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.

391 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/attributes.h"
  21. #include "libavutil/cpu.h"
  22. #include "libavutil/x86/asm.h"
  23. #include "libavutil/x86/cpu.h"
  24. #include "libavcodec/h264dsp.h"
  25. /***********************************/
  26. /* IDCT */
  27. #define IDCT_ADD_FUNC(NUM, DEPTH, OPT) \
  28. void ff_h264_idct ## NUM ## _add_ ## DEPTH ## _ ## OPT(uint8_t *dst, \
  29. int16_t *block, \
  30. int stride);
  31. IDCT_ADD_FUNC(, 8, mmx)
  32. IDCT_ADD_FUNC(, 10, sse2)
  33. IDCT_ADD_FUNC(_dc, 8, mmxext)
  34. IDCT_ADD_FUNC(_dc, 10, mmxext)
  35. IDCT_ADD_FUNC(8_dc, 8, mmxext)
  36. IDCT_ADD_FUNC(8_dc, 10, sse2)
  37. IDCT_ADD_FUNC(8, 8, mmx)
  38. IDCT_ADD_FUNC(8, 8, sse2)
  39. IDCT_ADD_FUNC(8, 10, sse2)
  40. IDCT_ADD_FUNC(, 10, avx)
  41. IDCT_ADD_FUNC(8_dc, 10, avx)
  42. IDCT_ADD_FUNC(8, 10, avx)
  43. #define IDCT_ADD_REP_FUNC(NUM, REP, DEPTH, OPT) \
  44. void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \
  45. (uint8_t *dst, const int *block_offset, \
  46. int16_t *block, int stride, const uint8_t nnzc[6 * 8]);
  47. IDCT_ADD_REP_FUNC(8, 4, 8, mmx)
  48. IDCT_ADD_REP_FUNC(8, 4, 8, mmxext)
  49. IDCT_ADD_REP_FUNC(8, 4, 8, sse2)
  50. IDCT_ADD_REP_FUNC(8, 4, 10, sse2)
  51. IDCT_ADD_REP_FUNC(8, 4, 10, avx)
  52. IDCT_ADD_REP_FUNC(, 16, 8, mmx)
  53. IDCT_ADD_REP_FUNC(, 16, 8, mmxext)
  54. IDCT_ADD_REP_FUNC(, 16, 8, sse2)
  55. IDCT_ADD_REP_FUNC(, 16, 10, sse2)
  56. IDCT_ADD_REP_FUNC(, 16intra, 8, mmx)
  57. IDCT_ADD_REP_FUNC(, 16intra, 8, mmxext)
  58. IDCT_ADD_REP_FUNC(, 16intra, 8, sse2)
  59. IDCT_ADD_REP_FUNC(, 16intra, 10, sse2)
  60. IDCT_ADD_REP_FUNC(, 16, 10, avx)
  61. IDCT_ADD_REP_FUNC(, 16intra, 10, avx)
  62. #define IDCT_ADD_REP_FUNC2(NUM, REP, DEPTH, OPT) \
  63. void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \
  64. (uint8_t **dst, const int *block_offset, \
  65. int16_t *block, int stride, const uint8_t nnzc[6 * 8]);
  66. IDCT_ADD_REP_FUNC2(, 8, 8, mmx)
  67. IDCT_ADD_REP_FUNC2(, 8, 8, mmxext)
  68. IDCT_ADD_REP_FUNC2(, 8, 8, sse2)
  69. IDCT_ADD_REP_FUNC2(, 8, 10, sse2)
  70. IDCT_ADD_REP_FUNC2(, 8, 10, avx)
  71. IDCT_ADD_REP_FUNC2(, 8_422, 8, mmx)
  72. IDCT_ADD_REP_FUNC2(, 8_422, 10, sse2)
  73. IDCT_ADD_REP_FUNC2(, 8_422, 10, avx)
  74. void ff_h264_luma_dc_dequant_idct_mmx(int16_t *output, int16_t *input, int qmul);
  75. void ff_h264_luma_dc_dequant_idct_sse2(int16_t *output, int16_t *input, int qmul);
  76. /***********************************/
  77. /* deblocking */
  78. void ff_h264_loop_filter_strength_mmxext(int16_t bS[2][4][4], uint8_t nnz[40],
  79. int8_t ref[2][40],
  80. int16_t mv[2][40][2],
  81. int bidir, int edges, int step,
  82. int mask_mv0, int mask_mv1, int field);
  83. #define LF_FUNC(DIR, TYPE, DEPTH, OPT) \
  84. void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix, \
  85. int stride, \
  86. int alpha, \
  87. int beta, \
  88. int8_t *tc0);
  89. #define LF_IFUNC(DIR, TYPE, DEPTH, OPT) \
  90. void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix, \
  91. int stride, \
  92. int alpha, \
  93. int beta);
  94. #define LF_FUNCS(type, depth) \
  95. LF_FUNC(h, chroma, depth, mmxext) \
  96. LF_IFUNC(h, chroma_intra, depth, mmxext) \
  97. LF_FUNC(v, chroma, depth, mmxext) \
  98. LF_IFUNC(v, chroma_intra, depth, mmxext) \
  99. LF_FUNC(h, luma, depth, mmxext) \
  100. LF_IFUNC(h, luma_intra, depth, mmxext) \
  101. LF_FUNC(h, luma, depth, sse2) \
  102. LF_IFUNC(h, luma_intra, depth, sse2) \
  103. LF_FUNC(v, luma, depth, sse2) \
  104. LF_IFUNC(v, luma_intra, depth, sse2) \
  105. LF_FUNC(h, chroma, depth, sse2) \
  106. LF_IFUNC(h, chroma_intra, depth, sse2) \
