval__), [bval]"=r"(bval__) \ + : [ap]LDRD_CONS(a), [bp]"m"(b) \ + ); \ + (m).r = SHL32(mr__, 1); \ + (m).i = SHL32(SUB32(mi1__, mi2__), 1); \ + } \ + while(0) + +#endif /* FIXED_POINT */ + +#endif /* KISS_FFT_GUTS_H */ diff --git a/src/opus-1.1/celt/arm/pitch_arm.h b/src/opus-1.1/celt/arm/pitch_arm.h new file mode 100644 index 00000000..a07f8ac2 --- /dev/null +++ b/src/opus-1.1/celt/arm/pitch_arm.h @@ -0,0 +1,57 @@ +/* Copyright (c) 2010 Xiph.Org Foundation + * Copyright (c) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(PITCH_ARM_H) +# define PITCH_ARM_H + +# include "armcpu.h" + +# if defined(FIXED_POINT) + +# if defined(OPUS_ARM_MAY_HAVE_NEON) +opus_val32 celt_pitch_xcorr_neon(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch); +# endif + +# if defined(OPUS_ARM_MAY_HAVE_MEDIA) +# define celt_pitch_xcorr_media MAY_HAVE_EDSP(celt_pitch_xcorr) +# endif + +# if defined(OPUS_ARM_MAY_HAVE_EDSP) +opus_val32 celt_pitch_xcorr_edsp(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch); +# endif + +# if !defined(OPUS_HAVE_RTCD) +# define OVERRIDE_PITCH_XCORR (1) +# define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \ + ((void)(arch),PRESUME_NEON(celt_pitch_xcorr)(_x, _y, xcorr, len, max_pitch)) +# endif + +# endif + +#endif diff --git a/src/opus-1.0.2/celt/bands.c b/src/opus-1.1/celt/bands.c index 3be543c3..cce56e2f 100644 --- a/src/opus-1.0.2/celt/bands.c +++ b/src/opus-1.1/celt/bands.c @@ -40,6 +40,23 @@ #include "os_support.h" #include "mathops.h" #include "rate.h" +#include "quant_bands.h" +#include "pitch.h" + +int hysteresis_decision(opus_val16 val, const opus_val16 *thresholds, const opus_val16 *hysteresis, int N, int prev) +{ + int i; + for (i=0;i<N;i++) + { + if (val < thresholds[i]) + break; + } + if (i>prev && val < thresholds[prev]+hysteresis[prev]) + i=prev; + if (i<prev && val > thresholds[prev-1]-hysteresis[prev-1]) + i=prev; + return i; +} opus_uint32 celt_lcg_rand(opus_uint32 seed) { @@ -172,7 +189,8 @@ void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, cel #endif /* FIXED_POINT */ /* De-normalise the energy to produce the synthesis from the unit-energy bands */ -void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X, celt_sig * OPUS_RESTRICT freq, const celt_ener *bandE, int end, int C, int M) +void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X, + celt_sig * OPUS_RESTRICT freq, const opus_val16 *bandLogE, int start, int end, int C, int M) { int i, c, N; const opus_int16 *eBands = m->eBands; @@ -182,18 +200,55 @@ void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X, cel celt_sig * OPUS_RESTRICT f; const celt_norm * OPUS_RESTRICT x; f = freq+c*N; - x = X+c*N; - for (i=0;i<end;i++) + x = X+c*N+M*eBands[start]; + for (i=0;i<M*eBands[start];i++) + *f++ = 0; + for (i=start;i<end;i++) { int j, band_end; - opus_val32 g = SHR32(bandE[i+c*m->nbEBands],1); + opus_val16 g; + opus_val16 lg; +#ifdef FIXED_POINT + int shift; +#endif j=M*eBands[i]; band_end = M*eBands[i+1]; + lg = ADD16(bandLogE[i+c*m->nbEBands], SHL16((opus_val16)eMeans[i],6)); +#ifndef FIXED_POINT + g = celt_exp2(lg); +#else + /* Handle the integer part of the log energy */ + shift = 16-(lg>>DB_SHIFT); + if (shift>31) + { + shift=0; + g=0; + } else { + /* Handle the fractional part. */ + g = celt_exp2_frac(lg&((1<<DB_SHIFT)-1)); + } + /* Handle extreme gains with negative shift. */ + if (shift<0) + { + /* For shift < -2 we'd be likely to overflow, so we're capping + the gain here. This shouldn't happen unless the bitstream is + already corrupted. */ + if (shift < -2) + { + g = 32767; + shift = -2; + } + do { + *f++ = SHL32(MULT16_16(*x++, g), -shift); + } while (++j<band_end); + } else +#endif + /* Be careful of the fixed-point "else" just above when changing this code */ do { - *f++ = SHL32(MULT16_32_Q15(*x, g),2); - x++; + *f++ = SHR32(MULT16_16(*x++, g), shift); } while (++j<band_end); } + celt_assert(start <= end); for (i=M*eBands[end];i<N;i++) *f++ = 0; } while (++c<C); @@ -345,11 +400,7 @@ static void stereo_merge(celt_norm *X, celt_norm *Y, opus_val16 mid, int N) opus_val32 t, lgain, rgain; /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */ - for (j=0;j<N;j++) - { - xp = MAC16_16(xp, X[j], Y[j]); - side = MAC16_16(side, Y[j], Y[j]); - } + dual_inner_prod(Y, X, Y, N, &xp, &side); /* Compensating for the mid normalization */ xp = MULT16_32_Q15(mid, xp); /* mid and side are in Q15, not Q14 like X and Y */ @@ -457,7 +508,7 @@ int spreading_decision(const CELTMode *m, celt_norm *X, int *average, *tapset_decision=0; } /*printf("%d %d %d\n", hf_sum, *hf_average, *tapset_decision);*/ - celt_assert(nbBands>0); /*M*(eBands[end]-eBands[end-1]) <= 8 assures this*/ + celt_assert(nbBands>0); /* end has to be non-zero */ sum /= nbBands; /* Recursive averaging */ sum = (sum+*average)>>1; @@ -483,50 +534,6 @@ int spreading_decision(const CELTMode *m, celt_norm *X, int *average, return decision; } -#ifdef MEASURE_NORM_MSE - -float MSE[30] = {0}; -int nbMSEBands = 0; -int MSECount[30] = {0}; - -void dump_norm_mse(void) -{ - int i; - for (i=0;i<nbMSEBands;i++) - { - printf ("%g ", MSE[i]/MSECount[i]); - } - printf ("\n"); -} - -void measure_norm_mse(const CELTMode *m, float *X, float *X0, float *bandE, float *bandE0, int M, int N, int C) -{ - static int init = 0; - int i; - if (!init) - { - atexit(dump_norm_mse); - init = 1; - } - for (i=0;i<m->nbEBands;i++) - { - int j; - int c; - float g; - if (bandE0[i]<10 || (C==2 && bandE0[i+m->nbEBands]<1)) - continue; - c=0; do { - g = bandE[i+c*m->nbEBands]/(1e-15+bandE0[i+c*m->nbEBands]); - for (j=M*m->eBands[i];j<M*m->eBands[i+1];j++) - MSE[i] += (g*X[j+c*N]-X0[j+c*N])*(g*X[j+c*N]-X0[j+c*N]); - } while (++c<C); - MSECount[i]+=C; - } - nbMSEBands = m->nbEBands; -} - -#endif - /* Indexing table for converting from natural Hadamard to ordery Hadamard This is essentially a bit-reversed Gray, on top of which we've added an inversion of the order because we want the DC at the end rather than @@ -629,289 +636,301 @@ static int compute_qn(int N, int b, int offset, int pulse_cap, int stereo) return qn; } -/* This function is responsible for encoding and decoding a band for both - the mono and stereo case. Even in the mono case, it can split the band - in two and transmit the energy difference with the two half-bands. It - can be called recursively so bands can end up being split in 8 parts. */ -static unsigned quant_band(int encode, const CELTMode *m, int i, celt_norm *X, celt_norm *Y, - int N, int b, int spread, int B, int intensity, int tf_change, celt_norm *lowband, ec_ctx *ec, - opus_int32 *remaining_bits, int LM, celt_norm *lowband_out, const celt_ener *bandE, int level, - opus_uint32 *seed, opus_val16 gain, celt_norm *lowband_scratch, int fill) +struct band_ctx { + int encode; + const CELTMode *m; + int i; + int intensity; + int spread; + int tf_change; + ec_ctx *ec; + opus_int32 remaining_bits; + const celt_ener *bandE; + opus_uint32 seed; +}; + +struct split_ctx { + int inv; + int imid; + int iside; + int delta; + int itheta; + int qalloc; +}; + +static void compute_theta(struct band_ctx *ctx, struct split_ctx *sctx, + celt_norm *X, celt_norm *Y, int N, int *b, int B, int B0, + int LM, + int stereo, int *fill) { - const unsigned char *cache; - int q; - int curr_bits; - int stereo, split; - int imid=0, iside=0; - int N0=N; - int N_B=N; - int N_B0; - int B0=B; - int time_divide=0; - int recombine=0; - int inv = 0; - opus_val16 mid=0, side=0; - int longBlocks; - unsigned cm=0; -#ifdef RESYNTH - int resynth = 1; -#else - int resynth = !encode; -#endif + int qn; + int itheta=0; + int delta; + int imid, iside; + int qalloc; + int pulse_cap; + int offset; + opus_int32 tell; + int inv=0; + int encode; + const CELTMode *m; + int i; + int intensity; + ec_ctx *ec; + const celt_ener *bandE; + + encode = ctx->encode; + m = ctx->m; + i = ctx->i; + intensity = ctx->intensity; + ec = ctx->ec; + bandE = ctx->bandE; + + /* Decide on the resolution to give to the split parameter theta */ + pulse_cap = m->logN[i]+LM*(1<<BITRES); + offset = (pulse_cap>>1) - (stereo&&N==2 ? QTHETA_OFFSET_TWOPHASE : QTHETA_OFFSET); + qn = compute_qn(N, *b, offset, pulse_cap, stereo); + if (stereo && i>=intensity) + qn = 1; + if (encode) + { + /* theta is the atan() of the ratio between the (normalized) + side and mid. With just that parameter, we can re-scale both + mid and side because we know that 1) they have unit norm and + 2) they are orthogonal. */ + itheta = stereo_itheta(X, Y, stereo, N); + } + tell = ec_tell_frac(ec); + if (qn!=1) + { + if (encode) + itheta = (itheta*qn+8192)>>14; - longBlocks = B0==1; + /* Entropy coding of the angle. We use a uniform pdf for the + time split, a step for stereo, and a triangular one for the rest. */ + if (stereo && N>2) + { + int p0 = 3; + int x = itheta; + int x0 = qn/2; + int ft = p0*(x0+1) + x0; + /* Use a probability of p0 up to itheta=8192 and then use 1 after */ + if (encode) + { + ec_encode(ec,x<=x0?p0*x:(x-1-x0)+(x0+1)*p0,x<=x0?p0*(x+1):(x-x0)+(x0+1)*p0,ft); + } else { + int fs; + fs=ec_decode(ec,ft); + if (fs<(x0+1)*p0) + x=fs/p0; + else + x=x0+1+(fs-(x0+1)*p0); + ec_dec_update(ec,x<=x0?p0*x:(x-1-x0)+(x0+1)*p0,x<=x0?p0*(x+1):(x-x0)+(x0+1)*p0,ft); + itheta = x; + } + } else if (B0>1 || stereo) { + /* Uniform pdf */ + if (encode) + ec_enc_uint(ec, itheta, qn+1); + else + itheta = ec_dec_uint(ec, qn+1); + } else { + int fs=1, ft; + ft = ((qn>>1)+1)*((qn>>1)+1); + if (encode) + { + int fl; - N_B /= B; - N_B0 = N_B; + fs = itheta <= (qn>>1) ? itheta + 1 : qn + 1 - itheta; + fl = itheta <= (qn>>1) ? itheta*(itheta + 1)>>1 : + ft - ((qn + 1 - itheta)*(qn + 2 - itheta)>>1); - split = stereo = Y != NULL; + ec_encode(ec, fl, fl+fs, ft); + } else { + /* Triangular pdf */ + int fl=0; + int fm; + fm = ec_decode(ec, ft); - /* Special case for one sample */ - if (N==1) - { - int c; - celt_norm *x = X; - c=0; do { - int sign=0; - if (*remaining_bits>=1<<BITRES) - { - if (encode) + if (fm < ((qn>>1)*((qn>>1) + 1)>>1)) { - sign = x[0]<0; - ec_enc_bits(ec, sign, 1); - } else { - sign = ec_dec_bits(ec, 1); + itheta = (isqrt32(8*(opus_uint32)fm + 1) - 1)>>1; + fs = itheta + 1; + fl = itheta*(itheta + 1)>>1; + } + else + { + itheta = (2*(qn + 1) + - isqrt32(8*(opus_uint32)(ft - fm - 1) + 1))>>1; + fs = qn + 1 - itheta; + fl = ft - ((qn + 1 - itheta)*(qn + 2 - itheta)>>1); } - *remaining_bits -= 1<<BITRES; - b-=1<<BITRES; - } - if (resynth) - x[0] = sign ? -NORM_SCALING : NORM_SCALING; - x = Y; - } while (++c<1+stereo); - if (lowband_out) - lowband_out[0] = SHR16(X[0],4); - return 1; - } - - if (!stereo && level == 0) - { - int k; - if (tf_change>0) - recombine = tf_change; - /* Band recombining to increase frequency resolution */ - if (lowband && (recombine || ((N_B&1) == 0 && tf_change<0) || B0>1)) - { - int j; - for (j=0;j<N;j++) - lowband_scratch[j] = lowband[j]; - lowband = lowband_scratch; + ec_dec_update(ec, fl, fl+fs, ft); + } } - - for (k=0;k<recombine;k++) + itheta = (opus_int32)itheta*16384/qn; + if (encode && stereo) { - static const unsigned char bit_interleave_table[16]={ - 0,1,1,1,2,3,3,3,2,3,3,3,2,3,3,3 - }; - if (encode) - haar1(X, N>>k, 1<<k); - if (lowband) - haar1(lowband, N>>k, 1<<k); - fill = bit_interleave_table[fill&0xF]|bit_interleave_table[fill>>4]<<2; + if (itheta==0) + intensity_stereo(m, X, Y, bandE, i, N); + else + stereo_split(X, Y, N); } - B>>=recombine; - N_B<<=recombine; - - /* Increasing the time resolution */ - while ((N_B&1) == 0 && tf_change<0) + /* NOTE: Renormalising X and Y *may* help fixed-point a bit at very high rate. + Let's do that at higher complexity */ + } else if (stereo) { + if (encode) { - if (encode) - haar1(X, N_B, B); - if (lowband) - haar1(lowband, N_B, B); - fill |= fill<<B; - B <<= 1; - N_B >>= 1; - time_divide++; - tf_change++; + inv = itheta > 8192; + if (inv) + { + int j; + for (j=0;j<N;j++) + Y[j] = -Y[j]; + } + intensity_stereo(m, X, Y, bandE, i, N); } - B0=B; - N_B0 = N_B; - - /* Reorganize the samples in time order instead of frequency order */ - if (B0>1) + if (*b>2<<BITRES && ctx->remaining_bits > 2<<BITRES) { if (encode) - deinterleave_hadamard(X, N_B>>recombine, B0<<recombine, longBlocks); - if (lowband) - deinterleave_hadamard(lowband, N_B>>recombine, B0<<recombine, longBlocks); - } + ec_enc_bit_logp(ec, inv, 2); + else + inv = ec_dec_bit_logp(ec, 2); + } else + inv = 0; + itheta = 0; } + qalloc = ec_tell_frac(ec) - tell; + *b -= qalloc; - /* If we need 1.5 more bit than we can produce, split the band in two. */ - cache = m->cache.bits + m->cache.index[(LM+1)*m->nbEBands+i]; - if (!stereo && LM != -1 && b > cache[cache[0]]+12 && N>2) + if (itheta == 0) { - N >>= 1; - Y = X+N; - split = 1; - LM -= 1; - if (B==1) - fill = (fill&1)|(fill<<1); - B = (B+1)>>1; + imid = 32767; + iside = 0; + *fill &= (1<<B)-1; + delta = -16384; + } else if (itheta == 16384) + { + imid = 0; + iside = 32767; + *fill &= ((1<<B)-1)<<B; + delta = 16384; + } else { + imid = bitexact_cos((opus_int16)itheta); + iside = bitexact_cos((opus_int16)(16384-itheta)); + /* This is the mid vs side allocation that minimizes squared error + in that band. */ + delta = FRAC_MUL16((N-1)<<7,bitexact_log2tan(iside,imid)); } - if (split) - { - int qn; - int itheta=0; - int mbits, sbits, delta; - int qalloc; - int pulse_cap; - int offset; - int orig_fill; - opus_int32 tell; + sctx->inv = inv; + sctx->imid = imid; + sctx->iside = iside; + sctx->delta = delta; + sctx->itheta = itheta; + sctx->qalloc = qalloc; +} +static unsigned quant_band_n1(struct band_ctx *ctx, celt_norm *X, celt_norm *Y, int b, + celt_norm *lowband_out) +{ +#ifdef RESYNTH + int resynth = 1; +#else + int resynth = !ctx->encode; +#endif + int c; + int stereo; + celt_norm *x = X; + int encode; + ec_ctx *ec; - /* Decide on the resolution to give to the split parameter theta */ - pulse_cap = m->logN[i]+LM*(1<<BITRES); - offset = (pulse_cap>>1) - (stereo&&N==2 ? QTHETA_OFFSET_TWOPHASE : QTHETA_OFFSET); - qn = compute_qn(N, b, offset, pulse_cap, stereo); - if (stereo && i>=intensity) - qn = 1; - if (encode) - { - /* theta is the atan() of the ratio between the (normalized) - side and mid. With just that parameter, we can re-scale both - mid and side because we know that 1) they have unit norm and - 2) they are orthogonal. */ - itheta = stereo_itheta(X, Y, stereo, N); - } - tell = ec_tell_frac(ec); - if (qn!=1) + encode = ctx->encode; + ec = ctx->ec; + + stereo = Y != NULL; + c=0; do { + int sign=0; + if (ctx->remaining_bits>=1<<BITRES) { if (encode) - itheta = (itheta*qn+8192)>>14; - - /* Entropy coding of the angle. We use a uniform pdf for the - time split, a step for stereo, and a triangular one for the rest. */ - if (stereo && N>2) { - int p0 = 3; - int x = itheta; - int x0 = qn/2; - int ft = p0*(x0+1) + x0; - /* Use a probability of p0 up to itheta=8192 and then use 1 after */ - if (encode) - { - ec_encode(ec,x<=x0?p0*x:(x-1-x0)+(x0+1)*p0,x<=x0?p0*(x+1):(x-x0)+(x0+1)*p0,ft); - } else { - int fs; - fs=ec_decode(ec,ft); - if (fs<(x0+1)*p0) - x=fs/p0; - else - x=x0+1+(fs-(x0+1)*p0); - ec_dec_update(ec,x<=x0?p0*x:(x-1-x0)+(x0+1)*p0,x<=x0?p0*(x+1):(x-x0)+(x0+1)*p0,ft); - itheta = x; - } - } else if (B0>1 || stereo) { - /* Uniform pdf */ - if (encode) - ec_enc_uint(ec, itheta, qn+1); - else - itheta = ec_dec_uint(ec, qn+1); + sign = x[0]<0; + ec_enc_bits(ec, sign, 1); } else { - int fs=1, ft; - ft = ((qn>>1)+1)*((qn>>1)+1); - if (encode) - { - int fl; - - fs = itheta <= (qn>>1) ? itheta + 1 : qn + 1 - itheta; - fl = itheta <= (qn>>1) ? itheta*(itheta + 1)>>1 : - ft - ((qn + 1 - itheta)*(qn + 2 - itheta)>>1); + sign = ec_dec_bits(ec, 1); + } + ctx->remaining_bits -= 1<<BITRES; + b-=1<<BITRES; + } + if (resynth) + x[0] = sign ? -NORM_SCALING : NORM_SCALING; + x = Y; + } while (++c<1+stereo); + if (lowband_out) + lowband_out[0] = SHR16(X[0],4); + return 1; +} - ec_encode(ec, fl, fl+fs, ft); - } else { - /* Triangular pdf */ - int fl=0; - int fm; - fm = ec_decode(ec, ft); +/* This function is responsible for encoding and decoding a mono partition. + It can split the band in two and transmit the energy difference with + the two half-bands. It can be called recursively so bands can end up being + split in 8 parts. */ +static unsigned quant_partition(struct band_ctx *ctx, celt_norm *X, + int N, int b, int B, celt_norm *lowband, + int LM, + opus_val16 gain, int fill) +{ + const unsigned char *cache; + int q; + int curr_bits; + int imid=0, iside=0; + int B0=B; + opus_val16 mid=0, side=0; + unsigned cm=0; +#ifdef RESYNTH + int resynth = 1; +#else + int resynth = !ctx->encode; +#endif + celt_norm *Y=NULL; + int encode; + const CELTMode *m; + int i; + int spread; + ec_ctx *ec; - if (fm < ((qn>>1)*((qn>>1) + 1)>>1)) - { - itheta = (isqrt32(8*(opus_uint32)fm + 1) - 1)>>1; - fs = itheta + 1; - fl = itheta*(itheta + 1)>>1; - } - else - { - itheta = (2*(qn + 1) - - isqrt32(8*(opus_uint32)(ft - fm - 1) + 1))>>1; - fs = qn + 1 - itheta; - fl = ft - ((qn + 1 - itheta)*(qn + 2 - itheta)>>1); - } + encode = ctx->encode; + m = ctx->m; + i = ctx->i; + spread = ctx->spread; + ec = ctx->ec; - ec_dec_update(ec, fl, fl+fs, ft); - } - } - itheta = (opus_int32)itheta*16384/qn; - if (encode && stereo) - { - if (itheta==0) - intensity_stereo(m, X, Y, bandE, i, N); - else - stereo_split(X, Y, N); - } - /* NOTE: Renormalising X and Y *may* help fixed-point a bit at very high rate. - Let's do that at higher complexity */ - } else if (stereo) { - if (encode) - { - inv = itheta > 8192; - if (inv) - { - int j; - for (j=0;j<N;j++) - Y[j] = -Y[j]; - } - intensity_stereo(m, X, Y, bandE, i, N); - } - if (b>2<<BITRES && *remaining_bits > 2<<BITRES) - { - if (encode) - ec_enc_bit_logp(ec, inv, 2); - else - inv = ec_dec_bit_logp(ec, 2); - } else - inv = 0; - itheta = 0; - } - qalloc = ec_tell_frac(ec) - tell; - b -= qalloc; + /* If we need 1.5 more bit than we can produce, split the band in two. */ + cache = m->cache.bits + m->cache.index[(LM+1)*m->nbEBands+i]; + if (LM != -1 && b > cache[cache[0]]+12 && N>2) + { + int mbits, sbits, delta; + int itheta; + int qalloc; + struct split_ctx sctx; + celt_norm *next_lowband2=NULL; + opus_int32 rebalance; - orig_fill = fill; - if (itheta == 0) - { - imid = 32767; - iside = 0; - fill &= (1<<B)-1; - delta = -16384; - } else if (itheta == 16384) - { - imid = 0; - iside = 32767; - fill &= ((1<<B)-1)<<B; - delta = 16384; - } else { - imid = bitexact_cos((opus_int16)itheta); - iside = bitexact_cos((opus_int16)(16384-itheta)); - /* This is the mid vs side allocation that minimizes squared error - in that band. */ - delta = FRAC_MUL16((N-1)<<7,bitexact_log2tan(iside,imid)); - } + N >>= 1; + Y = X+N; + LM -= 1; + if (B==1) + fill = (fill&1)|(fill<<1); + B = (B+1)>>1; + compute_theta(ctx, &sctx, X, Y, N, &b, B, B0, + LM, 0, &fill); + imid = sctx.imid; + iside = sctx.iside; + delta = sctx.delta; + itheta = sctx.itheta; + qalloc = sctx.qalloc; #ifdef FIXED_POINT mid = imid; side = iside; @@ -920,136 +939,59 @@ static unsigned quant_band(int encode, const CELTMode *m, int i, celt_norm *X, c side = (1.f/32768)*iside; #endif - /* This is a special case for N=2 that only works for stereo and takes - advantage of the fact that mid and side are orthogonal to encode - the side with just one bit. */ - if (N==2 && stereo) + /* Give more bits to low-energy MDCTs than they would otherwise deserve */ + if (B0>1 && (itheta&0x3fff)) { - int c; - int sign=0; - celt_norm *x2, *y2; - mbits = b; - sbits = 0; - /* Only need one bit for the side */ - if (itheta != 0 && itheta != 16384) - sbits = 1<<BITRES; - mbits -= sbits; - c = itheta > 8192; - *remaining_bits -= qalloc+sbits; - - x2 = c ? Y : X; - y2 = c ? X : Y; - if (sbits) - { - if (encode) - { - /* Here we only need to encode a sign for the side */ - sign = x2[0]*y2[1] - x2[1]*y2[0] < 0; - ec_enc_bits(ec, sign, 1); - } else { - sign = ec_dec_bits(ec, 1); - } - } - sign = 1-2*sign; - /* We use orig_fill here because we want to fold the side, but if - itheta==16384, we'll have cleared the low bits of fill. */ - cm = quant_band(encode, m, i, x2, NULL, N, mbits, spread, B, intensity, tf_change, lowband, ec, remaining_bits, LM, lowband_out, NULL, level, seed, gain, lowband_scratch, orig_fill); - /* We don't split N=2 bands, so cm is either 1 or 0 (for a fold-collapse), - and there's no need to worry about mixing with the other channel. */ - y2[0] = -sign*x2[1]; - y2[1] = sign*x2[0]; - if (resynth) - { - celt_norm tmp; - X[0] = MULT16_16_Q15(mid, X[0]); - X[1] = MULT16_16_Q15(mid, X[1]); - Y[0] = MULT16_16_Q15(side, Y[0]); - Y[1] = MULT16_16_Q15(side, Y[1]); - tmp = X[0]; - X[0] = SUB16(tmp,Y[0]); - Y[0] = ADD16(tmp,Y[0]); - tmp = X[1]; - X[1] = SUB16(tmp,Y[1]); - Y[1] = ADD16(tmp,Y[1]); - } - } else { - /* "Normal" split code */ - celt_norm *next_lowband2=NULL; - celt_norm *next_lowband_out1=NULL; - int next_level=0; - opus_int32 rebalance; - - /* Give more bits to low-energy MDCTs than they would otherwise deserve */ - if (B0>1 && !stereo && (itheta&0x3fff)) - { - if (itheta > 8192) - /* Rough approximation for pre-echo masking */ - delta -= delta>>(4-LM); - else - /* Corresponds to a forward-masking slope of 1.5 dB per 10 ms */ - delta = IMIN(0, delta + (N<<BITRES>>(5-LM))); - } - mbits = IMAX(0, IMIN(b, (b-delta)/2)); - sbits = b-mbits; - *remaining_bits -= qalloc; - - if (lowband && !stereo) - next_lowband2 = lowband+N; /* >32-bit split case */ - - /* Only stereo needs to pass on lowband_out. Otherwise, it's - handled at the end */ - if (stereo) - next_lowband_out1 = lowband_out; + if (itheta > 8192) + /* Rough approximation for pre-echo masking */ + delta -= delta>>(4-LM); else - next_level = level+1; - - rebalance = *remaining_bits; - if (mbits >= sbits) - { - /* In stereo mode, we do not apply a scaling to the mid because we need the normalized - mid for folding later */ - cm = quant_band(encode, m, i, X, NULL, N, mbits, spread, B, intensity, tf_change, - lowband, ec, remaining_bits, LM, next_lowband_out1, - NULL, next_level, seed, stereo ? Q15ONE : MULT16_16_P15(gain,mid), lowband_scratch, fill); - rebalance = mbits - (rebalance-*remaining_bits); - if (rebalance > 3<<BITRES && itheta!=0) - sbits += rebalance - (3<<BITRES); - - /* For a stereo split, the high bits of fill are always zero, so no - folding will be done to the side. */ - cm |= quant_band(encode, m, i, Y, NULL, N, sbits, spread, B, intensity, tf_change, - next_lowband2, ec, remaining_bits, LM, NULL, - NULL, next_level, seed, MULT16_16_P15(gain,side), NULL, fill>>B)<<((B0>>1)&(stereo-1)); - } else { - /* For a stereo split, the high bits of fill are always zero, so no - folding will be done to the side. */ - cm = quant_band(encode, m, i, Y, NULL, N, sbits, spread, B, intensity, tf_change, - next_lowband2, ec, remaining_bits, LM, NULL, - NULL, next_level, seed, MULT16_16_P15(gain,side), NULL, fill>>B)<<((B0>>1)&(stereo-1)); - rebalance = sbits - (rebalance-*remaining_bits); - if (rebalance > 3<<BITRES && itheta!=16384) - mbits += rebalance - (3<<BITRES); - /* In stereo mode, we do not apply a scaling to the mid because we need the normalized - mid for folding later */ - cm |= quant_band(encode, m, i, X, NULL, N, mbits, spread, B, intensity, tf_change, - lowband, ec, remaining_bits, LM, next_lowband_out1, - NULL, next_level, seed, stereo ? Q15ONE : MULT16_16_P15(gain,mid), lowband_scratch, fill); - } + /* Corresponds to a forward-masking slope of 1.5 dB per 10 ms */ + delta = IMIN(0, delta + (N<<BITRES>>(5-LM))); } + mbits = IMAX(0, IMIN(b, (b-delta)/2)); + sbits = b-mbits; + ctx->remaining_bits -= qalloc; + + if (lowband) + next_lowband2 = lowband+N; /* >32-bit split case */ + rebalance = ctx->remaining_bits; + if (mbits >= sbits) + { + cm = quant_partition(ctx, X, N, mbits, B, + lowband, LM, + MULT16_16_P15(gain,mid), fill); + rebalance = mbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=0) + sbits += rebalance - (3<<BITRES); + cm |= quant_partition(ctx, Y, N, sbits, B, + next_lowband2, LM, + MULT16_16_P15(gain,side), fill>>B)<<(B0>>1); + } else { + cm = quant_partition(ctx, Y, N, sbits, B, + next_lowband2, LM, + MULT16_16_P15(gain,side), fill>>B)<<(B0>>1); + rebalance = sbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=16384) + mbits += rebalance - (3<<BITRES); + cm |= quant_partition(ctx, X, N, mbits, B, + lowband, LM, + MULT16_16_P15(gain,mid), fill); + } } else { /* This is the basic no-split case */ q = bits2pulses(m, i, LM, b); curr_bits = pulses2bits(m, i, LM, q); - *remaining_bits -= curr_bits; + ctx->remaining_bits -= curr_bits; /* Ensures we can never bust the budget */ - while (*remaining_bits < 0 && q > 0) + while (ctx->remaining_bits < 0 && q > 0) { - *remaining_bits += curr_bits; + ctx->remaining_bits += curr_bits; q--; curr_bits = pulses2bits(m, i, LM, q); - *remaining_bits -= curr_bits; + ctx->remaining_bits -= curr_bits; } if (q!=0) @@ -1073,7 +1015,7 @@ static unsigned quant_band(int encode, const CELTMode *m, int i, celt_norm *X, c if (resynth) { unsigned cm_mask; - /*B can be as large as 16, so this shift might overflow an int on a + /* B can be as large as 16, so this shift might overflow an int on a 16-bit platform; use a long to get defined behavior.*/ cm_mask = (unsigned)(1UL<<B)-1; fill &= cm_mask; @@ -1087,8 +1029,8 @@ static unsigned quant_band(int encode, const CELTMode *m, int i, celt_norm *X, c /* Noise */ for (j=0;j<N;j++) { - *seed = celt_lcg_rand(*seed); - X[j] = (celt_norm)((opus_int32)*seed>>20); + ctx->seed = celt_lcg_rand(ctx->seed); + X[j] = (celt_norm)((opus_int32)ctx->seed>>20); } cm = cm_mask; } else { @@ -1096,10 +1038,10 @@ static unsigned quant_band(int encode, const CELTMode *m, int i, celt_norm *X, c for (j=0;j<N;j++) { opus_val16 tmp; - *seed = celt_lcg_rand(*seed); + ctx->seed = celt_lcg_rand(ctx->seed); /* About 48 dB below the "normal" folding level */ tmp = QCONST16(1.0f/256, 10); - tmp = (*seed)&0x8000 ? tmp : -tmp; + tmp = (ctx->seed)&0x8000 ? tmp : -tmp; X[j] = lowband[j]+tmp; } cm = fill; @@ -1110,64 +1052,306 @@ static unsigned quant_band(int encode, const CELTMode *m, int i, celt_norm *X, c } } + return cm; +} + + +/* This function is responsible for encoding and decoding a band for the mono case. */ +static unsigned quant_band(struct band_ctx *ctx, celt_norm *X, + int N, int b, int B, celt_norm *lowband, + int LM, celt_norm *lowband_out, + opus_val16 gain, celt_norm *lowband_scratch, int fill) +{ + int N0=N; + int N_B=N; + int N_B0; + int B0=B; + int time_divide=0; + int recombine=0; + int longBlocks; + unsigned cm=0; +#ifdef RESYNTH + int resynth = 1; +#else + int resynth = !ctx->encode; +#endif + int k; + int encode; + int tf_change; + + encode = ctx->encode; + tf_change = ctx->tf_change; + + longBlocks = B0==1; + + N_B /= B; + + /* Special case for one sample */ + if (N==1) + { + return quant_band_n1(ctx, X, NULL, b, lowband_out); + } + + if (tf_change>0) + recombine = tf_change; + /* Band recombining to increase frequency resolution */ + + if (lowband_scratch && lowband && (recombine || ((N_B&1) == 0 && tf_change<0) || B0>1)) + { + int j; + for (j=0;j<N;j++) + lowband_scratch[j] = lowband[j]; + lowband = lowband_scratch; + } + + for (k=0;k<recombine;k++) + { + static const unsigned char bit_interleave_table[16]={ + 0,1,1,1,2,3,3,3,2,3,3,3,2,3,3,3 + }; + if (encode) + haar1(X, N>>k, 1<<k); + if (lowband) + haar1(lowband, N>>k, 1<<k); + fill = bit_interleave_table[fill&0xF]|bit_interleave_table[fill>>4]<<2; + } + B>>=recombine; + N_B<<=recombine; + + /* Increasing the time resolution */ + while ((N_B&1) == 0 && tf_change<0) + { + if (encode) + haar1(X, N_B, B); + if (lowband) + haar1(lowband, N_B, B); + fill |= fill<<B; + B <<= 1; + N_B >>= 1; + time_divide++; + tf_change++; + } + B0=B; + N_B0 = N_B; + + /* Reorganize the samples in time order instead of frequency order */ + if (B0>1) + { + if (encode) + deinterleave_hadamard(X, N_B>>recombine, B0<<recombine, longBlocks); + if (lowband) + deinterleave_hadamard(lowband, N_B>>recombine, B0<<recombine, longBlocks); + } + + cm = quant_partition(ctx, X, N, b, B, lowband, + LM, gain, fill); + /* This code is used by the decoder and by the resynthesis-enabled encoder */ if (resynth) { - if (stereo) + /* Undo the sample reorganization going from time order to frequency order */ + if (B0>1) + interleave_hadamard(X, N_B>>recombine, B0<<recombine, longBlocks); + + /* Undo time-freq changes that we did earlier */ + N_B = N_B0; + B = B0; + for (k=0;k<time_divide;k++) { - if (N!=2) - stereo_merge(X, Y, mid, N); - if (inv) - { - int j; - for (j=0;j<N;j++) - Y[j] = -Y[j]; - } - } else if (level == 0) + B >>= 1; + N_B <<= 1; + cm |= cm>>B; + haar1(X, N_B, B); + } + + for (k=0;k<recombine;k++) { - int k; + static const unsigned char bit_deinterleave_table[16]={ + 0x00,0x03,0x0C,0x0F,0x30,0x33,0x3C,0x3F, + 0xC0,0xC3,0xCC,0xCF,0xF0,0xF3,0xFC,0xFF + }; + cm = bit_deinterleave_table[cm]; + haar1(X, N0>>k, 1<<k); + } + B<<=recombine; - /* Undo the sample reorganization going from time order to frequency order */ - if (B0>1) - interleave_hadamard(X, N_B>>recombine, B0<<recombine, longBlocks); + /* Scale output for later folding */ + if (lowband_out) + { + int j; + opus_val16 n; + n = celt_sqrt(SHL32(EXTEND32(N0),22)); + for (j=0;j<N0;j++) + lowband_out[j] = MULT16_16_Q15(n,X[j]); + } + cm &= (1<<B)-1; + } + return cm; +} - /* Undo time-freq changes that we did earlier */ - N_B = N_B0; - B = B0; - for (k=0;k<time_divide;k++) - { - B >>= 1; - N_B <<= 1; - cm |= cm>>B; - haar1(X, N_B, B); - } - for (k=0;k<recombine;k++) - { - static const unsigned char bit_deinterleave_table[16]={ - 0x00,0x03,0x0C,0x0F,0x30,0x33,0x3C,0x3F, - 0xC0,0xC3,0xCC,0xCF,0xF0,0xF3,0xFC,0xFF - }; - cm = bit_deinterleave_table[cm]; - haar1(X, N0>>k, 1<<k); - } - B<<=recombine; +/* This function is responsible for encoding and decoding a band for the stereo case. */ +static unsigned quant_band_stereo(struct band_ctx *ctx, celt_norm *X, celt_norm *Y, + int N, int b, int B, celt_norm *lowband, + int LM, celt_norm *lowband_out, + celt_norm *lowband_scratch, int fill) +{ + int imid=0, iside=0; + int inv = 0; + opus_val16 mid=0, side=0; + unsigned cm=0; +#ifdef RESYNTH + int resynth = 1; +#else + int resynth = !ctx->encode; +#endif + int mbits, sbits, delta; + int itheta; + int qalloc; + struct split_ctx sctx; + int orig_fill; + int encode; + ec_ctx *ec; + + encode = ctx->encode; + ec = ctx->ec; + + /* Special case for one sample */ + if (N==1) + { + return quant_band_n1(ctx, X, Y, b, lowband_out); + } + + orig_fill = fill; + + compute_theta(ctx, &sctx, X, Y, N, &b, B, B, + LM, 1, &fill); + inv = sctx.inv; + imid = sctx.imid; + iside = sctx.iside; + delta = sctx.delta; + itheta = sctx.itheta; + qalloc = sctx.qalloc; +#ifdef FIXED_POINT + mid = imid; + side = iside; +#else + mid = (1.f/32768)*imid; + side = (1.f/32768)*iside; +#endif - /* Scale output for later folding */ - if (lowband_out) + /* This is a special case for N=2 that only works for stereo and takes + advantage of the fact that mid and side are orthogonal to encode + the side with just one bit. */ + if (N==2) + { + int c; + int sign=0; + celt_norm *x2, *y2; + mbits = b; + sbits = 0; + /* Only need one bit for the side. */ + if (itheta != 0 && itheta != 16384) + sbits = 1<<BITRES; + mbits -= sbits; + c = itheta > 8192; + ctx->remaining_bits -= qalloc+sbits; + + x2 = c ? Y : X; + y2 = c ? X : Y; + if (sbits) + { + if (encode) { - int j; - opus_val16 n; - n = celt_sqrt(SHL32(EXTEND32(N0),22)); - for (j=0;j<N0;j++) - lowband_out[j] = MULT16_16_Q15(n,X[j]); + /* Here we only need to encode a sign for the side. */ + sign = x2[0]*y2[1] - x2[1]*y2[0] < 0; + ec_enc_bits(ec, sign, 1); + } else { + sign = ec_dec_bits(ec, 1); } - cm &= (1<<B)-1; + } + sign = 1-2*sign; + /* We use orig_fill here because we want to fold the side, but if + itheta==16384, we'll have cleared the low bits of fill. */ + cm = quant_band(ctx, x2, N, mbits, B, lowband, + LM, lowband_out, Q15ONE, lowband_scratch, orig_fill); + /* We don't split N=2 bands, so cm is either 1 or 0 (for a fold-collapse), + and there's no need to worry about mixing with the other channel. */ + y2[0] = -sign*x2[1]; + y2[1] = sign*x2[0]; + if (resynth) + { + celt_norm tmp; + X[0] = MULT16_16_Q15(mid, X[0]); + X[1] = MULT16_16_Q15(mid, X[1]); + Y[0] = MULT16_16_Q15(side, Y[0]); + Y[1] = MULT16_16_Q15(side, Y[1]); + tmp = X[0]; + X[0] = SUB16(tmp,Y[0]); + Y[0] = ADD16(tmp,Y[0]); + tmp = X[1]; + X[1] = SUB16(tmp,Y[1]); + Y[1] = ADD16(tmp,Y[1]); + } + } else { + /* "Normal" split code */ + opus_int32 rebalance; + + mbits = IMAX(0, IMIN(b, (b-delta)/2)); + sbits = b-mbits; + ctx->remaining_bits -= qalloc; + + rebalance = ctx->remaining_bits; + if (mbits >= sbits) + { + /* In stereo mode, we do not apply a scaling to the mid because we need the normalized + mid for folding later. */ + cm = quant_band(ctx, X, N, mbits, B, + lowband, LM, lowband_out, + Q15ONE, lowband_scratch, fill); + rebalance = mbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=0) + sbits += rebalance - (3<<BITRES); + + /* For a stereo split, the high bits of fill are always zero, so no + folding will be done to the side. */ + cm |= quant_band(ctx, Y, N, sbits, B, + NULL, LM, NULL, + side, NULL, fill>>B); + } else { + /* For a stereo split, the high bits of fill are always zero, so no + folding will be done to the side. */ + cm = quant_band(ctx, Y, N, sbits, B, + NULL, LM, NULL, + side, NULL, fill>>B); + rebalance = sbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=16384) + mbits += rebalance - (3<<BITRES); + /* In stereo mode, we do not apply a scaling to the mid because we need the normalized + mid for folding later. */ + cm |= quant_band(ctx, X, N, mbits, B, + lowband, LM, lowband_out, + Q15ONE, lowband_scratch, fill); + } + } + + + /* This code is used by the decoder and by the resynthesis-enabled encoder */ + if (resynth) + { + if (N!=2) + stereo_merge(X, Y, mid, N); + if (inv) + { + int j; + for (j=0;j<N;j++) + Y[j] = -Y[j]; } } return cm; } + void quant_all_bands(int encode, const CELTMode *m, int start, int end, celt_norm *X_, celt_norm *Y_, unsigned char *collapse_masks, const celt_ener *bandE, int *pulses, int shortBlocks, int spread, int dual_stereo, int intensity, int *tf_res, @@ -1178,27 +1362,41 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end, const opus_int16 * OPUS_RESTRICT eBands = m->eBands; celt_norm * OPUS_RESTRICT norm, * OPUS_RESTRICT norm2; VARDECL(celt_norm, _norm); - VARDECL(celt_norm, lowband_scratch); + celt_norm *lowband_scratch; int B; int M; int lowband_offset; int update_lowband = 1; int C = Y_ != NULL ? 2 : 1; + int norm_offset; #ifdef RESYNTH int resynth = 1; #else int resynth = !encode; #endif + struct band_ctx ctx; SAVE_STACK; M = 1<<LM; B = shortBlocks ? M : 1; - ALLOC(_norm, C*M*eBands[m->nbEBands], celt_norm); - ALLOC(lowband_scratch, M*(eBands[m->nbEBands]-eBands[m->nbEBands-1]), celt_norm); + norm_offset = M*eBands[start]; + /* No need to allocate norm for the last band because we don't need an + output in that band. */ + ALLOC(_norm, C*(M*eBands[m->nbEBands-1]-norm_offset), celt_norm); norm = _norm; - norm2 = norm + M*eBands[m->nbEBands]; + norm2 = norm + M*eBands[m->nbEBands-1]-norm_offset; + /* We can use the last band as scratch space because we don't need that + scratch space for the last band. */ + lowband_scratch = X_+M*eBands[m->nbEBands-1]; lowband_offset = 0; + ctx.bandE = bandE; + ctx.ec = ec; + ctx.encode = encode; + ctx.intensity = intensity; + ctx.m = m; + ctx.seed = *seed; + ctx.spread = spread; for (i=start;i<end;i++) { opus_int32 tell; @@ -1210,6 +1408,10 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end, int tf_change=0; unsigned x_cm; unsigned y_cm; + int last; + + ctx.i = i; + last = (i==end-1); X = X_+M*eBands[i]; if (Y_!=NULL) @@ -1223,6 +1425,7 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end, if (i != start) balance -= tell; remaining_bits = total_bits-tell-1; + ctx.remaining_bits = remaining_bits; if (i <= codedBands-1) { curr_balance = balance / IMIN(3, codedBands-i); @@ -1235,26 +1438,30 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end, lowband_offset = i; tf_change = tf_res[i]; + ctx.tf_change = tf_change; if (i>=m->effEBands) { X=norm; if (Y_!=NULL) Y = norm; + lowband_scratch = NULL; } + if (i==end-1) + lowband_scratch = NULL; /* Get a conservative estimate of the collapse_mask's for the bands we're - going to be folding from. */ + going to be folding from. */ if (lowband_offset != 0 && (spread!=SPREAD_AGGRESSIVE || B>1 || tf_change<0)) { int fold_start; int fold_end; int fold_i; /* This ensures we never repeat spectral content within one band */ - effective_lowband = IMAX(M*eBands[start], M*eBands[lowband_offset]-N); + effective_lowband = IMAX(0, M*eBands[lowband_offset]-norm_offset-N); fold_start = lowband_offset; - while(M*eBands[--fold_start] > effective_lowband); + while(M*eBands[--fold_start] > effective_lowband+norm_offset); fold_end = lowband_offset-1; - while(M*eBands[++fold_end] < effective_lowband+N); + while(M*eBands[++fold_end] < effective_lowband+norm_offset+N); x_cm = y_cm = 0; fold_i = fold_start; do { x_cm |= collapse_masks[fold_i*C+0]; @@ -1262,7 +1469,7 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end, } while (++fold_i<fold_end); } /* Otherwise, we'll be using the LCG to fold, so all blocks will (almost - always) be non-zero.*/ + always) be non-zero. */ else x_cm = y_cm = (1<<B)-1; @@ -1270,33 +1477,42 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end, { int j; - /* Switch off dual stereo to do intensity */ + /* Switch off dual stereo to do intensity. */ dual_stereo = 0; if (resynth) - for (j=M*eBands[start];j<M*eBands[i];j++) + for (j=0;j<M*eBands[i]-norm_offset;j++) norm[j] = HALF32(norm[j]+norm2[j]); } if (dual_stereo) { - x_cm = quant_band(encode, m, i, X, NULL, N, b/2, spread, B, intensity, tf_change, - effective_lowband != -1 ? norm+effective_lowband : NULL, ec, &remaining_bits, LM, - norm+M*eBands[i], bandE, 0, seed, Q15ONE, lowband_scratch, x_cm); - y_cm = quant_band(encode, m, i, Y, NULL, N, b/2, spread, B, intensity, tf_change, - effective_lowband != -1 ? norm2+effective_lowband : NULL, ec, &remaining_bits, LM, - norm2+M*eBands[i], bandE, 0, seed, Q15ONE, lowband_scratch, y_cm); + x_cm = quant_band(&ctx, X, N, b/2, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, Q15ONE, lowband_scratch, x_cm); + y_cm = quant_band(&ctx, Y, N, b/2, B, + effective_lowband != -1 ? norm2+effective_lowband : NULL, LM, + last?NULL:norm2+M*eBands[i]-norm_offset, Q15ONE, lowband_scratch, y_cm); } else { - x_cm = quant_band(encode, m, i, X, Y, N, b, spread, B, intensity, tf_change, - effective_lowband != -1 ? norm+effective_lowband : NULL, ec, &remaining_bits, LM, - norm+M*eBands[i], bandE, 0, seed, Q15ONE, lowband_scratch, x_cm|y_cm); + if (Y!=NULL) + { + x_cm = quant_band_stereo(&ctx, X, Y, N, b, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, lowband_scratch, x_cm|y_cm); + } else { + x_cm = quant_band(&ctx, X, N, b, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, Q15ONE, lowband_scratch, x_cm|y_cm); + } y_cm = x_cm; } collapse_masks[i*C+0] = (unsigned char)x_cm; collapse_masks[i*C+C-1] = (unsigned char)y_cm; balance += pulses[i] + tell; - /* Update the folding position only as long as we have 1 bit/sample depth */ + /* Update the folding position only as long as we have 1 bit/sample depth. */ update_lowband = b>(N<<BITRES); } + *seed = ctx.seed; + RESTORE_STACK; } diff --git a/src/opus-1.0.2/celt/bands.h b/src/opus-1.1/celt/bands.h index 9ff8ffd7..96ba52a6 100644 --- a/src/opus-1.0.2/celt/bands.h +++ b/src/opus-1.1/celt/bands.h @@ -39,7 +39,7 @@ /** Compute the amplitude (sqrt energy) in each of the bands * @param m Mode data * @param X Spectrum - * @param bands Square root of the energy for each band (returned) + * @param bandE Square root of the energy for each band (returned) */ void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int M); @@ -49,16 +49,17 @@ void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *band equal to 1 * @param m Mode data * @param X Spectrum (returned normalised) - * @param bands Square root of the energy for each band + * @param bandE Square root of the energy for each band */ void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, celt_norm * OPUS_RESTRICT X, const celt_ener *bandE, int end, int C, int M); /** Denormalise each band of X to restore full amplitude * @param m Mode data * @param X Spectrum (returned de-normalised) - * @param bands Square root of the energy for each band + * @param bandE Square root of the energy for each band */ -void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X, celt_sig * OPUS_RESTRICT freq, const celt_ener *bandE, int end, int C, int M); +void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X, + celt_sig * OPUS_RESTRICT freq, const opus_val16 *bandE, int start, int end, int C, int M); #define SPREAD_NONE (0) #define SPREAD_LIGHT (1) @@ -76,14 +77,30 @@ void measure_norm_mse(const CELTMode *m, float *X, float *X0, float *bandE, floa void haar1(celt_norm *X, int N0, int stride); /** Quantisation/encoding of the residual spectrum + * @param encode flag that indicates whether we're encoding (1) or decoding (0) * @param m Mode data + * @param start First band to process + * @param end Last band to process + 1 * @param X Residual (normalised) + * @param Y Residual (normalised) for second channel (or NULL for mono) + * @param collapse_masks Anti-collapse tracking mask + * @param bandE Square root of the energy for each band + * @param pulses Bit allocation (per band) for PVQ + * @param shortBlocks Zero for long blocks, non-zero for short blocks + * @param spread Amount of spreading to use + * @param dual_stereo Zero for MS stereo, non-zero for dual stereo + * @param intensity First band to use intensity stereo + * @param tf_res Time-frequency resolution change * @param total_bits Total number of bits that can be used for the frame (including the ones already spent) - * @param enc Entropy encoder + * @param balance Number of unallocated bits + * @param en Entropy coder state + * @param LM log2() of the number of 2.5 subframes in the frame + * @param codedBands Last band to receive bits + 1 + * @param seed Random generator seed */ void quant_all_bands(int encode, const CELTMode *m, int start, int end, celt_norm * X, celt_norm * Y, unsigned char *collapse_masks, const celt_ener *bandE, int *pulses, - int time_domain, int fold, int dual_stereo, int intensity, int *tf_res, + int shortBlocks, int spread, int dual_stereo, int intensity, int *tf_res, opus_int32 total_bits, opus_int32 balance, ec_ctx *ec, int M, int codedBands, opus_uint32 *seed); void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_masks, int LM, int C, int size, @@ -92,4 +109,6 @@ void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_mas opus_uint32 celt_lcg_rand(opus_uint32 seed); +int hysteresis_decision(opus_val16 val, const opus_val16 *thresholds, const opus_val16 *hysteresis, int N, int prev); + #endif /* BANDS_H */ diff --git a/src/opus-1.1/celt/celt.c b/src/opus-1.1/celt/celt.c new file mode 100644 index 00000000..3e0ce6e6 --- /dev/null +++ b/src/opus-1.1/celt/celt.c @@ -0,0 +1,223 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2010 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#define CELT_C + +#include "os_support.h" +#include "mdct.h" +#include <math.h> +#include "celt.h" +#include "pitch.h" +#include "bands.h" +#include "modes.h" +#include "entcode.h" +#include "quant_bands.h" +#include "rate.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "float_cast.h" +#include <stdarg.h> +#include "celt_lpc.h" +#include "vq.h" + +#ifndef PACKAGE_VERSION +#define PACKAGE_VERSION "unknown" +#endif + + +int resampling_factor(opus_int32 rate) +{ + int ret; + switch (rate) + { + case 48000: + ret = 1; + break; + case 24000: + ret = 2; + break; + case 16000: + ret = 3; + break; + case 12000: + ret = 4; + break; + case 8000: + ret = 6; + break; + default: +#ifndef CUSTOM_MODES + celt_assert(0); +#endif + ret = 0; + break; + } + return ret; +} + +#ifndef OVERRIDE_COMB_FILTER_CONST +static void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N, + opus_val16 g10, opus_val16 g11, opus_val16 g12) +{ + opus_val32 x0, x1, x2, x3, x4; + int i; + x4 = x[-T-2]; + x3 = x[-T-1]; + x2 = x[-T]; + x1 = x[-T+1]; + for (i=0;i<N;i++) + { + x0=x[i-T+2]; + y[i] = x[i] + + MULT16_32_Q15(g10,x2) + + MULT16_32_Q15(g11,ADD32(x1,x3)) + + MULT16_32_Q15(g12,ADD32(x0,x4)); + x4=x3; + x3=x2; + x2=x1; + x1=x0; + } + +} +#endif + +void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N, + opus_val16 g0, opus_val16 g1, int tapset0, int tapset1, + const opus_val16 *window, int overlap) +{ + int i; + /* printf ("%d %d %f %f\n", T0, T1, g0, g1); */ + opus_val16 g00, g01, g02, g10, g11, g12; + opus_val32 x0, x1, x2, x3, x4; + static const opus_val16 gains[3][3] = { + {QCONST16(0.3066406250f, 15), QCONST16(0.2170410156f, 15), QCONST16(0.1296386719f, 15)}, + {QCONST16(0.4638671875f, 15), QCONST16(0.2680664062f, 15), QCONST16(0.f, 15)}, + {QCONST16(0.7998046875f, 15), QCONST16(0.1000976562f, 15), QCONST16(0.f, 15)}}; + + if (g0==0 && g1==0) + { + /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */ + if (x!=y) + OPUS_MOVE(y, x, N); + return; + } + g00 = MULT16_16_Q15(g0, gains[tapset0][0]); + g01 = MULT16_16_Q15(g0, gains[tapset0][1]); + g02 = MULT16_16_Q15(g0, gains[tapset0][2]); + g10 = MULT16_16_Q15(g1, gains[tapset1][0]); + g11 = MULT16_16_Q15(g1, gains[tapset1][1]); + g12 = MULT16_16_Q15(g1, gains[tapset1][2]); + x1 = x[-T1+1]; + x2 = x[-T1 ]; + x3 = x[-T1-1]; + x4 = x[-T1-2]; + for (i=0;i<overlap;i++) + { + opus_val16 f; + x0=x[i-T1+2]; + f = MULT16_16_Q15(window[i],window[i]); + y[i] = x[i] + + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g00),x[i-T0]) + + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g01),ADD32(x[i-T0+1],x[i-T0-1])) + + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g02),ADD32(x[i-T0+2],x[i-T0-2])) + + MULT16_32_Q15(MULT16_16_Q15(f,g10),x2) + + MULT16_32_Q15(MULT16_16_Q15(f,g11),ADD32(x1,x3)) + + MULT16_32_Q15(MULT16_16_Q15(f,g12),ADD32(x0,x4)); + x4=x3; + x3=x2; + x2=x1; + x1=x0; + + } + if (g1==0) + { + /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */ + if (x!=y) + OPUS_MOVE(y+overlap, x+overlap, N-overlap); + return; + } + + /* Compute the part with the constant filter. */ + comb_filter_const(y+i, x+i, T1, N-i, g10, g11, g12); +} + +const signed char tf_select_table[4][8] = { + {0, -1, 0, -1, 0,-1, 0,-1}, + {0, -1, 0, -2, 1, 0, 1,-1}, + {0, -2, 0, -3, 2, 0, 1,-1}, + {0, -2, 0, -3, 3, 0, 1,-1}, +}; + + +void init_caps(const CELTMode *m,int *cap,int LM,int C) +{ + int i; + for (i=0;i<m->nbEBands;i++) + { + int N; + N=(m->eBands[i+1]-m->eBands[i])<<LM; + cap[i] = (m->cache.caps[m->nbEBands*(2*LM+C-1)+i]+64)*C*N>>2; + } +} + + + +const char *opus_strerror(int error) +{ + static const char * const error_strings[8] = { + "success", + "invalid argument", + "buffer too small", + "internal error", + "corrupted stream", + "request not implemented", + "invalid state", + "memory allocation failed" + }; + if (error > 0 || error < -7) + return "unknown error"; + else + return error_strings[-error]; +} + +const char *opus_get_version_string(void) +{ + return "libopus " PACKAGE_VERSION +#ifdef FIXED_POINT + "-fixed" +#endif +#ifdef FUZZING + "-fuzzing" +#endif + ; +} diff --git a/src/opus-1.0.2/celt/celt.h b/src/opus-1.1/celt/celt.h index 218cd883..5deea1f0 100644 --- a/src/opus-1.0.2/celt/celt.h +++ b/src/opus-1.1/celt/celt.h @@ -50,10 +50,26 @@ extern "C" { #define CELTDecoder OpusCustomDecoder #define CELTMode OpusCustomMode -#define _celt_check_mode_ptr_ptr(ptr) ((ptr) + ((ptr) - (const CELTMode**)(ptr))) +typedef struct { + int valid; + float tonality; + float tonality_slope; + float noisiness; + float activity; + float music_prob; + int bandwidth; +}AnalysisInfo; + +#define __celt_check_mode_ptr_ptr(ptr) ((ptr) + ((ptr) - (const CELTMode**)(ptr))) + +#define __celt_check_analysis_ptr(ptr) ((ptr) + ((ptr) - (const AnalysisInfo*)(ptr))) /* Encoder/decoder Requests */ +/* Expose this option again when variable framesize actually works */ +#define OPUS_FRAMESIZE_VARIABLE 5010 /**< Optimize the frame size dynamically */ + + #define CELT_SET_PREDICTION_REQUEST 10002 /** Controls the use of interframe prediction. 0=Independent frames @@ -81,12 +97,24 @@ extern "C" { #define CELT_GET_MODE_REQUEST 10015 /** Get the CELTMode used by an encoder or decoder */ -#define CELT_GET_MODE(x) CELT_GET_MODE_REQUEST, _celt_check_mode_ptr_ptr(x) +#define CELT_GET_MODE(x) CELT_GET_MODE_REQUEST, __celt_check_mode_ptr_ptr(x) #define CELT_SET_SIGNALLING_REQUEST 10016 #define CELT_SET_SIGNALLING(x) CELT_SET_SIGNALLING_REQUEST, __opus_check_int(x) +#define CELT_SET_TONALITY_REQUEST 10018 +#define CELT_SET_TONALITY(x) CELT_SET_TONALITY_REQUEST, __opus_check_int(x) +#define CELT_SET_TONALITY_SLOPE_REQUEST 10020 +#define CELT_SET_TONALITY_SLOPE(x) CELT_SET_TONALITY_SLOPE_REQUEST, __opus_check_int(x) + +#define CELT_SET_ANALYSIS_REQUEST 10022 +#define CELT_SET_ANALYSIS(x) CELT_SET_ANALYSIS_REQUEST, __celt_check_analysis_ptr(x) + +#define OPUS_SET_LFE_REQUEST 10024 +#define OPUS_SET_LFE(x) OPUS_SET_LFE_REQUEST, __opus_check_int(x) +#define OPUS_SET_ENERGY_MASK_REQUEST 10026 +#define OPUS_SET_ENERGY_MASK(x) OPUS_SET_ENERGY_MASK_REQUEST, __opus_check_val16_ptr(x) /* Encoder stuff */ @@ -94,7 +122,8 @@ int celt_encoder_get_size(int channels); int celt_encode_with_ec(OpusCustomEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes, ec_enc *enc); -int celt_encoder_init(CELTEncoder *st, opus_int32 sampling_rate, int channels); +int celt_encoder_init(CELTEncoder *st, opus_int32 sampling_rate, int channels, + int arch); @@ -110,6 +139,78 @@ int celt_decode_with_ec(OpusCustomDecoder * OPUS_RESTRICT st, const unsigned cha #define celt_encoder_ctl opus_custom_encoder_ctl #define celt_decoder_ctl opus_custom_decoder_ctl + +#ifdef CUSTOM_MODES +#define OPUS_CUSTOM_NOSTATIC +#else +#define OPUS_CUSTOM_NOSTATIC static OPUS_INLINE +#endif + +static const unsigned char trim_icdf[11] = {126, 124, 119, 109, 87, 41, 19, 9, 4, 2, 0}; +/* Probs: NONE: 21.875%, LIGHT: 6.25%, NORMAL: 65.625%, AGGRESSIVE: 6.25% */ +static const unsigned char spread_icdf[4] = {25, 23, 2, 0}; + +static const unsigned char tapset_icdf[3]={2,1,0}; + +#ifdef CUSTOM_MODES +static const unsigned char toOpusTable[20] = { + 0xE0, 0xE8, 0xF0, 0xF8, + 0xC0, 0xC8, 0xD0, 0xD8, + 0xA0, 0xA8, 0xB0, 0xB8, + 0x00, 0x00, 0x00, 0x00, + 0x80, 0x88, 0x90, 0x98, +}; + +static const unsigned char fromOpusTable[16] = { + 0x80, 0x88, 0x90, 0x98, + 0x40, 0x48, 0x50, 0x58, + 0x20, 0x28, 0x30, 0x38, + 0x00, 0x08, 0x10, 0x18 +}; + +static OPUS_INLINE int toOpus(unsigned char c) +{ + int ret=0; + if (c<0xA0) + ret = toOpusTable[c>>3]; + if (ret == 0) + return -1; + else + return ret|(c&0x7); +} + +static OPUS_INLINE int fromOpus(unsigned char c) +{ + if (c<0x80) + return -1; + else + return fromOpusTable[(c>>3)-16] | (c&0x7); +} +#endif /* CUSTOM_MODES */ + +#define COMBFILTER_MAXPERIOD 1024 +#define COMBFILTER_MINPERIOD 15 + +extern const signed char tf_select_table[4][8]; + +int resampling_factor(opus_int32 rate); + +void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RESTRICT inp, + int N, int CC, int upsample, const opus_val16 *coef, celt_sig *mem, int clip); + +void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N, + opus_val16 g0, opus_val16 g1, int tapset0, int tapset1, + const opus_val16 *window, int overlap); + +void init_caps(const CELTMode *m,int *cap,int LM,int C); + +#ifdef RESYNTH +void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, const opus_val16 *coef, celt_sig *mem, celt_sig * OPUS_RESTRICT scratch); + +void compute_inv_mdcts(const CELTMode *mode, int shortBlocks, celt_sig *X, + celt_sig * OPUS_RESTRICT out_mem[], int C, int LM); +#endif + #ifdef __cplusplus } #endif diff --git a/src/opus-1.1/celt/celt_decoder.c b/src/opus-1.1/celt/celt_decoder.c new file mode 100644 index 00000000..830398ee --- /dev/null +++ b/src/opus-1.1/celt/celt_decoder.c @@ -0,0 +1,1195 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2010 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#define CELT_DECODER_C + +#include "cpu_support.h" +#include "os_support.h" +#include "mdct.h" +#include <math.h> +#include "celt.h" +#include "pitch.h" +#include "bands.h" +#include "modes.h" +#include "entcode.h" +#include "quant_bands.h" +#include "rate.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "float_cast.h" +#include <stdarg.h> +#include "celt_lpc.h" +#include "vq.h" + +/**********************************************************************/ +/* */ +/* DECODER */ +/* */ +/**********************************************************************/ +#define DECODE_BUFFER_SIZE 2048 + +/** Decoder state + @brief Decoder state + */ +struct OpusCustomDecoder { + const OpusCustomMode *mode; + int overlap; + int channels; + int stream_channels; + + int downsample; + int start, end; + int signalling; + int arch; + + /* Everything beyond this point gets cleared on a reset */ +#define DECODER_RESET_START rng + + opus_uint32 rng; + int error; + int last_pitch_index; + int loss_count; + int postfilter_period; + int postfilter_period_old; + opus_val16 postfilter_gain; + opus_val16 postfilter_gain_old; + int postfilter_tapset; + int postfilter_tapset_old; + + celt_sig preemph_memD[2]; + + celt_sig _decode_mem[1]; /* Size = channels*(DECODE_BUFFER_SIZE+mode->overlap) */ + /* opus_val16 lpc[], Size = channels*LPC_ORDER */ + /* opus_val16 oldEBands[], Size = 2*mode->nbEBands */ + /* opus_val16 oldLogE[], Size = 2*mode->nbEBands */ + /* opus_val16 oldLogE2[], Size = 2*mode->nbEBands */ + /* opus_val16 backgroundLogE[], Size = 2*mode->nbEBands */ +}; + +int celt_decoder_get_size(int channels) +{ + const CELTMode *mode = opus_custom_mode_create(48000, 960, NULL); + return opus_custom_decoder_get_size(mode, channels); +} + +OPUS_CUSTOM_NOSTATIC int opus_custom_decoder_get_size(const CELTMode *mode, int channels) +{ + int size = sizeof(struct CELTDecoder) + + (channels*(DECODE_BUFFER_SIZE+mode->overlap)-1)*sizeof(celt_sig) + + channels*LPC_ORDER*sizeof(opus_val16) + + 4*2*mode->nbEBands*sizeof(opus_val16); + return size; +} + +#ifdef CUSTOM_MODES +CELTDecoder *opus_custom_decoder_create(const CELTMode *mode, int channels, int *error) +{ + int ret; + CELTDecoder *st = (CELTDecoder *)opus_alloc(opus_custom_decoder_get_size(mode, channels)); + ret = opus_custom_decoder_init(st, mode, channels); + if (ret != OPUS_OK) + { + opus_custom_decoder_destroy(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} +#endif /* CUSTOM_MODES */ + +int celt_decoder_init(CELTDecoder *st, opus_int32 sampling_rate, int channels) +{ + int ret; + ret = opus_custom_decoder_init(st, opus_custom_mode_create(48000, 960, NULL), channels); + if (ret != OPUS_OK) + return ret; + st->downsample = resampling_factor(sampling_rate); + if (st->downsample==0) + return OPUS_BAD_ARG; + else + return OPUS_OK; +} + +OPUS_CUSTOM_NOSTATIC int opus_custom_decoder_init(CELTDecoder *st, const CELTMode *mode, int channels) +{ + if (channels < 0 || channels > 2) + return OPUS_BAD_ARG; + + if (st==NULL) + return OPUS_ALLOC_FAIL; + + OPUS_CLEAR((char*)st, opus_custom_decoder_get_size(mode, channels)); + + st->mode = mode; + st->overlap = mode->overlap; + st->stream_channels = st->channels = channels; + + st->downsample = 1; + st->start = 0; + st->end = st->mode->effEBands; + st->signalling = 1; + st->arch = opus_select_arch(); + + st->loss_count = 0; + + opus_custom_decoder_ctl(st, OPUS_RESET_STATE); + + return OPUS_OK; +} + +#ifdef CUSTOM_MODES +void opus_custom_decoder_destroy(CELTDecoder *st) +{ + opus_free(st); +} +#endif /* CUSTOM_MODES */ + +static OPUS_INLINE opus_val16 SIG2WORD16(celt_sig x) +{ +#ifdef FIXED_POINT + x = PSHR32(x, SIG_SHIFT); + x = MAX32(x, -32768); + x = MIN32(x, 32767); + return EXTRACT16(x); +#else + return (opus_val16)x; +#endif +} + +#ifndef RESYNTH +static +#endif +void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, const opus_val16 *coef, celt_sig *mem, celt_sig * OPUS_RESTRICT scratch) +{ + int c; + int Nd; + int apply_downsampling=0; + opus_val16 coef0; + + coef0 = coef[0]; + Nd = N/downsample; + c=0; do { + int j; + celt_sig * OPUS_RESTRICT x; + opus_val16 * OPUS_RESTRICT y; + celt_sig m = mem[c]; + x =in[c]; + y = pcm+c; +#ifdef CUSTOM_MODES + if (coef[1] != 0) + { + opus_val16 coef1 = coef[1]; + opus_val16 coef3 = coef[3]; + for (j=0;j<N;j++) + { + celt_sig tmp = x[j] + m + VERY_SMALL; + m = MULT16_32_Q15(coef0, tmp) + - MULT16_32_Q15(coef1, x[j]); + tmp = SHL32(MULT16_32_Q15(coef3, tmp), 2); + scratch[j] = tmp; + } + apply_downsampling=1; + } else +#endif + if (downsample>1) + { + /* Shortcut for the standard (non-custom modes) case */ + for (j=0;j<N;j++) + { + celt_sig tmp = x[j] + m + VERY_SMALL; + m = MULT16_32_Q15(coef0, tmp); + scratch[j] = tmp; + } + apply_downsampling=1; + } else { + /* Shortcut for the standard (non-custom modes) case */ + for (j=0;j<N;j++) + { + celt_sig tmp = x[j] + m + VERY_SMALL; + m = MULT16_32_Q15(coef0, tmp); + y[j*C] = SCALEOUT(SIG2WORD16(tmp)); + } + } + mem[c] = m; + + if (apply_downsampling) + { + /* Perform down-sampling */ + for (j=0;j<Nd;j++) + y[j*C] = SCALEOUT(SIG2WORD16(scratch[j*downsample])); + } + } while (++c<C); +} + +/** Compute the IMDCT and apply window for all sub-frames and + all channels in a frame */ +#ifndef RESYNTH +static +#endif +void compute_inv_mdcts(const CELTMode *mode, int shortBlocks, celt_sig *X, + celt_sig * OPUS_RESTRICT out_mem[], int C, int LM) +{ + int b, c; + int B; + int N; + int shift; + const int overlap = OVERLAP(mode); + + if (shortBlocks) + { + B = shortBlocks; + N = mode->shortMdctSize; + shift = mode->maxLM; + } else { + B = 1; + N = mode->shortMdctSize<<LM; + shift = mode->maxLM-LM; + } + c=0; do { + /* IMDCT on the interleaved the sub-frames, overlap-add is performed by the IMDCT */ + for (b=0;b<B;b++) + clt_mdct_backward(&mode->mdct, &X[b+c*N*B], out_mem[c]+N*b, mode->window, overlap, shift, B); + } while (++c<C); +} + +static void tf_decode(int start, int end, int isTransient, int *tf_res, int LM, ec_dec *dec) +{ + int i, curr, tf_select; + int tf_select_rsv; + int tf_changed; + int logp; + opus_uint32 budget; + opus_uint32 tell; + + budget = dec->storage*8; + tell = ec_tell(dec); + logp = isTransient ? 2 : 4; + tf_select_rsv = LM>0 && tell+logp+1<=budget; + budget -= tf_select_rsv; + tf_changed = curr = 0; + for (i=start;i<end;i++) + { + if (tell+logp<=budget) + { + curr ^= ec_dec_bit_logp(dec, logp); + tell = ec_tell(dec); + tf_changed |= curr; + } + tf_res[i] = curr; + logp = isTransient ? 4 : 5; + } + tf_select = 0; + if (tf_select_rsv && + tf_select_table[LM][4*isTransient+0+tf_changed] != + tf_select_table[LM][4*isTransient+2+tf_changed]) + { + tf_select = ec_dec_bit_logp(dec, 1); + } + for (i=start;i<end;i++) + { + tf_res[i] = tf_select_table[LM][4*isTransient+2*tf_select+tf_res[i]]; + } +} + +/* The maximum pitch lag to allow in the pitch-based PLC. It's possible to save + CPU time in the PLC pitch search by making this smaller than MAX_PERIOD. The + current value corresponds to a pitch of 66.67 Hz. */ +#define PLC_PITCH_LAG_MAX (720) +/* The minimum pitch lag to allow in the pitch-based PLC. This corresponds to a + pitch of 480 Hz. */ +#define PLC_PITCH_LAG_MIN (100) + +static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, opus_val16 * OPUS_RESTRICT pcm, int N, int LM) +{ + int c; + int i; + const int C = st->channels; + celt_sig *decode_mem[2]; + celt_sig *out_syn[2]; + opus_val16 *lpc; + opus_val16 *oldBandE, *oldLogE, *oldLogE2, *backgroundLogE; + const OpusCustomMode *mode; + int nbEBands; + int overlap; + int start; + int downsample; + int loss_count; + int noise_based; + const opus_int16 *eBands; + VARDECL(celt_sig, scratch); + SAVE_STACK; + + mode = st->mode; + nbEBands = mode->nbEBands; + overlap = mode->overlap; + eBands = mode->eBands; + + c=0; do { + decode_mem[c] = st->_decode_mem + c*(DECODE_BUFFER_SIZE+overlap); + out_syn[c] = decode_mem[c]+DECODE_BUFFER_SIZE-N; + } while (++c<C); + lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+overlap)*C); + oldBandE = lpc+C*LPC_ORDER; + oldLogE = oldBandE + 2*nbEBands; + oldLogE2 = oldLogE + 2*nbEBands; + backgroundLogE = oldLogE2 + 2*nbEBands; + + loss_count = st->loss_count; + start = st->start; + downsample = st->downsample; + noise_based = loss_count >= 5 || start != 0; + ALLOC(scratch, noise_based?N*C:N, celt_sig); + if (noise_based) + { + /* Noise-based PLC/CNG */ + celt_sig *freq; + VARDECL(celt_norm, X); + opus_uint32 seed; + opus_val16 *plcLogE; + int end; + int effEnd; + + end = st->end; + effEnd = IMAX(start, IMIN(end, mode->effEBands)); + + /* Share the interleaved signal MDCT coefficient buffer with the + deemphasis scratch buffer. */ + freq = scratch; + ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ + + if (loss_count >= 5) + plcLogE = backgroundLogE; + else { + /* Energy decay */ + opus_val16 decay = loss_count==0 ? + QCONST16(1.5f, DB_SHIFT) : QCONST16(.5f, DB_SHIFT); + c=0; do + { + for (i=start;i<end;i++) + oldBandE[c*nbEBands+i] -= decay; + } while (++c<C); + plcLogE = oldBandE; + } + seed = st->rng; + for (c=0;c<C;c++) + { + for (i=start;i<effEnd;i++) + { + int j; + int boffs; + int blen; + boffs = N*c+(eBands[i]<<LM); + blen = (eBands[i+1]-eBands[i])<<LM; + for (j=0;j<blen;j++) + { + seed = celt_lcg_rand(seed); + X[boffs+j] = (celt_norm)((opus_int32)seed>>20); + } + renormalise_vector(X+boffs, blen, Q15ONE); + } + } + st->rng = seed; + + denormalise_bands(mode, X, freq, plcLogE, start, effEnd, C, 1<<LM); + + c=0; do { + int bound = eBands[effEnd]<<LM; + if (downsample!=1) + bound = IMIN(bound, N/downsample); + for (i=bound;i<N;i++) + freq[c*N+i] = 0; + } while (++c<C); + c=0; do { + OPUS_MOVE(decode_mem[c], decode_mem[c]+N, + DECODE_BUFFER_SIZE-N+(overlap>>1)); + } while (++c<C); + compute_inv_mdcts(mode, 0, freq, out_syn, C, LM); + } else { + /* Pitch-based PLC */ + const opus_val16 *window; + opus_val16 fade = Q15ONE; + int pitch_index; + VARDECL(opus_val32, etmp); + VARDECL(opus_val16, exc); + + if (loss_count == 0) + { + VARDECL( opus_val16, lp_pitch_buf ); + ALLOC( lp_pitch_buf, DECODE_BUFFER_SIZE>>1, opus_val16 ); + pitch_downsample(decode_mem, lp_pitch_buf, + DECODE_BUFFER_SIZE, C, st->arch); + pitch_search(lp_pitch_buf+(PLC_PITCH_LAG_MAX>>1), lp_pitch_buf, + DECODE_BUFFER_SIZE-PLC_PITCH_LAG_MAX, + PLC_PITCH_LAG_MAX-PLC_PITCH_LAG_MIN, &pitch_index, st->arch); + pitch_index = PLC_PITCH_LAG_MAX-pitch_index; + st->last_pitch_index = pitch_index; + } else { + pitch_index = st->last_pitch_index; + fade = QCONST16(.8f,15); + } + + ALLOC(etmp, overlap, opus_val32); + ALLOC(exc, MAX_PERIOD, opus_val16); + window = mode->window; + c=0; do { + opus_val16 decay; + opus_val16 attenuation; + opus_val32 S1=0; + celt_sig *buf; + int extrapolation_offset; + int extrapolation_len; + int exc_length; + int j; + + buf = decode_mem[c]; + for (i=0;i<MAX_PERIOD;i++) { + exc[i] = ROUND16(buf[DECODE_BUFFER_SIZE-MAX_PERIOD+i], SIG_SHIFT); + } + + if (loss_count == 0) + { + opus_val32 ac[LPC_ORDER+1]; + /* Compute LPC coefficients for the last MAX_PERIOD samples before + the first loss so we can work in the excitation-filter domain. */ + _celt_autocorr(exc, ac, window, overlap, + LPC_ORDER, MAX_PERIOD, st->arch); + /* Add a noise floor of -40 dB. */ +#ifdef FIXED_POINT + ac[0] += SHR32(ac[0],13); +#else + ac[0] *= 1.0001f; +#endif + /* Use lag windowing to stabilize the Levinson-Durbin recursion. */ + for (i=1;i<=LPC_ORDER;i++) + { + /*ac[i] *= exp(-.5*(2*M_PI*.002*i)*(2*M_PI*.002*i));*/ +#ifdef FIXED_POINT + ac[i] -= MULT16_32_Q15(2*i*i, ac[i]); +#else + ac[i] -= ac[i]*(0.008f*0.008f)*i*i; +#endif + } + _celt_lpc(lpc+c*LPC_ORDER, ac, LPC_ORDER); + } + /* We want the excitation for 2 pitch periods in order to look for a + decaying signal, but we can't get more than MAX_PERIOD. */ + exc_length = IMIN(2*pitch_index, MAX_PERIOD); + /* Initialize the LPC history with the samples just before the start + of the region for which we're computing the excitation. */ + { + opus_val16 lpc_mem[LPC_ORDER]; + for (i=0;i<LPC_ORDER;i++) + { + lpc_mem[i] = + ROUND16(buf[DECODE_BUFFER_SIZE-exc_length-1-i], SIG_SHIFT); + } + /* Compute the excitation for exc_length samples before the loss. */ + celt_fir(exc+MAX_PERIOD-exc_length, lpc+c*LPC_ORDER, + exc+MAX_PERIOD-exc_length, exc_length, LPC_ORDER, lpc_mem); + } + + /* Check if the waveform is decaying, and if so how fast. + We do this to avoid adding energy when concealing in a segment + with decaying energy. */ + { + opus_val32 E1=1, E2=1; + int decay_length; +#ifdef FIXED_POINT + int shift = IMAX(0,2*celt_zlog2(celt_maxabs16(&exc[MAX_PERIOD-exc_length], exc_length))-20); +#endif + decay_length = exc_length>>1; + for (i=0;i<decay_length;i++) + { + opus_val16 e; + e = exc[MAX_PERIOD-decay_length+i]; + E1 += SHR32(MULT16_16(e, e), shift); + e = exc[MAX_PERIOD-2*decay_length+i]; + E2 += SHR32(MULT16_16(e, e), shift); + } + E1 = MIN32(E1, E2); + decay = celt_sqrt(frac_div32(SHR32(E1, 1), E2)); + } + + /* Move the decoder memory one frame to the left to give us room to + add the data for the new frame. We ignore the overlap that extends + past the end of the buffer, because we aren't going to use it. */ + OPUS_MOVE(buf, buf+N, DECODE_BUFFER_SIZE-N); + + /* Extrapolate from the end of the excitation with a period of + "pitch_index", scaling down each period by an additional factor of + "decay". */ + extrapolation_offset = MAX_PERIOD-pitch_index; + /* We need to extrapolate enough samples to cover a complete MDCT + window (including overlap/2 samples on both sides). */ + extrapolation_len = N+overlap; + /* We also apply fading if this is not the first loss. */ + attenuation = MULT16_16_Q15(fade, decay); + for (i=j=0;i<extrapolation_len;i++,j++) + { + opus_val16 tmp; + if (j >= pitch_index) { + j -= pitch_index; + attenuation = MULT16_16_Q15(attenuation, decay); + } + buf[DECODE_BUFFER_SIZE-N+i] = + SHL32(EXTEND32(MULT16_16_Q15(attenuation, + exc[extrapolation_offset+j])), SIG_SHIFT); + /* Compute the energy of the previously decoded signal whose + excitation we're copying. */ + tmp = ROUND16( + buf[DECODE_BUFFER_SIZE-MAX_PERIOD-N+extrapolation_offset+j], + SIG_SHIFT); + S1 += SHR32(MULT16_16(tmp, tmp), 8); + } + + { + opus_val16 lpc_mem[LPC_ORDER]; + /* Copy the last decoded samples (prior to the overlap region) to + synthesis filter memory so we can have a continuous signal. */ + for (i=0;i<LPC_ORDER;i++) + lpc_mem[i] = ROUND16(buf[DECODE_BUFFER_SIZE-N-1-i], SIG_SHIFT); + /* Apply the synthesis filter to convert the excitation back into + the signal domain. */ + celt_iir(buf+DECODE_BUFFER_SIZE-N, lpc+c*LPC_ORDER, + buf+DECODE_BUFFER_SIZE-N, extrapolation_len, LPC_ORDER, + lpc_mem); + } + + /* Check if the synthesis energy is higher than expected, which can + happen with the signal changes during our window. If so, + attenuate. */ + { + opus_val32 S2=0; + for (i=0;i<extrapolation_len;i++) + { + opus_val16 tmp = ROUND16(buf[DECODE_BUFFER_SIZE-N+i], SIG_SHIFT); + S2 += SHR32(MULT16_16(tmp, tmp), 8); + } + /* This checks for an "explosion" in the synthesis. */ +#ifdef FIXED_POINT + if (!(S1 > SHR32(S2,2))) +#else + /* The float test is written this way to catch NaNs in the output + of the IIR filter at the same time. */ + if (!(S1 > 0.2f*S2)) +#endif + { + for (i=0;i<extrapolation_len;i++) + buf[DECODE_BUFFER_SIZE-N+i] = 0; + } else if (S1 < S2) + { + opus_val16 ratio = celt_sqrt(frac_div32(SHR32(S1,1)+1,S2+1)); + for (i=0;i<overlap;i++) + { + opus_val16 tmp_g = Q15ONE + - MULT16_16_Q15(window[i], Q15ONE-ratio); + buf[DECODE_BUFFER_SIZE-N+i] = + MULT16_32_Q15(tmp_g, buf[DECODE_BUFFER_SIZE-N+i]); + } + for (i=overlap;i<extrapolation_len;i++) + { + buf[DECODE_BUFFER_SIZE-N+i] = + MULT16_32_Q15(ratio, buf[DECODE_BUFFER_SIZE-N+i]); + } + } + } + + /* Apply the pre-filter to the MDCT overlap for the next frame because + the post-filter will be re-applied in the decoder after the MDCT + overlap. */ + comb_filter(etmp, buf+DECODE_BUFFER_SIZE, + st->postfilter_period, st->postfilter_period, overlap, + -st->postfilter_gain, -st->postfilter_gain, + st->postfilter_tapset, st->postfilter_tapset, NULL, 0); + + /* Simulate TDAC on the concealed audio so that it blends with the + MDCT of the next frame. */ + for (i=0;i<overlap/2;i++) + { + buf[DECODE_BUFFER_SIZE+i] = + MULT16_32_Q15(window[i], etmp[overlap-1-i]) + + MULT16_32_Q15(window[overlap-i-1], etmp[i]); + } + } while (++c<C); + } + + deemphasis(out_syn, pcm, N, C, downsample, + mode->preemph, st->preemph_memD, scratch); + + st->loss_count = loss_count+1; + + RESTORE_STACK; +} + +int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_val16 * OPUS_RESTRICT pcm, int frame_size, ec_dec *dec) +{ + int c, i, N; + int spread_decision; + opus_int32 bits; + ec_dec _dec; + VARDECL(celt_sig, freq); + VARDECL(celt_norm, X); + VARDECL(int, fine_quant); + VARDECL(int, pulses); + VARDECL(int, cap); + VARDECL(int, offsets); + VARDECL(int, fine_priority); + VARDECL(int, tf_res); + VARDECL(unsigned char, collapse_masks); + celt_sig *decode_mem[2]; + celt_sig *out_syn[2]; + opus_val16 *lpc; + opus_val16 *oldBandE, *oldLogE, *oldLogE2, *backgroundLogE; + + int shortBlocks; + int isTransient; + int intra_ener; + const int CC = st->channels; + int LM, M; + int effEnd; + int codedBands; + int alloc_trim; + int postfilter_pitch; + opus_val16 postfilter_gain; + int intensity=0; + int dual_stereo=0; + opus_int32 total_bits; + opus_int32 balance; + opus_int32 tell; + int dynalloc_logp; + int postfilter_tapset; + int anti_collapse_rsv; + int anti_collapse_on=0; + int silence; + int C = st->stream_channels; + const OpusCustomMode *mode; + int nbEBands; + int overlap; + const opus_int16 *eBands; + ALLOC_STACK; + + mode = st->mode; + nbEBands = mode->nbEBands; + overlap = mode->overlap; + eBands = mode->eBands; + frame_size *= st->downsample; + + c=0; do { + decode_mem[c] = st->_decode_mem + c*(DECODE_BUFFER_SIZE+overlap); + } while (++c<CC); + lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+overlap)*CC); + oldBandE = lpc+CC*LPC_ORDER; + oldLogE = oldBandE + 2*nbEBands; + oldLogE2 = oldLogE + 2*nbEBands; + backgroundLogE = oldLogE2 + 2*nbEBands; + +#ifdef CUSTOM_MODES + if (st->signalling && data!=NULL) + { + int data0=data[0]; + /* Convert "standard mode" to Opus header */ + if (mode->Fs==48000 && mode->shortMdctSize==120) + { + data0 = fromOpus(data0); + if (data0<0) + return OPUS_INVALID_PACKET; + } + st->end = IMAX(1, mode->effEBands-2*(data0>>5)); + LM = (data0>>3)&0x3; + C = 1 + ((data0>>2)&0x1); + data++; + len--; + if (LM>mode->maxLM) + return OPUS_INVALID_PACKET; + if (frame_size < mode->shortMdctSize<<LM) + return OPUS_BUFFER_TOO_SMALL; + else + frame_size = mode->shortMdctSize<<LM; + } else { +#else + { +#endif + for (LM=0;LM<=mode->maxLM;LM++) + if (mode->shortMdctSize<<LM==frame_size) + break; + if (LM>mode->maxLM) + return OPUS_BAD_ARG; + } + M=1<<LM; + + if (len<0 || len>1275 || pcm==NULL) + return OPUS_BAD_ARG; + + N = M*mode->shortMdctSize; + + effEnd = st->end; + if (effEnd > mode->effEBands) + effEnd = mode->effEBands; + + if (data == NULL || len<=1) + { + celt_decode_lost(st, pcm, N, LM); + RESTORE_STACK; + return frame_size/st->downsample; + } + + if (dec == NULL) + { + ec_dec_init(&_dec,(unsigned char*)data,len); + dec = &_dec; + } + + if (C==1) + { + for (i=0;i<nbEBands;i++) + oldBandE[i]=MAX16(oldBandE[i],oldBandE[nbEBands+i]); + } + + total_bits = len*8; + tell = ec_tell(dec); + + if (tell >= total_bits) + silence = 1; + else if (tell==1) + silence = ec_dec_bit_logp(dec, 15); + else + silence = 0; + if (silence) + { + /* Pretend we've read all the remaining bits */ + tell = len*8; + dec->nbits_total+=tell-ec_tell(dec); + } + + postfilter_gain = 0; + postfilter_pitch = 0; + postfilter_tapset = 0; + if (st->start==0 && tell+16 <= total_bits) + { + if(ec_dec_bit_logp(dec, 1)) + { + int qg, octave; + octave = ec_dec_uint(dec, 6); + postfilter_pitch = (16<<octave)+ec_dec_bits(dec, 4+octave)-1; + qg = ec_dec_bits(dec, 3); + if (ec_tell(dec)+2<=total_bits) + postfilter_tapset = ec_dec_icdf(dec, tapset_icdf, 2); + postfilter_gain = QCONST16(.09375f,15)*(qg+1); + } + tell = ec_tell(dec); + } + + if (LM > 0 && tell+3 <= total_bits) + { + isTransient = ec_dec_bit_logp(dec, 3); + tell = ec_tell(dec); + } + else + isTransient = 0; + + if (isTransient) + shortBlocks = M; + else + shortBlocks = 0; + + /* Decode the global flags (first symbols in the stream) */ + intra_ener = tell+3<=total_bits ? ec_dec_bit_logp(dec, 3) : 0; + /* Get band energies */ + unquant_coarse_energy(mode, st->start, st->end, oldBandE, + intra_ener, dec, C, LM); + + ALLOC(tf_res, nbEBands, int); + tf_decode(st->start, st->end, isTransient, tf_res, LM, dec); + + tell = ec_tell(dec); + spread_decision = SPREAD_NORMAL; + if (tell+4 <= total_bits) + spread_decision = ec_dec_icdf(dec, spread_icdf, 5); + + ALLOC(cap, nbEBands, int); + + init_caps(mode,cap,LM,C); + + ALLOC(offsets, nbEBands, int); + + dynalloc_logp = 6; + total_bits<<=BITRES; + tell = ec_tell_frac(dec); + for (i=st->start;i<st->end;i++) + { + int width, quanta; + int dynalloc_loop_logp; + int boost; + width = C*(eBands[i+1]-eBands[i])<<LM; + /* quanta is 6 bits, but no more than 1 bit/sample + and no less than 1/8 bit/sample */ + quanta = IMIN(width<<BITRES, IMAX(6<<BITRES, width)); + dynalloc_loop_logp = dynalloc_logp; + boost = 0; + while (tell+(dynalloc_loop_logp<<BITRES) < total_bits && boost < cap[i]) + { + int flag; + flag = ec_dec_bit_logp(dec, dynalloc_loop_logp); + tell = ec_tell_frac(dec); + if (!flag) + break; + boost += quanta; + total_bits -= quanta; + dynalloc_loop_logp = 1; + } + offsets[i] = boost; + /* Making dynalloc more likely */ + if (boost>0) + dynalloc_logp = IMAX(2, dynalloc_logp-1); + } + + ALLOC(fine_quant, nbEBands, int); + alloc_trim = tell+(6<<BITRES) <= total_bits ? + ec_dec_icdf(dec, trim_icdf, 7) : 5; + + bits = (((opus_int32)len*8)<<BITRES) - ec_tell_frac(dec) - 1; + anti_collapse_rsv = isTransient&&LM>=2&&bits>=((LM+2)<<BITRES) ? (1<<BITRES) : 0; + bits -= anti_collapse_rsv; + + ALLOC(pulses, nbEBands, int); + ALLOC(fine_priority, nbEBands, int); + + codedBands = compute_allocation(mode, st->start, st->end, offsets, cap, + alloc_trim, &intensity, &dual_stereo, bits, &balance, pulses, + fine_quant, fine_priority, C, LM, dec, 0, 0, 0); + + unquant_fine_energy(mode, st->start, st->end, oldBandE, fine_quant, dec, C); + + /* Decode fixed codebook */ + ALLOC(collapse_masks, C*nbEBands, unsigned char); + ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ + + quant_all_bands(0, mode, st->start, st->end, X, C==2 ? X+N : NULL, collapse_masks, + NULL, pulses, shortBlocks, spread_decision, dual_stereo, intensity, tf_res, + len*(8<<BITRES)-anti_collapse_rsv, balance, dec, LM, codedBands, &st->rng); + + if (anti_collapse_rsv > 0) + { + anti_collapse_on = ec_dec_bits(dec, 1); + } + + unquant_energy_finalise(mode, st->start, st->end, oldBandE, + fine_quant, fine_priority, len*8-ec_tell(dec), dec, C); + + if (anti_collapse_on) + anti_collapse(mode, X, collapse_masks, LM, C, N, + st->start, st->end, oldBandE, oldLogE, oldLogE2, pulses, st->rng); + + ALLOC(freq, IMAX(CC,C)*N, celt_sig); /**< Interleaved signal MDCTs */ + + if (silence) + { + for (i=0;i<C*nbEBands;i++) + oldBandE[i] = -QCONST16(28.f,DB_SHIFT); + for (i=0;i<C*N;i++) + freq[i] = 0; + } else { + /* Synthesis */ + denormalise_bands(mode, X, freq, oldBandE, st->start, effEnd, C, M); + } + c=0; do { + OPUS_MOVE(decode_mem[c], decode_mem[c]+N, DECODE_BUFFER_SIZE-N+overlap/2); + } while (++c<CC); + + c=0; do { + int bound = M*eBands[effEnd]; + if (st->downsample!=1) + bound = IMIN(bound, N/st->downsample); + for (i=bound;i<N;i++) + freq[c*N+i] = 0; + } while (++c<C); + + c=0; do { + out_syn[c] = decode_mem[c]+DECODE_BUFFER_SIZE-N; + } while (++c<CC); + + if (CC==2&&C==1) + { + for (i=0;i<N;i++) + freq[N+i] = freq[i]; + } + if (CC==1&&C==2) + { + for (i=0;i<N;i++) + freq[i] = HALF32(ADD32(freq[i],freq[N+i])); + } + + /* Compute inverse MDCTs */ + compute_inv_mdcts(mode, shortBlocks, freq, out_syn, CC, LM); + + c=0; do { + st->postfilter_period=IMAX(st->postfilter_period, COMBFILTER_MINPERIOD); + st->postfilter_period_old=IMAX(st->postfilter_period_old, COMBFILTER_MINPERIOD); + comb_filter(out_syn[c], out_syn[c], st->postfilter_period_old, st->postfilter_period, mode->shortMdctSize, + st->postfilter_gain_old, st->postfilter_gain, st->postfilter_tapset_old, st->postfilter_tapset, + mode->window, overlap); + if (LM!=0) + comb_filter(out_syn[c]+mode->shortMdctSize, out_syn[c]+mode->shortMdctSize, st->postfilter_period, postfilter_pitch, N-mode->shortMdctSize, + st->postfilter_gain, postfilter_gain, st->postfilter_tapset, postfilter_tapset, + mode->window, overlap); + + } while (++c<CC); + st->postfilter_period_old = st->postfilter_period; + st->postfilter_gain_old = st->postfilter_gain; + st->postfilter_tapset_old = st->postfilter_tapset; + st->postfilter_period = postfilter_pitch; + st->postfilter_gain = postfilter_gain; + st->postfilter_tapset = postfilter_tapset; + if (LM!=0) + { + st->postfilter_period_old = st->postfilter_period; + st->postfilter_gain_old = st->postfilter_gain; + st->postfilter_tapset_old = st->postfilter_tapset; + } + + if (C==1) { + for (i=0;i<nbEBands;i++) + oldBandE[nbEBands+i]=oldBandE[i]; + } + + /* In case start or end were to change */ + if (!isTransient) + { + for (i=0;i<2*nbEBands;i++) + oldLogE2[i] = oldLogE[i]; + for (i=0;i<2*nbEBands;i++) + oldLogE[i] = oldBandE[i]; + for (i=0;i<2*nbEBands;i++) + backgroundLogE[i] = MIN16(backgroundLogE[i] + M*QCONST16(0.001f,DB_SHIFT), oldBandE[i]); + } else { + for (i=0;i<2*nbEBands;i++) + oldLogE[i] = MIN16(oldLogE[i], oldBandE[i]); + } + c=0; do + { + for (i=0;i<st->start;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + for (i=st->end;i<nbEBands;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + } while (++c<2); + st->rng = dec->rng; + + /* We reuse freq[] as scratch space for the de-emphasis */ + deemphasis(out_syn, pcm, N, CC, st->downsample, mode->preemph, st->preemph_memD, freq); + st->loss_count = 0; + RESTORE_STACK; + if (ec_tell(dec) > 8*len) + return OPUS_INTERNAL_ERROR; + if(ec_get_error(dec)) + st->error = 1; + return frame_size/st->downsample; +} + + +#ifdef CUSTOM_MODES + +#ifdef FIXED_POINT +int opus_custom_decode(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_int16 * OPUS_RESTRICT pcm, int frame_size) +{ + return celt_decode_with_ec(st, data, len, pcm, frame_size, NULL); +} + +#ifndef DISABLE_FLOAT_API +int opus_custom_decode_float(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, float * OPUS_RESTRICT pcm, int frame_size) +{ + int j, ret, C, N; + VARDECL(opus_int16, out); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C = st->channels; + N = frame_size; + + ALLOC(out, C*N, opus_int16); + ret=celt_decode_with_ec(st, data, len, out, frame_size, NULL); + if (ret>0) + for (j=0;j<C*ret;j++) + pcm[j]=out[j]*(1.f/32768.f); + + RESTORE_STACK; + return ret; +} +#endif /* DISABLE_FLOAT_API */ + +#else + +int opus_custom_decode_float(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, float * OPUS_RESTRICT pcm, int frame_size) +{ + return celt_decode_with_ec(st, data, len, pcm, frame_size, NULL); +} + +int opus_custom_decode(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_int16 * OPUS_RESTRICT pcm, int frame_size) +{ + int j, ret, C, N; + VARDECL(celt_sig, out); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C = st->channels; + N = frame_size; + ALLOC(out, C*N, celt_sig); + + ret=celt_decode_with_ec(st, data, len, out, frame_size, NULL); + + if (ret>0) + for (j=0;j<C*ret;j++) + pcm[j] = FLOAT2INT16 (out[j]); + + RESTORE_STACK; + return ret; +} + +#endif +#endif /* CUSTOM_MODES */ + +int opus_custom_decoder_ctl(CELTDecoder * OPUS_RESTRICT st, int request, ...) +{ + va_list ap; + + va_start(ap, request); + switch (request) + { + case CELT_SET_START_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<0 || value>=st->mode->nbEBands) + goto bad_arg; + st->start = value; + } + break; + case CELT_SET_END_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>st->mode->nbEBands) + goto bad_arg; + st->end = value; + } + break; + case CELT_SET_CHANNELS_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>2) + goto bad_arg; + st->stream_channels = value; + } + break; + case CELT_GET_AND_CLEAR_ERROR_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (value==NULL) + goto bad_arg; + *value=st->error; + st->error = 0; + } + break; + case OPUS_GET_LOOKAHEAD_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (value==NULL) + goto bad_arg; + *value = st->overlap/st->downsample; + } + break; + case OPUS_RESET_STATE: + { + int i; + opus_val16 *lpc, *oldBandE, *oldLogE, *oldLogE2; + lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+st->overlap)*st->channels); + oldBandE = lpc+st->channels*LPC_ORDER; + oldLogE = oldBandE + 2*st->mode->nbEBands; + oldLogE2 = oldLogE + 2*st->mode->nbEBands; + OPUS_CLEAR((char*)&st->DECODER_RESET_START, + opus_custom_decoder_get_size(st->mode, st->channels)- + ((char*)&st->DECODER_RESET_START - (char*)st)); + for (i=0;i<2*st->mode->nbEBands;i++) + oldLogE[i]=oldLogE2[i]=-QCONST16(28.f,DB_SHIFT); + } + break; + case OPUS_GET_PITCH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (value==NULL) + goto bad_arg; + *value = st->postfilter_period; + } + break; + case CELT_GET_MODE_REQUEST: + { + const CELTMode ** value = va_arg(ap, const CELTMode**); + if (value==0) + goto bad_arg; + *value=st->mode; + } + break; + case CELT_SET_SIGNALLING_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->signalling = value; + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + opus_uint32 * value = va_arg(ap, opus_uint32 *); + if (value==0) + goto bad_arg; + *value=st->rng; + } + break; + default: + goto bad_request; + } + va_end(ap); + return OPUS_OK; +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +bad_request: + va_end(ap); + return OPUS_UNIMPLEMENTED; +} diff --git a/src/opus-1.1/celt/celt_encoder.c b/src/opus-1.1/celt/celt_encoder.c new file mode 100644 index 00000000..ffff0775 --- /dev/null +++ b/src/opus-1.1/celt/celt_encoder.c @@ -0,0 +1,2353 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2010 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#define CELT_ENCODER_C + +#include "cpu_support.h" +#include "os_support.h" +#include "mdct.h" +#include <math.h> +#include "celt.h" +#include "pitch.h" +#include "bands.h" +#include "modes.h" +#include "entcode.h" +#include "quant_bands.h" +#include "rate.