Parent Directory
| 
Revision Log
Revision 257 - (view) (download)
| 1 : | Isibaar | 225 | /************************************************************************** |
| 2 : | * | ||
| 3 : | * XVID MPEG-4 VIDEO CODEC - Unit tests and benches | ||
| 4 : | * | ||
| 5 : | * This program is free software; you can redistribute it and/or modify | ||
| 6 : | * it under the terms of the GNU General Public License as published by | ||
| 7 : | * the Free Software Foundation; either version 2 of the License, or | ||
| 8 : | * (at your option) any later version. | ||
| 9 : | * | ||
| 10 : | * This program is distributed in the hope that it will be useful, | ||
| 11 : | * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 12 : | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 13 : | * GNU General Public License for more details. | ||
| 14 : | * | ||
| 15 : | * You should have received a copy of the GNU General Public License | ||
| 16 : | * along with this program; if not, write to the Free Software | ||
| 17 : | * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. | ||
| 18 : | * | ||
| 19 : | *************************************************************************/ | ||
| 20 : | |||
| 21 : | /************************************************************************ | ||
| 22 : | * | ||
| 23 : | * 'Reference' output is at the end of file. | ||
| 24 : | * Don't take the checksums and crc too seriouly, they aren't | ||
| 25 : | * bullet-proof... | ||
| 26 : | * | ||
| 27 : | * compiles with something like: | ||
| 28 : | * gcc -o xvid_bench xvid_bench.c -I../src/ -lxvidcore -lm | ||
| 29 : | * | ||
| 30 : | * History: | ||
| 31 : | * | ||
| 32 : | * 06.06.2002 initial coding -Skal- | ||
| 33 : | * | ||
| 34 : | *************************************************************************/ | ||
| 35 : | |||
| 36 : | #include <stdio.h> | ||
| 37 : | #include <stdlib.h> | ||
| 38 : | #include <sys/time.h> // for gettimeofday | ||
| 39 : | #include <string.h> // for memset | ||
| 40 : | #include <assert.h> | ||
| 41 : | |||
| 42 : | #include "xvid.h" | ||
| 43 : | |||
| 44 : | // inner guts | ||
| 45 : | #include "dct/idct.h" | ||
| 46 : | #include "dct/fdct.h" | ||
| 47 : | #include "image/colorspace.h" | ||
| 48 : | #include "image/interpolate8x8.h" | ||
| 49 : | #include "utils/mem_transfer.h" | ||
| 50 : | #include "quant/quant_h263.h" | ||
| 51 : | #include "quant/quant_mpeg4.h" | ||
| 52 : | #include "motion/sad.h" | ||
| 53 : | #include "utils/emms.h" | ||
| 54 : | #include "utils/timer.h" | ||
| 55 : | #include "quant/quant_matrix.c" | ||
| 56 : | #include "bitstream/cbp.h" | ||
| 57 : | |||
| 58 : | const int speed_ref = 100; // on slow machines, decrease this value | ||
| 59 : | |||
| 60 : | /********************************************************************* | ||
| 61 : | * misc | ||
| 62 : | *********************************************************************/ | ||
| 63 : | |||
| 64 : | /* returns time in micro-s*/ | ||
| 65 : | double gettime_usec() | ||
| 66 : | { | ||
| 67 : | struct timeval tv; | ||
| 68 : | gettimeofday(&tv, 0); | ||
| 69 : | ia64p | 257 | return tv.tv_sec*1.0e6 + tv.tv_usec; |
| 70 : | Isibaar | 225 | } |
| 71 : | |||
| 72 : | /* returns squared deviates (mean(v*v)-mean(v)^2) of a 8x8 block */ | ||
| 73 : | double sqr_dev(uint8_t v[8*8]) | ||
| 74 : | { | ||
| 75 : | double sum=0.; | ||
| 76 : | double sum2=0.; | ||
| 77 : | int n; | ||
| 78 : | for (n=0;n<8*8;n++) | ||
| 79 : | { | ||
| 80 : | sum += v[n]; | ||
| 81 : | sum2 += v[n]*v[n]; | ||
| 82 : | } | ||
| 83 : | sum2 /= n; | ||
| 84 : | sum /= n; | ||
| 85 : | return sum2-sum*sum; | ||
| 86 : | } | ||
| 87 : | |||
| 88 : | /********************************************************************* | ||
| 89 : | * cpu init | ||
| 90 : | *********************************************************************/ | ||
| 91 : | |||
| 92 : | typedef struct { | ||
| 93 : | const char *name; | ||
| 94 : | unsigned int cpu; | ||
| 95 : | } CPU; | ||
| 96 : | |||
| 97 : | CPU cpu_list[] = | ||
| 98 : | { { "PLAINC", 0 } | ||
| 99 : | , { "MMX ", XVID_CPU_MMX } | ||
| 100 : | , { "MMXEXT", XVID_CPU_MMXEXT | XVID_CPU_MMX } | ||
| 101 : | , { "SSE2 ", XVID_CPU_SSE2 | XVID_CPU_MMX } | ||
| 102 : | , { "3DNOW ", XVID_CPU_3DNOW } | ||
| 103 : | , { "3DNOWE", XVID_CPU_3DNOWEXT } | ||
| 104 : | ia64p | 257 | |
| 105 : | Isibaar | 225 | //, { "TSC ", XVID_CPU_TSC } |
| 106 : | , { 0, 0 } } | ||
| 107 : | |||
| 108 : | , cpu_short_list[] = | ||
| 109 : | { { "PLAINC", 0 } | ||
| 110 : | , { "MMX ", XVID_CPU_MMX } | ||
| 111 : | , { "MMXEXT", XVID_CPU_MMXEXT | XVID_CPU_MMX } | ||
| 112 : | ia64p | 257 | , { "IA64 ", XVID_CPU_IA64 } |
| 113 : | Isibaar | 225 | , { 0, 0 } } |
| 114 : | |||
| 115 : | , cpu_short_list2[] = | ||
| 116 : | { { "PLAINC", 0 } | ||
| 117 : | , { "MMX ", XVID_CPU_MMX } | ||
| 118 : | , { "SSE2 ", XVID_CPU_SSE2 | XVID_CPU_MMX } | ||
