1/* 2 * AC-3 DSP functions 3 * Copyright (c) 2011 Justin Ruggles 4 * 5 * This file is part of FFmpeg. 6 * 7 * FFmpeg is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU Lesser General Public 9 * License as published by the Free Software Foundation; either 10 * version 2.1 of the License, or (at your option) any later version. 11 * 12 * FFmpeg is distributed in the hope that it will be useful, 13 * but WITHOUT ANY WARRANTY; without even the implied warranty of 14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 15 * Lesser General Public License for more details. 16 * 17 * You should have received a copy of the GNU Lesser General Public 18 * License along with FFmpeg; if not, write to the Free Software 19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 20 */ 21 22#include <math.h> 23#include <stdlib.h> 24#include <string.h> 25 26#include "config.h" 27#include "libavutil/attributes.h" 28#include "libavutil/common.h" 29#include "libavutil/intmath.h" 30#include "libavutil/mem_internal.h" 31 32#include "ac3defs.h" 33#include "ac3dsp.h" 34#include "ac3tab.h" 35#include "mathops.h" 36 37static void ac3_exponent_min_c(uint8_t *exp, int num_reuse_blocks, int nb_coefs) 38{ 39 int blk, i; 40 41 if (!num_reuse_blocks) 42 return; 43 44 for (i = 0; i < nb_coefs; i++) { 45 uint8_t min_exp = *exp; 46 uint8_t *exp1 = exp + 256; 47 for (blk = 0; blk < num_reuse_blocks; blk++) { 48 uint8_t next_exp = *exp1; 49 if (next_exp < min_exp) 50 min_exp = next_exp; 51 exp1 += 256; 52 } 53 *exp++ = min_exp; 54 } 55} 56 57static void float_to_fixed24_c(int32_t *dst, const float *src, unsigned int len) 58{ 59 const float scale = 1 << 24; 60 do { 61 *dst++ = lrintf(*src++ * scale); 62 *dst++ = lrintf(*src++ * scale); 63 *dst++ = lrintf(*src++ * scale); 64 *dst++ = lrintf(*src++ * scale); 65 *dst++ = lrintf(*src++ * scale); 66 *dst++ = lrintf(*src++ * scale); 67 *dst++ = lrintf(*src++ * scale); 68 *dst++ = lrintf(*src++ * scale); 69 len -= 8; 70 } while (len > 0); 71} 72 73static void ac3_bit_alloc_calc_bap_c(int16_t *mask, int16_t *psd, 74 int start, int end, 75 int snr_offset, int floor, 76 const uint8_t *bap_tab, uint8_t *bap) 77{ 78 int bin, band, band_end; 79 80 /* special case, if snr offset is -960, set all bap's to zero */ 81 if (snr_offset == -960) { 82 memset(bap, 0, AC3_MAX_COEFS); 83 return; 84 } 85 86 bin = start; 87 band = ff_ac3_bin_to_band_tab[start]; 88 do { 89 int m = (FFMAX(mask[band] - snr_offset - floor, 0) & 0x1FE0) + floor; 90 band_end = ff_ac3_band_start_tab[++band]; 91 band_end = FFMIN(band_end, end); 92 93 for (; bin < band_end; bin++) { 94 int address = av_clip_uintp2((psd[bin] - m) >> 5, 6); 95 bap[bin] = bap_tab[address]; 96 } 97 } while (end > band_end); 98} 99 100static void ac3_update_bap_counts_c(uint16_t mant_cnt[16], uint8_t *bap, 101 int len) 102{ 103 while (len-- > 0) 104 mant_cnt[bap[len]]++; 105} 106 107DECLARE_ALIGNED(16, const uint16_t, ff_ac3_bap_bits)[16] = { 108 0, 0, 0, 3, 0, 4, 5, 6, 7, 8, 9, 10, 11, 12, 14, 16 109}; 110 111static int ac3_compute_mantissa_size_c(uint16_t mant_cnt[6][16]) 112{ 113 int blk, bap; 114 int bits = 0; 115 116 for (blk = 0; blk < AC3_MAX_BLOCKS; blk++) { 117 // bap=1 : 3 mantissas in 5 bits 118 bits += (mant_cnt[blk][1] / 3) * 5; 119 // bap=2 : 3 mantissas in 7 bits 120 // bap=4 : 2 mantissas in 7 bits 121 bits += ((mant_cnt[blk][2] / 3) + (mant_cnt[blk][4] >> 1)) * 7; 122 // bap=3 : 1 mantissa in 3 bits 123 bits += mant_cnt[blk][3] * 3; 124 // bap=5 to 15 : get bits per mantissa from table 125 for (bap = 5; bap < 16; bap++) 126 bits += mant_cnt[blk][bap] * ff_ac3_bap_bits[bap]; 127 } 128 return bits; 129} 130 131static void ac3_extract_exponents_c(uint8_t *exp, int32_t *coef, int nb_coefs) 132{ 133 int i; 134 135 for (i = 0; i < nb_coefs; i++) { 136 int v = abs(coef[i]); 137 exp[i] = v ? 