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "float_cast.h" +#include <stdarg.h> +#include "celt_lpc.h" +#include "vq.h" + + +/** Encoder state + @brief Encoder state + */ +struct OpusCustomEncoder { + const OpusCustomMode *mode; /**< Mode used by the encoder */ + int overlap; + int channels; + int stream_channels; + + int force_intra; + int clip; + int disable_pf; + int complexity; + int upsample; + int start, end; + + opus_int32 bitrate; + int vbr; + int signalling; + int constrained_vbr; /* If zero, VBR can do whatever it likes with the rate */ + int loss_rate; + int lsb_depth; + int variable_duration; + int lfe; + int arch; + + /* Everything beyond this point gets cleared on a reset */ +#define ENCODER_RESET_START rng + + opus_uint32 rng; + int spread_decision; + opus_val32 delayedIntra; + int tonal_average; + int lastCodedBands; + int hf_average; + int tapset_decision; + + int prefilter_period; + opus_val16 prefilter_gain; + int prefilter_tapset; +#ifdef RESYNTH + int prefilter_period_old; + opus_val16 prefilter_gain_old; + int prefilter_tapset_old; +#endif + int consec_transient; + AnalysisInfo analysis; + + opus_val32 preemph_memE[2]; + opus_val32 preemph_memD[2]; + + /* VBR-related parameters */ + opus_int32 vbr_reservoir; + opus_int32 vbr_drift; + opus_int32 vbr_offset; + opus_int32 vbr_count; + opus_val32 overlap_max; + opus_val16 stereo_saving; + int intensity; + opus_val16 *energy_mask; + opus_val16 spec_avg; + +#ifdef RESYNTH + /* +MAX_PERIOD/2 to make space for overlap */ + celt_sig syn_mem[2][2*MAX_PERIOD+MAX_PERIOD/2]; +#endif + + celt_sig in_mem[1]; /* Size = channels*mode->overlap */ + /* celt_sig prefilter_mem[], Size = channels*COMBFILTER_MAXPERIOD */ + /* opus_val16 oldBandE[], Size = channels*mode->nbEBands */ + /* opus_val16 oldLogE[], Size = channels*mode->nbEBands */ + /* opus_val16 oldLogE2[], Size = channels*mode->nbEBands */ +}; + +int celt_encoder_get_size(int channels) +{ + CELTMode *mode = opus_custom_mode_create(48000, 960, NULL); + return opus_custom_encoder_get_size(mode, channels); +} + +OPUS_CUSTOM_NOSTATIC int opus_custom_encoder_get_size(const CELTMode *mode, int channels) +{ + int size = sizeof(struct CELTEncoder) + + (channels*mode->overlap-1)*sizeof(celt_sig) /* celt_sig in_mem[channels*mode->overlap]; */ + + channels*COMBFILTER_MAXPERIOD*sizeof(celt_sig) /* celt_sig prefilter_mem[channels*COMBFILTER_MAXPERIOD]; */ + + 3*channels*mode->nbEBands*sizeof(opus_val16); /* opus_val16 oldBandE[channels*mode->nbEBands]; */ + /* opus_val16 oldLogE[channels*mode->nbEBands]; */ + /* opus_val16 oldLogE2[channels*mode->nbEBands]; */ + return size; +} + +#ifdef CUSTOM_MODES +CELTEncoder *opus_custom_encoder_create(const CELTMode *mode, int channels, int *error) +{ + int ret; + CELTEncoder *st = (CELTEncoder *)opus_alloc(opus_custom_encoder_get_size(mode, channels)); + /* init will handle the NULL case */ + ret = opus_custom_encoder_init(st, mode, channels); + if (ret != OPUS_OK) + { + opus_custom_encoder_destroy(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} +#endif /* CUSTOM_MODES */ + +static int opus_custom_encoder_init_arch(CELTEncoder *st, const CELTMode *mode, + int channels, int arch) +{ + if (channels < 0 || channels > 2) + return OPUS_BAD_ARG; + + if (st==NULL || mode==NULL) + return OPUS_ALLOC_FAIL; + + OPUS_CLEAR((char*)st, opus_custom_encoder_get_size(mode, channels)); + + st->mode = mode; + st->overlap = mode->overlap; + st->stream_channels = st->channels = channels; + + st->upsample = 1; + st->start = 0; + st->end = st->mode->effEBands; + st->signalling = 1; + + st->arch = arch; + + st->constrained_vbr = 1; + st->clip = 1; + + st->bitrate = OPUS_BITRATE_MAX; + st->vbr = 0; + st->force_intra = 0; + st->complexity = 5; + st->lsb_depth=24; + + opus_custom_encoder_ctl(st, OPUS_RESET_STATE); + + return OPUS_OK; +} + +#ifdef CUSTOM_MODES +int opus_custom_encoder_init(CELTEncoder *st, const CELTMode *mode, int channels) +{ + return opus_custom_encoder_init_arch(st, mode, channels, opus_select_arch()); +} +#endif + +int celt_encoder_init(CELTEncoder *st, opus_int32 sampling_rate, int channels, + int arch) +{ + int ret; + ret = opus_custom_encoder_init_arch(st, + opus_custom_mode_create(48000, 960, NULL), channels, arch); + if (ret != OPUS_OK) + return ret; + st->upsample = resampling_factor(sampling_rate); + return OPUS_OK; +} + +#ifdef CUSTOM_MODES +void opus_custom_encoder_destroy(CELTEncoder *st) +{ + opus_free(st); +} +#endif /* CUSTOM_MODES */ + + +static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int C, + opus_val16 *tf_estimate, int *tf_chan) +{ + int i; + VARDECL(opus_val16, tmp); + opus_val32 mem0,mem1; + int is_transient = 0; + opus_int32 mask_metric = 0; + int c; + opus_val16 tf_max; + int len2; + /* Table of 6*64/x, trained on real data to minimize the average error */ + static const unsigned char inv_table[128] = { + 255,255,156,110, 86, 70, 59, 51, 45, 40, 37, 33, 31, 28, 26, 25, + 23, 22, 21, 20, 19, 18, 17, 16, 16, 15, 15, 14, 13, 13, 12, 12, + 12, 12, 11, 11, 11, 10, 10, 10, 9, 9, 9, 9, 9, 9, 8, 8, + 8, 8, 8, 7, 7, 7, 7, 7, 7, 6, 6, 6, 6, 6, 6, 6, + 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, + 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 3, 3, + 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 2, + }; + SAVE_STACK; + ALLOC(tmp, len, opus_val16); + + len2=len/2; + for (c=0;c<C;c++) + { + opus_val32 mean; + opus_int32 unmask=0; + opus_val32 norm; + opus_val16 maxE; + mem0=0; + mem1=0; + /* High-pass filter: (1 - 2*z^-1 + z^-2) / (1 - z^-1 + .5*z^-2) */ + for (i=0;i<len;i++) + { + opus_val32 x,y; + x = SHR32(in[i+c*len],SIG_SHIFT); + y = ADD32(mem0, x); +#ifdef FIXED_POINT + mem0 = mem1 + y - SHL32(x,1); + mem1 = x - SHR32(y,1); +#else + mem0 = mem1 + y - 2*x; + mem1 = x - .5f*y; +#endif + tmp[i] = EXTRACT16(SHR32(y,2)); + /*printf("%f ", tmp[i]);*/ + } + /*printf("\n");*/ + /* First few samples are bad because we don't propagate the memory */ + for (i=0;i<12;i++) + tmp[i] = 0; + +#ifdef FIXED_POINT + /* Normalize tmp to max range */ + { + int shift=0; + shift = 14-celt_ilog2(1+celt_maxabs16(tmp, len)); + if (shift!=0) + { + for (i=0;i<len;i++) + tmp[i] = SHL16(tmp[i], shift); + } + } +#endif + + mean=0; + mem0=0; + /* Grouping by two to reduce complexity */ + /* Forward pass to compute the post-echo threshold*/ + for (i=0;i<len2;i++) + { + opus_val16 x2 = PSHR32(MULT16_16(tmp[2*i],tmp[2*i]) + MULT16_16(tmp[2*i+1],tmp[2*i+1]),16); + mean += x2; +#ifdef FIXED_POINT + /* FIXME: Use PSHR16() instead */ + tmp[i] = mem0 + PSHR32(x2-mem0,4); +#else + tmp[i] = mem0 + MULT16_16_P15(QCONST16(.0625f,15),x2-mem0); +#endif + mem0 = tmp[i]; + } + + mem0=0; + maxE=0; + /* Backward pass to compute the pre-echo threshold */ + for (i=len2-1;i>=0;i--) + { +#ifdef FIXED_POINT + /* FIXME: Use PSHR16() instead */ + tmp[i] = mem0 + PSHR32(tmp[i]-mem0,3); +#else + tmp[i] = mem0 + MULT16_16_P15(QCONST16(0.125f,15),tmp[i]-mem0); +#endif + mem0 = tmp[i]; + maxE = MAX16(maxE, mem0); + } + /*for (i=0;i<len2;i++)printf("%f ", tmp[i]/mean);printf("\n");*/ + + /* Compute the ratio of the "frame energy" over the harmonic mean of the energy. + This essentially corresponds to a bitrate-normalized temporal noise-to-mask + ratio */ + + /* As a compromise with the old transient detector, frame energy is the + geometric mean of the energy and half the max */ +#ifdef FIXED_POINT + /* Costs two sqrt() to avoid overflows */ + mean = MULT16_16(celt_sqrt(mean), celt_sqrt(MULT16_16(maxE,len2>>1))); +#else + mean = celt_sqrt(mean * maxE*.5*len2); +#endif + /* Inverse of the mean energy in Q15+6 */ + norm = SHL32(EXTEND32(len2),6+14)/ADD32(EPSILON,SHR32(mean,1)); + /* Compute harmonic mean discarding the unreliable boundaries + The data is smooth, so we only take 1/4th of the samples */ + unmask=0; + for (i=12;i<len2-5;i+=4) + { + int id; +#ifdef FIXED_POINT + id = IMAX(0,IMIN(127,MULT16_32_Q15(tmp[i],norm))); /* Do not round to nearest */ +#else + id = IMAX(0,IMIN(127,(int)floor(64*norm*tmp[i]))); /* Do not round to nearest */ +#endif + unmask += inv_table[id]; + } + /*printf("%d\n", unmask);*/ + /* Normalize, compensate for the 1/4th of the sample and the factor of 6 in the inverse table */ + unmask = 64*unmask*4/(6*(len2-17)); + if (unmask>mask_metric) + { + *tf_chan = c; + mask_metric = unmask; + } + } + is_transient = mask_metric>200; + + /* Arbitrary metric for VBR boost */ + tf_max = MAX16(0,celt_sqrt(27*mask_metric)-42); + /* *tf_estimate = 1 + MIN16(1, sqrt(MAX16(0, tf_max-30))/20); */ + *tf_estimate = celt_sqrt(MAX16(0, SHL32(MULT16_16(QCONST16(0.0069,14),MIN16(163,tf_max)),14)-QCONST32(0.139,28))); + /*printf("%d %f\n", tf_max, mask_metric);*/ + RESTORE_STACK; +#ifdef FUZZING + is_transient = rand()&0x1; +#endif + /*printf("%d %f %d\n", is_transient, (float)*tf_estimate, tf_max);*/ + return is_transient; +} + +/* Looks for sudden increases of energy to decide whether we need to patch + the transient decision */ +int patch_transient_decision(opus_val16 *newE, opus_val16 *oldE, int nbEBands, + int end, int C) +{ + int i, c; + opus_val32 mean_diff=0; + opus_val16 spread_old[26]; + /* Apply an aggressive (-6 dB/Bark) spreading function to the old frame to + avoid false detection caused by irrelevant bands */ + if (C==1) + { + spread_old[0] = oldE[0]; + for (i=1;i<end;i++) + spread_old[i] = MAX16(spread_old[i-1]-QCONST16(1.0f, DB_SHIFT), oldE[i]); + } else { + spread_old[0] = MAX16(oldE[0],oldE[nbEBands]); + for (i=1;i<end;i++) + spread_old[i] = MAX16(spread_old[i-1]-QCONST16(1.0f, DB_SHIFT), + MAX16(oldE[i],oldE[i+nbEBands])); + } + for (i=end-2;i>=0;i--) + spread_old[i] = MAX16(spread_old[i], spread_old[i+1]-QCONST16(1.0f, DB_SHIFT)); + /* Compute mean increase */ + c=0; do { + for (i=2;i<end-1;i++) + { + opus_val16 x1, x2; + x1 = MAX16(0, newE[i]); + x2 = MAX16(0, spread_old[i]); + mean_diff = ADD32(mean_diff, EXTEND32(MAX16(0, SUB16(x1, x2)))); + } + } while (++c<C); + mean_diff = DIV32(mean_diff, C*(end-3)); + /*printf("%f %f %d\n", mean_diff, max_diff, count);*/ + return mean_diff > QCONST16(1.f, DB_SHIFT); +} + +/** Apply window and compute the MDCT for all sub-frames and + all channels in a frame */ +static void compute_mdcts(const CELTMode *mode, int shortBlocks, celt_sig * OPUS_RESTRICT in, + celt_sig * OPUS_RESTRICT out, int C, int CC, int LM, int upsample) +{ + const int overlap = OVERLAP(mode); + int N; + int B; + int shift; + int i, b, c; + if (shortBlocks) + { + B = shortBlocks; + N = mode->shortMdctSize; + shift = mode->maxLM; + } else { + B = 1; + N = mode->shortMdctSize<<LM; + shift = mode->maxLM-LM; + } + c=0; do { + for (b=0;b<B;b++) + { + /* Interleaving the sub-frames while doing the MDCTs */ + clt_mdct_forward(&mode->mdct, in+c*(B*N+overlap)+b*N, &out[b+c*N*B], mode->window, overlap, shift, B); + } + } while (++c<CC); + if (CC==2&&C==1) + { + for (i=0;i<B*N;i++) + out[i] = ADD32(HALF32(out[i]), HALF32(out[B*N+i])); + } + if (upsample != 1) + { + c=0; do + { + int bound = B*N/upsample; + for (i=0;i<bound;i++) + out[c*B*N+i] *= upsample; + for (;i<B*N;i++) + out[c*B*N+i] = 0; + } while (++c<C); + } +} + + +void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RESTRICT inp, + int N, int CC, int upsample, const opus_val16 *coef, celt_sig *mem, int clip) +{ + int i; + opus_val16 coef0; + celt_sig m; + int Nu; + + coef0 = coef[0]; + + + Nu = N/upsample; + if (upsample!=1) + { + for (i=0;i<N;i++) + inp[i] = 0; + } + for (i=0;i<Nu;i++) + { + celt_sig x; + + x = SCALEIN(pcmp[CC*i]); +#ifndef FIXED_POINT + /* Replace NaNs with zeros */ + if (!(x==x)) + x = 0; +#endif + inp[i*upsample] = x; + } + +#ifndef FIXED_POINT + if (clip) + { + /* Clip input to avoid encoding non-portable files */ + for (i=0;i<Nu;i++) + inp[i*upsample] = MAX32(-65536.f, MIN32(65536.f,inp[i*upsample])); + } +#else + (void)clip; /* Avoids a warning about clip being unused. */ +#endif + m = *mem; +#ifdef CUSTOM_MODES + if (coef[1] != 0) + { + opus_val16 coef1 = coef[1]; + opus_val16 coef2 = coef[2]; + for (i=0;i<N;i++) + { + celt_sig x, tmp; + x = inp[i]; + /* Apply pre-emphasis */ + tmp = MULT16_16(coef2, x); + inp[i] = tmp + m; + m = MULT16_32_Q15(coef1, inp[i]) - MULT16_32_Q15(coef0, tmp); + } + } else +#endif + { + for (i=0;i<N;i++) + { + celt_sig x; + x = SHL32(inp[i], SIG_SHIFT); + /* Apply pre-emphasis */ + inp[i] = x + m; + m = - MULT16_32_Q15(coef0, x); + } + } + *mem = m; +} + + + +static opus_val32 l1_metric(const celt_norm *tmp, int N, int LM, opus_val16 bias) +{ + int i; + opus_val32 L1; + L1 = 0; + for (i=0;i<N;i++) + L1 += EXTEND32(ABS16(tmp[i])); + /* When in doubt, prefer good freq resolution */ + L1 = MAC16_32_Q15(L1, LM*bias, L1); + return L1; + +} + +static int tf_analysis(const CELTMode *m, int len, int isTransient, + int *tf_res, int lambda, celt_norm *X, int N0, int LM, + int *tf_sum, opus_val16 tf_estimate, int tf_chan) +{ + int i; + VARDECL(int, metric); + int cost0; + int cost1; + VARDECL(int, path0); + VARDECL(int, path1); + VARDECL(celt_norm, tmp); + VARDECL(celt_norm, tmp_1); + int sel; + int selcost[2]; + int tf_select=0; + opus_val16 bias; + + SAVE_STACK; + bias = MULT16_16_Q14(QCONST16(.04f,15), MAX16(-QCONST16(.25f,14), QCONST16(.5f,14)-tf_estimate)); + /*printf("%f ", bias);*/ + + ALLOC(metric, len, int); + ALLOC(tmp, (m->eBands[len]-m->eBands[len-1])<<LM, celt_norm); + ALLOC(tmp_1, (m->eBands[len]-m->eBands[len-1])<<LM, celt_norm); + ALLOC(path0, len, int); + ALLOC(path1, len, int); + + *tf_sum = 0; + for (i=0;i<len;i++) + { + int j, k, N; + int narrow; + opus_val32 L1, best_L1; + int best_level=0; + N = (m->eBands[i+1]-m->eBands[i])<<LM; + /* band is too narrow to be split down to LM=-1 */ + narrow = (m->eBands[i+1]-m->eBands[i])==1; + for (j=0;j<N;j++) + tmp[j] = X[tf_chan*N0 + j+(m->eBands[i]<<LM)]; + /* Just add the right channel if we're in stereo */ + /*if (C==2) + for (j=0;j<N;j++) + tmp[j] = ADD16(SHR16(tmp[j], 1),SHR16(X[N0+j+(m->eBands[i]<<LM)], 1));*/ + L1 = l1_metric(tmp, N, isTransient ? LM : 0, bias); + best_L1 = L1; + /* Check the -1 case for transients */ + if (isTransient && !narrow) + { + for (j=0;j<N;j++) + tmp_1[j] = tmp[j]; + haar1(tmp_1, N>>LM, 1<<LM); + L1 = l1_metric(tmp_1, N, LM+1, bias); + if (L1<best_L1) + { + best_L1 = L1; + best_level = -1; + } + } + /*printf ("%f ", L1);*/ + for (k=0;k<LM+!(isTransient||narrow);k++) + { + int B; + + if (isTransient) + B = (LM-k-1); + else + B = k+1; + + haar1(tmp, N>>k, 1<<k); + + L1 = l1_metric(tmp, N, B, bias); + + if (L1 < best_L1) + { + best_L1 = L1; + best_level = k+1; + } + } + /*printf ("%d ", isTransient ? LM-best_level : best_level);*/ + /* metric is in Q1 to be able to select the mid-point (-0.5) for narrower bands */ + if (isTransient) + metric[i] = 2*best_level; + else + metric[i] = -2*best_level; + *tf_sum += (isTransient ? LM : 0) - metric[i]/2; + /* For bands that can't be split to -1, set the metric to the half-way point to avoid + biasing the decision */ + if (narrow && (metric[i]==0 || metric[i]==-2*LM)) + metric[i]-=1; + /*printf("%d ", metric[i]);*/ + } + /*printf("\n");*/ + /* Search for the optimal tf resolution, including tf_select */ + tf_select = 0; + for (sel=0;sel<2;sel++) + { + cost0 = 0; + cost1 = isTransient ? 0 : lambda; + for (i=1;i<len;i++) + { + int curr0, curr1; + curr0 = IMIN(cost0, cost1 + lambda); + curr1 = IMIN(cost0 + lambda, cost1); + cost0 = curr0 + abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*sel+0]); + cost1 = curr1 + abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*sel+1]); + } + cost0 = IMIN(cost0, cost1); + selcost[sel]=cost0; + } + /* For now, we're conservative and only allow tf_select=1 for transients. + * If tests confirm it's useful for non-transients, we could allow it. */ + if (selcost[1]<selcost[0] && isTransient) + tf_select=1; + cost0 = 0; + cost1 = isTransient ? 0 : lambda; + /* Viterbi forward pass */ + for (i=1;i<len;i++) + { + int curr0, curr1; + int from0, from1; + + from0 = cost0; + from1 = cost1 + lambda; + if (from0 < from1) + { + curr0 = from0; + path0[i]= 0; + } else { + curr0 = from1; + path0[i]= 1; + } + + from0 = cost0 + lambda; + from1 = cost1; + if (from0 < from1) + { + curr1 = from0; + path1[i]= 0; + } else { + curr1 = from1; + path1[i]= 1; + } + cost0 = curr0 + abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*tf_select+0]); + cost1 = curr1 + abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*tf_select+1]); + } + tf_res[len-1] = cost0 < cost1 ? 0 : 1; + /* Viterbi backward pass to check the decisions */ + for (i=len-2;i>=0;i--) + { + if (tf_res[i+1] == 1) + tf_res[i] = path1[i+1]; + else + tf_res[i] = path0[i+1]; + } + /*printf("%d %f\n", *tf_sum, tf_estimate);*/ + RESTORE_STACK; +#ifdef FUZZING + tf_select = rand()&0x1; + tf_res[0] = rand()&0x1; + for (i=1;i<len;i++) + tf_res[i] = tf_res[i-1] ^ ((rand()&0xF) == 0); +#endif + return tf_select; +} + +static void tf_encode(int start, int end, int isTransient, int *tf_res, int LM, int tf_select, ec_enc *enc) +{ + int curr, i; + int tf_select_rsv; + int tf_changed; + int logp; + opus_uint32 budget; + opus_uint32 tell; + budget = enc->storage*8; + tell = ec_tell(enc); + logp = isTransient ? 2 : 4; + /* Reserve space to code the tf_select decision. */ + tf_select_rsv = LM>0 && tell+logp+1 <= budget; + budget -= tf_select_rsv; + curr = tf_changed = 0; + for (i=start;i<end;i++) + { + if (tell+logp<=budget) + { + ec_enc_bit_logp(enc, tf_res[i] ^ curr, logp); + tell = ec_tell(enc); + curr = tf_res[i]; + tf_changed |= curr; + } + else + tf_res[i] = curr; + logp = isTransient ? 4 : 5; + } + /* Only code tf_select if it would actually make a difference. */ + if (tf_select_rsv && + tf_select_table[LM][4*isTransient+0+tf_changed]!= + tf_select_table[LM][4*isTransient+2+tf_changed]) + ec_enc_bit_logp(enc, tf_select, 1); + else + tf_select = 0; + for (i=start;i<end;i++) + tf_res[i] = tf_select_table[LM][4*isTransient+2*tf_select+tf_res[i]]; + /*for(i=0;i<end;i++)printf("%d ", isTransient ? tf_res[i] : LM+tf_res[i]);printf("\n");*/ +} + + +static int alloc_trim_analysis(const CELTMode *m, const celt_norm *X, + const opus_val16 *bandLogE, int end, int LM, int C, int N0, + AnalysisInfo *analysis, opus_val16 *stereo_saving, opus_val16 tf_estimate, + int intensity, opus_val16 surround_trim) +{ + int i; + opus_val32 diff=0; + int c; + int trim_index = 5; + opus_val16 trim = QCONST16(5.f, 8); + opus_val16 logXC, logXC2; + if (C==2) + { + opus_val16 sum = 0; /* Q10 */ + opus_val16 minXC; /* Q10 */ + /* Compute inter-channel correlation for low frequencies */ + for (i=0;i<8;i++) + { + int j; + opus_val32 partial = 0; + for (j=m->eBands[i]<<LM;j<m->eBands[i+1]<<LM;j++) + partial = MAC16_16(partial, X[j], X[N0+j]); + sum = ADD16(sum, EXTRACT16(SHR32(partial, 18))); + } + sum = MULT16_16_Q15(QCONST16(1.f/8, 15), sum); + sum = MIN16(QCONST16(1.f, 10), ABS16(sum)); + minXC = sum; + for (i=8;i<intensity;i++) + { + int j; + opus_val32 partial = 0; + for (j=m->eBands[i]<<LM;j<m->eBands[i+1]<<LM;j++) + partial = MAC16_16(partial, X[j], X[N0+j]); + minXC = MIN16(minXC, ABS16(EXTRACT16(SHR32(partial, 18)))); + } + minXC = MIN16(QCONST16(1.f, 10), ABS16(minXC)); + /*printf ("%f\n", sum);*/ + if (sum > QCONST16(.995f,10)) + trim_index-=4; + else if (sum > QCONST16(.92f,10)) + trim_index-=3; + else if (sum > QCONST16(.85f,10)) + trim_index-=2; + else if (sum > QCONST16(.8f,10)) + trim_index-=1; + /* mid-side savings estimations based on the LF average*/ + logXC = celt_log2(QCONST32(1.001f, 20)-MULT16_16(sum, sum)); + /* mid-side savings estimations based on min correlation */ + logXC2 = MAX16(HALF16(logXC), celt_log2(QCONST32(1.001f, 20)-MULT16_16(minXC, minXC))); +#ifdef FIXED_POINT + /* Compensate for Q20 vs Q14 input and convert output to Q8 */ + logXC = PSHR32(logXC-QCONST16(6.f, DB_SHIFT),DB_SHIFT-8); + logXC2 = PSHR32(logXC2-QCONST16(6.f, DB_SHIFT),DB_SHIFT-8); +#endif + + trim += MAX16(-QCONST16(4.f, 8), MULT16_16_Q15(QCONST16(.75f,15),logXC)); + *stereo_saving = MIN16(*stereo_saving + QCONST16(0.25f, 8), -HALF16(logXC2)); + } + + /* Estimate spectral tilt */ + c=0; do { + for (i=0;i<end-1;i++) + { + diff += bandLogE[i+c*m->nbEBands]*(opus_int32)(2+2*i-end); + } + } while (++c<C); + diff /= C*(end-1); + /*printf("%f\n", diff);*/ + if (diff > QCONST16(2.f, DB_SHIFT)) + trim_index--; + if (diff > QCONST16(8.f, DB_SHIFT)) + trim_index--; + if (diff < -QCONST16(4.f, DB_SHIFT)) + trim_index++; + if (diff < -QCONST16(10.f, DB_SHIFT)) + trim_index++; + trim -= MAX16(-QCONST16(2.f, 8), MIN16(QCONST16(2.f, 8), SHR16(diff+QCONST16(1.f, DB_SHIFT),DB_SHIFT-8)/6 )); + trim -= SHR16(surround_trim, DB_SHIFT-8); + trim -= 2*SHR16(tf_estimate, 14-8); +#ifndef DISABLE_FLOAT_API + if (analysis->valid) + { + trim -= MAX16(-QCONST16(2.f, 8), MIN16(QCONST16(2.f, 8), + (opus_val16)(QCONST16(2.f, 8)*(analysis->tonality_slope+.05f)))); + } +#endif + +#ifdef FIXED_POINT + trim_index = PSHR32(trim, 8); +#else + trim_index = (int)floor(.5f+trim); +#endif + if (trim_index<0) + trim_index = 0; + if (trim_index>10) + trim_index = 10; + /*printf("%d\n", trim_index);*/ +#ifdef FUZZING + trim_index = rand()%11; +#endif + return trim_index; +} + +static int stereo_analysis(const CELTMode *m, const celt_norm *X, + int LM, int N0) +{ + int i; + int thetas; + opus_val32 sumLR = EPSILON, sumMS = EPSILON; + + /* Use the L1 norm to model the entropy of the L/R signal vs the M/S signal */ + for (i=0;i<13;i++) + { + int j; + for (j=m->eBands[i]<<LM;j<m->eBands[i+1]<<LM;j++) + { + opus_val32 L, R, M, S; + /* We cast to 32-bit first because of the -32768 case */ + L = EXTEND32(X[j]); + R = EXTEND32(X[N0+j]); + M = ADD32(L, R); + S = SUB32(L, R); + sumLR = ADD32(sumLR, ADD32(ABS32(L), ABS32(R))); + sumMS = ADD32(sumMS, ADD32(ABS32(M), ABS32(S))); + } + } + sumMS = MULT16_32_Q15(QCONST16(0.707107f, 15), sumMS); + thetas = 13; + /* We don't need thetas for lower bands with LM<=1 */ + if (LM<=1) + thetas -= 8; + return MULT16_32_Q15((m->eBands[13]<<(LM+1))+thetas, sumMS) + > MULT16_32_Q15(m->eBands[13]<<(LM+1), sumLR); +} + +static opus_val16 dynalloc_analysis(const opus_val16 *bandLogE, const opus_val16 *bandLogE2, + int nbEBands, int start, int end, int C, int *offsets, int lsb_depth, const opus_int16 *logN, + int isTransient, int vbr, int constrained_vbr, const opus_int16 *eBands, int LM, + int effectiveBytes, opus_int32 *tot_boost_, int lfe, opus_val16 *surround_dynalloc) +{ + int i, c; + opus_int32 tot_boost=0; + opus_val16 maxDepth; + VARDECL(opus_val16, follower); + VARDECL(opus_val16, noise_floor); + SAVE_STACK; + ALLOC(follower, C*nbEBands, opus_val16); + ALLOC(noise_floor, C*nbEBands, opus_val16); + for (i=0;i<nbEBands;i++) + offsets[i] = 0; + /* Dynamic allocation code */ + maxDepth=-QCONST16(31.9f, DB_SHIFT); + for (i=0;i<end;i++) + { + /* Noise floor must take into account eMeans, the depth, the width of the bands + and the preemphasis filter (approx. square of bark band ID) */ + noise_floor[i] = MULT16_16(QCONST16(0.0625f, DB_SHIFT),logN[i]) + +QCONST16(.5f,DB_SHIFT)+SHL16(9-lsb_depth,DB_SHIFT)-SHL16(eMeans[i],6) + +MULT16_16(QCONST16(.0062,DB_SHIFT),(i+5)*(i+5)); + } + c=0;do + { + for (i=0;i<end;i++) + maxDepth = MAX16(maxDepth, bandLogE[c*nbEBands+i]-noise_floor[i]); + } while (++c<C); + /* Make sure that dynamic allocation can't make us bust the budget */ + if (effectiveBytes > 50 && LM>=1 && !lfe) + { + int last=0; + c=0;do + { + follower[c*nbEBands] = bandLogE2[c*nbEBands]; + for (i=1;i<end;i++) + { + /* The last band to be at least 3 dB higher than the previous one + is the last we'll consider. Otherwise, we run into problems on + bandlimited signals. */ + if (bandLogE2[c*nbEBands+i] > bandLogE2[c*nbEBands+i-1]+QCONST16(.5f,DB_SHIFT)) + last=i; + follower[c*nbEBands+i] = MIN16(follower[c*nbEBands+i-1]+QCONST16(1.5f,DB_SHIFT), bandLogE2[c*nbEBands+i]); + } + for (i=last-1;i>=0;i--) + follower[c*nbEBands+i] = MIN16(follower[c*nbEBands+i], MIN16(follower[c*nbEBands+i+1]+QCONST16(2.f,DB_SHIFT), bandLogE2[c*nbEBands+i])); + for (i=0;i<end;i++) + follower[c*nbEBands+i] = MAX16(follower[c*nbEBands+i], noise_floor[i]); + } while (++c<C); + if (C==2) + { + for (i=start;i<end;i++) + { + /* Consider 24 dB "cross-talk" */ + follower[nbEBands+i] = MAX16(follower[nbEBands+i], follower[ i]-QCONST16(4.f,DB_SHIFT)); + follower[ i] = MAX16(follower[ i], follower[nbEBands+i]-QCONST16(4.f,DB_SHIFT)); + follower[i] = HALF16(MAX16(0, bandLogE[i]-follower[i]) + MAX16(0, bandLogE[nbEBands+i]-follower[nbEBands+i])); + } + } else { + for (i=start;i<end;i++) + { + follower[i] = MAX16(0, bandLogE[i]-follower[i]); + } + } + for (i=start;i<end;i++) + follower[i] = MAX16(follower[i], surround_dynalloc[i]); + /* For non-transient CBR/CVBR frames, halve the dynalloc contribution */ + if ((!vbr || constrained_vbr)&&!isTransient) + { + for (i=start;i<end;i++) + follower[i] = HALF16(follower[i]); + } + for (i=start;i<end;i++) + { + int width; + int boost; + int boost_bits; + + if (i<8) + follower[i] *= 2; + if (i>=12) + follower[i] = HALF16(follower[i]); + follower[i] = MIN16(follower[i], QCONST16(4, DB_SHIFT)); + + width = C*(eBands[i+1]-eBands[i])<<LM; + if (width<6) + { + boost = (int)SHR32(EXTEND32(follower[i]),DB_SHIFT); + boost_bits = boost*width<<BITRES; + } else if (width > 48) { + boost = (int)SHR32(EXTEND32(follower[i])*8,DB_SHIFT); + boost_bits = (boost*width<<BITRES)/8; + } else { + boost = (int)SHR32(EXTEND32(follower[i])*width/6,DB_SHIFT); + boost_bits = boost*6<<BITRES; + } + /* For CBR and non-transient CVBR frames, limit dynalloc to 1/4 of the bits */ + if ((!vbr || (constrained_vbr&&!isTransient)) + && (tot_boost+boost_bits)>>BITRES>>3 > effectiveBytes/4) + { + opus_int32 cap = ((effectiveBytes/4)<<BITRES<<3); + offsets[i] = cap-tot_boost; + tot_boost = cap; + break; + } else { + offsets[i] = boost; + tot_boost += boost_bits; + } + } + } + *tot_boost_ = tot_boost; + RESTORE_STACK; + return maxDepth; +} + + +static int run_prefilter(CELTEncoder *st, celt_sig *in, celt_sig *prefilter_mem, int CC, int N, + int prefilter_tapset, int *pitch, opus_val16 *gain, int *qgain, int enabled, int nbAvailableBytes) +{ + int c; + VARDECL(celt_sig, _pre); + celt_sig *pre[2]; + const CELTMode *mode; + int pitch_index; + opus_val16 gain1; + opus_val16 pf_threshold; + int pf_on; + int qg; + SAVE_STACK; + + mode = st->mode; + ALLOC(_pre, CC*(N+COMBFILTER_MAXPERIOD), celt_sig); + + pre[0] = _pre; + pre[1] = _pre + (N+COMBFILTER_MAXPERIOD); + + + c=0; do { + OPUS_COPY(pre[c], prefilter_mem+c*COMBFILTER_MAXPERIOD, COMBFILTER_MAXPERIOD); + OPUS_COPY(pre[c]+COMBFILTER_MAXPERIOD, in+c*(N+st->overlap)+st->overlap, N); + } while (++c<CC); + + if (enabled) + { + VARDECL(opus_val16, pitch_buf); + ALLOC(pitch_buf, (COMBFILTER_MAXPERIOD+N)>>1, opus_val16); + + pitch_downsample(pre, pitch_buf, COMBFILTER_MAXPERIOD+N, CC, st->arch); + /* Don't search for the fir last 1.5 octave of the range because + there's too many false-positives due to short-term correlation */ + pitch_search(pitch_buf+(COMBFILTER_MAXPERIOD>>1), pitch_buf, N, + COMBFILTER_MAXPERIOD-3*COMBFILTER_MINPERIOD, &pitch_index, + st->arch); + pitch_index = COMBFILTER_MAXPERIOD-pitch_index; + + gain1 = remove_doubling(pitch_buf, COMBFILTER_MAXPERIOD, COMBFILTER_MINPERIOD, + N, &pitch_index, st->prefilter_period, st->prefilter_gain); + if (pitch_index > COMBFILTER_MAXPERIOD-2) + pitch_index = COMBFILTER_MAXPERIOD-2; + gain1 = MULT16_16_Q15(QCONST16(.7f,15),gain1); + /*printf("%d %d %f %f\n", pitch_change, pitch_index, gain1, st->analysis.tonality);*/ + if (st->loss_rate>2) + gain1 = HALF32(gain1); + if (st->loss_rate>4) + gain1 = HALF32(gain1); + if (st->loss_rate>8) + gain1 = 0; + } else { + gain1 = 0; + pitch_index = COMBFILTER_MINPERIOD; + } + + /* Gain threshold for enabling the prefilter/postfilter */ + pf_threshold = QCONST16(.2f,15); + + /* Adjusting the threshold based on rate and continuity */ + if (abs(pitch_index-st->prefilter_period)*10>pitch_index) + pf_threshold += QCONST16(.2f,15); + if (nbAvailableBytes<25) + pf_threshold += QCONST16(.1f,15); + if (nbAvailableBytes<35) + pf_threshold += QCONST16(.1f,15); + if (st->prefilter_gain > QCONST16(.4f,15)) + pf_threshold -= QCONST16(.1f,15); + if (st->prefilter_gain > QCONST16(.55f,15)) + pf_threshold -= QCONST16(.1f,15); + + /* Hard threshold at 0.2 */ + pf_threshold = MAX16(pf_threshold, QCONST16(.2f,15)); + if (gain1<pf_threshold) + { + gain1 = 0; + pf_on = 0; + qg = 0; + } else { + /*This block is not gated by a total bits check only because + of the nbAvailableBytes check above.*/ + if (ABS16(gain1-st->prefilter_gain)<QCONST16(.1f,15)) + gain1=st->prefilter_gain; + +#ifdef FIXED_POINT + qg = ((gain1+1536)>>10)/3-1; +#else + qg = (int)floor(.5f+gain1*32/3)-1; +#endif + qg = IMAX(0, IMIN(7, qg)); + gain1 = QCONST16(0.09375f,15)*(qg+1); + pf_on = 1; + } + /*printf("%d %f\n", pitch_index, gain1);*/ + + c=0; do { + int offset = mode->shortMdctSize-st->overlap; + st->prefilter_period=IMAX(st->prefilter_period, COMBFILTER_MINPERIOD); + OPUS_COPY(in+c*(N+st->overlap), st->in_mem+c*(st->overlap), st->overlap); + if (offset) + comb_filter(in+c*(N+st->overlap)+st->overlap, pre[c]+COMBFILTER_MAXPERIOD, + st->prefilter_period, st->prefilter_period, offset, -st->prefilter_gain, -st->prefilter_gain, + st->prefilter_tapset, st->prefilter_tapset, NULL, 0); + + comb_filter(in+c*(N+st->overlap)+st->overlap+offset, pre[c]+COMBFILTER_MAXPERIOD+offset, + st->prefilter_period, pitch_index, N-offset, -st->prefilter_gain, -gain1, + st->prefilter_tapset, prefilter_tapset, mode->window, st->overlap); + OPUS_COPY(st->in_mem+c*(st->overlap), in+c*(N+st->overlap)+N, st->overlap); + + if (N>COMBFILTER_MAXPERIOD) + { + OPUS_MOVE(prefilter_mem+c*COMBFILTER_MAXPERIOD, pre[c]+N, COMBFILTER_MAXPERIOD); + } else { + OPUS_MOVE(prefilter_mem+c*COMBFILTER_MAXPERIOD, prefilter_mem+c*COMBFILTER_MAXPERIOD+N, COMBFILTER_MAXPERIOD-N); + OPUS_MOVE(prefilter_mem+c*COMBFILTER_MAXPERIOD+COMBFILTER_MAXPERIOD-N, pre[c]+COMBFILTER_MAXPERIOD, N); + } + } while (++c<CC); + + RESTORE_STACK; + *gain = gain1; + *pitch = pitch_index; + *qgain = qg; + return pf_on; +} + +static int compute_vbr(const CELTMode *mode, AnalysisInfo *analysis, opus_int32 base_target, + int LM, opus_int32 bitrate, int lastCodedBands, int C, int intensity, + int constrained_vbr, opus_val16 stereo_saving, int tot_boost, + opus_val16 tf_estimate, int pitch_change, opus_val16 maxDepth, + int variable_duration, int lfe, int has_surround_mask, opus_val16 surround_masking, + opus_val16 temporal_vbr) +{ + /* The target rate in 8th bits per frame */ + opus_int32 target; + int coded_bins; + int coded_bands; + opus_val16 tf_calibration; + int nbEBands; + const opus_int16 *eBands; + + nbEBands = mode->nbEBands; + eBands = mode->eBands; + + coded_bands = lastCodedBands ? lastCodedBands : nbEBands; + coded_bins = eBands[coded_bands]<<LM; + if (C==2) + coded_bins += eBands[IMIN(intensity, coded_bands)]<<LM; + + target = base_target; + + /*printf("%f %f %f %f %d %d ", st->analysis.activity, st->analysis.tonality, tf_estimate, st->stereo_saving, tot_boost, coded_bands);*/ +#ifndef DISABLE_FLOAT_API + if (analysis->valid && analysis->activity<.4) + target -= (opus_int32)((coded_bins<<BITRES)*(.4f-analysis->activity)); +#endif + /* Stereo savings */ + if (C==2) + { + int coded_stereo_bands; + int coded_stereo_dof; + opus_val16 max_frac; + coded_stereo_bands = IMIN(intensity, coded_bands); + coded_stereo_dof = (eBands[coded_stereo_bands]<<LM)-coded_stereo_bands; + /* Maximum fraction of the bits we can save if the signal is mono. */ + max_frac = DIV32_16(MULT16_16(QCONST16(0.8f, 15), coded_stereo_dof), coded_bins); + stereo_saving = MIN16(stereo_saving, QCONST16(1.f, 8)); + /*printf("%d %d %d ", coded_stereo_dof, coded_bins, tot_boost);*/ + target -= (opus_int32)MIN32(MULT16_32_Q15(max_frac,target), + SHR32(MULT16_16(stereo_saving-QCONST16(0.1f,8),(coded_stereo_dof<<BITRES)),8)); + } + /* Boost the rate according to dynalloc (minus the dynalloc average for calibration). */ + target += tot_boost-(16<<LM); + /* Apply transient boost, compensating for average boost. */ + tf_calibration = variable_duration==OPUS_FRAMESIZE_VARIABLE ? + QCONST16(0.02f,14) : QCONST16(0.04f,14); + target += (opus_int32)SHL32(MULT16_32_Q15(tf_estimate-tf_calibration, target),1); + +#ifndef DISABLE_FLOAT_API + /* Apply tonality boost */ + if (analysis->valid && !lfe) + { + opus_int32 tonal_target; + float tonal; + + /* Tonality boost (compensating for the average). */ + tonal = MAX16(0.f,analysis->tonality-.15f)-0.09f; + tonal_target = target + (opus_int32)((coded_bins<<BITRES)*1.2f*tonal); + if (pitch_change) + tonal_target += (opus_int32)((coded_bins<<BITRES)*.8f); + /*printf("%f %f ", analysis->tonality, tonal);*/ + target = tonal_target; + } +#endif + + if (has_surround_mask&&!lfe) + { + opus_int32 surround_target = target + (opus_int32)SHR32(MULT16_16(surround_masking,coded_bins<<BITRES), DB_SHIFT); + /*printf("%f %d %d %d %d %d %d ", surround_masking, coded_bins, st->end, st->intensity, surround_target, target, st->bitrate);*/ + target = IMAX(target/4, surround_target); + } + + { + opus_int32 floor_depth; + int bins; + bins = eBands[nbEBands-2]<<LM; + /*floor_depth = SHR32(MULT16_16((C*bins<<BITRES),celt_log2(SHL32(MAX16(1,sample_max),13))), DB_SHIFT);*/ + floor_depth = (opus_int32)SHR32(MULT16_16((C*bins<<BITRES),maxDepth), DB_SHIFT); + floor_depth = IMAX(floor_depth, target>>2); + target = IMIN(target, floor_depth); + /*printf("%f %d\n", maxDepth, floor_depth);*/ + } + + if ((!has_surround_mask||lfe) && (constrained_vbr || bitrate<64000)) + { + opus_val16 rate_factor; +#ifdef FIXED_POINT + rate_factor = MAX16(0,(bitrate-32000)); +#else + rate_factor = MAX16(0,(1.f/32768)*(bitrate-32000)); +#endif + if (constrained_vbr) + rate_factor = MIN16(rate_factor, QCONST16(0.67f, 15)); + target = base_target + (opus_int32)MULT16_32_Q15(rate_factor, target-base_target); + + } + + if (!has_surround_mask && tf_estimate < QCONST16(.2f, 14)) + { + opus_val16 amount; + opus_val16 tvbr_factor; + amount = MULT16_16_Q15(QCONST16(.0000031f, 30), IMAX(0, IMIN(32000, 96000-bitrate))); + tvbr_factor = SHR32(MULT16_16(temporal_vbr, amount), DB_SHIFT); + target += (opus_int32)MULT16_32_Q15(tvbr_factor, target); + } + + /* Don't allow more than doubling the rate */ + target = IMIN(2*base_target, target); + + return target; +} + +int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes, ec_enc *enc) +{ + int i, c, N; + opus_int32 bits; + ec_enc _enc; + VARDECL(celt_sig, in); + VARDECL(celt_sig, freq); + VARDECL(celt_norm, X); + VARDECL(celt_ener, bandE); + VARDECL(opus_val16, bandLogE); + VARDECL(opus_val16, bandLogE2); + VARDECL(int, fine_quant); + VARDECL(opus_val16, error); + VARDECL(int, pulses); + VARDECL(int, cap); + VARDECL(int, offsets); + VARDECL(int, fine_priority); + VARDECL(int, tf_res); + VARDECL(unsigned char, collapse_masks); + celt_sig *prefilter_mem; + opus_val16 *oldBandE, *oldLogE, *oldLogE2; + int shortBlocks=0; + int isTransient=0; + const int CC = st->channels; + const int C = st->stream_channels; + int LM, M; + int tf_select; + int nbFilledBytes, nbAvailableBytes; + int effEnd; + int codedBands; + int tf_sum; + int alloc_trim; + int pitch_index=COMBFILTER_MINPERIOD; + opus_val16 gain1 = 0; + int dual_stereo=0; + int effectiveBytes; + int dynalloc_logp; + opus_int32 vbr_rate; + opus_int32 total_bits; + opus_int32 total_boost; + opus_int32 balance; + opus_int32 tell; + int prefilter_tapset=0; + int pf_on; + int anti_collapse_rsv; + int anti_collapse_on=0; + int silence=0; + int tf_chan = 0; + opus_val16 tf_estimate; + int pitch_change=0; + opus_int32 tot_boost; + opus_val32 sample_max; + opus_val16 maxDepth; + const OpusCustomMode *mode; + int nbEBands; + int overlap; + const opus_int16 *eBands; + int secondMdct; + int signalBandwidth; + int transient_got_disabled=0; + opus_val16 surround_masking=0; + opus_val16 temporal_vbr=0; + opus_val16 surround_trim = 0; + opus_int32 equiv_rate = 510000; + VARDECL(opus_val16, surround_dynalloc); + ALLOC_STACK; + + mode = st->mode; + nbEBands = mode->nbEBands; + overlap = mode->overlap; + eBands = mode->eBands; + tf_estimate = 0; + if (nbCompressedBytes<2 || pcm==NULL) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + + frame_size *= st->upsample; + for (LM=0;LM<=mode->maxLM;LM++) + if (mode->shortMdctSize<<LM==frame_size) + break; + if (LM>mode->maxLM) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + M=1<<LM; + N = M*mode->shortMdctSize; + + prefilter_mem = st->in_mem+CC*(st->overlap); + oldBandE = (opus_val16*)(st->in_mem+CC*(st->overlap+COMBFILTER_MAXPERIOD)); + oldLogE = oldBandE + CC*nbEBands; + oldLogE2 = oldLogE + CC*nbEBands; + + if (enc==NULL) + { + tell=1; + nbFilledBytes=0; + } else { + tell=ec_tell(enc); + nbFilledBytes=(tell+4)>>3; + } + +#ifdef CUSTOM_MODES + if (st->signalling && enc==NULL) + { + int tmp = (mode->effEBands-st->end)>>1; + st->end = IMAX(1, mode->effEBands-tmp); + compressed[0] = tmp<<5; + compressed[0] |= LM<<3; + compressed[0] |= (C==2)<<2; + /* Convert "standard mode" to Opus header */ + if (mode->Fs==48000 && mode->shortMdctSize==120) + { + int c0 = toOpus(compressed[0]); + if (c0<0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + compressed[0] = c0; + } + compressed++; + nbCompressedBytes--; + } +#else + celt_assert(st->signalling==0); +#endif + + /* Can't produce more than 1275 output bytes */ + nbCompressedBytes = IMIN(nbCompressedBytes,1275); + nbAvailableBytes = nbCompressedBytes - nbFilledBytes; + + if (st->vbr && st->bitrate!=OPUS_BITRATE_MAX) + { + opus_int32 den=mode->Fs>>BITRES; + vbr_rate=(st->bitrate*frame_size+(den>>1))/den; +#ifdef CUSTOM_MODES + if (st->signalling) + vbr_rate -= 8<<BITRES; +#endif + effectiveBytes = vbr_rate>>(3+BITRES); + } else { + opus_int32 tmp; + vbr_rate = 0; + tmp = st->bitrate*frame_size; + if (tell>1) + tmp += tell; + if (st->bitrate!=OPUS_BITRATE_MAX) + nbCompressedBytes = IMAX(2, IMIN(nbCompressedBytes, + (tmp+4*mode->Fs)/(8*mode->Fs)-!!st->signalling)); + effectiveBytes = nbCompressedBytes; + } + if (st->bitrate != OPUS_BITRATE_MAX) + equiv_rate = st->bitrate - (40*C+20)*((400>>LM) - 50); + + if (enc==NULL) + { + ec_enc_init(&_enc, compressed, nbCompressedBytes); + enc = &_enc; + } + + if (vbr_rate>0) + { + /* Computes the max bit-rate allowed in VBR mode to avoid violating the + target rate and buffering. + We must do this up front so that bust-prevention logic triggers + correctly if we don't have enough bits. */ + if (st->constrained_vbr) + { + opus_int32 vbr_bound; + opus_int32 max_allowed; + /* We could use any multiple of vbr_rate as bound (depending on the + delay). + This is clamped to ensure we use at least two bytes if the encoder + was entirely empty, but to allow 0 in hybrid mode. */ + vbr_bound = vbr_rate; + max_allowed = IMIN(IMAX(tell==1?2:0, + (vbr_rate+vbr_bound-st->vbr_reservoir)>>(BITRES+3)), + nbAvailableBytes); + if(max_allowed < nbAvailableBytes) + { + nbCompressedBytes = nbFilledBytes+max_allowed; + nbAvailableBytes = max_allowed; + ec_enc_shrink(enc, nbCompressedBytes); + } + } + } + total_bits = nbCompressedBytes*8; + + effEnd = st->end; + if (effEnd > mode->effEBands) + effEnd = mode->effEBands; + + ALLOC(in, CC*(N+st->overlap), celt_sig); + + sample_max=MAX32(st->overlap_max, celt_maxabs16(pcm, C*(N-overlap)/st->upsample)); + st->overlap_max=celt_maxabs16(pcm+C*(N-overlap)/st->upsample, C*overlap/st->upsample); + sample_max=MAX32(sample_max, st->overlap_max); +#ifdef FIXED_POINT + silence = (sample_max==0); +#else + silence = (sample_max <= (opus_val16)1/(1<<st->lsb_depth)); +#endif +#ifdef FUZZING + if ((rand()&0x3F)==0) + silence = 1; +#endif + if (tell==1) + ec_enc_bit_logp(enc, silence, 15); + else + silence=0; + if (silence) + { + /*In VBR mode there is no need to send more than the minimum. */ + if (vbr_rate>0) + { + effectiveBytes=nbCompressedBytes=IMIN(nbCompressedBytes, nbFilledBytes+2); + total_bits=nbCompressedBytes*8; + nbAvailableBytes=2; + ec_enc_shrink(enc, nbCompressedBytes); + } + /* Pretend we've filled all the remaining bits with zeros + (that's what the initialiser did anyway) */ + tell = nbCompressedBytes*8; + enc->nbits_total+=tell-ec_tell(enc); + } + c=0; do { + celt_preemphasis(pcm+c, in+c*(N+st->overlap)+st->overlap, N, CC, st->upsample, + mode->preemph, st->preemph_memE+c, st->clip); + } while (++c<CC); + + + + /* Find pitch period and gain */ + { + int enabled; + int qg; + enabled = ((st->lfe&&nbAvailableBytes>3) || nbAvailableBytes>12*C) && st->start==0 && !silence && !st->disable_pf + && st->complexity >= 5 && !(st->consec_transient && LM!=3 && st->variable_duration==OPUS_FRAMESIZE_VARIABLE); + + prefilter_tapset = st->tapset_decision; + pf_on = run_prefilter(st, in, prefilter_mem, CC, N, prefilter_tapset, &pitch_index, &gain1, &qg, enabled, nbAvailableBytes); + if ((gain1 > QCONST16(.4f,15) || st->prefilter_gain > QCONST16(.4f,15)) && (!st->analysis.valid || st->analysis.tonality > .3) + && (pitch_index > 1.26*st->prefilter_period || pitch_index < .79*st->prefilter_period)) + pitch_change = 1; + if (pf_on==0) + { + if(st->start==0 && tell+16<=total_bits) + ec_enc_bit_logp(enc, 0, 1); + } else { + /*This block is not gated by a total bits check only because + of the nbAvailableBytes check above.*/ + int octave; + ec_enc_bit_logp(enc, 1, 1); + pitch_index += 1; + octave = EC_ILOG(pitch_index)-5; + ec_enc_uint(enc, octave, 6); + ec_enc_bits(enc, pitch_index-(16<<octave), 4+octave); + pitch_index -= 1; + ec_enc_bits(enc, qg, 3); + ec_enc_icdf(enc, prefilter_tapset, tapset_icdf, 2); + } + } + + isTransient = 0; + shortBlocks = 0; + if (st->complexity >= 1 && !st->lfe) + { + isTransient = transient_analysis(in, N+st->overlap, CC, + &tf_estimate, &tf_chan); + } + if (LM>0 && ec_tell(enc)+3<=total_bits) + { + if (isTransient) + shortBlocks = M; + } else { + isTransient = 0; + transient_got_disabled=1; + } + + ALLOC(freq, CC*N, celt_sig); /**< Interleaved signal MDCTs */ + ALLOC(bandE,nbEBands*CC, celt_ener); + ALLOC(bandLogE,nbEBands*CC, opus_val16); + + secondMdct = shortBlocks && st->complexity>=8; + ALLOC(bandLogE2, C*nbEBands, opus_val16); + if (secondMdct) + { + compute_mdcts(mode, 0, in, freq, C, CC, LM, st->upsample); + compute_band_energies(mode, freq, bandE, effEnd, C, M); + amp2Log2(mode, effEnd, st->end, bandE, bandLogE2, C); + for (i=0;i<C*nbEBands;i++) + bandLogE2[i] += HALF16(SHL16(LM, DB_SHIFT)); + } + + compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample); + if (CC==2&&C==1) + tf_chan = 0; + compute_band_energies(mode, freq, bandE, effEnd, C, M); + + if (st->lfe) + { + for (i=2;i<st->end;i++) + { + bandE[i] = IMIN(bandE[i], MULT16_32_Q15(QCONST16(1e-4f,15),bandE[0])); + bandE[i] = MAX32(bandE[i], EPSILON); + } + } + amp2Log2(mode, effEnd, st->end, bandE, bandLogE, C); + + ALLOC(surround_dynalloc, C*nbEBands, opus_val16); + for(i=0;i<st->end;i++) + surround_dynalloc[i] = 0; + /* This computes how much masking takes place between surround channels */ + if (st->start==0&&st->energy_mask&&!st->lfe) + { + int mask_end; + int midband; + int count_dynalloc; + opus_val32 mask_avg=0; + opus_val32 diff=0; + int count=0; + mask_end = IMAX(2,st->lastCodedBands); + for (c=0;c<C;c++) + { + for(i=0;i<mask_end;i++) + { + opus_val16 mask; + mask = MAX16(MIN16(st->energy_mask[nbEBands*c+i], + QCONST16(.25f, DB_SHIFT)), -QCONST16(2.0f, DB_SHIFT)); + if (mask > 0) + mask = HALF16(mask); + mask_avg += MULT16_16(mask, eBands[i+1]-eBands[i]); + count += eBands[i+1]-eBands[i]; + diff += MULT16_16(mask, 1+2*i-mask_end); + } + } + mask_avg = DIV32_16(mask_avg,count); + mask_avg += QCONST16(.2f, DB_SHIFT); + diff = diff*6/(C*(mask_end-1)*(mask_end+1)*mask_end); + /* Again, being conservative */ + diff = HALF32(diff); + diff = MAX32(MIN32(diff, QCONST32(.031f, DB_SHIFT)), -QCONST32(.031f, DB_SHIFT)); + /* Find the band that's in the middle of the coded spectrum */ + for (midband=0;eBands[midband+1] < eBands[mask_end]/2;midband++); + count_dynalloc=0; + for(i=0;i<mask_end;i++) + { + opus_val32 lin; + opus_val16 unmask; + lin = mask_avg + diff*(i-midband); + if (C==2) + unmask = MAX16(st->energy_mask[i], st->energy_mask[nbEBands+i]); + else + unmask = st->energy_mask[i]; + unmask = MIN16(unmask, QCONST16(.0f, DB_SHIFT)); + unmask -= lin; + if (unmask > QCONST16(.25f, DB_SHIFT)) + { + surround_dynalloc[i] = unmask - QCONST16(.25f, DB_SHIFT); + count_dynalloc++; + } + } + if (count_dynalloc>=3) + { + /* If we need dynalloc in many bands, it's probably because our + initial masking rate was too low. */ + mask_avg += QCONST16(.25f, DB_SHIFT); + if (mask_avg>0) + { + /* Something went really wrong in the original calculations, + disabling masking. */ + mask_avg = 0; + diff = 0; + for(i=0;i<mask_end;i++) + surround_dynalloc[i] = 0; + } else { + for(i=0;i<mask_end;i++) + surround_dynalloc[i] = MAX16(0, surround_dynalloc[i]-QCONST16(.25f, DB_SHIFT)); + } + } + mask_avg += QCONST16(.2f, DB_SHIFT); + /* Convert to 1/64th units used for the trim */ + surround_trim = 64*diff; + /*printf("%d %d ", mask_avg, surround_trim);*/ + surround_masking = mask_avg; + } + /* Temporal VBR (but not for LFE) */ + if (!st->lfe) + { + opus_val16 follow=-QCONST16(10.0f,DB_SHIFT); + opus_val32 frame_avg=0; + opus_val16 offset = shortBlocks?HALF16(SHL16(LM, DB_SHIFT)):0; + for(i=st->start;i<st->end;i++) + { + follow = MAX16(follow-QCONST16(1.f, DB_SHIFT), bandLogE[i]-offset); + if (C==2) + follow = MAX16(follow, bandLogE[i+nbEBands]-offset); + frame_avg += follow; + } + frame_avg /= (st->end-st->start); + temporal_vbr = SUB16(frame_avg,st->spec_avg); + temporal_vbr = MIN16(QCONST16(3.f, DB_SHIFT), MAX16(-QCONST16(1.5f, DB_SHIFT), temporal_vbr)); + st->spec_avg += MULT16_16_Q15(QCONST16(.02f, 15), temporal_vbr); + } + /*for (i=0;i<21;i++) + printf("%f ", bandLogE[i]); + printf("\n");*/ + + if (!secondMdct) + { + for (i=0;i<C*nbEBands;i++) + bandLogE2[i] = bandLogE[i]; + } + + /* Last chance to catch any transient we might have missed in the + time-domain analysis */ + if (LM>0 && ec_tell(enc)+3<=total_bits && !isTransient && st->complexity>=5 && !st->lfe) + { + if (patch_transient_decision(bandLogE, oldBandE, nbEBands, st->end, C)) + { + isTransient = 1; + shortBlocks = M; + compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample); + compute_band_energies(mode, freq, bandE, effEnd, C, M); + amp2Log2(mode, effEnd, st->end, bandE, bandLogE, C); + /* Compensate for the scaling of short vs long mdcts */ + for (i=0;i<C*nbEBands;i++) + bandLogE2[i] += HALF16(SHL16(LM, DB_SHIFT)); + tf_estimate = QCONST16(.2f,14); + } + } + + if (LM>0 && ec_tell(enc)+3<=total_bits) + ec_enc_bit_logp(enc, isTransient, 3); + + ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ + + /* Band normalisation */ + normalise_bands(mode, freq, X, bandE, effEnd, C, M); + + ALLOC(tf_res, nbEBands, int); + /* Disable variable tf resolution for hybrid and at very low bitrate */ + if (effectiveBytes>=15*C && st->start==0 && st->complexity>=2 && !st->lfe) + { + int lambda; + if (effectiveBytes<40) + lambda = 12; + else if (effectiveBytes<60) + lambda = 6; + else if (effectiveBytes<100) + lambda = 4; + else + lambda = 3; + lambda*=2; + tf_select = tf_analysis(mode, effEnd, isTransient, tf_res, lambda, X, N, LM, &tf_sum, tf_estimate, tf_chan); + for (i=effEnd;i<st->end;i++) + tf_res[i] = tf_res[effEnd-1]; + } else { + tf_sum = 0; + for (i=0;i<st->end;i++) + tf_res[i] = isTransient; + tf_select=0; + } + + ALLOC(error, C*nbEBands, opus_val16); + quant_coarse_energy(mode, st->start, st->end, effEnd, bandLogE, + oldBandE, total_bits, error, enc, + C, LM, nbAvailableBytes, st->force_intra, + &st->delayedIntra, st->complexity >= 4, st->loss_rate, st->lfe); + + tf_encode(st->start, st->end, isTransient, tf_res, LM, tf_select, enc); + + if (ec_tell(enc)+4<=total_bits) + { + if (st->lfe) + { + st->tapset_decision = 0; + st->spread_decision = SPREAD_NORMAL; + } else if (shortBlocks || st->complexity < 3 || nbAvailableBytes < 10*C || st->start != 0) + { + if (st->complexity == 0) + st->spread_decision = SPREAD_NONE; + else + st->spread_decision = SPREAD_NORMAL; + } else { + /* Disable new spreading+tapset estimator until we can show it works + better than the old one. So far it seems like spreading_decision() + works best. */ +#if 0 + if (st->analysis.valid) + { + static const opus_val16 spread_thresholds[3] = {-QCONST16(.6f, 15), -QCONST16(.2f, 15), -QCONST16(.07f, 15)}; + static const opus_val16 spread_histeresis[3] = {QCONST16(.15f, 15), QCONST16(.07f, 15), QCONST16(.02f, 15)}; + static const opus_val16 tapset_thresholds[2] = {QCONST16(.0f, 15), QCONST16(.15f, 15)}; + static const opus_val16 tapset_histeresis[2] = {QCONST16(.1f, 15), QCONST16(.05f, 15)}; + st->spread_decision = hysteresis_decision(-st->analysis.tonality, spread_thresholds, spread_histeresis, 3, st->spread_decision); + st->tapset_decision = hysteresis_decision(st->analysis.tonality_slope, tapset_thresholds, tapset_histeresis, 2, st->tapset_decision); + } else +#endif + { + st->spread_decision = spreading_decision(mode, X, + &st->tonal_average, st->spread_decision, &st->hf_average, + &st->tapset_decision, pf_on&&!shortBlocks, effEnd, C, M); + } + /*printf("%d %d\n", st->tapset_decision, st->spread_decision);*/ + /*printf("%f %d %f %d\n\n", st->analysis.tonality, st->spread_decision, st->analysis.tonality_slope, st->tapset_decision);*/ + } + ec_enc_icdf(enc, st->spread_decision, spread_icdf, 5); + } + + ALLOC(offsets, nbEBands, int); + + maxDepth = dynalloc_analysis(bandLogE, bandLogE2, nbEBands, st->start, st->end, C, offsets, + st->lsb_depth, mode->logN, isTransient, st->vbr, st->constrained_vbr, + eBands, LM, effectiveBytes, &tot_boost, st->lfe, surround_dynalloc); + /* For LFE, everything interesting is in the first band */ + if (st->lfe) + offsets[0] = IMIN(8, effectiveBytes/3); + ALLOC(cap, nbEBands, int); + init_caps(mode,cap,LM,C); + + dynalloc_logp = 6; + total_bits<<=BITRES; + total_boost = 0; + tell = ec_tell_frac(enc); + for (i=st->start;i<st->end;i++) + { + int width, quanta; + int dynalloc_loop_logp; + int boost; + int j; + width = C*(eBands[i+1]-eBands[i])<<LM; + /* quanta is 6 bits, but no more than 1 bit/sample + and no less than 1/8 bit/sample */ + quanta = IMIN(width<<BITRES, IMAX(6<<BITRES, width)); + dynalloc_loop_logp = dynalloc_logp; + boost = 0; + for (j = 0; tell+(dynalloc_loop_logp<<BITRES) < total_bits-total_boost + && boost < cap[i]; j++) + { + int flag; + flag = j<offsets[i]; + ec_enc_bit_logp(enc, flag, dynalloc_loop_logp); + tell = ec_tell_frac(enc); + if (!flag) + break; + boost += quanta; + total_boost += quanta; + dynalloc_loop_logp = 1; + } + /* Making dynalloc more likely */ + if (j) + dynalloc_logp = IMAX(2, dynalloc_logp-1); + offsets[i] = boost; + } + + if (C==2) + { + static const opus_val16 intensity_thresholds[21]= + /* 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 off*/ + { 1, 2, 3, 4, 5, 6, 7, 8,16,24,36,44,50,56,62,67,72,79,88,106,134}; + static const opus_val16 intensity_histeresis[21]= + { 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 3, 3, 4, 5, 6, 8, 8}; + + /* Always use MS for 2.5 ms frames until we can do a better analysis */ + if (LM!=0) + dual_stereo = stereo_analysis(mode, X, LM, N); + + st->intensity = hysteresis_decision((opus_val16)(equiv_rate/1000), + intensity_thresholds, intensity_histeresis, 21, st->intensity); + st->intensity = IMIN(st->end,IMAX(st->start, st->intensity)); + } + + alloc_trim = 5; + if (tell+(6<<BITRES) <= total_bits - total_boost) + { + if (st->lfe) + alloc_trim = 5; + else + alloc_trim = alloc_trim_analysis(mode, X, bandLogE, + st->end, LM, C, N, &st->analysis, &st->stereo_saving, tf_estimate, st->intensity, surround_trim); + ec_enc_icdf(enc, alloc_trim, trim_icdf, 7); + tell = ec_tell_frac(enc); + } + + /* Variable bitrate */ + if (vbr_rate>0) + { + opus_val16 alpha; + opus_int32 delta; + /* The target rate in 8th bits per frame */ + opus_int32 target, base_target; + opus_int32 min_allowed; + int lm_diff = mode->maxLM - LM; + + /* Don't attempt to use more than 510 kb/s, even for frames smaller than 20 ms. + The CELT allocator will just not be able to use more than that anyway. */ + nbCompressedBytes = IMIN(nbCompressedBytes,1275>>(3-LM)); + base_target = vbr_rate - ((40*C+20)<<BITRES); + + if (st->constrained_vbr) + base_target += (st->vbr_offset>>lm_diff); + + target = compute_vbr(mode, &st->analysis, base_target, LM, equiv_rate, + st->lastCodedBands, C, st->intensity, st->constrained_vbr, + st->stereo_saving, tot_boost, tf_estimate, pitch_change, maxDepth, + st->variable_duration, st->lfe, st->energy_mask!=NULL, surround_masking, + temporal_vbr); + + /* The current offset is removed from the target and the space used + so far is added*/ + target=target+tell; + /* In VBR mode the frame size must not be reduced so much that it would + result in the encoder running out of bits. + The margin of 2 bytes ensures that none of the bust-prevention logic + in the decoder will have triggered so far. */ + min_allowed = ((tell+total_boost+(1<<(BITRES+3))-1)>>(BITRES+3)) + 2 - nbFilledBytes; + + nbAvailableBytes = (target+(1<<(BITRES+2)))>>(BITRES+3); + nbAvailableBytes = IMAX(min_allowed,nbAvailableBytes); + nbAvailableBytes = IMIN(nbCompressedBytes,nbAvailableBytes+nbFilledBytes) - nbFilledBytes; + + /* By how much did we "miss" the target on that frame */ + delta = target - vbr_rate; + + target=nbAvailableBytes<<(BITRES+3); + + /*If the frame is silent we don't adjust our drift, otherwise + the encoder will shoot to very high rates after hitting a + span of silence, but we do allow the bitres to refill. + This means that we'll undershoot our target in CVBR/VBR modes + on files with lots of silence. */ + if(silence) + { + nbAvailableBytes = 2; + target = 2*8<<BITRES; + delta = 0; + } + + if (st->vbr_count < 970) + { + st->vbr_count++; + alpha = celt_rcp(SHL32(EXTEND32(st->vbr_count+20),16)); + } else + alpha = QCONST16(.001f,15); + /* How many bits have we used in excess of what we're allowed */ + if (st->constrained_vbr) + st->vbr_reservoir += target - vbr_rate; + /*printf ("%d\n", st->vbr_reservoir);*/ + + /* Compute the offset we need to apply in order to reach the target */ + if (st->constrained_vbr) + { + st->vbr_drift += (opus_int32)MULT16_32_Q15(alpha,(delta*(1<<lm_diff))-st->vbr_offset-st->vbr_drift); + st->vbr_offset = -st->vbr_drift; + } + /*printf ("%d\n", st->vbr_drift);*/ + + if (st->constrained_vbr && st->vbr_reservoir < 0) + { + /* We're under the min value -- increase rate */ + int adjust = (-st->vbr_reservoir)/(8<<BITRES); + /* Unless we're just coding silence */ + nbAvailableBytes += silence?0:adjust; + st->vbr_reservoir = 0; + /*printf ("+%d\n", adjust);*/ + } + nbCompressedBytes = IMIN(nbCompressedBytes,nbAvailableBytes+nbFilledBytes); + /*printf("%d\n", nbCompressedBytes*50*8);*/ + /* This moves the raw bits to take into account the new compressed size */ + ec_enc_shrink(enc, nbCompressedBytes); + } + + /* Bit allocation */ + ALLOC(fine_quant, nbEBands, int); + ALLOC(pulses, nbEBands, int); + ALLOC(fine_priority, nbEBands, int); + + /* bits = packet size - where we are - safety*/ + bits = (((opus_int32)nbCompressedBytes*8)<<BITRES) - ec_tell_frac(enc) - 1; + anti_collapse_rsv = isTransient&&LM>=2&&bits>=((LM+2)<<BITRES) ? (1<<BITRES) : 0; + bits -= anti_collapse_rsv; + signalBandwidth = st->end-1; +#ifndef DISABLE_FLOAT_API + if (st->analysis.valid) + { + int min_bandwidth; + if (equiv_rate < (opus_int32)32000*C) + min_bandwidth = 13; + else if (equiv_rate < (opus_int32)48000*C) + min_bandwidth = 16; + else if (equiv_rate < (opus_int32)60000*C) + min_bandwidth = 18; + else if (equiv_rate < (opus_int32)80000*C) + min_bandwidth = 19; + else + min_bandwidth = 20; + signalBandwidth = IMAX(st->analysis.bandwidth, min_bandwidth); + } +#endif + if (st->lfe) + signalBandwidth = 1; + codedBands = compute_allocation(mode, st->start, st->end, offsets, cap, + alloc_trim, &st->intensity, &dual_stereo, bits, &balance, pulses, + fine_quant, fine_priority, C, LM, enc, 1, st->lastCodedBands, signalBandwidth); + if (st->lastCodedBands) + st->lastCodedBands = IMIN(st->lastCodedBands+1,IMAX(st->lastCodedBands-1,codedBands)); + else + st->lastCodedBands = codedBands; + + quant_fine_energy(mode, st->start, st->end, oldBandE, error, fine_quant, enc, C); + + /* Residual quantisation */ + ALLOC(collapse_masks, C*nbEBands, unsigned char); + quant_all_bands(1, mode, st->start, st->end, X, C==2 ? X+N : NULL, collapse_masks, + bandE, pulses, shortBlocks, st->spread_decision, dual_stereo, st->intensity, tf_res, + nbCompressedBytes*(8<<BITRES)-anti_collapse_rsv, balance, enc, LM, codedBands, &st->rng); + + if (anti_collapse_rsv > 0) + { + anti_collapse_on = st->consec_transient<2; +#ifdef FUZZING + anti_collapse_on = rand()&0x1; +#endif + ec_enc_bits(enc, anti_collapse_on, 1); + } + quant_energy_finalise(mode, st->start, st->end, oldBandE, error, fine_quant, fine_priority, nbCompressedBytes*8-ec_tell(enc), enc, C); + + if (silence) + { + for (i=0;i<C*nbEBands;i++) + oldBandE[i] = -QCONST16(28.f,DB_SHIFT); + } + +#ifdef RESYNTH + /* Re-synthesis of the coded audio if required */ + { + celt_sig *out_mem[2]; + + if (anti_collapse_on) + { + anti_collapse(mode, X, collapse_masks, LM, C, N, + st->start, st->end, oldBandE, oldLogE, oldLogE2, pulses, st->rng); + } + + if (silence) + { + for (i=0;i<C*N;i++) + freq[i] = 0; + } else { + /* Synthesis */ + denormalise_bands(mode, X, freq, oldBandE, st->start, effEnd, C, M); + } + + c=0; do { + OPUS_MOVE(st->syn_mem[c], st->syn_mem[c]+N, 2*MAX_PERIOD-N+overlap/2); + } while (++c<CC); + + if (CC==2&&C==1) + { + for (i=0;i<N;i++) + freq[N+i] = freq[i]; + } + + c=0; do { + out_mem[c] = st->syn_mem[c]+2*MAX_PERIOD-N; + } while (++c<CC); + + compute_inv_mdcts(mode, shortBlocks, freq, out_mem, CC, LM); + + c=0; do { + st->prefilter_period=IMAX(st->prefilter_period, COMBFILTER_MINPERIOD); + st->prefilter_period_old=IMAX(st->prefilter_period_old, COMBFILTER_MINPERIOD); + comb_filter(out_mem[c], out_mem[c], st->prefilter_period_old, st->prefilter_period, mode->shortMdctSize, + st->prefilter_gain_old, st->prefilter_gain, st->prefilter_tapset_old, st->prefilter_tapset, + mode->window, st->overlap); + if (LM!=0) + comb_filter(out_mem[c]+mode->shortMdctSize, out_mem[c]+mode->shortMdctSize, st->prefilter_period, pitch_index, N-mode->shortMdctSize, + st->prefilter_gain, gain1, st->prefilter_tapset, prefilter_tapset, + mode->window, overlap); + } while (++c<CC); + + /* We reuse freq[] as scratch space for the de-emphasis */ + deemphasis(out_mem, (opus_val16*)pcm, N, CC, st->upsample, mode->preemph, st->preemph_memD, freq); + st->prefilter_period_old = st->prefilter_period; + st->prefilter_gain_old = st->prefilter_gain; + st->prefilter_tapset_old = st->prefilter_tapset; + } +#endif + + st->prefilter_period = pitch_index; + st->prefilter_gain = gain1; + st->prefilter_tapset = prefilter_tapset; +#ifdef RESYNTH + if (LM!=0) + { + st->prefilter_period_old = st->prefilter_period; + st->prefilter_gain_old = st->prefilter_gain; + st->prefilter_tapset_old = st->prefilter_tapset; + } +#endif + + if (CC==2&&C==1) { + for (i=0;i<nbEBands;i++) + oldBandE[nbEBands+i]=oldBandE[i]; + } + + if (!isTransient) + { + for (i=0;i<CC*nbEBands;i++) + oldLogE2[i] = oldLogE[i]; + for (i=0;i<CC*nbEBands;i++) + oldLogE[i] = oldBandE[i]; + } else { + for (i=0;i<CC*nbEBands;i++) + oldLogE[i] = MIN16(oldLogE[i], oldBandE[i]); + } + /* In case start or end were to change */ + c=0; do + { + for (i=0;i<st->start;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + for (i=st->end;i<nbEBands;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + } while (++c<CC); + + if (isTransient || transient_got_disabled) + st->consec_transient++; + else + st->consec_transient=0; + st->rng = enc->rng; + + /* If there's any room left (can only happen for very high rates), + it's already filled with zeros */ + ec_enc_done(enc); + +#ifdef CUSTOM_MODES + if (st->signalling) + nbCompressedBytes++; +#endif + + RESTORE_STACK; + if (ec_get_error(enc)) + return OPUS_INTERNAL_ERROR; + else + return nbCompressedBytes; +} + + +#ifdef CUSTOM_MODES + +#ifdef FIXED_POINT +int opus_custom_encode(CELTEncoder * OPUS_RESTRICT st, const opus_int16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + return celt_encode_with_ec(st, pcm, frame_size, compressed, nbCompressedBytes, NULL); +} + +#ifndef DISABLE_FLOAT_API +int opus_custom_encode_float(CELTEncoder * OPUS_RESTRICT st, const float * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + int j, ret, C, N; + VARDECL(opus_int16, in); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C = st->channels; + N = frame_size; + ALLOC(in, C*N, opus_int16); + + for (j=0;j<C*N;j++) + in[j] = FLOAT2INT16(pcm[j]); + + ret=celt_encode_with_ec(st,in,frame_size,compressed,nbCompressedBytes, NULL); +#ifdef RESYNTH + for (j=0;j<C*N;j++) + ((float*)pcm)[j]=in[j]*(1.f/32768.f); +#endif + RESTORE_STACK; + return ret; +} +#endif /* DISABLE_FLOAT_API */ +#else + +int opus_custom_encode(CELTEncoder * OPUS_RESTRICT st, const opus_int16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + int j, ret, C, N; + VARDECL(celt_sig, in); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C=st->channels; + N=frame_size; + ALLOC(in, C*N, celt_sig); + for (j=0;j<C*N;j++) { + in[j] = SCALEOUT(pcm[j]); + } + + ret = celt_encode_with_ec(st,in,frame_size,compressed,nbCompressedBytes, NULL); +#ifdef RESYNTH + for (j=0;j<C*N;j++) + ((opus_int16*)pcm)[j] = FLOAT2INT16(in[j]); +#endif + RESTORE_STACK; + return ret; +} + +int opus_custom_encode_float(CELTEncoder * OPUS_RESTRICT st, const float * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + return celt_encode_with_ec(st, pcm, frame_size, compressed, nbCompressedBytes, NULL); +} + +#endif + +#endif /* CUSTOM_MODES */ + +int opus_custom_encoder_ctl(CELTEncoder * OPUS_RESTRICT st, int request, ...) +{ + va_list ap; + + va_start(ap, request); + switch (request) + { + case OPUS_SET_COMPLEXITY_REQUEST: + { + int value = va_arg(ap, opus_int32); + if (value<0 || value>10) + goto bad_arg; + st->complexity = value; + } + break; + case CELT_SET_START_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<0 || value>=st->mode->nbEBands) + goto bad_arg; + st->start = value; + } + break; + case CELT_SET_END_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>st->mode->nbEBands) + goto bad_arg; + st->end = value; + } + break; + case CELT_SET_PREDICTION_REQUEST: + { + int value = va_arg(ap, opus_int32); + if (value<0 || value>2) + goto bad_arg; + st->disable_pf = value<=1; + st->force_intra = value==0; + } + break; + case OPUS_SET_PACKET_LOSS_PERC_REQUEST: + { + int value = va_arg(ap, opus_int32); + if (value<0 || value>100) + goto bad_arg; + st->loss_rate = value; + } + break; + case OPUS_SET_VBR_CONSTRAINT_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->constrained_vbr = value; + } + break; + case OPUS_SET_VBR_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->vbr = value; + } + break; + case OPUS_SET_BITRATE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<=500 && value!=OPUS_BITRATE_MAX) + goto bad_arg; + value = IMIN(value, 260000*st->channels); + st->bitrate = value; + } + break; + case CELT_SET_CHANNELS_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>2) + goto bad_arg; + st->stream_channels = value; + } + break; + case OPUS_SET_LSB_DEPTH_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<8 || value>24) + goto bad_arg; + st->lsb_depth=value; + } + break; + case OPUS_GET_LSB_DEPTH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + *value=st->lsb_depth; + } + break; + case OPUS_SET_EXPERT_FRAME_DURATION_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->variable_duration = value; + } + break; + case OPUS_RESET_STATE: + { + int i; + opus_val16 *oldBandE, *oldLogE, *oldLogE2; + oldBandE = (opus_val16*)(st->in_mem+st->channels*(st->overlap+COMBFILTER_MAXPERIOD)); + oldLogE = oldBandE + st->channels*st->mode->nbEBands; + oldLogE2 = oldLogE + st->channels*st->mode->nbEBands; + OPUS_CLEAR((char*)&st->ENCODER_RESET_START, + opus_custom_encoder_get_size(st->mode, st->channels)- + ((char*)&st->ENCODER_RESET_START - (char*)st)); + for (i=0;i<st->channels*st->mode->nbEBands;i++) + oldLogE[i]=oldLogE2[i]=-QCONST16(28.f,DB_SHIFT); + st->vbr_offset = 0; + st->delayedIntra = 1; + st->spread_decision = SPREAD_NORMAL; + st->tonal_average = 256; + st->hf_average = 0; + st->tapset_decision = 0; + } + break; +#ifdef CUSTOM_MODES + case CELT_SET_INPUT_CLIPPING_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->clip = value; + } + break; +#endif + case CELT_SET_SIGNALLING_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->signalling = value; + } + break; + case CELT_SET_ANALYSIS_REQUEST: + { + AnalysisInfo *info = va_arg(ap, AnalysisInfo *); + if (info) + OPUS_COPY(&st->analysis, info, 1); + } + break; + case CELT_GET_MODE_REQUEST: + { + const CELTMode ** value = va_arg(ap, const CELTMode**); + if (value==0) + goto bad_arg; + *value=st->mode; + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + opus_uint32 * value = va_arg(ap, opus_uint32 *); + if (value==0) + goto bad_arg; + *value=st->rng; + } + break; + case OPUS_SET_LFE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->lfe = value; + } + break; + case OPUS_SET_ENERGY_MASK_REQUEST: + { + opus_val16 *value = va_arg(ap, opus_val16*); + st->energy_mask = value; + } + break; + default: + goto bad_request; + } + va_end(ap); + return OPUS_OK; +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +bad_request: + va_end(ap); + return OPUS_UNIMPLEMENTED; +} diff --git a/src/opus-1.1/celt/celt_lpc.c b/src/opus-1.1/celt/celt_lpc.c new file mode 100644 index 00000000..fa29d626 --- /dev/null +++ b/src/opus-1.1/celt/celt_lpc.c @@ -0,0 +1,309 @@ +/* Copyright (c) 2009-2010 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "celt_lpc.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "pitch.h" + +void _celt_lpc( + opus_val16 *_lpc, /* out: [0...p-1] LPC coefficients */ +const opus_val32 *ac, /* in: [0...p] autocorrelation values */ +int p +) +{ + int i, j; + opus_val32 r; + opus_val32 error = ac[0]; +#ifdef FIXED_POINT + opus_val32 lpc[LPC_ORDER]; +#else + float *lpc = _lpc; +#endif + + for (i = 0; i < p; i++) + lpc[i] = 0; + if (ac[0] != 0) + { + for (i = 0; i < p; i++) { + /* Sum up this iteration's reflection coefficient */ + opus_val32 rr = 0; + for (j = 0; j < i; j++) + rr += MULT32_32_Q31(lpc[j],ac[i - j]); + rr += SHR32(ac[i + 1],3); + r = -frac_div32(SHL32(rr,3), error); + /* Update LPC coefficients and total error */ + lpc[i] = SHR32(r,3); + for (j = 0; j < (i+1)>>1; j++) + { + opus_val32 tmp1, tmp2; + tmp1 = lpc[j]; + tmp2 = lpc[i-1-j]; + lpc[j] = tmp1 + MULT32_32_Q31(r,tmp2); + lpc[i-1-j] = tmp2 + MULT32_32_Q31(r,tmp1); + } + + error = error - MULT32_32_Q31(MULT32_32_Q31(r,r),error); + /* Bail out once we get 30 dB gain */ +#ifdef FIXED_POINT + if (error<SHR32(ac[0],10)) + break; +#else + if (error<.001f*ac[0]) + break; +#endif + } + } +#ifdef FIXED_POINT + for (i=0;i<p;i++) + _lpc[i] = ROUND16(lpc[i],16); +#endif +} + +void celt_fir(const opus_val16 *_x, + const opus_val16 *num, + opus_val16 *_y, + int N, + int ord, + opus_val16 *mem) +{ + int i,j; + VARDECL(opus_val16, rnum); + VARDECL(opus_val16, x); + SAVE_STACK; + + ALLOC(rnum, ord, opus_val16); + ALLOC(x, N+ord, opus_val16); + for(i=0;i<ord;i++) + rnum[i] = num[ord-i-1]; + for(i=0;i<ord;i++) + x[i] = mem[ord-i-1]; + for (i=0;i<N;i++) + x[i+ord]=_x[i]; + for(i=0;i<ord;i++) + mem[i] = _x[N-i-1]; +#ifdef SMALL_FOOTPRINT + for (i=0;i<N;i++) + { + opus_val32 sum = SHL32(EXTEND32(_x[i]), SIG_SHIFT); + for (j=0;j<ord;j++) + { + sum = MAC16_16(sum,rnum[j],x[i+j]); + } + _y[i] = SATURATE16(PSHR32(sum, SIG_SHIFT)); + } +#else + for (i=0;i<N-3;i+=4) + { + opus_val32 sum[4]={0,0,0,0}; + xcorr_kernel(rnum, x+i, sum, ord); + _y[i ] = SATURATE16(ADD32(EXTEND32(_x[i ]), PSHR32(sum[0], SIG_SHIFT))); + _y[i+1] = SATURATE16(ADD32(EXTEND32(_x[i+1]), PSHR32(sum[1], SIG_SHIFT))); + _y[i+2] = SATURATE16(ADD32(EXTEND32(_x[i+2]), PSHR32(sum[2], SIG_SHIFT))); + _y[i+3] = SATURATE16(ADD32(EXTEND32(_x[i+3]), PSHR32(sum[3], SIG_SHIFT))); + } + for (;i<N;i++) + { + opus_val32 sum = 0; + for (j=0;j<ord;j++) + sum = MAC16_16(sum,rnum[j],x[i+j]); + _y[i] = SATURATE16(ADD32(EXTEND32(_x[i]), PSHR32(sum, SIG_SHIFT))); + } +#endif + RESTORE_STACK; +} + +void celt_iir(const opus_val32 *_x, + const opus_val16 *den, + opus_val32 *_y, + int N, + int ord, + opus_val16 *mem) +{ +#ifdef SMALL_FOOTPRINT + int i,j; + for (i=0;i<N;i++) + { + opus_val32 sum = _x[i]; + for (j=0;j<ord;j++) + { + sum -= MULT16_16(den[j],mem[j]); + } + for (j=ord-1;j>=1;j--) + { + mem[j]=mem[j-1]; + } + mem[0] = ROUND16(sum,SIG_SHIFT); + _y[i] = sum; + } +#else + int i,j; + VARDECL(opus_val16, rden); + VARDECL(opus_val16, y); + SAVE_STACK; + + celt_assert((ord&3)==0); + ALLOC(rden, ord, opus_val16); + ALLOC(y, N+ord, opus_val16); + for(i=0;i<ord;i++) + rden[i] = den[ord-i-1]; + for(i=0;i<ord;i++) + y[i] = -mem[ord-i-1]; + for(;i<N+ord;i++) + y[i]=0; + for (i=0;i<N-3;i+=4) + { + /* Unroll by 4 as if it were an FIR filter */ + opus_val32 sum[4]; + sum[0]=_x[i]; + sum[1]=_x[i+1]; + sum[2]=_x[i+2]; + sum[3]=_x[i+3]; + xcorr_kernel(rden, y+i, sum, ord); + + /* Patch up the result to compensate for the fact that this is an IIR */ + y[i+ord ] = -ROUND16(sum[0],SIG_SHIFT); + _y[i ] = sum[0]; + sum[1] = MAC16_16(sum[1], y[i+ord ], den[0]); + y[i+ord+1] = -ROUND16(sum[1],SIG_SHIFT); + _y[i+1] = sum[1]; + sum[2] = MAC16_16(sum[2], y[i+ord+1], den[0]); + sum[2] = MAC16_16(sum[2], y[i+ord ], den[1]); + y[i+ord+2] = -ROUND16(sum[2],SIG_SHIFT); + _y[i+2] = sum[2]; + + sum[3] = MAC16_16(sum[3], y[i+ord+2], den[0]); + sum[3] = MAC16_16(sum[3], y[i+ord+1], den[1]); + sum[3] = MAC16_16(sum[3], y[i+ord ], den[2]); + y[i+ord+3] = -ROUND16(sum[3],SIG_SHIFT); + _y[i+3] = sum[3]; + } + for (;i<N;i++) + { + opus_val32 sum = _x[i]; + for (j=0;j<ord;j++) + sum -= MULT16_16(rden[j],y[i+j]); + y[i+ord] = ROUND16(sum,SIG_SHIFT); + _y[i] = sum; + } + for(i=0;i<ord;i++) + mem[i] = _y[N-i-1]; + RESTORE_STACK; +#endif +} + +int _celt_autocorr( + const opus_val16 *x, /* in: [0...n-1] samples x */ + opus_val32 *ac, /* out: [0...lag-1] ac values */ + const opus_val16 *window, + int overlap, + int lag, + int n, + int arch + ) +{ + opus_val32 d; + int i, k; + int fastN=n-lag; + int shift; + const opus_val16 *xptr; + VARDECL(opus_val16, xx); + SAVE_STACK; + ALLOC(xx, n, opus_val16); + celt_assert(n>0); + celt_assert(overlap>=0); + if (overlap == 0) + { + xptr = x; + } else { + for (i=0;i<n;i++) + xx[i] = x[i]; + for (i=0;i<overlap;i++) + { + xx[i] = MULT16_16_Q15(x[i],window[i]); + xx[n-i-1] = MULT16_16_Q15(x[n-i-1],window[i]); + } + xptr = xx; + } + shift=0; +#ifdef FIXED_POINT + { + opus_val32 ac0; + ac0 = 1+(n<<7); + if (n&1) ac0 += SHR32(MULT16_16(xptr[0],xptr[0]),9); + for(i=(n&1);i<n;i+=2) + { + ac0 += SHR32(MULT16_16(xptr[i],xptr[i]),9); + ac0 += SHR32(MULT16_16(xptr[i+1],xptr[i+1]),9); + } + + shift = celt_ilog2(ac0)-30+10; + shift = (shift)/2; + if (shift>0) + { + for(i=0;i<n;i++) + xx[i] = PSHR32(xptr[i], shift); + xptr = xx; + } else + shift = 0; + } +#endif + celt_pitch_xcorr(xptr, xptr, ac, fastN, lag+1, arch); + for (k=0;k<=lag;k++) + { + for (i = k+fastN, d = 0; i < n; i++) + d = MAC16_16(d, xptr[i], xptr[i-k]); + ac[k] += d; + } +#ifdef FIXED_POINT + shift = 2*shift; + if (shift<=0) + ac[0] += SHL32((opus_int32)1, -shift); + if (ac[0] < 268435456) + { + int shift2 = 29 - EC_ILOG(ac[0]); + for (i=0;i<=lag;i++) + ac[i] = SHL32(ac[i], shift2); + shift -= shift2; + } else if (ac[0] >= 536870912) + { + int shift2=1; + if (ac[0] >= 1073741824) + shift2++; + for (i=0;i<=lag;i++) + ac[i] = SHR32(ac[i], shift2); + shift += shift2; + } +#endif + + RESTORE_STACK; + return shift; +} diff --git a/src/opus-1.0.2/celt/celt_lpc.h b/src/opus-1.1/celt/celt_lpc.h index 2baa77ed..dc2a0a3d 100644 --- a/src/opus-1.0.2/celt/celt_lpc.h +++ b/src/opus-1.1/celt/celt_lpc.h @@ -48,6 +48,7 @@ void celt_iir(const opus_val32 *x, int ord, opus_val16 *mem); -void _celt_autocorr(const opus_val16 *x, opus_val32 *ac, const opus_val16 *window, int overlap, int lag, int n); +int _celt_autocorr(const opus_val16 *x, opus_val32 *ac, + const opus_val16 *window, int overlap, int lag, int n, int arch); #endif /* PLC_H */ diff --git a/src/opus-1.0.2/celt/pitch.h b/src/opus-1.1/celt/cpu_support.h index 2757071a..d68dbe62 100644 --- a/src/opus-1.0.2/celt/pitch.h +++ b/src/opus-1.1/celt/cpu_support.h @@ -1,11 +1,5 @@ -/* Copyright (c) 2007-2008 CSIRO - Copyright (c) 2007-2009 Xiph.Org Foundation - Written by Jean-Marc Valin */ -/** - @file pitch.h - @brief Pitch analysis - */ - +/* Copyright (c) 2010 Xiph.Org Foundation + * Copyright (c) 2013 Parrot */ /* Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions @@ -31,18 +25,30 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ -#ifndef PITCH_H -#define PITCH_H +#ifndef CPU_SUPPORT_H +#define CPU_SUPPORT_H -#include "modes.h" +#include "opus_types.h" +#include "opus_defines.h" -void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x_lp, - int len, int C); +#if defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_ASM) +#include "arm/armcpu.h" -void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTRICT y, - int len, int max_pitch, int *pitch); +/* We currently support 4 ARM variants: + * arch[0] -> ARMv4 + * arch[1] -> ARMv5E + * arch[2] -> ARMv6 + * arch[3] -> NEON + */ +#define OPUS_ARCHMASK 3 -opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod, - int N, int *T0, int prev_period, opus_val16 prev_gain); +#else +#define OPUS_ARCHMASK 0 + +static OPUS_INLINE int opus_select_arch(void) +{ + return 0; +} +#endif #endif diff --git a/src/opus-1.1/celt/cwrs.c b/src/opus-1.1/celt/cwrs.c new file mode 100644 index 00000000..ad980cc7 --- /dev/null +++ b/src/opus-1.1/celt/cwrs.c @@ -0,0 +1,697 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2007-2009 Timothy B. Terriberry + Written by Timothy B. Terriberry and Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "os_support.h" +#include "cwrs.h" +#include "mathops.h" +#include "arch.h" + +#ifdef CUSTOM_MODES + +/*Guaranteed to return a conservatively large estimate of the binary logarithm + with frac bits of fractional precision. + Tested for all possible 32-bit inputs with frac=4, where the maximum + overestimation is 0.06254243 bits.*/ +int log2_frac(opus_uint32 val, int frac) +{ + int l; + l=EC_ILOG(val); + if(val&(val-1)){ + /*This is (val>>l-16), but guaranteed to round up, even if adding a bias + before the shift would cause overflow (e.g., for 0xFFFFxxxx). + Doesn't work for val=0, but that case fails the test above.*/ + if(l>16)val=((val-1)>>(l-16))+1; + else val<<=16-l; + l=(l-1)<<frac; + /*Note that we always need one iteration, since the rounding up above means + that we might need to adjust the integer part of the logarithm.*/ + do{ + int b; + b=(int)(val>>16); + l+=b<<frac; + val=(val+b)>>b; + val=(val*val+0x7FFF)>>15; + } + while(frac-->0); + /*If val is not exactly 0x8000, then we have to round up the remainder.*/ + return l+(val>0x8000); + } + /*Exact powers of two require no rounding.*/ + else return (l-1)<<frac; +} +#endif + +/*Although derived separately, the pulse vector coding scheme is equivalent to + a Pyramid Vector Quantizer \cite{Fis86}. + Some additional notes about an early version appear at + http://people.xiph.org/~tterribe/notes/cwrs.html, but the codebook ordering + and the definitions of some terms have evolved since that was written. + + The conversion from a pulse vector to an integer index (encoding) and back + (decoding) is governed by two related functions, V(N,K) and U(N,K). + + V(N,K) = the number of combinations, with replacement, of N items, taken K + at a time, when a sign bit is added to each item taken at least once (i.e., + the number of N-dimensional unit pulse vectors with K pulses). + One way to compute this is via + V(N,K) = K>0 ? sum(k=1...K,2**k*choose(N,k)*choose(K-1,k-1)) : 1, + where choose() is the binomial function. + A table of values for N<10 and K<10 looks like: + V[10][10] = { + {1, 0, 0, 0, 0, 0, 0, 0, 0, 0}, + {1, 2, 2, 2, 2, 2, 2, 2, 2, 2}, + {1, 4, 8, 12, 16, 20, 24, 28, 32, 36}, + {1, 6, 18, 38, 66, 102, 146, 198, 258, 326}, + {1, 8, 32, 88, 192, 360, 608, 952, 1408, 1992}, + {1, 10, 50, 170, 450, 1002, 1970, 3530, 5890, 9290}, + {1, 12, 72, 292, 912, 2364, 5336, 10836, 20256, 35436}, + {1, 14, 98, 462, 1666, 4942, 12642, 28814, 59906, 115598}, + {1, 16, 128, 688, 2816, 9424, 27008, 68464, 157184, 332688}, + {1, 18, 162, 978, 4482, 16722, 53154, 148626, 374274, 864146} + }; + + U(N,K) = the number of such combinations wherein N-1 objects are taken at + most K-1 at a time. + This is given by + U(N,K) = sum(k=0...K-1,V(N-1,k)) + = K>0 ? (V(N-1,K-1) + V(N,K-1))/2 : 0. + The latter expression also makes clear that U(N,K) is half the number of such + combinations wherein the first object is taken at least once. + Although it may not be clear from either of these definitions, U(N,K) is the + natural function to work with when enumerating the pulse vector codebooks, + not V(N,K). + U(N,K) is not well-defined for N=0, but with the extension + U(0,K) = K>0 ? 0 : 1, + the function becomes symmetric: U(N,K) = U(K,N), with a similar table: + U[10][10] = { + {1, 0, 0, 0, 0, 0, 0, 0, 0, 0}, + {0, 1, 1, 1, 1, 1, 1, 1, 1, 1}, + {0, 1, 3, 5, 7, 9, 11, 13, 15, 17}, + {0, 1, 5, 13, 25, 41, 61, 85, 113, 145}, + {0, 1, 7, 25, 63, 129, 231, 377, 575, 833}, + {0, 1, 9, 41, 129, 321, 681, 1289, 2241, 3649}, + {0, 1, 11, 61, 231, 681, 1683, 3653, 7183, 13073}, + {0, 1, 13, 85, 377, 1289, 3653, 8989, 19825, 40081}, + {0, 1, 15, 113, 575, 2241, 7183, 19825, 48639, 108545}, + {0, 1, 17, 145, 833, 3649, 13073, 40081, 108545, 265729} + }; + + With this extension, V(N,K) may be written in terms of U(N,K): + V(N,K) = U(N,K) + U(N,K+1) + for all N>=0, K>=0. + Thus U(N,K+1) represents the number of combinations where the first element + is positive or zero, and U(N,K) represents the number of combinations where + it is negative. + With a large enough table of U(N,K) values, we could write O(N) encoding + and O(min(N*log(K),N+K)) decoding routines, but such a table would be + prohibitively large for small embedded devices (K may be as large as 32767 + for small N, and N may be as large as 200). + + Both functions obey the same recurrence relation: + V(N,K) = V(N-1,K) + V(N,K-1) + V(N-1,K-1), + U(N,K) = U(N-1,K) + U(N,K-1) + U(N-1,K-1), + for all N>0, K>0, with different initial conditions at N=0 or K=0. + This allows us to construct a row of one of the tables above given the + previous row or the next row. + Thus we can derive O(NK) encoding and decoding routines with O(K) memory + using only addition and subtraction. + + When encoding, we build up from the U(2,K) row and work our way forwards. + When decoding, we need to start at the U(N,K) row and work our way backwards, + which requires a means of computing U(N,K). + U(N,K) may be computed from two previous values with the same N: + U(N,K) = ((2*N-1)*U(N,K-1) - U(N,K-2))/(K-1) + U(N,K-2) + for all N>1, and since U(N,K) is symmetric, a similar relation holds for two + previous values with the same K: + U(N,K>1) = ((2*K-1)*U(N-1,K) - U(N-2,K))/(N-1) + U(N-2,K) + for all K>1. + This allows us to construct an arbitrary row of the U(N,K) table by starting + with the first two values, which are constants. + This saves roughly 2/3 the work in our O(NK) decoding routine, but costs O(K) + multiplications. + Similar relations can be derived for V(N,K), but are not used here. + + For N>0 and K>0, U(N,K) and V(N,K) take on the form of an (N-1)-degree + polynomial for fixed N. + The first few are + U(1,K) = 1, + U(2,K) = 2*K-1, + U(3,K) = (2*K-2)*K+1, + U(4,K) = (((4*K-6)*K+8)*K-3)/3, + U(5,K) = ((((2*K-4)*K+10)*K-8)*K+3)/3, + and + V(1,K) = 2, + V(2,K) = 4*K, + V(3,K) = 4*K*K+2, + V(4,K) = 8*(K*K+2)*K/3, + V(5,K) = ((4*K*K+20)*K*K+6)/3, + for all K>0. + This allows us to derive O(N) encoding and O(N*log(K)) decoding routines for + small N (and indeed decoding is also O(N) for N<3). + + @ARTICLE{Fis86, + author="Thomas R. Fischer", + title="A Pyramid Vector Quantizer", + journal="IEEE Transactions on Information Theory", + volume="IT-32", + number=4, + pages="568--583", + month=Jul, + year=1986 + }*/ + +#if !defined(SMALL_FOOTPRINT) + +/*U(N,K) = U(K,N) := N>0?K>0?U(N-1,K)+U(N,K-1)+U(N-1,K-1):0:K>0?1:0*/ +# define CELT_PVQ_U(_n,_k) (CELT_PVQ_U_ROW[IMIN(_n,_k)][IMAX(_n,_k)]) +/*V(N,K) := U(N,K)+U(N,K+1) = the number of PVQ codewords for a band of size N + with K pulses allocated to it.*/ +# define CELT_PVQ_V(_n,_k) (CELT_PVQ_U(_n,_k)+CELT_PVQ_U(_n,(_k)+1)) + +/*For each V(N,K) supported, we will access element U(min(N,K+1),max(N,K+1)). + Thus, the number of entries in row I is the larger of the maximum number of + pulses we will ever allocate for a given N=I (K=128, or however many fit in + 32 bits, whichever is smaller), plus one, and the maximum N for which + K=I-1 pulses fit in 32 bits. + The largest band size in an Opus Custom mode is 208. + Otherwise, we can limit things to the set of N which can be achieved by + splitting a band from a standard Opus mode: 176, 144, 96, 88, 72, 64, 48, + 44, 36, 32, 24, 22, 18, 16, 8, 4, 2).