You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

186 lines
4.6KB

  1. /* DCT test. (c) 2001 Gerard Lantau.
  2. Started from sample code by Juan J. Sierralta P.
  3. */
  4. #include <stdlib.h>
  5. #include <stdio.h>
  6. #include <string.h>
  7. #include <sys/time.h>
  8. #include <unistd.h>
  9. #include <getopt.h>
  10. #include "dsputil.h"
  11. #include "i386/mmx.h"
  12. /* reference fdct/idct */
  13. extern void fdct(DCTELEM *block);
  14. extern void idct(DCTELEM *block);
  15. extern void init_fdct();
  16. extern void j_rev_dct(DCTELEM *data);
  17. extern void ff_mmx_idct(DCTELEM *data);
  18. extern void ff_mmxext_idct(DCTELEM *data);
  19. #define AANSCALE_BITS 12
  20. static const unsigned short aanscales[64] = {
  21. /* precomputed values scaled up by 14 bits */
  22. 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
  23. 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270,
  24. 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906,
  25. 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315,
  26. 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520,
  27. 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552,
  28. 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446,
  29. 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247
  30. };
  31. INT64 gettime(void)
  32. {
  33. struct timeval tv;
  34. gettimeofday(&tv,NULL);
  35. return (INT64)tv.tv_sec * 1000000 + tv.tv_usec;
  36. }
  37. #define NB_ITS 20000
  38. #define NB_ITS_SPEED 50000
  39. static short idct_mmx_perm[64];
  40. void idct_mmx_init(void)
  41. {
  42. int i;
  43. /* the mmx/mmxext idct uses a reordered input, so we patch scan tables */
  44. for (i = 0; i < 64; i++) {
  45. idct_mmx_perm[i] = (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2);
  46. }
  47. }
  48. static DCTELEM block[64] __attribute__ ((aligned (8)));
  49. static DCTELEM block1[64] __attribute__ ((aligned (8)));
  50. void dct_error(const char *name, int is_idct,
  51. void (*fdct_func)(DCTELEM *block),
  52. void (*fdct_ref)(DCTELEM *block))
  53. {
  54. int it, i, scale;
  55. int err_inf, v;
  56. INT64 err2, ti, ti1, it1;
  57. srandom(0);
  58. err_inf = 0;
  59. err2 = 0;
  60. for(it=0;it<NB_ITS;it++) {
  61. for(i=0;i<64;i++)
  62. block1[i] = random() % 256;
  63. /* for idct test, generate inverse idct data */
  64. if (is_idct)
  65. fdct(block1);
  66. if (fdct_func == ff_mmx_idct ||
  67. fdct_func == j_rev_dct) {
  68. for(i=0;i<64;i++)
  69. block[idct_mmx_perm[i]] = block1[i];
  70. } else {
  71. memcpy(block, block1, sizeof(DCTELEM) * 64);
  72. }
  73. fdct_func(block);
  74. emms(); /* for ff_mmx_idct */
  75. if (fdct_func == jpeg_fdct_ifast) {
  76. for(i=0; i<64; i++) {
  77. scale = (1 << (AANSCALE_BITS + 11)) / aanscales[i];
  78. block[i] = (block[i] * scale) >> AANSCALE_BITS;
  79. }
  80. }
  81. fdct_ref(block1);
  82. for(i=0;i<64;i++) {
  83. v = abs(block[i] - block1[i]);
  84. if (v > err_inf)
  85. err_inf = v;
  86. err2 += v * v;
  87. }
  88. }
  89. printf("%s %s: err_inf=%d err2=%0.2f\n",
  90. is_idct ? "IDCT" : "DCT",
  91. name, err_inf, (double)err2 / NB_ITS / 64.0);
  92. /* speed test */
  93. for(i=0;i<64;i++)
  94. block1[i] = 255 - 63 + i;
  95. /* for idct test, generate inverse idct data */
  96. if (is_idct)
  97. fdct(block1);
  98. if (fdct_func == ff_mmx_idct ||
  99. fdct_func == j_rev_dct) {
  100. for(i=0;i<64;i++)
  101. block[idct_mmx_perm[i]] = block1[i];
  102. }
  103. ti = gettime();
  104. it1 = 0;
  105. do {
  106. for(it=0;it<NB_ITS_SPEED;it++) {
  107. memcpy(block, block1, sizeof(DCTELEM) * 64);
  108. fdct_func(block);
  109. }
  110. it1 += NB_ITS_SPEED;
  111. ti1 = gettime() - ti;
  112. } while (ti1 < 1000000);
  113. emms();
  114. printf("%s %s: %0.1f kdct/s\n",
  115. is_idct ? "IDCT" : "DCT",
  116. name, (double)it1 * 1000.0 / (double)ti1);
  117. }
  118. void help(void)
  119. {
  120. printf("dct-test [-i]\n"
  121. "test DCT implementations\n");
  122. exit(1);
  123. }
  124. int main(int argc, char **argv)
  125. {
  126. int test_idct = 0;
  127. int c;
  128. init_fdct();
  129. idct_mmx_init();
  130. for(;;) {
  131. c = getopt(argc, argv, "ih");
  132. if (c == -1)
  133. break;
  134. switch(c) {
  135. case 'i':
  136. test_idct = 1;
  137. break;
  138. case 'h':
  139. help();
  140. break;
  141. }
  142. }
  143. printf("ffmpeg DCT/IDCT test\n");
  144. if (!test_idct) {
  145. dct_error("REF", 0, fdct, fdct); /* only to verify code ! */
  146. dct_error("AAN", 0, jpeg_fdct_ifast, fdct);
  147. dct_error("MMX", 0, fdct_mmx, fdct);
  148. } else {
  149. dct_error("REF", 1, idct, idct);
  150. dct_error("INT", 1, j_rev_dct, idct);
  151. dct_error("MMX", 1, ff_mmx_idct, idct);
  152. // dct_error("MMX", 1, ff_mmxext_idct, idct);
  153. }
  154. return 0;
  155. }