| 1 | /* |
| 2 | * Copyright (c) 2001, 2002 Fabrice Bellard |
| 3 | * |
| 4 | * This file is part of FFmpeg. |
| 5 | * |
| 6 | * FFmpeg is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU Lesser General Public |
| 8 | * License as published by the Free Software Foundation; either |
| 9 | * version 2.1 of the License, or (at your option) any later version. |
| 10 | * |
| 11 | * FFmpeg is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 14 | * Lesser General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU Lesser General Public |
| 17 | * License along with FFmpeg; if not, write to the Free Software |
| 18 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 19 | */ |
| 20 | |
| 21 | #include <stdint.h> |
| 22 | |
| 23 | #include "libavutil/attributes.h" |
| 24 | #include "libavutil/mem.h" |
| 25 | #include "dct32.h" |
| 26 | #include "mathops.h" |
| 27 | #include "mpegaudiodsp.h" |
| 28 | #include "mpegaudio.h" |
| 29 | |
| 30 | #if USE_FLOATS |
| 31 | #define RENAME(n) n##_float |
| 32 | |
| 33 | static inline float round_sample(float *sum) |
| 34 | { |
| 35 | float sum1=*sum; |
| 36 | *sum = 0; |
| 37 | return sum1; |
| 38 | } |
| 39 | |
| 40 | #define MACS(rt, ra, rb) rt+=(ra)*(rb) |
| 41 | #define MULS(ra, rb) ((ra)*(rb)) |
| 42 | #define MULH3(x, y, s) ((s)*(y)*(x)) |
| 43 | #define MLSS(rt, ra, rb) rt-=(ra)*(rb) |
| 44 | #define MULLx(x, y, s) ((y)*(x)) |
| 45 | #define FIXHR(x) ((float)(x)) |
| 46 | #define FIXR(x) ((float)(x)) |
| 47 | #define SHR(a,b) ((a)*(1.0f/(1<<(b)))) |
| 48 | |
| 49 | #else |
| 50 | |
| 51 | #define RENAME(n) n##_fixed |
| 52 | #define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15) |
| 53 | |
| 54 | static inline int round_sample(int64_t *sum) |
| 55 | { |
| 56 | int sum1; |
| 57 | sum1 = (int)((*sum) >> OUT_SHIFT); |
| 58 | *sum &= (1<<OUT_SHIFT)-1; |
| 59 | return av_clip_int16(sum1); |
| 60 | } |
| 61 | |
| 62 | # define MULS(ra, rb) MUL64(ra, rb) |
| 63 | # define MACS(rt, ra, rb) MAC64(rt, ra, rb) |
| 64 | # define MLSS(rt, ra, rb) MLS64(rt, ra, rb) |
| 65 | # define MULH3(x, y, s) MULH((s)*(x), y) |
| 66 | # define MULLx(x, y, s) MULL(x,y,s) |
| 67 | # define SHR(a,b) ((a)>>(b)) |
| 68 | # define FIXR(a) ((int)((a) * FRAC_ONE + 0.5)) |
| 69 | # define FIXHR(a) ((int)((a) * (1LL<<32) + 0.5)) |
| 70 | #endif |
| 71 | |
| 72 | /** Window for MDCT. Actually only the elements in [0,17] and |
| 73 | [MDCT_BUF_SIZE/2, MDCT_BUF_SIZE/2 + 17] are actually used. The rest |
| 74 | is just to preserve alignment for SIMD implementations. |
| 75 | */ |
| 76 | DECLARE_ALIGNED(16, INTFLOAT, RENAME(ff_mdct_win))[8][MDCT_BUF_SIZE]; |
| 77 | |
| 78 | DECLARE_ALIGNED(16, MPA_INT, RENAME(ff_mpa_synth_window))[512+256]; |
| 79 | |
| 80 | #define SUM8(op, sum, w, p) \ |
| 81 | { \ |