*/ +#if defined(CUSTOM_MODES) +static const opus_uint32 CELT_PVQ_U_DATA[1488]={ +#else +static const opus_uint32 CELT_PVQ_U_DATA[1272]={ +#endif + /*N=0, K=0...176:*/ + 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +#if defined(CUSTOM_MODES) + /*...208:*/ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, +#endif + /*N=1, K=1...176:*/ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, +#if defined(CUSTOM_MODES) + /*...208:*/ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, +#endif + /*N=2, K=2...176:*/ + 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29, 31, 33, 35, 37, 39, 41, + 43, 45, 47, 49, 51, 53, 55, 57, 59, 61, 63, 65, 67, 69, 71, 73, 75, 77, 79, + 81, 83, 85, 87, 89, 91, 93, 95, 97, 99, 101, 103, 105, 107, 109, 111, 113, + 115, 117, 119, 121, 123, 125, 127, 129, 131, 133, 135, 137, 139, 141, 143, + 145, 147, 149, 151, 153, 155, 157, 159, 161, 163, 165, 167, 169, 171, 173, + 175, 177, 179, 181, 183, 185, 187, 189, 191, 193, 195, 197, 199, 201, 203, + 205, 207, 209, 211, 213, 215, 217, 219, 221, 223, 225, 227, 229, 231, 233, + 235, 237, 239, 241, 243, 245, 247, 249, 251, 253, 255, 257, 259, 261, 263, + 265, 267, 269, 271, 273, 275, 277, 279, 281, 283, 285, 287, 289, 291, 293, + 295, 297, 299, 301, 303, 305, 307, 309, 311, 313, 315, 317, 319, 321, 323, + 325, 327, 329, 331, 333, 335, 337, 339, 341, 343, 345, 347, 349, 351, +#if defined(CUSTOM_MODES) + /*...208:*/ + 353, 355, 357, 359, 361, 363, 365, 367, 369, 371, 373, 375, 377, 379, 381, + 383, 385, 387, 389, 391, 393, 395, 397, 399, 401, 403, 405, 407, 409, 411, + 413, 415, +#endif + /*N=3, K=3...176:*/ + 13, 25, 41, 61, 85, 113, 145, 181, 221, 265, 313, 365, 421, 481, 545, 613, + 685, 761, 841, 925, 1013, 1105, 1201, 1301, 1405, 1513, 1625, 1741, 1861, + 1985, 2113, 2245, 2381, 2521, 2665, 2813, 2965, 3121, 3281, 3445, 3613, 3785, + 3961, 4141, 4325, 4513, 4705, 4901, 5101, 5305, 5513, 5725, 5941, 6161, 6385, + 6613, 6845, 7081, 7321, 7565, 7813, 8065, 8321, 8581, 8845, 9113, 9385, 9661, + 9941, 10225, 10513, 10805, 11101, 11401, 11705, 12013, 12325, 12641, 12961, + 13285, 13613, 13945, 14281, 14621, 14965, 15313, 15665, 16021, 16381, 16745, + 17113, 17485, 17861, 18241, 18625, 19013, 19405, 19801, 20201, 20605, 21013, + 21425, 21841, 22261, 22685, 23113, 23545, 23981, 24421, 24865, 25313, 25765, + 26221, 26681, 27145, 27613, 28085, 28561, 29041, 29525, 30013, 30505, 31001, + 31501, 32005, 32513, 33025, 33541, 34061, 34585, 35113, 35645, 36181, 36721, + 37265, 37813, 38365, 38921, 39481, 40045, 40613, 41185, 41761, 42341, 42925, + 43513, 44105, 44701, 45301, 45905, 46513, 47125, 47741, 48361, 48985, 49613, + 50245, 50881, 51521, 52165, 52813, 53465, 54121, 54781, 55445, 56113, 56785, + 57461, 58141, 58825, 59513, 60205, 60901, 61601, +#if defined(CUSTOM_MODES) + /*...208:*/ + 62305, 63013, 63725, 64441, 65161, 65885, 66613, 67345, 68081, 68821, 69565, + 70313, 71065, 71821, 72581, 73345, 74113, 74885, 75661, 76441, 77225, 78013, + 78805, 79601, 80401, 81205, 82013, 82825, 83641, 84461, 85285, 86113, +#endif + /*N=4, K=4...176:*/ + 63, 129, 231, 377, 575, 833, 1159, 1561, 2047, 2625, 3303, 4089, 4991, 6017, + 7175, 8473, 9919, 11521, 13287, 15225, 17343, 19649, 22151, 24857, 27775, + 30913, 34279, 37881, 41727, 45825, 50183, 54809, 59711, 64897, 70375, 76153, + 82239, 88641, 95367, 102425, 109823, 117569, 125671, 134137, 142975, 152193, + 161799, 171801, 182207, 193025, 204263, 215929, 228031, 240577, 253575, + 267033, 280959, 295361, 310247, 325625, 341503, 357889, 374791, 392217, + 410175, 428673, 447719, 467321, 487487, 508225, 529543, 551449, 573951, + 597057, 620775, 645113, 670079, 695681, 721927, 748825, 776383, 804609, + 833511, 863097, 893375, 924353, 956039, 988441, 1021567, 1055425, 1090023, + 1125369, 1161471, 1198337, 1235975, 1274393, 1313599, 1353601, 1394407, + 1436025, 1478463, 1521729, 1565831, 1610777, 1656575, 1703233, 1750759, + 1799161, 1848447, 1898625, 1949703, 2001689, 2054591, 2108417, 2163175, + 2218873, 2275519, 2333121, 2391687, 2451225, 2511743, 2573249, 2635751, + 2699257, 2763775, 2829313, 2895879, 2963481, 3032127, 3101825, 3172583, + 3244409, 3317311, 3391297, 3466375, 3542553, 3619839, 3698241, 3777767, + 3858425, 3940223, 4023169, 4107271, 4192537, 4278975, 4366593, 4455399, + 4545401, 4636607, 4729025, 4822663, 4917529, 5013631, 5110977, 5209575, + 5309433, 5410559, 5512961, 5616647, 5721625, 5827903, 5935489, 6044391, + 6154617, 6266175, 6379073, 6493319, 6608921, 6725887, 6844225, 6963943, + 7085049, 7207551, +#if defined(CUSTOM_MODES) + /*...208:*/ + 7331457, 7456775, 7583513, 7711679, 7841281, 7972327, 8104825, 8238783, + 8374209, 8511111, 8649497, 8789375, 8930753, 9073639, 9218041, 9363967, + 9511425, 9660423, 9810969, 9963071, 10116737, 10271975, 10428793, 10587199, + 10747201, 10908807, 11072025, 11236863, 11403329, 11571431, 11741177, + 11912575, +#endif + /*N=5, K=5...176:*/ + 321, 681, 1289, 2241, 3649, 5641, 8361, 11969, 16641, 22569, 29961, 39041, + 50049, 63241, 78889, 97281, 118721, 143529, 172041, 204609, 241601, 283401, + 330409, 383041, 441729, 506921, 579081, 658689, 746241, 842249, 947241, + 1061761, 1186369, 1321641, 1468169, 1626561, 1797441, 1981449, 2179241, + 2391489, 2618881, 2862121, 3121929, 3399041, 3694209, 4008201, 4341801, + 4695809, 5071041, 5468329, 5888521, 6332481, 6801089, 7295241, 7815849, + 8363841, 8940161, 9545769, 10181641, 10848769, 11548161, 12280841, 13047849, + 13850241, 14689089, 15565481, 16480521, 17435329, 18431041, 19468809, + 20549801, 21675201, 22846209, 24064041, 25329929, 26645121, 28010881, + 29428489, 30899241, 32424449, 34005441, 35643561, 37340169, 39096641, + 40914369, 42794761, 44739241, 46749249, 48826241, 50971689, 53187081, + 55473921, 57833729, 60268041, 62778409, 65366401, 68033601, 70781609, + 73612041, 76526529, 79526721, 82614281, 85790889, 89058241, 92418049, + 95872041, 99421961, 103069569, 106816641, 110664969, 114616361, 118672641, + 122835649, 127107241, 131489289, 135983681, 140592321, 145317129, 150160041, + 155123009, 160208001, 165417001, 170752009, 176215041, 181808129, 187533321, + 193392681, 199388289, 205522241, 211796649, 218213641, 224775361, 231483969, + 238341641, 245350569, 252512961, 259831041, 267307049, 274943241, 282741889, + 290705281, 298835721, 307135529, 315607041, 324252609, 333074601, 342075401, + 351257409, 360623041, 370174729, 379914921, 389846081, 399970689, 410291241, + 420810249, 431530241, 442453761, 453583369, 464921641, 476471169, 488234561, + 500214441, 512413449, 524834241, 537479489, 550351881, 563454121, 576788929, + 590359041, 604167209, 618216201, 632508801, +#if defined(CUSTOM_MODES) + /*...208:*/ + 647047809, 661836041, 676876329, 692171521, 707724481, 723538089, 739615241, + 755958849, 772571841, 789457161, 806617769, 824056641, 841776769, 859781161, + 878072841, 896654849, 915530241, 934702089, 954173481, 973947521, 994027329, + 1014416041, 1035116809, 1056132801, 1077467201, 1099123209, 1121104041, + 1143412929, 1166053121, 1189027881, 1212340489, 1235994241, +#endif + /*N=6, K=6...96:*/ + 1683, 3653, 7183, 13073, 22363, 36365, 56695, 85305, 124515, 177045, 246047, + 335137, 448427, 590557, 766727, 982729, 1244979, 1560549, 1937199, 2383409, + 2908411, 3522221, 4235671, 5060441, 6009091, 7095093, 8332863, 9737793, + 11326283, 13115773, 15124775, 17372905, 19880915, 22670725, 25765455, + 29189457, 32968347, 37129037, 41699767, 46710137, 52191139, 58175189, + 64696159, 71789409, 79491819, 87841821, 96879431, 106646281, 117185651, + 128542501, 140763503, 153897073, 167993403, 183104493, 199284183, 216588185, + 235074115, 254801525, 275831935, 298228865, 322057867, 347386557, 374284647, + 402823977, 433078547, 465124549, 499040399, 534906769, 572806619, 612825229, + 655050231, 699571641, 746481891, 795875861, 847850911, 902506913, 959946283, + 1020274013, 1083597703, 1150027593, 1219676595, 1292660325, 1369097135, + 1449108145, 1532817275, 1620351277, 1711839767, 1807415257, 1907213187, + 2011371957, 2120032959, +#if defined(CUSTOM_MODES) + /*...109:*/ + 2233340609U, 2351442379U, 2474488829U, 2602633639U, 2736033641U, 2874848851U, + 3019242501U, 3169381071U, 3325434321U, 3487575323U, 3655980493U, 3830829623U, + 4012305913U, +#endif + /*N=7, K=7...54*/ + 8989, 19825, 40081, 75517, 134245, 227305, 369305, 579125, 880685, 1303777, + 1884961, 2668525, 3707509, 5064793, 6814249, 9041957, 11847485, 15345233, + 19665841, 24957661, 31388293, 39146185, 48442297, 59511829, 72616013, + 88043969, 106114625, 127178701, 151620757, 179861305, 212358985, 249612805, + 292164445, 340600625, 395555537, 457713341, 527810725, 606639529, 695049433, + 793950709, 904317037, 1027188385, 1163673953, 1314955181, 1482288821, + 1667010073, 1870535785, 2094367717, +#if defined(CUSTOM_MODES) + /*...60:*/ + 2340095869U, 2609401873U, 2904062449U, 3225952925U, 3577050821U, 3959439497U, +#endif + /*N=8, K=8...37*/ + 48639, 108545, 224143, 433905, 795455, 1392065, 2340495, 3800305, 5984767, + 9173505, 13726991, 20103025, 28875327, 40754369, 56610575, 77500017, + 104692735, 139703809, 184327311, 240673265, 311207743, 398796225, 506750351, + 638878193, 799538175, 993696769, 1226990095, 1505789553, 1837271615, + 2229491905U, +#if defined(CUSTOM_MODES) + /*...40:*/ + 2691463695U, 3233240945U, 3866006015U, +#endif + /*N=9, K=9...28:*/ + 265729, 598417, 1256465, 2485825, 4673345, 8405905, 14546705, 24331777, + 39490049, 62390545, 96220561, 145198913, 214828609, 312193553, 446304145, + 628496897, 872893441, 1196924561, 1621925137, 2173806145U, +#if defined(CUSTOM_MODES) + /*...29:*/ + 2883810113U, +#endif + /*N=10, K=10...24:*/ + 1462563, 3317445, 7059735, 14218905, 27298155, 50250765, 89129247, 152951073, + 254831667, 413442773, 654862247, 1014889769, 1541911931, 2300409629U, + 3375210671U, + /*N=11, K=11...19:*/ + 8097453, 18474633, 39753273, 81270333, 158819253, 298199265, 540279585, + 948062325, 1616336765, +#if defined(CUSTOM_MODES) + /*...20:*/ + 2684641785U, +#endif + /*N=12, K=12...18:*/ + 45046719, 103274625, 224298231, 464387817, 921406335, 1759885185, + 3248227095U, + /*N=13, K=13...16:*/ + 251595969, 579168825, 1267854873, 2653649025U, + /*N=14, K=14:*/ + 1409933619 +}; + +#if defined(CUSTOM_MODES) +static const opus_uint32 *const CELT_PVQ_U_ROW[15]={ + CELT_PVQ_U_DATA+ 0,CELT_PVQ_U_DATA+ 208,CELT_PVQ_U_DATA+ 415, + CELT_PVQ_U_DATA+ 621,CELT_PVQ_U_DATA+ 826,CELT_PVQ_U_DATA+1030, + CELT_PVQ_U_DATA+1233,CELT_PVQ_U_DATA+1336,CELT_PVQ_U_DATA+1389, + CELT_PVQ_U_DATA+1421,CELT_PVQ_U_DATA+1441,CELT_PVQ_U_DATA+1455, + CELT_PVQ_U_DATA+1464,CELT_PVQ_U_DATA+1470,CELT_PVQ_U_DATA+1473 +}; +#else +static const opus_uint32 *const CELT_PVQ_U_ROW[15]={ + CELT_PVQ_U_DATA+ 0,CELT_PVQ_U_DATA+ 176,CELT_PVQ_U_DATA+ 351, + CELT_PVQ_U_DATA+ 525,CELT_PVQ_U_DATA+ 698,CELT_PVQ_U_DATA+ 870, + CELT_PVQ_U_DATA+1041,CELT_PVQ_U_DATA+1131,CELT_PVQ_U_DATA+1178, + CELT_PVQ_U_DATA+1207,CELT_PVQ_U_DATA+1226,CELT_PVQ_U_DATA+1240, + CELT_PVQ_U_DATA+1248,CELT_PVQ_U_DATA+1254,CELT_PVQ_U_DATA+1257 +}; +#endif + +#if defined(CUSTOM_MODES) +void get_required_bits(opus_int16 *_bits,int _n,int _maxk,int _frac){ + int k; + /*_maxk==0 => there's nothing to do.*/ + celt_assert(_maxk>0); + _bits[0]=0; + for(k=1;k<=_maxk;k++)_bits[k]=log2_frac(CELT_PVQ_V(_n,k),_frac); +} +#endif + +static opus_uint32 icwrs(int _n,const int *_y){ + opus_uint32 i; + int j; + int k; + celt_assert(_n>=2); + j=_n-1; + i=_y[j]<0; + k=abs(_y[j]); + do{ + j--; + i+=CELT_PVQ_U(_n-j,k); + k+=abs(_y[j]); + if(_y[j]<0)i+=CELT_PVQ_U(_n-j,k+1); + } + while(j>0); + return i; +} + +void encode_pulses(const int *_y,int _n,int _k,ec_enc *_enc){ + celt_assert(_k>0); + ec_enc_uint(_enc,icwrs(_n,_y),CELT_PVQ_V(_n,_k)); +} + +static void cwrsi(int _n,int _k,opus_uint32 _i,int *_y){ + opus_uint32 p; + int s; + int k0; + celt_assert(_k>0); + celt_assert(_n>1); + while(_n>2){ + opus_uint32 q; + /*Lots of pulses case:*/ + if(_k>=_n){ + const opus_uint32 *row; + row=CELT_PVQ_U_ROW[_n]; + /*Are the pulses in this dimension negative?*/ + p=row[_k+1]; + s=-(_i>=p); + _i-=p&s; + /*Count how many pulses were placed in this dimension.*/ + k0=_k; + q=row[_n]; + if(q>_i){ + celt_assert(p>q); + _k=_n; + do p=CELT_PVQ_U_ROW[--_k][_n]; + while(p>_i); + } + else for(p=row[_k];p>_i;p=row[_k])_k--; + _i-=p; + *_y++=(k0-_k+s)^s; + } + /*Lots of dimensions case:*/ + else{ + /*Are there any pulses in this dimension at all?*/ + p=CELT_PVQ_U_ROW[_k][_n]; + q=CELT_PVQ_U_ROW[_k+1][_n]; + if(p<=_i&&_i<q){ + _i-=p; + *_y++=0; + } + else{ + /*Are the pulses in this dimension negative?*/ + s=-(_i>=q); + _i-=q&s; + /*Count how many pulses were placed in this dimension.*/ + k0=_k; + do p=CELT_PVQ_U_ROW[--_k][_n]; + while(p>_i); + _i-=p; + *_y++=(k0-_k+s)^s; + } + } + _n--; + } + /*_n==2*/ + p=2*_k+1; + s=-(_i>=p); + _i-=p&s; + k0=_k; + _k=(_i+1)>>1; + if(_k)_i-=2*_k-1; + *_y++=(k0-_k+s)^s; + /*_n==1*/ + s=-(int)_i; + *_y=(_k+s)^s; +} + +void decode_pulses(int *_y,int _n,int _k,ec_dec *_dec){ + cwrsi(_n,_k,ec_dec_uint(_dec,CELT_PVQ_V(_n,_k)),_y); +} + +#else /* SMALL_FOOTPRINT */ + +/*Computes the next row/column of any recurrence that obeys the relation + u[i][j]=u[i-1][j]+u[i][j-1]+u[i-1][j-1]. + _ui0 is the base case for the new row/column.*/ +static OPUS_INLINE void unext(opus_uint32 *_ui,unsigned _len,opus_uint32 _ui0){ + opus_uint32 ui1; + unsigned j; + /*This do-while will overrun the array if we don't have storage for at least + 2 values.*/ + j=1; do { + ui1=UADD32(UADD32(_ui[j],_ui[j-1]),_ui0); + _ui[j-1]=_ui0; + _ui0=ui1; + } while (++j<_len); + _ui[j-1]=_ui0; +} + +/*Computes the previous row/column of any recurrence that obeys the relation + u[i-1][j]=u[i][j]-u[i][j-1]-u[i-1][j-1]. + _ui0 is the base case for the new row/column.*/ +static OPUS_INLINE void uprev(opus_uint32 *_ui,unsigned _n,opus_uint32 _ui0){ + opus_uint32 ui1; + unsigned j; + /*This do-while will overrun the array if we don't have storage for at least + 2 values.*/ + j=1; do { + ui1=USUB32(USUB32(_ui[j],_ui[j-1]),_ui0); + _ui[j-1]=_ui0; + _ui0=ui1; + } while (++j<_n); + _ui[j-1]=_ui0; +} + +/*Compute V(_n,_k), as well as U(_n,0..._k+1). + _u: On exit, _u[i] contains U(_n,i) for i in [0..._k+1].*/ +static opus_uint32 ncwrs_urow(unsigned _n,unsigned _k,opus_uint32 *_u){ + opus_uint32 um2; + unsigned len; + unsigned k; + len=_k+2; + /*We require storage at least 3 values (e.g., _k>0).*/ + celt_assert(len>=3); + _u[0]=0; + _u[1]=um2=1; + /*If _n==0, _u[0] should be 1 and the rest should be 0.*/ + /*If _n==1, _u[i] should be 1 for i>1.*/ + celt_assert(_n>=2); + /*If _k==0, the following do-while loop will overflow the buffer.*/ + celt_assert(_k>0); + k=2; + do _u[k]=(k<<1)-1; + while(++k<len); + for(k=2;k<_n;k++)unext(_u+1,_k+1,1); + return _u[_k]+_u[_k+1]; +} + +/*Returns the _i'th combination of _k elements chosen from a set of size _n + with associated sign bits. + _y: Returns the vector of pulses. + _u: Must contain entries [0..._k+1] of row _n of U() on input. + Its contents will be destructively modified.*/ +static void cwrsi(int _n,int _k,opus_uint32 _i,int *_y,opus_uint32 *_u){ + int j; + celt_assert(_n>0); + j=0; + do{ + opus_uint32 p; + int s; + int yj; + p=_u[_k+1]; + s=-(_i>=p); + _i-=p&s; + yj=_k; + p=_u[_k]; + while(p>_i)p=_u[--_k]; + _i-=p; + yj-=_k; + _y[j]=(yj+s)^s; + uprev(_u,_k+2,0); + } + while(++j<_n); +} + +/*Returns the index of the given combination of K elements chosen from a set + of size 1 with associated sign bits. + _y: The vector of pulses, whose sum of absolute values is K. + _k: Returns K.*/ +static OPUS_INLINE opus_uint32 icwrs1(const int *_y,int *_k){ + *_k=abs(_y[0]); + return _y[0]<0; +} + +/*Returns the index of the given combination of K elements chosen from a set + of size _n with associated sign bits. + _y: The vector of pulses, whose sum of absolute values must be _k. + _nc: Returns V(_n,_k).*/ +static OPUS_INLINE opus_uint32 icwrs(int _n,int _k,opus_uint32 *_nc,const int *_y, + opus_uint32 *_u){ + opus_uint32 i; + int j; + int k; + /*We can't unroll the first two iterations of the loop unless _n>=2.*/ + celt_assert(_n>=2); + _u[0]=0; + for(k=1;k<=_k+1;k++)_u[k]=(k<<1)-1; + i=icwrs1(_y+_n-1,&k); + j=_n-2; + i+=_u[k]; + k+=abs(_y[j]); + if(_y[j]<0)i+=_u[k+1]; + while(j-->0){ + unext(_u,_k+2,0); + i+=_u[k]; + k+=abs(_y[j]); + if(_y[j]<0)i+=_u[k+1]; + } + *_nc=_u[k]+_u[k+1]; + return i; +} + +#ifdef CUSTOM_MODES +void get_required_bits(opus_int16 *_bits,int _n,int _maxk,int _frac){ + int k; + /*_maxk==0 => there's nothing to do.*/ + celt_assert(_maxk>0); + _bits[0]=0; + if (_n==1) + { + for (k=1;k<=_maxk;k++) + _bits[k] = 1<<_frac; + } + else { + VARDECL(opus_uint32,u); + SAVE_STACK; + ALLOC(u,_maxk+2U,opus_uint32); + ncwrs_urow(_n,_maxk,u); + for(k=1;k<=_maxk;k++) + _bits[k]=log2_frac(u[k]+u[k+1],_frac); + RESTORE_STACK; + } +} +#endif /* CUSTOM_MODES */ + +void encode_pulses(const int *_y,int _n,int _k,ec_enc *_enc){ + opus_uint32 i; + VARDECL(opus_uint32,u); + opus_uint32 nc; + SAVE_STACK; + celt_assert(_k>0); + ALLOC(u,_k+2U,opus_uint32); + i=icwrs(_n,_k,&nc,_y,u); + ec_enc_uint(_enc,i,nc); + RESTORE_STACK; +} + +void decode_pulses(int *_y,int _n,int _k,ec_dec *_dec){ + VARDECL(opus_uint32,u); + SAVE_STACK; + celt_assert(_k>0); + ALLOC(u,_k+2U,opus_uint32); + cwrsi(_n,_k,ec_dec_uint(_dec,ncwrs_urow(_n,_k,u)),_y,u); + RESTORE_STACK; +} + +#endif /* SMALL_FOOTPRINT */ diff --git a/src/opus-1.0.2/celt/cwrs.h b/src/opus-1.1/celt/cwrs.h index 7dfbd076..7dfbd076 100644 --- a/src/opus-1.0.2/celt/cwrs.h +++ b/src/opus-1.1/celt/cwrs.h diff --git a/src/opus-1.0.2/celt/ecintrin.h b/src/opus-1.1/celt/ecintrin.h index be57dd40..2263cff6 100644 --- a/src/opus-1.0.2/celt/ecintrin.h +++ b/src/opus-1.1/celt/ecintrin.h @@ -33,7 +33,7 @@ #if !defined(_ecintrin_H) # define _ecintrin_H (1) -/*Some specific platforms may have optimized intrinsic or inline assembly +/*Some specific platforms may have optimized intrinsic or OPUS_INLINE assembly versions of these functions which can substantially improve performance. We define macros for them to allow easy incorporation of these non-ANSI features.*/ diff --git a/src/opus-1.0.2/celt/entcode.c b/src/opus-1.1/celt/entcode.c index fa5d7c7c..fa5d7c7c 100644 --- a/src/opus-1.0.2/celt/entcode.c +++ b/src/opus-1.1/celt/entcode.c diff --git a/src/opus-1.0.2/celt/entcode.h b/src/opus-1.1/celt/entcode.h index aebecc06..dd13e49e 100644 --- a/src/opus-1.0.2/celt/entcode.h +++ b/src/opus-1.1/celt/entcode.h @@ -26,6 +26,7 @@ */ #include "opus_types.h" +#include "opus_defines.h" #if !defined(_entcode_H) # define _entcode_H (1) @@ -83,15 +84,15 @@ struct ec_ctx{ int error; }; -static inline opus_uint32 ec_range_bytes(ec_ctx *_this){ +static OPUS_INLINE opus_uint32 ec_range_bytes(ec_ctx *_this){ return _this->offs; } -static inline unsigned char *ec_get_buffer(ec_ctx *_this){ +static OPUS_INLINE unsigned char *ec_get_buffer(ec_ctx *_this){ return _this->buf; } -static inline int ec_get_error(ec_ctx *_this){ +static OPUS_INLINE int ec_get_error(ec_ctx *_this){ return _this->error; } @@ -101,7 +102,7 @@ static inline int ec_get_error(ec_ctx *_this){ Return: The number of bits. This will always be slightly larger than the exact value (e.g., all rounding error is in the positive direction).*/ -static inline int ec_tell(ec_ctx *_this){ +static OPUS_INLINE int ec_tell(ec_ctx *_this){ return _this->nbits_total-EC_ILOG(_this->rng); } diff --git a/src/opus-1.0.2/celt/entdec.c b/src/opus-1.1/celt/entdec.c index 75e3e45a..3c264685 100644 --- a/src/opus-1.0.2/celt/entdec.c +++ b/src/opus-1.1/celt/entdec.c @@ -85,7 +85,7 @@ number=3, pages="256--294", month=Jul, - URL="http://www.stanford.edu/class/ee398/handouts/papers/Moffat98ArithmCoding.pdf" + URL="http://www.stanford.edu/class/ee398a/handouts/papers/Moffat98ArithmCoding.pdf" }*/ static int ec_read_byte(ec_dec *_this){ diff --git a/src/opus-1.0.2/celt/entdec.h b/src/opus-1.1/celt/entdec.h index d8ab3187..d8ab3187 100644 --- a/src/opus-1.0.2/celt/entdec.h +++ b/src/opus-1.1/celt/entdec.h diff --git a/src/opus-1.0.2/celt/entenc.c b/src/opus-1.1/celt/entenc.c index a7e34ece..a7e34ece 100644 --- a/src/opus-1.0.2/celt/entenc.c +++ b/src/opus-1.1/celt/entenc.c diff --git a/src/opus-1.0.2/celt/entenc.h b/src/opus-1.1/celt/entenc.h index 796bc4d5..796bc4d5 100644 --- a/src/opus-1.0.2/celt/entenc.h +++ b/src/opus-1.1/celt/entenc.h diff --git a/src/opus-1.0.2/celt/fixed_debug.h b/src/opus-1.1/celt/fixed_debug.h index f11d890d..80bc9491 100644 --- a/src/opus-1.0.2/celt/fixed_debug.h +++ b/src/opus-1.1/celt/fixed_debug.h @@ -33,9 +33,9 @@ #define FIXED_DEBUG_H #include <stdio.h> +#include "opus_defines.h" #ifdef CELT_C -#include "opus_defines.h" OPUS_EXPORT opus_int64 celt_mips=0; #else extern opus_int64 celt_mips; @@ -59,7 +59,7 @@ extern opus_int64 celt_mips; #define SHR(a,b) SHR32(a,b) #define PSHR(a,b) PSHR32(a,b) -static inline short NEG16(int x) +static OPUS_INLINE short NEG16(int x) { int res; if (!VERIFY_SHORT(x)) @@ -80,7 +80,7 @@ static inline short NEG16(int x) celt_mips++; return res; } -static inline int NEG32(opus_int64 x) +static OPUS_INLINE int NEG32(opus_int64 x) { opus_int64 res; if (!VERIFY_INT(x)) @@ -103,7 +103,7 @@ static inline int NEG32(opus_int64 x) } #define EXTRACT16(x) EXTRACT16_(x, __FILE__, __LINE__) -static inline short EXTRACT16_(int x, char *file, int line) +static OPUS_INLINE short EXTRACT16_(int x, char *file, int line) { int res; if (!VERIFY_SHORT(x)) @@ -119,7 +119,7 @@ static inline short EXTRACT16_(int x, char *file, int line) } #define EXTEND32(x) EXTEND32_(x, __FILE__, __LINE__) -static inline int EXTEND32_(int x, char *file, int line) +static OPUS_INLINE int EXTEND32_(int x, char *file, int line) { int res; if (!VERIFY_SHORT(x)) @@ -135,7 +135,7 @@ static inline int EXTEND32_(int x, char *file, int line) } #define SHR16(a, shift) SHR16_(a, shift, __FILE__, __LINE__) -static inline short SHR16_(int a, int shift, char *file, int line) +static OPUS_INLINE short SHR16_(int a, int shift, char *file, int line) { int res; if (!VERIFY_SHORT(a) || !VERIFY_SHORT(shift)) @@ -157,7 +157,7 @@ static inline short SHR16_(int a, int shift, char *file, int line) return res; } #define SHL16(a, shift) SHL16_(a, shift, __FILE__, __LINE__) -static inline short SHL16_(int a, int shift, char *file, int line) +static OPUS_INLINE short SHL16_(int a, int shift, char *file, int line) { int res; if (!VERIFY_SHORT(a) || !VERIFY_SHORT(shift)) @@ -179,7 +179,7 @@ static inline short SHL16_(int a, int shift, char *file, int line) return res; } -static inline int SHR32(opus_int64 a, int shift) +static OPUS_INLINE int SHR32(opus_int64 a, int shift) { opus_int64 res; if (!VERIFY_INT(a) || !VERIFY_SHORT(shift)) @@ -201,7 +201,7 @@ static inline int SHR32(opus_int64 a, int shift) return res; } #define SHL32(a, shift) SHL32_(a, shift, __FILE__, __LINE__) -static inline int SHL32_(opus_int64 a, int shift, char *file, int line) +static OPUS_INLINE int SHL32_(opus_int64 a, int shift, char *file, int line) { opus_int64 res; if (!VERIFY_INT(a) || !VERIFY_SHORT(shift)) @@ -234,7 +234,7 @@ static inline int SHL32_(opus_int64 a, int shift, char *file, int line) //#define SHL(a,shift) ((a) << (shift)) #define ADD16(a, b) ADD16_(a, b, __FILE__, __LINE__) -static inline short ADD16_(int a, int b, char *file, int line) +static OPUS_INLINE short ADD16_(int a, int b, char *file, int line) { int res; if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) @@ -257,7 +257,7 @@ static inline short ADD16_(int a, int b, char *file, int line) } #define SUB16(a, b) SUB16_(a, b, __FILE__, __LINE__) -static inline short SUB16_(int a, int b, char *file, int line) +static OPUS_INLINE short SUB16_(int a, int b, char *file, int line) { int res; if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) @@ -280,7 +280,7 @@ static inline short SUB16_(int a, int b, char *file, int line) } #define ADD32(a, b) ADD32_(a, b, __FILE__, __LINE__) -static inline int ADD32_(opus_int64 a, opus_int64 b, char *file, int line) +static OPUS_INLINE int ADD32_(opus_int64 a, opus_int64 b, char *file, int line) { opus_int64 res; if (!VERIFY_INT(a) || !VERIFY_INT(b)) @@ -303,7 +303,7 @@ static inline int ADD32_(opus_int64 a, opus_int64 b, char *file, int line) } #define SUB32(a, b) SUB32_(a, b, __FILE__, __LINE__) -static inline int SUB32_(opus_int64 a, opus_int64 b, char *file, int line) +static OPUS_INLINE int SUB32_(opus_int64 a, opus_int64 b, char *file, int line) { opus_int64 res; if (!VERIFY_INT(a) || !VERIFY_INT(b)) @@ -327,7 +327,7 @@ static inline int SUB32_(opus_int64 a, opus_int64 b, char *file, int line) #undef UADD32 #define UADD32(a, b) UADD32_(a, b, __FILE__, __LINE__) -static inline unsigned int UADD32_(opus_uint64 a, opus_uint64 b, char *file, int line) +static OPUS_INLINE unsigned int UADD32_(opus_uint64 a, opus_uint64 b, char *file, int line) { opus_uint64 res; if (!VERIFY_UINT(a) || !VERIFY_UINT(b)) @@ -351,7 +351,7 @@ static inline unsigned int UADD32_(opus_uint64 a, opus_uint64 b, char *file, int #undef USUB32 #define USUB32(a, b) USUB32_(a, b, __FILE__, __LINE__) -static inline unsigned int USUB32_(opus_uint64 a, opus_uint64 b, char *file, int line) +static OPUS_INLINE unsigned int USUB32_(opus_uint64 a, opus_uint64 b, char *file, int line) { opus_uint64 res; if (!VERIFY_UINT(a) || !VERIFY_UINT(b)) @@ -381,7 +381,7 @@ static inline unsigned int USUB32_(opus_uint64 a, opus_uint64 b, char *file, int } /* result fits in 16 bits */ -static inline short MULT16_16_16(int a, int b) +static OPUS_INLINE short MULT16_16_16(int a, int b) { int res; if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) @@ -404,7 +404,7 @@ static inline short MULT16_16_16(int a, int b) } #define MULT16_16(a, b) MULT16_16_(a, b, __FILE__, __LINE__) -static inline int MULT16_16_(int a, int b, char *file, int line) +static OPUS_INLINE int MULT16_16_(int a, int b, char *file, int line) { opus_int64 res; if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) @@ -429,7 +429,7 @@ static inline int MULT16_16_(int a, int b, char *file, int line) #define MAC16_16(c,a,b) (celt_mips-=2,ADD32((c),MULT16_16((a),(b)))) #define MULT16_32_QX(a, b, Q) MULT16_32_QX_(a, b, Q, __FILE__, __LINE__) -static inline int MULT16_32_QX_(int a, opus_int64 b, int Q, char *file, int line) +static OPUS_INLINE int MULT16_32_QX_(int a, opus_int64 b, int Q, char *file, int line) { opus_int64 res; if (!VERIFY_SHORT(a) || !VERIFY_INT(b)) @@ -462,7 +462,7 @@ static inline int MULT16_32_QX_(int a, opus_int64 b, int Q, char *file, int line } #define MULT16_32_PX(a, b, Q) MULT16_32_PX_(a, b, Q, __FILE__, __LINE__) -static inline int MULT16_32_PX_(int a, opus_int64 b, int Q, char *file, int line) +static OPUS_INLINE int MULT16_32_PX_(int a, opus_int64 b, int Q, char *file, int line) { opus_int64 res; if (!VERIFY_SHORT(a) || !VERIFY_INT(b)) @@ -497,7 +497,7 @@ static inline int MULT16_32_PX_(int a, opus_int64 b, int Q, char *file, int line #define MULT16_32_Q15(a,b) MULT16_32_QX(a,b,15) #define MAC16_32_Q15(c,a,b) (celt_mips-=2,ADD32((c),MULT16_32_Q15((a),(b)))) -static inline int SATURATE(int a, int b) +static OPUS_INLINE int SATURATE(int a, int b) { if (a>b) a=b; @@ -507,7 +507,17 @@ static inline int SATURATE(int a, int b) return a; } -static inline int MULT16_16_Q11_32(int a, int b) +static OPUS_INLINE opus_int16 SATURATE16(opus_int32 a) +{ + celt_mips+=3; + if (a>32767) + return 32767; + else if (a<-32768) + return -32768; + else return a; +} + +static OPUS_INLINE int MULT16_16_Q11_32(int a, int b) { opus_int64 res; if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) @@ -529,7 +539,7 @@ static inline int MULT16_16_Q11_32(int a, int b) celt_mips+=3; return res; } -static inline short MULT16_16_Q13(int a, int b) +static OPUS_INLINE short MULT16_16_Q13(int a, int b) { opus_int64 res; if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) @@ -551,7 +561,7 @@ static inline short MULT16_16_Q13(int a, int b) celt_mips+=3; return res; } -static inline short MULT16_16_Q14(int a, int b) +static OPUS_INLINE short MULT16_16_Q14(int a, int b) { opus_int64 res; if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) @@ -575,7 +585,7 @@ static inline short MULT16_16_Q14(int a, int b) } #define MULT16_16_Q15(a, b) MULT16_16_Q15_(a, b, __FILE__, __LINE__) -static inline short MULT16_16_Q15_(int a, int b, char *file, int line) +static OPUS_INLINE short MULT16_16_Q15_(int a, int b, char *file, int line) { opus_int64 res; if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) @@ -598,7 +608,7 @@ static inline short MULT16_16_Q15_(int a, int b, char *file, int line) return res; } -static inline short MULT16_16_P13(int a, int b) +static OPUS_INLINE short MULT16_16_P13(int a, int b) { opus_int64 res; if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) @@ -628,7 +638,7 @@ static inline short MULT16_16_P13(int a, int b) celt_mips+=4; return res; } -static inline short MULT16_16_P14(int a, int b) +static OPUS_INLINE short MULT16_16_P14(int a, int b) { opus_int64 res; if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) @@ -658,7 +668,7 @@ static inline short MULT16_16_P14(int a, int b) celt_mips+=4; return res; } -static inline short MULT16_16_P15(int a, int b) +static OPUS_INLINE short MULT16_16_P15(int a, int b) { opus_int64 res; if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) @@ -691,7 +701,7 @@ static inline short MULT16_16_P15(int a, int b) #define DIV32_16(a, b) DIV32_16_(a, b, __FILE__, __LINE__) -static inline int DIV32_16_(opus_int64 a, opus_int64 b, char *file, int line) +static OPUS_INLINE int DIV32_16_(opus_int64 a, opus_int64 b, char *file, int line) { opus_int64 res; if (b==0) @@ -726,7 +736,7 @@ static inline int DIV32_16_(opus_int64 a, opus_int64 b, char *file, int line) } #define DIV32(a, b) DIV32_(a, b, __FILE__, __LINE__) -static inline int DIV32_(opus_int64 a, opus_int64 b, char *file, int line) +static OPUS_INLINE int DIV32_(opus_int64 a, opus_int64 b, char *file, int line) { opus_int64 res; if (b==0) diff --git a/src/opus-1.0.2/celt/fixed_generic.h b/src/opus-1.1/celt/fixed_generic.h index 71e28d62..ecf018a2 100644 --- a/src/opus-1.0.2/celt/fixed_generic.h +++ b/src/opus-1.1/celt/fixed_generic.h @@ -40,7 +40,7 @@ #define MULT16_32_Q16(a,b) ADD32(MULT16_16((a),SHR((b),16)), SHR(MULT16_16SU((a),((b)&0x0000ffff)),16)) /** 16x32 multiplication, followed by a 16-bit shift right (round-to-nearest). Results fits in 32 bits */ -#define MULT16_32_P16(a,b) ADD32(MULT16_16((a),SHR((b),16)), PSHR(MULT16_16((a),((b)&0x0000ffff)),16)) +#define MULT16_32_P16(a,b) ADD32(MULT16_16((a),SHR((b),16)), PSHR(MULT16_16SU((a),((b)&0x0000ffff)),16)) /** 16x32 multiplication, followed by a 15-bit shift right. Results fits in 32 bits */ #define MULT16_32_Q15(a,b) ADD32(SHL(MULT16_16((a),SHR((b),16)),1), SHR(MULT16_16SU((a),((b)&0x0000ffff)),15)) @@ -84,6 +84,8 @@ #define PSHR(a,shift) (SHR((a)+((EXTEND32(1)<<((shift))>>1)),shift)) #define SATURATE(x,a) (((x)>(a) ? (a) : (x)<-(a) ? -(a) : (x))) +#define SATURATE16(x) (EXTRACT16((x)>32767 ? 32767 : (x)<-32768 ? -32768 : (x))) + /** Shift by a and round-to-neareast 32-bit value. Result is a 16-bit value */ #define ROUND16(x,a) (EXTRACT16(PSHR32((x),(a)))) /** Divide by two */ @@ -108,10 +110,13 @@ /** 16x16 multiply-add where the result fits in 32 bits */ #define MAC16_16(c,a,b) (ADD32((c),MULT16_16((a),(b)))) -/** 16x32 multiply-add, followed by a 15-bit shift right. Results fits in 32 bits */ +/** 16x32 multiply, followed by a 15-bit shift right and 32-bit add. + b must fit in 31 bits. + Result fits in 32 bits. */ #define MAC16_32_Q15(c,a,b) ADD32(c,ADD32(MULT16_16((a),SHR((b),15)), SHR(MULT16_16((a),((b)&0x00007fff)),15))) #define MULT16_16_Q11_32(a,b) (SHR(MULT16_16((a),(b)),11)) +#define MULT16_16_Q11(a,b) (SHR(MULT16_16((a),(b)),11)) #define MULT16_16_Q13(a,b) (SHR(MULT16_16((a),(b)),13)) #define MULT16_16_Q14(a,b) (SHR(MULT16_16((a),(b)),14)) #define MULT16_16_Q15(a,b) (SHR(MULT16_16((a),(b)),15)) diff --git a/src/opus-1.0.2/celt/float_cast.h b/src/opus-1.1/celt/float_cast.h index 5ded2915..ede65748 100644 --- a/src/opus-1.0.2/celt/float_cast.h +++ b/src/opus-1.1/celt/float_cast.h @@ -101,7 +101,7 @@ #include <math.h> /* Win32 doesn't seem to have these functions. - ** Therefore implement inline versions of these functions here. + ** Therefore implement OPUS_INLINE versions of these functions here. */ __inline long int @@ -128,7 +128,7 @@ #endif #ifndef DISABLE_FLOAT_API -static inline opus_int16 FLOAT2INT16(float x) +static OPUS_INLINE opus_int16 FLOAT2INT16(float x) { x = x*CELT_SIG_SCALE; x = MAX32(x, -32768); diff --git a/src/opus-1.0.2/celt/kiss_fft.c b/src/opus-1.1/celt/kiss_fft.c index dcd69686..ad706c73 100644 --- a/src/opus-1.0.2/celt/kiss_fft.c +++ b/src/opus-1.1/celt/kiss_fft.c @@ -40,7 +40,6 @@ #include "os_support.h" #include "mathops.h" #include "stack_alloc.h" -#include "os_support.h" /* The guts header contains all the multiplication and addition macros that are defined for complex numbers. It also delares the kf_ internal functions. @@ -142,8 +141,6 @@ static void kf_bfly4( C_ADDTO(*Fout, scratch[1]); C_ADD( scratch[3] , scratch[0] , scratch[2] ); C_SUB( scratch[4] , scratch[0] , scratch[2] ); - Fout[m2].r = PSHR32(Fout[m2].r, 2); - Fout[m2].i = PSHR32(Fout[m2].i, 2); C_SUB( Fout[m2], *Fout, scratch[3] ); tw1 += fstride; tw2 += fstride*2; diff --git a/src/opus-1.0.2/celt/kiss_fft.h b/src/opus-1.1/celt/kiss_fft.h index 66332e3b..66332e3b 100644 --- a/src/opus-1.0.2/celt/kiss_fft.h +++ b/src/opus-1.1/celt/kiss_fft.h diff --git a/src/opus-1.0.2/celt/laplace.c b/src/opus-1.1/celt/laplace.c index a7bca874..a7bca874 100644 --- a/src/opus-1.0.2/celt/laplace.c +++ b/src/opus-1.1/celt/laplace.c diff --git a/src/opus-1.0.2/celt/laplace.h b/src/opus-1.1/celt/laplace.h index 46c14b5d..46c14b5d 100644 --- a/src/opus-1.0.2/celt/laplace.h +++ b/src/opus-1.1/celt/laplace.h diff --git a/src/opus-1.0.2/celt/mathops.c b/src/opus-1.1/celt/mathops.c index ce472c9f..3f8c5dcc 100644 --- a/src/opus-1.0.2/celt/mathops.c +++ b/src/opus-1.1/celt/mathops.c @@ -123,6 +123,8 @@ opus_val32 celt_sqrt(opus_val32 x) static const opus_val16 C[5] = {23175, 11561, -3011, 1699, -664}; if (x==0) return 0; + else if (x>=1073741824) + return 32767; k = (celt_ilog2(x)>>1)-7; x = VSHR32(x, 2*k); n = x-32768; @@ -137,7 +139,7 @@ opus_val32 celt_sqrt(opus_val32 x) #define L3 8277 #define L4 -626 -static inline opus_val16 _celt_cos_pi_2(opus_val16 x) +static OPUS_INLINE opus_val16 _celt_cos_pi_2(opus_val16 x) { opus_val16 x2; diff --git a/src/opus-1.0.2/celt/mathops.h b/src/opus-1.1/celt/mathops.h index 4e977956..a0525a96 100644 --- a/src/opus-1.0.2/celt/mathops.h +++ b/src/opus-1.1/celt/mathops.h @@ -43,6 +43,41 @@ unsigned isqrt32(opus_uint32 _val); +#ifndef OVERRIDE_CELT_MAXABS16 +static OPUS_INLINE opus_val32 celt_maxabs16(const opus_val16 *x, int len) +{ + int i; + opus_val16 maxval = 0; + opus_val16 minval = 0; + for (i=0;i<len;i++) + { + maxval = MAX16(maxval, x[i]); + minval = MIN16(minval, x[i]); + } + return MAX32(EXTEND32(maxval),-EXTEND32(minval)); +} +#endif + +#ifndef OVERRIDE_CELT_MAXABS32 +#ifdef FIXED_POINT +static OPUS_INLINE opus_val32 celt_maxabs32(const opus_val32 *x, int len) +{ + int i; + opus_val32 maxval = 0; + opus_val32 minval = 0; + for (i=0;i<len;i++) + { + maxval = MAX32(maxval, x[i]); + minval = MIN32(minval, x[i]); + } + return MAX32(maxval, -minval); +} +#else +#define celt_maxabs32(x,len) celt_maxabs16(x,len) +#endif +#endif + + #ifndef FIXED_POINT #define PI 3.141592653f @@ -60,7 +95,7 @@ unsigned isqrt32(opus_uint32 _val); denorm, +/- inf and NaN are *not* handled */ /** Base-2 log approximation (log2(x)). */ -static inline float celt_log2(float x) +static OPUS_INLINE float celt_log2(float x) { int integer; float frac; @@ -78,7 +113,7 @@ static inline float celt_log2(float x) } /** Base-2 exponential approximation (2^x). */ -static inline float celt_exp2(float x) +static OPUS_INLINE float celt_exp2(float x) { int integer; float frac; @@ -110,37 +145,16 @@ static inline float celt_exp2(float x) #ifndef OVERRIDE_CELT_ILOG2 /** Integer log in base2. Undefined for zero and negative numbers */ -static inline opus_int16 celt_ilog2(opus_int32 x) +static OPUS_INLINE opus_int16 celt_ilog2(opus_int32 x) { celt_assert2(x>0, "celt_ilog2() only defined for strictly positive numbers"); return EC_ILOG(x)-1; } #endif -#ifndef OVERRIDE_CELT_MAXABS16 -static inline opus_val16 celt_maxabs16(opus_val16 *x, int len) -{ - int i; - opus_val16 maxval = 0; - for (i=0;i<len;i++) - maxval = MAX16(maxval, ABS16(x[i])); - return maxval; -} -#endif - -#ifndef OVERRIDE_CELT_MAXABS32 -static inline opus_val32 celt_maxabs32(opus_val32 *x, int len) -{ - int i; - opus_val32 maxval = 0; - for (i=0;i<len;i++) - maxval = MAX32(maxval, ABS32(x[i])); - return maxval; -} -#endif /** Integer log in base2. Defined for zero, but not for negative numbers */ -static inline opus_int16 celt_zlog2(opus_val32 x) +static OPUS_INLINE opus_int16 celt_zlog2(opus_val32 x) { return x <= 0 ? 0 : celt_ilog2(x); } @@ -151,7 +165,8 @@ opus_val32 celt_sqrt(opus_val32 x); opus_val16 celt_cos_norm(opus_val32 x); -static inline opus_val16 celt_log2(opus_val32 x) +/** Base-2 logarithm approximation (log2(x)). (Q14 input, Q10 output) */ +static OPUS_INLINE opus_val16 celt_log2(opus_val32 x) { int i; opus_val16 n, frac; @@ -176,8 +191,15 @@ static inline opus_val16 celt_log2(opus_val32 x) #define D1 22804 #define D2 14819 #define D3 10204 + +static OPUS_INLINE opus_val32 celt_exp2_frac(opus_val16 x) +{ + opus_val16 frac; + frac = SHL16(x, 4); + return ADD16(D0, MULT16_16_Q15(frac, ADD16(D1, MULT16_16_Q15(frac, ADD16(D2 , MULT16_16_Q15(D3,frac)))))); +} /** Base-2 exponential approximation (2^x). (Q10 input, Q16 output) */ -static inline opus_val32 celt_exp2(opus_val16 x) +static OPUS_INLINE opus_val32 celt_exp2(opus_val16 x) { int integer; opus_val16 frac; @@ -186,8 +208,7 @@ static inline opus_val32 celt_exp2(opus_val16 x) return 0x7f000000; else if (integer < -15) return 0; - frac = SHL16(x-SHL16(integer,10),4); - frac = ADD16(D0, MULT16_16_Q15(frac, ADD16(D1, MULT16_16_Q15(frac, ADD16(D2 , MULT16_16_Q15(D3,frac)))))); + frac = celt_exp2_frac(x-SHL16(integer,10)); return VSHR32(EXTEND32(frac), -integer-2); } @@ -204,7 +225,7 @@ opus_val32 frac_div32(opus_val32 a, opus_val32 b); /* Atan approximation using a 4th order polynomial. Input is in Q15 format and normalized by pi/4. Output is in Q15 format */ -static inline opus_val16 celt_atan01(opus_val16 x) +static OPUS_INLINE opus_val16 celt_atan01(opus_val16 x) { return MULT16_16_P15(x, ADD32(M1, MULT16_16_P15(x, ADD32(M2, MULT16_16_P15(x, ADD32(M3, MULT16_16_P15(M4, x))))))); } @@ -215,7 +236,7 @@ static inline opus_val16 celt_atan01(opus_val16 x) #undef M4 /* atan2() approximation valid for positive input values */ -static inline opus_val16 celt_atan2p(opus_val16 y, opus_val16 x) +static OPUS_INLINE opus_val16 celt_atan2p(opus_val16 y, opus_val16 x) { if (y < x) { diff --git a/src/opus-1.0.2/celt/mdct.c b/src/opus-1.1/celt/mdct.c index 16a36c69..90a214ad 100644 --- a/src/opus-1.0.2/celt/mdct.c +++ b/src/opus-1.1/celt/mdct.c @@ -109,12 +109,14 @@ void clt_mdct_forward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar int N, N2, N4; kiss_twiddle_scalar sine; VARDECL(kiss_fft_scalar, f); + VARDECL(kiss_fft_scalar, f2); SAVE_STACK; N = l->n; N >>= shift; N2 = N>>1; N4 = N>>2; ALLOC(f, N2, kiss_fft_scalar); + ALLOC(f2, N2, kiss_fft_scalar); /* sin(x) ~= x here */ #ifdef FIXED_POINT sine = TRIG_UPSCALE*(QCONST16(0.7853981f, 15)+N2)/N; @@ -131,7 +133,7 @@ void clt_mdct_forward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar kiss_fft_scalar * OPUS_RESTRICT yp = f; const opus_val16 * OPUS_RESTRICT wp1 = window+(overlap>>1); const opus_val16 * OPUS_RESTRICT wp2 = window+(overlap>>1)-1; - for(i=0;i<(overlap>>2);i++) + for(i=0;i<((overlap+3)>>2);i++) { /* Real part arranged as -d-cR, Imag part arranged as -b+aR*/ *yp++ = MULT16_32_Q15(*wp2, xp1[N2]) + MULT16_32_Q15(*wp1,*xp2); @@ -143,7 +145,7 @@ void clt_mdct_forward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar } wp1 = window; wp2 = window+overlap-1; - for(;i<N4-(overlap>>2);i++) + for(;i<N4-((overlap+3)>>2);i++) { /* Real part arranged as a-bR, Imag part arranged as -c-dR */ *yp++ = *xp2; @@ -180,12 +182,12 @@ void clt_mdct_forward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar } /* N/4 complex FFT, down-scales by 4/N */ - opus_fft(l->kfft[shift], (kiss_fft_cpx *)f, (kiss_fft_cpx *)in); + opus_fft(l->kfft[shift], (kiss_fft_cpx *)f, (kiss_fft_cpx *)f2); /* Post-rotate */ { /* Temp pointers to make it really clear to the compiler what we're doing */ - const kiss_fft_scalar * OPUS_RESTRICT fp = in; + const kiss_fft_scalar * OPUS_RESTRICT fp = f2; kiss_fft_scalar * OPUS_RESTRICT yp1 = out; kiss_fft_scalar * OPUS_RESTRICT yp2 = out+stride*(N2-1); const kiss_twiddle_scalar *t = &l->trig[0]; @@ -212,14 +214,12 @@ void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scala int i; int N, N2, N4; kiss_twiddle_scalar sine; - VARDECL(kiss_fft_scalar, f); VARDECL(kiss_fft_scalar, f2); SAVE_STACK; N = l->n; N >>= shift; N2 = N>>1; N4 = N>>2; - ALLOC(f, N2, kiss_fft_scalar); ALLOC(f2, N2, kiss_fft_scalar); /* sin(x) ~= x here */ #ifdef FIXED_POINT @@ -249,81 +249,60 @@ void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scala } /* Inverse N/4 complex FFT. This one should *not* downscale even in fixed-point */ - opus_ifft(l->kfft[shift], (kiss_fft_cpx *)f2, (kiss_fft_cpx *)f); + opus_ifft(l->kfft[shift], (kiss_fft_cpx *)f2, (kiss_fft_cpx *)(out+(overlap>>1))); - /* Post-rotate */ + /* Post-rotate and de-shuffle from both ends of the buffer at once to make + it in-place. */ { - kiss_fft_scalar * OPUS_RESTRICT fp = f; + kiss_fft_scalar * OPUS_RESTRICT yp0 = out+(overlap>>1); + kiss_fft_scalar * OPUS_RESTRICT yp1 = out+(overlap>>1)+N2-2; const kiss_twiddle_scalar *t = &l->trig[0]; - - for(i=0;i<N4;i++) + /* Loop to (N4+1)>>1 to handle odd N4. When N4 is odd, the + middle pair will be computed twice. */ + for(i=0;i<(N4+1)>>1;i++) { kiss_fft_scalar re, im, yr, yi; - re = fp[0]; - im = fp[1]; + kiss_twiddle_scalar t0, t1; + re = yp0[0]; + im = yp0[1]; + t0 = t[i<<shift]; + t1 = t[(N4-i)<<shift]; /* We'd scale up by 2 here, but instead it's done when mixing the windows */ - yr = S_MUL(re,t[i<<shift]) - S_MUL(im,t[(N4-i)<<shift]); - yi = S_MUL(im,t[i<<shift]) + S_MUL(re,t[(N4-i)<<shift]); + yr = S_MUL(re,t0) - S_MUL(im,t1); + yi = S_MUL(im,t0) + S_MUL(re,t1); + re = yp1[0]; + im = yp1[1]; /* works because the cos is nearly one */ - *fp++ = yr - S_MUL(yi,sine); - *fp++ = yi + S_MUL(yr,sine); - } - } - /* De-shuffle the components for the middle of the window only */ - { - const kiss_fft_scalar * OPUS_RESTRICT fp1 = f; - const kiss_fft_scalar * OPUS_RESTRICT fp2 = f+N2-1; - kiss_fft_scalar * OPUS_RESTRICT yp = f2; - for(i = 0; i < N4; i++) - { - *yp++ =-*fp1; - *yp++ = *fp2; - fp1 += 2; - fp2 -= 2; + yp0[0] = -(yr - S_MUL(yi,sine)); + yp1[1] = yi + S_MUL(yr,sine); + + t0 = t[(N4-i-1)<<shift]; + t1 = t[(i+1)<<shift]; + /* We'd scale up by 2 here, but instead it's done when mixing the windows */ + yr = S_MUL(re,t0) - S_MUL(im,t1); + yi = S_MUL(im,t0) + S_MUL(re,t1); + /* works because the cos is nearly one */ + yp1[0] = -(yr - S_MUL(yi,sine)); + yp0[1] = yi + S_MUL(yr,sine); + yp0 += 2; + yp1 -= 2; } } - out -= (N2-overlap)>>1; + /* Mirror on both sides for TDAC */ { - kiss_fft_scalar * OPUS_RESTRICT fp1 = f2+N4-1; - kiss_fft_scalar * OPUS_RESTRICT xp1 = out+N2-1; - kiss_fft_scalar * OPUS_RESTRICT yp1 = out+N4-overlap/2; - const opus_val16 * OPUS_RESTRICT wp1 = window; - const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1; - for(i = 0; i< N4-overlap/2; i++) - { - *xp1 = *fp1; - xp1--; - fp1--; - } - for(; i < N4; i++) - { - kiss_fft_scalar x1; - x1 = *fp1--; - *yp1++ +=-MULT16_32_Q15(*wp1, x1); - *xp1-- += MULT16_32_Q15(*wp2, x1); - wp1++; - wp2--; - } - } - { - kiss_fft_scalar * OPUS_RESTRICT fp2 = f2+N4; - kiss_fft_scalar * OPUS_RESTRICT xp2 = out+N2; - kiss_fft_scalar * OPUS_RESTRICT yp2 = out+N-1-(N4-overlap/2); + kiss_fft_scalar * OPUS_RESTRICT xp1 = out+overlap-1; + kiss_fft_scalar * OPUS_RESTRICT yp1 = out; const opus_val16 * OPUS_RESTRICT wp1 = window; const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1; - for(i = 0; i< N4-overlap/2; i++) - { - *xp2 = *fp2; - xp2++; - fp2++; - } - for(; i < N4; i++) + + for(i = 0; i < overlap/2; i++) { - kiss_fft_scalar x2; - x2 = *fp2++; - *yp2-- = MULT16_32_Q15(*wp1, x2); - *xp2++ = MULT16_32_Q15(*wp2, x2); + kiss_fft_scalar x1, x2; + x1 = *xp1; + x2 = *yp1; + *yp1++ = MULT16_32_Q15(*wp2, x2) - MULT16_32_Q15(*wp1, x1); + *xp1-- = MULT16_32_Q15(*wp1, x2) + MULT16_32_Q15(*wp2, x1); wp1++; wp2--; } diff --git a/src/opus-1.0.2/celt/mdct.h b/src/opus-1.1/celt/mdct.h index d7218213..d7218213 100644 --- a/src/opus-1.0.2/celt/mdct.h +++ b/src/opus-1.1/celt/mdct.h diff --git a/src/opus-1.0.2/celt/mfrngcod.h b/src/opus-1.1/celt/mfrngcod.h index 809152a5..809152a5 100644 --- a/src/opus-1.0.2/celt/mfrngcod.h +++ b/src/opus-1.1/celt/mfrngcod.h diff --git a/src/opus-1.0.2/celt/modes.c b/src/opus-1.1/celt/modes.c index ed204d7d..42e68e1c 100644 --- a/src/opus-1.0.2/celt/modes.c +++ b/src/opus-1.1/celt/modes.c @@ -345,6 +345,14 @@ CELTMode *opus_custom_mode_create(opus_int32 Fs, int frame_size, int *error) mode->eBands = compute_ebands(Fs, mode->shortMdctSize, res, &mode->nbEBands); if (mode->eBands==NULL) goto failure; +#if !defined(SMALL_FOOTPRINT) + /* Make sure we don't allocate a band larger than our PVQ table. + 208 should be enough, but let's be paranoid. */ + if ((mode->eBands[mode->nbEBands] - mode->eBands[mode->nbEBands-1])<<LM > + 208) { + goto failure; + } +#endif mode->effEBands = mode->nbEBands; while (mode->eBands[mode->effEBands] > mode->shortMdctSize) diff --git a/src/opus-1.0.2/celt/modes.h b/src/opus-1.1/celt/modes.h index c8340f98..c8340f98 100644 --- a/src/opus-1.0.2/celt/modes.h +++ b/src/opus-1.1/celt/modes.h diff --git a/src/opus-1.0.2/celt/opus_custom_demo.c b/src/opus-1.1/celt/opus_custom_demo.c index ae41c0de..ae41c0de 100644 --- a/src/opus-1.0.2/celt/opus_custom_demo.c +++ b/src/opus-1.1/celt/opus_custom_demo.c diff --git a/src/opus-1.0.2/celt/os_support.h b/src/opus-1.1/celt/os_support.h index 2484f0b2..5e47e3cf 100644 --- a/src/opus-1.0.2/celt/os_support.h +++ b/src/opus-1.1/celt/os_support.h @@ -35,13 +35,16 @@ # include "custom_support.h" #endif +#include "opus_types.h" +#include "opus_defines.h" + #include <string.h> #include <stdio.h> #include <stdlib.h> /** Opus wrapper for malloc(). To do your own dynamic allocation, all you need to do is replace this function and opus_free */ #ifndef OVERRIDE_OPUS_ALLOC -static inline void *opus_alloc (size_t size) +static OPUS_INLINE void *opus_alloc (size_t size) { return malloc(size); } @@ -49,7 +52,7 @@ static inline void *opus_alloc (size_t size) /** Same as celt_alloc(), except that the area is only needed inside a CELT call (might cause problem with wideband though) */ #ifndef OVERRIDE_OPUS_ALLOC_SCRATCH -static inline void *opus_alloc_scratch (size_t size) +static OPUS_INLINE void *opus_alloc_scratch (size_t size) { /* Scratch space doesn't need to be cleared */ return opus_alloc(size); @@ -58,7 +61,7 @@ static inline void *opus_alloc_scratch (size_t size) /** Opus wrapper for free(). To do your own dynamic allocation, all you need to do is replace this function and opus_alloc */ #ifndef OVERRIDE_OPUS_FREE -static inline void opus_free (void *ptr) +static OPUS_INLINE void opus_free (void *ptr) { free(ptr); } diff --git a/src/opus-1.0.2/celt/pitch.c b/src/opus-1.1/celt/pitch.c index ca0f523e..d2b30544 100644 --- a/src/opus-1.0.2/celt/pitch.c +++ b/src/opus-1.1/celt/pitch.c @@ -102,13 +102,57 @@ static void find_best_pitch(opus_val32 *xcorr, opus_val16 *y, int len, } } +static void celt_fir5(const opus_val16 *x, + const opus_val16 *num, + opus_val16 *y, + int N, + opus_val16 *mem) +{ + int i; + opus_val16 num0, num1, num2, num3, num4; + opus_val32 mem0, mem1, mem2, mem3, mem4; + num0=num[0]; + num1=num[1]; + num2=num[2]; + num3=num[3]; + num4=num[4]; + mem0=mem[0]; + mem1=mem[1]; + mem2=mem[2]; + mem3=mem[3]; + mem4=mem[4]; + for (i=0;i<N;i++) + { + opus_val32 sum = SHL32(EXTEND32(x[i]), SIG_SHIFT); + sum = MAC16_16(sum,num0,mem0); + sum = MAC16_16(sum,num1,mem1); + sum = MAC16_16(sum,num2,mem2); + sum = MAC16_16(sum,num3,mem3); + sum = MAC16_16(sum,num4,mem4); + mem4 = mem3; + mem3 = mem2; + mem2 = mem1; + mem1 = mem0; + mem0 = x[i]; + y[i] = ROUND16(sum, SIG_SHIFT); + } + mem[0]=mem0; + mem[1]=mem1; + mem[2]=mem2; + mem[3]=mem3; + mem[4]=mem4; +} + + void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x_lp, - int len, int C) + int len, int C, int arch) { int i; opus_val32 ac[5]; opus_val16 tmp=Q15ONE; - opus_val16 lpc[4], mem[4]={0,0,0,0}; + opus_val16 lpc[4], mem[5]={0,0,0,0,0}; + opus_val16 lpc2[5]; + opus_val16 c1 = QCONST16(.8f,15); #ifdef FIXED_POINT int shift; opus_val32 maxabs = celt_maxabs32(x[0], len); @@ -136,7 +180,7 @@ void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x } _celt_autocorr(x_lp, ac, NULL, 0, - 4, len>>1); + 4, len>>1, arch); /* Noise floor -40 dB */ #ifdef FIXED_POINT @@ -161,16 +205,96 @@ void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x tmp = MULT16_16_Q15(QCONST16(.9f,15), tmp); lpc[i] = MULT16_16_Q15(lpc[i], tmp); } - celt_fir(x_lp, lpc, x_lp, len>>1, 4, mem); + /* Add a zero */ + lpc2[0] = lpc[0] + QCONST16(.8f,SIG_SHIFT); + lpc2[1] = lpc[1] + MULT16_16_Q15(c1,lpc[0]); + lpc2[2] = lpc[2] + MULT16_16_Q15(c1,lpc[1]); + lpc2[3] = lpc[3] + MULT16_16_Q15(c1,lpc[2]); + lpc2[4] = MULT16_16_Q15(c1,lpc[3]); + celt_fir5(x_lp, lpc2, x_lp, len>>1, mem); +} - mem[0]=0; - lpc[0]=QCONST16(.8f,12); - celt_fir(x_lp, lpc, x_lp, len>>1, 1, mem); +#if 0 /* This is a simple version of the pitch correlation that should work + well on DSPs like Blackfin and TI C5x/C6x */ +#ifdef FIXED_POINT +opus_val32 +#else +void +#endif +celt_pitch_xcorr(opus_val16 *x, opus_val16 *y, opus_val32 *xcorr, int len, int max_pitch) +{ + int i, j; +#ifdef FIXED_POINT + opus_val32 maxcorr=1; +#endif + for (i=0;i<max_pitch;i++) + { + opus_val32 sum = 0; + for (j=0;j<len;j++) + sum = MAC16_16(sum, x[j],y[i+j]); + xcorr[i] = sum; +#ifdef FIXED_POINT + maxcorr = MAX32(maxcorr, sum); +#endif + } +#ifdef FIXED_POINT + return maxcorr; +#endif +} + +#else /* Unrolled version of the pitch correlation -- runs faster on x86 and ARM */ + +#ifdef FIXED_POINT +opus_val32 +#else +void +#endif +celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, opus_val32 *xcorr, int len, int max_pitch) +{ + int i,j; + /*The EDSP version requires that max_pitch is at least 1, and that _x is + 32-bit aligned. + Since it's hard to put asserts in assembly, put them here.*/ + celt_assert(max_pitch>0); + celt_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0); +#ifdef FIXED_POINT + opus_val32 maxcorr=1; +#endif + for (i=0;i<max_pitch-3;i+=4) + { + opus_val32 sum[4]={0,0,0,0}; + xcorr_kernel(_x, _y+i, sum, len); + xcorr[i]=sum[0]; + xcorr[i+1]=sum[1]; + xcorr[i+2]=sum[2]; + xcorr[i+3]=sum[3]; +#ifdef FIXED_POINT + sum[0] = MAX32(sum[0], sum[1]); + sum[2] = MAX32(sum[2], sum[3]); + sum[0] = MAX32(sum[0], sum[2]); + maxcorr = MAX32(maxcorr, sum[0]); +#endif + } + /* In case max_pitch isn't a multiple of 4, do non-unrolled version. */ + for (;i<max_pitch;i++) + { + opus_val32 sum = 0; + for (j=0;j<len;j++) + sum = MAC16_16(sum, _x[j],_y[i+j]); + xcorr[i] = sum; +#ifdef FIXED_POINT + maxcorr = MAX32(maxcorr, sum); +#endif + } +#ifdef FIXED_POINT + return maxcorr; +#endif } +#endif void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTRICT y, - int len, int max_pitch, int *pitch) + int len, int max_pitch, int *pitch, int arch) { int i, j; int lag; @@ -179,8 +303,8 @@ void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTR VARDECL(opus_val16, y_lp4); VARDECL(opus_val32, xcorr); #ifdef FIXED_POINT - opus_val32 maxcorr=1; - opus_val16 xmax, ymax; + opus_val32 maxcorr; + opus_val32 xmax, ymax; int shift=0; #endif int offset; @@ -204,7 +328,7 @@ void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTR #ifdef FIXED_POINT xmax = celt_maxabs16(x_lp4, len>>2); ymax = celt_maxabs16(y_lp4, lag>>2); - shift = celt_ilog2(MAX16(1, MAX16(xmax, ymax)))-11; + shift = celt_ilog2(MAX32(1, MAX32(xmax, ymax)))-11; if (shift>0) { for (j=0;j<len>>2;j++) @@ -220,16 +344,11 @@ void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTR /* Coarse search with 4x decimation */ - for (i=0;i<max_pitch>>2;i++) - { - opus_val32 sum = 0; - for (j=0;j<len>>2;j++) - sum = MAC16_16(sum, x_lp4[j],y_lp4[i+j]); - xcorr[i] = MAX32(-1, sum); #ifdef FIXED_POINT - maxcorr = MAX32(maxcorr, sum); + maxcorr = #endif - } + celt_pitch_xcorr(x_lp4, y_lp4, xcorr, len>>2, max_pitch>>2, arch); + find_best_pitch(xcorr, y_lp4, len>>2, max_pitch>>2, best_pitch #ifdef FIXED_POINT , 0, maxcorr @@ -287,11 +406,13 @@ opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod, int k, i, T, T0; opus_val16 g, g0; opus_val16 pg; - opus_val32 xy,xx,yy; + opus_val32 xy,xx,yy,xy2; opus_val32 xcorr[3]; opus_val32 best_xy, best_yy; int offset; int minperiod0; + VARDECL(opus_val32, yy_lookup); + SAVE_STACK; minperiod0 = minperiod; maxperiod /= 2; @@ -304,13 +425,16 @@ opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod, *T0_=maxperiod-1; T = T0 = *T0_; - xx=xy=yy=0; - for (i=0;i<N;i++) + ALLOC(yy_lookup, maxperiod+1, opus_val32); + dual_inner_prod(x, x, x-T0, N, &xx, &xy); + yy_lookup[0] = xx; + yy=xx; + for (i=1;i<=maxperiod;i++) { - xy = MAC16_16(xy, x[i], x[i-T0]); - xx = MAC16_16(xx, x[i], x[i]); - yy = MAC16_16(yy, x[i-T0],x[i-T0]); + yy = yy+MULT16_16(x[-i],x[-i])-MULT16_16(x[N-i],x[N-i]); + yy_lookup[i] = MAX32(0, yy); } + yy = yy_lookup[T0]; best_xy = xy; best_yy = yy; #ifdef FIXED_POINT @@ -331,6 +455,7 @@ opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod, int T1, T1b; opus_val16 g1; opus_val16 cont=0; + opus_val16 thresh; T1 = (2*T0+k)/(2*k); if (T1 < minperiod) break; @@ -345,15 +470,9 @@ opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod, { T1b = (2*second_check[k]*T0+k)/(2*k); } - xy=yy=0; - for (i=0;i<N;i++) - { - xy = MAC16_16(xy, x[i], x[i-T1]); - yy = MAC16_16(yy, x[i-T1], x[i-T1]); - - xy = MAC16_16(xy, x[i], x[i-T1b]); - yy = MAC16_16(yy, x[i-T1b], x[i-T1b]); - } + dual_inner_prod(x, &x[-T1], &x[-T1b], N, &xy, &xy2); + xy += xy2; + yy = yy_lookup[T1] + yy_lookup[T1b]; #ifdef FIXED_POINT { opus_val32 x2y2; @@ -372,7 +491,14 @@ opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod, cont = HALF32(prev_gain); else cont = 0; - if (g1 > QCONST16(.3f,15) + MULT16_16_Q15(QCONST16(.4f,15),g0)-cont) + thresh = MAX16(QCONST16(.3f,15), MULT16_16_Q15(QCONST16(.7f,15),g0)-cont); + /* Bias against very high pitch (very short period) to avoid false-positives + due to short-term correlation */ + if (T1<3*minperiod) + thresh = MAX16(QCONST16(.4f,15), MULT16_16_Q15(QCONST16(.85f,15),g0)-cont); + else if (T1<2*minperiod) + thresh = MAX16(QCONST16(.5f,15), MULT16_16_Q15(QCONST16(.9f,15),g0)-cont); + if (g1 > thresh) { best_xy = xy; best_yy = yy; @@ -406,5 +532,6 @@ opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod, if (*T0_<minperiod0) *T0_=minperiod0; + RESTORE_STACK; return pg; } diff --git a/src/opus-1.1/celt/pitch.h b/src/opus-1.1/celt/pitch.h new file mode 100644 index 00000000..df317ecc --- /dev/null +++ b/src/opus-1.1/celt/pitch.h @@ -0,0 +1,173 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file pitch.h + @brief Pitch analysis + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef PITCH_H +#define PITCH_H + +#include "modes.h" +#include "cpu_support.h" + +#if defined(__SSE__) && !defined(FIXED_POINT) +#include "x86/pitch_sse.h" +#endif + +#if defined(OPUS_ARM_ASM) && defined(FIXED_POINT) +# include "arm/pitch_arm.h" +#endif + +void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x_lp, + int len, int C, int arch); + +void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTRICT y, + int len, int max_pitch, int *pitch, int arch); + +opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod, + int N, int *T0, int prev_period, opus_val16 prev_gain); + +/* OPT: This is the kernel you really want to optimize. It gets used a lot + by the prefilter and by the PLC. */ +#ifndef OVERRIDE_XCORR_KERNEL +static OPUS_INLINE void xcorr_kernel(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[4], int len) +{ + int j; + opus_val16 y_0, y_1, y_2, y_3; + celt_assert(len>=3); + y_3=0; /* gcc doesn't realize that y_3 can't be used uninitialized */ + y_0=*y++; + y_1=*y++; + y_2=*y++; + for (j=0;j<len-3;j+=4) + { + opus_val16 tmp; + tmp = *x++; + y_3=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_0); + sum[1] = MAC16_16(sum[1],tmp,y_1); + sum[2] = MAC16_16(sum[2],tmp,y_2); + sum[3] = MAC16_16(sum[3],tmp,y_3); + tmp=*x++; + y_0=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_1); + sum[1] = MAC16_16(sum[1],tmp,y_2); + sum[2] = MAC16_16(sum[2],tmp,y_3); + sum[3] = MAC16_16(sum[3],tmp,y_0); + tmp=*x++; + y_1=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_2); + sum[1] = MAC16_16(sum[1],tmp,y_3); + sum[2] = MAC16_16(sum[2],tmp,y_0); + sum[3] = MAC16_16(sum[3],tmp,y_1); + tmp=*x++; + y_2=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_3); + sum[1] = MAC16_16(sum[1],tmp,y_0); + sum[2] = MAC16_16(sum[2],tmp,y_1); + sum[3] = MAC16_16(sum[3],tmp,y_2); + } + if (j++<len) + { + opus_val16 tmp = *x++; + y_3=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_0); + sum[1] = MAC16_16(sum[1],tmp,y_1); + sum[2] = MAC16_16(sum[2],tmp,y_2); + sum[3] = MAC16_16(sum[3],tmp,y_3); + } + if (j++<len) + { + opus_val16 tmp=*x++; + y_0=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_1); + sum[1] = MAC16_16(sum[1],tmp,y_2); + sum[2] = MAC16_16(sum[2],tmp,y_3); + sum[3] = MAC16_16(sum[3],tmp,y_0); + } + if (j<len) + { + opus_val16 tmp=*x++; + y_1=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_2); + sum[1] = MAC16_16(sum[1],tmp,y_3); + sum[2] = MAC16_16(sum[2],tmp,y_0); + sum[3] = MAC16_16(sum[3],tmp,y_1); + } +} +#endif /* OVERRIDE_XCORR_KERNEL */ + +#ifndef OVERRIDE_DUAL_INNER_PROD +static OPUS_INLINE void dual_inner_prod(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2) +{ + int i; + opus_val32 xy01=0; + opus_val32 xy02=0; + for (i=0;i<N;i++) + { + xy01 = MAC16_16(xy01, x[i], y01[i]); + xy02 = MAC16_16(xy02, x[i], y02[i]); + } + *xy1 = xy01; + *xy2 = xy02; +} +#endif + +#ifdef FIXED_POINT +opus_val32 +#else +void +#endif +celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch); + +#if !defined(OVERRIDE_PITCH_XCORR) +/*Is run-time CPU detection enabled on this platform?*/ +# if defined(OPUS_HAVE_RTCD) +extern +# if defined(FIXED_POINT) +opus_val32 +# else +void +# endif +(*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *, + const opus_val16 *, opus_val32 *, int, int); + +# define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \ + ((*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK])(_x, _y, \ + xcorr, len, max_pitch)) +# else +# define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \ + ((void)(arch),celt_pitch_xcorr_c(_x, _y, xcorr, len, max_pitch)) +# endif +#endif + +#endif diff --git a/src/opus-1.0.2/celt/quant_bands.c b/src/opus-1.1/celt/quant_bands.c index 66f1f5fc..ac6952c2 100644 --- a/src/opus-1.0.2/celt/quant_bands.c +++ b/src/opus-1.1/celt/quant_bands.c @@ -40,8 +40,8 @@ #include "rate.h" #ifdef FIXED_POINT -/* Mean energy in each band quantized in Q6 */ -static const signed char eMeans[25] = { +/* Mean energy in each band quantized in Q4 */ +const signed char eMeans[25] = { 103,100, 92, 85, 81, 77, 72, 70, 78, 75, 73, 71, 78, 74, 69, @@ -49,8 +49,8 @@ static const signed char eMeans[25] = { 60, 60, 60, 60, 60 }; #else -/* Mean energy in each band quantized in Q6 and converted back to float */ -static const opus_val16 eMeans[25] = { +/* Mean energy in each band quantized in Q4 and converted back to float */ +const opus_val16 eMeans[25] = { 6.437500f, 6.250000f, 5.750000f, 5.312500f, 5.062500f, 4.812500f, 4.500000f, 4.375000f, 4.875000f, 4.687500f, 4.562500f, 4.437500f, 4.875000f, 4.625000f, 4.312500f, @@ -157,7 +157,7 @@ static int quant_coarse_energy_impl(const CELTMode *m, int start, int end, const opus_val16 *eBands, opus_val16 *oldEBands, opus_int32 budget, opus_int32 tell, const unsigned char *prob_model, opus_val16 *error, ec_enc *enc, - int C, int LM, int intra, opus_val16 max_decay) + int C, int LM, int intra, opus_val16 max_decay, int lfe) { int i, c; int badness = 0; @@ -222,6 +222,8 @@ static int quant_coarse_energy_impl(const CELTMode *m, int start, int end, if (bits_left < 16) qi = IMAX(-1, qi); } + if (lfe && i>=2) + qi = IMIN(qi, 0); if (budget-tell >= 15) { int pi; @@ -253,13 +255,13 @@ static int quant_coarse_energy_impl(const CELTMode *m, int start, int end, prev[c] = prev[c] + SHL32(q,7) - MULT16_16(beta,PSHR32(q,8)); } while (++c < C); } - return badness; + return lfe ? 0 : badness; } void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd, const opus_val16 *eBands, opus_val16 *oldEBands, opus_uint32 budget, opus_val16 *error, ec_enc *enc, int C, int LM, int nbAvailableBytes, - int force_intra, opus_val32 *delayedIntra, int two_pass, int loss_rate) + int force_intra, opus_val32 *delayedIntra, int two_pass, int loss_rate, int lfe) { int intra; opus_val16 max_decay; @@ -280,9 +282,6 @@ void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd, if (tell+3 > budget) two_pass = intra = 0; - /* Encode the global flags using a simple probability model - (first symbols in the stream) */ - max_decay = QCONST16(16.f,DB_SHIFT); if (end-start>10) { @@ -292,6 +291,8 @@ void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd, max_decay = MIN32(max_decay, .125f*nbAvailableBytes); #endif } + if (lfe) + max_decay=3; enc_start_state = *enc; ALLOC(oldEBands_intra, C*m->nbEBands, opus_val16); @@ -301,7 +302,7 @@ void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd, if (two_pass || intra) { badness1 = quant_coarse_energy_impl(m, start, end, eBands, oldEBands_intra, budget, - tell, e_prob_model[LM][1], error_intra, enc, C, LM, 1, max_decay); + tell, e_prob_model[LM][1], error_intra, enc, C, LM, 1, max_decay, lfe); } if (!intra) @@ -311,6 +312,7 @@ void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd, opus_int32 tell_intra; opus_uint32 nstart_bytes; opus_uint32 nintra_bytes; + opus_uint32 save_bytes; int badness2; VARDECL(unsigned char, intra_bits); @@ -321,14 +323,17 @@ void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd, nstart_bytes = ec_range_bytes(&enc_start_state); nintra_bytes = ec_range_bytes(&enc_intra_state); intra_buf = ec_get_buffer(&enc_intra_state) + nstart_bytes; - ALLOC(intra_bits, nintra_bytes-nstart_bytes, unsigned char); + save_bytes = nintra_bytes-nstart_bytes; + if (save_bytes == 0) + save_bytes = ALLOC_NONE; + ALLOC(intra_bits, save_bytes, unsigned char); /* Copy bits from intra bit-stream */ OPUS_COPY(intra_bits, intra_buf, nintra_bytes - nstart_bytes); *enc = enc_start_state; badness2 = quant_coarse_energy_impl(m, start, end, eBands, oldEBands, budget, - tell, e_prob_model[LM][intra], error, enc, C, LM, 0, max_decay); + tell, e_prob_model[LM][intra], error, enc, C, LM, 0, max_decay, lfe); if (two_pass && (badness1 < badness2 || (badness1 == badness2 && ((opus_int32)ec_tell_frac(enc))+intra_bias > tell_intra))) { @@ -535,25 +540,6 @@ void unquant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 * } } -void log2Amp(const CELTMode *m, int start, int end, - celt_ener *eBands, const opus_val16 *oldEBands, int C) -{ - int c, i; - c=0; - do { - for (i=0;i<start;i++) - eBands[i+c*m->nbEBands] = 0; - for (;i<end;i++) - { - opus_val16 lg = ADD16(oldEBands[i+c*m->nbEBands], - SHL16((opus_val16)eMeans[i],6)); - eBands[i+c*m->nbEBands] = PSHR32(celt_exp2(lg),4); - } - for (;i<m->nbEBands;i++) - eBands[i+c*m->nbEBands] = 0; - } while (++c < C); -} - void amp2Log2(const CELTMode *m, int effEnd, int end, celt_ener *bandE, opus_val16 *bandLogE, int C) { diff --git a/src/opus-1.0.2/celt/quant_bands.h b/src/opus-1.1/celt/quant_bands.h index bec2855c..0490bca4 100644 --- a/src/opus-1.0.2/celt/quant_bands.h +++ b/src/opus-1.1/celt/quant_bands.h @@ -35,6 +35,12 @@ #include "entdec.h" #include "mathops.h" +#ifdef FIXED_POINT +extern const signed char eMeans[25]; +#else +extern const opus_val16 eMeans[25]; +#endif + void amp2Log2(const CELTMode *m, int effEnd, int end, celt_ener *bandE, opus_val16 *bandLogE, int C); @@ -45,7 +51,7 @@ void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd, const opus_val16 *eBands, opus_val16 *oldEBands, opus_uint32 budget, opus_val16 *error, ec_enc *enc, int C, int LM, int nbAvailableBytes, int force_intra, opus_val32 *delayedIntra, - int two_pass, int loss_rate); + int two_pass, int loss_rate, int lfe); void quant_fine_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, opus_val16 *error, int *fine_quant, ec_enc *enc, int C); diff --git a/src/opus-1.0.2/celt/rate.c b/src/opus-1.1/celt/rate.c index 4e96787f..e13d839d 100644 --- a/src/opus-1.0.2/celt/rate.c +++ b/src/opus-1.1/celt/rate.c @@ -245,10 +245,10 @@ void compute_pulse_cache(CELTMode *m, int LM) #define ALLOC_STEPS 6 -static inline int interp_bits2pulses(const CELTMode *m, int start, int end, int skip_start, +static OPUS_INLINE int interp_bits2pulses(const CELTMode *m, int start, int end, int skip_start, const int *bits1, const int *bits2, const int *thresh, const int *cap, opus_int32 total, opus_int32 *_balance, int skip_rsv, int *intensity, int intensity_rsv, int *dual_stereo, int dual_stereo_rsv, int *bits, - int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev) + int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev, int signalBandwidth) { opus_int32 psum; int lo, hi; @@ -353,7 +353,7 @@ static inline int interp_bits2pulses(const CELTMode *m, int start, int end, int #ifdef FUZZING if ((rand()&0x1) == 0) #else - if (codedBands<=start+2 || band_bits > ((j<prev?7:9)*band_width<<LM<<BITRES)>>4) + if (codedBands<=start+2 || (band_bits > ((j<prev?7:9)*band_width<<LM<<BITRES)>>4 && j<=signalBandwidth)) #endif { ec_enc_bit_logp(ec, 1, 1); @@ -524,7 +524,7 @@ static inline int interp_bits2pulses(const CELTMode *m, int start, int end, int } int compute_allocation(const CELTMode *m, int start, int end, const int *offsets, const int *cap, int alloc_trim, int *intensity, int *dual_stereo, - opus_int32 total, opus_int32 *balance, int *pulses, int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev) + opus_int32 total, opus_int32 *balance, int *pulses, int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev, int signalBandwidth) { int lo, hi, len, j; int codedBands; @@ -631,7 +631,7 @@ int compute_allocation(const CELTMode *m, int start, int end, const int *offsets } codedBands = interp_bits2pulses(m, start, end, skip_start, bits1, bits2, thresh, cap, total, balance, skip_rsv, intensity, intensity_rsv, dual_stereo, dual_stereo_rsv, - pulses, ebits, fine_priority, C, LM, ec, encode, prev); + pulses, ebits, fine_priority, C, LM, ec, encode, prev, signalBandwidth); RESTORE_STACK; return codedBands; } diff --git a/src/opus-1.0.2/celt/rate.h b/src/opus-1.1/celt/rate.h index e0d50223..f1e06611 100644 --- a/src/opus-1.0.2/celt/rate.h +++ b/src/opus-1.1/celt/rate.h @@ -45,12 +45,12 @@ void compute_pulse_cache(CELTMode *m, int LM); -static inline int get_pulses(int i) +static OPUS_INLINE int get_pulses(int i) { return i<8 ? i : (8 + (i&7)) << ((i>>3)-1); } -static inline int bits2pulses(const CELTMode *m, int band, int LM, int bits) +static OPUS_INLINE int bits2pulses(const CELTMode *m, int band, int LM, int bits) { int i; int lo, hi; @@ -77,7 +77,7 @@ static inline int bits2pulses(const CELTMode *m, int band, int LM, int bits) return hi; } -static inline int pulses2bits(const CELTMode *m, int band, int LM, int pulses) +static OPUS_INLINE int pulses2bits(const CELTMode *m, int band, int LM, int pulses) { const unsigned char *cache; @@ -96,6 +96,6 @@ static inline int pulses2bits(const CELTMode *m, int band, int LM, int pulses) @return Total number of bits allocated */ int compute_allocation(const CELTMode *m, int start, int end, const int *offsets, const int *cap, int alloc_trim, int *intensity, int *dual_stero, - opus_int32 total, opus_int32 *balance, int *pulses, int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev); + opus_int32 total, opus_int32 *balance, int *pulses, int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev, int signalBandwidth); #endif diff --git a/src/opus-1.0.2/celt/stack_alloc.h b/src/opus-1.1/celt/stack_alloc.h index a6f06d22..316a6ce1 100644 --- a/src/opus-1.0.2/celt/stack_alloc.h +++ b/src/opus-1.1/celt/stack_alloc.h @@ -32,6 +32,9 @@ #ifndef STACK_ALLOC_H #define STACK_ALLOC_H +#include "opus_types.h" +#include "opus_defines.h" + #if (!defined (VAR_ARRAYS) && !defined (USE_ALLOCA) && !defined (NONTHREADSAFE_PSEUDOSTACK)) #error "Opus requires one of VAR_ARRAYS, USE_ALLOCA, or NONTHREADSAFE_PSEUDOSTACK be defined to select the temporary allocation mode." #endif @@ -92,6 +95,8 @@ #define SAVE_STACK #define RESTORE_STACK #define ALLOC_STACK +/* C99 does not allow VLAs of size zero */ +#define ALLOC_NONE 1 #elif defined(USE_ALLOCA) @@ -106,6 +111,7 @@ #define SAVE_STACK #define RESTORE_STACK #define ALLOC_STACK +#define ALLOC_NONE 0 #else @@ -143,7 +149,30 @@ extern char *global_stack_top; #define VARDECL(type, var) type *var #define ALLOC(var, size, type) var = PUSH(global_stack, size, type) #define SAVE_STACK char *_saved_stack = global_stack; +#define ALLOC_NONE 0 #endif /* VAR_ARRAYS */ + +#ifdef ENABLE_VALGRIND + +#include <valgrind/memcheck.h> +#define OPUS_CHECK_ARRAY(ptr, len) VALGRIND_CHECK_MEM_IS_DEFINED(ptr, len*sizeof(*ptr)) +#define OPUS_CHECK_VALUE(value) VALGRIND_CHECK_VALUE_IS_DEFINED(value) +#define OPUS_CHECK_ARRAY_COND(ptr, len) VALGRIND_CHECK_MEM_IS_DEFINED(ptr, len*sizeof(*ptr)) +#define OPUS_CHECK_VALUE_COND(value) VALGRIND_CHECK_VALUE_IS_DEFINED(value) +#define OPUS_PRINT_INT(value) do {fprintf(stderr, #value " = %d at %s:%d\n", value, __FILE__, __LINE__);}while(0) +#define OPUS_FPRINTF fprintf + +#else + +static OPUS_INLINE int _opus_false(void) {return 0;} +#define OPUS_CHECK_ARRAY(ptr, len) _opus_false() +#define OPUS_CHECK_VALUE(value) _opus_false() +#define OPUS_PRINT_INT(value) do{}while(0) +#define OPUS_FPRINTF (void) + +#endif + + #endif /* STACK_ALLOC_H */ diff --git a/src/opus-1.0.2/celt/static_modes_fixed.h b/src/opus-1.1/celt/static_modes_fixed.h index 216df9e6..216df9e6 100644 --- a/src/opus-1.0.2/celt/static_modes_fixed.h +++ b/src/opus-1.1/celt/static_modes_fixed.h diff --git a/src/opus-1.0.2/celt/static_modes_float.h b/src/opus-1.1/celt/static_modes_float.h index 5d7e7b8e..5d7e7b8e 100644 --- a/src/opus-1.0.2/celt/static_modes_float.h +++ b/src/opus-1.1/celt/static_modes_float.h diff --git a/src/opus-1.0.2/celt/vq.c b/src/opus-1.1/celt/vq.c index 98a0f36c..98a0f36c 100644 --- a/src/opus-1.0.2/celt/vq.c +++ b/src/opus-1.1/celt/vq.c diff --git a/src/opus-1.0.2/celt/vq.h b/src/opus-1.1/celt/vq.h index 1ceeeeb2..ffdc69cd 100644 --- a/src/opus-1.0.2/celt/vq.h +++ b/src/opus-1.1/celt/vq.h @@ -40,11 +40,9 @@ /** Algebraic pulse-vector quantiser. The signal x is replaced by the sum of * the pitch and a combination of pulses such that its norm is still equal * to 1. This is the function that will typically require the most CPU. - * @param x Residual signal to quantise/encode (returns quantised version) - * @param W Perceptual weight to use when optimising (currently unused) + * @param X Residual signal to quantise/encode (returns quantised version) * @param N Number of samples to encode * @param K Number of pulses to use - * @param p Pitch vector (it is assumed that p+x is a unit vector) * @param enc Entropy encoder state * @ret A mask indicating which blocks in the band received pulses */ @@ -56,10 +54,9 @@ unsigned alg_quant(celt_norm *X, int N, int K, int spread, int B, ); /** Algebraic pulse decoder - * @param x Decoded normalised spectrum (returned) + * @param X Decoded normalised spectrum (returned) * @param N Number of samples to decode * @param K Number of pulses to use - * @param p Pitch vector (automatically added to x) * @param dec Entropy decoder state * @ret A mask indicating which blocks in the band received pulses */ diff --git a/src/opus-1.1/celt/x86/pitch_sse.h b/src/opus-1.1/celt/x86/pitch_sse.h new file mode 100644 index 00000000..695122a5 --- /dev/null +++ b/src/opus-1.1/celt/x86/pitch_sse.h @@ -0,0 +1,156 @@ +/* Copyright (c) 2013 Jean-Marc Valin and John Ridges */ +/** + @file pitch_sse.h + @brief Pitch analysis + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef PITCH_SSE_H +#define PITCH_SSE_H + +#include <xmmintrin.h> +#include "arch.h" + +#define OVERRIDE_XCORR_KERNEL +static OPUS_INLINE void xcorr_kernel(const opus_val16 *x, const opus_val16 *y, opus_val32 sum[4], int len) +{ + int j; + __m128 xsum1, xsum2; + xsum1 = _mm_loadu_ps(sum); + xsum2 = _mm_setzero_ps(); + + for (j = 0; j < len-3; j += 4) + { + __m128 x0 = _mm_loadu_ps(x+j); + __m128 yj = _mm_loadu_ps(y+j); + __m128 y3 = _mm_loadu_ps(y+j+3); + + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x00),yj)); + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x55), + _mm_shuffle_ps(yj,y3,0x49))); + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xaa), + _mm_shuffle_ps(yj,y3,0x9e))); + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xff),y3)); + } + if (j < len) + { + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j))); + if (++j < len) + { + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j))); + if (++j < len) + { + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j))); + } + } + } + _mm_storeu_ps(sum,_mm_add_ps(xsum1,xsum2)); +} + +#define OVERRIDE_DUAL_INNER_PROD +static OPUS_INLINE void dual_inner_prod(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2) +{ + int i; + __m128 xsum1, xsum2; + xsum1 = _mm_setzero_ps(); + xsum2 = _mm_setzero_ps(); + for (i=0;i<N-3;i+=4) + { + __m128 xi = _mm_loadu_ps(x+i); + __m128 y1i = _mm_loadu_ps(y01+i); + __m128 y2i = _mm_loadu_ps(y02+i); + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(xi, y1i)); + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(xi, y2i)); + } + /* Horizontal sum */ + xsum1 = _mm_add_ps(xsum1, _mm_movehl_ps(xsum1, xsum1)); + xsum1 = _mm_add_ss(xsum1, _mm_shuffle_ps(xsum1, xsum1, 0x55)); + _mm_store_ss(xy1, xsum1); + xsum2 = _mm_add_ps(xsum2, _mm_movehl_ps(xsum2, xsum2)); + xsum2 = _mm_add_ss(xsum2, _mm_shuffle_ps(xsum2, xsum2, 0x55)); + _mm_store_ss(xy2, xsum2); + for (;i<N;i++) + { + *xy1 = MAC16_16(*xy1, x[i], y01[i]); + *xy2 = MAC16_16(*xy2, x[i], y02[i]); + } +} + +#define OVERRIDE_COMB_FILTER_CONST +static OPUS_INLINE void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N, + opus_val16 g10, opus_val16 g11, opus_val16 g12) +{ + int i; + __m128 x0v; + __m128 g10v, g11v, g12v; + g10v = _mm_load1_ps(&g10); + g11v = _mm_load1_ps(&g11); + g12v = _mm_load1_ps(&g12); + x0v = _mm_loadu_ps(&x[-T-2]); + for (i=0;i<N-3;i+=4) + { + __m128 yi, yi2, x1v, x2v, x3v, x4v; + const opus_val32 *xp = &x[i-T-2]; + yi = _mm_loadu_ps(x+i); + x4v = _mm_loadu_ps(xp+4); +#if 0 + /* Slower version with all loads */ + x1v = _mm_loadu_ps(xp+1); + x2v = _mm_loadu_ps(xp+2); + x3v = _mm_loadu_ps(xp+3); +#else + x2v = _mm_shuffle_ps(x0v, x4v, 0x4e); + x1v = _mm_shuffle_ps(x0v, x2v, 0x99); + x3v = _mm_shuffle_ps(x2v, x4v, 0x99); +#endif + + yi = _mm_add_ps(yi, _mm_mul_ps(g10v,x2v)); +#if 0 /* Set to 1 to make it bit-exact with the non-SSE version */ + yi = _mm_add_ps(yi, _mm_mul_ps(g11v,_mm_add_ps(x3v,x1v))); + yi = _mm_add_ps(yi, _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v))); +#else + /* Use partial sums */ + yi2 = _mm_add_ps(_mm_mul_ps(g11v,_mm_add_ps(x3v,x1v)), + _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v))); + yi = _mm_add_ps(yi, yi2); +#endif + x0v=x4v; + _mm_storeu_ps(y+i, yi); + } +#ifdef CUSTOM_MODES + for (;i<N;i++) + { + y[i] = x[i] + + MULT16_32_Q15(g10,x[i-T]) + + MULT16_32_Q15(g11,ADD32(x[i-T+1],x[i-T-1])) + + MULT16_32_Q15(g12,ADD32(x[i-T+2],x[i-T-2])); + } +#endif +} + +#endif diff --git a/src/opus-1.0.2/include/opus.h b/src/opus-1.1/include/opus.h index 847a07c1..93a53a2f 100644 --- a/src/opus-1.0.2/include/opus.h +++ b/src/opus-1.1/include/opus.h @@ -520,7 +520,7 @@ OPUS_EXPORT void opus_decoder_destroy(OpusDecoder *st); * @param [in] len <tt>opus_int32</tt>: size of data * @param [out] out_toc <tt>char*</tt>: TOC pointer * @param [out] frames <tt>char*[48]</tt> encapsulated frames - * @param [out] size <tt>short[48]</tt> sizes of the encapsulated frames + * @param [out] size <tt>opus_int16[48]</tt> sizes of the encapsulated frames * @param [out] payload_offset <tt>int*</tt>: returns the position of the payload within the packet (in bytes) * @returns number of frames */ @@ -529,7 +529,7 @@ OPUS_EXPORT int opus_packet_parse( opus_int32 len, unsigned char *out_toc, const unsigned char *frames[48], - short size[48], + opus_int16 size[48], int *payload_offset ) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(4); @@ -566,6 +566,7 @@ OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_packet_get_nb_channels(const unsign * @param [in] packet <tt>char*</tt>: Opus packet * @param [in] len <tt>opus_int32</tt>: Length of packet * @returns Number of frames + * @retval OPUS_BAD_ARG Insufficient data was passed to the function * @retval OPUS_INVALID_PACKET The compressed data passed is corrupted or of an unsupported type */ OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_packet_get_nb_frames(const unsigned char packet[], opus_int32 len) OPUS_ARG_NONNULL(1); @@ -577,6 +578,7 @@ OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_packet_get_nb_frames(const unsigned * This must be a multiple of 400, or * inaccurate results will be returned. * @returns Number of samples + * @retval OPUS_BAD_ARG Insufficient data was passed to the function * @retval OPUS_INVALID_PACKET The compressed data passed is corrupted or of an unsupported type */ OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_packet_get_nb_samples(const unsigned char packet[], opus_int32 len, opus_int32 Fs) OPUS_ARG_NONNULL(1); @@ -586,9 +588,24 @@ OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_packet_get_nb_samples(const unsigne * @param [in] packet <tt>char*</tt>: Opus packet * @param [in] len <tt>opus_int32</tt>: Length of packet * @returns Number of samples + * @retval OPUS_BAD_ARG Insufficient data was passed to the function * @retval OPUS_INVALID_PACKET The compressed data passed is corrupted or of an unsupported type */ OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_decoder_get_nb_samples(const OpusDecoder *dec, const unsigned char packet[], opus_int32 len) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2); + +/** Applies soft-clipping to bring a float signal within the [-1,1] range. If + * the signal is already in that range, nothing is done. If there are values + * outside of [-1,1], then the signal is clipped as smoothly as possible to + * both fit in the range and avoid creating excessive distortion in the + * process. + * @param [in,out] pcm <tt>float*</tt>: Input PCM and modified PCM + * @param [in] frame_size <tt>int</tt> Number of samples per channel to process + * @param [in] channels <tt>int</tt>: Number of channels + * @param [in,out] softclip_mem <tt>float*</tt>: State memory for the soft clipping process (one float per channel, initialized to zero) + */ +OPUS_EXPORT void opus_pcm_soft_clip(float *pcm, int frame_size, int channels, float *softclip_mem); + + /**@}*/ /** @defgroup opus_repacketizer Repacketizer @@ -894,6 +911,64 @@ OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_repacketizer_get_nb_frames(OpusRepa */ OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_repacketizer_out(OpusRepacketizer *rp, unsigned char *data, opus_int32 maxlen) OPUS_ARG_NONNULL(1); +/** Pads a given Opus packet to a larger size (possibly changing the TOC sequence). + * @param[in,out] data <tt>const unsigned char*</tt>: The buffer containing the + * packet to pad. + * @param len <tt>opus_int32</tt>: The size of the packet. + * This must be at least 1. + * @param new_len <tt>opus_int32</tt>: The desired size of the packet after padding. + * This must be at least as large as len. + * @returns an error code + * @retval #OPUS_OK \a on success. + * @retval #OPUS_BAD_ARG \a len was less than 1 or new_len was less than len. + * @retval #OPUS_INVALID_PACKET \a data did not contain a valid Opus packet. + */ +OPUS_EXPORT int opus_packet_pad(unsigned char *data, opus_int32 len, opus_int32 new_len); + +/** Remove all padding from a given Opus packet and rewrite the TOC sequence to + * minimize space usage. + * @param[in,out] data <tt>const unsigned char*</tt>: The buffer containing the + * packet to strip. + * @param len <tt>opus_int32</tt>: The size of the packet. + * This must be at least 1. + * @returns The new size of the output packet on success, or an error code + * on failure. + * @retval #OPUS_BAD_ARG \a len was less than 1. + * @retval #OPUS_INVALID_PACKET \a data did not contain a valid Opus packet. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_packet_unpad(unsigned char *data, opus_int32 len); + +/** Pads a given Opus multi-stream packet to a larger size (possibly changing the TOC sequence). + * @param[in,out] data <tt>const unsigned char*</tt>: The buffer containing the + * packet to pad. + * @param len <tt>opus_int32</tt>: The size of the packet. + * This must be at least 1. + * @param new_len <tt>opus_int32</tt>: The desired size of the packet after padding. + * This must be at least 1. + * @param nb_streams <tt>opus_int32</tt>: The number of streams (not channels) in the packet. + * This must be at least as large as len. + * @returns an error code + * @retval #OPUS_OK \a on success. + * @retval #OPUS_BAD_ARG \a len was less than 1. + * @retval #OPUS_INVALID_PACKET \a data did not contain a valid Opus packet. + */ +OPUS_EXPORT int opus_multistream_packet_pad(unsigned char *data, opus_int32 len, opus_int32 new_len, int nb_streams); + +/** Remove all padding from a given Opus multi-stream packet and rewrite the TOC sequence to + * minimize space usage. + * @param[in,out] data <tt>const unsigned char*</tt>: The buffer containing the + * packet to strip. + * @param len <tt>opus_int32</tt>: The size of the packet. + * This must be at least 1. + * @param nb_streams <tt>opus_int32</tt>: The number of streams (not channels) in the packet. + * This must be at least 1. + * @returns The new size of the output packet on success, or an error code + * on failure. + * @retval #OPUS_BAD_ARG \a len was less than 1 or new_len was less than len. + * @retval #OPUS_INVALID_PACKET \a data did not contain a valid Opus packet. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_multistream_packet_unpad(unsigned char *data, opus_int32 len, int nb_streams); + /**@}*/ #ifdef __cplusplus diff --git a/src/opus-1.0.2/include/opus_custom.h b/src/opus-1.1/include/opus_custom.h index e7861d6f..41f36bf2 100644 --- a/src/opus-1.0.2/include/opus_custom.h +++ b/src/opus-1.1/include/opus_custom.h @@ -42,15 +42,15 @@ extern "C" { #endif #ifdef CUSTOM_MODES -#define OPUS_CUSTOM_EXPORT OPUS_EXPORT -#define OPUS_CUSTOM_EXPORT_STATIC OPUS_EXPORT +# define OPUS_CUSTOM_EXPORT OPUS_EXPORT +# define OPUS_CUSTOM_EXPORT_STATIC OPUS_EXPORT #else -#define OPUS_CUSTOM_EXPORT -#ifdef CELT_C -#define OPUS_CUSTOM_EXPORT_STATIC static inline -#else -#define OPUS_CUSTOM_EXPORT_STATIC -#endif +# define OPUS_CUSTOM_EXPORT +# ifdef OPUS_BUILD +# define OPUS_CUSTOM_EXPORT_STATIC static OPUS_INLINE +# else +# define OPUS_CUSTOM_EXPORT_STATIC +# endif #endif /** @defgroup opus_custom Opus Custom @@ -126,6 +126,9 @@ OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT OpusCustomMode *opus_custom_mode_crea */ OPUS_CUSTOM_EXPORT void opus_custom_mode_destroy(OpusCustomMode *mode); + +#if !defined(OPUS_BUILD) || defined(CELT_ENCODER_C) + /* Encoder */ /** Gets the size of an OpusCustomEncoder structure. * @param [in] mode <tt>OpusCustomMode *</tt>: Mode configuration @@ -137,21 +140,7 @@ OPUS_CUSTOM_EXPORT_STATIC OPUS_WARN_UNUSED_RESULT int opus_custom_encoder_get_si int channels ) OPUS_ARG_NONNULL(1); -/** Creates a new encoder state. Each stream needs its own encoder - * state (can't be shared across simultaneous streams). - * @param [in] mode <tt>OpusCustomMode*</tt>: Contains all the information about the characteristics of - * the stream (must be the same characteristics as used for the - * decoder) - * @param [in] channels <tt>int</tt>: Number of channels - * @param [out] error <tt>int*</tt>: Returns an error code - * @return Newly created encoder state. -*/ -OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT OpusCustomEncoder *opus_custom_encoder_create( - const OpusCustomMode *mode, - int channels, - int *error -) OPUS_ARG_NONNULL(1); - +# ifdef CUSTOM_MODES /** Initializes a previously allocated encoder state * The memory pointed to by st must be the size returned by opus_custom_encoder_get_size. * This is intended for applications which use their own allocator instead of malloc. @@ -164,11 +153,30 @@ OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT OpusCustomEncoder *opus_custom_encode * @param [in] channels <tt>int</tt>: Number of channels * @return OPUS_OK Success or @ref opus_errorcodes */ -OPUS_CUSTOM_EXPORT_STATIC int opus_custom_encoder_init( +OPUS_CUSTOM_EXPORT int opus_custom_encoder_init( OpusCustomEncoder *st, const OpusCustomMode *mode, int channels ) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2); +# endif +#endif + + +/** Creates a new encoder state. Each stream needs its own encoder + * state (can't be shared across simultaneous streams). + * @param [in] mode <tt>OpusCustomMode*</tt>: Contains all the information about the characteristics of + * the stream (must be the same characteristics as used for the + * decoder) + * @param [in] channels <tt>int</tt>: Number of channels + * @param [out] error <tt>int*</tt>: Returns an error code + * @return Newly created encoder state. +*/ +OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT OpusCustomEncoder *opus_custom_encoder_create( + const OpusCustomMode *mode, + int channels, + int *error +) OPUS_ARG_NONNULL(1); + /** Destroys a an encoder state. * @param[in] st <tt>OpusCustomEncoder*</tt>: State to be freed. @@ -229,6 +237,8 @@ OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT int opus_custom_encode( */ OPUS_CUSTOM_EXPORT int opus_custom_encoder_ctl(OpusCustomEncoder * OPUS_RESTRICT st, int request, ...) OPUS_ARG_NONNULL(1); + +#if !defined(OPUS_BUILD) || defined(CELT_DECODER_C) /* Decoder */ /** Gets the size of an OpusCustomDecoder structure. @@ -241,20 +251,6 @@ OPUS_CUSTOM_EXPORT_STATIC OPUS_WARN_UNUSED_RESULT int opus_custom_decoder_get_si int channels ) OPUS_ARG_NONNULL(1); -/** Creates a new decoder state. Each stream needs its own decoder state (can't - * be shared across simultaneous streams). - * @param [in] mode <tt>OpusCustomMode</tt>: Contains all the information about the characteristics of the - * stream (must be the same characteristics as used for the encoder) - * @param [in] channels <tt>int</tt>: Number of channels - * @param [out] error <tt>int*</tt>: Returns an error code - * @return Newly created decoder state. - */ -OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT OpusCustomDecoder *opus_custom_decoder_create( - const OpusCustomMode *mode, - int channels, - int *error -) OPUS_ARG_NONNULL(1); - /** Initializes a previously allocated decoder state * The memory pointed to by st must be the size returned by opus_custom_decoder_get_size. * This is intended for applications which use their own allocator instead of malloc. @@ -273,6 +269,23 @@ OPUS_CUSTOM_EXPORT_STATIC int opus_custom_decoder_init( int channels ) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2); +#endif + + +/** Creates a new decoder state. Each stream needs its own decoder state (can't + * be shared across simultaneous streams). + * @param [in] mode <tt>OpusCustomMode</tt>: Contains all the information about the characteristics of the + * stream (must be the same characteristics as used for the encoder) + * @param [in] channels <tt>int</tt>: Number of channels + * @param [out] error <tt>int*</tt>: Returns an error code + * @return Newly created decoder state. + */ +OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT OpusCustomDecoder *opus_custom_decoder_create( + const OpusCustomMode *mode, + int channels, + int *error +) OPUS_ARG_NONNULL(1); + /** Destroys a an decoder state. * @param[in] st <tt>OpusCustomDecoder*</tt>: State to be freed. */ diff --git a/src/opus-1.0.2/include/opus_defines.h b/src/opus-1.1/include/opus_defines.h index cdde061a..265089f6 100644 --- a/src/opus-1.0.2/include/opus_defines.h +++ b/src/opus-1.1/include/opus_defines.h @@ -64,14 +64,14 @@ extern "C" { /**Export control for opus functions */ #ifndef OPUS_EXPORT -# if defined(__GNUC__) && defined(OPUS_BUILD) -# define OPUS_EXPORT __attribute__ ((visibility ("default"))) -# elif defined(WIN32) && !defined(__MINGW32__) +# if defined(WIN32) # ifdef OPUS_BUILD # define OPUS_EXPORT __declspec(dllexport) # else # define OPUS_EXPORT # endif +# elif defined(__GNUC__) && defined(OPUS_BUILD) +# define OPUS_EXPORT __attribute__ ((visibility ("default"))) # else # define OPUS_EXPORT # endif @@ -98,6 +98,18 @@ extern "C" { # define OPUS_RESTRICT restrict #endif +#if (!defined(__STDC_VERSION__) || (__STDC_VERSION__ < 199901L) ) +# if OPUS_GNUC_PREREQ(2,7) +# define OPUS_INLINE __inline__ +# elif (defined(_MSC_VER)) +# define OPUS_INLINE __inline +# else +# define OPUS_INLINE +# endif +#else +# define OPUS_INLINE inline +#endif + /**Warning attributes for opus functions * NONNULL is not used in OPUS_BUILD to avoid the compiler optimizing out * some paranoid null checks. */ @@ -148,8 +160,11 @@ extern "C" { #define OPUS_GET_GAIN_REQUEST 4045 /* Should have been 4035 */ #define OPUS_SET_LSB_DEPTH_REQUEST 4036 #define OPUS_GET_LSB_DEPTH_REQUEST 4037 - #define OPUS_GET_LAST_PACKET_DURATION_REQUEST 4039 +#define OPUS_SET_EXPERT_FRAME_DURATION_REQUEST 4040 +#define OPUS_GET_EXPERT_FRAME_DURATION_REQUEST 4041 +#define OPUS_SET_PREDICTION_DISABLED_REQUEST 4042 +#define OPUS_GET_PREDICTION_DISABLED_REQUEST 4043 /* Don't use 4045, it's already taken by OPUS_GET_GAIN_REQUEST */ @@ -157,6 +172,7 @@ extern "C" { #define __opus_check_int(x) (((void)((x) == (opus_int32)0)), (opus_int32)(x)) #define __opus_check_int_ptr(ptr) ((ptr) + ((ptr) - (opus_int32*)(ptr))) #define __opus_check_uint_ptr(ptr) ((ptr) + ((ptr) - (opus_uint32*)(ptr))) +#define __opus_check_val16_ptr(ptr) ((ptr) + ((ptr) - (opus_val16*)(ptr))) /** @endcond */ /** @defgroup opus_ctlvalues Pre-defined values for CTL interface @@ -185,6 +201,14 @@ extern "C" { #define OPUS_BANDWIDTH_SUPERWIDEBAND 1104 /**<12 kHz bandpass @hideinitializer*/ #define OPUS_BANDWIDTH_FULLBAND 1105 /**<20 kHz bandpass @hideinitializer*/ +#define OPUS_FRAMESIZE_ARG 5000 /**< Select frame size from the argument (default) */ +#define OPUS_FRAMESIZE_2_5_MS 5001 /**< Use 2.5 ms frames */ +#define OPUS_FRAMESIZE_5_MS 5002 /**< Use 5 ms frames */ +#define OPUS_FRAMESIZE_10_MS 5003 /**< Use 10 ms frames */ +#define OPUS_FRAMESIZE_20_MS 5004 /**< Use 20 ms frames */ +#define OPUS_FRAMESIZE_40_MS 5005 /**< Use 40 ms frames */ +#define OPUS_FRAMESIZE_60_MS 5006 /**< Use 60 ms frames */ + /**@}*/ @@ -525,6 +549,53 @@ extern "C" { * @param[out] x <tt>opus_int32 *</tt>: Number of samples (at current sampling rate). * @hideinitializer */ #define OPUS_GET_LAST_PACKET_DURATION(x) OPUS_GET_LAST_PACKET_DURATION_REQUEST, __opus_check_int_ptr(x) + +/** Configures the encoder's use of variable duration frames. + * When variable duration is enabled, the encoder is free to use a shorter frame + * size than the one requested in the opus_encode*() call. + * It is then the user's responsibility + * to verify how much audio was encoded by checking the ToC byte of the encoded + * packet. The part of the audio that was not encoded needs to be resent to the + * encoder for the next call. Do not use this option unless you <b>really</b> + * know what you are doing. + * @see OPUS_GET_EXPERT_VARIABLE_DURATION + * @param[in] x <tt>opus_int32</tt>: Allowed values: + * <dl> + * <dt>OPUS_FRAMESIZE_ARG</dt><dd>Select frame size from the argument (default).