| 119 : | , { 0, 0 } }; | ||
| 120 : | |||
| 121 : | |||
| 122 : | int init_cpu(CPU *cpu) | ||
| 123 : | { | ||
| 124 : | int xerr, cpu_type; | ||
| 125 : | XVID_INIT_PARAM xinit; | ||
| 126 : | |||
| 127 : | cpu_type = check_cpu_features() & cpu->cpu; | ||
| 128 : | xinit.cpu_flags = cpu_type | XVID_CPU_FORCE; | ||
| 129 : | // xinit.cpu_flags = XVID_CPU_MMX | XVID_CPU_FORCE; | ||
| 130 : | xerr = xvid_init(NULL, 0, &xinit, NULL); | ||
| 131 : | if (cpu->cpu>0 && (cpu_type==0 || xerr!=XVID_ERR_OK)) { | ||
| 132 : | printf( "%s - skipped...\n", cpu->name ); | ||
| 133 : | return 0; | ||
| 134 : | } | ||
| 135 : | return 1; | ||
| 136 : | } | ||
| 137 : | |||
| 138 : | /********************************************************************* | ||
| 139 : | * test DCT | ||
| 140 : | *********************************************************************/ | ||
| 141 : | |||
| 142 : | #define ABS(X) ((X)<0 ? -(X) : (X)) | ||
| 143 : | |||
| 144 : | void test_dct() | ||
| 145 : | { | ||
| 146 : | const int nb_tests = 300*speed_ref; | ||
| 147 : | int tst; | ||
| 148 : | CPU *cpu; | ||
| 149 : | int i; | ||
| 150 : | short iDst0[8*8], iDst[8*8], fDst[8*8]; | ||
| 151 : | double overhead; | ||
| 152 : | |||
| 153 : | printf( "\n ===== test fdct/idct =====\n" ); | ||
| 154 : | |||
| 155 : | for(i=0; i<8*8; ++i) iDst0[i] = (i*7-i*i) & 0x7f; | ||
| 156 : | overhead = gettime_usec(); | ||
| 157 : | for(tst=0; tst<nb_tests; ++tst) | ||
| 158 : | { | ||
| 159 : | for(i=0; i<8*8; ++i) fDst[i] = iDst0[i]; | ||
| 160 : | for(i=0; i<8*8; ++i) iDst[i] = fDst[i]; | ||
| 161 : | } | ||
| 162 : | overhead = gettime_usec() - overhead; | ||
| 163 : | |||
| 164 : | for(cpu = cpu_list; cpu->name!=0; ++cpu) | ||
| 165 : | { | ||
| 166 : | double t; | ||
| 167 : | int iCrc, fCrc; | ||
| 168 : | |||
| 169 : | if (!init_cpu(cpu)) | ||
| 170 : | continue; | ||
| 171 : | |||
| 172 : | t = gettime_usec(); | ||
| 173 : | emms(); | ||
| 174 : | for(tst=0; tst<nb_tests; ++tst) | ||
| 175 : | { | ||
| 176 : | for(i=0; i<8*8; ++i) fDst[i] = iDst0[i]; | ||
| 177 : | fdct(fDst); | ||
| 178 : | for(i=0; i<8*8; ++i) iDst[i] = fDst[i]; | ||
| 179 : | idct(iDst); | ||
| 180 : | } | ||
| 181 : | emms(); | ||
| 182 : | t = (gettime_usec() - t - overhead) / nb_tests; | ||
| 183 : | iCrc=0; fCrc=0; | ||
| 184 : | for(i=0; i<8*8; ++i) { | ||
| 185 : | iCrc += ABS(iDst[i] - iDst0[i]); | ||
| 186 : | fCrc += fDst[i]^i; | ||
| 187 : | } | ||
| 188 : | printf( "%s - %.3f usec iCrc=%d fCrc=%d\n", | ||
| 189 : | cpu->name, t, iCrc, fCrc ); | ||
| 190 : | // the norm tolerates ~1 bit of diff per coeff | ||
| 191 : | if (ABS(iCrc)>=64) printf( "*** CRC ERROR! ***\n" ); | ||
| 192 : | } | ||
| 193 : | } | ||
| 194 : | |||
| 195 : | /********************************************************************* | ||
| 196 : | * test SAD | ||
| 197 : | *********************************************************************/ | ||
| 198 : | |||
| 199 : | void test_sad() | ||
| 200 : | { | ||
| 201 : | const int nb_tests = 2000*speed_ref; | ||
| 202 : | int tst; | ||
| 203 : | CPU *cpu; | ||
| 204 : | int i; | ||
| 205 : | uint8_t Cur[16*16], Ref1[16*16], Ref2[16*16]; | ||
| 206 : | |||
| 207 : | printf( "\n ====== test SAD ======\n" ); | ||
| 208 : | for(i=0; i<16*16;++i) { | ||
| 209 : | Cur[i] = (i/5) ^ 0x05; | ||
| 210 : | Ref1[i] = (i + 0x0b) & 0xff; | ||
| 211 : | Ref2[i] = i ^ 0x76; | ||
| 212 : | } | ||
| 213 : | |||
| 214 : | for(cpu = cpu_list; cpu->name!=0; ++cpu) | ||
| 215 : | { | ||
| 216 : | double t; | ||
| 217 : | uint32_t s; | ||
| 218 : | if (!init_cpu(cpu)) | ||
| 219 : | continue; | ||
| 220 : | |||
| 221 : | t = gettime_usec(); | ||
| 222 : | emms(); | ||
| 223 : | for(tst=0; tst<nb_tests; ++tst) s = sad8(Cur, Ref1, 16); | ||
| 224 : | emms(); | ||
| 225 : | t = (gettime_usec() - t) / nb_tests; | ||
| 226 : | printf( "%s - sad8 %.3f usec sad=%d\n", cpu->name, t, s ); | ||
| 227 : | if (s!=3776) printf( "*** CRC ERROR! ***\n" ); | ||
| 228 : | |||
| 229 : | t = gettime_usec(); | ||
| 230 : | emms(); | ||
| 231 : | for(tst=0; tst<nb_tests; ++tst) s = sad16(Cur, Ref1, 16, -1); | ||
| 232 : | emms(); | ||
| 233 : | t = (gettime_usec() - t) / nb_tests; | ||
| 234 : | printf( "%s - sad16 %.3f usec sad=%d\n", cpu->name, t, s ); | ||
| 235 : | if (s!=27214) printf( "*** CRC ERROR! ***\n" ); | ||
| 236 : | |||
| 237 : | t = gettime_usec(); | ||
| 238 : | emms(); | ||
| 239 : | for(tst=0; tst<nb_tests; ++tst) s = sad16bi(Cur, Ref1, Ref2, 16); | ||
| 240 : | emms(); | ||
| 241 : | t = (gettime_usec() - t) / nb_tests; | ||
| 242 : | printf( "%s - sad16bi %.3f usec sad=%d\n", cpu->name, t, s ); | ||
| 243 : | if (s!=26274) printf( "*** CRC ERROR! ***\n" ); | ||
| 244 : | |||
| 245 : | t = gettime_usec(); | ||
| 246 : | emms(); | ||
| 247 : | for(tst=0; tst<nb_tests; ++tst) s = dev16(Cur, 16); | ||
| 248 : | emms(); | ||