  107. LF_FUNC(v, chroma, depth, sse2) \
  108. LF_IFUNC(v, chroma_intra, depth, sse2) \
  109. LF_FUNC(h, luma, depth, avx) \
  110. LF_IFUNC(h, luma_intra, depth, avx) \
  111. LF_FUNC(v, luma, depth, avx) \
  112. LF_IFUNC(v, luma_intra, depth, avx) \
  113. LF_FUNC(h, chroma, depth, avx) \
  114. LF_IFUNC(h, chroma_intra, depth, avx) \
  115. LF_FUNC(v, chroma, depth, avx) \
  116. LF_IFUNC(v, chroma_intra, depth, avx)
  117. LF_FUNCS(uint8_t, 8)
  118. LF_FUNCS(uint16_t, 10)
  119. void ff_deblock_h_chroma422_8_mmxext(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
  120. LF_IFUNC(h, chroma422_intra, 8, mmxext)
  121. #if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL
  122. LF_FUNC(v8, luma, 8, mmxext)
  123. static void deblock_v_luma_8_mmxext(uint8_t *pix, int stride, int alpha,
  124. int beta, int8_t *tc0)
  125. {
  126. if ((tc0[0] & tc0[1]) >= 0)
  127. ff_deblock_v8_luma_8_mmxext(pix + 0, stride, alpha, beta, tc0);
  128. if ((tc0[2] & tc0[3]) >= 0)
  129. ff_deblock_v8_luma_8_mmxext(pix + 8, stride, alpha, beta, tc0 + 2);
  130. }
  131. LF_IFUNC(v8, luma_intra, 8, mmxext)
  132. static void deblock_v_luma_intra_8_mmxext(uint8_t *pix, int stride,
  133. int alpha, int beta)
  134. {
  135. ff_deblock_v8_luma_intra_8_mmxext(pix + 0, stride, alpha, beta);
  136. ff_deblock_v8_luma_intra_8_mmxext(pix + 8, stride, alpha, beta);
  137. }
  138. #endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */
  139. LF_FUNC(v, luma, 10, mmxext)
  140. LF_IFUNC(v, luma_intra, 10, mmxext)
  141. /***********************************/
  142. /* weighted prediction */
  143. #define H264_WEIGHT(W, OPT) \
  144. void ff_h264_weight_ ## W ## _ ## OPT(uint8_t *dst, ptrdiff_t stride, \
  145. int height, int log2_denom, \
  146. int weight, int offset);
  147. #define H264_BIWEIGHT(W, OPT) \
  148. void ff_h264_biweight_ ## W ## _ ## OPT(uint8_t *dst, uint8_t *src, \
  149. ptrdiff_t stride, int height, \
  150. int log2_denom, int weightd, \
  151. int weights, int offset);
  152. #define H264_BIWEIGHT_MMX(W) \
  153. H264_WEIGHT(W, mmxext) \
  154. H264_BIWEIGHT(W, mmxext)
  155. #define H264_BIWEIGHT_MMX_SSE(W) \
  156. H264_BIWEIGHT_MMX(W) \
  157. H264_WEIGHT(W, sse2) \
  158. H264_BIWEIGHT(W, sse2) \
  159. H264_BIWEIGHT(W, ssse3)
  160. H264_BIWEIGHT_MMX_SSE(16)
  161. H264_BIWEIGHT_MMX_SSE(8)
  162. H264_BIWEIGHT_MMX(4)
  163. #define H264_WEIGHT_10(W, DEPTH, OPT) \
  164. void ff_h264_weight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \
  165. ptrdiff_t stride, \
  166. int height, \
  167. int log2_denom, \
  168. int weight, \
  169. int offset);
  170. #define H264_BIWEIGHT_10(W, DEPTH, OPT) \
  171. void ff_h264_biweight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \
  172. uint8_t *src, \
  173. ptrdiff_t stride, \
  174. int height, \
  175. int log2_denom, \
  176. int weightd, \
  177. int weights, \
  178. int offset);
  179. #define H264_BIWEIGHT_10_SSE(W, DEPTH) \
  180. H264_WEIGHT_10(W, DEPTH, sse2) \
  181. H264_WEIGHT_10(W, DEPTH, sse4) \
  182. H264_BIWEIGHT_10(W, DEPTH, sse2) \
  183. H264_BIWEIGHT_10(W, DEPTH, sse4)
  184. H264_BIWEIGHT_10_SSE(16, 10)
  185. H264_BIWEIGHT_10_SSE(8, 10)
  186. H264_BIWEIGHT_10_SSE(4, 10)
  187. av_cold void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
  188. const int chroma_format_idc)
  189. {
  190. #if HAVE_YASM
  191. int cpu_flags = av_get_cpu_flags();
  192. if (EXTERNAL_MMXEXT(cpu_flags) && chroma_format_idc <= 1)
  193. c->h264_loop_filter_strength = ff_h264_loop_filter_strength_mmxext;
  194. if (bit_depth == 8) {
  195. if (EXTERNAL_MMX(cpu_flags)) {
  196. c->h264_idct_dc_add =
  197. c->h264_idct_add = ff_h264_idct_add_8_mmx;
  198. c->h264_idct8_dc_add =
