Mercurial > libavcodec.hg
annotate dct-test.c @ 3592:6a358dccf2ab libavcodec
SIMD vector optimizations. 3% faster overall decoding.
author | banan |
---|---|
date | Sat, 19 Aug 2006 08:46:52 +0000 |
parents | e0927bc44a10 |
children | c537a97eec66 |
rev | line source |
---|---|
1106 | 1 /** |
2 * @file dct-test.c | |
2967 | 3 * DCT test. (c) 2001 Fabrice Bellard. |
1106 | 4 * Started from sample code by Juan J. Sierralta P. |
5 */ | |
6 | |
0 | 7 #include <stdlib.h> |
8 #include <stdio.h> | |
9 #include <string.h> | |
10 #include <sys/time.h> | |
11 #include <unistd.h> | |
12 | |
13 #include "dsputil.h" | |
14 | |
633 | 15 #include "simple_idct.h" |
1557 | 16 #include "faandct.h" |
33 | 17 |
980 | 18 #ifndef MAX |
19 #define MAX(a, b) (((a) > (b)) ? (a) : (b)) | |
20 #endif | |
21 | |
2872 | 22 #undef printf |
23 | |
24 void *fast_memcpy(void *a, const void *b, size_t c){return memcpy(a,b,c);}; | |
25 | |
33 | 26 /* reference fdct/idct */ |
0 | 27 extern void fdct(DCTELEM *block); |
33 | 28 extern void idct(DCTELEM *block); |
2872 | 29 extern void ff_idct_xvid_mmx(DCTELEM *block); |
30 extern void ff_idct_xvid_mmx2(DCTELEM *block); | |
0 | 31 extern void init_fdct(); |
32 | |
33 | 33 extern void j_rev_dct(DCTELEM *data); |
34 extern void ff_mmx_idct(DCTELEM *data); | |
35 extern void ff_mmxext_idct(DCTELEM *data); | |
36 | |
633 | 37 extern void odivx_idct_c (short *block); |
38 | |
0 | 39 #define AANSCALE_BITS 12 |
40 static const unsigned short aanscales[64] = { | |
41 /* precomputed values scaled up by 14 bits */ | |
42 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
43 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270, | |
44 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906, | |
45 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315, | |
46 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
47 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552, | |
48 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446, | |
49 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247 | |
50 }; | |
51 | |
1064 | 52 uint8_t cropTbl[256 + 2 * MAX_NEG_CROP]; |
633 | 53 |
1064 | 54 int64_t gettime(void) |
0 | 55 { |
56 struct timeval tv; | |
57 gettimeofday(&tv,NULL); | |
1064 | 58 return (int64_t)tv.tv_sec * 1000000 + tv.tv_usec; |
0 | 59 } |
60 | |
61 #define NB_ITS 20000 | |
62 #define NB_ITS_SPEED 50000 | |
63 | |
33 | 64 static short idct_mmx_perm[64]; |
65 | |
633 | 66 static short idct_simple_mmx_perm[64]={ |
2979 | 67 0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D, |
68 0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D, | |
69 0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D, | |
70 0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F, | |
71 0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F, | |
72 0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D, | |
73 0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F, | |
74 0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F, | |
633 | 75 }; |
76 | |
33 | 77 void idct_mmx_init(void) |
78 { | |
79 int i; | |
80 | |
81 /* the mmx/mmxext idct uses a reordered input, so we patch scan tables */ | |
82 for (i = 0; i < 64; i++) { | |
2979 | 83 idct_mmx_perm[i] = (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2); |
84 // idct_simple_mmx_perm[i] = simple_block_permute_op(i); | |