| 249 : | t = (gettime_usec() - t) / nb_tests; | ||
| 250 : | printf( "%s - dev16 %.3f usec sad=%d\n", cpu->name, t, s ); | ||
| 251 : | if (s!=3344) printf( "*** CRC ERROR! ***\n" ); | ||
| 252 : | |||
| 253 : | printf( " --- \n" ); | ||
| 254 : | } | ||
| 255 : | } | ||
| 256 : | |||
| 257 : | /********************************************************************* | ||
| 258 : | * test interpolation | ||
| 259 : | *********************************************************************/ | ||
| 260 : | |||
| 261 : | #define ENTER \ | ||
| 262 : | for(i=0; i<16*8; ++i) Dst[i] = 0; \ | ||
| 263 : | t = gettime_usec(); \ | ||
| 264 : | emms(); | ||
| 265 : | |||
| 266 : | #define LEAVE \ | ||
| 267 : | emms(); \ | ||
| 268 : | t = (gettime_usec() - t) / nb_tests; \ | ||
| 269 : | iCrc = 0; \ | ||
| 270 : | for(i=0; i<16*8; ++i) { iCrc += Dst[i]^i; } | ||
| 271 : | |||
| 272 : | #define TEST_MB(FUNC, R) \ | ||
| 273 : | ENTER \ | ||
| 274 : | for(tst=0; tst<nb_tests; ++tst) (FUNC)(Dst, Src0, 16, (R)); \ | ||
| 275 : | LEAVE | ||
| 276 : | |||
| 277 : | #define TEST_MB2(FUNC) \ | ||
| 278 : | ENTER \ | ||
| 279 : | for(tst=0; tst<nb_tests; ++tst) (FUNC)(Dst, Src0, 16); \ | ||
| 280 : | LEAVE | ||
| 281 : | |||
| 282 : | |||
| 283 : | void test_mb() | ||
| 284 : | { | ||
| 285 : | const int nb_tests = 2000*speed_ref; | ||
| 286 : | CPU *cpu; | ||
| 287 : | const uint8_t Src0[16*9] = { | ||
| 288 : | // try to have every possible combinaison of rounding... | ||
| 289 : | 0, 0, 1, 0, 2, 0, 3, 0, 4 ,0,0,0, 0,0,0,0 | ||
| 290 : | , 0, 1, 1, 1, 2, 1, 3, 1, 3 ,0,0,0, 0,0,0,0 | ||
| 291 : | , 0, 2, 1, 2, 2, 2, 3, 2, 2 ,0,0,0, 0,0,0,0 | ||
| 292 : | , 0, 3, 1, 3, 2, 3, 3, 3, 1 ,0,0,0, 0,0,0,0 | ||
| 293 : | , 1, 3, 0, 2, 1, 0, 2, 3, 4 ,0,0,0, 0,0,0,0 | ||
| 294 : | , 2, 2, 1, 2, 0, 1, 3, 5, 3 ,0,0,0, 0,0,0,0 | ||
| 295 : | , 3, 1, 2, 3, 1, 2, 2, 6, 2 ,0,0,0, 0,0,0,0 | ||
| 296 : | , 1, 0, 1, 3, 0, 3, 1, 6, 1 ,0,0,0, 0,0,0,0 | ||
| 297 : | , 4, 3, 2, 1, 2, 3, 4, 0, 3 ,0,0,0, 0,0,0,0 | ||
| 298 : | }; | ||
| 299 : | uint8_t Dst[16*8] = {0}; | ||
| 300 : | |||
| 301 : | printf( "\n === test block motion ===\n" ); | ||
| 302 : | |||
| 303 : | for(cpu = cpu_list; cpu->name!=0; ++cpu) | ||
| 304 : | { | ||
| 305 : | double t; | ||
| 306 : | int tst, i, iCrc; | ||
| 307 : | |||
| 308 : | if (!init_cpu(cpu)) | ||
| 309 : | continue; | ||
| 310 : | |||
| 311 : | TEST_MB(interpolate8x8_halfpel_h, 0); | ||
| 312 : | printf( "%s - interp- h-round0 %.3f usec iCrc=%d\n", cpu->name, t, iCrc ); | ||
| 313 : | if (iCrc!=8107) printf( "*** CRC ERROR! ***\n" ); | ||
| 314 : | |||
| 315 : | TEST_MB(interpolate8x8_halfpel_h, 1); | ||
| 316 : | printf( "%s - round1 %.3f usec iCrc=%d\n", cpu->name, t, iCrc ); | ||
| 317 : | if (iCrc!=8100) printf( "*** CRC ERROR! ***\n" ); | ||
| 318 : | |||
| 319 : | |||
| 320 : | TEST_MB(interpolate8x8_halfpel_v, 0); | ||
| 321 : | printf( "%s - interp- v-round0 %.3f usec iCrc=%d\n", cpu->name, t, iCrc ); | ||
| 322 : | if (iCrc!=8108) printf( "*** CRC ERROR! ***\n" ); | ||
| 323 : | |||
| 324 : | TEST_MB(interpolate8x8_halfpel_v, 1); | ||
| 325 : | printf( "%s - round1 %.3f usec iCrc=%d\n", cpu->name, t, iCrc ); | ||
| 326 : | if (iCrc!=8105) printf( "*** CRC ERROR! ***\n" ); | ||
| 327 : | |||
| 328 : | |||
| 329 : | TEST_MB(interpolate8x8_halfpel_hv, 0); | ||
| 330 : | printf( "%s - interp-hv-round0 %.3f usec iCrc=%d\n", cpu->name, t, iCrc ); | ||
| 331 : | if (iCrc!=8112) printf( "*** CRC ERROR! ***\n" ); | ||
| 332 : | |||
| 333 : | TEST_MB(interpolate8x8_halfpel_hv, 1); | ||
| 334 : | printf( "%s - round1 %.3f usec iCrc=%d\n", cpu->name, t, iCrc ); | ||
| 335 : | if (iCrc!=8103) printf( "*** CRC ERROR! ***\n" ); | ||
| 336 : | |||
| 337 : | printf( " --- \n" ); | ||
| 338 : | } | ||
| 339 : | } | ||
| 340 : | |||
| 341 : | /********************************************************************* | ||
| 342 : | * test transfer | ||
| 343 : | *********************************************************************/ | ||
| 344 : | |||
| 345 : | #define INIT_TRANSFER \ | ||
| 346 : | for(i=0; i<8*32; ++i) { \ | ||
| 347 : | Src8[i] = i; Src16[i] = i; \ | ||
| 348 : | Dst8[i] = 0; Dst16[i] = 0; \ | ||
| 349 : | Ref1[i] = i^0x27; \ | ||
| 350 : | Ref2[i] = i^0x51; \ | ||
| 351 : | } | ||
| 352 : | |||
| 353 : | #define TEST_TRANSFER_BEGIN(DST) \ | ||
| 354 : | INIT_TRANSFER \ | ||
| 355 : | overhead = -gettime_usec(); \ | ||
| 356 : | for(tst=0; tst<nb_tests; ++tst) { \ | ||
| 357 : | for(i=0; i<8*32; ++i) (DST)[i] = i^0x6a;\ | ||
| 358 : | } \ | ||
| 359 : | overhead += gettime_usec(); \ | ||
| 360 : | t = gettime_usec(); \ | ||
| 361 : | emms(); \ | ||
| 362 : | for(tst=0; tst<nb_tests; ++tst) { \ | ||
| 363 : | for(i=0; i<8*32; ++i) (DST)[i] = i^0x6a; | ||
| 364 : | |||
| 365 : | |||
| 366 : | #define TEST_TRANSFER_END(DST) \ | ||
| 367 : | } \ | ||
| 368 : | emms(); \ | ||
| 369 : | t = (gettime_usec()-t -overhead) / nb_tests;\ | ||
