Mercurial > libavcodec.hg
annotate dct-test.c @ 4048:bf6791303fa0 libavcodec
1 instruction less
author | michael |
---|---|
date | Fri, 20 Oct 2006 08:46:33 +0000 |
parents | 34fdffe98bd0 |
children | fbac0859753d |
rev | line source |
---|---|
3699
c537a97eec66
Add official LGPL license headers to the files that were missing them.
diego
parents:
3398
diff
changeset
|
1 /* |
c537a97eec66
Add official LGPL license headers to the files that were missing them.
diego
parents:
3398
diff
changeset
|
2 * (c) 2001 Fabrice Bellard |
c537a97eec66
Add official LGPL license headers to the files that were missing them.
diego
parents:
3398
diff
changeset
|
3 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3699
diff
changeset
|
4 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3699
diff
changeset
|
5 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3699
diff
changeset
|
6 * FFmpeg is free software; you can redistribute it and/or |
3699
c537a97eec66
Add official LGPL license headers to the files that were missing them.
diego
parents:
3398
diff
changeset
|
7 * modify it under the terms of the GNU Lesser General Public |
c537a97eec66
Add official LGPL license headers to the files that were missing them.
diego
parents:
3398
diff
changeset
|
8 * License as published by the Free Software Foundation; either |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3699
diff
changeset
|
9 * version 2.1 of the License, or (at your option) any later version. |
3699
c537a97eec66
Add official LGPL license headers to the files that were missing them.
diego
parents:
3398
diff
changeset
|
10 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3699
diff
changeset
|
11 * FFmpeg is distributed in the hope that it will be useful, |
3699
c537a97eec66
Add official LGPL license headers to the files that were missing them.
diego
parents:
3398
diff
changeset
|
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
c537a97eec66
Add official LGPL license headers to the files that were missing them.
diego
parents:
3398
diff
changeset
|
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
c537a97eec66
Add official LGPL license headers to the files that were missing them.
diego
parents:
3398
diff
changeset
|
14 * Lesser General Public License for more details. |
c537a97eec66
Add official LGPL license headers to the files that were missing them.
diego
parents:
3398
diff
changeset
|
15 * |
c537a97eec66
Add official LGPL license headers to the files that were missing them.
diego
parents:
3398
diff
changeset
|
16 * You should have received a copy of the GNU Lesser General Public |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3699
diff
changeset
|
17 * License along with FFmpeg; if not, write to the Free Software |
3699
c537a97eec66
Add official LGPL license headers to the files that were missing them.
diego
parents:
3398
diff
changeset
|
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
c537a97eec66
Add official LGPL license headers to the files that were missing them.
diego
parents:
3398
diff
changeset
|
19 */ |
c537a97eec66
Add official LGPL license headers to the files that were missing them.
diego
parents:
3398
diff
changeset
|
20 |
1106 | 21 /** |
22 * @file dct-test.c | |
2967 | 23 * DCT test. (c) 2001 Fabrice Bellard. |
1106 | 24 * Started from sample code by Juan J. Sierralta P. |
25 */ | |
26 | |
0 | 27 #include <stdlib.h> |
28 #include <stdio.h> | |
