dct-test.c 14 KB
Newer Older
1
/* DCT test. (c) 2001 Fabrice Bellard. 
Fabrice Bellard's avatar
Fabrice Bellard committed
2 3 4 5 6 7 8
   Started from sample code by Juan J. Sierralta P.
*/
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <sys/time.h>
#include <unistd.h>
9
#include <getopt.h>
Fabrice Bellard's avatar
Fabrice Bellard committed
10 11 12

#include "dsputil.h"

13
#include "i386/mmx.h"
Michael Niedermayer's avatar
Michael Niedermayer committed
14
#include "simple_idct.h"
15 16

/* reference fdct/idct */
Fabrice Bellard's avatar
Fabrice Bellard committed
17
extern void fdct(DCTELEM *block);
18
extern void idct(DCTELEM *block);
Fabrice Bellard's avatar
Fabrice Bellard committed
19 20
extern void init_fdct();

21 22 23 24
extern void j_rev_dct(DCTELEM *data);
extern void ff_mmx_idct(DCTELEM *data);
extern void ff_mmxext_idct(DCTELEM *data);

Michael Niedermayer's avatar
Michael Niedermayer committed
25 26
extern void odivx_idct_c (short *block);

Fabrice Bellard's avatar
Fabrice Bellard committed
27 28 29 30 31 32
void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, 
                           UINT8 *pixels/*align 8*/, int line_size);

void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, 
                           UINT8 *pixels/*align 8*/, int line_size);

Fabrice Bellard's avatar
Fabrice Bellard committed
33 34 35 36 37 38 39 40 41 42 43 44 45
#define AANSCALE_BITS 12
static const unsigned short aanscales[64] = {
    /* precomputed values scaled up by 14 bits */
    16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
    22725, 31521, 29692, 26722, 22725, 17855, 12299,  6270,
    21407, 29692, 27969, 25172, 21407, 16819, 11585,  5906,
    19266, 26722, 25172, 22654, 19266, 15137, 10426,  5315,
    16384, 22725, 21407, 19266, 16384, 12873,  8867,  4520,
    12873, 17855, 16819, 15137, 12873, 10114,  6967,  3552,
    8867, 12299, 11585, 10426,  8867,  6967,  4799,  2446,
    4520,  6270,  5906,  5315,  4520,  3552,  2446,  1247
};

Michael Niedermayer's avatar
Michael Niedermayer committed
46 47
UINT8 cropTbl[256 + 2 * MAX_NEG_CROP];

Fabrice Bellard's avatar
Fabrice Bellard committed
48 49 50 51 52 53 54 55 56 57
INT64 gettime(void)
{
    struct timeval tv;
    gettimeofday(&tv,NULL);
    return (INT64)tv.tv_sec * 1000000 + tv.tv_usec;
}

#define NB_ITS 20000
#define NB_ITS_SPEED 50000

58 59
static short idct_mmx_perm[64];

Michael Niedermayer's avatar
Michael Niedermayer committed
60 61 62 63 64 65 66 67 68 69 70
static short idct_simple_mmx_perm[64]={
	0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D, 
	0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D, 
	0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D, 
	0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F, 
	0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F, 
	0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D, 
	0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F, 
	0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,
};

71 72 73 74 75 76 77
void idct_mmx_init(void)
{
    int i;

    /* the mmx/mmxext idct uses a reordered input, so we patch scan tables */
    for (i = 0; i < 64; i++) {
	idct_mmx_perm[i] = (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2);
Michael Niedermayer's avatar
Michael Niedermayer committed
78
//	idct_simple_mmx_perm[i] = simple_block_permute_op(i);
79 80 81 82 83
    }
}

static DCTELEM block[64] __attribute__ ((aligned (8)));
static DCTELEM block1[64] __attribute__ ((aligned (8)));
Michael Niedermayer's avatar
Michael Niedermayer committed
84
static DCTELEM block_org[64] __attribute__ ((aligned (8)));
85 86 87

