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.

271 lines
11KB

  1. /*
  2. * Copyright (C) 2002-2012 Michael Niedermayer
  3. * Copyright (C) 2012 Ronald S. Bultje
  4. *
  5. * This file is part of FFmpeg.
  6. *
  7. * FFmpeg is free software; you can redistribute it and/or
  8. * modify it under the terms of the GNU Lesser General Public
  9. * License as published by the Free Software Foundation; either
  10. * version 2.1 of the License, or (at your option) any later version.
  11. *
  12. * FFmpeg is distributed in the hope that it will be useful,
  13. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  15. * Lesser General Public License for more details.
  16. *
  17. * You should have received a copy of the GNU Lesser General Public
  18. * License along with FFmpeg; if not, write to the Free Software
  19. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20. */
  21. #include "config.h"
  22. #include "libavutil/attributes.h"
  23. #include "libavutil/avassert.h"
  24. #include "libavutil/common.h"
  25. #include "libavutil/cpu.h"
  26. #include "libavutil/mem.h"
  27. #include "libavutil/x86/asm.h"
  28. #include "libavutil/x86/cpu.h"
  29. #include "libavcodec/videodsp.h"
  30. #if HAVE_YASM
  31. typedef void emu_edge_vfix_func(uint8_t *dst, x86_reg dst_stride,
  32. const uint8_t *src, x86_reg src_stride,
  33. x86_reg start_y, x86_reg end_y, x86_reg bh);
  34. typedef void emu_edge_vvar_func(uint8_t *dst, x86_reg dst_stride,
  35. const uint8_t *src, x86_reg src_stride,
  36. x86_reg start_y, x86_reg end_y, x86_reg bh,
  37. x86_reg w);
  38. extern emu_edge_vfix_func ff_emu_edge_vfix1_mmx;
  39. extern emu_edge_vfix_func ff_emu_edge_vfix2_mmx;
  40. extern emu_edge_vfix_func ff_emu_edge_vfix3_mmx;
  41. extern emu_edge_vfix_func ff_emu_edge_vfix4_mmx;
  42. extern emu_edge_vfix_func ff_emu_edge_vfix5_mmx;
  43. extern emu_edge_vfix_func ff_emu_edge_vfix6_mmx;
  44. extern emu_edge_vfix_func ff_emu_edge_vfix7_mmx;
  45. extern emu_edge_vfix_func ff_emu_edge_vfix8_mmx;
  46. extern emu_edge_vfix_func ff_emu_edge_vfix9_mmx;
  47. extern emu_edge_vfix_func ff_emu_edge_vfix10_mmx;
  48. extern emu_edge_vfix_func ff_emu_edge_vfix11_mmx;
  49. extern emu_edge_vfix_func ff_emu_edge_vfix12_mmx;
  50. extern emu_edge_vfix_func ff_emu_edge_vfix13_mmx;
  51. extern emu_edge_vfix_func ff_emu_edge_vfix14_mmx;
  52. extern emu_edge_vfix_func ff_emu_edge_vfix15_mmx;
  53. extern emu_edge_vfix_func ff_emu_edge_vfix16_mmx;
  54. extern emu_edge_vfix_func ff_emu_edge_vfix17_mmx;
  55. extern emu_edge_vfix_func ff_emu_edge_vfix18_mmx;
  56. extern emu_edge_vfix_func ff_emu_edge_vfix19_mmx;
  57. extern emu_edge_vfix_func ff_emu_edge_vfix20_mmx;
  58. extern emu_edge_vfix_func ff_emu_edge_vfix21_mmx;
