main_sse.h (20671B)
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 void silk_VQ_WMat_EC_sse4_1( 38 opus_int8 *ind, /* O index of best codebook vector */ 39 opus_int32 *res_nrg_Q15, /* O best residual energy */ 40 opus_int32 *rate_dist_Q8, /* O best total bitrate */ 41 opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ 42 const opus_int32 *XX_Q17, /* I correlation matrix */ 43 const opus_int32 *xX_Q17, /* I correlation vector */ 44 const opus_int8 *cb_Q7, /* I codebook */ 45 const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ 46 const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ 47 const opus_int subfr_len, /* I number of samples per subframe */ 48 const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ 49 const opus_int L /* I number of vectors in codebook */ 50 ); 51 52 # if defined OPUS_X86_PRESUME_SSE4_1 53 54 # define OVERRIDE_silk_VQ_WMat_EC 55 # define silk_VQ_WMat_EC(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \ 56 subfr_len, max_gain_Q7, L, arch) \ 57 ((void)(arch),silk_VQ_WMat_EC_sse4_1(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \ 58 subfr_len, max_gain_Q7, L)) 59 60 # elif defined(OPUS_HAVE_RTCD) 61 62 extern void (*const SILK_VQ_WMAT_EC_IMPL[OPUS_ARCHMASK + 1])( 63 opus_int8 *ind, /* O index of best codebook vector */ 64 opus_int32 *res_nrg_Q15, /* O best residual energy */ 65 opus_int32 *rate_dist_Q8, /* O best total bitrate */ 66 opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ 67 const opus_int32 *XX_Q17, /* I correlation matrix */ 68 const opus_int32 *xX_Q17, /* I correlation vector */ 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 subfr_len, /* I number of samples per subframe */ 73 const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ 74 const opus_int L /* I number of vectors in codebook */ 75 ); 76 77 # define OVERRIDE_silk_VQ_WMat_EC 78 # define silk_VQ_WMat_EC(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \ 79 subfr_len, max_gain_Q7, L, arch) \ 80 ((*SILK_VQ_WMAT_EC_IMPL[(arch) & OPUS_ARCHMASK])(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \ 81 subfr_len, max_gain_Q7, L)) 82 83 # endif 84 85 void silk_NSQ_sse4_1( 86 const silk_encoder_state *psEncC, /* I Encoder State */ 87 silk_nsq_state *NSQ, /* I/O NSQ state */ 88 SideInfoIndices *psIndices, /* I/O Quantization Indices */ 89 const opus_int16 x16[], /* I Input */ 90 opus_int8 pulses[], /* O Quantized pulse signal */ 91 const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ 92 const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ 93 const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ 94 const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ 95 const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ 96 const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ 97 const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ 98 const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ 99 const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ 100 const opus_int LTP_scale_Q14 /* I LTP state scaling */ 101 ); 102 103 # if defined OPUS_X86_PRESUME_SSE4_1 104 105 # define OVERRIDE_silk_NSQ 106 # define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 107 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ 108 ((void)(arch),silk_NSQ_sse4_1(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)) 110 111 # elif defined(OPUS_HAVE_RTCD) 112 113 extern void (*const SILK_NSQ_IMPL[OPUS_ARCHMASK + 1])( 114 const silk_encoder_state *psEncC, /* I Encoder State */ 115 silk_nsq_state *NSQ, /* I/O NSQ state */ 116 SideInfoIndices *psIndices, /* I/O Quantization Indices */ 117 const opus_int16 x16[], /* I Input */ 118 opus_int8 pulses[], /* O Quantized pulse signal */ 119 const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ 120 const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ 121 const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ 122 const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ 123 const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ 124 const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ 125 const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ 126 const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ 127 const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ 128 const opus_int LTP_scale_Q14 /* I LTP state scaling */ 129 ); 130 131 # define OVERRIDE_silk_NSQ 132 # define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 133 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ 134 ((*SILK_NSQ_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 135 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) 136 137 # endif 138 139 void silk_NSQ_del_dec_sse4_1( 140 const silk_encoder_state *psEncC, /* I Encoder State */ 141 silk_nsq_state *NSQ, /* I/O NSQ state */ 142 SideInfoIndices *psIndices, /* I/O Quantization Indices */ 143 const opus_int16 x16[], /* I Input */ 144 opus_int8 pulses[], /* O Quantized pulse signal */ 145 const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ 146 const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ 147 const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ 148 const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ 149 const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ 150 const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ 151 const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ 152 const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ 153 const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ 154 const opus_int LTP_scale_Q14 /* I LTP state scaling */ 155 ); 156 157 void silk_NSQ_del_dec_avx2( 158 const silk_encoder_state *psEncC, /* I Encoder State */ 159 silk_nsq_state *NSQ, /* I/O NSQ state */ 160 SideInfoIndices *psIndices, /* I/O