1 /* 2 * Copyright (c) 2013 The WebRTC project authors. All Rights Reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 11 /* 12 * The core AEC algorithm, which is presented with time-aligned signals. 13 */ 14 15 #include "webrtc/modules/audio_processing/aec/aec_core.h" 16 17 #include <math.h> 18 19 #include "webrtc/common_audio/signal_processing/include/signal_processing_library.h" 20 #include "webrtc/modules/audio_processing/aec/aec_core_internal.h" 21 #include "webrtc/modules/audio_processing/aec/aec_rdft.h" 22 23 extern const float WebRtcAec_weightCurve[65]; 24 extern const float WebRtcAec_overDriveCurve[65]; 25 26 void WebRtcAec_ComfortNoise_mips(AecCore* aec, 27 float efw[2][PART_LEN1], 28 float comfortNoiseHband[2][PART_LEN1], 29 const float* noisePow, 30 const float* lambda) { 31 int i, num; 32 float rand[PART_LEN]; 33 float noise, noiseAvg, tmp, tmpAvg; 34 int16_t randW16[PART_LEN]; 35 complex_t u[PART_LEN1]; 36 37 const float pi2 = 6.28318530717959f; 38 const float pi2t = pi2 / 32768; 39 40 // Generate a uniform random array on [0 1] 41 WebRtcSpl_RandUArray(randW16, PART_LEN, &aec->seed); 42 43 int16_t* randWptr = randW16; 44 float randTemp, randTemp2, randTemp3, randTemp4; 45 int32_t tmp1s, tmp2s, tmp3s, tmp4s; 46 47 for (i = 0; i < PART_LEN; i+=4) { 48 __asm __volatile ( 49 ".set push \n\t" 50 ".set noreorder \n\t" 51 "lh %[tmp1s], 0(%[randWptr]) \n\t" 52 "lh %[tmp2s], 2(%[randWptr]) \n\t" 53 "lh %[tmp3s], 4(%[randWptr]) \n\t" 54 "lh %[tmp4s], 6(%[randWptr]) \n\t" 55 "mtc1 %[tmp1s], %[randTemp] \n\t" 56 "mtc1 %[tmp2s], %[randTemp2] \n\t" 57 "mtc1 %[tmp3s], %[randTemp3] \n\t" 58 "mtc1 %[tmp4s], %[randTemp4] \n\t" 59 "cvt.s.w %[randTemp], %[randTemp] \n\t" 60 "cvt.s.w %[randTemp2], %[randTemp2] \n\t" 61 "cvt.s.w %[randTemp3], %[randTemp3] \n\t" 62 "cvt.s.w %[randTemp4], %[randTemp4] \n\t" 63 "addiu %[randWptr], %[randWptr], 8 \n\t" 64 "mul.s %[randTemp], %[randTemp], %[pi2t] \n\t" 65 "mul.s %[randTemp2], %[randTemp2], %[pi2t] \n\t" 66 "mul.s %[randTemp3], %[randTemp3], %[pi2t] \n\t" 67 "mul.s %[randTemp4], %[randTemp4], %[pi2t] \n\t" 68 ".set pop \n\t" 69 : [randWptr] "+r" (randWptr), [randTemp] "=&f" (randTemp), 70 [randTemp2] "=&f" (randTemp2), [randTemp3] "=&f" (randTemp3), 71 [randTemp4] "=&f" (randTemp4), [tmp1s] "=&r" (tmp1s), 72 [tmp2s] "=&r" (tmp2s), [tmp3s] "=&r" (tmp3s), 73 [tmp4s] "=&r" (tmp4s) 74 : [pi2t] "f" (pi2t) 75 : "memory" 76 ); 77 78 u[i+1][0] = cosf(randTemp); 79 u[i+1][1] = sinf(randTemp); 80 u[i+2][0] = cosf(randTemp2); 81 u[i+2][1] = sinf(randTemp2); 82 u[i+3][0] = cosf(randTemp3); 83 u[i+3][1] = sinf(randTemp3); 84 u[i+4][0] = cosf(randTemp4); 85 u[i+4][1] = sinf(randTemp4); 86 } 87 88 // Reject LF noise 89 float* u_ptr = &u[1][0]; 90 float noise2, noise3, noise4; 91 float tmp1f, tmp2f, tmp3f, tmp4f, tmp5f, tmp6f, tmp7f, tmp8f; 92 93 u[0][0] = 0; 94 u[0][1] = 0; 95 for (i = 1; i < PART_LEN1; i+=4) { 96 __asm __volatile ( 97 ".set push \n\t" 98 ".set noreorder \n\t" 99 "lwc1 %[noise], 4(%[noisePow]) \n\t" 100 "lwc1 %[noise2], 8(%[noisePow]) \n\t" 101 "lwc1 %[noise3], 12(%[noisePow]) \n\t" 102 "lwc1 %[noise4], 16(%[noisePow]) \n\t" 103 "sqrt.s %[noise], %[noise] \n\t" 104 "sqrt.s %[noise2], %[noise2] \n\t" 105 "sqrt.s %[noise3], %[noise3] \n\t" 106 "sqrt.s %[noise4], %[noise4] \n\t" 107 "lwc1 %[tmp1f], 0(%[u_ptr]) \n\t" 108 "lwc1 %[tmp2f], 