| 1 | /* |
| 2 | * (c) 2001 Fabrice Bellard |
| 3 | * 2007 Marc Hoffman <marc.hoffman@analog.com> |
| 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 | /** |
| 23 | * @file |
| 24 | * DCT test (c) 2001 Fabrice Bellard |
| 25 | * Started from sample code by Juan J. Sierralta P. |
| 26 | */ |
| 27 | |
| 28 | #include "config.h" |
| 29 | #include <stdlib.h> |
| 30 | #include <stdio.h> |
| 31 | #include <string.h> |
| 32 | #if HAVE_UNISTD_H |
| 33 | #include <unistd.h> |
| 34 | #endif |
| 35 | #include <math.h> |
| 36 | |
| 37 | #include "libavutil/cpu.h" |
| 38 | #include "libavutil/common.h" |
| 39 | #include "libavutil/lfg.h" |
| 40 | #include "libavutil/time.h" |
| 41 | |
| 42 | #include "dct.h" |
| 43 | #include "idctdsp.h" |
| 44 | #include "simple_idct.h" |
| 45 | #include "xvididct.h" |
| 46 | #include "aandcttab.h" |
| 47 | #include "faandct.h" |
| 48 | #include "faanidct.h" |
| 49 | #include "dctref.h" |
| 50 | |
| 51 | struct algo { |
| 52 | const char *name; |
| 53 | void (*func)(int16_t *block); |
| 54 | enum idct_permutation_type perm_type; |
| 55 | int cpu_flag; |
| 56 | int nonspec; |
| 57 | }; |
| 58 | |
| 59 | static const struct algo fdct_tab[] = { |
| 60 | { "REF-DBL", ff_ref_fdct, FF_IDCT_PERM_NONE }, |
| 61 | { "IJG-AAN-INT", ff_fdct_ifast, FF_IDCT_PERM_NONE }, |
| 62 | { "IJG-LLM-INT", ff_jpeg_fdct_islow_8, FF_IDCT_PERM_NONE }, |
| 63 | #if CONFIG_FAANDCT |
| 64 | { "FAAN", ff_faandct, FF_IDCT_PERM_NONE }, |
| 65 | #endif /* CONFIG_FAANDCT */ |
| 66 | }; |
| 67 | |
| 68 | static void ff_prores_idct_wrap(int16_t *dst){ |
| 69 | DECLARE_ALIGNED(16, static int16_t, qmat)[64]; |
| 70 | int i; |
| 71 | |
| 72 | for(i=0; i<64; i++){ |
| 73 | qmat[i]=4; |
| 74 | } |
| 75 | ff_prores_idct(dst, qmat); |
| 76 | for(i=0; i<64; i++) { |
| 77 | dst[i] -= 512; |
| 78 | } |
| 79 | } |
| 80 | |
| 81 | static const struct algo idct_tab[] = { |
| 82 | { "REF-DBL", ff_ref_idct, FF_IDCT_PERM_NONE }, |
| 83 | { "INT", ff_j_rev_dct, FF_IDCT_PERM_LIBMPEG2 }, |
| 84 | { "SIMPLE-C", ff_simple_idct_8, FF_IDCT_PERM_NONE }, |
| 85 | { "PR-C", ff_prores_idct_wrap, FF_IDCT_PERM_NONE, 0, 1 }, |
| 86 | #if CONFIG_FAANIDCT |
| 87 | { "FAANI", ff_faanidct, FF_IDCT_PERM_NONE }, |
| 88 | #endif /* CONFIG_FAANIDCT */ |
| 89 | #if CONFIG_MPEG4_DECODER |
| 90 | { "XVID", ff_xvid_idct, FF_IDCT_PERM_NONE, 0, 1 }, |
| 91 | #endif /* CONFIG_MPEG4_DECODER */ |
| 92 | }; |
| 93 | |
| 94 | #if ARCH_ARM |
| 95 | #include "arm/dct-test.c" |
| 96 | #elif ARCH_PPC |
| 97 | #include "ppc/dct-test.c" |
| 98 | #elif ARCH_X86 |
| 99 | #include "x86/dct-test.c" |
| 100 | #else |
| 101 | static const struct algo fdct_tab_arch[] = { 0 }; |
| 102 | static const struct algo idct_tab_arch[] = { 0 }; |
| 103 | #endif |
| 104 | |
| 105 | #define AANSCALE_BITS 12 |
| 106 | |
| 107 | #define NB_ITS 20000 |
| 108 | #define NB_ITS_SPEED 50000 |
| 109 | |
| 110 | DECLARE_ALIGNED(16, static int16_t, block)[64]; |
| 111 | DECLARE_ALIGNED(8, static int16_t, block1)[64]; |
| 112 | |
| 113 | static void init_block(int16_t block[64], int test, int is_idct, AVLFG *prng, int vals) |
