falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
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.
60152 Commits
32 Branches
322MB
Tree: d43efa68bd
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'd43efa68bd'
${ noResults }
FFmpeg/libavcodec/x86/vp9dsp_init.c

300 lines
11KB
Raw Blame History 

  1. /*

  2.  * VP9 SIMD optimizations

  3.  *

  4.  * Copyright (c) 2013 Ronald S. Bultje <rsbultje gmail com>

  5.  *

  6.  * This file is part of FFmpeg.

  7.  *

  8.  * FFmpeg is free software; you can redistribute it and/or

  9.  * modify it under the terms of the GNU Lesser General Public

  10.  * License as published by the Free Software Foundation; either

  11.  * version 2.1 of the License, or (at your option) any later version.

  12.  *

  13.  * FFmpeg is distributed in the hope that it will be useful,

  14.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  15.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  16.  * Lesser General Public License for more details.

  17.  *

  18.  * You should have received a copy of the GNU Lesser General Public

  19.  * License along with FFmpeg; if not, write to the Free Software

  20.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  21.  */

  22. 

  23. #include "libavutil/attributes.h"

  24. #include "libavutil/cpu.h"

  25. #include "libavutil/mem.h"

  26. #include "libavutil/x86/asm.h"

  27. #include "libavutil/x86/cpu.h"

  28. #include "libavcodec/vp9dsp.h"

  29. 

  30. #if HAVE_YASM

  31. 

  32. #define fpel_func(avg, sz, opt) \

  33. void ff_##avg##sz##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \

  34.                           const uint8_t *src, ptrdiff_t src_stride, \

  35.                           int h, int mx, int my)

  36. fpel_func(put,  4, mmx);

  37. fpel_func(put,  8, mmx);

  38. fpel_func(put, 16, sse);

  39. fpel_func(put, 32, sse);

  40. fpel_func(put, 64, sse);

  41. fpel_func(avg,  4, mmxext);

  42. fpel_func(avg,  8, mmxext);

  43. fpel_func(avg, 16, sse2);

  44. fpel_func(avg, 32, sse2);

  45. fpel_func(avg, 64, sse2);

  46. #undef fpel_func

  47. 

  48. #define mc_func(avg, sz, dir, opt) \

  49. void ff_##avg##_8tap_1d_##dir##_##sz##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \

  50.                                              const uint8_t *src, ptrdiff_t src_stride, \

  51.                                              int h, const int8_t (*filter)[16])

  52. #define mc_funcs(sz) \

  53. mc_func(put, sz, h, ssse3); \

  54. mc_func(avg, sz, h, ssse3); \

  55. mc_func(put, sz, v, ssse3); \

  56. mc_func(avg, sz, v, ssse3)

  57. 

  58. mc_funcs(4);

  59. mc_funcs(8);

  60. #if ARCH_X86_64

  61. mc_funcs(16);

  62. #endif

  63. 

  64. #undef mc_funcs

  65. #undef mc_func

  66. 

  67. #define mc_rep_func(avg, sz, hsz, dir, opt) \

  68. static av_always_inline void \

  69. ff_##avg##_8tap_1d_##dir##_##sz##_##opt(uint8_t *dst, ptrdiff_t dst_stride, \

  70.                                         const uint8_t *src, ptrdiff_t src_stride, \

  71.                                         int h, const int8_t (*filter)[16]) \

  72. { \

  73.     ff_##avg##_8tap_1d_##dir##_##hsz##_##opt(dst,       dst_stride, src, \

  74.                                              src_stride, h, filter); \

  75.     ff_##avg##_8tap_1d_##dir##_##hsz##_##opt(dst + hsz, dst_stride, src + hsz, \

  76.                                              src_stride, h, filter); \

  77. }

  78. 