| 370 : | s = 0; for(i=0; i<8*32; ++i) { s += (DST)[i]^i; } | ||
| 371 : | |||
| 372 : | #define TEST_TRANSFER(FUNC, DST, SRC) \ | ||
| 373 : | TEST_TRANSFER_BEGIN(DST); \ | ||
| 374 : | (FUNC)((DST), (SRC), 32); \ | ||
| 375 : | TEST_TRANSFER_END(DST) | ||
| 376 : | |||
| 377 : | |||
| 378 : | #define TEST_TRANSFER2_BEGIN(DST, SRC) \ | ||
| 379 : | INIT_TRANSFER \ | ||
| 380 : | overhead = -gettime_usec(); \ | ||
| 381 : | for(tst=0; tst<nb_tests; ++tst) { \ | ||
| 382 : | for(i=0; i<8*32; ++i) (DST)[i] = i^0x6a;\ | ||
| 383 : | for(i=0; i<8*32; ++i) (SRC)[i] = i^0x3e;\ | ||
| 384 : | } \ | ||
| 385 : | overhead += gettime_usec(); \ | ||
| 386 : | t = gettime_usec(); \ | ||
| 387 : | emms(); \ | ||
| 388 : | for(tst=0; tst<nb_tests; ++tst) { \ | ||
| 389 : | for(i=0; i<8*32; ++i) (DST)[i] = i^0x6a;\ | ||
| 390 : | for(i=0; i<8*32; ++i) (SRC)[i] = i^0x3e; | ||
| 391 : | |||
| 392 : | #define TEST_TRANSFER2_END(DST) \ | ||
| 393 : | } \ | ||
| 394 : | emms(); \ | ||
| 395 : | t = (gettime_usec()-t -overhead) / nb_tests;\ | ||
| 396 : | s = 0; for(i=0; i<8*32; ++i) { s += (DST)[i]; } | ||
| 397 : | |||
| 398 : | #define TEST_TRANSFER2(FUNC, DST, SRC, R1) \ | ||
| 399 : | TEST_TRANSFER2_BEGIN(DST,SRC); \ | ||
| 400 : | (FUNC)((DST), (SRC), (R1), 32); \ | ||
| 401 : | TEST_TRANSFER2_END(DST) | ||
| 402 : | |||
| 403 : | #define TEST_TRANSFER3(FUNC, DST, SRC, R1, R2)\ | ||
| 404 : | TEST_TRANSFER_BEGIN(DST); \ | ||
| 405 : | (FUNC)((DST), (SRC), (R1), (R2), 32); \ | ||
| 406 : | TEST_TRANSFER_END(DST) | ||
| 407 : | |||
| 408 : | void test_transfer() | ||
| 409 : | { | ||
| 410 : | const int nb_tests = 4000*speed_ref; | ||
| 411 : | int i; | ||
| 412 : | CPU *cpu; | ||
| 413 : | uint8_t Src8[8*32], Dst8[8*32], Ref1[8*32], Ref2[8*32]; | ||
| 414 : | int16_t Src16[8*32], Dst16[8*32]; | ||
| 415 : | |||
| 416 : | printf( "\n === test transfer ===\n" ); | ||
| 417 : | |||
| 418 : | for(cpu = cpu_short_list; cpu->name!=0; ++cpu) | ||
| 419 : | { | ||
| 420 : | double t, overhead; | ||
| 421 : | int tst, s; | ||
| 422 : | |||
| 423 : | if (!init_cpu(cpu)) | ||
| 424 : | continue; | ||
| 425 : | |||
| 426 : | TEST_TRANSFER(transfer_8to16copy, Dst16, Src8); | ||
| 427 : | printf( "%s - 8to16 %.3f usec crc=%d\n", cpu->name, t, s ); | ||
| 428 : | if (s!=28288) printf( "*** CRC ERROR! ***\n" ); | ||
| 429 : | |||
| 430 : | TEST_TRANSFER(transfer_16to8copy, Dst8, Src16); | ||
| 431 : | printf( "%s - 16to8 %.3f usec crc=%d\n", cpu->name, t, s ); | ||
| 432 : | if (s!=28288) printf( "*** CRC ERROR! ***\n" ); | ||
| 433 : | |||
| 434 : | TEST_TRANSFER(transfer8x8_copy, Dst8, Src8); | ||
| 435 : | printf( "%s - 8to8 %.3f usec crc=%d\n", cpu->name, t, s ); | ||
| 436 : | if (s!=20352) printf( "*** CRC ERROR! ***\n" ); | ||
| 437 : | |||
| 438 : | TEST_TRANSFER(transfer_16to8add, Dst8, Src16); | ||
| 439 : | printf( "%s - 16to8add %.3f usec crc=%d\n", cpu->name, t, s ); | ||
| 440 : | if (s!=25536) printf( "*** CRC ERROR! ***\n" ); | ||
| 441 : | |||
| 442 : | TEST_TRANSFER2(transfer_8to16sub, Dst16, Src8, Ref1); | ||
| 443 : | printf( "%s - 8to16sub %.3f usec crc1=%d ", cpu->name, t, s ); | ||
| 444 : | if (s!=28064) printf( "*** CRC ERROR! ***\n" ); | ||
| 445 : | s = 0; for(i=0; i<8*32; ++i) { s += (Src8[i]-Ref1[i])&i; } | ||
| 446 : | printf( "crc2=%d\n", s); | ||
| 447 : | if (s!=16256) printf( "*** CRC ERROR! ***\n" ); | ||
| 448 : | |||
| 449 : | TEST_TRANSFER3(transfer_8to16sub2, Dst16, Src8, Ref1, Ref2); | ||
| 450 : | printf( "%s - 8to16sub2 %.3f usec crc=%d\n", cpu->name, t, s ); | ||
| 451 : | if (s!=20384) printf( "*** CRC ERROR! ***\n" ); | ||
| 452 : | |||
| 453 : | printf( " --- \n" ); | ||
| 454 : | } | ||
| 455 : | } | ||
| 456 : | |||
| 457 : | /********************************************************************* | ||
| 458 : | * test quantization | ||
| 459 : | *********************************************************************/ | ||
| 460 : | |||
| 461 : | #define TEST_QUANT(FUNC, DST, SRC) \ | ||
| 462 : | t = gettime_usec(); \ | ||
| 463 : | emms(); \ | ||
| 464 : | for(tst=0; tst<nb_tests; ++tst) \ | ||
| 465 : | for(s=0, q=1; q<=max_Q; ++q) { \ | ||
| 466 : | (FUNC)((DST), (SRC), q); \ | ||
| 467 : | for(i=0; i<64; ++i) s+=(DST)[i]^i; \ | ||
| 468 : | } \ | ||
| 469 : | emms(); \ | ||
| 470 : | t = (gettime_usec()-t-overhead)/nb_tests; | ||
| 471 : | |||
| 472 : | #define TEST_QUANT2(FUNC, DST, SRC, MULT) \ | ||
| 473 : | t = gettime_usec(); \ | ||
| 474 : | emms(); \ | ||
| 475 : | for(tst=0; tst<nb_tests; ++tst) \ | ||
| 476 : | for(s=0, q=1; q<=max_Q; ++q) { \ | ||
| 477 : | (FUNC)((DST), (SRC), q, MULT); \ | ||
| 478 : | for(i=0; i<64; ++i) s+=(DST)[i]^i; \ | ||
| 479 : | } \ | ||
| 480 : | emms(); \ | ||
| 481 : | t = (gettime_usec()-t-overhead)/nb_tests; | ||
| 482 : | |||
| 483 : | void test_quant() | ||
| 484 : | { | ||
| 485 : | const int nb_tests = 150*speed_ref; | ||