| 82 | op(sum, (w)[0 * 64], (p)[0 * 64]); \ |
| 83 | op(sum, (w)[1 * 64], (p)[1 * 64]); \ |
| 84 | op(sum, (w)[2 * 64], (p)[2 * 64]); \ |
| 85 | op(sum, (w)[3 * 64], (p)[3 * 64]); \ |
| 86 | op(sum, (w)[4 * 64], (p)[4 * 64]); \ |
| 87 | op(sum, (w)[5 * 64], (p)[5 * 64]); \ |
| 88 | op(sum, (w)[6 * 64], (p)[6 * 64]); \ |
| 89 | op(sum, (w)[7 * 64], (p)[7 * 64]); \ |
| 90 | } |
| 91 | |
| 92 | #define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \ |
| 93 | { \ |
| 94 | INTFLOAT tmp;\ |
| 95 | tmp = p[0 * 64];\ |
| 96 | op1(sum1, (w1)[0 * 64], tmp);\ |
| 97 | op2(sum2, (w2)[0 * 64], tmp);\ |
| 98 | tmp = p[1 * 64];\ |
| 99 | op1(sum1, (w1)[1 * 64], tmp);\ |
| 100 | op2(sum2, (w2)[1 * 64], tmp);\ |
| 101 | tmp = p[2 * 64];\ |
| 102 | op1(sum1, (w1)[2 * 64], tmp);\ |
| 103 | op2(sum2, (w2)[2 * 64], tmp);\ |
| 104 | tmp = p[3 * 64];\ |
| 105 | op1(sum1, (w1)[3 * 64], tmp);\ |
| 106 | op2(sum2, (w2)[3 * 64], tmp);\ |
| 107 | tmp = p[4 * 64];\ |
| 108 | op1(sum1, (w1)[4 * 64], tmp);\ |
| 109 | op2(sum2, (w2)[4 * 64], tmp);\ |
| 110 | tmp = p[5 * 64];\ |
| 111 | op1(sum1, (w1)[5 * 64], tmp);\ |
| 112 | op2(sum2, (w2)[5 * 64], tmp);\ |
| 113 | tmp = p[6 * 64];\ |
| 114 | op1(sum1, (w1)[6 * 64], tmp);\ |
| 115 | op2(sum2, (w2)[6 * 64], tmp);\ |
| 116 | tmp = p[7 * 64];\ |
| 117 | op1(sum1, (w1)[7 * 64], tmp);\ |
| 118 | op2(sum2, (w2)[7 * 64], tmp);\ |
| 119 | } |
| 120 | |
| 121 | void RENAME(ff_mpadsp_apply_window)(MPA_INT *synth_buf, MPA_INT *window, |
| 122 | int *dither_state, OUT_INT *samples, |
| 123 | int incr) |
| 124 | { |
| 125 | register const MPA_INT *w, *w2, *p; |
| 126 | int j; |
| 127 | OUT_INT *samples2; |
| 128 | #if USE_FLOATS |
| 129 | float sum, sum2; |
| 130 | #else |
| 131 | int64_t sum, sum2; |
| 132 | #endif |
| 133 | |
| 134 | /* copy to avoid wrap */ |
| 135 | memcpy(synth_buf + 512, synth_buf, 32 * sizeof(*synth_buf)); |
| 136 | |
| 137 | samples2 = samples + 31 * incr; |
| 138 | w = window; |
| 139 | w2 = window + 31; |
| 140 | |
| 141 | sum = *dither_state; |
| 142 | p = synth_buf + 16; |
| 143 | SUM8(MACS, sum, w, p); |
| 144 | p = synth_buf + 48; |
| 145 | SUM8(MLSS, sum, w + 32, p); |
| 146 | *samples = round_sample(&sum); |
| 147 | samples += incr; |
| 148 | w++; |
| 149 | |
| 150 | /* we calculate two samples at the same time to avoid one memory |
| 151 | access per two sample */ |
| 152 | for(j=1;j<16;j++) { |
| 153 | sum2 = 0; |
| 154 | p = synth_buf + 16 + j; |
| 155 | SUM8P2(sum, MACS, sum2, MLSS, w, w2, p); |
| 156 | p = synth_buf + 48 - j; |
| 157 | SUM8P2(sum, MLSS, sum2, MLSS, w + 32, w2 + 32, p); |
| 158 | |
| 159 | *samples = round_sample(&sum); |
| 160 | samples += incr; |
| 161 | sum += sum2; |
| 162 | *samples2 = round_sample(&sum); |
| 163 | samples2 -= incr; |
| 164 | w++; |
| 165 | w2--; |
| 166 | } |