  59. extern emu_edge_vfix_func ff_emu_edge_vfix22_mmx;
  60. #if ARCH_X86_32
  61. static emu_edge_vfix_func *vfixtbl_mmx[22] = {
  62. &ff_emu_edge_vfix1_mmx, &ff_emu_edge_vfix2_mmx, &ff_emu_edge_vfix3_mmx,
  63. &ff_emu_edge_vfix4_mmx, &ff_emu_edge_vfix5_mmx, &ff_emu_edge_vfix6_mmx,
  64. &ff_emu_edge_vfix7_mmx, &ff_emu_edge_vfix8_mmx, &ff_emu_edge_vfix9_mmx,
  65. &ff_emu_edge_vfix10_mmx, &ff_emu_edge_vfix11_mmx, &ff_emu_edge_vfix12_mmx,
  66. &ff_emu_edge_vfix13_mmx, &ff_emu_edge_vfix14_mmx, &ff_emu_edge_vfix15_mmx,
  67. &ff_emu_edge_vfix16_mmx, &ff_emu_edge_vfix17_mmx, &ff_emu_edge_vfix18_mmx,
  68. &ff_emu_edge_vfix19_mmx, &ff_emu_edge_vfix20_mmx, &ff_emu_edge_vfix21_mmx,
  69. &ff_emu_edge_vfix22_mmx
  70. };
  71. #endif
  72. extern emu_edge_vvar_func ff_emu_edge_vvar_mmx;
  73. extern emu_edge_vfix_func ff_emu_edge_vfix16_sse;
  74. extern emu_edge_vfix_func ff_emu_edge_vfix17_sse;
  75. extern emu_edge_vfix_func ff_emu_edge_vfix18_sse;
  76. extern emu_edge_vfix_func ff_emu_edge_vfix19_sse;
  77. extern emu_edge_vfix_func ff_emu_edge_vfix20_sse;
  78. extern emu_edge_vfix_func ff_emu_edge_vfix21_sse;
  79. extern emu_edge_vfix_func ff_emu_edge_vfix22_sse;
  80. static emu_edge_vfix_func *vfixtbl_sse[22] = {
  81. ff_emu_edge_vfix1_mmx, ff_emu_edge_vfix2_mmx, ff_emu_edge_vfix3_mmx,
  82. ff_emu_edge_vfix4_mmx, ff_emu_edge_vfix5_mmx, ff_emu_edge_vfix6_mmx,
  83. ff_emu_edge_vfix7_mmx, ff_emu_edge_vfix8_mmx, ff_emu_edge_vfix9_mmx,
  84. ff_emu_edge_vfix10_mmx, ff_emu_edge_vfix11_mmx, ff_emu_edge_vfix12_mmx,
  85. ff_emu_edge_vfix13_mmx, ff_emu_edge_vfix14_mmx, ff_emu_edge_vfix15_mmx,
  86. ff_emu_edge_vfix16_sse, ff_emu_edge_vfix17_sse, ff_emu_edge_vfix18_sse,
  87. ff_emu_edge_vfix19_sse, ff_emu_edge_vfix20_sse, ff_emu_edge_vfix21_sse,
  88. ff_emu_edge_vfix22_sse
  89. };
  90. extern emu_edge_vvar_func ff_emu_edge_vvar_sse;
  91. typedef void emu_edge_hfix_func(uint8_t *dst, x86_reg dst_stride,
  92. x86_reg start_x, x86_reg bh);
  93. typedef void emu_edge_hvar_func(uint8_t *dst, x86_reg dst_stride,
  94. x86_reg start_x, x86_reg n_words, x86_reg bh);
  95. extern emu_edge_hfix_func ff_emu_edge_hfix2_mmx;
  96. extern emu_edge_hfix_func ff_emu_edge_hfix4_mmx;
  97. extern emu_edge_hfix_func ff_emu_edge_hfix6_mmx;
  98. extern emu_edge_hfix_func ff_emu_edge_hfix8_mmx;
  99. extern emu_edge_hfix_func ff_emu_edge_hfix10_mmx;
  100. extern emu_edge_hfix_func ff_emu_edge_hfix12_mmx;
  101. extern emu_edge_hfix_func ff_emu_edge_hfix14_mmx;
  102. extern emu_edge_hfix_func ff_emu_edge_hfix16_mmx;
  103. extern emu_edge_hfix_func ff_emu_edge_hfix18_mmx;
  104. extern emu_edge_hfix_func ff_emu_edge_hfix20_mmx;
  105. extern emu_edge_hfix_func ff_emu_edge_hfix22_mmx;
  106. #if ARCH_X86_32