| 486 : | const int max_Q = 31; | ||
| 487 : | int i; | ||
| 488 : | CPU *cpu; | ||
| 489 : | int16_t Src[8*8], Dst[8*8]; | ||
| 490 : | |||
| 491 : | printf( "\n ===== test quant =====\n" ); | ||
| 492 : | |||
| 493 : | for(i=0; i<64; ++i) { | ||
| 494 : | Src[i] = i-32; | ||
| 495 : | Dst[i] = 0; | ||
| 496 : | } | ||
| 497 : | |||
| 498 : | |||
| 499 : | for(cpu = cpu_short_list; cpu->name!=0; ++cpu) | ||
| 500 : | { | ||
| 501 : | double t, overhead; | ||
| 502 : | int tst, s, q; | ||
| 503 : | |||
| 504 : | if (!init_cpu(cpu)) | ||
| 505 : | continue; | ||
| 506 : | |||
| 507 : | set_inter_matrix( get_default_inter_matrix() ); | ||
| 508 : | set_intra_matrix( get_default_intra_matrix() ); | ||
| 509 : | overhead = -gettime_usec(); | ||
| 510 : | for(tst=0; tst<nb_tests; ++tst) | ||
| 511 : | for(s=0, q=1; q<=max_Q; ++q) | ||
| 512 : | for(i=0; i<64; ++i) s+=Dst[i]^i; | ||
| 513 : | overhead += gettime_usec(); | ||
| 514 : | |||
| 515 : | TEST_QUANT2(quant4_intra, Dst, Src, 7); | ||
| 516 : | printf( "%s - quant4_intra %.3f usec crc=%d\n", cpu->name, t, s ); | ||
| 517 : | if (s!=55827) printf( "*** CRC ERROR! ***\n" ); | ||
| 518 : | |||
| 519 : | TEST_QUANT(quant4_inter, Dst, Src); | ||
| 520 : | printf( "%s - quant4_inter %.3f usec crc=%d\n", cpu->name, t, s ); | ||
| 521 : | if (s!=58201) printf( "*** CRC ERROR! ***\n" ); | ||
| 522 : | |||
| 523 : | |||
| 524 : | TEST_QUANT2(dequant4_intra, Dst, Src, 7); | ||
| 525 : | printf( "%s - dequant4_intra %.3f usec crc=%d\n", cpu->name, t, s ); | ||
| 526 : | if (s!=193340) printf( "*** CRC ERROR! ***\n" ); | ||
| 527 : | |||
| 528 : | TEST_QUANT(dequant4_inter, Dst, Src); | ||
| 529 : | printf( "%s - dequant4_inter %.3f usec crc=%d\n", cpu->name, t, s ); | ||
| 530 : | if (s!=116483) printf( "*** CRC ERROR! ***\n" ); | ||
| 531 : | |||
| 532 : | TEST_QUANT2(quant_intra, Dst, Src, 7); | ||
| 533 : | printf( "%s - quant_intra %.3f usec crc=%d\n", cpu->name, t, s ); | ||
| 534 : | if (s!=56885) printf( "*** CRC ERROR! ***\n" ); | ||
| 535 : | |||
| 536 : | TEST_QUANT(quant_inter, Dst, Src); | ||
| 537 : | printf( "%s - quant_inter %.3f usec crc=%d\n", cpu->name, t, s ); | ||
| 538 : | if (s!=58056) printf( "*** CRC ERROR! ***\n" ); | ||
| 539 : | |||
| 540 : | TEST_QUANT2(dequant_intra, Dst, Src, 7); | ||
| 541 : | printf( "%s - dequant_intra %.3f usec crc=%d\n", cpu->name, t, s ); | ||
| 542 : | if (s!=-7936) printf( "*** CRC ERROR! ***\n" ); | ||
| 543 : | |||
| 544 : | TEST_QUANT(dequant_inter, Dst, Src); | ||
| 545 : | printf( "%s - dequant_inter %.3f usec crc=%d\n", cpu->name, t, s ); | ||
| 546 : | // { int k,l; for(k=0; k<8; ++k) { for(l=0; l<8; ++l) printf( "[%.4d]", Dst[k*8+l]); printf("\n"); } } | ||
| 547 : | if (s!=-33217) printf( "*** CRC ERROR! ***\n" ); | ||
| 548 : | |||
| 549 : | printf( " --- \n" ); | ||
| 550 : | } | ||
| 551 : | } | ||
| 552 : | |||
| 553 : | /********************************************************************* | ||
| 554 : | * test non-zero AC counting | ||
| 555 : | *********************************************************************/ | ||
| 556 : | |||
| 557 : | #define TEST_CBP(FUNC, SRC) \ | ||
| 558 : | t = gettime_usec(); \ | ||
| 559 : | emms(); \ | ||
| 560 : | for(tst=0; tst<nb_tests; ++tst) { \ | ||
| 561 : | cbp = (FUNC)((SRC)); \ | ||
| 562 : | } \ | ||
| 563 : | emms(); \ | ||
| 564 : | t = (gettime_usec()-t ) / nb_tests; | ||
| 565 : | |||
| 566 : | void test_cbp() | ||
| 567 : | { | ||
| 568 : | const int nb_tests = 10000*speed_ref; | ||
| 569 : | int i; | ||
| 570 : | CPU *cpu; | ||
| 571 : | int16_t Src1[6*64], Src2[6*64], Src3[6*64], Src4[6*64]; | ||
| 572 : | |||
| 573 : | printf( "\n ===== test cbp =====\n" ); | ||
| 574 : | |||
| 575 : | for(i=0; i<6*64; ++i) { | ||
| 576 : | Src1[i] = (i*i*3/8192)&(i/64)&1; // 'random' | ||
| 577 : | Src2[i] = (i<3*64); // half-full | ||
| 578 : | Src3[i] = ((i+32)>3*64); | ||
| 579 : | Src4[i] = (i==(3*64+2) || i==(5*64+9)); | ||
| 580 : | } | ||
| 581 : | |||
| 582 : | for(cpu = cpu_short_list2; cpu->name!=0; ++cpu) | ||
| 583 : | { | ||
| 584 : | double t; | ||
| 585 : | int tst, cbp; | ||
| 586 : | |||
| 587 : | if (!init_cpu(cpu)) | ||
| 588 : | continue; | ||
| 589 : | |||
| 590 : | TEST_CBP(calc_cbp, Src1); | ||
| 591 : | printf( "%s - calc_cbp#1 %.3f usec cbp=0x%x\n", cpu->name, t, cbp ); | ||
| 592 : | if (cbp!=0x15) printf( "*** CRC ERROR! ***\n" ); | ||
| 593 : | TEST_CBP(calc_cbp, Src2); | ||
| 594 : | printf( "%s - calc_cbp#2 %.3f usec cbp=0x%x\n", cpu->name, t, cbp ); | ||
| 595 : | if (cbp!=0x38) printf( "*** CRC ERROR! ***\n" ); | ||
| 596 : | TEST_CBP(calc_cbp, Src3); | ||
| 597 : | printf( "%s - calc_cbp#3 %.3f usec cbp=0x%x\n", cpu->name, t, cbp ); | ||
| 598 : | if (cbp!=0x0f) printf( "*** CRC ERROR! ***\n" ); | ||
| 599 : | TEST_CBP(calc_cbp, Src4); | ||
| 600 : | printf( "%s - calc_cbp#4 %.3f usec cbp=0x%x\n", cpu->name, t, cbp ); | ||