| 167 | |
| 168 | p = synth_buf + 32; |
| 169 | SUM8(MLSS, sum, w + 32, p); |
| 170 | *samples = round_sample(&sum); |
| 171 | *dither_state= sum; |
| 172 | } |
| 173 | |
| 174 | /* 32 sub band synthesis filter. Input: 32 sub band samples, Output: |
| 175 | 32 samples. */ |
| 176 | void RENAME(ff_mpa_synth_filter)(MPADSPContext *s, MPA_INT *synth_buf_ptr, |
| 177 | int *synth_buf_offset, |
| 178 | MPA_INT *window, int *dither_state, |
| 179 | OUT_INT *samples, int incr, |
| 180 | MPA_INT *sb_samples) |
| 181 | { |
| 182 | MPA_INT *synth_buf; |
| 183 | int offset; |
| 184 | |
| 185 | offset = *synth_buf_offset; |
| 186 | synth_buf = synth_buf_ptr + offset; |
| 187 | |
| 188 | s->RENAME(dct32)(synth_buf, sb_samples); |
| 189 | s->RENAME(apply_window)(synth_buf, window, dither_state, samples, incr); |
| 190 | |
| 191 | offset = (offset - 32) & 511; |
| 192 | *synth_buf_offset = offset; |
| 193 | } |
| 194 | |
| 195 | av_cold void RENAME(ff_mpa_synth_init)(MPA_INT *window) |
| 196 | { |
| 197 | int i, j; |
| 198 | |
| 199 | /* max = 18760, max sum over all 16 coefs : 44736 */ |
| 200 | for(i=0;i<257;i++) { |
| 201 | INTFLOAT v; |
| 202 | v = ff_mpa_enwindow[i]; |
| 203 | #if USE_FLOATS |
| 204 | v *= 1.0 / (1LL<<(16 + FRAC_BITS)); |
| 205 | #endif |
| 206 | window[i] = v; |
| 207 | if ((i & 63) != 0) |
| 208 | v = -v; |
| 209 | if (i != 0) |
| 210 | window[512 - i] = v; |
| 211 | } |
| 212 | |
| 213 | |
| 214 | // Needed for avoiding shuffles in ASM implementations |
| 215 | for(i=0; i < 8; i++) |
| 216 | for(j=0; j < 16; j++) |
| 217 | window[512+16*i+j] = window[64*i+32-j]; |
| 218 | |
| 219 | for(i=0; i < 8; i++) |
| 220 | for(j=0; j < 16; j++) |
| 221 | window[512+128+16*i+j] = window[64*i+48-j]; |
| 222 | } |
| 223 | |
| 224 | av_cold void RENAME(ff_init_mpadsp_tabs)(void) |
| 225 | { |
| 226 | int i, j; |
| 227 | /* compute mdct windows */ |
| 228 | for (i = 0; i < 36; i++) { |
| 229 | for (j = 0; j < 4; j++) { |
| 230 | double d; |
| 231 | |
| 232 | if (j == 2 && i % 3 != 1) |
| 233 | continue; |
| 234 | |
| 235 | d = sin(M_PI * (i + 0.5) / 36.0); |
| 236 | if (j == 1) { |
| 237 | if (i >= 30) d = 0; |
| 238 | else if (i >= 24) d = sin(M_PI * (i - 18 + 0.5) / 12.0); |
| 239 | else if (i >= 18) d = 1; |
| 240 | } else if (j == 3) { |
| 241 | if (i < 6) d = 0; |
| 242 | else if (i < 12) d = sin(M_PI * (i - 6 + 0.5) / 12.0); |
| 243 | else if (i < 18) d = 1; |
| 244 | } |
| 245 | //merge last stage of imdct into the window coefficients |
| 246 | d *= 0.5 * IMDCT_SCALAR / cos(M_PI * (2 * i + 19) / 72); |
| 247 | |
| 248 | if (j == 2) |
| 249 | RENAME(ff_mdct_win)[j][i/3] = FIXHR((d / (1<<5))); |
| 250 | else { |
| 251 | int idx = i < 18 ? i : i + (MDCT_BUF_SIZE/2 - 18); |
| 252 | RENAME(ff_mdct_win)[j][idx] = FIXHR((d / (1<<5))); |
| 253 | } |