  79. #define mc_rep_funcs(sz, hsz) \

  80. mc_rep_func(put, sz, hsz, h, ssse3); \

  81. mc_rep_func(avg, sz, hsz, h, ssse3); \

  82. mc_rep_func(put, sz, hsz, v, ssse3); \

  83. mc_rep_func(avg, sz, hsz, v, ssse3)

  84. 

  85. #if ARCH_X86_32

  86. mc_rep_funcs(16, 8);

  87. #endif

  88. mc_rep_funcs(32, 16);

  89. mc_rep_funcs(64, 32);

  90. 

  91. #undef mc_rep_funcs

  92. #undef mc_rep_func

  93. 

  94. extern const int8_t ff_filters_ssse3[3][15][4][16];

  95. 

  96. #define filter_8tap_2d_fn(op, sz, f, fname) \

  97. static void op##_8tap_##fname##_##sz##hv_ssse3(uint8_t *dst, ptrdiff_t dst_stride, \

  98.                                                const uint8_t *src, ptrdiff_t src_stride, \

  99.                                                int h, int mx, int my) \

  100. { \

  101.     LOCAL_ALIGNED_16(uint8_t, temp, [71 * 64]); \

  102.     ff_put_8tap_1d_h_##sz##_ssse3(temp, 64, src - 3 * src_stride, src_stride, \

  103.                                   h + 7, ff_filters_ssse3[f][mx - 1]); \

  104.     ff_##op##_8tap_1d_v_##sz##_ssse3(dst, dst_stride, temp + 3 * 64, 64, \

  105.                                      h, ff_filters_ssse3[f][my - 1]); \

  106. }

  107. 

  108. #define filters_8tap_2d_fn(op, sz) \

  109. filter_8tap_2d_fn(op, sz, FILTER_8TAP_REGULAR, regular) \

  110. filter_8tap_2d_fn(op, sz, FILTER_8TAP_SHARP,   sharp) \

  111. filter_8tap_2d_fn(op, sz, FILTER_8TAP_SMOOTH,  smooth)

  112. 

  113. #define filters_8tap_2d_fn2(op) \

  114. filters_8tap_2d_fn(op, 64) \

  115. filters_8tap_2d_fn(op, 32) \

  116. filters_8tap_2d_fn(op, 16) \

  117. filters_8tap_2d_fn(op, 8) \

  118. filters_8tap_2d_fn(op, 4)

  119. 

  120. filters_8tap_2d_fn2(put)

  121. filters_8tap_2d_fn2(avg)

  122. 

  123. #undef filters_8tap_2d_fn2

  124. #undef filters_8tap_2d_fn

  125. #undef filter_8tap_2d_fn

  126. 

  127. #define filter_8tap_1d_fn(op, sz, f, fname, dir, dvar) \

  128. static void op##_8tap_##fname##_##sz##dir##_ssse3(uint8_t *dst, ptrdiff_t dst_stride, \

  129.                                                   const uint8_t *src, ptrdiff_t src_stride, \

  130.                                                   int h, int mx, int my) \

  131. { \

  132.     ff_##op##_8tap_1d_##dir##_##sz##_ssse3(dst, dst_stride, src, src_stride, \

  133.                                            h, ff_filters_ssse3[f][dvar - 1]); \

  134. }

  135. 

  136. #define filters_8tap_1d_fn(op, sz, dir, dvar) \

  137. filter_8tap_1d_fn(op, sz, FILTER_8TAP_REGULAR, regular, dir, dvar) \

  138. filter_8tap_1d_fn(op, sz, FILTER_8TAP_SHARP,   sharp,   dir, dvar) \

  139. filter_8tap_1d_fn(op, sz, FILTER_8TAP_SMOOTH,  smooth,  dir, dvar)

  140. 

  141. #define filters_8tap_1d_fn2(op, sz) \

  142. filters_8tap_1d_fn(op, sz, h, mx) \

  143. filters_8tap_1d_fn(op, sz, v, my)

  144. 