| 601 : | if (cbp!=0x05) printf( "*** CRC ERROR! ***\n" ); | ||
| 602 : | printf( " --- \n" ); | ||
| 603 : | } | ||
| 604 : | } | ||
| 605 : | |||
| 606 : | /********************************************************************* | ||
| 607 : | * measure raw decoding speed | ||
| 608 : | *********************************************************************/ | ||
| 609 : | |||
| 610 : | void test_dec(const char *name, int width, int height, int with_chksum) | ||
| 611 : | { | ||
| 612 : | FILE *f = 0; | ||
| 613 : | void *dechandle = 0; | ||
| 614 : | int xerr; | ||
| 615 : | XVID_INIT_PARAM xinit; | ||
| 616 : | XVID_DEC_PARAM xparam; | ||
| 617 : | XVID_DEC_FRAME xframe; | ||
| 618 : | double t = 0.; | ||
| 619 : | int nb = 0; | ||
| 620 : | uint8_t *buf = 0; | ||
| 621 : | uint8_t *rgb_out = 0; | ||
| 622 : | int buf_size, pos; | ||
| 623 : | uint32_t chksum = 0; | ||
| 624 : | |||
| 625 : | xinit.cpu_flags = 0; | ||
| 626 : | xvid_init(NULL, 0, &xinit, NULL); | ||
| 627 : | printf( "API version: %d, core build:%d\n", xinit.api_version, xinit.core_build); | ||
| 628 : | |||
| 629 : | |||
| 630 : | xparam.width = width; | ||
| 631 : | xparam.height = height; | ||
| 632 : | xerr = xvid_decore(NULL, XVID_DEC_CREATE, &xparam, NULL); | ||
| 633 : | if (xerr!=XVID_ERR_OK) { | ||
| 634 : | printf("can't init decoder (err=%d)\n", xerr); | ||
| 635 : | return; | ||
| 636 : | } | ||
| 637 : | dechandle = xparam.handle; | ||
| 638 : | |||
| 639 : | |||
| 640 : | f = fopen(name, "rb"); | ||
| 641 : | if (f==0) { | ||
| 642 : | printf( "can't open file '%s'\n", name); | ||
| 643 : | return; | ||
| 644 : | } | ||
| 645 : | fseek(f, 0, SEEK_END); | ||
| 646 : | buf_size = ftell(f); | ||
| 647 : | fseek(f, 0, SEEK_SET); | ||
| 648 : | if (buf_size<=0) { | ||
| 649 : | printf("error while stating file\n"); | ||
| 650 : | goto End; | ||
| 651 : | } | ||
| 652 : | else printf( "Input size: %d\n", buf_size); | ||
| 653 : | |||
| 654 : | buf = malloc(buf_size); // should be enuf' | ||
| 655 : | rgb_out = calloc(4, width*height); // <-room for _RGB24 | ||
| 656 : | if (buf==0 || rgb_out==0) { | ||
| 657 : | printf( "malloc failed!\n" ); | ||
| 658 : | goto End; | ||
| 659 : | } | ||
| 660 : | |||
| 661 : | if (fread(buf, buf_size, 1, f)!=1) { | ||
| 662 : | printf( "file-read failed\n" ); | ||
| 663 : | goto End; | ||
| 664 : | } | ||
| 665 : | |||
| 666 : | nb = 0; | ||
| 667 : | pos = 0; | ||
| 668 : | t = -gettime_usec(); | ||
| 669 : | while(1) { | ||
| 670 : | xframe.bitstream = buf + pos; | ||
| 671 : | xframe.length = buf_size - pos; | ||
| 672 : | xframe.image = rgb_out; | ||
| 673 : | xframe.stride = width; | ||
| 674 : | xframe.colorspace = XVID_CSP_RGB24; | ||
| 675 : | xerr = xvid_decore(dechandle, XVID_DEC_DECODE, &xframe, 0); | ||
| 676 : | nb++; | ||
| 677 : | pos += xframe.length; | ||
| 678 : | if (with_chksum) { | ||
| 679 : | int k = width*height; | ||
| 680 : | uint32_t *ptr = (uint32_t *)rgb_out; | ||
| 681 : | while(k-->0) chksum += *ptr++; | ||
| 682 : | } | ||
| 683 : | if (pos==buf_size) | ||
| 684 : | break; | ||
| 685 : | if (xerr!=XVID_ERR_OK) { | ||
| 686 : | printf("decoding failed for frame #%d (err=%d)!\n", nb, xerr); | ||
| 687 : | break; | ||
| 688 : | } | ||
| 689 : | } | ||
| 690 : | t += gettime_usec(); | ||
| 691 : | if (t>0.) | ||
| 692 : | printf( "%d frames decoded in %.3f s -> %.1f FPS\n", nb, t*1.e-6f, (float)(nb*1.e6f/t) ); | ||
| 693 : | if (with_chksum) | ||
| 694 : | printf("checksum: 0x%.8x\n", chksum); | ||
| 695 : | |||
| 696 : | End: | ||
| 697 : | if (rgb_out!=0) free(rgb_out); | ||
| 698 : | if (buf!=0) free(buf); | ||
| 699 : | if (dechandle!=0) { | ||
| 700 : | xerr= xvid_decore(dechandle, XVID_DEC_DESTROY, NULL, NULL); | ||
| 701 : | if (xerr!=XVID_ERR_OK) | ||
| 702 : | printf("destroy-decoder failed (err=%d)!\n", xerr); | ||
| 703 : | } | ||
| 704 : | if (f!=0) fclose(f); | ||
| 705 : | } | ||
| 706 : | |||
| 707 : | /********************************************************************* | ||
| 708 : | * non-regression tests | ||
| 709 : | *********************************************************************/ | ||
| 710 : | |||
| 711 : | void test_bugs1() | ||
| 712 : | { | ||
| 713 : | CPU *cpu; | ||
| 714 : | |||
| 715 : | printf( "\n ===== (de)quant4_intra saturation bug? =====\n" ); | ||
| 716 : | |||
| 717 : | for(cpu = cpu_short_list; cpu->name!=0; ++cpu) | ||
| 718 : | { | ||
| 719 : | int i; | ||
| 720 : | int16_t Src[8*8], Dst[8*8]; | ||
| 721 : | |||
| 722 : | if (!init_cpu(cpu)) | ||
| 723 : | continue; | ||
| 724 : | |||
| 725 : | for(i=0; i<64; ++i) Src[i] = i-32; | ||
| 726 : | set_intra_matrix( get_default_intra_matrix() ); | ||
| 727 : | dequant4_intra(Dst, Src, 32, 5); | ||
| 728 : | printf( "dequant4_intra with CPU=%s: ", cpu->name); | ||
| 729 : | printf( " Out[]= " ); | ||
| 730 : | for(i=0; i<64; ++i) printf( "[%d]", Dst[i]); | ||