</dd> + * <dt>OPUS_FRAMESIZE_2_5_MS</dt><dd>Use 2.5 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_5_MS</dt><dd>Use 2.5 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_10_MS</dt><dd>Use 10 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_20_MS</dt><dd>Use 20 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_40_MS</dt><dd>Use 40 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_60_MS</dt><dd>Use 60 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_VARIABLE</dt><dd>Optimize the frame size dynamically.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_SET_EXPERT_FRAME_DURATION(x) OPUS_SET_EXPERT_FRAME_DURATION_REQUEST, __opus_check_int(x) +/** Gets the encoder's configured use of variable duration frames. + * @see OPUS_SET_EXPERT_VARIABLE_DURATION + * @param[out] x <tt>opus_int32 *</tt>: Returns one of the following values: + * <dl> + * <dt>OPUS_FRAMESIZE_ARG</dt><dd>Select frame size from the argument (default).</dd> + * <dt>OPUS_FRAMESIZE_2_5_MS</dt><dd>Use 2.5 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_5_MS</dt><dd>Use 2.5 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_10_MS</dt><dd>Use 10 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_20_MS</dt><dd>Use 20 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_40_MS</dt><dd>Use 40 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_60_MS</dt><dd>Use 60 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_VARIABLE</dt><dd>Optimize the frame size dynamically.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_GET_EXPERT_FRAME_DURATION(x) OPUS_GET_EXPERT_FRAME_DURATION_REQUEST, __opus_check_int_ptr(x) + +/** If set to 1, disables almost all use of prediction, making frames almost + completely independent. This reduces quality. (default : 0) + * @hideinitializer */ +#define OPUS_SET_PREDICTION_DISABLED(x) OPUS_SET_PREDICTION_DISABLED_REQUEST, __opus_check_int(x) +/** Gets the encoder's configured prediction status. + * @hideinitializer */ +#define OPUS_GET_PREDICTION_DISABLED(x) OPUS_GET_PREDICTION_DISABLED_REQUEST, __opus_check_int_ptr(x) + /**@}*/ /** @defgroup opus_genericctls Generic CTLs diff --git a/src/opus-1.0.2/include/opus_multistream.h b/src/opus-1.1/include/opus_multistream.h index 658067f7..ae599793 100644 --- a/src/opus-1.0.2/include/opus_multistream.h +++ b/src/opus-1.1/include/opus_multistream.h @@ -205,6 +205,12 @@ OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_multistream_encoder_get_size int coupled_streams ); +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_multistream_surround_encoder_get_size( + int channels, + int mapping_family +); + + /** Allocates and initializes a multistream encoder state. * Call opus_multistream_encoder_destroy() to release * this object when finished. @@ -258,6 +264,17 @@ OPUS_EXPORT OPUS_WARN_UNUSED_RESULT OpusMSEncoder *opus_multistream_encoder_crea int *error ) OPUS_ARG_NONNULL(5); +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT OpusMSEncoder *opus_multistream_surround_encoder_create( + opus_int32 Fs, + int channels, + int mapping_family, + int *streams, + int *coupled_streams, + unsigned char *mapping, + int application, + int *error +) OPUS_ARG_NONNULL(5); + /** Initialize a previously allocated multistream encoder state. * The memory pointed to by \a st must be at least the size returned by * opus_multistream_encoder_get_size(). @@ -316,6 +333,17 @@ OPUS_EXPORT int opus_multistream_encoder_init( int application ) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(6); +OPUS_EXPORT int opus_multistream_surround_encoder_init( + OpusMSEncoder *st, + opus_int32 Fs, + int channels, + int mapping_family, + int *streams, + int *coupled_streams, + unsigned char *mapping, + int application +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(6); + /** Encodes a multistream Opus frame. * @param st <tt>OpusMSEncoder*</tt>: Multistream encoder state. * @param[in] pcm <tt>const opus_int16*</tt>: The input signal as interleaved diff --git a/src/opus-1.0.2/include/opus_types.h b/src/opus-1.1/include/opus_types.h index b28e03ae..b28e03ae 100644 --- a/src/opus-1.0.2/include/opus_types.h +++ b/src/opus-1.1/include/opus_types.h diff --git a/src/opus-1.0.2/silk/A2NLSF.c b/src/opus-1.1/silk/A2NLSF.c index 49d5d9d9..74b1b95d 100644 --- a/src/opus-1.0.2/silk/A2NLSF.c +++ b/src/opus-1.1/silk/A2NLSF.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -44,7 +44,7 @@ POSSIBILITY OF SUCH DAMAGE. /* Helper function for A2NLSF(..) */ /* Transforms polynomials from cos(n*f) to cos(f)^n */ -static inline void silk_A2NLSF_trans_poly( +static OPUS_INLINE void silk_A2NLSF_trans_poly( opus_int32 *p, /* I/O Polynomial */ const opus_int dd /* I Polynomial order (= filter order / 2 ) */ ) @@ -60,7 +60,7 @@ static inline void silk_A2NLSF_trans_poly( } /* Helper function for A2NLSF(..) */ /* Polynomial evaluation */ -static inline opus_int32 silk_A2NLSF_eval_poly( /* return the polynomial evaluation, in Q16 */ +static OPUS_INLINE opus_int32 silk_A2NLSF_eval_poly( /* return the polynomial evaluation, in Q16 */ opus_int32 *p, /* I Polynomial, Q16 */ const opus_int32 x, /* I Evaluation point, Q12 */ const opus_int dd /* I Order */ @@ -77,7 +77,7 @@ static inline opus_int32 silk_A2NLSF_eval_poly( /* return the polynomial evaluat return y32; } -static inline void silk_A2NLSF_init( +static OPUS_INLINE void silk_A2NLSF_init( const opus_int32 *a_Q16, opus_int32 *P, opus_int32 *Q, diff --git a/src/opus-1.0.2/silk/API.h b/src/opus-1.1/silk/API.h index 4b8ca12a..f0601bcf 100644 --- a/src/opus-1.0.2/silk/API.h +++ b/src/opus-1.1/silk/API.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -64,6 +64,7 @@ opus_int silk_Get_Encoder_Size( /* O Returns error co /*************************/ opus_int silk_InitEncoder( /* O Returns error code */ void *encState, /* I/O State */ + int arch, /* I Run-time architecture */ silk_EncControlStruct *encStatus /* O Encoder Status */ ); diff --git a/src/opus-1.0.2/silk/CNG.c b/src/opus-1.1/silk/CNG.c index d0a619c1..8481d95d 100644 --- a/src/opus-1.0.2/silk/CNG.c +++ b/src/opus-1.1/silk/CNG.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,9 +30,10 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main.h" +#include "stack_alloc.h" /* Generates excitation for CNG LPC synthesis */ -static inline void silk_CNG_exc( +static OPUS_INLINE void silk_CNG_exc( opus_int32 residual_Q10[], /* O CNG residual signal Q10 */ opus_int32 exc_buf_Q14[], /* I Random samples buffer Q10 */ opus_int32 Gain_Q16, /* I Gain to apply */ @@ -86,8 +87,8 @@ void silk_CNG( opus_int i, subfr; opus_int32 sum_Q6, max_Gain_Q16; opus_int16 A_Q12[ MAX_LPC_ORDER ]; - opus_int32 CNG_sig_Q10[ MAX_FRAME_LENGTH + MAX_LPC_ORDER ]; silk_CNG_struct *psCNG = &psDec->sCNG; + SAVE_STACK; if( psDec->fs_kHz != psCNG->fs_kHz ) { /* Reset state */ @@ -123,6 +124,9 @@ void silk_CNG( /* Add CNG when packet is lost or during DTX */ if( psDec->lossCnt ) { + VARDECL( opus_int32, CNG_sig_Q10 ); + + ALLOC( CNG_sig_Q10, length + MAX_LPC_ORDER, opus_int32 ); /* Generate CNG excitation */ silk_CNG_exc( CNG_sig_Q10 + MAX_LPC_ORDER, psCNG->CNG_exc_buf_Q14, psCNG->CNG_smth_Gain_Q16, length, &psCNG->rand_seed ); @@ -164,4 +168,5 @@ void silk_CNG( } else { silk_memset( psCNG->CNG_synth_state, 0, psDec->LPC_order * sizeof( opus_int32 ) ); } + RESTORE_STACK; } diff --git a/src/opus-1.0.2/silk/HP_variable_cutoff.c b/src/opus-1.1/silk/HP_variable_cutoff.c index 199dbb34..bbe10f04 100644 --- a/src/opus-1.0.2/silk/HP_variable_cutoff.c +++ b/src/opus-1.1/silk/HP_variable_cutoff.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/Inlines.h b/src/opus-1.1/silk/Inlines.h index 87ac2e20..ec986cdf 100644 --- a/src/opus-1.0.2/silk/Inlines.h +++ b/src/opus-1.1/silk/Inlines.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -26,7 +26,7 @@ POSSIBILITY OF SUCH DAMAGE. ***********************************************************************/ /*! \file silk_Inlines.h - * \brief silk_Inlines.h defines inline signal processing functions. + * \brief silk_Inlines.h defines OPUS_INLINE signal processing functions. */ #ifndef SILK_FIX_INLINES_H @@ -38,7 +38,7 @@ extern "C" #endif /* count leading zeros of opus_int64 */ -static inline opus_int32 silk_CLZ64( opus_int64 in ) +static OPUS_INLINE opus_int32 silk_CLZ64( opus_int64 in ) { opus_int32 in_upper; @@ -53,7 +53,7 @@ static inline opus_int32 silk_CLZ64( opus_int64 in ) } /* get number of leading zeros and fractional part (the bits right after the leading one */ -static inline void silk_CLZ_FRAC( +static OPUS_INLINE void silk_CLZ_FRAC( opus_int32 in, /* I input */ opus_int32 *lz, /* O number of leading zeros */ opus_int32 *frac_Q7 /* O the 7 bits right after the leading one */ @@ -68,7 +68,7 @@ static inline void silk_CLZ_FRAC( /* Approximation of square root */ /* Accuracy: < +/- 10% for output values > 15 */ /* < +/- 2.5% for output values > 120 */ -static inline opus_int32 silk_SQRT_APPROX( opus_int32 x ) +static OPUS_INLINE opus_int32 silk_SQRT_APPROX( opus_int32 x ) { opus_int32 y, lz, frac_Q7; @@ -94,7 +94,7 @@ static inline opus_int32 silk_SQRT_APPROX( opus_int32 x ) } /* Divide two int32 values and return result as int32 in a given Q-domain */ -static inline opus_int32 silk_DIV32_varQ( /* O returns a good approximation of "(a32 << Qres) / b32" */ +static OPUS_INLINE opus_int32 silk_DIV32_varQ( /* O returns a good approximation of "(a32 << Qres) / b32" */ const opus_int32 a32, /* I numerator (Q0) */ const opus_int32 b32, /* I denominator (Q0) */ const opus_int Qres /* I Q-domain of result (>= 0) */ @@ -140,7 +140,7 @@ static inline opus_int32 silk_DIV32_varQ( /* O returns a good approximation } /* Invert int32 value and return result as int32 in a given Q-domain */ -static inline opus_int32 silk_INVERSE32_varQ( /* O returns a good approximation of "(1 << Qres) / b32" */ +static OPUS_INLINE opus_int32 silk_INVERSE32_varQ( /* O returns a good approximation of "(1 << Qres) / b32" */ const opus_int32 b32, /* I denominator (Q0) */ const opus_int Qres /* I Q-domain of result (> 0) */ ) diff --git a/src/opus-1.0.2/silk/LPC_analysis_filter.c b/src/opus-1.1/silk/LPC_analysis_filter.c index 421dba0b..9d1f16cb 100644 --- a/src/opus-1.0.2/silk/LPC_analysis_filter.c +++ b/src/opus-1.1/silk/LPC_analysis_filter.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,6 +30,7 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "SigProc_FIX.h" +#include "celt_lpc.h" /*******************************************/ /* LPC analysis filter */ @@ -46,14 +47,33 @@ void silk_LPC_analysis_filter( const opus_int32 d /* I Filter order */ ) { - opus_int ix, j; + opus_int j; +#ifdef FIXED_POINT + opus_int16 mem[SILK_MAX_ORDER_LPC]; + opus_int16 num[SILK_MAX_ORDER_LPC]; +#else + int ix; opus_int32 out32_Q12, out32; const opus_int16 *in_ptr; +#endif silk_assert( d >= 6 ); silk_assert( (d & 1) == 0 ); silk_assert( d <= len ); +#ifdef FIXED_POINT + silk_assert( d <= SILK_MAX_ORDER_LPC ); + for ( j = 0; j < d; j++ ) { + num[ j ] = -B[ j ]; + } + for (j=0;j<d;j++) { + mem[ j ] = in[ d - j - 1 ]; + } + celt_fir( in + d, num, out + d, len - d, d, mem ); + for ( j = 0; j < d; j++ ) { + out[ j ] = 0; + } +#else for( ix = d; ix < len; ix++ ) { in_ptr = &in[ ix - 1 ]; @@ -82,4 +102,5 @@ void silk_LPC_analysis_filter( /* Set first d output samples to zero */ silk_memset( out, 0, d * sizeof( opus_int16 ) ); +#endif } diff --git a/src/opus-1.0.2/silk/LPC_inv_pred_gain.c b/src/opus-1.1/silk/LPC_inv_pred_gain.c index c36f3422..4af89aa5 100644 --- a/src/opus-1.0.2/silk/LPC_inv_pred_gain.c +++ b/src/opus-1.1/silk/LPC_inv_pred_gain.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/LP_variable_cutoff.c b/src/opus-1.1/silk/LP_variable_cutoff.c index d0912a64..f639e1f8 100644 --- a/src/opus-1.0.2/silk/LP_variable_cutoff.c +++ b/src/opus-1.1/silk/LP_variable_cutoff.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -38,7 +38,7 @@ POSSIBILITY OF SUCH DAMAGE. #include "main.h" /* Helper function, interpolates the filter taps */ -static inline void silk_LP_interpolate_filter_taps( +static OPUS_INLINE void silk_LP_interpolate_filter_taps( opus_int32 B_Q28[ TRANSITION_NB ], opus_int32 A_Q28[ TRANSITION_NA ], const opus_int ind, diff --git a/src/opus-1.0.2/silk/MacroCount.h b/src/opus-1.1/silk/MacroCount.h index 2829e8cc..834817d0 100644 --- a/src/opus-1.0.2/silk/MacroCount.h +++ b/src/opus-1.1/silk/MacroCount.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -34,11 +34,11 @@ POSSIBILITY OF SUCH DAMAGE. extern opus_int64 ops_count; -static inline opus_int64 silk_SaveCount(){ +static OPUS_INLINE opus_int64 silk_SaveCount(){ return(ops_count); } -static inline opus_int64 silk_SaveResetCount(){ +static OPUS_INLINE opus_int64 silk_SaveResetCount(){ opus_int64 ret; ret = ops_count; @@ -46,12 +46,12 @@ static inline opus_int64 silk_SaveResetCount(){ return(ret); } -static inline silk_PrintCount(){ +static OPUS_INLINE silk_PrintCount(){ printf("ops_count = %d \n ", (opus_int32)ops_count); } #undef silk_MUL -static inline opus_int32 silk_MUL(opus_int32 a32, opus_int32 b32){ +static OPUS_INLINE opus_int32 silk_MUL(opus_int32 a32, opus_int32 b32){ opus_int32 ret; ops_count += 4; ret = a32 * b32; @@ -59,14 +59,14 @@ static inline opus_int32 silk_MUL(opus_int32 a32, opus_int32 b32){ } #undef silk_MUL_uint -static inline opus_uint32 silk_MUL_uint(opus_uint32 a32, opus_uint32 b32){ +static OPUS_INLINE opus_uint32 silk_MUL_uint(opus_uint32 a32, opus_uint32 b32){ opus_uint32 ret; ops_count += 4; ret = a32 * b32; return ret; } #undef silk_MLA -static inline opus_int32 silk_MLA(opus_int32 a32, opus_int32 b32, opus_int32 c32){ +static OPUS_INLINE opus_int32 silk_MLA(opus_int32 a32, opus_int32 b32, opus_int32 c32){ opus_int32 ret; ops_count += 4; ret = a32 + b32 * c32; @@ -74,7 +74,7 @@ static inline opus_int32 silk_MLA(opus_int32 a32, opus_int32 b32, opus_int32 c32 } #undef silk_MLA_uint -static inline opus_int32 silk_MLA_uint(opus_uint32 a32, opus_uint32 b32, opus_uint32 c32){ +static OPUS_INLINE opus_int32 silk_MLA_uint(opus_uint32 a32, opus_uint32 b32, opus_uint32 c32){ opus_uint32 ret; ops_count += 4; ret = a32 + b32 * c32; @@ -82,14 +82,14 @@ static inline opus_int32 silk_MLA_uint(opus_uint32 a32, opus_uint32 b32, opus_ui } #undef silk_SMULWB -static inline opus_int32 silk_SMULWB(opus_int32 a32, opus_int32 b32){ +static OPUS_INLINE opus_int32 silk_SMULWB(opus_int32 a32, opus_int32 b32){ opus_int32 ret; ops_count += 5; ret = (a32 >> 16) * (opus_int32)((opus_int16)b32) + (((a32 & 0x0000FFFF) * (opus_int32)((opus_int16)b32)) >> 16); return ret; } #undef silk_SMLAWB -static inline opus_int32 silk_SMLAWB(opus_int32 a32, opus_int32 b32, opus_int32 c32){ +static OPUS_INLINE opus_int32 silk_SMLAWB(opus_int32 a32, opus_int32 b32, opus_int32 c32){ opus_int32 ret; ops_count += 5; ret = ((a32) + ((((b32) >> 16) * (opus_int32)((opus_int16)(c32))) + ((((b32) & 0x0000FFFF) * (opus_int32)((opus_int16)(c32))) >> 16))); @@ -97,14 +97,14 @@ static inline opus_int32 silk_SMLAWB(opus_int32 a32, opus_int32 b32, opus_int32 } #undef silk_SMULWT -static inline opus_int32 silk_SMULWT(opus_int32 a32, opus_int32 b32){ +static OPUS_INLINE opus_int32 silk_SMULWT(opus_int32 a32, opus_int32 b32){ opus_int32 ret; ops_count += 4; ret = (a32 >> 16) * (b32 >> 16) + (((a32 & 0x0000FFFF) * (b32 >> 16)) >> 16); return ret; } #undef silk_SMLAWT -static inline opus_int32 silk_SMLAWT(opus_int32 a32, opus_int32 b32, opus_int32 c32){ +static OPUS_INLINE opus_int32 silk_SMLAWT(opus_int32 a32, opus_int32 b32, opus_int32 c32){ opus_int32 ret; ops_count += 4; ret = a32 + ((b32 >> 16) * (c32 >> 16)) + (((b32 & 0x0000FFFF) * ((c32 >> 16)) >> 16)); @@ -112,14 +112,14 @@ static inline opus_int32 silk_SMLAWT(opus_int32 a32, opus_int32 b32, opus_int32 } #undef silk_SMULBB -static inline opus_int32 silk_SMULBB(opus_int32 a32, opus_int32 b32){ +static OPUS_INLINE opus_int32 silk_SMULBB(opus_int32 a32, opus_int32 b32){ opus_int32 ret; ops_count += 1; ret = (opus_int32)((opus_int16)a32) * (opus_int32)((opus_int16)b32); return ret; } #undef silk_SMLABB -static inline opus_int32 silk_SMLABB(opus_int32 a32, opus_int32 b32, opus_int32 c32){ +static OPUS_INLINE opus_int32 silk_SMLABB(opus_int32 a32, opus_int32 b32, opus_int32 c32){ opus_int32 ret; ops_count += 1; ret = a32 + (opus_int32)((opus_int16)b32) * (opus_int32)((opus_int16)c32); @@ -127,7 +127,7 @@ static inline opus_int32 silk_SMLABB(opus_int32 a32, opus_int32 b32, opus_int32 } #undef silk_SMULBT -static inline opus_int32 silk_SMULBT(opus_int32 a32, opus_int32 b32 ){ +static OPUS_INLINE opus_int32 silk_SMULBT(opus_int32 a32, opus_int32 b32 ){ opus_int32 ret; ops_count += 4; ret = ((opus_int32)((opus_int16)a32)) * (b32 >> 16); @@ -135,7 +135,7 @@ static inline opus_int32 silk_SMULBT(opus_int32 a32, opus_int32 b32 ){ } #undef silk_SMLABT -static inline opus_int32 silk_SMLABT(opus_int32 a32, opus_int32 b32, opus_int32 c32){ +static OPUS_INLINE opus_int32 silk_SMLABT(opus_int32 a32, opus_int32 b32, opus_int32 c32){ opus_int32 ret; ops_count += 1; ret = a32 + ((opus_int32)((opus_int16)b32)) * (c32 >> 16); @@ -143,7 +143,7 @@ static inline opus_int32 silk_SMLABT(opus_int32 a32, opus_int32 b32, opus_int32 } #undef silk_SMULTT -static inline opus_int32 silk_SMULTT(opus_int32 a32, opus_int32 b32){ +static OPUS_INLINE opus_int32 silk_SMULTT(opus_int32 a32, opus_int32 b32){ opus_int32 ret; ops_count += 1; ret = (a32 >> 16) * (b32 >> 16); @@ -151,7 +151,7 @@ static inline opus_int32 silk_SMULTT(opus_int32 a32, opus_int32 b32){ } #undef silk_SMLATT -static inline opus_int32 silk_SMLATT(opus_int32 a32, opus_int32 b32, opus_int32 c32){ +static OPUS_INLINE opus_int32 silk_SMLATT(opus_int32 a32, opus_int32 b32, opus_int32 c32){ opus_int32 ret; ops_count += 1; ret = a32 + (b32 >> 16) * (c32 >> 16); @@ -179,7 +179,7 @@ static inline opus_int32 silk_SMLATT(opus_int32 a32, opus_int32 b32, opus_int32 #define silk_SMLAWT_ovflw silk_SMLAWT #undef silk_SMULL -static inline opus_int64 silk_SMULL(opus_int32 a32, opus_int32 b32){ +static OPUS_INLINE opus_int64 silk_SMULL(opus_int32 a32, opus_int32 b32){ opus_int64 ret; ops_count += 8; ret = ((opus_int64)(a32) * /*(opus_int64)*/(b32)); @@ -187,14 +187,14 @@ static inline opus_int64 silk_SMULL(opus_int32 a32, opus_int32 b32){ } #undef silk_SMLAL -static inline opus_int64 silk_SMLAL(opus_int64 a64, opus_int32 b32, opus_int32 c32){ +static OPUS_INLINE opus_int64 silk_SMLAL(opus_int64 a64, opus_int32 b32, opus_int32 c32){ opus_int64 ret; ops_count += 8; ret = a64 + ((opus_int64)(b32) * /*(opus_int64)*/(c32)); return ret; } #undef silk_SMLALBB -static inline opus_int64 silk_SMLALBB(opus_int64 a64, opus_int16 b16, opus_int16 c16){ +static OPUS_INLINE opus_int64 silk_SMLALBB(opus_int64 a64, opus_int16 b16, opus_int16 c16){ opus_int64 ret; ops_count += 4; ret = a64 + ((opus_int64)(b16) * /*(opus_int64)*/(c16)); @@ -202,7 +202,7 @@ static inline opus_int64 silk_SMLALBB(opus_int64 a64, opus_int16 b16, opus_int16 } #undef SigProcFIX_CLZ16 -static inline opus_int32 SigProcFIX_CLZ16(opus_int16 in16) +static OPUS_INLINE opus_int32 SigProcFIX_CLZ16(opus_int16 in16) { opus_int32 out32 = 0; ops_count += 10; @@ -240,7 +240,7 @@ static inline opus_int32 SigProcFIX_CLZ16(opus_int16 in16) } #undef SigProcFIX_CLZ32 -static inline opus_int32 SigProcFIX_CLZ32(opus_int32 in32) +static OPUS_INLINE opus_int32 SigProcFIX_CLZ32(opus_int32 in32) { /* test highest 16 bits and convert to opus_int16 */ ops_count += 2; @@ -252,19 +252,19 @@ static inline opus_int32 SigProcFIX_CLZ32(opus_int32 in32) } #undef silk_DIV32 -static inline opus_int32 silk_DIV32(opus_int32 a32, opus_int32 b32){ +static OPUS_INLINE opus_int32 silk_DIV32(opus_int32 a32, opus_int32 b32){ ops_count += 64; return a32 / b32; } #undef silk_DIV32_16 -static inline opus_int32 silk_DIV32_16(opus_int32 a32, opus_int32 b32){ +static OPUS_INLINE opus_int32 silk_DIV32_16(opus_int32 a32, opus_int32 b32){ ops_count += 32; return a32 / b32; } #undef silk_SAT8 -static inline opus_int8 silk_SAT8(opus_int64 a){ +static OPUS_INLINE opus_int8 silk_SAT8(opus_int64 a){ opus_int8 tmp; ops_count += 1; tmp = (opus_int8)((a) > silk_int8_MAX ? silk_int8_MAX : \ @@ -273,7 +273,7 @@ static inline opus_int8 silk_SAT8(opus_int64 a){ } #undef silk_SAT16 -static inline opus_int16 silk_SAT16(opus_int64 a){ +static OPUS_INLINE opus_int16 silk_SAT16(opus_int64 a){ opus_int16 tmp; ops_count += 1; tmp = (opus_int16)((a) > silk_int16_MAX ? silk_int16_MAX : \ @@ -281,7 +281,7 @@ static inline opus_int16 silk_SAT16(opus_int64 a){ return(tmp); } #undef silk_SAT32 -static inline opus_int32 silk_SAT32(opus_int64 a){ +static OPUS_INLINE opus_int32 silk_SAT32(opus_int64 a){ opus_int32 tmp; ops_count += 1; tmp = (opus_int32)((a) > silk_int32_MAX ? silk_int32_MAX : \ @@ -289,7 +289,7 @@ static inline opus_int32 silk_SAT32(opus_int64 a){ return(tmp); } #undef silk_POS_SAT32 -static inline opus_int32 silk_POS_SAT32(opus_int64 a){ +static OPUS_INLINE opus_int32 silk_POS_SAT32(opus_int64 a){ opus_int32 tmp; ops_count += 1; tmp = (opus_int32)((a) > silk_int32_MAX ? silk_int32_MAX : (a)); @@ -297,14 +297,14 @@ static inline opus_int32 silk_POS_SAT32(opus_int64 a){ } #undef silk_ADD_POS_SAT8 -static inline opus_int8 silk_ADD_POS_SAT8(opus_int64 a, opus_int64 b){ +static OPUS_INLINE opus_int8 silk_ADD_POS_SAT8(opus_int64 a, opus_int64 b){ opus_int8 tmp; ops_count += 1; tmp = (opus_int8)((((a)+(b)) & 0x80) ? silk_int8_MAX : ((a)+(b))); return(tmp); } #undef silk_ADD_POS_SAT16 -static inline opus_int16 silk_ADD_POS_SAT16(opus_int64 a, opus_int64 b){ +static OPUS_INLINE opus_int16 silk_ADD_POS_SAT16(opus_int64 a, opus_int64 b){ opus_int16 tmp; ops_count += 1; tmp = (opus_int16)((((a)+(b)) & 0x8000) ? silk_int16_MAX : ((a)+(b))); @@ -312,7 +312,7 @@ static inline opus_int16 silk_ADD_POS_SAT16(opus_int64 a, opus_int64 b){ } #undef silk_ADD_POS_SAT32 -static inline opus_int32 silk_ADD_POS_SAT32(opus_int64 a, opus_int64 b){ +static OPUS_INLINE opus_int32 silk_ADD_POS_SAT32(opus_int64 a, opus_int64 b){ opus_int32 tmp; ops_count += 1; tmp = (opus_int32)((((a)+(b)) & 0x80000000) ? silk_int32_MAX : ((a)+(b))); @@ -320,7 +320,7 @@ static inline opus_int32 silk_ADD_POS_SAT32(opus_int64 a, opus_int64 b){ } #undef silk_ADD_POS_SAT64 -static inline opus_int64 silk_ADD_POS_SAT64(opus_int64 a, opus_int64 b){ +static OPUS_INLINE opus_int64 silk_ADD_POS_SAT64(opus_int64 a, opus_int64 b){ opus_int64 tmp; ops_count += 1; tmp = ((((a)+(b)) & 0x8000000000000000LL) ? silk_int64_MAX : ((a)+(b))); @@ -328,40 +328,40 @@ static inline opus_int64 silk_ADD_POS_SAT64(opus_int64 a, opus_int64 b){ } #undef silk_LSHIFT8 -static inline opus_int8 silk_LSHIFT8(opus_int8 a, opus_int32 shift){ +static OPUS_INLINE opus_int8 silk_LSHIFT8(opus_int8 a, opus_int32 shift){ opus_int8 ret; ops_count += 1; ret = a << shift; return ret; } #undef silk_LSHIFT16 -static inline opus_int16 silk_LSHIFT16(opus_int16 a, opus_int32 shift){ +static OPUS_INLINE opus_int16 silk_LSHIFT16(opus_int16 a, opus_int32 shift){ opus_int16 ret; ops_count += 1; ret = a << shift; return ret; } #undef silk_LSHIFT32 -static inline opus_int32 silk_LSHIFT32(opus_int32 a, opus_int32 shift){ +static OPUS_INLINE opus_int32 silk_LSHIFT32(opus_int32 a, opus_int32 shift){ opus_int32 ret; ops_count += 1; ret = a << shift; return ret; } #undef silk_LSHIFT64 -static inline opus_int64 silk_LSHIFT64(opus_int64 a, opus_int shift){ +static OPUS_INLINE opus_int64 silk_LSHIFT64(opus_int64 a, opus_int shift){ ops_count += 1; return a << shift; } #undef silk_LSHIFT_ovflw -static inline opus_int32 silk_LSHIFT_ovflw(opus_int32 a, opus_int32 shift){ +static OPUS_INLINE opus_int32 silk_LSHIFT_ovflw(opus_int32 a, opus_int32 shift){ ops_count += 1; return a << shift; } #undef silk_LSHIFT_uint -static inline opus_uint32 silk_LSHIFT_uint(opus_uint32 a, opus_int32 shift){ +static OPUS_INLINE opus_uint32 silk_LSHIFT_uint(opus_uint32 a, opus_int32 shift){ opus_uint32 ret; ops_count += 1; ret = a << shift; @@ -369,83 +369,83 @@ static inline opus_uint32 silk_LSHIFT_uint(opus_uint32 a, opus_int32 shift){ } #undef silk_RSHIFT8 -static inline opus_int8 silk_RSHIFT8(opus_int8 a, opus_int32 shift){ +static OPUS_INLINE opus_int8 silk_RSHIFT8(opus_int8 a, opus_int32 shift){ ops_count += 1; return a >> shift; } #undef silk_RSHIFT16 -static inline opus_int16 silk_RSHIFT16(opus_int16 a, opus_int32 shift){ +static OPUS_INLINE opus_int16 silk_RSHIFT16(opus_int16 a, opus_int32 shift){ ops_count += 1; return a >> shift; } #undef silk_RSHIFT32 -static inline opus_int32 silk_RSHIFT32(opus_int32 a, opus_int32 shift){ +static OPUS_INLINE opus_int32 silk_RSHIFT32(opus_int32 a, opus_int32 shift){ ops_count += 1; return a >> shift; } #undef silk_RSHIFT64 -static inline opus_int64 silk_RSHIFT64(opus_int64 a, opus_int64 shift){ +static OPUS_INLINE opus_int64 silk_RSHIFT64(opus_int64 a, opus_int64 shift){ ops_count += 1; return a >> shift; } #undef silk_RSHIFT_uint -static inline opus_uint32 silk_RSHIFT_uint(opus_uint32 a, opus_int32 shift){ +static OPUS_INLINE opus_uint32 silk_RSHIFT_uint(opus_uint32 a, opus_int32 shift){ ops_count += 1; return a >> shift; } #undef silk_ADD_LSHIFT -static inline opus_int32 silk_ADD_LSHIFT(opus_int32 a, opus_int32 b, opus_int32 shift){ +static OPUS_INLINE opus_int32 silk_ADD_LSHIFT(opus_int32 a, opus_int32 b, opus_int32 shift){ opus_int32 ret; ops_count += 1; ret = a + (b << shift); return ret; /* shift >= 0*/ } #undef silk_ADD_LSHIFT32 -static inline opus_int32 silk_ADD_LSHIFT32(opus_int32 a, opus_int32 b, opus_int32 shift){ +static OPUS_INLINE opus_int32 silk_ADD_LSHIFT32(opus_int32 a, opus_int32 b, opus_int32 shift){ opus_int32 ret; ops_count += 1; ret = a + (b << shift); return ret; /* shift >= 0*/ } #undef silk_ADD_LSHIFT_uint -static inline opus_uint32 silk_ADD_LSHIFT_uint(opus_uint32 a, opus_uint32 b, opus_int32 shift){ +static OPUS_INLINE opus_uint32 silk_ADD_LSHIFT_uint(opus_uint32 a, opus_uint32 b, opus_int32 shift){ opus_uint32 ret; ops_count += 1; ret = a + (b << shift); return ret; /* shift >= 0*/ } #undef silk_ADD_RSHIFT -static inline opus_int32 silk_ADD_RSHIFT(opus_int32 a, opus_int32 b, opus_int32 shift){ +static OPUS_INLINE opus_int32 silk_ADD_RSHIFT(opus_int32 a, opus_int32 b, opus_int32 shift){ opus_int32 ret; ops_count += 1; ret = a + (b >> shift); return ret; /* shift > 0*/ } #undef silk_ADD_RSHIFT32 -static inline opus_int32 silk_ADD_RSHIFT32(opus_int32 a, opus_int32 b, opus_int32 shift){ +static OPUS_INLINE opus_int32 silk_ADD_RSHIFT32(opus_int32 a, opus_int32 b, opus_int32 shift){ opus_int32 ret; ops_count += 1; ret = a + (b >> shift); return ret; /* shift > 0*/ } #undef silk_ADD_RSHIFT_uint -static inline opus_uint32 silk_ADD_RSHIFT_uint(opus_uint32 a, opus_uint32 b, opus_int32 shift){ +static OPUS_INLINE opus_uint32 silk_ADD_RSHIFT_uint(opus_uint32 a, opus_uint32 b, opus_int32 shift){ opus_uint32 ret; ops_count += 1; ret = a + (b >> shift); return ret; /* shift > 0*/ } #undef silk_SUB_LSHIFT32 -static inline opus_int32 silk_SUB_LSHIFT32(opus_int32 a, opus_int32 b, opus_int32 shift){ +static OPUS_INLINE opus_int32 silk_SUB_LSHIFT32(opus_int32 a, opus_int32 b, opus_int32 shift){ opus_int32 ret; ops_count += 1; ret = a - (b << shift); return ret; /* shift >= 0*/ } #undef silk_SUB_RSHIFT32 -static inline opus_int32 silk_SUB_RSHIFT32(opus_int32 a, opus_int32 b, opus_int32 shift){ +static OPUS_INLINE opus_int32 silk_SUB_RSHIFT32(opus_int32 a, opus_int32 b, opus_int32 shift){ opus_int32 ret; ops_count += 1; ret = a - (b >> shift); @@ -453,7 +453,7 @@ static inline opus_int32 silk_SUB_RSHIFT32(opus_int32 a, opus_int32 b, opus_int3 } #undef silk_RSHIFT_ROUND -static inline opus_int32 silk_RSHIFT_ROUND(opus_int32 a, opus_int32 shift){ +static OPUS_INLINE opus_int32 silk_RSHIFT_ROUND(opus_int32 a, opus_int32 shift){ opus_int32 ret; ops_count += 3; ret = shift == 1 ? (a >> 1) + (a & 1) : ((a >> (shift - 1)) + 1) >> 1; @@ -461,7 +461,7 @@ static inline opus_int32 silk_RSHIFT_ROUND(opus_int32 a, opus_int32 shift){ } #undef silk_RSHIFT_ROUND64 -static inline opus_int64 silk_RSHIFT_ROUND64(opus_int64 a, opus_int32 shift){ +static OPUS_INLINE opus_int64 silk_RSHIFT_ROUND64(opus_int64 a, opus_int32 shift){ opus_int64 ret; ops_count += 6; ret = shift == 1 ? (a >> 1) + (a & 1) : ((a >> (shift - 1)) + 1) >> 1; @@ -469,13 +469,13 @@ static inline opus_int64 silk_RSHIFT_ROUND64(opus_int64 a, opus_int32 shift){ } #undef silk_abs_int64 -static inline opus_int64 silk_abs_int64(opus_int64 a){ +static OPUS_INLINE opus_int64 silk_abs_int64(opus_int64 a){ ops_count += 1; return (((a) > 0) ? (a) : -(a)); /* Be careful, silk_abs returns wrong when input equals to silk_intXX_MIN*/ } #undef silk_abs_int32 -static inline opus_int32 silk_abs_int32(opus_int32 a){ +static OPUS_INLINE opus_int32 silk_abs_int32(opus_int32 a){ ops_count += 1; return silk_abs(a); } @@ -498,7 +498,7 @@ static silk_sign(a){ } #undef silk_ADD16 -static inline opus_int16 silk_ADD16(opus_int16 a, opus_int16 b){ +static OPUS_INLINE opus_int16 silk_ADD16(opus_int16 a, opus_int16 b){ opus_int16 ret; ops_count += 1; ret = a + b; @@ -506,7 +506,7 @@ static inline opus_int16 silk_ADD16(opus_int16 a, opus_int16 b){ } #undef silk_ADD32 -static inline opus_int32 silk_ADD32(opus_int32 a, opus_int32 b){ +static OPUS_INLINE opus_int32 silk_ADD32(opus_int32 a, opus_int32 b){ opus_int32 ret; ops_count += 1; ret = a + b; @@ -514,7 +514,7 @@ static inline opus_int32 silk_ADD32(opus_int32 a, opus_int32 b){ } #undef silk_ADD64 -static inline opus_int64 silk_ADD64(opus_int64 a, opus_int64 b){ +static OPUS_INLINE opus_int64 silk_ADD64(opus_int64 a, opus_int64 b){ opus_int64 ret; ops_count += 2; ret = a + b; @@ -522,7 +522,7 @@ static inline opus_int64 silk_ADD64(opus_int64 a, opus_int64 b){ } #undef silk_SUB16 -static inline opus_int16 silk_SUB16(opus_int16 a, opus_int16 b){ +static OPUS_INLINE opus_int16 silk_SUB16(opus_int16 a, opus_int16 b){ opus_int16 ret; ops_count += 1; ret = a - b; @@ -530,7 +530,7 @@ static inline opus_int16 silk_SUB16(opus_int16 a, opus_int16 b){ } #undef silk_SUB32 -static inline opus_int32 silk_SUB32(opus_int32 a, opus_int32 b){ +static OPUS_INLINE opus_int32 silk_SUB32(opus_int32 a, opus_int32 b){ opus_int32 ret; ops_count += 1; ret = a - b; @@ -538,7 +538,7 @@ static inline opus_int32 silk_SUB32(opus_int32 a, opus_int32 b){ } #undef silk_SUB64 -static inline opus_int64 silk_SUB64(opus_int64 a, opus_int64 b){ +static OPUS_INLINE opus_int64 silk_SUB64(opus_int64 a, opus_int64 b){ opus_int64 ret; ops_count += 2; ret = a - b; @@ -546,7 +546,7 @@ static inline opus_int64 silk_SUB64(opus_int64 a, opus_int64 b){ } #undef silk_ADD_SAT16 -static inline opus_int16 silk_ADD_SAT16( opus_int16 a16, opus_int16 b16 ) { +static OPUS_INLINE opus_int16 silk_ADD_SAT16( opus_int16 a16, opus_int16 b16 ) { opus_int16 res; /* Nb will be counted in AKP_add32 and silk_SAT16*/ res = (opus_int16)silk_SAT16( silk_ADD32( (opus_int32)(a16), (b16) ) ); @@ -554,7 +554,7 @@ static inline opus_int16 silk_ADD_SAT16( opus_int16 a16, opus_int16 b16 ) { } #undef silk_ADD_SAT32 -static inline opus_int32 silk_ADD_SAT32(opus_int32 a32, opus_int32 b32){ +static OPUS_INLINE opus_int32 silk_ADD_SAT32(opus_int32 a32, opus_int32 b32){ opus_int32 res; ops_count += 1; res = ((((a32) + (b32)) & 0x80000000) == 0 ? \ @@ -564,7 +564,7 @@ static inline opus_int32 silk_ADD_SAT32(opus_int32 a32, opus_int32 b32){ } #undef silk_ADD_SAT64 -static inline opus_int64 silk_ADD_SAT64( opus_int64 a64, opus_int64 b64 ) { +static OPUS_INLINE opus_int64 silk_ADD_SAT64( opus_int64 a64, opus_int64 b64 ) { opus_int64 res; ops_count += 1; res = ((((a64) + (b64)) & 0x8000000000000000LL) == 0 ? \ @@ -574,7 +574,7 @@ static inline opus_int64 silk_ADD_SAT64( opus_int64 a64, opus_int64 b64 ) { } #undef silk_SUB_SAT16 -static inline opus_int16 silk_SUB_SAT16( opus_int16 a16, opus_int16 b16 ) { +static OPUS_INLINE opus_int16 silk_SUB_SAT16( opus_int16 a16, opus_int16 b16 ) { opus_int16 res; silk_assert(0); /* Nb will be counted in sub-macros*/ @@ -583,7 +583,7 @@ static inline opus_int16 silk_SUB_SAT16( opus_int16 a16, opus_int16 b16 ) { } #undef silk_SUB_SAT32 -static inline opus_int32 silk_SUB_SAT32( opus_int32 a32, opus_int32 b32 ) { +static OPUS_INLINE opus_int32 silk_SUB_SAT32( opus_int32 a32, opus_int32 b32 ) { opus_int32 res; ops_count += 1; res = ((((a32)-(b32)) & 0x80000000) == 0 ? \ @@ -593,7 +593,7 @@ static inline opus_int32 silk_SUB_SAT32( opus_int32 a32, opus_int32 b32 ) { } #undef silk_SUB_SAT64 -static inline opus_int64 silk_SUB_SAT64( opus_int64 a64, opus_int64 b64 ) { +static OPUS_INLINE opus_int64 silk_SUB_SAT64( opus_int64 a64, opus_int64 b64 ) { opus_int64 res; ops_count += 1; res = ((((a64)-(b64)) & 0x8000000000000000LL) == 0 ? \ @@ -604,7 +604,7 @@ static inline opus_int64 silk_SUB_SAT64( opus_int64 a64, opus_int64 b64 ) { } #undef silk_SMULWW -static inline opus_int32 silk_SMULWW(opus_int32 a32, opus_int32 b32){ +static OPUS_INLINE opus_int32 silk_SMULWW(opus_int32 a32, opus_int32 b32){ opus_int32 ret; /* Nb will be counted in sub-macros*/ ret = silk_MLA(silk_SMULWB((a32), (b32)), (a32), silk_RSHIFT_ROUND((b32), 16)); @@ -612,7 +612,7 @@ static inline opus_int32 silk_SMULWW(opus_int32 a32, opus_int32 b32){ } #undef silk_SMLAWW -static inline opus_int32 silk_SMLAWW(opus_int32 a32, opus_int32 b32, opus_int32 c32){ +static OPUS_INLINE opus_int32 silk_SMLAWW(opus_int32 a32, opus_int32 b32, opus_int32 c32){ opus_int32 ret; /* Nb will be counted in sub-macros*/ ret = silk_MLA(silk_SMLAWB((a32), (b32), (c32)), (b32), silk_RSHIFT_ROUND((c32), 16)); @@ -620,26 +620,26 @@ static inline opus_int32 silk_SMLAWW(opus_int32 a32, opus_int32 b32, opus_int32 } #undef silk_min_int -static inline opus_int silk_min_int(opus_int a, opus_int b) +static OPUS_INLINE opus_int silk_min_int(opus_int a, opus_int b) { ops_count += 1; return (((a) < (b)) ? (a) : (b)); } #undef silk_min_16 -static inline opus_int16 silk_min_16(opus_int16 a, opus_int16 b) +static OPUS_INLINE opus_int16 silk_min_16(opus_int16 a, opus_int16 b) { ops_count += 1; return (((a) < (b)) ? (a) : (b)); } #undef silk_min_32 -static inline opus_int32 silk_min_32(opus_int32 a, opus_int32 b) +static OPUS_INLINE opus_int32 silk_min_32(opus_int32 a, opus_int32 b) { ops_count += 1; return (((a) < (b)) ? (a) : (b)); } #undef silk_min_64 -static inline opus_int64 silk_min_64(opus_int64 a, opus_int64 b) +static OPUS_INLINE opus_int64 silk_min_64(opus_int64 a, opus_int64 b) { ops_count += 1; return (((a) < (b)) ? (a) : (b)); @@ -647,26 +647,26 @@ static inline opus_int64 silk_min_64(opus_int64 a, opus_int64 b) /* silk_min() versions with typecast in the function call */ #undef silk_max_int -static inline opus_int silk_max_int(opus_int a, opus_int b) +static OPUS_INLINE opus_int silk_max_int(opus_int a, opus_int b) { ops_count += 1; return (((a) > (b)) ? (a) : (b)); } #undef silk_max_16 -static inline opus_int16 silk_max_16(opus_int16 a, opus_int16 b) +static OPUS_INLINE opus_int16 silk_max_16(opus_int16 a, opus_int16 b) { ops_count += 1; return (((a) > (b)) ? (a) : (b)); } #undef silk_max_32 -static inline opus_int32 silk_max_32(opus_int32 a, opus_int32 b) +static OPUS_INLINE opus_int32 silk_max_32(opus_int32 a, opus_int32 b) { ops_count += 1; return (((a) > (b)) ? (a) : (b)); } #undef silk_max_64 -static inline opus_int64 silk_max_64(opus_int64 a, opus_int64 b) +static OPUS_INLINE opus_int64 silk_max_64(opus_int64 a, opus_int64 b) { ops_count += 1; return (((a) > (b)) ? (a) : (b)); @@ -674,7 +674,7 @@ static inline opus_int64 silk_max_64(opus_int64 a, opus_int64 b) #undef silk_LIMIT_int -static inline opus_int silk_LIMIT_int(opus_int a, opus_int limit1, opus_int limit2) +static OPUS_INLINE opus_int silk_LIMIT_int(opus_int a, opus_int limit1, opus_int limit2) { opus_int ret; ops_count += 6; @@ -686,7 +686,7 @@ static inline opus_int silk_LIMIT_int(opus_int a, opus_int limit1, opus_int limi } #undef silk_LIMIT_16 -static inline opus_int16 silk_LIMIT_16(opus_int16 a, opus_int16 limit1, opus_int16 limit2) +static OPUS_INLINE opus_int16 silk_LIMIT_16(opus_int16 a, opus_int16 limit1, opus_int16 limit2) { opus_int16 ret; ops_count += 6; @@ -699,7 +699,7 @@ return(ret); #undef silk_LIMIT_32 -static inline opus_int silk_LIMIT_32(opus_int32 a, opus_int32 limit1, opus_int32 limit2) +static OPUS_INLINE opus_int silk_LIMIT_32(opus_int32 a, opus_int32 limit1, opus_int32 limit2) { opus_int32 ret; ops_count += 6; diff --git a/src/opus-1.0.2/silk/MacroDebug.h b/src/opus-1.1/silk/MacroDebug.h index ecd90bc4..35aedc5c 100644 --- a/src/opus-1.0.2/silk/MacroDebug.h +++ b/src/opus-1.1/silk/MacroDebug.h @@ -9,11 +9,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -36,7 +36,7 @@ POSSIBILITY OF SUCH DAMAGE. #undef silk_ADD16 #define silk_ADD16(a,b) silk_ADD16_((a), (b), __FILE__, __LINE__) -static inline opus_int16 silk_ADD16_(opus_int16 a, opus_int16 b, char *file, int line){ +static OPUS_INLINE opus_int16 silk_ADD16_(opus_int16 a, opus_int16 b, char *file, int line){ opus_int16 ret; ret = a + b; @@ -52,7 +52,7 @@ static inline opus_int16 silk_ADD16_(opus_int16 a, opus_int16 b, char *file, int #undef silk_ADD32 #define silk_ADD32(a,b) silk_ADD32_((a), (b), __FILE__, __LINE__) -static inline opus_int32 silk_ADD32_(opus_int32 a, opus_int32 b, char *file, int line){ +static OPUS_INLINE opus_int32 silk_ADD32_(opus_int32 a, opus_int32 b, char *file, int line){ opus_int32 ret; ret = a + b; @@ -68,7 +68,7 @@ static inline opus_int32 silk_ADD32_(opus_int32 a, opus_int32 b, char *file, int #undef silk_ADD64 #define silk_ADD64(a,b) silk_ADD64_((a), (b), __FILE__, __LINE__) -static inline opus_int64 silk_ADD64_(opus_int64 a, opus_int64 b, char *file, int line){ +static OPUS_INLINE opus_int64 silk_ADD64_(opus_int64 a, opus_int64 b, char *file, int line){ opus_int64 ret; ret = a + b; @@ -84,7 +84,7 @@ static inline opus_int64 silk_ADD64_(opus_int64 a, opus_int64 b, char *file, int #undef silk_SUB16 #define silk_SUB16(a,b) silk_SUB16_((a), (b), __FILE__, __LINE__) -static inline opus_int16 silk_SUB16_(opus_int16 a, opus_int16 b, char *file, int line){ +static OPUS_INLINE opus_int16 silk_SUB16_(opus_int16 a, opus_int16 b, char *file, int line){ opus_int16 ret; ret = a - b; @@ -100,7 +100,7 @@ static inline opus_int16 silk_SUB16_(opus_int16 a, opus_int16 b, char *file, int #undef silk_SUB32 #define silk_SUB32(a,b) silk_SUB32_((a), (b), __FILE__, __LINE__) -static inline opus_int32 silk_SUB32_(opus_int32 a, opus_int32 b, char *file, int line){ +static OPUS_INLINE opus_int32 silk_SUB32_(opus_int32 a, opus_int32 b, char *file, int line){ opus_int32 ret; ret = a - b; @@ -116,7 +116,7 @@ static inline opus_int32 silk_SUB32_(opus_int32 a, opus_int32 b, char *file, int #undef silk_SUB64 #define silk_SUB64(a,b) silk_SUB64_((a), (b), __FILE__, __LINE__) -static inline opus_int64 silk_SUB64_(opus_int64 a, opus_int64 b, char *file, int line){ +static OPUS_INLINE opus_int64 silk_SUB64_(opus_int64 a, opus_int64 b, char *file, int line){ opus_int64 ret; ret = a - b; @@ -132,7 +132,7 @@ static inline opus_int64 silk_SUB64_(opus_int64 a, opus_int64 b, char *file, int #undef silk_ADD_SAT16 #define silk_ADD_SAT16(a,b) silk_ADD_SAT16_((a), (b), __FILE__, __LINE__) -static inline opus_int16 silk_ADD_SAT16_( opus_int16 a16, opus_int16 b16, char *file, int line) { +static OPUS_INLINE opus_int16 silk_ADD_SAT16_( opus_int16 a16, opus_int16 b16, char *file, int line) { opus_int16 res; res = (opus_int16)silk_SAT16( silk_ADD32( (opus_int32)(a16), (b16) ) ); if ( res != silk_SAT16( (opus_int32)a16 + (opus_int32)b16 ) ) @@ -147,7 +147,7 @@ static inline opus_int16 silk_ADD_SAT16_( opus_int16 a16, opus_int16 b16, char * #undef silk_ADD_SAT32 #define silk_ADD_SAT32(a,b) silk_ADD_SAT32_((a), (b), __FILE__, __LINE__) -static inline opus_int32 silk_ADD_SAT32_(opus_int32 a32, opus_int32 b32, char *file, int line){ +static OPUS_INLINE opus_int32 silk_ADD_SAT32_(opus_int32 a32, opus_int32 b32, char *file, int line){ opus_int32 res; res = ((((opus_uint32)(a32) + (opus_uint32)(b32)) & 0x80000000) == 0 ? \ ((((a32) & (b32)) & 0x80000000) != 0 ? silk_int32_MIN : (a32)+(b32)) : \ @@ -164,7 +164,7 @@ static inline opus_int32 silk_ADD_SAT32_(opus_int32 a32, opus_int32 b32, char *f #undef silk_ADD_SAT64 #define silk_ADD_SAT64(a,b) silk_ADD_SAT64_((a), (b), __FILE__, __LINE__) -static inline opus_int64 silk_ADD_SAT64_( opus_int64 a64, opus_int64 b64, char *file, int line) { +static OPUS_INLINE opus_int64 silk_ADD_SAT64_( opus_int64 a64, opus_int64 b64, char *file, int line) { opus_int64 res; int fail = 0; res = ((((a64) + (b64)) & 0x8000000000000000LL) == 0 ? \ @@ -193,7 +193,7 @@ static inline opus_int64 silk_ADD_SAT64_( opus_int64 a64, opus_int64 b64, char * #undef silk_SUB_SAT16 #define silk_SUB_SAT16(a,b) silk_SUB_SAT16_((a), (b), __FILE__, __LINE__) -static inline opus_int16 silk_SUB_SAT16_( opus_int16 a16, opus_int16 b16, char *file, int line ) { +static OPUS_INLINE opus_int16 silk_SUB_SAT16_( opus_int16 a16, opus_int16 b16, char *file, int line ) { opus_int16 res; res = (opus_int16)silk_SAT16( silk_SUB32( (opus_int32)(a16), (b16) ) ); if ( res != silk_SAT16( (opus_int32)a16 - (opus_int32)b16 ) ) @@ -208,7 +208,7 @@ static inline opus_int16 silk_SUB_SAT16_( opus_int16 a16, opus_int16 b16, char * #undef silk_SUB_SAT32 #define silk_SUB_SAT32(a,b) silk_SUB_SAT32_((a), (b), __FILE__, __LINE__) -static inline opus_int32 silk_SUB_SAT32_( opus_int32 a32, opus_int32 b32, char *file, int line ) { +static OPUS_INLINE opus_int32 silk_SUB_SAT32_( opus_int32 a32, opus_int32 b32, char *file, int line ) { opus_int32 res; res = ((((opus_uint32)(a32)-(opus_uint32)(b32)) & 0x80000000) == 0 ? \ (( (a32) & ((b32)^0x80000000) & 0x80000000) ? silk_int32_MIN : (a32)-(b32)) : \ @@ -225,7 +225,7 @@ static inline opus_int32 silk_SUB_SAT32_( opus_int32 a32, opus_int32 b32, char * #undef silk_SUB_SAT64 #define silk_SUB_SAT64(a,b) silk_SUB_SAT64_((a), (b), __FILE__, __LINE__) -static inline opus_int64 silk_SUB_SAT64_( opus_int64 a64, opus_int64 b64, char *file, int line ) { +static OPUS_INLINE opus_int64 silk_SUB_SAT64_( opus_int64 a64, opus_int64 b64, char *file, int line ) { opus_int64 res; int fail = 0; res = ((((a64)-(b64)) & 0x8000000000000000LL) == 0 ? \ @@ -254,7 +254,7 @@ static inline opus_int64 silk_SUB_SAT64_( opus_int64 a64, opus_int64 b64, char * #undef silk_MUL #define silk_MUL(a,b) silk_MUL_((a), (b), __FILE__, __LINE__) -static inline opus_int32 silk_MUL_(opus_int32 a32, opus_int32 b32, char *file, int line){ +static OPUS_INLINE opus_int32 silk_MUL_(opus_int32 a32, opus_int32 b32, char *file, int line){ opus_int32 ret; opus_int64 ret64; ret = a32 * b32; @@ -271,7 +271,7 @@ static inline opus_int32 silk_MUL_(opus_int32 a32, opus_int32 b32, char *file, i #undef silk_MUL_uint #define silk_MUL_uint(a,b) silk_MUL_uint_((a), (b), __FILE__, __LINE__) -static inline opus_uint32 silk_MUL_uint_(opus_uint32 a32, opus_uint32 b32, char *file, int line){ +static OPUS_INLINE opus_uint32 silk_MUL_uint_(opus_uint32 a32, opus_uint32 b32, char *file, int line){ opus_uint32 ret; ret = a32 * b32; if ( (opus_uint64)ret != (opus_uint64)a32 * (opus_uint64)b32 ) @@ -286,7 +286,7 @@ static inline opus_uint32 silk_MUL_uint_(opus_uint32 a32, opus_uint32 b32, char #undef silk_MLA #define silk_MLA(a,b,c) silk_MLA_((a), (b), (c), __FILE__, __LINE__) -static inline opus_int32 silk_MLA_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ +static OPUS_INLINE opus_int32 silk_MLA_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ opus_int32 ret; ret = a32 + b32 * c32; if ( (opus_int64)ret != (opus_int64)a32 + (opus_int64)b32 * (opus_int64)c32 ) @@ -301,7 +301,7 @@ static inline opus_int32 silk_MLA_(opus_int32 a32, opus_int32 b32, opus_int32 c3 #undef silk_MLA_uint #define silk_MLA_uint(a,b,c) silk_MLA_uint_((a), (b), (c), __FILE__, __LINE__) -static inline opus_int32 silk_MLA_uint_(opus_uint32 a32, opus_uint32 b32, opus_uint32 c32, char *file, int line){ +static OPUS_INLINE opus_int32 silk_MLA_uint_(opus_uint32 a32, opus_uint32 b32, opus_uint32 c32, char *file, int line){ opus_uint32 ret; ret = a32 + b32 * c32; if ( (opus_int64)ret != (opus_int64)a32 + (opus_int64)b32 * (opus_int64)c32 ) @@ -316,7 +316,7 @@ static inline opus_int32 silk_MLA_uint_(opus_uint32 a32, opus_uint32 b32, opus_u #undef silk_SMULWB #define silk_SMULWB(a,b) silk_SMULWB_((a), (b), __FILE__, __LINE__) -static inline opus_int32 silk_SMULWB_(opus_int32 a32, opus_int32 b32, char *file, int line){ +static OPUS_INLINE opus_int32 silk_SMULWB_(opus_int32 a32, opus_int32 b32, char *file, int line){ opus_int32 ret; ret = (a32 >> 16) * (opus_int32)((opus_int16)b32) + (((a32 & 0x0000FFFF) * (opus_int32)((opus_int16)b32)) >> 16); if ( (opus_int64)ret != ((opus_int64)a32 * (opus_int16)b32) >> 16 ) @@ -331,7 +331,7 @@ static inline opus_int32 silk_SMULWB_(opus_int32 a32, opus_int32 b32, char *file #undef silk_SMLAWB #define silk_SMLAWB(a,b,c) silk_SMLAWB_((a), (b), (c), __FILE__, __LINE__) -static inline opus_int32 silk_SMLAWB_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ +static OPUS_INLINE opus_int32 silk_SMLAWB_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ opus_int32 ret; ret = silk_ADD32( a32, silk_SMULWB( b32, c32 ) ); if ( silk_ADD32( a32, silk_SMULWB( b32, c32 ) ) != silk_ADD_SAT32( a32, silk_SMULWB( b32, c32 ) ) ) @@ -346,7 +346,7 @@ static inline opus_int32 silk_SMLAWB_(opus_int32 a32, opus_int32 b32, opus_int32 #undef silk_SMULWT #define silk_SMULWT(a,b) silk_SMULWT_((a), (b), __FILE__, __LINE__) -static inline opus_int32 silk_SMULWT_(opus_int32 a32, opus_int32 b32, char *file, int line){ +static OPUS_INLINE opus_int32 silk_SMULWT_(opus_int32 a32, opus_int32 b32, char *file, int line){ opus_int32 ret; ret = (a32 >> 16) * (b32 >> 16) + (((a32 & 0x0000FFFF) * (b32 >> 16)) >> 16); if ( (opus_int64)ret != ((opus_int64)a32 * (b32 >> 16)) >> 16 ) @@ -361,7 +361,7 @@ static inline opus_int32 silk_SMULWT_(opus_int32 a32, opus_int32 b32, char *file #undef silk_SMLAWT #define silk_SMLAWT(a,b,c) silk_SMLAWT_((a), (b), (c), __FILE__, __LINE__) -static inline opus_int32 silk_SMLAWT_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ +static OPUS_INLINE opus_int32 silk_SMLAWT_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ opus_int32 ret; ret = a32 + ((b32 >> 16) * (c32 >> 16)) + (((b32 & 0x0000FFFF) * ((c32 >> 16)) >> 16)); if ( (opus_int64)ret != (opus_int64)a32 + (((opus_int64)b32 * (c32 >> 16)) >> 16) ) @@ -376,7 +376,7 @@ static inline opus_int32 silk_SMLAWT_(opus_int32 a32, opus_int32 b32, opus_int32 #undef silk_SMULL #define silk_SMULL(a,b) silk_SMULL_((a), (b), __FILE__, __LINE__) -static inline opus_int64 silk_SMULL_(opus_int64 a64, opus_int64 b64, char *file, int line){ +static OPUS_INLINE opus_int64 silk_SMULL_(opus_int64 a64, opus_int64 b64, char *file, int line){ opus_int64 ret64; int fail = 0; ret64 = a64 * b64; @@ -398,7 +398,7 @@ static inline opus_int64 silk_SMULL_(opus_int64 a64, opus_int64 b64, char *file, /* no checking needed for silk_SMULBB */ #undef silk_SMLABB #define silk_SMLABB(a,b,c) silk_SMLABB_((a), (b), (c), __FILE__, __LINE__) -static inline opus_int32 silk_SMLABB_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ +static OPUS_INLINE opus_int32 silk_SMLABB_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ opus_int32 ret; ret = a32 + (opus_int32)((opus_int16)b32) * (opus_int32)((opus_int16)c32); if ( (opus_int64)ret != (opus_int64)a32 + (opus_int64)b32 * (opus_int16)c32 ) @@ -414,7 +414,7 @@ static inline opus_int32 silk_SMLABB_(opus_int32 a32, opus_int32 b32, opus_int32 /* no checking needed for silk_SMULBT */ #undef silk_SMLABT #define silk_SMLABT(a,b,c) silk_SMLABT_((a), (b), (c), __FILE__, __LINE__) -static inline opus_int32 silk_SMLABT_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ +static OPUS_INLINE opus_int32 silk_SMLABT_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ opus_int32 ret; ret = a32 + ((opus_int32)((opus_int16)b32)) * (c32 >> 16); if ( (opus_int64)ret != (opus_int64)a32 + (opus_int64)b32 * (c32 >> 16) ) @@ -430,7 +430,7 @@ static inline opus_int32 silk_SMLABT_(opus_int32 a32, opus_int32 b32, opus_int32 /* no checking needed for silk_SMULTT */ #undef silk_SMLATT #define silk_SMLATT(a,b,c) silk_SMLATT_((a), (b), (c), __FILE__, __LINE__) -static inline opus_int32 silk_SMLATT_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ +static OPUS_INLINE opus_int32 silk_SMLATT_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ opus_int32 ret; ret = a32 + (b32 >> 16) * (c32 >> 16); if ( (opus_int64)ret != (opus_int64)a32 + (b32 >> 16) * (c32 >> 16) ) @@ -445,7 +445,7 @@ static inline opus_int32 silk_SMLATT_(opus_int32 a32, opus_int32 b32, opus_int32 #undef silk_SMULWW #define silk_SMULWW(a,b) silk_SMULWW_((a), (b), __FILE__, __LINE__) -static inline opus_int32 silk_SMULWW_(opus_int32 a32, opus_int32 b32, char *file, int line){ +static OPUS_INLINE opus_int32 silk_SMULWW_(opus_int32 a32, opus_int32 b32, char *file, int line){ opus_int32 ret, tmp1, tmp2; opus_int64 ret64; int fail = 0; @@ -476,7 +476,7 @@ static inline opus_int32 silk_SMULWW_(opus_int32 a32, opus_int32 b32, char *file #undef silk_SMLAWW #define silk_SMLAWW(a,b,c) silk_SMLAWW_((a), (b), (c), __FILE__, __LINE__) -static inline opus_int32 silk_SMLAWW_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ +static OPUS_INLINE opus_int32 silk_SMLAWW_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ opus_int32 ret, tmp; tmp = silk_SMULWW( b32, c32 ); @@ -505,7 +505,7 @@ static inline opus_int32 silk_SMLAWW_(opus_int32 a32, opus_int32 b32, opus_int32 #undef silk_DIV32 #define silk_DIV32(a,b) silk_DIV32_((a), (b), __FILE__, __LINE__) -static inline opus_int32 silk_DIV32_(opus_int32 a32, opus_int32 b32, char *file, int line){ +static OPUS_INLINE opus_int32 silk_DIV32_(opus_int32 a32, opus_int32 b32, char *file, int line){ if ( b32 == 0 ) { fprintf (stderr, "silk_DIV32(%d, %d) in %s: line %d\n", a32, b32, file, line); @@ -518,7 +518,7 @@ static inline opus_int32 silk_DIV32_(opus_int32 a32, opus_int32 b32, char *file, #undef silk_DIV32_16 #define silk_DIV32_16(a,b) silk_DIV32_16_((a), (b), __FILE__, __LINE__) -static inline opus_int32 silk_DIV32_16_(opus_int32 a32, opus_int32 b32, char *file, int line){ +static OPUS_INLINE opus_int32 silk_DIV32_16_(opus_int32 a32, opus_int32 b32, char *file, int line){ int fail = 0; fail |= b32 == 0; fail |= b32 > silk_int16_MAX; @@ -544,7 +544,7 @@ static inline opus_int32 silk_DIV32_16_(opus_int32 a32, opus_int32 b32, char *fi #undef silk_LSHIFT8 #define silk_LSHIFT8(a,b) silk_LSHIFT8_((a), (b), __FILE__, __LINE__) -static inline opus_int8 silk_LSHIFT8_(opus_int8 a, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_int8 silk_LSHIFT8_(opus_int8 a, opus_int32 shift, char *file, int line){ opus_int8 ret; int fail = 0; ret = a << shift; @@ -563,7 +563,7 @@ static inline opus_int8 silk_LSHIFT8_(opus_int8 a, opus_int32 shift, char *file, #undef silk_LSHIFT16 #define silk_LSHIFT16(a,b) silk_LSHIFT16_((a), (b), __FILE__, __LINE__) -static inline opus_int16 silk_LSHIFT16_(opus_int16 a, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_int16 silk_LSHIFT16_(opus_int16 a, opus_int32 shift, char *file, int line){ opus_int16 ret; int fail = 0; ret = a << shift; @@ -582,7 +582,7 @@ static inline opus_int16 silk_LSHIFT16_(opus_int16 a, opus_int32 shift, char *fi #undef silk_LSHIFT32 #define silk_LSHIFT32(a,b) silk_LSHIFT32_((a), (b), __FILE__, __LINE__) -static inline opus_int32 silk_LSHIFT32_(opus_int32 a, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_int32 silk_LSHIFT32_(opus_int32 a, opus_int32 shift, char *file, int line){ opus_int32 ret; int fail = 0; ret = a << shift; @@ -601,7 +601,7 @@ static inline opus_int32 silk_LSHIFT32_(opus_int32 a, opus_int32 shift, char *fi #undef silk_LSHIFT64 #define silk_LSHIFT64(a,b) silk_LSHIFT64_((a), (b), __FILE__, __LINE__) -static inline opus_int64 silk_LSHIFT64_(opus_int64 a, opus_int shift, char *file, int line){ +static OPUS_INLINE opus_int64 silk_LSHIFT64_(opus_int64 a, opus_int shift, char *file, int line){ opus_int64 ret; int fail = 0; ret = a << shift; @@ -620,7 +620,7 @@ static inline opus_int64 silk_LSHIFT64_(opus_int64 a, opus_int shift, char *file #undef silk_LSHIFT_ovflw #define silk_LSHIFT_ovflw(a,b) silk_LSHIFT_ovflw_((a), (b), __FILE__, __LINE__) -static inline opus_int32 silk_LSHIFT_ovflw_(opus_int32 a, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_int32 silk_LSHIFT_ovflw_(opus_int32 a, opus_int32 shift, char *file, int line){ if ( (shift < 0) || (shift >= 32) ) /* no check for overflow */ { fprintf (stderr, "silk_LSHIFT_ovflw(%d, %d) in %s: line %d\n", a, shift, file, line); @@ -633,7 +633,7 @@ static inline opus_int32 silk_LSHIFT_ovflw_(opus_int32 a, opus_int32 shift, char #undef silk_LSHIFT_uint #define silk_LSHIFT_uint(a,b) silk_LSHIFT_uint_((a), (b), __FILE__, __LINE__) -static inline opus_uint32 silk_LSHIFT_uint_(opus_uint32 a, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_uint32 silk_LSHIFT_uint_(opus_uint32 a, opus_int32 shift, char *file, int line){ opus_uint32 ret; ret = a << shift; if ( (shift < 0) || ((opus_int64)ret != ((opus_int64)a) << shift)) @@ -648,7 +648,7 @@ static inline opus_uint32 silk_LSHIFT_uint_(opus_uint32 a, opus_int32 shift, cha #undef silk_RSHIFT8 #define silk_RSHITF8(a,b) silk_RSHIFT8_((a), (b), __FILE__, __LINE__) -static inline opus_int8 silk_RSHIFT8_(opus_int8 a, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_int8 silk_RSHIFT8_(opus_int8 a, opus_int32 shift, char *file, int line){ if ( (shift < 0) || (shift>=8) ) { fprintf (stderr, "silk_RSHITF8(%d, %d) in %s: line %d\n", a, shift, file, line); @@ -661,7 +661,7 @@ static inline opus_int8 silk_RSHIFT8_(opus_int8 a, opus_int32 shift, char *file, #undef silk_RSHIFT16 #define silk_RSHITF16(a,b) silk_RSHIFT16_((a), (b), __FILE__, __LINE__) -static inline opus_int16 silk_RSHIFT16_(opus_int16 a, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_int16 silk_RSHIFT16_(opus_int16 a, opus_int32 shift, char *file, int line){ if ( (shift < 0) || (shift>=16) ) { fprintf (stderr, "silk_RSHITF16(%d, %d) in %s: line %d\n", a, shift, file, line); @@ -674,7 +674,7 @@ static inline opus_int16 silk_RSHIFT16_(opus_int16 a, opus_int32 shift, char *fi #undef silk_RSHIFT32 #define silk_RSHIFT32(a,b) silk_RSHIFT32_((a), (b), __FILE__, __LINE__) -static inline opus_int32 silk_RSHIFT32_(opus_int32 a, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_int32 silk_RSHIFT32_(opus_int32 a, opus_int32 shift, char *file, int line){ if ( (shift < 0) || (shift>=32) ) { fprintf (stderr, "silk_RSHITF32(%d, %d) in %s: line %d\n", a, shift, file, line); @@ -687,7 +687,7 @@ static inline opus_int32 silk_RSHIFT32_(opus_int32 a, opus_int32 shift, char *fi #undef silk_RSHIFT64 #define silk_RSHIFT64(a,b) silk_RSHIFT64_((a), (b), __FILE__, __LINE__) -static inline opus_int64 silk_RSHIFT64_(opus_int64 a, opus_int64 shift, char *file, int line){ +static OPUS_INLINE opus_int64 silk_RSHIFT64_(opus_int64 a, opus_int64 shift, char *file, int line){ if ( (shift < 0) || (shift>=64) ) { fprintf (stderr, "silk_RSHITF64(%lld, %lld) in %s: line %d\n", (long long)a, (long long)shift, file, line); @@ -700,7 +700,7 @@ static inline opus_int64 silk_RSHIFT64_(opus_int64 a, opus_int64 shift, char *fi #undef silk_RSHIFT_uint #define silk_RSHIFT_uint(a,b) silk_RSHIFT_uint_((a), (b), __FILE__, __LINE__) -static inline opus_uint32 silk_RSHIFT_uint_(opus_uint32 a, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_uint32 silk_RSHIFT_uint_(opus_uint32 a, opus_int32 shift, char *file, int line){ if ( (shift < 0) || (shift>32) ) { fprintf (stderr, "silk_RSHIFT_uint(%u, %d) in %s: line %d\n", a, shift, file, line); @@ -713,7 +713,7 @@ static inline opus_uint32 silk_RSHIFT_uint_(opus_uint32 a, opus_int32 shift, cha #undef silk_ADD_LSHIFT #define silk_ADD_LSHIFT(a,b,c) silk_ADD_LSHIFT_((a), (b), (c), __FILE__, __LINE__) -static inline int silk_ADD_LSHIFT_(int a, int b, int shift, char *file, int line){ +static OPUS_INLINE int silk_ADD_LSHIFT_(int a, int b, int shift, char *file, int line){ opus_int16 ret; ret = a + (b << shift); if ( (shift < 0) || (shift>15) || ((opus_int64)ret != (opus_int64)a + (((opus_int64)b) << shift)) ) @@ -728,7 +728,7 @@ static inline int silk_ADD_LSHIFT_(int a, int b, int shift, char *file, int line #undef silk_ADD_LSHIFT32 #define silk_ADD_LSHIFT32(a,b,c) silk_ADD_LSHIFT32_((a), (b), (c), __FILE__, __LINE__) -static inline opus_int32 silk_ADD_LSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_int32 silk_ADD_LSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ opus_int32 ret; ret = a + (b << shift); if ( (shift < 0) || (shift>31) || ((opus_int64)ret != (opus_int64)a + (((opus_int64)b) << shift)) ) @@ -743,7 +743,7 @@ static inline opus_int32 silk_ADD_LSHIFT32_(opus_int32 a, opus_int32 b, opus_int #undef silk_ADD_LSHIFT_uint #define silk_ADD_LSHIFT_uint(a,b,c) silk_ADD_LSHIFT_uint_((a), (b), (c), __FILE__, __LINE__) -static inline opus_uint32 silk_ADD_LSHIFT_uint_(opus_uint32 a, opus_uint32 b, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_uint32 silk_ADD_LSHIFT_uint_(opus_uint32 a, opus_uint32 b, opus_int32 shift, char *file, int line){ opus_uint32 ret; ret = a + (b << shift); if ( (shift < 0) || (shift>32) || ((opus_int64)ret != (opus_int64)a + (((opus_int64)b) << shift)) ) @@ -758,7 +758,7 @@ static inline opus_uint32 silk_ADD_LSHIFT_uint_(opus_uint32 a, opus_uint32 b, op #undef silk_ADD_RSHIFT #define silk_ADD_RSHIFT(a,b,c) silk_ADD_RSHIFT_((a), (b), (c), __FILE__, __LINE__) -static inline int silk_ADD_RSHIFT_(int a, int b, int shift, char *file, int line){ +static OPUS_INLINE int silk_ADD_RSHIFT_(int a, int b, int shift, char *file, int line){ opus_int16 ret; ret = a + (b >> shift); if ( (shift < 0) || (shift>15) || ((opus_int64)ret != (opus_int64)a + (((opus_int64)b) >> shift)) ) @@ -773,7 +773,7 @@ static inline int silk_ADD_RSHIFT_(int a, int b, int shift, char *file, int line #undef silk_ADD_RSHIFT32 #define silk_ADD_RSHIFT32(a,b,c) silk_ADD_RSHIFT32_((a), (b), (c), __FILE__, __LINE__) -static inline opus_int32 silk_ADD_RSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_int32 silk_ADD_RSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ opus_int32 ret; ret = a + (b >> shift); if ( (shift < 0) || (shift>31) || ((opus_int64)ret != (opus_int64)a + (((opus_int64)b) >> shift)) ) @@ -788,7 +788,7 @@ static inline opus_int32 silk_ADD_RSHIFT32_(opus_int32 a, opus_int32 b, opus_int #undef silk_ADD_RSHIFT_uint #define silk_ADD_RSHIFT_uint(a,b,c) silk_ADD_RSHIFT_uint_((a), (b), (c), __FILE__, __LINE__) -static inline opus_uint32 silk_ADD_RSHIFT_uint_(opus_uint32 a, opus_uint32 b, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_uint32 silk_ADD_RSHIFT_uint_(opus_uint32 a, opus_uint32 b, opus_int32 shift, char *file, int line){ opus_uint32 ret; ret = a + (b >> shift); if ( (shift < 0) || (shift>32) || ((opus_int64)ret != (opus_int64)a + (((opus_int64)b) >> shift)) ) @@ -803,7 +803,7 @@ static inline opus_uint32 silk_ADD_RSHIFT_uint_(opus_uint32 a, opus_uint32 b, op #undef silk_SUB_LSHIFT32 #define silk_SUB_LSHIFT32(a,b,c) silk_SUB_LSHIFT32_((a), (b), (c), __FILE__, __LINE__) -static inline opus_int32 silk_SUB_LSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_int32 silk_SUB_LSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ opus_int32 ret; ret = a - (b << shift); if ( (shift < 0) || (shift>31) || ((opus_int64)ret != (opus_int64)a - (((opus_int64)b) << shift)) ) @@ -818,7 +818,7 @@ static inline opus_int32 silk_SUB_LSHIFT32_(opus_int32 a, opus_int32 b, opus_int #undef silk_SUB_RSHIFT32 #define silk_SUB_RSHIFT32(a,b,c) silk_SUB_RSHIFT32_((a), (b), (c), __FILE__, __LINE__) -static inline opus_int32 silk_SUB_RSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_int32 silk_SUB_RSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ opus_int32 ret; ret = a - (b >> shift); if ( (shift < 0) || (shift>31) || ((opus_int64)ret != (opus_int64)a - (((opus_int64)b) >> shift)) ) @@ -833,7 +833,7 @@ static inline opus_int32 silk_SUB_RSHIFT32_(opus_int32 a, opus_int32 b, opus_int #undef silk_RSHIFT_ROUND #define silk_RSHIFT_ROUND(a,b) silk_RSHIFT_ROUND_((a), (b), __FILE__, __LINE__) -static inline opus_int32 silk_RSHIFT_ROUND_(opus_int32 a, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_int32 silk_RSHIFT_ROUND_(opus_int32 a, opus_int32 shift, char *file, int line){ opus_int32 ret; ret = shift == 1 ? (a >> 1) + (a & 1) : ((a >> (shift - 1)) + 1) >> 1; /* the marco definition can't handle a shift of zero */ @@ -849,7 +849,7 @@ static inline opus_int32 silk_RSHIFT_ROUND_(opus_int32 a, opus_int32 shift, char #undef silk_RSHIFT_ROUND64 #define silk_RSHIFT_ROUND64(a,b) silk_RSHIFT_ROUND64_((a), (b), __FILE__, __LINE__) -static inline opus_int64 silk_RSHIFT_ROUND64_(opus_int64 a, opus_int32 shift, char *file, int line){ +static OPUS_INLINE opus_int64 silk_RSHIFT_ROUND64_(opus_int64 a, opus_int32 shift, char *file, int line){ opus_int64 ret; /* the marco definition can't handle a shift of zero */ if ( (shift <= 0) || (shift>=64) ) @@ -865,14 +865,14 @@ static inline opus_int64 silk_RSHIFT_ROUND64_(opus_int64 a, opus_int32 shift, ch /* silk_abs is used on floats also, so doesn't work... */ /*#undef silk_abs -static inline opus_int32 silk_abs(opus_int32 a){ +static OPUS_INLINE opus_int32 silk_abs(opus_int32 a){ silk_assert(a != 0x80000000); return (((a) > 0) ? (a) : -(a)); // Be careful, silk_abs returns wrong when input equals to silk_intXX_MIN }*/ #undef silk_abs_int64 #define silk_abs_int64(a) silk_abs_int64_((a), __FILE__, __LINE__) -static inline opus_int64 silk_abs_int64_(opus_int64 a, char *file, int line){ +static OPUS_INLINE opus_int64 silk_abs_int64_(opus_int64 a, char *file, int line){ if ( a == silk_int64_MIN ) { fprintf (stderr, "silk_abs_int64(%lld) in %s: line %d\n", (long long)a, file, line); @@ -885,7 +885,7 @@ static inline opus_int64 silk_abs_int64_(opus_int64 a, char *file, int line){ #undef silk_abs_int32 #define silk_abs_int32(a) silk_abs_int32_((a), __FILE__, __LINE__) -static inline opus_int32 silk_abs_int32_(opus_int32 a, char *file, int line){ +static OPUS_INLINE opus_int32 silk_abs_int32_(opus_int32 a, char *file, int line){ if ( a == silk_int32_MIN ) { fprintf (stderr, "silk_abs_int32(%d) in %s: line %d\n", a, file, line); @@ -898,7 +898,7 @@ static inline opus_int32 silk_abs_int32_(opus_int32 a, char *file, int line){ #undef silk_CHECK_FIT8 #define silk_CHECK_FIT8(a) silk_CHECK_FIT8_((a), __FILE__, __LINE__) -static inline opus_int8 silk_CHECK_FIT8_( opus_int64 a, char *file, int line ){ +static OPUS_INLINE opus_int8 silk_CHECK_FIT8_( opus_int64 a, char *file, int line ){ opus_int8 ret; ret = (opus_int8)a; if ( (opus_int64)ret != a ) @@ -913,7 +913,7 @@ static inline opus_int8 silk_CHECK_FIT8_( opus_int64 a, char *file, int line ){ #undef silk_CHECK_FIT16 #define silk_CHECK_FIT16(a) silk_CHECK_FIT16_((a), __FILE__, __LINE__) -static inline opus_int16 silk_CHECK_FIT16_( opus_int64 a, char *file, int line ){ +static OPUS_INLINE opus_int16 silk_CHECK_FIT16_( opus_int64 a, char *file, int line ){ opus_int16 ret; ret = (opus_int16)a; if ( (opus_int64)ret != a ) @@ -928,7 +928,7 @@ static inline opus_int16 silk_CHECK_FIT16_( opus_int64 a, char *file, int line ) #undef silk_CHECK_FIT32 #define silk_CHECK_FIT32(a) silk_CHECK_FIT32_((a), __FILE__, __LINE__) -static inline opus_int32 silk_CHECK_FIT32_( opus_int64 a, char *file, int line ){ +static OPUS_INLINE opus_int32 silk_CHECK_FIT32_( opus_int64 a, char *file, int line ){ opus_int32 ret; ret = (opus_int32)a; if ( (opus_int64)ret != a ) diff --git a/src/opus-1.0.2/silk/NLSF2A.c b/src/opus-1.1/silk/NLSF2A.c index 10b66b64..b1c559ea 100644 --- a/src/opus-1.0.2/silk/NLSF2A.c +++ b/src/opus-1.1/silk/NLSF2A.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -41,7 +41,7 @@ POSSIBILITY OF SUCH DAMAGE. #define QA 16 /* helper function for NLSF2A(..) */ -static inline void silk_NLSF2A_find_poly( +static OPUS_INLINE void silk_NLSF2A_find_poly( opus_int32 *out, /* O intermediate polynomial, QA [dd+1] */ const opus_int32 *cLSF, /* I vector of interleaved 2*cos(LSFs), QA [d] */ opus_int dd /* I polynomial order (= 1/2 * filter order) */ diff --git a/src/opus-1.0.2/silk/NLSF_VQ.c b/src/opus-1.1/silk/NLSF_VQ.c index 352dda26..69b6e22e 100644 --- a/src/opus-1.0.2/silk/NLSF_VQ.c +++ b/src/opus-1.1/silk/NLSF_VQ.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/NLSF_VQ_weights_laroia.c b/src/opus-1.1/silk/NLSF_VQ_weights_laroia.c index 05bb17af..04894c59 100644 --- a/src/opus-1.0.2/silk/NLSF_VQ_weights_laroia.c +++ b/src/opus-1.1/silk/NLSF_VQ_weights_laroia.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/NLSF_decode.c b/src/opus-1.1/silk/NLSF_decode.c index e007c49a..9f715060 100644 --- a/src/opus-1.0.2/silk/NLSF_decode.c +++ b/src/opus-1.1/silk/NLSF_decode.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -32,7 +32,7 @@ POSSIBILITY OF SUCH DAMAGE. #include "main.h" /* Predictive dequantizer for NLSF residuals */ -static inline void silk_NLSF_residual_dequant( /* O Returns RD value in Q30 */ +static OPUS_INLINE void silk_NLSF_residual_dequant( /* O Returns RD value in Q30 */ opus_int16 x_Q10[], /* O Output [ order ] */ const opus_int8 indices[], /* I Quantization indices [ order ] */ const opus_uint8 pred_coef_Q8[], /* I Backward predictor coefs [ order ] */ diff --git a/src/opus-1.0.2/silk/NLSF_del_dec_quant.c b/src/opus-1.1/silk/NLSF_del_dec_quant.c index 78870de5..504dbbd0 100644 --- a/src/opus-1.0.2/silk/NLSF_del_dec_quant.c +++ b/src/opus-1.1/silk/NLSF_del_dec_quant.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -121,7 +121,7 @@ opus_int32 silk_NLSF_del_dec_quant( /* O Returns RD_Q25[ j + nStates ] = silk_SMLABB( silk_MLA( RD_tmp_Q25, silk_SMULBB( diff_Q10, diff_Q10 ), w_Q5[ i ] ), mu_Q20, rate1_Q5 ); } - if( nStates < NLSF_QUANT_DEL_DEC_STATES ) { + if( nStates <= ( NLSF_QUANT_DEL_DEC_STATES >> 1 ) ) { /* double number of states and copy */ for( j = 0; j < nStates; j++ ) { ind[ j + nStates ][ i ] = ind[ j ][ i ] + 1; diff --git a/src/opus-1.0.2/silk/NLSF_encode.c b/src/opus-1.1/silk/NLSF_encode.c index 52a263d9..03a036fd 100644 --- a/src/opus-1.0.2/silk/NLSF_encode.c +++ b/src/opus-1.1/silk/NLSF_encode.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,6 +30,7 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main.h" +#include "stack_alloc.h" /***********************/ /* NLSF vector encoder */ @@ -46,10 +47,10 @@ opus_int32 silk_NLSF_encode( /* O Returns { opus_int i, s, ind1, bestIndex, prob_Q8, bits_q7; opus_int32 W_tmp_Q9; - opus_int32 err_Q26[ NLSF_VQ_MAX_VECTORS ]; - opus_int32 RD_Q25[ NLSF_VQ_MAX_SURVIVORS ]; - opus_int tempIndices1[ NLSF_VQ_MAX_SURVIVORS ]; - opus_int8 tempIndices2[ NLSF_VQ_MAX_SURVIVORS * MAX_LPC_ORDER ]; + VARDECL( opus_int32, err_Q26 ); + VARDECL( opus_int32, RD_Q25 ); + VARDECL( opus_int, tempIndices1 ); + VARDECL( opus_int8, tempIndices2 ); opus_int16 res_Q15[ MAX_LPC_ORDER ]; opus_int16 res_Q10[ MAX_LPC_ORDER ]; opus_int16 NLSF_tmp_Q15[ MAX_LPC_ORDER ]; @@ -58,6 +59,7 @@ opus_int32 silk_NLSF_encode( /* O Returns opus_uint8 pred_Q8[ MAX_LPC_ORDER ]; opus_int16 ec_ix[ MAX_LPC_ORDER ]; const opus_uint8 *pCB_element, *iCDF_ptr; + SAVE_STACK; silk_assert( nSurvivors <= NLSF_VQ_MAX_SURVIVORS ); silk_assert( signalType >= 0 && signalType <= 2 ); @@ -67,11 +69,16 @@ opus_int32 silk_NLSF_encode( /* O Returns silk_NLSF_stabilize( pNLSF_Q15, psNLSF_CB->deltaMin_Q15, psNLSF_CB->order ); /* First stage: VQ */ + ALLOC( err_Q26, psNLSF_CB->nVectors, opus_int32 ); silk_NLSF_VQ( err_Q26, pNLSF_Q15, psNLSF_CB->CB1_NLSF_Q8, psNLSF_CB->nVectors, psNLSF_CB->order ); /* Sort the quantization errors */ + ALLOC( tempIndices1, nSurvivors, opus_int ); silk_insertion_sort_increasing( err_Q26, tempIndices1, psNLSF_CB->nVectors, nSurvivors ); + ALLOC( RD_Q25, nSurvivors, opus_int32 ); + ALLOC( tempIndices2, nSurvivors * MAX_LPC_ORDER, opus_int8 ); + /* Loop over survivors */ for( s = 0; s < nSurvivors; s++ ) { ind1 = tempIndices1[ s ]; @@ -124,5 +131,6 @@ opus_int32 silk_NLSF_encode( /* O Returns /* Decode */ silk_NLSF_decode( pNLSF_Q15, NLSFIndices, psNLSF_CB ); + RESTORE_STACK; return RD_Q25[ 0 ]; } diff --git a/src/opus-1.0.2/silk/NLSF_stabilize.c b/src/opus-1.1/silk/NLSF_stabilize.c index 7498b54a..1fa1ea37 100644 --- a/src/opus-1.0.2/silk/NLSF_stabilize.c +++ b/src/opus-1.1/silk/NLSF_stabilize.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/NLSF_unpack.c b/src/opus-1.1/silk/NLSF_unpack.c index 47f6cfe8..17bd23f7 100644 --- a/src/opus-1.0.2/silk/NLSF_unpack.c +++ b/src/opus-1.1/silk/NLSF_unpack.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/NSQ.c b/src/opus-1.1/silk/NSQ.c index b49cdf58..cf5b3fd5 100644 --- a/src/opus-1.0.2/silk/NSQ.c +++ b/src/opus-1.1/silk/NSQ.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,8 +30,9 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main.h" +#include "stack_alloc.h" -static inline void silk_nsq_scale_states( +static OPUS_INLINE void silk_nsq_scale_states( const silk_encoder_state *psEncC, /* I Encoder State */ silk_nsq_state *NSQ, /* I/O NSQ state */ const opus_int32 x_Q3[], /* I input in Q3 */ @@ -45,7 +46,7 @@ static inline void silk_nsq_scale_states( const opus_int signal_type /* I Signal type */ ); -static inline void silk_noise_shape_quantizer( +static OPUS_INLINE void silk_noise_shape_quantizer( silk_nsq_state *NSQ, /* I/O NSQ state */ opus_int signalType, /* I Signal type */ const opus_int32 x_sc_Q10[], /* I */ @@ -88,11 +89,12 @@ void silk_NSQ( opus_int k, lag, start_idx, LSF_interpolation_flag; const opus_int16 *A_Q12, *B_Q14, *AR_shp_Q13; opus_int16 *pxq; - opus_int32 sLTP_Q15[ 2 * MAX_FRAME_LENGTH ]; - opus_int16 sLTP[ 2 * MAX_FRAME_LENGTH ]; + VARDECL( opus_int32, sLTP_Q15 ); + VARDECL( opus_int16, sLTP ); opus_int32 HarmShapeFIRPacked_Q14; opus_int offset_Q10; - opus_int32 x_sc_Q10[ MAX_SUB_FRAME_LENGTH ]; + VARDECL( opus_int32, x_sc_Q10 ); + SAVE_STACK; NSQ->rand_seed = psIndices->Seed; @@ -109,6 +111,10 @@ void silk_NSQ( LSF_interpolation_flag = 1; } + ALLOC( sLTP_Q15, + psEncC->ltp_mem_length + psEncC->frame_length, opus_int32 ); + ALLOC( sLTP, psEncC->ltp_mem_length + psEncC->frame_length, opus_int16 ); + ALLOC( x_sc_Q10, psEncC->subfr_length, opus_int32 ); /* Set up pointers to start of sub frame */ NSQ->sLTP_shp_buf_idx = psEncC->ltp_mem_length; NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; @@ -160,12 +166,13 @@ void silk_NSQ( /* DEBUG_STORE_DATA( enc.pcm, &NSQ->xq[ psEncC->ltp_mem_length ], psEncC->frame_length * sizeof( opus_int16 ) ) */ silk_memmove( NSQ->xq, &NSQ->xq[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int16 ) ); silk_memmove( NSQ->sLTP_shp_Q14, &NSQ->sLTP_shp_Q14[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int32 ) ); + RESTORE_STACK; } /***********************************/ /* silk_noise_shape_quantizer */ /***********************************/ -static inline void silk_noise_shape_quantizer( +static OPUS_INLINE void silk_noise_shape_quantizer( silk_nsq_state *NSQ, /* I/O NSQ state */ opus_int signalType, /* I Signal type */ const opus_int32 x_sc_Q10[], /* I */ @@ -363,7 +370,7 @@ static inline void silk_noise_shape_quantizer( silk_memcpy( NSQ->sLPC_Q14, &NSQ->sLPC_Q14[ length ], NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); } -static inline void silk_nsq_scale_states( +static OPUS_INLINE void silk_nsq_scale_states( const silk_encoder_state *psEncC, /* I Encoder State */ silk_nsq_state *NSQ, /* I/O NSQ state */ const opus_int32 x_Q3[], /* I input in Q3 */ diff --git a/src/opus-1.0.2/silk/NSQ_del_dec.c b/src/opus-1.1/silk/NSQ_del_dec.c index b877fa96..522be406 100644 --- a/src/opus-1.0.2/silk/NSQ_del_dec.c +++ b/src/opus-1.1/silk/NSQ_del_dec.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,6 +30,7 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main.h" +#include "stack_alloc.h" typedef struct { opus_int32 sLPC_Q14[ MAX_SUB_FRAME_LENGTH + NSQ_LPC_BUF_LENGTH ]; @@ -54,7 +55,9 @@ typedef struct { opus_int32 LPC_exc_Q14; } NSQ_sample_struct; -static inline void silk_nsq_del_dec_scale_states( +typedef NSQ_sample_struct NSQ_sample_pair[ 2 ]; + +static OPUS_INLINE void silk_nsq_del_dec_scale_states( const silk_encoder_state *psEncC, /* I Encoder State */ silk_nsq_state *NSQ, /* I/O NSQ state */ NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ @@ -74,7 +77,7 @@ static inline void silk_nsq_del_dec_scale_states( /******************************************/ /* Noise shape quantizer for one subframe */ /******************************************/ -static inline void silk_noise_shape_quantizer_del_dec( +static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( silk_nsq_state *NSQ, /* I/O NSQ state */ NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ opus_int signalType, /* I Signal type */ @@ -123,17 +126,18 @@ void silk_NSQ_del_dec( { opus_int i, k, lag, start_idx, LSF_interpolation_flag, Winner_ind, subfr; opus_int last_smple_idx, smpl_buf_idx, decisionDelay; - const opus_int16 *A_Q12, *B_Q14, *AR_shp_Q13; + const opus_int16 *A_Q12, *B_Q14, *AR_shp_Q13; opus_int16 *pxq; - opus_int32 sLTP_Q15[ 2 * MAX_FRAME_LENGTH ]; - opus_int16 sLTP[ 2 * MAX_FRAME_LENGTH ]; + VARDECL( opus_int32, sLTP_Q15 ); + VARDECL( opus_int16, sLTP ); opus_int32 HarmShapeFIRPacked_Q14; opus_int offset_Q10; opus_int32 RDmin_Q10, Gain_Q10; - opus_int32 x_sc_Q10[ MAX_SUB_FRAME_LENGTH ]; - opus_int32 delayedGain_Q10[ DECISION_DELAY ]; - NSQ_del_dec_struct psDelDec[ MAX_DEL_DEC_STATES ]; + VARDECL( opus_int32, x_sc_Q10 ); + VARDECL( opus_int32, delayedGain_Q10 ); + VARDECL( NSQ_del_dec_struct, psDelDec ); NSQ_del_dec_struct *psDD; + SAVE_STACK; /* Set unvoiced lag to the previous one, overwrite later for voiced */ lag = NSQ->lagPrev; @@ -141,6 +145,7 @@ void silk_NSQ_del_dec( silk_assert( NSQ->prev_gain_Q16 != 0 ); /* Initialize delayed decision states */ + ALLOC( psDelDec, psEncC->nStatesDelayedDecision, NSQ_del_dec_struct ); silk_memset( psDelDec, 0, psEncC->nStatesDelayedDecision * sizeof( NSQ_del_dec_struct ) ); for( k = 0; k < psEncC->nStatesDelayedDecision; k++ ) { psDD = &psDelDec[ k ]; @@ -175,6 +180,11 @@ void silk_NSQ_del_dec( LSF_interpolation_flag = 1; } + ALLOC( sLTP_Q15, + psEncC->ltp_mem_length + psEncC->frame_length, opus_int32 ); + ALLOC( sLTP, psEncC->ltp_mem_length + psEncC->frame_length, opus_int16 ); + ALLOC( x_sc_Q10, psEncC->subfr_length, opus_int32 ); + ALLOC( delayedGain_Q10, DECISION_DELAY, opus_int32 ); /* Set up pointers to start of sub frame */ pxq = &NSQ->xq[ psEncC->ltp_mem_length ]; NSQ->sLTP_shp_buf_idx = psEncC->ltp_mem_length; @@ -287,12 +297,13 @@ void silk_NSQ_del_dec( /* DEBUG_STORE_DATA( enc.pcm, &NSQ->xq[psEncC->ltp_mem_length], psEncC->frame_length * sizeof( opus_int16 ) ) */ silk_memmove( NSQ->xq, &NSQ->xq[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int16 ) ); silk_memmove( NSQ->sLTP_shp_Q14, &NSQ->sLTP_shp_Q14[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int32 ) ); + RESTORE_STACK; } /******************************************/ /* Noise shape quantizer for one subframe */ /******************************************/ -static inline void silk_noise_shape_quantizer_del_dec( +static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( silk_nsq_state *NSQ, /* I/O NSQ state */ NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ opus_int signalType, /* I Signal type */ @@ -328,11 +339,13 @@ static inline void silk_noise_shape_quantizer_del_dec( opus_int32 q1_Q0, q1_Q10, q2_Q10, exc_Q14, LPC_exc_Q14, xq_Q14, Gain_Q10; opus_int32 tmp1, tmp2, sLF_AR_shp_Q14; opus_int32 *pred_lag_ptr, *shp_lag_ptr, *psLPC_Q14; - NSQ_sample_struct psSampleState[ MAX_DEL_DEC_STATES ][ 2 ]; + VARDECL( NSQ_sample_pair, psSampleState ); NSQ_del_dec_struct *psDD; NSQ_sample_struct *psSS; + SAVE_STACK; silk_assert( nStatesDelayedDecision > 0 ); + ALLOC( psSampleState, nStatesDelayedDecision, NSQ_sample_pair ); shp_lag_ptr = &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - lag + HARM_SHAPE_FIR_TAPS / 2 ]; pred_lag_ptr = &sLTP_Q15[ NSQ->sLTP_buf_idx - lag + LTP_ORDER / 2 ]; @@ -614,9 +627,10 @@ static inline void silk_noise_shape_quantizer_del_dec( psDD = &psDelDec[ k ]; silk_memcpy( psDD->sLPC_Q14, &psDD->sLPC_Q14[ length ], NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); } + RESTORE_STACK; } -static inline void silk_nsq_del_dec_scale_states( +static OPUS_INLINE void silk_nsq_del_dec_scale_states( const silk_encoder_state *psEncC, /* I Encoder State */ silk_nsq_state *NSQ, /* I/O NSQ state */ NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ diff --git a/src/opus-1.0.2/silk/PLC.c b/src/opus-1.1/silk/PLC.c index 8d547295..01f40014 100644 --- a/src/opus-1.0.2/silk/PLC.c +++ b/src/opus-1.1/silk/PLC.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -38,12 +38,12 @@ static const opus_int16 HARM_ATT_Q15[NB_ATT] = { 32440, 31130 }; /* static const opus_int16 PLC_RAND_ATTENUATE_V_Q15[NB_ATT] = { 31130, 26214 }; /* 0.95, 0.8 */ static const opus_int16 PLC_RAND_ATTENUATE_UV_Q15[NB_ATT] = { 32440, 29491 }; /* 0.99, 0.9 */ -static inline void silk_PLC_update( +static OPUS_INLINE void silk_PLC_update( silk_decoder_state *psDec, /* I/O Decoder state */ silk_decoder_control *psDecCtrl /* I/O Decoder control */ ); -static inline void silk_PLC_conceal( +static OPUS_INLINE void silk_PLC_conceal( silk_decoder_state *psDec, /* I/O Decoder state */ silk_decoder_control *psDecCtrl, /* I/O Decoder control */ opus_int16 frame[] /* O LPC residual signal */ @@ -92,7 +92,7 @@ void silk_PLC( /**************************************************/ /* Update state of PLC */ /**************************************************/ -static inline void silk_PLC_update( +static OPUS_INLINE void silk_PLC_update( silk_decoder_state *psDec, /* I/O Decoder state */ silk_decoder_control *psDecCtrl /* I/O Decoder control */ ) @@ -165,7 +165,7 @@ static inline void silk_PLC_update( psPLC->nb_subfr = psDec->nb_subfr; } -static inline void silk_PLC_conceal( +static OPUS_INLINE void silk_PLC_conceal( silk_decoder_state *psDec, /* I/O Decoder state */ silk_decoder_control *psDecCtrl, /* I/O Decoder control */ opus_int16 frame[] /* O LPC residual signal */ diff --git a/src/opus-1.0.2/silk/PLC.h b/src/opus-1.1/silk/PLC.h index 1d2d9061..f1e2eccc 100644 --- a/src/opus-1.0.2/silk/PLC.h +++ b/src/opus-1.1/silk/PLC.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/SigProc_FIX.h b/src/opus-1.1/silk/SigProc_FIX.h index daa5fd04..1b580579 100644 --- a/src/opus-1.0.2/silk/SigProc_FIX.h +++ b/src/opus-1.1/silk/SigProc_FIX.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -168,12 +168,6 @@ opus_int32 silk_log2lin( const opus_int32 inLog_Q7 /* I input on log scale */ ); -/* Function that returns the maximum absolut value of the input vector */ -opus_int16 silk_int16_array_maxabs( /* O Maximum absolute value, max: 2^15-1 */ - const opus_int16 *vec, /* I Input vector [len] */ - const opus_int32 len /* I Length of input vector */ -); - /* Compute number of bits to right shift the sum of squares of a vector */ /* of int16s to make it fit in an int32 */ void silk_sum_sqr_shift( @@ -233,7 +227,8 @@ void silk_autocorr( opus_int *scale, /* O Scaling of the correlation vector */ const opus_int16 *inputData, /* I Input data to correlate */ const opus_int inputDataSize, /* I Length of input */ - const opus_int correlationCount /* I Number of correlation taps to compute */ + const opus_int correlationCount, /* I Number of correlation taps to compute */ + int arch /* I Run-time architecture */ ); void silk_decode_pitch( @@ -252,10 +247,11 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 opus_int *LTPCorr_Q15, /* I/O Normalized correlation; input: value from previous frame */ opus_int prevLag, /* I Last lag of previous frame; set to zero is unvoiced */ const opus_int32 search_thres1_Q16, /* I First stage threshold for lag candidates 0 - 1 */ - const opus_int search_thres2_Q15, /* I Final threshold for lag candidates 0 - 1 */ + const opus_int search_thres2_Q13, /* I Final threshold for lag candidates 0 - 1 */ const opus_int Fs_kHz, /* I Sample frequency (kHz) */ const opus_int complexity, /* I Complexity setting, 0-2, where 2 is highest */ - const opus_int nb_subfr /* I number of 5 ms subframes */ + const opus_int nb_subfr, /* I number of 5 ms subframes */ + int arch /* I Run-time architecture */ ); /* Compute Normalized Line Spectral Frequencies (NLSFs) from whitening filter coefficients */ @@ -315,7 +311,8 @@ void silk_burg_modified( const opus_int32 minInvGain_Q30, /* I Inverse of max prediction gain */ const opus_int subfr_length, /* I Input signal subframe length (incl. D preceding samples) */ const opus_int nb_subfr, /* I Number of subframes stacked in x */ - const opus_int D /* I Order */ + const opus_int D, /* I Order */ + int arch /* I Run-time architecture */ ); /* Copy and multiply a vector by a constant */ @@ -364,8 +361,8 @@ opus_int64 silk_inner_prod16_aligned_64( /* Rotate a32 right by 'rot' bits. Negative rot values result in rotating left. Output is 32bit int. Note: contemporary compilers recognize the C expression below and - compile it into a 'ror' instruction if available. No need for inline ASM! */ -static inline opus_int32 silk_ROR32( opus_int32 a32, opus_int rot ) + compile it into a 'ror' instruction if available. No need for OPUS_INLINE ASM! */ +static OPUS_INLINE opus_int32 silk_ROR32( opus_int32 a32, opus_int rot ) { opus_uint32 x = (opus_uint32) a32; opus_uint32 r = (opus_uint32) rot; @@ -514,37 +511,37 @@ static inline opus_int32 silk_ROR32( opus_int32 a32, opus_int rot ) #define SILK_FIX_CONST( C, Q ) ((opus_int32)((C) * ((opus_int64)1 << (Q)) + 0.5)) /* silk_min() versions with typecast in the function call */ -static inline opus_int silk_min_int(opus_int a, opus_int b) +static OPUS_INLINE opus_int silk_min_int(opus_int a, opus_int b) { return (((a) < (b)) ? (a) : (b)); } -static inline opus_int16 silk_min_16(opus_int16 a, opus_int16 b) +static OPUS_INLINE opus_int16 silk_min_16(opus_int16 a, opus_int16 b) { return (((a) < (b)) ? (a) : (b)); } -static inline opus_int32 silk_min_32(opus_int32 a, opus_int32 b) +static OPUS_INLINE opus_int32 silk_min_32(opus_int32 a, opus_int32 b) { return (((a) < (b)) ? (a) : (b)); } -static inline opus_int64 silk_min_64(opus_int64 a, opus_int64 b) +static OPUS_INLINE opus_int64 silk_min_64(opus_int64 a, opus_int64 b) { return (((a) < (b)) ? (a) : (b)); } /* silk_min() versions with typecast in the function call */ -static inline opus_int silk_max_int(opus_int a, opus_int b) +static OPUS_INLINE opus_int silk_max_int(opus_int a, opus_int b) { return (((a) > (b)) ? (a) : (b)); } -static inline opus_int16 silk_max_16(opus_int16 a, opus_int16 b) +static OPUS_INLINE opus_int16 silk_max_16(opus_int16 a, opus_int16 b) { return (((a) > (b)) ? (a) : (b)); } -static inline opus_int32 silk_max_32(opus_int32 a, opus_int32 b) +static OPUS_INLINE opus_int32 silk_max_32(opus_int32 a, opus_int32 b) { return (((a) > (b)) ? (a) : (b)); } -static inline opus_int64 silk_max_64(opus_int64 a, opus_int64 b) +static OPUS_INLINE opus_int64 silk_max_64(opus_int64 a, opus_int64 b) { return (((a) > (b)) ? (a) : (b)); } @@ -582,6 +579,14 @@ static inline opus_int64 silk_max_64(opus_int64 a, opus_int64 b) #include "MacroCount.h" #include "MacroDebug.h" +#ifdef OPUS_ARM_INLINE_ASM +#include "arm/SigProc_FIX_armv4.h" +#endif + +#ifdef OPUS_ARM_INLINE_EDSP +#include "arm/SigProc_FIX_armv5e.h" +#endif + #ifdef __cplusplus } #endif diff --git a/src/opus-1.0.2/silk/VAD.c b/src/opus-1.1/silk/VAD.c index bac89b44..a8090981 100644 --- a/src/opus-1.0.2/silk/VAD.c +++ b/src/opus-1.1/silk/VAD.