summaryrefslogtreecommitdiffstats
path: root/lib/rbcodec/codecs
diff options
context:
space:
mode:
authorWilliam Wilgus <me.theuser@yahoo.com>2019-01-04 02:01:18 -0600
committerSolomon Peachy <pizza@shaftnet.org>2021-04-23 14:23:04 +0000
commit14c6bb798d6bebc80f07e863236adbaf8d156a9c (patch)
tree551a4b641906c2626af844fa3239c1b2b1ff0ad3 /lib/rbcodec/codecs
parent75d93937965ec4df70d37df6d4feea04577c996b (diff)
downloadrockbox-14c6bb798d6bebc80f07e863236adbaf8d156a9c.tar.gz
rockbox-14c6bb798d6bebc80f07e863236adbaf8d156a9c.tar.bz2
rockbox-14c6bb798d6bebc80f07e863236adbaf8d156a9c.zip
Sync opus codec to upstream git
Change-Id: I0cfcc0005c4ad7bfbb1aaf454188ce70fb043dc1
Diffstat (limited to 'lib/rbcodec/codecs')
-rw-r--r--lib/rbcodec/codecs/libopus/README.rockbox16
-rw-r--r--lib/rbcodec/codecs/libopus/SOURCES2
-rw-r--r--lib/rbcodec/codecs/libopus/analysis.c945
-rw-r--r--lib/rbcodec/codecs/libopus/analysis.h102
-rw-r--r--lib/rbcodec/codecs/libopus/celt/_kiss_fft_guts.h21
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arch.h66
-rwxr-xr-xlib/rbcodec/codecs/libopus/celt/arm/arm2gnu.pl353
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arm/arm_celt_map.c160
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arm/armcpu.c185
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arm/armcpu.h77
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arm/armopts.s.in37
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arm/celt_fft_ne10.c173
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arm/celt_mdct_ne10.c258
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arm/celt_neon_intr.c211
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arm/celt_pitch_xcorr_arm.s551
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arm/fft_arm.h71
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arm/fixed_arm64.h35
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arm/fixed_armv4.h6
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arm/fixed_armv5e.h4
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arm/mdct_arm.h59
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arm/pitch_arm.h160
-rw-r--r--lib/rbcodec/codecs/libopus/celt/arm/pitch_neon_intr.c290
-rw-r--r--lib/rbcodec/codecs/libopus/celt/bands.c351
-rw-r--r--lib/rbcodec/codecs/libopus/celt/bands.h26
-rw-r--r--lib/rbcodec/codecs/libopus/celt/celt.c44
-rw-r--r--lib/rbcodec/codecs/libopus/celt/celt.h44
-rw-r--r--lib/rbcodec/codecs/libopus/celt/celt_decoder.c263
-rw-r--r--lib/rbcodec/codecs/libopus/celt/celt_encoder.c2607
-rw-r--r--lib/rbcodec/codecs/libopus/celt/celt_lpc.c73
-rw-r--r--lib/rbcodec/codecs/libopus/celt/celt_lpc.h18
-rw-r--r--lib/rbcodec/codecs/libopus/celt/cpu_support.h20
-rw-r--r--lib/rbcodec/codecs/libopus/celt/cwrs.c8
-rw-r--r--lib/rbcodec/codecs/libopus/celt/dump_modes/Makefile32
-rw-r--r--lib/rbcodec/codecs/libopus/celt/dump_modes/dump_modes.c353
-rw-r--r--lib/rbcodec/codecs/libopus/celt/dump_modes/dump_modes_arch.h45
-rw-r--r--lib/rbcodec/codecs/libopus/celt/dump_modes/dump_modes_arm_ne10.c152
-rw-r--r--lib/rbcodec/codecs/libopus/celt/entcode.c2
-rw-r--r--lib/rbcodec/codecs/libopus/celt/entcode.h4
-rw-r--r--lib/rbcodec/codecs/libopus/celt/entdec.h2
-rw-r--r--lib/rbcodec/codecs/libopus/celt/entenc.c2
-rw-r--r--lib/rbcodec/codecs/libopus/celt/entenc.h2
-rw-r--r--lib/rbcodec/codecs/libopus/celt/fixed_c5x.h79
-rw-r--r--lib/rbcodec/codecs/libopus/celt/fixed_c6x.h70
-rw-r--r--lib/rbcodec/codecs/libopus/celt/fixed_debug.h791
-rw-r--r--lib/rbcodec/codecs/libopus/celt/fixed_generic.h27
-rw-r--r--lib/rbcodec/codecs/libopus/celt/float_cast.h16
-rw-r--r--lib/rbcodec/codecs/libopus/celt/kiss_fft.c91
-rw-r--r--lib/rbcodec/codecs/libopus/celt/kiss_fft.h67
-rw-r--r--lib/rbcodec/codecs/libopus/celt/mathops.c7
-rw-r--r--lib/rbcodec/codecs/libopus/celt/mathops.h40
-rw-r--r--lib/rbcodec/codecs/libopus/celt/mdct.c38
-rw-r--r--lib/rbcodec/codecs/libopus/celt/mdct.h56
-rw-r--r--lib/rbcodec/codecs/libopus/celt/mips/celt_mipsr1.h151
-rw-r--r--lib/rbcodec/codecs/libopus/celt/mips/fixed_generic_mipsr1.h126
-rw-r--r--lib/rbcodec/codecs/libopus/celt/mips/kiss_fft_mipsr1.h167
-rw-r--r--lib/rbcodec/codecs/libopus/celt/mips/mdct_mipsr1.h288
-rw-r--r--lib/rbcodec/codecs/libopus/celt/mips/pitch_mipsr1.h161
-rw-r--r--lib/rbcodec/codecs/libopus/celt/mips/vq_mipsr1.h122
-rw-r--r--lib/rbcodec/codecs/libopus/celt/modes.c10
-rw-r--r--lib/rbcodec/codecs/libopus/celt/opus_custom_demo.c210
-rw-r--r--lib/rbcodec/codecs/libopus/celt/os_support.h6
-rw-r--r--lib/rbcodec/codecs/libopus/celt/pitch.c138
-rw-r--r--lib/rbcodec/codecs/libopus/celt/pitch.h65
-rw-r--r--lib/rbcodec/codecs/libopus/celt/quant_bands.c13
-rw-r--r--lib/rbcodec/codecs/libopus/celt/rate.c17
-rw-r--r--lib/rbcodec/codecs/libopus/celt/rate.h4
-rw-r--r--lib/rbcodec/codecs/libopus/celt/static_modes_fixed.h65
-rw-r--r--lib/rbcodec/codecs/libopus/celt/static_modes_fixed_arm_ne10.h388
-rw-r--r--lib/rbcodec/codecs/libopus/celt/static_modes_float.h888
-rw-r--r--lib/rbcodec/codecs/libopus/celt/static_modes_float_arm_ne10.h404
-rw-r--r--lib/rbcodec/codecs/libopus/celt/tests/test_unit_cwrs32.c161
-rw-r--r--lib/rbcodec/codecs/libopus/celt/tests/test_unit_dft.c179
-rw-r--r--lib/rbcodec/codecs/libopus/celt/tests/test_unit_entropy.c383
-rw-r--r--lib/rbcodec/codecs/libopus/celt/tests/test_unit_laplace.c93
-rw-r--r--lib/rbcodec/codecs/libopus/celt/tests/test_unit_mathops.c266
-rw-r--r--lib/rbcodec/codecs/libopus/celt/tests/test_unit_mdct.c227
-rw-r--r--lib/rbcodec/codecs/libopus/celt/tests/test_unit_rotation.c86
-rw-r--r--lib/rbcodec/codecs/libopus/celt/tests/test_unit_types.c50
-rw-r--r--lib/rbcodec/codecs/libopus/celt/vq.c130
-rw-r--r--lib/rbcodec/codecs/libopus/celt/vq.h29
-rw-r--r--lib/rbcodec/codecs/libopus/celt/x86/celt_lpc_sse.h66
-rw-r--r--lib/rbcodec/codecs/libopus/celt/x86/celt_lpc_sse4_1.c89
-rw-r--r--lib/rbcodec/codecs/libopus/celt/x86/pitch_sse.c185
-rw-r--r--lib/rbcodec/codecs/libopus/celt/x86/pitch_sse.h192
-rw-r--r--lib/rbcodec/codecs/libopus/celt/x86/pitch_sse2.c95
-rw-r--r--lib/rbcodec/codecs/libopus/celt/x86/pitch_sse4_1.c195
-rw-r--r--lib/rbcodec/codecs/libopus/celt/x86/vq_sse.h50
-rw-r--r--lib/rbcodec/codecs/libopus/celt/x86/vq_sse2.c217
-rw-r--r--lib/rbcodec/codecs/libopus/celt/x86/x86_celt_map.c167
-rw-r--r--lib/rbcodec/codecs/libopus/celt/x86/x86cpu.c157
-rw-r--r--lib/rbcodec/codecs/libopus/celt/x86/x86cpu.h95
-rw-r--r--lib/rbcodec/codecs/libopus/mapping_matrix.c378
-rw-r--r--lib/rbcodec/codecs/libopus/mapping_matrix.h133
-rw-r--r--lib/rbcodec/codecs/libopus/mlp.c144
-rw-r--r--lib/rbcodec/codecs/libopus/mlp.h60
-rw-r--r--lib/rbcodec/codecs/libopus/mlp_data.c672
-rw-r--r--lib/rbcodec/codecs/libopus/ogg/crctable.h278
-rw-r--r--lib/rbcodec/codecs/libopus/ogg/framing.c1294
-rw-r--r--lib/rbcodec/codecs/libopus/ogg/ogg.h1
-rw-r--r--lib/rbcodec/codecs/libopus/opus.c14
-rw-r--r--lib/rbcodec/codecs/libopus/opus.h9
-rw-r--r--lib/rbcodec/codecs/libopus/opus_compare.c382
-rw-r--r--lib/rbcodec/codecs/libopus/opus_decoder.c134
-rw-r--r--lib/rbcodec/codecs/libopus/opus_defines.h90
-rw-r--r--lib/rbcodec/codecs/libopus/opus_demo.c892
-rw-r--r--lib/rbcodec/codecs/libopus/opus_encoder.c2754
-rw-r--r--lib/rbcodec/codecs/libopus/opus_header.c117
-rw-r--r--lib/rbcodec/codecs/libopus/opus_header.h4
-rw-r--r--lib/rbcodec/codecs/libopus/opus_multistream.c92
-rw-r--r--lib/rbcodec/codecs/libopus/opus_multistream.h660
-rw-r--r--lib/rbcodec/codecs/libopus/opus_multistream_decoder.c549
-rw-r--r--lib/rbcodec/codecs/libopus/opus_multistream_encoder.c1328
-rw-r--r--lib/rbcodec/codecs/libopus/opus_private.h96
-rw-r--r--lib/rbcodec/codecs/libopus/opus_projection.h568
-rw-r--r--lib/rbcodec/codecs/libopus/opus_projection_decoder.c258
-rw-r--r--lib/rbcodec/codecs/libopus/opus_projection_encoder.c468
-rw-r--r--lib/rbcodec/codecs/libopus/opus_types.h27
-rw-r--r--lib/rbcodec/codecs/libopus/repacketizer.c349
-rw-r--r--lib/rbcodec/codecs/libopus/repacketizer_demo.c217
-rw-r--r--lib/rbcodec/codecs/libopus/silk/A2NLSF.c267
-rw-r--r--lib/rbcodec/codecs/libopus/silk/API.h6
-rw-r--r--lib/rbcodec/codecs/libopus/silk/CNG.c83
-rw-r--r--lib/rbcodec/codecs/libopus/silk/HP_variable_cutoff.c77
-rw-r--r--lib/rbcodec/codecs/libopus/silk/LPC_analysis_filter.c29
-rw-r--r--lib/rbcodec/codecs/libopus/silk/LPC_fit.c81
-rw-r--r--lib/rbcodec/codecs/libopus/silk/LPC_inv_pred_gain.c99
-rw-r--r--lib/rbcodec/codecs/libopus/silk/LP_variable_cutoff.c135
-rw-r--r--lib/rbcodec/codecs/libopus/silk/MacroCount.h10
-rw-r--r--lib/rbcodec/codecs/libopus/silk/MacroDebug.h3
-rw-r--r--lib/rbcodec/codecs/libopus/silk/NLSF2A.c59
-rw-r--r--lib/rbcodec/codecs/libopus/silk/NLSF_VQ.c76
-rw-r--r--lib/rbcodec/codecs/libopus/silk/NLSF_VQ_weights_laroia.c4
-rw-r--r--lib/rbcodec/codecs/libopus/silk/NLSF_decode.c22
-rw-r--r--lib/rbcodec/codecs/libopus/silk/NLSF_del_dec_quant.c215
-rw-r--r--lib/rbcodec/codecs/libopus/silk/NLSF_encode.c124
-rw-r--r--lib/rbcodec/codecs/libopus/silk/NLSF_stabilize.c2
-rw-r--r--lib/rbcodec/codecs/libopus/silk/NSQ.c437
-rw-r--r--lib/rbcodec/codecs/libopus/silk/NSQ.h101
-rw-r--r--lib/rbcodec/codecs/libopus/silk/NSQ_del_dec.c733
-rw-r--r--lib/rbcodec/codecs/libopus/silk/PLC.c28
-rw-r--r--lib/rbcodec/codecs/libopus/silk/PLC.h3
-rw-r--r--lib/rbcodec/codecs/libopus/silk/SigProc_FIX.h80
-rw-r--r--lib/rbcodec/codecs/libopus/silk/VAD.c360
-rw-r--r--lib/rbcodec/codecs/libopus/silk/VQ_WMat_EC.c131
-rw-r--r--lib/rbcodec/codecs/libopus/silk/ana_filt_bank_1.c74
-rw-r--r--lib/rbcodec/codecs/libopus/silk/arm/LPC_inv_pred_gain_arm.h57
-rw-r--r--lib/rbcodec/codecs/libopus/silk/arm/LPC_inv_pred_gain_neon_intr.c280
-rw-r--r--lib/rbcodec/codecs/libopus/silk/arm/NSQ_del_dec_arm.h100
-rw-r--r--lib/rbcodec/codecs/libopus/silk/arm/NSQ_del_dec_neon_intr.c1124
-rw-r--r--lib/rbcodec/codecs/libopus/silk/arm/NSQ_neon.c112
-rw-r--r--lib/rbcodec/codecs/libopus/silk/arm/NSQ_neon.h114
-rw-r--r--lib/rbcodec/codecs/libopus/silk/arm/arm_silk_map.c123
-rw-r--r--lib/rbcodec/codecs/libopus/silk/arm/biquad_alt_arm.h68
-rw-r--r--lib/rbcodec/codecs/libopus/silk/arm/biquad_alt_neon_intr.c156
-rw-r--r--lib/rbcodec/codecs/libopus/silk/arm/macros_arm64.h39
-rw-r--r--lib/rbcodec/codecs/libopus/silk/arm/macros_armv4.h13
-rw-r--r--lib/rbcodec/codecs/libopus/silk/arm/macros_armv5e.h9
-rw-r--r--lib/rbcodec/codecs/libopus/silk/biquad_alt.c121
-rw-r--r--lib/rbcodec/codecs/libopus/silk/bwexpander.c2
-rw-r--r--lib/rbcodec/codecs/libopus/silk/check_control_input.c106
-rw-r--r--lib/rbcodec/codecs/libopus/silk/code_signs.c2
-rw-r--r--lib/rbcodec/codecs/libopus/silk/control.h8
-rw-r--r--lib/rbcodec/codecs/libopus/silk/control_SNR.c113
-rw-r--r--lib/rbcodec/codecs/libopus/silk/control_audio_bandwidth.c132
-rw-r--r--lib/rbcodec/codecs/libopus/silk/control_codec.c423
-rw-r--r--lib/rbcodec/codecs/libopus/silk/debug.c170
-rw-r--r--lib/rbcodec/codecs/libopus/silk/debug.h266
-rw-r--r--lib/rbcodec/codecs/libopus/silk/dec_API.c11
-rw-r--r--lib/rbcodec/codecs/libopus/silk/decode_core.c13
-rw-r--r--lib/rbcodec/codecs/libopus/silk/decode_frame.c16
-rw-r--r--lib/rbcodec/codecs/libopus/silk/decode_indices.c2
-rw-r--r--lib/rbcodec/codecs/libopus/silk/decode_parameters.c4
-rw-r--r--lib/rbcodec/codecs/libopus/silk/decode_pitch.c4
-rw-r--r--lib/rbcodec/codecs/libopus/silk/decode_pulses.c6
-rw-r--r--lib/rbcodec/codecs/libopus/silk/decoder_set_fs.c8
-rw-r--r--lib/rbcodec/codecs/libopus/silk/define.h19
-rw-r--r--lib/rbcodec/codecs/libopus/silk/enc_API.c576
-rw-r--r--lib/rbcodec/codecs/libopus/silk/encode_indices.c181
-rw-r--r--lib/rbcodec/codecs/libopus/silk/encode_pulses.c206
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/LTP_analysis_filter_FIX.c90
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/LTP_scale_ctrl_FIX.c53
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/apply_sine_window_FIX.c101
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/arm/warped_autocorrelation_FIX_arm.h68
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/arm/warped_autocorrelation_FIX_neon_intr.c260
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/autocorr_FIX.c48
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/burg_modified_FIX.c280
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/corrMatrix_FIX.c150
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/encode_frame_FIX.c448
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/find_LPC_FIX.c151
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/find_LTP_FIX.c99
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/find_pitch_lags_FIX.c143
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/find_pred_coefs_FIX.c145
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/k2a_FIX.c54
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/k2a_Q16_FIX.c54
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/main_FIX.h244
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/mips/noise_shape_analysis_FIX_mipsr1.h336
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/mips/prefilter_FIX_mipsr1.h184
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/mips/warped_autocorrelation_FIX_mipsr1.h166
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/noise_shape_analysis_FIX.c407
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/pitch_analysis_core_FIX.c721
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/process_gains_FIX.c117
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/regularize_correlations_FIX.c47
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/residual_energy16_FIX.c103
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/residual_energy_FIX.c98
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/schur64_FIX.c93
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/schur_FIX.c107
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/structs_FIX.h116
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/vector_ops_FIX.c102
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/warped_autocorrelation_FIX.c90
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/x86/burg_modified_FIX_sse4_1.c377
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/x86/prefilter_FIX_sse.c160
-rw-r--r--lib/rbcodec/codecs/libopus/silk/fixed/x86/vector_ops_FIX_sse4_1.c88
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/LPC_analysis_filter_FLP.c249
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/LPC_inv_pred_gain_FLP.c73
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/LTP_analysis_filter_FLP.c75
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/LTP_scale_ctrl_FLP.c52
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/SigProc_FLP.h197
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/apply_sine_window_FLP.c81
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/autocorrelation_FLP.c52
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/burg_modified_FLP.c186
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/bwexpander_FLP.c49
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/corrMatrix_FLP.c93
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/encode_frame_FLP.c435
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/energy_FLP.c59
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/find_LPC_FLP.c104
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/find_LTP_FLP.c64
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/find_pitch_lags_FLP.c132
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/find_pred_coefs_FLP.c116
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/inner_product_FLP.c59
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/k2a_FLP.c54
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/main_FLP.h286
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/noise_shape_analysis_FLP.c350
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/pitch_analysis_core_FLP.c630
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/process_gains_FLP.c103
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/regularize_correlations_FLP.c48
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/residual_energy_FLP.c117
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/scale_copy_vector_FLP.c57
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/scale_vector_FLP.c56
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/schur_FLP.c70
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/sort_FLP.c83
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/structs_FLP.h112
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/warped_autocorrelation_FLP.c73
-rw-r--r--lib/rbcodec/codecs/libopus/silk/float/wrappers_FLP.c207
-rw-r--r--lib/rbcodec/codecs/libopus/silk/gain_quant.c5
-rw-r--r--lib/rbcodec/codecs/libopus/silk/init_decoder.c1
-rw-r--r--lib/rbcodec/codecs/libopus/silk/init_encoder.c64
-rw-r--r--lib/rbcodec/codecs/libopus/silk/inner_prod_aligned.c47
-rw-r--r--lib/rbcodec/codecs/libopus/silk/interpolate.c51
-rw-r--r--lib/rbcodec/codecs/libopus/silk/lin2log.c46
-rw-r--r--lib/rbcodec/codecs/libopus/silk/log2lin.c6
-rw-r--r--lib/rbcodec/codecs/libopus/silk/macros.h29
-rw-r--r--lib/rbcodec/codecs/libopus/silk/main.h108
-rw-r--r--lib/rbcodec/codecs/libopus/silk/mips/NSQ_del_dec_mipsr1.h410
-rw-r--r--lib/rbcodec/codecs/libopus/silk/mips/macros_mipsr1.h92
-rw-r--r--lib/rbcodec/codecs/libopus/silk/mips/sigproc_fix_mipsr1.h60
-rw-r--r--lib/rbcodec/codecs/libopus/silk/process_NLSFs.c107
-rw-r--r--lib/rbcodec/codecs/libopus/silk/quant_LTP_gains.c132
-rw-r--r--lib/rbcodec/codecs/libopus/silk/resampler.c10
-rw-r--r--lib/rbcodec/codecs/libopus/silk/resampler_down2.c74
-rw-r--r--lib/rbcodec/codecs/libopus/silk/resampler_down2_3.c103
-rw-r--r--lib/rbcodec/codecs/libopus/silk/resampler_private_down_FIR.c2
-rw-r--r--lib/rbcodec/codecs/libopus/silk/resampler_rom.c54
-rw-r--r--lib/rbcodec/codecs/libopus/silk/shell_coder.c4
-rw-r--r--lib/rbcodec/codecs/libopus/silk/sigm_Q15.c76
-rw-r--r--lib/rbcodec/codecs/libopus/silk/sort.c18
-rw-r--r--lib/rbcodec/codecs/libopus/silk/stereo_LR_to_MS.c229
-rw-r--r--lib/rbcodec/codecs/libopus/silk/stereo_encode_pred.c62
-rw-r--r--lib/rbcodec/codecs/libopus/silk/stereo_find_predictor.c79
-rw-r--r--lib/rbcodec/codecs/libopus/silk/stereo_quant_pred.c73
-rw-r--r--lib/rbcodec/codecs/libopus/silk/structs.h8
-rw-r--r--lib/rbcodec/codecs/libopus/silk/sum_sqr_shift.c51
-rw-r--r--lib/rbcodec/codecs/libopus/silk/tables.h14
-rw-r--r--lib/rbcodec/codecs/libopus/silk/tables_LTP.c8
-rw-r--r--lib/rbcodec/codecs/libopus/silk/tables_NLSF_CB_NB_MB.c36
-rw-r--r--lib/rbcodec/codecs/libopus/silk/tables_NLSF_CB_WB.c36
-rw-r--r--lib/rbcodec/codecs/libopus/silk/tables_other.c14
-rw-r--r--lib/rbcodec/codecs/libopus/silk/tests/test_unit_LPC_inv_pred_gain.c129
-rw-r--r--lib/rbcodec/codecs/libopus/silk/tuning_parameters.h155
-rw-r--r--lib/rbcodec/codecs/libopus/silk/x86/NSQ_del_dec_sse4_1.c859
-rw-r--r--lib/rbcodec/codecs/libopus/silk/x86/NSQ_sse4_1.c719
-rw-r--r--lib/rbcodec/codecs/libopus/silk/x86/SigProc_FIX_sse.h94
-rw-r--r--lib/rbcodec/codecs/libopus/silk/x86/VAD_sse4_1.c277
-rw-r--r--lib/rbcodec/codecs/libopus/silk/x86/VQ_WMat_EC_sse4_1.c142
-rw-r--r--lib/rbcodec/codecs/libopus/silk/x86/main_sse.h248
-rw-r--r--lib/rbcodec/codecs/libopus/silk/x86/x86_silk_map.c164
-rw-r--r--lib/rbcodec/codecs/libopus/tansig_table.h45
286 files changed, 48927 insertions, 1274 deletions
diff --git a/lib/rbcodec/codecs/libopus/README.rockbox b/lib/rbcodec/codecs/libopus/README.rockbox
index 370a5823f2..dd0ce70795 100644
--- a/lib/rbcodec/codecs/libopus/README.rockbox
+++ b/lib/rbcodec/codecs/libopus/README.rockbox
@@ -1,5 +1,5 @@
Libraries: Opus (snapshot) / Opus-tools (snapshot) / libogg 1.3
-Imported: September 15th, 2012
+Imported: January 1st, 2019
Steps taken to adapt original opus/opus-tool/ogg source files to rockbox
(useful when for example syncing a new snapshot)
@@ -10,13 +10,21 @@ Opus:
* copied .c/.h files from opus/celt to lib/rbcodec/codecs/libopus/celt
* copied .c/.h files from opus/silk to lib/rbcodec/codecs/libopus/silk
+Celt:
+* changed #if FIXED_POINT to #ifdef FIXED_POINT in bands.c
+* changed #elif OPUS_ARM_INLINE_EDSP to #elif defined (OPUS_ARM_INLINE_EDSP)
+* add #define ABS(a)(((a) < 0) ? - (a) :(a)) to mathops.h
+
Opus-tools:
* copied src/opus_header.h and src/opus_header.c to lib/rbcodec/codecs/libopus
-* changed #include <ogg/ogg.h> to #include "ogg/ogg.h" in opus_header.c
+* changed #include <ogg/ogg.h> to #include "ogg/ogg.h" in opus_header.h
Ogg:
* copied libogg/src/framing.c to lib/rbcodec/codecs/libopus/ogg
* copied libogg/include/ogg.h to lib/rbcodec/codecs/libopus/ogg
+* copied libogg/include/crctable.h to lib/rbcodec/codecs/libopus/ogg
* changed #include "ogg/ogg.h" to #include "ogg.h" in framing.c
-* added os_config.h to lib/rbcodec/codecs/libopus/ogg
-
+* -- added os_config.h to lib/rbcodec/codecs/libopus/ogg
+* added our own os_types.h
+* changed #include <ogg/os_types.h> to #include "os_types.h" in ogg.h
+* changed #include <ogg/os_types.h> to #include "os_types.h" in crctable.h
diff --git a/lib/rbcodec/codecs/libopus/SOURCES b/lib/rbcodec/codecs/libopus/SOURCES
index 417cf01d88..53806545c7 100644
--- a/lib/rbcodec/codecs/libopus/SOURCES
+++ b/lib/rbcodec/codecs/libopus/SOURCES
@@ -32,8 +32,10 @@ silk/decode_pulses.c
silk/decoder_set_fs.c
silk/gain_quant.c
silk/init_decoder.c
+silk/lin2log.c
silk/log2lin.c
silk/LPC_analysis_filter.c
+silk/LPC_fit.c
silk/LPC_inv_pred_gain.c
silk/NLSF2A.c
silk/NLSF_decode.c
diff --git a/lib/rbcodec/codecs/libopus/analysis.c b/lib/rbcodec/codecs/libopus/analysis.c
new file mode 100644
index 0000000000..b192ae4e8d
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/analysis.c
@@ -0,0 +1,945 @@
+/* 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
+
+#define ANALYSIS_C
+
+#include <stdio.h>
+
+#include "mathops.h"
+#include "kiss_fft.h"
+#include "celt.h"
+#include "modes.h"
+#include "arch.h"
+#include "quant_bands.h"
+#include "analysis.h"
+#include "mlp.h"
+#include "stack_alloc.h"
+#include "float_cast.h"
+
+#ifndef M_PI
+#define M_PI 3.141592653
+#endif
+
+#ifndef DISABLE_FLOAT_API
+
+#define TRANSITION_PENALTY 10
+
+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] = {
+ 4, 8, 12, 16, 20, 24, 28, 32, 40, 48, 56, 64, 80, 96, 112, 136, 160, 192, 240
+};
+
+#define NB_TONAL_SKIP_BANDS 9
+
+static opus_val32 silk_resampler_down2_hp(
+ opus_val32 *S, /* I/O State vector [ 2 ] */
+ opus_val32 *out, /* O Output signal [ floor(len/2) ] */
+ const opus_val32 *in, /* I Input signal [ len ] */
+ int inLen /* I Number of input samples */
+)
+{
+ int k, len2 = inLen/2;
+ opus_val32 in32, out32, out32_hp, Y, X;
+ opus_val64 hp_ener = 0;
+ /* Internal variables and state are in Q10 format */
+ for( k = 0; k < len2; k++ ) {
+ /* Convert to Q10 */
+ in32 = in[ 2 * k ];
+
+ /* All-pass section for even input sample */
+ Y = SUB32( in32, S[ 0 ] );
+ X = MULT16_32_Q15(QCONST16(0.6074371f, 15), Y);
+ out32 = ADD32( S[ 0 ], X );
+ S[ 0 ] = ADD32( in32, X );
+ out32_hp = out32;
+ /* Convert to Q10 */
+ in32 = in[ 2 * k + 1 ];
+
+ /* All-pass section for odd input sample, and add to output of previous section */
+ Y = SUB32( in32, S[ 1 ] );
+ X = MULT16_32_Q15(QCONST16(0.15063f, 15), Y);
+ out32 = ADD32( out32, S[ 1 ] );
+ out32 = ADD32( out32, X );
+ S[ 1 ] = ADD32( in32, X );
+
+ Y = SUB32( -in32, S[ 2 ] );
+ X = MULT16_32_Q15(QCONST16(0.15063f, 15), Y);
+ out32_hp = ADD32( out32_hp, S[ 2 ] );
+ out32_hp = ADD32( out32_hp, X );
+ S[ 2 ] = ADD32( -in32, X );
+
+ hp_ener += out32_hp*(opus_val64)out32_hp;
+ /* Add, convert back to int16 and store to output */
+ out[ k ] = HALF32(out32);
+ }
+#ifdef FIXED_POINT
+ /* len2 can be up to 480, so we shift by 8 more to make it fit. */
+ hp_ener = hp_ener >> (2*SIG_SHIFT + 8);
+#endif
+ return (opus_val32)hp_ener;
+}
+
+static opus_val32 downmix_and_resample(downmix_func downmix, const void *_x, opus_val32 *y, opus_val32 S[3], int subframe, int offset, int c1, int c2, int C, int Fs)
+{
+ VARDECL(opus_val32, tmp);
+ opus_val32 scale;
+ int j;
+ opus_val32 ret = 0;
+ SAVE_STACK;
+
+ if (subframe==0) return 0;
+ if (Fs == 48000)
+ {
+ subframe *= 2;
+ offset *= 2;
+ } else if (Fs == 16000) {
+ subframe = subframe*2/3;
+ offset = offset*2/3;
+ }
+ ALLOC(tmp, subframe, opus_val32);
+
+ downmix(_x, tmp, subframe, offset, c1, c2, C);
+#ifdef FIXED_POINT
+ scale = (1<<SIG_SHIFT);
+#else
+ scale = 1.f/32768;
+#endif
+ if (c2==-2)
+ scale /= C;
+ else if (c2>-1)
+ scale /= 2;
+ for (j=0;j<subframe;j++)
+ tmp[j] *= scale;
+ if (Fs == 48000)
+ {
+ ret = silk_resampler_down2_hp(S, y, tmp, subframe);
+ } else if (Fs == 24000) {
+ OPUS_COPY(y, tmp, subframe);
+ } else if (Fs == 16000) {
+ VARDECL(opus_val32, tmp3x);
+ ALLOC(tmp3x, 3*subframe, opus_val32);
+ /* Don't do this at home! This resampler is horrible and it's only (barely)
+ usable for the purpose of the analysis because we don't care about all
+ the aliasing between 8 kHz and 12 kHz. */
+ for (j=0;j<subframe;j++)
+ {
+ tmp3x[3*j] = tmp[j];
+ tmp3x[3*j+1] = tmp[j];
+ tmp3x[3*j+2] = tmp[j];
+ }
+ silk_resampler_down2_hp(S, y, tmp3x, 3*subframe);
+ }
+ RESTORE_STACK;
+ return ret;
+}
+
+void tonality_analysis_init(TonalityAnalysisState *tonal, opus_int32 Fs)
+{
+ /* Initialize reusable fields. */
+ tonal->arch = opus_select_arch();
+ tonal->Fs = Fs;
+ /* Clear remaining fields. */
+ tonality_analysis_reset(tonal);
+}
+
+void tonality_analysis_reset(TonalityAnalysisState *tonal)
+{
+ /* Clear non-reusable fields. */
+ char *start = (char*)&tonal->TONALITY_ANALYSIS_RESET_START;
+ OPUS_CLEAR(start, sizeof(TonalityAnalysisState) - (start - (char*)tonal));
+}
+
+void tonality_get_info(TonalityAnalysisState *tonal, AnalysisInfo *info_out, int len)
+{
+ int pos;
+ int curr_lookahead;
+ float tonality_max;
+ float tonality_avg;
+ int tonality_count;
+ int i;
+ int pos0;
+ float prob_avg;
+ float prob_count;
+ float prob_min, prob_max;
+ float vad_prob;
+ int mpos, vpos;
+ int bandwidth_span;
+
+ pos = tonal->read_pos;
+ curr_lookahead = tonal->write_pos-tonal->read_pos;
+ if (curr_lookahead<0)
+ curr_lookahead += DETECT_SIZE;
+
+ /* On long frames, look at the second analysis window rather than the first. */
+ if (len > tonal->Fs/50 && pos != tonal->write_pos)
+ {
+ pos++;
+ if (pos==DETECT_SIZE)
+ pos=0;
+ }
+ if (pos == tonal->write_pos)
+ pos--;
+ if (pos<0)
+ pos = DETECT_SIZE-1;
+ pos0 = pos;
+ OPUS_COPY(info_out, &tonal->info[pos], 1);
+ tonality_max = tonality_avg = info_out->tonality;
+ tonality_count = 1;
+ /* Look at the neighbouring frames and pick largest bandwidth found (to be safe). */
+ bandwidth_span = 6;
+ /* If possible, look ahead for a tone to compensate for the delay in the tone detector. */
+ for (i=0;i<3;i++)
+ {
+ pos++;
+ if (pos==DETECT_SIZE)
+ pos = 0;
+ if (pos == tonal->write_pos)
+ break;
+ tonality_max = MAX32(tonality_max, tonal->info[pos].tonality);
+ tonality_avg += tonal->info[pos].tonality;
+ tonality_count++;
+ info_out->bandwidth = IMAX(info_out->bandwidth, tonal->info[pos].bandwidth);
+ bandwidth_span--;
+ }
+ pos = pos0;
+ /* Look back in time to see if any has a wider bandwidth than the current frame. */
+ for (i=0;i<bandwidth_span;i++)
+ {
+ pos--;
+ if (pos < 0)
+ pos = DETECT_SIZE-1;
+ if (pos == tonal->write_pos)
+ break;
+ info_out->bandwidth = IMAX(info_out->bandwidth, tonal->info[pos].bandwidth);
+ }
+ info_out->tonality = MAX32(tonality_avg/tonality_count, tonality_max-.2f);
+
+ mpos = vpos = pos0;
+ /* If we have enough look-ahead, compensate for the ~5-frame delay in the music prob and
+ ~1 frame delay in the VAD prob. */
+ if (curr_lookahead > 15)
+ {
+ mpos += 5;
+ if (mpos>=DETECT_SIZE)
+ mpos -= DETECT_SIZE;
+ vpos += 1;
+ if (vpos>=DETECT_SIZE)
+ vpos -= DETECT_SIZE;
+ }
+
+ /* The following calculations attempt to minimize a "badness function"
+ for the transition. When switching from speech to music, the badness
+ of switching at frame k is
+ b_k = S*v_k + \sum_{i=0}^{k-1} v_i*(p_i - T)
+ where
+ v_i is the activity probability (VAD) at frame i,
+ p_i is the music probability at frame i
+ T is the probability threshold for switching
+ S is the penalty for switching during active audio rather than silence
+ the current frame has index i=0
+
+ Rather than apply badness to directly decide when to switch, what we compute
+ instead is the threshold for which the optimal switching point is now. When
+ considering whether to switch now (frame 0) or at frame k, we have:
+ S*v_0 = S*v_k + \sum_{i=0}^{k-1} v_i*(p_i - T)
+ which gives us:
+ T = ( \sum_{i=0}^{k-1} v_i*p_i + S*(v_k-v_0) ) / ( \sum_{i=0}^{k-1} v_i )
+ We take the min threshold across all positive values of k (up to the maximum
+ amount of lookahead we have) to give us the threshold for which the current
+ frame is the optimal switch point.
+
+ The last step is that we need to consider whether we want to switch at all.
+ For that we use the average of the music probability over the entire window.
+ If the threshold is higher than that average we're not going to
+ switch, so we compute a min with the average as well. The result of all these
+ min operations is music_prob_min, which gives the threshold for switching to music
+ if we're currently encoding for speech.
+
+ We do the exact opposite to compute music_prob_max which is used for switching
+ from music to speech.
+ */
+ prob_min = 1.f;
+ prob_max = 0.f;
+ vad_prob = tonal->info[vpos].activity_probability;
+ prob_count = MAX16(.1f, vad_prob);
+ prob_avg = MAX16(.1f, vad_prob)*tonal->info[mpos].music_prob;
+ while (1)
+ {
+ float pos_vad;
+ mpos++;
+ if (mpos==DETECT_SIZE)
+ mpos = 0;
+ if (mpos == tonal->write_pos)
+ break;
+ vpos++;
+ if (vpos==DETECT_SIZE)
+ vpos = 0;
+ if (vpos == tonal->write_pos)
+ break;
+ pos_vad = tonal->info[vpos].activity_probability;
+ prob_min = MIN16((prob_avg - TRANSITION_PENALTY*(vad_prob - pos_vad))/prob_count, prob_min);
+ prob_max = MAX16((prob_avg + TRANSITION_PENALTY*(vad_prob - pos_vad))/prob_count, prob_max);
+ prob_count += MAX16(.1f, pos_vad);
+ prob_avg += MAX16(.1f, pos_vad)*tonal->info[mpos].music_prob;
+ }
+ info_out->music_prob = prob_avg/prob_count;
+ prob_min = MIN16(prob_avg/prob_count, prob_min);
+ prob_max = MAX16(prob_avg/prob_count, prob_max);
+ prob_min = MAX16(prob_min, 0.f);
+ prob_max = MIN16(prob_max, 1.f);
+
+ /* If we don't have enough look-ahead, do our best to make a decent decision. */
+ if (curr_lookahead < 10)
+ {
+ float pmin, pmax;
+ pmin = prob_min;
+ pmax = prob_max;
+ pos = pos0;
+ /* Look for min/max in the past. */
+ for (i=0;i<IMIN(tonal->count-1, 15);i++)
+ {
+ pos--;
+ if (pos < 0)
+ pos = DETECT_SIZE-1;
+ pmin = MIN16(pmin, tonal->info[pos].music_prob);
+ pmax = MAX16(pmax, tonal->info[pos].music_prob);
+ }
+ /* Bias against switching on active audio. */
+ pmin = MAX16(0.f, pmin - .1f*vad_prob);
+ pmax = MIN16(1.f, pmax + .1f*vad_prob);
+ prob_min += (1.f-.1f*curr_lookahead)*(pmin - prob_min);
+ prob_max += (1.f-.1f*curr_lookahead)*(pmax - prob_max);
+ }
+ info_out->music_prob_min = prob_min;
+ info_out->music_prob_max = prob_max;
+
+ /* printf("%f %f %f %f %f\n", prob_min, prob_max, prob_avg/prob_count, vad_prob, info_out->music_prob); */
+ tonal->read_subframe += len/(tonal->Fs/400);
+ while (tonal->read_subframe>=8)
+ {
+ tonal->read_subframe -= 8;
+ tonal->read_pos++;
+ }
+ if (tonal->read_pos>=DETECT_SIZE)
+ tonal->read_pos-=DETECT_SIZE;
+}
+
+static const float std_feature_bias[9] = {
+ 5.684947f, 3.475288f, 1.770634f, 1.599784f, 3.773215f,
+ 2.163313f, 1.260756f, 1.116868f, 1.918795f
+};
+
+#define LEAKAGE_OFFSET 2.5f
+#define LEAKAGE_SLOPE 2.f
+
+#ifdef FIXED_POINT
+/* For fixed-point, the input is +/-2^15 shifted up by SIG_SHIFT, so we need to
+ compensate for that in the energy. */
+#define SCALE_COMPENS (1.f/((opus_int32)1<<(15+SIG_SHIFT)))
+#define SCALE_ENER(e) ((SCALE_COMPENS*SCALE_COMPENS)*(e))
+#else
+#define SCALE_ENER(e) (e)
+#endif
+
+static void tonality_analysis(TonalityAnalysisState *tonal, 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 is_masked[NB_TBANDS+1];
+ int bandwidth=0;
+ float maxE = 0;
+ float noise_floor;
+ int remaining;
+ AnalysisInfo *info;
+ float hp_ener;
+ float tonality2[240];
+ float midE[8];
+ float spec_variability=0;
+ float band_log2[NB_TBANDS+1];
+ float leakage_from[NB_TBANDS+1];
+ float leakage_to[NB_TBANDS+1];
+ float layer_out[MAX_NEURONS];
+ float below_max_pitch;
+ float above_max_pitch;
+ SAVE_STACK;
+
+ alpha = 1.f/IMIN(10, 1+tonal->count);
+ alphaE = 1.f/IMIN(25, 1+tonal->count);
+ /* Noise floor related decay for bandwidth detection: -2.2 dB/second */
+ alphaE2 = 1.f/IMIN(100, 1+tonal->count);
+ if (tonal->count <= 1) alphaE2 = 1;
+
+ if (tonal->Fs == 48000)
+ {
+ /* len and offset are now at 24 kHz. */
+ len/= 2;
+ offset /= 2;
+ } else if (tonal->Fs == 16000) {
+ len = 3*len/2;
+ offset = 3*offset/2;
+ }
+
+ kfft = celt_mode->mdct.kfft[0];
+ if (tonal->count==0)
+ tonal->mem_fill = 240;
+ tonal->hp_ener_accum += (float)downmix_and_resample(downmix, x,
+ &tonal->inmem[tonal->mem_fill], tonal->downmix_state,
+ IMIN(len, ANALYSIS_BUF_SIZE-tonal->mem_fill), offset, c1, c2, C, tonal->Fs);
+ if (tonal->mem_fill+len < ANALYSIS_BUF_SIZE)
+ {
+ tonal->mem_fill += len;
+ /* Don't have enough to update the analysis */
+ RESTORE_STACK;
+ return;
+ }
+ hp_ener = tonal->hp_ener_accum;
+ 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);
+ tonal->hp_ener_accum = (float)downmix_and_resample(downmix, x,
+ &tonal->inmem[240], tonal->downmix_state, remaining,
+ offset+ANALYSIS_BUF_SIZE-tonal->mem_fill, c1, c2, C, tonal->Fs);
+ tonal->mem_fill = 240 + remaining;
+ opus_fft(kfft, in, out, tonal->arch);
+#ifndef FIXED_POINT
+ /* If there's any NaN on the input, the entire output will be NaN, so we only need to check one value. */
+ if (celt_isnan(out[0].r))
+ {
+ info->valid = 0;
+ RESTORE_STACK;
+ return;
+ }
+#endif
+
+ 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)float2int(d2_angle);
+ noisiness[i] = ABS16(mod1);
+ mod1 *= mod1;
+ mod1 *= mod1;
+
+ mod2 = d2_angle2 - (float)float2int(d2_angle2);
+ noisiness[i] += ABS16(mod2);
+ mod2 *= mod2;
+ mod2 *= mod2;
+
+ avg_mod = .25f*(d2A[i]+mod1+2*mod2);
+ /* This introduces an extra delay of 2 frames in the detection. */
+ tonality[i] = 1.f/(1.f+40.f*16.f*pi4*avg_mod)-.015f;
+ /* No delay on this detection, but it's less reliable. */
+ tonality2[i] = 1.f/(1.f+40.f*16.f*pi4*mod2)-.015f;
+
+ A[i] = angle2;
+ dA[i] = d_angle2;
+ d2A[i] = mod2;
+ }
+ for (i=2;i<N2-1;i++)
+ {
+ float tt = MIN32(tonality2[i], MAX32(tonality2[i-1], tonality2[i+1]));
+ tonality[i] = .9f*MAX32(tonality[i], tt-.1f);
+ }
+ 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;
+ /* The energy of the very first band is special because of DC. */
+ {
+ float E = 0;
+ float X1r, X2r;
+ X1r = 2*(float)out[0].r;
+ X2r = 2*(float)out[0].i;
+ E = X1r*X1r + X2r*X2r;
+ for (i=1;i<4;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;
+ }
+ E = SCALE_ENER(E);
+ band_log2[0] = .5f*1.442695f*(float)log(E+1e-10f);
+ }
+ 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;
+ binE = SCALE_ENER(binE);
+ E += binE;
+ tE += binE*MAX32(0, tonality[i]);
+ nE += binE*2.f*(.5f-noisiness[i]);
+ }
+#ifndef FIXED_POINT
+ /* Check for extreme band energies that could cause NaNs later. */
+ if (!(E<1e9f) || celt_isnan(E))
+ {
+ info->valid = 0;
+ RESTORE_STACK;
+ return;
+ }
+#endif
+
+ 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);
+ band_log2[b+1] = .5f*1.442695f*(float)log(E+1e-10f);
+ tonal->logE[tonal->E_count][b] = logE[b];
+ if (tonal->count==0)
+ tonal->highE[b] = tonal->lowE[b] = logE[b];
+ if (tonal->highE[b] > tonal->lowE[b] + 7.5)
+ {
+ if (tonal->highE[b] - logE[b] > logE[b] - tonal->lowE[b])
+ tonal->highE[b] -= .01f;
+ else
+ tonal->lowE[b] += .01f;
+ }
+ if (logE[b] > tonal->highE[b])
+ {
+ tonal->highE[b] = logE[b];
+ tonal->lowE[b] = MAX32(tonal->highE[b]-15, tonal->lowE[b]);
+ } else if (logE[b] < tonal->lowE[b])
+ {
+ tonal->lowE[b] = logE[b];
+ tonal->highE[b] = MIN32(tonal->lowE[b]+15, tonal->highE[b]);
+ }
+ 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];
+ }
+
+ leakage_from[0] = band_log2[0];
+ leakage_to[0] = band_log2[0] - LEAKAGE_OFFSET;
+ for (b=1;b<NB_TBANDS+1;b++)
+ {
+ float leak_slope = LEAKAGE_SLOPE*(tbands[b]-tbands[b-1])/4;
+ leakage_from[b] = MIN16(leakage_from[b-1]+leak_slope, band_log2[b]);
+ leakage_to[b] = MAX16(leakage_to[b-1]-leak_slope, band_log2[b]-LEAKAGE_OFFSET);
+ }
+ for (b=NB_TBANDS-2;b>=0;b--)
+ {
+ float leak_slope = LEAKAGE_SLOPE*(tbands[b+1]-tbands[b])/4;
+ leakage_from[b] = MIN16(leakage_from[b+1]+leak_slope, leakage_from[b]);
+ leakage_to[b] = MAX16(leakage_to[b+1]-leak_slope, leakage_to[b]);
+ }
+ celt_assert(NB_TBANDS+1 <= LEAK_BANDS);
+ for (b=0;b<NB_TBANDS+1;b++)
+ {
+ /* leak_boost[] is made up of two terms. The first, based on leakage_to[],
+ represents the boost needed to overcome the amount of analysis leakage
+ cause in a weaker band b by louder neighbouring bands.
+ The second, based on leakage_from[], applies to a loud band b for
+ which the quantization noise causes synthesis leakage to the weaker
+ neighbouring bands. */
+ float boost = MAX16(0, leakage_to[b] - band_log2[b]) +
+ MAX16(0, band_log2[b] - (leakage_from[b]+LEAKAGE_OFFSET));
+ info->leak_boost[b] = IMIN(255, (int)floor(.5 + 64.f*boost));
+ }
+ for (;b<LEAK_BANDS;b++) info->leak_boost[b] = 0;
+
+ for (i=0;i<NB_FRAMES;i++)
+ {
+ int j;
+ float mindist = 1e15f;
+ for (j=0;j<NB_FRAMES;j++)
+ {
+ int k;
+ float dist=0;
+ for (k=0;k<NB_TBANDS;k++)
+ {
+ float tmp;
+ tmp = tonal->logE[i][k] - tonal->logE[j][k];
+ dist += tmp*tmp;
+ }
+ if (j!=i)
+ mindist = MIN32(mindist, dist);
+ }
+ spec_variability += mindist;
+ }
+ spec_variability = (float)sqrt(spec_variability/NB_FRAMES/NB_TBANDS);
+ bandwidth_mask = 0;
+ bandwidth = 0;
+ maxE = 0;
+ noise_floor = 5.7e-4f/(1<<(IMAX(0,lsb_depth-8)));
+ noise_floor *= noise_floor;
+ below_max_pitch=0;
+ above_max_pitch=0;
+ for (b=0;b<NB_TBANDS;b++)
+ {
+ float E=0;
+ float Em;
+ int band_start, band_end;
+ /* Keep a margin of 300 Hz for aliasing */
+ band_start = tbands[b];
+ band_end = tbands[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;
+ }
+ E = SCALE_ENER(E);
+ maxE = MAX32(maxE, E);
+ if (band_start < 64)
+ {
+ below_max_pitch += E;
+ } else {
+ above_max_pitch += E;
+ }
+ tonal->meanE[b] = MAX32((1-alphaE2)*tonal->meanE[b], E);
+ Em = MAX32(E, tonal->meanE[b]);
+ /* Consider the band "active" only if all these conditions are met:
+ 1) less than 90 dB below the peak band (maximal masking possible considering
+ both the ATH and the loudness-dependent slope of the spreading function)
+ 2) above the PCM quantization noise floor
+ We use b+1 because the first CELT band isn't included in tbands[]
+ */
+ if (E*1e9f > maxE && (Em > 3*noise_floor*(band_end-band_start) || E > noise_floor*(band_end-band_start)))
+ bandwidth = b+1;
+ /* Check if the band is masked (see below). */
+ is_masked[b] = E < (tonal->prev_bandwidth >= b+1 ? .01f : .05f)*bandwidth_mask;
+ /* Use a simple follower with 13 dB/Bark slope for spreading function. */
+ bandwidth_mask = MAX32(.05f*bandwidth_mask, E);
+ }
+ /* Special case for the last two bands, for which we don't have spectrum but only
+ the energy above 12 kHz. The difficulty here is that the high-pass we use
+ leaks some LF energy, so we need to increase the threshold without accidentally cutting
+ off the band. */
+ if (tonal->Fs == 48000) {
+ float noise_ratio;
+ float Em;
+ float E = hp_ener*(1.f/(60*60));
+ noise_ratio = tonal->prev_bandwidth==20 ? 10.f : 30.f;
+
+#ifdef FIXED_POINT
+ /* silk_resampler_down2_hp() shifted right by an extra 8 bits. */
+ E *= 256.f*(1.f/Q15ONE)*(1.f/Q15ONE);
+#endif
+ above_max_pitch += E;
+ tonal->meanE[b] = MAX32((1-alphaE2)*tonal->meanE[b], E);
+ Em = MAX32(E, tonal->meanE[b]);
+ if (Em > 3*noise_ratio*noise_floor*160 || E > noise_ratio*noise_floor*160)
+ bandwidth = 20;
+ /* Check if the band is masked (see below). */
+ is_masked[b] = E < (tonal->prev_bandwidth == 20 ? .01f : .05f)*bandwidth_mask;
+ }
+ if (above_max_pitch > below_max_pitch)
+ info->max_pitch_ratio = below_max_pitch/above_max_pitch;
+ else
+ info->max_pitch_ratio = 1;
+ /* In some cases, resampling aliasing can create a small amount of energy in the first band
+ being cut. So if the last band is masked, we don't include it. */
+ if (bandwidth == 20 && is_masked[NB_TBANDS])
+ bandwidth-=2;
+ else if (bandwidth > 0 && bandwidth <= NB_TBANDS && is_masked[bandwidth-1])
+ bandwidth--;
+ if (tonal->count<=2)
+ bandwidth = 20;
+ frame_loudness = 20*(float)log10(frame_loudness);
+ tonal->Etracker = MAX32(tonal->Etracker-.003f, 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;
+ }
+ for (i=0;i<8;i++)
+ {
+ float sum=0;
+ for (b=0;b<16;b++)
+ sum += dct_table[i*16+b]*.5f*(tonal->highE[b]+tonal->lowE[b]);
+ midE[i] = sum;
+ }
+
+ frame_stationarity /= NB_TBANDS;
+ relativeE /= NB_TBANDS;
+ if (tonal->count<10)
+ relativeE = .5f;
+ 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 = IMIN(tonal->count+1, ANALYSIS_COUNT_MAX);
+ 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<4;i++)
+ features[i] = BFCC[i]-midE[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]) - std_feature_bias[i];
+ features[18] = spec_variability - 0.78f;
+ features[20] = info->tonality - 0.154723f;
+ features[21] = info->activity - 0.724643f;
+ features[22] = frame_stationarity - 0.743717f;
+ features[23] = info->tonality_slope + 0.069216f;
+ features[24] = tonal->lowECount - 0.067930f;
+
+ compute_dense(&layer0, layer_out, features);
+ compute_gru(&layer1, tonal->rnn_state, layer_out);
+ compute_dense(&layer2, frame_probs, tonal->rnn_state);
+
+ /* Probability of speech or music vs noise */
+ info->activity_probability = frame_probs[1];
+ info->music_prob = frame_probs[0];
+
+ /*printf("%f %f %f\n", frame_probs[0], frame_probs[1], info->music_prob);*/
+#ifdef MLP_TRAINING
+ for (i=0;i<25;i++)
+ printf("%f ", features[i]);
+ printf("\n");
+#endif
+
+ info->bandwidth = bandwidth;
+ tonal->prev_bandwidth = bandwidth;
+ /*printf("%d %d\n", info->bandwidth, info->opus_bandwidth);*/
+ info->noisiness = frame_noisiness;
+ info->valid = 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;
+
+ analysis_frame_size -= analysis_frame_size&1;
+ if (analysis_pcm != NULL)
+ {
+ /* Avoid overflow/wrap-around of the analysis buffer */
+ analysis_frame_size = IMIN((DETECT_SIZE-5)*Fs/50, analysis_frame_size);
+
+ pcm_len = analysis_frame_size - analysis->analysis_offset;
+ offset = analysis->analysis_offset;
+ while (pcm_len>0) {
+ tonality_analysis(analysis, celt_mode, analysis_pcm, IMIN(Fs/50, pcm_len), offset, c1, c2, C, lsb_depth, downmix);
+ offset += Fs/50;
+ pcm_len -= Fs/50;
+ }
+ analysis->analysis_offset = analysis_frame_size;
+
+ analysis->analysis_offset -= frame_size;
+ }
+
+ analysis_info->valid = 0;
+ tonality_get_info(analysis, analysis_info, frame_size);
+}
+
+#endif /* DISABLE_FLOAT_API */
diff --git a/lib/rbcodec/codecs/libopus/analysis.h b/lib/rbcodec/codecs/libopus/analysis.h
new file mode 100644
index 0000000000..289c845e82
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/analysis.h
@@ -0,0 +1,102 @@
+/* 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"
+#include "mlp.h"
+
+#define NB_FRAMES 8
+#define NB_TBANDS 18
+#define ANALYSIS_BUF_SIZE 720 /* 30 ms at 24 kHz */
+
+/* At that point we can stop counting frames because it no longer matters. */
+#define ANALYSIS_COUNT_MAX 10000
+
+#define DETECT_SIZE 100
+
+/* Uncomment this to print the MLP features on stdout. */
+/*#define MLP_TRAINING*/
+
+typedef struct {
+ int arch;
+ int application;
+ opus_int32 Fs;
+#define TONALITY_ANALYSIS_RESET_START angle
+ 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;
+ int prev_bandwidth;
+ float E[NB_FRAMES][NB_TBANDS];
+ float logE[NB_FRAMES][NB_TBANDS];
+ float lowE[NB_TBANDS];
+ float highE[NB_TBANDS];
+ float meanE[NB_TBANDS+1];
+ float mem[32];
+ float cmean[8];
+ float std[9];
+ float Etracker;
+ float lowECount;
+ int E_count;
+ int count;
+ int analysis_offset;
+ int write_pos;
+ int read_pos;
+ int read_subframe;
+ float hp_ener_accum;
+ float rnn_state[MAX_NEURONS];
+ opus_val32 downmix_state[3];
+ AnalysisInfo info[DETECT_SIZE];
+} TonalityAnalysisState;
+
+/** Initialize a TonalityAnalysisState struct.
+ *
+ * This performs some possibly slow initialization steps which should
+ * not be repeated every analysis step. No allocated memory is retained
+ * by the state struct, so no cleanup call is required.
+ */
+void tonality_analysis_init(TonalityAnalysisState *analysis, opus_int32 Fs);
+
+/** Reset a TonalityAnalysisState stuct.
+ *
+ * Call this when there's a discontinuity in the data.
+ */
+void tonality_analysis_reset(TonalityAnalysisState *analysis);
+
+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/lib/rbcodec/codecs/libopus/celt/_kiss_fft_guts.h b/lib/rbcodec/codecs/libopus/celt/_kiss_fft_guts.h
index 8ddb9adc96..17392b3e90 100644
--- a/lib/rbcodec/codecs/libopus/celt/_kiss_fft_guts.h
+++ b/lib/rbcodec/codecs/libopus/celt/_kiss_fft_guts.h
@@ -58,12 +58,12 @@
# define S_MUL(a,b) MULT16_32_Q15(b, a)
# define C_MUL(m,a,b) \
- do{ (m).r = SUB32(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)); \
- (m).i = ADD32(S_MUL((a).r,(b).i) , S_MUL((a).i,(b).r)); }while(0)
+ do{ (m).r = SUB32_ovflw(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)); \
+ (m).i = ADD32_ovflw(S_MUL((a).r,(b).i) , S_MUL((a).i,(b).r)); }while(0)
# define C_MULC(m,a,b) \
- do{ (m).r = ADD32(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)); \
- (m).i = SUB32(S_MUL((a).i,(b).r) , S_MUL((a).r,(b).i)); }while(0)
+ do{ (m).r = ADD32_ovflw(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)); \
+ (m).i = SUB32_ovflw(S_MUL((a).i,(b).r) , S_MUL((a).r,(b).i)); }while(0)
# define C_MULBYSCALAR( c, s ) \
do{ (c).r = S_MUL( (c).r , s ) ;\
@@ -77,17 +77,17 @@
DIVSCALAR( (c).i , div); }while (0)
#define C_ADD( res, a,b)\
- do {(res).r=ADD32((a).r,(b).r); (res).i=ADD32((a).i,(b).i); \
+ do {(res).r=ADD32_ovflw((a).r,(b).r); (res).i=ADD32_ovflw((a).i,(b).i); \
}while(0)
#define C_SUB( res, a,b)\
- do {(res).r=SUB32((a).r,(b).r); (res).i=SUB32((a).i,(b).i); \
+ do {(res).r=SUB32_ovflw((a).r,(b).r); (res).i=SUB32_ovflw((a).i,(b).i); \
}while(0)
#define C_ADDTO( res , a)\
- do {(res).r = ADD32((res).r, (a).r); (res).i = ADD32((res).i,(a).i);\
+ do {(res).r = ADD32_ovflw((res).r, (a).r); (res).i = ADD32_ovflw((res).i,(a).i);\
}while(0)
#define C_SUBFROM( res , a)\
- do {(res).r = ADD32((res).r,(a).r); (res).i = SUB32((res).i,(a).i); \
+ do {(res).r = ADD32_ovflw((res).r,(a).r); (res).i = SUB32_ovflw((res).i,(a).i); \
}while(0)
#if defined(OPUS_ARM_INLINE_ASM)
@@ -97,9 +97,8 @@
#if defined(OPUS_ARM_INLINE_EDSP)
#include "arm/kiss_fft_armv5e.h"
#endif
-
-#if defined(OPUS_CF_INLINE_ASM)
-#include "cf/kiss_fft_cf.h"
+#if defined(MIPSr1_ASM)
+#include "mips/kiss_fft_mipsr1.h"
#endif
#else /* not FIXED_POINT*/
diff --git a/lib/rbcodec/codecs/libopus/celt/arch.h b/lib/rbcodec/codecs/libopus/celt/arch.h
index 035b92ff29..08b07db598 100644
--- a/lib/rbcodec/codecs/libopus/celt/arch.h
+++ b/lib/rbcodec/codecs/libopus/celt/arch.h
@@ -46,30 +46,54 @@
# endif
# endif
+#if OPUS_GNUC_PREREQ(3, 0)
+#define opus_likely(x) (__builtin_expect(!!(x), 1))
+#define opus_unlikely(x) (__builtin_expect(!!(x), 0))
+#else
+#define opus_likely(x) (!!(x))
+#define opus_unlikely(x) (!!(x))
+#endif
+
#define CELT_SIG_SCALE 32768.f
-#define celt_fatal(str) _celt_fatal(str, __FILE__, __LINE__);
-#ifdef ENABLE_ASSERTIONS
+#define CELT_FATAL(str) celt_fatal(str, __FILE__, __LINE__);
+
+#if defined(ENABLE_ASSERTIONS) || defined(ENABLE_HARDENING)
+#ifdef __GNUC__
+__attribute__((noreturn))
+#endif
+void celt_fatal(const char *str, const char *file, int line);
+
+#if defined(CELT_C) && !defined(OVERRIDE_celt_fatal)
#include <stdio.h>
#include <stdlib.h>
#ifdef __GNUC__
__attribute__((noreturn))
#endif
-static OPUS_INLINE void _celt_fatal(const char *str, const char *file, int line)
+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();
}
-#define celt_assert(cond) {if (!(cond)) {celt_fatal("assertion failed: " #cond);}}
-#define celt_assert2(cond, message) {if (!(cond)) {celt_fatal("assertion failed: " #cond "\n" message);}}
+#endif
+
+#define celt_assert(cond) {if (!(cond)) {CELT_FATAL("assertion failed: " #cond);}}
+#define celt_assert2(cond, message) {if (!(cond)) {CELT_FATAL("assertion failed: " #cond "\n" message);}}
+#define MUST_SUCCEED(call) celt_assert((call) == OPUS_OK)
#else
#define celt_assert(cond)
#define celt_assert2(cond, message)
+#define MUST_SUCCEED(call) do {if((call) != OPUS_OK) {RESTORE_STACK; return OPUS_INTERNAL_ERROR;} } while (0)
+#endif
+
+#if defined(ENABLE_ASSERTIONS)
+#define celt_sig_assert(cond) {if (!(cond)) {CELT_FATAL("signal assertion failed: " #cond);}}
+#else
+#define celt_sig_assert(cond)
#endif
#define IMUL32(a,b) ((a)*(b))
-#define ABS(x) ((x) < 0 ? (-(x)) : (x))
#define MIN16(a,b) ((a) < (b) ? (a) : (b)) /**< Minimum 16-bit value. */
#define MAX16(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum 16-bit value. */
#define MIN32(a,b) ((a) < (b) ? (a) : (b)) /**< Minimum 32-bit value. */
@@ -79,20 +103,35 @@ static OPUS_INLINE void _celt_fatal(const char *str, const char *file, int line)
#define UADD32(a,b) ((a)+(b))
#define USUB32(a,b) ((a)-(b))
+/* Set this if opus_int64 is a native type of the CPU. */
+/* Assume that all LP64 architectures have fast 64-bit types; also x86_64
+ (which can be ILP32 for x32) and Win64 (which is LLP64). */
+#if defined(__x86_64__) || defined(__LP64__) || defined(_WIN64)
+#define OPUS_FAST_INT64 1
+#else
+#define OPUS_FAST_INT64 0
+#endif
+
#define PRINT_MIPS(file)
#ifdef FIXED_POINT
typedef opus_int16 opus_val16;
typedef opus_int32 opus_val32;
+typedef opus_int64 opus_val64;
typedef opus_val32 celt_sig;
typedef opus_val16 celt_norm;
typedef opus_val32 celt_ener;
+#define celt_isnan(x) 0
+
#define Q15ONE 32767
#define SIG_SHIFT 12
+/* Safe saturation value for 32-bit signals. Should be less than
+ 2^31*(1-0.85) to avoid blowing up on DC at deemphasis.*/
+#define SIG_SAT (300000000)
#define NORM_SCALING 16384
@@ -119,7 +158,9 @@ static OPUS_INLINE opus_int16 SAT16(opus_int32 x) {
#include "fixed_generic.h"
-#ifdef OPUS_ARM_INLINE_EDSP
+#ifdef OPUS_ARM_PRESUME_AARCH64_NEON_INTR
+#include "arm/fixed_arm64.h"
+#elif defined (OPUS_ARM_INLINE_EDSP)
#include "arm/fixed_armv5e.h"
#elif defined (OPUS_ARM_INLINE_ASM)
#include "arm/fixed_armv4.h"
@@ -129,8 +170,6 @@ static OPUS_INLINE opus_int16 SAT16(opus_int32 x) {
#include "fixed_c5x.h"
#elif defined (TI_C6X_ASM)
#include "fixed_c6x.h"
-#elif defined (OPUS_CF_INLINE_ASM)
-#include "cf/fixed_cf.h"
#endif
#endif
@@ -139,6 +178,7 @@ static OPUS_INLINE opus_int16 SAT16(opus_int32 x) {
typedef float opus_val16;
typedef float opus_val32;
+typedef float opus_val64;
typedef float celt_sig;
typedef float celt_norm;
@@ -178,6 +218,7 @@ static OPUS_INLINE int celt_isnan(float x)
#define NEG16(x) (-(x))
#define NEG32(x) (-(x))
+#define NEG32_ovflw(x) (-(x))
#define EXTRACT16(x) (x)
#define EXTEND32(x) (x)
#define SHR16(a,shift) (a)
@@ -194,6 +235,7 @@ static OPUS_INLINE int celt_isnan(float x)
#define SATURATE16(x) (x)
#define ROUND16(a,shift) (a)
+#define SROUND16(a,shift) (a)
#define HALF16(x) (.5f*(x))
#define HALF32(x) (.5f*(x))
@@ -201,6 +243,8 @@ static OPUS_INLINE int celt_isnan(float x)
#define SUB16(a,b) ((a)-(b))
#define ADD32(a,b) ((a)+(b))
#define SUB32(a,b) ((a)-(b))
+#define ADD32_ovflw(a,b) ((a)+(b))
+#define SUB32_ovflw(a,b) ((a)-(b))
#define MULT16_16_16(a,b) ((a)*(b))
#define MULT16_16(a,b) ((opus_val32)(a)*(opus_val32)(b))
#define MAC16_16(c,a,b) ((c)+(opus_val32)(a)*(opus_val32)(b))
@@ -235,9 +279,9 @@ static OPUS_INLINE int celt_isnan(float x)
#ifndef GLOBAL_STACK_SIZE
#ifdef FIXED_POINT
-#define GLOBAL_STACK_SIZE 100000
+#define GLOBAL_STACK_SIZE 120000
#else
-#define GLOBAL_STACK_SIZE 100000
+#define GLOBAL_STACK_SIZE 120000
#endif
#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/arm/arm2gnu.pl b/lib/rbcodec/codecs/libopus/celt/arm/arm2gnu.pl
new file mode 100755
index 0000000000..a2895f7445
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/arm/arm2gnu.pl
@@ -0,0 +1,353 @@
+#!/usr/bin/perl
+# Copyright (C) 2002-2013 Xiph.org Foundation
+#
+# 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.