void dct_error(const char *name, int is_idct,
               void (*fdct_func)(DCTELEM *block),
Michael Niedermayer's avatar
Michael Niedermayer committed
88
               void (*fdct_ref)(DCTELEM *block), int test)
Fabrice Bellard's avatar
Fabrice Bellard committed
89 90 91 92
{
    int it, i, scale;
    int err_inf, v;
    INT64 err2, ti, ti1, it1;
Michael Niedermayer's avatar
Michael Niedermayer committed
93 94 95
    INT64 sysErr[64], sysErrMax=0;
    int maxout=0;
    int blockSumErrMax=0, blockSumErr;
Fabrice Bellard's avatar
Fabrice Bellard committed
96 97 98 99 100

    srandom(0);

    err_inf = 0;
    err2 = 0;
Michael Niedermayer's avatar
Michael Niedermayer committed
101
    for(i=0; i<64; i++) sysErr[i]=0;
Fabrice Bellard's avatar
Fabrice Bellard committed
102
    for(it=0;it<NB_ITS;it++) {
Michael Niedermayer's avatar
Michael Niedermayer committed
103 104 105 106 107 108
        for(i=0;i<64;i++)
            block1[i] = 0;
        switch(test){
        case 0: 
            for(i=0;i<64;i++)
                block1[i] = (random() % 512) -256;
Michael Niedermayer's avatar
Michael Niedermayer committed
109
            if (is_idct){
Michael Niedermayer's avatar
Michael Niedermayer committed
110
                fdct(block1);
Michael Niedermayer's avatar
Michael Niedermayer committed
111 112 113 114

                for(i=0;i<64;i++)
                    block1[i]>>=3;
            }
Michael Niedermayer's avatar
Michael Niedermayer committed
115 116 117 118 119 120 121 122 123 124 125
        break;
        case 1:{
            int num= (random()%10)+1;
            for(i=0;i<num;i++)
                block1[random()%64] = (random() % 512) -256;
        }break;
        case 2:
            block1[0]= (random()%4096)-2048;
            block1[63]= (block1[0]&1)^1;
        break;
        }
126

Michael Niedermayer's avatar
Michael Niedermayer committed
127 128 129 130 131 132 133 134 135 136 137
#if 0 // simulate mismatch control
{ int sum=0;
        for(i=0;i<64;i++)
           sum+=block1[i];

        if((sum&1)==0) block1[63]^=1; 
}
#endif

        for(i=0; i<64; i++)
            block_org[i]= block1[i];
138 139

        if (fdct_func == ff_mmx_idct ||
Michael Niedermayer's avatar
Michael Niedermayer committed
140 141
            fdct_func == j_rev_dct || fdct_func == ff_mmxext_idct) {
            for(i=0;i<64;i++)
142
                block[idct_mmx_perm[i]] = block1[i];
Fabrice Bellard's avatar
Fabrice Bellard committed
143
        } else if(fdct_func == ff_simple_idct_mmx ) {
Michael Niedermayer's avatar
Michael Niedermayer committed
144 145 146 147 148 149
            for(i=0;i<64;i++)
                block[idct_simple_mmx_perm[i]] = block1[i];

	} else {
            for(i=0; i<64; i++)
                block[i]= block1[i];
150
        }
Michael Niedermayer's avatar
Michael Niedermayer committed
151 152 153 154 155 156 157 158
#if 0 // simulate mismatch control for tested IDCT but not the ref
{ int sum=0;
        for(i=0;i<64;i++)
           sum+=block[i];

        if((sum&1)==0) block[63]^=1; 
}
#endif
159

Fabrice Bellard's avatar
Fabrice Bellard committed
160
        fdct_func(block);
161 162
        emms(); /* for ff_mmx_idct */

163
        if (fdct_func == fdct_ifast) {
Fabrice Bellard's avatar
Fabrice Bellard committed
164
            for(i=0; i<64; i++) {
Michael Niedermayer's avatar
Michael Niedermayer committed
165
                scale = 8*(1 << (AANSCALE_BITS + 11)) / aanscales[i];
Michael Niedermayer's avatar
Michael Niedermayer committed
166 167 168 169
                block[i] = (block[i] * scale /*+ (1<<(AANSCALE_BITS-1))*/) >> AANSCALE_BITS;
            }
        }

170
        fdct_ref(block1);
Fabrice Bellard's avatar
Fabrice Bellard committed
171

Michael Niedermayer's avatar
Michael Niedermayer committed
172
        blockSumErr=0;
Fabrice Bellard's avatar
Fabrice Bellard committed
173 174 175 176 177
        for(i=0;i<64;i++) {
            v = abs(block[i] - block1[i]);
            if (v > err_inf)
                err_inf = v;
            err2 += v * v;
Michael Niedermayer's avatar
Michael Niedermayer committed
178 179 180
	    sysErr[i] += block[i] - block1[i];
	    blockSumErr += v;
	    if( abs(block[i])>maxout) maxout=abs(block[i]);
Fabrice Bellard's avatar
Fabrice Bellard committed
181
        }
Michael Niedermayer's avatar
Michael Niedermayer committed
182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202
        if(blockSumErrMax < blockSumErr) blockSumErrMax= blockSumErr;
#if 0 // print different matrix pairs
        if(blockSumErr){
            printf("\n");
            for(i=0; i<64; i++){
                if((i&7)==0) printf("\n");
                printf("%4d ", block_org[i]);
            }
            for(i=0; i<64; i++){
                if((i&7)==0) printf("\n");
                printf("%4d ", block[i] - block1[i]);
            }
        }
#endif
    }
    for(i=0; i<64; i++) sysErrMax= MAX(sysErrMax, ABS(sysErr[i]));
    
#if 1 // dump systematic errors
    for(i=0; i<64; i++){
	if(i%8==0) printf("\n");
        printf("%5d ", (int)sysErr[i]);
Fabrice Bellard's avatar
Fabrice Bellard committed
203
    }
Michael Niedermayer's avatar
Michael Niedermayer committed
204 205 206 207
    printf("\n");
#endif
    
    printf("%s %s: err_inf=%d err2=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
208
           is_idct ? "IDCT" : "DCT",
Michael Niedermayer's avatar
Michael Niedermayer committed
209 210
           name, err_inf, (double)err2 / NB_ITS / 64.0, (double)sysErrMax / NB_ITS, maxout, blockSumErrMax);
#if 1 //Speed test
Fabrice Bellard's avatar
Fabrice Bellard committed
211
    /* speed test */
Michael Niedermayer's avatar
Michael Niedermayer committed
212 213 214 215 216 217
    for(i=0;i<64;i++)
        block1[i] = 0;
    switch(test){
    case 0: 
        for(i=0;i<64;i++)
            block1[i] = (random() % 512) -256;
Michael Niedermayer's avatar
Michael Niedermayer committed
218
        if (is_idct){
Michael Niedermayer's avatar
Michael Niedermayer committed
219
            fdct(block1);
Michael Niedermayer's avatar
Michael Niedermayer committed
220 221 222 223

            for(i=0;i<64;i++)
                block1[i]>>=3;
        }
Michael Niedermayer's avatar
Michael Niedermayer committed
224 225 226 227 228 229 230 231 232
    break;
    case 1:{
    case 2:
        block1[0] = (random() % 512) -256;
        block1[1] = (random() % 512) -256;
        block1[2] = (random() % 512) -256;
        block1[3] = (random() % 512) -256;
    }break;
    }
Fabrice Bellard's avatar
Fabrice Bellard committed
233

234
    if (fdct_func == ff_mmx_idct ||
Michael Niedermayer's avatar
Michael Niedermayer committed
235 236
        fdct_func == j_rev_dct || fdct_func == ff_mmxext_idct) {
        for(i=0;i<64;i++)
237
            block[idct_mmx_perm[i]] = block1[i];
Fabrice Bellard's avatar
Fabrice Bellard committed
238
    } else if(fdct_func == ff_simple_idct_mmx ) {
Michael Niedermayer's avatar
Michael Niedermayer committed
239 240 241 242 243
        for(i=0;i<64;i++)
            block[idct_simple_mmx_perm[i]] = block1[i];
    } else {
        for(i=0; i<64; i++)
            block[i]= block1[i];
244 245
    }

Fabrice Bellard's avatar
Fabrice Bellard committed
246 247 248 249
    ti = gettime();
    it1 = 0;
    do {
        for(it=0;it<NB_ITS_SPEED;it++) {
Michael Niedermayer's avatar
Michael Niedermayer committed
250 251 252 253
            for(i=0; i<64; i++)
                block[i]= block1[i];
//            memcpy(block, block1, sizeof(DCTELEM) * 64);
// dont memcpy especially not fastmemcpy because it does movntq !!!
Fabrice Bellard's avatar
Fabrice Bellard committed
254 255 256 257 258
            fdct_func(block);
        }
        it1 += NB_ITS_SPEED;
        ti1 = gettime() - ti;
    } while (ti1 < 1000000);
259
    emms();
Fabrice Bellard's avatar
Fabrice Bellard committed
260

Michael Niedermayer's avatar
Michael Niedermayer committed
261
    printf("%s %s: %0.1f kdct/s\n",
262
           is_idct ? "IDCT" : "DCT",
Fabrice Bellard's avatar
Fabrice Bellard committed
263
           name, (double)it1 * 1000.0 / (double)ti1);
Michael Niedermayer's avatar
Michael Niedermayer committed
264
#endif
Fabrice Bellard's avatar
Fabrice Bellard committed
265 266
}

Fabrice Bellard's avatar
Fabrice Bellard committed
267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301
static UINT8 img_dest[64] __attribute__ ((aligned (8)));
static UINT8 img_dest1[64] __attribute__ ((aligned (8)));

void idct248_ref(UINT8 *dest, int linesize, INT16 *block)
{
    static int init;
    static double c8[8][8];
    static double c4[4][4];
    double block1[64], block2[64], block3[64];
    double s, sum, v;
    int i, j, k;

    if (!init) {
        init = 1;

        for(i=0;i<8;i++) {
            sum = 0;
            for(j=0;j<8;j++) {
                s = (i==0) ? sqrt(1.0/8.0) : sqrt(1.0/4.0);
                c8[i][j] = s * cos(M_PI * i * (j + 0.5) / 8.0);
                sum += c8[i][j] * c8[i][j];
            }
        }
        
        for(i=0;i<4;i++) {
            sum = 0;
            for(j=0;j<4;j++) {
                s = (i==0) ? sqrt(1.0/4.0) : sqrt(1.0/2.0);
                c4[i][j] = s * cos(M_PI * i * (j + 0.5) / 4.0);
                sum += c4[i][j] * c4[i][j];
            }
        }
    }

    /* butterfly */
302
    s = 0.5 * sqrt(2.0);
Fabrice Bellard's avatar
Fabrice Bellard committed
303 304
    for(i=0;i<4;i++) {
        for(j=0;j<8;j++) {
305 306
            block1[8*(2*i)+j] = (block[8*(2*i)+j] + block[8*(2*i+1)+j]) * s;
            block1[8*(2*i+1)+j] = (block[8*(2*i)+j] - block[8*(2*i+1)+j]) * s;
Fabrice Bellard's avatar
Fabrice Bellard committed
307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339
        }
    }

    /* idct8 on lines */
    for(i=0;i<8;i++) {
        for(j=0;j<8;j++) {
            sum = 0;
            for(k=0;k<8;k++)
                sum += c8[k][j] * block1[8*i+k];
            block2[8*i+j] = sum;
        }
    }