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,9 +30,10 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main.h" +#include "stack_alloc.h" /* Silk VAD noise level estimation */ -static inline void silk_VAD_GetNoiseLevels( +static OPUS_INLINE void silk_VAD_GetNoiseLevels( const opus_int32 pX[ VAD_N_BANDS ], /* I subband energies */ silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ ); @@ -82,15 +83,19 @@ opus_int silk_VAD_GetSA_Q8( /* O Return v ) { opus_int SA_Q15, pSNR_dB_Q7, input_tilt; - opus_int decimated_framelength, dec_subframe_length, dec_subframe_offset, SNR_Q7, i, b, s; + opus_int decimated_framelength1, decimated_framelength2; + opus_int decimated_framelength; + opus_int dec_subframe_length, dec_subframe_offset, SNR_Q7, i, b, s; opus_int32 sumSquared, smooth_coef_Q16; opus_int16 HPstateTmp; - opus_int16 X[ VAD_N_BANDS ][ MAX_FRAME_LENGTH / 2 ]; + VARDECL( opus_int16, X ); opus_int32 Xnrg[ VAD_N_BANDS ]; opus_int32 NrgToNoiseRatio_Q8[ VAD_N_BANDS ]; opus_int32 speech_nrg, x_tmp; + opus_int X_offset[ VAD_N_BANDS ]; opus_int ret = 0; silk_VAD_state *psSilk_VAD = &psEncC->sVAD; + SAVE_STACK; /* Safety checks */ silk_assert( VAD_N_BANDS == 4 ); @@ -101,26 +106,46 @@ opus_int silk_VAD_GetSA_Q8( /* O Return v /***********************/ /* Filter and Decimate */ /***********************/ + decimated_framelength1 = silk_RSHIFT( psEncC->frame_length, 1 ); + decimated_framelength2 = silk_RSHIFT( psEncC->frame_length, 2 ); + decimated_framelength = silk_RSHIFT( psEncC->frame_length, 3 ); + /* Decimate into 4 bands: + 0 L 3L L 3L 5L + - -- - -- -- + 8 8 2 4 4 + + [0-1 kHz| temp. |1-2 kHz| 2-4 kHz | 4-8 kHz | + + They're arranged to allow the minimal ( frame_length / 4 ) extra + scratch space during the downsampling process */ + X_offset[ 0 ] = 0; + X_offset[ 1 ] = decimated_framelength + decimated_framelength2; + X_offset[ 2 ] = X_offset[ 1 ] + decimated_framelength; + X_offset[ 3 ] = X_offset[ 2 ] + decimated_framelength2; + ALLOC( X, X_offset[ 3 ] + decimated_framelength1, opus_int16 ); + /* 0-8 kHz to 0-4 kHz and 4-8 kHz */ - silk_ana_filt_bank_1( pIn, &psSilk_VAD->AnaState[ 0 ], &X[ 0 ][ 0 ], &X[ 3 ][ 0 ], psEncC->frame_length ); + silk_ana_filt_bank_1( pIn, &psSilk_VAD->AnaState[ 0 ], + X, &X[ X_offset[ 3 ] ], psEncC->frame_length ); /* 0-4 kHz to 0-2 kHz and 2-4 kHz */ - silk_ana_filt_bank_1( &X[ 0 ][ 0 ], &psSilk_VAD->AnaState1[ 0 ], &X[ 0 ][ 0 ], &X[ 2 ][ 0 ], silk_RSHIFT( psEncC->frame_length, 1 ) ); + silk_ana_filt_bank_1( X, &psSilk_VAD->AnaState1[ 0 ], + X, &X[ X_offset[ 2 ] ], decimated_framelength1 ); /* 0-2 kHz to 0-1 kHz and 1-2 kHz */ - silk_ana_filt_bank_1( &X[ 0 ][ 0 ], &psSilk_VAD->AnaState2[ 0 ], &X[ 0 ][ 0 ], &X[ 1 ][ 0 ], silk_RSHIFT( psEncC->frame_length, 2 ) ); + silk_ana_filt_bank_1( X, &psSilk_VAD->AnaState2[ 0 ], + X, &X[ X_offset[ 1 ] ], decimated_framelength2 ); /*********************************************/ /* HP filter on lowest band (differentiator) */ /*********************************************/ - decimated_framelength = silk_RSHIFT( psEncC->frame_length, 3 ); - X[ 0 ][ decimated_framelength - 1 ] = silk_RSHIFT( X[ 0 ][ decimated_framelength - 1 ], 1 ); - HPstateTmp = X[ 0 ][ decimated_framelength - 1 ]; + X[ decimated_framelength - 1 ] = silk_RSHIFT( X[ decimated_framelength - 1 ], 1 ); + HPstateTmp = X[ decimated_framelength - 1 ]; for( i = decimated_framelength - 1; i > 0; i-- ) { - X[ 0 ][ i - 1 ] = silk_RSHIFT( X[ 0 ][ i - 1 ], 1 ); - X[ 0 ][ i ] -= X[ 0 ][ i - 1 ]; + X[ i - 1 ] = silk_RSHIFT( X[ i - 1 ], 1 ); + X[ i ] -= X[ i - 1 ]; } - X[ 0 ][ 0 ] -= psSilk_VAD->HPstate; + X[ 0 ] -= psSilk_VAD->HPstate; psSilk_VAD->HPstate = HPstateTmp; /*************************************/ @@ -142,7 +167,8 @@ opus_int silk_VAD_GetSA_Q8( /* O Return v for( i = 0; i < dec_subframe_length; i++ ) { /* The energy will be less than dec_subframe_length * ( silk_int16_MIN / 8 ) ^ 2. */ /* Therefore we can accumulate with no risk of overflow (unless dec_subframe_length > 128) */ - x_tmp = silk_RSHIFT( X[ b ][ i + dec_subframe_offset ], 3 ); + x_tmp = silk_RSHIFT( + X[ X_offset[ b ] + i + dec_subframe_offset ], 3 ); sumSquared = silk_SMLABB( sumSquared, x_tmp, x_tmp ); /* Safety check */ @@ -263,13 +289,14 @@ opus_int silk_VAD_GetSA_Q8( /* O Return v psEncC->input_quality_bands_Q15[ b ] = silk_sigm_Q15( silk_RSHIFT( SNR_Q7 - 16 * 128, 4 ) ); } + RESTORE_STACK; return( ret ); } /**************************/ /* Noise level estimation */ /**************************/ -static inline void silk_VAD_GetNoiseLevels( +static OPUS_INLINE void silk_VAD_GetNoiseLevels( const opus_int32 pX[ VAD_N_BANDS ], /* I subband energies */ silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ ) diff --git a/src/opus-1.0.2/silk/VQ_WMat_EC.c b/src/opus-1.1/silk/VQ_WMat_EC.c index a308cfbf..13d5d34e 100644 --- a/src/opus-1.0.2/silk/VQ_WMat_EC.c +++ b/src/opus-1.1/silk/VQ_WMat_EC.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -35,15 +35,18 @@ POSSIBILITY OF SUCH DAMAGE. void silk_VQ_WMat_EC( opus_int8 *ind, /* O index of best codebook vector */ opus_int32 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ + opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ const opus_int16 *in_Q14, /* I input vector to be quantized */ const opus_int32 *W_Q18, /* I weighting matrix */ const opus_int8 *cb_Q7, /* I codebook */ + const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ const opus_int mu_Q9, /* I tradeoff betw. weighted error and rate */ + const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ opus_int L /* I number of vectors in codebook */ ) { - opus_int k; + opus_int k, gain_tmp_Q7; const opus_int8 *cb_row_Q7; opus_int16 diff_Q14[ 5 ]; opus_int32 sum1_Q14, sum2_Q16; @@ -52,6 +55,8 @@ void silk_VQ_WMat_EC( *rate_dist_Q14 = silk_int32_MAX; cb_row_Q7 = cb_Q7; for( k = 0; k < L; k++ ) { + gain_tmp_Q7 = cb_gain_Q7[k]; + diff_Q14[ 0 ] = in_Q14[ 0 ] - silk_LSHIFT( cb_row_Q7[ 0 ], 7 ); diff_Q14[ 1 ] = in_Q14[ 1 ] - silk_LSHIFT( cb_row_Q7[ 1 ], 7 ); diff_Q14[ 2 ] = in_Q14[ 2 ] - silk_LSHIFT( cb_row_Q7[ 2 ], 7 ); @@ -61,6 +66,9 @@ void silk_VQ_WMat_EC( /* Weighted rate */ sum1_Q14 = silk_SMULBB( mu_Q9, cl_Q5[ k ] ); + /* Penalty for too large gain */ + sum1_Q14 = silk_ADD_LSHIFT32( sum1_Q14, silk_max( silk_SUB32( gain_tmp_Q7, max_gain_Q7 ), 0 ), 10 ); + silk_assert( sum1_Q14 >= 0 ); /* first row of W_Q18 */ @@ -103,6 +111,7 @@ void silk_VQ_WMat_EC( if( sum1_Q14 < *rate_dist_Q14 ) { *rate_dist_Q14 = sum1_Q14; *ind = (opus_int8)k; + *gain_Q7 = gain_tmp_Q7; } /* Go to next cbk vector */ diff --git a/src/opus-1.0.2/silk/ana_filt_bank_1.c b/src/opus-1.1/silk/ana_filt_bank_1.c index 4e04bef3..24cfb03f 100644 --- a/src/opus-1.0.2/silk/ana_filt_bank_1.c +++ b/src/opus-1.1/silk/ana_filt_bank_1.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.1/silk/arm/SigProc_FIX_armv4.h b/src/opus-1.1/silk/arm/SigProc_FIX_armv4.h new file mode 100644 index 00000000..ff62b1e5 --- /dev/null +++ b/src/opus-1.1/silk/arm/SigProc_FIX_armv4.h @@ -0,0 +1,47 @@ +/*********************************************************************** +Copyright (C) 2013 Xiph.Org Foundation and contributors +Copyright (c) 2013 Parrot +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_SIGPROC_FIX_ARMv4_H +#define SILK_SIGPROC_FIX_ARMv4_H + +#undef silk_MLA +static OPUS_INLINE opus_int32 silk_MLA_armv4(opus_int32 a, opus_int32 b, + opus_int32 c) +{ + opus_int32 res; + __asm__( + "#silk_MLA\n\t" + "mla %0, %1, %2, %3\n\t" + : "=&r"(res) + : "r"(b), "r"(c), "r"(a) + ); + return res; +} +#define silk_MLA(a, b, c) (silk_MLA_armv4(a, b, c)) + +#endif diff --git a/src/opus-1.1/silk/arm/SigProc_FIX_armv5e.h b/src/opus-1.1/silk/arm/SigProc_FIX_armv5e.h new file mode 100644 index 00000000..617a09ca --- /dev/null +++ b/src/opus-1.1/silk/arm/SigProc_FIX_armv5e.h @@ -0,0 +1,61 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Copyright (c) 2013 Parrot +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_SIGPROC_FIX_ARMv5E_H +#define SILK_SIGPROC_FIX_ARMv5E_H + +#undef silk_SMULTT +static OPUS_INLINE opus_int32 silk_SMULTT_armv5e(opus_int32 a, opus_int32 b) +{ + opus_int32 res; + __asm__( + "#silk_SMULTT\n\t" + "smultt %0, %1, %2\n\t" + : "=r"(res) + : "%r"(a), "r"(b) + ); + return res; +} +#define silk_SMULTT(a, b) (silk_SMULTT_armv5e(a, b)) + +#undef silk_SMLATT +static OPUS_INLINE opus_int32 silk_SMLATT_armv5e(opus_int32 a, opus_int32 b, + opus_int32 c) +{ + opus_int32 res; + __asm__( + "#silk_SMLATT\n\t" + "smlatt %0, %1, %2, %3\n\t" + : "=r"(res) + : "%r"(b), "r"(c), "r"(a) + ); + return res; +} +#define silk_SMLATT(a, b, c) (silk_SMLATT_armv5e(a, b, c)) + +#endif diff --git a/src/opus-1.1/silk/arm/macros_armv4.h b/src/opus-1.1/silk/arm/macros_armv4.h new file mode 100644 index 00000000..3f30e972 --- /dev/null +++ b/src/opus-1.1/silk/arm/macros_armv4.h @@ -0,0 +1,103 @@ +/*********************************************************************** +Copyright (C) 2013 Xiph.Org Foundation and contributors. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_MACROS_ARMv4_H +#define SILK_MACROS_ARMv4_H + +/* (a32 * (opus_int32)((opus_int16)(b32))) >> 16 output have to be 32bit int */ +#undef silk_SMULWB +static OPUS_INLINE opus_int32 silk_SMULWB_armv4(opus_int32 a, opus_int16 b) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#silk_SMULWB\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(a), "r"(b<<16) + ); + return rd_hi; +} +#define silk_SMULWB(a, b) (silk_SMULWB_armv4(a, b)) + +/* a32 + (b32 * (opus_int32)((opus_int16)(c32))) >> 16 output have to be 32bit int */ +#undef silk_SMLAWB +#define silk_SMLAWB(a, b, c) ((a) + silk_SMULWB(b, c)) + +/* (a32 * (b32 >> 16)) >> 16 */ +#undef silk_SMULWT +static OPUS_INLINE opus_int32 silk_SMULWT_armv4(opus_int32 a, opus_int32 b) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#silk_SMULWT\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(a), "r"(b&~0xFFFF) + ); + return rd_hi; +} +#define silk_SMULWT(a, b) (silk_SMULWT_armv4(a, b)) + +/* a32 + (b32 * (c32 >> 16)) >> 16 */ +#undef silk_SMLAWT +#define silk_SMLAWT(a, b, c) ((a) + silk_SMULWT(b, c)) + +/* (a32 * b32) >> 16 */ +#undef silk_SMULWW +static OPUS_INLINE opus_int32 silk_SMULWW_armv4(opus_int32 a, opus_int32 b) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#silk_SMULWW\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(a), "r"(b) + ); + return (rd_hi<<16)+(rd_lo>>16); +} +#define silk_SMULWW(a, b) (silk_SMULWW_armv4(a, b)) + +#undef silk_SMLAWW +static OPUS_INLINE opus_int32 silk_SMLAWW_armv4(opus_int32 a, opus_int32 b, + opus_int32 c) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#silk_SMLAWW\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(b), "r"(c) + ); + return a+(rd_hi<<16)+(rd_lo>>16); +} +#define silk_SMLAWW(a, b, c) (silk_SMLAWW_armv4(a, b, c)) + +#endif /* SILK_MACROS_ARMv4_H */ diff --git a/src/opus-1.1/silk/arm/macros_armv5e.h b/src/opus-1.1/silk/arm/macros_armv5e.h new file mode 100644 index 00000000..aad4117e --- /dev/null +++ b/src/opus-1.1/silk/arm/macros_armv5e.h @@ -0,0 +1,213 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Copyright (c) 2013 Parrot +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_MACROS_ARMv5E_H +#define SILK_MACROS_ARMv5E_H + +/* (a32 * (opus_int32)((opus_int16)(b32))) >> 16 output have to be 32bit int */ +#undef silk_SMULWB +static OPUS_INLINE opus_int32 silk_SMULWB_armv5e(opus_int32 a, opus_int16 b) +{ + int res; + __asm__( + "#silk_SMULWB\n\t" + "smulwb %0, %1, %2\n\t" + : "=r"(res) + : "r"(a), "r"(b) + ); + return res; +} +#define silk_SMULWB(a, b) (silk_SMULWB_armv5e(a, b)) + +/* a32 + (b32 * (opus_int32)((opus_int16)(c32))) >> 16 output have to be 32bit int */ +#undef silk_SMLAWB +static OPUS_INLINE opus_int32 silk_SMLAWB_armv5e(opus_int32 a, opus_int32 b, + opus_int16 c) +{ + int res; + __asm__( + "#silk_SMLAWB\n\t" + "smlawb %0, %1, %2, %3\n\t" + : "=r"(res) + : "r"(b), "r"(c), "r"(a) + ); + return res; +} +#define silk_SMLAWB(a, b, c) (silk_SMLAWB_armv5e(a, b, c)) + +/* (a32 * (b32 >> 16)) >> 16 */ +#undef silk_SMULWT +static OPUS_INLINE opus_int32 silk_SMULWT_armv5e(opus_int32 a, opus_int32 b) +{ + int res; + __asm__( + "#silk_SMULWT\n\t" + "smulwt %0, %1, %2\n\t" + : "=r"(res) + : "r"(a), "r"(b) + ); + return res; +} +#define silk_SMULWT(a, b) (silk_SMULWT_armv5e(a, b)) + +/* a32 + (b32 * (c32 >> 16)) >> 16 */ +#undef silk_SMLAWT +static OPUS_INLINE opus_int32 silk_SMLAWT_armv5e(opus_int32 a, opus_int32 b, + opus_int32 c) +{ + int res; + __asm__( + "#silk_SMLAWT\n\t" + "smlawt %0, %1, %2, %3\n\t" + : "=r"(res) + : "r"(b), "r"(c), "r"(a) + ); + return res; +} +#define silk_SMLAWT(a, b, c) (silk_SMLAWT_armv5e(a, b, c)) + +/* (opus_int32)((opus_int16)(a3))) * (opus_int32)((opus_int16)(b32)) output have to be 32bit int */ +#undef silk_SMULBB +static OPUS_INLINE opus_int32 silk_SMULBB_armv5e(opus_int32 a, opus_int32 b) +{ + int res; + __asm__( + "#silk_SMULBB\n\t" + "smulbb %0, %1, %2\n\t" + : "=r"(res) + : "%r"(a), "r"(b) + ); + return res; +} +#define silk_SMULBB(a, b) (silk_SMULBB_armv5e(a, b)) + +/* a32 + (opus_int32)((opus_int16)(b32)) * (opus_int32)((opus_int16)(c32)) output have to be 32bit int */ +#undef silk_SMLABB +static OPUS_INLINE opus_int32 silk_SMLABB_armv5e(opus_int32 a, opus_int32 b, + opus_int32 c) +{ + int res; + __asm__( + "#silk_SMLABB\n\t" + "smlabb %0, %1, %2, %3\n\t" + : "=r"(res) + : "%r"(b), "r"(c), "r"(a) + ); + return res; +} +#define silk_SMLABB(a, b, c) (silk_SMLABB_armv5e(a, b, c)) + +/* (opus_int32)((opus_int16)(a32)) * (b32 >> 16) */ +#undef silk_SMULBT +static OPUS_INLINE opus_int32 silk_SMULBT_armv5e(opus_int32 a, opus_int32 b) +{ + int res; + __asm__( + "#silk_SMULBT\n\t" + "smulbt %0, %1, %2\n\t" + : "=r"(res) + : "r"(a), "r"(b) + ); + return res; +} +#define silk_SMULBT(a, b) (silk_SMULBT_armv5e(a, b)) + +/* a32 + (opus_int32)((opus_int16)(b32)) * (c32 >> 16) */ +#undef silk_SMLABT +static OPUS_INLINE opus_int32 silk_SMLABT_armv5e(opus_int32 a, opus_int32 b, + opus_int32 c) +{ + int res; + __asm__( + "#silk_SMLABT\n\t" + "smlabt %0, %1, %2, %3\n\t" + : "=r"(res) + : "r"(b), "r"(c), "r"(a) + ); + return res; +} +#define silk_SMLABT(a, b, c) (silk_SMLABT_armv5e(a, b, c)) + +/* add/subtract with output saturated */ +#undef silk_ADD_SAT32 +static OPUS_INLINE opus_int32 silk_ADD_SAT32_armv5e(opus_int32 a, opus_int32 b) +{ + int res; + __asm__( + "#silk_ADD_SAT32\n\t" + "qadd %0, %1, %2\n\t" + : "=r"(res) + : "%r"(a), "r"(b) + ); + return res; +} +#define silk_ADD_SAT32(a, b) (silk_ADD_SAT32_armv5e(a, b)) + +#undef silk_SUB_SAT32 +static OPUS_INLINE opus_int32 silk_SUB_SAT32_armv5e(opus_int32 a, opus_int32 b) +{ + int res; + __asm__( + "#silk_SUB_SAT32\n\t" + "qsub %0, %1, %2\n\t" + : "=r"(res) + : "r"(a), "r"(b) + ); + return res; +} +#define silk_SUB_SAT32(a, b) (silk_SUB_SAT32_armv5e(a, b)) + +#undef silk_CLZ16 +static OPUS_INLINE opus_int32 silk_CLZ16_armv5(opus_int16 in16) +{ + int res; + __asm__( + "#silk_CLZ16\n\t" + "clz %0, %1;\n" + : "=r"(res) + : "r"(in16<<16|0x8000) + ); + return res; +} +#define silk_CLZ16(in16) (silk_CLZ16_armv5(in16)) + +#undef silk_CLZ32 +static OPUS_INLINE opus_int32 silk_CLZ32_armv5(opus_int32 in32) +{ + int res; + __asm__( + "#silk_CLZ32\n\t" + "clz %0, %1\n\t" + : "=r"(res) + : "r"(in32) + ); + return res; +} +#define silk_CLZ32(in32) (silk_CLZ32_armv5(in32)) + +#endif /* SILK_MACROS_ARMv5E_H */ diff --git a/src/opus-1.0.2/silk/biquad_alt.c b/src/opus-1.1/silk/biquad_alt.c index a639e21a..d55f5ee9 100644 --- a/src/opus-1.0.2/silk/biquad_alt.c +++ b/src/opus-1.1/silk/biquad_alt.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/bwexpander.c b/src/opus-1.1/silk/bwexpander.c index 77ea1163..2eb44566 100644 --- a/src/opus-1.0.2/silk/bwexpander.c +++ b/src/opus-1.1/silk/bwexpander.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/bwexpander_32.c b/src/opus-1.1/silk/bwexpander_32.c index 5ad92dd4..d0010f73 100644 --- a/src/opus-1.0.2/silk/bwexpander_32.c +++ b/src/opus-1.1/silk/bwexpander_32.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/check_control_input.c b/src/opus-1.1/silk/check_control_input.c index 972a480d..b5de9ce4 100644 --- a/src/opus-1.0.2/silk/check_control_input.c +++ b/src/opus-1.1/silk/check_control_input.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/code_signs.c b/src/opus-1.1/silk/code_signs.c index 9893cdd8..0419ea26 100644 --- a/src/opus-1.0.2/silk/code_signs.c +++ b/src/opus-1.1/silk/code_signs.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/control.h b/src/opus-1.1/silk/control.h index c52ec3fe..747e5426 100644 --- a/src/opus-1.0.2/silk/control.h +++ b/src/opus-1.1/silk/control.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -92,6 +92,9 @@ typedef struct { /* I: Opus encoder is allowing us to switch bandwidth */ opus_int opusCanSwitch; + /* I: Make frames as independent as possible (but still use LPC) */ + opus_int reducedDependency; + /* O: Internal sampling rate used, in Hertz; 8000/12000/16000 */ opus_int32 internalSampleRate; diff --git a/src/opus-1.0.2/silk/control_SNR.c b/src/opus-1.1/silk/control_SNR.c index 08e4e1a1..f04e69fc 100644 --- a/src/opus-1.0.2/silk/control_SNR.c +++ b/src/opus-1.1/silk/control_SNR.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/control_audio_bandwidth.c b/src/opus-1.1/silk/control_audio_bandwidth.c index b645dd57..4f9bc5cb 100644 --- a/src/opus-1.0.2/silk/control_audio_bandwidth.c +++ b/src/opus-1.1/silk/control_audio_bandwidth.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -115,6 +115,9 @@ opus_int silk_control_audio_bandwidth( psEncC->sLP.mode = 1; } } + } else { + if (psEncC->sLP.mode<0) + psEncC->sLP.mode = 1; } } } diff --git a/src/opus-1.0.2/silk/control_codec.c b/src/opus-1.1/silk/control_codec.c index ecc338ce..1f674bdd 100644 --- a/src/opus-1.0.2/silk/control_codec.c +++ b/src/opus-1.1/silk/control_codec.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -35,6 +35,7 @@ POSSIBILITY OF SUCH DAMAGE. #include "main_FLP.h" #define silk_encoder_state_Fxx silk_encoder_state_FLP #endif +#include "stack_alloc.h" #include "tuning_parameters.h" #include "pitch_est_defines.h" @@ -54,7 +55,7 @@ static opus_int silk_setup_complexity( opus_int Complexity /* I */ ); -static inline opus_int silk_setup_LBRR( +static OPUS_INLINE opus_int silk_setup_LBRR( silk_encoder_state *psEncC, /* I/O */ const opus_int32 TargetRate_bps /* I */ ); @@ -137,7 +138,7 @@ static opus_int silk_setup_resamplers( ) { opus_int ret = SILK_NO_ERROR; - opus_int32 nSamples_temp; + SAVE_STACK; if( psEnc->sCmn.fs_kHz != fs_kHz || psEnc->sCmn.prev_API_fs_Hz != psEnc->sCmn.API_fs_Hz ) { @@ -145,44 +146,54 @@ static opus_int silk_setup_resamplers( /* Initialize the resampler for enc_API.c preparing resampling from API_fs_Hz to fs_kHz */ ret += silk_resampler_init( &psEnc->sCmn.resampler_state, psEnc->sCmn.API_fs_Hz, fs_kHz * 1000, 1 ); } else { - /* Allocate worst case space for temporary upsampling, 8 to 48 kHz, so a factor 6 */ - opus_int16 x_buf_API_fs_Hz[ ( 2 * MAX_FRAME_LENGTH_MS + LA_SHAPE_MS ) * MAX_API_FS_KHZ ]; - silk_resampler_state_struct temp_resampler_state; + VARDECL( opus_int16, x_buf_API_fs_Hz ); + VARDECL( silk_resampler_state_struct, temp_resampler_state ); #ifdef FIXED_POINT opus_int16 *x_bufFIX = psEnc->x_buf; #else - opus_int16 x_bufFIX[ 2 * MAX_FRAME_LENGTH + LA_SHAPE_MAX ]; + VARDECL( opus_int16, x_bufFIX ); + opus_int32 new_buf_samples; #endif + opus_int32 api_buf_samples; + opus_int32 old_buf_samples; + opus_int32 buf_length_ms; - nSamples_temp = silk_LSHIFT( psEnc->sCmn.frame_length, 1 ) + LA_SHAPE_MS * psEnc->sCmn.fs_kHz; + buf_length_ms = silk_LSHIFT( psEnc->sCmn.nb_subfr * 5, 1 ) + LA_SHAPE_MS; + old_buf_samples = buf_length_ms * psEnc->sCmn.fs_kHz; #ifndef FIXED_POINT - silk_float2short_array( x_bufFIX, psEnc->x_buf, nSamples_temp ); + new_buf_samples = buf_length_ms * fs_kHz; + ALLOC( x_bufFIX, silk_max( old_buf_samples, new_buf_samples ), + opus_int16 ); + silk_float2short_array( x_bufFIX, psEnc->x_buf, old_buf_samples ); #endif /* Initialize resampler for temporary resampling of x_buf data to API_fs_Hz */ - ret += silk_resampler_init( &temp_resampler_state, silk_SMULBB( psEnc->sCmn.fs_kHz, 1000 ), psEnc->sCmn.API_fs_Hz, 0 ); + ALLOC( temp_resampler_state, 1, silk_resampler_state_struct ); + ret += silk_resampler_init( temp_resampler_state, silk_SMULBB( psEnc->sCmn.fs_kHz, 1000 ), psEnc->sCmn.API_fs_Hz, 0 ); - /* Temporary resampling of x_buf data to API_fs_Hz */ - ret += silk_resampler( &temp_resampler_state, x_buf_API_fs_Hz, x_bufFIX, nSamples_temp ); + /* Calculate number of samples to temporarily upsample */ + api_buf_samples = buf_length_ms * silk_DIV32_16( psEnc->sCmn.API_fs_Hz, 1000 ); - /* Calculate number of samples that has been temporarily upsampled */ - nSamples_temp = silk_DIV32_16( nSamples_temp * psEnc->sCmn.API_fs_Hz, silk_SMULBB( psEnc->sCmn.fs_kHz, 1000 ) ); + /* Temporary resampling of x_buf data to API_fs_Hz */ + ALLOC( x_buf_API_fs_Hz, api_buf_samples, opus_int16 ); + ret += silk_resampler( temp_resampler_state, x_buf_API_fs_Hz, x_bufFIX, old_buf_samples ); /* Initialize the resampler for enc_API.c preparing resampling from API_fs_Hz to fs_kHz */ ret += silk_resampler_init( &psEnc->sCmn.resampler_state, psEnc->sCmn.API_fs_Hz, silk_SMULBB( fs_kHz, 1000 ), 1 ); /* Correct resampler state by resampling buffered data from API_fs_Hz to fs_kHz */ - ret += silk_resampler( &psEnc->sCmn.resampler_state, x_bufFIX, x_buf_API_fs_Hz, nSamples_temp ); + ret += silk_resampler( &psEnc->sCmn.resampler_state, x_bufFIX, x_buf_API_fs_Hz, api_buf_samples ); #ifndef FIXED_POINT - silk_short2float_array( psEnc->x_buf, x_bufFIX, ( 2 * MAX_FRAME_LENGTH_MS + LA_SHAPE_MS ) * fs_kHz ); + silk_short2float_array( psEnc->x_buf, x_bufFIX, new_buf_samples); #endif } } psEnc->sCmn.prev_API_fs_Hz = psEnc->sCmn.API_fs_Hz; + RESTORE_STACK; return ret; } @@ -381,7 +392,7 @@ static opus_int silk_setup_complexity( return ret; } -static inline opus_int silk_setup_LBRR( +static OPUS_INLINE opus_int silk_setup_LBRR( silk_encoder_state *psEncC, /* I/O */ const opus_int32 TargetRate_bps /* I */ ) diff --git a/src/opus-1.0.2/silk/debug.c b/src/opus-1.1/silk/debug.c index 9aa16cc8..9253faf7 100644 --- a/src/opus-1.0.2/silk/debug.c +++ b/src/opus-1.1/silk/debug.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/debug.h b/src/opus-1.1/silk/debug.h index 8ae7094f..efb6d3e9 100644 --- a/src/opus-1.0.2/silk/debug.h +++ b/src/opus-1.1/silk/debug.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -28,10 +28,6 @@ POSSIBILITY OF SUCH DAMAGE. #ifndef SILK_DEBUG_H #define SILK_DEBUG_H -#ifdef _WIN32 -#define _CRT_SECURE_NO_DEPRECATE 1 -#endif - #include "typedef.h" #include <stdio.h> /* file writing */ #include <string.h> /* strcpy, strcmp */ @@ -69,7 +65,6 @@ unsigned long GetHighResolutionTime(void); /* O time in usec*/ #if (defined(_WIN32) || defined(_WINCE)) #include <windows.h> /* timer */ -#pragma warning( disable : 4996 ) /* stop bitching about strcpy in TIC()*/ #else /* Linux or Mac*/ #include <sys/time.h> #endif diff --git a/src/opus-1.0.2/silk/dec_API.c b/src/opus-1.1/silk/dec_API.c index 68403b7c..4cbcf715 100644 --- a/src/opus-1.0.2/silk/dec_API.c +++ b/src/opus-1.1/silk/dec_API.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -69,6 +69,9 @@ opus_int silk_InitDecoder( /* O Returns error co for( n = 0; n < DECODER_NUM_CHANNELS; n++ ) { ret = silk_init_decoder( &channel_state[ n ] ); } + silk_memset(&((silk_decoder *)decState)->sStereo, 0, sizeof(((silk_decoder *)decState)->sStereo)); + /* Not strictly needed, but it's cleaner that way */ + ((silk_decoder *)decState)->prev_decode_only_middle = 0; return ret; } @@ -97,6 +100,8 @@ opus_int silk_Decode( /* O Returns error co opus_int stereo_to_mono; SAVE_STACK; + silk_assert( decControl->nChannelsInternal == 1 || decControl->nChannelsInternal == 2 ); + /**********************************/ /* Test if first frame in payload */ /**********************************/ @@ -300,7 +305,7 @@ opus_int silk_Decode( /* O Returns error co /* Set up pointers to temp buffers */ ALLOC( samplesOut2_tmp, - decControl->nChannelsAPI == 2 ? *nSamplesOut : 0, opus_int16 ); + decControl->nChannelsAPI == 2 ? *nSamplesOut : ALLOC_NONE, opus_int16 ); if( decControl->nChannelsAPI == 2 ) { resample_out_ptr = samplesOut2_tmp; } else { diff --git a/src/opus-1.0.2/silk/decode_core.c b/src/opus-1.1/silk/decode_core.c index 0365ffdf..a820bf11 100644 --- a/src/opus-1.0.2/silk/decode_core.c +++ b/src/opus-1.1/silk/decode_core.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/decode_frame.c b/src/opus-1.1/silk/decode_frame.c index 3e4a6e2b..abc00a3d 100644 --- a/src/opus-1.0.2/silk/decode_frame.c +++ b/src/opus-1.1/silk/decode_frame.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/decode_indices.c b/src/opus-1.1/silk/decode_indices.c index 69172102..7afe5c26 100644 --- a/src/opus-1.0.2/silk/decode_indices.c +++ b/src/opus-1.1/silk/decode_indices.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/decode_parameters.c b/src/opus-1.1/silk/decode_parameters.c index e4c7e7a4..e345b1dc 100644 --- a/src/opus-1.0.2/silk/decode_parameters.c +++ b/src/opus-1.1/silk/decode_parameters.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/decode_pitch.c b/src/opus-1.1/silk/decode_pitch.c index 80fb4d9f..fedbc6a5 100644 --- a/src/opus-1.0.2/silk/decode_pitch.c +++ b/src/opus-1.1/silk/decode_pitch.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/decode_pulses.c b/src/opus-1.1/silk/decode_pulses.c index 1c781a0b..e8a87c2a 100644 --- a/src/opus-1.0.2/silk/decode_pulses.c +++ b/src/opus-1.1/silk/decode_pulses.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/decoder_set_fs.c b/src/opus-1.1/silk/decoder_set_fs.c index 38ac249c..eef0fd25 100644 --- a/src/opus-1.0.2/silk/decoder_set_fs.c +++ b/src/opus-1.1/silk/decoder_set_fs.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/define.h b/src/opus-1.1/silk/define.h index f74f4869..c47aca9f 100644 --- a/src/opus-1.0.2/silk/define.h +++ b/src/opus-1.1/silk/define.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/enc_API.c b/src/opus-1.1/silk/enc_API.c index ec7915ce..43739efc 100644 --- a/src/opus-1.0.2/silk/enc_API.c +++ b/src/opus-1.1/silk/enc_API.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -32,6 +32,7 @@ POSSIBILITY OF SUCH DAMAGE. #include "API.h" #include "control.h" #include "typedef.h" +#include "stack_alloc.h" #include "structs.h" #include "tuning_parameters.h" #ifdef FIXED_POINT @@ -68,6 +69,7 @@ opus_int silk_Get_Encoder_Size( /* O Returns error co /*************************/ opus_int silk_InitEncoder( /* O Returns error code */ void *encState, /* I/O State */ + int arch, /* I Run-time architecture */ silk_EncControlStruct *encStatus /* O Encoder Status */ ) { @@ -79,7 +81,7 @@ opus_int silk_InitEncoder( /* O Returns error co /* Reset encoder */ silk_memset( psEnc, 0, sizeof( silk_encoder ) ); for( n = 0; n < ENCODER_NUM_CHANNELS; n++ ) { - if( ret += silk_init_encoder( &psEnc->state_Fxx[ n ] ) ) { + if( ret += silk_init_encoder( &psEnc->state_Fxx[ n ], arch ) ) { silk_assert( 0 ); } } @@ -146,18 +148,26 @@ opus_int silk_Encode( /* O Returns error co ) { opus_int n, i, nBits, flags, tmp_payloadSize_ms = 0, tmp_complexity = 0, ret = 0; - opus_int nSamplesToBuffer, nBlocksOf10ms, nSamplesFromInput = 0; + opus_int nSamplesToBuffer, nSamplesToBufferMax, nBlocksOf10ms; + opus_int nSamplesFromInput = 0, nSamplesFromInputMax; opus_int speech_act_thr_for_switch_Q8; opus_int32 TargetRate_bps, MStargetRates_bps[ 2 ], channelRate_bps, LBRR_symbol, sum; silk_encoder *psEnc = ( silk_encoder * )encState; - opus_int16 buf[ MAX_FRAME_LENGTH_MS * MAX_API_FS_KHZ ]; + VARDECL( opus_int16, buf ); opus_int transition, curr_block, tot_blocks; + SAVE_STACK; + if (encControl->reducedDependency) + { + psEnc->state_Fxx[0].sCmn.first_frame_after_reset = 1; + psEnc->state_Fxx[1].sCmn.first_frame_after_reset = 1; + } psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded = psEnc->state_Fxx[ 1 ].sCmn.nFramesEncoded = 0; /* Check values in encoder control structure */ if( ( ret = check_control_input( encControl ) != 0 ) ) { silk_assert( 0 ); + RESTORE_STACK; return ret; } @@ -165,7 +175,7 @@ opus_int silk_Encode( /* O Returns error co if( encControl->nChannelsInternal > psEnc->nChannelsInternal ) { /* Mono -> Stereo transition: init state of second channel and stereo state */ - ret += silk_init_encoder( &psEnc->state_Fxx[ 1 ] ); + ret += silk_init_encoder( &psEnc->state_Fxx[ 1 ], psEnc->state_Fxx[ 0 ].sCmn.arch ); silk_memset( psEnc->sStereo.pred_prev_Q13, 0, sizeof( psEnc->sStereo.pred_prev_Q13 ) ); silk_memset( psEnc->sStereo.sSide, 0, sizeof( psEnc->sStereo.sSide ) ); psEnc->sStereo.mid_side_amp_Q0[ 0 ] = 0; @@ -191,15 +201,14 @@ opus_int silk_Encode( /* O Returns error co if( prefillFlag ) { /* Only accept input length of 10 ms */ if( nBlocksOf10ms != 1 ) { - ret = SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES; silk_assert( 0 ); - return ret; + RESTORE_STACK; + return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES; } /* Reset Encoder */ for( n = 0; n < encControl->nChannelsInternal; n++ ) { - if( (ret = silk_init_encoder( &psEnc->state_Fxx[ n ] ) ) != 0 ) { - silk_assert( 0 ); - } + ret = silk_init_encoder( &psEnc->state_Fxx[ n ], psEnc->state_Fxx[ n ].sCmn.arch ); + silk_assert( !ret ); } tmp_payloadSize_ms = encControl->payloadSize_ms; encControl->payloadSize_ms = 10; @@ -212,15 +221,15 @@ opus_int silk_Encode( /* O Returns error co } else { /* Only accept input lengths that are a multiple of 10 ms */ if( nBlocksOf10ms * encControl->API_sampleRate != 100 * nSamplesIn || nSamplesIn < 0 ) { - ret = SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES; silk_assert( 0 ); - return ret; + RESTORE_STACK; + return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES; } /* Make sure no more than one packet can be produced */ if( 1000 * (opus_int32)nSamplesIn > encControl->payloadSize_ms * encControl->API_sampleRate ) { - ret = SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES; silk_assert( 0 ); - return ret; + RESTORE_STACK; + return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES; } } @@ -230,6 +239,7 @@ opus_int silk_Encode( /* O Returns error co opus_int force_fs_kHz = (n==1) ? psEnc->state_Fxx[0].sCmn.fs_kHz : 0; if( ( ret = silk_control_encoder( &psEnc->state_Fxx[ n ], encControl, TargetRate_bps, psEnc->allowBandwidthSwitch, n, force_fs_kHz ) ) != 0 ) { silk_assert( 0 ); + RESTORE_STACK; return ret; } if( psEnc->state_Fxx[n].sCmn.first_frame_after_reset || transition ) { @@ -242,9 +252,16 @@ opus_int silk_Encode( /* O Returns error co silk_assert( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 0 ].sCmn.fs_kHz == psEnc->state_Fxx[ 1 ].sCmn.fs_kHz ); /* Input buffering/resampling and encoding */ + nSamplesToBufferMax = + 10 * nBlocksOf10ms * psEnc->state_Fxx[ 0 ].sCmn.fs_kHz; + nSamplesFromInputMax = + silk_DIV32_16( nSamplesToBufferMax * + psEnc->state_Fxx[ 0 ].sCmn.API_fs_Hz, + psEnc->state_Fxx[ 0 ].sCmn.fs_kHz * 1000 ); + ALLOC( buf, nSamplesFromInputMax, opus_int16 ); while( 1 ) { nSamplesToBuffer = psEnc->state_Fxx[ 0 ].sCmn.frame_length - psEnc->state_Fxx[ 0 ].sCmn.inputBufIx; - nSamplesToBuffer = silk_min( nSamplesToBuffer, 10 * nBlocksOf10ms * psEnc->state_Fxx[ 0 ].sCmn.fs_kHz ); + nSamplesToBuffer = silk_min( nSamplesToBuffer, nSamplesToBufferMax ); nSamplesFromInput = silk_DIV32_16( nSamplesToBuffer * psEnc->state_Fxx[ 0 ].sCmn.API_fs_Hz, psEnc->state_Fxx[ 0 ].sCmn.fs_kHz * 1000 ); /* Resample and write to buffer */ if( encControl->nChannelsAPI == 2 && encControl->nChannelsInternal == 2 ) { @@ -533,6 +550,7 @@ opus_int silk_Encode( /* O Returns error co } } + RESTORE_STACK; return ret; } diff --git a/src/opus-1.0.2/silk/encode_indices.c b/src/opus-1.1/silk/encode_indices.c index 91e28aa9..666c8c0b 100644 --- a/src/opus-1.0.2/silk/encode_indices.c +++ b/src/opus-1.1/silk/encode_indices.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/encode_pulses.c b/src/opus-1.1/silk/encode_pulses.c index b01b5853..a4501438 100644 --- a/src/opus-1.0.2/silk/encode_pulses.c +++ b/src/opus-1.1/silk/encode_pulses.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,12 +30,13 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main.h" +#include "stack_alloc.h" /*********************************************/ /* Encode quantization indices of excitation */ /*********************************************/ -static inline opus_int combine_and_check( /* return ok */ +static OPUS_INLINE opus_int combine_and_check( /* return ok */ opus_int *pulses_comb, /* O */ const opus_int *pulses_in, /* I */ opus_int max_pulses, /* I max value for sum of pulses */ @@ -66,14 +67,15 @@ void silk_encode_pulses( { opus_int i, k, j, iter, bit, nLS, scale_down, RateLevelIndex = 0; opus_int32 abs_q, minSumBits_Q5, sumBits_Q5; - opus_int abs_pulses[ MAX_FRAME_LENGTH ]; - opus_int sum_pulses[ MAX_NB_SHELL_BLOCKS ]; - opus_int nRshifts[ MAX_NB_SHELL_BLOCKS ]; + VARDECL( opus_int, abs_pulses ); + VARDECL( opus_int, sum_pulses ); + VARDECL( opus_int, nRshifts ); opus_int pulses_comb[ 8 ]; opus_int *abs_pulses_ptr; const opus_int8 *pulses_ptr; const opus_uint8 *cdf_ptr; const opus_uint8 *nBits_ptr; + SAVE_STACK; silk_memset( pulses_comb, 0, 8 * sizeof( opus_int ) ); /* Fixing Valgrind reported problem*/ @@ -90,6 +92,8 @@ void silk_encode_pulses( } /* Take the absolute value of the pulses */ + ALLOC( abs_pulses, iter * SHELL_CODEC_FRAME_LENGTH, opus_int ); + silk_assert( !( SHELL_CODEC_FRAME_LENGTH & 3 ) ); for( i = 0; i < iter * SHELL_CODEC_FRAME_LENGTH; i+=4 ) { abs_pulses[i+0] = ( opus_int )silk_abs( pulses[ i + 0 ] ); abs_pulses[i+1] = ( opus_int )silk_abs( pulses[ i + 1 ] ); @@ -98,6 +102,8 @@ void silk_encode_pulses( } /* Calc sum pulses per shell code frame */ + ALLOC( sum_pulses, iter, opus_int ); + ALLOC( nRshifts, iter, opus_int ); abs_pulses_ptr = abs_pulses; for( i = 0; i < iter; i++ ) { nRshifts[ i ] = 0; @@ -196,4 +202,5 @@ void silk_encode_pulses( /* Encode signs */ /****************/ silk_encode_signs( psRangeEnc, pulses, frame_length, signalType, quantOffsetType, sum_pulses ); + RESTORE_STACK; } diff --git a/src/opus-1.0.2/silk/errors.h b/src/opus-1.1/silk/errors.h index 0591e009..45070800 100644 --- a/src/opus-1.0.2/silk/errors.h +++ b/src/opus-1.1/silk/errors.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/fixed/LTP_analysis_filter_FIX.c b/src/opus-1.1/silk/fixed/LTP_analysis_filter_FIX.c index a8fee555..a9419080 100644 --- a/src/opus-1.0.2/silk/fixed/LTP_analysis_filter_FIX.c +++ b/src/opus-1.1/silk/fixed/LTP_analysis_filter_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/fixed/LTP_scale_ctrl_FIX.c b/src/opus-1.1/silk/fixed/LTP_scale_ctrl_FIX.c index ac2fba17..3dcedef8 100644 --- a/src/opus-1.0.2/silk/fixed/LTP_scale_ctrl_FIX.c +++ b/src/opus-1.1/silk/fixed/LTP_scale_ctrl_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/fixed/apply_sine_window_FIX.c b/src/opus-1.1/silk/fixed/apply_sine_window_FIX.c index 897fdc30..4502b713 100644 --- a/src/opus-1.0.2/silk/fixed/apply_sine_window_FIX.c +++ b/src/opus-1.1/silk/fixed/apply_sine_window_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/fixed/autocorr_FIX.c b/src/opus-1.1/silk/fixed/autocorr_FIX.c index c2ebb6a9..de95c986 100644 --- a/src/opus-1.0.2/silk/fixed/autocorr_FIX.c +++ b/src/opus-1.1/silk/fixed/autocorr_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,6 +30,7 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "SigProc_FIX.h" +#include "celt_lpc.h" /* Compute autocorrelation */ void silk_autocorr( @@ -37,40 +38,11 @@ void silk_autocorr( opus_int *scale, /* O Scaling of the correlation vector */ const opus_int16 *inputData, /* I Input data to correlate */ const opus_int inputDataSize, /* I Length of input */ - const opus_int correlationCount /* I Number of correlation taps to compute */ + const opus_int correlationCount, /* I Number of correlation taps to compute */ + int arch /* I Run-time architecture */ ) { - opus_int i, lz, nRightShifts, corrCount; - opus_int64 corr64; - + opus_int corrCount; corrCount = silk_min_int( inputDataSize, correlationCount ); - - /* compute energy (zero-lag correlation) */ - corr64 = silk_inner_prod16_aligned_64( inputData, inputData, inputDataSize ); - - /* deal with all-zero input data */ - corr64 += 1; - - /* number of leading zeros */ - lz = silk_CLZ64( corr64 ); - - /* scaling: number of right shifts applied to correlations */ - nRightShifts = 35 - lz; - *scale = nRightShifts; - - if( nRightShifts <= 0 ) { - results[ 0 ] = silk_LSHIFT( (opus_int32)silk_CHECK_FIT32( corr64 ), -nRightShifts ); - - /* compute remaining correlations based on int32 inner product */ - for( i = 1; i < corrCount; i++ ) { - results[ i ] = silk_LSHIFT( silk_inner_prod_aligned( inputData, inputData + i, inputDataSize - i ), -nRightShifts ); - } - } else { - results[ 0 ] = (opus_int32)silk_CHECK_FIT32( silk_RSHIFT64( corr64, nRightShifts ) ); - - /* compute remaining correlations based on int64 inner product */ - for( i = 1; i < corrCount; i++ ) { - results[ i ] = (opus_int32)silk_CHECK_FIT32( silk_RSHIFT64( silk_inner_prod16_aligned_64( inputData, inputData + i, inputDataSize - i ), nRightShifts ) ); - } - } + *scale = _celt_autocorr(inputData, results, NULL, 0, corrCount-1, inputDataSize, arch); } diff --git a/src/opus-1.0.2/silk/fixed/burg_modified_FIX.c b/src/opus-1.1/silk/fixed/burg_modified_FIX.c index 26a66b1c..db348295 100644 --- a/src/opus-1.0.2/silk/fixed/burg_modified_FIX.c +++ b/src/opus-1.1/silk/fixed/burg_modified_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -32,6 +32,7 @@ POSSIBILITY OF SUCH DAMAGE. #include "SigProc_FIX.h" #include "define.h" #include "tuning_parameters.h" +#include "pitch.h" #define MAX_FRAME_SIZE 384 /* subfr_length * nb_subfr = ( 0.005 * 16000 + 16 ) * 4 = 384 */ @@ -49,7 +50,8 @@ void silk_burg_modified( const opus_int32 minInvGain_Q30, /* I Inverse of max prediction gain */ const opus_int subfr_length, /* I Input signal subframe length (incl. D preceding samples) */ const opus_int nb_subfr, /* I Number of subframes stacked in x */ - const opus_int D /* I Order */ + const opus_int D, /* I Order */ + int arch /* I Run-time architecture */ ) { opus_int k, n, s, lz, rshifts, rshifts_extra, reached_max_gain; @@ -60,6 +62,7 @@ void silk_burg_modified( opus_int32 Af_QA[ SILK_MAX_ORDER_LPC ]; opus_int32 CAf[ SILK_MAX_ORDER_LPC + 1 ]; opus_int32 CAb[ SILK_MAX_ORDER_LPC + 1 ]; + opus_int32 xcorr[ SILK_MAX_ORDER_LPC ]; silk_assert( subfr_length * nb_subfr <= MAX_FRAME_SIZE ); @@ -93,10 +96,17 @@ void silk_burg_modified( } } else { for( s = 0; s < nb_subfr; s++ ) { + int i; + opus_int32 d; x_ptr = x + s * subfr_length; + celt_pitch_xcorr(x_ptr, x_ptr + 1, xcorr, subfr_length - D, D, arch ); for( n = 1; n < D + 1; n++ ) { - C_first_row[ n - 1 ] += silk_LSHIFT32( - silk_inner_prod_aligned( x_ptr, x_ptr + n, subfr_length - n ), -rshifts ); + for ( i = n + subfr_length - D, d = 0; i < subfr_length; i++ ) + d = MAC16_16( d, x_ptr[ i ], x_ptr[ i - n ] ); + xcorr[ n - 1 ] += d; + } + for( n = 1; n < D + 1; n++ ) { + C_first_row[ n - 1 ] += silk_LSHIFT32( xcorr[ n - 1 ], -rshifts ); } } } @@ -263,7 +273,7 @@ void silk_burg_modified( tmp1 = silk_SMLAWW( tmp1, Atmp1, Atmp1 ); /* Q16 */ A_Q16[ k ] = -Atmp1; } - *res_nrg = silk_SMLAWW( nrg, silk_SMMUL( FIND_LPC_COND_FAC, C0 ), -tmp1 ); /* Q( -rshifts ) */ + *res_nrg = silk_SMLAWW( nrg, silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ), -tmp1 );/* Q( -rshifts ) */ *res_nrg_Q = -rshifts; - } + } } diff --git a/src/opus-1.0.2/silk/fixed/corrMatrix_FIX.c b/src/opus-1.1/silk/fixed/corrMatrix_FIX.c index 21502499..c6172705 100644 --- a/src/opus-1.0.2/silk/fixed/corrMatrix_FIX.c +++ b/src/opus-1.1/silk/fixed/corrMatrix_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/fixed/encode_frame_FIX.c b/src/opus-1.1/silk/fixed/encode_frame_FIX.c index a37a9f21..b490986b 100644 --- a/src/opus-1.0.2/silk/fixed/encode_frame_FIX.c +++ b/src/opus-1.1/silk/fixed/encode_frame_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,10 +30,11 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main_FIX.h" +#include "stack_alloc.h" #include "tuning_parameters.h" /* Low Bitrate Redundancy (LBRR) encoding. Reuse all parameters but encode with lower bitrate */ -static inline void silk_LBRR_encode_FIX( +static OPUS_INLINE void silk_LBRR_encode_FIX( silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ silk_encoder_control_FIX *psEncCtrl, /* I/O Pointer to Silk FIX encoder control struct */ const opus_int32 xfw_Q3[], /* I Input signal */ @@ -84,9 +85,7 @@ opus_int silk_encode_frame_FIX( { silk_encoder_control_FIX sEncCtrl; opus_int i, iter, maxIter, found_upper, found_lower, ret = 0; - opus_int16 *x_frame, *res_pitch_frame; - opus_int32 xfw_Q3[ MAX_FRAME_LENGTH ]; - opus_int16 res_pitch[ 2 * MAX_FRAME_LENGTH + LA_PITCH_MAX ]; + opus_int16 *x_frame; ec_enc sRangeEnc_copy, sRangeEnc_copy2; silk_nsq_state sNSQ_copy, sNSQ_copy2; opus_int32 seed_copy, nBits, nBits_lower, nBits_upper, gainMult_lower, gainMult_upper; @@ -95,7 +94,7 @@ opus_int silk_encode_frame_FIX( opus_int16 ec_prevLagIndex_copy; opus_int ec_prevSignalType_copy; opus_int8 LastGainIndex_copy2; - opus_uint8 ec_buf_copy[ 1275 ]; + SAVE_STACK; /* This is totally unnecessary but many compilers (including gcc) are too dumb to realise it */ LastGainIndex_copy2 = nBits_lower = nBits_upper = gainMult_lower = gainMult_upper = 0; @@ -105,9 +104,8 @@ opus_int silk_encode_frame_FIX( /**************************************************************/ /* Set up Input Pointers, and insert frame in input buffer */ /*************************************************************/ - /* pointers aligned with start of frame to encode */ - x_frame = psEnc->x_buf + psEnc->sCmn.ltp_mem_length; /* start of frame to encode */ - res_pitch_frame = res_pitch + psEnc->sCmn.ltp_mem_length; /* start of pitch LPC residual frame */ + /* start of frame to encode */ + x_frame = psEnc->x_buf + psEnc->sCmn.ltp_mem_length; /***************************************/ /* Ensure smooth bandwidth transitions */ @@ -120,15 +118,26 @@ opus_int silk_encode_frame_FIX( silk_memcpy( x_frame + LA_SHAPE_MS * psEnc->sCmn.fs_kHz, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.frame_length * sizeof( opus_int16 ) ); if( !psEnc->sCmn.prefillFlag ) { + VARDECL( opus_int32, xfw_Q3 ); + VARDECL( opus_int16, res_pitch ); + VARDECL( opus_uint8, ec_buf_copy ); + opus_int16 *res_pitch_frame; + + ALLOC( res_pitch, + psEnc->sCmn.la_pitch + psEnc->sCmn.frame_length + + psEnc->sCmn.ltp_mem_length, opus_int16 ); + /* start of pitch LPC residual frame */ + res_pitch_frame = res_pitch + psEnc->sCmn.ltp_mem_length; + /*****************************************/ /* Find pitch lags, initial LPC analysis */ /*****************************************/ - silk_find_pitch_lags_FIX( psEnc, &sEncCtrl, res_pitch, x_frame ); + silk_find_pitch_lags_FIX( psEnc, &sEncCtrl, res_pitch, x_frame, psEnc->sCmn.arch ); /************************/ /* Noise shape analysis */ /************************/ - silk_noise_shape_analysis_FIX( psEnc, &sEncCtrl, res_pitch_frame, x_frame ); + silk_noise_shape_analysis_FIX( psEnc, &sEncCtrl, res_pitch_frame, x_frame, psEnc->sCmn.arch ); /***************************************************/ /* Find linear prediction coefficients (LPC + LTP) */ @@ -143,6 +152,7 @@ opus_int silk_encode_frame_FIX( /*****************************************/ /* Prefiltering for noise shaper */ /*****************************************/ + ALLOC( xfw_Q3, psEnc->sCmn.frame_length, opus_int32 ); silk_prefilter_FIX( psEnc, &sEncCtrl, xfw_Q3, x_frame ); /****************************************/ @@ -164,6 +174,7 @@ opus_int silk_encode_frame_FIX( seed_copy = psEnc->sCmn.indices.Seed; ec_prevLagIndex_copy = psEnc->sCmn.ec_prevLagIndex; ec_prevSignalType_copy = psEnc->sCmn.ec_prevSignalType; + ALLOC( ec_buf_copy, 1275, opus_uint8 ); for( iter = 0; ; iter++ ) { if( gainsID == gainsID_lower ) { nBits = nBits_lower; @@ -291,17 +302,18 @@ opus_int silk_encode_frame_FIX( silk_memmove( psEnc->x_buf, &psEnc->x_buf[ psEnc->sCmn.frame_length ], ( psEnc->sCmn.ltp_mem_length + LA_SHAPE_MS * psEnc->sCmn.fs_kHz ) * sizeof( opus_int16 ) ); - /* Parameters needed for next frame */ - psEnc->sCmn.prevLag = sEncCtrl.pitchL[ psEnc->sCmn.nb_subfr - 1 ]; - psEnc->sCmn.prevSignalType = psEnc->sCmn.indices.signalType; - /* Exit without entropy coding */ if( psEnc->sCmn.prefillFlag ) { /* No payload */ *pnBytesOut = 0; + RESTORE_STACK; return ret; } + /* Parameters needed for next frame */ + psEnc->sCmn.prevLag = sEncCtrl.pitchL[ psEnc->sCmn.nb_subfr - 1 ]; + psEnc->sCmn.prevSignalType = psEnc->sCmn.indices.signalType; + /****************************************/ /* Finalize payload */ /****************************************/ @@ -309,11 +321,12 @@ opus_int silk_encode_frame_FIX( /* Payload size */ *pnBytesOut = silk_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 ); + RESTORE_STACK; return ret; } /* Low-Bitrate Redundancy (LBRR) encoding. Reuse all parameters but encode excitation at lower bitrate */ -static inline void silk_LBRR_encode_FIX( +static OPUS_INLINE void silk_LBRR_encode_FIX( silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ silk_encoder_control_FIX *psEncCtrl, /* I/O Pointer to Silk FIX encoder control struct */ const opus_int32 xfw_Q3[], /* I Input signal */ diff --git a/src/opus-1.0.2/silk/fixed/find_LPC_FIX.c b/src/opus-1.1/silk/fixed/find_LPC_FIX.c index 0ed7e846..783d32e2 100644 --- a/src/opus-1.0.2/silk/fixed/find_LPC_FIX.c +++ b/src/opus-1.1/silk/fixed/find_LPC_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,6 +30,7 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main_FIX.h" +#include "stack_alloc.h" #include "tuning_parameters.h" /* Finds LPC vector from correlations, and converts to NLSF */ @@ -51,7 +52,7 @@ void silk_find_LPC_FIX( opus_int res_nrg_interp_Q, res_nrg_Q, res_tmp_nrg_Q; opus_int16 a_tmp_Q12[ MAX_LPC_ORDER ]; opus_int16 NLSF0_Q15[ MAX_LPC_ORDER ]; - opus_int16 LPC_res[ MAX_FRAME_LENGTH + MAX_NB_SUBFR * MAX_LPC_ORDER ]; + SAVE_STACK; subfr_length = psEncC->subfr_length + psEncC->predictLPCOrder; @@ -59,11 +60,13 @@ void silk_find_LPC_FIX( psEncC->indices.NLSFInterpCoef_Q2 = 4; /* Burg AR analysis for the full frame */ - silk_burg_modified( &res_nrg, &res_nrg_Q, a_Q16, x, minInvGain_Q30, subfr_length, psEncC->nb_subfr, psEncC->predictLPCOrder ); + silk_burg_modified( &res_nrg, &res_nrg_Q, a_Q16, x, minInvGain_Q30, subfr_length, psEncC->nb_subfr, psEncC->predictLPCOrder, psEncC->arch ); if( psEncC->useInterpolatedNLSFs && !psEncC->first_frame_after_reset && psEncC->nb_subfr == MAX_NB_SUBFR ) { + VARDECL( opus_int16, LPC_res ); + /* Optimal solution for last 10 ms */ - silk_burg_modified( &res_tmp_nrg, &res_tmp_nrg_Q, a_tmp_Q16, x + 2 * subfr_length, minInvGain_Q30, subfr_length, 2, psEncC->predictLPCOrder ); + silk_burg_modified( &res_tmp_nrg, &res_tmp_nrg_Q, a_tmp_Q16, x + 2 * subfr_length, minInvGain_Q30, subfr_length, 2, psEncC->predictLPCOrder, psEncC->arch ); /* subtract residual energy here, as that's easier than adding it to the */ /* residual energy of the first 10 ms in each iteration of the search below */ @@ -81,6 +84,8 @@ void silk_find_LPC_FIX( /* Convert to NLSFs */ silk_A2NLSF( NLSF_Q15, a_tmp_Q16, psEncC->predictLPCOrder ); + ALLOC( LPC_res, 2 * subfr_length, opus_int16 ); + /* Search over interpolation indices to find the one with lowest residual energy */ for( k = 3; k >= 0; k-- ) { /* Interpolate NLSFs for first half */ @@ -142,4 +147,5 @@ void silk_find_LPC_FIX( } silk_assert( psEncC->indices.NLSFInterpCoef_Q2 == 4 || ( psEncC->useInterpolatedNLSFs && !psEncC->first_frame_after_reset && psEncC->nb_subfr == MAX_NB_SUBFR ) ); + RESTORE_STACK; } diff --git a/src/opus-1.0.2/silk/fixed/find_LTP_FIX.c b/src/opus-1.1/silk/fixed/find_LTP_FIX.c index bd210874..8c4d7037 100644 --- a/src/opus-1.0.2/silk/fixed/find_LTP_FIX.c +++ b/src/opus-1.1/silk/fixed/find_LTP_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/fixed/find_pitch_lags_FIX.c b/src/opus-1.1/silk/fixed/find_pitch_lags_FIX.c index 39c30487..620f8dcd 100644 --- a/src/opus-1.0.2/silk/fixed/find_pitch_lags_FIX.c +++ b/src/opus-1.1/silk/fixed/find_pitch_lags_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,6 +30,7 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main_FIX.h" +#include "stack_alloc.h" #include "tuning_parameters.h" /* Find pitch lags */ @@ -37,17 +38,20 @@ void silk_find_pitch_lags_FIX( silk_encoder_state_FIX *psEnc, /* I/O encoder state */ silk_encoder_control_FIX *psEncCtrl, /* I/O encoder control */ opus_int16 res[], /* O residual */ - const opus_int16 x[] /* I Speech signal */ + const opus_int16 x[], /* I Speech signal */ + int arch /* I Run-time architecture */ ) { opus_int buf_len, i, scale; - opus_int32 thrhld_Q15, res_nrg; + opus_int32 thrhld_Q13, res_nrg; const opus_int16 *x_buf, *x_buf_ptr; - opus_int16 Wsig[ FIND_PITCH_LPC_WIN_MAX ], *Wsig_ptr; + VARDECL( opus_int16, Wsig ); + opus_int16 *Wsig_ptr; opus_int32 auto_corr[ MAX_FIND_PITCH_LPC_ORDER + 1 ]; opus_int16 rc_Q15[ MAX_FIND_PITCH_LPC_ORDER ]; opus_int32 A_Q24[ MAX_FIND_PITCH_LPC_ORDER ]; opus_int16 A_Q12[ MAX_FIND_PITCH_LPC_ORDER ]; + SAVE_STACK; /******************************************/ /* Set up buffer lengths etc based on Fs */ @@ -65,6 +69,8 @@ void silk_find_pitch_lags_FIX( /* Calculate windowed signal */ + ALLOC( Wsig, psEnc->sCmn.pitch_LPC_win_length, opus_int16 ); + /* First LA_LTP samples */ x_buf_ptr = x_buf + buf_len - psEnc->sCmn.pitch_LPC_win_length; Wsig_ptr = Wsig; @@ -81,7 +87,7 @@ void silk_find_pitch_lags_FIX( silk_apply_sine_window( Wsig_ptr, x_buf_ptr, 2, psEnc->sCmn.la_pitch ); /* Calculate autocorrelation sequence */ - silk_autocorr( auto_corr, &scale, Wsig, psEnc->sCmn.pitch_LPC_win_length, psEnc->sCmn.pitchEstimationLPCOrder + 1 ); + silk_autocorr( auto_corr, &scale, Wsig, psEnc->sCmn.pitch_LPC_win_length, psEnc->sCmn.pitchEstimationLPCOrder + 1, arch ); /* Add white noise, as fraction of energy */ auto_corr[ 0 ] = silk_SMLAWB( auto_corr[ 0 ], auto_corr[ 0 ], SILK_FIX_CONST( FIND_PITCH_WHITE_NOISE_FRACTION, 16 ) ) + 1; @@ -110,19 +116,20 @@ void silk_find_pitch_lags_FIX( if( psEnc->sCmn.indices.signalType != TYPE_NO_VOICE_ACTIVITY && psEnc->sCmn.first_frame_after_reset == 0 ) { /* Threshold for pitch estimator */ - thrhld_Q15 = SILK_FIX_CONST( 0.6, 15 ); - thrhld_Q15 = silk_SMLABB( thrhld_Q15, SILK_FIX_CONST( -0.004, 15 ), psEnc->sCmn.pitchEstimationLPCOrder ); - thrhld_Q15 = silk_SMLABB( thrhld_Q15, SILK_FIX_CONST( -0.1, 7 ), psEnc->sCmn.speech_activity_Q8 ); - thrhld_Q15 = silk_SMLABB( thrhld_Q15, SILK_FIX_CONST( -0.15, 15 ), silk_RSHIFT( psEnc->sCmn.prevSignalType, 1 ) ); - thrhld_Q15 = silk_SMLAWB( thrhld_Q15, SILK_FIX_CONST( -0.1, 16 ), psEnc->sCmn.input_tilt_Q15 ); - thrhld_Q15 = silk_SAT16( thrhld_Q15 ); + thrhld_Q13 = SILK_FIX_CONST( 0.6, 13 ); + thrhld_Q13 = silk_SMLABB( thrhld_Q13, SILK_FIX_CONST( -0.004, 13 ), psEnc->sCmn.pitchEstimationLPCOrder ); + thrhld_Q13 = silk_SMLAWB( thrhld_Q13, SILK_FIX_CONST( -0.1, 21 ), psEnc->sCmn.speech_activity_Q8 ); + thrhld_Q13 = silk_SMLABB( thrhld_Q13, SILK_FIX_CONST( -0.15, 13 ), silk_RSHIFT( psEnc->sCmn.prevSignalType, 1 ) ); + thrhld_Q13 = silk_SMLAWB( thrhld_Q13, SILK_FIX_CONST( -0.1, 14 ), psEnc->sCmn.input_tilt_Q15 ); + thrhld_Q13 = silk_SAT16( thrhld_Q13 ); /*****************************************/ /* Call pitch estimator */ /*****************************************/ if( silk_pitch_analysis_core( res, psEncCtrl->pitchL, &psEnc->sCmn.indices.lagIndex, &psEnc->sCmn.indices.contourIndex, &psEnc->LTPCorr_Q15, psEnc->sCmn.prevLag, psEnc->sCmn.pitchEstimationThreshold_Q16, - (opus_int16)thrhld_Q15, psEnc->sCmn.fs_kHz, psEnc->sCmn.pitchEstimationComplexity, psEnc->sCmn.nb_subfr ) == 0 ) + (opus_int)thrhld_Q13, psEnc->sCmn.fs_kHz, psEnc->sCmn.pitchEstimationComplexity, psEnc->sCmn.nb_subfr, + psEnc->sCmn.arch) == 0 ) { psEnc->sCmn.indices.signalType = TYPE_VOICED; } else { @@ -134,4 +141,5 @@ void silk_find_pitch_lags_FIX( psEnc->sCmn.indices.contourIndex = 0; psEnc->LTPCorr_Q15 = 0; } + RESTORE_STACK; } diff --git a/src/opus-1.0.2/silk/fixed/find_pred_coefs_FIX.c b/src/opus-1.1/silk/fixed/find_pred_coefs_FIX.c index 997989b5..5c22f828 100644 --- a/src/opus-1.0.2/silk/fixed/find_pred_coefs_FIX.c +++ b/src/opus-1.1/silk/fixed/find_pred_coefs_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,6 +30,7 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main_FIX.h" +#include "stack_alloc.h" void silk_find_pred_coefs_FIX( silk_encoder_state_FIX *psEnc, /* I/O encoder state */ @@ -40,13 +41,14 @@ void silk_find_pred_coefs_FIX( ) { opus_int i; - opus_int32 WLTP[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ]; opus_int32 invGains_Q16[ MAX_NB_SUBFR ], local_gains[ MAX_NB_SUBFR ], Wght_Q15[ MAX_NB_SUBFR ]; opus_int16 NLSF_Q15[ MAX_LPC_ORDER ]; const opus_int16 *x_ptr; - opus_int16 *x_pre_ptr, LPC_in_pre[ MAX_NB_SUBFR * MAX_LPC_ORDER + MAX_FRAME_LENGTH ]; + opus_int16 *x_pre_ptr; + VARDECL( opus_int16, LPC_in_pre ); opus_int32 tmp, min_gain_Q16, minInvGain_Q30; opus_int LTP_corrs_rshift[ MAX_NB_SUBFR ]; + SAVE_STACK; /* weighting for weighted least squares */ min_gain_Q16 = silk_int32_MAX >> 6; @@ -71,12 +73,19 @@ void silk_find_pred_coefs_FIX( local_gains[ i ] = silk_DIV32( ( (opus_int32)1 << 16 ), invGains_Q16[ i ] ); } + ALLOC( LPC_in_pre, + psEnc->sCmn.nb_subfr * psEnc->sCmn.predictLPCOrder + + psEnc->sCmn.frame_length, opus_int16 ); if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + VARDECL( opus_int32, WLTP ); + /**********/ /* VOICED */ /**********/ silk_assert( psEnc->sCmn.ltp_mem_length - psEnc->sCmn.predictLPCOrder >= psEncCtrl->pitchL[ 0 ] + LTP_ORDER / 2 ); + ALLOC( WLTP, psEnc->sCmn.nb_subfr * LTP_ORDER * LTP_ORDER, opus_int32 ); + /* LTP analysis */ silk_find_LTP_FIX( psEncCtrl->LTPCoef_Q14, WLTP, &psEncCtrl->LTPredCodGain_Q7, res_pitch, psEncCtrl->pitchL, Wght_Q15, psEnc->sCmn.subfr_length, @@ -84,7 +93,7 @@ void silk_find_pred_coefs_FIX( /* Quantize LTP gain parameters */ silk_quant_LTP_gains( psEncCtrl->LTPCoef_Q14, psEnc->sCmn.indices.LTPIndex, &psEnc->sCmn.indices.PERIndex, - WLTP, psEnc->sCmn.mu_LTP_Q9, psEnc->sCmn.LTPQuantLowComplexity, psEnc->sCmn.nb_subfr); + &psEnc->sCmn.sum_log_gain_Q7, WLTP, psEnc->sCmn.mu_LTP_Q9, psEnc->sCmn.LTPQuantLowComplexity, psEnc->sCmn.nb_subfr); /* Control LTP scaling */ silk_LTP_scale_ctrl_FIX( psEnc, psEncCtrl, condCoding ); @@ -109,6 +118,7 @@ void silk_find_pred_coefs_FIX( silk_memset( psEncCtrl->LTPCoef_Q14, 0, psEnc->sCmn.nb_subfr * LTP_ORDER * sizeof( opus_int16 ) ); psEncCtrl->LTPredCodGain_Q7 = 0; + psEnc->sCmn.sum_log_gain_Q7 = 0; } /* Limit on total predictive coding gain */ @@ -133,4 +143,5 @@ void silk_find_pred_coefs_FIX( /* Copy to prediction struct for use in next frame for interpolation */ silk_memcpy( psEnc->sCmn.prev_NLSFq_Q15, NLSF_Q15, sizeof( psEnc->sCmn.prev_NLSFq_Q15 ) ); + RESTORE_STACK; } diff --git a/src/opus-1.0.2/silk/fixed/k2a_FIX.c b/src/opus-1.1/silk/fixed/k2a_FIX.c index cadc9274..5fee599b 100644 --- a/src/opus-1.0.2/silk/fixed/k2a_FIX.c +++ b/src/opus-1.1/silk/fixed/k2a_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/fixed/k2a_Q16_FIX.c b/src/opus-1.1/silk/fixed/k2a_Q16_FIX.c index f96f3064..3b039875 100644 --- a/src/opus-1.0.2/silk/fixed/k2a_Q16_FIX.c +++ b/src/opus-1.1/silk/fixed/k2a_Q16_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/fixed/main_FIX.h b/src/opus-1.1/silk/fixed/main_FIX.h index 369b31ee..a56ca07a 100644 --- a/src/opus-1.0.2/silk/fixed/main_FIX.h +++ b/src/opus-1.1/silk/fixed/main_FIX.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -73,7 +73,8 @@ opus_int silk_encode_frame_FIX( /* Initializes the Silk encoder state */ opus_int silk_init_encoder( - silk_encoder_state_Fxx *psEnc /* I/O Pointer to Silk FIX encoder state */ + silk_encoder_state_Fxx *psEnc, /* I/O Pointer to Silk FIX encoder state */ + int arch /* I Run-time architecture */ ); /* Control the Silk encoder */ @@ -104,7 +105,8 @@ void silk_noise_shape_analysis_FIX( silk_encoder_state_FIX *psEnc, /* I/O Encoder state FIX */ silk_encoder_control_FIX *psEncCtrl, /* I/O Encoder control FIX */ const opus_int16 *pitch_res, /* I LPC residual from pitch analysis */ - const opus_int16 *x /* I Input signal [ frame_length + la_shape ] */ + const opus_int16 *x, /* I Input signal [ frame_length + la_shape ] */ + int arch /* I Run-time architecture */ ); /* Autocorrelations for a warped frequency axis */ @@ -132,7 +134,8 @@ void silk_find_pitch_lags_FIX( silk_encoder_state_FIX *psEnc, /* I/O encoder state */ silk_encoder_control_FIX *psEncCtrl, /* I/O encoder control */ opus_int16 res[], /* O residual */ - const opus_int16 x[] /* I Speech signal */ + const opus_int16 x[], /* I Speech signal */ + int arch /* I Run-time architecture */ ); /* Find LPC and LTP coefficients */ diff --git a/src/opus-1.0.2/silk/fixed/noise_shape_analysis_FIX.c b/src/opus-1.1/silk/fixed/noise_shape_analysis_FIX.c index d230e48d..e24d2e9d 100644 --- a/src/opus-1.0.2/silk/fixed/noise_shape_analysis_FIX.c +++ b/src/opus-1.1/silk/fixed/noise_shape_analysis_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,13 +30,14 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main_FIX.h" +#include "stack_alloc.h" #include "tuning_parameters.h" /* Compute gain to make warped filter coefficients have a zero mean log frequency response on a */ /* non-warped frequency scale. (So that it can be implemented with a minimum-phase monic filter.) */ /* Note: A monic filter is one with the first coefficient equal to 1.0. In Silk we omit the first */ /* coefficient in an array of coefficients, for monic filters. */ -static inline opus_int32 warped_gain( /* gain in Q16*/ +static OPUS_INLINE opus_int32 warped_gain( /* gain in Q16*/ const opus_int32 *coefs_Q24, opus_int lambda_Q16, opus_int order @@ -55,7 +56,7 @@ static inline opus_int32 warped_gain( /* gain in Q16*/ /* Convert warped filter coefficients to monic pseudo-warped coefficients and limit maximum */ /* amplitude of monic warped coefficients by using bandwidth expansion on the true coefficients */ -static inline void limit_warped_coefs( +static OPUS_INLINE void limit_warped_coefs( opus_int32 *coefs_syn_Q24, opus_int32 *coefs_ana_Q24, opus_int lambda_Q16, @@ -144,7 +145,8 @@ void silk_noise_shape_analysis_FIX( silk_encoder_state_FIX *psEnc, /* I/O Encoder state FIX */ silk_encoder_control_FIX *psEncCtrl, /* I/O Encoder control FIX */ const opus_int16 *pitch_res, /* I LPC residual from pitch analysis */ - const opus_int16 *x /* I Input signal [ frame_length + la_shape ] */ + const opus_int16 *x, /* I Input signal [ frame_length + la_shape ] */ + int arch /* I Run-time architecture */ ) { silk_shape_state_FIX *psShapeSt = &psEnc->sShape; @@ -156,8 +158,9 @@ void silk_noise_shape_analysis_FIX( opus_int32 refl_coef_Q16[ MAX_SHAPE_LPC_ORDER ]; opus_int32 AR1_Q24[ MAX_SHAPE_LPC_ORDER ]; opus_int32 AR2_Q24[ MAX_SHAPE_LPC_ORDER ]; - opus_int16 x_windowed[ SHAPE_LPC_WIN_MAX ]; + VARDECL( opus_int16, x_windowed ); const opus_int16 *x_ptr, *pitch_res_ptr; + SAVE_STACK; /* Point to start of first LPC analysis block */ x_ptr = x - psEnc->sCmn.la_shape; @@ -258,6 +261,7 @@ void silk_noise_shape_analysis_FIX( /********************************************/ /* Compute noise shaping AR coefs and gains */ /********************************************/ + ALLOC( x_windowed, psEnc->sCmn.shapeWinLength, opus_int16 ); for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { /* Apply window: sine slope followed by flat part followed by cosine slope */ opus_int shift, slope_part, flat_part; @@ -278,7 +282,7 @@ void silk_noise_shape_analysis_FIX( silk_warped_autocorrelation_FIX( auto_corr, &scale, x_windowed, warping_Q16, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder ); } else { /* Calculate regular auto correlation */ - silk_autocorr( auto_corr, &scale, x_windowed, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder + 1 ); + silk_autocorr( auto_corr, &scale, x_windowed, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder + 1, arch ); } /* Add white noise, as a fraction of energy */ @@ -437,4 +441,5 @@ void silk_noise_shape_analysis_FIX( psEncCtrl->HarmShapeGain_Q14[ k ] = ( opus_int )silk_RSHIFT_ROUND( psShapeSt->HarmShapeGain_smth_Q16, 2 ); psEncCtrl->Tilt_Q14[ k ] = ( opus_int )silk_RSHIFT_ROUND( psShapeSt->Tilt_smth_Q16, 2 ); } + RESTORE_STACK; } diff --git a/src/opus-1.0.2/silk/fixed/pitch_analysis_core_FIX.c b/src/opus-1.1/silk/fixed/pitch_analysis_core_FIX.c index d43f444d..1641a0fb 100644 --- a/src/opus-1.0.2/silk/fixed/pitch_analysis_core_FIX.c +++ b/src/opus-1.1/silk/fixed/pitch_analysis_core_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -34,24 +34,40 @@ POSSIBILITY OF SUCH DAMAGE. ********************************************************** */ #include "SigProc_FIX.h" #include "pitch_est_defines.h" +#include "stack_alloc.h" #include "debug.h" +#include "pitch.h" #define SCRATCH_SIZE 22 +#define SF_LENGTH_4KHZ ( PE_SUBFR_LENGTH_MS * 4 ) +#define SF_LENGTH_8KHZ ( PE_SUBFR_LENGTH_MS * 8 ) +#define MIN_LAG_4KHZ ( PE_MIN_LAG_MS * 4 ) +#define MIN_LAG_8KHZ ( PE_MIN_LAG_MS * 8 ) +#define MAX_LAG_4KHZ ( PE_MAX_LAG_MS * 4 ) +#define MAX_LAG_8KHZ ( PE_MAX_LAG_MS * 8 - 1 ) +#define CSTRIDE_4KHZ ( MAX_LAG_4KHZ + 1 - MIN_LAG_4KHZ ) +#define CSTRIDE_8KHZ ( MAX_LAG_8KHZ + 3 - ( MIN_LAG_8KHZ - 2 ) ) +#define D_COMP_MIN ( MIN_LAG_8KHZ - 3 ) +#define D_COMP_MAX ( MAX_LAG_8KHZ + 4 ) +#define D_COMP_STRIDE ( D_COMP_MAX - D_COMP_MIN ) + +typedef opus_int32 silk_pe_stage3_vals[ PE_NB_STAGE3_LAGS ]; /************************************************************/ /* Internally used functions */ /************************************************************/ -void silk_P_Ana_calc_corr_st3( - opus_int32 cross_corr_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ],/* (O) 3 DIM correlation array */ +static void silk_P_Ana_calc_corr_st3( + silk_pe_stage3_vals cross_corr_st3[], /* O 3 DIM correlation array */ const opus_int16 frame[], /* I vector to correlate */ opus_int start_lag, /* I lag offset to search around */ opus_int sf_length, /* I length of a 5 ms subframe */ opus_int nb_subfr, /* I number of subframes */ - opus_int complexity /* I Complexity setting */ + opus_int complexity, /* I Complexity setting */ + int arch /* I Run-time architecture */ ); -void silk_P_Ana_calc_energy_st3( - opus_int32 energies_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ],/* (O) 3 DIM energy array */ +static void silk_P_Ana_calc_energy_st3( + silk_pe_stage3_vals energies_st3[], /* O 3 DIM energy array */ const opus_int16 frame[], /* I vector to calc energy in */ opus_int start_lag, /* I lag offset to search around */ opus_int sf_length, /* I length of one 5 ms subframe */ @@ -59,12 +75,6 @@ void silk_P_Ana_calc_energy_st3( opus_int complexity /* I Complexity setting */ ); -opus_int32 silk_P_Ana_find_scaling( - const opus_int16 *frame, - const opus_int frame_length, - const opus_int sum_sqr_len -); - /*************************************************************/ /* FIXED POINT CORE PITCH ANALYSIS FUNCTION */ /*************************************************************/ @@ -76,36 +86,38 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 opus_int *LTPCorr_Q15, /* I/O Normalized correlation; input: value from previous frame */ opus_int prevLag, /* I Last lag of previous frame; set to zero is unvoiced */ const opus_int32 search_thres1_Q16, /* I First stage threshold for lag candidates 0 - 1 */ - const opus_int search_thres2_Q15, /* I Final threshold for lag candidates 0 - 1 */ + const opus_int search_thres2_Q13, /* I Final threshold for lag candidates 0 - 1 */ const opus_int Fs_kHz, /* I Sample frequency (kHz) */ const opus_int complexity, /* I Complexity setting, 0-2, where 2 is highest */ - const opus_int nb_subfr /* I number of 5 ms subframes */ + const opus_int nb_subfr, /* I number of 5 ms subframes */ + int arch /* I Run-time architecture */ ) { - opus_int16 frame_8kHz[ PE_MAX_FRAME_LENGTH_ST_2 ]; - opus_int16 frame_4kHz[ PE_MAX_FRAME_LENGTH_ST_1 ]; + VARDECL( opus_int16, frame_8kHz ); + VARDECL( opus_int16, frame_4kHz ); opus_int32 filt_state[ 6 ]; - opus_int32 scratch_mem[ 3 * PE_MAX_FRAME_LENGTH ]; - opus_int16 *input_frame_ptr; + const opus_int16 *input_frame_ptr; opus_int i, k, d, j; - opus_int16 C[ PE_MAX_NB_SUBFR ][ ( PE_MAX_LAG >> 1 ) + 5 ]; + VARDECL( opus_int16, C ); + VARDECL( opus_int32, xcorr32 ); const opus_int16 *target_ptr, *basis_ptr; opus_int32 cross_corr, normalizer, energy, shift, energy_basis, energy_target; opus_int d_srch[ PE_D_SRCH_LENGTH ], Cmax, length_d_srch, length_d_comp; - opus_int16 d_comp[ ( PE_MAX_LAG >> 1 ) + 5 ]; - opus_int32 sum, threshold, temp32, lag_counter; + VARDECL( opus_int16, d_comp ); + opus_int32 sum, threshold, lag_counter; opus_int CBimax, CBimax_new, CBimax_old, lag, start_lag, end_lag, lag_new; opus_int32 CC[ PE_NB_CBKS_STAGE2_EXT ], CCmax, CCmax_b, CCmax_new_b, CCmax_new; - opus_int32 energies_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ]; - opus_int32 crosscorr_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ]; - opus_int frame_length, frame_length_8kHz, frame_length_4kHz, max_sum_sq_length; - opus_int sf_length, sf_length_8kHz, sf_length_4kHz; - opus_int min_lag, min_lag_8kHz, min_lag_4kHz; - opus_int max_lag, max_lag_8kHz, max_lag_4kHz; - opus_int32 contour_bias_Q20, diff, lz, lshift; + VARDECL( silk_pe_stage3_vals, energies_st3 ); + VARDECL( silk_pe_stage3_vals, cross_corr_st3 ); + opus_int frame_length, frame_length_8kHz, frame_length_4kHz; + opus_int sf_length; + opus_int min_lag; + opus_int max_lag; + opus_int32 contour_bias_Q15, diff; opus_int nb_cbk_search, cbk_size; - opus_int32 delta_lag_log2_sqr_Q7, lag_log2_Q7, prevLag_log2_Q7, prev_lag_bias_Q15, corr_thres_Q15; + opus_int32 delta_lag_log2_sqr_Q7, lag_log2_Q7, prevLag_log2_Q7, prev_lag_bias_Q13; const opus_int8 *Lag_CB_ptr; + SAVE_STACK; /* Check for valid sampling frequency */ silk_assert( Fs_kHz == 8 || Fs_kHz == 12 || Fs_kHz == 16 ); @@ -114,25 +126,18 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 silk_assert( complexity <= SILK_PE_MAX_COMPLEX ); silk_assert( search_thres1_Q16 >= 0 && search_thres1_Q16 <= (1<<16) ); - silk_assert( search_thres2_Q15 >= 0 && search_thres2_Q15 <= (1<<15) ); + silk_assert( search_thres2_Q13 >= 0 && search_thres2_Q13 <= (1<<13) ); /* Set up frame lengths max / min lag for the sampling frequency */ frame_length = ( PE_LTP_MEM_LENGTH_MS + nb_subfr * PE_SUBFR_LENGTH_MS ) * Fs_kHz; frame_length_4kHz = ( PE_LTP_MEM_LENGTH_MS + nb_subfr * PE_SUBFR_LENGTH_MS ) * 4; frame_length_8kHz = ( PE_LTP_MEM_LENGTH_MS + nb_subfr * PE_SUBFR_LENGTH_MS ) * 8; sf_length = PE_SUBFR_LENGTH_MS * Fs_kHz; - sf_length_4kHz = PE_SUBFR_LENGTH_MS * 4; - sf_length_8kHz = PE_SUBFR_LENGTH_MS * 8; min_lag = PE_MIN_LAG_MS * Fs_kHz; - min_lag_4kHz = PE_MIN_LAG_MS * 4; - min_lag_8kHz = PE_MIN_LAG_MS * 8; max_lag = PE_MAX_LAG_MS * Fs_kHz - 1; - max_lag_4kHz = PE_MAX_LAG_MS * 4; - max_lag_8kHz = PE_MAX_LAG_MS * 8 - 1; - - silk_memset( C, 0, sizeof( opus_int16 ) * nb_subfr * ( ( PE_MAX_LAG >> 1 ) + 5) ); /* Resample from input sampled at Fs_kHz to 8 kHz */ + ALLOC( frame_8kHz, frame_length_8kHz, opus_int16 ); if( Fs_kHz == 16 ) { silk_memset( filt_state, 0, 2 * sizeof( opus_int32 ) ); silk_resampler_down2( filt_state, frame_8kHz, frame, frame_length ); @@ -146,6 +151,7 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 /* Decimate again to 4 kHz */ silk_memset( filt_state, 0, 2 * sizeof( opus_int32 ) );/* Set state to zero */ + ALLOC( frame_4kHz, frame_length_4kHz, opus_int16 ); silk_resampler_down2( filt_state, frame_4kHz, frame_8kHz, frame_length_8kHz ); /* Low-pass filter */ @@ -159,9 +165,9 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 *******************************************************************************/ /* Inner product is calculated with different lengths, so scale for the worst case */ - max_sum_sq_length = silk_max_32( sf_length_8kHz, silk_LSHIFT( sf_length_4kHz, 2 ) ); - shift = silk_P_Ana_find_scaling( frame_4kHz, frame_length_4kHz, max_sum_sq_length ); + silk_sum_sqr_shift( &energy, &shift, frame_4kHz, frame_length_4kHz ); if( shift > 0 ) { + shift = silk_RSHIFT( shift, 1 ); for( i = 0; i < frame_length_4kHz; i++ ) { frame_4kHz[ i ] = silk_RSHIFT( frame_4kHz[ i ], shift ); } @@ -170,94 +176,93 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 /****************************************************************************** * FIRST STAGE, operating in 4 khz ******************************************************************************/ - target_ptr = &frame_4kHz[ silk_LSHIFT( sf_length_4kHz, 2 ) ]; + ALLOC( C, nb_subfr * CSTRIDE_8KHZ, opus_int16 ); + ALLOC( xcorr32, MAX_LAG_4KHZ-MIN_LAG_4KHZ+1, opus_int32 ); + silk_memset( C, 0, (nb_subfr >> 1) * CSTRIDE_4KHZ * sizeof( opus_int16 ) ); + target_ptr = &frame_4kHz[ silk_LSHIFT( SF_LENGTH_4KHZ, 2 ) ]; for( k = 0; k < nb_subfr >> 1; k++ ) { /* Check that we are within range of the array */ silk_assert( target_ptr >= frame_4kHz ); - silk_assert( target_ptr + sf_length_8kHz <= frame_4kHz + frame_length_4kHz ); + silk_assert( target_ptr + SF_LENGTH_8KHZ <= frame_4kHz + frame_length_4kHz ); - basis_ptr = target_ptr - min_lag_4kHz; + basis_ptr = target_ptr - MIN_LAG_4KHZ; /* Check that we are within range of the array */ silk_assert( basis_ptr >= frame_4kHz ); - silk_assert( basis_ptr + sf_length_8kHz <= frame_4kHz + frame_length_4kHz ); + silk_assert( basis_ptr + SF_LENGTH_8KHZ <= frame_4kHz + frame_length_4kHz ); + + celt_pitch_xcorr( target_ptr, target_ptr - MAX_LAG_4KHZ, xcorr32, SF_LENGTH_8KHZ, MAX_LAG_4KHZ - MIN_LAG_4KHZ + 1, arch ); /* Calculate first vector products before loop */ - cross_corr = silk_inner_prod_aligned( target_ptr, basis_ptr, sf_length_8kHz ); - normalizer = silk_inner_prod_aligned( basis_ptr, basis_ptr, sf_length_8kHz ); - normalizer = silk_ADD_SAT32( normalizer, silk_SMULBB( sf_length_8kHz, 4000 ) ); + cross_corr = xcorr32[ MAX_LAG_4KHZ - MIN_LAG_4KHZ ]; + normalizer = silk_inner_prod_aligned( target_ptr, target_ptr, SF_LENGTH_8KHZ ); + normalizer = silk_ADD32( normalizer, silk_inner_prod_aligned( basis_ptr, basis_ptr, SF_LENGTH_8KHZ ) ); + normalizer = silk_ADD32( normalizer, silk_SMULBB( SF_LENGTH_8KHZ, 4000 ) ); - temp32 = silk_DIV32( cross_corr, silk_SQRT_APPROX( normalizer ) + 1 ); - C[ k ][ min_lag_4kHz ] = (opus_int16)silk_SAT16( temp32 ); /* Q0 */ + matrix_ptr( C, k, 0, CSTRIDE_4KHZ ) = + (opus_int16)silk_DIV32_varQ( cross_corr, normalizer, 13 + 1 ); /* Q13 */ /* From now on normalizer is computed recursively */ - for( d = min_lag_4kHz + 1; d <= max_lag_4kHz; d++ ) { + for( d = MIN_LAG_4KHZ + 1; d <= MAX_LAG_4KHZ; d++ ) { basis_ptr--; /* Check that we are within range of the array */ silk_assert( basis_ptr >= frame_4kHz ); - silk_assert( basis_ptr + sf_length_8kHz <= frame_4kHz + frame_length_4kHz ); + silk_assert( basis_ptr + SF_LENGTH_8KHZ <= frame_4kHz + frame_length_4kHz ); - cross_corr = silk_inner_prod_aligned( target_ptr, basis_ptr, sf_length_8kHz ); + cross_corr = xcorr32[ MAX_LAG_4KHZ - d ]; /* Add contribution of new sample and remove contribution from oldest sample */ - normalizer += + normalizer = silk_ADD32( normalizer, silk_SMULBB( basis_ptr[ 0 ], basis_ptr[ 0 ] ) - - silk_SMULBB( basis_ptr[ sf_length_8kHz ], basis_ptr[ sf_length_8kHz ] ); + silk_SMULBB( basis_ptr[ SF_LENGTH_8KHZ ], basis_ptr[ SF_LENGTH_8KHZ ] ) ); - temp32 = silk_DIV32( cross_corr, silk_SQRT_APPROX( normalizer ) + 1 ); - C[ k ][ d ] = (opus_int16)silk_SAT16( temp32 ); /* Q0 */ + matrix_ptr( C, k, d - MIN_LAG_4KHZ, CSTRIDE_4KHZ) = + (opus_int16)silk_DIV32_varQ( cross_corr, normalizer, 13 + 1 ); /* Q13 */ } /* Update target pointer */ - target_ptr += sf_length_8kHz; + target_ptr += SF_LENGTH_8KHZ; } /* Combine two subframes into single correlation measure and apply short-lag bias */ if( nb_subfr == PE_MAX_NB_SUBFR ) { - for( i = max_lag_4kHz; i >= min_lag_4kHz; i-- ) { - sum = (opus_int32)C[ 0 ][ i ] + (opus_int32)C[ 1 ][ i ]; /* Q0 */ - silk_assert( silk_RSHIFT( sum, 1 ) == silk_SAT16( silk_RSHIFT( sum, 1 ) ) ); - sum = silk_RSHIFT( sum, 1 ); /* Q-1 */ - silk_assert( silk_LSHIFT( (opus_int32)-i, 4 ) == silk_SAT16( silk_LSHIFT( (opus_int32)-i, 4 ) ) ); - sum = silk_SMLAWB( sum, sum, silk_LSHIFT( -i, 4 ) ); /* Q-1 */ - silk_assert( sum == silk_SAT16( sum ) ); - C[ 0 ][ i ] = (opus_int16)sum; /* Q-1 */ + for( i = MAX_LAG_4KHZ; i >= MIN_LAG_4KHZ; i-- ) { + sum = (opus_int32)matrix_ptr( C, 0, i - MIN_LAG_4KHZ, CSTRIDE_4KHZ ) + + (opus_int32)matrix_ptr( C, 1, i - MIN_LAG_4KHZ, CSTRIDE_4KHZ ); /* Q14 */ + sum = silk_SMLAWB( sum, sum, silk_LSHIFT( -i, 4 ) ); /* Q14 */ + C[ i - MIN_LAG_4KHZ ] = (opus_int16)sum; /* Q14 */ } } else { /* Only short-lag bias */ - for( i = max_lag_4kHz; i >= min_lag_4kHz; i-- ) { - sum = (opus_int32)C[ 0 ][ i ]; - sum = silk_SMLAWB( sum, sum, silk_LSHIFT( -i, 4 ) ); /* Q-1 */ - C[ 0 ][ i ] = (opus_int16)sum; /* Q-1 */ + for( i = MAX_LAG_4KHZ; i >= MIN_LAG_4KHZ; i-- ) { + sum = silk_LSHIFT( (opus_int32)C[ i - MIN_LAG_4KHZ ], 1 ); /* Q14 */ + sum = silk_SMLAWB( sum, sum, silk_LSHIFT( -i, 4 ) ); /* Q14 */ + C[ i - MIN_LAG_4KHZ ] = (opus_int16)sum; /* Q14 */ } } /* Sort */ length_d_srch = silk_ADD_LSHIFT32( 4, complexity, 1 ); silk_assert( 3 * length_d_srch <= PE_D_SRCH_LENGTH ); - silk_insertion_sort_decreasing_int16( &C[ 0 ][ min_lag_4kHz ], d_srch, max_lag_4kHz - min_lag_4kHz + 1, length_d_srch ); + silk_insertion_sort_decreasing_int16( C, d_srch, CSTRIDE_4KHZ, + length_d_srch ); /* Escape if correlation is very low already here */ - target_ptr = &frame_4kHz[ silk_SMULBB( sf_length_4kHz, nb_subfr ) ]; - energy = silk_inner_prod_aligned( target_ptr, target_ptr, silk_LSHIFT( sf_length_4kHz, 2 ) ); - energy = silk_ADD_SAT32( energy, 1000 ); /* Q0 */ - Cmax = (opus_int)C[ 0 ][ min_lag_4kHz ]; /* Q-1 */ - threshold = silk_SMULBB( Cmax, Cmax ); /* Q-2 */ - - /* Compare in Q-2 domain */ - if( silk_RSHIFT( energy, 4 + 2 ) > threshold ) { + Cmax = (opus_int)C[ 0 ]; /* Q14 */ + if( Cmax < SILK_FIX_CONST( 0.2, 14 ) ) { silk_memset( pitch_out, 0, nb_subfr * sizeof( opus_int ) ); *LTPCorr_Q15 = 0; *lagIndex = 0; *contourIndex = 0; + RESTORE_STACK; return 1; } threshold = silk_SMULWB( search_thres1_Q16, Cmax ); for( i = 0; i < length_d_srch; i++ ) { /* Convert to 8 kHz indices for the sorted correlation that exceeds the threshold */ - if( C[ 0 ][ min_lag_4kHz + i ] > threshold ) { - d_srch[ i ] = silk_LSHIFT( d_srch[ i ] + min_lag_4kHz, 1 ); + if( C[ i ] > threshold ) { + d_srch[ i ] = silk_LSHIFT( d_srch[ i ] + MIN_LAG_4KHZ, 1 ); } else { length_d_srch = i; break; @@ -265,34 +270,37 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 } silk_assert( length_d_srch > 0 ); - for( i = min_lag_8kHz - 5; i < max_lag_8kHz + 5; i++ ) { - d_comp[ i ] = 0; + ALLOC( d_comp, D_COMP_STRIDE, opus_int16 ); + for( i = D_COMP_MIN; i < D_COMP_MAX; i++ ) { + d_comp[ i - D_COMP_MIN ] = 0; } for( i = 0; i < length_d_srch; i++ ) { - d_comp[ d_srch[ i ] ] = 1; + d_comp[ d_srch[ i ] - D_COMP_MIN ] = 1; } /* Convolution */ - for( i = max_lag_8kHz + 3; i >= min_lag_8kHz; i-- ) { - d_comp[ i ] += d_comp[ i - 1 ] + d_comp[ i - 2 ]; + for( i = D_COMP_MAX - 1; i >= MIN_LAG_8KHZ; i-- ) { + d_comp[ i - D_COMP_MIN ] += + d_comp[ i - 1 - D_COMP_MIN ] + d_comp[ i - 2 - D_COMP_MIN ]; } length_d_srch = 0; - for( i = min_lag_8kHz; i < max_lag_8kHz + 1; i++ ) { - if( d_comp[ i + 1 ] > 0 ) { + for( i = MIN_LAG_8KHZ; i < MAX_LAG_8KHZ + 1; i++ ) { + if( d_comp[ i + 1 - D_COMP_MIN ] > 0 ) { d_srch[ length_d_srch ] = i; length_d_srch++; } } /* Convolution */ - for( i = max_lag_8kHz + 3; i >= min_lag_8kHz; i-- ) { - d_comp[ i ] += d_comp[ i - 1 ] + d_comp[ i - 2 ] + d_comp[ i - 3 ]; + for( i = D_COMP_MAX - 1; i >= MIN_LAG_8KHZ; i-- ) { + d_comp[ i - D_COMP_MIN ] += d_comp[ i - 1 - D_COMP_MIN ] + + d_comp[ i - 2 - D_COMP_MIN ] + d_comp[ i - 3 - D_COMP_MIN ]; } length_d_comp = 0; - for( i = min_lag_8kHz; i < max_lag_8kHz + 4; i++ ) { - if( d_comp[ i ] > 0 ) { + for( i = MIN_LAG_8KHZ; i < D_COMP_MAX; i++ ) { + if( d_comp[ i - D_COMP_MIN ] > 0 ) { d_comp[ length_d_comp ] = i - 2; length_d_comp++; } @@ -306,8 +314,9 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 ** Scale signal down to avoid correlations measures from overflowing *******************************************************************************/ /* find scaling as max scaling for each subframe */ - shift = silk_P_Ana_find_scaling( frame_8kHz, frame_length_8kHz, sf_length_8kHz ); + silk_sum_sqr_shift( &energy, &shift, frame_8kHz, frame_length_8kHz ); if( shift > 0 ) { + shift = silk_RSHIFT( shift, 1 ); for( i = 0; i < frame_length_8kHz; i++ ) { frame_8kHz[ i ] = silk_RSHIFT( frame_8kHz[ i ], shift ); } @@ -316,43 +325,37 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 /********************************************************************************* * Find energy of each subframe projected onto its history, for a range of delays *********************************************************************************/ - silk_memset( C, 0, PE_MAX_NB_SUBFR * ( ( PE_MAX_LAG >> 1 ) + 5 ) * sizeof( opus_int16 ) ); + silk_memset( C, 0, nb_subfr * CSTRIDE_8KHZ * sizeof( opus_int16 ) ); target_ptr = &frame_8kHz[ PE_LTP_MEM_LENGTH_MS * 8 ]; for( k = 0; k < nb_subfr; k++ ) { /* Check that we are within range of the array */ silk_assert( target_ptr >= frame_8kHz ); - silk_assert( target_ptr + sf_length_8kHz <= frame_8kHz + frame_length_8kHz ); + silk_assert( target_ptr + SF_LENGTH_8KHZ <= frame_8kHz + frame_length_8kHz ); - energy_target = silk_inner_prod_aligned( target_ptr, target_ptr, sf_length_8kHz ); + energy_target = silk_ADD32( silk_inner_prod_aligned( target_ptr, target_ptr, SF_LENGTH_8KHZ ), 1 ); for( j = 0; j < length_d_comp; j++ ) { d = d_comp[ j ]; basis_ptr = target_ptr - d; /* Check that we are within range of the array */ silk_assert( basis_ptr >= frame_8kHz ); - silk_assert( basis_ptr + sf_length_8kHz <= frame_8kHz + frame_length_8kHz ); + silk_assert( basis_ptr + SF_LENGTH_8KHZ <= frame_8kHz + frame_length_8kHz ); - cross_corr = silk_inner_prod_aligned( target_ptr, basis_ptr, sf_length_8kHz ); - energy_basis = silk_inner_prod_aligned( basis_ptr, basis_ptr, sf_length_8kHz ); + cross_corr = silk_inner_prod_aligned( target_ptr, basis_ptr, SF_LENGTH_8KHZ ); if( cross_corr > 0 ) { - energy = silk_max( energy_target, energy_basis ); /* Find max to make sure first division < 1.0 */ - lz = silk_CLZ32( cross_corr ); - lshift = silk_LIMIT_32( lz - 1, 0, 15 ); - temp32 = silk_DIV32( silk_LSHIFT( cross_corr, lshift ), silk_RSHIFT( energy, 15 - lshift ) + 1 ); /* Q15 */ - silk_assert( temp32 == silk_SAT16( temp32 ) ); - temp32 = silk_SMULWB( cross_corr, temp32 ); /* Q(-1), cc * ( cc / max(b, t) ) */ - temp32 = silk_ADD_SAT32( temp32, temp32 ); /* Q(0) */ - lz = silk_CLZ32( temp32 ); - lshift = silk_LIMIT_32( lz - 1, 0, 15 ); - energy = silk_min( energy_target, energy_basis ); - C[ k ][ d ] = silk_DIV32( silk_LSHIFT( temp32, lshift ), silk_RSHIFT( energy, 15 - lshift ) + 1 ); /* Q15*/ + energy_basis = silk_inner_prod_aligned( basis_ptr, basis_ptr, SF_LENGTH_8KHZ ); + matrix_ptr( C, k, d - ( MIN_LAG_8KHZ - 2 ), CSTRIDE_8KHZ ) = + (opus_int16)silk_DIV32_varQ( cross_corr, + silk_ADD32( energy_target, + energy_basis ), + 13 + 1 ); /* Q13 */ } else { - C[ k ][ d ] = 0; + matrix_ptr( C, k, d - ( MIN_LAG_8KHZ - 2 ), CSTRIDE_8KHZ ) = 0; } } - target_ptr += sf_length_8kHz; + target_ptr += SF_LENGTH_8KHZ; } /* search over lag range and lags codebook */ @@ -374,7 +377,7 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 } else { prevLag_log2_Q7 = 0; } - silk_assert( search_thres2_Q15 == silk_SAT16( search_thres2_Q15 ) ); + silk_assert( search_thres2_Q13 == silk_SAT16( search_thres2_Q13 ) ); /* Set up stage 2 codebook based on number of subframes */ if( nb_subfr == PE_MAX_NB_SUBFR ) { cbk_size = PE_NB_CBKS_STAGE2_EXT; @@ -385,12 +388,10 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 } else { nb_cbk_search = PE_NB_CBKS_STAGE2; } - corr_thres_Q15 = silk_RSHIFT( silk_SMULBB( search_thres2_Q15, search_thres2_Q15 ), 13 ); } else { cbk_size = PE_NB_CBKS_STAGE2_10MS; Lag_CB_ptr = &silk_CB_lags_stage2_10_ms[ 0 ][ 0 ]; nb_cbk_search = PE_NB_CBKS_STAGE2_10MS; - corr_thres_Q15 = silk_RSHIFT( silk_SMULBB( search_thres2_Q15, search_thres2_Q15 ), 14 ); } for( k = 0; k < length_d_srch; k++ ) { @@ -398,8 +399,13 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 for( j = 0; j < nb_cbk_search; j++ ) { CC[ j ] = 0; for( i = 0; i < nb_subfr; i++ ) { + opus_int d_subfr; /* Try all codebooks */ - CC[ j ] = CC[ j ] + (opus_int32)C[ i ][ d + matrix_ptr( Lag_CB_ptr, i, j, cbk_size )]; + d_subfr = d + matrix_ptr( Lag_CB_ptr, i, j, cbk_size ); + CC[ j ] = CC[ j ] + + (opus_int32)matrix_ptr( C, i, + d_subfr - ( MIN_LAG_8KHZ - 2 ), + CSTRIDE_8KHZ ); } } /* Find best codebook */ @@ -413,25 +419,25 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 } /* Bias towards shorter lags */ - lag_log2_Q7 = silk_lin2log( (opus_int32)d ); /* Q7 */ + lag_log2_Q7 = silk_lin2log( d ); /* Q7 */ silk_assert( lag_log2_Q7 == silk_SAT16( lag_log2_Q7 ) ); - silk_assert( nb_subfr * SILK_FIX_CONST( PE_SHORTLAG_BIAS, 15 ) == silk_SAT16( nb_subfr * SILK_FIX_CONST( PE_SHORTLAG_BIAS, 15 ) ) ); - CCmax_new_b = CCmax_new - silk_RSHIFT( silk_SMULBB( nb_subfr * SILK_FIX_CONST( PE_SHORTLAG_BIAS, 15 ), lag_log2_Q7 ), 7 ); /* Q15 */ + silk_assert( nb_subfr * SILK_FIX_CONST( PE_SHORTLAG_BIAS, 13 ) == silk_SAT16( nb_subfr * SILK_FIX_CONST( PE_SHORTLAG_BIAS, 13 ) ) ); + CCmax_new_b = CCmax_new - silk_RSHIFT( silk_SMULBB( nb_subfr * SILK_FIX_CONST( PE_SHORTLAG_BIAS, 13 ), lag_log2_Q7 ), 7 ); /* Q13 */ /* Bias towards previous lag */ - silk_assert( nb_subfr * SILK_FIX_CONST( PE_PREVLAG_BIAS, 15 ) == silk_SAT16( nb_subfr * SILK_FIX_CONST( PE_PREVLAG_BIAS, 15 ) ) ); + silk_assert( nb_subfr * SILK_FIX_CONST( PE_PREVLAG_BIAS, 13 ) == silk_SAT16( nb_subfr * SILK_FIX_CONST( PE_PREVLAG_BIAS, 13 ) ) ); if( prevLag > 0 ) { delta_lag_log2_sqr_Q7 = lag_log2_Q7 - prevLag_log2_Q7; silk_assert( delta_lag_log2_sqr_Q7 == silk_SAT16( delta_lag_log2_sqr_Q7 ) ); delta_lag_log2_sqr_Q7 = silk_RSHIFT( silk_SMULBB( delta_lag_log2_sqr_Q7, delta_lag_log2_sqr_Q7 ), 7 ); - prev_lag_bias_Q15 = silk_RSHIFT( silk_SMULBB( nb_subfr * SILK_FIX_CONST( PE_PREVLAG_BIAS, 15 ), *LTPCorr_Q15 ), 15 ); /* Q15 */ - prev_lag_bias_Q15 = silk_DIV32( silk_MUL( prev_lag_bias_Q15, delta_lag_log2_sqr_Q7 ), delta_lag_log2_sqr_Q7 + ( 1 << 6 ) ); - CCmax_new_b -= prev_lag_bias_Q15; /* Q15 */ + prev_lag_bias_Q13 = silk_RSHIFT( silk_SMULBB( nb_subfr * SILK_FIX_CONST( PE_PREVLAG_BIAS, 13 ), *LTPCorr_Q15 ), 15 ); /* Q13 */ + prev_lag_bias_Q13 = silk_DIV32( silk_MUL( prev_lag_bias_Q13, delta_lag_log2_sqr_Q7 ), delta_lag_log2_sqr_Q7 + SILK_FIX_CONST( 0.5, 7 ) ); + CCmax_new_b -= prev_lag_bias_Q13; /* Q13 */ } if( CCmax_new_b > CCmax_b && /* Find maximum biased correlation */ - CCmax_new > corr_thres_Q15 && /* Correlation needs to be high enough to be voiced */ - silk_CB_lags_stage2[ 0 ][ CBimax_new ] <= min_lag_8kHz /* Lag must be in range */ + CCmax_new > silk_SMULBB( nb_subfr, search_thres2_Q13 ) && /* Correlation needs to be high enough to be voiced */ + silk_CB_lags_stage2[ 0 ][ CBimax_new ] <= MIN_LAG_8KHZ /* Lag must be in range */ ) { CCmax_b = CCmax_new_b; CCmax = CCmax_new; @@ -446,24 +452,31 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 *LTPCorr_Q15 = 0; *lagIndex = 0; *contourIndex = 0; + RESTORE_STACK; return 1; } + /* Output normalized correlation */ + *LTPCorr_Q15 = (opus_int)silk_LSHIFT( silk_DIV32_16( CCmax, nb_subfr ), 2 ); + silk_assert( *LTPCorr_Q15 >= 0 ); + if( Fs_kHz > 8 ) { + VARDECL( opus_int16, scratch_mem ); /***************************************************************************/ /* Scale input signal down to avoid correlations measures from overflowing */ /***************************************************************************/ /* find scaling as max scaling for each subframe */ - shift = silk_P_Ana_find_scaling( frame, frame_length, sf_length ); + silk_sum_sqr_shift( &energy, &shift, frame, frame_length ); + ALLOC( scratch_mem, shift > 0 ? frame_length : ALLOC_NONE, opus_int16 ); if( shift > 0 ) { /* Move signal to scratch mem because the input signal should be unchanged */ - /* Reuse the 32 bit scratch mem vector, use a 16 bit pointer from now */ - input_frame_ptr = (opus_int16*)scratch_mem; + shift = silk_RSHIFT( shift, 1 ); for( i = 0; i < frame_length; i++ ) { - input_frame_ptr[ i ] = silk_RSHIFT( frame[ i ], shift ); + scratch_mem[ i ] = silk_RSHIFT( frame[ i ], shift ); } + input_frame_ptr = scratch_mem; } else { - input_frame_ptr = (opus_int16*)frame; + input_frame_ptr = frame; } /* Search in original signal */ @@ -483,22 +496,13 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 start_lag = silk_max_int( lag - 2, min_lag ); end_lag = silk_min_int( lag + 2, max_lag ); lag_new = lag; /* to avoid undefined lag */ - CBimax = 0; /* to avoid undefined lag */ - silk_assert( silk_LSHIFT( CCmax, 13 ) >= 0 ); - *LTPCorr_Q15 = (opus_int)silk_SQRT_APPROX( silk_LSHIFT( CCmax, 13 ) ); /* Output normalized correlation */ + CBimax = 0; /* to avoid undefined lag */ CCmax = silk_int32_MIN; /* pitch lags according to second stage */ for( k = 0; k < nb_subfr; k++ ) { pitch_out[ k ] = lag + 2 * silk_CB_lags_stage2[ k ][ CBimax_old ]; } - /* Calculate the correlations and energies needed in stage 3 */ - silk_P_Ana_calc_corr_st3( crosscorr_st3, input_frame_ptr, start_lag, sf_length, nb_subfr, complexity ); - silk_P_Ana_calc_energy_st3( energies_st3, input_frame_ptr, start_lag, sf_length, nb_subfr, complexity ); - - lag_counter = 0; - silk_assert( lag == silk_SAT16( lag ) ); - contour_bias_Q20 = silk_DIV32_16( SILK_FIX_CONST( PE_FLATCONTOUR_BIAS, 20 ), lag ); /* Set up codebook parameters according to complexity setting and frame length */ if( nb_subfr == PE_MAX_NB_SUBFR ) { @@ -510,41 +514,43 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 cbk_size = PE_NB_CBKS_STAGE3_10MS; Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; } + + /* Calculate the correlations and energies needed in stage 3 */ + ALLOC( energies_st3, nb_subfr * nb_cbk_search, silk_pe_stage3_vals ); + ALLOC( cross_corr_st3, nb_subfr * nb_cbk_search, silk_pe_stage3_vals ); + silk_P_Ana_calc_corr_st3( cross_corr_st3, input_frame_ptr, start_lag, sf_length, nb_subfr, complexity, arch ); + silk_P_Ana_calc_energy_st3( energies_st3, input_frame_ptr, start_lag, sf_length, nb_subfr, complexity ); + + lag_counter = 0; + silk_assert( lag == silk_SAT16( lag ) ); + contour_bias_Q15 = silk_DIV32_16( SILK_FIX_CONST( PE_FLATCONTOUR_BIAS, 15 ), lag ); + + target_ptr = &input_frame_ptr[ PE_LTP_MEM_LENGTH_MS * Fs_kHz ]; + energy_target = silk_ADD32( silk_inner_prod_aligned( target_ptr, target_ptr, nb_subfr * sf_length ), 1 ); for( d = start_lag; d <= end_lag; d++ ) { for( j = 0; j < nb_cbk_search; j++ ) { cross_corr = 0; - energy = 0; + energy = energy_target; for( k = 0; k < nb_subfr; k++ ) { - silk_assert( PE_MAX_NB_SUBFR == 4 ); - energy += silk_RSHIFT( energies_st3[ k ][ j ][ lag_counter ], 2 ); /* use mean, to avoid overflow */ + cross_corr = silk_ADD32( cross_corr, + matrix_ptr( cross_corr_st3, k, j, + nb_cbk_search )[ lag_counter ] ); + energy = silk_ADD32( energy, + matrix_ptr( energies_st3, k, j, + nb_cbk_search )[ lag_counter ] ); silk_assert( energy >= 0 ); - cross_corr += silk_RSHIFT( crosscorr_st3[ k ][ j ][ lag_counter ], 2 ); /* use mean, to avoid overflow */ } if( cross_corr > 0 ) { - /* Divide cross_corr / energy and get result in Q15 */ - lz = silk_CLZ32( cross_corr ); - /* Divide with result in Q13, cross_corr could be larger than energy */ - lshift = silk_LIMIT_32( lz - 1, 0, 13 ); - CCmax_new = silk_DIV32( silk_LSHIFT( cross_corr, lshift ), silk_RSHIFT( energy, 13 - lshift ) + 1 ); - CCmax_new = silk_SAT16( CCmax_new ); - CCmax_new = silk_SMULWB( cross_corr, CCmax_new ); - /* Saturate */ - if( CCmax_new > silk_RSHIFT( silk_int32_MAX, 3 ) ) { - CCmax_new = silk_int32_MAX; - } else { - CCmax_new = silk_LSHIFT( CCmax_new, 3 ); - } + CCmax_new = silk_DIV32_varQ( cross_corr, energy, 13 + 1 ); /* Q13 */ /* Reduce depending on flatness of contour */ - diff = silk_int16_MAX - silk_RSHIFT( silk_MUL( contour_bias_Q20, j ), 5 ); /* Q20 -> Q15 */ + diff = silk_int16_MAX - silk_MUL( contour_bias_Q15, j ); /* Q15 */ silk_assert( diff == silk_SAT16( diff ) ); - CCmax_new = silk_LSHIFT( silk_SMULWB( CCmax_new, diff ), 1 ); + CCmax_new = silk_SMULWB( CCmax_new, diff ); /* Q14 */ } else { CCmax_new = 0; } - if( CCmax_new > CCmax && - ( d + silk_CB_lags_stage3[ 0 ][ j ] ) <= max_lag - ) { + if( CCmax_new > CCmax && ( d + silk_CB_lags_stage3[ 0 ][ j ] ) <= max_lag ) { CCmax = CCmax_new; lag_new = d; CBimax = j; @@ -560,40 +566,50 @@ opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 *lagIndex = (opus_int16)( lag_new - min_lag); *contourIndex = (opus_int8)CBimax; } else { /* Fs_kHz == 8 */ - /* Save Lags and correlation */ - CCmax = silk_max( CCmax, 0 ); - *LTPCorr_Q15 = (opus_int)silk_SQRT_APPROX( silk_LSHIFT( CCmax, 13 ) ); /* Output normalized correlation */ + /* Save Lags */ for( k = 0; k < nb_subfr; k++ ) { pitch_out[ k ] = lag + matrix_ptr( Lag_CB_ptr, k, CBimax, cbk_size ); - pitch_out[ k ] = silk_LIMIT( pitch_out[ k ], min_lag_8kHz, PE_MAX_LAG_MS * Fs_kHz ); + pitch_out[ k ] = silk_LIMIT( pitch_out[ k ], MIN_LAG_8KHZ, PE_MAX_LAG_MS * 8 ); } - *lagIndex = (opus_int16)( lag - min_lag_8kHz ); + *lagIndex = (opus_int16)( lag - MIN_LAG_8KHZ ); *contourIndex = (opus_int8)CBimax; } silk_assert( *lagIndex >= 0 ); /* return as voiced */ + RESTORE_STACK; return 0; } -/*************************************************************************/ -/* Calculates the correlations used in stage 3 search. In order to cover */ -/* the whole lag codebook for all the searched offset lags (lag +- 2), */ -/*************************************************************************/ -void silk_P_Ana_calc_corr_st3( - opus_int32 cross_corr_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ],/* (O) 3 DIM correlation array */ +/*********************************************************************** + * Calculates the correlations used in stage 3 search. In order to cover + * the whole lag codebook for all the searched offset lags (lag +- 2), + * the following correlations are needed in each sub frame: + * + * sf1: lag range [-8,...,7] total 16 correlations + * sf2: lag range [-4,...,4] total 9 correlations + * sf3: lag range [-3,....4] total 8 correltions + * sf4: lag range [-6,....8] total 15 correlations + * + * In total 48 correlations. The direct implementation computed in worst + * case 4*12*5 = 240 correlations, but more likely around 120. + ***********************************************************************/ +static void silk_P_Ana_calc_corr_st3( + silk_pe_stage3_vals cross_corr_st3[], /* O 3 DIM correlation array */ const opus_int16 frame[], /* I vector to correlate */ opus_int start_lag, /* I lag offset to search around */ opus_int sf_length, /* I length of a 5 ms subframe */ opus_int nb_subfr, /* I number of subframes */ - opus_int complexity /* I Complexity setting */ + opus_int complexity, /* I Complexity setting */ + int arch /* I Run-time architecture */ ) { - const opus_int16 *target_ptr, *basis_ptr; - opus_int32 cross_corr; + const opus_int16 *target_ptr; opus_int i, j, k, lag_counter, lag_low, lag_high; opus_int nb_cbk_search, delta, idx, cbk_size; - opus_int32 scratch_mem[ SCRATCH_SIZE ]; + VARDECL( opus_int32, scratch_mem ); + VARDECL( opus_int32, xcorr32 ); const opus_int8 *Lag_range_ptr, *Lag_CB_ptr; + SAVE_STACK; silk_assert( complexity >= SILK_PE_MIN_COMPLEX ); silk_assert( complexity <= SILK_PE_MAX_COMPLEX ); @@ -610,6 +626,8 @@ void silk_P_Ana_calc_corr_st3( nb_cbk_search = PE_NB_CBKS_STAGE3_10MS; cbk_size = PE_NB_CBKS_STAGE3_10MS; } + ALLOC( scratch_mem, SCRATCH_SIZE, opus_int32 ); + ALLOC( xcorr32, SCRATCH_SIZE, opus_int32 ); target_ptr = &frame[ silk_LSHIFT( sf_length, 2 ) ]; /* Pointer to middle of frame */ for( k = 0; k < nb_subfr; k++ ) { @@ -618,11 +636,11 @@ void silk_P_Ana_calc_corr_st3( /* Calculate the correlations for each subframe */ lag_low = matrix_ptr( Lag_range_ptr, k, 0, 2 ); lag_high = matrix_ptr( Lag_range_ptr, k, 1, 2 ); + silk_assert(lag_high-lag_low+1 <= SCRATCH_SIZE); + celt_pitch_xcorr( target_ptr, target_ptr - start_lag - lag_high, xcorr32, sf_length, lag_high - lag_low + 1, arch ); for( j = lag_low; j <= lag_high; j++ ) { - basis_ptr = target_ptr - ( start_lag + j ); - cross_corr = silk_inner_prod_aligned( (opus_int16*)target_ptr, (opus_int16*)basis_ptr, sf_length ); silk_assert( lag_counter < SCRATCH_SIZE ); - scratch_mem[ lag_counter ] = cross_corr; + scratch_mem[ lag_counter ] = xcorr32[ lag_high - j ]; lag_counter++; } @@ -634,32 +652,35 @@ void silk_P_Ana_calc_corr_st3( for( j = 0; j < PE_NB_STAGE3_LAGS; j++ ) { silk_assert( idx + j < SCRATCH_SIZE ); silk_assert( idx + j < lag_counter ); - cross_corr_st3[ k ][ i ][ j ] = scratch_mem[ idx + j ]; + matrix_ptr( cross_corr_st3, k, i, nb_cbk_search )[ j ] = + scratch_mem[ idx + j ]; } } target_ptr += sf_length; } + RESTORE_STACK; } /********************************************************************/ /* Calculate the energies for first two subframes. The energies are */ /* calculated recursively. */ /********************************************************************/ -void silk_P_Ana_calc_energy_st3( - opus_int32 energies_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ],/* (O) 3 DIM energy array */ - const opus_int16 frame[], /* I vector to calc energy in */ - opus_int start_lag, /* I lag offset to search around */ - opus_int sf_length, /* I length of one 5 ms subframe */ - opus_int nb_subfr, /* I number of subframes */ - opus_int complexity /* I Complexity setting */ +static void silk_P_Ana_calc_energy_st3( + silk_pe_stage3_vals energies_st3[], /* O 3 DIM energy array */ + const opus_int16 frame[], /* I vector to calc energy in */ + opus_int start_lag, /* I lag offset to search around */ + opus_int sf_length, /* I length of one 5 ms subframe */ + opus_int nb_subfr, /* I number of subframes */ + opus_int complexity /* I Complexity setting */ ) { const opus_int16 *target_ptr, *basis_ptr; opus_int32 energy; opus_int k, i, j, lag_counter; opus_int nb_cbk_search, delta, idx, cbk_size, lag_diff; - opus_int32 scratch_mem[ SCRATCH_SIZE ]; + VARDECL( opus_int32, scratch_mem ); const opus_int8 *Lag_range_ptr, *Lag_CB_ptr; + SAVE_STACK; silk_assert( complexity >= SILK_PE_MIN_COMPLEX ); silk_assert( complexity <= SILK_PE_MAX_COMPLEX ); @@ -676,6 +697,8 @@ void silk_P_Ana_calc_energy_st3( nb_cbk_search = PE_NB_CBKS_STAGE3_10MS; cbk_size = PE_NB_CBKS_STAGE3_10MS; } + ALLOC( scratch_mem, SCRATCH_SIZE, opus_int32 ); + target_ptr = &frame[ silk_LSHIFT( sf_length, 2 ) ]; for( k = 0; k < nb_subfr; k++ ) { lag_counter = 0; @@ -709,37 +732,13 @@ void silk_P_Ana_calc_energy_st3( for( j = 0; j < PE_NB_STAGE3_LAGS; j++ ) { silk_assert( idx + j < SCRATCH_SIZE ); silk_assert( idx + j < lag_counter ); - energies_st3[ k ][ i ][ j ] = scratch_mem[ idx + j ]; - silk_assert( energies_st3[ k ][ i ][ j ] >= 0 ); + matrix_ptr( energies_st3, k, i, nb_cbk_search )[ j ] = + scratch_mem[ idx + j ]; + silk_assert( + matrix_ptr( energies_st3, k, i, nb_cbk_search )[ j ] >= 0 ); } } target_ptr += sf_length; } -} - -opus_int32 silk_P_Ana_find_scaling( - const opus_int16 *frame, - const opus_int frame_length, - const opus_int sum_sqr_len -) -{ - opus_int32 nbits, x_max; - - x_max = silk_int16_array_maxabs( frame, frame_length ); - - if( x_max < silk_int16_MAX ) { - /* Number of bits needed for the sum of the squares */ - nbits = 32 - silk_CLZ32( silk_SMULBB( x_max, x_max ) ); - } else { - /* Here we don't know if x_max should have been silk_int16_MAX + 1, so we expect the worst case */ - nbits = 30; - } - nbits += 17 - silk_CLZ16( sum_sqr_len ); - - /* Without a guarantee of saturation, we need to keep the 31st bit free */ - if( nbits < 31 ) { - return 0; - } else { - return( nbits - 30 ); - } + RESTORE_STACK; } diff --git a/src/opus-1.0.2/silk/fixed/prefilter_FIX.c b/src/opus-1.1/silk/fixed/prefilter_FIX.c index a96f5118..d381730c 100644 --- a/src/opus-1.0.2/silk/fixed/prefilter_FIX.c +++ b/src/opus-1.1/silk/fixed/prefilter_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,10 +30,11 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main_FIX.h" +#include "stack_alloc.h" #include "tuning_parameters.h" /* Prefilter for finding Quantizer input signal */ -static inline void silk_prefilt_FIX( +static OPUS_INLINE void silk_prefilt_FIX( silk_prefilter_state_FIX *P, /* I/O state */ opus_int32 st_res_Q12[], /* I short term residual signal */ opus_int32 xw_Q3[], /* O prefiltered signal */ @@ -101,14 +102,17 @@ void silk_prefilter_FIX( opus_int32 *pxw_Q3; opus_int HarmShapeGain_Q12, Tilt_Q14; opus_int32 HarmShapeFIRPacked_Q12, LF_shp_Q14; - opus_int32 x_filt_Q12[ MAX_SUB_FRAME_LENGTH ]; - opus_int32 st_res_Q2[ MAX_SUB_FRAME_LENGTH + MAX_LPC_ORDER ]; + VARDECL( opus_int32, x_filt_Q12 ); + VARDECL( opus_int32, st_res_Q2 ); opus_int16 B_Q10[ 2 ]; + SAVE_STACK; /* Set up pointers */ px = x; pxw_Q3 = xw_Q3; lag = P->lagPrev; + ALLOC( x_filt_Q12, psEnc->sCmn.subfr_length, opus_int32 ); + ALLOC( st_res_Q2, psEnc->sCmn.subfr_length, opus_int32 ); for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { /* Update Variables that change per sub frame */ if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { @@ -148,10 +152,11 @@ void silk_prefilter_FIX( } P->lagPrev = psEncCtrl->pitchL[ psEnc->sCmn.nb_subfr - 1 ]; + RESTORE_STACK; } /* Prefilter for finding Quantizer input signal */ -static inline void silk_prefilt_FIX( +static OPUS_INLINE void silk_prefilt_FIX( silk_prefilter_state_FIX *P, /* I/O state */ opus_int32 st_res_Q12[], /* I short term residual signal */ opus_int32 xw_Q3[], /* O prefiltered signal */ diff --git a/src/opus-1.0.2/silk/fixed/process_gains_FIX.c b/src/opus-1.1/silk/fixed/process_gains_FIX.c index 22d3a71a..05aba317 100644 --- a/src/opus-1.0.2/silk/fixed/process_gains_FIX.c +++ b/src/opus-1.1/silk/fixed/process_gains_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/fixed/regularize_correlations_FIX.c b/src/opus-1.1/silk/fixed/regularize_correlations_FIX.c index 098c1509..a2836b05 100644 --- a/src/opus-1.0.2/silk/fixed/regularize_correlations_FIX.c +++ b/src/opus-1.1/silk/fixed/regularize_correlations_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/fixed/residual_energy16_FIX.c b/src/opus-1.1/silk/fixed/residual_energy16_FIX.c index d61e8493..ebffb2a6 100644 --- a/src/opus-1.0.2/silk/fixed/residual_energy16_FIX.c +++ b/src/opus-1.1/silk/fixed/residual_energy16_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/fixed/residual_energy_FIX.c b/src/opus-1.1/silk/fixed/residual_energy_FIX.c index f284e51f..105ae318 100644 --- a/src/opus-1.0.2/silk/fixed/residual_energy_FIX.c +++ b/src/opus-1.1/silk/fixed/residual_energy_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,6 +30,7 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main_FIX.h" +#include "stack_alloc.h" /* Calculates residual energies of input subframes where all subframes have LPC_order */ /* of preceding samples */ @@ -45,14 +46,18 @@ void silk_residual_energy_FIX( ) { opus_int offset, i, j, rshift, lz1, lz2; - opus_int16 *LPC_res_ptr, LPC_res[ ( MAX_FRAME_LENGTH + MAX_NB_SUBFR * MAX_LPC_ORDER ) / 2 ]; + opus_int16 *LPC_res_ptr; + VARDECL( opus_int16, LPC_res ); const opus_int16 *x_ptr; opus_int32 tmp32; + SAVE_STACK; x_ptr = x; offset = LPC_order + subfr_length; /* Filter input to create the LPC residual for each frame half, and measure subframe energies */ + ALLOC( LPC_res, ( MAX_NB_SUBFR >> 1 ) * offset, opus_int16 ); + silk_assert( ( nb_subfr >> 1 ) * ( MAX_NB_SUBFR >> 1 ) == nb_subfr ); for( i = 0; i < nb_subfr >> 1; i++ ) { /* Calculate half frame LPC residual signal including preceding samples */ silk_LPC_analysis_filter( LPC_res, x_ptr, a_Q12[ i ], ( MAX_NB_SUBFR >> 1 ) * offset, LPC_order ); @@ -88,4 +93,5 @@ void silk_residual_energy_FIX( nrgs[ i ] = silk_SMMUL( tmp32, silk_LSHIFT32( nrgs[ i ], lz1 ) ); /* Q( nrgsQ[ i ] + lz1 + 2 * lz2 - 32 - 32 )*/ nrgsQ[ i ] += lz1 + 2 * lz2 - 32 - 32; } + RESTORE_STACK; } diff --git a/src/opus-1.0.2/silk/fixed/schur64_FIX.c b/src/opus-1.1/silk/fixed/schur64_FIX.c index 5ff27567..764a10ef 100644 --- a/src/opus-1.0.2/silk/fixed/schur64_FIX.c +++ b/src/opus-1.1/silk/fixed/schur64_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -56,6 +56,17 @@ opus_int32 silk_schur64( /* O returns residual ene } for( k = 0; k < order; k++ ) { + /* Check that we won't be getting an unstable rc, otherwise stop here. */ + if (silk_abs_int32(C[ k + 1 ][ 0 ]) >= C[ 0 ][ 1 ]) { + if ( C[ k + 1 ][ 0 ] > 0 ) { + rc_Q16[ k ] = -SILK_FIX_CONST( .99f, 16 ); + } else { + rc_Q16[ k ] = SILK_FIX_CONST( .99f, 16 ); + } + k++; + break; + } + /* Get reflection coefficient: divide two Q30 values and get result in Q31 */ rc_tmp_Q31 = silk_DIV32_varQ( -C[ k + 1 ][ 0 ], C[ 0 ][ 1 ], 31 ); @@ -73,5 +84,9 @@ opus_int32 silk_schur64( /* O returns residual ene } } - return( C[ 0 ][ 1 ] ); + for(; k < order; k++ ) { + rc_Q16[ k ] = 0; + } + + return silk_max_32( 1, C[ 0 ][ 1 ] ); } diff --git a/src/opus-1.0.2/silk/fixed/schur_FIX.c b/src/opus-1.1/silk/fixed/schur_FIX.c index 43db5018..c4c0ef23 100644 --- a/src/opus-1.0.2/silk/fixed/schur_FIX.c +++ b/src/opus-1.1/silk/fixed/schur_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -68,6 +68,16 @@ opus_int32 silk_schur( /* O Returns residual ene } for( k = 0; k < order; k++ ) { + /* Check that we won't be getting an unstable rc, otherwise stop here. */ + if (silk_abs_int32(C[ k + 1 ][ 0 ]) >= C[ 0 ][ 1 ]) { + if ( C[ k + 1 ][ 0 ] > 0 ) { + rc_Q15[ k ] = -SILK_FIX_CONST( .99f, 15 ); + } else { + rc_Q15[ k ] = SILK_FIX_CONST( .99f, 15 ); + } + k++; + break; + } /* Get reflection coefficient */ rc_tmp_Q15 = -silk_DIV32_16( C[ k + 1 ][ 0 ], silk_max_32( silk_RSHIFT( C[ 0 ][ 1 ], 15 ), 1 ) ); @@ -87,6 +97,10 @@ opus_int32 silk_schur( /* O Returns residual ene } } + for(; k < order; k++ ) { + rc_Q15[ k ] = 0; + } + /* return residual energy */ - return C[ 0 ][ 1 ]; + return silk_max_32( 1, C[ 0 ][ 1 ] ); } diff --git a/src/opus-1.0.2/silk/fixed/solve_LS_FIX.c b/src/opus-1.1/silk/fixed/solve_LS_FIX.c index fb913abe..51d7d49d 100644 --- a/src/opus-1.0.2/silk/fixed/solve_LS_FIX.c +++ b/src/opus-1.1/silk/fixed/solve_LS_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,6 +30,7 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main_FIX.h" +#include "stack_alloc.h" #include "tuning_parameters.h" /*****************************/ @@ -42,7 +43,7 @@ typedef struct { } inv_D_t; /* Factorize square matrix A into LDL form */ -static inline void silk_LDL_factorize_FIX( +static OPUS_INLINE void silk_LDL_factorize_FIX( opus_int32 *A, /* I/O Pointer to Symetric Square Matrix */ opus_int M, /* I Size of Matrix */ opus_int32 *L_Q16, /* I/O Pointer to Square Upper triangular Matrix */ @@ -50,7 +51,7 @@ static inline void silk_LDL_factorize_FIX( ); /* Solve Lx = b, when L is lower triangular and has ones on the diagonal */ -static inline void silk_LS_SolveFirst_FIX( +static OPUS_INLINE void silk_LS_SolveFirst_FIX( const opus_int32 *L_Q16, /* I Pointer to Lower Triangular Matrix */ opus_int M, /* I Dim of Matrix equation */ const opus_int32 *b, /* I b Vector */ @@ -58,14 +59,14 @@ static inline void silk_LS_SolveFirst_FIX( ); /* Solve L^t*x = b, where L is lower triangular with ones on the diagonal */ -static inline void silk_LS_SolveLast_FIX( +static OPUS_INLINE void silk_LS_SolveLast_FIX( const opus_int32 *L_Q16, /* I Pointer to Lower Triangular Matrix */ const opus_int M, /* I Dim of Matrix equation */ const opus_int32 *b, /* I b Vector */ opus_int32 *x_Q16 /* O x Vector */ ); -static inline void silk_LS_divide_Q16_FIX( +static OPUS_INLINE void silk_LS_divide_Q16_FIX( opus_int32 T[], /* I/O Numenator vector */ inv_D_t *inv_D, /* I 1 / D vector */ opus_int M /* I dimension */ @@ -79,11 +80,13 @@ void silk_solve_LDL_FIX( opus_int32 *x_Q16 /* O Pointer to x solution vector */ ) { - opus_int32 L_Q16[ MAX_MATRIX_SIZE * MAX_MATRIX_SIZE ]; + VARDECL( opus_int32, L_Q16 ); opus_int32 Y[ MAX_MATRIX_SIZE ]; inv_D_t inv_D[ MAX_MATRIX_SIZE ]; + SAVE_STACK; silk_assert( M <= MAX_MATRIX_SIZE ); + ALLOC( L_Q16, M * M, opus_int32 ); /*************************************************** Factorize A by LDL such that A = L*D*L', @@ -107,9 +110,10 @@ void silk_solve_LDL_FIX( x = inv(L') * inv(D) * Y *****************************************************/ silk_LS_SolveLast_FIX( L_Q16, M, Y, x_Q16 ); + RESTORE_STACK; } -static inline void silk_LDL_factorize_FIX( +static OPUS_INLINE void silk_LDL_factorize_FIX( opus_int32 *A, /* I/O Pointer to Symetric Square Matrix */ opus_int M, /* I Size of Matrix */ opus_int32 *L_Q16, /* I/O Pointer to Square Upper triangular Matrix */ @@ -181,7 +185,7 @@ static inline void silk_LDL_factorize_FIX( silk_assert( status == 0 ); } -static inline void silk_LS_divide_Q16_FIX( +static OPUS_INLINE void silk_LS_divide_Q16_FIX( opus_int32 T[], /* I/O Numenator vector */ inv_D_t *inv_D, /* I 1 / D vector */ opus_int M /* I dimension */ @@ -201,7 +205,7 @@ static inline void silk_LS_divide_Q16_FIX( } /* Solve Lx = b, when L is lower triangular and has ones on the diagonal */ -static inline void silk_LS_SolveFirst_FIX( +static OPUS_INLINE void silk_LS_SolveFirst_FIX( const opus_int32 *L_Q16, /* I Pointer to Lower Triangular Matrix */ opus_int M, /* I Dim of Matrix equation */ const opus_int32 *b, /* I b Vector */ @@ -223,7 +227,7 @@ static inline void silk_LS_SolveFirst_FIX( } /* Solve L^t*x = b, where L is lower triangular with ones on the diagonal */ -static inline void silk_LS_SolveLast_FIX( +static OPUS_INLINE void silk_LS_SolveLast_FIX( const opus_int32 *L_Q16, /* I Pointer to Lower Triangular Matrix */ const opus_int M, /* I Dim of Matrix equation */ const opus_int32 *b, /* I b Vector */ diff --git a/src/opus-1.0.2/silk/fixed/structs_FIX.h b/src/opus-1.1/silk/fixed/structs_FIX.h index 4162608b..244b4793 100644 --- a/src/opus-1.0.2/silk/fixed/structs_FIX.h +++ b/src/opus-1.1/silk/fixed/structs_FIX.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/fixed/vector_ops_FIX.c b/src/opus-1.1/silk/fixed/vector_ops_FIX.c index d6206024..509c8b35 100644 --- a/src/opus-1.0.2/silk/fixed/vector_ops_FIX.c +++ b/src/opus-1.1/silk/fixed/vector_ops_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -94,34 +94,3 @@ opus_int64 silk_inner_prod16_aligned_64( } return sum; } - -/* Function that returns the maximum absolut value of the input vector */ -opus_int16 silk_int16_array_maxabs( /* O Maximum absolute value, max: 2^15-1 */ - const opus_int16 *vec, /* I Input vector [len] */ - const opus_int32 len /* I Length of input vector */ -) -{ - opus_int32 max = 0, i, lvl = 0, ind; - if( len == 0 ) return 0; - - ind = len - 1; - max = silk_SMULBB( vec[ ind ], vec[ ind ] ); - for( i = len - 2; i >= 0; i-- ) { - lvl = silk_SMULBB( vec[ i ], vec[ i ] ); - if( lvl > max ) { - max = lvl; - ind = i; - } - } - - /* Do not return 32768, as it will not fit in an int16 so may lead to problems later on */ - if( max >= 1073676289 ) { /* (2^15-1)^2 = 1073676289 */ - return( silk_int16_MAX ); - } else { - if( vec[ ind ] < 0 ) { - return( -vec[ ind ] ); - } else { - return( vec[ ind ] ); - } - } -} diff --git a/src/opus-1.0.2/silk/fixed/warped_autocorrelation_FIX.c b/src/opus-1.1/silk/fixed/warped_autocorrelation_FIX.c index d7a3944b..a4a579b1 100644 --- a/src/opus-1.0.2/silk/fixed/warped_autocorrelation_FIX.c +++ b/src/opus-1.1/silk/fixed/warped_autocorrelation_FIX.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/LPC_analysis_filter_FLP.c b/src/opus-1.1/silk/float/LPC_analysis_filter_FLP.c index 9845655b..cae89a0a 100644 --- a/src/opus-1.0.2/silk/float/LPC_analysis_filter_FLP.c +++ b/src/opus-1.1/silk/float/LPC_analysis_filter_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -40,7 +40,7 @@ POSSIBILITY OF SUCH DAMAGE. /************************************************/ /* 16th order LPC analysis filter, does not write first 16 samples */ -static inline void silk_LPC_analysis_filter16_FLP( +static OPUS_INLINE void silk_LPC_analysis_filter16_FLP( silk_float r_LPC[], /* O LPC residual signal */ const silk_float PredCoef[], /* I LPC coefficients */ const silk_float s[], /* I Input signal */ @@ -78,7 +78,7 @@ static inline void silk_LPC_analysis_filter16_FLP( } /* 12th order LPC analysis filter, does not write first 12 samples */ -static inline void silk_LPC_analysis_filter12_FLP( +static OPUS_INLINE void silk_LPC_analysis_filter12_FLP( silk_float r_LPC[], /* O LPC residual signal */ const silk_float PredCoef[], /* I LPC coefficients */ const silk_float s[], /* I Input signal */ @@ -112,7 +112,7 @@ static inline void silk_LPC_analysis_filter12_FLP( } /* 10th order LPC analysis filter, does not write first 10 samples */ -static inline void silk_LPC_analysis_filter10_FLP( +static OPUS_INLINE void silk_LPC_analysis_filter10_FLP( silk_float r_LPC[], /* O LPC residual signal */ const silk_float PredCoef[], /* I LPC coefficients */ const silk_float s[], /* I Input signal */ @@ -144,7 +144,7 @@ static inline void silk_LPC_analysis_filter10_FLP( } /* 8th order LPC analysis filter, does not write first 8 samples */ -static inline void silk_LPC_analysis_filter8_FLP( +static OPUS_INLINE void silk_LPC_analysis_filter8_FLP( silk_float r_LPC[], /* O LPC residual signal */ const silk_float PredCoef[], /* I LPC coefficients */ const silk_float s[], /* I Input signal */ @@ -174,7 +174,7 @@ static inline void silk_LPC_analysis_filter8_FLP( } /* 6th order LPC analysis filter, does not write first 6 samples */ -static inline void silk_LPC_analysis_filter6_FLP( +static OPUS_INLINE void silk_LPC_analysis_filter6_FLP( silk_float r_LPC[], /* O LPC residual signal */ const silk_float PredCoef[], /* I LPC coefficients */ const silk_float s[], /* I Input signal */ diff --git a/src/opus-1.0.2/silk/float/LPC_inv_pred_gain_FLP.c b/src/opus-1.1/silk/float/LPC_inv_pred_gain_FLP.c index 8645f77f..25178bac 100644 --- a/src/opus-1.0.2/silk/float/LPC_inv_pred_gain_FLP.c +++ b/src/opus-1.1/silk/float/LPC_inv_pred_gain_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/LTP_analysis_filter_FLP.c b/src/opus-1.1/silk/float/LTP_analysis_filter_FLP.c index d3a6a5ae..849b7c1c 100644 --- a/src/opus-1.0.2/silk/float/LTP_analysis_filter_FLP.c +++ b/src/opus-1.1/silk/float/LTP_analysis_filter_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/LTP_scale_ctrl_FLP.c b/src/opus-1.1/silk/float/LTP_scale_ctrl_FLP.c index f3f0c572..8dbe29d0 100644 --- a/src/opus-1.0.2/silk/float/LTP_scale_ctrl_FLP.c +++ b/src/opus-1.1/silk/float/LTP_scale_ctrl_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/SigProc_FLP.h b/src/opus-1.1/silk/float/SigProc_FLP.h index 036b46da..f0cb3733 100644 --- a/src/opus-1.0.2/silk/float/SigProc_FLP.h +++ b/src/opus-1.1/silk/float/SigProc_FLP.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -94,7 +94,8 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, const silk_float search_thres2, /* I Final threshold for lag candidates 0 - 1 */ const opus_int Fs_kHz, /* I sample frequency (kHz) */ const opus_int complexity, /* I Complexity setting, 0-2, where 2 is highest */ - const opus_int nb_subfr /* I Number of 5 ms subframes */ + const opus_int nb_subfr, /* I Number of 5 ms subframes */ + int arch /* I Run-time architecture */ ); void silk_insertion_sort_decreasing_FLP( @@ -153,19 +154,19 @@ double silk_energy_FLP( #define silk_abs_float( a ) ((silk_float)fabs(a)) /* sigmoid function */ -static inline silk_float silk_sigmoid( silk_float x ) +static OPUS_INLINE silk_float silk_sigmoid( silk_float x ) { return (silk_float)(1.0 / (1.0 + exp(-x))); } /* floating-point to integer conversion (rounding) */ -static inline opus_int32 silk_float2int( silk_float x ) +static OPUS_INLINE opus_int32 silk_float2int( silk_float x ) { return (opus_int32)float2int( x ); } /* floating-point to integer conversion (rounding) */ -static inline void silk_float2short_array( +static OPUS_INLINE void silk_float2short_array( opus_int16 *out, const silk_float *in, opus_int32 length @@ -178,7 +179,7 @@ static inline void silk_float2short_array( } /* integer to floating-point conversion */ -static inline void silk_short2float_array( +static OPUS_INLINE void silk_short2float_array( silk_float *out, const opus_int16 *in, opus_int32 length @@ -191,7 +192,7 @@ static inline void silk_short2float_array( } /* using log2() helps the fixed-point conversion */ -static inline silk_float silk_log2( double x ) +static OPUS_INLINE silk_float silk_log2( double x ) { return ( silk_float )( 3.32192809488736 * log10( x ) ); } diff --git a/src/opus-1.0.2/silk/float/apply_sine_window_FLP.c b/src/opus-1.1/silk/float/apply_sine_window_FLP.c index e06333f7..6aae57c0 100644 --- a/src/opus-1.0.2/silk/float/apply_sine_window_FLP.c +++ b/src/opus-1.1/silk/float/apply_sine_window_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/autocorrelation_FLP.c b/src/opus-1.1/silk/float/autocorrelation_FLP.c index 9ce709e2..8b8a9e65 100644 --- a/src/opus-1.0.2/silk/float/autocorrelation_FLP.c +++ b/src/opus-1.1/silk/float/autocorrelation_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/burg_modified_FLP.c b/src/opus-1.1/silk/float/burg_modified_FLP.c index 31c9b228..ea5dc25a 100644 --- a/src/opus-1.0.2/silk/float/burg_modified_FLP.c +++ b/src/opus-1.1/silk/float/burg_modified_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/bwexpander_FLP.c b/src/opus-1.1/silk/float/bwexpander_FLP.c index 59ca4eaf..d55a4d79 100644 --- a/src/opus-1.0.2/silk/float/bwexpander_FLP.c +++ b/src/opus-1.1/silk/float/bwexpander_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/corrMatrix_FLP.c b/src/opus-1.1/silk/float/corrMatrix_FLP.c index c59f73c3..eae6a1cf 100644 --- a/src/opus-1.0.2/silk/float/corrMatrix_FLP.c +++ b/src/opus-1.1/silk/float/corrMatrix_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/encode_frame_FLP.c b/src/opus-1.1/silk/float/encode_frame_FLP.c index 23260bc7..d54e2686 100644 --- a/src/opus-1.0.2/silk/float/encode_frame_FLP.c +++ b/src/opus-1.1/silk/float/encode_frame_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -33,7 +33,7 @@ POSSIBILITY OF SUCH DAMAGE. #include "tuning_parameters.h" /* Low Bitrate Redundancy (LBRR) encoding. Reuse all parameters but encode with lower bitrate */ -static inline void silk_LBRR_encode_FLP( +static OPUS_INLINE void silk_LBRR_encode_FLP( silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ const silk_float xfw[], /* I Input signal */ @@ -129,7 +129,7 @@ opus_int silk_encode_frame_FLP( /*****************************************/ /* Find pitch lags, initial LPC analysis */ /*****************************************/ - silk_find_pitch_lags_FLP( psEnc, &sEncCtrl, res_pitch, x_frame ); + silk_find_pitch_lags_FLP( psEnc, &sEncCtrl, res_pitch, x_frame, psEnc->sCmn.arch ); /************************/ /* Noise shape analysis */ @@ -294,10 +294,6 @@ opus_int silk_encode_frame_FLP( silk_memmove( psEnc->x_buf, &psEnc->x_buf[ psEnc->sCmn.frame_length ], ( psEnc->sCmn.ltp_mem_length + LA_SHAPE_MS * psEnc->sCmn.fs_kHz ) * sizeof( silk_float ) ); - /* Parameters needed for next frame */ - psEnc->sCmn.prevLag = sEncCtrl.pitchL[ psEnc->sCmn.nb_subfr - 1 ]; - psEnc->sCmn.prevSignalType = psEnc->sCmn.indices.signalType; - /* Exit without entropy coding */ if( psEnc->sCmn.prefillFlag ) { /* No payload */ @@ -305,6 +301,10 @@ opus_int silk_encode_frame_FLP( return ret; } + /* Parameters needed for next frame */ + psEnc->sCmn.prevLag = sEncCtrl.pitchL[ psEnc->sCmn.nb_subfr - 1 ]; + psEnc->sCmn.prevSignalType = psEnc->sCmn.indices.signalType; + /****************************************/ /* Finalize payload */ /****************************************/ @@ -316,7 +316,7 @@ opus_int silk_encode_frame_FLP( } /* Low-Bitrate Redundancy (LBRR) encoding. Reuse all parameters but encode excitation at lower bitrate */ -static inline void silk_LBRR_encode_FLP( +static OPUS_INLINE void silk_LBRR_encode_FLP( silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ const silk_float xfw[], /* I Input signal */ diff --git a/src/opus-1.0.2/silk/float/energy_FLP.c b/src/opus-1.1/silk/float/energy_FLP.c index e3eedf97..24b8179f 100644 --- a/src/opus-1.0.2/silk/float/energy_FLP.c +++ b/src/opus-1.1/silk/float/energy_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/find_LPC_FLP.c b/src/opus-1.1/silk/float/find_LPC_FLP.c index 66fa7dd4..61c1ad95 100644 --- a/src/opus-1.0.2/silk/float/find_LPC_FLP.c +++ b/src/opus-1.1/silk/float/find_LPC_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/find_LTP_FLP.c b/src/opus-1.1/silk/float/find_LTP_FLP.c index 0a3c71bb..72299960 100644 --- a/src/opus-1.0.2/silk/float/find_LTP_FLP.c +++ b/src/opus-1.1/silk/float/find_LTP_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/find_pitch_lags_FLP.c b/src/opus-1.1/silk/float/find_pitch_lags_FLP.c index 00862a6d..f3b22d25 100644 --- a/src/opus-1.0.2/silk/float/find_pitch_lags_FLP.c +++ b/src/opus-1.1/silk/float/find_pitch_lags_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -37,7 +37,8 @@ void silk_find_pitch_lags_FLP( silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ silk_float res[], /* O Residual */ - const silk_float x[] /* I Speech signal */ + const silk_float x[], /* I Speech signal */ + int arch /* I Run-time architecture */ ) { opus_int buf_len; @@ -116,7 +117,7 @@ void silk_find_pitch_lags_FLP( /*****************************************/ if( silk_pitch_analysis_core_FLP( res, psEncCtrl->pitchL, &psEnc->sCmn.indices.lagIndex, &psEnc->sCmn.indices.contourIndex, &psEnc->LTPCorr, psEnc->sCmn.prevLag, psEnc->sCmn.pitchEstimationThreshold_Q16 / 65536.0f, - thrhld, psEnc->sCmn.fs_kHz, psEnc->sCmn.pitchEstimationComplexity, psEnc->sCmn.nb_subfr ) == 0 ) + thrhld, psEnc->sCmn.fs_kHz, psEnc->sCmn.pitchEstimationComplexity, psEnc->sCmn.nb_subfr, arch ) == 0 ) { psEnc->sCmn.indices.signalType = TYPE_VOICED; } else { diff --git a/src/opus-1.0.2/silk/float/find_pred_coefs_FLP.c b/src/opus-1.1/silk/float/find_pred_coefs_FLP.c index 2156893a..ea2c6c43 100644 --- a/src/opus-1.0.2/silk/float/find_pred_coefs_FLP.c +++ b/src/opus-1.1/silk/float/find_pred_coefs_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -67,7 +67,7 @@ void silk_find_pred_coefs_FLP( /* Quantize LTP gain parameters */ silk_quant_LTP_gains_FLP( psEncCtrl->LTPCoef, psEnc->sCmn.indices.LTPIndex, &psEnc->sCmn.indices.PERIndex, - WLTP, psEnc->sCmn.mu_LTP_Q9, psEnc->sCmn.LTPQuantLowComplexity, psEnc->sCmn.nb_subfr ); + &psEnc->sCmn.sum_log_gain_Q7, WLTP, psEnc->sCmn.mu_LTP_Q9, psEnc->sCmn.LTPQuantLowComplexity, psEnc->sCmn.nb_subfr ); /* Control LTP scaling */ silk_LTP_scale_ctrl_FLP( psEnc, psEncCtrl, condCoding ); @@ -90,6 +90,7 @@ void silk_find_pred_coefs_FLP( } silk_memset( psEncCtrl->LTPCoef, 0, psEnc->sCmn.nb_subfr * LTP_ORDER * sizeof( silk_float ) ); psEncCtrl->LTPredCodGain = 0.0f; + psEnc->sCmn.sum_log_gain_Q7 = 0; } /* Limit on total predictive coding gain */ diff --git a/src/opus-1.0.2/silk/float/inner_product_FLP.c b/src/opus-1.1/silk/float/inner_product_FLP.c index 60823d6e..029c0129 100644 --- a/src/opus-1.0.2/silk/float/inner_product_FLP.c +++ b/src/opus-1.1/silk/float/inner_product_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/k2a_FLP.c b/src/opus-1.1/silk/float/k2a_FLP.c index 6f05d4b9..12af4e76 100644 --- a/src/opus-1.0.2/silk/float/k2a_FLP.c +++ b/src/opus-1.1/silk/float/k2a_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/levinsondurbin_FLP.c b/src/opus-1.1/silk/float/levinsondurbin_FLP.c index b4cd34e2..f0ba6069 100644 --- a/src/opus-1.0.2/silk/float/levinsondurbin_FLP.c +++ b/src/opus-1.1/silk/float/levinsondurbin_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/main_FLP.h b/src/opus-1.1/silk/float/main_FLP.h index 93455d4d..fb553b61 100644 --- a/src/opus-1.0.2/silk/float/main_FLP.h +++ b/src/opus-1.1/silk/float/main_FLP.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -71,7 +71,8 @@ opus_int silk_encode_frame_FLP( /* Initializes the Silk encoder state */ opus_int silk_init_encoder( - silk_encoder_state_FLP *psEnc /* I/O Encoder state FLP */ + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + int arch /* I Run-tim architecture */ ); /* Control the Silk encoder */ @@ -129,7 +130,8 @@ void silk_find_pitch_lags_FLP( silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ silk_float res[], /* O Residual */ - const silk_float x[] /* I Speech signal */ + const silk_float x[], /* I Speech signal */ + int arch /* I Run-time architecture */ ); /* Find LPC and LTP coefficients */ @@ -199,6 +201,7 @@ void silk_quant_LTP_gains_FLP( silk_float B[ MAX_NB_SUBFR * LTP_ORDER ], /* I/O (Un-)quantized LTP gains */ opus_int8 cbk_index[ MAX_NB_SUBFR ], /* O Codebook index */ opus_int8 *periodicity_index, /* O Periodicity index */ + opus_int32 *sum_log_gain_Q7, /* I/O Cumulative max prediction gain */ const silk_float W[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ], /* I Error weights */ const opus_int mu_Q10, /* I Mu value (R/D tradeoff) */ const opus_int lowComplexity, /* I Flag for low complexity */ diff --git a/src/opus-1.0.2/silk/float/noise_shape_analysis_FLP.c b/src/opus-1.1/silk/float/noise_shape_analysis_FLP.c index 33bfd20d..65f6ea58 100644 --- a/src/opus-1.0.2/silk/float/noise_shape_analysis_FLP.c +++ b/src/opus-1.1/silk/float/noise_shape_analysis_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -36,7 +36,7 @@ POSSIBILITY OF SUCH DAMAGE. /* non-warped frequency scale. (So that it can be implemented with a minimum-phase monic filter.) */ /* Note: A monic filter is one with the first coefficient equal to 1.0. In Silk we omit the first */ /* coefficient in an array of coefficients, for monic filters. */ -static inline silk_float warped_gain( +static OPUS_INLINE silk_float warped_gain( const silk_float *coefs, silk_float lambda, opus_int order @@ -54,7 +54,7 @@ static inline silk_float warped_gain( /* Convert warped filter coefficients to monic pseudo-warped coefficients and limit maximum */ /* amplitude of monic warped coefficients by using bandwidth expansion on the true coefficients */ -static inline void warped_true2monic_coefs( +static OPUS_INLINE void warped_true2monic_coefs( silk_float *coefs_syn, silk_float *coefs_ana, silk_float lambda, diff --git a/src/opus-1.0.2/silk/float/pitch_analysis_core_FLP.c b/src/opus-1.1/silk/float/pitch_analysis_core_FLP.c index fbff90c3..e58f041b 100644 --- a/src/opus-1.0.2/silk/float/pitch_analysis_core_FLP.c +++ b/src/opus-1.1/silk/float/pitch_analysis_core_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -35,9 +35,9 @@ POSSIBILITY OF SUCH DAMAGE. #include "SigProc_FLP.h" #include "SigProc_FIX.h" #include "pitch_est_defines.h" +#include "pitch.h" #define SCRATCH_SIZE 22 -#define eps 1.192092896e-07f /************************************************************/ /* Internally used functions */ @@ -48,7 +48,8 @@ static void silk_P_Ana_calc_corr_st3( opus_int start_lag, /* I start lag */ opus_int sf_length, /* I sub frame length */ opus_int nb_subfr, /* I number of subframes */ - opus_int complexity /* I Complexity setting */ + opus_int complexity, /* I Complexity setting */ + int arch /* I Run-time architecture */ ); static void silk_P_Ana_calc_energy_st3( @@ -74,7 +75,8 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, const silk_float search_thres2, /* I Final threshold for lag candidates 0 - 1 */ const opus_int Fs_kHz, /* I sample frequency (kHz) */ const opus_int complexity, /* I Complexity setting, 0-2, where 2 is highest */ - const opus_int nb_subfr /* I Number of 5 ms subframes */ + const opus_int nb_subfr, /* I Number of 5 ms subframes */ + int arch /* I Run-time architecture */ ) { opus_int i, k, d, j; @@ -85,6 +87,7 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, opus_int32 filt_state[ 6 ]; silk_float threshold, contour_bias; silk_float C[ PE_MAX_NB_SUBFR][ (PE_MAX_LAG >> 1) + 5 ]; + opus_val32 xcorr[ PE_MAX_LAG_MS * 4 - PE_MIN_LAG_MS * 4 + 1 ]; silk_float CC[ PE_NB_CBKS_STAGE2_EXT ]; const silk_float *target_ptr, *basis_ptr; double cross_corr, normalizer, energy, energy_tmp; @@ -129,8 +132,6 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, max_lag_4kHz = PE_MAX_LAG_MS * 4; max_lag_8kHz = PE_MAX_LAG_MS * 8 - 1; - silk_memset(C, 0, sizeof(silk_float) * nb_subfr * ((PE_MAX_LAG >> 1) + 5)); - /* Resample from input sampled at Fs_kHz to 8 kHz */ if( Fs_kHz == 16 ) { /* Resample to 16 -> 8 khz */ @@ -164,6 +165,7 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, /****************************************************************************** * FIRST STAGE, operating in 4 khz ******************************************************************************/ + silk_memset(C, 0, sizeof(silk_float) * nb_subfr * ((PE_MAX_LAG >> 1) + 5)); target_ptr = &frame_4kHz[ silk_LSHIFT( sf_length_4kHz, 2 ) ]; for( k = 0; k < nb_subfr >> 1; k++ ) { /* Check that we are within range of the array */ @@ -176,27 +178,31 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, silk_assert( basis_ptr >= frame_4kHz ); silk_assert( basis_ptr + sf_length_8kHz <= frame_4kHz + frame_length_4kHz ); + celt_pitch_xcorr( target_ptr, target_ptr-max_lag_4kHz, xcorr, sf_length_8kHz, max_lag_4kHz - min_lag_4kHz + 1, arch ); + /* Calculate first vector products before loop */ - cross_corr = silk_inner_product_FLP( target_ptr, basis_ptr, sf_length_8kHz ); - normalizer = silk_energy_FLP( basis_ptr, sf_length_8kHz ) + sf_length_8kHz * 4000.0f; + cross_corr = xcorr[ max_lag_4kHz - min_lag_4kHz ]; + normalizer = silk_energy_FLP( target_ptr, sf_length_8kHz ) + + silk_energy_FLP( basis_ptr, sf_length_8kHz ) + + sf_length_8kHz * 4000.0f; - C[ 0 ][ min_lag_4kHz ] += (silk_float)(cross_corr / sqrt(normalizer)); + C[ 0 ][ min_lag_4kHz ] += (silk_float)( 2 * cross_corr / normalizer ); /* From now on normalizer is computed recursively */ - for(d = min_lag_4kHz + 1; d <= max_lag_4kHz; d++) { + for( d = min_lag_4kHz + 1; d <= max_lag_4kHz; d++ ) { basis_ptr--; /* Check that we are within range of the array */ silk_assert( basis_ptr >= frame_4kHz ); silk_assert( basis_ptr + sf_length_8kHz <= frame_4kHz + frame_length_4kHz ); - cross_corr = silk_inner_product_FLP(target_ptr, basis_ptr, sf_length_8kHz); + cross_corr = xcorr[ max_lag_4kHz - d ]; /* Add contribution of new sample and remove contribution from oldest sample */ normalizer += basis_ptr[ 0 ] * (double)basis_ptr[ 0 ] - basis_ptr[ sf_length_8kHz ] * (double)basis_ptr[ sf_length_8kHz ]; - C[ 0 ][ d ] += (silk_float)(cross_corr / sqrt( normalizer )); + C[ 0 ][ d ] += (silk_float)( 2 * cross_corr / normalizer ); } /* Update target pointer */ target_ptr += sf_length_8kHz; @@ -214,13 +220,7 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, /* Escape if correlation is very low already here */ Cmax = C[ 0 ][ min_lag_4kHz ]; - target_ptr = &frame_4kHz[ silk_SMULBB( sf_length_4kHz, nb_subfr ) ]; - energy = 1000.0f; - for( i = 0; i < silk_LSHIFT( sf_length_4kHz, 2 ); i++ ) { - energy += target_ptr[i] * (double)target_ptr[i]; - } - threshold = Cmax * Cmax; - if( energy / 16.0f > threshold ) { + if( Cmax < 0.2f ) { silk_memset( pitch_out, 0, nb_subfr * sizeof( opus_int ) ); *LTPCorr = 0.0f; *lagIndex = 0; @@ -287,14 +287,14 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, target_ptr = &frame_8kHz[ PE_LTP_MEM_LENGTH_MS * 8 ]; } for( k = 0; k < nb_subfr; k++ ) { - energy_tmp = silk_energy_FLP( target_ptr, sf_length_8kHz ); + energy_tmp = silk_energy_FLP( target_ptr, sf_length_8kHz ) + 1.0; for( j = 0; j < length_d_comp; j++ ) { d = d_comp[ j ]; basis_ptr = target_ptr - d; cross_corr = silk_inner_product_FLP( basis_ptr, target_ptr, sf_length_8kHz ); - energy = silk_energy_FLP( basis_ptr, sf_length_8kHz ); if( cross_corr > 0.0f ) { - C[ k ][ d ] = (silk_float)(cross_corr * cross_corr / (energy * energy_tmp + eps)); + energy = silk_energy_FLP( basis_ptr, sf_length_8kHz ); + C[ k ][ d ] = (silk_float)( 2 * cross_corr / ( energy + energy_tmp ) ); } else { C[ k ][ d ] = 0.0f; } @@ -317,7 +317,7 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, } else if( Fs_kHz == 16 ) { prevLag = silk_RSHIFT( prevLag, 1 ); } - prevLag_log2 = silk_log2((silk_float)prevLag); + prevLag_log2 = silk_log2( (silk_float)prevLag ); } else { prevLag_log2 = 0; } @@ -356,23 +356,20 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, CBimax_new = i; } } - CCmax_new = silk_max_float(CCmax_new, 0.0f); /* To avoid taking square root of negative number later */ - CCmax_new_b = CCmax_new; /* Bias towards shorter lags */ - lag_log2 = silk_log2((silk_float)d); - CCmax_new_b -= PE_SHORTLAG_BIAS * nb_subfr * lag_log2; + lag_log2 = silk_log2( (silk_float)d ); + CCmax_new_b = CCmax_new - PE_SHORTLAG_BIAS * nb_subfr * lag_log2; /* Bias towards previous lag */ if( prevLag > 0 ) { delta_lag_log2_sqr = lag_log2 - prevLag_log2; delta_lag_log2_sqr *= delta_lag_log2_sqr; - CCmax_new_b -= PE_PREVLAG_BIAS * nb_subfr * (*LTPCorr) * delta_lag_log2_sqr / (delta_lag_log2_sqr + 0.5f); + CCmax_new_b -= PE_PREVLAG_BIAS * nb_subfr * (*LTPCorr) * delta_lag_log2_sqr / ( delta_lag_log2_sqr + 0.5f ); } - if( CCmax_new_b > CCmax_b && /* Find maximum biased correlation */ - CCmax_new > nb_subfr * search_thres2 * search_thres2 && /* Correlation needs to be high enough to be voiced */ - silk_CB_lags_stage2[ 0 ][ CBimax_new ] <= min_lag_8kHz /* Lag must be in range */ + if( CCmax_new_b > CCmax_b && /* Find maximum biased correlation */ + CCmax_new > nb_subfr * search_thres2 /* Correlation needs to be high enough to be voiced */ ) { CCmax_b = CCmax_new_b; CCmax = CCmax_new; @@ -390,6 +387,10 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, return 1; } + /* Output normalized correlation */ + *LTPCorr = (silk_float)( CCmax / nb_subfr ); + silk_assert( *LTPCorr >= 0.0f ); + if( Fs_kHz > 8 ) { /* Search in original signal */ @@ -406,13 +407,11 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, end_lag = silk_min_int( lag + 2, max_lag ); lag_new = lag; /* to avoid undefined lag */ CBimax = 0; /* to avoid undefined lag */ - silk_assert( CCmax >= 0.0f ); - *LTPCorr = (silk_float)sqrt( CCmax / nb_subfr ); /* Output normalized correlation */ CCmax = -1000.0f; /* Calculate the correlations and energies needed in stage 3 */ - silk_P_Ana_calc_corr_st3( cross_corr_st3, frame, start_lag, sf_length, nb_subfr, complexity ); + silk_P_Ana_calc_corr_st3( cross_corr_st3, frame, start_lag, sf_length, nb_subfr, complexity, arch ); silk_P_Ana_calc_energy_st3( energies_st3, frame, start_lag, sf_length, nb_subfr, complexity ); lag_counter = 0; @@ -430,25 +429,25 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; } + target_ptr = &frame[ PE_LTP_MEM_LENGTH_MS * Fs_kHz ]; + energy_tmp = silk_energy_FLP( target_ptr, nb_subfr * sf_length ) + 1.0; for( d = start_lag; d <= end_lag; d++ ) { for( j = 0; j < nb_cbk_search; j++ ) { cross_corr = 0.0; - energy = eps; + energy = energy_tmp; for( k = 0; k < nb_subfr; k++ ) { - energy += energies_st3[ k ][ j ][ lag_counter ]; cross_corr += cross_corr_st3[ k ][ j ][ lag_counter ]; + energy += energies_st3[ k ][ j ][ lag_counter ]; } if( cross_corr > 0.0 ) { - CCmax_new = (silk_float)(cross_corr * cross_corr / energy); + CCmax_new = (silk_float)( 2 * cross_corr / energy ); /* Reduce depending on flatness of contour */ CCmax_new *= 1.0f - contour_bias * j; } else { CCmax_new = 0.0f; } - if( CCmax_new > CCmax && - ( d + (opus_int)silk_CB_lags_stage3[ 0 ][ j ] ) <= max_lag - ) { + if( CCmax_new > CCmax && ( d + (opus_int)silk_CB_lags_stage3[ 0 ][ j ] ) <= max_lag ) { CCmax = CCmax_new; lag_new = d; CBimax = j; @@ -464,12 +463,10 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, *lagIndex = (opus_int16)( lag_new - min_lag ); *contourIndex = (opus_int8)CBimax; } else { /* Fs_kHz == 8 */ - /* Save Lags and correlation */ - silk_assert( CCmax >= 0.0f ); - *LTPCorr = (silk_float)sqrt( CCmax / nb_subfr ); /* Output normalized correlation */ + /* Save Lags */ for( k = 0; k < nb_subfr; k++ ) { pitch_out[ k ] = lag + matrix_ptr( Lag_CB_ptr, k, CBimax, cbk_size ); - pitch_out[ k ] = silk_LIMIT( pitch_out[ k ], min_lag_8kHz, PE_MAX_LAG_MS * Fs_kHz ); + pitch_out[ k ] = silk_LIMIT( pitch_out[ k ], min_lag_8kHz, PE_MAX_LAG_MS * 8 ); } *lagIndex = (opus_int16)( lag - min_lag_8kHz ); *contourIndex = (opus_int8)CBimax; @@ -479,32 +476,34 @@ opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, return 0; } +/*********************************************************************** + * Calculates the correlations used in stage 3 search. In order to cover + * the whole lag codebook for all the searched offset lags (lag +- 2), + * the following correlations are needed in each sub frame: + * + * sf1: lag range [-8,...,7] total 16 correlations + * sf2: lag range [-4,...,4] total 9 correlations + * sf3: lag range [-3,....4] total 8 correltions + * sf4: lag range [-6,....8] total 15 correlations + * + * In total 48 correlations. The direct implementation computed in worst + * case 4*12*5 = 240 correlations, but more likely around 120. + ***********************************************************************/ static void silk_P_Ana_calc_corr_st3( silk_float cross_corr_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ], /* O 3 DIM correlation array */ const silk_float frame[], /* I vector to correlate */ opus_int start_lag, /* I start lag */ opus_int sf_length, /* I sub frame length */ opus_int nb_subfr, /* I number of subframes */ - opus_int complexity /* I Complexity setting */ + opus_int complexity, /* I Complexity setting */ + int arch /* I Run-time architecture */ ) - /*********************************************************************** - Calculates the correlations used in stage 3 search. In order to cover - the whole lag codebook for all the searched offset lags (lag +- 2), - the following correlations are needed in each sub frame: - - sf1: lag range [-8,...,7] total 16 correlations - sf2: lag range [-4,...,4] total 9 correlations - sf3: lag range [-3,....4] total 8 correltions - sf4: lag range [-6,....8] total 15 correlations - - In total 48 correlations. The direct implementation computed in worst case - 4*12*5 = 240 correlations, but more likely around 120. - **********************************************************************/ { - const silk_float *target_ptr, *basis_ptr; + const silk_float *target_ptr; opus_int i, j, k, lag_counter, lag_low, lag_high; opus_int nb_cbk_search, delta, idx, cbk_size; silk_float scratch_mem[ SCRATCH_SIZE ]; + opus_val32 xcorr[ SCRATCH_SIZE ]; const opus_int8 *Lag_range_ptr, *Lag_CB_ptr; silk_assert( complexity >= SILK_PE_MIN_COMPLEX ); @@ -530,10 +529,11 @@ static void silk_P_Ana_calc_corr_st3( /* Calculate the correlations for each subframe */ lag_low = matrix_ptr( Lag_range_ptr, k, 0, 2 ); lag_high = matrix_ptr( Lag_range_ptr, k, 1, 2 ); + silk_assert(lag_high-lag_low+1 <= SCRATCH_SIZE); + celt_pitch_xcorr( target_ptr, target_ptr - start_lag - lag_high, xcorr, sf_length, lag_high - lag_low + 1, arch ); for( j = lag_low; j <= lag_high; j++ ) { - basis_ptr = target_ptr - ( start_lag + j ); silk_assert( lag_counter < SCRATCH_SIZE ); - scratch_mem[ lag_counter ] = (silk_float)silk_inner_product_FLP( target_ptr, basis_ptr, sf_length ); + scratch_mem[ lag_counter ] = xcorr[ lag_high - j ]; lag_counter++; } @@ -552,6 +552,10 @@ static void silk_P_Ana_calc_corr_st3( } } +/********************************************************************/ +/* Calculate the energies for first two subframes. The energies are */ +/* calculated recursively. */ +/********************************************************************/ static void silk_P_Ana_calc_energy_st3( silk_float energies_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ], /* O 3 DIM correlation array */ const silk_float frame[], /* I vector to correlate */ @@ -560,10 +564,6 @@ static void silk_P_Ana_calc_energy_st3( opus_int nb_subfr, /* I number of subframes */ opus_int complexity /* I Complexity setting */ ) -/**************************************************************** -Calculate the energies for first two subframes. The energies are -calculated recursively. -****************************************************************/ { const silk_float *target_ptr, *basis_ptr; double energy; diff --git a/src/opus-1.0.2/silk/float/prefilter_FLP.c b/src/opus-1.1/silk/float/prefilter_FLP.c index d6c84398..8bc32fb4 100644 --- a/src/opus-1.0.2/silk/float/prefilter_FLP.c +++ b/src/opus-1.1/silk/float/prefilter_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -35,7 +35,7 @@ POSSIBILITY OF SUCH DAMAGE. /* * Prefilter for finding Quantizer input signal */ -static inline void silk_prefilt_FLP( +static OPUS_INLINE void silk_prefilt_FLP( silk_prefilter_state_FLP *P, /* I/O state */ silk_float st_res[], /* I */ silk_float xw[], /* O */ @@ -153,7 +153,7 @@ void silk_prefilter_FLP( /* * Prefilter for finding Quantizer input signal */ -static inline void silk_prefilt_FLP( +static OPUS_INLINE void silk_prefilt_FLP( silk_prefilter_state_FLP *P, /* I/O state */ silk_float st_res[], /* I */ silk_float xw[], /* O */ diff --git a/src/opus-1.0.2/silk/float/process_gains_FLP.c b/src/opus-1.1/silk/float/process_gains_FLP.c index d572a4cd..c0da0dae 100644 --- a/src/opus-1.0.2/silk/float/process_gains_FLP.c +++ b/src/opus-1.1/silk/float/process_gains_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/regularize_correlations_FLP.c b/src/opus-1.1/silk/float/regularize_correlations_FLP.c index f5684637..df461260 100644 --- a/src/opus-1.0.2/silk/float/regularize_correlations_FLP.c +++ b/src/opus-1.1/silk/float/regularize_correlations_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/residual_energy_FLP.c b/src/opus-1.1/silk/float/residual_energy_FLP.c index e65457ab..b2e03a86 100644 --- a/src/opus-1.0.2/silk/float/residual_energy_FLP.c +++ b/src/opus-1.1/silk/float/residual_energy_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/scale_copy_vector_FLP.c b/src/opus-1.1/silk/float/scale_copy_vector_FLP.c index 988795a6..20db32b3 100644 --- a/src/opus-1.0.2/silk/float/scale_copy_vector_FLP.c +++ b/src/opus-1.1/silk/float/scale_copy_vector_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/scale_vector_FLP.c b/src/opus-1.1/silk/float/scale_vector_FLP.c index 387eb4ba..108fdcbe 100644 --- a/src/opus-1.0.2/silk/float/scale_vector_FLP.c +++ b/src/opus-1.1/silk/float/scale_vector_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/schur_FLP.c b/src/opus-1.1/silk/float/schur_FLP.c index 90c3a18b..ee436f83 100644 --- a/src/opus-1.0.2/silk/float/schur_FLP.c +++ b/src/opus-1.1/silk/float/schur_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/solve_LS_FLP.c b/src/opus-1.1/silk/float/solve_LS_FLP.c index a4bb0525..7c90d665 100644 --- a/src/opus-1.0.2/silk/float/solve_LS_FLP.c +++ b/src/opus-1.1/silk/float/solve_LS_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -37,7 +37,7 @@ POSSIBILITY OF SUCH DAMAGE. * Matrix D (only the diagonal elements returned in a vector)such that * the symmetric matric A is given by A = L*D*L'. **********************************************************************/ -static inline void silk_LDL_FLP( +static OPUS_INLINE void silk_LDL_FLP( silk_float *A, /* I/O Pointer to Symetric Square Matrix */ opus_int M, /* I Size of Matrix */ silk_float *L, /* I/O Pointer to Square Upper triangular Matrix */ @@ -48,7 +48,7 @@ static inline void silk_LDL_FLP( * Function to solve linear equation Ax = b, when A is a MxM lower * triangular matrix, with ones on the diagonal. **********************************************************************/ -static inline void silk_SolveWithLowerTriangularWdiagOnes_FLP( +static OPUS_INLINE void silk_SolveWithLowerTriangularWdiagOnes_FLP( const silk_float *L, /* I Pointer to Lower Triangular Matrix */ opus_int M, /* I Dim of Matrix equation */ const silk_float *b, /* I b Vector */ @@ -59,7 +59,7 @@ static inline void silk_SolveWithLowerTriangularWdiagOnes_FLP( * Function to solve linear equation (A^T)x = b, when A is a MxM lower * triangular, with ones on the diagonal. (ie then A^T is upper triangular) **********************************************************************/ -static inline void silk_SolveWithUpperTriangularFromLowerWdiagOnes_FLP( +static OPUS_INLINE void silk_SolveWithUpperTriangularFromLowerWdiagOnes_FLP( const silk_float *L, /* I Pointer to Lower Triangular Matrix */ opus_int M, /* I Dim of Matrix equation */ const silk_float *b, /* I b Vector */ @@ -109,7 +109,7 @@ void silk_solve_LDL_FLP( silk_SolveWithUpperTriangularFromLowerWdiagOnes_FLP( &L[ 0 ][ 0 ], M, T, x ); } -static inline void silk_SolveWithUpperTriangularFromLowerWdiagOnes_FLP( +static OPUS_INLINE void silk_SolveWithUpperTriangularFromLowerWdiagOnes_FLP( const silk_float *L, /* I Pointer to Lower Triangular Matrix */ opus_int M, /* I Dim of Matrix equation */ const silk_float *b, /* I b Vector */ @@ -131,7 +131,7 @@ static inline void silk_SolveWithUpperTriangularFromLowerWdiagOnes_FLP( } } -static inline void silk_SolveWithLowerTriangularWdiagOnes_FLP( +static OPUS_INLINE void silk_SolveWithLowerTriangularWdiagOnes_FLP( const silk_float *L, /* I Pointer to Lower Triangular Matrix */ opus_int M, /* I Dim of Matrix equation */ const silk_float *b, /* I b Vector */ @@ -153,7 +153,7 @@ static inline void silk_SolveWithLowerTriangularWdiagOnes_FLP( } } -static inline void silk_LDL_FLP( +static OPUS_INLINE void silk_LDL_FLP( silk_float *A, /* I/O Pointer to Symetric Square Matrix */ opus_int M, /* I Size of Matrix */ silk_float *L, /* I/O Pointer to Square Upper triangular Matrix */ diff --git a/src/opus-1.0.2/silk/float/sort_FLP.c b/src/opus-1.1/silk/float/sort_FLP.c index e290c380..f08d7592 100644 --- a/src/opus-1.0.2/silk/float/sort_FLP.c +++ b/src/opus-1.1/silk/float/sort_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/structs_FLP.h b/src/opus-1.1/silk/float/structs_FLP.h index c71e7bc3..bb529e71 100644 --- a/src/opus-1.0.2/silk/float/structs_FLP.h +++ b/src/opus-1.1/silk/float/structs_FLP.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/warped_autocorrelation_FLP.c b/src/opus-1.1/silk/float/warped_autocorrelation_FLP.c index e9ecc2a3..542414f4 100644 --- a/src/opus-1.0.2/silk/float/warped_autocorrelation_FLP.c +++ b/src/opus-1.1/silk/float/warped_autocorrelation_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/float/wrappers_FLP.c b/src/opus-1.1/silk/float/wrappers_FLP.c index 4259e90e..350599b2 100644 --- a/src/opus-1.0.2/silk/float/wrappers_FLP.c +++ b/src/opus-1.1/silk/float/wrappers_FLP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -175,6 +175,7 @@ void silk_quant_LTP_gains_FLP( silk_float B[ MAX_NB_SUBFR * LTP_ORDER ], /* I/O (Un-)quantized LTP gains */ opus_int8 cbk_index[ MAX_NB_SUBFR ], /* O Codebook index */ opus_int8 *periodicity_index, /* O Periodicity index */ + opus_int32 *sum_log_gain_Q7, /* I/O Cumulative max prediction gain */ const silk_float W[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ], /* I Error weights */ const opus_int mu_Q10, /* I Mu value (R/D tradeoff) */ const opus_int lowComplexity, /* I Flag for low complexity */ @@ -192,7 +193,7 @@ void silk_quant_LTP_gains_FLP( W_Q18[ i ] = (opus_int32)silk_float2int( W[ i ] * 262144.0f ); } - silk_quant_LTP_gains( B_Q14, cbk_index, periodicity_index, W_Q18, mu_Q10, lowComplexity, nb_subfr ); + silk_quant_LTP_gains( B_Q14, cbk_index, periodicity_index, sum_log_gain_Q7, W_Q18, mu_Q10, lowComplexity, nb_subfr ); for( i = 0; i < nb_subfr * LTP_ORDER; i++ ) { B[ i ] = (silk_float)B_Q14[ i ] * ( 1.0f / 16384.0f ); diff --git a/src/opus-1.0.2/silk/gain_quant.c b/src/opus-1.1/silk/gain_quant.c index b2f73735..64ccd061 100644 --- a/src/opus-1.0.2/silk/gain_quant.c +++ b/src/opus-1.1/silk/gain_quant.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/init_decoder.c b/src/opus-1.1/silk/init_decoder.c index 47834890..f887c678 100644 --- a/src/opus-1.0.2/silk/init_decoder.c +++ b/src/opus-1.1/silk/init_decoder.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/init_encoder.c b/src/opus-1.1/silk/init_encoder.c index fe3fe968..65995c33 100644 --- a/src/opus-1.0.2/silk/init_encoder.c +++ b/src/opus-1.1/silk/init_encoder.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -34,12 +34,14 @@ POSSIBILITY OF SUCH DAMAGE. #include "main_FLP.h" #endif #include "tuning_parameters.h" +#include "cpu_support.h" /*********************************/ /* Initialize Silk Encoder state */ /*********************************/ opus_int silk_init_encoder( - silk_encoder_state_Fxx *psEnc /* I/O Pointer to Silk FIX encoder state */ + silk_encoder_state_Fxx *psEnc, /* I/O Pointer to Silk FIX encoder state */ + int arch /* I Run-time architecture */ ) { opus_int ret = 0; @@ -47,6 +49,8 @@ opus_int silk_init_encoder( /* Clear the entire encoder state */ silk_memset( psEnc, 0, sizeof( silk_encoder_state_Fxx ) ); + psEnc->sCmn.arch = arch; + psEnc->sCmn.variable_HP_smth1_Q15 = silk_LSHIFT( silk_lin2log( SILK_FIX_CONST( VARIABLE_HP_MIN_CUTOFF_HZ, 16 ) ) - ( 16 << 7 ), 8 ); psEnc->sCmn.variable_HP_smth2_Q15 = psEnc->sCmn.variable_HP_smth1_Q15; diff --git a/src/opus-1.0.2/silk/inner_prod_aligned.c b/src/opus-1.1/silk/inner_prod_aligned.c index fe20a2b1..257ae9e0 100644 --- a/src/opus-1.0.2/silk/inner_prod_aligned.c +++ b/src/opus-1.1/silk/inner_prod_aligned.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/interpolate.c b/src/opus-1.1/silk/interpolate.c index 226488b9..1bd8ca4d 100644 --- a/src/opus-1.0.2/silk/interpolate.c +++ b/src/opus-1.1/silk/interpolate.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/lin2log.c b/src/opus-1.1/silk/lin2log.c index 212b670d..d4fe5153 100644 --- a/src/opus-1.0.2/silk/lin2log.c +++ b/src/opus-1.1/silk/lin2log.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/log2lin.c b/src/opus-1.1/silk/log2lin.c index 33a19ad1..a692e009 100644 --- a/src/opus-1.0.2/silk/log2lin.c +++ b/src/opus-1.1/silk/log2lin.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -41,7 +41,9 @@ opus_int32 silk_log2lin( if( inLog_Q7 < 0 ) { return 0; - } + } else if ( inLog_Q7 >= 3967 ) { + return silk_int32_MAX; + } out = silk_LSHIFT( 1, silk_RSHIFT( inLog_Q7, 7 ) ); frac_Q7 = inLog_Q7 & 0x7F; diff --git a/src/opus-1.0.2/silk/macros.h b/src/opus-1.1/silk/macros.h index 2612fc7a..a84e5a5d 100644 --- a/src/opus-1.0.2/silk/macros.h +++ b/src/opus-1.1/silk/macros.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -32,7 +32,10 @@ POSSIBILITY OF SUCH DAMAGE. #include "config.h" #endif -/* This is an inline header file for general platform. */ +#include "opus_types.h" +#include "opus_defines.h" + +/* This is an OPUS_INLINE header file for general platform. */ /* (a32 * (opus_int32)((opus_int16)(b32))) >> 16 output have to be 32bit int */ #define silk_SMULWB(a32, b32) ((((a32) >> 16) * (opus_int32)((opus_int16)(b32))) + ((((a32) & 0x0000FFFF) * (opus_int32)((opus_int16)(b32))) >> 16)) @@ -76,59 +79,36 @@ POSSIBILITY OF SUCH DAMAGE. (( (a) & ((b)^0x80000000) & 0x80000000) ? silk_int32_MIN : (a)-(b)) : \ ((((a)^0x80000000) & (b) & 0x80000000) ? silk_int32_MAX : (a)-(b)) ) -static inline opus_int32 silk_CLZ16(opus_int16 in16) +#include "ecintrin.h" + +static OPUS_INLINE opus_int32 silk_CLZ16(opus_int16 in16) { - opus_int32 out32 = 0; - if( in16 == 0 ) { - return 16; - } - /* test nibbles */ - if( in16 & 0xFF00 ) { - if( in16 & 0xF000 ) { - in16 >>= 12; - } else { - out32 += 4; - in16 >>= 8; - } - } else { - if( in16 & 0xFFF0 ) { - out32 += 8; - in16 >>= 4; - } else { - out32 += 12; - } - } - /* test bits and return */ - if( in16 & 0xC ) { - if( in16 & 0x8 ) - return out32 + 0; - else - return out32 + 1; - } else { - if( in16 & 0xE ) - return out32 + 2; - else - return out32 + 3; - } + return 32 - EC_ILOG(in16<<16|0x8000); } -static inline opus_int32 silk_CLZ32(opus_int32 in32) +static OPUS_INLINE opus_int32 silk_CLZ32(opus_int32 in32) { - /* test highest 16 bits and convert to opus_int16 */ - if( in32 & 0xFFFF0000 ) { - return silk_CLZ16((opus_int16)(in32 >> 16)); - } else { - return silk_CLZ16((opus_int16)in32) + 16; - } + return in32 ? 32 - EC_ILOG(in32) : 32; } /* Row based */ -#define matrix_ptr(Matrix_base_adr, row, column, N) *(Matrix_base_adr + ((row)*(N)+(column))) -#define matrix_adr(Matrix_base_adr, row, column, N) (Matrix_base_adr + ((row)*(N)+(column))) +#define matrix_ptr(Matrix_base_adr, row, column, N) \ + (*((Matrix_base_adr) + ((row)*(N)+(column)))) +#define matrix_adr(Matrix_base_adr, row, column, N) \ + ((Matrix_base_adr) + ((row)*(N)+(column))) /* Column based */ #ifndef matrix_c_ptr -# define matrix_c_ptr(Matrix_base_adr, row, column, M) *(Matrix_base_adr + ((row)+(M)*(column))) +# define matrix_c_ptr(Matrix_base_adr, row, column, M) \ + (*((Matrix_base_adr) + ((row)+(M)*(column)))) +#endif + +#ifdef OPUS_ARM_INLINE_ASM +#include "arm/macros_armv4.h" +#endif + +#ifdef OPUS_ARM_INLINE_EDSP +#include "arm/macros_armv5e.h" #endif #endif /* SILK_MACROS_H */ diff --git a/src/opus-1.0.2/silk/main.h b/src/opus-1.1/silk/main.h index 32675f69..2bdf8978 100644 --- a/src/opus-1.0.2/silk/main.h +++ b/src/opus-1.1/silk/main.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -204,6 +204,7 @@ void silk_quant_LTP_gains( opus_int16 B_Q14[ MAX_NB_SUBFR * LTP_ORDER ], /* I/O (un)quantized LTP gains */ opus_int8 cbk_index[ MAX_NB_SUBFR ], /* O Codebook Index */ opus_int8 *periodicity_index, /* O Periodicity Index */ + opus_int32 *sum_gain_dB_Q7, /* I/O Cumulative max prediction gain */ const opus_int32 W_Q18[ MAX_NB_SUBFR*LTP_ORDER*LTP_ORDER ], /* I Error Weights in Q18 */ opus_int mu_Q9, /* I Mu value (R/D tradeoff) */ opus_int lowComplexity, /* I Flag for low complexity */ @@ -214,11 +215,14 @@ void silk_quant_LTP_gains( void silk_VQ_WMat_EC( opus_int8 *ind, /* O index of best codebook vector */ opus_int32 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ + opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ const opus_int16 *in_Q14, /* I input vector to be quantized */ const opus_int32 *W_Q18, /* I weighting matrix */ const opus_int8 *cb_Q7, /* I codebook */ + const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ const opus_int mu_Q9, /* I tradeoff betw. weighted error and rate */ + const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ opus_int L /* I number of vectors in codebook */ ); diff --git a/src/opus-1.0.2/silk/pitch_est_defines.h b/src/opus-1.1/silk/pitch_est_defines.h index 0b6770eb..e1e4b5d7 100644 --- a/src/opus-1.0.2/silk/pitch_est_defines.h +++ b/src/opus-1.1/silk/pitch_est_defines.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/pitch_est_tables.c b/src/opus-1.1/silk/pitch_est_tables.c index 7b139ed2..81a8baca 100644 --- a/src/opus-1.0.2/silk/pitch_est_tables.c +++ b/src/opus-1.1/silk/pitch_est_tables.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/process_NLSFs.c b/src/opus-1.1/silk/process_NLSFs.c index 34ce7914..c27cf030 100644 --- a/src/opus-1.0.2/silk/process_NLSFs.c +++ b/src/opus-1.1/silk/process_NLSFs.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/quant_LTP_gains.c b/src/opus-1.1/silk/quant_LTP_gains.c index f73c0f50..fd0870da 100644 --- a/src/opus-1.0.2/silk/quant_LTP_gains.c +++ b/src/opus-1.1/silk/quant_LTP_gains.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,11 +30,13 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main.h" +#include "tuning_parameters.h" void silk_quant_LTP_gains( opus_int16 B_Q14[ MAX_NB_SUBFR * LTP_ORDER ], /* I/O (un)quantized LTP gains */ opus_int8 cbk_index[ MAX_NB_SUBFR ], /* O Codebook Index */ opus_int8 *periodicity_index, /* O Periodicity Index */ + opus_int32 *sum_log_gain_Q7, /* I/O Cumulative max prediction gain */ const opus_int32 W_Q18[ MAX_NB_SUBFR*LTP_ORDER*LTP_ORDER ], /* I Error Weights in Q18 */ opus_int mu_Q9, /* I Mu value (R/D tradeoff) */ opus_int lowComplexity, /* I Flag for low complexity */ @@ -45,18 +47,26 @@ void silk_quant_LTP_gains( opus_int8 temp_idx[ MAX_NB_SUBFR ]; const opus_uint8 *cl_ptr_Q5; const opus_int8 *cbk_ptr_Q7; + const opus_uint8 *cbk_gain_ptr_Q7; const opus_int16 *b_Q14_ptr; const opus_int32 *W_Q18_ptr; opus_int32 rate_dist_Q14_subfr, rate_dist_Q14, min_rate_dist_Q14; + opus_int32 sum_log_gain_tmp_Q7, best_sum_log_gain_Q7, max_gain_Q7, gain_Q7; /***************************************************/ /* iterate over different codebooks with different */ /* rates/distortions, and choose best */ /***************************************************/ min_rate_dist_Q14 = silk_int32_MAX; + best_sum_log_gain_Q7 = 0; for( k = 0; k < 3; k++ ) { + /* Safety margin for pitch gain control, to take into account factors + such as state rescaling/rewhitening. */ + opus_int32 gain_safety = SILK_FIX_CONST( 0.4, 7 ); + cl_ptr_Q5 = silk_LTP_gain_BITS_Q5_ptrs[ k ]; cbk_ptr_Q7 = silk_LTP_vq_ptrs_Q7[ k ]; + cbk_gain_ptr_Q7 = silk_LTP_vq_gain_ptrs_Q7[ k ]; cbk_size = silk_LTP_vq_sizes[ k ]; /* Set up pointer to first subframe */ @@ -64,19 +74,28 @@ void silk_quant_LTP_gains( b_Q14_ptr = B_Q14; rate_dist_Q14 = 0; + sum_log_gain_tmp_Q7 = *sum_log_gain_Q7; for( j = 0; j < nb_subfr; j++ ) { + max_gain_Q7 = silk_log2lin( ( SILK_FIX_CONST( MAX_SUM_LOG_GAIN_DB / 6.0, 7 ) - sum_log_gain_tmp_Q7 ) + + SILK_FIX_CONST( 7, 7 ) ) - gain_safety; + silk_VQ_WMat_EC( &temp_idx[ j ], /* O index of best codebook vector */ &rate_dist_Q14_subfr, /* O best weighted quantization error + mu * rate */ + &gain_Q7, /* O sum of absolute LTP coefficients */ b_Q14_ptr, /* I input vector to be quantized */ W_Q18_ptr, /* I weighting matrix */ cbk_ptr_Q7, /* I codebook */ + cbk_gain_ptr_Q7, /* I codebook effective gains */ cl_ptr_Q5, /* I code length for each codebook vector */ mu_Q9, /* I tradeoff between weighted error and rate */ + max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ cbk_size /* I number of vectors in codebook */ ); rate_dist_Q14 = silk_ADD_POS_SAT32( rate_dist_Q14, rate_dist_Q14_subfr ); + sum_log_gain_tmp_Q7 = silk_max(0, sum_log_gain_tmp_Q7 + + silk_lin2log( gain_safety + gain_Q7 ) - SILK_FIX_CONST( 7, 7 )); b_Q14_ptr += LTP_ORDER; W_Q18_ptr += LTP_ORDER * LTP_ORDER; @@ -89,6 +108,7 @@ void silk_quant_LTP_gains( min_rate_dist_Q14 = rate_dist_Q14; *periodicity_index = (opus_int8)k; silk_memcpy( cbk_index, temp_idx, nb_subfr * sizeof( opus_int8 ) ); + best_sum_log_gain_Q7 = sum_log_gain_tmp_Q7; } /* Break early in low-complexity mode if rate distortion is below threshold */ @@ -103,5 +123,6 @@ void silk_quant_LTP_gains( B_Q14[ j * LTP_ORDER + k ] = silk_LSHIFT( cbk_ptr_Q7[ cbk_index[ j ] * LTP_ORDER + k ], 7 ); } } + *sum_log_gain_Q7 = best_sum_log_gain_Q7; } diff --git a/src/opus-1.0.2/silk/resampler.c b/src/opus-1.1/silk/resampler.c index 7e58332f..374fbb37 100644 --- a/src/opus-1.0.2/silk/resampler.c +++ b/src/opus-1.1/silk/resampler.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/resampler_down2.c b/src/opus-1.1/silk/resampler_down2.c index 21d11992..cec36346 100644 --- a/src/opus-1.0.2/silk/resampler_down2.c +++ b/src/opus-1.1/silk/resampler_down2.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -35,8 +35,8 @@ POSSIBILITY OF SUCH DAMAGE. /* Downsample by a factor 2 */ void silk_resampler_down2( opus_int32 *S, /* I/O State vector [ 2 ] */ - opus_int16 *out, /* O Output signal [ len ] */ - const opus_int16 *in, /* I Input signal [ floor(len/2) ] */ + opus_int16 *out, /* O Output signal [ floor(len/2) ] */ + const opus_int16 *in, /* I Input signal [ len ] */ opus_int32 inLen /* I Number of input samples */ ) { diff --git a/src/opus-1.0.2/silk/resampler_down2_3.c b/src/opus-1.1/silk/resampler_down2_3.c index fe5b671d..4342614d 100644 --- a/src/opus-1.0.2/silk/resampler_down2_3.c +++ b/src/opus-1.1/silk/resampler_down2_3.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -31,6 +31,7 @@ POSSIBILITY OF SUCH DAMAGE. #include "SigProc_FIX.h" #include "resampler_private.h" +#include "stack_alloc.h" #define ORDER_FIR 4 @@ -43,8 +44,11 @@ void silk_resampler_down2_3( ) { opus_int32 nSamplesIn, counter, res_Q6; - opus_int32 buf[ RESAMPLER_MAX_BATCH_SIZE_IN + ORDER_FIR ]; + VARDECL( opus_int32, buf ); opus_int32 *buf_ptr; + SAVE_STACK; + + ALLOC( buf, RESAMPLER_MAX_BATCH_SIZE_IN + ORDER_FIR, opus_int32 ); /* Copy buffered samples to start of buffer */ silk_memcpy( buf, S, ORDER_FIR * sizeof( opus_int32 ) ); @@ -95,4 +99,5 @@ void silk_resampler_down2_3( /* Copy last part of filtered signal to the state for the next call */ silk_memcpy( S, &buf[ nSamplesIn ], ORDER_FIR * sizeof( opus_int32 ) ); + RESTORE_STACK; } diff --git a/src/opus-1.0.2/silk/resampler_private.h b/src/opus-1.1/silk/resampler_private.h index 45d342c7..422a7d9d 100644 --- a/src/opus-1.0.2/silk/resampler_private.h +++ b/src/opus-1.1/silk/resampler_private.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/resampler_private_AR2.c b/src/opus-1.1/silk/resampler_private_AR2.c index d069f2d8..5fff2371 100644 --- a/src/opus-1.0.2/silk/resampler_private_AR2.c +++ b/src/opus-1.1/silk/resampler_private_AR2.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/resampler_private_IIR_FIR.c b/src/opus-1.1/silk/resampler_private_IIR_FIR.c index d9e42ca0..6b2b3a2e 100644 --- a/src/opus-1.0.2/silk/resampler_private_IIR_FIR.c +++ b/src/opus-1.1/silk/resampler_private_IIR_FIR.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -31,8 +31,9 @@ POSSIBILITY OF SUCH DAMAGE. #include "SigProc_FIX.h" #include "resampler_private.h" +#include "stack_alloc.h" -static inline opus_int16 *silk_resampler_private_IIR_FIR_INTERPOL( +static OPUS_INLINE opus_int16 *silk_resampler_private_IIR_FIR_INTERPOL( opus_int16 *out, opus_int16 *buf, opus_int32 max_index_Q16, @@ -71,10 +72,13 @@ void silk_resampler_private_IIR_FIR( silk_resampler_state_struct *S = (silk_resampler_state_struct *)SS; opus_int32 nSamplesIn; opus_int32 max_index_Q16, index_increment_Q16; - opus_int16 buf[ RESAMPLER_MAX_BATCH_SIZE_IN + RESAMPLER_ORDER_FIR_12 ]; + VARDECL( opus_int16, buf ); + SAVE_STACK; + + ALLOC( buf, 2 * S->batchSize + RESAMPLER_ORDER_FIR_12, opus_int16 ); /* Copy buffered samples to start of buffer */ - silk_memcpy( buf, S->sFIR, RESAMPLER_ORDER_FIR_12 * sizeof( opus_int32 ) ); + silk_memcpy( buf, S->sFIR.i16, RESAMPLER_ORDER_FIR_12 * sizeof( opus_int16 ) ); /* Iterate over blocks of frameSizeIn input samples */ index_increment_Q16 = S->invRatio_Q16; @@ -91,13 +95,13 @@ void silk_resampler_private_IIR_FIR( if( inLen > 0 ) { /* More iterations to do; copy last part of filtered signal to beginning of buffer */ - silk_memcpy( buf, &buf[ nSamplesIn << 1 ], RESAMPLER_ORDER_FIR_12 * sizeof( opus_int32 ) ); + silk_memcpy( buf, &buf[ nSamplesIn << 1 ], RESAMPLER_ORDER_FIR_12 * sizeof( opus_int16 ) ); } else { break; } } /* Copy last part of filtered signal to the state for the next call */ - silk_memcpy( S->sFIR, &buf[ nSamplesIn << 1 ], RESAMPLER_ORDER_FIR_12 * sizeof( opus_int32 ) ); + silk_memcpy( S->sFIR.i16, &buf[ nSamplesIn << 1 ], RESAMPLER_ORDER_FIR_12 * sizeof( opus_int16 ) ); + RESTORE_STACK; } - diff --git a/src/opus-1.0.2/silk/resampler_private_down_FIR.c b/src/opus-1.1/silk/resampler_private_down_FIR.c index 5d24564c..783e42b3 100644 --- a/src/opus-1.0.2/silk/resampler_private_down_FIR.c +++ b/src/opus-1.1/silk/resampler_private_down_FIR.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -31,8 +31,9 @@ POSSIBILITY OF SUCH DAMAGE. #include "SigProc_FIX.h" #include "resampler_private.h" +#include "stack_alloc.h" -static inline opus_int16 *silk_resampler_private_down_FIR_INTERPOL( +static OPUS_INLINE opus_int16 *silk_resampler_private_down_FIR_INTERPOL( opus_int16 *out, opus_int32 *buf, const opus_int16 *FIR_Coefs, @@ -151,11 +152,14 @@ void silk_resampler_private_down_FIR( silk_resampler_state_struct *S = (silk_resampler_state_struct *)SS; opus_int32 nSamplesIn; opus_int32 max_index_Q16, index_increment_Q16; - opus_int32 buf[ RESAMPLER_MAX_BATCH_SIZE_IN + SILK_RESAMPLER_MAX_FIR_ORDER ]; + VARDECL( opus_int32, buf ); const opus_int16 *FIR_Coefs; + SAVE_STACK; + + ALLOC( buf, S->batchSize + S->FIR_Order, opus_int32 ); /* Copy buffered samples to start of buffer */ - silk_memcpy( buf, S->sFIR, S->FIR_Order * sizeof( opus_int32 ) ); + silk_memcpy( buf, S->sFIR.i32, S->FIR_Order * sizeof( opus_int32 ) ); FIR_Coefs = &S->Coefs[ 2 ]; @@ -185,5 +189,6 @@ void silk_resampler_private_down_FIR( } /* Copy last part of filtered signal to the state for the next call */ - silk_memcpy( S->sFIR, &buf[ nSamplesIn ], S->FIR_Order * sizeof( opus_int32 ) ); + silk_memcpy( S->sFIR.i32, &buf[ nSamplesIn ], S->FIR_Order * sizeof( opus_int32 ) ); + RESTORE_STACK; } diff --git a/src/opus-1.0.2/silk/resampler_private_up2_HQ.c b/src/opus-1.1/silk/resampler_private_up2_HQ.c index 9e6dfc9e..c7ec8de3 100644 --- a/src/opus-1.0.2/silk/resampler_private_up2_HQ.c +++ b/src/opus-1.1/silk/resampler_private_up2_HQ.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/resampler_rom.c b/src/opus-1.1/silk/resampler_rom.c index b50af2e2..2d502706 100644 --- a/src/opus-1.0.2/silk/resampler_rom.c +++ b/src/opus-1.1/silk/resampler_rom.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/resampler_rom.h b/src/opus-1.1/silk/resampler_rom.h index 473b24a2..490b3388 100644 --- a/src/opus-1.0.2/silk/resampler_rom.h +++ b/src/opus-1.1/silk/resampler_rom.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/resampler_structs.h b/src/opus-1.1/silk/resampler_structs.h index 4c28bd0a..9e9457d1 100644 --- a/src/opus-1.0.2/silk/resampler_structs.h +++ b/src/opus-1.1/silk/resampler_structs.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -37,7 +37,10 @@ extern "C" { typedef struct _silk_resampler_state_struct{ opus_int32 sIIR[ SILK_RESAMPLER_MAX_IIR_ORDER ]; /* this must be the first element of this struct */ - opus_int32 sFIR[ SILK_RESAMPLER_MAX_FIR_ORDER ]; + union{ + opus_int32 i32[ SILK_RESAMPLER_MAX_FIR_ORDER ]; + opus_int16 i16[ SILK_RESAMPLER_MAX_FIR_ORDER ]; + } sFIR; opus_int16 delayBuf[ 48 ]; opus_int resampler_function; opus_int batchSize; diff --git a/src/opus-1.0.2/silk/shell_coder.c b/src/opus-1.1/silk/shell_coder.c index 32d00129..796f57d6 100644 --- a/src/opus-1.0.2/silk/shell_coder.c +++ b/src/opus-1.1/silk/shell_coder.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -33,7 +33,7 @@ POSSIBILITY OF SUCH DAMAGE. /* shell coder; pulse-subframe length is hardcoded */ -static inline void combine_pulses( +static OPUS_INLINE void combine_pulses( opus_int *out, /* O combined pulses vector [len] */ const opus_int *in, /* I input vector [2 * len] */ const opus_int len /* I number of OUTPUT samples */ @@ -45,7 +45,7 @@ static inline void combine_pulses( } } -static inline void encode_split( +static OPUS_INLINE void encode_split( ec_enc *psRangeEnc, /* I/O compressor data structure */ const opus_int p_child1, /* I pulse amplitude of first child subframe */ const opus_int p, /* I pulse amplitude of current subframe */ @@ -57,7 +57,7 @@ static inline void encode_split( } } -static inline void decode_split( +static OPUS_INLINE void decode_split( opus_int *p_child1, /* O pulse amplitude of first child subframe */ opus_int *p_child2, /* O pulse amplitude of second child subframe */ ec_dec *psRangeDec, /* I/O Compressor data structure */ diff --git a/src/opus-1.0.2/silk/sigm_Q15.c b/src/opus-1.1/silk/sigm_Q15.c index cf5af6bc..3c507d25 100644 --- a/src/opus-1.0.2/silk/sigm_Q15.c +++ b/src/opus-1.1/silk/sigm_Q15.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/sort.c b/src/opus-1.1/silk/sort.c index a4072ec4..8670dbdd 100644 --- a/src/opus-1.0.2/silk/sort.c +++ b/src/opus-1.1/silk/sort.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/stereo_LR_to_MS.c b/src/opus-1.1/silk/stereo_LR_to_MS.c index 6a680e09..42906e6f 100644 --- a/src/opus-1.0.2/silk/stereo_LR_to_MS.c +++ b/src/opus-1.1/silk/stereo_LR_to_MS.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -30,6 +30,7 @@ POSSIBILITY OF SUCH DAMAGE. #endif #include "main.h" +#include "stack_alloc.h" /* Convert Left/Right stereo signal to adaptive Mid/Side representation */ void silk_stereo_LR_to_MS( @@ -49,11 +50,15 @@ void silk_stereo_LR_to_MS( opus_int n, is10msFrame, denom_Q16, delta0_Q13, delta1_Q13; opus_int32 sum, diff, smooth_coef_Q16, pred_Q13[ 2 ], pred0_Q13, pred1_Q13; opus_int32 LP_ratio_Q14, HP_ratio_Q14, frac_Q16, frac_3_Q16, min_mid_rate_bps, width_Q14, w_Q24, deltaw_Q24; - opus_int16 side[ MAX_FRAME_LENGTH + 2 ]; - opus_int16 LP_mid[ MAX_FRAME_LENGTH ], HP_mid[ MAX_FRAME_LENGTH ]; - opus_int16 LP_side[ MAX_FRAME_LENGTH ], HP_side[ MAX_FRAME_LENGTH ]; + VARDECL( opus_int16, side ); + VARDECL( opus_int16, LP_mid ); + VARDECL( opus_int16, HP_mid ); + VARDECL( opus_int16, LP_side ); + VARDECL( opus_int16, HP_side ); opus_int16 *mid = &x1[ -2 ]; + SAVE_STACK; + ALLOC( side, frame_length + 2, opus_int16 ); /* Convert to basic mid/side signals */ for( n = 0; n < frame_length + 2; n++ ) { sum = x1[ n - 2 ] + (opus_int32)x2[ n - 2 ]; @@ -69,6 +74,8 @@ void silk_stereo_LR_to_MS( silk_memcpy( state->sSide, &side[ frame_length ], 2 * sizeof( opus_int16 ) ); /* LP and HP filter mid signal */ + ALLOC( LP_mid, frame_length, opus_int16 ); + ALLOC( HP_mid, frame_length, opus_int16 ); for( n = 0; n < frame_length; n++ ) { sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT( mid[ n ] + mid[ n + 2 ], mid[ n + 1 ], 1 ), 2 ); LP_mid[ n ] = sum; @@ -76,6 +83,8 @@ void silk_stereo_LR_to_MS( } /* LP and HP filter side signal */ + ALLOC( LP_side, frame_length, opus_int16 ); + ALLOC( HP_side, frame_length, opus_int16 ); for( n = 0; n < frame_length; n++ ) { sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT( side[ n ] + side[ n + 2 ], side[ n + 1 ], 1 ), 2 ); LP_side[ n ] = sum; @@ -216,4 +225,5 @@ void silk_stereo_LR_to_MS( state->pred_prev_Q13[ 0 ] = (opus_int16)pred_Q13[ 0 ]; state->pred_prev_Q13[ 1 ] = (opus_int16)pred_Q13[ 1 ]; state->width_prev_Q14 = (opus_int16)width_Q14; + RESTORE_STACK; } diff --git a/src/opus-1.0.2/silk/stereo_MS_to_LR.c b/src/opus-1.1/silk/stereo_MS_to_LR.c index 23515870..62521a4f 100644 --- a/src/opus-1.0.2/silk/stereo_MS_to_LR.c +++ b/src/opus-1.1/silk/stereo_MS_to_LR.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/stereo_decode_pred.c b/src/opus-1.1/silk/stereo_decode_pred.c index 026aa7a0..56ba3925 100644 --- a/src/opus-1.0.2/silk/stereo_decode_pred.c +++ b/src/opus-1.1/silk/stereo_decode_pred.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/stereo_encode_pred.c b/src/opus-1.1/silk/stereo_encode_pred.c index 3cffd367..e6dd1950 100644 --- a/src/opus-1.0.2/silk/stereo_encode_pred.c +++ b/src/opus-1.1/silk/stereo_encode_pred.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/stereo_find_predictor.c b/src/opus-1.1/silk/stereo_find_predictor.c index aec58dab..e30e90bd 100644 --- a/src/opus-1.0.2/silk/stereo_find_predictor.c +++ b/src/opus-1.1/silk/stereo_find_predictor.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/stereo_quant_pred.c b/src/opus-1.1/silk/stereo_quant_pred.c index df97c9f6..d4ced6c3 100644 --- a/src/opus-1.0.2/silk/stereo_quant_pred.c +++ b/src/opus-1.1/silk/stereo_quant_pred.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/structs.h b/src/opus-1.1/silk/structs.h index 5d37f660..1826b36a 100644 --- a/src/opus-1.0.2/silk/structs.h +++ b/src/opus-1.1/silk/structs.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -171,6 +171,7 @@ typedef struct { opus_int32 pitchEstimationThreshold_Q16; /* Threshold for pitch estimator */ opus_int LTPQuantLowComplexity; /* Flag for low complexity LTP quantization */ opus_int mu_LTP_Q9; /* Rate-distortion tradeoff in LTP quantization */ + opus_int32 sum_log_gain_Q7; /* Cumulative max prediction gain */ opus_int NLSF_MSVQ_Survivors; /* Number of survivors in NLSF MSVQ */ opus_int first_frame_after_reset; /* Flag for deactivating NLSF interpolation, pitch prediction */ opus_int controlled_since_last_payload; /* Flag for ensuring codec_control only runs once per packet */ @@ -191,6 +192,8 @@ typedef struct { SideInfoIndices indices; opus_int8 pulses[ MAX_FRAME_LENGTH ]; + int arch; + /* Input/output buffering */ opus_int16 inputBuf[ MAX_FRAME_LENGTH + 2 ]; /* Buffer containing input signal */ opus_int inputBufIx; diff --git a/src/opus-1.0.2/silk/sum_sqr_shift.c b/src/opus-1.1/silk/sum_sqr_shift.c index 2eaf77b6..12514c99 100644 --- a/src/opus-1.0.2/silk/sum_sqr_shift.c +++ b/src/opus-1.1/silk/sum_sqr_shift.