+
+my $bigend; # little/big endian
+my $nxstack;
+my $apple = 0;
+my $symprefix = "";
+
+$nxstack = 0;
+
+eval 'exec /usr/local/bin/perl -S $0 ${1+"$@"}'
+ if $running_under_some_shell;
+
+while ($ARGV[0] =~ /^-/) {
+ $_ = shift;
+ last if /^--$/;
+ if (/^-n$/) {
+ $nflag++;
+ next;
+ }
+ if (/^--apple$/) {
+ $apple = 1;
+ $symprefix = "_";
+ 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 = ();
+
+printf (" .syntax unified\n");
+
+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\s*/.global $symprefix/;
+ 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/ unless ($apple);
+ }
+ }
+
+ 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", $proc) unless ($apple);
+ # Make sure we $prefix isn't empty here (for the $apple case).
+ # We handle mangling the label here, make sure it doesn't match
+ # the label handling below (if $prefix would be empty).
+ $prefix = $prefix."; ";
+ push(@proc_stack, $proc);
+ s/^[A-Za-z_\.]\w+/$symprefix$&:/;
+ }
+ $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 && !$apple);
+ }
+ 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 && !$apple) {
+ printf (" .section\t.note.GNU-stack,\"\",\%\%progbits\n");
+}
diff --git a/lib/rbcodec/codecs/libopus/celt/arm/arm_celt_map.c b/lib/rbcodec/codecs/libopus/celt/arm/arm_celt_map.c
new file mode 100644
index 0000000000..ca988b66f5
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/arm/arm_celt_map.c
@@ -0,0 +1,160 @@
+/* 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"
+#include "kiss_fft.h"
+#include "mdct.h"
+
+#if defined(OPUS_HAVE_RTCD)
+
+# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)
+opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *x, const opus_val16 *y, int N) = {
+ celt_inner_prod_c, /* ARMv4 */
+ celt_inner_prod_c, /* EDSP */
+ celt_inner_prod_c, /* Media */
+ celt_inner_prod_neon /* NEON */
+};
+
+void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02,
+ int N, opus_val32 *xy1, opus_val32 *xy2) = {
+ dual_inner_prod_c, /* ARMv4 */
+ dual_inner_prod_c, /* EDSP */
+ dual_inner_prod_c, /* Media */
+ dual_inner_prod_neon /* NEON */
+};
+# endif
+
+# if defined(FIXED_POINT)
+# if ((defined(OPUS_ARM_MAY_HAVE_NEON) && !defined(OPUS_ARM_PRESUME_NEON)) || \
+ (defined(OPUS_ARM_MAY_HAVE_MEDIA) && !defined(OPUS_ARM_PRESUME_MEDIA)) || \
+ (defined(OPUS_ARM_MAY_HAVE_EDSP) && !defined(OPUS_ARM_PRESUME_EDSP)))
+opus_val32 (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
+ const opus_val16 *, opus_val32 *, int, 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 */
+};
+
+# endif
+# else /* !FIXED_POINT */
+# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)
+void (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
+ const opus_val16 *, opus_val32 *, int, int, int) = {
+ celt_pitch_xcorr_c, /* ARMv4 */
+ celt_pitch_xcorr_c, /* EDSP */
+ celt_pitch_xcorr_c, /* Media */
+ celt_pitch_xcorr_float_neon /* Neon */
+};
+# endif
+# endif /* FIXED_POINT */
+
+#if defined(FIXED_POINT) && defined(OPUS_HAVE_RTCD) && \
+ defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)
+
+void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])(
+ const opus_val16 *x,
+ const opus_val16 *y,
+ opus_val32 sum[4],
+ int len
+) = {
+ xcorr_kernel_c, /* ARMv4 */
+ xcorr_kernel_c, /* EDSP */
+ xcorr_kernel_c, /* Media */
+ xcorr_kernel_neon_fixed, /* Neon */
+};
+
+#endif
+
+# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+# if defined(HAVE_ARM_NE10)
+# if defined(CUSTOM_MODES)
+int (*const OPUS_FFT_ALLOC_ARCH_IMPL[OPUS_ARCHMASK+1])(kiss_fft_state *st) = {
+ opus_fft_alloc_arch_c, /* ARMv4 */
+ opus_fft_alloc_arch_c, /* EDSP */
+ opus_fft_alloc_arch_c, /* Media */
+ opus_fft_alloc_arm_neon /* Neon with NE10 library support */
+};
+
+void (*const OPUS_FFT_FREE_ARCH_IMPL[OPUS_ARCHMASK+1])(kiss_fft_state *st) = {
+ opus_fft_free_arch_c, /* ARMv4 */
+ opus_fft_free_arch_c, /* EDSP */
+ opus_fft_free_arch_c, /* Media */
+ opus_fft_free_arm_neon /* Neon with NE10 */
+};
+# endif /* CUSTOM_MODES */
+
+void (*const OPUS_FFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg,
+ const kiss_fft_cpx *fin,
+ kiss_fft_cpx *fout) = {
+ opus_fft_c, /* ARMv4 */
+ opus_fft_c, /* EDSP */
+ opus_fft_c, /* Media */
+ opus_fft_neon /* Neon with NE10 */
+};
+
+void (*const OPUS_IFFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg,
+ const kiss_fft_cpx *fin,
+ kiss_fft_cpx *fout) = {
+ opus_ifft_c, /* ARMv4 */
+ opus_ifft_c, /* EDSP */
+ opus_ifft_c, /* Media */
+ opus_ifft_neon /* Neon with NE10 */
+};
+
+void (*const CLT_MDCT_FORWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l,
+ kiss_fft_scalar *in,
+ kiss_fft_scalar * OPUS_RESTRICT out,
+ const opus_val16 *window,
+ int overlap, int shift,
+ int stride, int arch) = {
+ clt_mdct_forward_c, /* ARMv4 */
+ clt_mdct_forward_c, /* EDSP */
+ clt_mdct_forward_c, /* Media */
+ clt_mdct_forward_neon /* Neon with NE10 */
+};
+
+void (*const CLT_MDCT_BACKWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l,
+ kiss_fft_scalar *in,
+ kiss_fft_scalar * OPUS_RESTRICT out,
+ const opus_val16 *window,
+ int overlap, int shift,
+ int stride, int arch) = {
+ clt_mdct_backward_c, /* ARMv4 */
+ clt_mdct_backward_c, /* EDSP */
+ clt_mdct_backward_c, /* Media */
+ clt_mdct_backward_neon /* Neon with NE10 */
+};
+
+# endif /* HAVE_ARM_NE10 */
+# endif /* OPUS_ARM_MAY_HAVE_NEON_INTR */
+
+#endif /* OPUS_HAVE_RTCD */
diff --git a/lib/rbcodec/codecs/libopus/celt/arm/armcpu.c b/lib/rbcodec/codecs/libopus/celt/arm/armcpu.c
new file mode 100644
index 0000000000..694a63b78e
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/arm/armcpu.c
@@ -0,0 +1,185 @@
+/* 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"
+#include "arch.h"
+
+#define OPUS_CPU_ARM_V4_FLAG (1<<OPUS_ARCH_ARM_V4)
+#define OPUS_CPU_ARM_EDSP_FLAG (1<<OPUS_ARCH_ARM_EDSP)
+#define OPUS_CPU_ARM_MEDIA_FLAG (1<<OPUS_ARCH_ARM_MEDIA)
+#define OPUS_CPU_ARM_NEON_FLAG (1<<OPUS_ARCH_ARM_NEON)
+
+#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) || defined(OPUS_ARM_MAY_HAVE_MEDIA) \
+ || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+ __try{
+ /*PLD [r13]*/
+ __emit(0xF5DDF000);
+ flags|=OPUS_CPU_ARM_EDSP_FLAG;
+ }
+ __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){
+ /*Ignore exception.*/
+ }
+# if defined(OPUS_ARM_MAY_HAVE_MEDIA) \
+ || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+ __try{
+ /*SHADD8 r3,r3,r3*/
+ __emit(0xE6333F93);
+ flags|=OPUS_CPU_ARM_MEDIA_FLAG;
+ }
+ __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){
+ /*Ignore exception.*/
+ }
+# if defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+ __try{
+ /*VORR q0,q0,q0*/
+ __emit(0xF2200150);
+ flags|=OPUS_CPU_ARM_NEON_FLAG;
+ }
+ __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_MEDIA) \
+ || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+ /* Search for edsp and neon flag */
+ if(memcmp(buf, "Features", 8) == 0)
+ {
+ char *p;
+ p = strstr(buf, " edsp");
+ if(p != NULL && (p[5] == ' ' || p[5] == '\n'))
+ flags |= OPUS_CPU_ARM_EDSP_FLAG;
+
+# if defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+ p = strstr(buf, " neon");
+ if(p != NULL && (p[5] == ' ' || p[5] == '\n'))
+ flags |= OPUS_CPU_ARM_NEON_FLAG;
+# endif
+ }
+# endif
+
+# if defined(OPUS_ARM_MAY_HAVE_MEDIA) \
+ || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+ /* 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_FLAG;
+ }
+# 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_FLAG)) {
+ /* Asserts ensure arch values are sequential */
+ celt_assert(arch == OPUS_ARCH_ARM_V4);
+ return arch;
+ }
+ arch++;
+
+ if(!(flags & OPUS_CPU_ARM_MEDIA_FLAG)) {
+ celt_assert(arch == OPUS_ARCH_ARM_EDSP);
+ return arch;
+ }
+ arch++;
+
+ if(!(flags & OPUS_CPU_ARM_NEON_FLAG)) {
+ celt_assert(arch == OPUS_ARCH_ARM_MEDIA);
+ return arch;
+ }
+ arch++;
+
+ celt_assert(arch == OPUS_ARCH_ARM_NEON);
+ return arch;
+}
+
+#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/arm/armcpu.h b/lib/rbcodec/codecs/libopus/celt/arm/armcpu.h
new file mode 100644
index 0000000000..820262ff5f
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/arm/armcpu.h
@@ -0,0 +1,77 @@
+/* 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);
+
+#define OPUS_ARCH_ARM_V4 (0)
+#define OPUS_ARCH_ARM_EDSP (1)
+#define OPUS_ARCH_ARM_MEDIA (2)
+#define OPUS_ARCH_ARM_NEON (3)
+
+# endif
+
+#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/arm/armopts.s.in b/lib/rbcodec/codecs/libopus/celt/arm/armopts.s.in
new file mode 100644
index 0000000000..3d8aaf2754
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/arm/armopts.s.in
@@ -0,0 +1,37 @@
+/* Copyright (C) 2013 Mozilla Corporation */
+/*
+ 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.
+*/
+
+; 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@
+
+; Set the following to 1 if we have ARMv6 media instructions.
+OPUS_ARM_MAY_HAVE_MEDIA * @OPUS_ARM_MAY_HAVE_MEDIA@
+
+; 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/lib/rbcodec/codecs/libopus/celt/arm/celt_fft_ne10.c b/lib/rbcodec/codecs/libopus/celt/arm/celt_fft_ne10.c
new file mode 100644
index 0000000000..ea5fd7808b
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/arm/celt_fft_ne10.c
@@ -0,0 +1,173 @@
+/* Copyright (c) 2015 Xiph.Org Foundation
+ Written by Viswanath Puttagunta */
+/**
+ @file celt_fft_ne10.c
+ @brief ARM Neon optimizations for fft using NE10 library
+ */
+
+/*
+ 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 SKIP_CONFIG_H
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+#endif
+
+#include <NE10_dsp.h>
+#include "os_support.h"
+#include "kiss_fft.h"
+#include "stack_alloc.h"
+
+#if !defined(FIXED_POINT)
+# define NE10_FFT_ALLOC_C2C_TYPE_NEON ne10_fft_alloc_c2c_float32_neon
+# define NE10_FFT_CFG_TYPE_T ne10_fft_cfg_float32_t
+# define NE10_FFT_STATE_TYPE_T ne10_fft_state_float32_t
+# define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_float32
+# define NE10_FFT_CPX_TYPE_T ne10_fft_cpx_float32_t
+# define NE10_FFT_C2C_1D_TYPE_NEON ne10_fft_c2c_1d_float32_neon
+#else
+# define NE10_FFT_ALLOC_C2C_TYPE_NEON(nfft) ne10_fft_alloc_c2c_int32_neon(nfft)
+# define NE10_FFT_CFG_TYPE_T ne10_fft_cfg_int32_t
+# define NE10_FFT_STATE_TYPE_T ne10_fft_state_int32_t
+# define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_int32
+# define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_int32
+# define NE10_FFT_CPX_TYPE_T ne10_fft_cpx_int32_t
+# define NE10_FFT_C2C_1D_TYPE_NEON ne10_fft_c2c_1d_int32_neon
+#endif
+
+#if defined(CUSTOM_MODES)
+
+/* nfft lengths in NE10 that support scaled fft */
+# define NE10_FFTSCALED_SUPPORT_MAX 4
+static const int ne10_fft_scaled_support[NE10_FFTSCALED_SUPPORT_MAX] = {
+ 480, 240, 120, 60
+};
+
+int opus_fft_alloc_arm_neon(kiss_fft_state *st)
+{
+ int i;
+ size_t memneeded = sizeof(struct arch_fft_state);
+
+ st->arch_fft = (arch_fft_state *)opus_alloc(memneeded);
+ if (!st->arch_fft)
+ return -1;
+
+ for (i = 0; i < NE10_FFTSCALED_SUPPORT_MAX; i++) {
+ if(st->nfft == ne10_fft_scaled_support[i])
+ break;
+ }
+ if (i == NE10_FFTSCALED_SUPPORT_MAX) {
+ /* This nfft length (scaled fft) is not supported in NE10 */
+ st->arch_fft->is_supported = 0;
+ st->arch_fft->priv = NULL;
+ }
+ else {
+ st->arch_fft->is_supported = 1;
+ st->arch_fft->priv = (void *)NE10_FFT_ALLOC_C2C_TYPE_NEON(st->nfft);
+ if (st->arch_fft->priv == NULL) {
+ return -1;
+ }
+ }
+ return 0;
+}
+
+void opus_fft_free_arm_neon(kiss_fft_state *st)
+{
+ NE10_FFT_CFG_TYPE_T cfg;
+
+ if (!st->arch_fft)
+ return;
+
+ cfg = (NE10_FFT_CFG_TYPE_T)st->arch_fft->priv;
+ if (cfg)
+ NE10_FFT_DESTROY_C2C_TYPE(cfg);
+ opus_free(st->arch_fft);
+}
+#endif
+
+void opus_fft_neon(const kiss_fft_state *st,
+ const kiss_fft_cpx *fin,
+ kiss_fft_cpx *fout)
+{
+ NE10_FFT_STATE_TYPE_T state;
+ NE10_FFT_CFG_TYPE_T cfg = &state;
+ VARDECL(NE10_FFT_CPX_TYPE_T, buffer);
+ SAVE_STACK;
+ ALLOC(buffer, st->nfft, NE10_FFT_CPX_TYPE_T);
+
+ if (!st->arch_fft->is_supported) {
+ /* This nfft length (scaled fft) not supported in NE10 */
+ opus_fft_c(st, fin, fout);
+ }
+ else {
+ memcpy((void *)cfg, st->arch_fft->priv, sizeof(NE10_FFT_STATE_TYPE_T));
+ state.buffer = (NE10_FFT_CPX_TYPE_T *)&buffer[0];
+#if !defined(FIXED_POINT)
+ state.is_forward_scaled = 1;
+
+ NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout,
+ (NE10_FFT_CPX_TYPE_T *)fin,
+ cfg, 0);
+#else
+ NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout,
+ (NE10_FFT_CPX_TYPE_T *)fin,
+ cfg, 0, 1);
+#endif
+ }
+ RESTORE_STACK;
+}
+
+void opus_ifft_neon(const kiss_fft_state *st,
+ const kiss_fft_cpx *fin,
+ kiss_fft_cpx *fout)
+{
+ NE10_FFT_STATE_TYPE_T state;
+ NE10_FFT_CFG_TYPE_T cfg = &state;
+ VARDECL(NE10_FFT_CPX_TYPE_T, buffer);
+ SAVE_STACK;
+ ALLOC(buffer, st->nfft, NE10_FFT_CPX_TYPE_T);
+
+ if (!st->arch_fft->is_supported) {
+ /* This nfft length (scaled fft) not supported in NE10 */
+ opus_ifft_c(st, fin, fout);
+ }
+ else {
+ memcpy((void *)cfg, st->arch_fft->priv, sizeof(NE10_FFT_STATE_TYPE_T));
+ state.buffer = (NE10_FFT_CPX_TYPE_T *)&buffer[0];
+#if !defined(FIXED_POINT)
+ state.is_backward_scaled = 0;
+
+ NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout,
+ (NE10_FFT_CPX_TYPE_T *)fin,
+ cfg, 1);
+#else
+ NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout,
+ (NE10_FFT_CPX_TYPE_T *)fin,
+ cfg, 1, 0);
+#endif
+ }
+ RESTORE_STACK;
+}
diff --git a/lib/rbcodec/codecs/libopus/celt/arm/celt_mdct_ne10.c b/lib/rbcodec/codecs/libopus/celt/arm/celt_mdct_ne10.c
new file mode 100644
index 0000000000..3531d02d10
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/arm/celt_mdct_ne10.c
@@ -0,0 +1,258 @@
+/* Copyright (c) 2015 Xiph.Org Foundation
+ Written by Viswanath Puttagunta */
+/**
+ @file celt_mdct_ne10.c
+ @brief ARM Neon optimizations for mdct using NE10 library
+ */
+
+/*
+ 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 SKIP_CONFIG_H
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+#endif
+
+#include "kiss_fft.h"
+#include "_kiss_fft_guts.h"
+#include "mdct.h"
+#include "stack_alloc.h"
+
+void clt_mdct_forward_neon(const mdct_lookup *l,
+ kiss_fft_scalar *in,
+ kiss_fft_scalar * OPUS_RESTRICT out,
+ const opus_val16 *window,
+ int overlap, int shift, int stride, int arch)
+{
+ int i;
+ int N, N2, N4;
+ VARDECL(kiss_fft_scalar, f);
+ VARDECL(kiss_fft_cpx, f2);
+ const kiss_fft_state *st = l->kfft[shift];
+ const kiss_twiddle_scalar *trig;
+
+ SAVE_STACK;
+
+ N = l->n;
+ trig = l->trig;
+ for (i=0;i<shift;i++)
+ {
+ N >>= 1;
+ trig += N;
+ }
+ N2 = N>>1;
+ N4 = N>>2;
+
+ ALLOC(f, N2, kiss_fft_scalar);
+ ALLOC(f2, N4, kiss_fft_cpx);
+
+ /* Consider the input to be composed of four blocks: [a, b, c, d] */
+ /* Window, shuffle, fold */
+ {
+ /* Temp pointers to make it really clear to the compiler what we're doing */
+ const kiss_fft_scalar * OPUS_RESTRICT xp1 = in+(overlap>>1);
+ const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+N2-1+(overlap>>1);
+ 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+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);
+ *yp++ = MULT16_32_Q15(*wp1, *xp1) - MULT16_32_Q15(*wp2, xp2[-N2]);
+ xp1+=2;
+ xp2-=2;
+ wp1+=2;
+ wp2-=2;
+ }
+ wp1 = window;
+ wp2 = window+overlap-1;
+ for(;i<N4-((overlap+3)>>2);i++)
+ {
+ /* Real part arranged as a-bR, Imag part arranged as -c-dR */
+ *yp++ = *xp2;
+ *yp++ = *xp1;
+ xp1+=2;
+ xp2-=2;
+ }
+ for(;i<N4;i++)
+ {
+ /* Real part arranged as a-bR, Imag part arranged as -c-dR */
+ *yp++ = -MULT16_32_Q15(*wp1, xp1[-N2]) + MULT16_32_Q15(*wp2, *xp2);
+ *yp++ = MULT16_32_Q15(*wp2, *xp1) + MULT16_32_Q15(*wp1, xp2[N2]);
+ xp1+=2;
+ xp2-=2;
+ wp1+=2;
+ wp2-=2;
+ }
+ }
+ /* Pre-rotation */
+ {
+ kiss_fft_scalar * OPUS_RESTRICT yp = f;
+ const kiss_twiddle_scalar *t = &trig[0];
+ for(i=0;i<N4;i++)
+ {
+ kiss_fft_cpx yc;
+ kiss_twiddle_scalar t0, t1;
+ kiss_fft_scalar re, im, yr, yi;
+ t0 = t[i];
+ t1 = t[N4+i];
+ re = *yp++;
+ im = *yp++;
+ yr = S_MUL(re,t0) - S_MUL(im,t1);
+ yi = S_MUL(im,t0) + S_MUL(re,t1);
+ yc.r = yr;
+ yc.i = yi;
+ f2[i] = yc;
+ }
+ }
+
+ opus_fft(st, f2, (kiss_fft_cpx *)f, arch);
+
+ /* Post-rotate */
+ {
+ /* Temp pointers to make it really clear to the compiler what we're doing */
+ const kiss_fft_cpx * OPUS_RESTRICT fp = (kiss_fft_cpx *)f;
+ kiss_fft_scalar * OPUS_RESTRICT yp1 = out;
+ kiss_fft_scalar * OPUS_RESTRICT yp2 = out+stride*(N2-1);
+ const kiss_twiddle_scalar *t = &trig[0];
+ /* Temp pointers to make it really clear to the compiler what we're doing */
+ for(i=0;i<N4;i++)
+ {
+ kiss_fft_scalar yr, yi;
+ yr = S_MUL(fp->i,t[N4+i]) - S_MUL(fp->r,t[i]);
+ yi = S_MUL(fp->r,t[N4+i]) + S_MUL(fp->i,t[i]);
+ *yp1 = yr;
+ *yp2 = yi;
+ fp++;
+ yp1 += 2*stride;
+ yp2 -= 2*stride;
+ }
+ }
+ RESTORE_STACK;
+}
+
+void clt_mdct_backward_neon(const mdct_lookup *l,
+ kiss_fft_scalar *in,
+ kiss_fft_scalar * OPUS_RESTRICT out,
+ const opus_val16 * OPUS_RESTRICT window,
+ int overlap, int shift, int stride, int arch)
+{
+ int i;
+ int N, N2, N4;
+ VARDECL(kiss_fft_scalar, f);
+ const kiss_twiddle_scalar *trig;
+ const kiss_fft_state *st = l->kfft[shift];
+
+ N = l->n;
+ trig = l->trig;
+ for (i=0;i<shift;i++)
+ {
+ N >>= 1;
+ trig += N;
+ }
+ N2 = N>>1;
+ N4 = N>>2;
+
+ ALLOC(f, N2, kiss_fft_scalar);
+
+ /* Pre-rotate */
+ {
+ /* Temp pointers to make it really clear to the compiler what we're doing */
+ const kiss_fft_scalar * OPUS_RESTRICT xp1 = in;
+ const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+stride*(N2-1);
+ kiss_fft_scalar * OPUS_RESTRICT yp = f;
+ const kiss_twiddle_scalar * OPUS_RESTRICT t = &trig[0];
+ for(i=0;i<N4;i++)
+ {
+ kiss_fft_scalar yr, yi;
+ yr = S_MUL(*xp2, t[i]) + S_MUL(*xp1, t[N4+i]);
+ yi = S_MUL(*xp1, t[i]) - S_MUL(*xp2, t[N4+i]);
+ yp[2*i] = yr;
+ yp[2*i+1] = yi;
+ xp1+=2*stride;
+ xp2-=2*stride;
+ }
+ }
+
+ opus_ifft(st, (kiss_fft_cpx *)f, (kiss_fft_cpx*)(out+(overlap>>1)), arch);
+
+ /* Post-rotate and de-shuffle from both ends of the buffer at once to make
+ it in-place. */
+ {
+ kiss_fft_scalar * yp0 = out+(overlap>>1);
+ kiss_fft_scalar * yp1 = out+(overlap>>1)+N2-2;
+ const kiss_twiddle_scalar *t = &trig[0];
+ /* 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;
+ kiss_twiddle_scalar t0, t1;
+ re = yp0[0];
+ im = yp0[1];
+ t0 = t[i];
+ t1 = t[N4+i];
+ /* 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(re,t1) - S_MUL(im,t0);
+ re = yp1[0];
+ im = yp1[1];
+ yp0[0] = yr;
+ yp1[1] = yi;
+
+ t0 = t[(N4-i-1)];
+ t1 = t[(N2-i-1)];
+ /* 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(re,t1) - S_MUL(im,t0);
+ yp1[0] = yr;
+ yp0[1] = yi;
+ yp0 += 2;
+ yp1 -= 2;
+ }
+ }
+
+ /* Mirror on both sides for TDAC */
+ {
+ 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 < overlap/2; i++)
+ {
+ 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--;
+ }
+ }
+ RESTORE_STACK;
+}
diff --git a/lib/rbcodec/codecs/libopus/celt/arm/celt_neon_intr.c b/lib/rbcodec/codecs/libopus/celt/arm/celt_neon_intr.c
new file mode 100644
index 0000000000..effda769d0
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/arm/celt_neon_intr.c
@@ -0,0 +1,211 @@
+/* Copyright (c) 2014-2015 Xiph.Org Foundation
+ Written by Viswanath Puttagunta */
+/**
+ @file celt_neon_intr.c
+ @brief ARM Neon Intrinsic optimizations for celt
+ */
+
+/*
+ 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 <arm_neon.h>
+#include "../pitch.h"
+
+#if defined(FIXED_POINT)
+void xcorr_kernel_neon_fixed(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[4], int len)
+{
+ int j;
+ int32x4_t a = vld1q_s32(sum);
+ /* Load y[0...3] */
+ /* This requires len>0 to always be valid (which we assert in the C code). */
+ int16x4_t y0 = vld1_s16(y);
+ y += 4;
+
+ for (j = 0; j + 8 <= len; j += 8)
+ {
+ /* Load x[0...7] */
+ int16x8_t xx = vld1q_s16(x);
+ int16x4_t x0 = vget_low_s16(xx);
+ int16x4_t x4 = vget_high_s16(xx);
+ /* Load y[4...11] */
+ int16x8_t yy = vld1q_s16(y);
+ int16x4_t y4 = vget_low_s16(yy);
+ int16x4_t y8 = vget_high_s16(yy);
+ int32x4_t a0 = vmlal_lane_s16(a, y0, x0, 0);
+ int32x4_t a1 = vmlal_lane_s16(a0, y4, x4, 0);
+
+ int16x4_t y1 = vext_s16(y0, y4, 1);
+ int16x4_t y5 = vext_s16(y4, y8, 1);
+ int32x4_t a2 = vmlal_lane_s16(a1, y1, x0, 1);
+ int32x4_t a3 = vmlal_lane_s16(a2, y5, x4, 1);
+
+ int16x4_t y2 = vext_s16(y0, y4, 2);
+ int16x4_t y6 = vext_s16(y4, y8, 2);
+ int32x4_t a4 = vmlal_lane_s16(a3, y2, x0, 2);
+ int32x4_t a5 = vmlal_lane_s16(a4, y6, x4, 2);
+
+ int16x4_t y3 = vext_s16(y0, y4, 3);
+ int16x4_t y7 = vext_s16(y4, y8, 3);
+ int32x4_t a6 = vmlal_lane_s16(a5, y3, x0, 3);
+ int32x4_t a7 = vmlal_lane_s16(a6, y7, x4, 3);
+
+ y0 = y8;
+ a = a7;
+ x += 8;
+ y += 8;
+ }
+
+ for (; j < len; j++)
+ {
+ int16x4_t x0 = vld1_dup_s16(x); /* load next x */
+ int32x4_t a0 = vmlal_s16(a, y0, x0);
+
+ int16x4_t y4 = vld1_dup_s16(y); /* load next y */
+ y0 = vext_s16(y0, y4, 1);
+ a = a0;
+ x++;
+ y++;
+ }
+
+ vst1q_s32(sum, a);
+}
+
+#else
+/*
+ * Function: xcorr_kernel_neon_float
+ * ---------------------------------
+ * Computes 4 correlation values and stores them in sum[4]
+ */
+static void xcorr_kernel_neon_float(const float32_t *x, const float32_t *y,
+ float32_t sum[4], int len) {
+ float32x4_t YY[3];
+ float32x4_t YEXT[3];
+ float32x4_t XX[2];
+ float32x2_t XX_2;
+ float32x4_t SUMM;
+ const float32_t *xi = x;
+ const float32_t *yi = y;
+
+ celt_assert(len>0);
+
+ YY[0] = vld1q_f32(yi);
+ SUMM = vdupq_n_f32(0);
+
+ /* Consume 8 elements in x vector and 12 elements in y
+ * vector. However, the 12'th element never really gets
+ * touched in this loop. So, if len == 8, then we only
+ * must access y[0] to y[10]. y[11] must not be accessed
+ * hence make sure len > 8 and not len >= 8
+ */
+ while (len > 8) {
+ yi += 4;
+ YY[1] = vld1q_f32(yi);
+ yi += 4;
+ YY[2] = vld1q_f32(yi);
+
+ XX[0] = vld1q_f32(xi);
+ xi += 4;
+ XX[1] = vld1q_f32(xi);
+ xi += 4;
+
+ SUMM = vmlaq_lane_f32(SUMM, YY[0], vget_low_f32(XX[0]), 0);
+ YEXT[0] = vextq_f32(YY[0], YY[1], 1);
+ SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[0]), 1);
+ YEXT[1] = vextq_f32(YY[0], YY[1], 2);
+ SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[0]), 0);
+ YEXT[2] = vextq_f32(YY[0], YY[1], 3);
+ SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[0]), 1);
+
+ SUMM = vmlaq_lane_f32(SUMM, YY[1], vget_low_f32(XX[1]), 0);
+ YEXT[0] = vextq_f32(YY[1], YY[2], 1);
+ SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[1]), 1);
+ YEXT[1] = vextq_f32(YY[1], YY[2], 2);
+ SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[1]), 0);
+ YEXT[2] = vextq_f32(YY[1], YY[2], 3);
+ SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[1]), 1);
+
+ YY[0] = YY[2];
+ len -= 8;
+ }
+
+ /* Consume 4 elements in x vector and 8 elements in y
+ * vector. However, the 8'th element in y never really gets
+ * touched in this loop. So, if len == 4, then we only
+ * must access y[0] to y[6]. y[7] must not be accessed
+ * hence make sure len>4 and not len>=4
+ */
+ if (len > 4) {
+ yi += 4;
+ YY[1] = vld1q_f32(yi);
+
+ XX[0] = vld1q_f32(xi);
+ xi += 4;
+
+ SUMM = vmlaq_lane_f32(SUMM, YY[0], vget_low_f32(XX[0]), 0);
+ YEXT[0] = vextq_f32(YY[0], YY[1], 1);
+ SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[0]), 1);
+ YEXT[1] = vextq_f32(YY[0], YY[1], 2);
+ SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[0]), 0);
+ YEXT[2] = vextq_f32(YY[0], YY[1], 3);
+ SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[0]), 1);
+
+ YY[0] = YY[1];
+ len -= 4;
+ }
+
+ while (--len > 0) {
+ XX_2 = vld1_dup_f32(xi++);
+ SUMM = vmlaq_lane_f32(SUMM, YY[0], XX_2, 0);
+ YY[0]= vld1q_f32(++yi);
+ }
+
+ XX_2 = vld1_dup_f32(xi);
+ SUMM = vmlaq_lane_f32(SUMM, YY[0], XX_2, 0);
+
+ vst1q_f32(sum, SUMM);
+}
+
+void celt_pitch_xcorr_float_neon(const opus_val16 *_x, const opus_val16 *_y,
+ opus_val32 *xcorr, int len, int max_pitch, int arch) {
+ int i;
+ (void)arch;
+ celt_assert(max_pitch > 0);
+ celt_sig_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0);
+
+ for (i = 0; i < (max_pitch-3); i += 4) {
+ xcorr_kernel_neon_float((const float32_t *)_x, (const float32_t *)_y+i,
+ (float32_t *)xcorr+i, len);
+ }
+
+ /* In case max_pitch isn't a multiple of 4, do non-unrolled version. */
+ for (; i < max_pitch; i++) {
+ xcorr[i] = celt_inner_prod_neon(_x, _y+i, len);
+ }
+}
+#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/arm/celt_pitch_xcorr_arm.s b/lib/rbcodec/codecs/libopus/celt/arm/celt_pitch_xcorr_arm.s
new file mode 100644
index 0000000000..6e873afc37
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/arm/celt_pitch_xcorr_arm.s
@@ -0,0 +1,551 @@
+; 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
+xcorr_kernel_neon_start
+ ; 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, int arch)
+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())
+ ; ignored:
+ ; int arch
+ 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_start
+ 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
+xcorr_kernel_edsp_start
+ ; 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)
+ LDRHGT 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)
+ LDRHGT 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)
+ LDRHGT 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
+ ; ignored:
+ ; int arch
+ 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
+ LDRHGE r6, [r4], #2
+ ; Stall
+ SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, *x, *y)
+ SUBSGE r12, r12, #1
+ LDRHGT 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_start ; 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)
+ LDRHGT 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
+ LDRHGE r6, [r4], #2
+ LDRHGE 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/lib/rbcodec/codecs/libopus/celt/arm/fft_arm.h b/lib/rbcodec/codecs/libopus/celt/arm/fft_arm.h
new file mode 100644
index 0000000000..0b78175f3a
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/arm/fft_arm.h
@@ -0,0 +1,71 @@
+/* Copyright (c) 2015 Xiph.Org Foundation
+ Written by Viswanath Puttagunta */
+/**
+ @file fft_arm.h
+ @brief ARM Neon Intrinsic optimizations for fft using NE10 library
+ */
+
+/*
+ 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(FFT_ARM_H)
+#define FFT_ARM_H
+
+#include "kiss_fft.h"
+
+#if defined(HAVE_ARM_NE10)
+
+int opus_fft_alloc_arm_neon(kiss_fft_state *st);
+void opus_fft_free_arm_neon(kiss_fft_state *st);
+
+void opus_fft_neon(const kiss_fft_state *st,
+ const kiss_fft_cpx *fin,
+ kiss_fft_cpx *fout);
+
+void opus_ifft_neon(const kiss_fft_state *st,
+ const kiss_fft_cpx *fin,
+ kiss_fft_cpx *fout);
+
+#if !defined(OPUS_HAVE_RTCD)
+#define OVERRIDE_OPUS_FFT (1)
+
+#define opus_fft_alloc_arch(_st, arch) \
+ ((void)(arch), opus_fft_alloc_arm_neon(_st))
+
+#define opus_fft_free_arch(_st, arch) \
+ ((void)(arch), opus_fft_free_arm_neon(_st))
+
+#define opus_fft(_st, _fin, _fout, arch) \
+ ((void)(arch), opus_fft_neon(_st, _fin, _fout))
+
+#define opus_ifft(_st, _fin, _fout, arch) \
+ ((void)(arch), opus_ifft_neon(_st, _fin, _fout))
+
+#endif /* OPUS_HAVE_RTCD */
+
+#endif /* HAVE_ARM_NE10 */
+
+#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/arm/fixed_arm64.h b/lib/rbcodec/codecs/libopus/celt/arm/fixed_arm64.h
new file mode 100644
index 0000000000..c6fbd3db2c
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/arm/fixed_arm64.h
@@ -0,0 +1,35 @@
+/* Copyright (C) 2015 Vidyo */
+/*
+ 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_ARM64_H
+#define FIXED_ARM64_H
+
+#include <arm_neon.h>
+
+#undef SIG2WORD16
+#define SIG2WORD16(x) (vqmovns_s32(PSHR32((x), SIG_SHIFT)))
+
+#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/arm/fixed_armv4.h b/lib/rbcodec/codecs/libopus/celt/arm/fixed_armv4.h
index efb3b1896a..d84888a772 100644
--- a/lib/rbcodec/codecs/libopus/celt/arm/fixed_armv4.h
+++ b/lib/rbcodec/codecs/libopus/celt/arm/fixed_armv4.h
@@ -37,7 +37,7 @@ static OPUS_INLINE opus_val32 MULT16_32_Q16_armv4(opus_val16 a, opus_val32 b)
"#MULT16_32_Q16\n\t"
"smull %0, %1, %2, %3\n\t"
: "=&r"(rd_lo), "=&r"(rd_hi)
- : "%r"(b),"r"(a<<16)
+ : "%r"(b),"r"(SHL32(a,16))
);
return rd_hi;
}
@@ -54,10 +54,10 @@ static OPUS_INLINE opus_val32 MULT16_32_Q15_armv4(opus_val16 a, opus_val32 b)
"#MULT16_32_Q15\n\t"
"smull %0, %1, %2, %3\n\t"
: "=&r"(rd_lo), "=&r"(rd_hi)
- : "%r"(b), "r"(a<<16)
+ : "%r"(b), "r"(SHL32(a,16))
);
/*We intentionally don't OR in the high bit of rd_lo for speed.*/
- return rd_hi<<1;
+ return SHL32(rd_hi,1);
}
#define MULT16_32_Q15(a, b) (MULT16_32_Q15_armv4(a, b))
diff --git a/lib/rbcodec/codecs/libopus/celt/arm/fixed_armv5e.h b/lib/rbcodec/codecs/libopus/celt/arm/fixed_armv5e.h
index 36a6321101..6bf73cbace 100644
--- a/lib/rbcodec/codecs/libopus/celt/arm/fixed_armv5e.h
+++ b/lib/rbcodec/codecs/libopus/celt/arm/fixed_armv5e.h
@@ -59,7 +59,7 @@ static OPUS_INLINE opus_val32 MULT16_32_Q15_armv5e(opus_val16 a, opus_val32 b)
: "=r"(res)
: "r"(b), "r"(a)
);
- return res<<1;
+ return SHL32(res,1);
}
#define MULT16_32_Q15(a, b) (MULT16_32_Q15_armv5e(a, b))
@@ -76,7 +76,7 @@ static OPUS_INLINE opus_val32 MAC16_32_Q15_armv5e(opus_val32 c, opus_val16 a,
"#MAC16_32_Q15\n\t"
"smlawb %0, %1, %2, %3;\n"
: "=r"(res)
- : "r"(b<<1), "r"(a), "r"(c)
+ : "r"(SHL32(b,1)), "r"(a), "r"(c)
);
return res;
}
diff --git a/lib/rbcodec/codecs/libopus/celt/arm/mdct_arm.h b/lib/rbcodec/codecs/libopus/celt/arm/mdct_arm.h
new file mode 100644
index 0000000000..14200bac4b
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/arm/mdct_arm.h
@@ -0,0 +1,59 @@
+/* Copyright (c) 2015 Xiph.Org Foundation
+ Written by Viswanath Puttagunta */
+/**
+ @file arm_mdct.h
+ @brief ARM Neon Intrinsic optimizations for mdct using NE10 library
+ */
+
+/*
+ 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(MDCT_ARM_H)
+#define MDCT_ARM_H
+
+#include "mdct.h"
+
+#if defined(HAVE_ARM_NE10)
+/** Compute a forward MDCT and scale by 4/N, trashes the input array */
+void clt_mdct_forward_neon(const mdct_lookup *l, kiss_fft_scalar *in,
+ kiss_fft_scalar * OPUS_RESTRICT out,
+ const opus_val16 *window, int overlap,
+ int shift, int stride, int arch);
+
+void clt_mdct_backward_neon(const mdct_lookup *l, kiss_fft_scalar *in,
+ kiss_fft_scalar * OPUS_RESTRICT out,
+ const opus_val16 *window, int overlap,
+ int shift, int stride, int arch);
+
+#if !defined(OPUS_HAVE_RTCD)
+#define OVERRIDE_OPUS_MDCT (1)
+#define clt_mdct_forward(_l, _in, _out, _window, _int, _shift, _stride, _arch) \
+ clt_mdct_forward_neon(_l, _in, _out, _window, _int, _shift, _stride, _arch)
+#define clt_mdct_backward(_l, _in, _out, _window, _int, _shift, _stride, _arch) \
+ clt_mdct_backward_neon(_l, _in, _out, _window, _int, _shift, _stride, _arch)
+#endif /* OPUS_HAVE_RTCD */
+#endif /* HAVE_ARM_NE10 */
+
+#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/arm/pitch_arm.h b/lib/rbcodec/codecs/libopus/celt/arm/pitch_arm.h
new file mode 100644
index 0000000000..bed8b04eac
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/arm/pitch_arm.h
@@ -0,0 +1,160 @@
+/* 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(OPUS_ARM_MAY_HAVE_NEON_INTR)
+opus_val32 celt_inner_prod_neon(const opus_val16 *x, const opus_val16 *y, int N);
+void dual_inner_prod_neon(const opus_val16 *x, const opus_val16 *y01,
+ const opus_val16 *y02, int N, opus_val32 *xy1, opus_val32 *xy2);
+
+# if !defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_PRESUME_NEON)
+# define OVERRIDE_CELT_INNER_PROD (1)
+# define OVERRIDE_DUAL_INNER_PROD (1)
+# define celt_inner_prod(x, y, N, arch) ((void)(arch), PRESUME_NEON(celt_inner_prod)(x, y, N))
+# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) ((void)(arch), PRESUME_NEON(dual_inner_prod)(x, y01, y02, N, xy1, xy2))
+# endif
+# endif
+
+# if !defined(OVERRIDE_CELT_INNER_PROD)
+# if defined(OPUS_HAVE_RTCD) && (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR))
+extern opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *x, const opus_val16 *y, int N);
+# define OVERRIDE_CELT_INNER_PROD (1)
+# define celt_inner_prod(x, y, N, arch) ((*CELT_INNER_PROD_IMPL[(arch)&OPUS_ARCHMASK])(x, y, N))
+# elif defined(OPUS_ARM_PRESUME_NEON_INTR)
+# define OVERRIDE_CELT_INNER_PROD (1)
+# define celt_inner_prod(x, y, N, arch) ((void)(arch), celt_inner_prod_neon(x, y, N))
+# endif
+# endif
+
+# if !defined(OVERRIDE_DUAL_INNER_PROD)
+# if defined(OPUS_HAVE_RTCD) && (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR))
+extern void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *x,
+ const opus_val16 *y01, const opus_val16 *y02, int N, opus_val32 *xy1, opus_val32 *xy2);
+# define OVERRIDE_DUAL_INNER_PROD (1)
+# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) ((*DUAL_INNER_PROD_IMPL[(arch)&OPUS_ARCHMASK])(x, y01, y02, N, xy1, xy2))
+# elif defined(OPUS_ARM_PRESUME_NEON_INTR)
+# define OVERRIDE_DUAL_INNER_PROD (1)
+# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) ((void)(arch), dual_inner_prod_neon(x, y01, y02, N, xy1, xy2))
+# endif
+# endif
+
+# 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, int arch);
+# 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, int arch);
+# endif
+
+# if defined(OPUS_HAVE_RTCD) && \
+ ((defined(OPUS_ARM_MAY_HAVE_NEON) && !defined(OPUS_ARM_PRESUME_NEON)) || \
+ (defined(OPUS_ARM_MAY_HAVE_MEDIA) && !defined(OPUS_ARM_PRESUME_MEDIA)) || \
+ (defined(OPUS_ARM_MAY_HAVE_EDSP) && !defined(OPUS_ARM_PRESUME_EDSP)))
+extern opus_val32
+(*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
+ const opus_val16 *, opus_val32 *, int, int, int);
+# define OVERRIDE_PITCH_XCORR (1)
+# define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \
+ ((*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK])(_x, _y, \
+ xcorr, len, max_pitch, arch))
+
+# elif defined(OPUS_ARM_PRESUME_EDSP) || \
+ defined(OPUS_ARM_PRESUME_MEDIA) || \
+ defined(OPUS_ARM_PRESUME_NEON)
+# define OVERRIDE_PITCH_XCORR (1)
+# define celt_pitch_xcorr (PRESUME_NEON(celt_pitch_xcorr))
+
+# endif
+
+# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+void xcorr_kernel_neon_fixed(
+ const opus_val16 *x,
+ const opus_val16 *y,
+ opus_val32 sum[4],
+ int len);
+# endif
+
+# if defined(OPUS_HAVE_RTCD) && \
+ (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR))
+
+extern void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])(
+ const opus_val16 *x,
+ const opus_val16 *y,
+ opus_val32 sum[4],
+ int len);
+
+# define OVERRIDE_XCORR_KERNEL (1)
+# define xcorr_kernel(x, y, sum, len, arch) \
+ ((*XCORR_KERNEL_IMPL[(arch) & OPUS_ARCHMASK])(x, y, sum, len))
+
+# elif defined(OPUS_ARM_PRESUME_NEON_INTR)
+# define OVERRIDE_XCORR_KERNEL (1)
+# define xcorr_kernel(x, y, sum, len, arch) \
+ ((void)arch, xcorr_kernel_neon_fixed(x, y, sum, len))
+
+# endif
+
+#else /* Start !FIXED_POINT */
+/* Float case */
+#if defined(OPUS_ARM_MAY_HAVE_NEON_INTR)
+void celt_pitch_xcorr_float_neon(const opus_val16 *_x, const opus_val16 *_y,
+ opus_val32 *xcorr, int len, int max_pitch, int arch);
+#endif
+
+# if defined(OPUS_HAVE_RTCD) && \
+ (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR))
+extern void
+(*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *,
+ const opus_val16 *, opus_val32 *, int, int, int);
+
+# define OVERRIDE_PITCH_XCORR (1)
+# define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \
+ ((*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK])(_x, _y, \
+ xcorr, len, max_pitch, arch))
+
+# elif defined(OPUS_ARM_PRESUME_NEON_INTR)
+
+# define OVERRIDE_PITCH_XCORR (1)
+# define celt_pitch_xcorr celt_pitch_xcorr_float_neon
+
+# endif
+
+#endif /* end !FIXED_POINT */
+
+#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/arm/pitch_neon_intr.c b/lib/rbcodec/codecs/libopus/celt/arm/pitch_neon_intr.c
new file mode 100644
index 0000000000..1ac38c433a
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/arm/pitch_neon_intr.c
@@ -0,0 +1,290 @@
+/***********************************************************************
+Copyright (c) 2017 Google Inc.
+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.
+***********************************************************************/
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include <arm_neon.h>
+#include "pitch.h"
+
+#ifdef FIXED_POINT
+
+opus_val32 celt_inner_prod_neon(const opus_val16 *x, const opus_val16 *y, int N)
+{
+ int i;
+ opus_val32 xy;
+ int16x8_t x_s16x8, y_s16x8;
+ int32x4_t xy_s32x4 = vdupq_n_s32(0);
+ int64x2_t xy_s64x2;
+ int64x1_t xy_s64x1;
+
+ for (i = 0; i < N - 7; i += 8) {
+ x_s16x8 = vld1q_s16(&x[i]);
+ y_s16x8 = vld1q_s16(&y[i]);
+ xy_s32x4 = vmlal_s16(xy_s32x4, vget_low_s16 (x_s16x8), vget_low_s16 (y_s16x8));
+ xy_s32x4 = vmlal_s16(xy_s32x4, vget_high_s16(x_s16x8), vget_high_s16(y_s16x8));
+ }
+
+ if (N - i >= 4) {
+ const int16x4_t x_s16x4 = vld1_s16(&x[i]);
+ const int16x4_t y_s16x4 = vld1_s16(&y[i]);
+ xy_s32x4 = vmlal_s16(xy_s32x4, x_s16x4, y_s16x4);
+ i += 4;
+ }
+
+ xy_s64x2 = vpaddlq_s32(xy_s32x4);
+ xy_s64x1 = vadd_s64(vget_low_s64(xy_s64x2), vget_high_s64(xy_s64x2));
+ xy = vget_lane_s32(vreinterpret_s32_s64(xy_s64x1), 0);
+
+ for (; i < N; i++) {
+ xy = MAC16_16(xy, x[i], y[i]);
+ }
+
+#ifdef OPUS_CHECK_ASM
+ celt_assert(celt_inner_prod_c(x, y, N) == xy);
+#endif
+
+ return xy;
+}
+
+void dual_inner_prod_neon(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, xy02;
+ int16x8_t x_s16x8, y01_s16x8, y02_s16x8;
+ int32x4_t xy01_s32x4 = vdupq_n_s32(0);
+ int32x4_t xy02_s32x4 = vdupq_n_s32(0);
+ int64x2_t xy01_s64x2, xy02_s64x2;
+ int64x1_t xy01_s64x1, xy02_s64x1;
+
+ for (i = 0; i < N - 7; i += 8) {
+ x_s16x8 = vld1q_s16(&x[i]);
+ y01_s16x8 = vld1q_s16(&y01[i]);
+ y02_s16x8 = vld1q_s16(&y02[i]);
+ xy01_s32x4 = vmlal_s16(xy01_s32x4, vget_low_s16 (x_s16x8), vget_low_s16 (y01_s16x8));
+ xy02_s32x4 = vmlal_s16(xy02_s32x4, vget_low_s16 (x_s16x8), vget_low_s16 (y02_s16x8));
+ xy01_s32x4 = vmlal_s16(xy01_s32x4, vget_high_s16(x_s16x8), vget_high_s16(y01_s16x8));
+ xy02_s32x4 = vmlal_s16(xy02_s32x4, vget_high_s16(x_s16x8), vget_high_s16(y02_s16x8));
+ }
+
+ if (N - i >= 4) {
+ const int16x4_t x_s16x4 = vld1_s16(&x[i]);
+ const int16x4_t y01_s16x4 = vld1_s16(&y01[i]);
+ const int16x4_t y02_s16x4 = vld1_s16(&y02[i]);
+ xy01_s32x4 = vmlal_s16(xy01_s32x4, x_s16x4, y01_s16x4);
+ xy02_s32x4 = vmlal_s16(xy02_s32x4, x_s16x4, y02_s16x4);
+ i += 4;
+ }
+
+ xy01_s64x2 = vpaddlq_s32(xy01_s32x4);
+ xy02_s64x2 = vpaddlq_s32(xy02_s32x4);
+ xy01_s64x1 = vadd_s64(vget_low_s64(xy01_s64x2), vget_high_s64(xy01_s64x2));
+ xy02_s64x1 = vadd_s64(vget_low_s64(xy02_s64x2), vget_high_s64(xy02_s64x2));
+ xy01 = vget_lane_s32(vreinterpret_s32_s64(xy01_s64x1), 0);
+ xy02 = vget_lane_s32(vreinterpret_s32_s64(xy02_s64x1), 0);
+
+ for (; i < N; i++) {
+ xy01 = MAC16_16(xy01, x[i], y01[i]);
+ xy02 = MAC16_16(xy02, x[i], y02[i]);
+ }
+ *xy1 = xy01;
+ *xy2 = xy02;
+
+#ifdef OPUS_CHECK_ASM
+ {
+ opus_val32 xy1_c, xy2_c;
+ dual_inner_prod_c(x, y01, y02, N, &xy1_c, &xy2_c);
+ celt_assert(xy1_c == *xy1);
+ celt_assert(xy2_c == *xy2);
+ }
+#endif
+}
+
+#else /* !FIXED_POINT */
+
+/* ========================================================================== */
+
+#ifdef OPUS_CHECK_ASM
+
+/* This part of code simulates floating-point NEON operations. */
+
+/* celt_inner_prod_neon_float_c_simulation() simulates the floating-point */
+/* operations of celt_inner_prod_neon(), and both functions should have bit */
+/* exact output. */
+static opus_val32 celt_inner_prod_neon_float_c_simulation(const opus_val16 *x, const opus_val16 *y, int N)
+{
+ int i;
+ opus_val32 xy, xy0 = 0, xy1 = 0, xy2 = 0, xy3 = 0;
+ for (i = 0; i < N - 3; i += 4) {
+ xy0 = MAC16_16(xy0, x[i + 0], y[i + 0]);
+ xy1 = MAC16_16(xy1, x[i + 1], y[i + 1]);
+ xy2 = MAC16_16(xy2, x[i + 2], y[i + 2]);
+ xy3 = MAC16_16(xy3, x[i + 3], y[i + 3]);
+ }
+ xy0 += xy2;
+ xy1 += xy3;
+ xy = xy0 + xy1;
+ for (; i < N; i++) {
+ xy = MAC16_16(xy, x[i], y[i]);
+ }
+ return xy;
+}
+
+/* dual_inner_prod_neon_float_c_simulation() simulates the floating-point */
+/* operations of dual_inner_prod_neon(), and both functions should have bit */
+/* exact output. */
+static void dual_inner_prod_neon_float_c_simulation(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, xy02, xy01_0 = 0, xy01_1 = 0, xy01_2 = 0, xy01_3 = 0, xy02_0 = 0, xy02_1 = 0, xy02_2 = 0, xy02_3 = 0;
+ for (i = 0; i < N - 3; i += 4) {
+ xy01_0 = MAC16_16(xy01_0, x[i + 0], y01[i + 0]);
+ xy01_1 = MAC16_16(xy01_1, x[i + 1], y01[i + 1]);
+ xy01_2 = MAC16_16(xy01_2, x[i + 2], y01[i + 2]);
+ xy01_3 = MAC16_16(xy01_3, x[i + 3], y01[i + 3]);
+ xy02_0 = MAC16_16(xy02_0, x[i + 0], y02[i + 0]);
+ xy02_1 = MAC16_16(xy02_1, x[i + 1], y02[i + 1]);
+ xy02_2 = MAC16_16(xy02_2, x[i + 2], y02[i + 2]);
+ xy02_3 = MAC16_16(xy02_3, x[i + 3], y02[i + 3]);
+ }
+ xy01_0 += xy01_2;
+ xy02_0 += xy02_2;
+ xy01_1 += xy01_3;
+ xy02_1 += xy02_3;
+ xy01 = xy01_0 + xy01_1;
+ xy02 = xy02_0 + xy02_1;
+ for (; i < N; i++) {
+ xy01 = MAC16_16(xy01, x[i], y01[i]);
+ xy02 = MAC16_16(xy02, x[i], y02[i]);
+ }
+ *xy1 = xy01;
+ *xy2 = xy02;
+}
+
+#endif /* OPUS_CHECK_ASM */
+
+/* ========================================================================== */
+
+opus_val32 celt_inner_prod_neon(const opus_val16 *x, const opus_val16 *y, int N)
+{
+ int i;
+ opus_val32 xy;
+ float32x4_t xy_f32x4 = vdupq_n_f32(0);
+ float32x2_t xy_f32x2;
+
+ for (i = 0; i < N - 7; i += 8) {
+ float32x4_t x_f32x4, y_f32x4;
+ x_f32x4 = vld1q_f32(&x[i]);
+ y_f32x4 = vld1q_f32(&y[i]);
+ xy_f32x4 = vmlaq_f32(xy_f32x4, x_f32x4, y_f32x4);
+ x_f32x4 = vld1q_f32(&x[i + 4]);
+ y_f32x4 = vld1q_f32(&y[i + 4]);
+ xy_f32x4 = vmlaq_f32(xy_f32x4, x_f32x4, y_f32x4);
+ }
+
+ if (N - i >= 4) {
+ const float32x4_t x_f32x4 = vld1q_f32(&x[i]);
+ const float32x4_t y_f32x4 = vld1q_f32(&y[i]);
+ xy_f32x4 = vmlaq_f32(xy_f32x4, x_f32x4, y_f32x4);
+ i += 4;
+ }
+
+ xy_f32x2 = vadd_f32(vget_low_f32(xy_f32x4), vget_high_f32(xy_f32x4));
+ xy_f32x2 = vpadd_f32(xy_f32x2, xy_f32x2);
+ xy = vget_lane_f32(xy_f32x2, 0);
+
+ for (; i < N; i++) {
+ xy = MAC16_16(xy, x[i], y[i]);
+ }
+
+#ifdef OPUS_CHECK_ASM
+ celt_assert(ABS32(celt_inner_prod_neon_float_c_simulation(x, y, N) - xy) <= VERY_SMALL);
+#endif
+
+ return xy;
+}
+
+void dual_inner_prod_neon(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, xy02;
+ float32x4_t xy01_f32x4 = vdupq_n_f32(0);
+ float32x4_t xy02_f32x4 = vdupq_n_f32(0);
+ float32x2_t xy01_f32x2, xy02_f32x2;
+
+ for (i = 0; i < N - 7; i += 8) {
+ float32x4_t x_f32x4, y01_f32x4, y02_f32x4;
+ x_f32x4 = vld1q_f32(&x[i]);
+ y01_f32x4 = vld1q_f32(&y01[i]);
+ y02_f32x4 = vld1q_f32(&y02[i]);
+ xy01_f32x4 = vmlaq_f32(xy01_f32x4, x_f32x4, y01_f32x4);
+ xy02_f32x4 = vmlaq_f32(xy02_f32x4, x_f32x4, y02_f32x4);
+ x_f32x4 = vld1q_f32(&x[i + 4]);
+ y01_f32x4 = vld1q_f32(&y01[i + 4]);
+ y02_f32x4 = vld1q_f32(&y02[i + 4]);
+ xy01_f32x4 = vmlaq_f32(xy01_f32x4, x_f32x4, y01_f32x4);
+ xy02_f32x4 = vmlaq_f32(xy02_f32x4, x_f32x4, y02_f32x4);
+ }
+
+ if (N - i >= 4) {
+ const float32x4_t x_f32x4 = vld1q_f32(&x[i]);
+ const float32x4_t y01_f32x4 = vld1q_f32(&y01[i]);
+ const float32x4_t y02_f32x4 = vld1q_f32(&y02[i]);
+ xy01_f32x4 = vmlaq_f32(xy01_f32x4, x_f32x4, y01_f32x4);
+ xy02_f32x4 = vmlaq_f32(xy02_f32x4, x_f32x4, y02_f32x4);
+ i += 4;
+ }
+
+ xy01_f32x2 = vadd_f32(vget_low_f32(xy01_f32x4), vget_high_f32(xy01_f32x4));
+ xy02_f32x2 = vadd_f32(vget_low_f32(xy02_f32x4), vget_high_f32(xy02_f32x4));
+ xy01_f32x2 = vpadd_f32(xy01_f32x2, xy01_f32x2);
+ xy02_f32x2 = vpadd_f32(xy02_f32x2, xy02_f32x2);
+ xy01 = vget_lane_f32(xy01_f32x2, 0);
+ xy02 = vget_lane_f32(xy02_f32x2, 0);
+
+ for (; i < N; i++) {
+ xy01 = MAC16_16(xy01, x[i], y01[i]);
+ xy02 = MAC16_16(xy02, x[i], y02[i]);
+ }
+ *xy1 = xy01;
+ *xy2 = xy02;
+
+#ifdef OPUS_CHECK_ASM
+ {
+ opus_val32 xy1_c, xy2_c;
+ dual_inner_prod_neon_float_c_simulation(x, y01, y02, N, &xy1_c, &xy2_c);
+ celt_assert(ABS32(xy1_c - *xy1) <= VERY_SMALL);
+ celt_assert(ABS32(xy2_c - *xy2) <= VERY_SMALL);
+ }
+#endif
+}
+
+#endif /* FIXED_POINT */
diff --git a/lib/rbcodec/codecs/libopus/celt/bands.c b/lib/rbcodec/codecs/libopus/celt/bands.c
index caa70163b4..2702963c37 100644
--- a/lib/rbcodec/codecs/libopus/celt/bands.c
+++ b/lib/rbcodec/codecs/libopus/celt/bands.c
@@ -65,19 +65,19 @@ opus_uint32 celt_lcg_rand(opus_uint32 seed)
/* This is a cos() approximation designed to be bit-exact on any platform. Bit exactness
with this approximation is important because it has an impact on the bit allocation */
-static opus_int16 bitexact_cos(opus_int16 x)
+opus_int16 bitexact_cos(opus_int16 x)
{
opus_int32 tmp;
opus_int16 x2;
tmp = (4096+((opus_int32)(x)*(x)))>>13;
- celt_assert(tmp<=32767);
+ celt_sig_assert(tmp<=32767);
x2 = tmp;
x2 = (32767-x2) + FRAC_MUL16(x2, (-7651 + FRAC_MUL16(x2, (8277 + FRAC_MUL16(-626, x2)))));
- celt_assert(x2<=32766);
+ celt_sig_assert(x2<=32766);
return 1+x2;
}
-static int bitexact_log2tan(int isin,int icos)
+int bitexact_log2tan(int isin,int icos)
{
int lc;
int ls;
@@ -90,13 +90,13 @@ static int bitexact_log2tan(int isin,int icos)
-FRAC_MUL16(icos, FRAC_MUL16(icos, -2597) + 7932);
}
-#if 0
#ifdef FIXED_POINT
/* Compute the amplitude (sqrt energy) in each of the bands */
-void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM)
+void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM, int arch)
{
int i, c, N;
const opus_int16 *eBands = m->eBands;
+ (void)arch;
N = m->shortMdctSize<<LM;
c=0; do {
for (i=0;i<end;i++)
@@ -156,7 +156,7 @@ void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, cel
#else /* FIXED_POINT */
/* Compute the amplitude (sqrt energy) in each of the bands */
-void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM)
+void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM, int arch)
{
int i, c, N;
const opus_int16 *eBands = m->eBands;
@@ -165,7 +165,7 @@ void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *band
for (i=0;i<end;i++)
{
opus_val32 sum;
- sum = 1e-27f + celt_inner_prod(&X[c*N+(eBands[i]<<LM)], &X[c*N+(eBands[i]<<LM)], (eBands[i+1]-eBands[i])<<LM);
+ sum = 1e-27f + celt_inner_prod(&X[c*N+(eBands[i]<<LM)], &X[c*N+(eBands[i]<<LM)], (eBands[i+1]-eBands[i])<<LM, arch);
bandE[i+c*m->nbEBands] = celt_sqrt(sum);
/*printf ("%f ", bandE[i+c*m->nbEBands]);*/
}
@@ -191,7 +191,6 @@ void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, cel
}
#endif /* FIXED_POINT */
-#endif
/* 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,
@@ -226,9 +225,9 @@ void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X,
#endif
j=M*eBands[i];
band_end = M*eBands[i+1];
- lg = ADD16(bandLogE[i], SHL16((opus_val16)eMeans[i],6));
+ lg = SATURATE16(ADD32(bandLogE[i], SHL32((opus_val32)eMeans[i],6)));
#ifndef FIXED_POINT
- g = celt_exp2(lg);
+ g = celt_exp2(MIN32(32.f, lg));
#else
/* Handle the integer part of the log energy */
shift = 16-(lg>>DB_SHIFT);
@@ -243,12 +242,12 @@ void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X,
/* 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)
+ /* For shift <= -2 and g > 16384 we'd be likely to overflow, so we're
+ capping the gain here, which is equivalent to a cap of 18 on lg.
