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 # define OVERRIDE_silk_VQ_WMat_EC 38 39 void silk_VQ_WMat_EC_sse4_1( 40 opus_int8 *ind, /* O index of best codebook vector */ 41 opus_int32 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ 42 opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ 43 const opus_int16 *in_Q14, /* I input vector to be quantized */ 44 const opus_int32 *W_Q18, /* I weighting matrix */ 45 const opus_int8 *cb_Q7, /* I codebook */ 46 const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ 47 const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ 48 const opus_int mu_Q9, /* I tradeoff betw. weighted error and rate */ 49 const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ 50 opus_int L /* I number of vectors in codebook */ 51 ); 52 53 #if defined OPUS_X86_PRESUME_SSE4_1 54 55 #define silk_VQ_WMat_EC(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \ 56 mu_Q9, max_gain_Q7, L, arch) \ 57 ((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, \ 58 mu_Q9, max_gain_Q7, L)) 59 60 #else 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 *rate_dist_Q14, /* O best weighted quant error + mu * rate */ 65 opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ 66 const opus_int16 *in_Q14, /* I input vector to be quantized */ 67 const opus_int32 *W_Q18, /* I weighting matrix */ 68 const opus_int8 *cb_Q7, /* I codebook */ 69 const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ 70 const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ 71 const opus_int mu_Q9, /* I tradeoff betw. weighted error and rate */ 72 const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ 73 opus_int L /* I number of vectors in codebook */ 74 ); 75 76 # define silk_VQ_WMat_EC(ind, rate_dist_Q14, gain_Q7, in_Q14, W_Q18, cb_Q7, cb_gain_Q7, cl_Q5, \ 77 mu_Q9, max_gain_Q7, L, arch) \ 78 ((*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, \ 79 mu_Q9, max_gain_Q7, L)) 80 81 #endif 82 83 # define OVERRIDE_silk_NSQ 84 85 void silk_NSQ_sse4_1( 86 const silk_encoder_state *psEncC, /* I/O Encoder State */ 87 silk_nsq_state *NSQ, /* I/O NSQ state */ 88 SideInfoIndices *psIndices, /* I/O Quantization Indices */ 89 const opus_int32 x_Q3[], /* I Prefiltered input signal */ 90 opus_int8 pulses[], /* O Quantized pulse signal */ 91 const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* 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 AR2_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 silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 106 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ 107 ((void)(arch),silk_NSQ_sse4_1(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 108 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) 109 110 #else 111 112 extern void (*const SILK_NSQ_IMPL[OPUS_ARCHMASK + 1])( 113 const silk_encoder_state *psEncC, /* I/O Encoder State */ 114 silk_nsq_state *NSQ, /* I/O NSQ state */ 115 SideInfoIndices *psIndices, /* I/O Quantization Indices */ 116 const opus_int32 x_Q3[], /* I Prefiltered input signal */ 117 opus_int8 pulses[], /* O Quantized pulse signal */ 118 const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ 119 const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ 120 const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ 121 const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ 122 const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ 123 const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ 124 const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ 125 const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ 126 const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ 127 const opus_int LTP_scale_Q14 /* I LTP state scaling */ 128 ); 129 130 # define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 131 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ 132 ((*SILK_NSQ_IMPL[(arch) & OPUS_ARCHMASK])(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)) 134 135 #endif 136 137 # define OVERRIDE_silk_NSQ_del_dec 138 139 void silk_NSQ_del_dec_sse4_1( 140 const silk_encoder_state *psEncC, /* I/O Encoder State */ 141 silk_nsq_state *NSQ, /* I/O NSQ state */ 142 SideInfoIndices *psIndices, /* I/O Quantization Indices */ 143 const opus_int32 x_Q3[], /* I Prefiltered input signal */ 144 opus_int8 pulses[], /* O Quantized pulse signal */ 145 const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* 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 AR2_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 #if defined OPUS_X86_PRESUME_SSE4_1 158 159 #define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 160 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ 161 ((void)(arch),silk_NSQ_del_dec_sse4_1(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 162 