33 | 85 } |
86 } | |
87 | |
88 static DCTELEM block[64] __attribute__ ((aligned (8))); | |
89 static DCTELEM block1[64] __attribute__ ((aligned (8))); | |
633 | 90 static DCTELEM block_org[64] __attribute__ ((aligned (8))); |
33 | 91 |
92 void dct_error(const char *name, int is_idct, | |
93 void (*fdct_func)(DCTELEM *block), | |
633 | 94 void (*fdct_ref)(DCTELEM *block), int test) |
0 | 95 { |
96 int it, i, scale; | |
97 int err_inf, v; | |
1064 | 98 int64_t err2, ti, ti1, it1; |
99 int64_t sysErr[64], sysErrMax=0; | |
633 | 100 int maxout=0; |
101 int blockSumErrMax=0, blockSumErr; | |
0 | 102 |
103 srandom(0); | |
104 | |
105 err_inf = 0; | |
106 err2 = 0; | |
633 | 107 for(i=0; i<64; i++) sysErr[i]=0; |
0 | 108 for(it=0;it<NB_ITS;it++) { |
633 | 109 for(i=0;i<64;i++) |
110 block1[i] = 0; | |
111 switch(test){ | |
2967 | 112 case 0: |
633 | 113 for(i=0;i<64;i++) |
114 block1[i] = (random() % 512) -256; | |
635 | 115 if (is_idct){ |
633 | 116 fdct(block1); |
635 | 117 |
118 for(i=0;i<64;i++) | |
119 block1[i]>>=3; | |
120 } | |
633 | 121 break; |
122 case 1:{ | |
123 int num= (random()%10)+1; | |
124 for(i=0;i<num;i++) | |
125 block1[random()%64] = (random() % 512) -256; | |
126 }break; | |
127 case 2: | |
128 block1[0]= (random()%4096)-2048; | |
129 block1[63]= (block1[0]&1)^1; | |
130 break; | |
131 } | |
33 | 132 |
633 | 133 #if 0 // simulate mismatch control |
134 { int sum=0; | |
135 for(i=0;i<64;i++) | |
136 sum+=block1[i]; | |
137 | |
2967 | 138 if((sum&1)==0) block1[63]^=1; |
633 | 139 } |
140 #endif | |
141 | |
142 for(i=0; i<64; i++) | |
143 block_org[i]= block1[i]; | |
33 | 144 |
145 if (fdct_func == ff_mmx_idct || | |
633 | 146 fdct_func == j_rev_dct || fdct_func == ff_mmxext_idct) { |
147 for(i=0;i<64;i++) | |
33 | 148 block[idct_mmx_perm[i]] = block1[i]; |
720 | 149 } else if(fdct_func == ff_simple_idct_mmx ) { |
633 | 150 for(i=0;i<64;i++) |
151 block[idct_simple_mmx_perm[i]] = block1[i]; | |
152 | |
2979 | 153 } else { |
633 | 154 for(i=0; i<64; i++) |
155 block[i]= block1[i]; | |
33 | 156 } |
633 | 157 #if 0 // simulate mismatch control for tested IDCT but not the ref |
158 { int sum=0; | |
159 for(i=0;i<64;i++) | |
160 sum+=block[i]; | |
161 | |
2967 | 162 if((sum&1)==0) block[63]^=1; |
633 | 163 } |
164 #endif | |
33 | 165 |
0 | 166 fdct_func(block); |
33 | 167 emms(); /* for ff_mmx_idct */ |
168 | |
2967 | 169 if (fdct_func == fdct_ifast |
170 #ifndef FAAN_POSTSCALE | |
1562
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1557
diff
changeset
|
171 || fdct_func == ff_faandct |
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1557
diff
changeset
|
172 #endif |
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1557
diff
changeset
|
173 ) { |
0 | 174 for(i=0; i<64; i++) { |
635 | 175 scale = 8*(1 << (AANSCALE_BITS + 11)) / aanscales[i]; |
633 | 176 block[i] = (block[i] * scale /*+ (1<<(AANSCALE_BITS-1))*/) >> AANSCALE_BITS; |
177 } | |
178 } | |
179 | |
33 | 180 fdct_ref(block1); |
0 | 181 |
633 | 182 blockSumErr=0; |
0 | 183 for(i=0;i<64;i++) { |
184 v = abs(block[i] - block1[i]); | |
185 if (v > err_inf) | |
186 err_inf = v; | |
187 err2 += v * v; | |
2979 | 188 sysErr[i] += block[i] - block1[i]; |
189 blockSumErr += v; | |
190 if( abs(block[i])>maxout) maxout=abs(block[i]); | |
0 | 191 } |
633 | 192 if(blockSumErrMax < blockSumErr) blockSumErrMax= blockSumErr; |