4(%[u_ptr]) \n\t" 109 "lwc1 %[tmp3f], 8(%[u_ptr]) \n\t" 110 "lwc1 %[tmp4f], 12(%[u_ptr]) \n\t" 111 "lwc1 %[tmp5f], 16(%[u_ptr]) \n\t" 112 "lwc1 %[tmp6f], 20(%[u_ptr]) \n\t" 113 "lwc1 %[tmp7f], 24(%[u_ptr]) \n\t" 114 "lwc1 %[tmp8f], 28(%[u_ptr]) \n\t" 115 "addiu %[noisePow], %[noisePow], 16 \n\t" 116 "mul.s %[tmp1f], %[tmp1f], %[noise] \n\t" 117 "mul.s %[tmp2f], %[tmp2f], %[noise] \n\t" 118 "mul.s %[tmp3f], %[tmp3f], %[noise2] \n\t" 119 "mul.s %[tmp4f], %[tmp4f], %[noise2] \n\t" 120 "mul.s %[tmp5f], %[tmp5f], %[noise3] \n\t" 121 "mul.s %[tmp6f], %[tmp6f], %[noise3] \n\t" 122 "swc1 %[tmp1f], 0(%[u_ptr]) \n\t" 123 "swc1 %[tmp3f], 8(%[u_ptr]) \n\t" 124 "mul.s %[tmp8f], %[tmp8f], %[noise4] \n\t" 125 "mul.s %[tmp7f], %[tmp7f], %[noise4] \n\t" 126 "neg.s %[tmp2f] \n\t" 127 "neg.s %[tmp4f] \n\t" 128 "neg.s %[tmp6f] \n\t" 129 "neg.s %[tmp8f] \n\t" 130 "swc1 %[tmp5f], 16(%[u_ptr]) \n\t" 131 "swc1 %[tmp7f], 24(%[u_ptr]) \n\t" 132 "swc1 %[tmp2f], 4(%[u_ptr]) \n\t" 133 "swc1 %[tmp4f], 12(%[u_ptr]) \n\t" 134 "swc1 %[tmp6f], 20(%[u_ptr]) \n\t" 135 "swc1 %[tmp8f], 28(%[u_ptr]) \n\t" 136 "addiu %[u_ptr], %[u_ptr], 32 \n\t" 137 ".set pop \n\t" 138 : [u_ptr] "+r" (u_ptr), [noisePow] "+r" (noisePow), 139 [noise] "=&f" (noise), [noise2] "=&f" (noise2), 140 [noise3] "=&f" (noise3), [noise4] "=&f" (noise4), 141 [tmp1f] "=&f" (tmp1f), [tmp2f] "=&f" (tmp2f), 142 [tmp3f] "=&f" (tmp3f), [tmp4f] "=&f" (tmp4f), 143 [tmp5f] "=&f" (tmp5f), [tmp6f] "=&f" (tmp6f), 144 [tmp7f] "=&f" (tmp7f), [tmp8f] "=&f" (tmp8f) 145 : 146 : "memory" 147 ); 148 } 149 u[PART_LEN][1] = 0; 150 noisePow -= PART_LEN; 151 152 u_ptr = &u[0][0]; 153 float* u_ptr_end = &u[PART_LEN][0]; 154 float* efw_ptr_0 = &efw[0][0]; 155 float* efw_ptr_1 = &efw[1][0]; 156 float tmp9f, tmp10f; 157 const float tmp1c = 1.0; 158 159 __asm __volatile ( 160 ".set push \n\t" 161 ".set noreorder \n\t" 162 "1: \n\t" 163 "lwc1 %[tmp1f], 0(%[lambda]) \n\t" 164 "lwc1 %[tmp6f], 4(%[lambda]) \n\t" 165 "addiu %[lambda], %[lambda], 8 \n\t" 166 "c.lt.s %[tmp1f], %[tmp1c] \n\t" 167 "bc1f 4f \n\t" 168 " nop \n\t" 169 "c.lt.s %[tmp6f], %[tmp1c] \n\t" 170 "bc1f 3f \n\t" 171 " nop \n\t" 172 "2: \n\t" 173 "mul.s %[tmp1f], %[tmp1f], %[tmp1f] \n\t" 174 "mul.s %[tmp6f], %[tmp6f], %[tmp6f] \n\t" 175 "sub.s %[tmp1f], %[tmp1c], %[tmp1f] \n\t" 176 "sub.s %[tmp6f], %[tmp1c], %[tmp6f] \n\t" 177 "sqrt.s %[tmp1f], %[tmp1f] \n\t" 178 "sqrt.s %[tmp6f], %[tmp6f] \n\t" 179 "lwc1 %[tmp2f], 0(%[efw_ptr_0]) \n\t" 180 "lwc1 %[tmp3f], 0(%[u_ptr]) \n\t" 181 "lwc1 %[tmp7f], 4(%[efw_ptr_0]) \n\t" 182 "lwc1 %[tmp8f], 8(%[u_ptr]) \n\t" 183 "lwc1 %[tmp4f], 0(%[efw_ptr_1]) \n\t" 184 "lwc1 %[tmp5f], 4(%[u_ptr]) \n\t" 185 "lwc1 %[tmp9f], 4(%[efw_ptr_1]) \n\t" 186 "lwc1 %[tmp10f], 12(%[u_ptr]) \n\t" 187 #if !defined(MIPS32_R2_LE) 188 "mul.s %[tmp3f], %[tmp1f], %[tmp3f] \n\t" 189 "add.s %[tmp2f], %[tmp2f], %[tmp3f] \n\t" 190 "mul.s %[tmp3f], %[tmp1f], %[tmp5f] \n\t" 191 "add.s %[tmp4f], %[tmp4f], %[tmp3f] \n\t" 192 "mul.s %[tmp3f], %[tmp6f], %[tmp8f] \n\t" 193 "add.s %[tmp7f], %[tmp7f], %[tmp3f] \n\t" 194 "mul.s %[tmp3f], %[tmp6f], %[tmp10f] \n\t" 195 "add.s %[tmp9f], %[tmp9f], %[tmp3f] \n\t" 196 #else // #if !defined(MIPS32_R2_LE) 197 "madd.s %[tmp2f], %[tmp2f], %[tmp1f], %[tmp3f] \n\t" 198 "madd.s %[tmp4f], %[tmp4f], %[tmp1f], %[tmp5f] \n\t" 199 "madd.s %[tmp7f], %[tmp7f], %[tmp6f], %[tmp8f] \n\t" 200 "madd.s %[tmp9f], %[tmp9f], %[tmp6f], %[tmp10f] \n\t" 201 #endif // #if !defined(MIPS32_R2_LE) 202 "swc1 %[tmp2f], 0(%[efw_ptr_0]) \n\t" 203 "swc1 %[tmp4f], 0(%[efw_ptr_1]) \n\t" 204 "swc1 %[tmp7f], 4(%[efw_ptr_0]) \n\t" 205 "b 5f \n\t" 206 " swc1 %[tmp9f], 4(%[efw_ptr_1]) \n\t" 207 "3: \n\t" 208 "mul.s %[tmp1f], %[tmp1f], %[tmp1f] \n\t" 209 "sub.s %[tmp1f], %[tmp1c], %[tmp1f] \n\t" 210 "sqrt.s %[tmp1f], %[tmp1f] \n\t" 211 "lwc1 %[tmp2f], 0(%[efw_ptr_0]) \n\t" 212 "lwc1 %[tmp3f], 0(%[u_ptr]) \n\t" 213 "lwc1 %[tmp4f], 0(%[efw_ptr_1]) \n\t" 214 "lwc1 %[tmp5f], 4(%[u_ptr]) \n\t" 215 #if !defined(MIPS32_R2_LE) 216 "mul.s %[tmp3f], %[tmp1f], %[tmp3f] \n\t" 217 "add.s %[tmp2f], %[tmp2f], %[tmp3f] \n\t" 218 "mul.s %[tmp3f], %[tmp1f], %[tmp5f] \n\t" 219 "add.s %[tmp4f], %[tmp4f], %[tmp3f] \n\t" 220 #else // #if !defined(MIPS32_R2_LE) 221 "madd.s %[tmp2f], %[tmp2f], %[tmp1f], %[tmp3f] \n\t" 222 "madd.s %[tmp4f], %[tmp4f], %[tmp1f], %[tmp5f] \n\t" 223 #endif // #if !defined(MIPS32_R2_LE) 224 "swc1 %[tmp2f], 0(%[efw_ptr_0]) \n\t" 225 "b 5f \n\t" 226 " swc1 %[tmp4f], 0(%[efw_ptr_1]) \n\t" 227 "4: \n\t" 228 "c.lt.s %[tmp6f], %[tmp1c] \n\t" 229 "bc1f 5f \n\t" 230 " nop \n\t" 231 "mul.s %[tmp6f], %[tmp6f], %[tmp6f] \n\t" 232 "sub.s %[tmp6f], %[tmp1c], %[tmp6f] \n\t" 233 "sqrt.s %[tmp6f], %[tmp6f] \n\t" 234 "lwc1 %[tmp7f], 4(%[efw_ptr_0]) \n\t" 235 "lwc1 %[tmp8f], 8(%[u_ptr]) \n\t" 236 "lwc1 %[tmp9f], 4(%[efw_ptr_1]) \n\t" 237 "lwc1 %[tmp10f], 12(%[u_ptr]) \n\t" 238 #if !defined(MIPS32_R2_LE) 239 "mul.s %[tmp3f], %[tmp6f], %[tmp8f] \n\t" 240 "add.s %[tmp7f], %[tmp7f], %[tmp3f] \n\t" 241 "mul.s %[tmp3f], %[tmp6f], %[tmp10f] \n\t" 242 "add.s %[tmp9f], %[tmp9f], %[tmp3f] \n\t" 243 #else // #if !defined(MIPS32_R2_LE) 244 "madd.s %[tmp7f], %[tmp7f], %[tmp6f], %[tmp8f] \n\t" 245 "madd.s %[tmp9f], %[tmp9f], %[tmp6f], %[tmp10f] \n\t" 246 #endif // #if !defined(MIPS32_R2_LE) 247 "swc1 %[tmp7f], 4(%[efw_ptr_0]) \n\t" 248 "swc1 %[tmp9f], 4(%[efw_ptr_1]) \n\t" 249 "5: \n\t" 250 "addiu %[u_ptr], %[u_ptr], 16 \n\t" 251 "addiu %[efw_ptr_0], %[efw_ptr_0], 8 \n\t" 252 "bne %[u_ptr], %[u_ptr_end], 1b \n\t" 253 " addiu %[efw_ptr_1], %[efw_ptr_1], 8 \n\t" 254 ".set pop \n\t" 255 : [lambda] "+r" (lambda), [u_ptr] "+r" (u_ptr), 256 [efw_ptr_0] "+r" (efw_ptr_0), [efw_ptr_1] "+r" (efw_ptr_1), 257 [tmp1f] "=&f" (tmp1f), [tmp2f] "=&f" (tmp2f), [tmp3f] "=&f" (tmp3f), 258 [tmp4f] "=&f" (tmp4f), [tmp5f] "=&f" (tmp5f), 259 [tmp6f] "=&f" (tmp6f), [tmp7f] "=&f" (tmp7f), [tmp8f] "=&f" (tmp8f), 260 [tmp9f] "=&f" (tmp9f), [tmp10f] "=&f" (tmp10f) 261 : [tmp1c] "f" (tmp1c), [u_ptr_end] "r" (u_ptr_end) 262 : "memory" 263 ); 264 265 lambda -= PART_LEN; 266 tmp = sqrtf(WEBRTC_SPL_MAX(1 - lambda[PART_LEN] * lambda[PART_LEN], 0)); 267 //tmp = 1 - lambda[i]; 268 efw[0][PART_LEN] += tmp * u[PART_LEN][0]; 269 efw[1][PART_LEN] += tmp * u[PART_LEN][1]; 270 271 // For H band comfort noise 272 // TODO: don't compute noise and "tmp" twice. Use the previous results. 273 noiseAvg = 0.0; 274 tmpAvg = 0.0; 275 num = 0; 276 if (aec->num_bands > 1) { 277 for (i = 0; i < PART_LEN; i++) { 278 rand[i] = ((float)randW16[i]) / 32768; 279 } 280 281 // average noise scale 282 // average over second half of freq spectrum (i.e., 4->8khz) 283 // TODO: we shouldn't need num. We know how many elements we're summing. 