| 731 : | printf( "\n" ); | ||
| 732 : | } | ||
| 733 : | |||
| 734 : | printf( "\n ===== (de)quant4_inter saturation bug? =====\n" ); | ||
| 735 : | |||
| 736 : | for(cpu = cpu_short_list; cpu->name!=0; ++cpu) | ||
| 737 : | { | ||
| 738 : | int i; | ||
| 739 : | int16_t Src[8*8], Dst[8*8]; | ||
| 740 : | |||
| 741 : | if (!init_cpu(cpu)) | ||
| 742 : | continue; | ||
| 743 : | |||
| 744 : | for(i=0; i<64; ++i) Src[i] = i-32; | ||
| 745 : | set_inter_matrix( get_default_inter_matrix() ); | ||
| 746 : | dequant4_inter(Dst, Src, 32); | ||
| 747 : | printf( "dequant4_inter with CPU=%s: ", cpu->name); | ||
| 748 : | printf( " Out[]= " ); | ||
| 749 : | for(i=0; i<64; ++i) printf( "[%d]", Dst[i]); | ||
| 750 : | printf( "\n" ); | ||
| 751 : | } | ||
| 752 : | } | ||
| 753 : | |||
| 754 : | void test_dct_precision_diffs() | ||
| 755 : | { | ||
| 756 : | CPU *cpu; | ||
| 757 : | short Blk[8*8], Blk0[8*8]; | ||
| 758 : | |||
| 759 : | printf( "\n ===== fdct/idct saturation diffs =====\n" ); | ||
| 760 : | |||
| 761 : | for(cpu = cpu_short_list; cpu->name!=0; ++cpu) | ||
| 762 : | { | ||
| 763 : | int i; | ||
| 764 : | |||
| 765 : | if (!init_cpu(cpu)) | ||
| 766 : | continue; | ||
| 767 : | |||
| 768 : | for(i=0; i<8*8; ++i) { | ||
| 769 : | Blk0[i] = (i*7-i*i) & 0x7f; | ||
| 770 : | Blk[i] = Blk0[i]; | ||
| 771 : | } | ||
| 772 : | |||
| 773 : | fdct(Blk); | ||
| 774 : | idct(Blk); | ||
| 775 : | printf( " fdct+idct diffs with CPU=%s: \n", cpu->name ); | ||
| 776 : | for(i=0; i<8; ++i) { | ||
| 777 : | int j; | ||
| 778 : | for(j=0; j<8; ++j) printf( " %d ", Blk[i*8+j]-Blk0[i*8+j]); | ||
| 779 : | printf("\n"); | ||
| 780 : | } | ||
| 781 : | printf("\n"); | ||
| 782 : | } | ||
| 783 : | } | ||
| 784 : | |||
| 785 : | |||
| 786 : | /********************************************************************* | ||
| 787 : | * main | ||
| 788 : | *********************************************************************/ | ||
| 789 : | |||
| 790 : | int main(int argc, char *argv[]) | ||
| 791 : | { | ||
| 792 : | int what = 0; | ||
| 793 : | if (argc>1) what = atoi(argv[1]); | ||
| 794 : | if (what==0 || what==1) test_dct(); | ||
| 795 : | if (what==0 || what==2) test_mb(); | ||
| 796 : | if (what==0 || what==3) test_sad(); | ||
| 797 : | if (what==0 || what==4) test_transfer(); | ||
| 798 : | if (what==0 || what==5) test_quant(); | ||
| 799 : | if (what==0 || what==6) test_cbp(); | ||
| 800 : | |||
| 801 : | if (what==8) { | ||
| 802 : | int width, height; | ||
| 803 : | if (argc<5) { | ||
| 804 : | printf("usage: %s %d [bitstream] [width] [height]\n", argv[0], what); | ||
| 805 : | return 1; | ||
| 806 : | } | ||
| 807 : | width = atoi(argv[3]); | ||
| 808 : | height = atoi(argv[4]); | ||
| 809 : | test_dec(argv[2], width, height, (argc>5)); | ||
| 810 : | } | ||
| 811 : | |||
| 812 : | if (what==-1) { | ||
| 813 : | test_bugs1(); | ||
| 814 : | test_dct_precision_diffs(); | ||
| 815 : | } | ||
| 816 : | return 0; | ||
| 817 : | } | ||
| 818 : | |||
| 819 : | /********************************************************************* | ||
| 820 : | * 'Reference' output (except for timing) on a PIII 1.13Ghz/linux | ||
| 821 : | *********************************************************************/ | ||
| 822 : | /* | ||
| 823 : | |||
| 824 : | ===== test fdct/idct ===== | ||
| 825 : | PLAINC - 2.631 usec iCrc=3 fCrc=-85 | ||
| 826 : | MMX - 0.596 usec iCrc=3 fCrc=-67 | ||
| 827 : | MMXEXT - 0.608 usec iCrc=3 fCrc=-67 | ||
| 828 : | SSE2 - 0.605 usec iCrc=3 fCrc=-67 | ||
| 829 : | 3DNOW - skipped... | ||
| 830 : | 3DNOWE - skipped... | ||
| 831 : | |||
| 832 : | === test block motion === | ||
| 833 : | PLAINC - interp- h-round0 1.031 usec iCrc=8107 | ||
| 834 : | PLAINC - round1 1.022 usec iCrc=8100 | ||
| 835 : | PLAINC - interp- v-round0 1.002 usec iCrc=8108 | ||
| 836 : | PLAINC - round1 1.011 usec iCrc=8105 | ||
| 837 : | PLAINC - interp-hv-round0 1.623 usec iCrc=8112 | ||
| 838 : | PLAINC - round1 1.621 usec iCrc=8103 | ||
| 839 : | PLAINC - interpolate8x8_c 0.229 usec iCrc=8107 | ||
| 840 : | --- | ||
| 841 : | MMX - interp- h-round0 0.105 usec iCrc=8107 | ||
| 842 : | MMX - round1 0.105 usec iCrc=8100 | ||
| 843 : | MMX - interp- v-round0 0.106 usec iCrc=8108 | ||
| 844 : | MMX - round1 0.107 usec iCrc=8105 | ||
| 845 : | MMX - interp-hv-round0 0.145 usec iCrc=8112 | ||
| 846 : | MMX - round1 0.145 usec iCrc=8103 | ||
| 847 : | MMX - interpolate8x8_c 0.229 usec iCrc=8107 | ||
| 848 : | --- | ||
| 849 : | MMXEXT - interp- h-round0 0.027 usec iCrc=8107 | ||
| 850 : | MMXEXT - round1 0.041 usec iCrc=8100 | ||
| 851 : | MMXEXT - interp- v-round0 0.027 usec iCrc=8108 | ||
| 852 : | MMXEXT - round1 0.040 usec iCrc=8105 | ||
| 853 : | MMXEXT - interp-hv-round0 0.070 usec iCrc=8112 | ||
| 854 : | MMXEXT - round1 0.066 usec iCrc=8103 | ||
| 855 : | MMXEXT - interpolate8x8_c 0.027 usec iCrc=8107 | ||