193 #if 0 // print different matrix pairs | |
194 if(blockSumErr){ | |
195 printf("\n"); | |
196 for(i=0; i<64; i++){ | |
197 if((i&7)==0) printf("\n"); | |
198 printf("%4d ", block_org[i]); | |
199 } | |
200 for(i=0; i<64; i++){ | |
201 if((i&7)==0) printf("\n"); | |
202 printf("%4d ", block[i] - block1[i]); | |
203 } | |
204 } | |
205 #endif | |
0 | 206 } |
633 | 207 for(i=0; i<64; i++) sysErrMax= MAX(sysErrMax, ABS(sysErr[i])); |
2967 | 208 |
633 | 209 #if 1 // dump systematic errors |
210 for(i=0; i<64; i++){ | |
2979 | 211 if(i%8==0) printf("\n"); |
633 | 212 printf("%5d ", (int)sysErr[i]); |
213 } | |
214 printf("\n"); | |
215 #endif | |
2967 | 216 |
633 | 217 printf("%s %s: err_inf=%d err2=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n", |
33 | 218 is_idct ? "IDCT" : "DCT", |
633 | 219 name, err_inf, (double)err2 / NB_ITS / 64.0, (double)sysErrMax / NB_ITS, maxout, blockSumErrMax); |
220 #if 1 //Speed test | |
0 | 221 /* speed test */ |
633 | 222 for(i=0;i<64;i++) |
223 block1[i] = 0; | |
224 switch(test){ | |
2967 | 225 case 0: |
633 | 226 for(i=0;i<64;i++) |
227 block1[i] = (random() % 512) -256; | |
635 | 228 if (is_idct){ |
633 | 229 fdct(block1); |
635 | 230 |
231 for(i=0;i<64;i++) | |
232 block1[i]>>=3; | |
233 } | |
633 | 234 break; |
235 case 1:{ | |
236 case 2: | |
237 block1[0] = (random() % 512) -256; | |
238 block1[1] = (random() % 512) -256; | |
239 block1[2] = (random() % 512) -256; | |
240 block1[3] = (random() % 512) -256; | |
241 }break; | |
242 } | |
0 | 243 |
33 | 244 if (fdct_func == ff_mmx_idct || |
633 | 245 fdct_func == j_rev_dct || fdct_func == ff_mmxext_idct) { |
246 for(i=0;i<64;i++) | |
33 | 247 block[idct_mmx_perm[i]] = block1[i]; |
720 | 248 } else if(fdct_func == ff_simple_idct_mmx ) { |
633 | 249 for(i=0;i<64;i++) |
250 block[idct_simple_mmx_perm[i]] = block1[i]; | |
251 } else { | |
252 for(i=0; i<64; i++) | |
253 block[i]= block1[i]; | |
33 | 254 } |
255 | |
0 | 256 ti = gettime(); |
257 it1 = 0; | |
258 do { | |
259 for(it=0;it<NB_ITS_SPEED;it++) { | |
633 | 260 for(i=0; i<64; i++) |
261 block[i]= block1[i]; | |
262 // memcpy(block, block1, sizeof(DCTELEM) * 64); | |
263 // dont memcpy especially not fastmemcpy because it does movntq !!! | |
0 | 264 fdct_func(block); |
265 } | |
266 it1 += NB_ITS_SPEED; | |
267 ti1 = gettime() - ti; | |
268 } while (ti1 < 1000000); | |
33 | 269 emms(); |
0 | 270 |
633 | 271 printf("%s %s: %0.1f kdct/s\n", |
33 | 272 is_idct ? "IDCT" : "DCT", |
0 | 273 name, (double)it1 * 1000.0 / (double)ti1); |
633 | 274 #endif |
0 | 275 } |
276 | |
1064 | 277 static uint8_t img_dest[64] __attribute__ ((aligned (8))); |
278 static uint8_t img_dest1[64] __attribute__ ((aligned (8))); | |
720 | 279 |
1064 | 280 void idct248_ref(uint8_t *dest, int linesize, int16_t *block) |
720 | 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 } | |
2967 | 300 |
720 | 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 */ | |
722
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
312 s = 0.5 * sqrt(2.0); |
720 | 313 for(i=0;i<4;i++) { |
314 for(j=0;j<8;j++) { | |
722
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
315 block1[8*(2*i)+j] = (block[8*(2*i)+j] + block[8*(2*i+1)+j]) * s; |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
316 block1[8*(2*i+1)+j] = (block[8*(2*i)+j] - block[8*(2*i+1)+j]) * s; |