    /* idct4 */
    for(i=0;i<8;i++) {
        for(j=0;j<4;j++) {
            /* top */
            sum = 0;
            for(k=0;k<4;k++)
                sum += c4[k][j] * block2[8*(2*k)+i];
            block3[8*(2*j)+i] = sum;

            /* bottom */
            sum = 0;
            for(k=0;k<4;k++)
                sum += c4[k][j] * block2[8*(2*k+1)+i];
            block3[8*(2*j+1)+i] = sum;
        }
    }

    /* clamp and store the result */
    for(i=0;i<8;i++) {
        for(j=0;j<8;j++) {
340
            v = block3[8*i+j];
Fabrice Bellard's avatar
Fabrice Bellard committed
341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361
            if (v < 0)
                v = 0;
            else if (v > 255)
                v = 255;
            dest[i * linesize + j] = (int)rint(v);
        }
    }
}

void idct248_error(const char *name, 
                    void (*idct248_put)(UINT8 *dest, int line_size, INT16 *block))
{
    int it, i, it1, ti, ti1, err_max, v;

    srandom(0);
    
    /* just one test to see if code is correct (precision is less
       important here) */
    err_max = 0;
    for(it=0;it<NB_ITS;it++) {
        
362 363 364 365 366
        /* XXX: use forward transform to generate values */
        for(i=0;i<64;i++)
            block1[i] = (random() % 256) - 128;
        block1[0] += 1024;

Fabrice Bellard's avatar
Fabrice Bellard committed
367 368 369 370
        for(i=0; i<64; i++)
            block[i]= block1[i];
        idct248_ref(img_dest1, 8, block);
        
371 372 373 374 375 376 377 378 379 380 381
        for(i=0; i<64; i++)
            block[i]= block1[i];
        idct248_put(img_dest, 8, block);
        
        for(i=0;i<64;i++) {
            v = abs((int)img_dest[i] - (int)img_dest1[i]);
            if (v == 255)
                printf("%d %d\n", img_dest[i], img_dest1[i]);
            if (v > err_max)
                err_max = v;
        }
Fabrice Bellard's avatar
Fabrice Bellard committed
382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425
#if 0
        printf("ref=\n");
        for(i=0;i<8;i++) {
            int j;
            for(j=0;j<8;j++) {
                printf(" %3d", img_dest1[i*8+j]);
            }
            printf("\n");
        }
        
        printf("out=\n");
        for(i=0;i<8;i++) {
            int j;
            for(j=0;j<8;j++) {
                printf(" %3d", img_dest[i*8+j]);
            }
            printf("\n");
        }
#endif
    }
    printf("%s %s: err_inf=%d\n",
           1 ? "IDCT248" : "DCT248",
           name, err_max);

    ti = gettime();
    it1 = 0;
    do {
        for(it=0;it<NB_ITS_SPEED;it++) {
            for(i=0; i<64; i++)
                block[i]= block1[i];
//            memcpy(block, block1, sizeof(DCTELEM) * 64);
// dont memcpy especially not fastmemcpy because it does movntq !!!
            idct248_put(img_dest, 8, block);
        }
        it1 += NB_ITS_SPEED;
        ti1 = gettime() - ti;
    } while (ti1 < 1000000);
    emms();

    printf("%s %s: %0.1f kdct/s\n",
           1 ? "IDCT248" : "DCT248",
           name, (double)it1 * 1000.0 / (double)ti1);
}

426 427
void help(void)
{
Michael Niedermayer's avatar
Michael Niedermayer committed
428 429 430 431
    printf("dct-test [-i] [<test-number>]\n"
           "test-number 0 -> test with random matrixes\n"
           "            1 -> test with random sparse matrixes\n"
           "            2 -> do 3. test from mpeg4 std\n"
Fabrice Bellard's avatar
Fabrice Bellard committed
432 433
           "-i          test IDCT implementations\n"
           "-4          test IDCT248 implementations\n");
434 435 436
    exit(1);
}