23 - av_log2(v) : 24; 138 } 139} 140 141static void ac3_sum_square_butterfly_int32_c(int64_t sum[4], 142 const int32_t *coef0, 143 const int32_t *coef1, 144 int len) 145{ 146 int i; 147 148 sum[0] = sum[1] = sum[2] = sum[3] = 0; 149 150 for (i = 0; i < len; i++) { 151 int lt = coef0[i]; 152 int rt = coef1[i]; 153 int md = lt + rt; 154 int sd = lt - rt; 155 MAC64(sum[0], lt, lt); 156 MAC64(sum[1], rt, rt); 157 MAC64(sum[2], md, md); 158 MAC64(sum[3], sd, sd); 159 } 160} 161 162static void ac3_sum_square_butterfly_float_c(float sum[4], 163 const float *coef0, 164 const float *coef1, 165 int len) 166{ 167 int i; 168 169 sum[0] = sum[1] = sum[2] = sum[3] = 0; 170 171 for (i = 0; i < len; i++) { 172 float lt = coef0[i]; 173 float rt = coef1[i]; 174 float md = lt + rt; 175 float sd = lt - rt; 176 sum[0] += lt * lt; 177 sum[1] += rt * rt; 178 sum[2] += md * md; 179 sum[3] += sd * sd; 180 } 181} 182 183static void ac3_downmix_5_to_2_symmetric_c(float **samples, float **matrix, 184 int len) 185{ 186 int i; 187 float v0, v1; 188 float front_mix = matrix[0][0]; 189 float center_mix = matrix[0][1]; 190 float surround_mix = matrix[0][3]; 191 192 for (i = 0; i < len; i++) { 193 v0 = samples[0][i] * front_mix + 194 samples[1][i] * center_mix + 195 samples[3][i] * surround_mix; 196 197 v1 = samples[1][i] * center_mix + 198 samples[2][i] * front_mix + 199 samples[4][i] * surround_mix; 200 201 samples[0][i] = v0; 202 samples[1][i] = v1; 203 } 204} 205 206static void ac3_downmix_5_to_1_symmetric_c(float **samples, float **matrix, 207 int len) 208{ 209 int i; 210 float front_mix = matrix[0][0]; 211 float center_mix = matrix[0][1]; 212 float surround_mix = matrix[0][3]; 213 214 for (i = 0; i < len; i++) { 215 samples[0][i] = samples[0][i] * front_mix + 216 samples[1][i] * center_mix + 217 samples[2][i] * front_mix + 218 samples[3][i] * surround_mix + 219 samples[4][i] * surround_mix; 220 } 221} 222 223static void ac3_downmix_c(float **samples, float **matrix, 224 int out_ch, int in_ch, int len) 225{ 226 int i, j; 227 float v0, v1; 228 229 if (out_ch == 2) { 230 for (i = 0; i < len; i++) { 231 v0 = v1 = 0.0f; 232 for (j = 0; j < in_ch; j++) { 233 v0 += samples[j][i] * matrix[0][j]; 234 v1 += samples[j][i] * matrix[1][j]; 235 } 236 samples[0][i] = v0; 237 samples[1][i] = v1; 238 } 239 } else if (out_ch == 1) { 240 for (i = 0; i < len; i++) { 241 v0 = 0.0f; 242 for (j = 0; j < in_ch; j++) 243 v0 += samples[j][i] * matrix[0][j]; 244 samples[0][i] = v0; 245 } 246 } 247} 248 249static void ac3_downmix_5_to_2_symmetric_c_fixed(int32_t **samples, int16_t **matrix, 250 int len) 251{ 252 int i; 253 int64_t v0, v1; 254 int16_t front_mix = matrix[0][0]; 255 int16_t center_mix = matrix[0][1]; 256 int16_t surround_mix = matrix[0][3]; 257 258 for (i = 0; i < len; i++) { 259 v0 = (int64_t)samples[0][i] * front_mix + 260 (int64_t)samples[1][i] * center_mix + 261 (int64_t)samples[3][i] * surround_mix; 262 263 v1 = (int64_t)samples[1][i] * center_mix + 264 (int64_t)samples[2][i] * front_mix + 265 (int64_t)samples[4][i] * surround_mix; 266 267 samples[0][i] = (v0+2048)>>12; 268 samples[1][i] = (v1+2048)>>12; 269 } 270} 271 272static void ac3_downmix_5_to_1_symmetric_c_fixed(int32_t **samples, int16_t **matrix, 273 int len) 274{ 275 int i; 276 int64_t v0; 277 int16_t front_mix = matrix[0][0]; 278 int16_t center_mix = matrix[0][1]; 279 int16_t surround_mix = matrix[0][3]; 