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.

380 lines
18KB

  1. /*
  2. * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt
  3. *
  4. * This file is part of FFmpeg.
  5. *
  6. * FFmpeg is free software; you can redistribute it and/or
  7. * modify it under the terms of the GNU Lesser General Public
  8. * License as published by the Free Software Foundation; either
  9. * version 2.1 of the License, or (at your option) any later version.
  10. *
  11. * FFmpeg is distributed in the hope that it will be useful,
  12. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14. * Lesser General Public License for more details.
  15. *
  16. * You should have received a copy of the GNU Lesser General Public
  17. * License along with FFmpeg; if not, write to the Free Software
  18. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19. */
  20. #include "libavutil/cpu.h"
  21. #include "libavutil/x86/asm.h"
  22. #include "libavcodec/h264dsp.h"
  23. #include "dsputil_mmx.h"
  24. /***********************************/
  25. /* IDCT */
  26. #define IDCT_ADD_FUNC(NUM, DEPTH, OPT) \
  27. void ff_h264_idct ## NUM ## _add_ ## DEPTH ## _ ## OPT(uint8_t *dst, \
  28. int16_t *block, \
  29. int stride);
  30. IDCT_ADD_FUNC(, 8, mmx)
  31. IDCT_ADD_FUNC(, 10, sse2)
  32. IDCT_ADD_FUNC(_dc, 8, mmx2)
  33. IDCT_ADD_FUNC(_dc, 10, mmx2)
  34. IDCT_ADD_FUNC(8_dc, 8, mmx2)
  35. IDCT_ADD_FUNC(8_dc, 10, sse2)
  36. IDCT_ADD_FUNC(8, 8, mmx)
  37. IDCT_ADD_FUNC(8, 8, sse2)
  38. IDCT_ADD_FUNC(8, 10, sse2)
  39. IDCT_ADD_FUNC(, 10, avx)
  40. IDCT_ADD_FUNC(8_dc, 10, avx)
  41. IDCT_ADD_FUNC(8, 10, avx)
  42. #define IDCT_ADD_REP_FUNC(NUM, REP, DEPTH, OPT) \
  43. void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \
  44. (uint8_t *dst, const int *block_offset, \
  45. DCTELEM *block, int stride, const uint8_t nnzc[6 * 8]);
  46. IDCT_ADD_REP_FUNC(8, 4, 8, mmx)
  47. IDCT_ADD_REP_FUNC(8, 4, 8, mmx2)
  48. IDCT_ADD_REP_FUNC(8, 4, 8, sse2)
  49. IDCT_ADD_REP_FUNC(8, 4, 10, sse2)
  50. IDCT_ADD_REP_FUNC(8, 4, 10, avx)
  51. IDCT_ADD_REP_FUNC(, 16, 8, mmx)
  52. IDCT_ADD_REP_FUNC(, 16, 8, mmx2)
  53. IDCT_ADD_REP_FUNC(, 16, 8, sse2)
  54. IDCT_ADD_REP_FUNC(, 16, 10, sse2)
  55. IDCT_ADD_REP_FUNC(, 16intra, 8, mmx)
  56. IDCT_ADD_REP_FUNC(, 16intra, 8, mmx2)
  57. IDCT_ADD_REP_FUNC(, 16intra, 8, sse2)
  58. IDCT_ADD_REP_FUNC(, 16intra, 10, sse2)
  59. IDCT_ADD_REP_FUNC(, 16, 10, avx)
  60. IDCT_ADD_REP_FUNC(, 16intra, 10, avx)
  61. #define IDCT_ADD_REP_FUNC2(NUM, REP, DEPTH, OPT) \
  62. void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \
  63. (uint8_t **dst, const int *block_offset, \
  64. DCTELEM *block, int stride, const uint8_t nnzc[6 * 8]);
  65. IDCT_ADD_REP_FUNC2(, 8, 8, mmx)
  66. IDCT_ADD_REP_FUNC2(, 8, 8, mmx2)
  67. IDCT_ADD_REP_FUNC2(, 8, 8, sse2)
  68. IDCT_ADD_REP_FUNC2(, 8, 10, sse2)
  69. IDCT_ADD_REP_FUNC2(, 8, 10, avx)
  70. void ff_h264_luma_dc_dequant_idct_mmx(DCTELEM *output, DCTELEM *input, int qmul);
  71. void ff_h264_luma_dc_dequant_idct_sse2(DCTELEM *output, DCTELEM *input, int qmul);
  72. /***********************************/
  73. /* deblocking */
  74. void ff_h264_loop_filter_strength_mmx2(int16_t bS[2][4][4], uint8_t nnz[40],
  75. int8_t ref[2][40], int16_t mv[2][40][2],
  76. int bidir, int edges, int step,
  77. int mask_mv0, int mask_mv1, int field);
  78. #define LF_FUNC(DIR, TYPE, DEPTH, OPT) \
  79. void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix, \
  80. int stride, \
  81. int alpha, \
  82. int beta, \
  83. int8_t *tc0);
  84. #define LF_IFUNC(DIR, TYPE, DEPTH, OPT) \
  85. void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix, \
  86. int stride, \
  87. int alpha, \
  88. int beta);
  89. #define LF_FUNCS(type, depth) \