+ This shouldn't trigger unless the bitstream is already corrupted. */
+ if (shift <= -2)
{
- g = 32767;
+ g = 16384;
shift = -2;
}
do {
@@ -268,7 +267,7 @@ void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X,
/* This prevents energy collapse for transients with multiple short MDCTs */
void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_masks, int LM, int C, int size,
int start, int end, const opus_val16 *logE, const opus_val16 *prev1logE,
- const opus_val16 *prev2logE, const int *pulses, opus_uint32 seed)
+ const opus_val16 *prev2logE, const int *pulses, opus_uint32 seed, int arch)
{
int c, i, j, k;
for (i=start;i<end;i++)
@@ -283,7 +282,7 @@ void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_mas
N0 = m->eBands[i+1]-m->eBands[i];
/* depth in 1/8 bits */
- celt_assert(pulses[i]>=0);
+ celt_sig_assert(pulses[i]>=0);
depth = celt_udiv(1+pulses[i], (m->eBands[i+1]-m->eBands[i]))>>LM;
#ifdef FIXED_POINT
@@ -357,11 +356,35 @@ void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_mas
}
/* We just added some energy, so we need to renormalise */
if (renormalize)
- renormalise_vector(X, N0<<LM, Q15ONE);
+ renormalise_vector(X, N0<<LM, Q15ONE, arch);
} while (++c<C);
}
}
+/* Compute the weights to use for optimizing normalized distortion across
+ channels. We use the amplitude to weight square distortion, which means
+ that we use the square root of the value we would have been using if we
+ wanted to minimize the MSE in the non-normalized domain. This roughly
+ corresponds to some quick-and-dirty perceptual experiments I ran to
+ measure inter-aural masking (there doesn't seem to be any published data
+ on the topic). */
+static void compute_channel_weights(celt_ener Ex, celt_ener Ey, opus_val16 w[2])
+{
+ celt_ener minE;
+#ifdef FIXED_POINT
+ int shift;
+#endif
+ minE = MIN32(Ex, Ey);
+ /* Adjustment to make the weights a bit more conservative. */
+ Ex = ADD32(Ex, minE/3);
+ Ey = ADD32(Ey, minE/3);
+#ifdef FIXED_POINT
+ shift = celt_ilog2(EPSILON+MAX32(Ex, Ey))-14;
+#endif
+ w[0] = VSHR32(Ex, shift);
+ w[1] = VSHR32(Ey, shift);
+}
+
static void intensity_stereo(const CELTMode *m, celt_norm * OPUS_RESTRICT X, const celt_norm * OPUS_RESTRICT Y, const celt_ener *bandE, int bandID, int N)
{
int i = bandID;
@@ -400,7 +423,7 @@ static void stereo_split(celt_norm * OPUS_RESTRICT X, celt_norm * OPUS_RESTRICT
}
}
-static void stereo_merge(celt_norm * OPUS_RESTRICT X, celt_norm * OPUS_RESTRICT Y, opus_val16 mid, int N)
+static void stereo_merge(celt_norm * OPUS_RESTRICT X, celt_norm * OPUS_RESTRICT Y, opus_val16 mid, int N, int arch)
{
int j;
opus_val32 xp=0, side=0;
@@ -412,11 +435,11 @@ static void stereo_merge(celt_norm * OPUS_RESTRICT X, celt_norm * OPUS_RESTRICT
opus_val32 t, lgain, rgain;
/* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */
- dual_inner_prod(Y, X, Y, N, &xp, &side);
+ dual_inner_prod(Y, X, Y, N, &xp, &side, arch);
/* Compensating for the mid normalization */
xp = MULT16_32_Q15(mid, xp);
/* mid and side are in Q15, not Q14 like X and Y */
- mid2 = SHR32(mid, 1);
+ mid2 = SHR16(mid, 1);
El = MULT16_16(mid2, mid2) + side - 2*xp;
Er = MULT16_16(mid2, mid2) + side + 2*xp;
if (Er < QCONST32(6e-4f, 28) || El < QCONST32(6e-4f, 28))
@@ -452,11 +475,10 @@ static void stereo_merge(celt_norm * OPUS_RESTRICT X, celt_norm * OPUS_RESTRICT
}
}
-#if 0
/* Decide whether we should spread the pulses in the current frame */
int spreading_decision(const CELTMode *m, const celt_norm *X, int *average,
int last_decision, int *hf_average, int *tapset_decision, int update_hf,
- int end, int C, int M)
+ int end, int C, int M, const int *spread_weight)
{
int i, c, N0;
int sum = 0, nbBands=0;
@@ -497,8 +519,8 @@ int spreading_decision(const CELTMode *m, const celt_norm *X, int *average,
if (i>m->nbEBands-4)
hf_sum += celt_udiv(32*(tcount[1]+tcount[0]), N);
tmp = (2*tcount[2] >= N) + (2*tcount[1] >= N) + (2*tcount[0] >= N);
- sum += tmp*256;
- nbBands++;
+ sum += tmp*spread_weight[i];
+ nbBands+=spread_weight[i];
}
} while (++c<C);
@@ -522,7 +544,7 @@ int spreading_decision(const CELTMode *m, const celt_norm *X, int *average,
/*printf("%d %d %d\n", hf_sum, *hf_average, *tapset_decision);*/
celt_assert(nbBands>0); /* end has to be non-zero */
celt_assert(sum>=0);
- sum = celt_udiv(sum, nbBands);
+ sum = celt_udiv((opus_int32)sum<<8, nbBands);
/* Recursive averaging */
sum = (sum+*average)>>1;
*average = sum;
@@ -546,7 +568,6 @@ int spreading_decision(const CELTMode *m, const celt_norm *X, int *average,
#endif
return decision;
}
-#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
@@ -651,6 +672,7 @@ static int compute_qn(int N, int b, int offset, int pulse_cap, int stereo)
struct band_ctx {
int encode;
+ int resynth;
const CELTMode *m;
int i;
int intensity;
@@ -660,6 +682,10 @@ struct band_ctx {
opus_int32 remaining_bits;
const celt_ener *bandE;
opus_uint32 seed;
+ int arch;
+ int theta_round;
+ int disable_inv;
+ int avoid_split_noise;
};
struct split_ctx {
@@ -711,14 +737,41 @@ static void compute_theta(struct band_ctx *ctx, struct split_ctx *sctx,
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);
+ itheta = stereo_itheta(X, Y, stereo, N, ctx->arch);
}
tell = ec_tell_frac(ec);
if (qn!=1)
{
if (encode)
- itheta = (itheta*qn+8192)>>14;
-
+ {
+ if (!stereo || ctx->theta_round == 0)
+ {
+ itheta = (itheta*(opus_int32)qn+8192)>>14;
+ if (!stereo && ctx->avoid_split_noise && itheta > 0 && itheta < qn)
+ {
+ /* Check if the selected value of theta will cause the bit allocation
+ to inject noise on one side. If so, make sure the energy of that side
+ is zero. */
+ int unquantized = celt_udiv((opus_int32)itheta*16384, qn);
+ imid = bitexact_cos((opus_int16)unquantized);
+ iside = bitexact_cos((opus_int16)(16384-unquantized));
+ delta = FRAC_MUL16((N-1)<<7,bitexact_log2tan(iside,imid));
+ if (delta > *b)
+ itheta = qn;
+ else if (delta < -*b)
+ itheta = 0;
+ }
+ } else {
+ int down;
+ /* Bias quantization towards itheta=0 and itheta=16384. */
+ int bias = itheta > 8192 ? 32767/qn : -32767/qn;
+ down = IMIN(qn-1, IMAX(0, (itheta*(opus_int32)qn + bias)>>14));
+ if (ctx->theta_round < 0)
+ itheta = down;
+ else
+ itheta = down+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)
@@ -796,7 +849,7 @@ static void compute_theta(struct band_ctx *ctx, struct split_ctx *sctx,
} else if (stereo) {
if (encode)
{
- inv = itheta > 8192;
+ inv = itheta > 8192 && !ctx->disable_inv;
if (inv)
{
int j;
@@ -813,6 +866,9 @@ static void compute_theta(struct band_ctx *ctx, struct split_ctx *sctx,
inv = ec_dec_bit_logp(ec, 2);
} else
inv = 0;
+ /* inv flag override to avoid problems with downmixing. */
+ if (ctx->disable_inv)
+ inv = 0;
itheta = 0;
}
qalloc = ec_tell_frac(ec) - tell;
@@ -848,11 +904,6 @@ static void compute_theta(struct band_ctx *ctx, struct split_ctx *sctx,
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;
@@ -877,7 +928,7 @@ static unsigned quant_band_n1(struct band_ctx *ctx, celt_norm *X, celt_norm *Y,
ctx->remaining_bits -= 1<<BITRES;
b-=1<<BITRES;
}
- if (resynth)
+ if (ctx->resynth)
x[0] = sign ? -NORM_SCALING : NORM_SCALING;
x = Y;
} while (++c<1+stereo);
@@ -902,11 +953,6 @@ static unsigned quant_partition(struct band_ctx *ctx, celt_norm *X,
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;
@@ -938,8 +984,7 @@ static unsigned quant_partition(struct band_ctx *ctx, celt_norm *X,
fill = (fill&1)|(fill<<1);
B = (B+1)>>1;
- compute_theta(ctx, &sctx, X, Y, N, &b, B, B0,
- LM, 0, &fill);
+ compute_theta(ctx, &sctx, X, Y, N, &b, B, B0, LM, 0, &fill);
imid = sctx.imid;
iside = sctx.iside;
delta = sctx.delta;
@@ -973,24 +1018,20 @@ static unsigned quant_partition(struct band_ctx *ctx, celt_norm *X,
rebalance = ctx->remaining_bits;
if (mbits >= sbits)
{
- cm = quant_partition(ctx, X, N, mbits, B,
- lowband, LM,
+ 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,
+ 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,
+ 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,
+ cm |= quant_partition(ctx, X, N, mbits, B, lowband, LM,
MULT16_16_P15(gain,mid), fill);
}
} else {
@@ -1015,18 +1056,14 @@ static unsigned quant_partition(struct band_ctx *ctx, celt_norm *X,
/* Finally do the actual quantization */
if (encode)
{
- cm = alg_quant(X, N, K, spread, B, ec
-#ifdef RESYNTH
- , gain
-#endif
- );
+ cm = alg_quant(X, N, K, spread, B, ec, gain, ctx->resynth, ctx->arch);
} else {
cm = alg_unquant(X, N, K, spread, B, ec, gain);
}
} else {
/* If there's no pulse, fill the band anyway */
int j;
- if (resynth)
+ if (ctx->resynth)
{
unsigned cm_mask;
/* B can be as large as 16, so this shift might overflow an int on a
@@ -1059,7 +1096,7 @@ static unsigned quant_partition(struct band_ctx *ctx, celt_norm *X,
}
cm = fill;
}
- renormalise_vector(X, N, gain);
+ renormalise_vector(X, N, gain, ctx->arch);
}
}
}
@@ -1083,11 +1120,6 @@ static unsigned quant_band(struct band_ctx *ctx, celt_norm *X,
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;
@@ -1154,11 +1186,10 @@ static unsigned quant_band(struct band_ctx *ctx, celt_norm *X,
deinterleave_hadamard(lowband, N_B>>recombine, B0<<recombine, longBlocks);
}
- cm = quant_partition(ctx, X, N, b, B, lowband,
- LM, gain, fill);
+ 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 (ctx->resynth)
{
/* Undo the sample reorganization going from time order to frequency order */
if (B0>1)
@@ -1211,11 +1242,6 @@ static unsigned quant_band_stereo(struct band_ctx *ctx, celt_norm *X, celt_norm
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;
@@ -1235,8 +1261,7 @@ static unsigned quant_band_stereo(struct band_ctx *ctx, celt_norm *X, celt_norm
orig_fill = fill;
- compute_theta(ctx, &sctx, X, Y, N, &b, B, B,
- LM, 1, &fill);
+ compute_theta(ctx, &sctx, X, Y, N, &b, B, B, LM, 1, &fill);
inv = sctx.inv;
imid = sctx.imid;
iside = sctx.iside;
@@ -1284,13 +1309,13 @@ static unsigned quant_band_stereo(struct band_ctx *ctx, celt_norm *X, celt_norm
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);
+ 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)
+ if (ctx->resynth)
{
celt_norm tmp;
X[0] = MULT16_16_Q15(mid, X[0]);
@@ -1317,41 +1342,35 @@ static unsigned quant_band_stereo(struct band_ctx *ctx, celt_norm *X, celt_norm
{
/* 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);
+ 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);
+ 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);
+ 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);
+ 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 (ctx->resynth)
{
if (N!=2)
- stereo_merge(X, Y, mid, N);
+ stereo_merge(X, Y, mid, N, ctx->arch);
if (inv)
{
int j;
@@ -1362,17 +1381,38 @@ static unsigned quant_band_stereo(struct band_ctx *ctx, celt_norm *X, celt_norm
return cm;
}
+static void special_hybrid_folding(const CELTMode *m, celt_norm *norm, celt_norm *norm2, int start, int M, int dual_stereo)
+{
+ int n1, n2;
+ const opus_int16 * OPUS_RESTRICT eBands = m->eBands;
+ n1 = M*(eBands[start+1]-eBands[start]);
+ n2 = M*(eBands[start+2]-eBands[start+1]);
+ /* Duplicate enough of the first band folding data to be able to fold the second band.
+ Copies no data for CELT-only mode. */
+ OPUS_COPY(&norm[n1], &norm[2*n1 - n2], n2-n1);
+ if (dual_stereo)
+ OPUS_COPY(&norm2[n1], &norm2[2*n1 - n2], n2-n1);
+}
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,
- opus_int32 total_bits, opus_int32 balance, ec_ctx *ec, int LM, int codedBands, opus_uint32 *seed)
+ 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, opus_int32 total_bits,
+ opus_int32 balance, ec_ctx *ec, int LM, int codedBands,
+ opus_uint32 *seed, int complexity, int arch, int disable_inv)
{
int i;
opus_int32 remaining_bits;
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);
+ VARDECL(celt_norm, X_save);
+ VARDECL(celt_norm, Y_save);
+ VARDECL(celt_norm, X_save2);
+ VARDECL(celt_norm, Y_save2);
+ VARDECL(celt_norm, norm_save2);
+ int resynth_alloc;
celt_norm *lowband_scratch;
int B;
int M;
@@ -1380,10 +1420,11 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end,
int update_lowband = 1;
int C = Y_ != NULL ? 2 : 1;
int norm_offset;
+ int theta_rdo = encode && Y_!=NULL && !dual_stereo && complexity>=8;
#ifdef RESYNTH
int resynth = 1;
#else
- int resynth = !encode;
+ int resynth = !encode || theta_rdo;
#endif
struct band_ctx ctx;
SAVE_STACK;
@@ -1396,9 +1437,24 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end,
ALLOC(_norm, C*(M*eBands[m->nbEBands-1]-norm_offset), celt_norm);
norm = _norm;
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];
+
+ /* For decoding, we can use the last band as scratch space because we don't need that
+ scratch space for the last band and we don't care about the data there until we're
+ decoding the last band. */
+ if (encode && resynth)
+ resynth_alloc = M*(eBands[m->nbEBands]-eBands[m->nbEBands-1]);
+ else
+ resynth_alloc = ALLOC_NONE;
+ ALLOC(_lowband_scratch, resynth_alloc, celt_norm);
+ if (encode && resynth)
+ lowband_scratch = _lowband_scratch;
+ else
+ lowband_scratch = X_+M*eBands[m->nbEBands-1];
+ ALLOC(X_save, resynth_alloc, celt_norm);
+ ALLOC(Y_save, resynth_alloc, celt_norm);
+ ALLOC(X_save2, resynth_alloc, celt_norm);
+ ALLOC(Y_save2, resynth_alloc, celt_norm);
+ ALLOC(norm_save2, resynth_alloc, celt_norm);
lowband_offset = 0;
ctx.bandE = bandE;
@@ -1408,6 +1464,12 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end,
ctx.m = m;
ctx.seed = *seed;
ctx.spread = spread;
+ ctx.arch = arch;
+ ctx.disable_inv = disable_inv;
+ ctx.resynth = resynth;
+ ctx.theta_round = 0;
+ /* Avoid injecting noise in the first band on transients. */
+ ctx.avoid_split_noise = B > 1;
for (i=start;i<end;i++)
{
opus_int32 tell;
@@ -1430,6 +1492,7 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end,
else
Y = NULL;
N = M*eBands[i+1]-M*eBands[i];
+ celt_assert(N > 0);
tell = ec_tell_frac(ec);
/* Compute how many bits we want to allocate to this band */
@@ -1445,8 +1508,15 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end,
b = 0;
}
+#ifndef DISABLE_UPDATE_DRAFT
+ if (resynth && (M*eBands[i]-N >= M*eBands[start] || i==start+1) && (update_lowband || lowband_offset==0))
+ lowband_offset = i;
+ if (i == start+1)
+ special_hybrid_folding(m, norm, norm2, start, M, dual_stereo);
+#else
if (resynth && M*eBands[i]-N >= M*eBands[start] && (update_lowband || lowband_offset==0))
lowband_offset = i;
+#endif
tf_change = tf_res[i];
ctx.tf_change = tf_change;
@@ -1457,7 +1527,7 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end,
Y = norm;
lowband_scratch = NULL;
}
- if (i==end-1)
+ if (last && !theta_rdo)
lowband_scratch = NULL;
/* Get a conservative estimate of the collapse_mask's for the bands we're
@@ -1472,7 +1542,11 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end,
fold_start = lowband_offset;
while(M*eBands[--fold_start] > effective_lowband+norm_offset);
fold_end = lowband_offset-1;
+#ifndef DISABLE_UPDATE_DRAFT
+ while(++fold_end < i && M*eBands[fold_end] < effective_lowband+norm_offset+N);
+#else
while(M*eBands[++fold_end] < effective_lowband+norm_offset+N);
+#endif
x_cm = y_cm = 0;
fold_i = fold_start; do {
x_cm |= collapse_masks[fold_i*C+0];
@@ -1505,13 +1579,79 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end,
} else {
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);
+ if (theta_rdo && i < intensity)
+ {
+ ec_ctx ec_save, ec_save2;
+ struct band_ctx ctx_save, ctx_save2;
+ opus_val32 dist0, dist1;
+ unsigned cm, cm2;
+ int nstart_bytes, nend_bytes, save_bytes;
+ unsigned char *bytes_buf;
+ unsigned char bytes_save[1275];
+ opus_val16 w[2];
+ compute_channel_weights(bandE[i], bandE[i+m->nbEBands], w);
+ /* Make a copy. */
+ cm = x_cm|y_cm;
+ ec_save = *ec;
+ ctx_save = ctx;
+ OPUS_COPY(X_save, X, N);
+ OPUS_COPY(Y_save, Y, N);
+ /* Encode and round down. */
+ ctx.theta_round = -1;
+ 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, cm);
+ dist0 = MULT16_32_Q15(w[0], celt_inner_prod(X_save, X, N, arch)) + MULT16_32_Q15(w[1], celt_inner_prod(Y_save, Y, N, arch));
+
+ /* Save first result. */
+ cm2 = x_cm;
+ ec_save2 = *ec;
+ ctx_save2 = ctx;
+ OPUS_COPY(X_save2, X, N);
+ OPUS_COPY(Y_save2, Y, N);
+ if (!last)
+ OPUS_COPY(norm_save2, norm+M*eBands[i]-norm_offset, N);
+ nstart_bytes = ec_save.offs;
+ nend_bytes = ec_save.storage;
+ bytes_buf = ec_save.buf+nstart_bytes;
+ save_bytes = nend_bytes-nstart_bytes;
+ OPUS_COPY(bytes_save, bytes_buf, save_bytes);
+
+ /* Restore */
+ *ec = ec_save;
+ ctx = ctx_save;
+ OPUS_COPY(X, X_save, N);
+ OPUS_COPY(Y, Y_save, N);
+#ifndef DISABLE_UPDATE_DRAFT
+ if (i == start+1)
+ special_hybrid_folding(m, norm, norm2, start, M, dual_stereo);
+#endif
+ /* Encode and round up. */
+ ctx.theta_round = 1;
+ 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, cm);
+ dist1 = MULT16_32_Q15(w[0], celt_inner_prod(X_save, X, N, arch)) + MULT16_32_Q15(w[1], celt_inner_prod(Y_save, Y, N, arch));
+ if (dist0 >= dist1) {
+ x_cm = cm2;
+ *ec = ec_save2;
+ ctx = ctx_save2;
+ OPUS_COPY(X, X_save2, N);
+ OPUS_COPY(Y, Y_save2, N);
+ if (!last)
+ OPUS_COPY(norm+M*eBands[i]-norm_offset, norm_save2, N);
+ OPUS_COPY(bytes_buf, bytes_save, save_bytes);
+ }
+ } else {
+ ctx.theta_round = 0;
+ 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);
+ last?NULL:norm+M*eBands[i]-norm_offset, Q15ONE, lowband_scratch, x_cm|y_cm);
}
y_cm = x_cm;
}
@@ -1521,6 +1661,9 @@ void quant_all_bands(int encode, const CELTMode *m, int start, int end,
/* Update the folding position only as long as we have 1 bit/sample depth. */
update_lowband = b>(N<<BITRES);
+ /* We only need to avoid noise on a split for the first band. After that, we
+ have folding. */
+ ctx.avoid_split_noise = 0;
}
*seed = ctx.seed;
diff --git a/lib/rbcodec/codecs/libopus/celt/bands.h b/lib/rbcodec/codecs/libopus/celt/bands.h
index 69901b1e33..422b32cf75 100644
--- a/lib/rbcodec/codecs/libopus/celt/bands.h
+++ b/lib/rbcodec/codecs/libopus/celt/bands.h
@@ -36,12 +36,15 @@
#include "entdec.h"
#include "rate.h"
+opus_int16 bitexact_cos(opus_int16 x);
+int bitexact_log2tan(int isin,int icos);
+
/** Compute the amplitude (sqrt energy) in each of the bands
* @param m Mode data
* @param X Spectrum
* @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 LM);
+void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM, int arch);
/*void compute_noise_energies(const CELTMode *m, const celt_sig *X, const opus_val16 *tonality, celt_ener *bandE);*/
@@ -69,7 +72,7 @@ void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X,
int spreading_decision(const CELTMode *m, const celt_norm *X, int *average,
int last_decision, int *hf_average, int *tapset_decision, int update_hf,
- int end, int C, int M);
+ int end, int C, int M, const int *spread_weight);
#ifdef MEASURE_NORM_MSE
void measure_norm_mse(const CELTMode *m, float *X, float *X0, float *bandE, float *bandE0, int M, int N, int C);
@@ -98,15 +101,20 @@ void haar1(celt_norm *X, int N0, int stride);
* @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
+ * @param arch Run-time architecture (see opus_select_arch())
*/
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,
- 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,
- int start, int end, const opus_val16 *logE, const opus_val16 *prev1logE,
- const opus_val16 *prev2logE, const int *pulses, opus_uint32 seed);
+ 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, opus_int32 total_bits,
+ opus_int32 balance, ec_ctx *ec, int M, int codedBands, opus_uint32 *seed,
+ int complexity, int arch, int disable_inv);
+
+void anti_collapse(const CELTMode *m, celt_norm *X_,
+ unsigned char *collapse_masks, int LM, int C, int size, int start,
+ int end, const opus_val16 *logE, const opus_val16 *prev1logE,
+ const opus_val16 *prev2logE, const int *pulses, opus_uint32 seed,
+ int arch);
opus_uint32 celt_lcg_rand(opus_uint32 seed);
diff --git a/lib/rbcodec/codecs/libopus/celt/celt.c b/lib/rbcodec/codecs/libopus/celt/celt.c
index c0a1e0dab9..9ce234695c 100644
--- a/lib/rbcodec/codecs/libopus/celt/celt.c
+++ b/lib/rbcodec/codecs/libopus/celt/celt.c
@@ -89,10 +89,13 @@ int resampling_factor(opus_int32 rate)
return ret;
}
-#ifndef OVERRIDE_COMB_FILTER_CONST
+#if !defined(OVERRIDE_COMB_FILTER_CONST) || defined(NON_STATIC_COMB_FILTER_CONST_C)
/* This version should be faster on ARM */
#ifdef OPUS_ARM_ASM
-static void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N,
+#ifndef NON_STATIC_COMB_FILTER_CONST_C
+static
+#endif
+void comb_filter_const_c(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;
@@ -108,26 +111,31 @@ static void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N,
t = MAC16_32_Q16(x[i], g10, x2);
t = MAC16_32_Q16(t, g11, ADD32(x1,x3));
t = MAC16_32_Q16(t, g12, ADD32(x0,x4));
+ t = SATURATE(t, SIG_SAT);
y[i] = t;
x4=SHL32(x[i-T+3],1);
t = MAC16_32_Q16(x[i+1], g10, x1);
t = MAC16_32_Q16(t, g11, ADD32(x0,x2));
t = MAC16_32_Q16(t, g12, ADD32(x4,x3));
+ t = SATURATE(t, SIG_SAT);
y[i+1] = t;
x3=SHL32(x[i-T+4],1);
t = MAC16_32_Q16(x[i+2], g10, x0);
t = MAC16_32_Q16(t, g11, ADD32(x4,x1));
t = MAC16_32_Q16(t, g12, ADD32(x3,x2));
+ t = SATURATE(t, SIG_SAT);
y[i+2] = t;
x2=SHL32(x[i-T+5],1);
t = MAC16_32_Q16(x[i+3], g10, x4);
t = MAC16_32_Q16(t, g11, ADD32(x3,x0));
t = MAC16_32_Q16(t, g12, ADD32(x2,x1));
+ t = SATURATE(t, SIG_SAT);
y[i+3] = t;
x1=SHL32(x[i-T+6],1);
t = MAC16_32_Q16(x[i+4], g10, x3);
t = MAC16_32_Q16(t, g11, ADD32(x2,x4));
t = MAC16_32_Q16(t, g12, ADD32(x1,x0));
+ t = SATURATE(t, SIG_SAT);
y[i+4] = t;
}
#ifdef CUSTOM_MODES
@@ -138,6 +146,7 @@ static void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N,
t = MAC16_32_Q16(x[i], g10, x2);
t = MAC16_32_Q16(t, g11, ADD32(x1,x3));
t = MAC16_32_Q16(t, g12, ADD32(x0,x4));
+ t = SATURATE(t, SIG_SAT);
y[i] = t;
x4=x3;
x3=x2;
@@ -147,7 +156,10 @@ static void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N,
#endif
}
#else
-static void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N,
+#ifndef NON_STATIC_COMB_FILTER_CONST_C
+static
+#endif
+void comb_filter_const_c(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;
@@ -163,6 +175,7 @@ static void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N,
+ MULT16_32_Q15(g10,x2)
+ MULT16_32_Q15(g11,ADD32(x1,x3))
+ MULT16_32_Q15(g12,ADD32(x0,x4));
+ y[i] = SATURATE(y[i], SIG_SAT);
x4=x3;
x3=x2;
x2=x1;
@@ -176,7 +189,7 @@ static void comb_filter_const(opus_val32 *y, opus_val32 *x, int T, int N,
#ifndef OVERRIDE_comb_filter
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)
+ const opus_val16 *window, int overlap, int arch)
{
int i;
/* printf ("%d %d %f %f\n", T0, T1, g0, g1); */
@@ -194,6 +207,10 @@ void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N,
OPUS_MOVE(y, x, N);
return;
}
+ /* When the gain is zero, T0 and/or T1 is set to zero. We need
+ to have then be at least 2 to avoid processing garbage data. */
+ T0 = IMAX(T0, COMBFILTER_MINPERIOD);
+ T1 = IMAX(T1, COMBFILTER_MINPERIOD);
g00 = MULT16_16_P15(g0, gains[tapset0][0]);
g01 = MULT16_16_P15(g0, gains[tapset0][1]);
g02 = MULT16_16_P15(g0, gains[tapset0][2]);
@@ -219,6 +236,7 @@ void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N,
+ 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));
+ y[i] = SATURATE(y[i], SIG_SAT);
x4=x3;
x3=x2;
x2=x1;
@@ -234,15 +252,20 @@ void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N,
}
/* Compute the part with the constant filter. */
- comb_filter_const(y+i, x+i, T1, N-i, g10, g11, g12);
+ comb_filter_const(y+i, x+i, T1, N-i, g10, g11, g12, arch);
}
#endif /* OVERRIDE_comb_filter */
+/* TF change table. Positive values mean better frequency resolution (longer
+ effective window), whereas negative values mean better time resolution
+ (shorter effective window). The second index is computed as:
+ 4*isTransient + 2*tf_select + per_band_flag */
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},
+ /*isTransient=0 isTransient=1 */
+ {0, -1, 0, -1, 0,-1, 0,-1}, /* 2.5 ms */
+ {0, -1, 0, -2, 1, 0, 1,-1}, /* 5 ms */
+ {0, -2, 0, -3, 2, 0, 1,-1}, /* 10 ms */
+ {0, -2, 0, -3, 3, 0, 1,-1}, /* 20 ms */
};
@@ -280,6 +303,9 @@ const char *opus_strerror(int error)
const char *opus_get_version_string(void)
{
return "libopus " PACKAGE_VERSION
+ /* Applications may rely on the presence of this substring in the version
+ string to determine if they have a fixed-point or floating-point build
+ at runtime. */
#ifdef FIXED_POINT
"-fixed"
#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/celt.h b/lib/rbcodec/codecs/libopus/celt/celt.h
index b1967516dc..24b6b2b520 100644
--- a/lib/rbcodec/codecs/libopus/celt/celt.h
+++ b/lib/rbcodec/codecs/libopus/celt/celt.h
@@ -50,6 +50,8 @@ extern "C" {
#define CELTDecoder OpusCustomDecoder
#define CELTMode OpusCustomMode
+#define LEAK_BANDS 19
+
typedef struct {
int valid;
float tonality;
@@ -57,17 +59,27 @@ typedef struct {
float noisiness;
float activity;
float music_prob;
- int bandwidth;
-}AnalysisInfo;
+ float music_prob_min;
+ float music_prob_max;
+ int bandwidth;
+ float activity_probability;
+ float max_pitch_ratio;
+ /* Store as Q6 char to save space. */
+ unsigned char leak_boost[LEAK_BANDS];
+} AnalysisInfo;
+
+typedef struct {
+ int signalType;
+ int offset;
+} SILKInfo;
#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 */
+#define __celt_check_silkinfo_ptr(ptr) ((ptr) + ((ptr) - (const SILKInfo*)(ptr)))
-/* Expose this option again when variable framesize actually works */
-#define OPUS_FRAMESIZE_VARIABLE 5010 /**< Optimize the frame size dynamically */
+/* Encoder/decoder Requests */
#define CELT_SET_PREDICTION_REQUEST 10002
@@ -116,6 +128,9 @@ typedef struct {
#define OPUS_SET_ENERGY_MASK_REQUEST 10026
#define OPUS_SET_ENERGY_MASK(x) OPUS_SET_ENERGY_MASK_REQUEST, __opus_check_val16_ptr(x)
+#define CELT_SET_SILK_INFO_REQUEST 10028
+#define CELT_SET_SILK_INFO(x) CELT_SET_SILK_INFO_REQUEST, __celt_check_silkinfo_ptr(x)
+
/* Encoder stuff */
int celt_encoder_get_size(int channels);
@@ -194,6 +209,13 @@ static OPUS_INLINE int fromOpus(unsigned char c)
extern const signed char tf_select_table[4][8];
+#if defined(ENABLE_HARDENING) || defined(ENABLE_ASSERTIONS)
+void validate_celt_decoder(CELTDecoder *st);
+#define VALIDATE_CELT_DECODER(st) validate_celt_decoder(st)
+#else
+#define VALIDATE_CELT_DECODER(st)
+#endif
+
int resampling_factor(opus_int32 rate);
void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RESTRICT inp,
@@ -201,7 +223,17 @@ void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RES
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);
+ const opus_val16 *window, int overlap, int arch);
+
+#ifdef NON_STATIC_COMB_FILTER_CONST_C
+void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N,
+ opus_val16 g10, opus_val16 g11, opus_val16 g12);
+#endif
+
+#ifndef OVERRIDE_COMB_FILTER_CONST
+# define comb_filter_const(y, x, T, N, g10, g11, g12, arch) \
+ ((void)(arch),comb_filter_const_c(y, x, T, N, g10, g11, g12))
+#endif
void init_caps(const CELTMode *m,int *cap,int LM,int C);
diff --git a/lib/rbcodec/codecs/libopus/celt/celt_decoder.c b/lib/rbcodec/codecs/libopus/celt/celt_decoder.c
index 8af96b7931..e6efce9358 100644
--- a/lib/rbcodec/codecs/libopus/celt/celt_decoder.c
+++ b/lib/rbcodec/codecs/libopus/celt/celt_decoder.c
@@ -51,6 +51,14 @@
#include "celt_lpc.h"
#include "vq.h"
+/* 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)
+
#if defined(SMALL_FOOTPRINT) && defined(FIXED_POINT)
#define NORM_ALIASING_HACK
#endif
@@ -73,6 +81,7 @@ struct OpusCustomDecoder {
int downsample;
int start, end;
int signalling;
+ int disable_inv;
int arch;
/* Everything beyond this point gets cleared on a reset */
@@ -82,6 +91,7 @@ struct OpusCustomDecoder {
int error;
int last_pitch_index;
int loss_count;
+ int skip_plc;
int postfilter_period;
int postfilter_period_old;
opus_val16 postfilter_gain;
@@ -99,6 +109,38 @@ struct OpusCustomDecoder {
/* opus_val16 backgroundLogE[], Size = 2*mode->nbEBands */
};
+#if defined(ENABLE_HARDENING) || defined(ENABLE_ASSERTIONS)
+/* Make basic checks on the CELT state to ensure we don't end
+ up writing all over memory. */
+void validate_celt_decoder(CELTDecoder *st)
+{
+#ifndef CUSTOM_MODES
+ celt_assert(st->mode == opus_custom_mode_create(48000, 960, NULL));
+ celt_assert(st->overlap == 120);
+#endif
+ celt_assert(st->channels == 1 || st->channels == 2);
+ celt_assert(st->stream_channels == 1 || st->stream_channels == 2);
+ celt_assert(st->downsample > 0);
+ celt_assert(st->start == 0 || st->start == 17);
+ celt_assert(st->start < st->end);
+ celt_assert(st->end <= 21);
+#ifdef OPUS_ARCHMASK
+ celt_assert(st->arch >= 0);
+ celt_assert(st->arch <= OPUS_ARCHMASK);
+#endif
+ celt_assert(st->last_pitch_index <= PLC_PITCH_LAG_MAX);
+ celt_assert(st->last_pitch_index >= PLC_PITCH_LAG_MIN || st->last_pitch_index == 0);
+ celt_assert(st->postfilter_period < MAX_PERIOD);
+ celt_assert(st->postfilter_period >= COMBFILTER_MINPERIOD || st->postfilter_period == 0);
+ celt_assert(st->postfilter_period_old < MAX_PERIOD);
+ celt_assert(st->postfilter_period_old >= COMBFILTER_MINPERIOD || st->postfilter_period_old == 0);
+ celt_assert(st->postfilter_tapset <= 2);
+ celt_assert(st->postfilter_tapset >= 0);
+ celt_assert(st->postfilter_tapset_old <= 2);
+ celt_assert(st->postfilter_tapset_old >= 0);
+}
+#endif
+
int celt_decoder_get_size(int channels)
{
const CELTMode *mode = opus_custom_mode_create(48000, 960, NULL);
@@ -162,10 +204,13 @@ OPUS_CUSTOM_NOSTATIC int opus_custom_decoder_init(CELTDecoder *st, const CELTMod
st->start = 0;
st->end = st->mode->effEBands;
st->signalling = 1;
+#ifndef DISABLE_UPDATE_DRAFT
+ st->disable_inv = channels == 1;
+#else
+ st->disable_inv = 0;
+#endif
st->arch = opus_select_arch();
- st->loss_count = 0;
-
opus_custom_decoder_ctl(st, OPUS_RESET_STATE);
return OPUS_OK;
@@ -178,6 +223,36 @@ void opus_custom_decoder_destroy(CELTDecoder *st)
}
#endif /* CUSTOM_MODES */
+#ifndef CUSTOM_MODES
+/* Special case for stereo with no downsampling and no accumulation. This is
+ quite common and we can make it faster by processing both channels in the
+ same loop, reducing overhead due to the dependency loop in the IIR filter. */
+static void deemphasis_stereo_simple(celt_sig *in[], opus_val16 *pcm, int N, const opus_val16 coef0,
+ celt_sig *mem)
+{
+ celt_sig * OPUS_RESTRICT x0;
+ celt_sig * OPUS_RESTRICT x1;
+ celt_sig m0, m1;
+ int j;
+ x0=in[0];
+ x1=in[1];
+ m0 = mem[0];
+ m1 = mem[1];
+ for (j=0;j<N;j++)
+ {
+ celt_sig tmp0, tmp1;
+ /* Add VERY_SMALL to x[] first to reduce dependency chain. */
+ tmp0 = x0[j] + VERY_SMALL + m0;
+ tmp1 = x1[j] + VERY_SMALL + m1;
+ m0 = MULT16_32_Q15(coef0, tmp0);
+ m1 = MULT16_32_Q15(coef0, tmp1);
+ pcm[2*j ] = SCALEOUT(SIG2WORD16(tmp0));
+ pcm[2*j+1] = SCALEOUT(SIG2WORD16(tmp1));
+ }
+ mem[0] = m0;
+ mem[1] = m1;
+}
+#endif
#ifndef RESYNTH
static
@@ -191,6 +266,14 @@ void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, c
opus_val16 coef0;
VARDECL(celt_sig, scratch);
SAVE_STACK;
+#ifndef CUSTOM_MODES
+ /* Short version for common case. */
+ if (downsample == 1 && C == 2 && !accum)
+ {
+ deemphasis_stereo_simple(in, pcm, N, coef[0], mem);
+ return;
+ }
+#endif
#ifndef FIXED_POINT
(void)accum;
celt_assert(accum==0);
@@ -226,7 +309,7 @@ void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, c
/* Shortcut for the standard (non-custom modes) case */
for (j=0;j<N;j++)
{
- celt_sig tmp = x[j] + m + VERY_SMALL;
+ celt_sig tmp = x[j] + VERY_SMALL + m;
m = MULT16_32_Q15(coef0, tmp);
scratch[j] = tmp;
}
@@ -247,7 +330,7 @@ void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, c
{
for (j=0;j<N;j++)
{
- celt_sig tmp = x[j] + m + VERY_SMALL;
+ celt_sig tmp = x[j] + VERY_SMALL + m;
m = MULT16_32_Q15(coef0, tmp);
y[j*C] = SCALEOUT(SIG2WORD16(tmp));
}
@@ -278,8 +361,9 @@ void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, c
static
#endif
void celt_synthesis(const CELTMode *mode, celt_norm *X, celt_sig * out_syn[],
- opus_val16 *oldBandE, int start, int effEnd, int C, int CC, int isTransient,
- int LM, int downsample, int silence)
+ opus_val16 *oldBandE, int start, int effEnd, int C, int CC,
+ int isTransient, int LM, int downsample,
+ int silence, int arch)
{
int c, i;
int M;
@@ -319,9 +403,9 @@ void celt_synthesis(const CELTMode *mode, celt_norm *X, celt_sig * out_syn[],
freq2 = out_syn[1]+overlap/2;
OPUS_COPY(freq2, freq, N);
for (b=0;b<B;b++)
- clt_mdct_backward(&mode->mdct, &freq2[b], out_syn[0]+NB*b, mode->window, overlap, shift, B);
+ clt_mdct_backward(&mode->mdct, &freq2[b], out_syn[0]+NB*b, mode->window, overlap, shift, B, arch);
for (b=0;b<B;b++)
- clt_mdct_backward(&mode->mdct, &freq[b], out_syn[1]+NB*b, mode->window, overlap, shift, B);
+ clt_mdct_backward(&mode->mdct, &freq[b], out_syn[1]+NB*b, mode->window, overlap, shift, B, arch);
} else if (CC==1&&C==2)
{
/* Downmixing a stereo stream to mono */
@@ -333,18 +417,24 @@ void celt_synthesis(const CELTMode *mode, celt_norm *X, celt_sig * out_syn[],
denormalise_bands(mode, X+N, freq2, oldBandE+nbEBands, start, effEnd, M,
downsample, silence);
for (i=0;i<N;i++)
- freq[i] = HALF32(ADD32(freq[i],freq2[i]));
+ freq[i] = ADD32(HALF32(freq[i]), HALF32(freq2[i]));
for (b=0;b<B;b++)
- clt_mdct_backward(&mode->mdct, &freq[b], out_syn[0]+NB*b, mode->window, overlap, shift, B);
+ clt_mdct_backward(&mode->mdct, &freq[b], out_syn[0]+NB*b, mode->window, overlap, shift, B, arch);
} else {
/* Normal case (mono or stereo) */
c=0; do {
denormalise_bands(mode, X+c*N, freq, oldBandE+c*nbEBands, start, effEnd, M,
downsample, silence);
for (b=0;b<B;b++)
- clt_mdct_backward(&mode->mdct, &freq[b], out_syn[c]+NB*b, mode->window, overlap, shift, B);
+ clt_mdct_backward(&mode->mdct, &freq[b], out_syn[c]+NB*b, mode->window, overlap, shift, B, arch);
} while (++c<CC);
}
+ /* Saturate IMDCT output so that we can't overflow in the pitch postfilter
+ or in the */
+ c=0; do {
+ for (i=0;i<N;i++)
+ out_syn[c][i] = SATURATE(out_syn[c][i], SIG_SAT);
+ } while (++c<CC);
RESTORE_STACK;
}
@@ -387,14 +477,6 @@ static void tf_decode(int start, int end, int isTransient, int *tf_res, int LM,
}
}
-/* 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 int celt_plc_pitch_search(celt_sig *decode_mem[2], int C, int arch)
{
int pitch_index;
@@ -446,7 +528,7 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM)
loss_count = st->loss_count;
start = st->start;
- noise_based = loss_count >= 5 || start != 0;
+ noise_based = loss_count >= 5 || start != 0 || st->skip_plc;
if (noise_based)
{
/* Noise-based PLC/CNG */
@@ -456,10 +538,9 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM)
VARDECL(celt_norm, X);
#endif
opus_uint32 seed;
- opus_val16 *plcLogE;
int end;
int effEnd;
-
+ opus_val16 decay;
end = st->end;
effEnd = IMAX(start, IMIN(end, mode->effEBands));
@@ -471,19 +552,13 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM)
ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */
#endif
- 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;
- }
+ /* Energy decay */
+ 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] = MAX16(backgroundLogE[c*nbEBands+i], oldBandE[c*nbEBands+i] - decay);
+ } while (++c<C);
seed = st->rng;
for (c=0;c<C;c++)
{
@@ -499,7 +574,7 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM)
seed = celt_lcg_rand(seed);
X[boffs+j] = (celt_norm)((opus_int32)seed>>20);
}
- renormalise_vector(X+boffs, blen, Q15ONE);
+ renormalise_vector(X+boffs, blen, Q15ONE, st->arch);
}
}
st->rng = seed;
@@ -509,14 +584,17 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM)
DECODE_BUFFER_SIZE-N+(overlap>>1));
} while (++c<C);
- celt_synthesis(mode, X, out_syn, plcLogE, start, effEnd, C, C, 0, LM, st->downsample, 0);
+ celt_synthesis(mode, X, out_syn, oldBandE, start, effEnd, C, C, 0, LM, st->downsample, 0, st->arch);
} else {
+ int exc_length;
/* Pitch-based PLC */
const opus_val16 *window;
+ opus_val16 *exc;
opus_val16 fade = Q15ONE;
int pitch_index;
VARDECL(opus_val32, etmp);
- VARDECL(opus_val16, exc);
+ VARDECL(opus_val16, _exc);
+ VARDECL(opus_val16, fir_tmp);
if (loss_count == 0)
{
@@ -526,8 +604,14 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM)
fade = QCONST16(.8f,15);
}
+ /* 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);
+
ALLOC(etmp, overlap, opus_val32);
- ALLOC(exc, MAX_PERIOD, opus_val16);
+ ALLOC(_exc, MAX_PERIOD+LPC_ORDER, opus_val16);
+ ALLOC(fir_tmp, exc_length, opus_val16);
+ exc = _exc+LPC_ORDER;
window = mode->window;
c=0; do {
opus_val16 decay;
@@ -536,13 +620,11 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM)
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);
- }
+ for (i=0;i<MAX_PERIOD+LPC_ORDER;i++)
+ exc[i-LPC_ORDER] = ROUND16(buf[DECODE_BUFFER_SIZE-MAX_PERIOD-LPC_ORDER+i], SIG_SHIFT);
if (loss_count == 0)
{
@@ -568,22 +650,32 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM)
#endif
}
_celt_lpc(lpc+c*LPC_ORDER, ac, LPC_ORDER);
+#ifdef FIXED_POINT
+ /* For fixed-point, apply bandwidth expansion until we can guarantee that
+ no overflow can happen in the IIR filter. This means:
+ 32768*sum(abs(filter)) < 2^31 */
+ while (1) {
+ opus_val16 tmp=Q15ONE;
+ opus_val32 sum=QCONST16(1., SIG_SHIFT);
+ for (i=0;i<LPC_ORDER;i++)
+ sum += ABS16(lpc[c*LPC_ORDER+i]);
+ if (sum < 65535) break;
+ for (i=0;i<LPC_ORDER;i++)
+ {
+ tmp = MULT16_16_Q15(QCONST16(.99f,15), tmp);
+ lpc[c*LPC_ORDER+i] = MULT16_16_Q15(lpc[c*LPC_ORDER+i], tmp);
+ }
+ }
+#endif
}
- /* 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. */
+ /* Compute the excitation for exc_length samples before the loss. We need the copy
+ because celt_fir() cannot filter in-place. */
celt_fir(exc+MAX_PERIOD-exc_length, lpc+c*LPC_ORDER,
- exc+MAX_PERIOD-exc_length, exc_length, LPC_ORDER, lpc_mem);
+ fir_tmp, exc_length, LPC_ORDER, st->arch);
+ OPUS_COPY(exc+MAX_PERIOD-exc_length, fir_tmp, exc_length);
}
/* Check if the waveform is decaying, and if so how fast.
@@ -637,9 +729,8 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM)
tmp = ROUND16(
buf[DECODE_BUFFER_SIZE-MAX_PERIOD-N+extrapolation_offset+j],
SIG_SHIFT);
- S1 += SHR32(MULT16_16(tmp, tmp), 8);
+ S1 += SHR32(MULT16_16(tmp, tmp), 10);
}
-
{
opus_val16 lpc_mem[LPC_ORDER];
/* Copy the last decoded samples (prior to the overlap region) to
@@ -650,7 +741,11 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM)
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);
+ lpc_mem, st->arch);
+#ifdef FIXED_POINT
+ for (i=0; i < extrapolation_len; i++)
+ buf[DECODE_BUFFER_SIZE-N+i] = SATURATE(buf[DECODE_BUFFER_SIZE-N+i], SIG_SAT);
+#endif
}
/* Check if the synthesis energy is higher than expected, which can
@@ -661,7 +756,7 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM)
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);
+ S2 += SHR32(MULT16_16(tmp, tmp), 10);
}
/* This checks for an "explosion" in the synthesis. */
#ifdef FIXED_POINT
@@ -698,7 +793,7 @@ static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM)
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);
+ st->postfilter_tapset, st->postfilter_tapset, NULL, 0, st->arch);
/* Simulate TDAC on the concealed audio so that it blends with the
MDCT of the next frame. */
@@ -769,6 +864,7 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
const opus_int16 *eBands;
ALLOC_STACK;
+ VALIDATE_CELT_DECODER(st);
mode = st->mode;
nbEBands = mode->nbEBands;
overlap = mode->overlap;
@@ -838,6 +934,10 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
return frame_size/st->downsample;
}
+ /* Check if there are at least two packets received consecutively before
+ * turning on the pitch-based PLC */
+ st->skip_plc = st->loss_count != 0;
+
if (dec == NULL)
{
ec_dec_init(&_dec,(unsigned char*)data,len);
@@ -959,7 +1059,7 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
ALLOC(pulses, nbEBands, int);
ALLOC(fine_priority, nbEBands, int);
- codedBands = compute_allocation(mode, start, end, offsets, cap,
+ codedBands = clt_compute_allocation(mode, start, end, offsets, cap,
alloc_trim, &intensity, &dual_stereo, bits, &balance, pulses,
fine_quant, fine_priority, C, LM, dec, 0, 0, 0);
@@ -982,7 +1082,8 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
quant_all_bands(0, mode, start, 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);
+ len*(8<<BITRES)-anti_collapse_rsv, balance, dec, LM, codedBands, &st->rng, 0,
+ st->arch, st->disable_inv);
if (anti_collapse_rsv > 0)
{
@@ -994,7 +1095,7 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
if (anti_collapse_on)
anti_collapse(mode, X, collapse_masks, LM, C, N,
- start, end, oldBandE, oldLogE, oldLogE2, pulses, st->rng);
+ start, end, oldBandE, oldLogE, oldLogE2, pulses, st->rng, st->arch);
if (silence)
{
@@ -1002,18 +1103,19 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
oldBandE[i] = -QCONST16(28.f,DB_SHIFT);
}
- celt_synthesis(mode, X, out_syn, oldBandE, start, effEnd, C, CC, isTransient, LM, st->downsample, silence);
+ celt_synthesis(mode, X, out_syn, oldBandE, start, effEnd,
+ C, CC, isTransient, LM, st->downsample, silence, st->arch);
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);
+ mode->window, overlap, st->arch);
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);
+ mode->window, overlap, st->arch);
} while (++c<CC);
st->postfilter_period_old = st->postfilter_period;
@@ -1035,10 +1137,18 @@ int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *dat
/* In case start or end were to change */
if (!isTransient)
{
+ opus_val16 max_background_increase;
OPUS_COPY(oldLogE2, oldLogE, 2*nbEBands);
OPUS_COPY(oldLogE, oldBandE, 2*nbEBands);
+ /* In normal circumstances, we only allow the noise floor to increase by
+ up to 2.4 dB/second, but when we're in DTX, we allow up to 6 dB
+ increase for each update.*/
+ if (st->loss_count < 10)
+ max_background_increase = M*QCONST16(0.001f,DB_SHIFT);
+ else
+ max_background_increase = QCONST16(1.f,DB_SHIFT);
for (i=0;i<2*nbEBands;i++)
- backgroundLogE[i] = MIN16(backgroundLogE[i] + M*QCONST16(0.001f,DB_SHIFT), oldBandE[i]);
+ backgroundLogE[i] = MIN16(backgroundLogE[i] + max_background_increase, oldBandE[i]);
} else {
for (i=0;i<2*nbEBands;i++)
oldLogE[i] = MIN16(oldLogE[i], oldBandE[i]);
@@ -1195,6 +1305,7 @@ int opus_custom_decoder_ctl(CELTDecoder * OPUS_RESTRICT st, int request, ...)
((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);
+ st->skip_plc = 1;
}
break;
case OPUS_GET_PITCH_REQUEST:
@@ -1227,6 +1338,26 @@ int opus_custom_decoder_ctl(CELTDecoder * OPUS_RESTRICT st, int request, ...)
*value=st->rng;
}
break;
+ case OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST:
+ {
+ opus_int32 value = va_arg(ap, opus_int32);
+ if(value<0 || value>1)
+ {
+ goto bad_arg;
+ }
+ st->disable_inv = value;
+ }
+ break;
+ case OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST:
+ {
+ opus_int32 *value = va_arg(ap, opus_int32*);
+ if (!value)
+ {
+ goto bad_arg;
+ }
+ *value = st->disable_inv;
+ }
+ break;
default:
goto bad_request;
}
diff --git a/lib/rbcodec/codecs/libopus/celt/celt_encoder.c b/lib/rbcodec/codecs/libopus/celt/celt_encoder.c
new file mode 100644
index 0000000000..44cb0850ab
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/celt_encoder.c
@@ -0,0 +1,2607 @@
+/* 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 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 lfe;
+ int disable_inv;
+ 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;
+ SILKInfo silk_info;
+
+ 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 */
+ /* opus_val16 energyError[], 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]; */
+ + 4*channels*mode->nbEBands*sizeof(opus_val16); /* opus_val16 oldBandE[channels*mode->nbEBands]; */
+ /* opus_val16 oldLogE[channels*mode->nbEBands]; */
+ /* opus_val16 oldLogE2[channels*mode->nbEBands]; */
+ /* opus_val16 energyError[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->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 allow_weak_transients,
+ int *weak_transient)
+{
+ 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;
+ /* Forward masking: 6.7 dB/ms. */
+#ifdef FIXED_POINT
+ int forward_shift = 4;
+#else
+ opus_val16 forward_decay = QCONST16(.0625f,15);
+#endif
+ /* 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);
+
+ *weak_transient = 0;
+ /* For lower bitrates, let's be more conservative and have a forward masking
+ decay of 3.3 dB/ms. This avoids having to code transients at very low
+ bitrate (mostly for hybrid), which can result in unstable energy and/or
+ partial collapse. */
+ if (allow_weak_transients)
+ {
+#ifdef FIXED_POINT
+ forward_shift = 5;
+#else
+ forward_decay = QCONST16(.03125f,15);
+#endif
+ }
+ 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] = SROUND16(y, 2);
+ /*printf("%f ", tmp[i]);*/
+ }
+ /*printf("\n");*/
+ /* First few samples are bad because we don't propagate the memory */
+ OPUS_CLEAR(tmp, 12);
+
+#ifdef FIXED_POINT
+ /* Normalize tmp to max range */
+ {
+ int shift=0;
+ shift = 14-celt_ilog2(MAX16(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,forward_shift);
+#else
+ tmp[i] = mem0 + MULT16_16_P15(forward_decay,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--)
+ {
+ /* Backward masking: 13.9 dB/ms. */
+#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;
+ /* We should never see NaNs here. If we find any, then something really bad happened and we better abort
+ before it does any damage later on. If these asserts are disabled (no hardening), then the table
+ lookup a few lines below (id = ...) is likely to crash dur to an out-of-bounds read. DO NOT FIX
+ that crash on NaN since it could result in a worse issue later on. */
+ celt_assert(!celt_isnan(tmp[0]));
+ celt_assert(!celt_isnan(norm));
+ for (i=12;i<len2-5;i+=4)
+ {
+ int id;
+#ifdef FIXED_POINT
+ id = MAX32(0,MIN32(127,MULT16_32_Q15(tmp[i]+EPSILON,norm))); /* Do not round to nearest */
+#else
+ id = (int)MAX32(0,MIN32(127,floor(64*norm*(tmp[i]+EPSILON)))); /* 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;
+ /* For low bitrates, define "weak transients" that need to be
+ handled differently to avoid partial collapse. */
+ if (allow_weak_transients && is_transient && mask_metric<600) {
+ is_transient = 0;
+ *weak_transient = 1;
+ }
+ /* 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(MAX32(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 */
+static int patch_transient_decision(opus_val16 *newE, opus_val16 *oldE, int nbEBands,
+ int start, 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[start] = oldE[start];
+ for (i=start+1;i<end;i++)
+ spread_old[i] = MAX16(spread_old[i-1]-QCONST16(1.0f, DB_SHIFT), oldE[i]);
+ } else {
+ spread_old[start] = MAX16(oldE[start],oldE[start+nbEBands]);
+ for (i=start+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>=start;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=IMAX(2,start);i<end-1;i++)
+ {
+ opus_val16 x1, x2;
+ x1 = MAX16(0, newE[i + c*nbEBands]);
+ 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-1-IMAX(2,start)));
+ /*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,
+ int arch)
+{
+ const int overlap = mode->overlap;
+ 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,
+ arch);
+ }
+ } 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;
+ OPUS_CLEAR(&out[c*B*N+bound], B*N-bound);
+ } 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];
+ m = *mem;
+
+ /* Fast path for the normal 48kHz case and no clipping */
+ if (coef[1] == 0 && upsample == 1 && !clip)
+ {
+ for (i=0;i<N;i++)
+ {
+ opus_val16 x;
+ x = SCALEIN(pcmp[CC*i]);
+ /* Apply pre-emphasis */
+ inp[i] = SHL32(x, SIG_SHIFT) - m;
+ m = SHR32(MULT16_16(coef0, x), 15-SIG_SHIFT);
+ }
+ *mem = m;
+ return;
+ }
+
+ Nu = N/upsample;
+ if (upsample!=1)
+ {
+ OPUS_CLEAR(inp, N);
+ }
+ for (i=0;i<Nu;i++)
+ inp[i*upsample] = SCALEIN(pcmp[CC*i]);
+
+#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
+#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++)
+ {
+ opus_val16 x;
+ x = inp[i];
+ /* Apply pre-emphasis */
+ inp[i] = SHL32(x, SIG_SHIFT) - m;
+ m = SHR32(MULT16_16(coef0, x), 15-SIG_SHIFT);
+ }
+ }
+ *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,
+ opus_val16 tf_estimate, int tf_chan, int *importance)
+{
+ 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);
+
+ for (i=0;i<len;i++)
+ {
+ int 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;
+ OPUS_COPY(tmp, &X[tf_chan*N0 + (m->eBands[i]<<LM)], N);
+ /* 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)
+ {
+ OPUS_COPY(tmp_1, tmp, N);
+ 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;
+ /* 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]/2 + (!isTransient)*LM);*/
+ }
+ /*printf("\n");*/
+ /* Search for the optimal tf resolution, including tf_select */
+ tf_select = 0;
+ for (sel=0;sel<2;sel++)
+ {
+ cost0 = importance[0]*abs(metric[0]-2*tf_select_table[LM][4*isTransient+2*sel+0]);
+ cost1 = importance[0]*abs(metric[0]-2*tf_select_table[LM][4*isTransient+2*sel+1]) + (isTransient ? 0 : lambda);
+ for (i=1;i<len;i++)
+ {
+ int curr0, curr1;
+ curr0 = IMIN(cost0, cost1 + lambda);
+ curr1 = IMIN(cost0 + lambda, cost1);
+ cost0 = curr0 + importance[i]*abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*sel+0]);
+ cost1 = curr1 + importance[i]*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 = importance[0]*abs(metric[0]-2*tf_select_table[LM][4*isTransient+2*tf_select+0]);
+ cost1 = importance[0]*abs(metric[0]-2*tf_select_table[LM][4*isTransient+2*tf_select+1]) + (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 + importance[i]*abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*tf_select+0]);
+ cost1 = curr1 + importance[i]*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, opus_int32 equiv_rate, int arch)
+{
+ int i;
+ opus_val32 diff=0;
+ int c;
+ int trim_index;
+ opus_val16 trim = QCONST16(5.f, 8);
+ opus_val16 logXC, logXC2;
+ /* At low bitrate, reducing the trim seems to help. At higher bitrates, it's less
+ clear what's best, so we're keeping it as it was before, at least for now. */
+ if (equiv_rate < 64000) {
+ trim = QCONST16(4.f, 8);
+ } else if (equiv_rate < 80000) {
+ opus_int32 frac = (equiv_rate-64000) >> 10;
+ trim = QCONST16(4.f, 8) + QCONST16(1.f/16.f, 8)*frac;
+ }
+ 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++)
+ {
+ opus_val32 partial;
+ partial = celt_inner_prod(&X[m->eBands[i]<<LM], &X[N0+(m->eBands[i]<<LM)],
+ (m->eBands[i+1]-m->eBands[i])<<LM, arch);
+ 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++)
+ {
+ opus_val32 partial;
+ partial = celt_inner_prod(&X[m->eBands[i]<<LM], &X[N0+(m->eBands[i]<<LM)],
+ (m->eBands[i+1]-m->eBands[i])<<LM, arch);
+ minXC = MIN16(minXC, ABS16(EXTRACT16(SHR32(partial, 18))));
+ }
+ minXC = MIN16(QCONST16(1.f, 10), ABS16(minXC));
+ /*printf ("%f\n", sum);*/
+ /* 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);*/
+ trim -= MAX32(-QCONST16(2.f, 8), MIN32(QCONST16(2.f, 8), SHR32(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))));
+ }
+#else
+ (void)analysis;
+#endif
+
+#ifdef FIXED_POINT
+ trim_index = PSHR32(trim, 8);
+#else
+ trim_index = (int)floor(.5f+trim);
+#endif
+ trim_index = IMAX(0, IMIN(10, trim_index));
+ /*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);
+}
+
+#define MSWAP(a,b) do {opus_val16 tmp = a;a=b;b=tmp;} while(0)
+static opus_val16 median_of_5(const opus_val16 *x)
+{
+ opus_val16 t0, t1, t2, t3, t4;
+ t2 = x[2];
+ if (x[0] > x[1])
+ {
+ t0 = x[1];
+ t1 = x[0];
+ } else {
+ t0 = x[0];
+ t1 = x[1];
+ }
+ if (x[3] > x[4])
+ {
+ t3 = x[4];
+ t4 = x[3];
+ } else {
+ t3 = x[3];
+ t4 = x[4];
+ }
+ if (t0 > t3)
+ {
+ MSWAP(t0, t3);
+ MSWAP(t1, t4);
+ }
+ if (t2 > t1)
+ {
+ if (t1 < t3)
+ return MIN16(t2, t3);
+ else
+ return MIN16(t4, t1);
+ } else {
+ if (t2 < t3)
+ return MIN16(t1, t3);
+ else
+ return MIN16(t2, t4);
+ }
+}
+
+static opus_val16 median_of_3(const opus_val16 *x)
+{
+ opus_val16 t0, t1, t2;
+ if (x[0] > x[1])
+ {
+ t0 = x[1];
+ t1 = x[0];
+ } else {
+ t0 = x[0];
+ t1 = x[1];
+ }
+ t2 = x[2];
+ if (t1 < t2)
+ return t1;
+ else if (t0 < t2)
+ return t2;
+ else
+ return t0;
+}
+
+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,
+ AnalysisInfo *analysis, int *importance, int *spread_weight)
+{
+ 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);
+ OPUS_CLEAR(offsets, nbEBands);
+ /* 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);
+ {
+ /* Compute a really simple masking model to avoid taking into account completely masked
+ bands when computing the spreading decision. */
+ VARDECL(opus_val16, mask);
+ VARDECL(opus_val16, sig);
+ ALLOC(mask, nbEBands, opus_val16);
+ ALLOC(sig, nbEBands, opus_val16);
+ for (i=0;i<end;i++)
+ mask[i] = bandLogE[i]-noise_floor[i];
+ if (C==2)
+ {
+ for (i=0;i<end;i++)
+ mask[i] = MAX16(mask[i], bandLogE[nbEBands+i]-noise_floor[i]);
+ }
+ OPUS_COPY(sig, mask, end);
+ for (i=1;i<end;i++)
+ mask[i] = MAX16(mask[i], mask[i-1] - QCONST16(2.f, DB_SHIFT));
+ for (i=end-2;i>=0;i--)
+ mask[i] = MAX16(mask[i], mask[i+1] - QCONST16(3.f, DB_SHIFT));
+ for (i=0;i<end;i++)
+ {
+ /* Compute SMR: Mask is never more than 72 dB below the peak and never below the noise floor.*/
+ opus_val16 smr = sig[i]-MAX16(MAX16(0, maxDepth-QCONST16(12.f, DB_SHIFT)), mask[i]);
+ /* Clamp SMR to make sure we're not shifting by something negative or too large. */
+#ifdef FIXED_POINT
+ /* FIXME: Use PSHR16() instead */
+ int shift = -PSHR32(MAX16(-QCONST16(5.f, DB_SHIFT), MIN16(0, smr)), DB_SHIFT);
+#else
+ int shift = IMIN(5, IMAX(0, -(int)floor(.5f + smr)));
+#endif
+ spread_weight[i] = 32 >> shift;
+ }
+ /*for (i=0;i<end;i++)
+ printf("%d ", spread_weight[i]);
+ printf("\n");*/
+ }
+ /* Make sure that dynamic allocation can't make us bust the budget */
+ if (effectiveBytes > 50 && LM>=1 && !lfe)
+ {
+ int last=0;
+ c=0;do
+ {
+ opus_val16 offset;
+ opus_val16 tmp;
+ opus_val16 *f;
+ f = &follower[c*nbEBands];
+ f[0] = 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;
+ f[i] = MIN16(f[i-1]+QCONST16(1.5f,DB_SHIFT), bandLogE2[c*nbEBands+i]);
+ }
+ for (i=last-1;i>=0;i--)
+ f[i] = MIN16(f[i], MIN16(f[i+1]+QCONST16(2.f,DB_SHIFT), bandLogE2[c*nbEBands+i]));
+
+ /* Combine with a median filter to avoid dynalloc triggering unnecessarily.
+ The "offset" value controls how conservative we are -- a higher offset
+ reduces the impact of the median filter and makes dynalloc use more bits. */
+ offset = QCONST16(1.f, DB_SHIFT);
+ for (i=2;i<end-2;i++)
+ f[i] = MAX16(f[i], median_of_5(&bandLogE2[c*nbEBands+i-2])-offset);
+ tmp = median_of_3(&bandLogE2[c*nbEBands])-offset;
+ f[0] = MAX16(f[0], tmp);
+ f[1] = MAX16(f[1], tmp);
+ tmp = median_of_3(&bandLogE2[c*nbEBands+end-3])-offset;
+ f[end-2] = MAX16(f[end-2], tmp);
+ f[end-1] = MAX16(f[end-1], tmp);
+
+ for (i=0;i<end;i++)
+ f[i] = MAX16(f[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 (i=start;i<end;i++)
+ {
+#ifdef FIXED_POINT
+ importance[i] = PSHR32(13*celt_exp2(MIN16(follower[i], QCONST16(4.f, DB_SHIFT))), 16);
+#else
+ importance[i] = (int)floor(.5f+13*celt_exp2(MIN16(follower[i], QCONST16(4.f, DB_SHIFT))));
+#endif
+ }
+ /* 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++)
+ {
+ if (i<8)
+ follower[i] *= 2;
+ if (i>=12)
+ follower[i] = HALF16(follower[i]);
+ }
+#ifdef DISABLE_FLOAT_API
+ (void)analysis;
+#else
+ if (analysis->valid)
+ {
+ for (i=start;i<IMIN(LEAK_BANDS, end);i++)
+ follower[i] = follower[i] + QCONST16(1.f/64.f, DB_SHIFT)*analysis->leak_boost[i];
+ }
+#endif
+ for (i=start;i<end;i++)
+ {
+ int width;
+ int boost;
+ int boost_bits;
+
+ 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 2/3 of the bits */
+ if ((!vbr || (constrained_vbr&&!isTransient))
+ && (tot_boost+boost_bits)>>BITRES>>3 > 2*effectiveBytes/3)
+ {
+ opus_int32 cap = ((2*effectiveBytes/3)<<BITRES<<3);
+ offsets[i] = cap-tot_boost;
+ tot_boost = cap;
+ break;
+ } else {
+ offsets[i] = boost;
+ tot_boost += boost_bits;
+ }
+ }
+ } else {
+ for (i=start;i<end;i++)
+ importance[i] = 13;
+ }
+ *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, AnalysisInfo *analysis)
+{
+ 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;
+ int overlap;
+ SAVE_STACK;
+
+ mode = st->mode;
+ overlap = mode->overlap;
+ 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+overlap)+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, st->arch);
+ 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;
+ }
+#ifndef DISABLE_FLOAT_API
+ if (analysis->valid)
+ gain1 = (opus_val16)(gain1 * analysis->max_pitch_ratio);
+#else
+ (void)analysis;
+#endif
+ /* 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-overlap;
+ st->prefilter_period=IMAX(st->prefilter_period, COMBFILTER_MINPERIOD);
+ OPUS_COPY(in+c*(N+overlap), st->in_mem+c*(overlap), overlap);
+ if (offset)
+ comb_filter(in+c*(N+overlap)+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, st->arch);
+
+ comb_filter(in+c*(N+overlap)+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, overlap, st->arch);
+ OPUS_COPY(st->in_mem+c*(overlap), in+c*(N+overlap)+N, overlap);
+
+ if (N>COMBFILTER_MAXPERIOD)
+ {
+ OPUS_COPY(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_COPY(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 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-(19<<LM);
+ /* Apply transient boost, compensating for average boost. */
+ tf_calibration = QCONST16(0.044f,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)-0.12f;
+ 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;
+ }
+#else
+ (void)analysis;
+ (void)pitch_change;
+#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);*/
+ }
+
+ /* Make VBR less aggressive for constrained VBR because we can't keep a higher bitrate
+ for long. Needs tuning. */
+ if ((!has_surround_mask||lfe) && constrained_vbr)
+ {
+ target = base_target + (opus_int32)MULT16_32_Q15(QCONST16(0.67f, 15), 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, importance);
+ VARDECL(int, spread_weight);
+ VARDECL(int, fine_priority);
+ VARDECL(int, tf_res);
+ VARDECL(unsigned char, collapse_masks);
+ celt_sig *prefilter_mem;
+ opus_val16 *oldBandE, *oldLogE, *oldLogE2, *energyError;
+ 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 start;
+ int end;
+ int effEnd;
+ int codedBands;
+ 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;
+ opus_int32 tell0_frac;
+ 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;
+ int hybrid;
+ int weak_transient = 0;
+ int enable_tf_analysis;
+ VARDECL(opus_val16, surround_dynalloc);
+ ALLOC_STACK;
+
+ mode = st->mode;
+ nbEBands = mode->nbEBands;
+ overlap = mode->overlap;
+ eBands = mode->eBands;
+ start = st->start;
+ end = st->end;
+ hybrid = start != 0;
+ 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*(overlap);
+ oldBandE = (opus_val16*)(st->in_mem+CC*(overlap+COMBFILTER_MAXPERIOD));
+ oldLogE = oldBandE + CC*nbEBands;
+ oldLogE2 = oldLogE + CC*nbEBands;
+ energyError = oldLogE2 + CC*nbEBands;
+
+ if (enc==NULL)
+ {
+ tell0_frac=tell=1;
+ nbFilledBytes=0;
+ } else {
+ tell0_frac=ec_tell_frac(enc);
+ tell=ec_tell(enc);
+ nbFilledBytes=(tell+4)>>3;
+ }
+
+#ifdef CUSTOM_MODES
+ if (st->signalling && enc==NULL)
+ {
+ int tmp = (mode->effEBands-end)>>1;
+ end = 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 - nbFilledBytes;
+ }
+ equiv_rate = ((opus_int32)nbCompressedBytes*8*50 >> (3-LM)) - (40*C+20)*((400>>LM) - 50);
+ if (st->bitrate != OPUS_BITRATE_MAX)
+ equiv_rate = IMIN(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 = end;
+ if (effEnd > mode->effEBands)
+ effEnd = mode->effEBands;
+
+ ALLOC(in, CC*(N+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 {
+ int need_clip=0;
+#ifndef FIXED_POINT
+ need_clip = st->clip && sample_max>65536.f;
+#endif
+ celt_preemphasis(pcm+c, in+c*(N+overlap)+overlap, N, CC, st->upsample,
+ mode->preemph, st->preemph_memE+c, need_clip);
+ } while (++c<CC);
+
+
+
+ /* Find pitch period and gain */
+ {
+ int enabled;
+ int qg;
+ enabled = ((st->lfe&&nbAvailableBytes>3) || nbAvailableBytes>12*C) && !hybrid && !silence && !st->disable_pf
+ && st->complexity >= 5;
+
+ prefilter_tapset = st->tapset_decision;
+ pf_on = run_prefilter(st, in, prefilter_mem, CC, N, prefilter_tapset, &pitch_index, &gain1, &qg, enabled, nbAvailableBytes, &st->analysis);
+ 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(!hybrid && 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)
+ {
+ /* Reduces the likelihood of energy instability on fricatives at low bitrate
+ in hybrid mode. It seems like we still want to have real transients on vowels
+ though (small SILK quantization offset value). */
+ int allow_weak_transients = hybrid && effectiveBytes<15 && st->silk_info.signalType != 2;
+ isTransient = transient_analysis(in, N+overlap, CC,
+ &tf_estimate, &tf_chan, allow_weak_transients, &weak_transient);
+ }
+ 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, st->arch);
+ compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch);
+ amp2Log2(mode, effEnd, 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, st->arch);
+ /* This should catch any NaN in the CELT input. Since we're not supposed to see any (they're filtered
+ at the Opus layer), just abort. */
+ celt_assert(!celt_isnan(freq[0]) && (C==1 || !celt_isnan(freq[N])));
+ if (CC==2&&C==1)
+ tf_chan = 0;
+ compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch);
+
+ if (st->lfe)
+ {
+ for (i=2;i<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, end, bandE, bandLogE, C);
+
+ ALLOC(surround_dynalloc, C*nbEBands, opus_val16);
+ OPUS_CLEAR(surround_dynalloc, end);
+ /* This computes how much masking takes place between surround channels */
+ if (!hybrid&&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);
+ }
+ }
+ celt_assert(count>0);
+ 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;
+ OPUS_CLEAR(surround_dynalloc, mask_end);
+ } 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=start;i<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 /= (end-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)
+ {
+ OPUS_COPY(bandLogE2, bandLogE, C*nbEBands);
+ }
+
+ /* 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 && !hybrid)
+ {
+ if (patch_transient_decision(bandLogE, oldBandE, nbEBands, start, end, C))
+ {
+ isTransient = 1;
+ shortBlocks = M;
+ compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample, st->arch);
+ compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch);
+ amp2Log2(mode, effEnd, 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);
+
+ enable_tf_analysis = effectiveBytes>=15*C && !hybrid && st->complexity>=2 && !st->lfe;
+
+ ALLOC(offsets, nbEBands, int);
+ ALLOC(importance, nbEBands, int);
+ ALLOC(spread_weight, nbEBands, int);
+
+ maxDepth = dynalloc_analysis(bandLogE, bandLogE2, nbEBands, start, end, C, offsets,
+ st->lsb_depth, mode->logN, isTransient, st->vbr, st->constrained_vbr,
+ eBands, LM, effectiveBytes, &tot_boost, st->lfe, surround_dynalloc, &st->analysis, importance, spread_weight);
+
+ ALLOC(tf_res, nbEBands, int);
+ /* Disable variable tf resolution for hybrid and at very low bitrate */
+ if (enable_tf_analysis)
+ {
+ int lambda;
+ lambda = IMAX(80, 20480/effectiveBytes + 2);
+ tf_select = tf_analysis(mode, effEnd, isTransient, tf_res, lambda, X, N, LM, tf_estimate, tf_chan, importance);
+ for (i=effEnd;i<end;i++)
+ tf_res[i] = tf_res[effEnd-1];
+ } else if (hybrid && weak_transient)
+ {
+ /* For weak transients, we rely on the fact that improving time resolution using
+ TF on a long window is imperfect and will not result in an energy collapse at
+ low bitrate. */
+ for (i=0;i<end;i++)
+ tf_res[i] = 1;
+ tf_select=0;
+ } else if (hybrid && effectiveBytes<15 && st->silk_info.signalType != 2)
+ {
+ /* For low bitrate hybrid, we force temporal resolution to 5 ms rather than 2.5 ms. */
+ for (i=0;i<end;i++)
+ tf_res[i] = 0;
+ tf_select=isTransient;
+ } else {
+ for (i=0;i<end;i++)
+ tf_res[i] = isTransient;
+ tf_select=0;
+ }
+
+ ALLOC(error, C*nbEBands, opus_val16);
+ c=0;
+ do {
+ for (i=start;i<end;i++)
+ {
+ /* When the energy is stable, slightly bias energy quantization towards
+ the previous error to make the gain more stable (a constant offset is
+ better than fluctuations). */
+ if (ABS32(SUB32(bandLogE[i+c*nbEBands], oldBandE[i+c*nbEBands])) < QCONST16(2.f, DB_SHIFT))
+ {
+ bandLogE[i+c*nbEBands] -= MULT16_16_Q15(energyError[i+c*nbEBands], QCONST16(0.25f, 15));
+ }
+ }
+ } while (++c < C);
+ quant_coarse_energy(mode, start, 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(start, 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 (hybrid)
+ {
+ if (st->complexity == 0)
+ st->spread_decision = SPREAD_NONE;
+ else if (isTransient)
+ st->spread_decision = SPREAD_NORMAL;
+ else
+ st->spread_decision = SPREAD_AGGRESSIVE;
+ } else if (shortBlocks || st->complexity < 3 || nbAvailableBytes < 10*C)
+ {
+ 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, spread_weight);
+ }
+ /*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);
+ }
+
+ /* 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=start;i<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(end,IMAX(start, st->intensity));
+ }
+
+ alloc_trim = 5;
+ if (tell+(6<<BITRES) <= total_bits - total_boost)
+ {
+ if (start > 0 || st->lfe)
+ {
+ st->stereo_saving = 0;
+ alloc_trim = 5;
+ } else {
+ alloc_trim = alloc_trim_analysis(mode, X, bandLogE,
+ end, LM, C, N, &st->analysis, &st->stereo_saving, tf_estimate,
+ st->intensity, surround_trim, equiv_rate, st->arch);
+ }
+ 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));
+ if (!hybrid)
+ {
+ base_target = vbr_rate - ((40*C+20)<<BITRES);
+ } else {
+ base_target = IMAX(0, vbr_rate - ((9*C+4)<<BITRES));
+ }
+
+ if (st->constrained_vbr)
+ base_target += (st->vbr_offset>>lm_diff);
+
+ if (!hybrid)
+ {
+ 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->lfe, st->energy_mask!=NULL, surround_masking,
+ temporal_vbr);
+ } else {
+ target = base_target;
+ /* Tonal frames (offset<100) need more bits than noisy (offset>100) ones. */
+ if (st->silk_info.offset < 100) target += 12 << BITRES >> (3-LM);
+ if (st->silk_info.offset > 100) target -= 18 << BITRES >> (3-LM);
+ /* Boosting bitrate on transients and vowels with significant temporal
+ spikes. */
+ target += (opus_int32)MULT16_16_Q14(tf_estimate-QCONST16(.25f,14), (50<<BITRES));
+ /* If we have a strong transient, let's make sure it has enough bits to code
+ the first two bands, so that it can use folding rather than noise. */
+ if (tf_estimate > QCONST16(.7f,14))
+ target = IMAX(target, 50<<BITRES);
+ }
+ /* 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;
+ /* Take into account the 37 bits we need to have left in the packet to
+ signal a redundant frame in hybrid mode. Creating a shorter packet would
+ create an entropy coder desync. */
+ if (hybrid)
+ min_allowed = IMAX(min_allowed, (tell0_frac+(37<<BITRES)+total_boost+(1<<(BITRES+3))-1)>>(BITRES+3));
+
+ nbAvailableBytes = (target+(1<<(BITRES+2)))>>(BITRES+3);
+ nbAvailableBytes = IMAX(min_allowed,nbAvailableBytes);
+ nbAvailableBytes = IMIN(nbCompressedBytes,nbAvailableBytes);
+
+ /* 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);
+ /*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 = 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 = clt_compute_allocation(mode, start, 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, start, end, oldBandE, error, fine_quant, enc, C);
+
+ /* Residual quantisation */
+ ALLOC(collapse_masks, C*nbEBands, unsigned char);
+ quant_all_bands(1, mode, start, 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, st->complexity, st->arch, st->disable_inv);
+
+ 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, start, end, oldBandE, error, fine_quant, fine_priority, nbCompressedBytes*8-ec_tell(enc), enc, C);
+ OPUS_CLEAR(energyError, nbEBands*CC);
+ c=0;
+ do {
+ for (i=start;i<end;i++)
+ {
+ energyError[i+c*nbEBands] = MAX16(-QCONST16(0.5f, 15), MIN16(QCONST16(0.5f, 15), error[i+c*nbEBands]));
+ }
+ } while (++c < 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,
+ start, end, oldBandE, oldLogE, oldLogE2, pulses, st->rng);
+ }
+
+ c=0; do {
+ OPUS_MOVE(st->syn_mem[c], st->syn_mem[c]+N, 2*MAX_PERIOD-N+overlap/2);
+ } while (++c<CC);
+
+ c=0; do {
+ out_mem[c] = st->syn_mem[c]+2*MAX_PERIOD-N;
+ } while (++c<CC);
+
+ celt_synthesis(mode, X, out_mem, oldBandE, start, effEnd,
+ C, CC, isTransient, LM, st->upsample, silence, st->arch);
+
+ 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, 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);
+ 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) {
+ OPUS_COPY(&oldBandE[nbEBands], oldBandE, nbEBands);
+ }
+
+ if (!isTransient)
+ {
+ OPUS_COPY(oldLogE2, oldLogE, CC*nbEBands);
+ OPUS_COPY(oldLogE, oldBandE, CC*nbEBands);
+ } 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<start;i++)
+ {
+ oldBandE[c*nbEBands+i]=0;
+ oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT);
+ }
+ for (i=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_PHASE_INVERSION_DISABLED_REQUEST:
+ {
+ opus_int32 value = va_arg(ap, opus_int32);
+ if(value<0 || value>1)
+ {
+ goto bad_arg;
+ }
+ st->disable_inv = value;
+ }
+ break;
+ case OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST:
+ {
+ opus_int32 *value = va_arg(ap, opus_int32*);
+ if (!value)
+ {
+ goto bad_arg;
+ }
+ *value = st->disable_inv;
+ }
+ break;
+ case OPUS_RESET_STATE:
+ {
+ int i;
+ opus_val16 *oldBandE, *oldLogE, *oldLogE2;
+ oldBandE = (opus_val16*)(st->in_mem+st->channels*(st->mode->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_SET_SILK_INFO_REQUEST:
+ {
+ SILKInfo *info = va_arg(ap, SILKInfo *);
+ if (info)
+ OPUS_COPY(&st->silk_info, 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/lib/rbcodec/codecs/libopus/celt/celt_lpc.c b/lib/rbcodec/codecs/libopus/celt/celt_lpc.c
index fa29d626ea..8ecb693ee9 100644
--- a/lib/rbcodec/codecs/libopus/celt/celt_lpc.c
+++ b/lib/rbcodec/codecs/libopus/celt/celt_lpc.c
@@ -49,8 +49,7 @@ int p
float *lpc = _lpc;
#endif
- for (i = 0; i < p; i++)
- lpc[i] = 0;
+ OPUS_CLEAR(lpc, p);
if (ac[0] != 0)
{
for (i = 0; i < p; i++) {
@@ -88,56 +87,42 @@ int p
#endif
}
-void celt_fir(const opus_val16 *_x,
+
+void celt_fir_c(
+ const opus_val16 *x,
const opus_val16 *num,
- opus_val16 *_y,
+ opus_val16 *y,
int N,
int ord,
- opus_val16 *mem)
+ int arch)
{
int i,j;
VARDECL(opus_val16, rnum);
- VARDECL(opus_val16, x);
SAVE_STACK;
-
+ celt_assert(x != y);
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)));
+ opus_val32 sum[4];
+ sum[0] = SHL32(EXTEND32(x[i ]), SIG_SHIFT);
+ sum[1] = SHL32(EXTEND32(x[i+1]), SIG_SHIFT);
+ sum[2] = SHL32(EXTEND32(x[i+2]), SIG_SHIFT);
+ sum[3] = SHL32(EXTEND32(x[i+3]), SIG_SHIFT);
+ xcorr_kernel(rnum, x+i-ord, sum, ord, arch);
+ y[i ] = ROUND16(sum[0], SIG_SHIFT);
+ y[i+1] = ROUND16(sum[1], SIG_SHIFT);
+ y[i+2] = ROUND16(sum[2], SIG_SHIFT);
+ y[i+3] = ROUND16(sum[3], SIG_SHIFT);
}
for (;i<N;i++)
{
- opus_val32 sum = 0;
+ 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(ADD32(EXTEND32(_x[i]), PSHR32(sum, SIG_SHIFT)));
+ sum = MAC16_16(sum,rnum[j],x[i+j-ord]);
+ y[i] = ROUND16(sum, SIG_SHIFT);
}
-#endif
RESTORE_STACK;
}
@@ -146,10 +131,12 @@ void celt_iir(const opus_val32 *_x,
opus_val32 *_y,
int N,
int ord,
- opus_val16 *mem)
+ opus_val16 *mem,
+ int arch)
{
#ifdef SMALL_FOOTPRINT
int i,j;
+ (void)arch;
for (i=0;i<N;i++)
{
opus_val32 sum = _x[i];
@@ -161,7 +148,7 @@ void celt_iir(const opus_val32 *_x,
{
mem[j]=mem[j-1];
}
- mem[0] = ROUND16(sum,SIG_SHIFT);
+ mem[0] = SROUND16(sum, SIG_SHIFT);
_y[i] = sum;
}
#else
@@ -187,23 +174,23 @@ void celt_iir(const opus_val32 *_x,
sum[1]=_x[i+1];
sum[2]=_x[i+2];
sum[3]=_x[i+3];
- xcorr_kernel(rden, y+i, sum, ord);
+ xcorr_kernel(rden, y+i, sum, ord, arch);
/* Patch up the result to compensate for the fact that this is an IIR */
- y[i+ord ] = -ROUND16(sum[0],SIG_SHIFT);
+ y[i+ord ] = -SROUND16(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+ord+1] = -SROUND16(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+ord+2] = -SROUND16(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+ord+3] = -SROUND16(sum[3],SIG_SHIFT);
_y[i+3] = sum[3];
}
for (;i<N;i++)
@@ -211,7 +198,7 @@ void celt_iir(const opus_val32 *_x,
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+ord] = SROUND16(sum,SIG_SHIFT);
_y[i] = sum;
}
for(i=0;i<ord;i++)
diff --git a/lib/rbcodec/codecs/libopus/celt/celt_lpc.h b/lib/rbcodec/codecs/libopus/celt/celt_lpc.h
index dc2a0a3d26..a4c5fd6ea5 100644
--- a/lib/rbcodec/codecs/libopus/celt/celt_lpc.h
+++ b/lib/rbcodec/codecs/libopus/celt/celt_lpc.h
@@ -29,24 +29,36 @@
#define PLC_H
#include "arch.h"
+#include "cpu_support.h"
+
+#if defined(OPUS_X86_MAY_HAVE_SSE4_1)
+#include "x86/celt_lpc_sse.h"
+#endif
#define LPC_ORDER 24
void _celt_lpc(opus_val16 *_lpc, const opus_val32 *ac, int p);
-void celt_fir(const opus_val16 *x,
+void celt_fir_c(
+ const opus_val16 *x,
const opus_val16 *num,
opus_val16 *y,
int N,
int ord,
- opus_val16 *mem);
+ int arch);
+
+#if !defined(OVERRIDE_CELT_FIR)
+#define celt_fir(x, num, y, N, ord, arch) \
+ (celt_fir_c(x, num, y, N, ord, arch))
+#endif
void celt_iir(const opus_val32 *x,
const opus_val16 *den,
opus_val32 *y,
int N,
int ord,
- opus_val16 *mem);
+ opus_val16 *mem,
+ int arch);
int _celt_autocorr(const opus_val16 *x, opus_val32 *ac,
const opus_val16 *window, int overlap, int lag, int n, int arch);
diff --git a/lib/rbcodec/codecs/libopus/celt/cpu_support.h b/lib/rbcodec/codecs/libopus/celt/cpu_support.h
index d68dbe62c5..68fc60678f 100644
--- a/lib/rbcodec/codecs/libopus/celt/cpu_support.h
+++ b/lib/rbcodec/codecs/libopus/celt/cpu_support.h
@@ -31,7 +31,8 @@
#include "opus_types.h"
#include "opus_defines.h"
-#if defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_ASM)
+#if defined(OPUS_HAVE_RTCD) && \
+ (defined(OPUS_ARM_ASM) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR))
#include "arm/armcpu.h"
/* We currently support 4 ARM variants:
@@ -42,6 +43,22 @@
*/
#define OPUS_ARCHMASK 3
+#elif (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE)) || \
+ (defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2)) || \
+ (defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)) || \
+ (defined(OPUS_X86_MAY_HAVE_AVX) && !defined(OPUS_X86_PRESUME_AVX))
+
+#include "x86/x86cpu.h"
+/* We currently support 5 x86 variants:
+ * arch[0] -> non-sse
+ * arch[1] -> sse
+ * arch[2] -> sse2
+ * arch[3] -> sse4.1
+ * arch[4] -> avx
+ */
+#define OPUS_ARCHMASK 7
+int opus_select_arch(void);
+
#else
#define OPUS_ARCHMASK 0
@@ -50,5 +67,4 @@ static OPUS_INLINE int opus_select_arch(void)
return 0;
}
#endif
-
#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/cwrs.c b/lib/rbcodec/codecs/libopus/celt/cwrs.c
index 031a875995..a552e4f0fb 100644
--- a/lib/rbcodec/codecs/libopus/celt/cwrs.c
+++ b/lib/rbcodec/codecs/libopus/celt/cwrs.c
@@ -74,7 +74,7 @@ int log2_frac(opus_uint32 val, int frac)
/*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
+ https://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
@@ -210,7 +210,7 @@ int log2_frac(opus_uint32 val, int frac)
#if defined(CUSTOM_MODES)
static const opus_uint32 CELT_PVQ_U_DATA[1488]={
#else
-static const opus_uint32 CELT_PVQ_U_DATA[1272] ICONST_ATTR ={
+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,
@@ -418,7 +418,7 @@ static const opus_uint32 *const CELT_PVQ_U_ROW[15]={
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] ICONST_ATTR ={
+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,
@@ -482,7 +482,7 @@ static opus_val32 cwrsi(int _n,int _k,opus_uint32 _i,int *_y){
k0=_k;
q=row[_n];
if(q>_i){
- celt_assert(p>q);
+ celt_sig_assert(p>q);
_k=_n;
do p=CELT_PVQ_U_ROW[--_k][_n];
while(p>_i);
diff --git a/lib/rbcodec/codecs/libopus/celt/dump_modes/Makefile b/lib/rbcodec/codecs/libopus/celt/dump_modes/Makefile
new file mode 100644
index 0000000000..93f599fb5b
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/dump_modes/Makefile
@@ -0,0 +1,32 @@
+
+CFLAGS=-O2 -Wall -Wextra -DHAVE_CONFIG_H
+INCLUDES=-I. -I../ -I../.. -I../../include
+
+SOURCES = dump_modes.c \
+ ../modes.c \
+ ../cwrs.c \
+ ../rate.c \
+ ../entcode.c \
+ ../entenc.c \
+ ../entdec.c \
+ ../mathops.c \
+ ../mdct.c \
+ ../kiss_fft.c
+
+ifdef HAVE_ARM_NE10
+CC = gcc
+CFLAGS += -mfpu=neon
+INCLUDES += -I$(NE10_INCDIR) -DHAVE_ARM_NE10 -DOPUS_ARM_PRESUME_NEON_INTR
+LIBS = -L$(NE10_LIBDIR) -lNE10
+SOURCES += ../arm/celt_ne10_fft.c \
+ dump_modes_arm_ne10.c \
+ ../arm/armcpu.c
+endif
+
+all: dump_modes
+
+dump_modes:
+ $(PREFIX)$(CC) $(CFLAGS) $(INCLUDES) -DCUSTOM_MODES_ONLY -DCUSTOM_MODES $(SOURCES) -o $@ $(LIBS) -lm
+
+clean:
+ rm -f dump_modes
diff --git a/lib/rbcodec/codecs/libopus/celt/dump_modes/dump_modes.c b/lib/rbcodec/codecs/libopus/celt/dump_modes/dump_modes.c
new file mode 100644
index 0000000000..9105a5344e
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/dump_modes/dump_modes.c
@@ -0,0 +1,353 @@
+/* Copyright (c) 2008 CSIRO
+ Copyright (c) 2008-2009 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 <stdlib.h>
+#include <stdio.h>
+#include "modes.h"
+#include "celt.h"
+#include "rate.h"
+#include "dump_modes_arch.h"
+
+#define INT16 "%d"
+#define INT32 "%d"
+#define FLOAT "%#0.8gf"
+
+#ifdef FIXED_POINT
+#define WORD16 INT16
+#define WORD32 INT32
+#else
+#define WORD16 FLOAT
+#define WORD32 FLOAT
+#endif
+
+void dump_modes(FILE *file, CELTMode **modes, int nb_modes)
+{
+ int i, j, k;
+ int mdct_twiddles_size;
+ fprintf(file, "/* The contents of this file was automatically generated by dump_modes.c\n");
+ fprintf(file, " with arguments:");
+ for (i=0;i<nb_modes;i++)
+ {
+ CELTMode *mode = modes[i];
+ fprintf(file, " %d %d",mode->Fs,mode->shortMdctSize*mode->nbShortMdcts);
+ }
+ fprintf(file, "\n It contains static definitions for some pre-defined modes. */\n");
+ fprintf(file, "#include \"modes.h\"\n");
+ fprintf(file, "#include \"rate.h\"\n");
+ fprintf(file, "\n#ifdef HAVE_ARM_NE10\n");
+ fprintf(file, "#define OVERRIDE_FFT 1\n");
+ fprintf(file, "#include \"%s\"\n", ARM_NE10_ARCH_FILE_NAME);
+ fprintf(file, "#endif\n");
+
+ fprintf(file, "\n");
+
+ for (i=0;i<nb_modes;i++)
+ {
+ CELTMode *mode = modes[i];
+ int mdctSize;
+ int standard, framerate;
+
+ mdctSize = mode->shortMdctSize*mode->nbShortMdcts;
+ standard = (mode->Fs == 400*(opus_int32)mode->shortMdctSize);
+ framerate = mode->Fs/mode->shortMdctSize;
+
+ if (!standard)
+ {
+ fprintf(file, "#ifndef DEF_EBANDS%d_%d\n", mode->Fs, mdctSize);
+ fprintf(file, "#define DEF_EBANDS%d_%d\n", mode->Fs, mdctSize);
+ fprintf (file, "static const opus_int16 eBands%d_%d[%d] = {\n", mode->Fs, mdctSize, mode->nbEBands+2);
+ for (j=0;j<mode->nbEBands+2;j++)
+ fprintf (file, "%d, ", mode->eBands[j]);
+ fprintf (file, "};\n");
+ fprintf(file, "#endif\n");
+ fprintf(file, "\n");
+ }
+
+ fprintf(file, "#ifndef DEF_WINDOW%d\n", mode->overlap);
+ fprintf(file, "#define DEF_WINDOW%d\n", mode->overlap);
+ fprintf (file, "static const opus_val16 window%d[%d] = {\n", mode->overlap, mode->overlap);
+ for (j=0;j<mode->overlap;j++)
+ fprintf (file, WORD16 ",%c", mode->window[j],(j+6)%5==0?'\n':' ');
+ fprintf (file, "};\n");
+ fprintf(file, "#endif\n");
+ fprintf(file, "\n");
+
+ if (!standard)
+ {
+ fprintf(file, "#ifndef DEF_ALLOC_VECTORS%d_%d\n", mode->Fs, mdctSize);
+ fprintf(file, "#define DEF_ALLOC_VECTORS%d_%d\n", mode->Fs, mdctSize);
+ fprintf (file, "static const unsigned char allocVectors%d_%d[%d] = {\n", mode->Fs, mdctSize, mode->nbEBands*mode->nbAllocVectors);
+ for (j=0;j<mode->nbAllocVectors;j++)
+ {
+ for (k=0;k<mode->nbEBands;k++)
+ fprintf (file, "%2d, ", mode->allocVectors[j*mode->nbEBands+k]);
+ fprintf (file, "\n");
+ }
+ fprintf (file, "};\n");
+ fprintf(file, "#endif\n");
+ fprintf(file, "\n");
+ }
+
+ fprintf(file, "#ifndef DEF_LOGN%d\n", framerate);
+ fprintf(file, "#define DEF_LOGN%d\n", framerate);
+ fprintf (file, "static const opus_int16 logN%d[%d] = {\n", framerate, mode->nbEBands);
+ for (j=0;j<mode->nbEBands;j++)
+ fprintf (file, "%d, ", mode->logN[j]);
+ fprintf (file, "};\n");
+ fprintf(file, "#endif\n");
+ fprintf(file, "\n");
+
+ /* Pulse cache */
+ fprintf(file, "#ifndef DEF_PULSE_CACHE%d\n", mode->Fs/mdctSize);
+ fprintf(file, "#define DEF_PULSE_CACHE%d\n", mode->Fs/mdctSize);
+ fprintf (file, "static const opus_int16 cache_index%d[%d] = {\n", mode->Fs/mdctSize, (mode->maxLM+2)*mode->nbEBands);
+ for (j=0;j<mode->nbEBands*(mode->maxLM+2);j++)
+ fprintf (file, "%d,%c", mode->cache.index[j],(j+16)%15==0?'\n':' ');
+ fprintf (file, "};\n");
+ fprintf (file, "static const unsigned char cache_bits%d[%d] = {\n", mode->Fs/mdctSize, mode->cache.size);
+ for (j=0;j<mode->cache.size;j++)
+ fprintf (file, "%d,%c", mode->cache.bits[j],(j+16)%15==0?'\n':' ');
+ fprintf (file, "};\n");
+ fprintf (file, "static const unsigned char cache_caps%d[%d] = {\n", mode->Fs/mdctSize, (mode->maxLM+1)*2*mode->nbEBands);
+ for (j=0;j<(mode->maxLM+1)*2*mode->nbEBands;j++)
+ fprintf (file, "%d,%c", mode->cache.caps[j],(j+16)%15==0?'\n':' ');
+ fprintf (file, "};\n");
+
+ fprintf(file, "#endif\n");
+ fprintf(file, "\n");
+
+ /* FFT twiddles */
+ fprintf(file, "#ifndef FFT_TWIDDLES%d_%d\n", mode->Fs, mdctSize);
+ fprintf(file, "#define FFT_TWIDDLES%d_%d\n", mode->Fs, mdctSize);
+ fprintf (file, "static const kiss_twiddle_cpx fft_twiddles%d_%d[%d] = {\n",
+ mode->Fs, mdctSize, mode->mdct.kfft[0]->nfft);
+ for (j=0;j<mode->mdct.kfft[0]->nfft;j++)
+ fprintf (file, "{" WORD16 ", " WORD16 "},%c", mode->mdct.kfft[0]->twiddles[j].r, mode->mdct.kfft[0]->twiddles[j].i,(j+3)%2==0?'\n':' ');
+ fprintf (file, "};\n");
+
+#ifdef OVERRIDE_FFT
+ dump_mode_arch(mode);
+#endif
+ /* FFT Bitrev tables */
+ for (k=0;k<=mode->mdct.maxshift;k++)
+ {
+ fprintf(file, "#ifndef FFT_BITREV%d\n", mode->mdct.kfft[k]->nfft);
+ fprintf(file, "#define FFT_BITREV%d\n", mode->mdct.kfft[k]->nfft);
+ fprintf (file, "static const opus_int16 fft_bitrev%d[%d] = {\n",
+ mode->mdct.kfft[k]->nfft, mode->mdct.kfft[k]->nfft);
+ for (j=0;j<mode->mdct.kfft[k]->nfft;j++)
+ fprintf (file, "%d,%c", mode->mdct.kfft[k]->bitrev[j],(j+16)%15==0?'\n':' ');
+ fprintf (file, "};\n");
+
+ fprintf(file, "#endif\n");
+ fprintf(file, "\n");
+ }
+
+ /* FFT States */
+ for (k=0;k<=mode->mdct.maxshift;k++)
+ {
+ fprintf(file, "#ifndef FFT_STATE%d_%d_%d\n", mode->Fs, mdctSize, k);
+ fprintf(file, "#define FFT_STATE%d_%d_%d\n", mode->Fs, mdctSize, k);
+ fprintf (file, "static const kiss_fft_state fft_state%d_%d_%d = {\n",
+ mode->Fs, mdctSize, k);
+ fprintf (file, "%d, /* nfft */\n", mode->mdct.kfft[k]->nfft);
+ fprintf (file, WORD16 ", /* scale */\n", mode->mdct.kfft[k]->scale);
+#ifdef FIXED_POINT
+ fprintf (file, "%d, /* scale_shift */\n", mode->mdct.kfft[k]->scale_shift);
+#endif
+ fprintf (file, "%d, /* shift */\n", mode->mdct.kfft[k]->shift);
+ fprintf (file, "{");
+ for (j=0;j<2*MAXFACTORS;j++)
+ fprintf (file, "%d, ", mode->mdct.kfft[k]->factors[j]);
+ fprintf (file, "}, /* factors */\n");
+ fprintf (file, "fft_bitrev%d, /* bitrev */\n", mode->mdct.kfft[k]->nfft);
+ fprintf (file, "fft_twiddles%d_%d, /* bitrev */\n", mode->Fs, mdctSize);
+
+ fprintf (file, "#ifdef OVERRIDE_FFT\n");
+ fprintf (file, "(arch_fft_state *)&cfg_arch_%d,\n", mode->mdct.kfft[k]->nfft);
+ fprintf (file, "#else\n");
+ fprintf (file, "NULL,\n");
+ fprintf(file, "#endif\n");
+
+ fprintf (file, "};\n");
+
+ fprintf(file, "#endif\n");
+ fprintf(file, "\n");
+ }
+
+ fprintf(file, "#endif\n");
+ fprintf(file, "\n");
+
+ /* MDCT twiddles */
+ mdct_twiddles_size = mode->mdct.n-(mode->mdct.n/2>>mode->mdct.maxshift);
+ fprintf(file, "#ifndef MDCT_TWIDDLES%d\n", mdctSize);
+ fprintf(file, "#define MDCT_TWIDDLES%d\n", mdctSize);
+ fprintf (file, "static const opus_val16 mdct_twiddles%d[%d] = {\n",
+ mdctSize, mdct_twiddles_size);
+ for (j=0;j<mdct_twiddles_size;j++)
+ fprintf (file, WORD16 ",%c", mode->mdct.trig[j],(j+6)%5==0?'\n':' ');
+ fprintf (file, "};\n");
+
+ fprintf(file, "#endif\n");
+ fprintf(file, "\n");
+
+
+ /* Print the actual mode data */
+ fprintf(file, "static const CELTMode mode%d_%d_%d = {\n", mode->Fs, mdctSize, mode->overlap);
+ fprintf(file, INT32 ", /* Fs */\n", mode->Fs);
+ fprintf(file, "%d, /* overlap */\n", mode->overlap);
+ fprintf(file, "%d, /* nbEBands */\n", mode->nbEBands);
+ fprintf(file, "%d, /* effEBands */\n", mode->effEBands);
+ fprintf(file, "{");
+ for (j=0;j<4;j++)
+ fprintf(file, WORD16 ", ", mode->preemph[j]);
+ fprintf(file, "}, /* preemph */\n");
+ if (standard)
+ fprintf(file, "eband5ms, /* eBands */\n");
+ else
+ fprintf(file, "eBands%d_%d, /* eBands */\n", mode->Fs, mdctSize);
+
+ fprintf(file, "%d, /* maxLM */\n", mode->maxLM);
+ fprintf(file, "%d, /* nbShortMdcts */\n", mode->nbShortMdcts);
+ fprintf(file, "%d, /* shortMdctSize */\n", mode->shortMdctSize);
+
+ fprintf(file, "%d, /* nbAllocVectors */\n", mode->nbAllocVectors);
+ if (standard)
+ fprintf(file, "band_allocation, /* allocVectors */\n");
+ else
+ fprintf(file, "allocVectors%d_%d, /* allocVectors */\n", mode->Fs, mdctSize);
+
+ fprintf(file, "logN%d, /* logN */\n", framerate);
+ fprintf(file, "window%d, /* window */\n", mode->overlap);
+ fprintf(file, "{%d, %d, {", mode->mdct.n, mode->mdct.maxshift);
+ for (k=0;k<=mode->mdct.maxshift;k++)
+ fprintf(file, "&fft_state%d_%d_%d, ", mode->Fs, mdctSize, k);
+ fprintf (file, "}, mdct_twiddles%d}, /* mdct */\n", mdctSize);
+
+ fprintf(file, "{%d, cache_index%d, cache_bits%d, cache_caps%d}, /* cache */\n",
+ mode->cache.size, mode->Fs/mdctSize, mode->Fs/mdctSize, mode->Fs/mdctSize);
+ fprintf(file, "};\n");
+ }
+ fprintf(file, "\n");
+ fprintf(file, "/* List of all the available modes */\n");
+ fprintf(file, "#define TOTAL_MODES %d\n", nb_modes);
+ fprintf(file, "static const CELTMode * const static_mode_list[TOTAL_MODES] = {\n");
+ for (i=0;i<nb_modes;i++)
+ {
+ CELTMode *mode = modes[i];
+ int mdctSize;
+ mdctSize = mode->shortMdctSize*mode->nbShortMdcts;
+ fprintf(file, "&mode%d_%d_%d,\n", mode->Fs, mdctSize, mode->overlap);
+ }
+ fprintf(file, "};\n");
+}
+
+void dump_header(FILE *file, CELTMode **modes, int nb_modes)
+{
+ int i;
+ int channels = 0;
+ int frame_size = 0;
+ int overlap = 0;
+ fprintf (file, "/* This header file is generated automatically*/\n");
+ for (i=0;i<nb_modes;i++)
+ {
+ CELTMode *mode = modes[i];
+ if (frame_size==0)
+ frame_size = mode->shortMdctSize*mode->nbShortMdcts;
+ else if (frame_size != mode->shortMdctSize*mode->nbShortMdcts)
+ frame_size = -1;
+ if (overlap==0)
+ overlap = mode->overlap;
+ else if (overlap != mode->overlap)
+ overlap = -1;
+ }
+ if (channels>0)
+ {
+ fprintf (file, "#define CHANNELS(mode) %d\n", channels);
+ if (channels==1)
+ fprintf (file, "#define DISABLE_STEREO\n");
+ }
+ if (frame_size>0)
+ {
+ fprintf (file, "#define FRAMESIZE(mode) %d\n", frame_size);
+ }
+ if (overlap>0)
+ {
+ fprintf (file, "#define OVERLAP(mode) %d\n", overlap);
+ }
+}
+
+#ifdef FIXED_POINT
+#define BASENAME "static_modes_fixed"
+#else
+#define BASENAME "static_modes_float"
+#endif
+
+int main(int argc, char **argv)
+{
+ int i, nb;
+ FILE *file;
+ CELTMode **m;
+ if (argc%2 != 1 || argc<3)
+ {
+ fprintf (stderr, "Usage: %s rate frame_size [rate frame_size] [rate frame_size]...\n",argv[0]);
+ return 1;
+ }
+ nb = (argc-1)/2;
+ m = malloc(nb*sizeof(CELTMode*));
+ for (i=0;i<nb;i++)
+ {
+ int Fs, frame;
+ Fs = atoi(argv[2*i+1]);
+ frame = atoi(argv[2*i+2]);
+ m[i] = opus_custom_mode_create(Fs, frame, NULL);
+ if (m[i]==NULL)
+ {
+ fprintf(stderr,"Error creating mode with Fs=%s, frame_size=%s\n",
+ argv[2*i+1],argv[2*i+2]);
+ return EXIT_FAILURE;
+ }
+ }
+ file = fopen(BASENAME ".h", "w");
+#ifdef OVERRIDE_FFT
+ dump_modes_arch_init(m, nb);
+#endif
+ dump_modes(file, m, nb);
+ fclose(file);
+#ifdef OVERRIDE_FFT
+ dump_modes_arch_finalize();
+#endif
+ for (i=0;i<nb;i++)
+ opus_custom_mode_destroy(m[i]);
+ free(m);
+ return 0;
+}
diff --git a/lib/rbcodec/codecs/libopus/celt/dump_modes/dump_modes_arch.h b/lib/rbcodec/codecs/libopus/celt/dump_modes/dump_modes_arch.h
new file mode 100644
index 0000000000..cc0d4be1ec
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/dump_modes/dump_modes_arch.h
@@ -0,0 +1,45 @@
+/* Copyright (c) 2015 Xiph.Org Foundation
+ Written by Viswanath Puttagunta */
+/*
+ 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 DUMP_MODE_ARCH_H
+#define DUMP_MODE_ARCH_H
+
+void dump_modes_arch_init();
+void dump_mode_arch(CELTMode *mode);
+void dump_modes_arch_finalize();
+
+#if !defined(FIXED_POINT)
+#define ARM_NE10_ARCH_FILE_NAME "static_modes_float_arm_ne10.h"
+#else
+#define ARM_NE10_ARCH_FILE_NAME "static_modes_fixed_arm_ne10.h"
+#endif
+
+#if defined(HAVE_ARM_NE10)
+#define OVERRIDE_FFT (1)
+#endif
+
+#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/dump_modes/dump_modes_arm_ne10.c b/lib/rbcodec/codecs/libopus/celt/dump_modes/dump_modes_arm_ne10.c
new file mode 100644
index 0000000000..828e7b9fff
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/dump_modes/dump_modes_arm_ne10.c
@@ -0,0 +1,152 @@
+/* Copyright (c) 2015 Xiph.Org Foundation
+ Written by Viswanath Puttagunta */
+/*
+ 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(HAVE_CONFIG_H)
+# include "config.h"
+#endif
+
+#include <stdio.h>
+#include <stdlib.h>
+#include "modes.h"
+#include "dump_modes_arch.h"
+#include <NE10_dsp.h>
+
+#if !defined(FIXED_POINT)
+# define NE10_FFT_CFG_TYPE_T ne10_fft_cfg_float32_t
+# define NE10_FFT_CPX_TYPE_T_STR "ne10_fft_cpx_float32_t"
+# define NE10_FFT_STATE_TYPE_T_STR "ne10_fft_state_float32_t"
+#else
+# define NE10_FFT_CFG_TYPE_T ne10_fft_cfg_int32_t
+# define NE10_FFT_CPX_TYPE_T_STR "ne10_fft_cpx_int32_t"
+# define NE10_FFT_STATE_TYPE_T_STR "ne10_fft_state_int32_t"
+#endif
+
+static FILE *file;
+
+void dump_modes_arch_init(CELTMode **modes, int nb_modes)
+{
+ int i;
+
+ file = fopen(ARM_NE10_ARCH_FILE_NAME, "w");
+ fprintf(file, "/* The contents of this file was automatically generated by\n");
+ fprintf(file, " * dump_mode_arm_ne10.c with arguments:");
+ for (i=0;i<nb_modes;i++)
+ {
+ CELTMode *mode = modes[i];
+ fprintf(file, " %d %d",mode->Fs,mode->shortMdctSize*mode->nbShortMdcts);
+ }
+ fprintf(file, "\n * It contains static definitions for some pre-defined modes. */\n");
+ fprintf(file, "#include <NE10_types.h>\n\n");
+}
+
+void dump_modes_arch_finalize()
+{
+ fclose(file);
+}
+
+void dump_mode_arch(CELTMode *mode)
+{
+ int k, j;
+ int mdctSize;
+
+ mdctSize = mode->shortMdctSize*mode->nbShortMdcts;
+
+ fprintf(file, "#ifndef NE10_FFT_PARAMS%d_%d\n", mode->Fs, mdctSize);
+ fprintf(file, "#define NE10_FFT_PARAMS%d_%d\n", mode->Fs, mdctSize);
+ /* cfg->factors */
+ for(k=0;k<=mode->mdct.maxshift;k++) {
+ NE10_FFT_CFG_TYPE_T cfg;
+ cfg = (NE10_FFT_CFG_TYPE_T)mode->mdct.kfft[k]->arch_fft->priv;
+ if (!cfg)
+ continue;
+ fprintf(file, "static const ne10_int32_t ne10_factors_%d[%d] = {\n",
+ mode->mdct.kfft[k]->nfft, (NE10_MAXFACTORS * 2));
+ for(j=0;j<(NE10_MAXFACTORS * 2);j++) {
+ fprintf(file, "%d,%c", cfg->factors[j],(j+16)%15==0?'\n':' ');
+ }
+ fprintf (file, "};\n");
+ }
+
+ /* cfg->twiddles */
+ for(k=0;k<=mode->mdct.maxshift;k++) {
+ NE10_FFT_CFG_TYPE_T cfg;
+ cfg = (NE10_FFT_CFG_TYPE_T)mode->mdct.kfft[k]->arch_fft->priv;
+ if (!cfg)
+ continue;
+ fprintf(file, "static const %s ne10_twiddles_%d[%d] = {\n",
+ NE10_FFT_CPX_TYPE_T_STR, mode->mdct.kfft[k]->nfft,
+ mode->mdct.kfft[k]->nfft);
+ for(j=0;j<mode->mdct.kfft[k]->nfft;j++) {
+#if !defined(FIXED_POINT)
+ fprintf(file, "{%#0.8gf,%#0.8gf},%c",
+ cfg->twiddles[j].r, cfg->twiddles[j].i,(j+4)%3==0?'\n':' ');
+#else
+ fprintf(file, "{%d,%d},%c",
+ cfg->twiddles[j].r, cfg->twiddles[j].i,(j+4)%3==0?'\n':' ');
+#endif
+ }
+ fprintf (file, "};\n");
+ }
+
+ for(k=0;k<=mode->mdct.maxshift;k++) {
+ NE10_FFT_CFG_TYPE_T cfg;
+ cfg = (NE10_FFT_CFG_TYPE_T)mode->mdct.kfft[k]->arch_fft->priv;
+ if (!cfg) {
+ fprintf(file, "/* Ne10 does not support scaled FFT for length = %d */\n",
+ mode->mdct.kfft[k]->nfft);
+ fprintf(file, "static const arch_fft_state cfg_arch_%d = {\n", mode->mdct.kfft[k]->nfft);
+ fprintf(file, "0,\n");
+ fprintf(file, "NULL\n");
+ fprintf(file, "};\n");
+ continue;
+ }
+ fprintf(file, "static const %s %s_%d = {\n", NE10_FFT_STATE_TYPE_T_STR,
+ NE10_FFT_STATE_TYPE_T_STR, mode->mdct.kfft[k]->nfft);
+ fprintf(file, "%d,\n", cfg->nfft);
+ fprintf(file, "(ne10_int32_t *)ne10_factors_%d,\n", mode->mdct.kfft[k]->nfft);
+ fprintf(file, "(%s *)ne10_twiddles_%d,\n",
+ NE10_FFT_CPX_TYPE_T_STR, mode->mdct.kfft[k]->nfft);
+ fprintf(file, "NULL,\n"); /* buffer */
+ fprintf(file, "(%s *)&ne10_twiddles_%d[%d],\n",
+ NE10_FFT_CPX_TYPE_T_STR, mode->mdct.kfft[k]->nfft, cfg->nfft);
+#if !defined(FIXED_POINT)
+ fprintf(file, "/* is_forward_scaled = true */\n");
+ fprintf(file, "(ne10_int32_t) 1,\n");
+ fprintf(file, "/* is_backward_scaled = false */\n");
+ fprintf(file, "(ne10_int32_t) 0,\n");
+#endif
+ fprintf(file, "};\n");
+
+ fprintf(file, "static const arch_fft_state cfg_arch_%d = {\n",
+ mode->mdct.kfft[k]->nfft);
+ fprintf(file, "1,\n");
+ fprintf(file, "(void *)&%s_%d,\n",
+ NE10_FFT_STATE_TYPE_T_STR, mode->mdct.kfft[k]->nfft);
+ fprintf(file, "};\n\n");
+ }
+ fprintf(file, "#endif /* end NE10_FFT_PARAMS%d_%d */\n", mode->Fs, mdctSize);
+}
diff --git a/lib/rbcodec/codecs/libopus/celt/entcode.c b/lib/rbcodec/codecs/libopus/celt/entcode.c
index 461a36dd55..70f32016ec 100644
--- a/lib/rbcodec/codecs/libopus/celt/entcode.c
+++ b/lib/rbcodec/codecs/libopus/celt/entcode.c
@@ -116,7 +116,7 @@ opus_uint32 ec_tell_frac(ec_ctx *_this){
#ifdef USE_SMALL_DIV_TABLE
/* Result of 2^32/(2*i+1), except for i=0. */
-const opus_uint32 SMALL_DIV_TABLE[129] ICONST_ATTR = {
+const opus_uint32 SMALL_DIV_TABLE[129] = {
0xFFFFFFFF, 0x55555555, 0x33333333, 0x24924924,
0x1C71C71C, 0x1745D174, 0x13B13B13, 0x11111111,
0x0F0F0F0F, 0x0D79435E, 0x0C30C30C, 0x0B21642C,
diff --git a/lib/rbcodec/codecs/libopus/celt/entcode.h b/lib/rbcodec/codecs/libopus/celt/entcode.h
index 13d6c84ef0..3763e3f284 100644
--- a/lib/rbcodec/codecs/libopus/celt/entcode.h
+++ b/lib/rbcodec/codecs/libopus/celt/entcode.h
@@ -122,7 +122,7 @@ opus_uint32 ec_tell_frac(ec_ctx *_this);
/* Tested exhaustively for all n and for 1<=d<=256 */
static OPUS_INLINE opus_uint32 celt_udiv(opus_uint32 n, opus_uint32 d) {
- celt_assert(d>0);
+ celt_sig_assert(d>0);
#ifdef USE_SMALL_DIV_TABLE
if (d>256)
return n/d;
@@ -138,7 +138,7 @@ static OPUS_INLINE opus_uint32 celt_udiv(opus_uint32 n, opus_uint32 d) {
}
static OPUS_INLINE opus_int32 celt_sudiv(opus_int32 n, opus_int32 d) {
- celt_assert(d>0);
+ celt_sig_assert(d>0);
#ifdef USE_SMALL_DIV_TABLE
if (n<0)
return -(opus_int32)celt_udiv(-n, d);
diff --git a/lib/rbcodec/codecs/libopus/celt/entdec.h b/lib/rbcodec/codecs/libopus/celt/entdec.h
index d8ab318730..025fc1870d 100644
--- a/lib/rbcodec/codecs/libopus/celt/entdec.h
+++ b/lib/rbcodec/codecs/libopus/celt/entdec.h
@@ -85,7 +85,7 @@ int ec_dec_icdf(ec_dec *_this,const unsigned char *_icdf,unsigned _ftb);
The bits must have been encoded with ec_enc_uint().
No call to ec_dec_update() is necessary after this call.
_ft: The number of integers that can be decoded (one more than the max).
- This must be at least one, and no more than 2**32-1.
+ This must be at least 2, and no more than 2**32-1.
Return: The decoded bits.*/
opus_uint32 ec_dec_uint(ec_dec *_this,opus_uint32 _ft);
diff --git a/lib/rbcodec/codecs/libopus/celt/entenc.c b/lib/rbcodec/codecs/libopus/celt/entenc.c
index 271e4d30c5..f1750d25b8 100644
--- a/lib/rbcodec/codecs/libopus/celt/entenc.c
+++ b/lib/rbcodec/codecs/libopus/celt/entenc.c
@@ -98,7 +98,7 @@ static void ec_enc_carry_out(ec_enc *_this,int _c){
else _this->ext++;
}
-static void ec_enc_normalize(ec_enc *_this){
+static OPUS_INLINE void ec_enc_normalize(ec_enc *_this){
/*If the range is too small, output some bits and rescale it.*/
while(_this->rng<=EC_CODE_BOT){
ec_enc_carry_out(_this,(int)(_this->val>>EC_CODE_SHIFT));
diff --git a/lib/rbcodec/codecs/libopus/celt/entenc.h b/lib/rbcodec/codecs/libopus/celt/entenc.h
index 796bc4d572..f502eaf662 100644
--- a/lib/rbcodec/codecs/libopus/celt/entenc.h
+++ b/lib/rbcodec/codecs/libopus/celt/entenc.h
@@ -67,7 +67,7 @@ void ec_enc_icdf(ec_enc *_this,int _s,const unsigned char *_icdf,unsigned _ftb);
/*Encodes a raw unsigned integer in the stream.
_fl: The integer to encode.
_ft: The number of integers that can be encoded (one more than the max).