284 for (i = PART_LEN1 >> 1; i < PART_LEN1; i++) { 285 num++; 286 noiseAvg += sqrtf(noisePow[i]); 287 } 288 noiseAvg /= (float)num; 289 290 // average nlp scale 291 // average over second half of freq spectrum (i.e., 4->8khz) 292 // TODO: we shouldn't need num. We know how many elements we're summing. 293 num = 0; 294 for (i = PART_LEN1 >> 1; i < PART_LEN1; i++) { 295 num++; 296 tmpAvg += sqrtf(WEBRTC_SPL_MAX(1 - lambda[i] * lambda[i], 0)); 297 } 298 tmpAvg /= (float)num; 299 300 // Use average noise for H band 301 // TODO: we should probably have a new random vector here. 302 // Reject LF noise 303 u[0][0] = 0; 304 u[0][1] = 0; 305 for (i = 1; i < PART_LEN1; i++) { 306 tmp = pi2 * rand[i - 1]; 307 308 // Use average noise for H band 309 u[i][0] = noiseAvg * (float)cos(tmp); 310 u[i][1] = -noiseAvg * (float)sin(tmp); 311 } 312 u[PART_LEN][1] = 0; 313 314 for (i = 0; i < PART_LEN1; i++) { 315 // Use average NLP weight for H band 316 comfortNoiseHband[0][i] = tmpAvg * u[i][0]; 317 comfortNoiseHband[1][i] = tmpAvg * u[i][1]; 318 } 319 } else { 320 memset(comfortNoiseHband, 0, 321 2 * PART_LEN1 * sizeof(comfortNoiseHband[0][0])); 322 } 323 } 324 325 void WebRtcAec_FilterFar_mips( 326 int num_partitions, 327 int x_fft_buf_block_pos, 328 float x_fft_buf[2][kExtendedNumPartitions * PART_LEN1], 329 float h_fft_buf[2][kExtendedNumPartitions * PART_LEN1], 330 float y_fft[2][PART_LEN1]) { 331 int i; 332 for (i = 0; i < num_partitions; i++) { 333 int xPos = (i + x_fft_buf_block_pos) * PART_LEN1; 334 int pos = i * PART_LEN1; 335 // Check for wrap 336 if (i + x_fft_buf_block_pos >= num_partitions) { 337 xPos -= num_partitions * (PART_LEN1); 338 } 339 float* yf0 = y_fft[0]; 340 float* yf1 = y_fft[1]; 341 float* aRe = x_fft_buf[0] + xPos; 342 float* aIm = x_fft_buf[1] + xPos; 343 float* bRe = h_fft_buf[0] + pos; 344 float* bIm = h_fft_buf[1] + pos; 345 float f0, f1, f2, f3, f4, f5, f6, f7, f8, f9, f10, f11, f12, f13; 346 int len = PART_LEN1 >> 1; 347 348 __asm __volatile ( 349 ".set push \n\t" 350 ".set noreorder \n\t" 351 "1: \n\t" 352 "lwc1 %[f0], 0(%[aRe]) \n\t" 353 "lwc1 %[f1], 0(%[bRe]) \n\t" 354 "lwc1 %[f2], 0(%[bIm]) \n\t" 355 "lwc1 %[f3], 0(%[aIm]) \n\t" 356 "lwc1 %[f4], 4(%[aRe]) \n\t" 357 "lwc1 %[f5], 4(%[bRe]) \n\t" 358 "lwc1 %[f6], 4(%[bIm]) \n\t" 359 "mul.s %[f8], %[f0], %[f1] \n\t" 360 "mul.s %[f0], %[f0], %[f2] \n\t" 361 "mul.s %[f9], %[f4], %[f5] \n\t" 362 "mul.s %[f4], %[f4], %[f6] \n\t" 363 "lwc1 %[f7], 4(%[aIm]) \n\t" 364 #if !defined(MIPS32_R2_LE) 365 "mul.s %[f12], %[f2], %[f3] \n\t" 366 "mul.s %[f1], %[f3], %[f1] \n\t" 367 "mul.s %[f11], %[f6], %[f7] \n\t" 368 "addiu %[aRe], %[aRe], 8 \n\t" 369 "addiu %[aIm], %[aIm], 8 \n\t" 370 "addiu %[len], %[len], -1 \n\t" 371 "sub.s %[f8], %[f8], %[f12] \n\t" 372 "mul.s %[f12], %[f7], %[f5] \n\t" 373 "lwc1 %[f2], 0(%[yf0]) \n\t" 374 "add.s %[f1], %[f0], %[f1] \n\t" 375 "lwc1 %[f3], 0(%[yf1]) \n\t" 376 "sub.s %[f9], %[f9], %[f11] \n\t" 377 "lwc1 %[f6], 4(%[yf0]) \n\t" 378 "add.s %[f4], %[f4], %[f12] \n\t" 379 #else // #if !defined(MIPS32_R2_LE) 380 "addiu %[aRe], %[aRe], 8 \n\t" 381 "addiu %[aIm], %[aIm], 8 \n\t" 382 "addiu %[len], %[len], -1 \n\t" 383 "nmsub.s %[f8], %[f8], %[f2], %[f3] \n\t" 384 "lwc1 %[f2], 0(%[yf0]) \n\t" 385 "madd.s %[f1], %[f0], %[f3], %[f1] \n\t" 386 "lwc1 %[f3], 0(%[yf1]) \n\t" 387 "nmsub.s %[f9], %[f9], %[f6], %[f7] \n\t" 388 "lwc1 %[f6], 4(%[yf0]) \n\t" 