| 114 | { |
| 115 | int i, j; |
| 116 | |
| 117 | memset(block, 0, 64 * sizeof(*block)); |
| 118 | |
| 119 | switch (test) { |
| 120 | case 0: |
| 121 | for (i = 0; i < 64; i++) |
| 122 | block[i] = (av_lfg_get(prng) % (2*vals)) -vals; |
| 123 | if (is_idct) { |
| 124 | ff_ref_fdct(block); |
| 125 | for (i = 0; i < 64; i++) |
| 126 | block[i] >>= 3; |
| 127 | } |
| 128 | break; |
| 129 | case 1: |
| 130 | j = av_lfg_get(prng) % 10 + 1; |
| 131 | for (i = 0; i < j; i++) { |
| 132 | int idx = av_lfg_get(prng) % 64; |
| 133 | block[idx] = av_lfg_get(prng) % (2*vals) -vals; |
| 134 | } |
| 135 | break; |
| 136 | case 2: |
| 137 | block[ 0] = av_lfg_get(prng) % (16*vals) - (8*vals); |
| 138 | block[63] = (block[0] & 1) ^ 1; |
| 139 | break; |
| 140 | } |
| 141 | } |
| 142 | |
| 143 | static void permute(int16_t dst[64], const int16_t src[64], |
| 144 | enum idct_permutation_type perm_type) |
| 145 | { |
| 146 | int i; |
| 147 | |
| 148 | #if ARCH_X86 |
| 149 | if (permute_x86(dst, src, perm_type)) |
| 150 | return; |
| 151 | #endif |
| 152 | |
| 153 | switch (perm_type) { |
| 154 | case FF_IDCT_PERM_LIBMPEG2: |
| 155 | for (i = 0; i < 64; i++) |
| 156 | dst[(i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2)] = src[i]; |
| 157 | break; |
| 158 | case FF_IDCT_PERM_PARTTRANS: |
| 159 | for (i = 0; i < 64; i++) |
| 160 | dst[(i & 0x24) | ((i & 3) << 3) | ((i >> 3) & 3)] = src[i]; |
| 161 | break; |
| 162 | case FF_IDCT_PERM_TRANSPOSE: |
| 163 | for (i = 0; i < 64; i++) |
| 164 | dst[(i>>3) | ((i<<3)&0x38)] = src[i]; |
| 165 | break; |
| 166 | default: |
| 167 | for (i = 0; i < 64; i++) |
| 168 | dst[i] = src[i]; |
| 169 | break; |
| 170 | } |
| 171 | } |
| 172 | |
| 173 | static int dct_error(const struct algo *dct, int test, int is_idct, int speed, const int bits) |
| 174 | { |
| 175 | void (*ref)(int16_t *block) = is_idct ? ff_ref_idct : ff_ref_fdct; |
| 176 | int it, i, scale; |
| 177 | int err_inf, v; |
| 178 | int64_t err2, ti, ti1, it1, err_sum = 0; |
| 179 | int64_t sysErr[64], sysErrMax = 0; |
| 180 | int maxout = 0; |
| 181 | int blockSumErrMax = 0, blockSumErr; |
| 182 | AVLFG prng; |
| 183 | const int vals=1<<bits; |
| 184 | double omse, ome; |
| 185 | int spec_err; |
| 186 | |
| 187 | av_lfg_init(&prng, 1); |
| 188 | |
| 189 | err_inf = 0; |
| 190 | err2 = 0; |
| 191 | for (i = 0; i < 64; i++) |
| 192 | sysErr[i] = 0; |
| 193 | for (it = 0; it < NB_ITS; it++) { |
| 194 | init_block(block1, test, is_idct, &prng, vals); |
| 195 | permute(block, block1, dct->perm_type); |
| 196 | |
| 197 | dct->func(block); |
| 198 | emms_c(); |
| 199 | |
| 200 | if (!strcmp(dct->name, "IJG-AAN-INT")) { |
| 201 | for (i = 0; i < 64; i++) { |
| 202 | scale = 8 * (1 << (AANSCALE_BITS + 11)) / ff_aanscales[i]; |
| 203 | block[i] = (block[i] * scale) >> AANSCALE_BITS; |
| 204 | } |
| 205 | } |
| 206 | |
| 207 | ref(block1); |
| 208 | if (!strcmp(dct->name, "PR-SSE2")) |
| 209 | for (i = 0; i < 64; i++) |
| 210 | block1[i] = av_clip(block1[i], 4-512, 1019-512); |
| 211 | |
| 212 | blockSumErr = 0; |
| 213 | for (i = 0; i < 64; i++) { |