- This must be at least one, and no more than 2**32-1.*/
+ This must be at least 2, and no more than 2**32-1.*/
void ec_enc_uint(ec_enc *_this,opus_uint32 _fl,opus_uint32 _ft);
/*Encodes a sequence of raw bits in the stream.
diff --git a/lib/rbcodec/codecs/libopus/celt/fixed_c5x.h b/lib/rbcodec/codecs/libopus/celt/fixed_c5x.h
new file mode 100644
index 0000000000..ea95a998c3
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/fixed_c5x.h
@@ -0,0 +1,79 @@
+/* Copyright (C) 2003 Jean-Marc Valin */
+/**
+ @file fixed_c5x.h
+ @brief Fixed-point operations for the TI C5x DSP family
+*/
+/*
+ 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_C5X_H
+#define FIXED_C5X_H
+
+#include "dsplib.h"
+
+#undef IMUL32
+static OPUS_INLINE long IMUL32(long i, long j)
+{
+ long ac0, ac1;
+ ac0 = _lmpy(i>>16,j);
+ ac1 = ac0 + _lmpy(i,j>>16);
+ return _lmpyu(i,j) + (ac1<<16);
+}
+
+#undef MAX16
+#define MAX16(a,b) _max(a,b)
+
+#undef MIN16
+#define MIN16(a,b) _min(a,b)
+
+#undef MAX32
+#define MAX32(a,b) _lmax(a,b)
+
+#undef MIN32
+#define MIN32(a,b) _lmin(a,b)
+
+#undef VSHR32
+#define VSHR32(a, shift) _lshl(a,-(shift))
+
+#undef MULT16_16_Q15
+#define MULT16_16_Q15(a,b) (_smpy(a,b))
+
+#undef MULT16_16SU
+#define MULT16_16SU(a,b) _lmpysu(a,b)
+
+#undef MULT_16_16
+#define MULT_16_16(a,b) _lmpy(a,b)
+
+/* FIXME: This is technically incorrect and is bound to cause problems. Is there any cleaner solution? */
+#undef MULT16_32_Q15
+#define MULT16_32_Q15(a,b) ADD32(SHL(MULT16_16((a),SHR((b),16)),1), SHR(MULT16_16SU((a),(b)),15))
+
+#define celt_ilog2(x) (30 - _lnorm(x))
+#define OVERRIDE_CELT_ILOG2
+
+#define celt_maxabs16(x, len) MAX32(EXTEND32(maxval((DATA *)x, len)),-EXTEND32(minval((DATA *)x, len)))
+#define OVERRIDE_CELT_MAXABS16
+
+#endif /* FIXED_C5X_H */
diff --git a/lib/rbcodec/codecs/libopus/celt/fixed_c6x.h b/lib/rbcodec/codecs/libopus/celt/fixed_c6x.h
new file mode 100644
index 0000000000..bb6ad92780
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/fixed_c6x.h
@@ -0,0 +1,70 @@
+/* Copyright (C) 2008 CSIRO */
+/**
+ @file fixed_c6x.h
+ @brief Fixed-point operations for the TI C6x DSP family
+*/
+/*
+ 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_C6X_H
+#define FIXED_C6X_H
+
+#undef MULT16_16SU
+#define MULT16_16SU(a,b) _mpysu(a,b)
+
+#undef MULT_16_16
+#define MULT_16_16(a,b) _mpy(a,b)
+
+#define celt_ilog2(x) (30 - _norm(x))
+#define OVERRIDE_CELT_ILOG2
+
+#undef MULT16_32_Q15
+#define MULT16_32_Q15(a,b) (_mpylill(a, b) >> 15)
+
+#if 0
+#include "dsplib.h"
+
+#undef MAX16
+#define MAX16(a,b) _max(a,b)
+
+#undef MIN16
+#define MIN16(a,b) _min(a,b)
+
+#undef MAX32
+#define MAX32(a,b) _lmax(a,b)
+
+#undef MIN32
+#define MIN32(a,b) _lmin(a,b)
+
+#undef VSHR32
+#define VSHR32(a, shift) _lshl(a,-(shift))
+
+#undef MULT16_16_Q15
+#define MULT16_16_Q15(a,b) (_smpy(a,b))
+
+#define celt_maxabs16(x, len) MAX32(EXTEND32(maxval((DATA *)x, len)),-EXTEND32(minval((DATA *)x, len)))
+#define OVERRIDE_CELT_MAXABS16
+
+#endif /* FIXED_C6X_H */
diff --git a/lib/rbcodec/codecs/libopus/celt/fixed_debug.h b/lib/rbcodec/codecs/libopus/celt/fixed_debug.h
new file mode 100644
index 0000000000..f435295234
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/fixed_debug.h
@@ -0,0 +1,791 @@
+/* Copyright (C) 2003-2008 Jean-Marc Valin
+ Copyright (C) 2007-2012 Xiph.Org Foundation */
+/**
+ @file fixed_debug.h
+ @brief Fixed-point operations with debugging
+*/
+/*
+ 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_DEBUG_H
+#define FIXED_DEBUG_H
+
+#include <stdio.h>
+#include "opus_defines.h"
+
+#ifdef CELT_C
+OPUS_EXPORT opus_int64 celt_mips=0;
+#else
+extern opus_int64 celt_mips;
+#endif
+
+#define MULT16_16SU(a,b) ((opus_val32)(opus_val16)(a)*(opus_val32)(opus_uint16)(b))
+#define MULT32_32_Q31(a,b) ADD32(ADD32(SHL32(MULT16_16(SHR32((a),16),SHR((b),16)),1), SHR32(MULT16_16SU(SHR32((a),16),((b)&0x0000ffff)),15)), SHR32(MULT16_16SU(SHR32((b),16),((a)&0x0000ffff)),15))
+
+/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */
+#define MULT16_32_Q16(a,b) ADD32(MULT16_16((a),SHR32((b),16)), SHR32(MULT16_16SU((a),((b)&0x0000ffff)),16))
+
+#define MULT16_32_P16(a,b) MULT16_32_PX(a,b,16)
+
+#define QCONST16(x,bits) ((opus_val16)(.5+(x)*(((opus_val32)1)<<(bits))))
+#define QCONST32(x,bits) ((opus_val32)(.5+(x)*(((opus_val32)1)<<(bits))))
+
+#define VERIFY_SHORT(x) ((x)<=32767&&(x)>=-32768)
+#define VERIFY_INT(x) ((x)<=2147483647LL&&(x)>=-2147483648LL)
+#define VERIFY_UINT(x) ((x)<=(2147483647LLU<<1))
+
+#define SHR(a,b) SHR32(a,b)
+#define PSHR(a,b) PSHR32(a,b)
+
+/** Add two 32-bit values, ignore any overflows */
+#define ADD32_ovflw(a,b) (celt_mips+=2,(opus_val32)((opus_uint32)(a)+(opus_uint32)(b)))
+/** Subtract two 32-bit values, ignore any overflows */
+#define SUB32_ovflw(a,b) (celt_mips+=2,(opus_val32)((opus_uint32)(a)-(opus_uint32)(b)))
+/* Avoid MSVC warning C4146: unary minus operator applied to unsigned type */
+/** Negate 32-bit value, ignore any overflows */
+#define NEG32_ovflw(a) (celt_mips+=2,(opus_val32)(0-(opus_uint32)(a)))
+
+static OPUS_INLINE short NEG16(int x)
+{
+ int res;
+ if (!VERIFY_SHORT(x))
+ {
+ fprintf (stderr, "NEG16: input is not short: %d\n", (int)x);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = -x;
+ if (!VERIFY_SHORT(res))
+ {
+ fprintf (stderr, "NEG16: output is not short: %d\n", (int)res);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips++;
+ return res;
+}
+static OPUS_INLINE int NEG32(opus_int64 x)
+{
+ opus_int64 res;
+ if (!VERIFY_INT(x))
+ {
+ fprintf (stderr, "NEG16: input is not int: %d\n", (int)x);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = -x;
+ if (!VERIFY_INT(res))
+ {
+ fprintf (stderr, "NEG16: output is not int: %d\n", (int)res);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=2;
+ return res;
+}
+
+#define EXTRACT16(x) EXTRACT16_(x, __FILE__, __LINE__)
+static OPUS_INLINE short EXTRACT16_(int x, char *file, int line)
+{
+ int res;
+ if (!VERIFY_SHORT(x))
+ {
+ fprintf (stderr, "EXTRACT16: input is not short: %d in %s: line %d\n", x, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = x;
+ celt_mips++;
+ return res;
+}
+
+#define EXTEND32(x) EXTEND32_(x, __FILE__, __LINE__)
+static OPUS_INLINE int EXTEND32_(int x, char *file, int line)
+{
+ int res;
+ if (!VERIFY_SHORT(x))
+ {
+ fprintf (stderr, "EXTEND32: input is not short: %d in %s: line %d\n", x, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = x;
+ celt_mips++;
+ return res;
+}
+
+#define SHR16(a, shift) SHR16_(a, shift, __FILE__, __LINE__)
+static OPUS_INLINE short SHR16_(int a, int shift, char *file, int line)
+{
+ int res;
+ if (!VERIFY_SHORT(a) || !VERIFY_SHORT(shift))
+ {
+ fprintf (stderr, "SHR16: inputs are not short: %d >> %d in %s: line %d\n", a, shift, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = a>>shift;
+ if (!VERIFY_SHORT(res))
+ {
+ fprintf (stderr, "SHR16: output is not short: %d in %s: line %d\n", res, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips++;
+ return res;
+}
+#define SHL16(a, shift) SHL16_(a, shift, __FILE__, __LINE__)
+static OPUS_INLINE short SHL16_(int a, int shift, char *file, int line)
+{
+ int res;
+ if (!VERIFY_SHORT(a) || !VERIFY_SHORT(shift))
+ {
+ fprintf (stderr, "SHL16: inputs are not short: %d %d in %s: line %d\n", a, shift, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = a<<shift;
+ if (!VERIFY_SHORT(res))
+ {
+ fprintf (stderr, "SHL16: output is not short: %d in %s: line %d\n", res, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips++;
+ return res;
+}
+
+static OPUS_INLINE int SHR32(opus_int64 a, int shift)
+{
+ opus_int64 res;
+ if (!VERIFY_INT(a) || !VERIFY_SHORT(shift))
+ {
+ fprintf (stderr, "SHR32: inputs are not int: %d %d\n", (int)a, shift);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = a>>shift;
+ if (!VERIFY_INT(res))
+ {
+ fprintf (stderr, "SHR32: output is not int: %d\n", (int)res);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=2;
+ return res;
+}
+#define SHL32(a, shift) SHL32_(a, shift, __FILE__, __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))
+ {
+ fprintf (stderr, "SHL32: inputs are not int: %lld %d in %s: line %d\n", a, shift, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = a<<shift;
+ if (!VERIFY_INT(res))
+ {
+ fprintf (stderr, "SHL32: output is not int: %lld<<%d = %lld in %s: line %d\n", a, shift, res, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=2;
+ return res;
+}
+
+#define PSHR32(a,shift) (celt_mips--,SHR32(ADD32((a),(((opus_val32)(1)<<((shift))>>1))),shift))
+#define VSHR32(a, shift) (((shift)>0) ? SHR32(a, shift) : SHL32(a, -(shift)))
+
+#define ROUND16(x,a) (celt_mips--,EXTRACT16(PSHR32((x),(a))))
+#define SROUND16(x,a) (celt_mips--,EXTRACT16(SATURATE(PSHR32(x,a), 32767)));
+
+#define HALF16(x) (SHR16(x,1))
+#define HALF32(x) (SHR32(x,1))
+
+#define ADD16(a, b) ADD16_(a, b, __FILE__, __LINE__)
+static OPUS_INLINE short ADD16_(int a, int b, char *file, int line)
+{
+ int res;
+ if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b))
+ {
+ fprintf (stderr, "ADD16: inputs are not short: %d %d in %s: line %d\n", a, b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = a+b;
+ if (!VERIFY_SHORT(res))
+ {
+ fprintf (stderr, "ADD16: output is not short: %d+%d=%d in %s: line %d\n", a,b,res, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips++;
+ return res;
+}
+
+#define SUB16(a, b) SUB16_(a, b, __FILE__, __LINE__)
+static OPUS_INLINE short SUB16_(int a, int b, char *file, int line)
+{
+ int res;
+ if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b))
+ {
+ fprintf (stderr, "SUB16: inputs are not short: %d %d in %s: line %d\n", a, b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = a-b;
+ if (!VERIFY_SHORT(res))
+ {
+ fprintf (stderr, "SUB16: output is not short: %d in %s: line %d\n", res, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips++;
+ return res;
+}
+
+#define ADD32(a, b) ADD32_(a, b, __FILE__, __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))
+ {
+ fprintf (stderr, "ADD32: inputs are not int: %d %d in %s: line %d\n", (int)a, (int)b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = a+b;
+ if (!VERIFY_INT(res))
+ {
+ fprintf (stderr, "ADD32: output is not int: %d in %s: line %d\n", (int)res, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=2;
+ return res;
+}
+
+#define SUB32(a, b) SUB32_(a, b, __FILE__, __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))
+ {
+ fprintf (stderr, "SUB32: inputs are not int: %d %d in %s: line %d\n", (int)a, (int)b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = a-b;
+ if (!VERIFY_INT(res))
+ {
+ fprintf (stderr, "SUB32: output is not int: %d in %s: line %d\n", (int)res, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=2;
+ return res;
+}
+
+#undef UADD32
+#define UADD32(a, b) UADD32_(a, b, __FILE__, __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))
+ {
+ fprintf (stderr, "UADD32: inputs are not uint32: %llu %llu in %s: line %d\n", a, b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = a+b;
+ if (!VERIFY_UINT(res))
+ {
+ fprintf (stderr, "UADD32: output is not uint32: %llu in %s: line %d\n", res, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=2;
+ return res;
+}
+
+#undef USUB32
+#define USUB32(a, b) USUB32_(a, b, __FILE__, __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))
+ {
+ fprintf (stderr, "USUB32: inputs are not uint32: %llu %llu in %s: line %d\n", a, b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ if (a<b)
+ {
+ fprintf (stderr, "USUB32: inputs underflow: %llu < %llu in %s: line %d\n", a, b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = a-b;
+ if (!VERIFY_UINT(res))
+ {
+ fprintf (stderr, "USUB32: output is not uint32: %llu - %llu = %llu in %s: line %d\n", a, b, res, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=2;
+ return res;
+}
+
+/* result fits in 16 bits */
+static OPUS_INLINE short MULT16_16_16(int a, int b)
+{
+ int res;
+ if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b))
+ {
+ fprintf (stderr, "MULT16_16_16: inputs are not short: %d %d\n", a, b);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = a*b;
+ if (!VERIFY_SHORT(res))
+ {
+ fprintf (stderr, "MULT16_16_16: output is not short: %d\n", res);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips++;
+ return res;
+}
+
+#define MULT16_16(a, b) MULT16_16_(a, b, __FILE__, __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))
+ {
+ fprintf (stderr, "MULT16_16: inputs are not short: %d %d in %s: line %d\n", a, b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = ((opus_int64)a)*b;
+ if (!VERIFY_INT(res))
+ {
+ fprintf (stderr, "MULT16_16: output is not int: %d in %s: line %d\n", (int)res, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips++;
+ return res;
+}
+
+#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 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))
+ {
+ fprintf (stderr, "MULT16_32_Q%d: inputs are not short+int: %d %d in %s: line %d\n", Q, (int)a, (int)b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ if (ABS32(b)>=((opus_val32)(1)<<(15+Q)))
+ {
+ fprintf (stderr, "MULT16_32_Q%d: second operand too large: %d %d in %s: line %d\n", Q, (int)a, (int)b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = (((opus_int64)a)*(opus_int64)b) >> Q;
+ if (!VERIFY_INT(res))
+ {
+ fprintf (stderr, "MULT16_32_Q%d: output is not int: %d*%d=%d in %s: line %d\n", Q, (int)a, (int)b,(int)res, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ if (Q==15)
+ celt_mips+=3;
+ else
+ celt_mips+=4;
+ return res;
+}
+
+#define MULT16_32_PX(a, b, Q) MULT16_32_PX_(a, b, Q, __FILE__, __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))
+ {
+ fprintf (stderr, "MULT16_32_P%d: inputs are not short+int: %d %d in %s: line %d\n\n", Q, (int)a, (int)b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ if (ABS32(b)>=((opus_int64)(1)<<(15+Q)))
+ {
+ fprintf (stderr, "MULT16_32_Q%d: second operand too large: %d %d in %s: line %d\n\n", Q, (int)a, (int)b,file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = ((((opus_int64)a)*(opus_int64)b) + (((opus_val32)(1)<<Q)>>1))>> Q;
+ if (!VERIFY_INT(res))
+ {
+ fprintf (stderr, "MULT16_32_P%d: output is not int: %d*%d=%d in %s: line %d\n\n", Q, (int)a, (int)b,(int)res, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ if (Q==15)
+ celt_mips+=4;
+ else
+ celt_mips+=5;
+ return res;
+}
+
+#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))))
+#define MAC16_32_Q16(c,a,b) (celt_mips-=2,ADD32((c),MULT16_32_Q16((a),(b))))
+
+static OPUS_INLINE int SATURATE(int a, int b)
+{
+ if (a>b)
+ a=b;
+ if (a<-b)
+ a = -b;
+ celt_mips+=3;
+ return a;
+}
+
+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))
+ {
+ fprintf (stderr, "MULT16_16_Q11: inputs are not short: %d %d\n", a, b);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = ((opus_int64)a)*b;
+ res >>= 11;
+ if (!VERIFY_INT(res))
+ {
+ fprintf (stderr, "MULT16_16_Q11: output is not short: %d*%d=%d\n", (int)a, (int)b, (int)res);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=3;
+ return res;
+}
+static OPUS_INLINE short MULT16_16_Q13(int a, int b)
+{
+ opus_int64 res;
+ if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b))
+ {
+ fprintf (stderr, "MULT16_16_Q13: inputs are not short: %d %d\n", a, b);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = ((opus_int64)a)*b;
+ res >>= 13;
+ if (!VERIFY_SHORT(res))
+ {
+ fprintf (stderr, "MULT16_16_Q13: output is not short: %d*%d=%d\n", a, b, (int)res);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=3;
+ return res;
+}
+static OPUS_INLINE short MULT16_16_Q14(int a, int b)
+{
+ opus_int64 res;
+ if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b))
+ {
+ fprintf (stderr, "MULT16_16_Q14: inputs are not short: %d %d\n", a, b);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = ((opus_int64)a)*b;
+ res >>= 14;
+ if (!VERIFY_SHORT(res))
+ {
+ fprintf (stderr, "MULT16_16_Q14: output is not short: %d\n", (int)res);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=3;
+ return res;
+}
+
+#define MULT16_16_Q15(a, b) MULT16_16_Q15_(a, b, __FILE__, __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))
+ {
+ fprintf (stderr, "MULT16_16_Q15: inputs are not short: %d %d in %s: line %d\n", a, b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = ((opus_int64)a)*b;
+ res >>= 15;
+ if (!VERIFY_SHORT(res))
+ {
+ fprintf (stderr, "MULT16_16_Q15: output is not short: %d in %s: line %d\n", (int)res, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=1;
+ return res;
+}
+
+static OPUS_INLINE short MULT16_16_P13(int a, int b)
+{
+ opus_int64 res;
+ if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b))
+ {
+ fprintf (stderr, "MULT16_16_P13: inputs are not short: %d %d\n", a, b);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = ((opus_int64)a)*b;
+ res += 4096;
+ if (!VERIFY_INT(res))
+ {
+ fprintf (stderr, "MULT16_16_P13: overflow: %d*%d=%d\n", a, b, (int)res);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res >>= 13;
+ if (!VERIFY_SHORT(res))
+ {
+ fprintf (stderr, "MULT16_16_P13: output is not short: %d*%d=%d\n", a, b, (int)res);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=4;
+ return res;
+}
+static OPUS_INLINE short MULT16_16_P14(int a, int b)
+{
+ opus_int64 res;
+ if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b))
+ {
+ fprintf (stderr, "MULT16_16_P14: inputs are not short: %d %d\n", a, b);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = ((opus_int64)a)*b;
+ res += 8192;
+ if (!VERIFY_INT(res))
+ {
+ fprintf (stderr, "MULT16_16_P14: overflow: %d*%d=%d\n", a, b, (int)res);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res >>= 14;
+ if (!VERIFY_SHORT(res))
+ {
+ fprintf (stderr, "MULT16_16_P14: output is not short: %d*%d=%d\n", a, b, (int)res);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=4;
+ return res;
+}
+static OPUS_INLINE short MULT16_16_P15(int a, int b)
+{
+ opus_int64 res;
+ if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b))
+ {
+ fprintf (stderr, "MULT16_16_P15: inputs are not short: %d %d\n", a, b);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = ((opus_int64)a)*b;
+ res += 16384;
+ if (!VERIFY_INT(res))
+ {
+ fprintf (stderr, "MULT16_16_P15: overflow: %d*%d=%d\n", a, b, (int)res);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res >>= 15;
+ if (!VERIFY_SHORT(res))
+ {
+ fprintf (stderr, "MULT16_16_P15: output is not short: %d*%d=%d\n", a, b, (int)res);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=2;
+ return res;
+}
+
+#define DIV32_16(a, b) DIV32_16_(a, b, __FILE__, __LINE__)
+
+static OPUS_INLINE int DIV32_16_(opus_int64 a, opus_int64 b, char *file, int line)
+{
+ opus_int64 res;
+ if (b==0)
+ {
+ fprintf(stderr, "DIV32_16: divide by zero: %d/%d in %s: line %d\n", (int)a, (int)b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ return 0;
+ }
+ if (!VERIFY_INT(a) || !VERIFY_SHORT(b))
+ {
+ fprintf (stderr, "DIV32_16: inputs are not int/short: %d %d in %s: line %d\n", (int)a, (int)b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = a/b;
+ if (!VERIFY_SHORT(res))
+ {
+ fprintf (stderr, "DIV32_16: output is not short: %d / %d = %d in %s: line %d\n", (int)a,(int)b,(int)res, file, line);
+ if (res>32767)
+ res = 32767;
+ if (res<-32768)
+ res = -32768;
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=35;
+ return res;
+}
+
+#define DIV32(a, b) DIV32_(a, b, __FILE__, __LINE__)
+static OPUS_INLINE int DIV32_(opus_int64 a, opus_int64 b, char *file, int line)
+{
+ opus_int64 res;
+ if (b==0)
+ {
+ fprintf(stderr, "DIV32: divide by zero: %d/%d in %s: line %d\n", (int)a, (int)b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ return 0;
+ }
+
+ if (!VERIFY_INT(a) || !VERIFY_INT(b))
+ {
+ fprintf (stderr, "DIV32: inputs are not int/short: %d %d in %s: line %d\n", (int)a, (int)b, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ res = a/b;
+ if (!VERIFY_INT(res))
+ {
+ fprintf (stderr, "DIV32: output is not int: %d in %s: line %d\n", (int)res, file, line);
+#ifdef FIXED_DEBUG_ASSERT
+ celt_assert(0);
+#endif
+ }
+ celt_mips+=70;
+ return res;
+}
+
+static OPUS_INLINE opus_val16 SIG2WORD16_generic(celt_sig x)
+{
+ x = PSHR32(x, SIG_SHIFT);
+ x = MAX32(x, -32768);
+ x = MIN32(x, 32767);
+ return EXTRACT16(x);
+}
+#define SIG2WORD16(x) (SIG2WORD16_generic(x))
+
+
+#undef PRINT_MIPS
+#define PRINT_MIPS(file) do {fprintf (file, "total complexity = %llu MIPS\n", celt_mips);} while (0);
+
+#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/fixed_generic.h b/lib/rbcodec/codecs/libopus/celt/fixed_generic.h
index ac67d37ce8..5f4abda76e 100644
--- a/lib/rbcodec/codecs/libopus/celt/fixed_generic.h
+++ b/lib/rbcodec/codecs/libopus/celt/fixed_generic.h
@@ -37,16 +37,32 @@
#define MULT16_16SU(a,b) ((opus_val32)(opus_val16)(a)*(opus_val32)(opus_uint16)(b))
/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */
+#if OPUS_FAST_INT64
+#define MULT16_32_Q16(a,b) ((opus_val32)SHR((opus_int64)((opus_val16)(a))*(b),16))
+#else
#define MULT16_32_Q16(a,b) ADD32(MULT16_16((a),SHR((b),16)), SHR(MULT16_16SU((a),((b)&0x0000ffff)),16))
+#endif
/** 16x32 multiplication, followed by a 16-bit shift right (round-to-nearest). Results fits in 32 bits */
+#if OPUS_FAST_INT64
+#define MULT16_32_P16(a,b) ((opus_val32)PSHR((opus_int64)((opus_val16)(a))*(b),16))
+#else
#define MULT16_32_P16(a,b) ADD32(MULT16_16((a),SHR((b),16)), PSHR(MULT16_16SU((a),((b)&0x0000ffff)),16))
+#endif
/** 16x32 multiplication, followed by a 15-bit shift right. Results fits in 32 bits */
+#if OPUS_FAST_INT64
+#define MULT16_32_Q15(a,b) ((opus_val32)SHR((opus_int64)((opus_val16)(a))*(b),15))
+#else
#define MULT16_32_Q15(a,b) ADD32(SHL(MULT16_16((a),SHR((b),16)),1), SHR(MULT16_16SU((a),((b)&0x0000ffff)),15))
+#endif
/** 32x32 multiplication, followed by a 31-bit shift right. Results fits in 32 bits */
+#if OPUS_FAST_INT64
+#define MULT32_32_Q31(a,b) ((opus_val32)SHR((opus_int64)(a)*(opus_int64)(b),31))
+#else
#define MULT32_32_Q31(a,b) ADD32(ADD32(SHL(MULT16_16(SHR((a),16),SHR((b),16)),1), SHR(MULT16_16SU(SHR((a),16),((b)&0x0000ffff)),15)), SHR(MULT16_16SU(SHR((b),16),((a)&0x0000ffff)),15))
+#endif
/** Compile-time conversion of float constant to 16-bit value */
#define QCONST16(x,bits) ((opus_val16)(.5+(x)*(((opus_val32)1)<<(bits))))
@@ -88,6 +104,9 @@
/** Shift by a and round-to-neareast 32-bit value. Result is a 16-bit value */
#define ROUND16(x,a) (EXTRACT16(PSHR32((x),(a))))
+/** Shift by a and round-to-neareast 32-bit value. Result is a saturated 16-bit value */
+#define SROUND16(x,a) EXTRACT16(SATURATE(PSHR32(x,a), 32767));
+
/** Divide by two */
#define HALF16(x) (SHR16(x,1))
#define HALF32(x) (SHR32(x,1))
@@ -101,6 +120,14 @@
/** Subtract two 32-bit values */
#define SUB32(a,b) ((opus_val32)(a)-(opus_val32)(b))
+/** Add two 32-bit values, ignore any overflows */
+#define ADD32_ovflw(a,b) ((opus_val32)((opus_uint32)(a)+(opus_uint32)(b)))
+/** Subtract two 32-bit values, ignore any overflows */
+#define SUB32_ovflw(a,b) ((opus_val32)((opus_uint32)(a)-(opus_uint32)(b)))
+/* Avoid MSVC warning C4146: unary minus operator applied to unsigned type */
+/** Negate 32-bit value, ignore any overflows */
+#define NEG32_ovflw(a) ((opus_val32)(0-(opus_uint32)(a)))
+
/** 16x16 multiplication where the result fits in 16 bits */
#define MULT16_16_16(a,b) ((((opus_val16)(a))*((opus_val16)(b))))
diff --git a/lib/rbcodec/codecs/libopus/celt/float_cast.h b/lib/rbcodec/codecs/libopus/celt/float_cast.h
index ede6574860..889dae965f 100644
--- a/lib/rbcodec/codecs/libopus/celt/float_cast.h
+++ b/lib/rbcodec/codecs/libopus/celt/float_cast.h
@@ -61,7 +61,13 @@
** the config.h file.
*/
-#if (HAVE_LRINTF)
+/* With GCC, when SSE is available, the fastest conversion is cvtss2si. */
+#if defined(__GNUC__) && defined(__SSE__)
+
+#include <xmmintrin.h>
+static OPUS_INLINE opus_int32 float2int(float x) {return _mm_cvt_ss2si(_mm_set_ss(x));}
+
+#elif defined(HAVE_LRINTF)
/* These defines enable functionality introduced with the 1999 ISO C
** standard. They must be defined before the inclusion of math.h to
@@ -90,21 +96,21 @@
#include <math.h>
#define float2int(x) lrint(x)
-#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && (defined (WIN64) || defined (_WIN64))
+#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && (defined(_M_X64) || (defined(_M_IX86_FP) && _M_IX86_FP >= 1))
#include <xmmintrin.h>
- __inline long int float2int(float value)
+ static __inline long int float2int(float value)
{
return _mm_cvtss_si32(_mm_load_ss(&value));
}
-#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && (defined (WIN32) || defined (_WIN32))
+#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && defined (_M_IX86)
#include <math.h>
/* Win32 doesn't seem to have these functions.
** Therefore implement OPUS_INLINE versions of these functions here.
*/
- __inline long int
+ static __inline long int
float2int (float flt)
{ int intgr;
diff --git a/lib/rbcodec/codecs/libopus/celt/kiss_fft.c b/lib/rbcodec/codecs/libopus/celt/kiss_fft.c
index 833ef5a71f..83775165d8 100644
--- a/lib/rbcodec/codecs/libopus/celt/kiss_fft.c
+++ b/lib/rbcodec/codecs/libopus/celt/kiss_fft.c
@@ -82,8 +82,8 @@ static void kf_bfly2(
C_SUB( Fout2[0] , Fout[0] , t );
C_ADDTO( Fout[0] , t );
- t.r = S_MUL(Fout2[1].r+Fout2[1].i, tw);
- t.i = S_MUL(Fout2[1].i-Fout2[1].r, tw);
+ t.r = S_MUL(ADD32_ovflw(Fout2[1].r, Fout2[1].i), tw);
+ t.i = S_MUL(SUB32_ovflw(Fout2[1].i, Fout2[1].r), tw);
C_SUB( Fout2[1] , Fout[1] , t );
C_ADDTO( Fout[1] , t );
@@ -92,8 +92,8 @@ static void kf_bfly2(
C_SUB( Fout2[2] , Fout[2] , t );
C_ADDTO( Fout[2] , t );
- t.r = S_MUL(Fout2[3].i-Fout2[3].r, tw);
- t.i = S_MUL(-Fout2[3].i-Fout2[3].r, tw);
+ t.r = S_MUL(SUB32_ovflw(Fout2[3].i, Fout2[3].r), tw);
+ t.i = S_MUL(NEG32_ovflw(ADD32_ovflw(Fout2[3].i, Fout2[3].r)), tw);
C_SUB( Fout2[3] , Fout[3] , t );
C_ADDTO( Fout[3] , t );
Fout += 8;
@@ -126,10 +126,10 @@ static void kf_bfly4(
C_ADDTO( *Fout , scratch1 );
C_SUB( scratch1 , Fout[1] , Fout[3] );
- Fout[1].r = scratch0.r + scratch1.i;
- Fout[1].i = scratch0.i - scratch1.r;
- Fout[3].r = scratch0.r - scratch1.i;
- Fout[3].i = scratch0.i + scratch1.r;
+ Fout[1].r = ADD32_ovflw(scratch0.r, scratch1.i);
+ Fout[1].i = SUB32_ovflw(scratch0.i, scratch1.r);
+ Fout[3].r = SUB32_ovflw(scratch0.r, scratch1.i);
+ Fout[3].i = ADD32_ovflw(scratch0.i, scratch1.r);
Fout+=4;
}
} else {
@@ -160,10 +160,10 @@ static void kf_bfly4(
tw3 += fstride*3;
C_ADDTO( *Fout , scratch[3] );
- Fout[m].r = scratch[5].r + scratch[4].i;
- Fout[m].i = scratch[5].i - scratch[4].r;
- Fout[m3].r = scratch[5].r - scratch[4].i;
- Fout[m3].i = scratch[5].i + scratch[4].r;
+ Fout[m].r = ADD32_ovflw(scratch[5].r, scratch[4].i);
+ Fout[m].i = SUB32_ovflw(scratch[5].i, scratch[4].r);
+ Fout[m3].r = SUB32_ovflw(scratch[5].r, scratch[4].i);
+ Fout[m3].i = ADD32_ovflw(scratch[5].i, scratch[4].r);
++Fout;
}
}
@@ -191,7 +191,7 @@ static void kf_bfly3(
kiss_fft_cpx * Fout_beg = Fout;
#ifdef FIXED_POINT
- epi3.r = -16384;
+ /*epi3.r = -16384;*/ /* Unused */
epi3.i = -28378;
#else
epi3 = st->twiddles[fstride*m];
@@ -212,18 +212,18 @@ static void kf_bfly3(
tw1 += fstride;
tw2 += fstride*2;
- Fout[m].r = Fout->r - HALF_OF(scratch[3].r);
- Fout[m].i = Fout->i - HALF_OF(scratch[3].i);
+ Fout[m].r = SUB32_ovflw(Fout->r, HALF_OF(scratch[3].r));
+ Fout[m].i = SUB32_ovflw(Fout->i, HALF_OF(scratch[3].i));
C_MULBYSCALAR( scratch[0] , epi3.i );
C_ADDTO(*Fout,scratch[3]);
- Fout[m2].r = Fout[m].r + scratch[0].i;
- Fout[m2].i = Fout[m].i - scratch[0].r;
+ Fout[m2].r = ADD32_ovflw(Fout[m].r, scratch[0].i);
+ Fout[m2].i = SUB32_ovflw(Fout[m].i, scratch[0].r);
- Fout[m].r -= scratch[0].i;
- Fout[m].i += scratch[0].r;
+ Fout[m].r = SUB32_ovflw(Fout[m].r, scratch[0].i);
+ Fout[m].i = ADD32_ovflw(Fout[m].i, scratch[0].r);
++Fout;
} while(--k);
@@ -282,22 +282,22 @@ static void kf_bfly5(
C_ADD( scratch[8],scratch[2],scratch[3]);
C_SUB( scratch[9],scratch[2],scratch[3]);
- Fout0->r += scratch[7].r + scratch[8].r;
- Fout0->i += scratch[7].i + scratch[8].i;
+ Fout0->r = ADD32_ovflw(Fout0->r, ADD32_ovflw(scratch[7].r, scratch[8].r));
+ Fout0->i = ADD32_ovflw(Fout0->i, ADD32_ovflw(scratch[7].i, scratch[8].i));
- scratch[5].r = scratch[0].r + S_MUL(scratch[7].r,ya.r) + S_MUL(scratch[8].r,yb.r);
- scratch[5].i = scratch[0].i + S_MUL(scratch[7].i,ya.r) + S_MUL(scratch[8].i,yb.r);
+ scratch[5].r = ADD32_ovflw(scratch[0].r, ADD32_ovflw(S_MUL(scratch[7].r,ya.r), S_MUL(scratch[8].r,yb.r)));
+ scratch[5].i = ADD32_ovflw(scratch[0].i, ADD32_ovflw(S_MUL(scratch[7].i,ya.r), S_MUL(scratch[8].i,yb.r)));
- scratch[6].r = S_MUL(scratch[10].i,ya.i) + S_MUL(scratch[9].i,yb.i);
- scratch[6].i = -S_MUL(scratch[10].r,ya.i) - S_MUL(scratch[9].r,yb.i);
+ scratch[6].r = ADD32_ovflw(S_MUL(scratch[10].i,ya.i), S_MUL(scratch[9].i,yb.i));
+ scratch[6].i = NEG32_ovflw(ADD32_ovflw(S_MUL(scratch[10].r,ya.i), S_MUL(scratch[9].r,yb.i)));
C_SUB(*Fout1,scratch[5],scratch[6]);
C_ADD(*Fout4,scratch[5],scratch[6]);
- scratch[11].r = scratch[0].r + S_MUL(scratch[7].r,yb.r) + S_MUL(scratch[8].r,ya.r);
- scratch[11].i = scratch[0].i + S_MUL(scratch[7].i,yb.r) + S_MUL(scratch[8].i,ya.r);
- scratch[12].r = - S_MUL(scratch[10].i,yb.i) + S_MUL(scratch[9].i,ya.i);
- scratch[12].i = S_MUL(scratch[10].r,yb.i) - S_MUL(scratch[9].r,ya.i);
+ scratch[11].r = ADD32_ovflw(scratch[0].r, ADD32_ovflw(S_MUL(scratch[7].r,yb.r), S_MUL(scratch[8].r,ya.r)));
+ scratch[11].i = ADD32_ovflw(scratch[0].i, ADD32_ovflw(S_MUL(scratch[7].i,yb.r), S_MUL(scratch[8].i,ya.r)));
+ scratch[12].r = SUB32_ovflw(S_MUL(scratch[9].i,ya.i), S_MUL(scratch[10].i,yb.i));
+ scratch[12].i = SUB32_ovflw(S_MUL(scratch[10].r,yb.i), S_MUL(scratch[9].r,ya.i));
C_ADD(*Fout2,scratch[11],scratch[12]);
C_SUB(*Fout3,scratch[11],scratch[12]);
@@ -423,13 +423,19 @@ static void compute_twiddles(kiss_twiddle_cpx *twiddles, int nfft)
#endif
}
+int opus_fft_alloc_arch_c(kiss_fft_state *st) {
+ (void)st;
+ return 0;
+}
+
/*
*
* Allocates all necessary storage space for the fft and ifft.
* The return value is a contiguous block of memory. As such,
* It can be freed with free().
* */
-kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem, const kiss_fft_state *base)
+kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem,
+ const kiss_fft_state *base, int arch)
{
kiss_fft_state *st=NULL;
size_t memneeded = sizeof(struct kiss_fft_state); /* twiddle factors*/
@@ -478,22 +484,31 @@ kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem, co
if (st->bitrev==NULL)
goto fail;
compute_bitrev_table(0, bitrev, 1,1, st->factors,st);
+
+ /* Initialize architecture specific fft parameters */
+ if (opus_fft_alloc_arch(st, arch))
+ goto fail;
}
return st;
fail:
- opus_fft_free(st);
+ opus_fft_free(st, arch);
return NULL;
}
-kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem )
+kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem, int arch)
{
- return opus_fft_alloc_twiddles(nfft, mem, lenmem, NULL);
+ return opus_fft_alloc_twiddles(nfft, mem, lenmem, NULL, arch);
}
-void opus_fft_free(const kiss_fft_state *cfg)
+void opus_fft_free_arch_c(kiss_fft_state *st) {
+ (void)st;
+}
+
+void opus_fft_free(const kiss_fft_state *cfg, int arch)
{
if (cfg)
{
+ opus_fft_free_arch((kiss_fft_state *)cfg, arch);
opus_free((opus_int16*)cfg->bitrev);
if (cfg->shift < 0)
opus_free((kiss_twiddle_cpx*)cfg->twiddles);
@@ -551,8 +566,7 @@ void opus_fft_impl(const kiss_fft_state *st,kiss_fft_cpx *fout)
}
}
-#if 0
-void opus_fft(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout)
+void opus_fft_c(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout)
{
int i;
opus_val16 scale;
@@ -573,11 +587,9 @@ void opus_fft(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fou
}
opus_fft_impl(st, fout);
}
-#endif
-#ifdef TEST_UNIT_DFT_C
-void opus_ifft(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout)
+void opus_ifft_c(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout)
{
int i;
celt_assert2 (fin != fout, "In-place FFT not supported");
@@ -590,4 +602,3 @@ void opus_ifft(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fo
for (i=0;i<st->nfft;i++)
fout[i].i = -fout[i].i;
}
-#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/kiss_fft.h b/lib/rbcodec/codecs/libopus/celt/kiss_fft.h
index 390b54d948..bffa2bfad6 100644
--- a/lib/rbcodec/codecs/libopus/celt/kiss_fft.h
+++ b/lib/rbcodec/codecs/libopus/celt/kiss_fft.h
@@ -32,6 +32,7 @@
#include <stdlib.h>
#include <math.h>
#include "arch.h"
+#include "cpu_support.h"
#ifdef __cplusplus
extern "C" {
@@ -77,6 +78,11 @@ typedef struct {
4*4*4*2
*/
+typedef struct arch_fft_state{
+ int is_supported;
+ void *priv;
+} arch_fft_state;
+
typedef struct kiss_fft_state{
int nfft;
opus_val16 scale;
@@ -87,8 +93,13 @@ typedef struct kiss_fft_state{
opus_int16 factors[2*MAXFACTORS];
const opus_int16 *bitrev;
const kiss_twiddle_cpx *twiddles;
+ arch_fft_state *arch_fft;
} kiss_fft_state;
+#if defined(HAVE_ARM_NE10)
+#include "arm/fft_arm.h"
+#endif
+
/*typedef struct kiss_fft_state* kiss_fft_cfg;*/
/**
@@ -114,9 +125,9 @@ typedef struct kiss_fft_state{
* buffer size in *lenmem.
* */
-kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem, const kiss_fft_state *base);
+kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem, const kiss_fft_state *base, int arch);
-kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem);
+kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem, int arch);
/**
* opus_fft(cfg,in_out_buf)
@@ -128,13 +139,59 @@ kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem);
* Note that each element is complex and can be accessed like
f[k].r and f[k].i
* */
-void opus_fft(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout);
-void opus_ifft(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout);
+void opus_fft_c(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout);
+void opus_ifft_c(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout);
void opus_fft_impl(const kiss_fft_state *st,kiss_fft_cpx *fout);
void opus_ifft_impl(const kiss_fft_state *st,kiss_fft_cpx *fout);
-void opus_fft_free(const kiss_fft_state *cfg);
+void opus_fft_free(const kiss_fft_state *cfg, int arch);
+
+
+void opus_fft_free_arch_c(kiss_fft_state *st);
+int opus_fft_alloc_arch_c(kiss_fft_state *st);
+
+#if !defined(OVERRIDE_OPUS_FFT)
+/* Is run-time CPU detection enabled on this platform? */
+#if defined(OPUS_HAVE_RTCD) && (defined(HAVE_ARM_NE10))
+
+extern int (*const OPUS_FFT_ALLOC_ARCH_IMPL[OPUS_ARCHMASK+1])(
+ kiss_fft_state *st);
+
+#define opus_fft_alloc_arch(_st, arch) \
+ ((*OPUS_FFT_ALLOC_ARCH_IMPL[(arch)&OPUS_ARCHMASK])(_st))
+
+extern void (*const OPUS_FFT_FREE_ARCH_IMPL[OPUS_ARCHMASK+1])(
+ kiss_fft_state *st);
+#define opus_fft_free_arch(_st, arch) \
+ ((*OPUS_FFT_FREE_ARCH_IMPL[(arch)&OPUS_ARCHMASK])(_st))
+
+extern void (*const OPUS_FFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg,
+ const kiss_fft_cpx *fin, kiss_fft_cpx *fout);
+#define opus_fft(_cfg, _fin, _fout, arch) \
+ ((*OPUS_FFT[(arch)&OPUS_ARCHMASK])(_cfg, _fin, _fout))
+
+extern void (*const OPUS_IFFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg,
+ const kiss_fft_cpx *fin, kiss_fft_cpx *fout);
+#define opus_ifft(_cfg, _fin, _fout, arch) \
+ ((*OPUS_IFFT[(arch)&OPUS_ARCHMASK])(_cfg, _fin, _fout))
+
+#else /* else for if defined(OPUS_HAVE_RTCD) && (defined(HAVE_ARM_NE10)) */
+
+#define opus_fft_alloc_arch(_st, arch) \
+ ((void)(arch), opus_fft_alloc_arch_c(_st))
+
+#define opus_fft_free_arch(_st, arch) \
+ ((void)(arch), opus_fft_free_arch_c(_st))
+
+#define opus_fft(_cfg, _fin, _fout, arch) \
+ ((void)(arch), opus_fft_c(_cfg, _fin, _fout))
+
+#define opus_ifft(_cfg, _fin, _fout, arch) \
+ ((void)(arch), opus_ifft_c(_cfg, _fin, _fout))
+
+#endif /* end if defined(OPUS_HAVE_RTCD) && (defined(HAVE_ARM_NE10)) */
+#endif /* end if !defined(OVERRIDE_OPUS_FFT) */
#ifdef __cplusplus
}
diff --git a/lib/rbcodec/codecs/libopus/celt/mathops.c b/lib/rbcodec/codecs/libopus/celt/mathops.c
index 3f8c5dcc0e..6ee9b9e101 100644
--- a/lib/rbcodec/codecs/libopus/celt/mathops.c
+++ b/lib/rbcodec/codecs/libopus/celt/mathops.c
@@ -38,7 +38,8 @@
#include "mathops.h"
/*Compute floor(sqrt(_val)) with exact arithmetic.
- This has been tested on all possible 32-bit inputs.*/
+ _val must be greater than 0.
+ This has been tested on all possible 32-bit inputs greater than 0.*/
unsigned isqrt32(opus_uint32 _val){
unsigned b;
unsigned g;
@@ -164,7 +165,7 @@ opus_val16 celt_cos_norm(opus_val32 x)
{
return _celt_cos_pi_2(EXTRACT16(x));
} else {
- return NEG32(_celt_cos_pi_2(EXTRACT16(65536-x)));
+ return NEG16(_celt_cos_pi_2(EXTRACT16(65536-x)));
}
} else {
if (x&0x0000ffff)
@@ -182,7 +183,7 @@ opus_val32 celt_rcp(opus_val32 x)
int i;
opus_val16 n;
opus_val16 r;
- celt_assert2(x>0, "celt_rcp() only defined for positive values");
+ celt_sig_assert(x>0);
i = celt_ilog2(x);
/* n is Q15 with range [0,1). */
n = VSHR32(x,i-15)-32768;
diff --git a/lib/rbcodec/codecs/libopus/celt/mathops.h b/lib/rbcodec/codecs/libopus/celt/mathops.h
index a0525a9610..f3e5246a39 100644
--- a/lib/rbcodec/codecs/libopus/celt/mathops.h
+++ b/lib/rbcodec/codecs/libopus/celt/mathops.h
@@ -38,11 +38,48 @@
#include "entcode.h"
#include "os_support.h"
+#define PI 3.141592653f
+
+#ifndef ABS
+#define ABS(a)(((a) < 0) ? - (a) :(a))
+#endif
+
/* Multiplies two 16-bit fractional values. Bit-exactness of this macro is important */
#define FRAC_MUL16(a,b) ((16384+((opus_int32)(opus_int16)(a)*(opus_int16)(b)))>>15)
unsigned isqrt32(opus_uint32 _val);
+/* CELT doesn't need it for fixed-point, by analysis.c does. */
+#if !defined(FIXED_POINT) || defined(ANALYSIS_C)
+#define cA 0.43157974f
+#define cB 0.67848403f
+#define cC 0.08595542f
+#define cE ((float)PI/2)
+static OPUS_INLINE float fast_atan2f(float y, float x) {
+ float x2, y2;
+ x2 = x*x;
+ y2 = y*y;
+ /* For very small values, we don't care about the answer, so
+ we can just return 0. */
+ if (x2 + y2 < 1e-18f)
+ {
+ return 0;
+ }
+ if(x2<y2){
+ float den = (y2 + cB*x2) * (y2 + cC*x2);
+ return -x*y*(y2 + cA*x2) / den + (y<0 ? -cE : cE);
+ }else{
+ float den = (x2 + cB*y2) * (x2 + cC*y2);
+ return x*y*(x2 + cA*y2) / den + (y<0 ? -cE : cE) - (x*y<0 ? -cE : cE);
+ }
+}
+#undef cA
+#undef cB
+#undef cC
+#undef cE
+#endif
+
+
#ifndef OVERRIDE_CELT_MAXABS16
static OPUS_INLINE opus_val32 celt_maxabs16(const opus_val16 *x, int len)
{
@@ -80,7 +117,6 @@ static OPUS_INLINE opus_val32 celt_maxabs32(const opus_val32 *x, int len)
#ifndef FIXED_POINT
-#define PI 3.141592653f
#define celt_sqrt(x) ((float)sqrt(x))
#define celt_rsqrt(x) (1.f/celt_sqrt(x))
#define celt_rsqrt_norm(x) (celt_rsqrt(x))
@@ -147,7 +183,7 @@ static OPUS_INLINE float celt_exp2(float x)
/** Integer log in base2. Undefined for zero and negative numbers */
static OPUS_INLINE opus_int16 celt_ilog2(opus_int32 x)
{
- celt_assert2(x>0, "celt_ilog2() only defined for strictly positive numbers");
+ celt_sig_assert(x>0);
return EC_ILOG(x)-1;
}
#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/mdct.c b/lib/rbcodec/codecs/libopus/celt/mdct.c
index 7fa8eaf6bf..5c6dab5b75 100644
--- a/lib/rbcodec/codecs/libopus/celt/mdct.c
+++ b/lib/rbcodec/codecs/libopus/celt/mdct.c
@@ -60,7 +60,7 @@
#ifdef CUSTOM_MODES
-int clt_mdct_init(mdct_lookup *l,int N, int maxshift)
+int clt_mdct_init(mdct_lookup *l,int N, int maxshift, int arch)
{
int i;
kiss_twiddle_scalar *trig;
@@ -71,9 +71,9 @@ int clt_mdct_init(mdct_lookup *l,int N, int maxshift)
for (i=0;i<=maxshift;i++)
{
if (i==0)
- l->kfft[i] = opus_fft_alloc(N>>2>>i, 0, 0);
+ l->kfft[i] = opus_fft_alloc(N>>2>>i, 0, 0, arch);
else
- l->kfft[i] = opus_fft_alloc_twiddles(N>>2>>i, 0, 0, l->kfft[0]);
+ l->kfft[i] = opus_fft_alloc_twiddles(N>>2>>i, 0, 0, l->kfft[0], arch);
#ifndef ENABLE_TI_DSPLIB55
if (l->kfft[i]==NULL)
return 0;
@@ -104,21 +104,20 @@ int clt_mdct_init(mdct_lookup *l,int N, int maxshift)
return 1;
}
-void clt_mdct_clear(mdct_lookup *l)
+void clt_mdct_clear(mdct_lookup *l, int arch)
{
int i;
for (i=0;i<=l->maxshift;i++)
- opus_fft_free(l->kfft[i]);
+ opus_fft_free(l->kfft[i], arch);
opus_free((kiss_twiddle_scalar*)l->trig);
}
#endif /* CUSTOM_MODES */
-#if 0
/* Forward MDCT trashes the input array */
#ifndef OVERRIDE_clt_mdct_forward
-void clt_mdct_forward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
- const opus_val16 *window, int overlap, int shift, int stride)
+void clt_mdct_forward_c(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
+ const opus_val16 *window, int overlap, int shift, int stride, int arch)
{
int i;
int N, N2, N4;
@@ -133,6 +132,7 @@ void clt_mdct_forward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar
int scale_shift = st->scale_shift-1;
#endif
SAVE_STACK;
+ (void)arch;
scale = st->scale;
N = l->n;
@@ -237,15 +237,15 @@ void clt_mdct_forward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar
RESTORE_STACK;
}
#endif /* OVERRIDE_clt_mdct_forward */
-#endif
#ifndef OVERRIDE_clt_mdct_backward
-void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
- const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride)
+void clt_mdct_backward_c(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
+ const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride, int arch)
{
int i;
int N, N2, N4;
const kiss_twiddle_scalar *trig;
+ (void) arch;
N = l->n;
trig = l->trig;
@@ -270,8 +270,8 @@ void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scala
int rev;
kiss_fft_scalar yr, yi;
rev = *bitrev++;
- yr = S_MUL(*xp2, t[i]) + S_MUL(*xp1, t[N4+i]);
- yi = S_MUL(*xp1, t[i]) - S_MUL(*xp2, t[N4+i]);
+ yr = ADD32_ovflw(S_MUL(*xp2, t[i]), S_MUL(*xp1, t[N4+i]));
+ yi = SUB32_ovflw(S_MUL(*xp1, t[i]), S_MUL(*xp2, t[N4+i]));
/* We swap real and imag because we use an FFT instead of an IFFT. */
yp[2*rev+1] = yr;
yp[2*rev] = yi;
@@ -301,8 +301,8 @@ void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scala
t0 = t[i];
t1 = t[N4+i];
/* 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(re,t1) - S_MUL(im,t0);
+ yr = ADD32_ovflw(S_MUL(re,t0), S_MUL(im,t1));
+ yi = SUB32_ovflw(S_MUL(re,t1), S_MUL(im,t0));
/* We swap real and imag because we're using an FFT instead of an IFFT. */
re = yp1[1];
im = yp1[0];
@@ -312,8 +312,8 @@ void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scala
t0 = t[(N4-i-1)];
t1 = t[(N2-i-1)];
/* 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(re,t1) - S_MUL(im,t0);
+ yr = ADD32_ovflw(S_MUL(re,t0), S_MUL(im,t1));
+ yi = SUB32_ovflw(S_MUL(re,t1), S_MUL(im,t0));
yp1[0] = yr;
yp0[1] = yi;
yp0 += 2;
@@ -333,8 +333,8 @@ void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scala
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);
+ *yp1++ = SUB32_ovflw(MULT16_32_Q15(*wp2, x2), MULT16_32_Q15(*wp1, x1));
+ *xp1-- = ADD32_ovflw(MULT16_32_Q15(*wp1, x2), MULT16_32_Q15(*wp2, x1));
wp1++;
wp2--;
}
diff --git a/lib/rbcodec/codecs/libopus/celt/mdct.h b/lib/rbcodec/codecs/libopus/celt/mdct.h
index d72182138a..160ae4e0f3 100644
--- a/lib/rbcodec/codecs/libopus/celt/mdct.h
+++ b/lib/rbcodec/codecs/libopus/celt/mdct.h
@@ -53,18 +53,60 @@ typedef struct {
const kiss_twiddle_scalar * OPUS_RESTRICT trig;
} mdct_lookup;
-int clt_mdct_init(mdct_lookup *l,int N, int maxshift);
-void clt_mdct_clear(mdct_lookup *l);
+#if defined(HAVE_ARM_NE10)
+#include "arm/mdct_arm.h"
+#endif
+
+
+int clt_mdct_init(mdct_lookup *l,int N, int maxshift, int arch);
+void clt_mdct_clear(mdct_lookup *l, int arch);
/** Compute a forward MDCT and scale by 4/N, trashes the input array */
-void clt_mdct_forward(const mdct_lookup *l, kiss_fft_scalar *in,
- kiss_fft_scalar * OPUS_RESTRICT out,
- const opus_val16 *window, int overlap, int shift, int stride);
+void clt_mdct_forward_c(const mdct_lookup *l, kiss_fft_scalar *in,
+ kiss_fft_scalar * OPUS_RESTRICT out,
+ const opus_val16 *window, int overlap,
+ int shift, int stride, int arch);
/** Compute a backward MDCT (no scaling) and performs weighted overlap-add
(scales implicitly by 1/2) */
-void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in,
+void clt_mdct_backward_c(const mdct_lookup *l, kiss_fft_scalar *in,
kiss_fft_scalar * OPUS_RESTRICT out,
- const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride);
+ const opus_val16 * OPUS_RESTRICT window,
+ int overlap, int shift, int stride, int arch);
+
+#if !defined(OVERRIDE_OPUS_MDCT)
+/* Is run-time CPU detection enabled on this platform? */
+#if defined(OPUS_HAVE_RTCD) && defined(HAVE_ARM_NE10)
+
+extern void (*const CLT_MDCT_FORWARD_IMPL[OPUS_ARCHMASK+1])(
+ const mdct_lookup *l, kiss_fft_scalar *in,
+ kiss_fft_scalar * OPUS_RESTRICT out, const opus_val16 *window,
+ int overlap, int shift, int stride, int arch);
+
+#define clt_mdct_forward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \
+ ((*CLT_MDCT_FORWARD_IMPL[(arch)&OPUS_ARCHMASK])(_l, _in, _out, \
+ _window, _overlap, _shift, \
+ _stride, _arch))
+
+extern void (*const CLT_MDCT_BACKWARD_IMPL[OPUS_ARCHMASK+1])(
+ const mdct_lookup *l, kiss_fft_scalar *in,
+ kiss_fft_scalar * OPUS_RESTRICT out, const opus_val16 *window,
+ int overlap, int shift, int stride, int arch);
+
+#define clt_mdct_backward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \
+ (*CLT_MDCT_BACKWARD_IMPL[(arch)&OPUS_ARCHMASK])(_l, _in, _out, \
+ _window, _overlap, _shift, \
+ _stride, _arch)
+
+#else /* if defined(OPUS_HAVE_RTCD) && defined(HAVE_ARM_NE10) */
+
+#define clt_mdct_forward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \
+ clt_mdct_forward_c(_l, _in, _out, _window, _overlap, _shift, _stride, _arch)
+
+#define clt_mdct_backward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \
+ clt_mdct_backward_c(_l, _in, _out, _window, _overlap, _shift, _stride, _arch)
+
+#endif /* end if defined(OPUS_HAVE_RTCD) && defined(HAVE_ARM_NE10) && !defined(FIXED_POINT) */
+#endif /* end if !defined(OVERRIDE_OPUS_MDCT) */
#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/mips/celt_mipsr1.h b/lib/rbcodec/codecs/libopus/celt/mips/celt_mipsr1.h
new file mode 100644
index 0000000000..e85661a661
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/mips/celt_mipsr1.h
@@ -0,0 +1,151 @@
+/* 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.
+*/
+
+#ifndef __CELT_MIPSR1_H__
+#define __CELT_MIPSR1_H__
+
+#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"
+
+#define OVERRIDE_comb_filter
+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 arch)
+{
+ int i;
+ opus_val32 x0, x1, x2, x3, x4;
+
+ (void)arch;
+
+ /* 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)}};
+
+ 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_P15(g0, gains[tapset0][0]);
+ g01 = MULT16_16_P15(g0, gains[tapset0][1]);
+ g02 = MULT16_16_P15(g0, gains[tapset0][2]);
+ g10 = MULT16_16_P15(g1, gains[tapset1][0]);
+ g11 = MULT16_16_P15(g1, gains[tapset1][1]);
+ g12 = MULT16_16_P15(g1, gains[tapset1][2]);
+ x1 = x[-T1+1];
+ x2 = x[-T1 ];
+ x3 = x[-T1-1];
+ x4 = x[-T1-2];
+ /* If the filter didn't change, we don't need the overlap */
+ if (g0==g1 && T0==T1 && tapset0==tapset1)
+ overlap=0;
+
+ for (i=0;i<overlap;i++)
+ {
+ opus_val16 f;
+ opus_val32 res;
+ f = MULT16_16_Q15(window[i],window[i]);
+ x0= x[i-T1+2];
+
+ asm volatile("MULT $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15((Q15ONE-f),g00)), "r" ((int)x[i-T0]));
+
+ asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15((Q15ONE-f),g01)), "r" ((int)ADD32(x[i-T0-1],x[i-T0+1])));
+ asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15((Q15ONE-f),g02)), "r" ((int)ADD32(x[i-T0-2],x[i-T0+2])));
+ asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15(f,g10)), "r" ((int)x2));
+ asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15(f,g11)), "r" ((int)ADD32(x3,x1)));
+ asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15(f,g12)), "r" ((int)ADD32(x4,x0)));
+
+ asm volatile("EXTR.W %0,$ac1, %1" : "=r" (res): "i" (15));
+
+ y[i] = x[i] + res;
+
+ x4=x3;
+ x3=x2;
+ x2=x1;
+ x1=x0;
+ }
+
+ x4 = x[i-T1-2];
+ x3 = x[i-T1-1];
+ x2 = x[i-T1];
+ x1 = x[i-T1+1];
+
+ 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;
+ }
+
+ for (i=overlap;i<N;i++)
+ {
+ opus_val32 res;
+ x0=x[i-T1+2];
+
+ asm volatile("MULT $ac1, %0, %1" : : "r" ((int)g10), "r" ((int)x2));
+
+ asm volatile("MADD $ac1, %0, %1" : : "r" ((int)g11), "r" ((int)ADD32(x3,x1)));
+ asm volatile("MADD $ac1, %0, %1" : : "r" ((int)g12), "r" ((int)ADD32(x4,x0)));
+ asm volatile("EXTR.W %0,$ac1, %1" : "=r" (res): "i" (15));
+ y[i] = x[i] + res;
+ x4=x3;
+ x3=x2;
+ x2=x1;
+ x1=x0;
+ }
+}
+
+#endif /* __CELT_MIPSR1_H__ */
diff --git a/lib/rbcodec/codecs/libopus/celt/mips/fixed_generic_mipsr1.h b/lib/rbcodec/codecs/libopus/celt/mips/fixed_generic_mipsr1.h
new file mode 100644
index 0000000000..4a05efbf85
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/mips/fixed_generic_mipsr1.h
@@ -0,0 +1,126 @@
+/* Copyright (C) 2007-2009 Xiph.Org Foundation
+ Copyright (C) 2003-2008 Jean-Marc Valin
+ Copyright (C) 2007-2008 CSIRO */
+/**
+ @file fixed_generic.h
+ @brief Generic fixed-point operations
+*/
+/*
+ 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 CELT_FIXED_GENERIC_MIPSR1_H
+#define CELT_FIXED_GENERIC_MIPSR1_H
+
+#undef MULT16_32_Q15_ADD
+static inline int MULT16_32_Q15_ADD(int a, int b, int c, int d) {
+ int m;
+ asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b));
+ asm volatile("madd $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d));
+ asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15));
+ return m;
+}
+
+#undef MULT16_32_Q15_SUB
+static inline int MULT16_32_Q15_SUB(int a, int b, int c, int d) {
+ int m;
+ asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b));
+ asm volatile("msub $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d));
+ asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15));
+ return m;
+}
+
+#undef MULT16_16_Q15_ADD
+static inline int MULT16_16_Q15_ADD(int a, int b, int c, int d) {
+ int m;
+ asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b));
+ asm volatile("madd $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d));
+ asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15));
+ return m;
+}
+
+#undef MULT16_16_Q15_SUB
+static inline int MULT16_16_Q15_SUB(int a, int b, int c, int d) {
+ int m;
+ asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b));
+ asm volatile("msub $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d));
+ asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15));
+ return m;
+}
+
+
+#undef MULT16_32_Q16
+static inline int MULT16_32_Q16(int a, int b)
+{
+ int c;
+ asm volatile("MULT $ac1,%0, %1" : : "r" (a), "r" (b));
+ asm volatile("EXTR.W %0,$ac1, %1" : "=r" (c): "i" (16));
+ return c;
+}
+
+#undef MULT16_32_P16
+static inline int MULT16_32_P16(int a, int b)
+{
+ int c;
+ asm volatile("MULT $ac1, %0, %1" : : "r" (a), "r" (b));
+ asm volatile("EXTR_R.W %0,$ac1, %1" : "=r" (c): "i" (16));
+ return c;
+}
+
+#undef MULT16_32_Q15
+static inline int MULT16_32_Q15(int a, int b)
+{
+ int c;
+ asm volatile("MULT $ac1, %0, %1" : : "r" (a), "r" (b));
+ asm volatile("EXTR.W %0,$ac1, %1" : "=r" (c): "i" (15));
+ return c;
+}
+
+#undef MULT32_32_Q31
+static inline int MULT32_32_Q31(int a, int b)
+{
+ int r;
+ asm volatile("MULT $ac1, %0, %1" : : "r" (a), "r" (b));
+ asm volatile("EXTR.W %0,$ac1, %1" : "=r" (r): "i" (31));
+ return r;
+}
+
+#undef PSHR32
+static inline int PSHR32(int a, int shift)
+{
+ int r;
+ asm volatile ("SHRAV_R.W %0, %1, %2" :"=r" (r): "r" (a), "r" (shift));
+ return r;
+}
+
+#undef MULT16_16_P15
+static inline int MULT16_16_P15(int a, int b)
+{
+ int r;
+ asm volatile ("mul %0, %1, %2" :"=r" (r): "r" (a), "r" (b));
+ asm volatile ("SHRA_R.W %0, %1, %2" : "+r" (r): "0" (r), "i"(15));
+ return r;
+}
+
+#endif /* CELT_FIXED_GENERIC_MIPSR1_H */
diff --git a/lib/rbcodec/codecs/libopus/celt/mips/kiss_fft_mipsr1.h b/lib/rbcodec/codecs/libopus/celt/mips/kiss_fft_mipsr1.h
new file mode 100644
index 0000000000..400ca4de9c
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/mips/kiss_fft_mipsr1.h
@@ -0,0 +1,167 @@
+/*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_MIPSR1_H
+#define KISS_FFT_MIPSR1_H
+
+#if !defined(KISS_FFT_GUTS_H)
+#error "This file should only be included from _kiss_fft_guts.h"
+#endif
+
+#ifdef FIXED_POINT
+
+#define S_MUL_ADD(a, b, c, d) (S_MUL(a,b)+S_MUL(c,d))
+#define S_MUL_SUB(a, b, c, d) (S_MUL(a,b)-S_MUL(c,d))
+
+#undef S_MUL_ADD
+static inline int S_MUL_ADD(int a, int b, int c, int d) {
+ int m;
+ asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b));
+ asm volatile("madd $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d));
+ asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15));
+ return m;
+}
+
+#undef S_MUL_SUB
+static inline int S_MUL_SUB(int a, int b, int c, int d) {
+ int m;
+ asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b));
+ asm volatile("msub $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d));
+ asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15));
+ return m;
+}
+
+#undef C_MUL
+# define C_MUL(m,a,b) (m=C_MUL_fun(a,b))
+static inline kiss_fft_cpx C_MUL_fun(kiss_fft_cpx a, kiss_twiddle_cpx b) {
+ kiss_fft_cpx m;
+
+ asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.r));
+ asm volatile("msub $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.i));
+ asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.r): "i" (15));
+ asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.i));
+ asm volatile("madd $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.r));
+ asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.i): "i" (15));
+
+ return m;
+}
+#undef C_MULC
+# define C_MULC(m,a,b) (m=C_MULC_fun(a,b))
+static inline kiss_fft_cpx C_MULC_fun(kiss_fft_cpx a, kiss_twiddle_cpx b) {
+ kiss_fft_cpx m;
+
+ asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.r));
+ asm volatile("madd $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.i));
+ asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.r): "i" (15));
+ asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.r));
+ asm volatile("msub $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.i));
+ asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.i): "i" (15));
+
+ return m;
+}
+
+#endif /* FIXED_POINT */
+
+#define OVERRIDE_kf_bfly5
+static void kf_bfly5(
+ kiss_fft_cpx * Fout,
+ const size_t fstride,
+ const kiss_fft_state *st,
+ int m,
+ int N,
+ int mm
+ )
+{
+ kiss_fft_cpx *Fout0,*Fout1,*Fout2,*Fout3,*Fout4;
+ int i, u;
+ kiss_fft_cpx scratch[13];
+
+ const kiss_twiddle_cpx *tw;
+ kiss_twiddle_cpx ya,yb;
+ kiss_fft_cpx * Fout_beg = Fout;
+
+#ifdef FIXED_POINT
+ ya.r = 10126;
+ ya.i = -31164;
+ yb.r = -26510;
+ yb.i = -19261;
+#else
+ ya = st->twiddles[fstride*m];
+ yb = st->twiddles[fstride*2*m];
+#endif
+
+ tw=st->twiddles;
+
+ for (i=0;i<N;i++)
+ {
+ Fout = Fout_beg + i*mm;
+ Fout0=Fout;
+ Fout1=Fout0+m;
+ Fout2=Fout0+2*m;
+ Fout3=Fout0+3*m;
+ Fout4=Fout0+4*m;
+
+ /* For non-custom modes, m is guaranteed to be a multiple of 4. */
+ for ( u=0; u<m; ++u ) {
+ scratch[0] = *Fout0;
+
+
+ C_MUL(scratch[1] ,*Fout1, tw[u*fstride]);
+ C_MUL(scratch[2] ,*Fout2, tw[2*u*fstride]);
+ C_MUL(scratch[3] ,*Fout3, tw[3*u*fstride]);
+ C_MUL(scratch[4] ,*Fout4, tw[4*u*fstride]);
+
+ C_ADD( scratch[7],scratch[1],scratch[4]);
+ C_SUB( scratch[10],scratch[1],scratch[4]);
+ C_ADD( scratch[8],scratch[2],scratch[3]);
+ C_SUB( scratch[9],scratch[2],scratch[3]);
+
+ Fout0->r += scratch[7].r + scratch[8].r;
+ Fout0->i += scratch[7].i + scratch[8].i;
+ scratch[5].r = scratch[0].r + S_MUL_ADD(scratch[7].r,ya.r,scratch[8].r,yb.r);
+ scratch[5].i = scratch[0].i + S_MUL_ADD(scratch[7].i,ya.r,scratch[8].i,yb.r);
+
+ scratch[6].r = S_MUL_ADD(scratch[10].i,ya.i,scratch[9].i,yb.i);
+ scratch[6].i = -S_MUL_ADD(scratch[10].r,ya.i,scratch[9].r,yb.i);
+
+ C_SUB(*Fout1,scratch[5],scratch[6]);
+ C_ADD(*Fout4,scratch[5],scratch[6]);
+
+ scratch[11].r = scratch[0].r + S_MUL_ADD(scratch[7].r,yb.r,scratch[8].r,ya.r);
+ scratch[11].i = scratch[0].i + S_MUL_ADD(scratch[7].i,yb.r,scratch[8].i,ya.r);
+
+ scratch[12].r = S_MUL_SUB(scratch[9].i,ya.i,scratch[10].i,yb.i);
+ scratch[12].i = S_MUL_SUB(scratch[10].r,yb.i,scratch[9].r,ya.i);
+
+ C_ADD(*Fout2,scratch[11],scratch[12]);
+ C_SUB(*Fout3,scratch[11],scratch[12]);
+
+ ++Fout0;++Fout1;++Fout2;++Fout3;++Fout4;
+ }
+ }
+}
+
+
+#endif /* KISS_FFT_MIPSR1_H */
diff --git a/lib/rbcodec/codecs/libopus/celt/mips/mdct_mipsr1.h b/lib/rbcodec/codecs/libopus/celt/mips/mdct_mipsr1.h
new file mode 100644
index 0000000000..2934dab776
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/mips/mdct_mipsr1.h
@@ -0,0 +1,288 @@
+/* Copyright (c) 2007-2008 CSIRO
+ Copyright (c) 2007-2008 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.