| 856 : | --- | ||
| 857 : | SSE2 - interp- h-round0 0.106 usec iCrc=8107 | ||
| 858 : | SSE2 - round1 0.105 usec iCrc=8100 | ||
| 859 : | SSE2 - interp- v-round0 0.106 usec iCrc=8108 | ||
| 860 : | SSE2 - round1 0.106 usec iCrc=8105 | ||
| 861 : | SSE2 - interp-hv-round0 0.145 usec iCrc=8112 | ||
| 862 : | SSE2 - round1 0.145 usec iCrc=8103 | ||
| 863 : | SSE2 - interpolate8x8_c 0.237 usec iCrc=8107 | ||
| 864 : | --- | ||
| 865 : | 3DNOW - skipped... | ||
| 866 : | 3DNOWE - skipped... | ||
| 867 : | |||
| 868 : | ====== test SAD ====== | ||
| 869 : | PLAINC - sad8 0.296 usec sad=3776 | ||
| 870 : | PLAINC - sad16 1.599 usec sad=27214 | ||
| 871 : | PLAINC - sad16bi 2.350 usec sad=26274 | ||
| 872 : | PLAINC - dev16 1.610 usec sad=3344 | ||
| 873 : | --- | ||
| 874 : | MMX - sad8 0.057 usec sad=3776 | ||
| 875 : | MMX - sad16 0.178 usec sad=27214 | ||
| 876 : | MMX - sad16bi 2.381 usec sad=26274 | ||
| 877 : | MMX - dev16 0.312 usec sad=3344 | ||
| 878 : | --- | ||
| 879 : | MMXEXT - sad8 0.036 usec sad=3776 | ||
| 880 : | MMXEXT - sad16 0.106 usec sad=27214 | ||
| 881 : | MMXEXT - sad16bi 0.182 usec sad=26274 | ||
| 882 : | MMXEXT - dev16 0.193 usec sad=3344 | ||
| 883 : | --- | ||
| 884 : | SSE2 - sad8 0.057 usec sad=3776 | ||
| 885 : | SSE2 - sad16 0.178 usec sad=27214 | ||
| 886 : | SSE2 - sad16bi 2.427 usec sad=26274 | ||
| 887 : | SSE2 - dev16 0.313 usec sad=3344 | ||
| 888 : | --- | ||
| 889 : | 3DNOW - skipped... | ||
| 890 : | 3DNOWE - skipped... | ||
| 891 : | |||
| 892 : | === test transfer === | ||
| 893 : | PLAINC - 8to16 0.124 usec crc=28288 | ||
| 894 : | PLAINC - 16to8 0.753 usec crc=28288 | ||
| 895 : | PLAINC - 8to8 0.041 usec crc=20352 | ||
| 896 : | PLAINC - 16to8add 0.916 usec crc=25536 | ||
| 897 : | PLAINC - 8to16sub 0.812 usec crc1=28064 crc2=16256 | ||
| 898 : | PLAINC - 8to16sub2 0.954 usec crc=20384 | ||
| 899 : | --- | ||
| 900 : | MMX - 8to16 0.037 usec crc=28288 | ||
| 901 : | MMX - 16to8 0.016 usec crc=28288 | ||
| 902 : | MMX - 8to8 0.018 usec crc=20352 | ||
| 903 : | MMX - 16to8add 0.044 usec crc=25536 | ||
| 904 : | MMX - 8to16sub 0.065 usec crc1=28064 crc2=16256 | ||
| 905 : | MMX - 8to16sub2 0.110 usec crc=20384 | ||
| 906 : | --- | ||
| 907 : | MMXEXT - 8to16 0.032 usec crc=28288 | ||
| 908 : | MMXEXT - 16to8 0.023 usec crc=28288 | ||
| 909 : | MMXEXT - 8to8 0.018 usec crc=20352 | ||
| 910 : | MMXEXT - 16to8add 0.041 usec crc=25536 | ||
| 911 : | MMXEXT - 8to16sub 0.065 usec crc1=28064 crc2=16256 | ||
| 912 : | MMXEXT - 8to16sub2 0.069 usec crc=20384 | ||
| 913 : | --- | ||
| 914 : | |||
| 915 : | ===== test quant ===== | ||
| 916 : | PLAINC - quant4_intra 78.889 usec crc=55827 | ||
| 917 : | PLAINC - quant4_inter 71.957 usec crc=58201 | ||
| 918 : | PLAINC - dequant4_intra 34.968 usec crc=193340 | ||
| 919 : | PLAINC - dequant4_inter 40.792 usec crc=116483 | ||
| 920 : | PLAINC - quant_intra 30.845 usec crc=56885 | ||
| 921 : | PLAINC - quant_inter 34.842 usec crc=58056 | ||
| 922 : | PLAINC - dequant_intra 33.211 usec crc=-7936 | ||
| 923 : | PLAINC - dequant_inter 45.486 usec crc=-33217 | ||
| 924 : | --- | ||
| 925 : | MMX - quant4_intra 9.030 usec crc=55827 | ||
| 926 : | MMX - quant4_inter 8.234 usec crc=58201 | ||
| 927 : | MMX - dequant4_intra 18.330 usec crc=193340 | ||
| 928 : | MMX - dequant4_inter 19.181 usec crc=116483 | ||
| 929 : | MMX - quant_intra 7.124 usec crc=56885 | ||
| 930 : | MMX - quant_inter 6.861 usec crc=58056 | ||
| 931 : | MMX - dequant_intra 9.048 usec crc=-7936 | ||
| 932 : | MMX - dequant_inter 8.203 usec crc=-33217 | ||
| 933 : | --- | ||
| 934 : | MMXEXT - quant4_intra 9.045 usec crc=55827 | ||
| 935 : | MMXEXT - quant4_inter 8.232 usec crc=58201 | ||
| 936 : | MMXEXT - dequant4_intra 18.250 usec crc=193340 | ||
| 937 : | MMXEXT - dequant4_inter 19.256 usec crc=116483 | ||
| 938 : | MMXEXT - quant_intra 7.121 usec crc=56885 | ||
| 939 : | MMXEXT - quant_inter 6.855 usec crc=58056 | ||
| 940 : | MMXEXT - dequant_intra 9.034 usec crc=-7936 | ||
| 941 : | MMXEXT - dequant_inter 8.202 usec crc=-33217 | ||
| 942 : | --- | ||
| 943 : | |||
| 944 : | ===== test cbp ===== | ||
| 945 : | PLAINC - calc_cbp#1 0.545 usec cbp=0x15 | ||
| 946 : | PLAINC - calc_cbp#2 0.540 usec cbp=0x38 | ||
| 947 : | PLAINC - calc_cbp#3 0.477 usec cbp=0xf | ||
| 948 : | PLAINC - calc_cbp#4 0.739 usec cbp=0x5 | ||
| 949 : | --- | ||
| 950 : | MMX - calc_cbp#1 0.136 usec cbp=0x15 | ||
| 951 : | MMX - calc_cbp#2 0.131 usec cbp=0x38 | ||
| 952 : | MMX - calc_cbp#3 0.132 usec cbp=0xf | ||
| 953 : | MMX - calc_cbp#4 0.135 usec cbp=0x5 | ||
| 954 : | --- | ||
| 955 : | SSE2 - calc_cbp#1 0.135 usec cbp=0x15 | ||
| 956 : | SSE2 - calc_cbp#2 0.131 usec cbp=0x38 | ||
| 957 : | SSE2 - calc_cbp#3 0.134 usec cbp=0xf | ||
| 958 : | SSE2 - calc_cbp#4 0.136 usec cbp=0x5 | ||
| 959 : | --- | ||
| 960 : | */ |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |