sbrdsp.c 9.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298
  1. /*
  2. * This file is part of FFmpeg.
  3. *
  4. * FFmpeg is free software; you can redistribute it and/or modify
  5. * it under the terms of the GNU General Public License as published by
  6. * the Free Software Foundation; either version 2 of the License, or
  7. * (at your option) any later version.
  8. *
  9. * FFmpeg is distributed in the hope that it will be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  12. * GNU General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU General Public License along
  15. * with FFmpeg; if not, write to the Free Software Foundation, Inc.,
  16. * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  17. */
  18. #include "libavcodec/sbrdsp.h"
  19. #include "checkasm.h"
  20. #define randomize(buf, len) do { \
  21. int i; \
  22. for (i = 0; i < len; i++) { \
  23. const INTFLOAT f = (INTFLOAT)rnd() / UINT_MAX; \
  24. (buf)[i] = f; \
  25. } \
  26. } while (0)
  27. #define EPS 0.0001
  28. static void test_sum64x5(void)
  29. {
  30. LOCAL_ALIGNED_16(INTFLOAT, dst0, [64 + 256]);
  31. LOCAL_ALIGNED_16(INTFLOAT, dst1, [64 + 256]);
  32. declare_func(void, INTFLOAT *z);
  33. randomize((INTFLOAT *)dst0, 64 + 256);
  34. memcpy(dst1, dst0, (64 + 256) * sizeof(INTFLOAT));
  35. call_ref(dst0);
  36. call_new(dst1);
  37. if (!float_near_abs_eps_array(dst0, dst1, EPS, 64 + 256))
  38. fail();
  39. bench_new(dst1);
  40. }
  41. static void test_sum_square(void)
  42. {
  43. INTFLOAT res0;
  44. INTFLOAT res1;
  45. LOCAL_ALIGNED_16(INTFLOAT, src, [256], [2]);
  46. declare_func_float(INTFLOAT, INTFLOAT (*x)[2], int n);
  47. randomize((INTFLOAT *)src, 256 * 2);
  48. res0 = call_ref(src, 256);
  49. res1 = call_new(src, 256);
  50. if (!float_near_abs_eps(res0, res1, EPS))
  51. fail();
  52. bench_new(src, 256);
  53. }
  54. static void test_neg_odd_64(void)
  55. {
  56. LOCAL_ALIGNED_16(INTFLOAT, dst0, [64]);
  57. LOCAL_ALIGNED_16(INTFLOAT, dst1, [64]);
  58. declare_func(void, INTFLOAT *x);
  59. randomize((INTFLOAT *)dst0, 64);
  60. memcpy(dst1, dst0, (64) * sizeof(INTFLOAT));
  61. call_ref(dst0);
  62. call_new(dst1);
  63. if (!float_near_abs_eps_array(dst0, dst1, EPS, 64))
  64. fail();
  65. bench_new(dst1);
  66. }
  67. static void test_qmf_pre_shuffle(void)
  68. {
  69. LOCAL_ALIGNED_16(INTFLOAT, dst0, [128]);
  70. LOCAL_ALIGNED_16(INTFLOAT, dst1, [128]);
  71. declare_func(void, INTFLOAT *z);
  72. randomize((INTFLOAT *)dst0, 128);
  73. memcpy(dst1, dst0, (128) * sizeof(INTFLOAT));
  74. call_ref(dst0);
  75. call_new(dst1);
  76. if (!float_near_abs_eps_array(dst0, dst1, EPS, 128))
  77. fail();
  78. bench_new(dst1);
  79. }
  80. static void test_qmf_post_shuffle(void)
  81. {
  82. LOCAL_ALIGNED_16(INTFLOAT, src, [64]);
  83. LOCAL_ALIGNED_16(INTFLOAT, dst0, [32], [2]);
  84. LOCAL_ALIGNED_16(INTFLOAT, dst1, [32], [2]);
  85. declare_func(void, INTFLOAT W[32][2], const INTFLOAT *z);
  86. randomize((INTFLOAT *)src, 64);
  87. call_ref(dst0, src);
  88. call_new(dst1, src);
  89. if (!float_near_abs_eps_array((INTFLOAT *)dst0, (INTFLOAT *)dst1, EPS, 64))
  90. fail();
  91. bench_new(dst1, src);
  92. }
  93. static void test_qmf_deint_neg(void)
  94. {
  95. LOCAL_ALIGNED_16(INTFLOAT, src, [64]);
  96. LOCAL_ALIGNED_16(INTFLOAT, dst0, [64]);
  97. LOCAL_ALIGNED_16(INTFLOAT, dst1, [64]);
  98. declare_func(void, INTFLOAT *v, const INTFLOAT *src);
  99. randomize((INTFLOAT *)src, 64);
  100. call_ref(dst0, src);
  101. call_new(dst1, src);
  102. if (!float_near_abs_eps_array(dst0, dst1, EPS, 64))
  103. fail();
  104. bench_new(dst1, src);
  105. }
  106. static void test_qmf_deint_bfly(void)
  107. {
  108. LOCAL_ALIGNED_16(INTFLOAT, src0, [64]);
  109. LOCAL_ALIGNED_16(INTFLOAT, src1, [64]);
  110. LOCAL_ALIGNED_16(INTFLOAT, dst0, [128]);
  111. LOCAL_ALIGNED_16(INTFLOAT, dst1, [128]);
  112. declare_func(void, INTFLOAT *v, const INTFLOAT *src0, const INTFLOAT *src1);
  113. memset(dst0, 0, 128 * sizeof(INTFLOAT));
  114. memset(dst1, 0, 128 * sizeof(INTFLOAT));
  115. randomize((INTFLOAT *)src0, 64);
  116. randomize((INTFLOAT *)src1, 64);
  117. call_ref(dst0, src0, src1);
  118. call_new(dst1, src0, src1);
  119. if (!float_near_abs_eps_array(dst0, dst1, EPS, 128))
  120. fail();
  121. bench_new(dst1, src0, src1);
  122. }
  123. static void test_autocorrelate(void)
  124. {
  125. LOCAL_ALIGNED_16(INTFLOAT, src, [40], [2]);
  126. LOCAL_ALIGNED_16(INTFLOAT, dst0, [3], [2][2]);
  127. LOCAL_ALIGNED_16(INTFLOAT, dst1, [3], [2][2]);
  128. declare_func(void, const INTFLOAT x[40][2], INTFLOAT phi[3][2][2]);
  129. memset(dst0, 0, 3 * 2 * 2 * sizeof(INTFLOAT));
  130. memset(dst1, 0, 3 * 2 * 2 * sizeof(INTFLOAT));
  131. randomize((INTFLOAT *)src, 80);
  132. call_ref(src, dst0);
  133. call_new(src, dst1);
  134. if (!float_near_abs_eps_array((INTFLOAT *)dst0, (INTFLOAT *)dst1, EPS, 3 * 2 * 2))
  135. fail();
  136. bench_new(src, dst1);
  137. }
  138. static void test_hf_gen(void)
  139. {
  140. LOCAL_ALIGNED_16(INTFLOAT, low, [128], [2]);
  141. LOCAL_ALIGNED_16(INTFLOAT, alpha0, [2]);
  142. LOCAL_ALIGNED_16(INTFLOAT, alpha1, [2]);
  143. LOCAL_ALIGNED_16(INTFLOAT, dst0, [128], [2]);
  144. LOCAL_ALIGNED_16(INTFLOAT, dst1, [128], [2]);
  145. INTFLOAT bw = (INTFLOAT)rnd() / UINT_MAX;
  146. int i;
  147. declare_func(void, INTFLOAT (*X_high)[2], const INTFLOAT (*X_low)[2],
  148. const INTFLOAT alpha0[2], const INTFLOAT alpha1[2],
  149. INTFLOAT bw, int start, int end);
  150. randomize((INTFLOAT *)low, 128 * 2);
  151. randomize((INTFLOAT *)alpha0, 2);
  152. randomize((INTFLOAT *)alpha1, 2);
  153. for (i = 2; i < 64; i += 2) {
  154. memset(dst0, 0, 128 * 2 * sizeof(INTFLOAT));
  155. memset(dst1, 0, 128 * 2 * sizeof(INTFLOAT));
  156. call_ref(dst0, low, alpha0, alpha1, 0.0, i, 128);
  157. call_new(dst1, low, alpha0, alpha1, 0.0, i, 128);
  158. if (!float_near_abs_eps_array((INTFLOAT *)dst0, (INTFLOAT *)dst1, EPS, 128 * 2))
  159. fail();
  160. bench_new(dst1, low, alpha0, alpha1, bw, i, 128);
  161. }
  162. }
  163. static void test_hf_g_filt(void)
  164. {
  165. LOCAL_ALIGNED_16(INTFLOAT, high, [128], [40][2]);
  166. LOCAL_ALIGNED_16(INTFLOAT, g_filt, [128]);
