dct-test.c

ffmpeg的完整源代码和作者自己写的文档。不但有在Linux的工程哦

#if 1 // dump systematic errors
    for(i=0; i<64; i++){
        if(i%8==0) printf("\n");
        printf("%5d ", (int)sysErr[i]);
    }
    printf("\n");
#endif

    printf("%s %s: err_inf=%d err2=%0.8f syserr=%0.8f maxout=%d blockSumErr=%d\n",
           is_idct ? "IDCT" : "DCT",
           name, err_inf, (double)err2 / NB_ITS / 64.0, (double)sysErrMax / NB_ITS, maxout, blockSumErrMax);
#if 1 //Speed test
    /* speed test */
    for(i=0;i<64;i++)
        block1[i] = 0;
    switch(test){
    case 0:
        for(i=0;i<64;i++)
            block1[i] = (random() % 512) -256;
        if (is_idct){
            fdct(block1);

            for(i=0;i<64;i++)
                block1[i]>>=3;
        }
    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;
    }

    if (form == MMX_PERM) {
        for(i=0;i<64;i++)
            block[idct_mmx_perm[i]] = block1[i];
    } else if(form == MMX_SIMPLE_PERM) {
        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];
    }

    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);
// do not memcpy especially not fastmemcpy because it does movntq !!!
            fdct_func(block);
        }
        it1 += NB_ITS_SPEED;
        ti1 = gettime() - ti;
    } while (ti1 < 1000000);
    emms_c();

    printf("%s %s: %0.1f kdct/s\n",
           is_idct ? "IDCT" : "DCT",
           name, (double)it1 * 1000.0 / (double)ti1);
#endif
}

static uint8_t img_dest[64] __attribute__ ((aligned (8)));
static uint8_t img_dest1[64] __attribute__ ((aligned (8)));

void idct248_ref(uint8_t *dest, int linesize, int16_t *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 */
    s = 0.5 * sqrt(2.0);
    for(i=0;i<4;i++) {
        for(j=0;j<8;j++) {
            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;
        }
    }

    /* 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++) {
            v = block3[8*i+j];
            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_t *dest, int line_size, int16_t *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++) {

        /* XXX: use forward transform to generate values */
        for(i=0;i<64;i++)
            block1[i] = (random() % 256) - 128;
        block1[0] += 1024;

        for(i=0; i<64; i++)
            block[i]= block1[i];
        idct248_ref(img_dest1, 8, block);

        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;
        }
#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);
// do not 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_c();

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

void help(void)
{
    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"
           "-i          test IDCT implementations\n"
           "-4          test IDCT248 implementations\n");
}

int main(int argc, char **argv)
{
    int test_idct = 0, test_248_dct = 0;
    int c,i;
    int test=1;

    init_fdct();
    idct_mmx_init();

    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;
    }

    for(;;) {
        c = getopt(argc, argv, "ih4");
        if (c == -1)
            break;
        switch(c) {
        case 'i':
            test_idct = 1;
            break;
        case '4':
            test_248_dct = 1;
            break;
        default :
        case 'h':
            help();
            return 0;
        }
    }

    if(optind <argc) test= atoi(argv[optind]);

    printf("ffmpeg DCT/IDCT test\n");

    if (test_248_dct) {
        idct248_error("SIMPLE-C", simple_idct248_put);
    } else {
      for (i=0;algos[i].name;i++)
        if (algos[i].is_idct == test_idct) {
          dct_error (algos[i].name, algos[i].is_idct, algos[i].func, algos[i].ref, algos[i].format, test);
        }
    }
    return 0;
}