720 | 317 } |
318 } | |
319 | |
320 /* idct8 on lines */ | |
321 for(i=0;i<8;i++) { | |
322 for(j=0;j<8;j++) { | |
323 sum = 0; | |
324 for(k=0;k<8;k++) | |
325 sum += c8[k][j] * block1[8*i+k]; | |
326 block2[8*i+j] = sum; | |
327 } | |
328 } | |
329 | |
330 /* idct4 */ | |
331 for(i=0;i<8;i++) { | |
332 for(j=0;j<4;j++) { | |
333 /* top */ | |
334 sum = 0; | |
335 for(k=0;k<4;k++) | |
336 sum += c4[k][j] * block2[8*(2*k)+i]; | |
337 block3[8*(2*j)+i] = sum; | |
338 | |
339 /* bottom */ | |
340 sum = 0; | |
341 for(k=0;k<4;k++) | |
342 sum += c4[k][j] * block2[8*(2*k+1)+i]; | |
343 block3[8*(2*j+1)+i] = sum; | |
344 } | |
345 } | |
346 | |
347 /* clamp and store the result */ | |
348 for(i=0;i<8;i++) { | |
349 for(j=0;j<8;j++) { | |
722
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
350 v = block3[8*i+j]; |
720 | 351 if (v < 0) |
352 v = 0; | |
353 else if (v > 255) | |
354 v = 255; | |
355 dest[i * linesize + j] = (int)rint(v); | |
356 } | |
357 } | |
358 } | |
359 | |
2967 | 360 void idct248_error(const char *name, |
1064 | 361 void (*idct248_put)(uint8_t *dest, int line_size, int16_t *block)) |
720 | 362 { |
363 int it, i, it1, ti, ti1, err_max, v; | |
364 | |
365 srandom(0); | |
2967 | 366 |
720 | 367 /* just one test to see if code is correct (precision is less |
368 important here) */ | |
369 err_max = 0; | |
370 for(it=0;it<NB_ITS;it++) { | |
2967 | 371 |
722
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
372 /* XXX: use forward transform to generate values */ |
720 | 373 for(i=0;i<64;i++) |
722
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
374 block1[i] = (random() % 256) - 128; |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
375 block1[0] += 1024; |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
376 |
720 | 377 for(i=0; i<64; i++) |
378 block[i]= block1[i]; | |
379 idct248_ref(img_dest1, 8, block); | |
2967 | 380 |
722
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
381 for(i=0; i<64; i++) |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
382 block[i]= block1[i]; |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
383 idct248_put(img_dest, 8, block); |
2967 | 384 |
722
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
385 for(i=0;i<64;i++) { |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
386 v = abs((int)img_dest[i] - (int)img_dest1[i]); |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
387 if (v == 255) |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
388 printf("%d %d\n", img_dest[i], img_dest1[i]); |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
389 if (v > err_max) |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
390 err_max = v; |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
391 } |
720 | 392 #if 0 |
393 printf("ref=\n"); | |
394 for(i=0;i<8;i++) { | |
395 int j; | |
396 for(j=0;j<8;j++) { | |
397 printf(" %3d", img_dest1[i*8+j]); | |
398 } | |
399 printf("\n"); | |
400 } | |
2967 | 401 |
720 | 402 printf("out=\n"); |
403 for(i=0;i<8;i++) { | |
404 int j; | |
405 for(j=0;j<8;j++) { | |
406 printf(" %3d", img_dest[i*8+j]); | |
407 } | |
408 printf("\n"); | |
409 } | |
410 #endif | |
411 } | |
412 printf("%s %s: err_inf=%d\n", | |
413 1 ? "IDCT248" : "DCT248", | |
414 name, err_max); | |
415 | |
416 ti = gettime(); | |
417 it1 = 0; | |
418 do { | |