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/table_LSF_cos.c b/src/opus-1.1/silk/table_LSF_cos.c index 710537fb..ec9dc639 100644 --- a/src/opus-1.0.2/silk/table_LSF_cos.c +++ b/src/opus-1.1/silk/table_LSF_cos.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/tables.h b/src/opus-1.1/silk/tables.h index 072b7929..a91431e8 100644 --- a/src/opus-1.0.2/silk/tables.h +++ b/src/opus-1.1/silk/tables.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -78,6 +78,8 @@ extern const opus_uint8 * const silk_LTP_gain_iCDF_ptrs[ NB_LTP_CBKS ]; extern const opus_uint8 * const silk_LTP_gain_BITS_Q5_ptrs[ NB_LTP_CBKS ]; /* 3 */ extern const opus_int16 silk_LTP_gain_middle_avg_RD_Q14; extern const opus_int8 * const silk_LTP_vq_ptrs_Q7[ NB_LTP_CBKS ]; /* 168 */ +extern const opus_uint8 * const silk_LTP_vq_gain_ptrs_Q7[NB_LTP_CBKS]; + extern const opus_int8 silk_LTP_vq_sizes[ NB_LTP_CBKS ]; /* 3 */ extern const opus_uint8 silk_LTPscale_iCDF[ 3 ]; /* 4 */ diff --git a/src/opus-1.0.2/silk/tables_LTP.c b/src/opus-1.1/silk/tables_LTP.c index dd1fb556..0e6a0254 100644 --- a/src/opus-1.0.2/silk/tables_LTP.c +++ b/src/opus-1.1/silk/tables_LTP.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -267,6 +267,30 @@ const opus_int8 * const silk_LTP_vq_ptrs_Q7[NB_LTP_CBKS] = { (opus_int8 *)&silk_LTP_gain_vq_2[0][0] }; +/* Maximum frequency-dependent response of the pitch taps above, + computed as max(abs(freqz(taps))) */ +static const opus_uint8 silk_LTP_gain_vq_0_gain[8] = { + 46, 2, 90, 87, 93, 91, 82, 98 +}; + +static const opus_uint8 silk_LTP_gain_vq_1_gain[16] = { + 109, 120, 118, 12, 113, 115, 117, 119, + 99, 59, 87, 111, 63, 111, 112, 80 +}; + +static const opus_uint8 silk_LTP_gain_vq_2_gain[32] = { + 126, 124, 125, 124, 129, 121, 126, 23, + 132, 127, 127, 127, 126, 127, 122, 133, + 130, 134, 101, 118, 119, 145, 126, 86, + 124, 120, 123, 119, 170, 173, 107, 109 +}; + +const opus_uint8 * const silk_LTP_vq_gain_ptrs_Q7[NB_LTP_CBKS] = { + &silk_LTP_gain_vq_0_gain[0], + &silk_LTP_gain_vq_1_gain[0], + &silk_LTP_gain_vq_2_gain[0] +}; + const opus_int8 silk_LTP_vq_sizes[NB_LTP_CBKS] = { 8, 16, 32 }; diff --git a/src/opus-1.0.2/silk/tables_NLSF_CB_NB_MB.c b/src/opus-1.1/silk/tables_NLSF_CB_NB_MB.c index 75480526..8c59d207 100644 --- a/src/opus-1.0.2/silk/tables_NLSF_CB_NB_MB.c +++ b/src/opus-1.1/silk/tables_NLSF_CB_NB_MB.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/tables_NLSF_CB_WB.c b/src/opus-1.1/silk/tables_NLSF_CB_WB.c index 3d6052e4..50af87eb 100644 --- a/src/opus-1.0.2/silk/tables_NLSF_CB_WB.c +++ b/src/opus-1.1/silk/tables_NLSF_CB_WB.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/tables_gain.c b/src/opus-1.1/silk/tables_gain.c index fccef821..37e41d89 100644 --- a/src/opus-1.0.2/silk/tables_gain.c +++ b/src/opus-1.1/silk/tables_gain.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/tables_other.c b/src/opus-1.1/silk/tables_other.c index 3dc68d47..398686bf 100644 --- a/src/opus-1.0.2/silk/tables_other.c +++ b/src/opus-1.1/silk/tables_other.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/tables_pitch_lag.c b/src/opus-1.1/silk/tables_pitch_lag.c index 819b0ab3..e80cc59a 100644 --- a/src/opus-1.0.2/silk/tables_pitch_lag.c +++ b/src/opus-1.1/silk/tables_pitch_lag.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/tables_pulses_per_block.c b/src/opus-1.1/silk/tables_pulses_per_block.c index 521e6ff6..c7c01c88 100644 --- a/src/opus-1.0.2/silk/tables_pulses_per_block.c +++ b/src/opus-1.1/silk/tables_pulses_per_block.c @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE diff --git a/src/opus-1.0.2/silk/tuning_parameters.h b/src/opus-1.1/silk/tuning_parameters.h index a26de4d2..e1057bba 100644 --- a/src/opus-1.0.2/silk/tuning_parameters.h +++ b/src/opus-1.1/silk/tuning_parameters.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -50,7 +50,7 @@ extern "C" /* Linear prediction */ /*********************/ -/* LPC analysis defines: regularization and bandwidth expansion */ +/* LPC analysis regularization */ #define FIND_LPC_COND_FAC 1e-5f /* LTP analysis defines */ @@ -63,6 +63,9 @@ extern "C" #define MU_LTP_QUANT_MB 0.025f #define MU_LTP_QUANT_WB 0.02f +/* Max cumulative LTP gain */ +#define MAX_SUM_LOG_GAIN_DB 250.0f + /***********************/ /* High pass filtering */ /***********************/ diff --git a/src/opus-1.0.2/silk/typedef.h b/src/opus-1.1/silk/typedef.h index da981237..97b7e709 100644 --- a/src/opus-1.0.2/silk/typedef.h +++ b/src/opus-1.1/silk/typedef.h @@ -8,11 +8,11 @@ this list of conditions and the following disclaimer. - Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution. -- Neither the name of Internet Society, IETF or IETF Trust, nor the +- Neither the name of Internet Society, IETF or IETF Trust, nor the names of specific contributors, may be used to endorse or promote products derived from this software without specific prior written permission. -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE @@ -29,6 +29,7 @@ POSSIBILITY OF SUCH DAMAGE. #define SILK_TYPEDEF_H #include "opus_types.h" +#include "opus_defines.h" #ifndef FIXED_POINT # include <float.h> @@ -63,7 +64,7 @@ POSSIBILITY OF SUCH DAMAGE. #ifdef __GNUC__ __attribute__((noreturn)) #endif -static inline void _silk_fatal(const char *str, const char *file, int line) +static OPUS_INLINE void _silk_fatal(const char *str, const char *file, int line) { fprintf (stderr, "Fatal (internal) error in %s, line %d: %s\n", file, line, str); abort(); diff --git a/src/opus-1.1/src/analysis.c b/src/opus-1.1/src/analysis.c new file mode 100644 index 00000000..778a62aa --- /dev/null +++ b/src/opus-1.1/src/analysis.c @@ -0,0 +1,645 @@ +/* Copyright (c) 2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "kiss_fft.h" +#include "celt.h" +#include "modes.h" +#include "arch.h" +#include "quant_bands.h" +#include <stdio.h> +#include "analysis.h" +#include "mlp.h" +#include "stack_alloc.h" + +extern const MLP net; + +#ifndef M_PI +#define M_PI 3.141592653 +#endif + +static const float dct_table[128] = { + 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, + 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, + 0.351851f, 0.338330f, 0.311806f, 0.273300f, 0.224292f, 0.166664f, 0.102631f, 0.034654f, + -0.034654f,-0.102631f,-0.166664f,-0.224292f,-0.273300f,-0.311806f,-0.338330f,-0.351851f, + 0.346760f, 0.293969f, 0.196424f, 0.068975f,-0.068975f,-0.196424f,-0.293969f,-0.346760f, + -0.346760f,-0.293969f,-0.196424f,-0.068975f, 0.068975f, 0.196424f, 0.293969f, 0.346760f, + 0.338330f, 0.224292f, 0.034654f,-0.166664f,-0.311806f,-0.351851f,-0.273300f,-0.102631f, + 0.102631f, 0.273300f, 0.351851f, 0.311806f, 0.166664f,-0.034654f,-0.224292f,-0.338330f, + 0.326641f, 0.135299f,-0.135299f,-0.326641f,-0.326641f,-0.135299f, 0.135299f, 0.326641f, + 0.326641f, 0.135299f,-0.135299f,-0.326641f,-0.326641f,-0.135299f, 0.135299f, 0.326641f, + 0.311806f, 0.034654f,-0.273300f,-0.338330f,-0.102631f, 0.224292f, 0.351851f, 0.166664f, + -0.166664f,-0.351851f,-0.224292f, 0.102631f, 0.338330f, 0.273300f,-0.034654f,-0.311806f, + 0.293969f,-0.068975f,-0.346760f,-0.196424f, 0.196424f, 0.346760f, 0.068975f,-0.293969f, + -0.293969f, 0.068975f, 0.346760f, 0.196424f,-0.196424f,-0.346760f,-0.068975f, 0.293969f, + 0.273300f,-0.166664f,-0.338330f, 0.034654f, 0.351851f, 0.102631f,-0.311806f,-0.224292f, + 0.224292f, 0.311806f,-0.102631f,-0.351851f,-0.034654f, 0.338330f, 0.166664f,-0.273300f, +}; + +static const float analysis_window[240] = { + 0.000043f, 0.000171f, 0.000385f, 0.000685f, 0.001071f, 0.001541f, 0.002098f, 0.002739f, + 0.003466f, 0.004278f, 0.005174f, 0.006156f, 0.007222f, 0.008373f, 0.009607f, 0.010926f, + 0.012329f, 0.013815f, 0.015385f, 0.017037f, 0.018772f, 0.020590f, 0.022490f, 0.024472f, + 0.026535f, 0.028679f, 0.030904f, 0.033210f, 0.035595f, 0.038060f, 0.040604f, 0.043227f, + 0.045928f, 0.048707f, 0.051564f, 0.054497f, 0.057506f, 0.060591f, 0.063752f, 0.066987f, + 0.070297f, 0.073680f, 0.077136f, 0.080665f, 0.084265f, 0.087937f, 0.091679f, 0.095492f, + 0.099373f, 0.103323f, 0.107342f, 0.111427f, 0.115579f, 0.119797f, 0.124080f, 0.128428f, + 0.132839f, 0.137313f, 0.141849f, 0.146447f, 0.151105f, 0.155823f, 0.160600f, 0.165435f, + 0.170327f, 0.175276f, 0.180280f, 0.185340f, 0.190453f, 0.195619f, 0.200838f, 0.206107f, + 0.211427f, 0.216797f, 0.222215f, 0.227680f, 0.233193f, 0.238751f, 0.244353f, 0.250000f, + 0.255689f, 0.261421f, 0.267193f, 0.273005f, 0.278856f, 0.284744f, 0.290670f, 0.296632f, + 0.302628f, 0.308658f, 0.314721f, 0.320816f, 0.326941f, 0.333097f, 0.339280f, 0.345492f, + 0.351729f, 0.357992f, 0.364280f, 0.370590f, 0.376923f, 0.383277f, 0.389651f, 0.396044f, + 0.402455f, 0.408882f, 0.415325f, 0.421783f, 0.428254f, 0.434737f, 0.441231f, 0.447736f, + 0.454249f, 0.460770f, 0.467298f, 0.473832f, 0.480370f, 0.486912f, 0.493455f, 0.500000f, + 0.506545f, 0.513088f, 0.519630f, 0.526168f, 0.532702f, 0.539230f, 0.545751f, 0.552264f, + 0.558769f, 0.565263f, 0.571746f, 0.578217f, 0.584675f, 0.591118f, 0.597545f, 0.603956f, + 0.610349f, 0.616723f, 0.623077f, 0.629410f, 0.635720f, 0.642008f, 0.648271f, 0.654508f, + 0.660720f, 0.666903f, 0.673059f, 0.679184f, 0.685279f, 0.691342f, 0.697372f, 0.703368f, + 0.709330f, 0.715256f, 0.721144f, 0.726995f, 0.732807f, 0.738579f, 0.744311f, 0.750000f, + 0.755647f, 0.761249f, 0.766807f, 0.772320f, 0.777785f, 0.783203f, 0.788573f, 0.793893f, + 0.799162f, 0.804381f, 0.809547f, 0.814660f, 0.819720f, 0.824724f, 0.829673f, 0.834565f, + 0.839400f, 0.844177f, 0.848895f, 0.853553f, 0.858151f, 0.862687f, 0.867161f, 0.871572f, + 0.875920f, 0.880203f, 0.884421f, 0.888573f, 0.892658f, 0.896677f, 0.900627f, 0.904508f, + 0.908321f, 0.912063f, 0.915735f, 0.919335f, 0.922864f, 0.926320f, 0.929703f, 0.933013f, + 0.936248f, 0.939409f, 0.942494f, 0.945503f, 0.948436f, 0.951293f, 0.954072f, 0.956773f, + 0.959396f, 0.961940f, 0.964405f, 0.966790f, 0.969096f, 0.971321f, 0.973465f, 0.975528f, + 0.977510f, 0.979410f, 0.981228f, 0.982963f, 0.984615f, 0.986185f, 0.987671f, 0.989074f, + 0.990393f, 0.991627f, 0.992778f, 0.993844f, 0.994826f, 0.995722f, 0.996534f, 0.997261f, + 0.997902f, 0.998459f, 0.998929f, 0.999315f, 0.999615f, 0.999829f, 0.999957f, 1.000000f, +}; + +static const int tbands[NB_TBANDS+1] = { + 2, 4, 6, 8, 10, 12, 14, 16, 20, 24, 28, 32, 40, 48, 56, 68, 80, 96, 120 +}; + +static const int extra_bands[NB_TOT_BANDS+1] = { + 1, 2, 4, 6, 8, 10, 12, 14, 16, 20, 24, 28, 32, 40, 48, 56, 68, 80, 96, 120, 160, 200 +}; + +/*static const float tweight[NB_TBANDS+1] = { + .3, .4, .5, .6, .7, .8, .9, 1., 1., 1., 1., 1., 1., 1., .8, .7, .6, .5 +};*/ + +#define NB_TONAL_SKIP_BANDS 9 + +#define cA 0.43157974f +#define cB 0.67848403f +#define cC 0.08595542f +#define cE ((float)M_PI/2) +static OPUS_INLINE float fast_atan2f(float y, float x) { + float x2, y2; + /* Should avoid underflow on the values we'll get */ + if (ABS16(x)+ABS16(y)<1e-9f) + { + x*=1e12f; + y*=1e12f; + } + x2 = x*x; + y2 = y*y; + if(x2<y2){ + float den = (y2 + cB*x2) * (y2 + cC*x2); + if (den!=0) + return -x*y*(y2 + cA*x2) / den + (y<0 ? -cE : cE); + else + return (y<0 ? -cE : cE); + }else{ + float den = (x2 + cB*y2) * (x2 + cC*y2); + if (den!=0) + return x*y*(x2 + cA*y2) / den + (y<0 ? -cE : cE) - (x*y<0 ? -cE : cE); + else + return (y<0 ? -cE : cE) - (x*y<0 ? -cE : cE); + } +} + +void tonality_get_info(TonalityAnalysisState *tonal, AnalysisInfo *info_out, int len) +{ + int pos; + int curr_lookahead; + float psum; + int i; + + pos = tonal->read_pos; + curr_lookahead = tonal->write_pos-tonal->read_pos; + if (curr_lookahead<0) + curr_lookahead += DETECT_SIZE; + + if (len > 480 && pos != tonal->write_pos) + { + pos++; + if (pos==DETECT_SIZE) + pos=0; + } + if (pos == tonal->write_pos) + pos--; + if (pos<0) + pos = DETECT_SIZE-1; + OPUS_COPY(info_out, &tonal->info[pos], 1); + tonal->read_subframe += len/120; + while (tonal->read_subframe>=4) + { + tonal->read_subframe -= 4; + tonal->read_pos++; + } + if (tonal->read_pos>=DETECT_SIZE) + tonal->read_pos-=DETECT_SIZE; + + /* Compensate for the delay in the features themselves. + FIXME: Need a better estimate the 10 I just made up */ + curr_lookahead = IMAX(curr_lookahead-10, 0); + + psum=0; + /* Summing the probability of transition patterns that involve music at + time (DETECT_SIZE-curr_lookahead-1) */ + for (i=0;i<DETECT_SIZE-curr_lookahead;i++) + psum += tonal->pmusic[i]; + for (;i<DETECT_SIZE;i++) + psum += tonal->pspeech[i]; + psum = psum*tonal->music_confidence + (1-psum)*tonal->speech_confidence; + /*printf("%f %f %f\n", psum, info_out->music_prob, info_out->tonality);*/ + + info_out->music_prob = psum; +} + +void tonality_analysis(TonalityAnalysisState *tonal, AnalysisInfo *info_out, const CELTMode *celt_mode, const void *x, int len, int offset, int c1, int c2, int C, int lsb_depth, downmix_func downmix) +{ + int i, b; + const kiss_fft_state *kfft; + VARDECL(kiss_fft_cpx, in); + VARDECL(kiss_fft_cpx, out); + int N = 480, N2=240; + float * OPUS_RESTRICT A = tonal->angle; + float * OPUS_RESTRICT dA = tonal->d_angle; + float * OPUS_RESTRICT d2A = tonal->d2_angle; + VARDECL(float, tonality); + VARDECL(float, noisiness); + float band_tonality[NB_TBANDS]; + float logE[NB_TBANDS]; + float BFCC[8]; + float features[25]; + float frame_tonality; + float max_frame_tonality; + /*float tw_sum=0;*/ + float frame_noisiness; + const float pi4 = (float)(M_PI*M_PI*M_PI*M_PI); + float slope=0; + float frame_stationarity; + float relativeE; + float frame_probs[2]; + float alpha, alphaE, alphaE2; + float frame_loudness; + float bandwidth_mask; + int bandwidth=0; + float maxE = 0; + float noise_floor; + int remaining; + AnalysisInfo *info; + SAVE_STACK; + + tonal->last_transition++; + alpha = 1.f/IMIN(20, 1+tonal->count); + alphaE = 1.f/IMIN(50, 1+tonal->count); + alphaE2 = 1.f/IMIN(1000, 1+tonal->count); + + if (tonal->count<4) + tonal->music_prob = .5; + kfft = celt_mode->mdct.kfft[0]; + if (tonal->count==0) + tonal->mem_fill = 240; + downmix(x, &tonal->inmem[tonal->mem_fill], IMIN(len, ANALYSIS_BUF_SIZE-tonal->mem_fill), offset, c1, c2, C); + if (tonal->mem_fill+len < ANALYSIS_BUF_SIZE) + { + tonal->mem_fill += len; + /* Don't have enough to update the analysis */ + RESTORE_STACK; + return; + } + info = &tonal->info[tonal->write_pos++]; + if (tonal->write_pos>=DETECT_SIZE) + tonal->write_pos-=DETECT_SIZE; + + ALLOC(in, 480, kiss_fft_cpx); + ALLOC(out, 480, kiss_fft_cpx); + ALLOC(tonality, 240, float); + ALLOC(noisiness, 240, float); + for (i=0;i<N2;i++) + { + float w = analysis_window[i]; + in[i].r = (kiss_fft_scalar)(w*tonal->inmem[i]); + in[i].i = (kiss_fft_scalar)(w*tonal->inmem[N2+i]); + in[N-i-1].r = (kiss_fft_scalar)(w*tonal->inmem[N-i-1]); + in[N-i-1].i = (kiss_fft_scalar)(w*tonal->inmem[N+N2-i-1]); + } + OPUS_MOVE(tonal->inmem, tonal->inmem+ANALYSIS_BUF_SIZE-240, 240); + remaining = len - (ANALYSIS_BUF_SIZE-tonal->mem_fill); + downmix(x, &tonal->inmem[240], remaining, offset+ANALYSIS_BUF_SIZE-tonal->mem_fill, c1, c2, C); + tonal->mem_fill = 240 + remaining; + opus_fft(kfft, in, out); + + for (i=1;i<N2;i++) + { + float X1r, X2r, X1i, X2i; + float angle, d_angle, d2_angle; + float angle2, d_angle2, d2_angle2; + float mod1, mod2, avg_mod; + X1r = (float)out[i].r+out[N-i].r; + X1i = (float)out[i].i-out[N-i].i; + X2r = (float)out[i].i+out[N-i].i; + X2i = (float)out[N-i].r-out[i].r; + + angle = (float)(.5f/M_PI)*fast_atan2f(X1i, X1r); + d_angle = angle - A[i]; + d2_angle = d_angle - dA[i]; + + angle2 = (float)(.5f/M_PI)*fast_atan2f(X2i, X2r); + d_angle2 = angle2 - angle; + d2_angle2 = d_angle2 - d_angle; + + mod1 = d2_angle - (float)floor(.5+d2_angle); + noisiness[i] = ABS16(mod1); + mod1 *= mod1; + mod1 *= mod1; + + mod2 = d2_angle2 - (float)floor(.5+d2_angle2); + noisiness[i] += ABS16(mod2); + mod2 *= mod2; + mod2 *= mod2; + + avg_mod = .25f*(d2A[i]+2.f*mod1+mod2); + tonality[i] = 1.f/(1.f+40.f*16.f*pi4*avg_mod)-.015f; + + A[i] = angle2; + dA[i] = d_angle2; + d2A[i] = mod2; + } + + frame_tonality = 0; + max_frame_tonality = 0; + /*tw_sum = 0;*/ + info->activity = 0; + frame_noisiness = 0; + frame_stationarity = 0; + if (!tonal->count) + { + for (b=0;b<NB_TBANDS;b++) + { + tonal->lowE[b] = 1e10; + tonal->highE[b] = -1e10; + } + } + relativeE = 0; + frame_loudness = 0; + for (b=0;b<NB_TBANDS;b++) + { + float E=0, tE=0, nE=0; + float L1, L2; + float stationarity; + for (i=tbands[b];i<tbands[b+1];i++) + { + float binE = out[i].r*(float)out[i].r + out[N-i].r*(float)out[N-i].r + + out[i].i*(float)out[i].i + out[N-i].i*(float)out[N-i].i; +#ifdef FIXED_POINT + /* FIXME: It's probably best to change the BFCC filter initial state instead */ + binE *= 5.55e-17f; +#endif + E += binE; + tE += binE*tonality[i]; + nE += binE*2.f*(.5f-noisiness[i]); + } + tonal->E[tonal->E_count][b] = E; + frame_noisiness += nE/(1e-15f+E); + + frame_loudness += (float)sqrt(E+1e-10f); + logE[b] = (float)log(E+1e-10f); + tonal->lowE[b] = MIN32(logE[b], tonal->lowE[b]+.01f); + tonal->highE[b] = MAX32(logE[b], tonal->highE[b]-.1f); + if (tonal->highE[b] < tonal->lowE[b]+1.f) + { + tonal->highE[b]+=.5f; + tonal->lowE[b]-=.5f; + } + relativeE += (logE[b]-tonal->lowE[b])/(1e-15f+tonal->highE[b]-tonal->lowE[b]); + + L1=L2=0; + for (i=0;i<NB_FRAMES;i++) + { + L1 += (float)sqrt(tonal->E[i][b]); + L2 += tonal->E[i][b]; + } + + stationarity = MIN16(0.99f,L1/(float)sqrt(1e-15+NB_FRAMES*L2)); + stationarity *= stationarity; + stationarity *= stationarity; + frame_stationarity += stationarity; + /*band_tonality[b] = tE/(1e-15+E)*/; + band_tonality[b] = MAX16(tE/(1e-15f+E), stationarity*tonal->prev_band_tonality[b]); +#if 0 + if (b>=NB_TONAL_SKIP_BANDS) + { + frame_tonality += tweight[b]*band_tonality[b]; + tw_sum += tweight[b]; + } +#else + frame_tonality += band_tonality[b]; + if (b>=NB_TBANDS-NB_TONAL_SKIP_BANDS) + frame_tonality -= band_tonality[b-NB_TBANDS+NB_TONAL_SKIP_BANDS]; +#endif + max_frame_tonality = MAX16(max_frame_tonality, (1.f+.03f*(b-NB_TBANDS))*frame_tonality); + slope += band_tonality[b]*(b-8); + /*printf("%f %f ", band_tonality[b], stationarity);*/ + tonal->prev_band_tonality[b] = band_tonality[b]; + } + + bandwidth_mask = 0; + bandwidth = 0; + maxE = 0; + noise_floor = 5.7e-4f/(1<<(IMAX(0,lsb_depth-8))); +#ifdef FIXED_POINT + noise_floor *= 1<<(15+SIG_SHIFT); +#endif + noise_floor *= noise_floor; + for (b=0;b<NB_TOT_BANDS;b++) + { + float E=0; + int band_start, band_end; + /* Keep a margin of 300 Hz for aliasing */ + band_start = extra_bands[b]; + band_end = extra_bands[b+1]; + for (i=band_start;i<band_end;i++) + { + float binE = out[i].r*(float)out[i].r + out[N-i].r*(float)out[N-i].r + + out[i].i*(float)out[i].i + out[N-i].i*(float)out[N-i].i; + E += binE; + } + maxE = MAX32(maxE, E); + tonal->meanE[b] = MAX32((1-alphaE2)*tonal->meanE[b], E); + E = MAX32(E, tonal->meanE[b]); + /* Use a simple follower with 13 dB/Bark slope for spreading function */ + bandwidth_mask = MAX32(.05f*bandwidth_mask, E); + /* Consider the band "active" only if all these conditions are met: + 1) less than 10 dB below the simple follower + 2) less than 90 dB below the peak band (maximal masking possible considering + both the ATH and the loudness-dependent slope of the spreading function) + 3) above the PCM quantization noise floor + */ + if (E>.1*bandwidth_mask && E*1e9f > maxE && E > noise_floor*(band_end-band_start)) + bandwidth = b; + } + if (tonal->count<=2) + bandwidth = 20; + frame_loudness = 20*(float)log10(frame_loudness); + tonal->Etracker = MAX32(tonal->Etracker-.03f, frame_loudness); + tonal->lowECount *= (1-alphaE); + if (frame_loudness < tonal->Etracker-30) + tonal->lowECount += alphaE; + + for (i=0;i<8;i++) + { + float sum=0; + for (b=0;b<16;b++) + sum += dct_table[i*16+b]*logE[b]; + BFCC[i] = sum; + } + + frame_stationarity /= NB_TBANDS; + relativeE /= NB_TBANDS; + if (tonal->count<10) + relativeE = .5; + frame_noisiness /= NB_TBANDS; +#if 1 + info->activity = frame_noisiness + (1-frame_noisiness)*relativeE; +#else + info->activity = .5*(1+frame_noisiness-frame_stationarity); +#endif + frame_tonality = (max_frame_tonality/(NB_TBANDS-NB_TONAL_SKIP_BANDS)); + frame_tonality = MAX16(frame_tonality, tonal->prev_tonality*.8f); + tonal->prev_tonality = frame_tonality; + + slope /= 8*8; + info->tonality_slope = slope; + + tonal->E_count = (tonal->E_count+1)%NB_FRAMES; + tonal->count++; + info->tonality = frame_tonality; + + for (i=0;i<4;i++) + features[i] = -0.12299f*(BFCC[i]+tonal->mem[i+24]) + 0.49195f*(tonal->mem[i]+tonal->mem[i+16]) + 0.69693f*tonal->mem[i+8] - 1.4349f*tonal->cmean[i]; + + for (i=0;i<4;i++) + tonal->cmean[i] = (1-alpha)*tonal->cmean[i] + alpha*BFCC[i]; + + for (i=0;i<4;i++) + features[4+i] = 0.63246f*(BFCC[i]-tonal->mem[i+24]) + 0.31623f*(tonal->mem[i]-tonal->mem[i+16]); + for (i=0;i<3;i++) + features[8+i] = 0.53452f*(BFCC[i]+tonal->mem[i+24]) - 0.26726f*(tonal->mem[i]+tonal->mem[i+16]) -0.53452f*tonal->mem[i+8]; + + if (tonal->count > 5) + { + for (i=0;i<9;i++) + tonal->std[i] = (1-alpha)*tonal->std[i] + alpha*features[i]*features[i]; + } + + for (i=0;i<8;i++) + { + tonal->mem[i+24] = tonal->mem[i+16]; + tonal->mem[i+16] = tonal->mem[i+8]; + tonal->mem[i+8] = tonal->mem[i]; + tonal->mem[i] = BFCC[i]; + } + for (i=0;i<9;i++) + features[11+i] = (float)sqrt(tonal->std[i]); + features[20] = info->tonality; + features[21] = info->activity; + features[22] = frame_stationarity; + features[23] = info->tonality_slope; + features[24] = tonal->lowECount; + +#ifndef DISABLE_FLOAT_API + mlp_process(&net, features, frame_probs); + frame_probs[0] = .5f*(frame_probs[0]+1); + /* Curve fitting between the MLP probability and the actual probability */ + frame_probs[0] = .01f + 1.21f*frame_probs[0]*frame_probs[0] - .23f*(float)pow(frame_probs[0], 10); + /* Probability of active audio (as opposed to silence) */ + frame_probs[1] = .5f*frame_probs[1]+.5f; + /* Consider that silence has a 50-50 probability. */ + frame_probs[0] = frame_probs[1]*frame_probs[0] + (1-frame_probs[1])*.5f; + + /*printf("%f %f ", frame_probs[0], frame_probs[1]);*/ + { + /* Probability of state transition */ + float tau; + /* Represents independence of the MLP probabilities, where + beta=1 means fully independent. */ + float beta; + /* Denormalized probability of speech (p0) and music (p1) after update */ + float p0, p1; + /* Probabilities for "all speech" and "all music" */ + float s0, m0; + /* Probability sum for renormalisation */ + float psum; + /* Instantaneous probability of speech and music, with beta pre-applied. */ + float speech0; + float music0; + + /* One transition every 3 minutes of active audio */ + tau = .00005f*frame_probs[1]; + beta = .05f; + if (1) { + /* Adapt beta based on how "unexpected" the new prob is */ + float p, q; + p = MAX16(.05f,MIN16(.95f,frame_probs[0])); + q = MAX16(.05f,MIN16(.95f,tonal->music_prob)); + beta = .01f+.05f*ABS16(p-q)/(p*(1-q)+q*(1-p)); + } + /* p0 and p1 are the probabilities of speech and music at this frame + using only information from previous frame and applying the + state transition model */ + p0 = (1-tonal->music_prob)*(1-tau) + tonal->music_prob *tau; + p1 = tonal->music_prob *(1-tau) + (1-tonal->music_prob)*tau; + /* We apply the current probability with exponent beta to work around + the fact that the probability estimates aren't independent. */ + p0 *= (float)pow(1-frame_probs[0], beta); + p1 *= (float)pow(frame_probs[0], beta); + /* Normalise the probabilities to get the Marokv probability of music. */ + tonal->music_prob = p1/(p0+p1); + info->music_prob = tonal->music_prob; + + /* This chunk of code deals with delayed decision. */ + psum=1e-20f; + /* Instantaneous probability of speech and music, with beta pre-applied. */ + speech0 = (float)pow(1-frame_probs[0], beta); + music0 = (float)pow(frame_probs[0], beta); + if (tonal->count==1) + { + tonal->pspeech[0]=.5; + tonal->pmusic [0]=.5; + } + /* Updated probability of having only speech (s0) or only music (m0), + before considering the new observation. */ + s0 = tonal->pspeech[0] + tonal->pspeech[1]; + m0 = tonal->pmusic [0] + tonal->pmusic [1]; + /* Updates s0 and m0 with instantaneous probability. */ + tonal->pspeech[0] = s0*(1-tau)*speech0; + tonal->pmusic [0] = m0*(1-tau)*music0; + /* Propagate the transition probabilities */ + for (i=1;i<DETECT_SIZE-1;i++) + { + tonal->pspeech[i] = tonal->pspeech[i+1]*speech0; + tonal->pmusic [i] = tonal->pmusic [i+1]*music0; + } + /* Probability that the latest frame is speech, when all the previous ones were music. */ + tonal->pspeech[DETECT_SIZE-1] = m0*tau*speech0; + /* Probability that the latest frame is music, when all the previous ones were speech. */ + tonal->pmusic [DETECT_SIZE-1] = s0*tau*music0; + + /* Renormalise probabilities to 1 */ + for (i=0;i<DETECT_SIZE;i++) + psum += tonal->pspeech[i] + tonal->pmusic[i]; + psum = 1.f/psum; + for (i=0;i<DETECT_SIZE;i++) + { + tonal->pspeech[i] *= psum; + tonal->pmusic [i] *= psum; + } + psum = tonal->pmusic[0]; + for (i=1;i<DETECT_SIZE;i++) + psum += tonal->pspeech[i]; + + /* Estimate our confidence in the speech/music decisions */ + if (frame_probs[1]>.75) + { + if (tonal->music_prob>.9) + { + float adapt; + adapt = 1.f/(++tonal->music_confidence_count); + tonal->music_confidence_count = IMIN(tonal->music_confidence_count, 500); + tonal->music_confidence += adapt*MAX16(-.2f,frame_probs[0]-tonal->music_confidence); + } + if (tonal->music_prob<.1) + { + float adapt; + adapt = 1.f/(++tonal->speech_confidence_count); + tonal->speech_confidence_count = IMIN(tonal->speech_confidence_count, 500); + tonal->speech_confidence += adapt*MIN16(.2f,frame_probs[0]-tonal->speech_confidence); + } + } else { + if (tonal->music_confidence_count==0) + tonal->music_confidence = .9f; + if (tonal->speech_confidence_count==0) + tonal->speech_confidence = .1f; + } + } + if (tonal->last_music != (tonal->music_prob>.5f)) + tonal->last_transition=0; + tonal->last_music = tonal->music_prob>.5f; +#else + info->music_prob = 0; +#endif + /*for (i=0;i<25;i++) + printf("%f ", features[i]); + printf("\n");*/ + + info->bandwidth = bandwidth; + /*printf("%d %d\n", info->bandwidth, info->opus_bandwidth);*/ + info->noisiness = frame_noisiness; + info->valid = 1; + if (info_out!=NULL) + OPUS_COPY(info_out, info, 1); + RESTORE_STACK; +} + +void run_analysis(TonalityAnalysisState *analysis, const CELTMode *celt_mode, const void *analysis_pcm, + int analysis_frame_size, int frame_size, int c1, int c2, int C, opus_int32 Fs, + int lsb_depth, downmix_func downmix, AnalysisInfo *analysis_info) +{ + int offset; + int pcm_len; + + if (analysis_pcm != NULL) + { + /* Avoid overflow/wrap-around of the analysis buffer */ + analysis_frame_size = IMIN((DETECT_SIZE-5)*Fs/100, analysis_frame_size); + + pcm_len = analysis_frame_size - analysis->analysis_offset; + offset = analysis->analysis_offset; + do { + tonality_analysis(analysis, NULL, celt_mode, analysis_pcm, IMIN(480, pcm_len), offset, c1, c2, C, lsb_depth, downmix); + offset += 480; + pcm_len -= 480; + } while (pcm_len>0); + analysis->analysis_offset = analysis_frame_size; + + analysis->analysis_offset -= frame_size; + } + + analysis_info->valid = 0; + tonality_get_info(analysis, analysis_info, frame_size); +} diff --git a/src/opus-1.1/src/analysis.h b/src/opus-1.1/src/analysis.h new file mode 100644 index 00000000..be0388fa --- /dev/null +++ b/src/opus-1.1/src/analysis.h @@ -0,0 +1,90 @@ +/* Copyright (c) 2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef ANALYSIS_H +#define ANALYSIS_H + +#include "celt.h" +#include "opus_private.h" + +#define NB_FRAMES 8 +#define NB_TBANDS 18 +#define NB_TOT_BANDS 21 +#define ANALYSIS_BUF_SIZE 720 /* 15 ms at 48 kHz */ + +#define DETECT_SIZE 200 + +typedef struct { + float angle[240]; + float d_angle[240]; + float d2_angle[240]; + opus_val32 inmem[ANALYSIS_BUF_SIZE]; + int mem_fill; /* number of usable samples in the buffer */ + float prev_band_tonality[NB_TBANDS]; + float prev_tonality; + float E[NB_FRAMES][NB_TBANDS]; + float lowE[NB_TBANDS]; + float highE[NB_TBANDS]; + float meanE[NB_TOT_BANDS]; + float mem[32]; + float cmean[8]; + float std[9]; + float music_prob; + float Etracker; + float lowECount; + int E_count; + int last_music; + int last_transition; + int count; + float subframe_mem[3]; + int analysis_offset; + /** Probability of having speech for time i to DETECT_SIZE-1 (and music before). + pspeech[0] is the probability that all frames in the window are speech. */ + float pspeech[DETECT_SIZE]; + /** Probability of having music for time i to DETECT_SIZE-1 (and speech before). + pmusic[0] is the probability that all frames in the window are music. */ + float pmusic[DETECT_SIZE]; + float speech_confidence; + float music_confidence; + int speech_confidence_count; + int music_confidence_count; + int write_pos; + int read_pos; + int read_subframe; + AnalysisInfo info[DETECT_SIZE]; +} TonalityAnalysisState; + +void tonality_analysis(TonalityAnalysisState *tonal, AnalysisInfo *info, + const CELTMode *celt_mode, const void *x, int len, int offset, int c1, int c2, int C, int lsb_depth, downmix_func downmix); + +void tonality_get_info(TonalityAnalysisState *tonal, AnalysisInfo *info_out, int len); + +void run_analysis(TonalityAnalysisState *analysis, const CELTMode *celt_mode, const void *analysis_pcm, + int analysis_frame_size, int frame_size, int c1, int c2, int C, opus_int32 Fs, + int lsb_depth, downmix_func downmix, AnalysisInfo *analysis_info); + +#endif diff --git a/src/opus-1.1/src/mlp.c b/src/opus-1.1/src/mlp.c new file mode 100644 index 00000000..46386026 --- /dev/null +++ b/src/opus-1.1/src/mlp.c @@ -0,0 +1,140 @@ +/* Copyright (c) 2008-2011 Octasic Inc. + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "opus_types.h" +#include "opus_defines.h" + +#include <math.h> +#include "mlp.h" +#include "arch.h" +#include "tansig_table.h" +#define MAX_NEURONS 100 + +#if 0 +static OPUS_INLINE opus_val16 tansig_approx(opus_val32 _x) /* Q19 */ +{ + int i; + opus_val16 xx; /* Q11 */ + /*double x, y;*/ + opus_val16 dy, yy; /* Q14 */ + /*x = 1.9073e-06*_x;*/ + if (_x>=QCONST32(8,19)) + return QCONST32(1.,14); + if (_x<=-QCONST32(8,19)) + return -QCONST32(1.,14); + xx = EXTRACT16(SHR32(_x, 8)); + /*i = lrint(25*x);*/ + i = SHR32(ADD32(1024,MULT16_16(25, xx)),11); + /*x -= .04*i;*/ + xx -= EXTRACT16(SHR32(MULT16_16(20972,i),8)); + /*x = xx*(1./2048);*/ + /*y = tansig_table[250+i];*/ + yy = tansig_table[250+i]; + /*y = yy*(1./16384);*/ + dy = 16384-MULT16_16_Q14(yy,yy); + yy = yy + MULT16_16_Q14(MULT16_16_Q11(xx,dy),(16384 - MULT16_16_Q11(yy,xx))); + return yy; +} +#else +/*extern const float tansig_table[501];*/ +static OPUS_INLINE float tansig_approx(float x) +{ + int i; + float y, dy; + float sign=1; + /* Tests are reversed to catch NaNs */ + if (!(x<8)) + return 1; + if (!(x>-8)) + return -1; + if (x<0) + { + x=-x; + sign=-1; + } + i = (int)floor(.5f+25*x); + x -= .04f*i; + y = tansig_table[i]; + dy = 1-y*y; + y = y + x*dy*(1 - y*x); + return sign*y; +} +#endif + +#if 0 +void mlp_process(const MLP *m, const opus_val16 *in, opus_val16 *out) +{ + int j; + opus_val16 hidden[MAX_NEURONS]; + const opus_val16 *W = m->weights; + /* Copy to tmp_in */ + for (j=0;j<m->topo[1];j++) + { + int k; + opus_val32 sum = SHL32(EXTEND32(*W++),8); + for (k=0;k<m->topo[0];k++) + sum = MAC16_16(sum, in[k],*W++); + hidden[j] = tansig_approx(sum); + } + for (j=0;j<m->topo[2];j++) + { + int k; + opus_val32 sum = SHL32(EXTEND32(*W++),14); + for (k=0;k<m->topo[1];k++) + sum = MAC16_16(sum, hidden[k], *W++); + out[j] = tansig_approx(EXTRACT16(PSHR32(sum,17))); + } +} +#else +void mlp_process(const MLP *m, const float *in, float *out) +{ + int j; + float hidden[MAX_NEURONS]; + const float *W = m->weights; + /* Copy to tmp_in */ + for (j=0;j<m->topo[1];j++) + { + int k; + float sum = *W++; + for (k=0;k<m->topo[0];k++) + sum = sum + in[k]**W++; + hidden[j] = tansig_approx(sum); + } + for (j=0;j<m->topo[2];j++) + { + int k; + float sum = *W++; + for (k=0;k<m->topo[1];k++) + sum = sum + hidden[k]**W++; + out[j] = tansig_approx(sum); + } +} +#endif diff --git a/src/opus-1.1/src/mlp.h b/src/opus-1.1/src/mlp.h new file mode 100644 index 00000000..86c8e061 --- /dev/null +++ b/src/opus-1.1/src/mlp.h @@ -0,0 +1,41 @@ +/* Copyright (c) 2008-2011 Octasic Inc. + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef _MLP_H_ +#define _MLP_H_ + +#include "arch.h" + +typedef struct { + int layers; + const int *topo; + const float *weights; +} MLP; + +void mlp_process(const MLP *m, const float *in, float *out); + +#endif /* _MLP_H_ */ diff --git a/src/opus-1.1/src/mlp_data.c b/src/opus-1.1/src/mlp_data.c new file mode 100644 index 00000000..401c4c02 --- /dev/null +++ b/src/opus-1.1/src/mlp_data.c @@ -0,0 +1,105 @@ +/* The contents of this file was automatically generated by mlp_train.c + It contains multi-layer perceptron (MLP) weights. */ + +#include "mlp.h" + +/* RMS error was 0.138320, seed was 1361535663 */ + +static const float weights[422] = { + +/* hidden layer */ +-0.0941125f, -0.302976f, -0.603555f, -0.19393f, -0.185983f, +-0.601617f, -0.0465317f, -0.114563f, -0.103599f, -0.618938f, +-0.317859f, -0.169949f, -0.0702885f, 0.148065f, 0.409524f, +0.548432f, 0.367649f, -0.494393f, 0.764306f, -1.83957f, +0.170849f, 12.786f, -1.08848f, -1.27284f, -16.2606f, +24.1773f, -5.57454f, -0.17276f, -0.163388f, -0.224421f, +-0.0948944f, -0.0728695f, -0.26557f, -0.100283f, -0.0515459f, +-0.146142f, -0.120674f, -0.180655f, 0.12857f, 0.442138f, +-0.493735f, 0.167767f, 0.206699f, -0.197567f, 0.417999f, +1.50364f, -0.773341f, -10.0401f, 0.401872f, 2.97966f, +15.2165f, -1.88905f, -1.19254f, 0.0285397f, -0.00405139f, +0.0707565f, 0.00825699f, -0.0927269f, -0.010393f, -0.00428882f, +-0.00489743f, -0.0709731f, -0.00255992f, 0.0395619f, 0.226424f, +0.0325231f, 0.162175f, -0.100118f, 0.485789f, 0.12697f, +0.285937f, 0.0155637f, 0.10546f, 3.05558f, 1.15059f, +-1.00904f, -1.83088f, 3.31766f, -3.42516f, -0.119135f, +-0.0405654f, 0.00690068f, 0.0179877f, -0.0382487f, 0.00597941f, +-0.0183611f, 0.00190395f, -0.144322f, -0.0435671f, 0.000990594f, +0.221087f, 0.142405f, 0.484066f, 0.404395f, 0.511955f, +-0.237255f, 0.241742f, 0.35045f, -0.699428f, 10.3993f, +2.6507f, -2.43459f, -4.18838f, 1.05928f, 1.71067f, +0.00667811f, -0.0721335f, -0.0397346f, 0.0362704f, -0.11496f, +-0.0235776f, 0.0082161f, -0.0141741f, -0.0329699f, -0.0354253f, +0.00277404f, -0.290654f, -1.14767f, -0.319157f, -0.686544f, +0.36897f, 0.478899f, 0.182579f, -0.411069f, 0.881104f, +-4.60683f, 1.4697f, 0.335845f, -1.81905f, -30.1699f, +5.55225f, 0.0019508f, -0.123576f, -0.0727332f, -0.0641597f, +-0.0534458f, -0.108166f, -0.0937368f, -0.0697883f, -0.0275475f, +-0.192309f, -0.110074f, 0.285375f, -0.405597f, 0.0926724f, +-0.287881f, -0.851193f, -0.099493f, -0.233764f, -1.2852f, +1.13611f, 3.12168f, -0.0699f, -1.86216f, 2.65292f, +-7.31036f, 2.44776f, -0.00111802f, -0.0632786f, -0.0376296f, +-0.149851f, 0.142963f, 0.184368f, 0.123433f, 0.0756158f, +0.117312f, 0.0933395f, 0.0692163f, 0.0842592f, 0.0704683f, +0.0589963f, 0.0942205f, -0.448862f, 0.0262677f, 0.270352f, +-0.262317f, 0.172586f, 2.00227f, -0.159216f, 0.038422f, +10.2073f, 4.15536f, -2.3407f, -0.0550265f, 0.00964792f, +-0.141336f, 0.0274501f, 0.0343921f, -0.0487428f, 0.0950172f, +-0.00775017f, -0.0372492f, -0.00548121f, -0.0663695f, 0.0960506f, +-0.200008f, -0.0412827f, 0.58728f, 0.0515787f, 0.337254f, +0.855024f, 0.668371f, -0.114904f, -3.62962f, -0.467477f, +-0.215472f, 2.61537f, 0.406117f, -1.36373f, 0.0425394f, +0.12208f, 0.0934502f, 0.123055f, 0.0340935f, -0.142466f, +0.035037f, -0.0490666f, 0.0733208f, 0.0576672f, 0.123984f, +-0.0517194f, -0.253018f, 0.590565f, 0.145849f, 0.315185f, +0.221534f, -0.149081f, 0.216161f, -0.349575f, 24.5664f, +-0.994196f, 0.614289f, -18.7905f, -2.83277f, -0.716801f, +-0.347201f, 0.479515f, -0.246027f, 0.0758683f, 0.137293f, +-0.17781f, 0.118751f, -0.00108329f, -0.237334f, 0.355732f, +-0.12991f, -0.0547627f, -0.318576f, -0.325524f, 0.180494f, +-0.0625604f, 0.141219f, 0.344064f, 0.37658f, -0.591772f, +5.8427f, -0.38075f, 0.221894f, -1.41934f, -1.87943e+06f, +1.34114f, 0.0283355f, -0.0447856f, -0.0211466f, -0.0256927f, +0.0139618f, 0.0207934f, -0.0107666f, 0.0110969f, 0.0586069f, +-0.0253545f, -0.0328433f, 0.11872f, -0.216943f, 0.145748f, +0.119808f, -0.0915211f, -0.120647f, -0.0787719f, -0.143644f, +-0.595116f, -1.152f, -1.25335f, -1.17092f, 4.34023f, +-975268.f, -1.37033f, -0.0401123f, 0.210602f, -0.136656f, +0.135962f, -0.0523293f, 0.0444604f, 0.0143928f, 0.00412666f, +-0.0193003f, 0.218452f, -0.110204f, -2.02563f, 0.918238f, +-2.45362f, 1.19542f, -0.061362f, -1.92243f, 0.308111f, +0.49764f, 0.912356f, 0.209272f, -2.34525f, 2.19326f, +-6.47121f, 1.69771f, -0.725123f, 0.0118929f, 0.0377944f, +0.0554003f, 0.0226452f, -0.0704421f, -0.0300309f, 0.0122978f, +-0.0041782f, -0.0686612f, 0.0313115f, 0.039111f, 0.364111f, +-0.0945548f, 0.0229876f, -0.17414f, 0.329795f, 0.114714f, +0.30022f, 0.106997f, 0.132355f, 5.79932f, 0.908058f, +-0.905324f, -3.3561f, 0.190647f, 0.184211f, -0.673648f, +0.231807f, -0.0586222f, 0.230752f, -0.438277f, 0.245857f, +-0.17215f, 0.0876383f, -0.720512f, 0.162515f, 0.0170571f, +0.101781f, 0.388477f, 1.32931f, 1.08548f, -0.936301f, +-2.36958f, -6.71988f, -3.44376f, 2.13818f, 14.2318f, +4.91459f, -3.09052f, -9.69191f, -0.768234f, 1.79604f, +0.0549653f, 0.163399f, 0.0797025f, 0.0343933f, -0.0555876f, +-0.00505673f, 0.0187258f, 0.0326628f, 0.0231486f, 0.15573f, +0.0476223f, -0.254824f, 1.60155f, -0.801221f, 2.55496f, +0.737629f, -1.36249f, -0.695463f, -2.44301f, -1.73188f, +3.95279f, 1.89068f, 0.486087f, -11.3343f, 3.9416e+06f, + +/* output layer */ +-0.381439f, 0.12115f, -0.906927f, 2.93878f, 1.6388f, +0.882811f, 0.874344f, 1.21726f, -0.874545f, 0.321706f, +0.785055f, 0.946558f, -0.575066f, -3.46553f, 0.884905f, +0.0924047f, -9.90712f, 0.391338f, 0.160103f, -2.04954f, +4.1455f, 0.0684029f, -0.144761f, -0.285282f, 0.379244f, +-1.1584f, -0.0277241f, -9.85f, -4.82386f, 3.71333f, +3.87308f, 3.52558f}; + +static const int topo[3] = {25, 15, 2}; + +const MLP net = { + 3, + topo, + weights +}; diff --git a/src/opus-1.1/src/opus.c b/src/opus-1.1/src/opus.c new file mode 100644 index 00000000..30890b9c --- /dev/null +++ b/src/opus-1.1/src/opus.c @@ -0,0 +1,329 @@ +/* Copyright (c) 2011 Xiph.Org Foundation, Skype Limited + Written by Jean-Marc Valin and Koen Vos */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "opus.h" +#include "opus_private.h" + +#ifndef DISABLE_FLOAT_API +OPUS_EXPORT void opus_pcm_soft_clip(float *_x, int N, int C, float *declip_mem) +{ + int c; + int i; + float *x; + + if (C<1 || N<1 || !_x || !declip_mem) return; + + /* First thing: saturate everything to +/- 2 which is the highest level our + non-linearity can handle. At the point where the signal reaches +/-2, + the derivative will be zero anyway, so this doesn't introduce any + discontinuity in the derivative. */ + for (i=0;i<N*C;i++) + _x[i] = MAX16(-2.f, MIN16(2.f, _x[i])); + for (c=0;c<C;c++) + { + float a; + float x0; + int curr; + + x = _x+c; + a = declip_mem[c]; + /* Continue applying the non-linearity from the previous frame to avoid + any discontinuity. */ + for (i=0;i<N;i++) + { + if (x[i*C]*a>=0) + break; + x[i*C] = x[i*C]+a*x[i*C]*x[i*C]; + } + + curr=0; + x0 = x[0]; + while(1) + { + int start, end; + float maxval; + int special=0; + int peak_pos; + for (i=curr;i<N;i++) + { + if (x[i*C]>1 || x[i*C]<-1) + break; + } + if (i==N) + { + a=0; + break; + } + peak_pos = i; + start=end=i; + maxval=ABS16(x[i*C]); + /* Look for first zero crossing before clipping */ + while (start>0 && x[i*C]*x[(start-1)*C]>=0) + start--; + /* Look for first zero crossing after clipping */ + while (end<N && x[i*C]*x[end*C]>=0) + { + /* Look for other peaks until the next zero-crossing. */ + if (ABS16(x[end*C])>maxval) + { + maxval = ABS16(x[end*C]); + peak_pos = end; + } + end++; + } + /* Detect the special case where we clip before the first zero crossing */ + special = (start==0 && x[i*C]*x[0]>=0); + + /* Compute a such that maxval + a*maxval^2 = 1 */ + a=(maxval-1)/(maxval*maxval); + if (x[i*C]>0) + a = -a; + /* Apply soft clipping */ + for (i=start;i<end;i++) + x[i*C] = x[i*C]+a*x[i*C]*x[i*C]; + + if (special && peak_pos>=2) + { + /* Add a linear ramp from the first sample to the signal peak. + This avoids a discontinuity at the beginning of the frame. */ + float delta; + float offset = x0-x[0]; + delta = offset / peak_pos; + for (i=curr;i<peak_pos;i++) + { + offset -= delta; + x[i*C] += offset; + x[i*C] = MAX16(-1.f, MIN16(1.f, x[i*C])); + } + } + curr = end; + if (curr==N) + break; + } + declip_mem[c] = a; + } +} +#endif + +int encode_size(int size, unsigned char *data) +{ + if (size < 252) + { + data[0] = size; + return 1; + } else { + data[0] = 252+(size&0x3); + data[1] = (size-(int)data[0])>>2; + return 2; + } +} + +static int parse_size(const unsigned char *data, opus_int32 len, opus_int16 *size) +{ + if (len<1) + { + *size = -1; + return -1; + } else if (data[0]<252) + { + *size = data[0]; + return 1; + } else if (len<2) + { + *size = -1; + return -1; + } else { + *size = 4*data[1] + data[0]; + return 2; + } +} + +int opus_packet_parse_impl(const unsigned char *data, opus_int32 len, + int self_delimited, unsigned char *out_toc, + const unsigned char *frames[48], opus_int16 size[48], + int *payload_offset, opus_int32 *packet_offset) +{ + int i, bytes; + int count; + int cbr; + unsigned char ch, toc; + int framesize; + opus_int32 last_size; + opus_int32 pad = 0; + const unsigned char *data0 = data; + + if (size==NULL) + return OPUS_BAD_ARG; + + framesize = opus_packet_get_samples_per_frame(data, 48000); + + cbr = 0; + toc = *data++; + len--; + last_size = len; + switch (toc&0x3) + { + /* One frame */ + case 0: + count=1; + break; + /* Two CBR frames */ + case 1: + count=2; + cbr = 1; + if (!self_delimited) + { + if (len&0x1) + return OPUS_INVALID_PACKET; + last_size = len/2; + /* If last_size doesn't fit in size[0], we'll catch it later */ + size[0] = (opus_int16)last_size; + } + break; + /* Two VBR frames */ + case 2: + count = 2; + bytes = parse_size(data, len, size); + len -= bytes; + if (size[0]<0 || size[0] > len) + return OPUS_INVALID_PACKET; + data += bytes; + last_size = len-size[0]; + break; + /* Multiple CBR/VBR frames (from 0 to 120 ms) */ + default: /*case 3:*/ + if (len<1) + return OPUS_INVALID_PACKET; + /* Number of frames encoded in bits 0 to 5 */ + ch = *data++; + count = ch&0x3F; + if (count <= 0 || framesize*count > 5760) + return OPUS_INVALID_PACKET; + len--; + /* Padding flag is bit 6 */ + if (ch&0x40) + { + int p; + do { + int tmp; + if (len<=0) + return OPUS_INVALID_PACKET; + p = *data++; + len--; + tmp = p==255 ? 254: p; + len -= tmp; + pad += tmp; + } while (p==255); + } + if (len<0) + return OPUS_INVALID_PACKET; + /* VBR flag is bit 7 */ + cbr = !(ch&0x80); + if (!cbr) + { + /* VBR case */ + last_size = len; + for (i=0;i<count-1;i++) + { + bytes = parse_size(data, len, size+i); + len -= bytes; + if (size[i]<0 || size[i] > len) + return OPUS_INVALID_PACKET; + data += bytes; + last_size -= bytes+size[i]; + } + if (last_size<0) + return OPUS_INVALID_PACKET; + } else if (!self_delimited) + { + /* CBR case */ + last_size = len/count; + if (last_size*count!=len) + return OPUS_INVALID_PACKET; + for (i=0;i<count-1;i++) + size[i] = (opus_int16)last_size; + } + break; + } + /* Self-delimited framing has an extra size for the last frame. */ + if (self_delimited) + { + bytes = parse_size(data, len, size+count-1); + len -= bytes; + if (size[count-1]<0 || size[count-1] > len) + return OPUS_INVALID_PACKET; + data += bytes; + /* For CBR packets, apply the size to all the frames. */ + if (cbr) + { + if (size[count-1]*count > len) + return OPUS_INVALID_PACKET; + for (i=0;i<count-1;i++) + size[i] = size[count-1]; + } else if (bytes+size[count-1] > last_size) + return OPUS_INVALID_PACKET; + } else + { + /* Because it's not encoded explicitly, it's possible the size of the + last packet (or all the packets, for the CBR case) is larger than + 1275. Reject them here.*/ + if (last_size > 1275) + return OPUS_INVALID_PACKET; + size[count-1] = (opus_int16)last_size; + } + + if (payload_offset) + *payload_offset = (int)(data-data0); + + for (i=0;i<count;i++) + { + if (frames) + frames[i] = data; + data += size[i]; + } + + if (packet_offset) + *packet_offset = pad+(opus_int32)(data-data0); + + if (out_toc) + *out_toc = toc; + + return count; +} + +int opus_packet_parse(const unsigned char *data, opus_int32 len, + unsigned char *out_toc, const unsigned char *frames[48], + opus_int16 size[48], int *payload_offset) +{ + return opus_packet_parse_impl(data, len, 0, out_toc, + frames, size, payload_offset, NULL); +} + diff --git a/src/opus-1.0.2/src/opus_decoder.c b/src/opus-1.1/src/opus_decoder.c index ad5f7470..919ba521 100644 --- a/src/opus-1.0.2/src/opus_decoder.c +++ b/src/opus-1.1/src/opus_decoder.c @@ -26,11 +26,15 @@ */ #ifdef HAVE_CONFIG_H -#include "config.h" +# include "config.h" #endif #ifndef OPUS_BUILD -#error "OPUS_BUILD _MUST_ be defined to build Opus. This probably means you need other defines as well, as in a config.h. See the included build files for details." +# error "OPUS_BUILD _MUST_ be defined to build Opus. This probably means you need other defines as well, as in a config.h. See the included build files for details." +#endif + +#if defined(__GNUC__) && (__GNUC__ >= 2) && !defined(__OPTIMIZE__) +# pragma message "You appear to be compiling without optimization, if so opus will be very slow." #endif #include <stdarg.h> @@ -46,6 +50,7 @@ #include "structs.h" #include "define.h" #include "mathops.h" +#include "cpu_support.h" struct OpusDecoder { int celt_dec_offset; @@ -65,12 +70,15 @@ struct OpusDecoder { int frame_size; int prev_redundancy; int last_packet_duration; +#ifndef FIXED_POINT + opus_val16 softclip_mem[2]; +#endif opus_uint32 rangeFinal; }; #ifdef FIXED_POINT -static inline opus_int16 SAT16(opus_int32 x) { +static OPUS_INLINE opus_int16 SAT16(opus_int32 x) { return x > 32767 ? 32767 : x < -32768 ? -32768 : (opus_int16)x; } #endif @@ -201,8 +209,14 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, int i, silk_ret=0, celt_ret=0; ec_dec dec; opus_int32 silk_frame_size; + int pcm_silk_size; VARDECL(opus_int16, pcm_silk); - VARDECL(opus_val16, pcm_transition); + int pcm_transition_silk_size; + VARDECL(opus_val16, pcm_transition_silk); + int pcm_transition_celt_size; + VARDECL(opus_val16, pcm_transition_celt); + opus_val16 *pcm_transition; + int redundant_audio_size; VARDECL(opus_val16, redundant_audio); int audiosize; @@ -245,33 +259,61 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, ec_dec_init(&dec,(unsigned char*)data,len); } else { audiosize = frame_size; + mode = st->prev_mode; - if (st->prev_mode == 0) + if (mode == 0) { /* If we haven't got any packet yet, all we can do is return zeros */ for (i=0;i<audiosize*st->channels;i++) pcm[i] = 0; RESTORE_STACK; return audiosize; - } else { - mode = st->prev_mode; } - } - /* For CELT/hybrid PLC of more than 20 ms, opus_decode_native() will do - multiple calls */ - if (data==NULL && mode != MODE_SILK_ONLY) - frame_size = IMIN(frame_size, F20); - ALLOC(pcm_transition, F5*st->channels, opus_val16); + /* Avoids trying to run the PLC on sizes other than 2.5 (CELT), 5 (CELT), + 10, or 20 (e.g. 12.5 or 30 ms). */ + if (audiosize > F20) + { + do { + int ret = opus_decode_frame(st, NULL, 0, pcm, IMIN(audiosize, F20), 0); + if (ret<0) + { + RESTORE_STACK; + return ret; + } + pcm += ret*st->channels; + audiosize -= ret; + } while (audiosize > 0); + RESTORE_STACK; + return frame_size; + } else if (audiosize < F20) + { + if (audiosize > F10) + audiosize = F10; + else if (mode != MODE_SILK_ONLY && audiosize > F5 && audiosize < F10) + audiosize = F5; + } + } + pcm_transition_silk_size = ALLOC_NONE; + pcm_transition_celt_size = ALLOC_NONE; if (data!=NULL && st->prev_mode > 0 && ( (mode == MODE_CELT_ONLY && st->prev_mode != MODE_CELT_ONLY && !st->prev_redundancy) || (mode != MODE_CELT_ONLY && st->prev_mode == MODE_CELT_ONLY) ) ) { transition = 1; + /* Decide where to allocate the stack memory for pcm_transition */ if (mode == MODE_CELT_ONLY) - opus_decode_frame(st, NULL, 0, pcm_transition, IMIN(F5, audiosize), 0); + pcm_transition_celt_size = F5*st->channels; + else + pcm_transition_silk_size = F5*st->channels; + } + ALLOC(pcm_transition_celt, pcm_transition_celt_size, opus_val16); + if (transition && mode == MODE_CELT_ONLY) + { + pcm_transition = pcm_transition_celt; + opus_decode_frame(st, NULL, 0, pcm_transition, IMIN(F5, audiosize), 0); } if (audiosize > frame_size) { @@ -282,8 +324,9 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, frame_size = audiosize; } - ALLOC(pcm_silk, IMAX(F10, frame_size)*st->channels, opus_int16); - ALLOC(redundant_audio, F5*st->channels, opus_val16); + /* Don't allocate any memory when in CELT-only mode */ + pcm_silk_size = (mode != MODE_CELT_ONLY) ? IMAX(F10, frame_size)*st->channels : ALLOC_NONE; + ALLOC(pcm_silk, pcm_silk_size, opus_int16); /* SILK processing */ if (mode != MODE_CELT_ONLY) @@ -332,7 +375,7 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, pcm_ptr[i] = 0; } else { RESTORE_STACK; - return OPUS_INVALID_PACKET; + return OPUS_INTERNAL_ERROR; } } pcm_ptr += silk_frame_size * st->channels; @@ -397,10 +440,22 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, } if (redundancy) + { transition = 0; + pcm_transition_silk_size=ALLOC_NONE; + } + + ALLOC(pcm_transition_silk, pcm_transition_silk_size, opus_val16); if (transition && mode != MODE_CELT_ONLY) + { + pcm_transition = pcm_transition_silk; opus_decode_frame(st, NULL, 0, pcm_transition, IMIN(F5, audiosize), 0); + } + + /* Only allocation memory for redundancy if/when needed */ + redundant_audio_size = redundancy ? F5*st->channels : ALLOC_NONE; + ALLOC(redundant_audio, redundant_audio_size, opus_val16); /* 5 ms redundant frame for CELT->SILK*/ if (redundancy && celt_to_silk) @@ -514,197 +569,28 @@ static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, st->prev_mode = mode; st->prev_redundancy = redundancy && !celt_to_silk; - RESTORE_STACK; - return celt_ret < 0 ? celt_ret : audiosize; - -} - -static int parse_size(const unsigned char *data, opus_int32 len, short *size) -{ - if (len<1) - { - *size = -1; - return -1; - } else if (data[0]<252) - { - *size = data[0]; - return 1; - } else if (len<2) - { - *size = -1; - return -1; - } else { - *size = 4*data[1] + data[0]; - return 2; - } -} - -static int opus_packet_parse_impl(const unsigned char *data, opus_int32 len, - int self_delimited, unsigned char *out_toc, - const unsigned char *frames[48], short size[48], int *payload_offset) -{ - int i, bytes; - int count; - int cbr; - unsigned char ch, toc; - int framesize; - opus_int32 last_size; - const unsigned char *data0 = data; - - if (size==NULL) - return OPUS_BAD_ARG; - - framesize = opus_packet_get_samples_per_frame(data, 48000); - - cbr = 0; - toc = *data++; - len--; - last_size = len; - switch (toc&0x3) - { - /* One frame */ - case 0: - count=1; - break; - /* Two CBR frames */ - case 1: - count=2; - cbr = 1; - if (!self_delimited) - { - if (len&0x1) - return OPUS_INVALID_PACKET; - last_size = len/2; - /* If last_size doesn't fit in size[0], we'll catch it later */ - size[0] = (short)last_size; - } - break; - /* Two VBR frames */ - case 2: - count = 2; - bytes = parse_size(data, len, size); - len -= bytes; - if (size[0]<0 || size[0] > len) - return OPUS_INVALID_PACKET; - data += bytes; - last_size = len-size[0]; - break; - /* Multiple CBR/VBR frames (from 0 to 120 ms) */ - default: /*case 3:*/ - if (len<1) - return OPUS_INVALID_PACKET; - /* Number of frames encoded in bits 0 to 5 */ - ch = *data++; - count = ch&0x3F; - if (count <= 0 || framesize*count > 5760) - return OPUS_INVALID_PACKET; - len--; - /* Padding flag is bit 6 */ - if (ch&0x40) - { - int p; - do { - if (len<=0) - return OPUS_INVALID_PACKET; - p = *data++; - len--; - len -= p==255 ? 254: p; - } while (p==255); - } - if (len<0) - return OPUS_INVALID_PACKET; - /* VBR flag is bit 7 */ - cbr = !(ch&0x80); - if (!cbr) - { - /* VBR case */ - last_size = len; - for (i=0;i<count-1;i++) - { - bytes = parse_size(data, len, size+i); - len -= bytes; - if (size[i]<0 || size[i] > len) - return OPUS_INVALID_PACKET; - data += bytes; - last_size -= bytes+size[i]; - } - if (last_size<0) - return OPUS_INVALID_PACKET; - } else if (!self_delimited) - { - /* CBR case */ - last_size = len/count; - if (last_size*count!=len) - return OPUS_INVALID_PACKET; - for (i=0;i<count-1;i++) - size[i] = (short)last_size; - } - break; - } - /* Self-delimited framing has an extra size for the last frame. */ - if (self_delimited) - { - bytes = parse_size(data, len, size+count-1); - len -= bytes; - if (size[count-1]<0 || size[count-1] > len) - return OPUS_INVALID_PACKET; - data += bytes; - /* For CBR packets, apply the size to all the frames. */ - if (cbr) - { - if (size[count-1]*count > len) - return OPUS_INVALID_PACKET; - for (i=0;i<count-1;i++) - size[i] = size[count-1]; - } else if(size[count-1] > last_size) - return OPUS_INVALID_PACKET; - } else - { - /* Because it's not encoded explicitly, it's possible the size of the - last packet (or all the packets, for the CBR case) is larger than - 1275. Reject them here.*/ - if (last_size > 1275) - return OPUS_INVALID_PACKET; - size[count-1] = (short)last_size; - } - if (frames) + if (celt_ret>=0) { - for (i=0;i<count;i++) - { - frames[i] = data; - data += size[i]; - } + if (OPUS_CHECK_ARRAY(pcm, audiosize*st->channels)) + OPUS_PRINT_INT(audiosize); } - if (out_toc) - *out_toc = toc; - - if (payload_offset) - *payload_offset = data-data0; - - return count; -} + RESTORE_STACK; + return celt_ret < 0 ? celt_ret : audiosize; -int opus_packet_parse(const unsigned char *data, opus_int32 len, - unsigned char *out_toc, const unsigned char *frames[48], - short size[48], int *payload_offset) -{ - return opus_packet_parse_impl(data, len, 0, out_toc, - frames, size, payload_offset); } int opus_decode_native(OpusDecoder *st, const unsigned char *data, opus_int32 len, opus_val16 *pcm, int frame_size, int decode_fec, - int self_delimited, int *packet_offset) + int self_delimited, opus_int32 *packet_offset, int soft_clip) { int i, nb_samples; int count, offset; unsigned char toc; - int tot_offset; int packet_frame_size, packet_bandwidth, packet_mode, packet_stream_channels; /* 48 x 2.5 ms = 120 ms */ - short size[48]; + opus_int16 size[48]; if (decode_fec<0 || decode_fec>1) return OPUS_BAD_ARG; /* For FEC/PLC, frame_size has to be to have a multiple of 2.5 ms */ @@ -715,12 +601,14 @@ int opus_decode_native(OpusDecoder *st, const unsigned char *data, int pcm_count=0; do { int ret; - ret = opus_decode_frame(st, NULL, 0, pcm, frame_size-pcm_count, 0); + ret = opus_decode_frame(st, NULL, 0, pcm+pcm_count*st->channels, frame_size-pcm_count, 0); if (ret<0) return ret; - pcm += st->channels*ret; pcm_count += ret; } while (pcm_count < frame_size); + celt_assert(pcm_count == frame_size); + if (OPUS_CHECK_ARRAY(pcm, pcm_count*st->channels)) + OPUS_PRINT_INT(pcm_count); st->last_packet_duration = pcm_count; return pcm_count; } else if (len<0) @@ -731,7 +619,10 @@ int opus_decode_native(OpusDecoder *st, const unsigned char *data, packet_frame_size = opus_packet_get_samples_per_frame(data, st->Fs); packet_stream_channels = opus_packet_get_nb_channels(data); - count = opus_packet_parse_impl(data, len, self_delimited, &toc, NULL, size, &offset); + count = opus_packet_parse_impl(data, len, self_delimited, &toc, NULL, + size, &offset, packet_offset); + if (count<0) + return count; data += offset; @@ -740,15 +631,19 @@ int opus_decode_native(OpusDecoder *st, const unsigned char *data, int duration_copy; int ret; /* If no FEC can be present, run the PLC (recursive call) */ - if (frame_size <= packet_frame_size || packet_mode == MODE_CELT_ONLY || st->mode == MODE_CELT_ONLY) - return opus_decode_native(st, NULL, 0, pcm, frame_size, 0, 0, NULL); + if (frame_size < packet_frame_size || packet_mode == MODE_CELT_ONLY || st->mode == MODE_CELT_ONLY) + return opus_decode_native(st, NULL, 0, pcm, frame_size, 0, 0, NULL, soft_clip); /* Otherwise, run the PLC on everything except the size for which we might have FEC */ duration_copy = st->last_packet_duration; - ret = opus_decode_native(st, NULL, 0, pcm, frame_size-packet_frame_size, 0, 0, NULL); - if (ret<0) + if (frame_size-packet_frame_size!=0) { - st->last_packet_duration = duration_copy; - return ret; + ret = opus_decode_native(st, NULL, 0, pcm, frame_size-packet_frame_size, 0, 0, NULL, soft_clip); + if (ret<0) + { + st->last_packet_duration = duration_copy; + return ret; + } + celt_assert(ret==frame_size-packet_frame_size); } /* Complete with FEC */ st->mode = packet_mode; @@ -759,14 +654,13 @@ int opus_decode_native(OpusDecoder *st, const unsigned char *data, packet_frame_size, 1); if (ret<0) return ret; - st->last_packet_duration = frame_size; - return frame_size; + else { + if (OPUS_CHECK_ARRAY(pcm, frame_size*st->channels)) + OPUS_PRINT_INT(frame_size); + st->last_packet_duration = frame_size; + return frame_size; + } } - tot_offset = 0; - if (count < 0) - return count; - - tot_offset += offset; if (count*packet_frame_size > frame_size) return OPUS_BUFFER_TOO_SMALL; @@ -781,17 +675,22 @@ int opus_decode_native(OpusDecoder *st, const unsigned char *data, for (i=0;i<count;i++) { int ret; - ret = opus_decode_frame(st, data, size[i], pcm, frame_size-nb_samples, decode_fec); + ret = opus_decode_frame(st, data, size[i], pcm+nb_samples*st->channels, frame_size-nb_samples, 0); if (ret<0) return ret; + celt_assert(ret==packet_frame_size); data += size[i]; - tot_offset += size[i]; - pcm += ret*st->channels; nb_samples += ret; } - if (packet_offset != NULL) - *packet_offset = tot_offset; st->last_packet_duration = nb_samples; + if (OPUS_CHECK_ARRAY(pcm, nb_samples*st->channels)) + OPUS_PRINT_INT(nb_samples); +#ifndef FIXED_POINT + if (soft_clip) + opus_pcm_soft_clip(pcm, nb_samples, st->channels, st->softclip_mem); + else + st->softclip_mem[0]=st->softclip_mem[1]=0; +#endif return nb_samples; } @@ -800,7 +699,9 @@ int opus_decode_native(OpusDecoder *st, const unsigned char *data, int opus_decode(OpusDecoder *st, const unsigned char *data, opus_int32 len, opus_val16 *pcm, int frame_size, int decode_fec) { - return opus_decode_native(st, data, len, pcm, frame_size, decode_fec, 0, NULL); + if(frame_size<=0) + return OPUS_BAD_ARG; + return opus_decode_native(st, data, len, pcm, frame_size, decode_fec, 0, NULL, 0); } #ifndef DISABLE_FLOAT_API @@ -811,9 +712,14 @@ int opus_decode_float(OpusDecoder *st, const unsigned char *data, int ret, i; ALLOC_STACK; + if(frame_size<=0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } ALLOC(out, frame_size*st->channels, opus_int16); - ret = opus_decode_native(st, data, len, out, frame_size, decode_fec, 0, NULL); + ret = opus_decode_native(st, data, len, out, frame_size, decode_fec, 0, NULL, 0); if (ret > 0) { for (i=0;i<ret*st->channels;i++) @@ -833,7 +739,7 @@ int opus_decode(OpusDecoder *st, const unsigned char *data, int ret, i; ALLOC_STACK; - if(frame_size<0) + if(frame_size<=0) { RESTORE_STACK; return OPUS_BAD_ARG; @@ -841,7 +747,7 @@ int opus_decode(OpusDecoder *st, const unsigned char *data, ALLOC(out, frame_size*st->channels, float); - ret = opus_decode_native(st, data, len, out, frame_size, decode_fec, 0, NULL); + ret = opus_decode_native(st, data, len, out, frame_size, decode_fec, 0, NULL, 1); if (ret > 0) { for (i=0;i<ret*st->channels;i++) @@ -854,7 +760,9 @@ int opus_decode(OpusDecoder *st, const unsigned char *data, int opus_decode_float(OpusDecoder *st, const unsigned char *data, opus_int32 len, opus_val16 *pcm, int frame_size, int decode_fec) { - return opus_decode_native(st, data, len, pcm, frame_size, decode_fec, 0, NULL); + if(frame_size<=0) + return OPUS_BAD_ARG; + return opus_decode_native(st, data, len, pcm, frame_size, decode_fec, 0, NULL, 0); } #endif @@ -877,12 +785,20 @@ int opus_decoder_ctl(OpusDecoder *st, int request, ...) case OPUS_GET_BANDWIDTH_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } *value = st->bandwidth; } break; case OPUS_GET_FINAL_RANGE_REQUEST: { opus_uint32 *value = va_arg(ap, opus_uint32*); + if (!value) + { + goto bad_arg; + } *value = st->rangeFinal; } break; @@ -901,10 +817,9 @@ int opus_decoder_ctl(OpusDecoder *st, int request, ...) case OPUS_GET_SAMPLE_RATE_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); - if (value==NULL) + if (!value) { - ret = OPUS_BAD_ARG; - break; + goto bad_arg; } *value = st->Fs; } @@ -912,10 +827,9 @@ int opus_decoder_ctl(OpusDecoder *st, int request, ...) case OPUS_GET_PITCH_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); - if (value==NULL) + if (!value) { - ret = OPUS_BAD_ARG; - break; + goto bad_arg; } if (st->prev_mode == MODE_CELT_ONLY) celt_decoder_ctl(celt_dec, OPUS_GET_PITCH(value)); @@ -926,10 +840,9 @@ int opus_decoder_ctl(OpusDecoder *st, int request, ...) case OPUS_GET_GAIN_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); - if (value==NULL) + if (!value) { - ret = OPUS_BAD_ARG; - break; + goto bad_arg; } *value = st->decode_gain; } @@ -939,8 +852,7 @@ int opus_decoder_ctl(OpusDecoder *st, int request, ...) opus_int32 value = va_arg(ap, opus_int32); if (value<-32768 || value>32767) { - ret = OPUS_BAD_ARG; - break; + goto bad_arg; } st->decode_gain = value; } @@ -948,6 +860,10 @@ int opus_decoder_ctl(OpusDecoder *st, int request, ...) case OPUS_GET_LAST_PACKET_DURATION_REQUEST: { opus_uint32 *value = va_arg(ap, opus_uint32*); + if (!value) + { + goto bad_arg; + } *value = st->last_packet_duration; } break; @@ -959,6 +875,9 @@ int opus_decoder_ctl(OpusDecoder *st, int request, ...) va_end(ap); return ret; +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; } void opus_decoder_destroy(OpusDecoder *st) diff --git a/src/opus-1.0.2/src/opus_encoder.c b/src/opus-1.1/src/opus_encoder.c index aae31256..fbd3de63 100644 --- a/src/opus-1.0.2/src/opus_encoder.c +++ b/src/opus-1.1/src/opus_encoder.c @@ -40,7 +40,9 @@ #include "arch.h" #include "opus_private.h" #include "os_support.h" - +#include "cpu_support.h" +#include "analysis.h" +#include "mathops.h" #include "tuning_parameters.h" #ifdef FIXED_POINT #include "fixed/structs_FIX.h" @@ -50,6 +52,12 @@ #define MAX_ENCODER_BUFFER 480 +typedef struct { + opus_val32 XX, XY, YY; + opus_val16 smoothed_width; + opus_val16 max_follower; +} StereoWidthState; + struct OpusEncoder { int celt_enc_offset; int silk_enc_offset; @@ -66,14 +74,18 @@ struct OpusEncoder { opus_int32 Fs; int use_vbr; int vbr_constraint; + int variable_duration; opus_int32 bitrate_bps; opus_int32 user_bitrate_bps; + int lsb_depth; int encoder_buffer; + int lfe; #define OPUS_ENCODER_RESET_START stream_channels int stream_channels; opus_int16 hybrid_stereo_width_Q14; opus_int32 variable_HP_smth2_Q15; + opus_val16 prev_HB_gain; opus_val32 hp_mem[4]; int mode; int prev_mode; @@ -83,9 +95,16 @@ struct OpusEncoder { int silk_bw_switch; /* Sampling rate (at the API level) */ int first; + opus_val16 * energy_masking; + StereoWidthState width_mem; opus_val16 delay_buffer[MAX_ENCODER_BUFFER*2]; - +#ifndef DISABLE_FLOAT_API + TonalityAnalysisState analysis; + int detected_bandwidth; + int analysis_offset; +#endif opus_uint32 rangeFinal; + int arch; }; /* Transition tables for the voice and music. First column is the @@ -94,36 +113,36 @@ struct OpusEncoder { static const opus_int32 mono_voice_bandwidth_thresholds[8] = { 11000, 1000, /* NB<->MB */ 14000, 1000, /* MB<->WB */ - 21000, 2000, /* WB<->SWB */ - 29000, 2000, /* SWB<->FB */ + 17000, 1000, /* WB<->SWB */ + 21000, 2000, /* SWB<->FB */ }; static const opus_int32 mono_music_bandwidth_thresholds[8] = { - 14000, 1000, /* MB not allowed */ - 18000, 2000, /* MB<->WB */ - 24000, 2000, /* WB<->SWB */ - 33000, 2000, /* SWB<->FB */ + 12000, 1000, /* NB<->MB */ + 15000, 1000, /* MB<->WB */ + 18000, 2000, /* WB<->SWB */ + 22000, 2000, /* SWB<->FB */ }; static const opus_int32 stereo_voice_bandwidth_thresholds[8] = { 11000, 1000, /* NB<->MB */ 14000, 1000, /* MB<->WB */ 21000, 2000, /* WB<->SWB */ - 32000, 2000, /* SWB<->FB */ + 28000, 2000, /* SWB<->FB */ }; static const opus_int32 stereo_music_bandwidth_thresholds[8] = { - 14000, 1000, /* MB not allowed */ + 12000, 1000, /* NB<->MB */ 18000, 2000, /* MB<->WB */ - 24000, 2000, /* WB<->SWB */ - 48000, 2000, /* SWB<->FB */ + 21000, 2000, /* WB<->SWB */ + 30000, 2000, /* SWB<->FB */ }; /* Threshold bit-rates for switching between mono and stereo */ -static const opus_int32 stereo_voice_threshold = 26000; -static const opus_int32 stereo_music_threshold = 36000; +static const opus_int32 stereo_voice_threshold = 30000; +static const opus_int32 stereo_music_threshold = 30000; /* Threshold bit-rate for switching between SILK/hybrid and CELT-only */ static const opus_int32 mode_thresholds[2][2] = { /* voice */ /* music */ - { 48000, 24000}, /* mono */ - { 48000, 24000}, /* stereo */ + { 64000, 16000}, /* mono */ + { 36000, 16000}, /* stereo */ }; int opus_encoder_get_size(int channels) @@ -167,7 +186,9 @@ int opus_encoder_init(OpusEncoder* st, opus_int32 Fs, int channels, int applicat st->Fs = Fs; - ret = silk_InitEncoder( silk_enc, &st->silk_mode ); + st->arch = opus_select_arch(); + + ret = silk_InitEncoder( silk_enc, st->arch, &st->silk_mode ); if(ret)return OPUS_INTERNAL_ERROR; /* default SILK parameters */ @@ -180,18 +201,19 @@ int opus_encoder_init(OpusEncoder* st, opus_int32 Fs, int channels, int applicat st->silk_mode.payloadSize_ms = 20; st->silk_mode.bitRate = 25000; st->silk_mode.packetLossPercentage = 0; - st->silk_mode.complexity = 10; + st->silk_mode.complexity = 9; st->silk_mode.useInBandFEC = 0; st->silk_mode.useDTX = 0; st->silk_mode.useCBR = 0; + st->silk_mode.reducedDependency = 0; /* Create CELT encoder */ /* Initialize CELT encoder */ - err = celt_encoder_init(celt_enc, Fs, channels); + err = celt_encoder_init(celt_enc, Fs, channels, st->arch); if(err!=OPUS_OK)return OPUS_INTERNAL_ERROR; celt_encoder_ctl(celt_enc, CELT_SET_SIGNALLING(0)); - celt_encoder_ctl(celt_enc, OPUS_SET_COMPLEXITY(10)); + celt_encoder_ctl(celt_enc, OPUS_SET_COMPLEXITY(st->silk_mode.complexity)); st->use_vbr = 1; /* Makes constrained VBR the default (safer for real-time use) */ @@ -206,12 +228,15 @@ int opus_encoder_init(OpusEncoder* st, opus_int32 Fs, int channels, int applicat st->user_forced_mode = OPUS_AUTO; st->voice_ratio = -1; st->encoder_buffer = st->Fs/100; + st->lsb_depth = 24; + st->variable_duration = OPUS_FRAMESIZE_ARG; /* Delay compensation of 4 ms (2.5 ms for SILK's extra look-ahead + 1.5 ms for SILK resamplers and stereo prediction) */ st->delay_compensation = st->Fs/250; st->hybrid_stereo_width_Q14 = 1 << 14; + st->prev_HB_gain = Q15ONE; st->variable_HP_smth2_Q15 = silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 ); st->first = 1; st->mode = MODE_HYBRID; @@ -220,44 +245,6 @@ int opus_encoder_init(OpusEncoder* st, opus_int32 Fs, int channels, int applicat return OPUS_OK; } -static int pad_frame(unsigned char *data, opus_int32 len, opus_int32 new_len) -{ - if (len == new_len) - return 0; - if (len > new_len) - return 1; - - if ((data[0]&0x3)==0) - { - int i; - int padding, nb_255s; - - padding = new_len - len; - if (padding >= 2) - { - nb_255s = (padding-2)/255; - - for (i=len-1;i>=1;i--) - data[i+nb_255s+2] = data[i]; - data[0] |= 0x3; - data[1] = 0x41; - for (i=0;i<nb_255s;i++) - data[i+2] = 255; - data[nb_255s+2] = padding-255*nb_255s-2; - for (i=len+3+nb_255s;i<new_len;i++) - data[i] = 0; - } else { - for (i=len-1;i>=1;i--) - data[i+1] = data[i]; - data[0] |= 0x3; - data[1] = 1; - } - return 0; - } else { - return 1; - } -} - static unsigned char gen_toc(int mode, int framerate, int bandwidth, int channels) { int period; @@ -322,7 +309,7 @@ static void silk_biquad_float( S[ 0 ] = S[1] - vout*A[0] + B[1]*inval; - S[ 1 ] = - vout*A[1] + B[2]*inval; + S[ 1 ] = - vout*A[1] + B[2]*inval + VERY_SMALL; /* Scale back to Q0 and saturate */ out[ k*stride ] = vout; @@ -365,6 +352,56 @@ static void hp_cutoff(const opus_val16 *in, opus_int32 cutoff_Hz, opus_val16 *ou #endif } +#ifdef FIXED_POINT +static void dc_reject(const opus_val16 *in, opus_int32 cutoff_Hz, opus_val16 *out, opus_val32 *hp_mem, int len, int channels, opus_int32 Fs) +{ + int c, i; + int shift; + + /* Approximates -round(log2(4.*cutoff_Hz/Fs)) */ + shift=celt_ilog2(Fs/(cutoff_Hz*3)); + for (c=0;c<channels;c++) + { + for (i=0;i<len;i++) + { + opus_val32 x, tmp, y; + x = SHL32(EXTEND32(in[channels*i+c]), 15); + /* First stage */ + tmp = x-hp_mem[2*c]; + hp_mem[2*c] = hp_mem[2*c] + PSHR32(x - hp_mem[2*c], shift); + /* Second stage */ + y = tmp - hp_mem[2*c+1]; + hp_mem[2*c+1] = hp_mem[2*c+1] + PSHR32(tmp - hp_mem[2*c+1], shift); + out[channels*i+c] = EXTRACT16(SATURATE(PSHR32(y, 15), 32767)); + } + } +} + +#else +static void dc_reject(const opus_val16 *in, opus_int32 cutoff_Hz, opus_val16 *out, opus_val32 *hp_mem, int len, int channels, opus_int32 Fs) +{ + int c, i; + float coef; + + coef = 4.0f*cutoff_Hz/Fs; + for (c=0;c<channels;c++) + { + for (i=0;i<len;i++) + { + opus_val32 x, tmp, y; + x = in[channels*i+c]; + /* First stage */ + tmp = x-hp_mem[2*c]; + hp_mem[2*c] = hp_mem[2*c] + coef*(x - hp_mem[2*c]) + VERY_SMALL; + /* Second stage */ + y = tmp - hp_mem[2*c+1]; + hp_mem[2*c+1] = hp_mem[2*c+1] + coef*(tmp - hp_mem[2*c+1]) + VERY_SMALL; + out[channels*i+c] = y; + } + } +} +#endif + static void stereo_fade(const opus_val16 *in, opus_val16 *out, opus_val16 g1, opus_val16 g2, int overlap48, int frame_size, int channels, const opus_val16 *window, opus_int32 Fs) { @@ -397,6 +434,45 @@ static void stereo_fade(const opus_val16 *in, opus_val16 *out, opus_val16 g1, op } } +static void gain_fade(const opus_val16 *in, opus_val16 *out, opus_val16 g1, opus_val16 g2, + int overlap48, int frame_size, int channels, const opus_val16 *window, opus_int32 Fs) +{ + int i; + int inc; + int overlap; + int c; + inc = 48000/Fs; + overlap=overlap48/inc; + if (channels==1) + { + for (i=0;i<overlap;i++) + { + opus_val16 g, w; + w = MULT16_16_Q15(window[i*inc], window[i*inc]); + g = SHR32(MAC16_16(MULT16_16(w,g2), + Q15ONE-w, g1), 15); + out[i] = MULT16_16_Q15(g, in[i]); + } + } else { + for (i=0;i<overlap;i++) + { + opus_val16 g, w; + w = MULT16_16_Q15(window[i*inc], window[i*inc]); + g = SHR32(MAC16_16(MULT16_16(w,g2), + Q15ONE-w, g1), 15); + out[i*2] = MULT16_16_Q15(g, in[i*2]); + out[i*2+1] = MULT16_16_Q15(g, in[i*2+1]); + } + } + c=0;do { + for (i=overlap;i<frame_size;i++) + { + out[i*channels+c] = MULT16_16_Q15(g2, in[i*channels+c]); + } + } + while (++c<channels); +} + OpusEncoder *opus_encoder_create(opus_int32 Fs, int channels, int application, int *error) { int ret; @@ -438,15 +514,417 @@ static opus_int32 user_bitrate_to_bitrate(OpusEncoder *st, int frame_size, int m return st->user_bitrate_bps; } +#ifndef DISABLE_FLOAT_API +/* Don't use more than 60 ms for the frame size analysis */ +#define MAX_DYNAMIC_FRAMESIZE 24 +/* Estimates how much the bitrate will be boosted based on the sub-frame energy */ +static float transient_boost(const float *E, const float *E_1, int LM, int maxM) +{ + int i; + int M; + float sumE=0, sumE_1=0; + float metric; + + M = IMIN(maxM, (1<<LM)+1); + for (i=0;i<M;i++) + { + sumE += E[i]; + sumE_1 += E_1[i]; + } + metric = sumE*sumE_1/(M*M); + /*if (LM==3) + printf("%f\n", metric);*/ + /*return metric>10 ? 1 : 0;*/ + /*return MAX16(0,1-exp(-.25*(metric-2.)));*/ + return MIN16(1,(float)sqrt(MAX16(0,.05f*(metric-2)))); +} + +/* Viterbi decoding trying to find the best frame size combination using look-ahead + + State numbering: + 0: unused + 1: 2.5 ms + 2: 5 ms (#1) + 3: 5 ms (#2) + 4: 10 ms (#1) + 5: 10 ms (#2) + 6: 10 ms (#3) + 7: 10 ms (#4) + 8: 20 ms (#1) + 9: 20 ms (#2) + 10: 20 ms (#3) + 11: 20 ms (#4) + 12: 20 ms (#5) + 13: 20 ms (#6) + 14: 20 ms (#7) + 15: 20 ms (#8) +*/ +static int transient_viterbi(const float *E, const float *E_1, int N, int frame_cost, int rate) +{ + int i; + float cost[MAX_DYNAMIC_FRAMESIZE][16]; + int states[MAX_DYNAMIC_FRAMESIZE][16]; + float best_cost; + int best_state; + float factor; + /* Take into account that we damp VBR in the 32 kb/s to 64 kb/s range. */ + if (rate<80) + factor=0; + else if (rate>160) + factor=1; + else + factor = (rate-80.f)/80.f; + /* Makes variable framesize less aggressive at lower bitrates, but I can't + find any valid theoretical justification for this (other than it seems + to help) */ + for (i=0;i<16;i++) + { + /* Impossible state */ + states[0][i] = -1; + cost[0][i] = 1e10; + } + for (i=0;i<4;i++) + { + cost[0][1<<i] = (frame_cost + rate*(1<<i))*(1+factor*transient_boost(E, E_1, i, N+1)); + states[0][1<<i] = i; + } + for (i=1;i<N;i++) + { + int j; + + /* Follow continuations */ + for (j=2;j<16;j++) + { + cost[i][j] = cost[i-1][j-1]; + states[i][j] = j-1; + } + + /* New frames */ + for(j=0;j<4;j++) + { + int k; + float min_cost; + float curr_cost; + states[i][1<<j] = 1; + min_cost = cost[i-1][1]; + for(k=1;k<4;k++) + { + float tmp = cost[i-1][(1<<(k+1))-1]; + if (tmp < min_cost) + { + states[i][1<<j] = (1<<(k+1))-1; + min_cost = tmp; + } + } + curr_cost = (frame_cost + rate*(1<<j))*(1+factor*transient_boost(E+i, E_1+i, j, N-i+1)); + cost[i][1<<j] = min_cost; + /* If part of the frame is outside the analysis window, only count part of the cost */ + if (N-i < (1<<j)) + cost[i][1<<j] += curr_cost*(float)(N-i)/(1<<j); + else + cost[i][1<<j] += curr_cost; + } + } + + best_state=1; + best_cost = cost[N-1][1]; + /* Find best end state (doesn't force a frame to end at N-1) */ + for (i=2;i<16;i++) + { + if (cost[N-1][i]<best_cost) + { + best_cost = cost[N-1][i]; + best_state = i; + } + } + + /* Follow transitions back */ + for (i=N-1;i>=0;i--) + { + /*printf("%d ", best_state);*/ + best_state = states[i][best_state]; + } + /*printf("%d\n", best_state);*/ + return best_state; +} + +int optimize_framesize(const opus_val16 *x, int len, int C, opus_int32 Fs, + int bitrate, opus_val16 tonality, float *mem, int buffering, + downmix_func downmix) +{ + int N; + int i; + float e[MAX_DYNAMIC_FRAMESIZE+4]; + float e_1[MAX_DYNAMIC_FRAMESIZE+3]; + opus_val32 memx; + int bestLM=0; + int subframe; + int pos; + VARDECL(opus_val32, sub); + + subframe = Fs/400; + ALLOC(sub, subframe, opus_val32); + e[0]=mem[0]; + e_1[0]=1.f/(EPSILON+mem[0]); + if (buffering) + { + /* Consider the CELT delay when not in restricted-lowdelay */ + /* We assume the buffering is between 2.5 and 5 ms */ + int offset = 2*subframe - buffering; + celt_assert(offset>=0 && offset <= subframe); + x += C*offset; + len -= offset; + e[1]=mem[1]; + e_1[1]=1.f/(EPSILON+mem[1]); + e[2]=mem[2]; + e_1[2]=1.f/(EPSILON+mem[2]); + pos = 3; + } else { + pos=1; + } + N=IMIN(len/subframe, MAX_DYNAMIC_FRAMESIZE); + /* Just silencing a warning, it's really initialized later */ + memx = 0; + for (i=0;i<N;i++) + { + float tmp; + opus_val32 tmpx; + int j; + tmp=EPSILON; + + downmix(x, sub, subframe, i*subframe, 0, -2, C); + if (i==0) + memx = sub[0]; + for (j=0;j<subframe;j++) + { + tmpx = sub[j]; + tmp += (tmpx-memx)*(float)(tmpx-memx); + memx = tmpx; + } + e[i+pos] = tmp; + e_1[i+pos] = 1.f/tmp; + } + /* Hack to get 20 ms working with APPLICATION_AUDIO + The real problem is that the corresponding memory needs to use 1.5 ms + from this frame and 1 ms from the next frame */ + e[i+pos] = e[i+pos-1]; + if (buffering) + N=IMIN(MAX_DYNAMIC_FRAMESIZE, N+2); + bestLM = transient_viterbi(e, e_1, N, (int)((1.f+.5f*tonality)*(60*C+40)), bitrate/400); + mem[0] = e[1<<bestLM]; + if (buffering) + { + mem[1] = e[(1<<bestLM)+1]; + mem[2] = e[(1<<bestLM)+2]; + } + return bestLM; +} + +#endif + +#ifndef DISABLE_FLOAT_API #ifdef FIXED_POINT -#define opus_encode_native opus_encode -opus_int32 opus_encode(OpusEncoder *st, const opus_val16 *pcm, int frame_size, - unsigned char *data, opus_int32 out_data_bytes) +#define PCM2VAL(x) FLOAT2INT16(x) #else -#define opus_encode_native opus_encode_float -opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_size, - unsigned char *data, opus_int32 out_data_bytes) +#define PCM2VAL(x) SCALEIN(x) +#endif +void downmix_float(const void *_x, opus_val32 *sub, int subframe, int offset, int c1, int c2, int C) +{ + const float *x; + opus_val32 scale; + int j; + x = (const float *)_x; + for (j=0;j<subframe;j++) + sub[j] = PCM2VAL(x[(j+offset)*C+c1]); + if (c2>-1) + { + for (j=0;j<subframe;j++) + sub[j] += PCM2VAL(x[(j+offset)*C+c2]); + } else if (c2==-2) + { + int c; + for (c=1;c<C;c++) + { + for (j=0;j<subframe;j++) + sub[j] += PCM2VAL(x[(j+offset)*C+c]); + } + } +#ifdef FIXED_POINT + scale = (1<<SIG_SHIFT); +#else + scale = 1.f; +#endif + if (C==-2) + scale /= C; + else + scale /= 2; + for (j=0;j<subframe;j++) + sub[j] *= scale; +} +#endif + +void downmix_int(const void *_x, opus_val32 *sub, int subframe, int offset, int c1, int c2, int C) +{ + const opus_int16 *x; + opus_val32 scale; + int j; + x = (const opus_int16 *)_x; + for (j=0;j<subframe;j++) + sub[j] = x[(j+offset)*C+c1]; + if (c2>-1) + { + for (j=0;j<subframe;j++) + sub[j] += x[(j+offset)*C+c2]; + } else if (c2==-2) + { + int c; + for (c=1;c<C;c++) + { + for (j=0;j<subframe;j++) + sub[j] += x[(j+offset)*C+c]; + } + } +#ifdef FIXED_POINT + scale = (1<<SIG_SHIFT); +#else + scale = 1.f/32768; +#endif + if (C==-2) + scale /= C; + else + scale /= 2; + for (j=0;j<subframe;j++) + sub[j] *= scale; +} + +opus_int32 frame_size_select(opus_int32 frame_size, int variable_duration, opus_int32 Fs) +{ + int new_size; + if (frame_size<Fs/400) + return -1; + if (variable_duration == OPUS_FRAMESIZE_ARG) + new_size = frame_size; + else if (variable_duration == OPUS_FRAMESIZE_VARIABLE) + new_size = Fs/50; + else if (variable_duration >= OPUS_FRAMESIZE_2_5_MS && variable_duration <= OPUS_FRAMESIZE_60_MS) + new_size = IMIN(3*Fs/50, (Fs/400)<<(variable_duration-OPUS_FRAMESIZE_2_5_MS)); + else + return -1; + if (new_size>frame_size) + return -1; + if (400*new_size!=Fs && 200*new_size!=Fs && 100*new_size!=Fs && + 50*new_size!=Fs && 25*new_size!=Fs && 50*new_size!=3*Fs) + return -1; + return new_size; +} + +opus_int32 compute_frame_size(const void *analysis_pcm, int frame_size, + int variable_duration, int C, opus_int32 Fs, int bitrate_bps, + int delay_compensation, downmix_func downmix +#ifndef DISABLE_FLOAT_API + , float *subframe_mem +#endif + ) +{ +#ifndef DISABLE_FLOAT_API + if (variable_duration == OPUS_FRAMESIZE_VARIABLE && frame_size >= Fs/200) + { + int LM = 3; + LM = optimize_framesize(analysis_pcm, frame_size, C, Fs, bitrate_bps, + 0, subframe_mem, delay_compensation, downmix); + while ((Fs/400<<LM)>frame_size) + LM--; + frame_size = (Fs/400<<LM); + } else #endif + { + frame_size = frame_size_select(frame_size, variable_duration, Fs); + } + if (frame_size<0) + return -1; + return frame_size; +} + +opus_val16 compute_stereo_width(const opus_val16 *pcm, int frame_size, opus_int32 Fs, StereoWidthState *mem) +{ + opus_val16 corr; + opus_val16 ldiff; + opus_val16 width; + opus_val32 xx, xy, yy; + opus_val16 sqrt_xx, sqrt_yy; + opus_val16 qrrt_xx, qrrt_yy; + int frame_rate; + int i; + opus_val16 short_alpha; + + frame_rate = Fs/frame_size; + short_alpha = Q15ONE - 25*Q15ONE/IMAX(50,frame_rate); + xx=xy=yy=0; + for (i=0;i<frame_size;i+=4) + { + opus_val32 pxx=0; + opus_val32 pxy=0; + opus_val32 pyy=0; + opus_val16 x, y; + x = pcm[2*i]; + y = pcm[2*i+1]; + pxx = SHR32(MULT16_16(x,x),2); + pxy = SHR32(MULT16_16(x,y),2); + pyy = SHR32(MULT16_16(y,y),2); + x = pcm[2*i+2]; + y = pcm[2*i+3]; + pxx += SHR32(MULT16_16(x,x),2); + pxy += SHR32(MULT16_16(x,y),2); + pyy += SHR32(MULT16_16(y,y),2); + x = pcm[2*i+4]; + y = pcm[2*i+5]; + pxx += SHR32(MULT16_16(x,x),2); + pxy += SHR32(MULT16_16(x,y),2); + pyy += SHR32(MULT16_16(y,y),2); + x = pcm[2*i+6]; + y = pcm[2*i+7]; + pxx += SHR32(MULT16_16(x,x),2); + pxy += SHR32(MULT16_16(x,y),2); + pyy += SHR32(MULT16_16(y,y),2); + + xx += SHR32(pxx, 10); + xy += SHR32(pxy, 10); + yy += SHR32(pyy, 10); + } + mem->XX += MULT16_32_Q15(short_alpha, xx-mem->XX); + mem->XY += MULT16_32_Q15(short_alpha, xy-mem->XY); + mem->YY += MULT16_32_Q15(short_alpha, yy-mem->YY); + mem->XX = MAX32(0, mem->XX); + mem->XY = MAX32(0, mem->XY); + mem->YY = MAX32(0, mem->YY); + if (MAX32(mem->XX, mem->YY)>QCONST16(8e-4f, 18)) + { + sqrt_xx = celt_sqrt(mem->XX); + sqrt_yy = celt_sqrt(mem->YY); + qrrt_xx = celt_sqrt(sqrt_xx); + qrrt_yy = celt_sqrt(sqrt_yy); + /* Inter-channel correlation */ + mem->XY = MIN32(mem->XY, sqrt_xx*sqrt_yy); + corr = SHR32(frac_div32(mem->XY,EPSILON+MULT16_16(sqrt_xx,sqrt_yy)),16); + /* Approximate loudness difference */ + ldiff = Q15ONE*ABS16(qrrt_xx-qrrt_yy)/(EPSILON+qrrt_xx+qrrt_yy); + width = MULT16_16_Q15(celt_sqrt(QCONST32(1.f,30)-MULT16_16(corr,corr)), ldiff); + /* Smoothing over one second */ + mem->smoothed_width += (width-mem->smoothed_width)/frame_rate; + /* Peak follower */ + mem->max_follower = MAX16(mem->max_follower-QCONST16(.02f,15)/frame_rate, mem->smoothed_width); + } else { + width = 0; + corr=Q15ONE; + ldiff=0; + } + /*printf("%f %f %f %f %f ", corr/(float)Q15ONE, ldiff/(float)Q15ONE, width/(float)Q15ONE, mem->smoothed_width/(float)Q15ONE, mem->max_follower/(float)Q15ONE);*/ + return EXTRACT16(MIN32(Q15ONE,20*mem->max_follower)); +} + +opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_size, + unsigned char *data, opus_int32 out_data_bytes, int lsb_depth, + const void *analysis_pcm, opus_int32 analysis_size, int c1, int c2, int analysis_channels, downmix_func downmix) { void *silk_enc; CELTEncoder *celt_enc; @@ -471,7 +949,14 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s int frame_rate; opus_int32 max_rate; /* Max bitrate we're allowed to use */ int curr_bandwidth; + opus_val16 HB_gain; opus_int32 max_data_bytes; /* Max number of bytes we're allowed to use */ + int total_buffer; + opus_val16 stereo_width; + const CELTMode *celt_mode; + AnalysisInfo analysis_info; + int analysis_read_pos_bak=-1; + int analysis_read_subframe_bak=-1; VARDECL(opus_val16, tmp_prefill); ALLOC_STACK; @@ -479,25 +964,70 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s max_data_bytes = IMIN(1276, out_data_bytes); st->rangeFinal = 0; - if (400*frame_size != st->Fs && 200*frame_size != st->Fs && 100*frame_size != st->Fs && + if ((!st->variable_duration && 400*frame_size != st->Fs && 200*frame_size != st->Fs && 100*frame_size != st->Fs && 50*frame_size != st->Fs && 25*frame_size != st->Fs && 50*frame_size != 3*st->Fs) - { - RESTORE_STACK; - return OPUS_BAD_ARG; - } - if (max_data_bytes<=0) + || (400*frame_size < st->Fs) + || max_data_bytes<=0 + ) { RESTORE_STACK; return OPUS_BAD_ARG; } silk_enc = (char*)st+st->silk_enc_offset; celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset); - if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY) delay_compensation = 0; else delay_compensation = st->delay_compensation; + lsb_depth = IMIN(lsb_depth, st->lsb_depth); + + analysis_info.valid = 0; + celt_encoder_ctl(celt_enc, CELT_GET_MODE(&celt_mode)); +#ifndef DISABLE_FLOAT_API +#ifdef FIXED_POINT + if (st->silk_mode.complexity >= 10 && st->Fs==48000) +#else + if (st->silk_mode.complexity >= 7 && st->Fs==48000) +#endif + { + analysis_read_pos_bak = st->analysis.read_pos; + analysis_read_subframe_bak = st->analysis.read_subframe; + run_analysis(&st->analysis, celt_mode, analysis_pcm, analysis_size, frame_size, + c1, c2, analysis_channels, st->Fs, + lsb_depth, downmix, &analysis_info); + } +#endif + + st->voice_ratio = -1; + +#ifndef DISABLE_FLOAT_API + st->detected_bandwidth = 0; + if (analysis_info.valid) + { + int analysis_bandwidth; + if (st->signal_type == OPUS_AUTO) + st->voice_ratio = (int)floor(.5+100*(1-analysis_info.music_prob)); + + analysis_bandwidth = analysis_info.bandwidth; + if (analysis_bandwidth<=12) + st->detected_bandwidth = OPUS_BANDWIDTH_NARROWBAND; + else if (analysis_bandwidth<=14) + st->detected_bandwidth = OPUS_BANDWIDTH_MEDIUMBAND; + else if (analysis_bandwidth<=16) + st->detected_bandwidth = OPUS_BANDWIDTH_WIDEBAND; + else if (analysis_bandwidth<=18) + st->detected_bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND; + else + st->detected_bandwidth = OPUS_BANDWIDTH_FULLBAND; + } +#endif + + if (st->channels==2 && st->force_channels!