+*/
+
+/* This is a simple MDCT implementation that uses a N/4 complex FFT
+ to do most of the work. It should be relatively straightforward to
+ plug in pretty much and FFT here.
+
+ This replaces the Vorbis FFT (and uses the exact same API), which
+ was a bit too messy and that was ending up duplicating code
+ (might as well use the same FFT everywhere).
+
+ The algorithm is similar to (and inspired from) Fabrice Bellard's
+ MDCT implementation in FFMPEG, but has differences in signs, ordering
+ and scaling in many places.
+*/
+#ifndef __MDCT_MIPSR1_H__
+#define __MDCT_MIPSR1_H__
+
+#ifndef SKIP_CONFIG_H
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+#endif
+
+#include "mdct.h"
+#include "kiss_fft.h"
+#include "_kiss_fft_guts.h"
+#include <math.h>
+#include "os_support.h"
+#include "mathops.h"
+#include "stack_alloc.h"
+
+/* Forward MDCT trashes the input array */
+#define OVERRIDE_clt_mdct_forward
+void clt_mdct_forward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
+ const opus_val16 *window, int overlap, int shift, int stride, int arch)
+{
+ int i;
+ int N, N2, N4;
+ VARDECL(kiss_fft_scalar, f);
+ VARDECL(kiss_fft_cpx, f2);
+ const kiss_fft_state *st = l->kfft[shift];
+ const kiss_twiddle_scalar *trig;
+ opus_val16 scale;
+#ifdef FIXED_POINT
+ /* Allows us to scale with MULT16_32_Q16(), which is faster than
+ MULT16_32_Q15() on ARM. */
+ int scale_shift = st->scale_shift-1;
+#endif
+
+ (void)arch;
+
+ SAVE_STACK;
+ scale = st->scale;
+
+ N = l->n;
+ trig = l->trig;
+ for (i=0;i<shift;i++)
+ {
+ N >>= 1;
+ trig += N;
+ }
+ N2 = N>>1;
+ N4 = N>>2;
+
+ ALLOC(f, N2, kiss_fft_scalar);
+ ALLOC(f2, N4, kiss_fft_cpx);
+
+ /* Consider the input to be composed of four blocks: [a, b, c, d] */
+ /* Window, shuffle, fold */
+ {
+ /* Temp pointers to make it really clear to the compiler what we're doing */
+ const kiss_fft_scalar * OPUS_RESTRICT xp1 = in+(overlap>>1);
+ const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+N2-1+(overlap>>1);
+ 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+3)>>2);i++)
+ {
+ /* Real part arranged as -d-cR, Imag part arranged as -b+aR*/
+ *yp++ = S_MUL_ADD(*wp2, xp1[N2],*wp1,*xp2);
+ *yp++ = S_MUL_SUB(*wp1, *xp1,*wp2, xp2[-N2]);
+ xp1+=2;
+ xp2-=2;
+ wp1+=2;
+ wp2-=2;
+ }
+ wp1 = window;
+ wp2 = window+overlap-1;
+ for(;i<N4-((overlap+3)>>2);i++)
+ {
+ /* Real part arranged as a-bR, Imag part arranged as -c-dR */
+ *yp++ = *xp2;
+ *yp++ = *xp1;
+ xp1+=2;
+ xp2-=2;
+ }
+ for(;i<N4;i++)
+ {
+ /* Real part arranged as a-bR, Imag part arranged as -c-dR */
+ *yp++ = S_MUL_SUB(*wp2, *xp2, *wp1, xp1[-N2]);
+ *yp++ = S_MUL_ADD(*wp2, *xp1, *wp1, xp2[N2]);
+ xp1+=2;
+ xp2-=2;
+ wp1+=2;
+ wp2-=2;
+ }
+ }
+ /* Pre-rotation */
+ {
+ kiss_fft_scalar * OPUS_RESTRICT yp = f;
+ const kiss_twiddle_scalar *t = &trig[0];
+ for(i=0;i<N4;i++)
+ {
+ kiss_fft_cpx yc;
+ kiss_twiddle_scalar t0, t1;
+ kiss_fft_scalar re, im, yr, yi;
+ t0 = t[i];
+ t1 = t[N4+i];
+ re = *yp++;
+ im = *yp++;
+
+ yr = S_MUL_SUB(re,t0,im,t1);
+ yi = S_MUL_ADD(im,t0,re,t1);
+
+ yc.r = yr;
+ yc.i = yi;
+ yc.r = PSHR32(MULT16_32_Q16(scale, yc.r), scale_shift);
+ yc.i = PSHR32(MULT16_32_Q16(scale, yc.i), scale_shift);
+ f2[st->bitrev[i]] = yc;
+ }
+ }
+
+ /* N/4 complex FFT, does not downscale anymore */
+ opus_fft_impl(st, f2);
+
+ /* Post-rotate */
+ {
+ /* Temp pointers to make it really clear to the compiler what we're doing */
+ const kiss_fft_cpx * 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 = &trig[0];
+ /* Temp pointers to make it really clear to the compiler what we're doing */
+ for(i=0;i<N4;i++)
+ {
+ kiss_fft_scalar yr, yi;
+ yr = S_MUL_SUB(fp->i,t[N4+i] , fp->r,t[i]);
+ yi = S_MUL_ADD(fp->r,t[N4+i] ,fp->i,t[i]);
+ *yp1 = yr;
+ *yp2 = yi;
+ fp++;
+ yp1 += 2*stride;
+ yp2 -= 2*stride;
+ }
+ }
+ RESTORE_STACK;
+}
+
+#define OVERRIDE_clt_mdct_backward
+void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out,
+ const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride, int arch)
+{
+ int i;
+ int N, N2, N4;
+ const kiss_twiddle_scalar *trig;
+
+ (void)arch;
+
+ N = l->n;
+ trig = l->trig;
+ for (i=0;i<shift;i++)
+ {
+ N >>= 1;
+ trig += N;
+ }
+ N2 = N>>1;
+ N4 = N>>2;
+
+ /* Pre-rotate */
+ {
+ /* Temp pointers to make it really clear to the compiler what we're doing */
+ const kiss_fft_scalar * OPUS_RESTRICT xp1 = in;
+ const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+stride*(N2-1);
+ kiss_fft_scalar * OPUS_RESTRICT yp = out+(overlap>>1);
+ const kiss_twiddle_scalar * OPUS_RESTRICT t = &trig[0];
+ const opus_int16 * OPUS_RESTRICT bitrev = l->kfft[shift]->bitrev;
+ for(i=0;i<N4;i++)
+ {
+ int rev;
+ kiss_fft_scalar yr, yi;
+ rev = *bitrev++;
+ yr = S_MUL_ADD(*xp2, t[i] , *xp1, t[N4+i]);
+ yi = S_MUL_SUB(*xp1, t[i] , *xp2, t[N4+i]);
+ /* We swap real and imag because we use an FFT instead of an IFFT. */
+ yp[2*rev+1] = yr;
+ yp[2*rev] = yi;
+ /* Storing the pre-rotation directly in the bitrev order. */
+ xp1+=2*stride;
+ xp2-=2*stride;
+ }
+ }
+
+ opus_fft_impl(l->kfft[shift], (kiss_fft_cpx*)(out+(overlap>>1)));
+
+ /* Post-rotate and de-shuffle from both ends of the buffer at once to make
+ it in-place. */
+ {
+ 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 = &trig[0];
+ /* 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;
+ kiss_twiddle_scalar t0, t1;
+ /* We swap real and imag because we're using an FFT instead of an IFFT. */
+ re = yp0[1];
+ im = yp0[0];
+ t0 = t[i];
+ t1 = t[N4+i];
+ /* We'd scale up by 2 here, but instead it's done when mixing the windows */
+ yr = S_MUL_ADD(re,t0 , im,t1);
+ yi = S_MUL_SUB(re,t1 , im,t0);
+ /* We swap real and imag because we're using an FFT instead of an IFFT. */
+ re = yp1[1];
+ im = yp1[0];
+ yp0[0] = yr;
+ yp1[1] = yi;
+
+ t0 = t[(N4-i-1)];
+ t1 = t[(N2-i-1)];
+ /* We'd scale up by 2 here, but instead it's done when mixing the windows */
+ yr = S_MUL_ADD(re,t0,im,t1);
+ yi = S_MUL_SUB(re,t1,im,t0);
+ yp1[0] = yr;
+ yp0[1] = yi;
+ yp0 += 2;
+ yp1 -= 2;
+ }
+ }
+
+ /* Mirror on both sides for TDAC */
+ {
+ 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 < overlap/2; i++)
+ {
+ 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--;
+ }
+ }
+}
+#endif /* __MDCT_MIPSR1_H__ */
diff --git a/lib/rbcodec/codecs/libopus/celt/mips/pitch_mipsr1.h b/lib/rbcodec/codecs/libopus/celt/mips/pitch_mipsr1.h
new file mode 100644
index 0000000000..a9500aff58
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/mips/pitch_mipsr1.h
@@ -0,0 +1,161 @@
+/* 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_MIPSR1_H
+#define PITCH_MIPSR1_H
+
+#define OVERRIDE_DUAL_INNER_PROD
+static 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 arch)
+{
+ int j;
+ opus_val32 xy01=0;
+ opus_val32 xy02=0;
+
+ (void)arch;
+
+ asm volatile("MULT $ac1, $0, $0");
+ asm volatile("MULT $ac2, $0, $0");
+ /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */
+ for (j=0;j<N;j++)
+ {
+ asm volatile("MADD $ac1, %0, %1" : : "r" ((int)x[j]), "r" ((int)y01[j]));
+ asm volatile("MADD $ac2, %0, %1" : : "r" ((int)x[j]), "r" ((int)y02[j]));
+ ++j;
+ asm volatile("MADD $ac1, %0, %1" : : "r" ((int)x[j]), "r" ((int)y01[j]));
+ asm volatile("MADD $ac2, %0, %1" : : "r" ((int)x[j]), "r" ((int)y02[j]));
+ }
+ asm volatile ("mflo %0, $ac1": "=r"(xy01));
+ asm volatile ("mflo %0, $ac2": "=r"(xy02));
+ *xy1 = xy01;
+ *xy2 = xy02;
+}
+
+static inline void xcorr_kernel_mips(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;
+
+ opus_int64 sum_0, sum_1, sum_2, sum_3;
+ sum_0 = (opus_int64)sum[0];
+ sum_1 = (opus_int64)sum[1];
+ sum_2 = (opus_int64)sum[2];
+ sum_3 = (opus_int64)sum[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 = __builtin_mips_madd( sum_0, tmp, y_0);
+ sum_1 = __builtin_mips_madd( sum_1, tmp, y_1);
+ sum_2 = __builtin_mips_madd( sum_2, tmp, y_2);
+ sum_3 = __builtin_mips_madd( sum_3, tmp, y_3);
+
+ tmp=*x++;
+ y_0=*y++;
+
+ sum_0 = __builtin_mips_madd( sum_0, tmp, y_1 );
+ sum_1 = __builtin_mips_madd( sum_1, tmp, y_2 );
+ sum_2 = __builtin_mips_madd( sum_2, tmp, y_3);
+ sum_3 = __builtin_mips_madd( sum_3, tmp, y_0);
+
+ tmp=*x++;
+ y_1=*y++;
+
+ sum_0 = __builtin_mips_madd( sum_0, tmp, y_2 );
+ sum_1 = __builtin_mips_madd( sum_1, tmp, y_3 );
+ sum_2 = __builtin_mips_madd( sum_2, tmp, y_0);
+ sum_3 = __builtin_mips_madd( sum_3, tmp, y_1);
+
+
+ tmp=*x++;
+ y_2=*y++;
+
+ sum_0 = __builtin_mips_madd( sum_0, tmp, y_3 );
+ sum_1 = __builtin_mips_madd( sum_1, tmp, y_0 );
+ sum_2 = __builtin_mips_madd( sum_2, tmp, y_1);
+ sum_3 = __builtin_mips_madd( sum_3, tmp, y_2);
+
+ }
+ if (j++<len)
+ {
+ opus_val16 tmp = *x++;
+ y_3=*y++;
+
+ sum_0 = __builtin_mips_madd( sum_0, tmp, y_0 );
+ sum_1 = __builtin_mips_madd( sum_1, tmp, y_1 );
+ sum_2 = __builtin_mips_madd( sum_2, tmp, y_2);
+ sum_3 = __builtin_mips_madd( sum_3, tmp, y_3);
+ }
+
+ if (j++<len)
+ {
+ opus_val16 tmp=*x++;
+ y_0=*y++;
+
+ sum_0 = __builtin_mips_madd( sum_0, tmp, y_1 );
+ sum_1 = __builtin_mips_madd( sum_1, tmp, y_2 );
+ sum_2 = __builtin_mips_madd( sum_2, tmp, y_3);
+ sum_3 = __builtin_mips_madd( sum_3, tmp, y_0);
+ }
+
+ if (j<len)
+ {
+ opus_val16 tmp=*x++;
+ y_1=*y++;
+
+ sum_0 = __builtin_mips_madd( sum_0, tmp, y_2 );
+ sum_1 = __builtin_mips_madd( sum_1, tmp, y_3 );
+ sum_2 = __builtin_mips_madd( sum_2, tmp, y_0);
+ sum_3 = __builtin_mips_madd( sum_3, tmp, y_1);
+
+ }
+
+ sum[0] = (opus_val32)sum_0;
+ sum[1] = (opus_val32)sum_1;
+ sum[2] = (opus_val32)sum_2;
+ sum[3] = (opus_val32)sum_3;
+}
+
+#define OVERRIDE_XCORR_KERNEL
+#define xcorr_kernel(x, y, sum, len, arch) \
+ ((void)(arch), xcorr_kernel_mips(x, y, sum, len))
+
+#endif /* PITCH_MIPSR1_H */
diff --git a/lib/rbcodec/codecs/libopus/celt/mips/vq_mipsr1.h b/lib/rbcodec/codecs/libopus/celt/mips/vq_mipsr1.h
new file mode 100644
index 0000000000..fd18eab7a9
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/mips/vq_mipsr1.h
@@ -0,0 +1,122 @@
+/* Copyright (c) 2007-2008 CSIRO
+ Copyright (c) 2007-2009 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.
+*/
+
+#ifndef __VQ_MIPSR1_H__
+#define __VQ_MIPSR1_H__
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+
+#include "mathops.h"
+#include "arch.h"
+
+static void renormalise_vector_mips(celt_norm *X, int N, opus_val16 gain, int arch);
+
+#define OVERRIDE_vq_exp_rotation1
+static void exp_rotation1(celt_norm *X, int len, int stride, opus_val16 c, opus_val16 s)
+{
+ int i;
+ opus_val16 ms;
+ celt_norm *Xptr;
+ Xptr = X;
+ ms = NEG16(s);
+ for (i=0;i<len-stride;i++)
+ {
+ celt_norm x1, x2;
+ x1 = Xptr[0];
+ x2 = Xptr[stride];
+ Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2), s, x1), 15));
+ *Xptr++ = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15));
+ }
+ Xptr = &X[len-2*stride-1];
+ for (i=len-2*stride-1;i>=0;i--)
+ {
+ celt_norm x1, x2;
+ x1 = Xptr[0];
+ x2 = Xptr[stride];
+ Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2), s, x1), 15));
+ *Xptr-- = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15));
+ }
+}
+
+#define OVERRIDE_renormalise_vector
+
+#define renormalise_vector(X, N, gain, arch) \
+ (renormalise_vector_mips(X, N, gain, arch))
+
+void renormalise_vector_mips(celt_norm *X, int N, opus_val16 gain, int arch)
+{
+ int i;
+#ifdef FIXED_POINT
+ int k;
+#endif
+ opus_val32 E = EPSILON;
+ opus_val16 g;
+ opus_val32 t;
+ celt_norm *xptr = X;
+ int X0, X1;
+
+ (void)arch;
+
+ asm volatile("mult $ac1, $0, $0");
+ asm volatile("MTLO %0, $ac1" : :"r" (E));
+ /*if(N %4)
+ printf("error");*/
+ for (i=0;i<N-2;i+=2)
+ {
+ X0 = (int)*xptr++;
+ asm volatile("MADD $ac1, %0, %1" : : "r" (X0), "r" (X0));
+
+ X1 = (int)*xptr++;
+ asm volatile("MADD $ac1, %0, %1" : : "r" (X1), "r" (X1));
+ }
+
+ for (;i<N;i++)
+ {
+ X0 = (int)*xptr++;
+ asm volatile("MADD $ac1, %0, %1" : : "r" (X0), "r" (X0));
+ }
+
+ asm volatile("MFLO %0, $ac1" : "=r" (E));
+#ifdef FIXED_POINT
+ k = celt_ilog2(E)>>1;
+#endif
+ t = VSHR32(E, 2*(k-7));
+ g = MULT16_16_P15(celt_rsqrt_norm(t),gain);
+
+ xptr = X;
+ for (i=0;i<N;i++)
+ {
+ *xptr = EXTRACT16(PSHR32(MULT16_16(g, *xptr), k+1));
+ xptr++;
+ }
+ /*return celt_sqrt(E);*/
+}
+
+#endif /* __VQ_MIPSR1_H__ */
diff --git a/lib/rbcodec/codecs/libopus/celt/modes.c b/lib/rbcodec/codecs/libopus/celt/modes.c
index 42e68e1cb7..390c5e8aeb 100644
--- a/lib/rbcodec/codecs/libopus/celt/modes.c
+++ b/lib/rbcodec/codecs/libopus/celt/modes.c
@@ -37,6 +37,7 @@
#include "os_support.h"
#include "stack_alloc.h"
#include "quant_bands.h"
+#include "cpu_support.h"
static const opus_int16 eband5ms[] = {
/*0 200 400 600 800 1k 1.2 1.4 1.6 2k 2.4 2.8 3.2 4k 4.8 5.6 6.8 8k 9.6 12k 15.6 */
@@ -229,6 +230,7 @@ CELTMode *opus_custom_mode_create(opus_int32 Fs, int frame_size, int *error)
opus_val16 *window;
opus_int16 *logN;
int LM;
+ int arch = opus_select_arch();
ALLOC_STACK;
#if !defined(VAR_ARRAYS) && !defined(USE_ALLOCA)
if (global_stack==NULL)
@@ -389,7 +391,7 @@ CELTMode *opus_custom_mode_create(opus_int32 Fs, int frame_size, int *error)
compute_pulse_cache(mode, mode->maxLM);
if (clt_mdct_init(&mode->mdct, 2*mode->shortMdctSize*mode->nbShortMdcts,
- mode->maxLM) == 0)
+ mode->maxLM, arch) == 0)
goto failure;
if (error)
@@ -408,6 +410,8 @@ failure:
#ifdef CUSTOM_MODES
void opus_custom_mode_destroy(CELTMode *mode)
{
+ int arch = opus_select_arch();
+
if (mode == NULL)
return;
#ifndef CUSTOM_MODES_ONLY
@@ -423,7 +427,7 @@ void opus_custom_mode_destroy(CELTMode *mode)
}
#endif /* CUSTOM_MODES_ONLY */
opus_free((opus_int16*)mode->eBands);
- opus_free((opus_int16*)mode->allocVectors);
+ opus_free((unsigned char*)mode->allocVectors);
opus_free((opus_val16*)mode->window);
opus_free((opus_int16*)mode->logN);
@@ -431,7 +435,7 @@ void opus_custom_mode_destroy(CELTMode *mode)
opus_free((opus_int16*)mode->cache.index);
opus_free((unsigned char*)mode->cache.bits);
opus_free((unsigned char*)mode->cache.caps);
- clt_mdct_clear(&mode->mdct);
+ clt_mdct_clear(&mode->mdct, arch);
opus_free((CELTMode *)mode);
}
diff --git a/lib/rbcodec/codecs/libopus/celt/opus_custom_demo.c b/lib/rbcodec/codecs/libopus/celt/opus_custom_demo.c
new file mode 100644
index 0000000000..ae41c0de5a
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/opus_custom_demo.c
@@ -0,0 +1,210 @@
+/* Copyright (c) 2007-2008 CSIRO
+ Copyright (c) 2007-2009 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_custom.h"
+#include "arch.h"
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <string.h>
+
+#define MAX_PACKET 1275
+
+int main(int argc, char *argv[])
+{
+ int err;
+ char *inFile, *outFile;
+ FILE *fin, *fout;
+ OpusCustomMode *mode=NULL;
+ OpusCustomEncoder *enc;
+ OpusCustomDecoder *dec;
+ int len;
+ opus_int32 frame_size, channels, rate;
+ int bytes_per_packet;
+ unsigned char data[MAX_PACKET];
+ int complexity;
+#if !(defined (FIXED_POINT) && !defined(CUSTOM_MODES)) && defined(RESYNTH)
+ int i;
+ double rmsd = 0;
+#endif
+ int count = 0;
+ opus_int32 skip;
+ opus_int16 *in, *out;
+ if (argc != 9 && argc != 8 && argc != 7)
+ {
+ fprintf (stderr, "Usage: test_opus_custom <rate> <channels> <frame size> "
+ " <bytes per packet> [<complexity> [packet loss rate]] "
+ "<input> <output>\n");
+ return 1;
+ }
+
+ rate = (opus_int32)atol(argv[1]);
+ channels = atoi(argv[2]);
+ frame_size = atoi(argv[3]);
+ mode = opus_custom_mode_create(rate, frame_size, NULL);
+ if (mode == NULL)
+ {
+ fprintf(stderr, "failed to create a mode\n");
+ return 1;
+ }
+
+ bytes_per_packet = atoi(argv[4]);
+ if (bytes_per_packet < 0 || bytes_per_packet > MAX_PACKET)
+ {
+ fprintf (stderr, "bytes per packet must be between 0 and %d\n",
+ MAX_PACKET);
+ return 1;
+ }
+
+ inFile = argv[argc-2];
+ fin = fopen(inFile, "rb");
+ if (!fin)
+ {
+ fprintf (stderr, "Could not open input file %s\n", argv[argc-2]);
+ return 1;
+ }
+ outFile = argv[argc-1];
+ fout = fopen(outFile, "wb+");
+ if (!fout)
+ {
+ fprintf (stderr, "Could not open output file %s\n", argv[argc-1]);
+ fclose(fin);
+ return 1;
+ }
+
+ enc = opus_custom_encoder_create(mode, channels, &err);
+ if (err != 0)
+ {
+ fprintf(stderr, "Failed to create the encoder: %s\n", opus_strerror(err));
+ fclose(fin);
+ fclose(fout);
+ return 1;
+ }
+ dec = opus_custom_decoder_create(mode, channels, &err);
+ if (err != 0)
+ {
+ fprintf(stderr, "Failed to create the decoder: %s\n", opus_strerror(err));
+ fclose(fin);
+ fclose(fout);
+ return 1;
+ }
+ opus_custom_decoder_ctl(dec, OPUS_GET_LOOKAHEAD(&skip));
+
+ if (argc>7)
+ {
+ complexity=atoi(argv[5]);
+ opus_custom_encoder_ctl(enc,OPUS_SET_COMPLEXITY(complexity));
+ }
+
+ in = (opus_int16*)malloc(frame_size*channels*sizeof(opus_int16));
+ out = (opus_int16*)malloc(frame_size*channels*sizeof(opus_int16));
+
+ while (!feof(fin))
+ {
+ int ret;
+ err = fread(in, sizeof(short), frame_size*channels, fin);
+ if (feof(fin))
+ break;
+ len = opus_custom_encode(enc, in, frame_size, data, bytes_per_packet);
+ if (len <= 0)
+ fprintf (stderr, "opus_custom_encode() failed: %s\n", opus_strerror(len));
+
+ /* This is for simulating bit errors */
+#if 0
+ int errors = 0;
+ int eid = 0;
+ /* This simulates random bit error */
+ for (i=0;i<len*8;i++)
+ {
+ if (rand()%atoi(argv[8])==0)
+ {
+ if (i<64)
+ {
+ errors++;
+ eid = i;
+ }
+ data[i/8] ^= 1<<(7-(i%8));
+ }
+ }
+ if (errors == 1)
+ data[eid/8] ^= 1<<(7-(eid%8));
+ else if (errors%2 == 1)
+ data[rand()%8] ^= 1<<rand()%8;
+#endif
+
+#if 1 /* Set to zero to use the encoder's output instead */
+ /* This is to simulate packet loss */
+ if (argc==9 && rand()%1000<atoi(argv[argc-3]))
+ /*if (errors && (errors%2==0))*/
+ ret = opus_custom_decode(dec, NULL, len, out, frame_size);
+ else
+ ret = opus_custom_decode(dec, data, len, out, frame_size);
+ if (ret < 0)
+ fprintf(stderr, "opus_custom_decode() failed: %s\n", opus_strerror(ret));
+#else
+ for (i=0;i<ret*channels;i++)
+ out[i] = in[i];
+#endif
+#if !(defined (FIXED_POINT) && !defined(CUSTOM_MODES)) && defined(RESYNTH)
+ for (i=0;i<ret*channels;i++)
+ {
+ rmsd += (in[i]-out[i])*1.0*(in[i]-out[i]);
+ /*out[i] -= in[i];*/
+ }
+#endif
+ count++;
+ fwrite(out+skip*channels, sizeof(short), (ret-skip)*channels, fout);
+ skip = 0;
+ }
+ PRINT_MIPS(stderr);
+
+ opus_custom_encoder_destroy(enc);
+ opus_custom_decoder_destroy(dec);
+ fclose(fin);
+ fclose(fout);
+ opus_custom_mode_destroy(mode);
+ free(in);
+ free(out);
+#if !(defined (FIXED_POINT) && !defined(CUSTOM_MODES)) && defined(RESYNTH)
+ if (rmsd > 0)
+ {
+ rmsd = sqrt(rmsd/(1.0*frame_size*channels*count));
+ fprintf (stderr, "Error: encoder doesn't match decoder\n");
+ fprintf (stderr, "RMS mismatch is %f\n", rmsd);
+ return 1;
+ } else {
+ fprintf (stderr, "Encoder matches decoder!!\n");
+ }
+#endif
+ return 0;
+}
+
diff --git a/lib/rbcodec/codecs/libopus/celt/os_support.h b/lib/rbcodec/codecs/libopus/celt/os_support.h
index 5e47e3cff9..a2171971e9 100644
--- a/lib/rbcodec/codecs/libopus/celt/os_support.h
+++ b/lib/rbcodec/codecs/libopus/celt/os_support.h
@@ -67,18 +67,18 @@ static OPUS_INLINE void opus_free (void *ptr)
}
#endif
-/** Copy n bytes of memory from src to dst. The 0* term provides compile-time type checking */
+/** Copy n elements from src to dst. The 0* term provides compile-time type checking */
#ifndef OVERRIDE_OPUS_COPY
#define OPUS_COPY(dst, src, n) (memcpy((dst), (src), (n)*sizeof(*(dst)) + 0*((dst)-(src)) ))
#endif
-/** Copy n bytes of memory from src to dst, allowing overlapping regions. The 0* term
+/** Copy n elements from src to dst, allowing overlapping regions. The 0* term
provides compile-time type checking */
#ifndef OVERRIDE_OPUS_MOVE
#define OPUS_MOVE(dst, src, n) (memmove((dst), (src), (n)*sizeof(*(dst)) + 0*((dst)-(src)) ))
#endif
-/** Set n elements of dst to zero, starting at address s */
+/** Set n elements of dst to zero */
#ifndef OVERRIDE_OPUS_CLEAR
#define OPUS_CLEAR(dst, n) (memset((dst), 0, (n)*sizeof(*(dst))))
#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/pitch.c b/lib/rbcodec/codecs/libopus/celt/pitch.c
index ee56a434f0..872582a48a 100644
--- a/lib/rbcodec/codecs/libopus/celt/pitch.c
+++ b/lib/rbcodec/codecs/libopus/celt/pitch.c
@@ -102,11 +102,9 @@ static void find_best_pitch(opus_val32 *xcorr, opus_val16 *y, int len,
}
}
-static void celt_fir5(const opus_val16 *x,
+static void celt_fir5(opus_val16 *x,
const opus_val16 *num,
- opus_val16 *y,
- int N,
- opus_val16 *mem)
+ int N)
{
int i;
opus_val16 num0, num1, num2, num3, num4;
@@ -116,11 +114,11 @@ static void celt_fir5(const opus_val16 *x,
num2=num[2];
num3=num[3];
num4=num[4];
- mem0=mem[0];
- mem1=mem[1];
- mem2=mem[2];
- mem3=mem[3];
- mem4=mem[4];
+ mem0=0;
+ mem1=0;
+ mem2=0;
+ mem3=0;
+ mem4=0;
for (i=0;i<N;i++)
{
opus_val32 sum = SHL32(EXTEND32(x[i]), SIG_SHIFT);
@@ -134,13 +132,8 @@ static void celt_fir5(const opus_val16 *x,
mem2 = mem1;
mem1 = mem0;
mem0 = x[i];
- y[i] = ROUND16(sum, SIG_SHIFT);
+ x[i] = ROUND16(sum, SIG_SHIFT);
}
- mem[0]=mem0;
- mem[1]=mem1;
- mem[2]=mem2;
- mem[3]=mem3;
- mem[4]=mem4;
}
@@ -150,7 +143,7 @@ void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x
int i;
opus_val32 ac[5];
opus_val16 tmp=Q15ONE;
- opus_val16 lpc[4], mem[5]={0,0,0,0,0};
+ opus_val16 lpc[4];
opus_val16 lpc2[5];
opus_val16 c1 = QCONST16(.8f,15);
#ifdef FIXED_POINT
@@ -211,28 +204,33 @@ void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x
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);
+ celt_fir5(x_lp, lpc2, len>>1);
}
-#if 0 /* This is a simple version of the pitch correlation that should work
- well on DSPs like Blackfin and TI C5x/C6x */
-
+/* Pure C implementation. */
#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)
+celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y,
+ opus_val32 *xcorr, int len, int max_pitch, int arch)
{
+
+#if 0 /* This is a simple version of the pitch correlation that should work
+ well on DSPs like Blackfin and TI C5x/C6x */
int i, j;
#ifdef FIXED_POINT
opus_val32 maxcorr=1;
#endif
+#if !defined(OVERRIDE_PITCH_XCORR)
+ (void)arch;
+#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]);
+ sum = MAC16_16(sum, _x[j], _y[i+j]);
xcorr[i] = sum;
#ifdef FIXED_POINT
maxcorr = MAX32(maxcorr, sum);
@@ -241,17 +239,8 @@ celt_pitch_xcorr(opus_val16 *x, opus_val16 *y, opus_val32 *xcorr, int len, int m
#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;
/*The EDSP version requires that max_pitch is at least 1, and that _x is
32-bit aligned.
@@ -260,11 +249,11 @@ celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, opus_val32 *xcorr
opus_val32 maxcorr=1;
#endif
celt_assert(max_pitch>0);
- celt_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0);
+ celt_sig_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0);
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_kernel(_x, _y+i, sum, len, arch);
xcorr[i]=sum[0];
xcorr[i+1]=sum[1];
xcorr[i+2]=sum[2];
@@ -280,7 +269,7 @@ celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, opus_val32 *xcorr
for (;i<max_pitch;i++)
{
opus_val32 sum;
- sum = celt_inner_prod(_x, _y+i, len);
+ sum = celt_inner_prod(_x, _y+i, len, arch);
xcorr[i] = sum;
#ifdef FIXED_POINT
maxcorr = MAX32(maxcorr, sum);
@@ -289,9 +278,9 @@ celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, opus_val32 *xcorr
#ifdef FIXED_POINT
return maxcorr;
#endif
+#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 arch)
{
@@ -369,7 +358,7 @@ void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTR
for (j=0;j<len>>1;j++)
sum += SHR32(MULT16_16(x_lp[j],y[i+j]), shift);
#else
- sum = celt_inner_prod(x_lp, y+i, len>>1);
+ sum = celt_inner_prod(x_lp, y+i, len>>1, arch);
#endif
xcorr[i] = MAX32(-1, sum);
#ifdef FIXED_POINT
@@ -403,10 +392,44 @@ void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTR
RESTORE_STACK;
}
-#if 0
+#ifdef FIXED_POINT
+static opus_val16 compute_pitch_gain(opus_val32 xy, opus_val32 xx, opus_val32 yy)
+{
+ opus_val32 x2y2;
+ int sx, sy, shift;
+ opus_val32 g;
+ opus_val16 den;
+ if (xy == 0 || xx == 0 || yy == 0)
+ return 0;
+ sx = celt_ilog2(xx)-14;
+ sy = celt_ilog2(yy)-14;
+ shift = sx + sy;
+ x2y2 = SHR32(MULT16_16(VSHR32(xx, sx), VSHR32(yy, sy)), 14);
+ if (shift & 1) {
+ if (x2y2 < 32768)
+ {
+ x2y2 <<= 1;
+ shift--;
+ } else {
+ x2y2 >>= 1;
+ shift++;
+ }
+ }
+ den = celt_rsqrt_norm(x2y2);
+ g = MULT16_32_Q15(den, xy);
+ g = VSHR32(g, (shift>>1)-1);
+ return EXTRACT16(MIN32(g, Q15ONE));
+}
+#else
+static opus_val16 compute_pitch_gain(opus_val32 xy, opus_val32 xx, opus_val32 yy)
+{
+ return xy/celt_sqrt(1+xx*yy);
+}
+#endif
+
static const int second_check[16] = {0, 0, 3, 2, 3, 2, 5, 2, 3, 2, 3, 2, 5, 2, 3, 2};
opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod,
- int N, int *T0_, int prev_period, opus_val16 prev_gain)
+ int N, int *T0_, int prev_period, opus_val16 prev_gain, int arch)
{
int k, i, T, T0;
opus_val16 g, g0;
@@ -431,7 +454,7 @@ opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod,
T = T0 = *T0_;
ALLOC(yy_lookup, maxperiod+1, opus_val32);
- dual_inner_prod(x, x, x-T0, N, &xx, &xy);
+ dual_inner_prod(x, x, x-T0, N, &xx, &xy, arch);
yy_lookup[0] = xx;
yy=xx;
for (i=1;i<=maxperiod;i++)
@@ -442,18 +465,7 @@ opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod,
yy = yy_lookup[T0];
best_xy = xy;
best_yy = yy;
-#ifdef FIXED_POINT
- {
- opus_val32 x2y2;
- int sh, t;
- x2y2 = 1+HALF32(MULT32_32_Q31(xx,yy));
- sh = celt_ilog2(x2y2)>>1;
- t = VSHR32(x2y2, 2*(sh-7));
- g = g0 = VSHR32(MULT16_32_Q15(celt_rsqrt_norm(t), xy),sh+1);
- }
-#else
- g = g0 = xy/celt_sqrt(1+xx*yy);
-#endif
+ g = g0 = compute_pitch_gain(xy, xx, yy);
/* Look for any pitch at T/k */
for (k=2;k<=15;k++)
{
@@ -475,25 +487,14 @@ opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod,
{
T1b = celt_udiv(2*second_check[k]*T0+k, 2*k);
}
- 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;
- int sh, t;
- x2y2 = 1+MULT32_32_Q31(xx,yy);
- sh = celt_ilog2(x2y2)>>1;
- t = VSHR32(x2y2, 2*(sh-7));
- g1 = VSHR32(MULT16_32_Q15(celt_rsqrt_norm(t), xy),sh+1);
- }
-#else
- g1 = xy/celt_sqrt(1+2.f*xx*1.f*yy);
-#endif
+ dual_inner_prod(x, &x[-T1], &x[-T1b], N, &xy, &xy2, arch);
+ xy = HALF32(xy + xy2);
+ yy = HALF32(yy_lookup[T1] + yy_lookup[T1b]);
+ g1 = compute_pitch_gain(xy, xx, yy);
if (abs(T1-prev_period)<=1)
cont = prev_gain;
else if (abs(T1-prev_period)<=2 && 5*k*k < T0)
- cont = HALF32(prev_gain);
+ cont = HALF16(prev_gain);
else
cont = 0;
thresh = MAX16(QCONST16(.3f,15), MULT16_16_Q15(QCONST16(.7f,15),g0)-cont);
@@ -518,7 +519,7 @@ opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod,
pg = SHR32(frac_div32(best_xy,best_yy+1),16);
for (k=0;k<3;k++)
- xcorr[k] = celt_inner_prod(x, x-(T+k-1), N);
+ xcorr[k] = celt_inner_prod(x, x-(T+k-1), N, arch);
if ((xcorr[2]-xcorr[0]) > MULT16_32_Q15(QCONST16(.7f,15),xcorr[1]-xcorr[0]))
offset = 1;
else if ((xcorr[0]-xcorr[2]) > MULT16_32_Q15(QCONST16(.7f,15),xcorr[1]-xcorr[2]))
@@ -534,4 +535,3 @@ opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod,
RESTORE_STACK;
return pg;
}
-#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/pitch.h b/lib/rbcodec/codecs/libopus/celt/pitch.h
index 96dbc0d794..e425f56aea 100644
--- a/lib/rbcodec/codecs/libopus/celt/pitch.h
+++ b/lib/rbcodec/codecs/libopus/celt/pitch.h
@@ -37,7 +37,8 @@
#include "modes.h"
#include "cpu_support.h"
-#if defined(__SSE__) && !defined(FIXED_POINT)
+#if (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT)) \
+ || ((defined(OPUS_X86_MAY_HAVE_SSE4_1) || defined(OPUS_X86_MAY_HAVE_SSE2)) && defined(FIXED_POINT))
#include "x86/pitch_sse.h"
#endif
@@ -45,8 +46,8 @@
#include "mips/pitch_mipsr1.h"
#endif
-#if defined(OPUS_ARM_ASM) && defined(FIXED_POINT)
-//# include "arm/pitch_arm.h"
+#if (defined(OPUS_ARM_ASM) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR))
+# include "arm/pitch_arm.h"
#endif
void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x_lp,
@@ -56,12 +57,12 @@ void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTR
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);
+ int N, int *T0, int prev_period, opus_val16 prev_gain, int arch);
+
/* 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)
+static OPUS_INLINE void xcorr_kernel_c(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;
@@ -126,10 +127,14 @@ static OPUS_INLINE void xcorr_kernel(const opus_val16 * x, const opus_val16 * y,
sum[3] = MAC16_16(sum[3],tmp,y_1);
}
}
+
+#ifndef OVERRIDE_XCORR_KERNEL
+#define xcorr_kernel(x, y, sum, len, arch) \
+ ((void)(arch),xcorr_kernel_c(x, y, sum, len))
#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,
+
+static OPUS_INLINE void dual_inner_prod_c(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02,
int N, opus_val32 *xy1, opus_val32 *xy2)
{
int i;
@@ -143,11 +148,16 @@ static OPUS_INLINE void dual_inner_prod(const opus_val16 *x, const opus_val16 *y
*xy1 = xy01;
*xy2 = xy02;
}
+
+#ifndef OVERRIDE_DUAL_INNER_PROD
+# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) \
+ ((void)(arch),dual_inner_prod_c(x, y01, y02, N, xy1, xy2))
#endif
-#ifndef OVERRIDE_CELT_INNER_PROD
-static OPUS_INLINE opus_val32 celt_inner_prod(const opus_val16 *x, const opus_val16 *y,
- int N)
+/*We make sure a C version is always available for cases where the overhead of
+ vectorization and passing around an arch flag aren't worth it.*/
+static OPUS_INLINE opus_val32 celt_inner_prod_c(const opus_val16 *x,
+ const opus_val16 *y, int N)
{
int i;
opus_val32 xy=0;
@@ -155,35 +165,28 @@ static OPUS_INLINE opus_val32 celt_inner_prod(const opus_val16 *x, const opus_va
xy = MAC16_16(xy, x[i], y[i]);
return xy;
}
+
+#if !defined(OVERRIDE_CELT_INNER_PROD)
+# define celt_inner_prod(x, y, N, arch) \
+ ((void)(arch),celt_inner_prod_c(x, y, N))
+#endif
+
+#ifdef NON_STATIC_COMB_FILTER_CONST_C
+void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N,
+ opus_val16 g10, opus_val16 g11, opus_val16 g12);
#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);
+ opus_val32 *xcorr, int len, int max_pitch, int arch);
-#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
+#ifndef OVERRIDE_PITCH_XCORR
+# define celt_pitch_xcorr celt_pitch_xcorr_c
#endif
#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/quant_bands.c b/lib/rbcodec/codecs/libopus/celt/quant_bands.c
index ac6952c266..39a221eda5 100644
--- a/lib/rbcodec/codecs/libopus/celt/quant_bands.c
+++ b/lib/rbcodec/codecs/libopus/celt/quant_bands.c
@@ -292,7 +292,7 @@ void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd,
#endif
}
if (lfe)
- max_decay=3;
+ max_decay = QCONST16(3.f,DB_SHIFT);
enc_start_state = *enc;
ALLOC(oldEBands_intra, C*m->nbEBands, opus_val16);
@@ -418,6 +418,7 @@ void quant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 *ol
offset = (q2-.5f)*(1<<(14-fine_quant[i]-1))*(1.f/16384);
#endif
oldEBands[i+c*m->nbEBands] += offset;
+ error[i+c*m->nbEBands] -= offset;
bits_left--;
} while (++c < C);
}
@@ -456,7 +457,7 @@ void unquant_coarse_energy(const CELTMode *m, int start, int end, opus_val16 *ol
/* It would be better to express this invariant as a
test on C at function entry, but that isn't enough
to make the static analyzer happy. */
- celt_assert(c<2);
+ celt_sig_assert(c<2);
tell = ec_tell(dec);
if(budget-tell>=15)
{
@@ -547,9 +548,15 @@ void amp2Log2(const CELTMode *m, int effEnd, int end,
c=0;
do {
for (i=0;i<effEnd;i++)
+ {
bandLogE[i+c*m->nbEBands] =
- celt_log2(SHL32(bandE[i+c*m->nbEBands],2))
+ celt_log2(bandE[i+c*m->nbEBands])
- SHL16((opus_val16)eMeans[i],6);
+#ifdef FIXED_POINT
+ /* Compensate for bandE[] being Q12 but celt_log2() taking a Q14 input. */
+ bandLogE[i+c*m->nbEBands] += QCONST16(2.f, DB_SHIFT);
+#endif
+ }
for (i=effEnd;i<end;i++)
bandLogE[c*m->nbEBands+i] = -QCONST16(14.f,DB_SHIFT);
} while (++c < C);
diff --git a/lib/rbcodec/codecs/libopus/celt/rate.c b/lib/rbcodec/codecs/libopus/celt/rate.c
index f85c3ee63c..465e1ba26c 100644
--- a/lib/rbcodec/codecs/libopus/celt/rate.c
+++ b/lib/rbcodec/codecs/libopus/celt/rate.c
@@ -131,7 +131,7 @@ void compute_pulse_cache(CELTMode *m, int LM)
for (i=0;i<nbEntries;i++)
{
unsigned char *ptr = bits+entryI[i];
- opus_int16 tmp[MAX_PULSES+1];
+ opus_int16 tmp[CELT_MAX_PULSES+1];
get_required_bits(tmp, entryN[i], get_pulses(entryK[i]), BITRES);
for (j=1;j<=entryK[i];j++)
ptr[j] = tmp[get_pulses(j)]-1;
@@ -296,7 +296,7 @@ static OPUS_INLINE int interp_bits2pulses(const CELTMode *m, int start, int end,
done = 0;
for (j=end;j-->start;)
{
- int tmp = bits1[j] + (lo*bits2[j]>>ALLOC_STEPS);
+ int tmp = bits1[j] + ((opus_int32)lo*bits2[j]>>ALLOC_STEPS);
if (tmp < thresh[j] && !done)
{
if (tmp >= alloc_floor)
@@ -348,12 +348,17 @@ static OPUS_INLINE int interp_bits2pulses(const CELTMode *m, int start, int end,
/*This if() block is the only part of the allocation function that
is not a mandatory part of the bitstream: any bands we choose to
skip here must be explicitly signaled.*/
- /*Choose a threshold with some hysteresis to keep bands from
- fluctuating in and out.*/
+ int depth_threshold;
+ /*We choose a threshold with some hysteresis to keep bands from
+ fluctuating in and out, but we try not to fold below a certain point. */
+ if (codedBands > 17)
+ depth_threshold = j<prev ? 7 : 9;
+ else
+ depth_threshold = 0;
#ifdef FUZZING
if ((rand()&0x1) == 0)
#else
- if (codedBands<=start+2 || (band_bits > ((j<prev?7:9)*band_width<<LM<<BITRES)>>4 && j<=signalBandwidth))
+ if (codedBands<=start+2 || (band_bits > (depth_threshold*band_width<<LM<<BITRES)>>4 && j<=signalBandwidth))
#endif
{
ec_enc_bit_logp(ec, 1, 1);
@@ -524,7 +529,7 @@ static OPUS_INLINE int interp_bits2pulses(const CELTMode *m, int start, int end,
return codedBands;
}
-int compute_allocation(const CELTMode *m, int start, int end, const int *offsets, const int *cap, int alloc_trim, int *intensity, int *dual_stereo,
+int clt_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, int signalBandwidth)
{
int lo, hi, len, j;
diff --git a/lib/rbcodec/codecs/libopus/celt/rate.h b/lib/rbcodec/codecs/libopus/celt/rate.h
index f1e0661129..852b9d6f60 100644
--- a/lib/rbcodec/codecs/libopus/celt/rate.h
+++ b/lib/rbcodec/codecs/libopus/celt/rate.h
@@ -32,7 +32,7 @@
#define MAX_PSEUDO 40
#define LOG_MAX_PSEUDO 6
-#define MAX_PULSES 128
+#define CELT_MAX_PULSES 128
#define MAX_FINE_BITS 8
@@ -95,7 +95,7 @@ static OPUS_INLINE int pulses2bits(const CELTMode *m, int band, int LM, int puls
@param pulses Number of pulses per band (returned)
@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,
+int clt_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, int signalBandwidth);
#endif
diff --git a/lib/rbcodec/codecs/libopus/celt/static_modes_fixed.h b/lib/rbcodec/codecs/libopus/celt/static_modes_fixed.h
index 1f13497c69..8717d626cb 100644
--- a/lib/rbcodec/codecs/libopus/celt/static_modes_fixed.h
+++ b/lib/rbcodec/codecs/libopus/celt/static_modes_fixed.h
@@ -4,9 +4,14 @@
#include "modes.h"
#include "rate.h"
+#ifdef HAVE_ARM_NE10
+#define OVERRIDE_FFT 1
+#include "static_modes_fixed_arm_ne10.h"
+#endif
+
#ifndef DEF_WINDOW120
#define DEF_WINDOW120
-static const opus_val16 window120[120] ICONST_ATTR = {
+static const opus_val16 window120[120] = {
2, 20, 55, 108, 178,
266, 372, 494, 635, 792,
966, 1157, 1365, 1590, 1831,
@@ -36,13 +41,13 @@ static const opus_val16 window120[120] ICONST_ATTR = {
#ifndef DEF_LOGN400
#define DEF_LOGN400
-static const opus_int16 logN400[21] ICONST_ATTR = {
+static const opus_int16 logN400[21] = {
0, 0, 0, 0, 0, 0, 0, 0, 8, 8, 8, 8, 16, 16, 16, 21, 21, 24, 29, 34, 36, };
#endif
#ifndef DEF_PULSE_CACHE50
#define DEF_PULSE_CACHE50
-static const opus_int16 cache_index50[105] ICONST_ATTR = {
+static const opus_int16 cache_index50[105] = {
-1, -1, -1, -1, -1, -1, -1, -1, 0, 0, 0, 0, 41, 41, 41,
82, 82, 123, 164, 200, 222, 0, 0, 0, 0, 0, 0, 0, 0, 41,
41, 41, 41, 123, 123, 123, 164, 164, 240, 266, 283, 295, 41, 41, 41,
@@ -51,7 +56,7 @@ static const opus_int16 cache_index50[105] ICONST_ATTR = {
305, 305, 305, 318, 318, 343, 351, 358, 364, 240, 240, 240, 240, 240, 240,
240, 240, 305, 305, 305, 305, 343, 343, 343, 351, 351, 370, 376, 382, 387,
};
-static const unsigned char cache_bits50[392] ICONST_ATTR = {
+static const unsigned char cache_bits50[392] = {
40, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 40, 15, 23, 28,
@@ -79,7 +84,7 @@ static const unsigned char cache_bits50[392] ICONST_ATTR = {
106, 151, 192, 231, 5, 59, 111, 158, 202, 243, 5, 55, 103, 147, 187,
224, 5, 60, 113, 161, 206, 248, 4, 65, 122, 175, 224, 4, 67, 127,
182, 234, };
-static const unsigned char cache_caps50[168] ICONST_ATTR = {
+static const unsigned char cache_caps50[168] = {
224, 224, 224, 224, 224, 224, 224, 224, 160, 160, 160, 160, 185, 185, 185,
178, 178, 168, 134, 61, 37, 224, 224, 224, 224, 224, 224, 224, 224, 240,
240, 240, 240, 207, 207, 207, 198, 198, 183, 144, 66, 40, 160, 160, 160,
@@ -96,7 +101,7 @@ static const unsigned char cache_caps50[168] ICONST_ATTR = {
#ifndef FFT_TWIDDLES48000_960
#define FFT_TWIDDLES48000_960
-static const kiss_twiddle_cpx fft_twiddles48000_960[480] ICONST_ATTR = {
+static const kiss_twiddle_cpx fft_twiddles48000_960[480] = {
{32767, 0}, {32766, -429},
{32757, -858}, {32743, -1287},
{32724, -1715}, {32698, -2143},
@@ -424,53 +429,73 @@ static const opus_int16 fft_bitrev60[60] = {
#ifndef FFT_STATE48000_960_0
#define FFT_STATE48000_960_0
-static const kiss_fft_state fft_state48000_960_0 ICONST_ATTR = {
+static const kiss_fft_state fft_state48000_960_0 = {
480, /* nfft */
-17476, /* scale */
+17476, /* scale */
8, /* scale_shift */
-1, /* shift */
-{5, 96, 3, 32, 4, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, }, /* factors */
+{5, 96, 3, 32, 4, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, }, /* factors */
fft_bitrev480, /* bitrev */
fft_twiddles48000_960, /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_480,
+#else
+NULL,
+#endif
};
#endif
#ifndef FFT_STATE48000_960_1
#define FFT_STATE48000_960_1
-static const kiss_fft_state fft_state48000_960_1 ICONST_ATTR = {
+static const kiss_fft_state fft_state48000_960_1 = {
240, /* nfft */
-17476, /* scale */
+17476, /* scale */
7, /* scale_shift */
1, /* shift */
-{5, 48, 3, 16, 4, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */
+{5, 48, 3, 16, 4, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */
fft_bitrev240, /* bitrev */
fft_twiddles48000_960, /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_240,
+#else
+NULL,
+#endif
};
#endif
#ifndef FFT_STATE48000_960_2
#define FFT_STATE48000_960_2
-static const kiss_fft_state fft_state48000_960_2 ICONST_ATTR = {
+static const kiss_fft_state fft_state48000_960_2 = {
120, /* nfft */
-17476, /* scale */
+17476, /* scale */
6, /* scale_shift */
2, /* shift */
-{5, 24, 3, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */
+{5, 24, 3, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */
fft_bitrev120, /* bitrev */
fft_twiddles48000_960, /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_120,
+#else
+NULL,
+#endif
};
#endif
#ifndef FFT_STATE48000_960_3
#define FFT_STATE48000_960_3
-static const kiss_fft_state fft_state48000_960_3 ICONST_ATTR = {
+static const kiss_fft_state fft_state48000_960_3 = {
60, /* nfft */
-17476, /* scale */
+17476, /* scale */
5, /* scale_shift */
3, /* shift */
-{5, 12, 3, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */
+{5, 12, 3, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */
fft_bitrev60, /* bitrev */
fft_twiddles48000_960, /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_60,
+#else
+NULL,
+#endif
};
#endif
@@ -478,7 +503,7 @@ fft_twiddles48000_960, /* bitrev */
#ifndef MDCT_TWIDDLES960
#define MDCT_TWIDDLES960
-static const opus_val16 mdct_twiddles960[1800] ICONST_ATTR = {
+static const opus_val16 mdct_twiddles960[1800] = {
32767, 32767, 32767, 32766, 32765,
32763, 32761, 32759, 32756, 32753,
32750, 32746, 32742, 32738, 32733,
@@ -842,7 +867,7 @@ static const opus_val16 mdct_twiddles960[1800] ICONST_ATTR = {
};
#endif
-static const CELTMode mode48000_960_120 ICONST_ATTR = {
+static const CELTMode mode48000_960_120 = {
48000, /* Fs */
120, /* overlap */
21, /* nbEBands */
diff --git a/lib/rbcodec/codecs/libopus/celt/static_modes_fixed_arm_ne10.h b/lib/rbcodec/codecs/libopus/celt/static_modes_fixed_arm_ne10.h
new file mode 100644
index 0000000000..7623092192
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/static_modes_fixed_arm_ne10.h
@@ -0,0 +1,388 @@
+/* The contents of this file was automatically generated by
+ * dump_mode_arm_ne10.c with arguments: 48000 960
+ * It contains static definitions for some pre-defined modes. */
+#include <NE10_types.h>
+
+#ifndef NE10_FFT_PARAMS48000_960
+#define NE10_FFT_PARAMS48000_960
+static const ne10_int32_t ne10_factors_480[64] = {
+4, 40, 4, 30, 2, 15, 5, 3, 3, 1, 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, };
+static const ne10_int32_t ne10_factors_240[64] = {
+3, 20, 4, 15, 5, 3, 3, 1, 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, };
+static const ne10_int32_t ne10_factors_120[64] = {
+3, 10, 2, 15, 5, 3, 3, 1, 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, };
+static const ne10_int32_t ne10_factors_60[64] = {
+2, 5, 5, 3, 3, 1, 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, };
+static const ne10_fft_cpx_int32_t ne10_twiddles_480[480] = {
+{0,0}, {2147483647,0}, {2147483647,0},
+{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394},
+{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496},
+{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096},
+{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152},
+{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313},
+{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424},
+{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496},
+{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268},
+{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785},
+{2147483647,0}, {2144540595,-112390613}, {2135719506,-224473172},
+{2121044558,-335940465}, {2100555974,-446486968}, {2074309912,-555809682},
+{2042378310,-663608960}, {2004848691,-769589332}, {1961823921,-873460313},
+{1913421927,-974937199}, {1859775377,-1073741851}, {1801031311,-1169603450},
+{1737350743,-1262259248}, {1668908218,-1351455280}, {1595891331,-1436947067},
+{1518500216,-1518500282}, {1436946998,-1595891394}, {1351455207,-1668908277},
+{1262259172,-1737350799}, {1169603371,-1801031362}, {1073741769,-1859775424},
+{974937230,-1913421912}, {873460227,-1961823959}, {769589125,-2004848771},
+{663608871,-2042378339}, {555809715,-2074309903}, {446486876,-2100555994},
+{335940246,-2121044593}, {224473078,-2135719516}, {112390647,-2144540593},
+{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968},
+{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851},
+{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394},
+{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959},
+{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516},
+{-94,-2147483647}, {-224473265,-2135719496}, {-446487060,-2100555955},
+{-663609049,-2042378281}, {-873460398,-1961823883}, {-1073741932,-1859775330},
+{-1262259116,-1737350839}, {-1436947137,-1595891268}, {-1595891628,-1436946738},
+{-1737350854,-1262259096}, {-1859775343,-1073741910}, {-1961823997,-873460141},
+{-2042378447,-663608538}, {-2100556013,-446486785}, {-2135719499,-224473240},
+{2147483647,0}, {2121044558,-335940465}, {2042378310,-663608960},
+{1913421927,-974937199}, {1737350743,-1262259248}, {1518500216,-1518500282},
+{1262259172,-1737350799}, {974937230,-1913421912}, {663608871,-2042378339},
+{335940246,-2121044593}, {-94,-2147483647}, {-335940431,-2121044564},
+{-663609049,-2042378281}, {-974937397,-1913421827}, {-1262259116,-1737350839},
+{-1518500258,-1518500240}, {-1737350854,-1262259096}, {-1913422071,-974936918},
+{-2042378447,-663608538}, {-2121044568,-335940406}, {-2147483647,188},
+{-2121044509,335940777}, {-2042378331,663608895}, {-1913421900,974937252},
+{-1737350633,1262259400}, {-1518499993,1518500506}, {-1262258813,1737351059},
+{-974936606,1913422229}, {-663609179,2042378239}, {-335940566,2121044542},
+{2147483647,0}, {2147299667,-28109693}, {2146747758,-56214570},
+{2145828015,-84309815}, {2144540595,-112390613}, {2142885719,-140452154},
+{2140863671,-168489630}, {2138474797,-196498235}, {2135719506,-224473172},
+{2132598271,-252409646}, {2129111626,-280302871}, {2125260168,-308148068},
+{2121044558,-335940465}, {2116465518,-363675300}, {2111523833,-391347822},
+{2106220349,-418953288}, {2100555974,-446486968}, {2094531681,-473944146},
+{2088148500,-501320115}, {2081407525,-528610186}, {2074309912,-555809682},
+{2066856885,-582913912}, {2059049696,-609918325}, {2050889698,-636818231},
+{2042378310,-663608960}, {2033516972,-690285983}, {2024307180,-716844791},
+{2014750533,-743280770}, {2004848691,-769589332}, {1994603329,-795766029},
+{1984016179,-821806435}, {1973089077,-847706028}, {1961823921,-873460313},
+{1950222618,-899064934}, {1938287127,-924515564}, {1926019520,-949807783},
+{1913421927,-974937199}, {1900496481,-999899565}, {1887245364,-1024690661},
+{1873670877,-1049306180}, {1859775377,-1073741851}, {1845561215,-1097993541},
+{1831030826,-1122057097}, {1816186632,-1145928502}, {1801031311,-1169603450},
+{1785567394,-1193077993}, {1769797456,-1216348214}, {1753724345,-1239409914},
+{1737350743,-1262259248}, {1720679456,-1284892300}, {1703713340,-1307305194},
+{1686455222,-1329494189}, {1668908218,-1351455280}, {1651075255,-1373184807},
+{1632959307,-1394679144}, {1614563642,-1415934412}, {1595891331,-1436947067},
+{1576945572,-1457713510}, {1557729613,-1478230181}, {1538246655,-1498493658},
+{1518500216,-1518500282}, {1498493590,-1538246721}, {1478230113,-1557729677},
+{1457713441,-1576945636}, {1436946998,-1595891394}, {1415934341,-1614563704},
+{1394679073,-1632959368}, {1373184735,-1651075315}, {1351455207,-1668908277},
+{1329494115,-1686455280}, {1307305120,-1703713397}, {1284892225,-1720679512},
+{1262259172,-1737350799}, {1239409837,-1753724400}, {1216348136,-1769797510},
+{1193077915,-1785567446}, {1169603371,-1801031362}, {1145928423,-1816186682},
+{1122057017,-1831030875}, {1097993571,-1845561197}, {1073741769,-1859775424},
+{1049305987,-1873670985}, {1024690635,-1887245378}, {999899482,-1900496524},
+{974937230,-1913421912}, {949807699,-1926019561}, {924515422,-1938287195},
+{899064965,-1950222603}, {873460227,-1961823959}, {847705824,-1973089164},
+{821806407,-1984016190}, {795765941,-1994603364}, {769589125,-2004848771},
+{743280682,-2014750566}, {716844642,-2024307233}, {690286016,-2033516961},
+{663608871,-2042378339}, {636818019,-2050889764}, {609918296,-2059049705},
+{582913822,-2066856911}, {555809715,-2074309903}, {528610126,-2081407540},
+{501319962,-2088148536}, {473944148,-2094531680}, {446486876,-2100555994},
+{418953102,-2106220386}, {391347792,-2111523838}, {363675176,-2116465540},
+{335940246,-2121044593}, {308148006,-2125260177}, {280302715,-2129111646},
+{252409648,-2132598271}, {224473078,-2135719516}, {196498046,-2138474814},
+{168489600,-2140863674}, {140452029,-2142885728}, {112390647,-2144540593},
+{84309753,-2145828017}, {56214412,-2146747762}, {28109695,-2147299667},
+{2147483647,0}, {2146747758,-56214570}, {2144540595,-112390613},
+{2140863671,-168489630}, {2135719506,-224473172}, {2129111626,-280302871},
+{2121044558,-335940465}, {2111523833,-391347822}, {2100555974,-446486968},
+{2088148500,-501320115}, {2074309912,-555809682}, {2059049696,-609918325},
+{2042378310,-663608960}, {2024307180,-716844791}, {2004848691,-769589332},
+{1984016179,-821806435}, {1961823921,-873460313}, {1938287127,-924515564},
+{1913421927,-974937199}, {1887245364,-1024690661}, {1859775377,-1073741851},
+{1831030826,-1122057097}, {1801031311,-1169603450}, {1769797456,-1216348214},
+{1737350743,-1262259248}, {1703713340,-1307305194}, {1668908218,-1351455280},
+{1632959307,-1394679144}, {1595891331,-1436947067}, {1557729613,-1478230181},
+{1518500216,-1518500282}, {1478230113,-1557729677}, {1436946998,-1595891394},
+{1394679073,-1632959368}, {1351455207,-1668908277}, {1307305120,-1703713397},
+{1262259172,-1737350799}, {1216348136,-1769797510}, {1169603371,-1801031362},
+{1122057017,-1831030875}, {1073741769,-1859775424}, {1024690635,-1887245378},
+{974937230,-1913421912}, {924515422,-1938287195}, {873460227,-1961823959},
+{821806407,-1984016190}, {769589125,-2004848771}, {716844642,-2024307233},
+{663608871,-2042378339}, {609918296,-2059049705}, {555809715,-2074309903},
+{501319962,-2088148536}, {446486876,-2100555994}, {391347792,-2111523838},
+{335940246,-2121044593}, {280302715,-2129111646}, {224473078,-2135719516},
+{168489600,-2140863674}, {112390647,-2144540593}, {56214412,-2146747762},
+{-94,-2147483647}, {-56214600,-2146747757}, {-112390835,-2144540584},
+{-168489787,-2140863659}, {-224473265,-2135719496}, {-280302901,-2129111622},
+{-335940431,-2121044564}, {-391347977,-2111523804}, {-446487060,-2100555955},
+{-501320144,-2088148493}, {-555809896,-2074309855}, {-609918476,-2059049651},
+{-663609049,-2042378281}, {-716844819,-2024307170}, {-769589300,-2004848703},
+{-821806581,-1984016118}, {-873460398,-1961823883}, {-924515591,-1938287114},
+{-974937397,-1913421827}, {-1024690575,-1887245411}, {-1073741932,-1859775330},
+{-1122057395,-1831030643}, {-1169603421,-1801031330}, {-1216348291,-1769797403},
+{-1262259116,-1737350839}, {-1307305268,-1703713283}, {-1351455453,-1668908078},
+{-1394679021,-1632959413}, {-1436947137,-1595891268}, {-1478230435,-1557729372},
+{-1518500258,-1518500240}, {-1557729742,-1478230045}, {-1595891628,-1436946738},
+{-1632959429,-1394679001}, {-1668908417,-1351455035}, {-1703713298,-1307305248},
+{-1737350854,-1262259096}, {-1769797708,-1216347848}, {-1801031344,-1169603400},
+{-1831030924,-1122056937}, {-1859775343,-1073741910}, {-1887245423,-1024690552},
+{-1913422071,-974936918}, {-1938287125,-924515568}, {-1961823997,-873460141},
+{-1984016324,-821806084}, {-2004848713,-769589276}, {-2024307264,-716844553},
+{-2042378447,-663608538}, {-2059049731,-609918206}, {-2074309994,-555809377},
+{-2088148499,-501320119}, {-2100556013,-446486785}, {-2111523902,-391347448},
+{-2121044568,-335940406}, {-2129111659,-280302621}, {-2135719499,-224473240},
+{-2140863681,-168489506}, {-2144540612,-112390298}, {-2146747758,-56214574},
+{2147483647,0}, {2145828015,-84309815}, {2140863671,-168489630},
+{2132598271,-252409646}, {2121044558,-335940465}, {2106220349,-418953288},
+{2088148500,-501320115}, {2066856885,-582913912}, {2042378310,-663608960},
+{2014750533,-743280770}, {1984016179,-821806435}, {1950222618,-899064934},
+{1913421927,-974937199}, {1873670877,-1049306180}, {1831030826,-1122057097},
+{1785567394,-1193077993}, {1737350743,-1262259248}, {1686455222,-1329494189},
+{1632959307,-1394679144}, {1576945572,-1457713510}, {1518500216,-1518500282},
+{1457713441,-1576945636}, {1394679073,-1632959368}, {1329494115,-1686455280},
+{1262259172,-1737350799}, {1193077915,-1785567446}, {1122057017,-1831030875},
+{1049305987,-1873670985}, {974937230,-1913421912}, {899064965,-1950222603},
+{821806407,-1984016190}, {743280682,-2014750566}, {663608871,-2042378339},
+{582913822,-2066856911}, {501319962,-2088148536}, {418953102,-2106220386},
+{335940246,-2121044593}, {252409648,-2132598271}, {168489600,-2140863674},
+{84309753,-2145828017}, {-94,-2147483647}, {-84309940,-2145828010},
+{-168489787,-2140863659}, {-252409834,-2132598249}, {-335940431,-2121044564},
+{-418953286,-2106220349}, {-501320144,-2088148493}, {-582914003,-2066856860},
+{-663609049,-2042378281}, {-743280858,-2014750501}, {-821806581,-1984016118},
+{-899065136,-1950222525}, {-974937397,-1913421827}, {-1049306374,-1873670768},
+{-1122057395,-1831030643}, {-1193078284,-1785567199}, {-1262259116,-1737350839},
+{-1329494061,-1686455323}, {-1394679021,-1632959413}, {-1457713485,-1576945595},
+{-1518500258,-1518500240}, {-1576945613,-1457713466}, {-1632959429,-1394679001},
+{-1686455338,-1329494041}, {-1737350854,-1262259096}, {-1785567498,-1193077837},
+{-1831030924,-1122056937}, {-1873671031,-1049305905}, {-1913422071,-974936918},
+{-1950222750,-899064648}, {-1984016324,-821806084}, {-2014750687,-743280354},
+{-2042378447,-663608538}, {-2066856867,-582913978}, {-2088148499,-501320119},
+{-2106220354,-418953261}, {-2121044568,-335940406}, {-2132598282,-252409555},
+{-2140863681,-168489506}, {-2145828021,-84309659}, {-2147483647,188},
+{-2145828006,84310034}, {-2140863651,168489881}, {-2132598237,252409928},
+{-2121044509,335940777}, {-2106220281,418953629}, {-2088148411,501320484},
+{-2066856765,582914339}, {-2042378331,663608895}, {-2014750557,743280706},
+{-1984016181,821806431}, {-1950222593,899064989}, {-1913421900,974937252},
+{-1873670848,1049306232}, {-1831030728,1122057257}, {-1785567289,1193078149},
+{-1737350633,1262259400}, {-1686455106,1329494336}, {-1632959185,1394679287},
+{-1576945358,1457713742}, {-1518499993,1518500506}, {-1457713209,1576945850},
+{-1394678735,1632959656}, {-1329493766,1686455555}, {-1262258813,1737351059},
+{-1193077546,1785567692}, {-1122056638,1831031107}, {-1049305599,1873671202},
+{-974936606,1913422229}, {-899064330,1950222896}, {-821805761,1984016458},
+{-743280025,2014750808}, {-663609179,2042378239}, {-582914134,2066856823},
+{-501320277,2088148461}, {-418953420,2106220322}, {-335940566,2121044542},
+{-252409716,2132598263}, {-168489668,2140863668}, {-84309821,2145828015},
+};
+static const ne10_fft_cpx_int32_t ne10_twiddles_240[240] = {
+{0,0}, {2147483647,0}, {2147483647,0},
+{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394},
+{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496},
+{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096},
+{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152},
+{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968},
+{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851},
+{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394},
+{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959},
+{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516},
+{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313},
+{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424},
+{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496},
+{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268},
+{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785},
+{2147483647,0}, {2042378310,-663608960}, {1737350743,-1262259248},
+{1262259172,-1737350799}, {663608871,-2042378339}, {-94,-2147483647},
+{-663609049,-2042378281}, {-1262259116,-1737350839}, {-1737350854,-1262259096},
+{-2042378447,-663608538}, {-2147483647,188}, {-2042378331,663608895},
+{-1737350633,1262259400}, {-1262258813,1737351059}, {-663609179,2042378239},
+{2147483647,0}, {2146747758,-56214570}, {2144540595,-112390613},
+{2140863671,-168489630}, {2135719506,-224473172}, {2129111626,-280302871},
+{2121044558,-335940465}, {2111523833,-391347822}, {2100555974,-446486968},
+{2088148500,-501320115}, {2074309912,-555809682}, {2059049696,-609918325},
+{2042378310,-663608960}, {2024307180,-716844791}, {2004848691,-769589332},
+{1984016179,-821806435}, {1961823921,-873460313}, {1938287127,-924515564},
+{1913421927,-974937199}, {1887245364,-1024690661}, {1859775377,-1073741851},
+{1831030826,-1122057097}, {1801031311,-1169603450}, {1769797456,-1216348214},
+{1737350743,-1262259248}, {1703713340,-1307305194}, {1668908218,-1351455280},
+{1632959307,-1394679144}, {1595891331,-1436947067}, {1557729613,-1478230181},
+{1518500216,-1518500282}, {1478230113,-1557729677}, {1436946998,-1595891394},
+{1394679073,-1632959368}, {1351455207,-1668908277}, {1307305120,-1703713397},
+{1262259172,-1737350799}, {1216348136,-1769797510}, {1169603371,-1801031362},
+{1122057017,-1831030875}, {1073741769,-1859775424}, {1024690635,-1887245378},
+{974937230,-1913421912}, {924515422,-1938287195}, {873460227,-1961823959},
+{821806407,-1984016190}, {769589125,-2004848771}, {716844642,-2024307233},
+{663608871,-2042378339}, {609918296,-2059049705}, {555809715,-2074309903},
+{501319962,-2088148536}, {446486876,-2100555994}, {391347792,-2111523838},
+{335940246,-2121044593}, {280302715,-2129111646}, {224473078,-2135719516},
+{168489600,-2140863674}, {112390647,-2144540593}, {56214412,-2146747762},
+{2147483647,0}, {2144540595,-112390613}, {2135719506,-224473172},
+{2121044558,-335940465}, {2100555974,-446486968}, {2074309912,-555809682},
+{2042378310,-663608960}, {2004848691,-769589332}, {1961823921,-873460313},
+{1913421927,-974937199}, {1859775377,-1073741851}, {1801031311,-1169603450},
+{1737350743,-1262259248}, {1668908218,-1351455280}, {1595891331,-1436947067},
+{1518500216,-1518500282}, {1436946998,-1595891394}, {1351455207,-1668908277},
+{1262259172,-1737350799}, {1169603371,-1801031362}, {1073741769,-1859775424},
+{974937230,-1913421912}, {873460227,-1961823959}, {769589125,-2004848771},
+{663608871,-2042378339}, {555809715,-2074309903}, {446486876,-2100555994},
+{335940246,-2121044593}, {224473078,-2135719516}, {112390647,-2144540593},
+{-94,-2147483647}, {-112390835,-2144540584}, {-224473265,-2135719496},
+{-335940431,-2121044564}, {-446487060,-2100555955}, {-555809896,-2074309855},
+{-663609049,-2042378281}, {-769589300,-2004848703}, {-873460398,-1961823883},
+{-974937397,-1913421827}, {-1073741932,-1859775330}, {-1169603421,-1801031330},
+{-1262259116,-1737350839}, {-1351455453,-1668908078}, {-1436947137,-1595891268},
+{-1518500258,-1518500240}, {-1595891628,-1436946738}, {-1668908417,-1351455035},
+{-1737350854,-1262259096}, {-1801031344,-1169603400}, {-1859775343,-1073741910},
+{-1913422071,-974936918}, {-1961823997,-873460141}, {-2004848713,-769589276},
+{-2042378447,-663608538}, {-2074309994,-555809377}, {-2100556013,-446486785},
+{-2121044568,-335940406}, {-2135719499,-224473240}, {-2144540612,-112390298},
+{2147483647,0}, {2140863671,-168489630}, {2121044558,-335940465},
+{2088148500,-501320115}, {2042378310,-663608960}, {1984016179,-821806435},
+{1913421927,-974937199}, {1831030826,-1122057097}, {1737350743,-1262259248},
+{1632959307,-1394679144}, {1518500216,-1518500282}, {1394679073,-1632959368},
+{1262259172,-1737350799}, {1122057017,-1831030875}, {974937230,-1913421912},
+{821806407,-1984016190}, {663608871,-2042378339}, {501319962,-2088148536},
+{335940246,-2121044593}, {168489600,-2140863674}, {-94,-2147483647},
+{-168489787,-2140863659}, {-335940431,-2121044564}, {-501320144,-2088148493},
+{-663609049,-2042378281}, {-821806581,-1984016118}, {-974937397,-1913421827},
+{-1122057395,-1831030643}, {-1262259116,-1737350839}, {-1394679021,-1632959413},
+{-1518500258,-1518500240}, {-1632959429,-1394679001}, {-1737350854,-1262259096},
+{-1831030924,-1122056937}, {-1913422071,-974936918}, {-1984016324,-821806084},
+{-2042378447,-663608538}, {-2088148499,-501320119}, {-2121044568,-335940406},
+{-2140863681,-168489506}, {-2147483647,188}, {-2140863651,168489881},
+{-2121044509,335940777}, {-2088148411,501320484}, {-2042378331,663608895},
+{-1984016181,821806431}, {-1913421900,974937252}, {-1831030728,1122057257},
+{-1737350633,1262259400}, {-1632959185,1394679287}, {-1518499993,1518500506},
+{-1394678735,1632959656}, {-1262258813,1737351059}, {-1122056638,1831031107},
+{-974936606,1913422229}, {-821805761,1984016458}, {-663609179,2042378239},
+{-501320277,2088148461}, {-335940566,2121044542}, {-168489668,2140863668},
+};
+static const ne10_fft_cpx_int32_t ne10_twiddles_120[120] = {
+{0,0}, {2147483647,0}, {2147483647,0},
+{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394},
+{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496},
+{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096},
+{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152},
+{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313},
+{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424},
+{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496},
+{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268},
+{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785},
+{2147483647,0}, {2144540595,-112390613}, {2135719506,-224473172},
+{2121044558,-335940465}, {2100555974,-446486968}, {2074309912,-555809682},
+{2042378310,-663608960}, {2004848691,-769589332}, {1961823921,-873460313},
+{1913421927,-974937199}, {1859775377,-1073741851}, {1801031311,-1169603450},
+{1737350743,-1262259248}, {1668908218,-1351455280}, {1595891331,-1436947067},
+{1518500216,-1518500282}, {1436946998,-1595891394}, {1351455207,-1668908277},
+{1262259172,-1737350799}, {1169603371,-1801031362}, {1073741769,-1859775424},
+{974937230,-1913421912}, {873460227,-1961823959}, {769589125,-2004848771},
+{663608871,-2042378339}, {555809715,-2074309903}, {446486876,-2100555994},
+{335940246,-2121044593}, {224473078,-2135719516}, {112390647,-2144540593},
+{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968},
+{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851},
+{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394},
+{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959},
+{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516},
+{-94,-2147483647}, {-224473265,-2135719496}, {-446487060,-2100555955},
+{-663609049,-2042378281}, {-873460398,-1961823883}, {-1073741932,-1859775330},
+{-1262259116,-1737350839}, {-1436947137,-1595891268}, {-1595891628,-1436946738},
+{-1737350854,-1262259096}, {-1859775343,-1073741910}, {-1961823997,-873460141},
+{-2042378447,-663608538}, {-2100556013,-446486785}, {-2135719499,-224473240},
+{2147483647,0}, {2121044558,-335940465}, {2042378310,-663608960},
+{1913421927,-974937199}, {1737350743,-1262259248}, {1518500216,-1518500282},
+{1262259172,-1737350799}, {974937230,-1913421912}, {663608871,-2042378339},
+{335940246,-2121044593}, {-94,-2147483647}, {-335940431,-2121044564},
+{-663609049,-2042378281}, {-974937397,-1913421827}, {-1262259116,-1737350839},
+{-1518500258,-1518500240}, {-1737350854,-1262259096}, {-1913422071,-974936918},
+{-2042378447,-663608538}, {-2121044568,-335940406}, {-2147483647,188},
+{-2121044509,335940777}, {-2042378331,663608895}, {-1913421900,974937252},
+{-1737350633,1262259400}, {-1518499993,1518500506}, {-1262258813,1737351059},
+{-974936606,1913422229}, {-663609179,2042378239}, {-335940566,2121044542},
+};
+static const ne10_fft_cpx_int32_t ne10_twiddles_60[60] = {
+{0,0}, {2147483647,0}, {2147483647,0},
+{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394},
+{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496},
+{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096},
+{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152},
+{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968},
+{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851},
+{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394},
+{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959},
+{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516},
+{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313},
+{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424},
+{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496},
+{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268},
+{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785},
+{2147483647,0}, {2042378310,-663608960}, {1737350743,-1262259248},
+{1262259172,-1737350799}, {663608871,-2042378339}, {-94,-2147483647},
+{-663609049,-2042378281}, {-1262259116,-1737350839}, {-1737350854,-1262259096},
+{-2042378447,-663608538}, {-2147483647,188}, {-2042378331,663608895},
+{-1737350633,1262259400}, {-1262258813,1737351059}, {-663609179,2042378239},
+};
+static const ne10_fft_state_int32_t ne10_fft_state_int32_t_480 = {
+120,
+(ne10_int32_t *)ne10_factors_480,
+(ne10_fft_cpx_int32_t *)ne10_twiddles_480,
+NULL,
+(ne10_fft_cpx_int32_t *)&ne10_twiddles_480[120],
+};
+static const arch_fft_state cfg_arch_480 = {
+1,
+(void *)&ne10_fft_state_int32_t_480,
+};
+
+static const ne10_fft_state_int32_t ne10_fft_state_int32_t_240 = {
+60,
+(ne10_int32_t *)ne10_factors_240,
+(ne10_fft_cpx_int32_t *)ne10_twiddles_240,
+NULL,
+(ne10_fft_cpx_int32_t *)&ne10_twiddles_240[60],
+};
+static const arch_fft_state cfg_arch_240 = {
+1,
+(void *)&ne10_fft_state_int32_t_240,
+};
+
+static const ne10_fft_state_int32_t ne10_fft_state_int32_t_120 = {
+30,
+(ne10_int32_t *)ne10_factors_120,
+(ne10_fft_cpx_int32_t *)ne10_twiddles_120,
+NULL,
+(ne10_fft_cpx_int32_t *)&ne10_twiddles_120[30],
+};
+static const arch_fft_state cfg_arch_120 = {
+1,
+(void *)&ne10_fft_state_int32_t_120,
+};
+
+static const ne10_fft_state_int32_t ne10_fft_state_int32_t_60 = {
+15,
+(ne10_int32_t *)ne10_factors_60,
+(ne10_fft_cpx_int32_t *)ne10_twiddles_60,
+NULL,
+(ne10_fft_cpx_int32_t *)&ne10_twiddles_60[15],
+};
+static const arch_fft_state cfg_arch_60 = {
+1,
+(void *)&ne10_fft_state_int32_t_60,
+};
+
+#endif /* end NE10_FFT_PARAMS48000_960 */
diff --git a/lib/rbcodec/codecs/libopus/celt/static_modes_float.h b/lib/rbcodec/codecs/libopus/celt/static_modes_float.h
new file mode 100644
index 0000000000..e102a38391
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/static_modes_float.h
@@ -0,0 +1,888 @@
+/* The contents of this file was automatically generated by dump_modes.c
+ with arguments: 48000 960
+ It contains static definitions for some pre-defined modes. */
+#include "modes.h"
+#include "rate.h"
+
+#ifdef HAVE_ARM_NE10
+#define OVERRIDE_FFT 1
+#include "static_modes_float_arm_ne10.h"
+#endif
+
+#ifndef DEF_WINDOW120
+#define DEF_WINDOW120
+static const opus_val16 window120[120] = {
+6.7286966e-05f, 0.00060551348f, 0.0016815970f, 0.0032947962f, 0.0054439943f,
+0.0081276923f, 0.011344001f, 0.015090633f, 0.019364886f, 0.024163635f,
+0.029483315f, 0.035319905f, 0.041668911f, 0.048525347f, 0.055883718f,
+0.063737999f, 0.072081616f, 0.080907428f, 0.090207705f, 0.099974111f,
+0.11019769f, 0.12086883f, 0.13197729f, 0.14351214f, 0.15546177f,
+0.16781389f, 0.18055550f, 0.19367290f, 0.20715171f, 0.22097682f,
+0.23513243f, 0.24960208f, 0.26436860f, 0.27941419f, 0.29472040f,
+0.31026818f, 0.32603788f, 0.34200931f, 0.35816177f, 0.37447407f,
+0.39092462f, 0.40749142f, 0.42415215f, 0.44088423f, 0.45766484f,
+0.47447104f, 0.49127978f, 0.50806798f, 0.52481261f, 0.54149077f,
+0.55807973f, 0.57455701f, 0.59090049f, 0.60708841f, 0.62309951f,
+0.63891306f, 0.65450896f, 0.66986776f, 0.68497077f, 0.69980010f,
+0.71433873f, 0.72857055f, 0.74248043f, 0.75605424f, 0.76927895f,
+0.78214257f, 0.79463430f, 0.80674445f, 0.81846456f, 0.82978733f,
+0.84070669f, 0.85121779f, 0.86131698f, 0.87100183f, 0.88027111f,
+0.88912479f, 0.89756398f, 0.90559094f, 0.91320904f, 0.92042270f,
+0.92723738f, 0.93365955f, 0.93969656f, 0.94535671f, 0.95064907f,
+0.95558353f, 0.96017067f, 0.96442171f, 0.96834849f, 0.97196334f,
+0.97527906f, 0.97830883f, 0.98106616f, 0.98356480f, 0.98581869f,
+0.98784191f, 0.98964856f, 0.99125274f, 0.99266849f, 0.99390969f,
+0.99499004f, 0.99592297f, 0.99672162f, 0.99739874f, 0.99796667f,
+0.99843728f, 0.99882195f, 0.99913147f, 0.99937606f, 0.99956527f,
+0.99970802f, 0.99981248f, 0.99988613f, 0.99993565f, 0.99996697f,
+0.99998518f, 0.99999457f, 0.99999859f, 0.99999982f, 1.0000000f,
+};
+#endif
+
+#ifndef DEF_LOGN400
+#define DEF_LOGN400
+static const opus_int16 logN400[21] = {
+0, 0, 0, 0, 0, 0, 0, 0, 8, 8, 8, 8, 16, 16, 16, 21, 21, 24, 29, 34, 36, };
+#endif
+
+#ifndef DEF_PULSE_CACHE50
+#define DEF_PULSE_CACHE50
+static const opus_int16 cache_index50[105] = {
+-1, -1, -1, -1, -1, -1, -1, -1, 0, 0, 0, 0, 41, 41, 41,
+82, 82, 123, 164, 200, 222, 0, 0, 0, 0, 0, 0, 0, 0, 41,
+41, 41, 41, 123, 123, 123, 164, 164, 240, 266, 283, 295, 41, 41, 41,
+41, 41, 41, 41, 41, 123, 123, 123, 123, 240, 240, 240, 266, 266, 305,
+318, 328, 336, 123, 123, 123, 123, 123, 123, 123, 123, 240, 240, 240, 240,
+305, 305, 305, 318, 318, 343, 351, 358, 364, 240, 240, 240, 240, 240, 240,
+240, 240, 305, 305, 305, 305, 343, 343, 343, 351, 351, 370, 376, 382, 387,
+};
+static const unsigned char cache_bits50[392] = {
+40, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7,
+7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 40, 15, 23, 28,
+31, 34, 36, 38, 39, 41, 42, 43, 44, 45, 46, 47, 47, 49, 50,
+51, 52, 53, 54, 55, 55, 57, 58, 59, 60, 61, 62, 63, 63, 65,
+66, 67, 68, 69, 70, 71, 71, 40, 20, 33, 41, 48, 53, 57, 61,
+64, 66, 69, 71, 73, 75, 76, 78, 80, 82, 85, 87, 89, 91, 92,
+94, 96, 98, 101, 103, 105, 107, 108, 110, 112, 114, 117, 119, 121, 123,
+124, 126, 128, 40, 23, 39, 51, 60, 67, 73, 79, 83, 87, 91, 94,
+97, 100, 102, 105, 107, 111, 115, 118, 121, 124, 126, 129, 131, 135, 139,
+142, 145, 148, 150, 153, 155, 159, 163, 166, 169, 172, 174, 177, 179, 35,
+28, 49, 65, 78, 89, 99, 107, 114, 120, 126, 132, 136, 141, 145, 149,
+153, 159, 165, 171, 176, 180, 185, 189, 192, 199, 205, 211, 216, 220, 225,
+229, 232, 239, 245, 251, 21, 33, 58, 79, 97, 112, 125, 137, 148, 157,
+166, 174, 182, 189, 195, 201, 207, 217, 227, 235, 243, 251, 17, 35, 63,
+86, 106, 123, 139, 152, 165, 177, 187, 197, 206, 214, 222, 230, 237, 250,
+25, 31, 55, 75, 91, 105, 117, 128, 138, 146, 154, 161, 168, 174, 180,
+185, 190, 200, 208, 215, 222, 229, 235, 240, 245, 255, 16, 36, 65, 89,
+110, 128, 144, 159, 173, 185, 196, 207, 217, 226, 234, 242, 250, 11, 41,
+74, 103, 128, 151, 172, 191, 209, 225, 241, 255, 9, 43, 79, 110, 138,
+163, 186, 207, 227, 246, 12, 39, 71, 99, 123, 144, 164, 182, 198, 214,
+228, 241, 253, 9, 44, 81, 113, 142, 168, 192, 214, 235, 255, 7, 49,
+90, 127, 160, 191, 220, 247, 6, 51, 95, 134, 170, 203, 234, 7, 47,
+87, 123, 155, 184, 212, 237, 6, 52, 97, 137, 174, 208, 240, 5, 57,
+106, 151, 192, 231, 5, 59, 111, 158, 202, 243, 5, 55, 103, 147, 187,
+224, 5, 60, 113, 161, 206, 248, 4, 65, 122, 175, 224, 4, 67, 127,
+182, 234, };
+static const unsigned char cache_caps50[168] = {
+224, 224, 224, 224, 224, 224, 224, 224, 160, 160, 160, 160, 185, 185, 185,
+178, 178, 168, 134, 61, 37, 224, 224, 224, 224, 224, 224, 224, 224, 240,
+240, 240, 240, 207, 207, 207, 198, 198, 183, 144, 66, 40, 160, 160, 160,
+160, 160, 160, 160, 160, 185, 185, 185, 185, 193, 193, 193, 183, 183, 172,
+138, 64, 38, 240, 240, 240, 240, 240, 240, 240, 240, 207, 207, 207, 207,
+204, 204, 204, 193, 193, 180, 143, 66, 40, 185, 185, 185, 185, 185, 185,
+185, 185, 193, 193, 193, 193, 193, 193, 193, 183, 183, 172, 138, 65, 39,
+207, 207, 207, 207, 207, 207, 207, 207, 204, 204, 204, 204, 201, 201, 201,
+188, 188, 176, 141, 66, 40, 193, 193, 193, 193, 193, 193, 193, 193, 193,
+193, 193, 193, 194, 194, 194, 184, 184, 173, 139, 65, 39, 204, 204, 204,
+204, 204, 204, 204, 204, 201, 201, 201, 201, 198, 198, 198, 187, 187, 175,
+140, 66, 40, };
+#endif
+
+#ifndef FFT_TWIDDLES48000_960
+#define FFT_TWIDDLES48000_960
+static const kiss_twiddle_cpx fft_twiddles48000_960[480] = {
+{1.0000000f, -0.0000000f}, {0.99991433f, -0.013089596f},
+{0.99965732f, -0.026176948f}, {0.99922904f, -0.039259816f},
+{0.99862953f, -0.052335956f}, {0.99785892f, -0.065403129f},
+{0.99691733f, -0.078459096f}, {0.99580493f, -0.091501619f},
+{0.99452190f, -0.10452846f}, {0.99306846f, -0.11753740f},
+{0.99144486f, -0.13052619f}, {0.98965139f, -0.14349262f},
+{0.98768834f, -0.15643447f}, {0.98555606f, -0.16934950f},
+{0.98325491f, -0.18223553f}, {0.98078528f, -0.19509032f},
+{0.97814760f, -0.20791169f}, {0.97534232f, -0.22069744f},
+{0.97236992f, -0.23344536f}, {0.96923091f, -0.24615329f},
+{0.96592583f, -0.25881905f}, {0.96245524f, -0.27144045f},
+{0.95881973f, -0.28401534f}, {0.95501994f, -0.29654157f},
+{0.95105652f, -0.30901699f}, {0.94693013f, -0.32143947f},
+{0.94264149f, -0.33380686f}, {0.93819134f, -0.34611706f},
+{0.93358043f, -0.35836795f}, {0.92880955f, -0.37055744f},
+{0.92387953f, -0.38268343f}, {0.91879121f, -0.39474386f},
+{0.91354546f, -0.40673664f}, {0.90814317f, -0.41865974f},
+{0.90258528f, -0.43051110f}, {0.89687274f, -0.44228869f},
+{0.89100652f, -0.45399050f}, {0.88498764f, -0.46561452f},
+{0.87881711f, -0.47715876f}, {0.87249601f, -0.48862124f},
+{0.86602540f, -0.50000000f}, {0.85940641f, -0.51129309f},
+{0.85264016f, -0.52249856f}, {0.84572782f, -0.53361452f},
+{0.83867057f, -0.54463904f}, {0.83146961f, -0.55557023f},
+{0.82412619f, -0.56640624f}, {0.81664156f, -0.57714519f},
+{0.80901699f, -0.58778525f}, {0.80125381f, -0.59832460f},
+{0.79335334f, -0.60876143f}, {0.78531693f, -0.61909395f},
+{0.77714596f, -0.62932039f}, {0.76884183f, -0.63943900f},
+{0.76040597f, -0.64944805f}, {0.75183981f, -0.65934582f},
+{0.74314483f, -0.66913061f}, {0.73432251f, -0.67880075f},
+{0.72537437f, -0.68835458f}, {0.71630194f, -0.69779046f},
+{0.70710678f, -0.70710678f}, {0.69779046f, -0.71630194f},
+{0.68835458f, -0.72537437f}, {0.67880075f, -0.73432251f},
+{0.66913061f, -0.74314483f}, {0.65934582f, -0.75183981f},
+{0.64944805f, -0.76040597f}, {0.63943900f, -0.76884183f},
+{0.62932039f, -0.77714596f}, {0.61909395f, -0.78531693f},
+{0.60876143f, -0.79335334f}, {0.59832460f, -0.80125381f},
+{0.58778525f, -0.80901699f}, {0.57714519f, -0.81664156f},
+{0.56640624f, -0.82412619f}, {0.55557023f, -0.83146961f},
+{0.54463904f, -0.83867057f}, {0.53361452f, -0.84572782f},
+{0.52249856f, -0.85264016f}, {0.51129309f, -0.85940641f},
+{0.50000000f, -0.86602540f}, {0.48862124f, -0.87249601f},
+{0.47715876f, -0.87881711f}, {0.46561452f, -0.88498764f},
+{0.45399050f, -0.89100652f}, {0.44228869f, -0.89687274f},
+{0.43051110f, -0.90258528f}, {0.41865974f, -0.90814317f},
+{0.40673664f, -0.91354546f}, {0.39474386f, -0.91879121f},
+{0.38268343f, -0.92387953f}, {0.37055744f, -0.92880955f},
+{0.35836795f, -0.93358043f}, {0.34611706f, -0.93819134f},
+{0.33380686f, -0.94264149f}, {0.32143947f, -0.94693013f},
+{0.30901699f, -0.95105652f}, {0.29654157f, -0.95501994f},
+{0.28401534f, -0.95881973f}, {0.27144045f, -0.96245524f},
+{0.25881905f, -0.96592583f}, {0.24615329f, -0.96923091f},
+{0.23344536f, -0.97236992f}, {0.22069744f, -0.97534232f},
+{0.20791169f, -0.97814760f}, {0.19509032f, -0.98078528f},
+{0.18223553f, -0.98325491f}, {0.16934950f, -0.98555606f},
+{0.15643447f, -0.98768834f}, {0.14349262f, -0.98965139f},
+{0.13052619f, -0.99144486f}, {0.11753740f, -0.99306846f},
+{0.10452846f, -0.99452190f}, {0.091501619f, -0.99580493f},
+{0.078459096f, -0.99691733f}, {0.065403129f, -0.99785892f},
+{0.052335956f, -0.99862953f}, {0.039259816f, -0.99922904f},
+{0.026176948f, -0.99965732f}, {0.013089596f, -0.99991433f},
+{6.1230318e-17f, -1.0000000f}, {-0.013089596f, -0.99991433f},
+{-0.026176948f, -0.99965732f}, {-0.039259816f, -0.99922904f},
+{-0.052335956f, -0.99862953f}, {-0.065403129f, -0.99785892f},
+{-0.078459096f, -0.99691733f}, {-0.091501619f, -0.99580493f},
+{-0.10452846f, -0.99452190f}, {-0.11753740f, -0.99306846f},
+{-0.13052619f, -0.99144486f}, {-0.14349262f, -0.98965139f},
+{-0.15643447f, -0.98768834f}, {-0.16934950f, -0.98555606f},
+{-0.18223553f, -0.98325491f}, {-0.19509032f, -0.98078528f},
+{-0.20791169f, -0.97814760f}, {-0.22069744f, -0.97534232f},
+{-0.23344536f, -0.97236992f}, {-0.24615329f, -0.96923091f},
+{-0.25881905f, -0.96592583f}, {-0.27144045f, -0.96245524f},
+{-0.28401534f, -0.95881973f}, {-0.29654157f, -0.95501994f},
+{-0.30901699f, -0.95105652f}, {-0.32143947f, -0.94693013f},
+{-0.33380686f, -0.94264149f}, {-0.34611706f, -0.93819134f},
+{-0.35836795f, -0.93358043f}, {-0.37055744f, -0.92880955f},
+{-0.38268343f, -0.92387953f}, {-0.39474386f, -0.91879121f},
+{-0.40673664f, -0.91354546f}, {-0.41865974f, -0.90814317f},
+{-0.43051110f, -0.90258528f}, {-0.44228869f, -0.89687274f},
+{-0.45399050f, -0.89100652f}, {-0.46561452f, -0.88498764f},
+{-0.47715876f, -0.87881711f}, {-0.48862124f, -0.87249601f},
+{-0.50000000f, -0.86602540f}, {-0.51129309f, -0.85940641f},
+{-0.52249856f, -0.85264016f}, {-0.53361452f, -0.84572782f},
+{-0.54463904f, -0.83867057f}, {-0.55557023f, -0.83146961f},
+{-0.56640624f, -0.82412619f}, {-0.57714519f, -0.81664156f},
+{-0.58778525f, -0.80901699f}, {-0.59832460f, -0.80125381f},
+{-0.60876143f, -0.79335334f}, {-0.61909395f, -0.78531693f},
+{-0.62932039f, -0.77714596f}, {-0.63943900f, -0.76884183f},
+{-0.64944805f, -0.76040597f}, {-0.65934582f, -0.75183981f},
+{-0.66913061f, -0.74314483f}, {-0.67880075f, -0.73432251f},
+{-0.68835458f, -0.72537437f}, {-0.69779046f, -0.71630194f},
+{-0.70710678f, -0.70710678f}, {-0.71630194f, -0.69779046f},
+{-0.72537437f, -0.68835458f}, {-0.73432251f, -0.67880075f},
+{-0.74314483f, -0.66913061f}, {-0.75183981f, -0.65934582f},
+{-0.76040597f, -0.64944805f}, {-0.76884183f, -0.63943900f},
+{-0.77714596f, -0.62932039f}, {-0.78531693f, -0.61909395f},
+{-0.79335334f, -0.60876143f}, {-0.80125381f, -0.59832460f},
+{-0.80901699f, -0.58778525f}, {-0.81664156f, -0.57714519f},
+{-0.82412619f, -0.56640624f}, {-0.83146961f, -0.55557023f},
+{-0.83867057f, -0.54463904f}, {-0.84572782f, -0.53361452f},
+{-0.85264016f, -0.52249856f}, {-0.85940641f, -0.51129309f},
+{-0.86602540f, -0.50000000f}, {-0.87249601f, -0.48862124f},
+{-0.87881711f, -0.47715876f}, {-0.88498764f, -0.46561452f},
+{-0.89100652f, -0.45399050f}, {-0.89687274f, -0.44228869f},
+{-0.90258528f, -0.43051110f}, {-0.90814317f, -0.41865974f},
+{-0.91354546f, -0.40673664f}, {-0.91879121f, -0.39474386f},
+{-0.92387953f, -0.38268343f}, {-0.92880955f, -0.37055744f},
+{-0.93358043f, -0.35836795f}, {-0.93819134f, -0.34611706f},
+{-0.94264149f, -0.33380686f}, {-0.94693013f, -0.32143947f},
+{-0.95105652f, -0.30901699f}, {-0.95501994f, -0.29654157f},
+{-0.95881973f, -0.28401534f}, {-0.96245524f, -0.27144045f},
+{-0.96592583f, -0.25881905f}, {-0.96923091f, -0.24615329f},
+{-0.97236992f, -0.23344536f}, {-0.97534232f, -0.22069744f},
+{-0.97814760f, -0.20791169f}, {-0.98078528f, -0.19509032f},
+{-0.98325491f, -0.18223553f}, {-0.98555606f, -0.16934950f},
+{-0.98768834f, -0.15643447f}, {-0.98965139f, -0.14349262f},
+{-0.99144486f, -0.13052619f}, {-0.99306846f, -0.11753740f},
+{-0.99452190f, -0.10452846f}, {-0.99580493f, -0.091501619f},
+{-0.99691733f, -0.078459096f}, {-0.99785892f, -0.065403129f},
+{-0.99862953f, -0.052335956f}, {-0.99922904f, -0.039259816f},
+{-0.99965732f, -0.026176948f}, {-0.99991433f, -0.013089596f},
+{-1.0000000f, -1.2246064e-16f}, {-0.99991433f, 0.013089596f},
+{-0.99965732f, 0.026176948f}, {-0.99922904f, 0.039259816f},
+{-0.99862953f, 0.052335956f}, {-0.99785892f, 0.065403129f},
+{-0.99691733f, 0.078459096f}, {-0.99580493f, 0.091501619f},
+{-0.99452190f, 0.10452846f}, {-0.99306846f, 0.11753740f},
+{-0.99144486f, 0.13052619f}, {-0.98965139f, 0.14349262f},
+{-0.98768834f, 0.15643447f}, {-0.98555606f, 0.16934950f},
+{-0.98325491f, 0.18223553f}, {-0.98078528f, 0.19509032f},
+{-0.97814760f, 0.20791169f}, {-0.97534232f, 0.22069744f},
+{-0.97236992f, 0.23344536f}, {-0.96923091f, 0.24615329f},
+{-0.96592583f, 0.25881905f}, {-0.96245524f, 0.27144045f},
+{-0.95881973f, 0.28401534f}, {-0.95501994f, 0.29654157f},
+{-0.95105652f, 0.30901699f}, {-0.94693013f, 0.32143947f},
+{-0.94264149f, 0.33380686f}, {-0.93819134f, 0.34611706f},
+{-0.93358043f, 0.35836795f}, {-0.92880955f, 0.37055744f},
+{-0.92387953f, 0.38268343f}, {-0.91879121f, 0.39474386f},
+{-0.91354546f, 0.40673664f}, {-0.90814317f, 0.41865974f},
+{-0.90258528f, 0.43051110f}, {-0.89687274f, 0.44228869f},
+{-0.89100652f, 0.45399050f}, {-0.88498764f, 0.46561452f},
+{-0.87881711f, 0.47715876f}, {-0.87249601f, 0.48862124f},
+{-0.86602540f, 0.50000000f}, {-0.85940641f, 0.51129309f},
+{-0.85264016f, 0.52249856f}, {-0.84572782f, 0.53361452f},
+{-0.83867057f, 0.54463904f}, {-0.83146961f, 0.55557023f},
+{-0.82412619f, 0.56640624f}, {-0.81664156f, 0.57714519f},
+{-0.80901699f, 0.58778525f}, {-0.80125381f, 0.59832460f},
+{-0.79335334f, 0.60876143f}, {-0.78531693f, 0.61909395f},
+{-0.77714596f, 0.62932039f}, {-0.76884183f, 0.63943900f},
+{-0.76040597f, 0.64944805f}, {-0.75183981f, 0.65934582f},
+{-0.74314483f, 0.66913061f}, {-0.73432251f, 0.67880075f},
+{-0.72537437f, 0.68835458f}, {-0.71630194f, 0.69779046f},
+{-0.70710678f, 0.70710678f}, {-0.69779046f, 0.71630194f},
+{-0.68835458f, 0.72537437f}, {-0.67880075f, 0.73432251f},
+{-0.66913061f, 0.74314483f}, {-0.65934582f, 0.75183981f},
+{-0.64944805f, 0.76040597f}, {-0.63943900f, 0.76884183f},
+{-0.62932039f, 0.77714596f}, {-0.61909395f, 0.78531693f},
+{-0.60876143f, 0.79335334f}, {-0.59832460f, 0.80125381f},
+{-0.58778525f, 0.80901699f}, {-0.57714519f, 0.81664156f},
+{-0.56640624f, 0.82412619f}, {-0.55557023f, 0.83146961f},
+{-0.54463904f, 0.83867057f}, {-0.53361452f, 0.84572782f},
+{-0.52249856f, 0.85264016f}, {-0.51129309f, 0.85940641f},
+{-0.50000000f, 0.86602540f}, {-0.48862124f, 0.87249601f},
+{-0.47715876f, 0.87881711f}, {-0.46561452f, 0.88498764f},
+{-0.45399050f, 0.89100652f}, {-0.44228869f, 0.89687274f},
+{-0.43051110f, 0.90258528f}, {-0.41865974f, 0.90814317f},
+{-0.40673664f, 0.91354546f}, {-0.39474386f, 0.91879121f},
+{-0.38268343f, 0.92387953f}, {-0.37055744f, 0.92880955f},
+{-0.35836795f, 0.93358043f}, {-0.34611706f, 0.93819134f},
+{-0.33380686f, 0.94264149f}, {-0.32143947f, 0.94693013f},
+{-0.30901699f, 0.95105652f}, {-0.29654157f, 0.95501994f},
+{-0.28401534f, 0.95881973f}, {-0.27144045f, 0.96245524f},
+{-0.25881905f, 0.96592583f}, {-0.24615329f, 0.96923091f},
+{-0.23344536f, 0.97236992f}, {-0.22069744f, 0.97534232f},
+{-0.20791169f, 0.97814760f}, {-0.19509032f, 0.98078528f},
+{-0.18223553f, 0.98325491f}, {-0.16934950f, 0.98555606f},
+{-0.15643447f, 0.98768834f}, {-0.14349262f, 0.98965139f},
+{-0.13052619f, 0.99144486f}, {-0.11753740f, 0.99306846f},
+{-0.10452846f, 0.99452190f}, {-0.091501619f, 0.99580493f},
+{-0.078459096f, 0.99691733f}, {-0.065403129f, 0.99785892f},
+{-0.052335956f, 0.99862953f}, {-0.039259816f, 0.99922904f},
+{-0.026176948f, 0.99965732f}, {-0.013089596f, 0.99991433f},
+{-1.8369095e-16f, 1.0000000f}, {0.013089596f, 0.99991433f},
+{0.026176948f, 0.99965732f}, {0.039259816f, 0.99922904f},
+{0.052335956f, 0.99862953f}, {0.065403129f, 0.99785892f},
+{0.078459096f, 0.99691733f}, {0.091501619f, 0.99580493f},
+{0.10452846f, 0.99452190f}, {0.11753740f, 0.99306846f},
+{0.13052619f, 0.99144486f}, {0.14349262f, 0.98965139f},
+{0.15643447f, 0.98768834f}, {0.16934950f, 0.98555606f},
+{0.18223553f, 0.98325491f}, {0.19509032f, 0.98078528f},
+{0.20791169f, 0.97814760f}, {0.22069744f, 0.97534232f},
+{0.23344536f, 0.97236992f}, {0.24615329f, 0.96923091f},
+{0.25881905f, 0.96592583f}, {0.27144045f, 0.96245524f},
+{0.28401534f, 0.95881973f}, {0.29654157f, 0.95501994f},
+{0.30901699f, 0.95105652f}, {0.32143947f, 0.94693013f},
+{0.33380686f, 0.94264149f}, {0.34611706f, 0.93819134f},
+{0.35836795f, 0.93358043f}, {0.37055744f, 0.92880955f},
+{0.38268343f, 0.92387953f}, {0.39474386f, 0.91879121f},
+{0.40673664f, 0.91354546f}, {0.41865974f, 0.90814317f},
+{0.43051110f, 0.90258528f}, {0.44228869f, 0.89687274f},
+{0.45399050f, 0.89100652f}, {0.46561452f, 0.88498764f},
+{0.47715876f, 0.87881711f}, {0.48862124f, 0.87249601f},
+{0.50000000f, 0.86602540f}, {0.51129309f, 0.85940641f},
+{0.52249856f, 0.85264016f}, {0.53361452f, 0.84572782f},
+{0.54463904f, 0.83867057f}, {0.55557023f, 0.83146961f},
+{0.56640624f, 0.82412619f}, {0.57714519f, 0.81664156f},
+{0.58778525f, 0.80901699f}, {0.59832460f, 0.80125381f},
+{0.60876143f, 0.79335334f}, {0.61909395f, 0.78531693f},
+{0.62932039f, 0.77714596f}, {0.63943900f, 0.76884183f},
+{0.64944805f, 0.76040597f}, {0.65934582f, 0.75183981f},
+{0.66913061f, 0.74314483f}, {0.67880075f, 0.73432251f},
+{0.68835458f, 0.72537437f}, {0.69779046f, 0.71630194f},
+{0.70710678f, 0.70710678f}, {0.71630194f, 0.69779046f},
+{0.72537437f, 0.68835458f}, {0.73432251f, 0.67880075f},
+{0.74314483f, 0.66913061f}, {0.75183981f, 0.65934582f},
+{0.76040597f, 0.64944805f}, {0.76884183f, 0.63943900f},
+{0.77714596f, 0.62932039f}, {0.78531693f, 0.61909395f},
+{0.79335334f, 0.60876143f}, {0.80125381f, 0.59832460f},
+{0.80901699f, 0.58778525f}, {0.81664156f, 0.57714519f},
+{0.82412619f, 0.56640624f}, {0.83146961f, 0.55557023f},
+{0.83867057f, 0.54463904f}, {0.84572782f, 0.53361452f},
+{0.85264016f, 0.52249856f}, {0.85940641f, 0.51129309f},
+{0.86602540f, 0.50000000f}, {0.87249601f, 0.48862124f},
+{0.87881711f, 0.47715876f}, {0.88498764f, 0.46561452f},
+{0.89100652f, 0.45399050f}, {0.89687274f, 0.44228869f},
+{0.90258528f, 0.43051110f}, {0.90814317f, 0.41865974f},
+{0.91354546f, 0.40673664f}, {0.91879121f, 0.39474386f},
+{0.92387953f, 0.38268343f}, {0.92880955f, 0.37055744f},
+{0.93358043f, 0.35836795f}, {0.93819134f, 0.34611706f},
+{0.94264149f, 0.33380686f}, {0.94693013f, 0.32143947f},
+{0.95105652f, 0.30901699f}, {0.95501994f, 0.29654157f},
+{0.95881973f, 0.28401534f}, {0.96245524f, 0.27144045f},
+{0.96592583f, 0.25881905f}, {0.96923091f, 0.24615329f},
+{0.97236992f, 0.23344536f}, {0.97534232f, 0.22069744f},
+{0.97814760f, 0.20791169f}, {0.98078528f, 0.19509032f},
+{0.98325491f, 0.18223553f}, {0.98555606f, 0.16934950f},
+{0.98768834f, 0.15643447f}, {0.98965139f, 0.14349262f},
+{0.99144486f, 0.13052619f}, {0.99306846f, 0.11753740f},
+{0.99452190f, 0.10452846f}, {0.99580493f, 0.091501619f},
+{0.99691733f, 0.078459096f}, {0.99785892f, 0.065403129f},
+{0.99862953f, 0.052335956f}, {0.99922904f, 0.039259816f},
+{0.99965732f, 0.026176948f}, {0.99991433f, 0.013089596f},
+};
+#ifndef FFT_BITREV480
+#define FFT_BITREV480
+static const opus_int16 fft_bitrev480[480] = {
+0, 96, 192, 288, 384, 32, 128, 224, 320, 416, 64, 160, 256, 352, 448,
+8, 104, 200, 296, 392, 40, 136, 232, 328, 424, 72, 168, 264, 360, 456,
+16, 112, 208, 304, 400, 48, 144, 240, 336, 432, 80, 176, 272, 368, 464,
+24, 120, 216, 312, 408, 56, 152, 248, 344, 440, 88, 184, 280, 376, 472,
+4, 100, 196, 292, 388, 36, 132, 228, 324, 420, 68, 164, 260, 356, 452,
+12, 108, 204, 300, 396, 44, 140, 236, 332, 428, 76, 172, 268, 364, 460,
+20, 116, 212, 308, 404, 52, 148, 244, 340, 436, 84, 180, 276, 372, 468,
+28, 124, 220, 316, 412, 60, 156, 252, 348, 444, 92, 188, 284, 380, 476,
+1, 97, 193, 289, 385, 33, 129, 225, 321, 417, 65, 161, 257, 353, 449,
+9, 105, 201, 297, 393, 41, 137, 233, 329, 425, 73, 169, 265, 361, 457,
+17, 113, 209, 305, 401, 49, 145, 241, 337, 433, 81, 177, 273, 369, 465,
+25, 121, 217, 313, 409, 57, 153, 249, 345, 441, 89, 185, 281, 377, 473,
+5, 101, 197, 293, 389, 37, 133, 229, 325, 421, 69, 165, 261, 357, 453,
+13, 109, 205, 301, 397, 45, 141, 237, 333, 429, 77, 173, 269, 365, 461,
+21, 117, 213, 309, 405, 53, 149, 245, 341, 437, 85, 181, 277, 373, 469,
+29, 125, 221, 317, 413, 61, 157, 253, 349, 445, 93, 189, 285, 381, 477,
+2, 98, 194, 290, 386, 34, 130, 226, 322, 418, 66, 162, 258, 354, 450,
+10, 106, 202, 298, 394, 42, 138, 234, 330, 426, 74, 170, 266, 362, 458,
+18, 114, 210, 306, 402, 50, 146, 242, 338, 434, 82, 178, 274, 370, 466,
+26, 122, 218, 314, 410, 58, 154, 250, 346, 442, 90, 186, 282, 378, 474,
+6, 102, 198, 294, 390, 38, 134, 230, 326, 422, 70, 166, 262, 358, 454,
+14, 110, 206, 302, 398, 46, 142, 238, 334, 430, 78, 174, 270, 366, 462,
+22, 118, 214, 310, 406, 54, 150, 246, 342, 438, 86, 182, 278, 374, 470,
+30, 126, 222, 318, 414, 62, 158, 254, 350, 446, 94, 190, 286, 382, 478,
+3, 99, 195, 291, 387, 35, 131, 227, 323, 419, 67, 163, 259, 355, 451,
+11, 107, 203, 299, 395, 43, 139, 235, 331, 427, 75, 171, 267, 363, 459,
+19, 115, 211, 307, 403, 51, 147, 243, 339, 435, 83, 179, 275, 371, 467,
+27, 123, 219, 315, 411, 59, 155, 251, 347, 443, 91, 187, 283, 379, 475,
+7, 103, 199, 295, 391, 39, 135, 231, 327, 423, 71, 167, 263, 359, 455,
+15, 111, 207, 303, 399, 47, 143, 239, 335, 431, 79, 175, 271, 367, 463,
+23, 119, 215, 311, 407, 55, 151, 247, 343, 439, 87, 183, 279, 375, 471,
+31, 127, 223, 319, 415, 63, 159, 255, 351, 447, 95, 191, 287, 383, 479,
+};
+#endif
+
+#ifndef FFT_BITREV240
+#define FFT_BITREV240
+static const opus_int16 fft_bitrev240[240] = {
+0, 48, 96, 144, 192, 16, 64, 112, 160, 208, 32, 80, 128, 176, 224,
+4, 52, 100, 148, 196, 20, 68, 116, 164, 212, 36, 84, 132, 180, 228,
+8, 56, 104, 152, 200, 24, 72, 120, 168, 216, 40, 88, 136, 184, 232,
+12, 60, 108, 156, 204, 28, 76, 124, 172, 220, 44, 92, 140, 188, 236,
+1, 49, 97, 145, 193, 17, 65, 113, 161, 209, 33, 81, 129, 177, 225,
+5, 53, 101, 149, 197, 21, 69, 117, 165, 213, 37, 85, 133, 181, 229,
+9, 57, 105, 153, 201, 25, 73, 121, 169, 217, 41, 89, 137, 185, 233,
+13, 61, 109, 157, 205, 29, 77, 125, 173, 221, 45, 93, 141, 189, 237,
+2, 50, 98, 146, 194, 18, 66, 114, 162, 210, 34, 82, 130, 178, 226,
+6, 54, 102, 150, 198, 22, 70, 118, 166, 214, 38, 86, 134, 182, 230,
+10, 58, 106, 154, 202, 26, 74, 122, 170, 218, 42, 90, 138, 186, 234,
+14, 62, 110, 158, 206, 30, 78, 126, 174, 222, 46, 94, 142, 190, 238,
+3, 51, 99, 147, 195, 19, 67, 115, 163, 211, 35, 83, 131, 179, 227,
+7, 55, 103, 151, 199, 23, 71, 119, 167, 215, 39, 87, 135, 183, 231,
+11, 59, 107, 155, 203, 27, 75, 123, 171, 219, 43, 91, 139, 187, 235,
+15, 63, 111, 159, 207, 31, 79, 127, 175, 223, 47, 95, 143, 191, 239,
+};
+#endif
+
+#ifndef FFT_BITREV120
+#define FFT_BITREV120
+static const opus_int16 fft_bitrev120[120] = {
+0, 24, 48, 72, 96, 8, 32, 56, 80, 104, 16, 40, 64, 88, 112,
+4, 28, 52, 76, 100, 12, 36, 60, 84, 108, 20, 44, 68, 92, 116,
+1, 25, 49, 73, 97, 9, 33, 57, 81, 105, 17, 41, 65, 89, 113,
+5, 29, 53, 77, 101, 13, 37, 61, 85, 109, 21, 45, 69, 93, 117,
+2, 26, 50, 74, 98, 10, 34, 58, 82, 106, 18, 42, 66, 90, 114,
+6, 30, 54, 78, 102, 14, 38, 62, 86, 110, 22, 46, 70, 94, 118,
+3, 27, 51, 75, 99, 11, 35, 59, 83, 107, 19, 43, 67, 91, 115,
+7, 31, 55, 79, 103, 15, 39, 63, 87, 111, 23, 47, 71, 95, 119,
+};
+#endif
+
+#ifndef FFT_BITREV60
+#define FFT_BITREV60
+static const opus_int16 fft_bitrev60[60] = {
+0, 12, 24, 36, 48, 4, 16, 28, 40, 52, 8, 20, 32, 44, 56,
+1, 13, 25, 37, 49, 5, 17, 29, 41, 53, 9, 21, 33, 45, 57,
+2, 14, 26, 38, 50, 6, 18, 30, 42, 54, 10, 22, 34, 46, 58,
+3, 15, 27, 39, 51, 7, 19, 31, 43, 55, 11, 23, 35, 47, 59,
+};
+#endif
+
+#ifndef FFT_STATE48000_960_0
+#define FFT_STATE48000_960_0
+static const kiss_fft_state fft_state48000_960_0 = {
+480, /* nfft */
+0.002083333f, /* scale */
+-1, /* shift */
+{5, 96, 3, 32, 4, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, }, /* factors */
+fft_bitrev480, /* bitrev */
+fft_twiddles48000_960, /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_480,
+#else
+NULL,
+#endif
+};
+#endif
+
+#ifndef FFT_STATE48000_960_1
+#define FFT_STATE48000_960_1
+static const kiss_fft_state fft_state48000_960_1 = {
+240, /* nfft */
+0.004166667f, /* scale */
+1, /* shift */
+{5, 48, 3, 16, 4, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */
+fft_bitrev240, /* bitrev */
+fft_twiddles48000_960, /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_240,
+#else
+NULL,
+#endif
+};
+#endif
+
+#ifndef FFT_STATE48000_960_2
+#define FFT_STATE48000_960_2
+static const kiss_fft_state fft_state48000_960_2 = {
+120, /* nfft */
+0.008333333f, /* scale */
+2, /* shift */
+{5, 24, 3, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */
+fft_bitrev120, /* bitrev */
+fft_twiddles48000_960, /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_120,
+#else
+NULL,
+#endif
+};
+#endif
+
+#ifndef FFT_STATE48000_960_3
+#define FFT_STATE48000_960_3
+static const kiss_fft_state fft_state48000_960_3 = {
+60, /* nfft */
+0.016666667f, /* scale */
+3, /* shift */
+{5, 12, 3, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */
+fft_bitrev60, /* bitrev */
+fft_twiddles48000_960, /* bitrev */
+#ifdef OVERRIDE_FFT
+(arch_fft_state *)&cfg_arch_60,
+#else
+NULL,
+#endif
+};
+#endif
+
+#endif
+
+#ifndef MDCT_TWIDDLES960
+#define MDCT_TWIDDLES960
+static const opus_val16 mdct_twiddles960[1800] = {
+0.99999994f, 0.99999321f, 0.99997580f, 0.99994773f, 0.99990886f,
+0.99985933f, 0.99979913f, 0.99972820f, 0.99964654f, 0.99955416f,
+0.99945110f, 0.99933738f, 0.99921292f, 0.99907774f, 0.99893188f,
+0.99877530f, 0.99860805f, 0.99843007f, 0.99824142f, 0.99804211f,
+0.99783206f, 0.99761140f, 0.99737996f, 0.99713790f, 0.99688518f,
+0.99662173f, 0.99634761f, 0.99606287f, 0.99576741f, 0.99546129f,
+0.99514455f, 0.99481714f, 0.99447906f, 0.99413031f, 0.99377096f,
+0.99340093f, 0.99302030f, 0.99262899f, 0.99222708f, 0.99181455f,
+0.99139136f, 0.99095762f, 0.99051321f, 0.99005818f, 0.98959261f,
+0.98911643f, 0.98862964f, 0.98813224f, 0.98762429f, 0.98710573f,
+0.98657662f, 0.98603696f, 0.98548669f, 0.98492593f, 0.98435456f,
+0.98377270f, 0.98318028f, 0.98257732f, 0.98196387f, 0.98133987f,
+0.98070538f, 0.98006040f, 0.97940493f, 0.97873890f, 0.97806245f,
+0.97737551f, 0.97667813f, 0.97597027f, 0.97525197f, 0.97452319f,
+0.97378403f, 0.97303438f, 0.97227436f, 0.97150391f, 0.97072303f,
+0.96993178f, 0.96913016f, 0.96831810f, 0.96749574f, 0.96666300f,
+0.96581990f, 0.96496642f, 0.96410263f, 0.96322852f, 0.96234411f,
+0.96144938f, 0.96054435f, 0.95962906f, 0.95870346f, 0.95776761f,
+0.95682150f, 0.95586514f, 0.95489854f, 0.95392174f, 0.95293468f,
+0.95193744f, 0.95093000f, 0.94991243f, 0.94888461f, 0.94784665f,
+0.94679856f, 0.94574034f, 0.94467193f, 0.94359344f, 0.94250488f,
+0.94140619f, 0.94029742f, 0.93917859f, 0.93804967f, 0.93691075f,
+0.93576175f, 0.93460274f, 0.93343377f, 0.93225473f, 0.93106574f,
+0.92986679f, 0.92865789f, 0.92743903f, 0.92621022f, 0.92497152f,
+0.92372292f, 0.92246443f, 0.92119598f, 0.91991776f, 0.91862965f,
+0.91733170f, 0.91602397f, 0.91470635f, 0.91337901f, 0.91204184f,
+0.91069490f, 0.90933824f, 0.90797186f, 0.90659571f, 0.90520984f,
+0.90381432f, 0.90240908f, 0.90099424f, 0.89956969f, 0.89813554f,
+0.89669174f, 0.89523834f, 0.89377540f, 0.89230281f, 0.89082074f,
+0.88932908f, 0.88782793f, 0.88631725f, 0.88479710f, 0.88326746f,
+0.88172835f, 0.88017982f, 0.87862182f, 0.87705445f, 0.87547767f,
+0.87389153f, 0.87229604f, 0.87069118f, 0.86907703f, 0.86745358f,
+0.86582077f, 0.86417878f, 0.86252749f, 0.86086690f, 0.85919720f,
+0.85751826f, 0.85583007f, 0.85413277f, 0.85242635f, 0.85071075f,
+0.84898609f, 0.84725231f, 0.84550947f, 0.84375757f, 0.84199661f,
+0.84022665f, 0.83844769f, 0.83665979f, 0.83486289f, 0.83305705f,
+0.83124226f, 0.82941860f, 0.82758605f, 0.82574469f, 0.82389444f,
+0.82203537f, 0.82016748f, 0.81829083f, 0.81640542f, 0.81451124f,
+0.81260836f, 0.81069672f, 0.80877650f, 0.80684757f, 0.80490994f,
+0.80296379f, 0.80100900f, 0.79904562f, 0.79707366f, 0.79509324f,
+0.79310423f, 0.79110676f, 0.78910083f, 0.78708643f, 0.78506362f,
+0.78303236f, 0.78099275f, 0.77894479f, 0.77688843f, 0.77482378f,
+0.77275085f, 0.77066964f, 0.76858020f, 0.76648247f, 0.76437658f,
+0.76226246f, 0.76014024f, 0.75800985f, 0.75587130f, 0.75372469f,
+0.75157005f, 0.74940729f, 0.74723655f, 0.74505776f, 0.74287105f,
+0.74067634f, 0.73847371f, 0.73626316f, 0.73404479f, 0.73181850f,