29 #include <string.h> | |
30 #include <sys/time.h> | |
31 #include <unistd.h> | |
32 | |
33 #include "dsputil.h" | |
34 | |
633 | 35 #include "simple_idct.h" |
1557 | 36 #include "faandct.h" |
33 | 37 |
980 | 38 #ifndef MAX |
39 #define MAX(a, b) (((a) > (b)) ? (a) : (b)) | |
40 #endif | |
41 | |
2872 | 42 #undef printf |
43 | |
44 void *fast_memcpy(void *a, const void *b, size_t c){return memcpy(a,b,c);}; | |
45 | |
33 | 46 /* reference fdct/idct */ |
0 | 47 extern void fdct(DCTELEM *block); |
33 | 48 extern void idct(DCTELEM *block); |
2872 | 49 extern void ff_idct_xvid_mmx(DCTELEM *block); |
50 extern void ff_idct_xvid_mmx2(DCTELEM *block); | |
0 | 51 extern void init_fdct(); |
52 | |
33 | 53 extern void j_rev_dct(DCTELEM *data); |
54 extern void ff_mmx_idct(DCTELEM *data); | |
55 extern void ff_mmxext_idct(DCTELEM *data); | |
56 | |
633 | 57 extern void odivx_idct_c (short *block); |
58 | |
0 | 59 #define AANSCALE_BITS 12 |
60 static const unsigned short aanscales[64] = { | |
61 /* precomputed values scaled up by 14 bits */ | |
62 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
63 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270, | |
64 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906, | |
65 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315, | |
66 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, | |
67 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552, | |
68 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446, | |
69 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247 | |
70 }; | |
71 | |
1064 | 72 uint8_t cropTbl[256 + 2 * MAX_NEG_CROP]; |
633 | 73 |
1064 | 74 int64_t gettime(void) |
0 | 75 { |
76 struct timeval tv; | |
77 gettimeofday(&tv,NULL); | |
1064 | 78 return (int64_t)tv.tv_sec * 1000000 + tv.tv_usec; |
0 | 79 } |
80 | |
81 #define NB_ITS 20000 | |
82 #define NB_ITS_SPEED 50000 | |
83 | |
33 | 84 static short idct_mmx_perm[64]; |
85 | |
633 | 86 static short idct_simple_mmx_perm[64]={ |
2979 | 87 0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D, |
88 0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D, | |
89 0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D, | |
90 0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F, | |
91 0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F, | |
92 0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D, | |
93 0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F, | |
94 0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F, | |
633 | 95 }; |
96 | |
33 | 97 void idct_mmx_init(void) |
98 { | |
99 int i; | |
100 | |
101 /* the mmx/mmxext idct uses a reordered input, so we patch scan tables */ | |
102 for (i = 0; i < 64; i++) { | |
2979 | 103 idct_mmx_perm[i] = (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2); |
104 // idct_simple_mmx_perm[i] = simple_block_permute_op(i); | |
33 | 105 } |
106 } | |
107 | |
108 static DCTELEM block[64] __attribute__ ((aligned (8))); | |
109 static DCTELEM block1[64] __attribute__ ((aligned (8))); | |
633 | 110 static DCTELEM block_org[64] __attribute__ ((aligned (8))); |
33 | 111 |
112 void dct_error(const char *name, int is_idct, | |
113 void (*fdct_func)(DCTELEM *block), | |
633 | 114 void (*fdct_ref)(DCTELEM *block), int test) |
0 | 115 { |
116 int it, i, scale; | |
117 int err_inf, v; | |
1064 | 118 int64_t err2, ti, ti1, it1; |