| 254 | } |
| 255 | } |
| 256 | |
| 257 | /* NOTE: we do frequency inversion adter the MDCT by changing |
| 258 | the sign of the right window coefs */ |
| 259 | for (j = 0; j < 4; j++) { |
| 260 | for (i = 0; i < MDCT_BUF_SIZE; i += 2) { |
| 261 | RENAME(ff_mdct_win)[j + 4][i ] = RENAME(ff_mdct_win)[j][i ]; |
| 262 | RENAME(ff_mdct_win)[j + 4][i + 1] = -RENAME(ff_mdct_win)[j][i + 1]; |
| 263 | } |
| 264 | } |
| 265 | } |
| 266 | /* cos(pi*i/18) */ |
| 267 | #define C1 FIXHR(0.98480775301220805936/2) |
| 268 | #define C2 FIXHR(0.93969262078590838405/2) |
| 269 | #define C3 FIXHR(0.86602540378443864676/2) |
| 270 | #define C4 FIXHR(0.76604444311897803520/2) |
| 271 | #define C5 FIXHR(0.64278760968653932632/2) |
| 272 | #define C6 FIXHR(0.5/2) |
| 273 | #define C7 FIXHR(0.34202014332566873304/2) |
| 274 | #define C8 FIXHR(0.17364817766693034885/2) |
| 275 | |
| 276 | /* 0.5 / cos(pi*(2*i+1)/36) */ |
| 277 | static const INTFLOAT icos36[9] = { |
| 278 | FIXR(0.50190991877167369479), |
| 279 | FIXR(0.51763809020504152469), //0 |
| 280 | FIXR(0.55168895948124587824), |
| 281 | FIXR(0.61038729438072803416), |
| 282 | FIXR(0.70710678118654752439), //1 |
| 283 | FIXR(0.87172339781054900991), |
| 284 | FIXR(1.18310079157624925896), |
| 285 | FIXR(1.93185165257813657349), //2 |
| 286 | FIXR(5.73685662283492756461), |
| 287 | }; |
| 288 | |
| 289 | /* 0.5 / cos(pi*(2*i+1)/36) */ |
| 290 | static const INTFLOAT icos36h[9] = { |
| 291 | FIXHR(0.50190991877167369479/2), |
| 292 | FIXHR(0.51763809020504152469/2), //0 |
| 293 | FIXHR(0.55168895948124587824/2), |
| 294 | FIXHR(0.61038729438072803416/2), |
| 295 | FIXHR(0.70710678118654752439/2), //1 |
| 296 | FIXHR(0.87172339781054900991/2), |
| 297 | FIXHR(1.18310079157624925896/4), |
| 298 | FIXHR(1.93185165257813657349/4), //2 |
| 299 | // FIXHR(5.73685662283492756461), |
| 300 | }; |
| 301 | |
| 302 | /* using Lee like decomposition followed by hand coded 9 points DCT */ |
| 303 | static void imdct36(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in, INTFLOAT *win) |
| 304 | { |
| 305 | int i, j; |
| 306 | INTFLOAT t0, t1, t2, t3, s0, s1, s2, s3; |
| 307 | INTFLOAT tmp[18], *tmp1, *in1; |
| 308 | |
| 309 | for (i = 17; i >= 1; i--) |
| 310 | in[i] += in[i-1]; |
| 311 | for (i = 17; i >= 3; i -= 2) |
| 312 | in[i] += in[i-2]; |
| 313 | |
| 314 | for (j = 0; j < 2; j++) { |
| 315 | tmp1 = tmp + j; |
| 316 | in1 = in + j; |
| 317 | |
| 318 | t2 = in1[2*4] + in1[2*8] - in1[2*2]; |
| 319 | |
| 320 | t3 = in1[2*0] + SHR(in1[2*6],1); |
| 321 | t1 = in1[2*0] - in1[2*6]; |
| 322 | tmp1[ 6] = t1 - SHR(t2,1); |
| 323 | tmp1[16] = t1 + t2; |
| 324 | |
| 325 | t0 = MULH3(in1[2*2] + in1[2*4] , C2, 2); |
| 326 | t1 = MULH3(in1[2*4] - in1[2*8] , -2*C8, 1); |
| 327 | t2 = MULH3(in1[2*2] + in1[2*8] , -C4, 2); |