  167. LOCAL_ALIGNED_16(INTFLOAT, dst0, [128], [2]);
  168. LOCAL_ALIGNED_16(INTFLOAT, dst1, [128], [2]);
  169. declare_func(void, INTFLOAT (*Y)[2], const INTFLOAT (*X_high)[40][2],
  170. const INTFLOAT *g_filt, int m_max, intptr_t ixh);
  171. randomize((INTFLOAT *)high, 128 * 40 * 2);
  172. randomize((INTFLOAT *)g_filt, 128);
  173. call_ref(dst0, high, g_filt, 128, 20);
  174. call_new(dst1, high, g_filt, 128, 20);
  175. if (!float_near_abs_eps_array((INTFLOAT *)dst0, (INTFLOAT *)dst1, EPS, 128 * 2))
  176. fail();
  177. bench_new(dst1, high, g_filt, 128, 20);
  178. }
  179. static void test_hf_apply_noise(const SBRDSPContext *sbrdsp)
  180. {
  181. LOCAL_ALIGNED_16(AAC_FLOAT, s_m, [128]);
  182. LOCAL_ALIGNED_16(AAC_FLOAT, q_filt, [128]);
  183. LOCAL_ALIGNED_16(INTFLOAT, ref, [128], [2]);
  184. LOCAL_ALIGNED_16(INTFLOAT, dst0, [128], [2]);
  185. LOCAL_ALIGNED_16(INTFLOAT, dst1, [128], [2]);
  186. int noise = 0x2a;
  187. int i, j;
  188. declare_func(void, INTFLOAT (*Y)[2], const AAC_FLOAT *s_m,
  189. const AAC_FLOAT *q_filt, int noise,
  190. int kx, int m_max);
  191. randomize((INTFLOAT *)ref, 128 * 2);
  192. randomize((INTFLOAT *)s_m, 128);
  193. randomize((INTFLOAT *)q_filt, 128);
  194. for (i = 0; i < 4; i++) {
  195. if (check_func(sbrdsp->hf_apply_noise[i], "hf_apply_noise_%d", i)) {
  196. for (j = 0; j < 2; j++) {
  197. memcpy(dst0, ref, 128 * 2 * sizeof(INTFLOAT));
  198. memcpy(dst1, ref, 128 * 2 * sizeof(INTFLOAT));
  199. call_ref(dst0, s_m, q_filt, noise, j, 128);
  200. call_new(dst1, s_m, q_filt, noise, j, 128);
  201. if (!float_near_abs_eps_array((INTFLOAT *)dst0, (INTFLOAT *)dst1, EPS, 128 * 2))
  202. fail();
  203. bench_new(dst1, s_m, q_filt, noise, j, 128);
  204. }
  205. }
  206. }
  207. }
  208. void checkasm_check_sbrdsp(void)
  209. {
  210. SBRDSPContext sbrdsp;
  211. ff_sbrdsp_init(&sbrdsp);
  212. if (check_func(sbrdsp.sum64x5, "sum64x5"))
  213. test_sum64x5();
  214. report("sum64x5");
  215. if (check_func(sbrdsp.sum_square, "sum_square"))
  216. test_sum_square();
  217. report("sum_square");
  218. if (check_func(sbrdsp.neg_odd_64, "neg_odd_64"))
  219. test_neg_odd_64();
  220. report("neg_odd_64");
  221. if (check_func(sbrdsp.qmf_pre_shuffle, "qmf_pre_shuffle"))
  222. test_qmf_pre_shuffle();
  223. report("qmf_pre_shuffle");
  224. if (check_func(sbrdsp.qmf_post_shuffle, "qmf_post_shuffle"))
  225. test_qmf_post_shuffle();
  226. report("qmf_post_shuffle");
  227. if (check_func(sbrdsp.qmf_deint_neg, "qmf_deint_neg"))
  228. test_qmf_deint_neg();
  229. report("qmf_deint_neg");
  230. if (check_func(sbrdsp.qmf_deint_bfly, "qmf_deint_bfly"))
  231. test_qmf_deint_bfly();
  232. report("qmf_deint_bfly");
  233. if (check_func(sbrdsp.autocorrelate, "autocorrelate"))
  234. test_autocorrelate();
  235. report("autocorrelate");
  236. if (check_func(sbrdsp.hf_gen, "hf_gen"))
  237. test_hf_gen();
  238. report("hf_gen");
  239. if (check_func(sbrdsp.hf_g_filt, "hf_g_filt"))
  240. test_hf_g_filt();
  241. report("hf_g_filt");
  242. test_hf_apply_noise(&sbrdsp);
  243. report("hf_apply_noise");
  244. }