419 for(it=0;it<NB_ITS_SPEED;it++) { | |
420 for(i=0; i<64; i++) | |
421 block[i]= block1[i]; | |
422 // memcpy(block, block1, sizeof(DCTELEM) * 64); | |
423 // dont memcpy especially not fastmemcpy because it does movntq !!! | |
424 idct248_put(img_dest, 8, block); | |
425 } | |
426 it1 += NB_ITS_SPEED; | |
427 ti1 = gettime() - ti; | |
428 } while (ti1 < 1000000); | |
429 emms(); | |
430 | |
431 printf("%s %s: %0.1f kdct/s\n", | |
432 1 ? "IDCT248" : "DCT248", | |
433 name, (double)it1 * 1000.0 / (double)ti1); | |
434 } | |
435 | |
33 | 436 void help(void) |
437 { | |
633 | 438 printf("dct-test [-i] [<test-number>]\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" | |
720 | 442 "-i test IDCT implementations\n" |
443 "-4 test IDCT248 implementations\n"); | |
33 | 444 exit(1); |
445 } | |
446 | |
0 | 447 int main(int argc, char **argv) |
448 { | |
720 | 449 int test_idct = 0, test_248_dct = 0; |
633 | 450 int c,i; |
451 int test=1; | |
33 | 452 |
0 | 453 init_fdct(); |
33 | 454 idct_mmx_init(); |
0 | 455 |
633 | 456 for(i=0;i<256;i++) cropTbl[i + MAX_NEG_CROP] = i; |
457 for(i=0;i<MAX_NEG_CROP;i++) { | |
458 cropTbl[i] = 0; | |
459 cropTbl[i + MAX_NEG_CROP + 256] = 255; | |
460 } | |
2967 | 461 |
33 | 462 for(;;) { |
720 | 463 c = getopt(argc, argv, "ih4"); |
33 | 464 if (c == -1) |
465 break; | |
466 switch(c) { | |
467 case 'i': | |
468 test_idct = 1; | |
469 break; | |
720 | 470 case '4': |
471 test_248_dct = 1; | |
472 break; | |
633 | 473 default : |
33 | 474 case 'h': |
475 help(); | |
476 break; | |
477 } | |
478 } | |
2967 | 479 |
633 | 480 if(optind <argc) test= atoi(argv[optind]); |
2967 | 481 |
33 | 482 printf("ffmpeg DCT/IDCT test\n"); |
483 | |
720 | 484 if (test_248_dct) { |
485 idct248_error("SIMPLE-C", simple_idct248_put); | |
33 | 486 } else { |
720 | 487 if (!test_idct) { |
488 dct_error("REF-DBL", 0, fdct, fdct, test); /* only to verify code ! */ | |
489 dct_error("IJG-AAN-INT", 0, fdct_ifast, fdct, test); | |
490 dct_error("IJG-LLM-INT", 0, ff_jpeg_fdct_islow, fdct, test); | |
491 dct_error("MMX", 0, ff_fdct_mmx, fdct, test); | |
1574 | 492 dct_error("MMX2", 0, ff_fdct_mmx2, fdct, test); |
1557 | 493 dct_error("FAAN", 0, ff_faandct, fdct, test); |
720 | 494 } else { |
495 dct_error("REF-DBL", 1, idct, idct, test); | |
496 dct_error("INT", 1, j_rev_dct, idct, test); | |
497 dct_error("LIBMPEG2-MMX", 1, ff_mmx_idct, idct, test); | |
498 dct_error("LIBMPEG2-MMXEXT", 1, ff_mmxext_idct, idct, test); | |
499 dct_error("SIMPLE-C", 1, simple_idct, idct, test); | |
500 dct_error("SIMPLE-MMX", 1, ff_simple_idct_mmx, idct, test); | |
2872 | 501 dct_error("XVID-MMX", 1, ff_idct_xvid_mmx, idct, test); |
502 dct_error("XVID-MMX2", 1, ff_idct_xvid_mmx2, idct, test); | |
720 | 503 // dct_error("ODIVX-C", 1, odivx_idct_c, idct); |
504 //printf(" test against odivx idct\n"); | |
2979 | 505 // dct_error("REF", 1, idct, odivx_idct_c); |
720 | 506 // dct_error("INT", 1, j_rev_dct, odivx_idct_c); |
507 // dct_error("MMX", 1, ff_mmx_idct, odivx_idct_c); | |
508 // dct_error("MMXEXT", 1, ff_mmxext_idct, odivx_idct_c); | |
509 // dct_error("SIMPLE-C", 1, simple_idct, odivx_idct_c); | |
510 // dct_error("SIMPLE-MMX", 1, ff_simple_idct_mmx, odivx_idct_c); | |
511 // dct_error("ODIVX-C", 1, odivx_idct_c, odivx_idct_c); | |
512 } | |
33 | 513 } |
0 | 514 return 0; |
515 } |