Chris@69: /* Copyright (c) 2014, Cisco Systems, INC Chris@69: Written by XiangMingZhu WeiZhou MinPeng YanWang Chris@69: Chris@69: Redistribution and use in source and binary forms, with or without Chris@69: modification, are permitted provided that the following conditions Chris@69: are met: Chris@69: Chris@69: - Redistributions of source code must retain the above copyright Chris@69: notice, this list of conditions and the following disclaimer. Chris@69: Chris@69: - Redistributions in binary form must reproduce the above copyright Chris@69: notice, this list of conditions and the following disclaimer in the Chris@69: documentation and/or other materials provided with the distribution. Chris@69: Chris@69: THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS Chris@69: ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT Chris@69: LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR Chris@69: A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER Chris@69: OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, Chris@69: EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, Chris@69: PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR Chris@69: PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF Chris@69: LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING Chris@69: NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS Chris@69: SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. Chris@69: */ Chris@69: Chris@69: #ifndef MAIN_SSE_H Chris@69: #define MAIN_SSE_H Chris@69: Chris@69: #ifdef HAVE_CONFIG_H Chris@69: #include "config.h" Chris@69: #endif Chris@69: Chris@69: # if defined(OPUS_X86_MAY_HAVE_SSE4_1) Chris@69: Chris@69: #if 0 /* FIXME: SSE disabled until silk_VQ_WMat_EC_sse4_1() gets updated. */ Chris@69: # define OVERRIDE_silk_VQ_WMat_EC Chris@69: Chris@69: void silk_VQ_WMat_EC_sse4_1( Chris@69: opus_int8 *ind, /* O index of best codebook vector */ Chris@69: opus_int32 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ Chris@69: opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ Chris@69: const opus_int16 *in_Q14, /* I input vector to be quantized */ Chris@69: const opus_int32 *W_Q18, /* I weighting matrix */ Chris@69: const opus_int8 *cb_Q7, /* I codebook */ Chris@69: const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ Chris@69: const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ Chris@69: const opus_int mu_Q9, /* I tradeoff betw. weighted error and rate */ Chris@69: const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ Chris@69: opus_int L /* I number of vectors in codebook */ Chris@69: ); Chris@69: Chris@69: #if defined OPUS_X86_PRESUME_SSE4_1 Chris@69: Chris@69: #define silk_VQ_WMat_EC(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \ Chris@69: mu_Q9, max_gain_Q7, L, arch) \ Chris@69: ((void)(arch),silk_VQ_WMat_EC_sse4_1(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \ Chris@69: mu_Q9, max_gain_Q7, L)) Chris@69: Chris@69: #else Chris@69: Chris@69: extern void (*const SILK_VQ_WMAT_EC_IMPL[OPUS_ARCHMASK + 1])( Chris@69: opus_int8 *ind, /* O index of best codebook vector */ Chris@69: opus_int32 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ Chris@69: opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ Chris@69: const opus_int16 *in_Q14, /* I input vector to be quantized */ Chris@69: const opus_int32 *W_Q18, /* I weighting matrix */ Chris@69: const opus_int8 *cb_Q7, /* I codebook */ Chris@69: const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ Chris@69: const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ Chris@69: const opus_int mu_Q9, /* I tradeoff betw. weighted error and rate */ Chris@69: const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ Chris@69: opus_int L /* I number of vectors in codebook */ Chris@69: ); Chris@69: Chris@69: # define silk_VQ_WMat_EC(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \ Chris@69: mu_Q9, max_gain_Q7, L, arch) \ Chris@69: ((*SILK_VQ_WMAT_EC_IMPL[(arch) & OPUS_ARCHMASK])(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \ Chris@69: mu_Q9, max_gain_Q7, L)) Chris@69: Chris@69: #endif Chris@69: #endif Chris@69: Chris@69: #if 0 /* FIXME: SSE disabled until the NSQ code gets updated. */ Chris@69: # define OVERRIDE_silk_NSQ Chris@69: Chris@69: void silk_NSQ_sse4_1( Chris@69: const silk_encoder_state *psEncC, /* I Encoder State */ Chris@69: silk_nsq_state *NSQ, /* I/O NSQ state */ Chris@69: SideInfoIndices *psIndices, /* I/O Quantization Indices */ Chris@69: const opus_int32 x_Q3[], /* I Prefiltered input signal */ Chris@69: opus_int8 pulses[], /* O Quantized pulse signal */ Chris@69: const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ Chris@69: const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ Chris@69: const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ Chris@69: const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ Chris@69: const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ Chris@69: const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ Chris@69: const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ Chris@69: const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ Chris@69: const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ Chris@69: const opus_int LTP_scale_Q14 /* I LTP state scaling */ Chris@69: ); Chris@69: Chris@69: #if defined OPUS_X86_PRESUME_SSE4_1 Chris@69: Chris@69: #define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ Chris@69: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ Chris@69: ((void)(arch),silk_NSQ_sse4_1(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ Chris@69: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) Chris@69: Chris@69: #else Chris@69: Chris@69: extern void (*const SILK_NSQ_IMPL[OPUS_ARCHMASK + 1])( Chris@69: const silk_encoder_state *psEncC, /* I Encoder State */ Chris@69: silk_nsq_state *NSQ, /* I/O NSQ state */ Chris@69: SideInfoIndices *psIndices, /* I/O Quantization Indices */ Chris@69: const opus_int32 x_Q3[], /* I Prefiltered input signal */ Chris@69: opus_int8 pulses[], /* O Quantized pulse signal */ Chris@69: const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ Chris@69: const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ Chris@69: const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ Chris@69: const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ Chris@69: const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ Chris@69: const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ Chris@69: const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ Chris@69: const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ Chris@69: const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ Chris@69: const opus_int LTP_scale_Q14 /* I LTP state scaling */ Chris@69: ); Chris@69: Chris@69: # define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ Chris@69: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ Chris@69: ((*SILK_NSQ_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ Chris@69: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) Chris@69: Chris@69: #endif Chris@69: Chris@69: # define OVERRIDE_silk_NSQ_del_dec Chris@69: Chris@69: void silk_NSQ_del_dec_sse4_1( Chris@69: const silk_encoder_state *psEncC, /* I Encoder State */ Chris@69: silk_nsq_state *NSQ, /* I/O NSQ state */ Chris@69: SideInfoIndices *psIndices, /* I/O Quantization Indices */ Chris@69: const opus_int32 x_Q3[], /* I Prefiltered input signal */ Chris@69: opus_int8 pulses[], /* O Quantized pulse signal */ Chris@69: const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ Chris@69: const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ Chris@69: const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ Chris@69: const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ Chris@69: const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ Chris@69: const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ Chris@69: const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ Chris@69: const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ Chris@69: const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ Chris@69: const opus_int LTP_scale_Q14 /* I LTP state scaling */ Chris@69: ); Chris@69: Chris@69: #if defined OPUS_X86_PRESUME_SSE4_1 Chris@69: Chris@69: #define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ Chris@69: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ Chris@69: ((void)(arch),silk_NSQ_del_dec_sse4_1(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ Chris@69: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) Chris@69: Chris@69: #else Chris@69: Chris@69: extern void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( Chris@69: const silk_encoder_state *psEncC, /* I Encoder State */ Chris@69: silk_nsq_state *NSQ, /* I/O NSQ state */ Chris@69: SideInfoIndices *psIndices, /* I/O Quantization Indices */ Chris@69: const opus_int32 x_Q3[], /* I Prefiltered input signal */ Chris@69: opus_int8 pulses[], /* O Quantized pulse signal */ Chris@69: const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ Chris@69: const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ Chris@69: const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ Chris@69: const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ Chris@69: const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ Chris@69: const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ Chris@69: const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ Chris@69: const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ Chris@69: const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ Chris@69: const opus_int LTP_scale_Q14 /* I LTP state scaling */ Chris@69: ); Chris@69: Chris@69: # define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ Chris@69: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ Chris@69: ((*SILK_NSQ_DEL_DEC_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ Chris@69: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) Chris@69: Chris@69: #endif Chris@69: #endif Chris@69: Chris@69: void silk_noise_shape_quantizer( Chris@69: silk_nsq_state *NSQ, /* I/O NSQ state */ Chris@69: opus_int signalType, /* I Signal type */ Chris@69: const opus_int32 x_sc_Q10[], /* I */ Chris@69: opus_int8 pulses[], /* O */ Chris@69: opus_int16 xq[], /* O */ Chris@69: opus_int32 sLTP_Q15[], /* I/O LTP state */ Chris@69: const opus_int16 a_Q12[], /* I Short term prediction coefs */ Chris@69: const opus_int16 b_Q14[], /* I Long term prediction coefs */ Chris@69: const opus_int16 AR_shp_Q13[], /* I Noise shaping AR coefs */ Chris@69: opus_int lag, /* I Pitch lag */ Chris@69: opus_int32 HarmShapeFIRPacked_Q14, /* I */ Chris@69: opus_int Tilt_Q14, /* I Spectral tilt */ Chris@69: opus_int32 LF_shp_Q14, /* I */ Chris@69: opus_int32 Gain_Q16, /* I */ Chris@69: opus_int Lambda_Q10, /* I */ Chris@69: opus_int offset_Q10, /* I */ Chris@69: opus_int length, /* I Input length */ Chris@69: opus_int shapingLPCOrder, /* I Noise shaping AR filter order */ Chris@69: opus_int predictLPCOrder, /* I Prediction filter order */ Chris@69: int arch /* I Architecture */ Chris@69: ); Chris@69: Chris@69: /**************************/ Chris@69: /* Noise level estimation */ Chris@69: /**************************/ Chris@69: void silk_VAD_GetNoiseLevels( Chris@69: const opus_int32 pX[ VAD_N_BANDS ], /* I subband energies */ Chris@69: silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ Chris@69: ); Chris@69: Chris@69: # define OVERRIDE_silk_VAD_GetSA_Q8 Chris@69: Chris@69: opus_int silk_VAD_GetSA_Q8_sse4_1( Chris@69: silk_encoder_state *psEnC, Chris@69: const opus_int16 pIn[] Chris@69: ); Chris@69: Chris@69: #if defined(OPUS_X86_PRESUME_SSE4_1) Chris@69: #define silk_VAD_GetSA_Q8(psEnC, pIn, arch) ((void)(arch),silk_VAD_GetSA_Q8_sse4_1(psEnC, pIn)) Chris@69: Chris@69: #else Chris@69: Chris@69: # define silk_VAD_GetSA_Q8(psEnC, pIn, arch) \ Chris@69: ((*SILK_VAD_GETSA_Q8_IMPL[(arch) & OPUS_ARCHMASK])(psEnC, pIn)) Chris@69: Chris@69: extern opus_int (*const SILK_VAD_GETSA_Q8_IMPL[OPUS_ARCHMASK + 1])( Chris@69: silk_encoder_state *psEnC, Chris@69: const opus_int16 pIn[]); Chris@69: Chris@69: #endif Chris@69: Chris@69: # endif Chris@69: #endif