  107. static emu_edge_hfix_func *hfixtbl_mmx[11] = {
  108. ff_emu_edge_hfix2_mmx, ff_emu_edge_hfix4_mmx, ff_emu_edge_hfix6_mmx,
  109. ff_emu_edge_hfix8_mmx, ff_emu_edge_hfix10_mmx, ff_emu_edge_hfix12_mmx,
  110. ff_emu_edge_hfix14_mmx, ff_emu_edge_hfix16_mmx, ff_emu_edge_hfix18_mmx,
  111. ff_emu_edge_hfix20_mmx, ff_emu_edge_hfix22_mmx
  112. };
  113. #endif
  114. extern emu_edge_hvar_func ff_emu_edge_hvar_mmx;
  115. extern emu_edge_hfix_func ff_emu_edge_hfix16_sse2;
  116. extern emu_edge_hfix_func ff_emu_edge_hfix18_sse2;
  117. extern emu_edge_hfix_func ff_emu_edge_hfix20_sse2;
  118. extern emu_edge_hfix_func ff_emu_edge_hfix22_sse2;
  119. static emu_edge_hfix_func *hfixtbl_sse2[11] = {
  120. ff_emu_edge_hfix2_mmx, ff_emu_edge_hfix4_mmx, ff_emu_edge_hfix6_mmx,
  121. ff_emu_edge_hfix8_mmx, ff_emu_edge_hfix10_mmx, ff_emu_edge_hfix12_mmx,
  122. ff_emu_edge_hfix14_mmx, ff_emu_edge_hfix16_sse2, ff_emu_edge_hfix18_sse2,
  123. ff_emu_edge_hfix20_sse2, ff_emu_edge_hfix22_sse2
  124. };
  125. extern emu_edge_hvar_func ff_emu_edge_hvar_sse2;
  126. static av_always_inline void emulated_edge_mc(uint8_t *dst, const uint8_t *src,
  127. ptrdiff_t dst_stride,
  128. ptrdiff_t src_stride,
  129. x86_reg block_w, x86_reg block_h,
  130. x86_reg src_x, x86_reg src_y,
  131. x86_reg w, x86_reg h,
  132. emu_edge_vfix_func **vfix_tbl,
  133. emu_edge_vvar_func *v_extend_var,
  134. emu_edge_hfix_func **hfix_tbl,
  135. emu_edge_hvar_func *h_extend_var)
  136. {
  137. x86_reg start_y, start_x, end_y, end_x, src_y_add = 0, p;
  138. if (!w || !h)
  139. return;
  140. if (src_y >= h) {
  141. src -= src_y*src_stride;
  142. src_y_add = h - 1;
  143. src_y = h - 1;
  144. } else if (src_y <= -block_h) {
  145. src -= src_y*src_stride;
  146. src_y_add = 1 - block_h;
  147. src_y = 1 - block_h;
  148. }
  149. if (src_x >= w) {
  150. src += w - 1 - src_x;
  151. src_x = w - 1;
  152. } else if (src_x <= -block_w) {
  153. src += 1 - block_w - src_x;
  154. src_x = 1 - block_w;
  155. }
  156. start_y = FFMAX(0, -src_y);
  157. start_x = FFMAX(0, -src_x);
  158. end_y = FFMIN(block_h, h-src_y);
  159. end_x = FFMIN(block_w, w-src_x);
  160. av_assert2(start_x < end_x && block_w > 0);
  161. av_assert2(start_y < end_y && block_h > 0);
  162. // fill in the to-be-copied part plus all above/below
  163. src += (src_y_add + start_y) * src_stride + start_x;
  164. w = end_x - start_x;
  165. if (w <= 22) {
  166. vfix_tbl[w - 1](dst + start_x, dst_stride, src, src_stride,
  167. start_y, end_y, block_h);
  168. } else {
  169. v_extend_var(dst + start_x, dst_stride, src, src_stride,
  170. start_y, end_y, block_h, w);
  171. }
  172. // fill left
  173. if (start_x) {
  174. if (start_x <= 22) {