389 "madd.s %[f4], %[f4], %[f7], %[f5] \n\t" 390 #endif // #if !defined(MIPS32_R2_LE) 391 "lwc1 %[f5], 4(%[yf1]) \n\t" 392 "add.s %[f2], %[f2], %[f8] \n\t" 393 "addiu %[bRe], %[bRe], 8 \n\t" 394 "addiu %[bIm], %[bIm], 8 \n\t" 395 "add.s %[f3], %[f3], %[f1] \n\t" 396 "add.s %[f6], %[f6], %[f9] \n\t" 397 "add.s %[f5], %[f5], %[f4] \n\t" 398 "swc1 %[f2], 0(%[yf0]) \n\t" 399 "swc1 %[f3], 0(%[yf1]) \n\t" 400 "swc1 %[f6], 4(%[yf0]) \n\t" 401 "swc1 %[f5], 4(%[yf1]) \n\t" 402 "addiu %[yf0], %[yf0], 8 \n\t" 403 "bgtz %[len], 1b \n\t" 404 " addiu %[yf1], %[yf1], 8 \n\t" 405 "lwc1 %[f0], 0(%[aRe]) \n\t" 406 "lwc1 %[f1], 0(%[bRe]) \n\t" 407 "lwc1 %[f2], 0(%[bIm]) \n\t" 408 "lwc1 %[f3], 0(%[aIm]) \n\t" 409 "mul.s %[f8], %[f0], %[f1] \n\t" 410 "mul.s %[f0], %[f0], %[f2] \n\t" 411 #if !defined(MIPS32_R2_LE) 412 "mul.s %[f12], %[f2], %[f3] \n\t" 413 "mul.s %[f1], %[f3], %[f1] \n\t" 414 "sub.s %[f8], %[f8], %[f12] \n\t" 415 "lwc1 %[f2], 0(%[yf0]) \n\t" 416 "add.s %[f1], %[f0], %[f1] \n\t" 417 "lwc1 %[f3], 0(%[yf1]) \n\t" 418 #else // #if !defined(MIPS32_R2_LE) 419 "nmsub.s %[f8], %[f8], %[f2], %[f3] \n\t" 420 "lwc1 %[f2], 0(%[yf0]) \n\t" 421 "madd.s %[f1], %[f0], %[f3], %[f1] \n\t" 422 "lwc1 %[f3], 0(%[yf1]) \n\t" 423 #endif // #if !defined(MIPS32_R2_LE) 424 "add.s %[f2], %[f2], %[f8] \n\t" 425 "add.s %[f3], %[f3], %[f1] \n\t" 426 "swc1 %[f2], 0(%[yf0]) \n\t" 427 "swc1 %[f3], 0(%[yf1]) \n\t" 428 ".set pop \n\t" 429 : [f0] "=&f" (f0), [f1] "=&f" (f1), [f2] "=&f" (f2), 430 [f3] "=&f" (f3), [f4] "=&f" (f4), [f5] "=&f" (f5), 431 [f6] "=&f" (f6), [f7] "=&f" (f7), [f8] "=&f" (f8), 432 [f9] "=&f" (f9), [f10] "=&f" (f10), [f11] "=&f" (f11), 433 [f12] "=&f" (f12), [f13] "=&f" (f13), [aRe] "+r" (aRe), 434 [aIm] "+r" (aIm), [bRe] "+r" (bRe), [bIm] "+r" (bIm), 435 [yf0] "+r" (yf0), [yf1] "+r" (yf1), [len] "+r" (len) 436 : 437 : "memory" 438 ); 439 } 440 } 441 442 void WebRtcAec_FilterAdaptation_mips( 443 int num_partitions, 444 int x_fft_buf_block_pos, 445 float x_fft_buf[2][kExtendedNumPartitions * PART_LEN1], 446 float e_fft[2][PART_LEN1], 447 float h_fft_buf[2][kExtendedNumPartitions * PART_LEN1]) { 448 float fft[PART_LEN2]; 449 int i; 450 for (i = 0; i < num_partitions; i++) { 451 int xPos = (i + x_fft_buf_block_pos)*(PART_LEN1); 452 int pos; 453 // Check for wrap 454 if (i + x_fft_buf_block_pos >= num_partitions) { 455 xPos -= num_partitions * PART_LEN1; 456 } 457 458 pos = i * PART_LEN1; 459 float* aRe = x_fft_buf[0] + xPos; 460 float* aIm = x_fft_buf[1] + xPos; 461 float* bRe = e_fft[0]; 462 float* bIm = e_fft[1]; 463 float* fft_tmp; 464 465 float f0, f1, f2, f3, f4, f5, f6 ,f7, f8, f9, f10, f11, f12; 466 int len = PART_LEN >> 1; 467 468 __asm __volatile ( 469 ".set push \n\t" 470 ".set noreorder \n\t" 471 "addiu %[fft_tmp], %[fft], 0 \n\t" 472 "1: \n\t" 473 "lwc1 %[f0], 0(%[aRe]) \n\t" 474 "lwc1 %[f1], 0(%[bRe]) \n\t" 475 "lwc1 %[f2], 0(%[bIm]) \n\t" 476 "lwc1 %[f4], 4(%[aRe]) \n\t" 477 "lwc1 %[f5], 4(%[bRe]) \n\t" 478 "lwc1 %[f6], 4(%[bIm]) \n\t" 479 "addiu %[aRe], %[aRe], 8 \n\t" 480 "addiu %[bRe], %[bRe], 8 \n\t" 481 "mul.s %[f8], %[f0], %[f1] \n\t" 482 "mul.s %[f0], %[f0], %[f2] \n\t" 483 "lwc1 %[f3], 0(%[aIm]) \n\t" 484 "mul.s %[f9], %[f4], %[f5] \n\t" 485 "lwc1 %[f7], 4(%[aIm]) \n\t" 486 "mul.s %[f4], %[f4], %[f6] \n\t" 487 #if !defined(MIPS32_R2_LE) 488 "mul.s %[f10], %[f3], %[f2] \n\t" 489 "mul.s %[f1], %[f3], %[f1] \n\t" 490 "mul.s %[f11], %[f7], %[f6] \n\t" 491 "mul.s %[f5], %[f7], %[f5] \n\t" 492 "addiu %[aIm], %[aIm], 8 \n\t" 493 "addiu %[bIm], %[bIm], 8 \n\t" 494 "addiu %[len], %[len], -1 \n\t" 495 "add.s %[f8], %[f8], %[f10] \n\t" 496 "sub.s %[f1], %[f0], %[f1] \n\t" 497 "add.s %[f9], %[f9], %[f11] \n\t" 498 "sub.s %[f5], %[f4], %[f5] \n\t" 499 #else // #if !defined(MIPS32_R2_LE) 500 "addiu %[aIm], %[aIm], 8 \n\t" 501 "addiu %[bIm], %[bIm], 8 \n\t" 502 "addiu %[len], %[len], -1 \n\t" 503 "madd.s %[f8], %[f8], %[f3], %[f2] \n\t" 504 "nmsub.s %[f1], %[f0], %[f3], %[f1] \n\t" 505 "madd.s %[f9], %[f9], %[f7], %[f6] \n\t" 506 "nmsub.s %[f5], %[f4], %[f7], %[f5] \n\t" 507 #endif // #if !defined(MIPS32_R2_LE) 508 "swc1 %[f8], 0(%[fft_tmp]) \n\t" 509 "swc1 %[f1], 4(%[fft_tmp]) \n\t" 510 "swc1 %[f9], 8(%[fft_tmp]) \n\t" 511 "swc1 %[f5], 12(%[fft_tmp]) \n\t" 512 "bgtz %[len], 1b \n\t" 513 " addiu %[fft_tmp], %[fft_tmp], 16 \n\t" 514 "lwc1 %[f0], 0(%[aRe]) \n\t" 515 "lwc1 %[f1], 0(%[bRe]) \n\t" 516 "lwc1 %[f2], 0(%[bIm]) \n\t" 517 "lwc1 %[f3], 0(%[aIm]) \n\t" 518 "mul.s %[f8], %[f0], %[f1] \n\t" 519 #if !defined(MIPS32_R2_LE) 520 "mul.s %[f10], %[f3], %[f2] \n\t" 521 "add.s %[f8], %[f8], %[f10] \n\t" 522 #else // #if !defined(MIPS32_R2_LE) 523 "madd.s %[f8], %[f8], %[f3], %[f2] \n\t" 524 #endif // #if !defined(MIPS32_R2_LE) 525 "swc1 %[f8], 4(%[fft]) \n\t" 526 ".set pop \n\t" 527 : [f0] "=&f" (f0), [f1] "=&f" (f1), [f2] "=&f" (f2), 528 [f3] "=&f" (f3), [f4] "=&f" (f4), [f5] "=&f" (f5), 529 [f6] "=&f" (f6), [f7] "=&f" (f7), [f8] "=&f" (f8), 530 [f9] "=&f" (f9), [f10] "=&f" (f10), [f11] "=&f" (f11), 531 [f12] "=&f" (f12), [aRe] "+r" (aRe), [aIm] "+r" (aIm), 532 [bRe] "+r" (bRe), [bIm] "+r" (bIm), [fft_tmp] "=&r" (fft_tmp), 533 [len] "+r" (len) 534 : [fft] "r" (fft) 535 : "memory" 536 ); 537 538 aec_rdft_inverse_128(fft); 539 memset(fft + PART_LEN, 0, sizeof(float) * PART_LEN); 540 541 // fft scaling 542 { 543 float scale = 2.0f / PART_LEN2; 544 __asm __volatile ( 545 ".set push \n\t" 546 ".set noreorder \n\t" 547 "addiu %[fft_tmp], %[fft], 0 \n\t" 548 "addiu %[len], $zero, 8 \n\t" 549 "1: \n\t" 550 "addiu %[len], %[len], -1 \n\t" 551 "lwc1 %[f0], 0(%[fft_tmp]) \n\t" 552 "lwc1 %[f1], 4(%[fft_tmp]) \n\t" 553 "lwc1 %[f2], 8(%[fft_tmp]) \n\t" 554 "lwc1 %[f3], 12(%[fft_tmp]) \n\t" 555 "mul.s %[f0], %[f0], %[scale] \n\t" 556 "mul.s %[f1], %[f1], %[scale] \n\t" 557 "mul.s %[f2], %[f2], %[scale] \n\t" 558 "mul.s %[f3], %[f3], %[scale] \n\t" 559 "lwc1 %[f4], 16(%[fft_tmp]) \n\t" 560 "lwc1 %[f5], 20(%[fft_tmp]) \n\t" 561 "lwc1 %[f6], 24(%[fft_tmp]) \n\t" 562 "lwc1 %[f7], 28(%[fft_tmp]) \n\t" 563 "mul.s %[f4], %[f4], %[scale] \n\t" 564 "mul.s %[f5], %[f5], %[scale] \n\t" 565 "mul.s %[f6], %[f6], %[scale] \n\t" 566 "mul.s %[f7], %[f7], %[scale] \n\t" 567 "swc1 %[f0], 0(%[fft_tmp]) \n\t" 568 "swc1 %[f1], 4(%[fft_tmp]) \n\t" 569 "swc1 %[f2], 8(%[fft_tmp]) \n\t" 570 "swc1 %[f3], 12(%[fft_tmp]) \n\t" 571 "swc1 %[f4], 16(%[fft_tmp]) \n\t" 572 "swc1 %[f5], 20(%[fft_tmp]) \n\t" 573 "swc1 %[f6], 24(%[fft_tmp]) \n\t" 574 "swc1 %[f7], 28(%[fft_tmp]) \n\t" 575 "bgtz %[len], 1b \n\t" 576 " addiu %[fft_tmp], %[fft_tmp], 32 \n\t" 577 ".set pop \n\t" 578 : [f0] "=&f" (f0), [f1] "=&f" (f1), [f2] "=&f" (f2), 579 [f3] "=&f" (f3), [f4] "=&f" (f4), [f5] "=&f" (f5), 580 [f6] "=&f" (f6), [f7] "=&f" (f7), [len] "=&r" (len), 