  90. LF_FUNC(h, chroma, depth, mmx2) \
  91. LF_IFUNC(h, chroma_intra, depth, mmx2) \
  92. LF_FUNC(v, chroma, depth, mmx2) \
  93. LF_IFUNC(v, chroma_intra, depth, mmx2) \
  94. LF_FUNC(h, luma, depth, mmx2) \
  95. LF_IFUNC(h, luma_intra, depth, mmx2) \
  96. LF_FUNC(h, luma, depth, sse2) \
  97. LF_IFUNC(h, luma_intra, depth, sse2) \
  98. LF_FUNC(v, luma, depth, sse2) \
  99. LF_IFUNC(v, luma_intra, depth, sse2) \
  100. LF_FUNC(h, chroma, depth, sse2) \
  101. LF_IFUNC(h, chroma_intra, depth, sse2) \
  102. LF_FUNC(v, chroma, depth, sse2) \
  103. LF_IFUNC(v, chroma_intra, depth, sse2) \
  104. LF_FUNC(h, luma, depth, avx) \
  105. LF_IFUNC(h, luma_intra, depth, avx) \
  106. LF_FUNC(v, luma, depth, avx) \
  107. LF_IFUNC(v, luma_intra, depth, avx) \
  108. LF_FUNC(h, chroma, depth, avx) \
  109. LF_IFUNC(h, chroma_intra, depth, avx) \
  110. LF_FUNC(v, chroma, depth, avx) \
  111. LF_IFUNC(v, chroma_intra, depth, avx)
  112. LF_FUNCS(uint8_t, 8)
  113. LF_FUNCS(uint16_t, 10)
  114. #if ARCH_X86_32 && HAVE_YASM
  115. LF_FUNC(v8, luma, 8, mmx2)
  116. static void ff_deblock_v_luma_8_mmx2(uint8_t *pix, int stride, int alpha,
  117. int beta, int8_t *tc0)
  118. {
  119. if ((tc0[0] & tc0[1]) >= 0)
  120. ff_deblock_v8_luma_8_mmx2(pix + 0, stride, alpha, beta, tc0);
  121. if ((tc0[2] & tc0[3]) >= 0)
  122. ff_deblock_v8_luma_8_mmx2(pix + 8, stride, alpha, beta, tc0 + 2);
  123. }
  124. LF_IFUNC(v8, luma_intra, 8, mmx2)
  125. static void ff_deblock_v_luma_intra_8_mmx2(uint8_t *pix, int stride,
  126. int alpha, int beta)
  127. {
  128. ff_deblock_v8_luma_intra_8_mmx2(pix + 0, stride, alpha, beta);
  129. ff_deblock_v8_luma_intra_8_mmx2(pix + 8, stride, alpha, beta);
  130. }
  131. #endif /* ARCH_X86_32 */
  132. LF_FUNC(v, luma, 10, mmx2)
  133. LF_IFUNC(v, luma_intra, 10, mmx2)
  134. /***********************************/
  135. /* weighted prediction */
  136. #define H264_WEIGHT(W, OPT) \
  137. void ff_h264_weight_ ## W ## _ ## OPT(uint8_t *dst, int stride, \
  138. int height, int log2_denom, \
  139. int weight, int offset);
  140. #define H264_BIWEIGHT(W, OPT) \
  141. void ff_h264_biweight_ ## W ## _ ## OPT(uint8_t *dst, uint8_t *src, \
  142. int stride, int height, \
  143. int log2_denom, int weightd, \
  144. int weights, int offset);
  145. #define H264_BIWEIGHT_MMX(W) \
  146. H264_WEIGHT(W, mmx2) \
  147. H264_BIWEIGHT(W, mmx2)
  148. #define H264_BIWEIGHT_MMX_SSE(W) \
  149. H264_BIWEIGHT_MMX(W) \
  150. H264_WEIGHT(W, sse2) \
  151. H264_BIWEIGHT(W, sse2) \
  152. H264_BIWEIGHT(W, ssse3)
  153. H264_BIWEIGHT_MMX_SSE(16)
  154. H264_BIWEIGHT_MMX_SSE(8)
  155. H264_BIWEIGHT_MMX(4)
  156. #define H264_WEIGHT_10(W, DEPTH, OPT) \
  157. void ff_h264_weight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \
  158. int stride, \
  159. int height, \
  160. int log2_denom, \
  161. int weight, \
  162. int offset);
  163. #define H264_BIWEIGHT_10(W, DEPTH, OPT) \
  164. void ff_h264_biweight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \
  165. uint8_t *src, \
  166. int stride, \
  167. int height, \
  168. int log2_denom, \
  169. int weightd, \
  170. int weights, \
  171. int offset);
  172. #define H264_BIWEIGHT_10_SSE(W, DEPTH) \
  173. H264_WEIGHT_10(W, DEPTH, sse2) \
  174. H264_WEIGHT_10(W, DEPTH, sse4) \
  175. H264_BIWEIGHT_10(W, DEPTH, sse2) \
  176. H264_BIWEIGHT_10(W, DEPTH, sse4)
  177. H264_BIWEIGHT_10_SSE(16, 10)
  178. H264_BIWEIGHT_10_SSE(8, 10)
  179. H264_BIWEIGHT_10_SSE(4, 10)
  180. void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
  181. const int chroma_format_idc)
  182. {
  183. #if HAVE_YASM
  184. int mm_flags = av_get_cpu_flags();
  185. if (chroma_format_idc == 1 && mm_flags & AV_CPU_FLAG_MMXEXT)