=1) + stereo_width = compute_stereo_width(pcm, frame_size, st->Fs, &st->width_mem); + else + stereo_width = 0; + total_buffer = delay_compensation; st->bitrate_bps = user_bitrate_to_bitrate(st, frame_size, max_data_bytes); frame_rate = st->Fs/frame_size; @@ -533,15 +1063,19 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s max_rate = frame_rate*max_data_bytes*8; /* Equivalent 20-ms rate for mode/channel/bandwidth decisions */ - equiv_rate = st->bitrate_bps - 60*(st->Fs/frame_size - 50); + equiv_rate = st->bitrate_bps - (40*st->channels+20)*(st->Fs/frame_size - 50); if (st->signal_type == OPUS_SIGNAL_VOICE) voice_est = 127; else if (st->signal_type == OPUS_SIGNAL_MUSIC) voice_est = 0; else if (st->voice_ratio >= 0) + { voice_est = st->voice_ratio*327>>8; - else if (st->application == OPUS_APPLICATION_VOIP) + /* For AUDIO, never be more than 90% confident of having speech */ + if (st->application == OPUS_APPLICATION_AUDIO) + voice_est = IMIN(voice_est, 115); + } else if (st->application == OPUS_APPLICATION_VOIP) voice_est = 115; else voice_est = 48; @@ -561,15 +1095,16 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s opus_int32 stereo_threshold; stereo_threshold = stereo_music_threshold + ((voice_est*voice_est*(stereo_voice_threshold-stereo_music_threshold))>>14); if (st->stream_channels == 2) - stereo_threshold -= 4000; + stereo_threshold -= 1000; else - stereo_threshold += 4000; + stereo_threshold += 1000; st->stream_channels = (equiv_rate > stereo_threshold) ? 2 : 1; } else { st->stream_channels = st->channels; } #endif } + equiv_rate = st->bitrate_bps - (40*st->stream_channels+20)*(st->Fs/frame_size - 50); /* Mode selection depending on application and signal type */ if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY) @@ -592,15 +1127,21 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s st->mode = MODE_SILK_ONLY; } #else - int chan; opus_int32 mode_voice, mode_music; opus_int32 threshold; - chan = (st->channels==2) && st->force_channels!=1; - mode_voice = mode_thresholds[chan][0]; - mode_music = mode_thresholds[chan][1]; + /* Interpolate based on stereo width */ + mode_voice = (opus_int32)(MULT16_32_Q15(Q15ONE-stereo_width,mode_thresholds[0][0]) + + MULT16_32_Q15(stereo_width,mode_thresholds[1][0])); + mode_music = (opus_int32)(MULT16_32_Q15(Q15ONE-stereo_width,mode_thresholds[1][1]) + + MULT16_32_Q15(stereo_width,mode_thresholds[1][1])); + /* Interpolate based on speech/music probability */ threshold = mode_music + ((voice_est*voice_est*(mode_voice-mode_music))>>14); + /* Bias towards SILK for VoIP because of some useful features */ + if (st->application == OPUS_APPLICATION_VOIP) + threshold += 8000; + /*printf("%f %d\n", stereo_width/(float)Q15ONE, threshold);*/ /* Hysteresis */ if (st->prev_mode == MODE_CELT_ONLY) threshold -= 4000; @@ -623,6 +1164,11 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s /* Override the chosen mode to make sure we meet the requested frame size */ if (st->mode != MODE_CELT_ONLY && frame_size < st->Fs/100) st->mode = MODE_CELT_ONLY; + if (st->lfe) + st->mode = MODE_CELT_ONLY; + /* If max_data_bytes represents less than 8 kb/s, switch to CELT-only mode */ + if (max_data_bytes < (frame_rate > 50 ? 12000 : 8000)*frame_size / (st->Fs * 8)) + st->mode = MODE_CELT_ONLY; if (st->stream_channels == 1 && st->prev_channels ==2 && st->silk_mode.toMono==0 && st->mode != MODE_CELT_ONLY && st->prev_mode != MODE_CELT_ONLY) @@ -658,6 +1204,7 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s redundancy = 1; celt_to_silk = 1; st->silk_bw_switch = 0; + prefill=1; } if (redundancy) @@ -672,7 +1219,7 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s if (st->mode != MODE_CELT_ONLY && st->prev_mode == MODE_CELT_ONLY) { silk_EncControlStruct dummy; - silk_InitEncoder( silk_enc, &dummy); + silk_InitEncoder( silk_enc, st->arch, &dummy); prefill=1; } @@ -750,14 +1297,37 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s st->bandwidth = OPUS_BANDWIDTH_MEDIUMBAND; if (st->Fs <= 8000 && st->bandwidth > OPUS_BANDWIDTH_NARROWBAND) st->bandwidth = OPUS_BANDWIDTH_NARROWBAND; +#ifndef DISABLE_FLOAT_API + /* Use detected bandwidth to reduce the encoded bandwidth. */ + if (st->detected_bandwidth && st->user_bandwidth == OPUS_AUTO) + { + int min_detected_bandwidth; + /* Makes bandwidth detection more conservative just in case the detector + gets it wrong when we could have coded a high bandwidth transparently. + When operating in SILK/hybrid mode, we don't go below wideband to avoid + more complicated switches that require redundancy. */ + if (equiv_rate <= 18000*st->stream_channels && st->mode == MODE_CELT_ONLY) + min_detected_bandwidth = OPUS_BANDWIDTH_NARROWBAND; + else if (equiv_rate <= 24000*st->stream_channels && st->mode == MODE_CELT_ONLY) + min_detected_bandwidth = OPUS_BANDWIDTH_MEDIUMBAND; + else if (equiv_rate <= 30000*st->stream_channels) + min_detected_bandwidth = OPUS_BANDWIDTH_WIDEBAND; + else if (equiv_rate <= 44000*st->stream_channels) + min_detected_bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND; + else + min_detected_bandwidth = OPUS_BANDWIDTH_FULLBAND; - /* If max_data_bytes represents less than 8 kb/s, switch to CELT-only mode */ - if (max_data_bytes < (frame_rate > 50 ? 12000 : 8000)*frame_size / (st->Fs * 8)) - st->mode = MODE_CELT_ONLY; + st->detected_bandwidth = IMAX(st->detected_bandwidth, min_detected_bandwidth); + st->bandwidth = IMIN(st->bandwidth, st->detected_bandwidth); + } +#endif + celt_encoder_ctl(celt_enc, OPUS_SET_LSB_DEPTH(lsb_depth)); /* CELT mode doesn't support mediumband, use wideband instead */ if (st->mode == MODE_CELT_ONLY && st->bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) st->bandwidth = OPUS_BANDWIDTH_WIDEBAND; + if (st->lfe) + st->bandwidth = OPUS_BANDWIDTH_NARROWBAND; /* Can't support higher than wideband for >20 ms frames */ if (frame_size > st->Fs/50 && (st->mode == MODE_CELT_ONLY || st->bandwidth > OPUS_BANDWIDTH_WIDEBAND)) @@ -765,16 +1335,25 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s VARDECL(unsigned char, tmp_data); int nb_frames; int bak_mode, bak_bandwidth, bak_channels, bak_to_mono; - OpusRepacketizer rp; + VARDECL(OpusRepacketizer, rp); opus_int32 bytes_per_frame; + opus_int32 repacketize_len; +#ifndef DISABLE_FLOAT_API + if (analysis_read_pos_bak!= -1) + { + st->analysis.read_pos = analysis_read_pos_bak; + st->analysis.read_subframe = analysis_read_subframe_bak; + } +#endif nb_frames = frame_size > st->Fs/25 ? 3 : 2; bytes_per_frame = IMIN(1276,(out_data_bytes-3)/nb_frames); ALLOC(tmp_data, nb_frames*bytes_per_frame, unsigned char); - opus_repacketizer_init(&rp); + ALLOC(rp, 1, OpusRepacketizer); + opus_repacketizer_init(rp); bak_mode = st->user_forced_mode; bak_bandwidth = st->user_bandwidth; @@ -796,20 +1375,26 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s /* When switching from SILK/Hybrid to CELT, only ask for a switch at the last frame */ if (to_celt && i==nb_frames-1) st->user_forced_mode = MODE_CELT_ONLY; - tmp_len = opus_encode_native(st, pcm+i*(st->channels*st->Fs/50), st->Fs/50, tmp_data+i*bytes_per_frame, bytes_per_frame); + tmp_len = opus_encode_native(st, pcm+i*(st->channels*st->Fs/50), st->Fs/50, + tmp_data+i*bytes_per_frame, bytes_per_frame, lsb_depth, + NULL, 0, c1, c2, analysis_channels, downmix); if (tmp_len<0) { RESTORE_STACK; return OPUS_INTERNAL_ERROR; } - ret = opus_repacketizer_cat(&rp, tmp_data+i*bytes_per_frame, tmp_len); + ret = opus_repacketizer_cat(rp, tmp_data+i*bytes_per_frame, tmp_len); if (ret<0) { RESTORE_STACK; return OPUS_INTERNAL_ERROR; } } - ret = opus_repacketizer_out(&rp, data, out_data_bytes); + if (st->use_vbr) + repacketize_len = out_data_bytes; + else + repacketize_len = IMIN(3*st->bitrate_bps/(3*8*50/nb_frames), out_data_bytes); + ret = opus_repacketizer_out_range_impl(rp, 0, nb_frames, data, repacketize_len, 0, !st->use_vbr); if (ret<0) { RESTORE_STACK; @@ -822,7 +1407,6 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s RESTORE_STACK; return ret; } - curr_bandwidth = st->bandwidth; /* Chooses the appropriate mode for speech @@ -839,9 +1423,9 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s ec_enc_init(&enc, data, max_data_bytes-1); - ALLOC(pcm_buf, (delay_compensation+frame_size)*st->channels, opus_val16); - for (i=0;i<delay_compensation*st->channels;i++) - pcm_buf[i] = st->delay_buffer[(st->encoder_buffer-delay_compensation)*st->channels+i]; + ALLOC(pcm_buf, (total_buffer+frame_size)*st->channels, opus_val16); + for (i=0;i<total_buffer*st->channels;i++) + pcm_buf[i] = st->delay_buffer[(st->encoder_buffer-total_buffer)*st->channels+i]; if (st->mode == MODE_CELT_ONLY) hp_freq_smth1 = silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 ); @@ -856,46 +1440,96 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s if (st->application == OPUS_APPLICATION_VOIP) { - hp_cutoff(pcm, cutoff_Hz, &pcm_buf[delay_compensation*st->channels], st->hp_mem, frame_size, st->channels, st->Fs); + hp_cutoff(pcm, cutoff_Hz, &pcm_buf[total_buffer*st->channels], st->hp_mem, frame_size, st->channels, st->Fs); } else { - for (i=0;i<frame_size*st->channels;i++) - pcm_buf[delay_compensation*st->channels + i] = pcm[i]; + dc_reject(pcm, 3, &pcm_buf[total_buffer*st->channels], st->hp_mem, frame_size, st->channels, st->Fs); } + + /* SILK processing */ + HB_gain = Q15ONE; if (st->mode != MODE_CELT_ONLY) { + opus_int32 total_bitRate, celt_rate; #ifdef FIXED_POINT const opus_int16 *pcm_silk; #else VARDECL(opus_int16, pcm_silk); ALLOC(pcm_silk, st->channels*frame_size, opus_int16); #endif - st->silk_mode.bitRate = 8*bytes_target*frame_rate; + + /* Distribute bits between SILK and CELT */ + total_bitRate = 8 * bytes_target * frame_rate; if( st->mode == MODE_HYBRID ) { - st->silk_mode.bitRate /= st->stream_channels; + int HB_gain_ref; + /* Base rate for SILK */ + st->silk_mode.bitRate = st->stream_channels * ( 5000 + 1000 * ( st->Fs == 100 * frame_size ) ); if( curr_bandwidth == OPUS_BANDWIDTH_SUPERWIDEBAND ) { - if( st->Fs == 100 * frame_size ) { - /* 24 kHz, 10 ms */ - st->silk_mode.bitRate = ( ( st->silk_mode.bitRate + 2000 + st->use_vbr * 1000 ) * 2 ) / 3; - } else { - /* 24 kHz, 20 ms */ - st->silk_mode.bitRate = ( ( st->silk_mode.bitRate + 1000 + st->use_vbr * 1000 ) * 2 ) / 3; - } - } else { - if( st->Fs == 100 * frame_size ) { - /* 48 kHz, 10 ms */ - st->silk_mode.bitRate = ( st->silk_mode.bitRate + 8000 + st->use_vbr * 3000 ) / 2; - } else { - /* 48 kHz, 20 ms */ - st->silk_mode.bitRate = ( st->silk_mode.bitRate + 9000 + st->use_vbr * 1000 ) / 2; - } + /* SILK gets 2/3 of the remaining bits */ + st->silk_mode.bitRate += ( total_bitRate - st->silk_mode.bitRate ) * 2 / 3; + } else { /* FULLBAND */ + /* SILK gets 3/5 of the remaining bits */ + st->silk_mode.bitRate += ( total_bitRate - st->silk_mode.bitRate ) * 3 / 5; } - st->silk_mode.bitRate *= st->stream_channels; - /* don't let SILK use more than 80% */ - if( st->silk_mode.bitRate > ( st->bitrate_bps - 8*st->Fs/frame_size ) * 4/5 ) { - st->silk_mode.bitRate = ( st->bitrate_bps - 8*st->Fs/frame_size ) * 4/5; + /* Don't let SILK use more than 80% */ + if( st->silk_mode.bitRate > total_bitRate * 4/5 ) { + st->silk_mode.bitRate = total_bitRate * 4/5; } + if (!st->energy_masking) + { + /* Increasingly attenuate high band when it gets allocated fewer bits */ + celt_rate = total_bitRate - st->silk_mode.bitRate; + HB_gain_ref = (curr_bandwidth == OPUS_BANDWIDTH_SUPERWIDEBAND) ? 3000 : 3600; + HB_gain = SHL32((opus_val32)celt_rate, 9) / SHR32((opus_val32)celt_rate + st->stream_channels * HB_gain_ref, 6); + HB_gain = HB_gain < Q15ONE*6/7 ? HB_gain + Q15ONE/7 : Q15ONE; + } + } else { + /* SILK gets all bits */ + st->silk_mode.bitRate = total_bitRate; + } + + /* Surround masking for SILK */ + if (st->energy_masking && st->use_vbr && !st->lfe) + { + opus_val32 mask_sum=0; + opus_val16 masking_depth; + opus_int32 rate_offset; + int c; + int end = 17; + opus_int16 srate = 16000; + if (st->bandwidth == OPUS_BANDWIDTH_NARROWBAND) + { + end = 13; + srate = 8000; + } else if (st->bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) + { + end = 15; + srate = 12000; + } + for (c=0;c<st->channels;c++) + { + for(i=0;i<end;i++) + { + opus_val16 mask; + mask = MAX16(MIN16(st->energy_masking[21*c+i], + QCONST16(.5f, DB_SHIFT)), -QCONST16(2.0f, DB_SHIFT)); + if (mask > 0) + mask = HALF16(mask); + mask_sum += mask; + } + } + /* Conservative rate reduction, we cut the masking in half */ + masking_depth = mask_sum / end*st->channels; + masking_depth += QCONST16(.2f, DB_SHIFT); + rate_offset = (opus_int32)PSHR32(MULT16_16(srate, masking_depth), DB_SHIFT); + rate_offset = MAX32(rate_offset, -2*st->silk_mode.bitRate/3); + /* Split the rate change between the SILK and CELT part for hybrid. */ + if (st->bandwidth==OPUS_BANDWIDTH_SUPERWIDEBAND || st->bandwidth==OPUS_BANDWIDTH_FULLBAND) + st->silk_mode.bitRate += 3*rate_offset/5; + else + st->silk_mode.bitRate += rate_offset; + bytes_target += rate_offset * frame_size / (8 * st->Fs); } st->silk_mode.payloadSize_ms = 1000 * frame_size / st->Fs; @@ -955,6 +1589,18 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s if (prefill) { opus_int32 zero=0; + int prefill_offset; + /* Use a smooth onset for the SILK prefill to avoid the encoder trying to encode + a discontinuity. The exact location is what we need to avoid leaving any "gap" + in the audio when mixing with the redundant CELT frame. Here we can afford to + overwrite st->delay_buffer because the only thing that uses it before it gets + rewritten is tmp_prefill[] and even then only the part after the ramp really + gets used (rather than sent to the encoder and discarded) */ + prefill_offset = st->channels*(st->encoder_buffer-st->delay_compensation-st->Fs/400); + gain_fade(st->delay_buffer+prefill_offset, st->delay_buffer+prefill_offset, + 0, Q15ONE, celt_mode->overlap, st->Fs/400, st->channels, celt_mode->window, st->Fs); + for(i=0;i<prefill_offset;i++) + st->delay_buffer[i]=0; #ifdef FIXED_POINT pcm_silk = st->delay_buffer; #else @@ -965,10 +1611,10 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s } #ifdef FIXED_POINT - pcm_silk = pcm_buf+delay_compensation*st->channels; + pcm_silk = pcm_buf+total_buffer*st->channels; #else for (i=0;i<frame_size*st->channels;i++) - pcm_silk[i] = FLOAT2INT16(pcm_buf[delay_compensation*st->channels + i]); + pcm_silk[i] = FLOAT2INT16(pcm_buf[total_buffer*st->channels + i]); #endif ret = silk_Encode( silk_enc, &st->silk_mode, pcm_silk, frame_size, &enc, &nBytes, 0 ); if( ret ) { @@ -1033,9 +1679,12 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(OPUS_BITRATE_MAX)); if (st->mode != MODE_SILK_ONLY) { + opus_val32 celt_pred=2; celt_encoder_ctl(celt_enc, OPUS_SET_VBR(0)); - /* Allow prediction unless we decide to disable it later */ - celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(2)); + /* We may still decide to disable prediction later */ + if (st->silk_mode.reducedDependency) + celt_pred = 0; + celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(celt_pred)); if (st->mode == MODE_HYBRID) { @@ -1053,9 +1702,18 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s } else { if (st->use_vbr) { + opus_int32 bonus=0; +#ifndef DISABLE_FLOAT_API + if (st->variable_duration==OPUS_FRAMESIZE_VARIABLE && frame_size != st->Fs/50) + { + bonus = (60*st->stream_channels+40)*(st->Fs/frame_size-50); + if (analysis_info.valid) + bonus = (opus_int32)(bonus*(1.f+.5f*analysis_info.tonality)); + } +#endif celt_encoder_ctl(celt_enc, OPUS_SET_VBR(1)); celt_encoder_ctl(celt_enc, OPUS_SET_VBR_CONSTRAINT(st->vbr_constraint)); - celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(st->bitrate_bps)); + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(st->bitrate_bps+bonus)); nb_compr_bytes = max_data_bytes-1-redundancy_bytes; } else { nb_compr_bytes = bytes_target; @@ -1070,24 +1728,27 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s if (st->mode != MODE_SILK_ONLY && st->mode != st->prev_mode && st->prev_mode > 0) { for (i=0;i<st->channels*st->Fs/400;i++) - tmp_prefill[i] = st->delay_buffer[(st->encoder_buffer-st->delay_compensation-st->Fs/400)*st->channels + i]; + tmp_prefill[i] = st->delay_buffer[(st->encoder_buffer-total_buffer-st->Fs/400)*st->channels + i]; } - for (i=0;i<st->channels*(st->encoder_buffer-(frame_size+delay_compensation));i++) + for (i=0;i<st->channels*(st->encoder_buffer-(frame_size+total_buffer));i++) st->delay_buffer[i] = st->delay_buffer[i+st->channels*frame_size]; for (;i<st->encoder_buffer*st->channels;i++) - st->delay_buffer[i] = pcm_buf[(frame_size+delay_compensation-st->encoder_buffer)*st->channels+i]; - + st->delay_buffer[i] = pcm_buf[(frame_size+total_buffer-st->encoder_buffer)*st->channels+i]; + /* gain_fade() and stereo_fade() need to be after the buffer copying + because we don't want any of this to affect the SILK part */ + if( st->prev_HB_gain < Q15ONE || HB_gain < Q15ONE ) { + gain_fade(pcm_buf, pcm_buf, + st->prev_HB_gain, HB_gain, celt_mode->overlap, frame_size, st->channels, celt_mode->window, st->Fs); + } + st->prev_HB_gain = HB_gain; if (st->mode != MODE_HYBRID || st->stream_channels==1) - st->silk_mode.stereoWidth_Q14 = 1<<14; - if( st->channels == 2 ) { + st->silk_mode.stereoWidth_Q14 = IMIN((1<<14),2*IMAX(0,equiv_rate-30000)); + if( !st->energy_masking && st->channels == 2 ) { /* Apply stereo width reduction (at low bitrates) */ if( st->hybrid_stereo_width_Q14 < (1 << 14) || st->silk_mode.stereoWidth_Q14 < (1 << 14) ) { opus_val16 g1, g2; - const CELTMode *celt_mode; - - celt_encoder_ctl(celt_enc, CELT_GET_MODE(&celt_mode)); g1 = st->hybrid_stereo_width_Q14; g2 = (opus_val16)(st->silk_mode.stereoWidth_Q14); #ifdef FIXED_POINT @@ -1113,7 +1774,7 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s int max_redundancy; ec_enc_bit_logp(&enc, celt_to_silk, 1); if (st->mode == MODE_HYBRID) - max_redundancy = (max_data_bytes-1)-nb_compr_bytes-1; + max_redundancy = (max_data_bytes-1)-nb_compr_bytes; else max_redundancy = (max_data_bytes-1)-((ec_tell(&enc)+7)>>3); /* Target the same bit-rate for redundancy as for the rest, @@ -1144,6 +1805,10 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s ec_enc_shrink(&enc, nb_compr_bytes); } +#ifndef DISABLE_FLOAT_API + if (redundancy || st->mode != MODE_SILK_ONLY) + celt_encoder_ctl(celt_enc, CELT_SET_ANALYSIS(&analysis_info)); +#endif /* 5 ms redundant frame for CELT->SILK */ if (redundancy && celt_to_silk) @@ -1252,9 +1917,10 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s } /* Count ToC and redundancy */ ret += 1+redundancy_bytes; - if (!st->use_vbr && ret >= 3) + if (!st->use_vbr) { - if (pad_frame(data, ret, max_data_bytes)) + if (opus_packet_pad(data, ret, max_data_bytes) != OPUS_OK) + { RESTORE_STACK; return OPUS_INTERNAL_ERROR; @@ -1268,45 +1934,93 @@ opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_s #ifdef FIXED_POINT #ifndef DISABLE_FLOAT_API -opus_int32 opus_encode_float(OpusEncoder *st, const float *pcm, int frame_size, +opus_int32 opus_encode_float(OpusEncoder *st, const float *pcm, int analysis_frame_size, unsigned char *data, opus_int32 max_data_bytes) { int i, ret; + int frame_size; + int delay_compensation; VARDECL(opus_int16, in); ALLOC_STACK; - if(frame_size<0) - { - RESTORE_STACK; - return OPUS_BAD_ARG; - } + if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY) + delay_compensation = 0; + else + delay_compensation = st->delay_compensation; + frame_size = compute_frame_size(pcm, analysis_frame_size, + st->variable_duration, st->channels, st->Fs, st->bitrate_bps, + delay_compensation, downmix_float, st->analysis.subframe_mem); ALLOC(in, frame_size*st->channels, opus_int16); for (i=0;i<frame_size*st->channels;i++) in[i] = FLOAT2INT16(pcm[i]); - ret = opus_encode(st, in, frame_size, data, max_data_bytes); + ret = opus_encode_native(st, in, frame_size, data, max_data_bytes, 16, pcm, analysis_frame_size, 0, -2, st->channels, downmix_float); RESTORE_STACK; return ret; } #endif +opus_int32 opus_encode(OpusEncoder *st, const opus_int16 *pcm, int analysis_frame_size, + unsigned char *data, opus_int32 out_data_bytes) +{ + int frame_size; + int delay_compensation; + if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY) + delay_compensation = 0; + else + delay_compensation = st->delay_compensation; + frame_size = compute_frame_size(pcm, analysis_frame_size, + st->variable_duration, st->channels, st->Fs, st->bitrate_bps, + delay_compensation, downmix_int +#ifndef DISABLE_FLOAT_API + , st->analysis.subframe_mem +#endif + ); + return opus_encode_native(st, pcm, frame_size, data, out_data_bytes, 16, pcm, analysis_frame_size, 0, -2, st->channels, downmix_int); +} + #else -opus_int32 opus_encode(OpusEncoder *st, const opus_int16 *pcm, int frame_size, +opus_int32 opus_encode(OpusEncoder *st, const opus_int16 *pcm, int analysis_frame_size, unsigned char *data, opus_int32 max_data_bytes) { int i, ret; + int frame_size; + int delay_compensation; VARDECL(float, in); ALLOC_STACK; + if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY) + delay_compensation = 0; + else + delay_compensation = st->delay_compensation; + frame_size = compute_frame_size(pcm, analysis_frame_size, + st->variable_duration, st->channels, st->Fs, st->bitrate_bps, + delay_compensation, downmix_int, st->analysis.subframe_mem); + ALLOC(in, frame_size*st->channels, float); for (i=0;i<frame_size*st->channels;i++) in[i] = (1.0f/32768)*pcm[i]; - ret = opus_encode_float(st, in, frame_size, data, max_data_bytes); + ret = opus_encode_native(st, in, frame_size, data, max_data_bytes, 16, pcm, analysis_frame_size, 0, -2, st->channels, downmix_int); RESTORE_STACK; return ret; } +opus_int32 opus_encode_float(OpusEncoder *st, const float *pcm, int analysis_frame_size, + unsigned char *data, opus_int32 out_data_bytes) +{ + int frame_size; + int delay_compensation; + if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY) + delay_compensation = 0; + else + delay_compensation = st->delay_compensation; + frame_size = compute_frame_size(pcm, analysis_frame_size, + st->variable_duration, st->channels, st->Fs, st->bitrate_bps, + delay_compensation, downmix_float, st->analysis.subframe_mem); + return opus_encode_native(st, pcm, frame_size, data, out_data_bytes, 24, + pcm, analysis_frame_size, 0, -2, st->channels, downmix_float); +} #endif @@ -1339,6 +2053,10 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) case OPUS_GET_APPLICATION_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } *value = st->application; } break; @@ -1360,6 +2078,10 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) case OPUS_GET_BITRATE_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } *value = user_bitrate_to_bitrate(st, st->prev_framesize, 1276); } break; @@ -1367,21 +2089,29 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) { opus_int32 value = va_arg(ap, opus_int32); if((value<1 || value>st->channels) && value != OPUS_AUTO) - return OPUS_BAD_ARG; + { + goto bad_arg; + } st->force_channels = value; } break; case OPUS_GET_FORCE_CHANNELS_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } *value = st->force_channels; } break; case OPUS_SET_MAX_BANDWIDTH_REQUEST: { opus_int32 value = va_arg(ap, opus_int32); - if (value < OPUS_BANDWIDTH_NARROWBAND || value > OPUS_BANDWIDTH_FULLBAND) - return OPUS_BAD_ARG; + if (value < OPUS_BANDWIDTH_NARROWBAND || value > OPUS_BANDWIDTH_FULLBAND) + { + goto bad_arg; + } st->max_bandwidth = value; if (st->max_bandwidth == OPUS_BANDWIDTH_NARROWBAND) { st->silk_mode.maxInternalSampleRate = 8000; @@ -1395,6 +2125,10 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) case OPUS_GET_MAX_BANDWIDTH_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } *value = st->max_bandwidth; } break; @@ -1402,7 +2136,9 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) { opus_int32 value = va_arg(ap, opus_int32); if ((value < OPUS_BANDWIDTH_NARROWBAND || value > OPUS_BANDWIDTH_FULLBAND) && value != OPUS_AUTO) - return OPUS_BAD_ARG; + { + goto bad_arg; + } st->user_bandwidth = value; if (st->user_bandwidth == OPUS_BANDWIDTH_NARROWBAND) { st->silk_mode.maxInternalSampleRate = 8000; @@ -1416,6 +2152,10 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) case OPUS_GET_BANDWIDTH_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } *value = st->bandwidth; } break; @@ -1423,13 +2163,19 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) { opus_int32 value = va_arg(ap, opus_int32); if(value<0 || value>1) - return OPUS_BAD_ARG; + { + goto bad_arg; + } st->silk_mode.useDTX = value; } break; case OPUS_GET_DTX_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } *value = st->silk_mode.useDTX; } break; @@ -1437,7 +2183,9 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) { opus_int32 value = va_arg(ap, opus_int32); if(value<0 || value>10) - return OPUS_BAD_ARG; + { + goto bad_arg; + } st->silk_mode.complexity = value; celt_encoder_ctl(celt_enc, OPUS_SET_COMPLEXITY(value)); } @@ -1445,6 +2193,10 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) case OPUS_GET_COMPLEXITY_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } *value = st->silk_mode.complexity; } break; @@ -1452,13 +2204,19 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) { opus_int32 value = va_arg(ap, opus_int32); if(value<0 || value>1) - return OPUS_BAD_ARG; + { + goto bad_arg; + } st->silk_mode.useInBandFEC = value; } break; case OPUS_GET_INBAND_FEC_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } *value = st->silk_mode.useInBandFEC; } break; @@ -1466,7 +2224,9 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) { opus_int32 value = va_arg(ap, opus_int32); if (value < 0 || value > 100) - return OPUS_BAD_ARG; + { + goto bad_arg; + } st->silk_mode.packetLossPercentage = value; celt_encoder_ctl(celt_enc, OPUS_SET_PACKET_LOSS_PERC(value)); } @@ -1474,6 +2234,10 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) case OPUS_GET_PACKET_LOSS_PERC_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } *value = st->silk_mode.packetLossPercentage; } break; @@ -1481,7 +2245,9 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) { opus_int32 value = va_arg(ap, opus_int32); if(value<0 || value>1) - return OPUS_BAD_ARG; + { + goto bad_arg; + } st->use_vbr = value; st->silk_mode.useCBR = 1-value; } @@ -1489,20 +2255,30 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) case OPUS_GET_VBR_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } *value = st->use_vbr; } break; case OPUS_SET_VOICE_RATIO_REQUEST: { opus_int32 value = va_arg(ap, opus_int32); - if (value>100 || value<-1) - goto bad_arg; + if (value<-1 || value>100) + { + goto bad_arg; + } st->voice_ratio = value; } break; case OPUS_GET_VOICE_RATIO_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } *value = st->voice_ratio; } break; @@ -1510,13 +2286,19 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) { opus_int32 value = va_arg(ap, opus_int32); if(value<0 || value>1) - return OPUS_BAD_ARG; + { + goto bad_arg; + } st->vbr_constraint = value; } break; case OPUS_GET_VBR_CONSTRAINT_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } *value = st->vbr_constraint; } break; @@ -1524,19 +2306,29 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) { opus_int32 value = va_arg(ap, opus_int32); if(value!=OPUS_AUTO && value!=OPUS_SIGNAL_VOICE && value!=OPUS_SIGNAL_MUSIC) - return OPUS_BAD_ARG; + { + goto bad_arg; + } st->signal_type = value; } break; case OPUS_GET_SIGNAL_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } *value = st->signal_type; } break; case OPUS_GET_LOOKAHEAD_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } *value = st->Fs/400; if (st->application != OPUS_APPLICATION_RESTRICTED_LOWDELAY) *value += st->delay_compensation; @@ -1545,10 +2337,9 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) case OPUS_GET_SAMPLE_RATE_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); - if (value==NULL) + if (!value) { - ret = OPUS_BAD_ARG; - break; + goto bad_arg; } *value = st->Fs; } @@ -1556,19 +2347,71 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) case OPUS_GET_FINAL_RANGE_REQUEST: { opus_uint32 *value = va_arg(ap, opus_uint32*); + if (!value) + { + goto bad_arg; + } *value = st->rangeFinal; } break; case OPUS_SET_LSB_DEPTH_REQUEST: { opus_int32 value = va_arg(ap, opus_int32); - ret = celt_encoder_ctl(celt_enc, OPUS_SET_LSB_DEPTH(value)); + if (value<8 || value>24) + { + goto bad_arg; + } + st->lsb_depth=value; } break; case OPUS_GET_LSB_DEPTH_REQUEST: { opus_int32 *value = va_arg(ap, opus_int32*); - celt_encoder_ctl(celt_enc, OPUS_GET_LSB_DEPTH(value)); + if (!value) + { + goto bad_arg; + } + *value = st->lsb_depth; + } + break; + case OPUS_SET_EXPERT_FRAME_DURATION_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value != OPUS_FRAMESIZE_ARG && value != OPUS_FRAMESIZE_2_5_MS && + value != OPUS_FRAMESIZE_5_MS && value != OPUS_FRAMESIZE_10_MS && + value != OPUS_FRAMESIZE_20_MS && value != OPUS_FRAMESIZE_40_MS && + value != OPUS_FRAMESIZE_60_MS && value != OPUS_FRAMESIZE_VARIABLE) + { + goto bad_arg; + } + st->variable_duration = value; + celt_encoder_ctl(celt_enc, OPUS_SET_EXPERT_FRAME_DURATION(value)); + } + break; + case OPUS_GET_EXPERT_FRAME_DURATION_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->variable_duration; + } + break; + case OPUS_SET_PREDICTION_DISABLED_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value > 1 || value < 0) + goto bad_arg; + st->silk_mode.reducedDependency = value; + } + break; + case OPUS_GET_PREDICTION_DISABLED_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + goto bad_arg; + *value = st->silk_mode.reducedDependency; } break; case OPUS_RESET_STATE: @@ -1582,9 +2425,10 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) ((char*)&st->OPUS_ENCODER_RESET_START - (char*)st)); celt_encoder_ctl(celt_enc, OPUS_RESET_STATE); - silk_InitEncoder( silk_enc, &dummy ); + silk_InitEncoder( silk_enc, st->arch, &dummy ); st->stream_channels = st->channels; st->hybrid_stereo_width_Q14 = 1 << 14; + st->prev_HB_gain = Q15ONE; st->first = 1; st->mode = MODE_HYBRID; st->bandwidth = OPUS_BANDWIDTH_FULLBAND; @@ -1595,10 +2439,37 @@ int opus_encoder_ctl(OpusEncoder *st, int request, ...) { opus_int32 value = va_arg(ap, opus_int32); if ((value < MODE_SILK_ONLY || value > MODE_CELT_ONLY) && value != OPUS_AUTO) + { goto bad_arg; + } st->user_forced_mode = value; } break; + case OPUS_SET_LFE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->lfe = value; + ret = celt_encoder_ctl(celt_enc, OPUS_SET_LFE(value)); + } + break; + case OPUS_SET_ENERGY_MASK_REQUEST: + { + opus_val16 *value = va_arg(ap, opus_val16*); + st->energy_masking = value; + ret = celt_encoder_ctl(celt_enc, OPUS_SET_ENERGY_MASK(value)); + } + break; + + case CELT_GET_MODE_REQUEST: + { + const CELTMode ** value = va_arg(ap, const CELTMode**); + if (!value) + { + goto bad_arg; + } + ret = celt_encoder_ctl(celt_enc, CELT_GET_MODE(value)); + } + break; default: /* fprintf(stderr, "unknown opus_encoder_ctl() request: %d", request);*/ ret = OPUS_UNIMPLEMENTED; diff --git a/src/opus-1.1/src/opus_multistream.c b/src/opus-1.1/src/opus_multistream.c new file mode 100644 index 00000000..09c3639b --- /dev/null +++ b/src/opus-1.1/src/opus_multistream.c @@ -0,0 +1,92 @@ +/* Copyright (c) 2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "opus_multistream.h" +#include "opus.h" +#include "opus_private.h" +#include "stack_alloc.h" +#include <stdarg.h> +#include "float_cast.h" +#include "os_support.h" + + +int validate_layout(const ChannelLayout *layout) +{ + int i, max_channel; + + max_channel = layout->nb_streams+layout->nb_coupled_streams; + if (max_channel>255) + return 0; + for (i=0;i<layout->nb_channels;i++) + { + if (layout->mapping[i] >= max_channel && layout->mapping[i] != 255) + return 0; + } + return 1; +} + + +int get_left_channel(const ChannelLayout *layout, int stream_id, int prev) +{ + int i; + i = (prev<0) ? 0 : prev+1; + for (;i<layout->nb_channels;i++) + { + if (layout->mapping[i]==stream_id*2) + return i; + } + return -1; +} + +int get_right_channel(const ChannelLayout *layout, int stream_id, int prev) +{ + int i; + i = (prev<0) ? 0 : prev+1; + for (;i<layout->nb_channels;i++) + { + if (layout->mapping[i]==stream_id*2+1) + return i; + } + return -1; +} + +int get_mono_channel(const ChannelLayout *layout, int stream_id, int prev) +{ + int i; + i = (prev<0) ? 0 : prev+1; + for (;i<layout->nb_channels;i++) + { + if (layout->mapping[i]==stream_id+layout->nb_coupled_streams) + return i; + } + return -1; +} + diff --git a/src/opus-1.1/src/opus_multistream_decoder.c b/src/opus-1.1/src/opus_multistream_decoder.c new file mode 100644 index 00000000..a05fa1e7 --- /dev/null +++ b/src/opus-1.1/src/opus_multistream_decoder.c @@ -0,0 +1,537 @@ +/* Copyright (c) 2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "opus_multistream.h" +#include "opus.h" +#include "opus_private.h" +#include "stack_alloc.h" +#include <stdarg.h> +#include "float_cast.h" +#include "os_support.h" + +struct OpusMSDecoder { + ChannelLayout layout; + /* Decoder states go here */ +}; + + + + +/* DECODER */ + +opus_int32 opus_multistream_decoder_get_size(int nb_streams, int nb_coupled_streams) +{ + int coupled_size; + int mono_size; + + if(nb_streams<1||nb_coupled_streams>nb_streams||nb_coupled_streams<0)return 0; + coupled_size = opus_decoder_get_size(2); + mono_size = opus_decoder_get_size(1); + return align(sizeof(OpusMSDecoder)) + + nb_coupled_streams * align(coupled_size) + + (nb_streams-nb_coupled_streams) * align(mono_size); +} + +int opus_multistream_decoder_init( + OpusMSDecoder *st, + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping +) +{ + int coupled_size; + int mono_size; + int i, ret; + char *ptr; + + if ((channels>255) || (channels<1) || (coupled_streams>streams) || + (coupled_streams+streams>255) || (streams<1) || (coupled_streams<0)) + return OPUS_BAD_ARG; + + st->layout.nb_channels = channels; + st->layout.nb_streams = streams; + st->layout.nb_coupled_streams = coupled_streams; + + for (i=0;i<st->layout.nb_channels;i++) + st->layout.mapping[i] = mapping[i]; + if (!validate_layout(&st->layout)) + return OPUS_BAD_ARG; + + ptr = (char*)st + align(sizeof(OpusMSDecoder)); + coupled_size = opus_decoder_get_size(2); + mono_size = opus_decoder_get_size(1); + + for (i=0;i<st->layout.nb_coupled_streams;i++) + { + ret=opus_decoder_init((OpusDecoder*)ptr, Fs, 2); + if(ret!=OPUS_OK)return ret; + ptr += align(coupled_size); + } + for (;i<st->layout.nb_streams;i++) + { + ret=opus_decoder_init((OpusDecoder*)ptr, Fs, 1); + if(ret!=OPUS_OK)return ret; + ptr += align(mono_size); + } + return OPUS_OK; +} + + +OpusMSDecoder *opus_multistream_decoder_create( + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping, + int *error +) +{ + int ret; + OpusMSDecoder *st; + if ((channels>255) || (channels<1) || (coupled_streams>streams) || + (coupled_streams+streams>255) || (streams<1) || (coupled_streams<0)) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + st = (OpusMSDecoder *)opus_alloc(opus_multistream_decoder_get_size(streams, coupled_streams)); + if (st==NULL) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + ret = opus_multistream_decoder_init(st, Fs, channels, streams, coupled_streams, mapping); + if (error) + *error = ret; + if (ret != OPUS_OK) + { + opus_free(st); + st = NULL; + } + return st; +} + +typedef void (*opus_copy_channel_out_func)( + void *dst, + int dst_stride, + int dst_channel, + const opus_val16 *src, + int src_stride, + int frame_size +); + +static int opus_multistream_packet_validate(const unsigned char *data, + opus_int32 len, int nb_streams, opus_int32 Fs) +{ + int s; + int count; + unsigned char toc; + opus_int16 size[48]; + int samples=0; + opus_int32 packet_offset; + + for (s=0;s<nb_streams;s++) + { + int tmp_samples; + if (len<=0) + return OPUS_INVALID_PACKET; + count = opus_packet_parse_impl(data, len, s!=nb_streams-1, &toc, NULL, + size, NULL, &packet_offset); + if (count<0) + return count; + tmp_samples = opus_packet_get_nb_samples(data, packet_offset, Fs); + if (s!=0 && samples != tmp_samples) + return OPUS_INVALID_PACKET; + samples = tmp_samples; + data += packet_offset; + len -= packet_offset; + } + return samples; +} + +static int opus_multistream_decode_native( + OpusMSDecoder *st, + const unsigned char *data, + opus_int32 len, + void *pcm, + opus_copy_channel_out_func copy_channel_out, + int frame_size, + int decode_fec, + int soft_clip +) +{ + opus_int32 Fs; + int coupled_size; + int mono_size; + int s, c; + char *ptr; + int do_plc=0; + VARDECL(opus_val16, buf); + ALLOC_STACK; + + /* Limit frame_size to avoid excessive stack allocations. */ + opus_multistream_decoder_ctl(st, OPUS_GET_SAMPLE_RATE(&Fs)); + frame_size = IMIN(frame_size, Fs/25*3); + ALLOC(buf, 2*frame_size, opus_val16); + ptr = (char*)st + align(sizeof(OpusMSDecoder)); + coupled_size = opus_decoder_get_size(2); + mono_size = opus_decoder_get_size(1); + + if (len==0) + do_plc = 1; + if (len < 0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + if (!do_plc && len < 2*st->layout.nb_streams-1) + { + RESTORE_STACK; + return OPUS_INVALID_PACKET; + } + if (!do_plc) + { + int ret = opus_multistream_packet_validate(data, len, st->layout.nb_streams, Fs); + if (ret < 0) + { + RESTORE_STACK; + return ret; + } else if (ret > frame_size) + { + RESTORE_STACK; + return OPUS_BUFFER_TOO_SMALL; + } + } + for (s=0;s<st->layout.nb_streams;s++) + { + OpusDecoder *dec; + int packet_offset, ret; + + dec = (OpusDecoder*)ptr; + ptr += (s < st->layout.nb_coupled_streams) ? align(coupled_size) : align(mono_size); + + if (!do_plc && len<=0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + packet_offset = 0; + ret = opus_decode_native(dec, data, len, buf, frame_size, decode_fec, s!=st->layout.nb_streams-1, &packet_offset, soft_clip); + data += packet_offset; + len -= packet_offset; + if (ret <= 0) + { + RESTORE_STACK; + return ret; + } + frame_size = ret; + if (s < st->layout.nb_coupled_streams) + { + int chan, prev; + prev = -1; + /* Copy "left" audio to the channel(s) where it belongs */ + while ( (chan = get_left_channel(&st->layout, s, prev)) != -1) + { + (*copy_channel_out)(pcm, st->layout.nb_channels, chan, + buf, 2, frame_size); + prev = chan; + } + prev = -1; + /* Copy "right" audio to the channel(s) where it belongs */ + while ( (chan = get_right_channel(&st->layout, s, prev)) != -1) + { + (*copy_channel_out)(pcm, st->layout.nb_channels, chan, + buf+1, 2, frame_size); + prev = chan; + } + } else { + int chan, prev; + prev = -1; + /* Copy audio to the channel(s) where it belongs */ + while ( (chan = get_mono_channel(&st->layout, s, prev)) != -1) + { + (*copy_channel_out)(pcm, st->layout.nb_channels, chan, + buf, 1, frame_size); + prev = chan; + } + } + } + /* Handle muted channels */ + for (c=0;c<st->layout.nb_channels;c++) + { + if (st->layout.mapping[c] == 255) + { + (*copy_channel_out)(pcm, st->layout.nb_channels, c, + NULL, 0, frame_size); + } + } + RESTORE_STACK; + return frame_size; +} + +#if !defined(DISABLE_FLOAT_API) +static void opus_copy_channel_out_float( + void *dst, + int dst_stride, + int dst_channel, + const opus_val16 *src, + int src_stride, + int frame_size +) +{ + float *float_dst; + opus_int32 i; + float_dst = (float*)dst; + if (src != NULL) + { + for (i=0;i<frame_size;i++) +#if defined(FIXED_POINT) + float_dst[i*dst_stride+dst_channel] = (1/32768.f)*src[i*src_stride]; +#else + float_dst[i*dst_stride+dst_channel] = src[i*src_stride]; +#endif + } + else + { + for (i=0;i<frame_size;i++) + float_dst[i*dst_stride+dst_channel] = 0; + } +} +#endif + +static void opus_copy_channel_out_short( + void *dst, + int dst_stride, + int dst_channel, + const opus_val16 *src, + int src_stride, + int frame_size +) +{ + opus_int16 *short_dst; + opus_int32 i; + short_dst = (opus_int16*)dst; + if (src != NULL) + { + for (i=0;i<frame_size;i++) +#if defined(FIXED_POINT) + short_dst[i*dst_stride+dst_channel] = src[i*src_stride]; +#else + short_dst[i*dst_stride+dst_channel] = FLOAT2INT16(src[i*src_stride]); +#endif + } + else + { + for (i=0;i<frame_size;i++) + short_dst[i*dst_stride+dst_channel] = 0; + } +} + + + +#ifdef FIXED_POINT +int opus_multistream_decode( + OpusMSDecoder *st, + const unsigned char *data, + opus_int32 len, + opus_int16 *pcm, + int frame_size, + int decode_fec +) +{ + return opus_multistream_decode_native(st, data, len, + pcm, opus_copy_channel_out_short, frame_size, decode_fec, 0); +} + +#ifndef DISABLE_FLOAT_API +int opus_multistream_decode_float(OpusMSDecoder *st, const unsigned char *data, + opus_int32 len, float *pcm, int frame_size, int decode_fec) +{ + return opus_multistream_decode_native(st, data, len, + pcm, opus_copy_channel_out_float, frame_size, decode_fec, 0); +} +#endif + +#else + +int opus_multistream_decode(OpusMSDecoder *st, const unsigned char *data, + opus_int32 len, opus_int16 *pcm, int frame_size, int decode_fec) +{ + return opus_multistream_decode_native(st, data, len, + pcm, opus_copy_channel_out_short, frame_size, decode_fec, 1); +} + +int opus_multistream_decode_float( + OpusMSDecoder *st, + const unsigned char *data, + opus_int32 len, + float *pcm, + int frame_size, + int decode_fec +) +{ + return opus_multistream_decode_native(st, data, len, + pcm, opus_copy_channel_out_float, frame_size, decode_fec, 0); +} +#endif + +int opus_multistream_decoder_ctl(OpusMSDecoder *st, int request, ...) +{ + va_list ap; + int coupled_size, mono_size; + char *ptr; + int ret = OPUS_OK; + + va_start(ap, request); + + coupled_size = opus_decoder_get_size(2); + mono_size = opus_decoder_get_size(1); + ptr = (char*)st + align(sizeof(OpusMSDecoder)); + switch (request) + { + case OPUS_GET_BANDWIDTH_REQUEST: + case OPUS_GET_SAMPLE_RATE_REQUEST: + case OPUS_GET_GAIN_REQUEST: + case OPUS_GET_LAST_PACKET_DURATION_REQUEST: + { + OpusDecoder *dec; + /* For int32* GET params, just query the first stream */ + opus_int32 *value = va_arg(ap, opus_int32*); + dec = (OpusDecoder*)ptr; + ret = opus_decoder_ctl(dec, request, value); + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + int s; + opus_uint32 *value = va_arg(ap, opus_uint32*); + opus_uint32 tmp; + if (!value) + { + goto bad_arg; + } + *value = 0; + for (s=0;s<st->layout.nb_streams;s++) + { + OpusDecoder *dec; + dec = (OpusDecoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + ret = opus_decoder_ctl(dec, request, &tmp); + if (ret != OPUS_OK) break; + *value ^= tmp; + } + } + break; + case OPUS_RESET_STATE: + { + int s; + for (s=0;s<st->layout.nb_streams;s++) + { + OpusDecoder *dec; + + dec = (OpusDecoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + ret = opus_decoder_ctl(dec, OPUS_RESET_STATE); + if (ret != OPUS_OK) + break; + } + } + break; + case OPUS_MULTISTREAM_GET_DECODER_STATE_REQUEST: + { + int s; + opus_int32 stream_id; + OpusDecoder **value; + stream_id = va_arg(ap, opus_int32); + if (stream_id<0 || stream_id >= st->layout.nb_streams) + ret = OPUS_BAD_ARG; + value = va_arg(ap, OpusDecoder**); + if (!value) + { + goto bad_arg; + } + for (s=0;s<stream_id;s++) + { + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + } + *value = (OpusDecoder*)ptr; + } + break; + case OPUS_SET_GAIN_REQUEST: + { + int s; + /* This works for int32 params */ + opus_int32 value = va_arg(ap, opus_int32); + for (s=0;s<st->layout.nb_streams;s++) + { + OpusDecoder *dec; + + dec = (OpusDecoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + ret = opus_decoder_ctl(dec, request, value); + if (ret != OPUS_OK) + break; + } + } + break; + default: + ret = OPUS_UNIMPLEMENTED; + break; + } + + va_end(ap); + return ret; +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +} + + +void opus_multistream_decoder_destroy(OpusMSDecoder *st) +{ + opus_free(st); +} diff --git a/src/opus-1.1/src/opus_multistream_encoder.c b/src/opus-1.1/src/opus_multistream_encoder.c new file mode 100644 index 00000000..49e27913 --- /dev/null +++ b/src/opus-1.1/src/opus_multistream_encoder.c @@ -0,0 +1,1174 @@ +/* Copyright (c) 2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "opus_multistream.h" +#include "opus.h" +#include "opus_private.h" +#include "stack_alloc.h" +#include <stdarg.h> +#include "float_cast.h" +#include "os_support.h" +#include "mathops.h" +#include "mdct.h" +#include "modes.h" +#include "bands.h" +#include "quant_bands.h" + +typedef struct { + int nb_streams; + int nb_coupled_streams; + unsigned char mapping[8]; +} VorbisLayout; + +/* Index is nb_channel-1*/ +static const VorbisLayout vorbis_mappings[8] = { + {1, 0, {0}}, /* 1: mono */ + {1, 1, {0, 1}}, /* 2: stereo */ + {2, 1, {0, 2, 1}}, /* 3: 1-d surround */ + {2, 2, {0, 1, 2, 3}}, /* 4: quadraphonic surround */ + {3, 2, {0, 4, 1, 2, 3}}, /* 5: 5-channel surround */ + {4, 2, {0, 4, 1, 2, 3, 5}}, /* 6: 5.1 surround */ + {4, 3, {0, 4, 1, 2, 3, 5, 6}}, /* 7: 6.1 surround */ + {5, 3, {0, 6, 1, 2, 3, 4, 5, 7}}, /* 8: 7.1 surround */ +}; + +typedef void (*opus_copy_channel_in_func)( + opus_val16 *dst, + int dst_stride, + const void *src, + int src_stride, + int src_channel, + int frame_size +); + +struct OpusMSEncoder { + ChannelLayout layout; + int lfe_stream; + int application; + int variable_duration; + int surround; + opus_int32 bitrate_bps; + float subframe_mem[3]; + /* Encoder states go here */ + /* then opus_val32 window_mem[channels*120]; */ + /* then opus_val32 preemph_mem[channels]; */ +}; + +static opus_val32 *ms_get_preemph_mem(OpusMSEncoder *st) +{ + int s; + char *ptr; + int coupled_size, mono_size; + + coupled_size = opus_encoder_get_size(2); + mono_size = opus_encoder_get_size(1); + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + for (s=0;s<st->layout.nb_streams;s++) + { + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + } + return (opus_val32*)(ptr+st->layout.nb_channels*120*sizeof(opus_val32)); +} + +static opus_val32 *ms_get_window_mem(OpusMSEncoder *st) +{ + int s; + char *ptr; + int coupled_size, mono_size; + + coupled_size = opus_encoder_get_size(2); + mono_size = opus_encoder_get_size(1); + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + for (s=0;s<st->layout.nb_streams;s++) + { + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + } + return (opus_val32*)ptr; +} + +static int validate_encoder_layout(const ChannelLayout *layout) +{ + int s; + for (s=0;s<layout->nb_streams;s++) + { + if (s < layout->nb_coupled_streams) + { + if (get_left_channel(layout, s, -1)==-1) + return 0; + if (get_right_channel(layout, s, -1)==-1) + return 0; + } else { + if (get_mono_channel(layout, s, -1)==-1) + return 0; + } + } + return 1; +} + +static void channel_pos(int channels, int pos[8]) +{ + /* Position in the mix: 0 don't mix, 1: left, 2: center, 3:right */ + if (channels==4) + { + pos[0]=1; + pos[1]=3; + pos[2]=1; + pos[3]=3; + } else if (channels==3||channels==5||channels==6) + { + pos[0]=1; + pos[1]=2; + pos[2]=3; + pos[3]=1; + pos[4]=3; + pos[5]=0; + } else if (channels==7) + { + pos[0]=1; + pos[1]=2; + pos[2]=3; + pos[3]=1; + pos[4]=3; + pos[5]=2; + pos[6]=0; + } else if (channels==8) + { + pos[0]=1; + pos[1]=2; + pos[2]=3; + pos[3]=1; + pos[4]=3; + pos[5]=1; + pos[6]=3; + pos[7]=0; + } +} + +#if 1 +/* Computes a rough approximation of log2(2^a + 2^b) */ +static opus_val16 logSum(opus_val16 a, opus_val16 b) +{ + opus_val16 max; + opus_val32 diff; + opus_val16 frac; + static const opus_val16 diff_table[17] = { + QCONST16(0.5000000f, DB_SHIFT), QCONST16(0.2924813f, DB_SHIFT), QCONST16(0.1609640f, DB_SHIFT), QCONST16(0.0849625f, DB_SHIFT), + QCONST16(0.0437314f, DB_SHIFT), QCONST16(0.0221971f, DB_SHIFT), QCONST16(0.0111839f, DB_SHIFT), QCONST16(0.0056136f, DB_SHIFT), + QCONST16(0.0028123f, DB_SHIFT) + }; + int low; + if (a>b) + { + max = a; + diff = SUB32(EXTEND32(a),EXTEND32(b)); + } else { + max = b; + diff = SUB32(EXTEND32(b),EXTEND32(a)); + } + if (diff >= QCONST16(8.f, DB_SHIFT)) + return max; +#ifdef FIXED_POINT + low = SHR32(diff, DB_SHIFT-1); + frac = SHL16(diff - SHL16(low, DB_SHIFT-1), 16-DB_SHIFT); +#else + low = (int)floor(2*diff); + frac = 2*diff - low; +#endif + return max + diff_table[low] + MULT16_16_Q15(frac, SUB16(diff_table[low+1], diff_table[low])); +} +#else +opus_val16 logSum(opus_val16 a, opus_val16 b) +{ + return log2(pow(4, a)+ pow(4, b))/2; +} +#endif + +void surround_analysis(const CELTMode *celt_mode, const void *pcm, opus_val16 *bandLogE, opus_val32 *mem, opus_val32 *preemph_mem, + int len, int overlap, int channels, int rate, opus_copy_channel_in_func copy_channel_in +) +{ + int c; + int i; + int LM; + int pos[8] = {0}; + int upsample; + int frame_size; + opus_val16 channel_offset; + opus_val32 bandE[21]; + opus_val16 maskLogE[3][21]; + VARDECL(opus_val32, in); + VARDECL(opus_val16, x); + VARDECL(opus_val32, freq); + SAVE_STACK; + + upsample = resampling_factor(rate); + frame_size = len*upsample; + + for (LM=0;LM<celt_mode->maxLM;LM++) + if (celt_mode->shortMdctSize<<LM==frame_size) + break; + + ALLOC(in, frame_size+overlap, opus_val32); + ALLOC(x, len, opus_val16); + ALLOC(freq, frame_size, opus_val32); + + channel_pos(channels, pos); + + for (c=0;c<3;c++) + for (i=0;i<21;i++) + maskLogE[c][i] = -QCONST16(28.f, DB_SHIFT); + + for (c=0;c<channels;c++) + { + OPUS_COPY(in, mem+c*overlap, overlap); + (*copy_channel_in)(x, 1, pcm, channels, c, len); + celt_preemphasis(x, in+overlap, frame_size, 1, upsample, celt_mode->preemph, preemph_mem+c, 0); + clt_mdct_forward(&celt_mode->mdct, in, freq, celt_mode->window, overlap, celt_mode->maxLM-LM, 1); + if (upsample != 1) + { + int bound = len; + for (i=0;i<bound;i++) + freq[i] *= upsample; + for (;i<frame_size;i++) + freq[i] = 0; + } + + compute_band_energies(celt_mode, freq, bandE, 21, 1, 1<<LM); + amp2Log2(celt_mode, 21, 21, bandE, bandLogE+21*c, 1); + /* Apply spreading function with -6 dB/band going up and -12 dB/band going down. */ + for (i=1;i<21;i++) + bandLogE[21*c+i] = MAX16(bandLogE[21*c+i], bandLogE[21*c+i-1]-QCONST16(1.f, DB_SHIFT)); + for (i=19;i>=0;i--) + bandLogE[21*c+i] = MAX16(bandLogE[21*c+i], bandLogE[21*c+i+1]-QCONST16(2.f, DB_SHIFT)); + if (pos[c]==1) + { + for (i=0;i<21;i++) + maskLogE[0][i] = logSum(maskLogE[0][i], bandLogE[21*c+i]); + } else if (pos[c]==3) + { + for (i=0;i<21;i++) + maskLogE[2][i] = logSum(maskLogE[2][i], bandLogE[21*c+i]); + } else if (pos[c]==2) + { + for (i=0;i<21;i++) + { + maskLogE[0][i] = logSum(maskLogE[0][i], bandLogE[21*c+i]-QCONST16(.5f, DB_SHIFT)); + maskLogE[2][i] = logSum(maskLogE[2][i], bandLogE[21*c+i]-QCONST16(.5f, DB_SHIFT)); + } + } +#if 0 + for (i=0;i<21;i++) + printf("%f ", bandLogE[21*c+i]); + float sum=0; + for (i=0;i<21;i++) + sum += bandLogE[21*c+i]; + printf("%f ", sum/21); +#endif + OPUS_COPY(mem+c*overlap, in+frame_size, overlap); + } + for (i=0;i<21;i++) + maskLogE[1][i] = MIN32(maskLogE[0][i],maskLogE[2][i]); + channel_offset = HALF16(celt_log2(QCONST32(2.f,14)/(channels-1))); + for (c=0;c<3;c++) + for (i=0;i<21;i++) + maskLogE[c][i] += channel_offset; +#if 0 + for (c=0;c<3;c++) + { + for (i=0;i<21;i++) + printf("%f ", maskLogE[c][i]); + } +#endif + for (c=0;c<channels;c++) + { + opus_val16 *mask; + if (pos[c]!=0) + { + mask = &maskLogE[pos[c]-1][0]; + for (i=0;i<21;i++) + bandLogE[21*c+i] = bandLogE[21*c+i] - mask[i]; + } else { + for (i=0;i<21;i++) + bandLogE[21*c+i] = 0; + } +#if 0 + for (i=0;i<21;i++) + printf("%f ", bandLogE[21*c+i]); + printf("\n"); +#endif +#if 0 + float sum=0; + for (i=0;i<21;i++) + sum += bandLogE[21*c+i]; + printf("%f ", sum/(float)QCONST32(21.f, DB_SHIFT)); + printf("\n"); +#endif + } + RESTORE_STACK; +} + +opus_int32 opus_multistream_encoder_get_size(int nb_streams, int nb_coupled_streams) +{ + int coupled_size; + int mono_size; + + if(nb_streams<1||nb_coupled_streams>nb_streams||nb_coupled_streams<0)return 0; + coupled_size = opus_encoder_get_size(2); + mono_size = opus_encoder_get_size(1); + return align(sizeof(OpusMSEncoder)) + + nb_coupled_streams * align(coupled_size) + + (nb_streams-nb_coupled_streams) * align(mono_size); +} + +opus_int32 opus_multistream_surround_encoder_get_size(int channels, int mapping_family) +{ + int nb_streams; + int nb_coupled_streams; + opus_int32 size; + + if (mapping_family==0) + { + if (channels==1) + { + nb_streams=1; + nb_coupled_streams=0; + } else if (channels==2) + { + nb_streams=1; + nb_coupled_streams=1; + } else + return 0; + } else if (mapping_family==1 && channels<=8 && channels>=1) + { + nb_streams=vorbis_mappings[channels-1].nb_streams; + nb_coupled_streams=vorbis_mappings[channels-1].nb_coupled_streams; + } else if (mapping_family==255) + { + nb_streams=channels; + nb_coupled_streams=0; + } else + return 0; + size = opus_multistream_encoder_get_size(nb_streams, nb_coupled_streams); + if (channels>2) + { + size += channels*(120*sizeof(opus_val32) + sizeof(opus_val32)); + } + return size; +} + + +static int opus_multistream_encoder_init_impl( + OpusMSEncoder *st, + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping, + int application, + int surround +) +{ + int coupled_size; + int mono_size; + int i, ret; + char *ptr; + + if ((channels>255) || (channels<1) || (coupled_streams>streams) || + (coupled_streams+streams>255) || (streams<1) || (coupled_streams<0)) + return OPUS_BAD_ARG; + + st->layout.nb_channels = channels; + st->layout.nb_streams = streams; + st->layout.nb_coupled_streams = coupled_streams; + st->subframe_mem[0]=st->subframe_mem[1]=st->subframe_mem[2]=0; + if (!surround) + st->lfe_stream = -1; + st->bitrate_bps = OPUS_AUTO; + st->application = application; + st->variable_duration = OPUS_FRAMESIZE_ARG; + for (i=0;i<st->layout.nb_channels;i++) + st->layout.mapping[i] = mapping[i]; + if (!validate_layout(&st->layout) || !validate_encoder_layout(&st->layout)) + return OPUS_BAD_ARG; + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + coupled_size = opus_encoder_get_size(2); + mono_size = opus_encoder_get_size(1); + + for (i=0;i<st->layout.nb_coupled_streams;i++) + { + ret = opus_encoder_init((OpusEncoder*)ptr, Fs, 2, application); + if(ret!=OPUS_OK)return ret; + if (i==st->lfe_stream) + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_SET_LFE(1)); + ptr += align(coupled_size); + } + for (;i<st->layout.nb_streams;i++) + { + ret = opus_encoder_init((OpusEncoder*)ptr, Fs, 1, application); + if (i==st->lfe_stream) + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_SET_LFE(1)); + if(ret!=OPUS_OK)return ret; + ptr += align(mono_size); + } + if (surround) + { + OPUS_CLEAR(ms_get_preemph_mem(st), channels); + OPUS_CLEAR(ms_get_window_mem(st), channels*120); + } + st->surround = surround; + return OPUS_OK; +} + +int opus_multistream_encoder_init( + OpusMSEncoder *st, + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping, + int application +) +{ + return opus_multistream_encoder_init_impl(st, Fs, channels, streams, coupled_streams, mapping, application, 0); +} + +int opus_multistream_surround_encoder_init( + OpusMSEncoder *st, + opus_int32 Fs, + int channels, + int mapping_family, + int *streams, + int *coupled_streams, + unsigned char *mapping, + int application +) +{ + if ((channels>255) || (channels<1)) + return OPUS_BAD_ARG; + st->lfe_stream = -1; + if (mapping_family==0) + { + if (channels==1) + { + *streams=1; + *coupled_streams=0; + mapping[0]=0; + } else if (channels==2) + { + *streams=1; + *coupled_streams=1; + mapping[0]=0; + mapping[1]=1; + } else + return OPUS_UNIMPLEMENTED; + } else if (mapping_family==1 && channels<=8 && channels>=1) + { + int i; + *streams=vorbis_mappings[channels-1].nb_streams; + *coupled_streams=vorbis_mappings[channels-1].nb_coupled_streams; + for (i=0;i<channels;i++) + mapping[i] = vorbis_mappings[channels-1].mapping[i]; + if (channels>=6) + st->lfe_stream = *streams-1; + } else if (mapping_family==255) + { + int i; + *streams=channels; + *coupled_streams=0; + for(i=0;i<channels;i++) + mapping[i] = i; + } else + return OPUS_UNIMPLEMENTED; + return opus_multistream_encoder_init_impl(st, Fs, channels, *streams, *coupled_streams, + mapping, application, channels>2&&mapping_family==1); +} + +OpusMSEncoder *opus_multistream_encoder_create( + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping, + int application, + int *error +) +{ + int ret; + OpusMSEncoder *st; + if ((channels>255) || (channels<1) || (coupled_streams>streams) || + (coupled_streams+streams>255) || (streams<1) || (coupled_streams<0)) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + st = (OpusMSEncoder *)opus_alloc(opus_multistream_encoder_get_size(streams, coupled_streams)); + if (st==NULL) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + ret = opus_multistream_encoder_init(st, Fs, channels, streams, coupled_streams, mapping, application); + if (ret != OPUS_OK) + { + opus_free(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} + +OpusMSEncoder *opus_multistream_surround_encoder_create( + opus_int32 Fs, + int channels, + int mapping_family, + int *streams, + int *coupled_streams, + unsigned char *mapping, + int application, + int *error +) +{ + int ret; + OpusMSEncoder *st; + if ((channels>255) || (channels<1)) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + st = (OpusMSEncoder *)opus_alloc(opus_multistream_surround_encoder_get_size(channels, mapping_family)); + if (st==NULL) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + ret = opus_multistream_surround_encoder_init(st, Fs, channels, mapping_family, streams, coupled_streams, mapping, application); + if (ret != OPUS_OK) + { + opus_free(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} + +static void surround_rate_allocation( + OpusMSEncoder *st, + opus_int32 *rate, + int frame_size + ) +{ + int i; + opus_int32 channel_rate; + opus_int32 Fs; + char *ptr; + int stream_offset; + int lfe_offset; + int coupled_ratio; /* Q8 */ + int lfe_ratio; /* Q8 */ + + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_GET_SAMPLE_RATE(&Fs)); + + if (st->bitrate_bps > st->layout.nb_channels*40000) + stream_offset = 20000; + else + stream_offset = st->bitrate_bps/st->layout.nb_channels/2; + stream_offset += 60*(Fs/frame_size-50); + /* We start by giving each stream (coupled or uncoupled) the same bitrate. + This models the main saving of coupled channels over uncoupled. */ + /* The LFE stream is an exception to the above and gets fewer bits. */ + lfe_offset = 3500 + 60*(Fs/frame_size-50); + /* Coupled streams get twice the mono rate after the first 20 kb/s. */ + coupled_ratio = 512; + /* Should depend on the bitrate, for now we assume LFE gets 1/8 the bits of mono */ + lfe_ratio = 32; + + /* Compute bitrate allocation between streams */ + if (st->bitrate_bps==OPUS_AUTO) + { + channel_rate = Fs+60*Fs/frame_size; + } else if (st->bitrate_bps==OPUS_BITRATE_MAX) + { + channel_rate = 300000; + } else { + int nb_lfe; + int nb_uncoupled; + int nb_coupled; + int total; + nb_lfe = (st->lfe_stream!=-1); + nb_coupled = st->layout.nb_coupled_streams; + nb_uncoupled = st->layout.nb_streams-nb_coupled-nb_lfe; + total = (nb_uncoupled<<8) /* mono */ + + coupled_ratio*nb_coupled /* stereo */ + + nb_lfe*lfe_ratio; + channel_rate = 256*(st->bitrate_bps-lfe_offset*nb_lfe-stream_offset*(nb_coupled+nb_uncoupled))/total; + } +#ifndef FIXED_POINT + if (st->variable_duration==OPUS_FRAMESIZE_VARIABLE && frame_size != Fs/50) + { + opus_int32 bonus; + bonus = 60*(Fs/frame_size-50); + channel_rate += bonus; + } +#endif + + for (i=0;i<st->layout.nb_streams;i++) + { + if (i<st->layout.nb_coupled_streams) + rate[i] = stream_offset+(channel_rate*coupled_ratio>>8); + else if (i!=st->lfe_stream) + rate[i] = stream_offset+channel_rate; + else + rate[i] = lfe_offset+(channel_rate*lfe_ratio>>8); + } +} + +/* Max size in case the encoder decides to return three frames */ +#define MS_FRAME_TMP (3*1275+7) +static int opus_multistream_encode_native +( + OpusMSEncoder *st, + opus_copy_channel_in_func copy_channel_in, + const void *pcm, + int analysis_frame_size, + unsigned char *data, + opus_int32 max_data_bytes, + int lsb_depth, + downmix_func downmix +) +{ + opus_int32 Fs; + int coupled_size; + int mono_size; + int s; + char *ptr; + int tot_size; + VARDECL(opus_val16, buf); + VARDECL(opus_val16, bandSMR); + unsigned char tmp_data[MS_FRAME_TMP]; + OpusRepacketizer rp; + opus_int32 vbr; + const CELTMode *celt_mode; + opus_int32 bitrates[256]; + opus_val16 bandLogE[42]; + opus_val32 *mem = NULL; + opus_val32 *preemph_mem=NULL; + int frame_size; + ALLOC_STACK; + + if (st->surround) + { + preemph_mem = ms_get_preemph_mem(st); + mem = ms_get_window_mem(st); + } + + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_GET_SAMPLE_RATE(&Fs)); + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_GET_VBR(&vbr)); + opus_encoder_ctl((OpusEncoder*)ptr, CELT_GET_MODE(&celt_mode)); + + { + opus_int32 delay_compensation; + int channels; + + channels = st->layout.nb_streams + st->layout.nb_coupled_streams; + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_GET_LOOKAHEAD(&delay_compensation)); + delay_compensation -= Fs/400; + frame_size = compute_frame_size(pcm, analysis_frame_size, + st->variable_duration, channels, Fs, st->bitrate_bps, + delay_compensation, downmix +#ifndef DISABLE_FLOAT_API + , st->subframe_mem +#endif + ); + } + + if (400*frame_size < Fs) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + /* Validate frame_size before using it to allocate stack space. + This mirrors the checks in opus_encode[_float](). */ + if (400*frame_size != Fs && 200*frame_size != Fs && + 100*frame_size != Fs && 50*frame_size != Fs && + 25*frame_size != Fs && 50*frame_size != 3*Fs) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + ALLOC(buf, 2*frame_size, opus_val16); + coupled_size = opus_encoder_get_size(2); + mono_size = opus_encoder_get_size(1); + + ALLOC(bandSMR, 21*st->layout.nb_channels, opus_val16); + if (st->surround) + { + surround_analysis(celt_mode, pcm, bandSMR, mem, preemph_mem, frame_size, 120, st->layout.nb_channels, Fs, copy_channel_in); + } + + if (max_data_bytes < 4*st->layout.nb_streams-1) + { + RESTORE_STACK; + return OPUS_BUFFER_TOO_SMALL; + } + + /* Compute bitrate allocation between streams (this could be a lot better) */ + surround_rate_allocation(st, bitrates, frame_size); + + if (!vbr) + max_data_bytes = IMIN(max_data_bytes, 3*st->bitrate_bps/(3*8*Fs/frame_size)); + + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + for (s=0;s<st->layout.nb_streams;s++) + { + OpusEncoder *enc; + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + opus_encoder_ctl(enc, OPUS_SET_BITRATE(bitrates[s])); + if (st->surround) + { + opus_int32 equiv_rate; + equiv_rate = st->bitrate_bps; + if (frame_size*50 < Fs) + equiv_rate -= 60*(Fs/frame_size - 50)*st->layout.nb_channels; + if (equiv_rate > 10000*st->layout.nb_channels) + opus_encoder_ctl(enc, OPUS_SET_BANDWIDTH(OPUS_BANDWIDTH_FULLBAND)); + else if (equiv_rate > 7000*st->layout.nb_channels) + opus_encoder_ctl(enc, OPUS_SET_BANDWIDTH(OPUS_BANDWIDTH_SUPERWIDEBAND)); + else if (equiv_rate > 5000*st->layout.nb_channels) + opus_encoder_ctl(enc, OPUS_SET_BANDWIDTH(OPUS_BANDWIDTH_WIDEBAND)); + else + opus_encoder_ctl(enc, OPUS_SET_BANDWIDTH(OPUS_BANDWIDTH_NARROWBAND)); + if (s < st->layout.nb_coupled_streams) + { + /* To preserve the spatial image, force stereo CELT on coupled streams */ + opus_encoder_ctl(enc, OPUS_SET_FORCE_MODE(MODE_CELT_ONLY)); + opus_encoder_ctl(enc, OPUS_SET_FORCE_CHANNELS(2)); + } + } + } + + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + /* Counting ToC */ + tot_size = 0; + for (s=0;s<st->layout.nb_streams;s++) + { + OpusEncoder *enc; + int len; + int curr_max; + int c1, c2; + + opus_repacketizer_init(&rp); + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + { + int i; + int left, right; + left = get_left_channel(&st->layout, s, -1); + right = get_right_channel(&st->layout, s, -1); + (*copy_channel_in)(buf, 2, + pcm, st->layout.nb_channels, left, frame_size); + (*copy_channel_in)(buf+1, 2, + pcm, st->layout.nb_channels, right, frame_size); + ptr += align(coupled_size); + if (st->surround) + { + for (i=0;i<21;i++) + { + bandLogE[i] = bandSMR[21*left+i]; + bandLogE[21+i] = bandSMR[21*right+i]; + } + } + c1 = left; + c2 = right; + } else { + int i; + int chan = get_mono_channel(&st->layout, s, -1); + (*copy_channel_in)(buf, 1, + pcm, st->layout.nb_channels, chan, frame_size); + ptr += align(mono_size); + if (st->surround) + { + for (i=0;i<21;i++) + bandLogE[i] = bandSMR[21*chan+i]; + } + c1 = chan; + c2 = -1; + } + if (st->surround) + opus_encoder_ctl(enc, OPUS_SET_ENERGY_MASK(bandLogE)); + /* number of bytes left (+Toc) */ + curr_max = max_data_bytes - tot_size; + /* Reserve three bytes for the last stream and four for the others */ + curr_max -= IMAX(0,4*(st->layout.nb_streams-s-1)-1); + curr_max = IMIN(curr_max,MS_FRAME_TMP); + if (!vbr && s == st->layout.nb_streams-1) + opus_encoder_ctl(enc, OPUS_SET_BITRATE(curr_max*(8*Fs/frame_size))); + len = opus_encode_native(enc, buf, frame_size, tmp_data, curr_max, lsb_depth, + pcm, analysis_frame_size, c1, c2, st->layout.nb_channels, downmix); + if (len<0) + { + RESTORE_STACK; + return len; + } + /* We need to use the repacketizer to add the self-delimiting lengths + while taking into account the fact that the encoder can now return + more than one frame at a time (e.g. 60 ms CELT-only) */ + opus_repacketizer_cat(&rp, tmp_data, len); + len = opus_repacketizer_out_range_impl(&rp, 0, opus_repacketizer_get_nb_frames(&rp), + data, max_data_bytes-tot_size, s != st->layout.nb_streams-1, !vbr && s == st->layout.nb_streams-1); + data += len; + tot_size += len; + } + /*printf("\n");*/ + RESTORE_STACK; + return tot_size; +} + +#if !defined(DISABLE_FLOAT_API) +static void opus_copy_channel_in_float( + opus_val16 *dst, + int dst_stride, + const void *src, + int src_stride, + int src_channel, + int frame_size +) +{ + const float *float_src; + opus_int32 i; + float_src = (const float *)src; + for (i=0;i<frame_size;i++) +#if defined(FIXED_POINT) + dst[i*dst_stride] = FLOAT2INT16(float_src[i*src_stride+src_channel]); +#else + dst[i*dst_stride] = float_src[i*src_stride+src_channel]; +#endif +} +#endif + +static void opus_copy_channel_in_short( + opus_val16 *dst, + int dst_stride, + const void *src, + int src_stride, + int src_channel, + int frame_size +) +{ + const opus_int16 *short_src; + opus_int32 i; + short_src = (const opus_int16 *)src; + for (i=0;i<frame_size;i++) +#if defined(FIXED_POINT) + dst[i*dst_stride] = short_src[i*src_stride+src_channel]; +#else + dst[i*dst_stride] = (1/32768.f)*short_src[i*src_stride+src_channel]; +#endif +} + + +#ifdef FIXED_POINT +int opus_multistream_encode( + OpusMSEncoder *st, + const opus_val16 *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) +{ + return opus_multistream_encode_native(st, opus_copy_channel_in_short, + pcm, frame_size, data, max_data_bytes, 16, downmix_int); +} + +#ifndef DISABLE_FLOAT_API +int opus_multistream_encode_float( + OpusMSEncoder *st, + const float *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) +{ + return opus_multistream_encode_native(st, opus_copy_channel_in_float, + pcm, frame_size, data, max_data_bytes, 16, downmix_float); +} +#endif + +#else + +int opus_multistream_encode_float +( + OpusMSEncoder *st, + const opus_val16 *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) +{ + return opus_multistream_encode_native(st, opus_copy_channel_in_float, + pcm, frame_size, data, max_data_bytes, 24, downmix_float); +} + +int opus_multistream_encode( + OpusMSEncoder *st, + const opus_int16 *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) +{ + return opus_multistream_encode_native(st, opus_copy_channel_in_short, + pcm, frame_size, data, max_data_bytes, 16, downmix_int); +} +#endif + +int opus_multistream_encoder_ctl(OpusMSEncoder *st, int request, ...) +{ + va_list ap; + int coupled_size, mono_size; + char *ptr; + int ret = OPUS_OK; + + va_start(ap, request); + + coupled_size = opus_encoder_get_size(2); + mono_size = opus_encoder_get_size(1); + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + switch (request) + { + case OPUS_SET_BITRATE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<0 && value!=OPUS_AUTO && value!=OPUS_BITRATE_MAX) + { + goto bad_arg; + } + st->bitrate_bps = value; + } + break; + case OPUS_GET_BITRATE_REQUEST: + { + int s; + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = 0; + for (s=0;s<st->layout.nb_streams;s++) + { + opus_int32 rate; + OpusEncoder *enc; + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + opus_encoder_ctl(enc, request, &rate); + *value += rate; + } + } + break; + case OPUS_GET_LSB_DEPTH_REQUEST: + case OPUS_GET_VBR_REQUEST: + case OPUS_GET_APPLICATION_REQUEST: + case OPUS_GET_BANDWIDTH_REQUEST: + case OPUS_GET_COMPLEXITY_REQUEST: + case OPUS_GET_PACKET_LOSS_PERC_REQUEST: + case OPUS_GET_DTX_REQUEST: + case OPUS_GET_VOICE_RATIO_REQUEST: + case OPUS_GET_VBR_CONSTRAINT_REQUEST: + case OPUS_GET_SIGNAL_REQUEST: + case OPUS_GET_LOOKAHEAD_REQUEST: + case OPUS_GET_SAMPLE_RATE_REQUEST: + case OPUS_GET_INBAND_FEC_REQUEST: + case OPUS_GET_FORCE_CHANNELS_REQUEST: + case OPUS_GET_PREDICTION_DISABLED_REQUEST: + { + OpusEncoder *enc; + /* For int32* GET params, just query the first stream */ + opus_int32 *value = va_arg(ap, opus_int32*); + enc = (OpusEncoder*)ptr; + ret = opus_encoder_ctl(enc, request, value); + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + int s; + opus_uint32 *value = va_arg(ap, opus_uint32*); + opus_uint32 tmp; + if (!value) + { + goto bad_arg; + } + *value=0; + for (s=0;s<st->layout.nb_streams;s++) + { + OpusEncoder *enc; + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + ret = opus_encoder_ctl(enc, request, &tmp); + if (ret != OPUS_OK) break; + *value ^= tmp; + } + } + break; + case OPUS_SET_LSB_DEPTH_REQUEST: + case OPUS_SET_COMPLEXITY_REQUEST: + case OPUS_SET_VBR_REQUEST: + case OPUS_SET_VBR_CONSTRAINT_REQUEST: + case OPUS_SET_MAX_BANDWIDTH_REQUEST: + case OPUS_SET_BANDWIDTH_REQUEST: + case OPUS_SET_SIGNAL_REQUEST: + case OPUS_SET_APPLICATION_REQUEST: + case OPUS_SET_INBAND_FEC_REQUEST: + case OPUS_SET_PACKET_LOSS_PERC_REQUEST: + case OPUS_SET_DTX_REQUEST: + case OPUS_SET_FORCE_MODE_REQUEST: + case OPUS_SET_FORCE_CHANNELS_REQUEST: + case OPUS_SET_PREDICTION_DISABLED_REQUEST: + { + int s; + /* This works for int32 params */ + opus_int32 value = va_arg(ap, opus_int32); + for (s=0;s<st->layout.nb_streams;s++) + { + OpusEncoder *enc; + + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + ret = opus_encoder_ctl(enc, request, value); + if (ret != OPUS_OK) + break; + } + } + break; + case OPUS_MULTISTREAM_GET_ENCODER_STATE_REQUEST: + { + int s; + opus_int32 stream_id; + OpusEncoder **value; + stream_id = va_arg(ap, opus_int32); + if (stream_id<0 || stream_id >= st->layout.nb_streams) + ret = OPUS_BAD_ARG; + value = va_arg(ap, OpusEncoder**); + if (!value) + { + goto bad_arg; + } + for (s=0;s<stream_id;s++) + { + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + } + *value = (OpusEncoder*)ptr; + } + break; + case OPUS_SET_EXPERT_FRAME_DURATION_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->variable_duration = value; + } + break; + case OPUS_GET_EXPERT_FRAME_DURATION_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->variable_duration; + } + break; + case OPUS_RESET_STATE: + { + int s; + st->subframe_mem[0] = st->subframe_mem[1] = st->subframe_mem[2] = 0; + if (st->surround) + { + OPUS_CLEAR(ms_get_preemph_mem(st), st->layout.nb_channels); + OPUS_CLEAR(ms_get_window_mem(st), st->layout.nb_channels*120); + } + for (s=0;s<st->layout.nb_streams;s++) + { + OpusEncoder *enc; + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + ret = opus_encoder_ctl(enc, OPUS_RESET_STATE); + if (ret != OPUS_OK) + break; + } + } + break; + default: + ret = OPUS_UNIMPLEMENTED; + break; + } + + va_end(ap); + return ret; +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +} + +void opus_multistream_encoder_destroy(OpusMSEncoder *st) +{ + opus_free(st); +} diff --git a/src/opus-1.0.2/src/opus_private.h b/src/opus-1.1/src/opus_private.h index 52482bc1..83225f2b 100644 --- a/src/opus-1.0.2/src/opus_private.h +++ b/src/opus-1.1/src/opus_private.h @@ -31,15 +31,29 @@ #include "arch.h" #include "opus.h" +#include "celt.h" struct OpusRepacketizer { unsigned char toc; int nb_frames; const unsigned char *frames[48]; - short len[48]; + opus_int16 len[48]; int framesize; }; +typedef struct ChannelLayout { + int nb_channels; + int nb_streams; + int nb_coupled_streams; + unsigned char mapping[256]; +} ChannelLayout; + +int validate_layout(const ChannelLayout *layout); +int get_left_channel(const ChannelLayout *layout, int stream_id, int prev); +int get_right_channel(const ChannelLayout *layout, int stream_id, int prev); +int get_mono_channel(const ChannelLayout *layout, int stream_id, int prev); + + #define MODE_SILK_ONLY 1000 #define MODE_HYBRID 1001 @@ -68,18 +82,48 @@ struct OpusRepacketizer { #define OPUS_SET_FORCE_MODE_REQUEST 11002 #define OPUS_SET_FORCE_MODE(x) OPUS_SET_FORCE_MODE_REQUEST, __opus_check_int(x) +typedef void (*downmix_func)(const void *, opus_val32 *, int, int, int, int, int); +void downmix_float(const void *_x, opus_val32 *sub, int subframe, int offset, int c1, int c2, int C); +void downmix_int(const void *_x, opus_val32 *sub, int subframe, int offset, int c1, int c2, int C); + +int optimize_framesize(const opus_val16 *x, int len, int C, opus_int32 Fs, + int bitrate, opus_val16 tonality, float *mem, int buffering, + downmix_func downmix); int encode_size(int size, unsigned char *data); +opus_int32 frame_size_select(opus_int32 frame_size, int variable_duration, opus_int32 Fs); + +opus_int32 compute_frame_size(const void *analysis_pcm, int frame_size, + int variable_duration, int C, opus_int32 Fs, int bitrate_bps, + int delay_compensation, downmix_func downmix +#ifndef DISABLE_FLOAT_API + , float *subframe_mem +#endif + ); + +opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_size, + unsigned char *data, opus_int32 out_data_bytes, int lsb_depth, + const void *analysis_pcm, opus_int32 analysis_size, int c1, int c2, int analysis_channels, downmix_func downmix); + int opus_decode_native(OpusDecoder *st, const unsigned char *data, opus_int32 len, - opus_val16 *pcm, int frame_size, int decode_fec, int self_delimited, int *packet_offset); + opus_val16 *pcm, int frame_size, int decode_fec, int self_delimited, + opus_int32 *packet_offset, int soft_clip); /* Make sure everything's aligned to sizeof(void *) bytes */ -static inline int align(int i) +static OPUS_INLINE int align(int i) { - return (i+sizeof(void *)-1)&-((int)sizeof(void *)); + return (i+(int)sizeof(void *)-1)&-(int)sizeof(void *); } -opus_int32 opus_repacketizer_out_range_impl(OpusRepacketizer *rp, int begin, int end, unsigned char *data, opus_int32 maxlen, int self_delimited); +int opus_packet_parse_impl(const unsigned char *data, opus_int32 len, + int self_delimited, unsigned char *out_toc, + const unsigned char *frames[48], opus_int16 size[48], + int *payload_offset, opus_int32 *packet_offset); + +opus_int32 opus_repacketizer_out_range_impl(OpusRepacketizer *rp, int begin, int end, + unsigned char *data, opus_int32 maxlen, int self_delimited, int pad); + +int pad_frame(unsigned char *data, opus_int32 len, opus_int32 new_len); #endif /* OPUS_PRIVATE_H */ diff --git a/src/opus-1.0.2/src/repacketizer.c b/src/opus-1.1/src/repacketizer.c index 26315b62..a62675ce 100644 --- a/src/opus-1.0.2/src/repacketizer.c +++ b/src/opus-1.1/src/repacketizer.c @@ -58,7 +58,7 @@ void opus_repacketizer_destroy(OpusRepacketizer *rp) opus_free(rp); } -int opus_repacketizer_cat(OpusRepacketizer *rp, const unsigned char *data, opus_int32 len) +static int opus_repacketizer_cat_impl(OpusRepacketizer *rp, const unsigned char *data, opus_int32 len, int self_delimited) { unsigned char tmp_toc; int curr_nb_frames,ret; @@ -82,24 +82,31 @@ int opus_repacketizer_cat(OpusRepacketizer *rp, const unsigned char *data, opus_ return OPUS_INVALID_PACKET; } - ret=opus_packet_parse(data, len, &tmp_toc, &rp->frames[rp->nb_frames], &rp->len[rp->nb_frames], NULL); + ret=opus_packet_parse_impl(data, len, self_delimited, &tmp_toc, &rp->frames[rp->nb_frames], &rp->len[rp->nb_frames], NULL, NULL); if(ret<1)return ret; rp->nb_frames += curr_nb_frames; return OPUS_OK; } +int opus_repacketizer_cat(OpusRepacketizer *rp, const unsigned char *data, opus_int32 len) +{ + return opus_repacketizer_cat_impl(rp, data, len, 0); +} + int opus_repacketizer_get_nb_frames(OpusRepacketizer *rp) { return rp->nb_frames; } -opus_int32 opus_repacketizer_out_range_impl(OpusRepacketizer *rp, int begin, int end, unsigned char *data, opus_int32 maxlen, int self_delimited) +opus_int32 opus_repacketizer_out_range_impl(OpusRepacketizer *rp, int begin, int end, + unsigned char *data, opus_int32 maxlen, int self_delimited, int pad) { int i, count; opus_int32 tot_size; - short *len; + opus_int16 *len; const unsigned char **frames; + unsigned char * ptr; if (begin<0 || begin>=end || end>rp->nb_frames) { @@ -115,18 +122,15 @@ opus_int32 opus_repacketizer_out_range_impl(OpusRepacketizer *rp, int begin, int else tot_size = 0; - switch (count) - { - case 1: + ptr = data; + if (count==1) { /* Code 0 */ tot_size += len[0]+1; if (tot_size > maxlen) return OPUS_BUFFER_TOO_SMALL; - *data++ = rp->toc&0xFC; - } - break; - case 2: + *ptr++ = rp->toc&0xFC; + } else if (count==2) { if (len[1] == len[0]) { @@ -134,22 +138,28 @@ opus_int32 opus_repacketizer_out_range_impl(OpusRepacketizer *rp, int begin, int tot_size += 2*len[0]+1; if (tot_size > maxlen) return OPUS_BUFFER_TOO_SMALL; - *data++ = (rp->toc&0xFC) | 0x1; + *ptr++ = (rp->toc&0xFC) | 0x1; } else { /* Code 2 */ tot_size += len[0]+len[1]+2+(len[0]>=252); if (tot_size > maxlen) return OPUS_BUFFER_TOO_SMALL; - *data++ = (rp->toc&0xFC) | 0x2; - data += encode_size(len[0], data); + *ptr++ = (rp->toc&0xFC) | 0x2; + ptr += encode_size(len[0], ptr); } } - break; - default: + if (count > 2 || (pad && tot_size < maxlen)) { /* Code 3 */ int vbr; + int pad_amount=0; + /* Restart the process for the padding case */ + ptr = data; + if (self_delimited) + tot_size = 1 + (len[count-1]>=252); + else + tot_size = 0; vbr = 0; for (i=1;i<count;i++) { @@ -168,41 +178,168 @@ opus_int32 opus_repacketizer_out_range_impl(OpusRepacketizer *rp, int begin, int if (tot_size > maxlen) return OPUS_BUFFER_TOO_SMALL; - *data++ = (rp->toc&0xFC) | 0x3; - *data++ = count | 0x80; - for (i=0;i<count-1;i++) - data += encode_size(len[i], data); + *ptr++ = (rp->toc&0xFC) | 0x3; + *ptr++ = count | 0x80; } else { tot_size += count*len[0]+2; if (tot_size > maxlen) return OPUS_BUFFER_TOO_SMALL; - *data++ = (rp->toc&0xFC) | 0x3; - *data++ = count; + *ptr++ = (rp->toc&0xFC) | 0x3; + *ptr++ = count; + } + pad_amount = pad ? (maxlen-tot_size) : 0; + if (pad_amount != 0) + { + int nb_255s; + data[1] |= 0x40; + nb_255s = (pad_amount-1)/255; + for (i=0;i<nb_255s;i++) + *ptr++ = 255; + *ptr++ = pad_amount-255*nb_255s-1; + tot_size += pad_amount; + } + if (vbr) + { + for (i=0;i<count-1;i++) + ptr += encode_size(len[i], ptr); } - } - break; } if (self_delimited) { - int sdlen = encode_size(len[count-1], data); - data += sdlen; + int sdlen = encode_size(len[count-1], ptr); + ptr += sdlen; } /* Copy the actual data */ for (i=0;i<count;i++) { - OPUS_COPY(data, frames[i], len[i]); - data += len[i]; + /* Using OPUS_MOVE() instead of OPUS_COPY() in case we're doing in-place + padding from opus_packet_pad or opus_packet_unpad(). */ + celt_assert(frames[i] + len[i] <= data || ptr <= frames[i]); + OPUS_MOVE(ptr, frames[i], len[i]); + ptr += len[i]; + } + if (pad) + { + for (i=ptr-data;i<maxlen;i++) + data[i] = 0; } return tot_size; } opus_int32 opus_repacketizer_out_range(OpusRepacketizer *rp, int begin, int end, unsigned char *data, opus_int32 maxlen) { - return opus_repacketizer_out_range_impl(rp, begin, end, data, maxlen, 0); + return opus_repacketizer_out_range_impl(rp, begin, end, data, maxlen, 0, 0); } opus_int32 opus_repacketizer_out(OpusRepacketizer *rp, unsigned char *data, opus_int32 maxlen) { - return opus_repacketizer_out_range_impl(rp, 0, rp->nb_frames, data, maxlen, 0); + return opus_repacketizer_out_range_impl(rp, 0, rp->nb_frames, data, maxlen, 0, 0); } +int opus_packet_pad(unsigned char *data, opus_int32 len, opus_int32 new_len) +{ + OpusRepacketizer rp; + opus_int32 ret; + if (len < 1) + return OPUS_BAD_ARG; + if (len==new_len) + return OPUS_OK; + else if (len > new_len) + return OPUS_BAD_ARG; + opus_repacketizer_init(&rp); + /* Moving payload to the end of the packet so we can do in-place padding */ + OPUS_MOVE(data+new_len-len, data, len); + opus_repacketizer_cat(&rp, data+new_len-len, len); + ret = opus_repacketizer_out_range_impl(&rp, 0, rp.nb_frames, data, new_len, 0, 1); + if (ret > 0) + return OPUS_OK; + else + return ret; +} + +opus_int32 opus_packet_unpad(unsigned char *data, opus_int32 len) +{ + OpusRepacketizer rp; + opus_int32 ret; + if (len < 1) + return OPUS_BAD_ARG; + opus_repacketizer_init(&rp); + ret = opus_repacketizer_cat(&rp, data, len); + if (ret < 0) + return ret; + ret = opus_repacketizer_out_range_impl(&rp, 0, rp.nb_frames, data, len, 0, 0); + celt_assert(ret > 0 && ret <= len); + return ret; +} + +int opus_multistream_packet_pad(unsigned char *data, opus_int32 len, opus_int32 new_len, int nb_streams) +{ + int s; + int count; + unsigned char toc; + opus_int16 size[48]; + opus_int32 packet_offset; + opus_int32 amount; + + if (len < 1) + return OPUS_BAD_ARG; + if (len==new_len) + return OPUS_OK; + else if (len > new_len) + return OPUS_BAD_ARG; + amount = new_len - len; + /* Seek to last stream */ + for (s=0;s<nb_streams-1;s++) + { + if (len<=0) + return OPUS_INVALID_PACKET; + count = opus_packet_parse_impl(data, len, 1, &toc, NULL, + size, NULL, &packet_offset); + if (count<0) + return count; + data += packet_offset; + len -= packet_offset; + } + return opus_packet_pad(data, len, len+amount); +} + +opus_int32 opus_multistream_packet_unpad(unsigned char *data, opus_int32 len, int nb_streams) +{ + int s; + unsigned char toc; + opus_int16 size[48]; + opus_int32 packet_offset; + OpusRepacketizer rp; + unsigned char *dst; + opus_int32 dst_len; + + if (len < 1) + return OPUS_BAD_ARG; + dst = data; + dst_len = 0; + /* Unpad all frames */ + for (s=0;s<nb_streams;s++) + { + opus_int32 ret; + int self_delimited = s!=nb_streams-1; + if (len<=0) + return OPUS_INVALID_PACKET; + opus_repacketizer_init(&rp); + ret = opus_packet_parse_impl(data, len, self_delimited, &toc, NULL, + size, NULL, &packet_offset); + if (ret<0) + return ret; + ret = opus_repacketizer_cat_impl(&rp, data, packet_offset, self_delimited); + if (ret < 0) + return ret; + ret = opus_repacketizer_out_range_impl(&rp, 0, rp.nb_frames, dst, len, self_delimited, 0); + if (ret < 0) + return ret; + else + dst_len += ret; + dst += ret; + data += packet_offset; + len -= packet_offset; + } + return dst_len; +} diff --git a/src/opus-1.1/src/tansig_table.h b/src/opus-1.1/src/tansig_table.h new file mode 100644 index 00000000..c76f844a --- /dev/null +++ b/src/opus-1.1/src/tansig_table.h @@ -0,0 +1,45 @@ +/* This file is auto-generated by gen_tables */ + +static const float tansig_table[201] = { +0.000000f, 0.039979f, 0.079830f, 0.119427f, 0.158649f, +0.197375f, 0.235496f, 0.272905f, 0.309507f, 0.345214f, +0.379949f, 0.413644f, 0.446244f, 0.477700f, 0.507977f, +0.537050f, 0.564900f, 0.591519f, 0.616909f, 0.641077f, +0.664037f, 0.685809f, 0.706419f, 0.725897f, 0.744277f, +0.761594f, 0.777888f, 0.793199f, 0.807569f, 0.821040f, +0.833655f, 0.845456f, 0.856485f, 0.866784f, 0.876393f, +0.885352f, 0.893698f, 0.901468f, 0.908698f, 0.915420f, +0.921669f, 0.927473f, 0.932862f, 0.937863f, 0.942503f, +0.946806f, 0.950795f, 0.954492f, 0.957917f, 0.961090f, +0.964028f, 0.966747f, 0.969265f, 0.971594f, 0.973749f, +0.975743f, 0.977587f, 0.979293f, 0.980869f, 0.982327f, +0.983675f, 0.984921f, 0.986072f, 0.987136f, 0.988119f, +0.989027f, 0.989867f, 0.990642f, 0.991359f, 0.992020f, +0.992631f, 0.993196f, 0.993718f, 0.994199f, 0.994644f, +0.995055f, 0.995434f, 0.995784f, 0.996108f, 0.996407f, +0.996682f, 0.996937f, 0.997172f, 0.997389f, 0.997590f, +0.997775f, 0.997946f, 0.998104f, 0.998249f, 0.998384f, +0.998508f, 0.998623f, 0.998728f, 0.998826f, 0.998916f, +0.999000f, 0.999076f, 0.999147f, 0.999213f, 0.999273f, +0.999329f, 0.999381f, 0.999428f, 0.999472f, 0.999513f, +0.999550f, 0.999585f, 0.999617f, 0.999646f, 0.999673f, +0.999699f, 0.999722f, 0.999743f, 0.999763f, 0.999781f, +0.999798f, 0.999813f, 0.999828f, 0.999841f, 0.999853f, +0.999865f, 0.999875f, 0.999885f, 0.999893f, 0.999902f, +0.999909f, 0.999916f, 0.999923f, 0.999929f, 0.999934f, +0.999939f, 0.999944f, 0.999948f, 0.999952f, 0.999956f, +0.999959f, 0.999962f, 0.999965f, 0.999968f, 0.999970f, +0.999973f, 0.999975f, 0.999977f, 0.999978f, 0.999980f, +0.999982f, 0.999983f, 0.999984f, 0.999986f, 0.999987f, +0.999988f, 0.999989f, 0.999990f, 0.999990f, 0.999991f, +0.999992f, 0.999992f, 0.999993f, 0.999994f, 0.999994f, +0.999994f, 0.999995f, 0.999995f, 0.999996f, 0.999996f, +0.999996f, 0.999997f, 0.999997f, 0.999997f, 0.999997f, +0.999997f, 0.999998f, 0.999998f, 0.999998f, 0.999998f, +0.999998f, 0.999998f, 0.999999f, 0.999999f, 0.999999f, +0.999999f, 0.999999f, 0.999999f, 0.999999f, 0.999999f, +0.999999f, 0.999999f, 0.999999f, 0.999999f, 0.999999f, +1.000000f, 1.000000f, 1.000000f, 1.000000f, 1.000000f, +1.000000f, 1.000000f, 1.000000f, 1.000000f, 1.000000f, +1.000000f, +}; |