119 int64_t sysErr[64], sysErrMax=0; | |
633 | 120 int maxout=0; |
121 int blockSumErrMax=0, blockSumErr; | |
0 | 122 |
123 srandom(0); | |
124 | |
125 err_inf = 0; | |
126 err2 = 0; | |
633 | 127 for(i=0; i<64; i++) sysErr[i]=0; |
0 | 128 for(it=0;it<NB_ITS;it++) { |
633 | 129 for(i=0;i<64;i++) |
130 block1[i] = 0; | |
131 switch(test){ | |
2967 | 132 case 0: |
633 | 133 for(i=0;i<64;i++) |
134 block1[i] = (random() % 512) -256; | |
635 | 135 if (is_idct){ |
633 | 136 fdct(block1); |
635 | 137 |
138 for(i=0;i<64;i++) | |
139 block1[i]>>=3; | |
140 } | |
633 | 141 break; |
142 case 1:{ | |
143 int num= (random()%10)+1; | |
144 for(i=0;i<num;i++) | |
145 block1[random()%64] = (random() % 512) -256; | |
146 }break; | |
147 case 2: | |
148 block1[0]= (random()%4096)-2048; | |
149 block1[63]= (block1[0]&1)^1; | |
150 break; | |
151 } | |
33 | 152 |
633 | 153 #if 0 // simulate mismatch control |
154 { int sum=0; | |
155 for(i=0;i<64;i++) | |
156 sum+=block1[i]; | |
157 | |
2967 | 158 if((sum&1)==0) block1[63]^=1; |
633 | 159 } |
160 #endif | |
161 | |
162 for(i=0; i<64; i++) | |
163 block_org[i]= block1[i]; | |
33 | 164 |
165 if (fdct_func == ff_mmx_idct || | |
633 | 166 fdct_func == j_rev_dct || fdct_func == ff_mmxext_idct) { |
167 for(i=0;i<64;i++) | |
33 | 168 block[idct_mmx_perm[i]] = block1[i]; |
720 | 169 } else if(fdct_func == ff_simple_idct_mmx ) { |
633 | 170 for(i=0;i<64;i++) |
171 block[idct_simple_mmx_perm[i]] = block1[i]; | |
172 | |
2979 | 173 } else { |
633 | 174 for(i=0; i<64; i++) |
175 block[i]= block1[i]; | |
33 | 176 } |
633 | 177 #if 0 // simulate mismatch control for tested IDCT but not the ref |
178 { int sum=0; | |
179 for(i=0;i<64;i++) | |
180 sum+=block[i]; | |
181 | |
2967 | 182 if((sum&1)==0) block[63]^=1; |
633 | 183 } |
184 #endif | |
33 | 185 |
0 | 186 fdct_func(block); |
33 | 187 emms(); /* for ff_mmx_idct */ |
188 | |
2967 | 189 if (fdct_func == fdct_ifast |
190 #ifndef FAAN_POSTSCALE | |
1562
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1557
diff
changeset
|
191 || fdct_func == ff_faandct |
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1557
diff
changeset
|
192 #endif |
bf452704100f
optionally merge postscale into quantization table for the float aan dct
michael
parents:
1557
diff
changeset
|
193 ) { |
0 | 194 for(i=0; i<64; i++) { |
635 | 195 scale = 8*(1 << (AANSCALE_BITS + 11)) / aanscales[i]; |
633 | 196 block[i] = (block[i] * scale /*+ (1<<(AANSCALE_BITS-1))*/) >> AANSCALE_BITS; |
197 } | |
198 } | |
199 | |
33 | 200 fdct_ref(block1); |
0 | 201 |
633 | 202 blockSumErr=0; |
0 | 203 for(i=0;i<64;i++) { |
204 v = abs(block[i] - block1[i]); | |
205 if (v > err_inf) | |
206 err_inf = v; | |
207 err2 += v * v; | |
2979 | 208 sysErr[i] += block[i] - block1[i]; |
209 blockSumErr += v; | |
210 if( abs(block[i])>maxout) maxout=abs(block[i]); | |
0 | 211 } |
633 | 212 if(blockSumErrMax < blockSumErr) blockSumErrMax= blockSumErr; |
213 #if 0 // print different matrix pairs | |
214 if(blockSumErr){ | |
215 printf("\n"); | |
216 for(i=0; i<64; i++){ | |
217 if((i&7)==0) printf("\n"); | |
218 printf("%4d ", block_org[i]); | |
219 } | |
220 for(i=0; i<64; i++){ | |
221 if((i&7)==0) printf("\n"); | |
222 printf("%4d ", block[i] - block1[i]); | |
223 } | |
224 } | |
225 #endif | |
0 | 226 } |
4001 | 227 for(i=0; i<64; i++) sysErrMax= MAX(sysErrMax, FFABS(sysErr[i])); |
2967 | 228 |
