1 /* Copyright (c) 2014, Cisco Systems, INC 2 Written by XiangMingZhu WeiZhou MinPeng YanWang 3 4 Redistribution and use in source and binary forms, with or without 5 modification, are permitted provided that the following conditions 6 are met: 7 8 - Redistributions of source code must retain the above copyright 9 notice, this list of conditions and the following disclaimer. 10 11 - Redistributions in binary form must reproduce the above copyright 12 notice, this list of conditions and the following disclaimer in the 13 documentation and/or other materials provided with the distribution. 14 15 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 16 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 17 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 18 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER 19 OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 20 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 21 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 22 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 23 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 24 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 25 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 26 */ 27 28 #ifndef MAIN_SSE_H 29 #define MAIN_SSE_H 30 31 #ifdef HAVE_CONFIG_H 32 #include "config.h" 33 #endif 34 35 # if defined(OPUS_X86_MAY_HAVE_SSE4_1) 36 37 #if 0 /* FIXME: SSE disabled until silk_VQ_WMat_EC_sse4_1() gets updated. */ 38 # define OVERRIDE_silk_VQ_WMat_EC 39 40 void silk_VQ_WMat_EC_sse4_1( 41 opus_int8 *ind, /* O index of best codebook vector */ 42 opus_int32 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ 43 opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ 44 const opus_int16 *in_Q14, /* I input vector to be quantized */ 45 const opus_int32 *W_Q18, /* I weighting matrix */ 46 const opus_int8 *cb_Q7, /* I codebook */ 47 const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ 48 const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ 49 const opus_int mu_Q9, /* I tradeoff betw. weighted error and rate */ 50 const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ 51 opus_int L /* I number of vectors in codebook */ 52 ); 53 54 #if defined OPUS_X86_PRESUME_SSE4_1 55 56 #define silk_VQ_WMat_EC(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \ 57 mu_Q9, max_gain_Q7, L, arch) \ 58 ((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, \ 59 mu_Q9, max_gain_Q7, L)) 60 61 #else 62 63 extern void (*const SILK_VQ_WMAT_EC_IMPL[OPUS_ARCHMASK + 1])( 64 opus_int8 *ind, /* O index of best codebook vector */ 65 opus_int32 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ 66 opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ 67 const opus_int16 *in_Q14, /* I input vector to be quantized */ 68 const opus_int32 *W_Q18, /* I weighting matrix */ 69 const opus_int8 *cb_Q7, /* I codebook */ 70 const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ 71 const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ 72 const opus_int mu_Q9, /* I tradeoff betw. weighted error and rate */ 73 const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ 74 opus_int L /* I number of vectors in codebook */ 75 ); 76 77 # define silk_VQ_WMat_EC(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \ 78 mu_Q9, max_gain_Q7, L, arch) \ 79 ((*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, \ 80 mu_Q9, max_gain_Q7, L)) 81 82 #endif 83 #endif 84 85 #if 0 /* FIXME: SSE disabled until the NSQ code gets updated. */ 86 # define OVERRIDE_silk_NSQ 87 88 void silk_NSQ_sse4_1( 89 const silk_encoder_state *psEncC, /* I Encoder State */ 90 silk_nsq_state *NSQ, /* I/O NSQ state */ 91 SideInfoIndices *psIndices, /* I/O Quantization Indices */ 92 const opus_int32 x_Q3[], /* I Prefiltered input signal */ 93 opus_int8 pulses[], /* O Quantized pulse signal */ 94 const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ 95 const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ 96 const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ 97 const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ 98 const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ 99 const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ 100 const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ 101 const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ 102 const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ 103 const opus_int LTP_scale_Q14 /* I LTP state scaling */ 104 ); 105 106 #if defined OPUS_X86_PRESUME_SSE4_1 107 108 #define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 109 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ 110 ((void)(arch),silk_NSQ_sse4_1(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 111 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) 112 113 #else 114 115 extern void (*const SILK_NSQ_IMPL[OPUS_ARCHMASK + 1])( 116 const silk_encoder_state *psEncC, /* I Encoder State */ 117 silk_nsq_state *NSQ, /* I/O NSQ state */ 118 SideInfoIndices *psIndices, /* I/O Quantization Indices */ 119 const opus_int32 x_Q3[], /* I Prefiltered input signal */ 120 opus_int8 pulses[], /* O Quantized pulse signal */ 121 const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ 122 const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ 123 const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ 124 const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ 125 const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ 126 const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ 127 const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ 128 const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ 129 const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ 130 const opus_int LTP_scale_Q14 /* I LTP state scaling */ 131 ); 132 133 # define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 134 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ 135 ((*SILK_NSQ_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 136 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) 137 138 #endif 139 140 # define OVERRIDE_silk_NSQ_del_dec 141 142 void silk_NSQ_del_dec_sse4_1( 143 const silk_encoder_state *psEncC, /* I Encoder State */ 144 silk_nsq_state *NSQ, /* I/O NSQ state */ 145 SideInfoIndices *psIndices, /* I/O Quantization Indices */ 146 const opus_int32 x_Q3[], /* I Prefiltered input signal */ 147 opus_int8 pulses[], /* O Quantized pulse signal */ 148 const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ 149 const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ 150 const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ 151 const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ 152 const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ 153 const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ 154 const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ 155 const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ 156 const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ 157 const opus_int LTP_scale_Q14 /* I LTP state scaling */ 158 ); 159 160 #if defined OPUS_X86_PRESUME_SSE4_1 161 162 #define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 163 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ 164 ((void)(arch),silk_NSQ_del_dec_sse4_1(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 165 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) 166 167 #else 168 169 extern void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( 170 const silk_encoder_state *psEncC, /* I Encoder State */ 171 silk_nsq_state *NSQ, /* I/O NSQ state */ 172 SideInfoIndices *psIndices, /* I/O Quantization Indices */ 173 const opus_int32 x_Q3[], /* I Prefiltered input signal */ 174 opus_int8 pulses[], /* O Quantized pulse signal */ 175 const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ 176 const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ 177 const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ 178 const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ 179 const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ 180 const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ 181 const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ 182 const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ 183 const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ 184 const opus_int LTP_scale_Q14 /* I LTP state scaling */ 185 ); 186 187 # define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 188 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ 189 ((*SILK_NSQ_DEL_DEC_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 190 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) 191 192 #endif 193 #endif 194 195 void silk_noise_shape_quantizer( 196 silk_nsq_state *NSQ, /* I/O NSQ state */ 197 opus_int signalType, /* I Signal type */ 198 const opus_int32 x_sc_Q10[], /* I */ 199 opus_int8 pulses[], /* O */ 200 opus_int16 xq[], /* O */ 201 opus_int32 sLTP_Q15[], /* I/O LTP state */ 202 const opus_int16 a_Q12[], /* I Short term prediction coefs */ 203 const opus_int16 b_Q14[], /* I Long term prediction coefs */ 204 const opus_int16 AR_shp_Q13[], /* I Noise shaping AR coefs */ 205 opus_int lag, /* I Pitch lag */ 206 opus_int32 HarmShapeFIRPacked_Q14, /* I */ 207 opus_int Tilt_Q14, /* I Spectral tilt */ 208 opus_int32 LF_shp_Q14, /* I */ 209 opus_int32 Gain_Q16, /* I */ 210 opus_int Lambda_Q10, /* I */ 211 opus_int offset_Q10, /* I */ 212 opus_int length, /* I Input length */ 213 opus_int shapingLPCOrder, /* I Noise shaping AR filter order */ 214 opus_int predictLPCOrder, /* I Prediction filter order */ 215 int arch /* I Architecture */ 216 ); 217 218 /**************************/ 219 /* Noise level estimation */ 220 /**************************/ 221 void silk_VAD_GetNoiseLevels( 222 const opus_int32 pX[ VAD_N_BANDS ], /* I subband energies */ 223 silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ 224 ); 225 226 # define OVERRIDE_silk_VAD_GetSA_Q8 227 228 opus_int silk_VAD_GetSA_Q8_sse4_1( 229 silk_encoder_state *psEnC, 230 const opus_int16 pIn[] 231 ); 232 233 #if defined(OPUS_X86_PRESUME_SSE4_1) 234 #define silk_VAD_GetSA_Q8(psEnC, pIn, arch) ((void)(arch),silk_VAD_GetSA_Q8_sse4_1(psEnC, pIn)) 235 236 #else 237 238 # define silk_VAD_GetSA_Q8(psEnC, pIn, arch) \ 239 ((*SILK_VAD_GETSA_Q8_IMPL[(arch) & OPUS_ARCHMASK])(psEnC, pIn)) 240 241 extern opus_int (*const SILK_VAD_GETSA_Q8_IMPL[OPUS_ARCHMASK + 1])( 242 silk_encoder_state *psEnC, 243 const opus_int16 pIn[]); 244 245 #endif 246 247 # endif 248 #endif 249