| 328 | |
| 329 | tmp1[10] = t3 - t0 - t2; |
| 330 | tmp1[ 2] = t3 + t0 + t1; |
| 331 | tmp1[14] = t3 + t2 - t1; |
| 332 | |
| 333 | tmp1[ 4] = MULH3(in1[2*5] + in1[2*7] - in1[2*1], -C3, 2); |
| 334 | t2 = MULH3(in1[2*1] + in1[2*5], C1, 2); |
| 335 | t3 = MULH3(in1[2*5] - in1[2*7], -2*C7, 1); |
| 336 | t0 = MULH3(in1[2*3], C3, 2); |
| 337 | |
| 338 | t1 = MULH3(in1[2*1] + in1[2*7], -C5, 2); |
| 339 | |
| 340 | tmp1[ 0] = t2 + t3 + t0; |
| 341 | tmp1[12] = t2 + t1 - t0; |
| 342 | tmp1[ 8] = t3 - t1 - t0; |
| 343 | } |
| 344 | |
| 345 | i = 0; |
| 346 | for (j = 0; j < 4; j++) { |
| 347 | t0 = tmp[i]; |
| 348 | t1 = tmp[i + 2]; |
| 349 | s0 = t1 + t0; |
| 350 | s2 = t1 - t0; |
| 351 | |
| 352 | t2 = tmp[i + 1]; |
| 353 | t3 = tmp[i + 3]; |
| 354 | s1 = MULH3(t3 + t2, icos36h[ j], 2); |
| 355 | s3 = MULLx(t3 - t2, icos36 [8 - j], FRAC_BITS); |
| 356 | |
| 357 | t0 = s0 + s1; |
| 358 | t1 = s0 - s1; |
| 359 | out[(9 + j) * SBLIMIT] = MULH3(t1, win[ 9 + j], 1) + buf[4*(9 + j)]; |
| 360 | out[(8 - j) * SBLIMIT] = MULH3(t1, win[ 8 - j], 1) + buf[4*(8 - j)]; |
| 361 | buf[4 * ( 9 + j )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 9 + j], 1); |
| 362 | buf[4 * ( 8 - j )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 8 - j], 1); |
| 363 | |
| 364 | t0 = s2 + s3; |
| 365 | t1 = s2 - s3; |
| 366 | out[(9 + 8 - j) * SBLIMIT] = MULH3(t1, win[ 9 + 8 - j], 1) + buf[4*(9 + 8 - j)]; |
| 367 | out[ j * SBLIMIT] = MULH3(t1, win[ j], 1) + buf[4*( j)]; |
| 368 | buf[4 * ( 9 + 8 - j )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 9 + 8 - j], 1); |
| 369 | buf[4 * ( j )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + j], 1); |
| 370 | i += 4; |
| 371 | } |
| 372 | |
| 373 | s0 = tmp[16]; |
| 374 | s1 = MULH3(tmp[17], icos36h[4], 2); |
| 375 | t0 = s0 + s1; |
| 376 | t1 = s0 - s1; |
| 377 | out[(9 + 4) * SBLIMIT] = MULH3(t1, win[ 9 + 4], 1) + buf[4*(9 + 4)]; |
| 378 | out[(8 - 4) * SBLIMIT] = MULH3(t1, win[ 8 - 4], 1) + buf[4*(8 - 4)]; |
| 379 | buf[4 * ( 9 + 4 )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 9 + 4], 1); |
| 380 | buf[4 * ( 8 - 4 )] = MULH3(t0, win[MDCT_BUF_SIZE/2 + 8 - 4], 1); |
| 381 | } |
| 382 | |
| 383 | void RENAME(ff_imdct36_blocks)(INTFLOAT *out, INTFLOAT *buf, INTFLOAT *in, |
| 384 | int count, int switch_point, int block_type) |
| 385 | { |
| 386 | int j; |
| 387 | for (j=0 ; j < count; j++) { |
| 388 | /* apply window & overlap with previous buffer */ |
| 389 | |
| 390 | /* select window */ |
| 391 | int win_idx = (switch_point && j < 2) ? 0 : block_type; |
| 392 | INTFLOAT *win = RENAME(ff_mdct_win)[win_idx + (4 & -(j & 1))]; |
| 393 | |
| 394 | imdct36(out, buf, in, win); |
| 395 | |
| 396 | in += 18; |
| 397 | buf += ((j&3) != 3 ? 1 : (72-3)); |
| 398 | out++; |
| 399 | } |
| 400 | } |
| 401 | |