280 281 for (i = 0; i < len; i++) { 282 v0 = (int64_t)samples[0][i] * front_mix + 283 (int64_t)samples[1][i] * center_mix + 284 (int64_t)samples[2][i] * front_mix + 285 (int64_t)samples[3][i] * surround_mix + 286 (int64_t)samples[4][i] * surround_mix; 287 288 samples[0][i] = (v0+2048)>>12; 289 } 290} 291 292static void ac3_downmix_c_fixed(int32_t **samples, int16_t **matrix, 293 int out_ch, int in_ch, int len) 294{ 295 int i, j; 296 int64_t v0, v1; 297 if (out_ch == 2) { 298 for (i = 0; i < len; i++) { 299 v0 = v1 = 0; 300 for (j = 0; j < in_ch; j++) { 301 v0 += (int64_t)samples[j][i] * matrix[0][j]; 302 v1 += (int64_t)samples[j][i] * matrix[1][j]; 303 } 304 samples[0][i] = (v0+2048)>>12; 305 samples[1][i] = (v1+2048)>>12; 306 } 307 } else if (out_ch == 1) { 308 for (i = 0; i < len; i++) { 309 v0 = 0; 310 for (j = 0; j < in_ch; j++) 311 v0 += (int64_t)samples[j][i] * matrix[0][j]; 312 samples[0][i] = (v0+2048)>>12; 313 } 314 } 315} 316 317void ff_ac3dsp_downmix_fixed(AC3DSPContext *c, int32_t **samples, int16_t **matrix, 318 int out_ch, int in_ch, int len) 319{ 320 if (c->in_channels != in_ch || c->out_channels != out_ch) { 321 c->in_channels = in_ch; 322 c->out_channels = out_ch; 323 c->downmix_fixed = NULL; 324 325 if (in_ch == 5 && out_ch == 2 && 326 !(matrix[1][0] | matrix[0][2] | 327 matrix[1][3] | matrix[0][4] | 328 (matrix[0][1] ^ matrix[1][1]) | 329 (matrix[0][0] ^ matrix[1][2]))) { 330 c->downmix_fixed = ac3_downmix_5_to_2_symmetric_c_fixed; 331 } else if (in_ch == 5 && out_ch == 1 && 332 matrix[0][0] == matrix[0][2] && 333 matrix[0][3] == matrix[0][4]) { 334 c->downmix_fixed = ac3_downmix_5_to_1_symmetric_c_fixed; 335 } 336 } 337 338 if (c->downmix_fixed) 339 c->downmix_fixed(samples, matrix, len); 340 else 341 ac3_downmix_c_fixed(samples, matrix, out_ch, in_ch, len); 342} 343 344void ff_ac3dsp_downmix(AC3DSPContext *c, float **samples, float **matrix, 345 int out_ch, int in_ch, int len) 346{ 347 if (c->in_channels != in_ch || c->out_channels != out_ch) { 348 int **matrix_cmp = (int **)matrix; 349 350 c->in_channels = in_ch; 351 c->out_channels = out_ch; 352 c->downmix = NULL; 353 354 if (in_ch == 5 && out_ch == 2 && 355 !(matrix_cmp[1][0] | matrix_cmp[0][2] | 356 matrix_cmp[1][3] | matrix_cmp[0][4] | 357 (matrix_cmp[0][1] ^ matrix_cmp[1][1]) | 358 (matrix_cmp[0][0] ^ matrix_cmp[1][2]))) { 359 c->downmix = ac3_downmix_5_to_2_symmetric_c; 360 } else if (in_ch == 5 && out_ch == 1 && 361 matrix_cmp[0][0] == matrix_cmp[0][2] && 362 matrix_cmp[0][3] == matrix_cmp[0][4]) { 363 c->downmix = ac3_downmix_5_to_1_symmetric_c; 364 } 365 366#if ARCH_X86 367 ff_ac3dsp_set_downmix_x86(c); 368#endif 369 } 370 371 if (c->downmix) 372 c->downmix(samples, matrix, len); 373 else 374 ac3_downmix_c(samples, matrix, out_ch, in_ch, len); 375} 376 377av_cold void ff_ac3dsp_init(AC3DSPContext *c, int bit_exact) 378{ 379 c->ac3_exponent_min = ac3_exponent_min_c; 380 c->float_to_fixed24 = float_to_fixed24_c; 381 c->bit_alloc_calc_bap = ac3_bit_alloc_calc_bap_c; 382 c->update_bap_counts = ac3_update_bap_counts_c; 383 c->compute_mantissa_size = ac3_compute_mantissa_size_c; 384 c->extract_exponents = ac3_extract_exponents_c; 385 c->sum_square_butterfly_int32 = ac3_sum_square_butterfly_int32_c; 386 c->sum_square_butterfly_float = ac3_sum_square_butterfly_float_c; 387 c->in_channels = 0; 388 c->out_channels = 0; 389 c->downmix = NULL; 390 c->downmix_fixed = NULL; 391 392#if ARCH_ARM 393 ff_ac3dsp_init_arm(c, bit_exact); 394#elif ARCH_X86 395 ff_ac3dsp_init_x86(c, bit_exact); 396#elif ARCH_MIPS 397 ff_ac3dsp_init_mips(c, bit_exact); 398#endif 399} 400