  175. hfix_tbl[(start_x - 1) >> 1](dst, dst_stride, start_x, block_h);
  176. } else {
  177. h_extend_var(dst, dst_stride,
  178. start_x, (start_x + 1) >> 1, block_h);
  179. }
  180. }
  181. // fill right
  182. p = block_w - end_x;
  183. if (p) {
  184. if (p <= 22) {
  185. hfix_tbl[(p - 1) >> 1](dst + end_x - (p & 1), dst_stride,
  186. -!(p & 1), block_h);
  187. } else {
  188. h_extend_var(dst + end_x - (p & 1), dst_stride,
  189. -!(p & 1), (p + 1) >> 1, block_h);
  190. }
  191. }
  192. }
  193. #if ARCH_X86_32
  194. static av_noinline void emulated_edge_mc_mmx(uint8_t *buf, const uint8_t *src,
  195. ptrdiff_t buf_stride,
  196. ptrdiff_t src_stride,
  197. int block_w, int block_h,
  198. int src_x, int src_y, int w, int h)
  199. {
  200. emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
  201. src_x, src_y, w, h, vfixtbl_mmx, &ff_emu_edge_vvar_mmx,
  202. hfixtbl_mmx, &ff_emu_edge_hvar_mmx);
  203. }
  204. static av_noinline void emulated_edge_mc_sse(uint8_t *buf, const uint8_t *src,
  205. ptrdiff_t buf_stride,
  206. ptrdiff_t src_stride,
  207. int block_w, int block_h,
  208. int src_x, int src_y, int w, int h)
  209. {
  210. emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
  211. src_x, src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse,
  212. hfixtbl_mmx, &ff_emu_edge_hvar_mmx);
  213. }
  214. #endif
  215. static av_noinline void emulated_edge_mc_sse2(uint8_t *buf, const uint8_t *src,
  216. ptrdiff_t buf_stride,
  217. ptrdiff_t src_stride,
  218. int block_w, int block_h,
  219. int src_x, int src_y, int w, int h)
  220. {
  221. emulated_edge_mc(buf, src, buf_stride, src_stride, block_w, block_h,
  222. src_x, src_y, w, h, vfixtbl_sse, &ff_emu_edge_vvar_sse,
  223. hfixtbl_sse2, &ff_emu_edge_hvar_sse2);
  224. }
  225. #endif /* HAVE_YASM */
  226. void ff_prefetch_mmxext(uint8_t *buf, ptrdiff_t stride, int h);
  227. void ff_prefetch_3dnow(uint8_t *buf, ptrdiff_t stride, int h);
  228. av_cold void ff_videodsp_init_x86(VideoDSPContext *ctx, int bpc)
  229. {
  230. #if HAVE_YASM
  231. int cpu_flags = av_get_cpu_flags();
  232. #if ARCH_X86_32
  233. if (EXTERNAL_MMX(cpu_flags) && bpc <= 8) {
  234. ctx->emulated_edge_mc = emulated_edge_mc_mmx;
  235. }
  236. if (EXTERNAL_AMD3DNOW(cpu_flags)) {
  237. ctx->prefetch = ff_prefetch_3dnow;
  238. }
  239. #endif /* ARCH_X86_32 */
  240. if (EXTERNAL_MMXEXT(cpu_flags)) {
  241. ctx->prefetch = ff_prefetch_mmxext;
  242. }
  243. #if ARCH_X86_32
  244. if (EXTERNAL_SSE(cpu_flags) && bpc <= 8) {
  245. ctx->emulated_edge_mc = emulated_edge_mc_sse;
  246. }
  247. #endif /* ARCH_X86_32 */
  248. if (EXTERNAL_SSE2(cpu_flags) && bpc <= 8) {
  249. ctx->emulated_edge_mc = emulated_edge_mc_sse2;
  250. }
  251. #endif /* HAVE_YASM */
  252. }