633 | 229 #if 1 // dump systematic errors |
230 for(i=0; i<64; i++){ | |
2979 | 231 if(i%8==0) printf("\n"); |
633 | 232 printf("%5d ", (int)sysErr[i]); |
233 } | |
234 printf("\n"); | |
235 #endif | |
2967 | 236 |
633 | 237 printf("%s %s: err_inf=%d err2=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n", |
33 | 238 is_idct ? "IDCT" : "DCT", |
633 | 239 name, err_inf, (double)err2 / NB_ITS / 64.0, (double)sysErrMax / NB_ITS, maxout, blockSumErrMax); |
240 #if 1 //Speed test | |
0 | 241 /* speed test */ |
633 | 242 for(i=0;i<64;i++) |
243 block1[i] = 0; | |
244 switch(test){ | |
2967 | 245 case 0: |
633 | 246 for(i=0;i<64;i++) |
247 block1[i] = (random() % 512) -256; | |
635 | 248 if (is_idct){ |
633 | 249 fdct(block1); |
635 | 250 |
251 for(i=0;i<64;i++) | |
252 block1[i]>>=3; | |
253 } | |
633 | 254 break; |
255 case 1:{ | |
256 case 2: | |
257 block1[0] = (random() % 512) -256; | |
258 block1[1] = (random() % 512) -256; | |
259 block1[2] = (random() % 512) -256; | |
260 block1[3] = (random() % 512) -256; | |
261 }break; | |
262 } | |
0 | 263 |
33 | 264 if (fdct_func == ff_mmx_idct || |
633 | 265 fdct_func == j_rev_dct || fdct_func == ff_mmxext_idct) { |
266 for(i=0;i<64;i++) | |
33 | 267 block[idct_mmx_perm[i]] = block1[i]; |
720 | 268 } else if(fdct_func == ff_simple_idct_mmx ) { |
633 | 269 for(i=0;i<64;i++) |
270 block[idct_simple_mmx_perm[i]] = block1[i]; | |
271 } else { | |
272 for(i=0; i<64; i++) | |
273 block[i]= block1[i]; | |
33 | 274 } |
275 | |
0 | 276 ti = gettime(); |
277 it1 = 0; | |
278 do { | |
279 for(it=0;it<NB_ITS_SPEED;it++) { | |
633 | 280 for(i=0; i<64; i++) |
281 block[i]= block1[i]; | |
282 // memcpy(block, block1, sizeof(DCTELEM) * 64); | |
283 // dont memcpy especially not fastmemcpy because it does movntq !!! | |
0 | 284 fdct_func(block); |
285 } | |
286 it1 += NB_ITS_SPEED; | |
287 ti1 = gettime() - ti; | |
288 } while (ti1 < 1000000); | |
33 | 289 emms(); |
0 | 290 |
633 | 291 printf("%s %s: %0.1f kdct/s\n", |
33 | 292 is_idct ? "IDCT" : "DCT", |
0 | 293 name, (double)it1 * 1000.0 / (double)ti1); |
633 | 294 #endif |
0 | 295 } |
296 | |
1064 | 297 static uint8_t img_dest[64] __attribute__ ((aligned (8))); |
298 static uint8_t img_dest1[64] __attribute__ ((aligned (8))); | |
720 | 299 |
1064 | 300 void idct248_ref(uint8_t *dest, int linesize, int16_t *block) |
720 | 301 { |
302 static int init; | |
303 static double c8[8][8]; | |
304 static double c4[4][4]; | |
305 double block1[64], block2[64], block3[64]; | |
306 double s, sum, v; | |
307 int i, j, k; | |
308 | |
309 if (!init) { | |
310 init = 1; | |
311 | |
312 for(i=0;i<8;i++) { | |
313 sum = 0; | |
314 for(j=0;j<8;j++) { | |
315 s = (i==0) ? sqrt(1.0/8.0) : sqrt(1.0/4.0); | |
316 c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0); | |
317 sum += c8[i][j] * c8[i][j]; | |
318 } | |
319 } | |
2967 | 320 |
720 | 321 for(i=0;i<4;i++) { |
322 sum = 0; | |
323 for(j=0;j<4;j++) { | |
324 s = (i==0) ? sqrt(1.0/4.0) : sqrt(1.0/2.0); | |
325 c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0); | |
326 sum += c4[i][j] * c4[i][j]; | |
327 } | |
328 } | |
329 } | |
330 | |
331 /* butterfly */ | |
722
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
332 s = 0.5 * sqrt(2.0); |
720 | 333 for(i=0;i<4;i++) { |
334 for(j=0;j<8;j++) { | |
722
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
335 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
|