Fabrice Bellard's avatar
Fabrice Bellard committed
437 438
int main(int argc, char **argv)
{
Fabrice Bellard's avatar
Fabrice Bellard committed
439
    int test_idct = 0, test_248_dct = 0;
Michael Niedermayer's avatar
Michael Niedermayer committed
440 441
    int c,i;
    int test=1;
442

Fabrice Bellard's avatar
Fabrice Bellard committed
443
    init_fdct();
444
    idct_mmx_init();
Fabrice Bellard's avatar
Fabrice Bellard committed
445

Michael Niedermayer's avatar
Michael Niedermayer committed
446 447 448 449 450 451
    for(i=0;i<256;i++) cropTbl[i + MAX_NEG_CROP] = i;
    for(i=0;i<MAX_NEG_CROP;i++) {
        cropTbl[i] = 0;
        cropTbl[i + MAX_NEG_CROP + 256] = 255;
    }
    
452
    for(;;) {
Fabrice Bellard's avatar
Fabrice Bellard committed
453
        c = getopt(argc, argv, "ih4");
454 455 456 457 458 459
        if (c == -1)
            break;
        switch(c) {
        case 'i':
            test_idct = 1;
            break;
Fabrice Bellard's avatar
Fabrice Bellard committed
460 461 462
        case '4':
            test_248_dct = 1;
            break;
Michael Niedermayer's avatar
Michael Niedermayer committed
463
        default :
464 465 466 467 468
        case 'h':
            help();
            break;
        }
    }
Michael Niedermayer's avatar
Michael Niedermayer committed
469 470
    
    if(optind <argc) test= atoi(argv[optind]);
471 472 473
               
    printf("ffmpeg DCT/IDCT test\n");

Fabrice Bellard's avatar
Fabrice Bellard committed
474 475
    if (test_248_dct) {
        idct248_error("SIMPLE-C", simple_idct248_put);
476
    } else {
Fabrice Bellard's avatar
Fabrice Bellard committed
477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498
        if (!test_idct) {
            dct_error("REF-DBL", 0, fdct, fdct, test); /* only to verify code ! */
            dct_error("IJG-AAN-INT", 0, fdct_ifast, fdct, test);
            dct_error("IJG-LLM-INT", 0, ff_jpeg_fdct_islow, fdct, test);
            dct_error("MMX", 0, ff_fdct_mmx, fdct, test);
        } else {
            dct_error("REF-DBL", 1, idct, idct, test);
            dct_error("INT", 1, j_rev_dct, idct, test);
            dct_error("LIBMPEG2-MMX", 1, ff_mmx_idct, idct, test);
            dct_error("LIBMPEG2-MMXEXT", 1, ff_mmxext_idct, idct, test);
            dct_error("SIMPLE-C", 1, simple_idct, idct, test);
            dct_error("SIMPLE-MMX", 1, ff_simple_idct_mmx, idct, test);
            //        dct_error("ODIVX-C", 1, odivx_idct_c, idct);
            //printf(" test against odivx idct\n");
            //	dct_error("REF", 1, idct, odivx_idct_c);
            //        dct_error("INT", 1, j_rev_dct, odivx_idct_c);
            //        dct_error("MMX", 1, ff_mmx_idct, odivx_idct_c);
            //        dct_error("MMXEXT", 1, ff_mmxext_idct, odivx_idct_c);
            //        dct_error("SIMPLE-C", 1, simple_idct, odivx_idct_c);
            //        dct_error("SIMPLE-MMX", 1, ff_simple_idct_mmx, odivx_idct_c);
            //        dct_error("ODIVX-C", 1, odivx_idct_c, odivx_idct_c);
        }
499
    }
Fabrice Bellard's avatar
Fabrice Bellard committed
500 501
    return 0;
}