  186. c->h264_loop_filter_strength = ff_h264_loop_filter_strength_mmx2;
  187. if (bit_depth == 8) {
  188. if (mm_flags & AV_CPU_FLAG_MMX) {
  189. c->h264_idct_dc_add =
  190. c->h264_idct_add = ff_h264_idct_add_8_mmx;
  191. c->h264_idct8_dc_add =
  192. c->h264_idct8_add = ff_h264_idct8_add_8_mmx;
  193. c->h264_idct_add16 = ff_h264_idct_add16_8_mmx;
  194. c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx;
  195. if (chroma_format_idc == 1)
  196. c->h264_idct_add8 = ff_h264_idct_add8_8_mmx;
  197. c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx;
  198. if (mm_flags & AV_CPU_FLAG_CMOV)
  199. c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx;
  200. if (mm_flags & AV_CPU_FLAG_MMXEXT) {
  201. c->h264_idct_dc_add = ff_h264_idct_dc_add_8_mmx2;
  202. c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmx2;
  203. c->h264_idct_add16 = ff_h264_idct_add16_8_mmx2;
  204. c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx2;
  205. if (chroma_format_idc == 1)
  206. c->h264_idct_add8 = ff_h264_idct_add8_8_mmx2;
  207. c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx2;
  208. c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_8_mmx2;
  209. c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_8_mmx2;
  210. if (chroma_format_idc == 1) {
  211. c->h264_h_loop_filter_chroma = ff_deblock_h_chroma_8_mmx2;
  212. c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_8_mmx2;
  213. }
  214. #if ARCH_X86_32
  215. c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_mmx2;
  216. c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_mmx2;
  217. c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_mmx2;
  218. c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmx2;
  219. #endif /* ARCH_X86_32 */
  220. c->weight_h264_pixels_tab[0] = ff_h264_weight_16_mmx2;
  221. c->weight_h264_pixels_tab[1] = ff_h264_weight_8_mmx2;
  222. c->weight_h264_pixels_tab[2] = ff_h264_weight_4_mmx2;
  223. c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_mmx2;
  224. c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_mmx2;
  225. c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_mmx2;
  226. if (mm_flags & AV_CPU_FLAG_SSE2) {
  227. c->h264_idct8_add = ff_h264_idct8_add_8_sse2;
  228. c->h264_idct_add16 = ff_h264_idct_add16_8_sse2;
  229. c->h264_idct8_add4 = ff_h264_idct8_add4_8_sse2;
  230. if (chroma_format_idc == 1)
  231. c->h264_idct_add8 = ff_h264_idct_add8_8_sse2;
  232. c->h264_idct_add16intra = ff_h264_idct_add16intra_8_sse2;
  233. c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_sse2;
  234. c->weight_h264_pixels_tab[0] = ff_h264_weight_16_sse2;
  235. c->weight_h264_pixels_tab[1] = ff_h264_weight_8_sse2;
  236. c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_sse2;
  237. c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_sse2;
  238. #if HAVE_ALIGNED_STACK
  239. c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_sse2;
  240. c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_sse2;
  241. c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
  242. c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
  243. #endif /* HAVE_ALIGNED_STACK */
  244. }
  245. if (mm_flags & AV_CPU_FLAG_SSSE3) {
  246. c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_ssse3;
  247. c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_ssse3;
  248. }
  249. if (HAVE_AVX_EXTERNAL && mm_flags & AV_CPU_FLAG_AVX) {
  250. #if HAVE_ALIGNED_STACK
  251. c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_avx;
  252. c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_avx;
  253. c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
  254. c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
  255. #endif /* HAVE_ALIGNED_STACK */