336 block1[8*(2*i+1)+j] = (block[8*(2*i)+j] - block[8*(2*i+1)+j]) * s; |
720 | 337 } |
338 } | |
339 | |
340 /* idct8 on lines */ | |
341 for(i=0;i<8;i++) { | |
342 for(j=0;j<8;j++) { | |
343 sum = 0; | |
344 for(k=0;k<8;k++) | |
345 sum += c8[k][j] * block1[8*i+k]; | |
346 block2[8*i+j] = sum; | |
347 } | |
348 } | |
349 | |
350 /* idct4 */ | |
351 for(i=0;i<8;i++) { | |
352 for(j=0;j<4;j++) { | |
353 /* top */ | |
354 sum = 0; | |
355 for(k=0;k<4;k++) | |
356 sum += c4[k][j] * block2[8*(2*k)+i]; | |
357 block3[8*(2*j)+i] = sum; | |
358 | |
359 /* bottom */ | |
360 sum = 0; | |
361 for(k=0;k<4;k++) | |
362 sum += c4[k][j] * block2[8*(2*k+1)+i]; | |
363 block3[8*(2*j+1)+i] = sum; | |
364 } | |
365 } | |
366 | |
367 /* clamp and store the result */ | |
368 for(i=0;i<8;i++) { | |
369 for(j=0;j<8;j++) { | |
722
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
370 v = block3[8*i+j]; |
720 | 371 if (v < 0) |
372 v = 0; | |
373 else if (v > 255) | |
374 v = 255; | |
375 dest[i * linesize + j] = (int)rint(v); | |
376 } | |
377 } | |
378 } | |
379 | |
2967 | 380 void idct248_error(const char *name, |
1064 | 381 void (*idct248_put)(uint8_t *dest, int line_size, int16_t *block)) |
720 | 382 { |
383 int it, i, it1, ti, ti1, err_max, v; | |
384 | |
385 srandom(0); | |
2967 | 386 |
720 | 387 /* just one test to see if code is correct (precision is less |
388 important here) */ | |
389 err_max = 0; | |
390 for(it=0;it<NB_ITS;it++) { | |
2967 | 391 |
722
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
392 /* XXX: use forward transform to generate values */ |
720 | 393 for(i=0;i<64;i++) |
722
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
394 block1[i] = (random() % 256) - 128; |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
395 block1[0] += 1024; |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
396 |
720 | 397 for(i=0; i<64; i++) |
398 block[i]= block1[i]; | |
399 idct248_ref(img_dest1, 8, block); | |
2967 | 400 |
722
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
401 for(i=0; i<64; i++) |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
402 block[i]= block1[i]; |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
403 idct248_put(img_dest, 8, block); |
2967 | 404 |
722
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
405 for(i=0;i<64;i++) { |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
406 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
|
407 if (v == 255) |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
408 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
|
409 if (v > err_max) |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
410 err_max = v; |
ff90043f4a2d
in fact IDCT248 needs to be normalized as I suspected
bellard
parents:
720
diff
changeset
|
411 } |
720 | 412 #if 0 |
413 printf("ref=\n"); | |
414 for(i=0;i<8;i++) { | |
415 int j; | |
416 for(j=0;j<8;j++) { | |
417 printf(" %3d", img_dest1[i*8+j]); | |
418 } | |
419 printf("\n"); | |
420 } | |
2967 | 421 |
720 | 422 printf("out=\n"); |
423 for(i=0;i<8;i++) { | |
424 int j; | |
425 for(j=0;j<8;j++) { | |
426 printf(" %3d", img_dest[i*8+j]); | |
427 } | |
428 printf("\n"); | |
429 } | |
430 #endif | |
431 } | |
432 printf("%s %s: err_inf=%d\n", | |
433 1 ? "IDCT248" : "DCT248", | |
434 name, err_max); | |
435 | |
436 ti = gettime(); | |
437 it1 = 0; | |
438 do { | |
439 for(it=0;it<NB_ITS_SPEED;it++) { | |