  199. c->h264_idct8_add = ff_h264_idct8_add_8_mmx;
  200. c->h264_idct_add16 = ff_h264_idct_add16_8_mmx;
  201. c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx;
  202. if (chroma_format_idc <= 1) {
  203. c->h264_idct_add8 = ff_h264_idct_add8_8_mmx;
  204. } else {
  205. c->h264_idct_add8 = ff_h264_idct_add8_422_8_mmx;
  206. }
  207. c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx;
  208. if (cpu_flags & AV_CPU_FLAG_CMOV)
  209. c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx;
  210. }
  211. if (EXTERNAL_MMXEXT(cpu_flags)) {
  212. c->h264_idct_dc_add = ff_h264_idct_dc_add_8_mmxext;
  213. c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmxext;
  214. c->h264_idct_add16 = ff_h264_idct_add16_8_mmxext;
  215. c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmxext;
  216. if (chroma_format_idc <= 1)
  217. c->h264_idct_add8 = ff_h264_idct_add8_8_mmxext;
  218. c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmxext;
  219. c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_8_mmxext;
  220. c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_8_mmxext;
  221. if (chroma_format_idc <= 1) {
  222. c->h264_h_loop_filter_chroma = ff_deblock_h_chroma_8_mmxext;
  223. c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_8_mmxext;
  224. } else {
  225. c->h264_h_loop_filter_chroma = ff_deblock_h_chroma422_8_mmxext;
  226. c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma422_intra_8_mmxext;
  227. }
  228. #if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL
  229. c->h264_v_loop_filter_luma = deblock_v_luma_8_mmxext;
  230. c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_mmxext;
  231. c->h264_v_loop_filter_luma_intra = deblock_v_luma_intra_8_mmxext;
  232. c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmxext;
  233. #endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */
  234. c->weight_h264_pixels_tab[0] = ff_h264_weight_16_mmxext;
  235. c->weight_h264_pixels_tab[1] = ff_h264_weight_8_mmxext;
  236. c->weight_h264_pixels_tab[2] = ff_h264_weight_4_mmxext;
  237. c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_mmxext;
  238. c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_mmxext;
  239. c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_mmxext;
  240. }
  241. if (EXTERNAL_SSE2(cpu_flags)) {
  242. c->h264_idct8_add = ff_h264_idct8_add_8_sse2;
  243. c->h264_idct_add16 = ff_h264_idct_add16_8_sse2;
  244. c->h264_idct8_add4 = ff_h264_idct8_add4_8_sse2;
  245. if (chroma_format_idc <= 1)
  246. c->h264_idct_add8 = ff_h264_idct_add8_8_sse2;
  247. c->h264_idct_add16intra = ff_h264_idct_add16intra_8_sse2;
  248. c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_sse2;
  249. c->weight_h264_pixels_tab[0] = ff_h264_weight_16_sse2;
  250. c->weight_h264_pixels_tab[1] = ff_h264_weight_8_sse2;
  251. c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_sse2;
  252. c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_sse2;
  253. c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_sse2;
  254. c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_sse2;
  255. c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
  256. c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
  257. }
  258. if (EXTERNAL_SSSE3(cpu_flags)) {
  259. c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_ssse3;
  260. c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_ssse3;
  261. }
  262. if (EXTERNAL_AVX(cpu_flags)) {
  263. c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_avx;
  264. c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_avx;
  265. c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
  266. c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
  267. }