| 214 | int err = block[i] - block1[i]; |
| 215 | err_sum += err; |
| 216 | v = abs(err); |
| 217 | if (v > err_inf) |
| 218 | err_inf = v; |
| 219 | err2 += v * v; |
| 220 | sysErr[i] += block[i] - block1[i]; |
| 221 | blockSumErr += v; |
| 222 | if (abs(block[i]) > maxout) |
| 223 | maxout = abs(block[i]); |
| 224 | } |
| 225 | if (blockSumErrMax < blockSumErr) |
| 226 | blockSumErrMax = blockSumErr; |
| 227 | } |
| 228 | for (i = 0; i < 64; i++) |
| 229 | sysErrMax = FFMAX(sysErrMax, FFABS(sysErr[i])); |
| 230 | |
| 231 | for (i = 0; i < 64; i++) { |
| 232 | if (i % 8 == 0) |
| 233 | printf("\n"); |
| 234 | printf("%7d ", (int) sysErr[i]); |
| 235 | } |
| 236 | printf("\n"); |
| 237 | |
| 238 | omse = (double) err2 / NB_ITS / 64; |
| 239 | ome = (double) err_sum / NB_ITS / 64; |
| 240 | |
| 241 | spec_err = is_idct && (err_inf > 1 || omse > 0.02 || fabs(ome) > 0.0015); |
| 242 | |
| 243 | printf("%s %s: max_err=%d omse=%0.8f ome=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n", |
| 244 | is_idct ? "IDCT" : "DCT", dct->name, err_inf, |
| 245 | omse, ome, (double) sysErrMax / NB_ITS, |
| 246 | maxout, blockSumErrMax); |
| 247 | |
| 248 | if (spec_err && !dct->nonspec) |
| 249 | return 1; |
| 250 | |
| 251 | if (!speed) |
| 252 | return 0; |
| 253 | |
| 254 | /* speed test */ |
| 255 | |
| 256 | init_block(block, test, is_idct, &prng, vals); |
| 257 | permute(block1, block, dct->perm_type); |
| 258 | |
| 259 | ti = av_gettime_relative(); |
| 260 | it1 = 0; |
| 261 | do { |
| 262 | for (it = 0; it < NB_ITS_SPEED; it++) { |
| 263 | memcpy(block, block1, sizeof(block)); |
| 264 | dct->func(block); |
| 265 | } |
| 266 | emms_c(); |
| 267 | it1 += NB_ITS_SPEED; |
| 268 | ti1 = av_gettime_relative() - ti; |
| 269 | } while (ti1 < 1000000); |
| 270 | |
| 271 | printf("%s %s: %0.1f kdct/s\n", is_idct ? "IDCT" : "DCT", dct->name, |
| 272 | (double) it1 * 1000.0 / (double) ti1); |
| 273 | |
| 274 | return 0; |
| 275 | } |
| 276 | |
| 277 | DECLARE_ALIGNED(8, static uint8_t, img_dest)[64]; |
| 278 | DECLARE_ALIGNED(8, static uint8_t, img_dest1)[64]; |
| 279 | |
| 280 | static void idct248_ref(uint8_t *dest, int linesize, int16_t *block) |
| 281 | { |
| 282 | static int init; |
| 283 | static double c8[8][8]; |
| 284 | static double c4[4][4]; |
| 285 | double block1[64], block2[64], block3[64]; |
| 286 | double s, sum, v; |
| 287 | int i, j, k; |
| 288 | |
| 289 | if (!init) { |
| 290 | init = 1; |
| 291 | |
| 292 | for (i = 0; i < 8; i++) { |
| 293 | sum = 0; |
| 294 | for (j = 0; j < 8; j++) { |
| 295 | s = (i == 0) ? sqrt(1.0 / 8.0) : sqrt(1.0 / 4.0); |
| 296 | c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0); |
| 297 | sum += c8[i][j] * c8[i][j]; |
| 298 | } |
| 299 | } |
| 300 | |
| 301 | for (i = 0; i < 4; i++) { |
| 302 | sum = 0; |
| 303 | for (j = 0; j < 4; j++) { |
| 304 | s = (i == 0) ? sqrt(1.0 / 4.0) : sqrt(1.0 / 2.0); |
| 305 | c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0); |
| 306 | sum += c4[i][j] * c4[i][j]; |
| 307 | } |
| 308 | } |
| 309 | } |
| 310 | |
| 311 | /* butterfly */ |