440 for(i=0; i<64; i++) | |
441 block[i]= block1[i]; | |
442 // memcpy(block, block1, sizeof(DCTELEM) * 64); | |
443 // dont memcpy especially not fastmemcpy because it does movntq !!! | |
444 idct248_put(img_dest, 8, block); | |
445 } | |
446 it1 += NB_ITS_SPEED; | |
447 ti1 = gettime() - ti; | |
448 } while (ti1 < 1000000); | |
449 emms(); | |
450 | |
451 printf("%s %s: %0.1f kdct/s\n", | |
452 1 ? "IDCT248" : "DCT248", | |
453 name, (double)it1 * 1000.0 / (double)ti1); | |
454 } | |
455 | |
33 | 456 void help(void) |
457 { | |
633 | 458 printf("dct-test [-i] [<test-number>]\n" |
459 "test-number 0 -> test with random matrixes\n" | |
460 " 1 -> test with random sparse matrixes\n" | |
461 " 2 -> do 3. test from mpeg4 std\n" | |
720 | 462 "-i test IDCT implementations\n" |
463 "-4 test IDCT248 implementations\n"); | |
33 | 464 exit(1); |
465 } | |
466 | |
0 | 467 int main(int argc, char **argv) |
468 { | |
720 | 469 int test_idct = 0, test_248_dct = 0; |
633 | 470 int c,i; |
471 int test=1; | |
33 | 472 |
0 | 473 init_fdct(); |
33 | 474 idct_mmx_init(); |
0 | 475 |
633 | 476 for(i=0;i<256;i++) cropTbl[i + MAX_NEG_CROP] = i; |
477 for(i=0;i<MAX_NEG_CROP;i++) { | |
478 cropTbl[i] = 0; | |
479 cropTbl[i + MAX_NEG_CROP + 256] = 255; | |
480 } | |
2967 | 481 |
33 | 482 for(;;) { |
720 | 483 c = getopt(argc, argv, "ih4"); |
33 | 484 if (c == -1) |
485 break; | |
486 switch(c) { | |
487 case 'i': | |
488 test_idct = 1; | |
489 break; | |
720 | 490 case '4': |
491 test_248_dct = 1; | |
492 break; | |
633 | 493 default : |
33 | 494 case 'h': |
495 help(); | |
496 break; | |
497 } | |
498 } | |
2967 | 499 |
633 | 500 if(optind <argc) test= atoi(argv[optind]); |
2967 | 501 |
33 | 502 printf("ffmpeg DCT/IDCT test\n"); |
503 | |
720 | 504 if (test_248_dct) { |
505 idct248_error("SIMPLE-C", simple_idct248_put); | |
33 | 506 } else { |
720 | 507 if (!test_idct) { |
508 dct_error("REF-DBL", 0, fdct, fdct, test); /* only to verify code ! */ | |
509 dct_error("IJG-AAN-INT", 0, fdct_ifast, fdct, test); | |
510 dct_error("IJG-LLM-INT", 0, ff_jpeg_fdct_islow, fdct, test); | |
511 dct_error("MMX", 0, ff_fdct_mmx, fdct, test); | |
1574 | 512 dct_error("MMX2", 0, ff_fdct_mmx2, fdct, test); |
1557 | 513 dct_error("FAAN", 0, ff_faandct, fdct, test); |
720 | 514 } else { |
515 dct_error("REF-DBL", 1, idct, idct, test); | |
516 dct_error("INT", 1, j_rev_dct, idct, test); | |
517 dct_error("LIBMPEG2-MMX", 1, ff_mmx_idct, idct, test); | |
518 dct_error("LIBMPEG2-MMXEXT", 1, ff_mmxext_idct, idct, test); | |
519 dct_error("SIMPLE-C", 1, simple_idct, idct, test); | |
520 dct_error("SIMPLE-MMX", 1, ff_simple_idct_mmx, idct, test); | |
2872 | 521 dct_error("XVID-MMX", 1, ff_idct_xvid_mmx, idct, test); |
522 dct_error("XVID-MMX2", 1, ff_idct_xvid_mmx2, idct, test); | |
720 | 523 // dct_error("ODIVX-C", 1, odivx_idct_c, idct); |
524 //printf(" test against odivx idct\n"); | |
2979 | 525 // dct_error("REF", 1, idct, odivx_idct_c); |
720 | 526 // dct_error("INT", 1, j_rev_dct, odivx_idct_c); |
527 // dct_error("MMX", 1, ff_mmx_idct, odivx_idct_c); | |
528 // dct_error("MMXEXT", 1, ff_mmxext_idct, odivx_idct_c); | |
529 // dct_error("SIMPLE-C", 1, simple_idct, odivx_idct_c); | |
530 // dct_error("SIMPLE-MMX", 1, ff_simple_idct_mmx, odivx_idct_c); | |
531 // dct_error("ODIVX-C", 1, odivx_idct_c, odivx_idct_c); | |
532 } | |
33 | 533 } |
0 | 534 return 0; |
535 } |