+0.72958434f, 0.72734243f, 0.72509271f, 0.72283524f, 0.72057003f,
+0.71829706f, 0.71601641f, 0.71372813f, 0.71143216f, 0.70912862f,
+0.70681745f, 0.70449871f, 0.70217246f, 0.69983864f, 0.69749737f,
+0.69514859f, 0.69279242f, 0.69042879f, 0.68805778f, 0.68567938f,
+0.68329364f, 0.68090063f, 0.67850029f, 0.67609268f, 0.67367786f,
+0.67125577f, 0.66882652f, 0.66639012f, 0.66394657f, 0.66149592f,
+0.65903819f, 0.65657341f, 0.65410155f, 0.65162271f, 0.64913690f,
+0.64664418f, 0.64414448f, 0.64163786f, 0.63912445f, 0.63660413f,
+0.63407701f, 0.63154310f, 0.62900239f, 0.62645501f, 0.62390089f,
+0.62134010f, 0.61877263f, 0.61619854f, 0.61361790f, 0.61103064f,
+0.60843682f, 0.60583651f, 0.60322970f, 0.60061646f, 0.59799677f,
+0.59537065f, 0.59273821f, 0.59009939f, 0.58745426f, 0.58480281f,
+0.58214509f, 0.57948118f, 0.57681108f, 0.57413477f, 0.57145232f,
+0.56876373f, 0.56606907f, 0.56336832f, 0.56066155f, 0.55794877f,
+0.55523002f, 0.55250537f, 0.54977477f, 0.54703826f, 0.54429591f,
+0.54154772f, 0.53879374f, 0.53603399f, 0.53326851f, 0.53049731f,
+0.52772039f, 0.52493787f, 0.52214974f, 0.51935595f, 0.51655668f,
+0.51375180f, 0.51094145f, 0.50812566f, 0.50530440f, 0.50247771f,
+0.49964568f, 0.49680826f, 0.49396557f, 0.49111754f, 0.48826426f,
+0.48540577f, 0.48254207f, 0.47967321f, 0.47679919f, 0.47392011f,
+0.47103590f, 0.46814668f, 0.46525243f, 0.46235323f, 0.45944905f,
+0.45653993f, 0.45362595f, 0.45070711f, 0.44778344f, 0.44485497f,
+0.44192174f, 0.43898380f, 0.43604112f, 0.43309379f, 0.43014181f,
+0.42718524f, 0.42422408f, 0.42125839f, 0.41828820f, 0.41531351f,
+0.41233435f, 0.40935081f, 0.40636289f, 0.40337059f, 0.40037400f,
+0.39737311f, 0.39436796f, 0.39135858f, 0.38834500f, 0.38532731f,
+0.38230544f, 0.37927949f, 0.37624949f, 0.37321547f, 0.37017745f,
+0.36713544f, 0.36408952f, 0.36103970f, 0.35798600f, 0.35492846f,
+0.35186714f, 0.34880206f, 0.34573323f, 0.34266070f, 0.33958447f,
+0.33650464f, 0.33342120f, 0.33033419f, 0.32724363f, 0.32414958f,
+0.32105204f, 0.31795108f, 0.31484672f, 0.31173897f, 0.30862790f,
+0.30551350f, 0.30239585f, 0.29927495f, 0.29615086f, 0.29302359f,
+0.28989318f, 0.28675964f, 0.28362307f, 0.28048345f, 0.27734083f,
+0.27419522f, 0.27104670f, 0.26789525f, 0.26474094f, 0.26158381f,
+0.25842386f, 0.25526115f, 0.25209570f, 0.24892756f, 0.24575676f,
+0.24258332f, 0.23940729f, 0.23622867f, 0.23304754f, 0.22986393f,
+0.22667783f, 0.22348931f, 0.22029841f, 0.21710514f, 0.21390954f,
+0.21071166f, 0.20751151f, 0.20430915f, 0.20110460f, 0.19789790f,
+0.19468907f, 0.19147816f, 0.18826519f, 0.18505022f, 0.18183327f,
+0.17861435f, 0.17539354f, 0.17217083f, 0.16894630f, 0.16571994f,
+0.16249183f, 0.15926196f, 0.15603039f, 0.15279715f, 0.14956227f,
+0.14632578f, 0.14308774f, 0.13984816f, 0.13660708f, 0.13336454f,
+0.13012058f, 0.12687522f, 0.12362850f, 0.12038045f, 0.11713112f,
+0.11388054f, 0.11062872f, 0.10737573f, 0.10412160f, 0.10086634f,
+0.097609997f, 0.094352618f, 0.091094226f, 0.087834857f, 0.084574550f,
+0.081313334f, 0.078051247f, 0.074788325f, 0.071524605f, 0.068260118f,
+0.064994894f, 0.061728980f, 0.058462404f, 0.055195201f, 0.051927410f,
+0.048659060f, 0.045390189f, 0.042120833f, 0.038851023f, 0.035580799f,
+0.032310195f, 0.029039243f, 0.025767982f, 0.022496443f, 0.019224664f,
+0.015952680f, 0.012680525f, 0.0094082337f, 0.0061358409f, 0.0028633832f,
+-0.00040910527f, -0.0036815894f, -0.0069540343f, -0.010226404f, -0.013498665f,
+-0.016770782f, -0.020042717f, -0.023314439f, -0.026585912f, -0.029857099f,
+-0.033127967f, -0.036398482f, -0.039668605f, -0.042938303f, -0.046207540f,
+-0.049476285f, -0.052744497f, -0.056012146f, -0.059279196f, -0.062545612f,
+-0.065811358f, -0.069076397f, -0.072340697f, -0.075604223f, -0.078866936f,
+-0.082128808f, -0.085389800f, -0.088649876f, -0.091909006f, -0.095167145f,
+-0.098424271f, -0.10168034f, -0.10493532f, -0.10818918f, -0.11144188f,
+-0.11469338f, -0.11794366f, -0.12119267f, -0.12444039f, -0.12768677f,
+-0.13093179f, -0.13417540f, -0.13741758f, -0.14065829f, -0.14389749f,
+-0.14713514f, -0.15037122f, -0.15360570f, -0.15683852f, -0.16006967f,
+-0.16329910f, -0.16652679f, -0.16975269f, -0.17297678f, -0.17619900f,
+-0.17941935f, -0.18263777f, -0.18585424f, -0.18906870f, -0.19228116f,
+-0.19549155f, -0.19869985f, -0.20190603f, -0.20511003f, -0.20831184f,
+-0.21151142f, -0.21470875f, -0.21790376f, -0.22109644f, -0.22428675f,
+-0.22747467f, -0.23066014f, -0.23384315f, -0.23702365f, -0.24020162f,
+-0.24337701f, -0.24654980f, -0.24971995f, -0.25288740f, -0.25605217f,
+-0.25921419f, -0.26237345f, -0.26552987f, -0.26868346f, -0.27183419f,
+-0.27498198f, -0.27812684f, -0.28126872f, -0.28440759f, -0.28754342f,
+-0.29067615f, -0.29380578f, -0.29693225f, -0.30005556f, -0.30317566f,
+-0.30629250f, -0.30940607f, -0.31251630f, -0.31562322f, -0.31872672f,
+-0.32182685f, -0.32492352f, -0.32801670f, -0.33110636f, -0.33419248f,
+-0.33727503f, -0.34035397f, -0.34342924f, -0.34650084f, -0.34956875f,
+-0.35263291f, -0.35569328f, -0.35874987f, -0.36180258f, -0.36485144f,
+-0.36789638f, -0.37093741f, -0.37397444f, -0.37700745f, -0.38003644f,
+-0.38306138f, -0.38608220f, -0.38909888f, -0.39211139f, -0.39511973f,
+-0.39812380f, -0.40112361f, -0.40411916f, -0.40711036f, -0.41009718f,
+-0.41307965f, -0.41605768f, -0.41903123f, -0.42200032f, -0.42496487f,
+-0.42792490f, -0.43088034f, -0.43383113f, -0.43677729f, -0.43971881f,
+-0.44265559f, -0.44558764f, -0.44851488f, -0.45143735f, -0.45435500f,
+-0.45726776f, -0.46017563f, -0.46307856f, -0.46597654f, -0.46886954f,
+-0.47175750f, -0.47464043f, -0.47751826f, -0.48039100f, -0.48325855f,
+-0.48612097f, -0.48897815f, -0.49183011f, -0.49467680f, -0.49751821f,
+-0.50035429f, -0.50318497f, -0.50601029f, -0.50883019f, -0.51164466f,
+-0.51445359f, -0.51725709f, -0.52005500f, -0.52284735f, -0.52563411f,
+-0.52841520f, -0.53119069f, -0.53396046f, -0.53672451f, -0.53948283f,
+-0.54223537f, -0.54498214f, -0.54772300f, -0.55045801f, -0.55318713f,
+-0.55591035f, -0.55862761f, -0.56133890f, -0.56404412f, -0.56674337f,
+-0.56943649f, -0.57212353f, -0.57480448f, -0.57747924f, -0.58014780f,
+-0.58281022f, -0.58546633f, -0.58811617f, -0.59075975f, -0.59339696f,
+-0.59602785f, -0.59865236f, -0.60127044f, -0.60388207f, -0.60648727f,
+-0.60908598f, -0.61167812f, -0.61426371f, -0.61684275f, -0.61941516f,
+-0.62198097f, -0.62454009f, -0.62709254f, -0.62963831f, -0.63217729f,
+-0.63470948f, -0.63723493f, -0.63975352f, -0.64226526f, -0.64477009f,
+-0.64726806f, -0.64975911f, -0.65224314f, -0.65472025f, -0.65719032f,
+-0.65965337f, -0.66210932f, -0.66455823f, -0.66700000f, -0.66943461f,
+-0.67186207f, -0.67428231f, -0.67669535f, -0.67910111f, -0.68149966f,
+-0.68389088f, -0.68627477f, -0.68865126f, -0.69102043f, -0.69338220f,
+-0.69573659f, -0.69808346f, -0.70042288f, -0.70275480f, -0.70507920f,
+-0.70739603f, -0.70970529f, -0.71200693f, -0.71430099f, -0.71658736f,
+-0.71886611f, -0.72113711f, -0.72340041f, -0.72565591f, -0.72790372f,
+-0.73014367f, -0.73237586f, -0.73460019f, -0.73681659f, -0.73902518f,
+-0.74122584f, -0.74341851f, -0.74560326f, -0.74778003f, -0.74994880f,
+-0.75210953f, -0.75426215f, -0.75640678f, -0.75854325f, -0.76067162f,
+-0.76279181f, -0.76490390f, -0.76700771f, -0.76910341f, -0.77119076f,
+-0.77326995f, -0.77534080f, -0.77740335f, -0.77945763f, -0.78150350f,
+-0.78354102f, -0.78557014f, -0.78759086f, -0.78960317f, -0.79160696f,
+-0.79360235f, -0.79558921f, -0.79756755f, -0.79953730f, -0.80149853f,
+-0.80345118f, -0.80539525f, -0.80733067f, -0.80925739f, -0.81117553f,
+-0.81308490f, -0.81498563f, -0.81687760f, -0.81876087f, -0.82063532f,
+-0.82250100f, -0.82435787f, -0.82620591f, -0.82804507f, -0.82987541f,
+-0.83169687f, -0.83350939f, -0.83531296f, -0.83710766f, -0.83889335f,
+-0.84067005f, -0.84243774f, -0.84419644f, -0.84594607f, -0.84768665f,
+-0.84941816f, -0.85114056f, -0.85285389f, -0.85455805f, -0.85625303f,
+-0.85793889f, -0.85961550f, -0.86128294f, -0.86294121f, -0.86459017f,
+-0.86622989f, -0.86786032f, -0.86948150f, -0.87109333f, -0.87269586f,
+-0.87428904f, -0.87587279f, -0.87744725f, -0.87901229f, -0.88056785f,
+-0.88211405f, -0.88365078f, -0.88517809f, -0.88669586f, -0.88820416f,
+-0.88970292f, -0.89119220f, -0.89267188f, -0.89414203f, -0.89560264f,
+-0.89705360f, -0.89849502f, -0.89992678f, -0.90134889f, -0.90276134f,
+-0.90416414f, -0.90555727f, -0.90694070f, -0.90831441f, -0.90967834f,
+-0.91103262f, -0.91237706f, -0.91371179f, -0.91503674f, -0.91635185f,
+-0.91765714f, -0.91895264f, -0.92023826f, -0.92151409f, -0.92277998f,
+-0.92403603f, -0.92528218f, -0.92651838f, -0.92774469f, -0.92896110f,
+-0.93016750f, -0.93136400f, -0.93255049f, -0.93372697f, -0.93489349f,
+-0.93604994f, -0.93719643f, -0.93833286f, -0.93945926f, -0.94057560f,
+-0.94168180f, -0.94277799f, -0.94386405f, -0.94494003f, -0.94600588f,
+-0.94706154f, -0.94810712f, -0.94914252f, -0.95016778f, -0.95118284f,
+-0.95218778f, -0.95318246f, -0.95416695f, -0.95514119f, -0.95610523f,
+-0.95705903f, -0.95800257f, -0.95893586f, -0.95985889f, -0.96077162f,
+-0.96167403f, -0.96256620f, -0.96344805f, -0.96431959f, -0.96518075f,
+-0.96603161f, -0.96687216f, -0.96770233f, -0.96852213f, -0.96933156f,
+-0.97013056f, -0.97091925f, -0.97169751f, -0.97246534f, -0.97322279f,
+-0.97396982f, -0.97470641f, -0.97543252f, -0.97614825f, -0.97685349f,
+-0.97754824f, -0.97823256f, -0.97890645f, -0.97956979f, -0.98022264f,
+-0.98086500f, -0.98149687f, -0.98211825f, -0.98272908f, -0.98332942f,
+-0.98391914f, -0.98449844f, -0.98506713f, -0.98562527f, -0.98617285f,
+-0.98670989f, -0.98723638f, -0.98775226f, -0.98825759f, -0.98875231f,
+-0.98923647f, -0.98971003f, -0.99017298f, -0.99062532f, -0.99106705f,
+-0.99149817f, -0.99191868f, -0.99232858f, -0.99272782f, -0.99311644f,
+-0.99349445f, -0.99386179f, -0.99421853f, -0.99456459f, -0.99489999f,
+-0.99522477f, -0.99553883f, -0.99584228f, -0.99613506f, -0.99641716f,
+-0.99668860f, -0.99694937f, -0.99719942f, -0.99743885f, -0.99766755f,
+-0.99788558f, -0.99809295f, -0.99828959f, -0.99847561f, -0.99865085f,
+-0.99881548f, -0.99896932f, -0.99911255f, -0.99924499f, -0.99936682f,
+-0.99947786f, -0.99957830f, -0.99966794f, -0.99974692f, -0.99981517f,
+-0.99987274f, -0.99991959f, -0.99995571f, -0.99998116f, -0.99999589f,
+0.99999964f, 0.99997288f, 0.99990326f, 0.99979085f, 0.99963558f,
+0.99943751f, 0.99919659f, 0.99891287f, 0.99858636f, 0.99821711f,
+0.99780506f, 0.99735034f, 0.99685282f, 0.99631262f, 0.99572974f,
+0.99510419f, 0.99443603f, 0.99372530f, 0.99297196f, 0.99217612f,
+0.99133772f, 0.99045694f, 0.98953366f, 0.98856801f, 0.98756003f,
+0.98650974f, 0.98541719f, 0.98428243f, 0.98310548f, 0.98188645f,
+0.98062533f, 0.97932225f, 0.97797716f, 0.97659022f, 0.97516143f,
+0.97369087f, 0.97217858f, 0.97062469f, 0.96902919f, 0.96739221f,
+0.96571374f, 0.96399397f, 0.96223283f, 0.96043050f, 0.95858705f,
+0.95670253f, 0.95477700f, 0.95281059f, 0.95080340f, 0.94875544f,
+0.94666684f, 0.94453770f, 0.94236809f, 0.94015813f, 0.93790787f,
+0.93561745f, 0.93328691f, 0.93091643f, 0.92850608f, 0.92605597f,
+0.92356616f, 0.92103678f, 0.91846794f, 0.91585976f, 0.91321236f,
+0.91052586f, 0.90780038f, 0.90503591f, 0.90223277f, 0.89939094f,
+0.89651060f, 0.89359182f, 0.89063478f, 0.88763964f, 0.88460642f,
+0.88153529f, 0.87842643f, 0.87527996f, 0.87209594f, 0.86887461f,
+0.86561602f, 0.86232042f, 0.85898781f, 0.85561842f, 0.85221243f,
+0.84876984f, 0.84529096f, 0.84177583f, 0.83822471f, 0.83463764f,
+0.83101481f, 0.82735640f, 0.82366252f, 0.81993335f, 0.81616908f,
+0.81236988f, 0.80853581f, 0.80466717f, 0.80076402f, 0.79682660f,
+0.79285502f, 0.78884947f, 0.78481019f, 0.78073722f, 0.77663082f,
+0.77249116f, 0.76831841f, 0.76411277f, 0.75987434f, 0.75560343f,
+0.75130010f, 0.74696463f, 0.74259710f, 0.73819780f, 0.73376691f,
+0.72930455f, 0.72481096f, 0.72028631f, 0.71573079f, 0.71114463f,
+0.70652801f, 0.70188117f, 0.69720417f, 0.69249737f, 0.68776089f,
+0.68299496f, 0.67819971f, 0.67337549f, 0.66852236f, 0.66364062f,
+0.65873051f, 0.65379208f, 0.64882571f, 0.64383155f, 0.63880974f,
+0.63376063f, 0.62868434f, 0.62358117f, 0.61845124f, 0.61329484f,
+0.60811216f, 0.60290343f, 0.59766883f, 0.59240872f, 0.58712316f,
+0.58181250f, 0.57647687f, 0.57111657f, 0.56573176f, 0.56032276f,
+0.55488980f, 0.54943299f, 0.54395270f, 0.53844911f, 0.53292239f,
+0.52737290f, 0.52180082f, 0.51620632f, 0.51058978f, 0.50495136f,
+0.49929130f, 0.49360985f, 0.48790723f, 0.48218375f, 0.47643960f,
+0.47067502f, 0.46489030f, 0.45908567f, 0.45326138f, 0.44741765f,
+0.44155475f, 0.43567297f, 0.42977250f, 0.42385364f, 0.41791660f,
+0.41196167f, 0.40598908f, 0.39999911f, 0.39399201f, 0.38796803f,
+0.38192743f, 0.37587047f, 0.36979741f, 0.36370850f, 0.35760403f,
+0.35148421f, 0.34534934f, 0.33919969f, 0.33303553f, 0.32685706f,
+0.32066461f, 0.31445843f, 0.30823877f, 0.30200592f, 0.29576012f,
+0.28950164f, 0.28323078f, 0.27694780f, 0.27065292f, 0.26434645f,
+0.25802869f, 0.25169984f, 0.24536023f, 0.23901010f, 0.23264973f,
+0.22627939f, 0.21989937f, 0.21350993f, 0.20711134f, 0.20070387f,
+0.19428782f, 0.18786344f, 0.18143101f, 0.17499080f, 0.16854310f,
+0.16208819f, 0.15562633f, 0.14915779f, 0.14268288f, 0.13620184f,
+0.12971498f, 0.12322257f, 0.11672486f, 0.11022217f, 0.10371475f,
+0.097202882f, 0.090686858f, 0.084166944f, 0.077643424f, 0.071116582f,
+0.064586692f, 0.058054037f, 0.051518895f, 0.044981543f, 0.038442269f,
+0.031901345f, 0.025359053f, 0.018815678f, 0.012271495f, 0.0057267868f,
+-0.00081816671f, -0.0073630852f, -0.013907688f, -0.020451695f, -0.026994826f,
+-0.033536803f, -0.040077340f, -0.046616159f, -0.053152986f, -0.059687532f,
+-0.066219524f, -0.072748676f, -0.079274714f, -0.085797355f, -0.092316322f,
+-0.098831341f, -0.10534211f, -0.11184838f, -0.11834986f, -0.12484626f,
+-0.13133731f, -0.13782275f, -0.14430228f, -0.15077563f, -0.15724251f,
+-0.16370267f, -0.17015581f, -0.17660165f, -0.18303993f, -0.18947038f,
+-0.19589271f, -0.20230664f, -0.20871192f, -0.21510825f, -0.22149536f,
+-0.22787298f, -0.23424086f, -0.24059868f, -0.24694622f, -0.25328314f,
+-0.25960925f, -0.26592422f, -0.27222782f, -0.27851975f, -0.28479972f,
+-0.29106751f, -0.29732284f, -0.30356544f, -0.30979502f, -0.31601134f,
+-0.32221413f, -0.32840309f, -0.33457801f, -0.34073856f, -0.34688455f,
+-0.35301566f, -0.35913166f, -0.36523229f, -0.37131724f, -0.37738630f,
+-0.38343921f, -0.38947567f, -0.39549544f, -0.40149832f, -0.40748394f,
+-0.41345215f, -0.41940263f, -0.42533514f, -0.43124944f, -0.43714526f,
+-0.44302234f, -0.44888046f, -0.45471936f, -0.46053877f, -0.46633846f,
+-0.47211814f, -0.47787762f, -0.48361665f, -0.48933494f, -0.49503228f,
+-0.50070840f, -0.50636309f, -0.51199609f, -0.51760709f, -0.52319598f,
+-0.52876246f, -0.53430629f, -0.53982723f, -0.54532504f, -0.55079949f,
+-0.55625033f, -0.56167740f, -0.56708032f, -0.57245898f, -0.57781315f,
+-0.58314258f, -0.58844697f, -0.59372622f, -0.59897995f, -0.60420811f,
+-0.60941035f, -0.61458647f, -0.61973625f, -0.62485951f, -0.62995601f,
+-0.63502556f, -0.64006782f, -0.64508271f, -0.65007001f, -0.65502942f,
+-0.65996075f, -0.66486382f, -0.66973841f, -0.67458433f, -0.67940134f,
+-0.68418926f, -0.68894786f, -0.69367695f, -0.69837630f, -0.70304573f,
+-0.70768511f, -0.71229410f, -0.71687263f, -0.72142041f, -0.72593731f,
+-0.73042315f, -0.73487765f, -0.73930067f, -0.74369204f, -0.74805158f,
+-0.75237900f, -0.75667429f, -0.76093709f, -0.76516730f, -0.76936477f,
+-0.77352923f, -0.77766061f, -0.78175867f, -0.78582323f, -0.78985411f,
+-0.79385114f, -0.79781419f, -0.80174309f, -0.80563760f, -0.80949765f,
+-0.81332302f, -0.81711352f, -0.82086903f, -0.82458937f, -0.82827437f,
+-0.83192390f, -0.83553779f, -0.83911592f, -0.84265804f, -0.84616417f,
+-0.84963393f, -0.85306740f, -0.85646427f, -0.85982448f, -0.86314780f,
+-0.86643422f, -0.86968350f, -0.87289548f, -0.87607014f, -0.87920725f,
+-0.88230664f, -0.88536829f, -0.88839203f, -0.89137769f, -0.89432514f,
+-0.89723432f, -0.90010506f, -0.90293723f, -0.90573072f, -0.90848541f,
+-0.91120118f, -0.91387796f, -0.91651553f, -0.91911387f, -0.92167282f,
+-0.92419231f, -0.92667222f, -0.92911243f, -0.93151283f, -0.93387336f,
+-0.93619382f, -0.93847424f, -0.94071442f, -0.94291431f, -0.94507378f,
+-0.94719279f, -0.94927126f, -0.95130903f, -0.95330608f, -0.95526224f,
+-0.95717752f, -0.95905179f, -0.96088499f, -0.96267700f, -0.96442777f,
+-0.96613729f, -0.96780539f, -0.96943200f, -0.97101706f, -0.97256058f,
+-0.97406244f, -0.97552258f, -0.97694093f, -0.97831738f, -0.97965199f,
+-0.98094457f, -0.98219514f, -0.98340368f, -0.98457009f, -0.98569429f,
+-0.98677629f, -0.98781598f, -0.98881340f, -0.98976845f, -0.99068111f,
+-0.99155134f, -0.99237907f, -0.99316430f, -0.99390697f, -0.99460709f,
+-0.99526459f, -0.99587947f, -0.99645168f, -0.99698120f, -0.99746799f,
+-0.99791211f, -0.99831343f, -0.99867201f, -0.99898779f, -0.99926084f,
+-0.99949104f, -0.99967843f, -0.99982297f, -0.99992472f, -0.99998361f,
+0.99999869f, 0.99989158f, 0.99961317f, 0.99916345f, 0.99854255f,
+0.99775058f, 0.99678761f, 0.99565387f, 0.99434954f, 0.99287480f,
+0.99122995f, 0.98941529f, 0.98743105f, 0.98527765f, 0.98295540f,
+0.98046476f, 0.97780609f, 0.97497988f, 0.97198665f, 0.96882683f,
+0.96550101f, 0.96200979f, 0.95835376f, 0.95453346f, 0.95054960f,
+0.94640291f, 0.94209403f, 0.93762374f, 0.93299282f, 0.92820197f,
+0.92325211f, 0.91814411f, 0.91287869f, 0.90745693f, 0.90187967f,
+0.89614785f, 0.89026248f, 0.88422459f, 0.87803519f, 0.87169534f,
+0.86520612f, 0.85856867f, 0.85178405f, 0.84485358f, 0.83777827f,
+0.83055943f, 0.82319832f, 0.81569612f, 0.80805415f, 0.80027372f,
+0.79235619f, 0.78430289f, 0.77611518f, 0.76779449f, 0.75934225f,
+0.75075996f, 0.74204898f, 0.73321080f, 0.72424710f, 0.71515924f,
+0.70594883f, 0.69661748f, 0.68716675f, 0.67759830f, 0.66791373f,
+0.65811473f, 0.64820296f, 0.63818014f, 0.62804794f, 0.61780810f,
+0.60746247f, 0.59701276f, 0.58646071f, 0.57580817f, 0.56505698f,
+0.55420899f, 0.54326600f, 0.53222996f, 0.52110273f, 0.50988621f,
+0.49858227f, 0.48719296f, 0.47572014f, 0.46416581f, 0.45253196f,
+0.44082057f, 0.42903364f, 0.41717321f, 0.40524128f, 0.39323992f,
+0.38117120f, 0.36903715f, 0.35683987f, 0.34458145f, 0.33226398f,
+0.31988961f, 0.30746040f, 0.29497850f, 0.28244606f, 0.26986524f,
+0.25723818f, 0.24456702f, 0.23185398f, 0.21910121f, 0.20631088f,
+0.19348522f, 0.18062639f, 0.16773662f, 0.15481812f, 0.14187308f,
+0.12890373f, 0.11591230f, 0.10290100f, 0.089872077f, 0.076827750f,
+0.063770257f, 0.050701842f, 0.037624735f, 0.024541186f, 0.011453429f,
+-0.0016362892f, -0.014725727f, -0.027812643f, -0.040894791f, -0.053969935f,
+-0.067035832f, -0.080090240f, -0.093130924f, -0.10615565f, -0.11916219f,
+-0.13214831f, -0.14511178f, -0.15805040f, -0.17096193f, -0.18384418f,
+-0.19669491f, -0.20951195f, -0.22229309f, -0.23503613f, -0.24773891f,
+-0.26039925f, -0.27301496f, -0.28558388f, -0.29810387f, -0.31057280f,
+-0.32298848f, -0.33534884f, -0.34765175f, -0.35989508f, -0.37207675f,
+-0.38419467f, -0.39624676f, -0.40823093f, -0.42014518f, -0.43198743f,
+-0.44375566f, -0.45544785f, -0.46706200f, -0.47859612f, -0.49004826f,
+-0.50141639f, -0.51269865f, -0.52389306f, -0.53499764f, -0.54601061f,
+-0.55693001f, -0.56775403f, -0.57848072f, -0.58910829f, -0.59963489f,
+-0.61005878f, -0.62037814f, -0.63059121f, -0.64069623f, -0.65069145f,
+-0.66057515f, -0.67034572f, -0.68000144f, -0.68954057f, -0.69896162f,
+-0.70826286f, -0.71744281f, -0.72649974f, -0.73543227f, -0.74423873f,
+-0.75291771f, -0.76146764f, -0.76988715f, -0.77817470f, -0.78632891f,
+-0.79434842f, -0.80223179f, -0.80997771f, -0.81758487f, -0.82505190f,
+-0.83237761f, -0.83956063f, -0.84659988f, -0.85349399f, -0.86024189f,
+-0.86684239f, -0.87329435f, -0.87959671f, -0.88574833f, -0.89174819f,
+-0.89759529f, -0.90328854f, -0.90882701f, -0.91420978f, -0.91943592f,
+-0.92450452f, -0.92941469f, -0.93416560f, -0.93875647f, -0.94318646f,
+-0.94745487f, -0.95156091f, -0.95550388f, -0.95928317f, -0.96289814f,
+-0.96634805f, -0.96963239f, -0.97275060f, -0.97570217f, -0.97848648f,
+-0.98110318f, -0.98355180f, -0.98583186f, -0.98794299f, -0.98988485f,
+-0.99165714f, -0.99325943f, -0.99469161f, -0.99595332f, -0.99704438f,
+-0.99796462f, -0.99871385f, -0.99929196f, -0.99969882f, -0.99993443f,
+0.99999464f, 0.99956632f, 0.99845290f, 0.99665523f, 0.99417448f,
+0.99101239f, 0.98717111f, 0.98265326f, 0.97746199f, 0.97160077f,
+0.96507365f, 0.95788515f, 0.95004016f, 0.94154406f, 0.93240267f,
+0.92262226f, 0.91220951f, 0.90117162f, 0.88951606f, 0.87725091f,
+0.86438453f, 0.85092574f, 0.83688372f, 0.82226819f, 0.80708915f,
+0.79135692f, 0.77508235f, 0.75827658f, 0.74095112f, 0.72311783f,
+0.70478898f, 0.68597710f, 0.66669506f, 0.64695615f, 0.62677377f,
+0.60616189f, 0.58513457f, 0.56370622f, 0.54189157f, 0.51970547f,
+0.49716324f, 0.47428027f, 0.45107225f, 0.42755505f, 0.40374488f,
+0.37965798f, 0.35531086f, 0.33072025f, 0.30590299f, 0.28087607f,
+0.25565663f, 0.23026201f, 0.20470956f, 0.17901683f, 0.15320139f,
+0.12728097f, 0.10127331f, 0.075196236f, 0.049067631f, 0.022905400f,
+-0.0032725304f, -0.029448219f, -0.055603724f, -0.081721120f, -0.10778251f,
+-0.13377003f, -0.15966587f, -0.18545228f, -0.21111161f, -0.23662624f,
+-0.26197869f, -0.28715160f, -0.31212771f, -0.33688989f, -0.36142120f,
+-0.38570482f, -0.40972409f, -0.43346253f, -0.45690393f, -0.48003218f,
+-0.50283146f, -0.52528608f, -0.54738069f, -0.56910020f, -0.59042966f,
+-0.61135447f, -0.63186026f, -0.65193301f, -0.67155898f, -0.69072473f,
+-0.70941705f, -0.72762316f, -0.74533063f, -0.76252723f, -0.77920127f,
+-0.79534131f, -0.81093621f, -0.82597536f, -0.84044844f, -0.85434550f,
+-0.86765707f, -0.88037395f, -0.89248747f, -0.90398932f, -0.91487163f,
+-0.92512697f, -0.93474823f, -0.94372886f, -0.95206273f, -0.95974404f,
+-0.96676767f, -0.97312868f, -0.97882277f, -0.98384601f, -0.98819500f,
+-0.99186671f, -0.99485862f, -0.99716878f, -0.99879545f, -0.99973762f,
+};
+#endif
+
+static const CELTMode mode48000_960_120 = {
+48000, /* Fs */
+120, /* overlap */
+21, /* nbEBands */
+21, /* effEBands */
+{0.85000610f, 0.0000000f, 1.0000000f, 1.0000000f, }, /* preemph */
+eband5ms, /* eBands */
+3, /* maxLM */
+8, /* nbShortMdcts */
+120, /* shortMdctSize */
+11, /* nbAllocVectors */
+band_allocation, /* allocVectors */
+logN400, /* logN */
+window120, /* window */
+{1920, 3, {&fft_state48000_960_0, &fft_state48000_960_1, &fft_state48000_960_2, &fft_state48000_960_3, }, mdct_twiddles960}, /* mdct */
+{392, cache_index50, cache_bits50, cache_caps50}, /* cache */
+};
+
+/* List of all the available modes */
+#define TOTAL_MODES 1
+static const CELTMode * const static_mode_list[TOTAL_MODES] = {
+&mode48000_960_120,
+};
diff --git a/lib/rbcodec/codecs/libopus/celt/static_modes_float_arm_ne10.h b/lib/rbcodec/codecs/libopus/celt/static_modes_float_arm_ne10.h
new file mode 100644
index 0000000000..66e1abb101
--- /dev/null
+++ b/lib/rbcodec/codecs/libopus/celt/static_modes_float_arm_ne10.h
@@ -0,0 +1,404 @@
+/* The contents of this file was automatically generated by
+ * dump_mode_arm_ne10.c with arguments: 48000 960
+ * It contains static definitions for some pre-defined modes. */
+#include <NE10_types.h>
+
+#ifndef NE10_FFT_PARAMS48000_960
+#define NE10_FFT_PARAMS48000_960
+static const ne10_int32_t ne10_factors_480[64] = {
+4, 40, 4, 30, 2, 15, 5, 3, 3, 1, 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, };
+static const ne10_int32_t ne10_factors_240[64] = {
+3, 20, 4, 15, 5, 3, 3, 1, 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, };
+static const ne10_int32_t ne10_factors_120[64] = {
+3, 10, 2, 15, 5, 3, 3, 1, 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, };
+static const ne10_int32_t ne10_factors_60[64] = {
+2, 5, 5, 3, 3, 1, 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, };
+static const ne10_fft_cpx_float32_t ne10_twiddles_480[480] = {
+{1.0000000f,0.0000000f}, {1.0000000f,-0.0000000f}, {1.0000000f,-0.0000000f},
+{1.0000000f,-0.0000000f}, {0.91354543f,-0.40673664f}, {0.66913056f,-0.74314487f},
+{1.0000000f,-0.0000000f}, {0.66913056f,-0.74314487f}, {-0.10452851f,-0.99452192f},
+{1.0000000f,-0.0000000f}, {0.30901697f,-0.95105654f}, {-0.80901700f,-0.58778518f},
+{1.0000000f,-0.0000000f}, {-0.10452851f,-0.99452192f}, {-0.97814757f,0.20791179f},
+{1.0000000f,-0.0000000f}, {0.97814763f,-0.20791170f}, {0.91354543f,-0.40673664f},
+{0.80901700f,-0.58778524f}, {0.66913056f,-0.74314487f}, {0.49999997f,-0.86602545f},
+{0.30901697f,-0.95105654f}, {0.10452842f,-0.99452192f}, {-0.10452851f,-0.99452192f},
+{-0.30901703f,-0.95105648f}, {-0.50000006f,-0.86602533f}, {-0.66913068f,-0.74314475f},
+{-0.80901700f,-0.58778518f}, {-0.91354549f,-0.40673658f}, {-0.97814763f,-0.20791161f},
+{1.0000000f,-0.0000000f}, {0.99862951f,-0.052335959f}, {0.99452192f,-0.10452846f},
+{0.98768836f,-0.15643448f}, {0.97814763f,-0.20791170f}, {0.96592581f,-0.25881904f},
+{0.95105648f,-0.30901700f}, {0.93358040f,-0.35836795f}, {0.91354543f,-0.40673664f},
+{0.89100653f,-0.45399052f}, {0.86602545f,-0.50000000f}, {0.83867055f,-0.54463905f},
+{0.80901700f,-0.58778524f}, {0.77714598f,-0.62932038f}, {0.74314475f,-0.66913062f},
+{0.70710677f,-0.70710683f}, {0.66913056f,-0.74314487f}, {0.62932038f,-0.77714598f},
+{0.58778524f,-0.80901700f}, {0.54463899f,-0.83867055f}, {0.49999997f,-0.86602545f},
+{0.45399052f,-0.89100653f}, {0.40673661f,-0.91354549f}, {0.35836786f,-0.93358046f},
+{0.30901697f,-0.95105654f}, {0.25881907f,-0.96592581f}, {0.20791166f,-0.97814763f},
+{0.15643437f,-0.98768836f}, {0.10452842f,-0.99452192f}, {0.052335974f,-0.99862951f},
+{1.0000000f,-0.0000000f}, {0.99452192f,-0.10452846f}, {0.97814763f,-0.20791170f},
+{0.95105648f,-0.30901700f}, {0.91354543f,-0.40673664f}, {0.86602545f,-0.50000000f},
+{0.80901700f,-0.58778524f}, {0.74314475f,-0.66913062f}, {0.66913056f,-0.74314487f},
+{0.58778524f,-0.80901700f}, {0.49999997f,-0.86602545f}, {0.40673661f,-0.91354549f},
+{0.30901697f,-0.95105654f}, {0.20791166f,-0.97814763f}, {0.10452842f,-0.99452192f},
+{-4.3711388e-08f,-1.0000000f}, {-0.10452851f,-0.99452192f}, {-0.20791174f,-0.97814757f},
+{-0.30901703f,-0.95105648f}, {-0.40673670f,-0.91354543f}, {-0.50000006f,-0.86602533f},
+{-0.58778518f,-0.80901700f}, {-0.66913068f,-0.74314475f}, {-0.74314493f,-0.66913044f},
+{-0.80901700f,-0.58778518f}, {-0.86602539f,-0.50000006f}, {-0.91354549f,-0.40673658f},
+{-0.95105654f,-0.30901679f}, {-0.97814763f,-0.20791161f}, {-0.99452192f,-0.10452849f},
+{1.0000000f,-0.0000000f}, {0.98768836f,-0.15643448f}, {0.95105648f,-0.30901700f},
+{0.89100653f,-0.45399052f}, {0.80901700f,-0.58778524f}, {0.70710677f,-0.70710683f},
+{0.58778524f,-0.80901700f}, {0.45399052f,-0.89100653f}, {0.30901697f,-0.95105654f},
+{0.15643437f,-0.98768836f}, {-4.3711388e-08f,-1.0000000f}, {-0.15643445f,-0.98768836f},
+{-0.30901703f,-0.95105648f}, {-0.45399061f,-0.89100647f}, {-0.58778518f,-0.80901700f},
+{-0.70710677f,-0.70710677f}, {-0.80901700f,-0.58778518f}, {-0.89100659f,-0.45399037f},
+{-0.95105654f,-0.30901679f}, {-0.98768836f,-0.15643445f}, {-1.0000000f,8.7422777e-08f},
+{-0.98768830f,0.15643461f}, {-0.95105654f,0.30901697f}, {-0.89100653f,0.45399055f},
+{-0.80901694f,0.58778536f}, {-0.70710665f,0.70710689f}, {-0.58778507f,0.80901712f},
+{-0.45399022f,0.89100665f}, {-0.30901709f,0.95105648f}, {-0.15643452f,0.98768830f},
+{1.0000000f,-0.0000000f}, {0.99991435f,-0.013089596f}, {0.99965733f,-0.026176950f},
+{0.99922901f,-0.039259817f}, {0.99862951f,-0.052335959f}, {0.99785894f,-0.065403134f},
+{0.99691731f,-0.078459099f}, {0.99580491f,-0.091501623f}, {0.99452192f,-0.10452846f},
+{0.99306846f,-0.11753740f}, {0.99144489f,-0.13052620f}, {0.98965138f,-0.14349262f},
+{0.98768836f,-0.15643448f}, {0.98555607f,-0.16934951f}, {0.98325491f,-0.18223552f},
+{0.98078525f,-0.19509032f}, {0.97814763f,-0.20791170f}, {0.97534233f,-0.22069745f},
+{0.97236991f,-0.23344538f}, {0.96923089f,-0.24615330f}, {0.96592581f,-0.25881904f},
+{0.96245521f,-0.27144045f}, {0.95881975f,-0.28401536f}, {0.95501995f,-0.29654160f},
+{0.95105648f,-0.30901700f}, {0.94693011f,-0.32143945f}, {0.94264150f,-0.33380687f},
+{0.93819129f,-0.34611708f}, {0.93358040f,-0.35836795f}, {0.92880952f,-0.37055743f},
+{0.92387956f,-0.38268346f}, {0.91879117f,-0.39474389f}, {0.91354543f,-0.40673664f},
+{0.90814316f,-0.41865975f}, {0.90258527f,-0.43051112f}, {0.89687270f,-0.44228873f},
+{0.89100653f,-0.45399052f}, {0.88498765f,-0.46561453f}, {0.87881708f,-0.47715878f},
+{0.87249601f,-0.48862126f}, {0.86602545f,-0.50000000f}, {0.85940641f,-0.51129311f},
+{0.85264015f,-0.52249855f}, {0.84572786f,-0.53361452f}, {0.83867055f,-0.54463905f},
+{0.83146960f,-0.55557024f}, {0.82412618f,-0.56640625f}, {0.81664151f,-0.57714522f},
+{0.80901700f,-0.58778524f}, {0.80125380f,-0.59832460f}, {0.79335332f,-0.60876143f},
+{0.78531694f,-0.61909395f}, {0.77714598f,-0.62932038f}, {0.76884180f,-0.63943899f},
+{0.76040596f,-0.64944810f}, {0.75183982f,-0.65934587f}, {0.74314475f,-0.66913062f},
+{0.73432249f,-0.67880076f}, {0.72537434f,-0.68835455f}, {0.71630192f,-0.69779050f},
+{0.70710677f,-0.70710683f}, {0.69779044f,-0.71630198f}, {0.68835455f,-0.72537440f},
+{0.67880070f,-0.73432255f}, {0.66913056f,-0.74314487f}, {0.65934581f,-0.75183982f},
+{0.64944804f,-0.76040596f}, {0.63943899f,-0.76884186f}, {0.62932038f,-0.77714598f},
+{0.61909395f,-0.78531694f}, {0.60876137f,-0.79335338f}, {0.59832460f,-0.80125386f},
+{0.58778524f,-0.80901700f}, {0.57714516f,-0.81664151f}, {0.56640625f,-0.82412618f},
+{0.55557019f,-0.83146960f}, {0.54463899f,-0.83867055f}, {0.53361452f,-0.84572786f},
+{0.52249849f,-0.85264015f}, {0.51129311f,-0.85940641f}, {0.49999997f,-0.86602545f},
+{0.48862118f,-0.87249601f}, {0.47715876f,-0.87881708f}, {0.46561447f,-0.88498765f},
+{0.45399052f,-0.89100653f}, {0.44228867f,-0.89687276f}, {0.43051103f,-0.90258533f},
+{0.41865975f,-0.90814316f}, {0.40673661f,-0.91354549f}, {0.39474380f,-0.91879129f},
+{0.38268343f,-0.92387956f}, {0.37055740f,-0.92880958f}, {0.35836786f,-0.93358046f},
+{0.34611705f,-0.93819135f}, {0.33380681f,-0.94264150f}, {0.32143947f,-0.94693011f},
+{0.30901697f,-0.95105654f}, {0.29654151f,-0.95501995f}, {0.28401533f,-0.95881975f},
+{0.27144039f,-0.96245527f}, {0.25881907f,-0.96592581f}, {0.24615327f,-0.96923089f},
+{0.23344530f,-0.97236991f}, {0.22069745f,-0.97534233f}, {0.20791166f,-0.97814763f},
+{0.19509023f,-0.98078531f}, {0.18223552f,-0.98325491f}, {0.16934945f,-0.98555607f},
+{0.15643437f,-0.98768836f}, {0.14349259f,-0.98965138f}, {0.13052613f,-0.99144489f},
+{0.11753740f,-0.99306846f}, {0.10452842f,-0.99452192f}, {0.091501534f,-0.99580491f},
+{0.078459084f,-0.99691731f}, {0.065403074f,-0.99785894f}, {0.052335974f,-0.99862951f},
+{0.039259788f,-0.99922901f}, {0.026176875f,-0.99965733f}, {0.013089597f,-0.99991435f},
+{1.0000000f,-0.0000000f}, {0.99965733f,-0.026176950f}, {0.99862951f,-0.052335959f},
+{0.99691731f,-0.078459099f}, {0.99452192f,-0.10452846f}, {0.99144489f,-0.13052620f},
+{0.98768836f,-0.15643448f}, {0.98325491f,-0.18223552f}, {0.97814763f,-0.20791170f},
+{0.97236991f,-0.23344538f}, {0.96592581f,-0.25881904f}, {0.95881975f,-0.28401536f},
+{0.95105648f,-0.30901700f}, {0.94264150f,-0.33380687f}, {0.93358040f,-0.35836795f},
+{0.92387956f,-0.38268346f}, {0.91354543f,-0.40673664f}, {0.90258527f,-0.43051112f},
+{0.89100653f,-0.45399052f}, {0.87881708f,-0.47715878f}, {0.86602545f,-0.50000000f},
+{0.85264015f,-0.52249855f}, {0.83867055f,-0.54463905f}, {0.82412618f,-0.56640625f},
+{0.80901700f,-0.58778524f}, {0.79335332f,-0.60876143f}, {0.77714598f,-0.62932038f},
+{0.76040596f,-0.64944810f}, {0.74314475f,-0.66913062f}, {0.72537434f,-0.68835455f},
+{0.70710677f,-0.70710683f}, {0.68835455f,-0.72537440f}, {0.66913056f,-0.74314487f},
+{0.64944804f,-0.76040596f}, {0.62932038f,-0.77714598f}, {0.60876137f,-0.79335338f},
+{0.58778524f,-0.80901700f}, {0.56640625f,-0.82412618f}, {0.54463899f,-0.83867055f},
+{0.52249849f,-0.85264015f}, {0.49999997f,-0.86602545f}, {0.47715876f,-0.87881708f},
+{0.45399052f,-0.89100653f}, {0.43051103f,-0.90258533f}, {0.40673661f,-0.91354549f},
+{0.38268343f,-0.92387956f}, {0.35836786f,-0.93358046f}, {0.33380681f,-0.94264150f},
+{0.30901697f,-0.95105654f}, {0.28401533f,-0.95881975f}, {0.25881907f,-0.96592581f},
+{0.23344530f,-0.97236991f}, {0.20791166f,-0.97814763f}, {0.18223552f,-0.98325491f},
+{0.15643437f,-0.98768836f}, {0.13052613f,-0.99144489f}, {0.10452842f,-0.99452192f},
+{0.078459084f,-0.99691731f}, {0.052335974f,-0.99862951f}, {0.026176875f,-0.99965733f},
+{-4.3711388e-08f,-1.0000000f}, {-0.026176963f,-0.99965733f}, {-0.052336060f,-0.99862951f},
+{-0.078459173f,-0.99691731f}, {-0.10452851f,-0.99452192f}, {-0.13052621f,-0.99144489f},
+{-0.15643445f,-0.98768836f}, {-0.18223560f,-0.98325491f}, {-0.20791174f,-0.97814757f},
+{-0.23344538f,-0.97236991f}, {-0.25881916f,-0.96592581f}, {-0.28401542f,-0.95881969f},
+{-0.30901703f,-0.95105648f}, {-0.33380687f,-0.94264150f}, {-0.35836795f,-0.93358040f},
+{-0.38268352f,-0.92387950f}, {-0.40673670f,-0.91354543f}, {-0.43051112f,-0.90258527f},
+{-0.45399061f,-0.89100647f}, {-0.47715873f,-0.87881708f}, {-0.50000006f,-0.86602533f},
+{-0.52249867f,-0.85264009f}, {-0.54463905f,-0.83867055f}, {-0.56640631f,-0.82412612f},
+{-0.58778518f,-0.80901700f}, {-0.60876143f,-0.79335332f}, {-0.62932050f,-0.77714586f},
+{-0.64944804f,-0.76040596f}, {-0.66913068f,-0.74314475f}, {-0.68835467f,-0.72537428f},
+{-0.70710677f,-0.70710677f}, {-0.72537446f,-0.68835449f}, {-0.74314493f,-0.66913044f},
+{-0.76040596f,-0.64944804f}, {-0.77714604f,-0.62932026f}, {-0.79335332f,-0.60876143f},
+{-0.80901700f,-0.58778518f}, {-0.82412624f,-0.56640613f}, {-0.83867055f,-0.54463899f},
+{-0.85264021f,-0.52249849f}, {-0.86602539f,-0.50000006f}, {-0.87881714f,-0.47715873f},
+{-0.89100659f,-0.45399037f}, {-0.90258527f,-0.43051112f}, {-0.91354549f,-0.40673658f},
+{-0.92387956f,-0.38268328f}, {-0.93358040f,-0.35836792f}, {-0.94264150f,-0.33380675f},
+{-0.95105654f,-0.30901679f}, {-0.95881975f,-0.28401530f}, {-0.96592587f,-0.25881892f},
+{-0.97236991f,-0.23344538f}, {-0.97814763f,-0.20791161f}, {-0.98325491f,-0.18223536f},
+{-0.98768836f,-0.15643445f}, {-0.99144489f,-0.13052608f}, {-0.99452192f,-0.10452849f},
+{-0.99691737f,-0.078459039f}, {-0.99862957f,-0.052335810f}, {-0.99965733f,-0.026176952f},
+{1.0000000f,-0.0000000f}, {0.99922901f,-0.039259817f}, {0.99691731f,-0.078459099f},
+{0.99306846f,-0.11753740f}, {0.98768836f,-0.15643448f}, {0.98078525f,-0.19509032f},
+{0.97236991f,-0.23344538f}, {0.96245521f,-0.27144045f}, {0.95105648f,-0.30901700f},
+{0.93819129f,-0.34611708f}, {0.92387956f,-0.38268346f}, {0.90814316f,-0.41865975f},
+{0.89100653f,-0.45399052f}, {0.87249601f,-0.48862126f}, {0.85264015f,-0.52249855f},
+{0.83146960f,-0.55557024f}, {0.80901700f,-0.58778524f}, {0.78531694f,-0.61909395f},
+{0.76040596f,-0.64944810f}, {0.73432249f,-0.67880076f}, {0.70710677f,-0.70710683f},
+{0.67880070f,-0.73432255f}, {0.64944804f,-0.76040596f}, {0.61909395f,-0.78531694f},
+{0.58778524f,-0.80901700f}, {0.55557019f,-0.83146960f}, {0.52249849f,-0.85264015f},
+{0.48862118f,-0.87249601f}, {0.45399052f,-0.89100653f}, {0.41865975f,-0.90814316f},
+{0.38268343f,-0.92387956f}, {0.34611705f,-0.93819135f}, {0.30901697f,-0.95105654f},
+{0.27144039f,-0.96245527f}, {0.23344530f,-0.97236991f}, {0.19509023f,-0.98078531f},
+{0.15643437f,-0.98768836f}, {0.11753740f,-0.99306846f}, {0.078459084f,-0.99691731f},
+{0.039259788f,-0.99922901f}, {-4.3711388e-08f,-1.0000000f}, {-0.039259877f,-0.99922901f},
+{-0.078459173f,-0.99691731f}, {-0.11753749f,-0.99306846f}, {-0.15643445f,-0.98768836f},
+{-0.19509032f,-0.98078525f}, {-0.23344538f,-0.97236991f}, {-0.27144048f,-0.96245521f},
+{-0.30901703f,-0.95105648f}, {-0.34611711f,-0.93819129f}, {-0.38268352f,-0.92387950f},
+{-0.41865984f,-0.90814310f}, {-0.45399061f,-0.89100647f}, {-0.48862135f,-0.87249595f},
+{-0.52249867f,-0.85264009f}, {-0.55557036f,-0.83146954f}, {-0.58778518f,-0.80901700f},
+{-0.61909389f,-0.78531694f}, {-0.64944804f,-0.76040596f}, {-0.67880076f,-0.73432249f},
+{-0.70710677f,-0.70710677f}, {-0.73432249f,-0.67880070f}, {-0.76040596f,-0.64944804f},
+{-0.78531694f,-0.61909389f}, {-0.80901700f,-0.58778518f}, {-0.83146966f,-0.55557019f},
+{-0.85264021f,-0.52249849f}, {-0.87249607f,-0.48862115f}, {-0.89100659f,-0.45399037f},
+{-0.90814322f,-0.41865960f}, {-0.92387956f,-0.38268328f}, {-0.93819135f,-0.34611690f},
+{-0.95105654f,-0.30901679f}, {-0.96245521f,-0.27144048f}, {-0.97236991f,-0.23344538f},
+{-0.98078531f,-0.19509031f}, {-0.98768836f,-0.15643445f}, {-0.99306846f,-0.11753736f},
+{-0.99691737f,-0.078459039f}, {-0.99922901f,-0.039259743f}, {-1.0000000f,8.7422777e-08f},
+{-0.99922901f,0.039259918f}, {-0.99691731f,0.078459218f}, {-0.99306846f,0.11753753f},
+{-0.98768830f,0.15643461f}, {-0.98078525f,0.19509049f}, {-0.97236985f,0.23344554f},
+{-0.96245515f,0.27144065f}, {-0.95105654f,0.30901697f}, {-0.93819135f,0.34611705f},
+{-0.92387956f,0.38268346f}, {-0.90814316f,0.41865975f}, {-0.89100653f,0.45399055f},
+{-0.87249601f,0.48862129f}, {-0.85264015f,0.52249861f}, {-0.83146960f,0.55557030f},
+{-0.80901694f,0.58778536f}, {-0.78531688f,0.61909401f}, {-0.76040590f,0.64944816f},
+{-0.73432243f,0.67880082f}, {-0.70710665f,0.70710689f}, {-0.67880058f,0.73432261f},
+{-0.64944792f,0.76040608f}, {-0.61909378f,0.78531706f}, {-0.58778507f,0.80901712f},
+{-0.55557001f,0.83146977f}, {-0.52249837f,0.85264033f}, {-0.48862100f,0.87249613f},
+{-0.45399022f,0.89100665f}, {-0.41865945f,0.90814328f}, {-0.38268313f,0.92387968f},
+{-0.34611672f,0.93819147f}, {-0.30901709f,0.95105648f}, {-0.27144054f,0.96245521f},
+{-0.23344545f,0.97236991f}, {-0.19509038f,0.98078525f}, {-0.15643452f,0.98768830f},
+{-0.11753743f,0.99306846f}, {-0.078459114f,0.99691731f}, {-0.039259821f,0.99922901f},
+};
+static const ne10_fft_cpx_float32_t ne10_twiddles_240[240] = {
+{1.0000000f,0.0000000f}, {1.0000000f,-0.0000000f}, {1.0000000f,-0.0000000f},
+{1.0000000f,-0.0000000f}, {0.91354543f,-0.40673664f}, {0.66913056f,-0.74314487f},
+{1.0000000f,-0.0000000f}, {0.66913056f,-0.74314487f}, {-0.10452851f,-0.99452192f},
+{1.0000000f,-0.0000000f}, {0.30901697f,-0.95105654f}, {-0.80901700f,-0.58778518f},
+{1.0000000f,-0.0000000f}, {-0.10452851f,-0.99452192f}, {-0.97814757f,0.20791179f},
+{1.0000000f,-0.0000000f}, {0.99452192f,-0.10452846f}, {0.97814763f,-0.20791170f},
+{0.95105648f,-0.30901700f}, {0.91354543f,-0.40673664f}, {0.86602545f,-0.50000000f},
+{0.80901700f,-0.58778524f}, {0.74314475f,-0.66913062f}, {0.66913056f,-0.74314487f},
+{0.58778524f,-0.80901700f}, {0.49999997f,-0.86602545f}, {0.40673661f,-0.91354549f},
+{0.30901697f,-0.95105654f}, {0.20791166f,-0.97814763f}, {0.10452842f,-0.99452192f},
+{1.0000000f,-0.0000000f}, {0.97814763f,-0.20791170f}, {0.91354543f,-0.40673664f},
+{0.80901700f,-0.58778524f}, {0.66913056f,-0.74314487f}, {0.49999997f,-0.86602545f},
+{0.30901697f,-0.95105654f}, {0.10452842f,-0.99452192f}, {-0.10452851f,-0.99452192f},
+{-0.30901703f,-0.95105648f}, {-0.50000006f,-0.86602533f}, {-0.66913068f,-0.74314475f},
+{-0.80901700f,-0.58778518f}, {-0.91354549f,-0.40673658f}, {-0.97814763f,-0.20791161f},
+{1.0000000f,-0.0000000f}, {0.95105648f,-0.30901700f}, {0.80901700f,-0.58778524f},
+{0.58778524f,-0.80901700f}, {0.30901697f,-0.95105654f}, {-4.3711388e-08f,-1.0000000f},
+{-0.30901703f,-0.95105648f}, {-0.58778518f,-0.80901700f}, {-0.80901700f,-0.58778518f},
+{-0.95105654f,-0.30901679f}, {-1.0000000f,8.7422777e-08f}, {-0.95105654f,0.30901697f},
+{-0.80901694f,0.58778536f}, {-0.58778507f,0.80901712f}, {-0.30901709f,0.95105648f},
+{1.0000000f,-0.0000000f}, {0.99965733f,-0.026176950f}, {0.99862951f,-0.052335959f},
+{0.99691731f,-0.078459099f}, {0.99452192f,-0.10452846f}, {0.99144489f,-0.13052620f},
+{0.98768836f,-0.15643448f}, {0.98325491f,-0.18223552f}, {0.97814763f,-0.20791170f},
+{0.97236991f,-0.23344538f}, {0.96592581f,-0.25881904f}, {0.95881975f,-0.28401536f},
+{0.95105648f,-0.30901700f}, {0.94264150f,-0.33380687f}, {0.93358040f,-0.35836795f},
+{0.92387956f,-0.38268346f}, {0.91354543f,-0.40673664f}, {0.90258527f,-0.43051112f},
+{0.89100653f,-0.45399052f}, {0.87881708f,-0.47715878f}, {0.86602545f,-0.50000000f},
+{0.85264015f,-0.52249855f}, {0.83867055f,-0.54463905f}, {0.82412618f,-0.56640625f},
+{0.80901700f,-0.58778524f}, {0.79335332f,-0.60876143f}, {0.77714598f,-0.62932038f},
+{0.76040596f,-0.64944810f}, {0.74314475f,-0.66913062f}, {0.72537434f,-0.68835455f},
+{0.70710677f,-0.70710683f}, {0.68835455f,-0.72537440f}, {0.66913056f,-0.74314487f},
+{0.64944804f,-0.76040596f}, {0.62932038f,-0.77714598f}, {0.60876137f,-0.79335338f},
+{0.58778524f,-0.80901700f}, {0.56640625f,-0.82412618f}, {0.54463899f,-0.83867055f},
+{0.52249849f,-0.85264015f}, {0.49999997f,-0.86602545f}, {0.47715876f,-0.87881708f},
+{0.45399052f,-0.89100653f}, {0.43051103f,-0.90258533f}, {0.40673661f,-0.91354549f},
+{0.38268343f,-0.92387956f}, {0.35836786f,-0.93358046f}, {0.33380681f,-0.94264150f},
+{0.30901697f,-0.95105654f}, {0.28401533f,-0.95881975f}, {0.25881907f,-0.96592581f},
+{0.23344530f,-0.97236991f}, {0.20791166f,-0.97814763f}, {0.18223552f,-0.98325491f},
+{0.15643437f,-0.98768836f}, {0.13052613f,-0.99144489f}, {0.10452842f,-0.99452192f},
+{0.078459084f,-0.99691731f}, {0.052335974f,-0.99862951f}, {0.026176875f,-0.99965733f},
+{1.0000000f,-0.0000000f}, {0.99862951f,-0.052335959f}, {0.99452192f,-0.10452846f},
+{0.98768836f,-0.15643448f}, {0.97814763f,-0.20791170f}, {0.96592581f,-0.25881904f},
+{0.95105648f,-0.30901700f}, {0.93358040f,-0.35836795f}, {0.91354543f,-0.40673664f},
+{0.89100653f,-0.45399052f}, {0.86602545f,-0.50000000f}, {0.83867055f,-0.54463905f},
+{0.80901700f,-0.58778524f}, {0.77714598f,-0.62932038f}, {0.74314475f,-0.66913062f},
+{0.70710677f,-0.70710683f}, {0.66913056f,-0.74314487f}, {0.62932038f,-0.77714598f},
+{0.58778524f,-0.80901700f}, {0.54463899f,-0.83867055f}, {0.49999997f,-0.86602545f},
+{0.45399052f,-0.89100653f}, {0.40673661f,-0.91354549f}, {0.35836786f,-0.93358046f},
+{0.30901697f,-0.95105654f}, {0.25881907f,-0.96592581f}, {0.20791166f,-0.97814763f},
+{0.15643437f,-0.98768836f}, {0.10452842f,-0.99452192f}, {0.052335974f,-0.99862951f},
+{-4.3711388e-08f,-1.0000000f}, {-0.052336060f,-0.99862951f}, {-0.10452851f,-0.99452192f},
+{-0.15643445f,-0.98768836f}, {-0.20791174f,-0.97814757f}, {-0.25881916f,-0.96592581f},
+{-0.30901703f,-0.95105648f}, {-0.35836795f,-0.93358040f}, {-0.40673670f,-0.91354543f},
+{-0.45399061f,-0.89100647f}, {-0.50000006f,-0.86602533f}, {-0.54463905f,-0.83867055f},
+{-0.58778518f,-0.80901700f}, {-0.62932050f,-0.77714586f}, {-0.66913068f,-0.74314475f},
+{-0.70710677f,-0.70710677f}, {-0.74314493f,-0.66913044f}, {-0.77714604f,-0.62932026f},
+{-0.80901700f,-0.58778518f}, {-0.83867055f,-0.54463899f}, {-0.86602539f,-0.50000006f},
+{-0.89100659f,-0.45399037f}, {-0.91354549f,-0.40673658f}, {-0.93358040f,-0.35836792f},
+{-0.95105654f,-0.30901679f}, {-0.96592587f,-0.25881892f}, {-0.97814763f,-0.20791161f},
+{-0.98768836f,-0.15643445f}, {-0.99452192f,-0.10452849f}, {-0.99862957f,-0.052335810f},
+{1.0000000f,-0.0000000f}, {0.99691731f,-0.078459099f}, {0.98768836f,-0.15643448f},
+{0.97236991f,-0.23344538f}, {0.95105648f,-0.30901700f}, {0.92387956f,-0.38268346f},
+{0.89100653f,-0.45399052f}, {0.85264015f,-0.52249855f}, {0.80901700f,-0.58778524f},
+{0.76040596f,-0.64944810f}, {0.70710677f,-0.70710683f}, {0.64944804f,-0.76040596f},
+{0.58778524f,-0.80901700f}, {0.52249849f,-0.85264015f}, {0.45399052f,-0.89100653f},
+{0.38268343f,-0.92387956f}, {0.30901697f,-0.95105654f}, {0.23344530f,-0.97236991f},
+{0.15643437f,-0.98768836f}, {0.078459084f,-0.99691731f}, {-4.3711388e-08f,-1.0000000f},
+{-0.078459173f,-0.99691731f}, {-0.15643445f,-0.98768836f}, {-0.23344538f,-0.97236991f},
+{-0.30901703f,-0.95105648f}, {-0.38268352f,-0.92387950f}, {-0.45399061f,-0.89100647f},
+{-0.52249867f,-0.85264009f}, {-0.58778518f,-0.80901700f}, {-0.64944804f,-0.76040596f},
+{-0.70710677f,-0.70710677f}, {-0.76040596f,-0.64944804f}, {-0.80901700f,-0.58778518f},
+{-0.85264021f,-0.52249849f}, {-0.89100659f,-0.45399037f}, {-0.92387956f,-0.38268328f},
+{-0.95105654f,-0.30901679f}, {-0.97236991f,-0.23344538f}, {-0.98768836f,-0.15643445f},
+{-0.99691737f,-0.078459039f}, {-1.0000000f,8.7422777e-08f}, {-0.99691731f,0.078459218f},
+{-0.98768830f,0.15643461f}, {-0.97236985f,0.23344554f}, {-0.95105654f,0.30901697f},
+{-0.92387956f,0.38268346f}, {-0.89100653f,0.45399055f}, {-0.85264015f,0.52249861f},
+{-0.80901694f,0.58778536f}, {-0.76040590f,0.64944816f}, {-0.70710665f,0.70710689f},
+{-0.64944792f,0.76040608f}, {-0.58778507f,0.80901712f}, {-0.52249837f,0.85264033f},
+{-0.45399022f,0.89100665f}, {-0.38268313f,0.92387968f}, {-0.30901709f,0.95105648f},
+{-0.23344545f,0.97236991f}, {-0.15643452f,0.98768830f}, {-0.078459114f,0.99691731f},
+};
+static const ne10_fft_cpx_float32_t ne10_twiddles_120[120] = {
+{1.0000000f,0.0000000f}, {1.0000000f,-0.0000000f}, {1.0000000f,-0.0000000f},
+{1.0000000f,-0.0000000f}, {0.91354543f,-0.40673664f}, {0.66913056f,-0.74314487f},
+{1.0000000f,-0.0000000f}, {0.66913056f,-0.74314487f}, {-0.10452851f,-0.99452192f},
+{1.0000000f,-0.0000000f}, {0.30901697f,-0.95105654f}, {-0.80901700f,-0.58778518f},
+{1.0000000f,-0.0000000f}, {-0.10452851f,-0.99452192f}, {-0.97814757f,0.20791179f},
+{1.0000000f,-0.0000000f}, {0.97814763f,-0.20791170f}, {0.91354543f,-0.40673664f},
+{0.80901700f,-0.58778524f}, {0.66913056f,-0.74314487f}, {0.49999997f,-0.86602545f},
+{0.30901697f,-0.95105654f}, {0.10452842f,-0.99452192f}, {-0.10452851f,-0.99452192f},
+{-0.30901703f,-0.95105648f}, {-0.50000006f,-0.86602533f}, {-0.66913068f,-0.74314475f},
+{-0.80901700f,-0.58778518f}, {-0.91354549f,-0.40673658f}, {-0.97814763f,-0.20791161f},
+{1.0000000f,-0.0000000f}, {0.99862951f,-0.052335959f}, {0.99452192f,-0.10452846f},
+{0.98768836f,-0.15643448f}, {0.97814763f,-0.20791170f}, {0.96592581f,-0.25881904f},
+{0.95105648f,-0.30901700f}, {0.93358040f,-0.35836795f}, {0.91354543f,-0.40673664f},
+{0.89100653f,-0.45399052f}, {0.86602545f,-0.50000000f}, {0.83867055f,-0.54463905f},
+{0.80901700f,-0.58778524f}, {0.77714598f,-0.62932038f}, {0.74314475f,-0.66913062f},
+{0.70710677f,-0.70710683f}, {0.66913056f,-0.74314487f}, {0.62932038f,-0.77714598f},
+{0.58778524f,-0.80901700f}, {0.54463899f,-0.83867055f}, {0.49999997f,-0.86602545f},
+{0.45399052f,-0.89100653f}, {0.40673661f,-0.91354549f}, {0.35836786f,-0.93358046f},
+{0.30901697f,-0.95105654f}, {0.25881907f,-0.96592581f}, {0.20791166f,-0.97814763f},
+{0.15643437f,-0.98768836f}, {0.10452842f,-0.99452192f}, {0.052335974f,-0.99862951f},
+{1.0000000f,-0.0000000f}, {0.99452192f,-0.10452846f}, {0.97814763f,-0.20791170f},
+{0.95105648f,-0.30901700f}, {0.91354543f,-0.40673664f}, {0.86602545f,-0.50000000f},
+{0.80901700f,-0.58778524f}, {0.74314475f,-0.66913062f}, {0.66913056f,-0.74314487f},
+{0.58778524f,-0.80901700f}, {0.49999997f,-0.86602545f}, {0.40673661f,-0.91354549f},
+{0.30901697f,-0.95105654f}, {0.20791166f,-0.97814763f}, {0.10452842f,-0.99452192f},
+{-4.3711388e-08f,-1.0000000f}, {-0.10452851f,-0.99452192f}, {-0.20791174f,-0.97814757f},
+{-0.30901703f,-0.95105648f}, {-0.40673670f,-0.91354543f}, {-0.50000006f,-0.86602533f},
+{-0.58778518f,-0.80901700f}, {-0.66913068f,-0.74314475f}, {-0.74314493f,-0.66913044f},
+{-0.80901700f,-0.58778518f}, {-0.86602539f,-0.50