  268. } else if (bit_depth == 10) {
  269. if (EXTERNAL_MMXEXT(cpu_flags)) {
  270. #if ARCH_X86_32
  271. c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_mmxext;
  272. c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_mmxext;
  273. c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_mmxext;
  274. c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_mmxext;
  275. c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_mmxext;
  276. c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_mmxext;
  277. #endif /* ARCH_X86_32 */
  278. c->h264_idct_dc_add = ff_h264_idct_dc_add_10_mmxext;
  279. }
  280. if (EXTERNAL_SSE2(cpu_flags)) {
  281. c->h264_idct_add = ff_h264_idct_add_10_sse2;
  282. c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_sse2;
  283. c->h264_idct_add16 = ff_h264_idct_add16_10_sse2;
  284. if (chroma_format_idc <= 1) {
  285. c->h264_idct_add8 = ff_h264_idct_add8_10_sse2;
  286. } else {
  287. c->h264_idct_add8 = ff_h264_idct_add8_422_10_sse2;
  288. }
  289. c->h264_idct_add16intra = ff_h264_idct_add16intra_10_sse2;
  290. #if HAVE_ALIGNED_STACK
  291. c->h264_idct8_add = ff_h264_idct8_add_10_sse2;
  292. c->h264_idct8_add4 = ff_h264_idct8_add4_10_sse2;
  293. #endif /* HAVE_ALIGNED_STACK */
  294. c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse2;
  295. c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse2;
  296. c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse2;
  297. c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse2;
  298. c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse2;
  299. c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse2;
  300. c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_sse2;
  301. c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_sse2;
  302. #if HAVE_ALIGNED_STACK
  303. c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_sse2;
  304. c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_sse2;
  305. c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
  306. c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
  307. #endif /* HAVE_ALIGNED_STACK */
  308. }
  309. if (EXTERNAL_SSE4(cpu_flags)) {
  310. c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse4;
  311. c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse4;
  312. c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse4;
  313. c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse4;
  314. c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse4;
  315. c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse4;
  316. }
  317. if (EXTERNAL_AVX(cpu_flags)) {
  318. c->h264_idct_dc_add =
  319. c->h264_idct_add = ff_h264_idct_add_10_avx;
  320. c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_avx;
  321. c->h264_idct_add16 = ff_h264_idct_add16_10_avx;
  322. if (chroma_format_idc <= 1) {
  323. c->h264_idct_add8 = ff_h264_idct_add8_10_avx;
  324. } else {
  325. c->h264_idct_add8 = ff_h264_idct_add8_422_10_avx;
  326. }
  327. c->h264_idct_add16intra = ff_h264_idct_add16intra_10_avx;
  328. #if HAVE_ALIGNED_STACK
  329. c->h264_idct8_add = ff_h264_idct8_add_10_avx;
  330. c->h264_idct8_add4 = ff_h264_idct8_add4_10_avx;
  331. #endif /* HAVE_ALIGNED_STACK */
  332. c->h264_v_loop_filter_chroma = ff_deblock_v_chroma_10_avx;
  333. c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_avx;
  334. #if HAVE_ALIGNED_STACK
  335. c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_avx;
  336. c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_avx;
  337. c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_avx;
  338. c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_avx;
  339. #endif /* HAVE_ALIGNED_STACK */
  340. }
  341. }
  342. #endif
  343. }