581 [fft_tmp] "=&r" (fft_tmp) 582 : [scale] "f" (scale), [fft] "r" (fft) 583 : "memory" 584 ); 585 } 586 aec_rdft_forward_128(fft); 587 aRe = h_fft_buf[0] + pos; 588 aIm = h_fft_buf[1] + pos; 589 __asm __volatile ( 590 ".set push \n\t" 591 ".set noreorder \n\t" 592 "addiu %[fft_tmp], %[fft], 0 \n\t" 593 "addiu %[len], $zero, 31 \n\t" 594 "lwc1 %[f0], 0(%[aRe]) \n\t" 595 "lwc1 %[f1], 0(%[fft_tmp]) \n\t" 596 "lwc1 %[f2], 256(%[aRe]) \n\t" 597 "lwc1 %[f3], 4(%[fft_tmp]) \n\t" 598 "lwc1 %[f4], 4(%[aRe]) \n\t" 599 "lwc1 %[f5], 8(%[fft_tmp]) \n\t" 600 "lwc1 %[f6], 4(%[aIm]) \n\t" 601 "lwc1 %[f7], 12(%[fft_tmp]) \n\t" 602 "add.s %[f0], %[f0], %[f1] \n\t" 603 "add.s %[f2], %[f2], %[f3] \n\t" 604 "add.s %[f4], %[f4], %[f5] \n\t" 605 "add.s %[f6], %[f6], %[f7] \n\t" 606 "addiu %[fft_tmp], %[fft_tmp], 16 \n\t" 607 "swc1 %[f0], 0(%[aRe]) \n\t" 608 "swc1 %[f2], 256(%[aRe]) \n\t" 609 "swc1 %[f4], 4(%[aRe]) \n\t" 610 "addiu %[aRe], %[aRe], 8 \n\t" 611 "swc1 %[f6], 4(%[aIm]) \n\t" 612 "addiu %[aIm], %[aIm], 8 \n\t" 613 "1: \n\t" 614 "lwc1 %[f0], 0(%[aRe]) \n\t" 615 "lwc1 %[f1], 0(%[fft_tmp]) \n\t" 616 "lwc1 %[f2], 0(%[aIm]) \n\t" 617 "lwc1 %[f3], 4(%[fft_tmp]) \n\t" 618 "lwc1 %[f4], 4(%[aRe]) \n\t" 619 "lwc1 %[f5], 8(%[fft_tmp]) \n\t" 620 "lwc1 %[f6], 4(%[aIm]) \n\t" 621 "lwc1 %[f7], 12(%[fft_tmp]) \n\t" 622 "add.s %[f0], %[f0], %[f1] \n\t" 623 "add.s %[f2], %[f2], %[f3] \n\t" 624 "add.s %[f4], %[f4], %[f5] \n\t" 625 "add.s %[f6], %[f6], %[f7] \n\t" 626 "addiu %[len], %[len], -1 \n\t" 627 "addiu %[fft_tmp], %[fft_tmp], 16 \n\t" 628 "swc1 %[f0], 0(%[aRe]) \n\t" 629 "swc1 %[f2], 0(%[aIm]) \n\t" 630 "swc1 %[f4], 4(%[aRe]) \n\t" 631 "addiu %[aRe], %[aRe], 8 \n\t" 632 "swc1 %[f6], 4(%[aIm]) \n\t" 633 "bgtz %[len], 1b \n\t" 634 " addiu %[aIm], %[aIm], 8 \n\t" 635 ".set pop \n\t" 636 : [f0] "=&f" (f0), [f1] "=&f" (f1), [f2] "=&f" (f2), 637 [f3] "=&f" (f3), [f4] "=&f" (f4), [f5] "=&f" (f5), 638 [f6] "=&f" (f6), [f7] "=&f" (f7), [len] "=&r" (len), 639 [fft_tmp] "=&r" (fft_tmp), [aRe] "+r" (aRe), [aIm] "+r" (aIm) 640 : [fft] "r" (fft) 641 : "memory" 642 ); 643 } 644 } 645 646 void WebRtcAec_OverdriveAndSuppress_mips(AecCore* aec, 647 float hNl[PART_LEN1], 648 const float hNlFb, 649 float efw[2][PART_LEN1]) { 650 int i; 651 const float one = 1.0; 652 float* p_hNl; 653 float* p_efw0; 654 float* p_efw1; 655 float* p_WebRtcAec_wC; 656 float temp1, temp2, temp3, temp4; 657 658 p_hNl = &hNl[0]; 659 p_efw0 = &efw[0][0]; 660 p_efw1 = &efw[1][0]; 661 p_WebRtcAec_wC = (float*)&WebRtcAec_weightCurve[0]; 662 663 for (i = 0; i < PART_LEN1; i++) { 664 // Weight subbands 665 __asm __volatile ( 666 ".set push \n\t" 667 ".set noreorder \n\t" 668 "lwc1 %[temp1], 0(%[p_hNl]) \n\t" 669 "lwc1 %[temp2], 0(%[p_wC]) \n\t" 670 "c.lt.s %[hNlFb], %[temp1] \n\t" 671 "bc1f 1f \n\t" 672 " mul.s %[temp3], %[temp2], %[hNlFb] \n\t" 673 "sub.s %[temp4], %[one], %[temp2] \n\t" 674 #if !defined(MIPS32_R2_LE) 675 "mul.s %[temp1], %[temp1], %[temp4] \n\t" 676 "add.s %[temp1], %[temp3], %[temp1] \n\t" 677 #else // #if !defined(MIPS32_R2_LE) 678 "madd.s %[temp1], %[temp3], %[temp1], %[temp4] \n\t" 679 #endif // #if !defined(MIPS32_R2_LE) 680 "swc1 %[temp1], 0(%[p_hNl]) \n\t" 681 "1: \n\t" 682 "addiu %[p_wC], %[p_wC], 4 \n\t" 683 ".set pop \n\t" 684 : [temp1] "=&f" (temp1), [temp2] "=&f" (temp2), [temp3] "=&f" (temp3), 685 [temp4] "=&f" (temp4), [p_wC] "+r" (p_WebRtcAec_wC) 686 : [hNlFb] "f" (hNlFb), [one] "f" (one), [p_hNl] "r" (p_hNl) 687 : "memory" 688 ); 689 690 hNl[i] = powf(hNl[i], aec->overDriveSm * WebRtcAec_overDriveCurve[i]); 691 692 __asm __volatile ( 693 "lwc1 %[temp1], 0(%[p_hNl]) \n\t" 694 "lwc1 %[temp3], 0(%[p_efw1]) \n\t" 695 "lwc1 %[temp2], 0(%[p_efw0]) \n\t" 696 "addiu %[p_hNl], %[p_hNl], 4 \n\t" 697 "mul.s %[temp3], %[temp3], %[temp1] \n\t" 698 "mul.s %[temp2], %[temp2], %[temp1] \n\t" 699 "addiu %[p_efw0], %[p_efw0], 4 \n\t" 700 "addiu %[p_efw1], %[p_efw1], 4 \n\t" 701 "neg.s %[temp4], %[temp3] \n\t" 702 "swc1 %[temp2], -4(%[p_efw0]) \n\t" 703 "swc1 %[temp4], -4(%[p_efw1]) \n\t" 704 : [temp1] "=&f" (temp1), [temp2] "=&f" (temp2), [temp3] "=&f" (temp3), 705 [temp4] "=&f" (temp4), [p_efw0] "+r" (p_efw0), [p_efw1] "+r" (p_efw1), 706 [p_hNl] "+r" (p_hNl) 707 : 708 : "memory" 709 ); 710 } 711 } 712 713 void WebRtcAec_ScaleErrorSignal_mips(int extended_filter_enabled, 714 float normal_mu, 715 float normal_error_threshold, 716 float x_pow[PART_LEN1], 717 float ef[2][PART_LEN1]) { 718 const float mu = extended_filter_enabled ? kExtendedMu : normal_mu; 719 const float error_threshold = extended_filter_enabled 720 ? kExtendedErrorThreshold 721 : normal_error_threshold; 722 int len = (PART_LEN1); 723 float* ef0 = ef[0]; 724 float* ef1 = ef[1]; 725 float fac1 = 1e-10f; 726 float err_th2 = error_threshold * error_threshold; 727 float f0, f1, f2; 728 #if !defined(MIPS32_R2_LE) 729 float f3; 730 #endif 731 732 __asm __volatile ( 733 ".set push \n\t" 734 ".set noreorder \n\t" 735 "1: \n\t" 736 "lwc1 %[f0], 0(%[x_pow]) \n\t" 737 "lwc1 %[f1], 0(%[ef0]) \n\t" 738 "lwc1 %[f2], 0(%[ef1]) \n\t" 739 "add.s %[f0], %[f0], %[fac1] \n\t" 740 "div.s %[f1], %[f1], %[f0] \n\t" 741 "div.s %[f2], %[f2], %[f0] \n\t" 742 "mul.s %[f0], %[f1], %[f1] \n\t" 743 #if defined(MIPS32_R2_LE) 744 "madd.s %[f0], %[f0], %[f2], %[f2] \n\t" 745 #else 746 "mul.s %[f3], %[f2], %[f2] \n\t" 747 "add.s %[f0], %[f0], %[f3] \n\t" 748 #endif 749 "c.le.s %[f0], %[err_th2] \n\t" 750 "nop \n\t" 751 "bc1t 2f \n\t" 752 " nop \n\t" 753 "sqrt.s %[f0], %[f0] \n\t" 754 "add.s %[f0], %[f0], %[fac1] \n\t" 755 "div.s %[f0], %[err_th], %[f0] \n\t" 756 "mul.s %[f1], %[f1], %[f0] \n\t" 757 "mul.s %[f2], %[f2], %[f0] \n\t" 758 "2: \n\t" 759 "mul.s %[f1], %[f1], %[mu] \n\t" 760 "mul.s %[f2], %[f2], %[mu] \n\t" 761 "swc1 %[f1], 0(%[ef0]) \n\t" 762 "swc1 %[f2], 0(%[ef1]) \n\t" 763 "addiu %[len], %[len], -1 \n\t" 764 "addiu %[x_pow], %[x_pow], 4 \n\t" 765 "addiu %[ef0], %[ef0], 4 \n\t" 766 "bgtz %[len], 1b \n\t" 767 " addiu %[ef1], %[ef1], 4 \n\t" 768 ".set pop \n\t" 769 : [f0] "=&f" (f0), [f1] "=&f" (f1), [f2] "=&f" (f2), 770 #if !defined(MIPS32_R2_LE) 771 [f3] "=&f" (f3), 772 #endif 773 [x_pow] "+r" (x_pow), [ef0] "+r" (ef0), [ef1] "+r" (ef1), 774 [len] "+r" (len) 775 : [fac1] "f" (fac1), [err_th2] "f" (err_th2), [mu] "f" (mu), 776 [err_th] "f" (error_threshold) 777 : "memory" 778 ); 779 } 780 781 void WebRtcAec_InitAec_mips(void) { 782 WebRtcAec_FilterFar = WebRtcAec_FilterFar_mips; 783 WebRtcAec_FilterAdaptation = WebRtcAec_FilterAdaptation_mips; 784 WebRtcAec_ScaleErrorSignal = WebRtcAec_ScaleErrorSignal_mips; 785 WebRtcAec_ComfortNoise = WebRtcAec_ComfortNoise_mips; 786 WebRtcAec_OverdriveAndSuppress = WebRtcAec_OverdriveAndSuppress_mips; 787 } 788