| 312 | s = 0.5 * sqrt(2.0); |
| 313 | for (i = 0; i < 4; i++) { |
| 314 | for (j = 0; j < 8; j++) { |
| 315 | block1[8 * (2 * i) + j] = |
| 316 | (block[8 * (2 * i) + j] + block[8 * (2 * i + 1) + j]) * s; |
| 317 | block1[8 * (2 * i + 1) + j] = |
| 318 | (block[8 * (2 * i) + j] - block[8 * (2 * i + 1) + j]) * s; |
| 319 | } |
| 320 | } |
| 321 | |
| 322 | /* idct8 on lines */ |
| 323 | for (i = 0; i < 8; i++) { |
| 324 | for (j = 0; j < 8; j++) { |
| 325 | sum = 0; |
| 326 | for (k = 0; k < 8; k++) |
| 327 | sum += c8[k][j] * block1[8 * i + k]; |
| 328 | block2[8 * i + j] = sum; |
| 329 | } |
| 330 | } |
| 331 | |
| 332 | /* idct4 */ |
| 333 | for (i = 0; i < 8; i++) { |
| 334 | for (j = 0; j < 4; j++) { |
| 335 | /* top */ |
| 336 | sum = 0; |
| 337 | for (k = 0; k < 4; k++) |
| 338 | sum += c4[k][j] * block2[8 * (2 * k) + i]; |
| 339 | block3[8 * (2 * j) + i] = sum; |
| 340 | |
| 341 | /* bottom */ |
| 342 | sum = 0; |
| 343 | for (k = 0; k < 4; k++) |
| 344 | sum += c4[k][j] * block2[8 * (2 * k + 1) + i]; |
| 345 | block3[8 * (2 * j + 1) + i] = sum; |
| 346 | } |
| 347 | } |
| 348 | |
| 349 | /* clamp and store the result */ |
| 350 | for (i = 0; i < 8; i++) { |
| 351 | for (j = 0; j < 8; j++) { |
| 352 | v = block3[8 * i + j]; |
| 353 | if (v < 0) v = 0; |
| 354 | else if (v > 255) v = 255; |
| 355 | dest[i * linesize + j] = (int) rint(v); |
| 356 | } |
| 357 | } |
| 358 | } |
| 359 | |
| 360 | static void idct248_error(const char *name, |
| 361 | void (*idct248_put)(uint8_t *dest, int line_size, |
| 362 | int16_t *block), |
| 363 | int speed) |
| 364 | { |
| 365 | int it, i, it1, ti, ti1, err_max, v; |
| 366 | AVLFG prng; |
| 367 | |
| 368 | av_lfg_init(&prng, 1); |
| 369 | |
| 370 | /* just one test to see if code is correct (precision is less |
| 371 | important here) */ |
| 372 | err_max = 0; |
| 373 | for (it = 0; it < NB_ITS; it++) { |
| 374 | /* XXX: use forward transform to generate values */ |
| 375 | for (i = 0; i < 64; i++) |
| 376 | block1[i] = av_lfg_get(&prng) % 256 - 128; |
| 377 | block1[0] += 1024; |
| 378 | |
| 379 | for (i = 0; i < 64; i++) |
| 380 | block[i] = block1[i]; |
| 381 | idct248_ref(img_dest1, 8, block); |
| 382 | |
| 383 | for (i = 0; i < 64; i++) |
| 384 | block[i] = block1[i]; |
| 385 | idct248_put(img_dest, 8, block); |
| 386 | |
| 387 | for (i = 0; i < 64; i++) { |
| 388 | v = abs((int) img_dest[i] - (int) img_dest1[i]); |
| 389 | if (v == 255) |
| 390 | printf("%d %d\n", img_dest[i], img_dest1[i]); |
| 391 | if (v > err_max) |
| 392 | err_max = v; |
| 393 | } |
| 394 | #if 0 |
| 395 | printf("ref=\n"); |
| 396 | for(i=0;i<8;i++) { |
| 397 | int j; |
| 398 | for(j=0;j<8;j++) { |
| 399 | printf(" %3d", img_dest1[i*8+j]); |
| 400 | } |
| 401 | printf("\n"); |
| 402 | } |
| 403 | |
| 404 | printf("out=\n"); |
| 405 | for(i=0;i<8;i++) { |
| 406 | int j; |
| 407 | for(j=0;j<8;j++) { |
| 408 | printf(" %3d", img_dest[i*8+j]); |
| 409 | } |
| 410 | printf("\n"); |
| 411 | } |
| 412 | #endif |
| 413 | } |
| 414 | printf("%s %s: err_inf=%d\n", 1 ? "IDCT248" : "DCT248", name, err_max); |
| 415 | |