Quantization Indices */ 161 const opus_int16 x16[], /* I Input */ 162 opus_int8 pulses[], /* O Quantized pulse signal */ 163 const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ 164 const opus_int16 LTPCoef_Q14[LTP_ORDER * MAX_NB_SUBFR], /* I Long term prediction coefs */ 165 const opus_int16 AR_Q13[MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER], /* I Noise shaping coefs */ 166 const opus_int HarmShapeGain_Q14[MAX_NB_SUBFR], /* I Long term shaping coefs */ 167 const opus_int Tilt_Q14[MAX_NB_SUBFR], /* I Spectral tilt */ 168 const opus_int32 LF_shp_Q14[MAX_NB_SUBFR], /* I Low frequency shaping coefs */ 169 const opus_int32 Gains_Q16[MAX_NB_SUBFR], /* I Quantization step sizes */ 170 const opus_int32 pitchL[MAX_NB_SUBFR], /* I Pitch lags */ 171 const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ 172 const opus_int LTP_scale_Q14 /* I LTP state scaling */ 173 ); 174 175 # if defined (OPUS_X86_PRESUME_AVX2) 176 177 # define OVERRIDE_silk_NSQ_del_dec 178 # define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ 179 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ 180 ((void)(arch),silk_NSQ_del_dec_avx2(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ 181 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) 182 183 # elif defined (OPUS_X86_PRESUME_SSE4_1) && !defined(OPUS_X86_MAY_HAVE_AVX2) 184 185 # define OVERRIDE_silk_NSQ_del_dec 186 # define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ 187 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ 188 ((void)(arch),silk_NSQ_del_dec_sse4_1(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ 189 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) 190 191 # elif defined(OPUS_HAVE_RTCD) 192 193 extern void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( 194 const silk_encoder_state *psEncC, /* I Encoder State */ 195 silk_nsq_state *NSQ, /* I/O NSQ state */ 196 SideInfoIndices *psIndices, /* I/O Quantization Indices */ 197 const opus_int16 x16[], /* I Input */ 198 opus_int8 pulses[], /* O Quantized pulse signal */ 199 const opus_int16 *PredCoef_Q12, /* I Short term prediction coefs */ 200 const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ 201 const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ 202 const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ 203 const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ 204 const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ 205 const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ 206 const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ 207 const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ 208 const opus_int LTP_scale_Q14 /* I LTP state scaling */ 209 ); 210 211 # define OVERRIDE_silk_NSQ_del_dec 212 # define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ 213 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ 214 ((*SILK_NSQ_DEL_DEC_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ 215 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) 216 217 # endif 218 219 void silk_noise_shape_quantizer( 220 silk_nsq_state *NSQ, /* I/O NSQ state */ 221 opus_int signalType, /* I Signal type */ 222 const opus_int32 x_sc_Q10[], /* I */ 223 opus_int8 pulses[], /* O */ 224 opus_int16 xq[], /* O */ 225 opus_int32 sLTP_Q15[], /* I/O LTP state */ 226 const opus_int16 a_Q12[], /* I Short term prediction coefs */ 227 const opus_int16 b_Q14[], /* I Long term prediction coefs */ 228 const opus_int16 AR_shp_Q13[], /* I Noise shaping AR coefs */ 229 opus_int lag, /* I Pitch lag */ 230 opus_int32 HarmShapeFIRPacked_Q14, /* I */ 231 opus_int Tilt_Q14, /* I Spectral tilt */ 232 opus_int32 LF_shp_Q14, /* I */ 233 opus_int32 Gain_Q16, /* I */ 234 opus_int Lambda_Q10, /* I */ 235 opus_int offset_Q10, /* I */ 236 opus_int length, /* I Input length */ 237 opus_int shapingLPCOrder, /* I Noise shaping AR filter order */ 238 opus_int predictLPCOrder, /* I Prediction filter order */ 239 int arch /* I Architecture */ 240 ); 241 242 /**************************/ 243 /* Noise level estimation */ 244 /**************************/ 245 void silk_VAD_GetNoiseLevels( 246 const opus_int32 pX[ VAD_N_BANDS ], /* I subband energies */ 247 silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ 248 ); 249 250 opus_int silk_VAD_GetSA_Q8_sse4_1( 251 silk_encoder_state *psEnC, 252 const opus_int16 pIn[] 253 ); 254 255 # if defined(OPUS_X86_PRESUME_SSE4_1) 256 257 # define OVERRIDE_silk_VAD_GetSA_Q8 258 # define silk_VAD_GetSA_Q8(psEnC, pIn, arch) ((void)(arch),silk_VAD_GetSA_Q8_sse4_1(psEnC, pIn)) 259 260 # elif defined(OPUS_HAVE_RTCD) 261 262 extern opus_int (*const SILK_VAD_GETSA_Q8_IMPL[OPUS_ARCHMASK + 1])( 263 silk_encoder_state *psEnC, 264 const opus_int16 pIn[]); 265 266 # define OVERRIDE_silk_VAD_GetSA_Q8 267 # define silk_VAD_GetSA_Q8(psEnC, pIn, arch) \ 268 ((*SILK_VAD_GETSA_Q8_IMPL[(arch) & OPUS_ARCHMASK])(psEnC, pIn)) 269 270 # endif 271 272 #ifndef FIXED_POINT 273 double silk_inner_product_FLP_avx2( 274 const silk_float *data1, 275 const silk_float *data2, 276 opus_int dataSize 277 ); 278 279 #if defined (OPUS_X86_PRESUME_AVX2) 280 281 #define OVERRIDE_inner_product_FLP 282 #define silk_inner_product_FLP(data1, data2, dataSize, arch) ((void)arch,silk_inner_product_FLP_avx2(data1, data2, dataSize)) 283 284 #elif defined(OPUS_HAVE_RTCD) && defined(OPUS_X86_MAY_HAVE_AVX2) 285 286 #define OVERRIDE_inner_product_FLP 287 extern double (*const SILK_INNER_PRODUCT_FLP_IMPL[OPUS_ARCHMASK + 1])( 288 const silk_float *data1, 289 const silk_float *data2, 290 opus_int dataSize 291 ); 292 293 #define silk_inner_product_FLP(data1, data2, dataSize, arch) ((void)arch,(*SILK_INNER_PRODUCT_FLP_IMPL[(arch) & OPUS_ARCHMASK])(data1, data2, dataSize)) 294 295 #endif 296 #endif 297 298 # endif 299 #endif