cannam@154: /* Copyright (c) 2014, Cisco Systems, INC cannam@154: Written by XiangMingZhu WeiZhou MinPeng YanWang cannam@154: cannam@154: Redistribution and use in source and binary forms, with or without cannam@154: modification, are permitted provided that the following conditions cannam@154: are met: cannam@154: cannam@154: - Redistributions of source code must retain the above copyright cannam@154: notice, this list of conditions and the following disclaimer. cannam@154: cannam@154: - Redistributions in binary form must reproduce the above copyright cannam@154: notice, this list of conditions and the following disclaimer in the cannam@154: documentation and/or other materials provided with the distribution. cannam@154: cannam@154: THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS cannam@154: ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT cannam@154: LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR cannam@154: A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER cannam@154: OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, cannam@154: EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, cannam@154: PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR cannam@154: PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF cannam@154: LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING cannam@154: NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS cannam@154: SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. cannam@154: */ cannam@154: cannam@154: #ifndef MAIN_SSE_H cannam@154: #define MAIN_SSE_H cannam@154: cannam@154: #ifdef HAVE_CONFIG_H cannam@154: #include "config.h" cannam@154: #endif cannam@154: cannam@154: # if defined(OPUS_X86_MAY_HAVE_SSE4_1) cannam@154: cannam@154: #if 0 /* FIXME: SSE disabled until silk_VQ_WMat_EC_sse4_1() gets updated. */ cannam@154: # define OVERRIDE_silk_VQ_WMat_EC cannam@154: cannam@154: void silk_VQ_WMat_EC_sse4_1( cannam@154: opus_int8 *ind, /* O index of best codebook vector */ cannam@154: opus_int32 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ cannam@154: opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ cannam@154: const opus_int16 *in_Q14, /* I input vector to be quantized */ cannam@154: const opus_int32 *W_Q18, /* I weighting matrix */ cannam@154: const opus_int8 *cb_Q7, /* I codebook */ cannam@154: const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ cannam@154: const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ cannam@154: const opus_int mu_Q9, /* I tradeoff betw. weighted error and rate */ cannam@154: const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ cannam@154: opus_int L /* I number of vectors in codebook */ cannam@154: ); cannam@154: cannam@154: #if defined OPUS_X86_PRESUME_SSE4_1 cannam@154: cannam@154: #define silk_VQ_WMat_EC(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \ cannam@154: mu_Q9, max_gain_Q7, L, arch) \ cannam@154: ((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, \ cannam@154: mu_Q9, max_gain_Q7, L)) cannam@154: cannam@154: #else cannam@154: cannam@154: extern void (*const SILK_VQ_WMAT_EC_IMPL[OPUS_ARCHMASK + 1])( cannam@154: opus_int8 *ind, /* O index of best codebook vector */ cannam@154: opus_int32 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ cannam@154: opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ cannam@154: const opus_int16 *in_Q14, /* I input vector to be quantized */ cannam@154: const opus_int32 *W_Q18, /* I weighting matrix */ cannam@154: const opus_int8 *cb_Q7, /* I codebook */ cannam@154: const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ cannam@154: const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ cannam@154: const opus_int mu_Q9, /* I tradeoff betw. weighted error and rate */ cannam@154: const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ cannam@154: opus_int L /* I number of vectors in codebook */ cannam@154: ); cannam@154: cannam@154: # define silk_VQ_WMat_EC(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \ cannam@154: mu_Q9, max_gain_Q7, L, arch) \ cannam@154: ((*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, \ cannam@154: mu_Q9, max_gain_Q7, L)) cannam@154: cannam@154: #endif cannam@154: #endif cannam@154: cannam@154: #if 0 /* FIXME: SSE disabled until the NSQ code gets updated. */ cannam@154: # define OVERRIDE_silk_NSQ cannam@154: cannam@154: void silk_NSQ_sse4_1( cannam@154: const silk_encoder_state *psEncC, /* I Encoder State */ cannam@154: silk_nsq_state *NSQ, /* I/O NSQ state */ cannam@154: SideInfoIndices *psIndices, /* I/O Quantization Indices */ cannam@154: const opus_int32 x_Q3[], /* I Prefiltered input signal */ cannam@154: opus_int8 pulses[], /* O Quantized pulse signal */ cannam@154: const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ cannam@154: const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ cannam@154: const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ cannam@154: const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ cannam@154: const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ cannam@154: const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ cannam@154: const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ cannam@154: const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ cannam@154: const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ cannam@154: const opus_int LTP_scale_Q14 /* I LTP state scaling */ cannam@154: ); cannam@154: cannam@154: #if defined OPUS_X86_PRESUME_SSE4_1 cannam@154: cannam@154: #define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ cannam@154: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ cannam@154: ((void)(arch),silk_NSQ_sse4_1(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ cannam@154: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) cannam@154: cannam@154: #else cannam@154: cannam@154: extern void (*const SILK_NSQ_IMPL[OPUS_ARCHMASK + 1])( cannam@154: const silk_encoder_state *psEncC, /* I Encoder State */ cannam@154: silk_nsq_state *NSQ, /* I/O NSQ state */ cannam@154: SideInfoIndices *psIndices, /* I/O Quantization Indices */ cannam@154: const opus_int32 x_Q3[], /* I Prefiltered input signal */ cannam@154: opus_int8 pulses[], /* O Quantized pulse signal */ cannam@154: const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ cannam@154: const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ cannam@154: const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ cannam@154: const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ cannam@154: const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ cannam@154: const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ cannam@154: const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ cannam@154: const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ cannam@154: const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ cannam@154: const opus_int LTP_scale_Q14 /* I LTP state scaling */ cannam@154: ); cannam@154: cannam@154: # define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ cannam@154: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ cannam@154: ((*SILK_NSQ_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ cannam@154: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) cannam@154: cannam@154: #endif cannam@154: cannam@154: # define OVERRIDE_silk_NSQ_del_dec cannam@154: cannam@154: void silk_NSQ_del_dec_sse4_1( cannam@154: const silk_encoder_state *psEncC, /* I Encoder State */ cannam@154: silk_nsq_state *NSQ, /* I/O NSQ state */ cannam@154: SideInfoIndices *psIndices, /* I/O Quantization Indices */ cannam@154: const opus_int32 x_Q3[], /* I Prefiltered input signal */ cannam@154: opus_int8 pulses[], /* O Quantized pulse signal */ cannam@154: const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ cannam@154: const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ cannam@154: const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ cannam@154: const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ cannam@154: const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ cannam@154: const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ cannam@154: const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ cannam@154: const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ cannam@154: const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ cannam@154: const opus_int LTP_scale_Q14 /* I LTP state scaling */ cannam@154: ); cannam@154: cannam@154: #if defined OPUS_X86_PRESUME_SSE4_1 cannam@154: cannam@154: #define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ cannam@154: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ cannam@154: ((void)(arch),silk_NSQ_del_dec_sse4_1(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ cannam@154: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) cannam@154: cannam@154: #else cannam@154: cannam@154: extern void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( cannam@154: const silk_encoder_state *psEncC, /* I Encoder State */ cannam@154: silk_nsq_state *NSQ, /* I/O NSQ state */ cannam@154: SideInfoIndices *psIndices, /* I/O Quantization Indices */ cannam@154: const opus_int32 x_Q3[], /* I Prefiltered input signal */ cannam@154: opus_int8 pulses[], /* O Quantized pulse signal */ cannam@154: const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ cannam@154: const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ cannam@154: const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ cannam@154: const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ cannam@154: const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ cannam@154: const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ cannam@154: const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ cannam@154: const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ cannam@154: const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ cannam@154: const opus_int LTP_scale_Q14 /* I LTP state scaling */ cannam@154: ); cannam@154: cannam@154: # define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ cannam@154: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ cannam@154: ((*SILK_NSQ_DEL_DEC_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ cannam@154: HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) cannam@154: cannam@154: #endif cannam@154: #endif cannam@154: cannam@154: void silk_noise_shape_quantizer( cannam@154: silk_nsq_state *NSQ, /* I/O NSQ state */ cannam@154: opus_int signalType, /* I Signal type */ cannam@154: const opus_int32 x_sc_Q10[], /* I */ cannam@154: opus_int8 pulses[], /* O */ cannam@154: opus_int16 xq[], /* O */ cannam@154: opus_int32 sLTP_Q15[], /* I/O LTP state */ cannam@154: const opus_int16 a_Q12[], /* I Short term prediction coefs */ cannam@154: const opus_int16 b_Q14[], /* I Long term prediction coefs */ cannam@154: const opus_int16 AR_shp_Q13[], /* I Noise shaping AR coefs */ cannam@154: opus_int lag, /* I Pitch lag */ cannam@154: opus_int32 HarmShapeFIRPacked_Q14, /* I */ cannam@154: opus_int Tilt_Q14, /* I Spectral tilt */ cannam@154: opus_int32 LF_shp_Q14, /* I */ cannam@154: opus_int32 Gain_Q16, /* I */ cannam@154: opus_int Lambda_Q10, /* I */ cannam@154: opus_int offset_Q10, /* I */ cannam@154: opus_int length, /* I Input length */ cannam@154: opus_int shapingLPCOrder, /* I Noise shaping AR filter order */ cannam@154: opus_int predictLPCOrder, /* I Prediction filter order */ cannam@154: int arch /* I Architecture */ cannam@154: ); cannam@154: cannam@154: /**************************/ cannam@154: /* Noise level estimation */ cannam@154: /**************************/ cannam@154: void silk_VAD_GetNoiseLevels( cannam@154: const opus_int32 pX[ VAD_N_BANDS ], /* I subband energies */ cannam@154: silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ cannam@154: ); cannam@154: cannam@154: # define OVERRIDE_silk_VAD_GetSA_Q8 cannam@154: cannam@154: opus_int silk_VAD_GetSA_Q8_sse4_1( cannam@154: silk_encoder_state *psEnC, cannam@154: const opus_int16 pIn[] cannam@154: ); cannam@154: cannam@154: #if defined(OPUS_X86_PRESUME_SSE4_1) cannam@154: #define silk_VAD_GetSA_Q8(psEnC, pIn, arch) ((void)(arch),silk_VAD_GetSA_Q8_sse4_1(psEnC, pIn)) cannam@154: cannam@154: #else cannam@154: cannam@154: # define silk_VAD_GetSA_Q8(psEnC, pIn, arch) \ cannam@154: ((*SILK_VAD_GETSA_Q8_IMPL[(arch) & OPUS_ARCHMASK])(psEnC, pIn)) cannam@154: cannam@154: extern opus_int (*const SILK_VAD_GETSA_Q8_IMPL[OPUS_ARCHMASK + 1])( cannam@154: silk_encoder_state *psEnC, cannam@154: const opus_int16 pIn[]); cannam@154: cannam@154: #endif cannam@154: cannam@154: # endif cannam@154: #endif