  256. }
  257. }
  258. }
  259. } else if (bit_depth == 10) {
  260. if (mm_flags & AV_CPU_FLAG_MMX) {
  261. if (mm_flags & AV_CPU_FLAG_MMXEXT) {
  262. #if ARCH_X86_32
  263. c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_mmx2;
  264. c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_mmx2;
  265. c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_mmx2;
  266. c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_mmx2;
  267. c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_mmx2;
  268. c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_mmx2;
  269. #endif /* ARCH_X86_32 */
  270. c->h264_idct_dc_add = ff_h264_idct_dc_add_10_mmx2;
  271. if (mm_flags & AV_CPU_FLAG_SSE2) {
  272. c->h264_idct_add = ff_h264_idct_add_10_sse2;
  273. c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_sse2;
  274. c->h264_idct_add16 = ff_h264_idct_add16_10_sse2;
  275. if (chroma_format_idc == 1)
  276. c->h264_idct_add8 = ff_h264_idct_add8_10_sse2;
  277. c->h264_idct_add16intra = ff_h264_idct_add16intra_10_sse2;
  278. #if HAVE_ALIGNED_STACK
  279. c->h264_idct8_add = ff_h264_idct8_add_10_sse2;
  280. c->h264_idct8_add4 = ff_h264_idct8_add4_10_sse2;
  281. #endif /* HAVE_ALIGNED_STACK */
  282. c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse2;
  283. c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse2;
  284. c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse2;
  285. c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse2;
  286. c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse2;
  287. c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse2;
  288. c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_sse2;
  289. c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_sse2;
  290. #if HAVE_ALIGNED_STACK
  291. c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_sse2;
  292. c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_sse2;
  293. c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
  294. c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
  295. #endif /* HAVE_ALIGNED_STACK */
  296. }
  297. if (mm_flags & AV_CPU_FLAG_SSE4) {
  298. c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse4;
  299. c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse4;
  300. c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse4;
  301. c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse4;
  302. c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse4;
  303. c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse4;
  304. }
  305. #if HAVE_AVX_EXTERNAL
  306. if (mm_flags & AV_CPU_FLAG_AVX) {
  307. c->h264_idct_dc_add =
  308. c->h264_idct_add = ff_h264_idct_add_10_avx;
  309. c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_avx;
  310. c->h264_idct_add16 = ff_h264_idct_add16_10_avx;
  311. if (chroma_format_idc == 1)
  312. c->h264_idct_add8 = ff_h264_idct_add8_10_avx;
  313. c->h264_idct_add16intra = ff_h264_idct_add16intra_10_avx;
  314. #if HAVE_ALIGNED_STACK
  315. c->h264_idct8_add = ff_h264_idct8_add_10_avx;
  316. c->h264_idct8_add4 = ff_h264_idct8_add4_10_avx;
  317. #endif /* HAVE_ALIGNED_STACK */
  318. c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_avx;
  319. c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_avx;
  320. #if HAVE_ALIGNED_STACK
  321. c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_avx;
  322. c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_avx;
  323. c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_avx;
  324. c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_avx;
  325. #endif /* HAVE_ALIGNED_STACK */
  326. }
  327. #endif /* HAVE_AVX_EXTERNAL */
  328. }
  329. }
  330. }
  331. #endif /* HAVE_YASM */
  332. }