| 416 | if (!speed) |
| 417 | return; |
| 418 | |
| 419 | ti = av_gettime_relative(); |
| 420 | it1 = 0; |
| 421 | do { |
| 422 | for (it = 0; it < NB_ITS_SPEED; it++) { |
| 423 | for (i = 0; i < 64; i++) |
| 424 | block[i] = block1[i]; |
| 425 | idct248_put(img_dest, 8, block); |
| 426 | } |
| 427 | emms_c(); |
| 428 | it1 += NB_ITS_SPEED; |
| 429 | ti1 = av_gettime_relative() - ti; |
| 430 | } while (ti1 < 1000000); |
| 431 | |
| 432 | printf("%s %s: %0.1f kdct/s\n", 1 ? "IDCT248" : "DCT248", name, |
| 433 | (double) it1 * 1000.0 / (double) ti1); |
| 434 | } |
| 435 | |
| 436 | static void help(void) |
| 437 | { |
| 438 | printf("dct-test [-i] [<test-number>] [<bits>]\n" |
| 439 | "test-number 0 -> test with random matrixes\n" |
| 440 | " 1 -> test with random sparse matrixes\n" |
| 441 | " 2 -> do 3. test from mpeg4 std\n" |
| 442 | "bits Number of time domain bits to use, 8 is default\n" |
| 443 | "-i test IDCT implementations\n" |
| 444 | "-4 test IDCT248 implementations\n" |
| 445 | "-t speed test\n"); |
| 446 | } |
| 447 | |
| 448 | #if !HAVE_GETOPT |
| 449 | #include "compat/getopt.c" |
| 450 | #endif |
| 451 | |
| 452 | int main(int argc, char **argv) |
| 453 | { |
| 454 | int test_idct = 0, test_248_dct = 0; |
| 455 | int c, i; |
| 456 | int test = 1; |
| 457 | int speed = 0; |
| 458 | int err = 0; |
| 459 | int bits=8; |
| 460 | |
| 461 | ff_ref_dct_init(); |
| 462 | |
| 463 | for (;;) { |
| 464 | c = getopt(argc, argv, "ih4t"); |
| 465 | if (c == -1) |
| 466 | break; |
| 467 | switch (c) { |
| 468 | case 'i': |
| 469 | test_idct = 1; |
| 470 | break; |
| 471 | case '4': |
| 472 | test_248_dct = 1; |
| 473 | break; |
| 474 | case 't': |
| 475 | speed = 1; |
| 476 | break; |
| 477 | default: |
| 478 | case 'h': |
| 479 | help(); |
| 480 | return 0; |
| 481 | } |
| 482 | } |
| 483 | |
| 484 | if (optind < argc) |
| 485 | test = atoi(argv[optind]); |
| 486 | if(optind+1 < argc) bits= atoi(argv[optind+1]); |
| 487 | |
| 488 | printf("ffmpeg DCT/IDCT test\n"); |
| 489 | |
| 490 | if (test_248_dct) { |
| 491 | idct248_error("SIMPLE-C", ff_simple_idct248_put, speed); |
| 492 | } else { |
| 493 | const int cpu_flags = av_get_cpu_flags(); |
| 494 | if (test_idct) { |
| 495 | for (i = 0; i < FF_ARRAY_ELEMS(idct_tab); i++) |
| 496 | err |= dct_error(&idct_tab[i], test, test_idct, speed, bits); |
| 497 | |
| 498 | for (i = 0; idct_tab_arch[i].name; i++) |
| 499 | if (!(~cpu_flags & idct_tab_arch[i].cpu_flag)) |
| 500 | err |= dct_error(&idct_tab_arch[i], test, test_idct, speed, bits); |
| 501 | } |
| 502 | #if CONFIG_FDCTDSP |
| 503 | else { |
| 504 | for (i = 0; i < FF_ARRAY_ELEMS(fdct_tab); i++) |
| 505 | err |= dct_error(&fdct_tab[i], test, test_idct, speed, bits); |
| 506 | |
| 507 | for (i = 0; fdct_tab_arch[i].name; i++) |
| 508 | if (!(~cpu_flags & fdct_tab_arch[i].cpu_flag)) |
| 509 | err |= dct_error(&fdct_tab_arch[i], test, test_idct, speed, bits); |
| 510 | } |
| 511 | #endif /* CONFIG_FDCTDSP */ |
| 512 | } |
| 513 | |
| 514 | if (err) |
| 515 | printf("Error: %d.\n", err); |
| 516 | |
| 517 | return !!err; |
| 518 | } |