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) 163 164 #else 165 166 extern void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( 167 const silk_encoder_state *psEncC, /* I/O Encoder State */ 168 silk_nsq_state *NSQ, /* I/O NSQ state */ 169 SideInfoIndices *psIndices, /* I/O Quantization Indices */ 170 const opus_int32 x_Q3[], /* I Prefiltered input signal */ 171 opus_int8 pulses[], /* O Quantized pulse signal */ 172 const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ 173 const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ 174 const opus_int16 AR2_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ 175 const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ 176 const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ 177 const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ 178 const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ 179 const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ 180 const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ 181 const opus_int LTP_scale_Q14 /* I LTP state scaling */ 182 ); 183 184 # define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 185 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ 186 ((*SILK_NSQ_DEL_DEC_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ 187 HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) 188 189 #endif 190 191 void silk_noise_shape_quantizer( 192 silk_nsq_state *NSQ, /* I/O NSQ state */ 193 opus_int signalType, /* I Signal type */ 194 const opus_int32 x_sc_Q10[], /* I */ 195 opus_int8 pulses[], /* O */ 196 opus_int16 xq[], /* O */ 197 opus_int32 sLTP_Q15[], /* I/O LTP state */ 198 const opus_int16 a_Q12[], /* I Short term prediction coefs */ 199 const opus_int16 b_Q14[], /* I Long term prediction coefs */ 200 const opus_int16 AR_shp_Q13[], /* I Noise shaping AR coefs */ 201 opus_int lag, /* I Pitch lag */ 202 opus_int32 HarmShapeFIRPacked_Q14, /* I */ 203 opus_int Tilt_Q14, /* I Spectral tilt */ 204 opus_int32 LF_shp_Q14, /* I */ 205 opus_int32 Gain_Q16, /* I */ 206 opus_int Lambda_Q10, /* I */ 207 opus_int offset_Q10, /* I */ 208 opus_int length, /* I Input length */ 209 opus_int shapingLPCOrder, /* I Noise shaping AR filter order */ 210 opus_int predictLPCOrder, /* I Prediction filter order */ 211 int arch /* I Architecture */ 212 ); 213 214 /**************************/ 215 /* Noise level estimation */ 216 /**************************/ 217 void silk_VAD_GetNoiseLevels( 218 const opus_int32 pX[ VAD_N_BANDS ], /* I subband energies */ 219 silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ 220 ); 221 222 # define OVERRIDE_silk_VAD_GetSA_Q8 223 224 opus_int silk_VAD_GetSA_Q8_sse4_1( 225 silk_encoder_state *psEnC, 226 const opus_int16 pIn[] 227 ); 228 229 #if defined(OPUS_X86_PRESUME_SSE4_1) 230 #define silk_VAD_GetSA_Q8(psEnC, pIn, arch) ((void)(arch),silk_VAD_GetSA_Q8_sse4_1(psEnC, pIn)) 231 232 #else 233 234 # define silk_VAD_GetSA_Q8(psEnC, pIn, arch) \ 235 ((*SILK_VAD_GETSA_Q8_IMPL[(arch) & OPUS_ARCHMASK])(psEnC, pIn)) 236 237 extern opus_int (*const SILK_VAD_GETSA_Q8_IMPL[OPUS_ARCHMASK + 1])( 238 silk_encoder_state *psEnC, 239 const opus_int16 pIn[]); 240 241 # define OVERRIDE_silk_warped_LPC_analysis_filter_FIX 242 243 #endif 244 245 void silk_warped_LPC_analysis_filter_FIX_sse4_1( 246 opus_int32 state[], /* I/O State [order + 1] */ 247 opus_int32 res_Q2[], /* O Residual signal [length] */ 248 const opus_int16 coef_Q13[], /* I Coefficients [order] */ 249 const opus_int16 input[], /* I Input signal [length] */ 250 const opus_int16 lambda_Q16, /* I Warping factor */ 251 const opus_int length, /* I Length of input signal */ 252 const opus_int order /* I Filter order (even) */ 253 ); 254 255 #if defined(OPUS_X86_PRESUME_SSE4_1) 256 #define silk_warped_LPC_analysis_filter_FIX(state, res_Q2, coef_Q13, input, lambda_Q16, length, order, arch) \ 257 ((void)(arch),silk_warped_LPC_analysis_filter_FIX_c(state, res_Q2, coef_Q13, input, lambda_Q16, length, order)) 258 259 #else 260 261 extern void (*const SILK_WARPED_LPC_ANALYSIS_FILTER_FIX_IMPL[OPUS_ARCHMASK + 1])( 262 opus_int32 state[], /* I/O State [order + 1] */ 263 opus_int32 res_Q2[], /* O Residual signal [length] */ 264 const opus_int16 coef_Q13[], /* I Coefficients [order] */ 265 const opus_int16 input[], /* I Input signal [length] */ 266 const opus_int16 lambda_Q16, /* I Warping factor */ 267 const opus_int length, /* I Length of input signal */ 268 const opus_int order /* I Filter order (even) */ 269 ); 270 271 # define silk_warped_LPC_analysis_filter_FIX(state, res_Q2, coef_Q13, input, lambda_Q16, length, order, arch) \ 272 ((*SILK_WARPED_LPC_ANALYSIS_FILTER_FIX_IMPL[(arch) & OPUS_ARCHMASK])(state, res_Q2, coef_Q13, input, lambda_Q16, length, order)) 273 274 #endif 275 276 # endif 277 #endif 278