  145. #define filters_8tap_1d_fn3(op) \

  146. filters_8tap_1d_fn2(op, 64) \

  147. filters_8tap_1d_fn2(op, 32) \

  148. filters_8tap_1d_fn2(op, 16) \

  149. filters_8tap_1d_fn2(op, 8) \

  150. filters_8tap_1d_fn2(op, 4)

  151. 

  152. filters_8tap_1d_fn3(put)

  153. filters_8tap_1d_fn3(avg)

  154. 

  155. #undef filters_8tap_1d_fn

  156. #undef filters_8tap_1d_fn2

  157. #undef filters_8tap_1d_fn3

  158. #undef filter_8tap_1d_fn

  159. 

  160. #define itxfm_func(typea, typeb, size, opt) \

  161. void ff_vp9_##typea##_##typeb##_##size##x##size##_add_##opt(uint8_t *dst, ptrdiff_t stride, \

  162.                                                             int16_t *block, int eob)

  163. #define itxfm_funcs(size, opt) \

  164. itxfm_func(idct,  idct,  size, opt); \

  165. itxfm_func(iadst, idct,  size, opt); \

  166. itxfm_func(idct,  iadst, size, opt); \

  167. itxfm_func(iadst, iadst, size, opt)

  168. 

  169. itxfm_funcs(4, ssse3);

  170. itxfm_funcs(8, ssse3);

  171. itxfm_funcs(8, avx);

  172. itxfm_funcs(16, ssse3);

  173. itxfm_funcs(16, avx);

  174. itxfm_func(idct, idct, 32, ssse3);

  175. itxfm_func(idct, idct, 32, avx);

  176. 

  177. #undef itxfm_func

  178. #undef itxfm_funcs

  179. 

  180. #define lpf_funcs(size1, size2, opt) \

  181. void ff_vp9_loop_filter_v_##size1##_##size2##_##opt(uint8_t *dst, ptrdiff_t stride, \

  182.                                                     int E, int I, int H); \

  183. void ff_vp9_loop_filter_h_##size1##_##size2##_##opt(uint8_t *dst, ptrdiff_t stride, \

  184.                                                     int E, int I, int H)

  185. 

  186. lpf_funcs(16, 16, sse2);

  187. lpf_funcs(16, 16, ssse3);

  188. lpf_funcs(16, 16, avx);

  189. 

  190. #undef lpf_funcs

  191. 

  192. #endif /* HAVE_YASM */

  193. 

  194. av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp)

  195. {

  196. #if HAVE_YASM

  197.     int cpu_flags = av_get_cpu_flags();

  198. 

  199. #define init_fpel(idx1, idx2, sz, type, opt) \

  200.     dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][0][0] = \

  201.     dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][0][0] = \

  202.     dsp->mc[idx1][FILTER_8TAP_SHARP  ][idx2][0][0] = \

  203.     dsp->mc[idx1][FILTER_BILINEAR    ][idx2][0][0] = ff_##type##sz##_##opt

  204. 

  205. 

  206. #define init_subpel1(idx1, idx2, idxh, idxv, sz, dir, type, opt) \

  207.     dsp->mc[idx1][FILTER_8TAP_SMOOTH ][idx2][idxh][idxv] = type##_8tap_smooth_##sz##dir##_##opt; \

  208.     dsp->mc[idx1][FILTER_8TAP_REGULAR][idx2][idxh][idxv] = type##_8tap_regular_##sz##dir##_##opt; \

  209.     dsp->mc[idx1][FILTER_8TAP_SHARP  ][idx2][idxh][idxv] = type##_8tap_sharp_##sz##dir##_##opt

  210. 

  211. #define init_subpel2(idx, idxh, idxv, dir, type, opt) \

  212.     init_subpel1(0, idx, idxh, idxv, 64, dir, type, opt); \

  213.     init_subpel1(1, idx, idxh, idxv, 32, dir, type, opt); \

  214.     init_subpel1(2, idx, idxh, idxv, 16, dir, type, opt); \

  215.     init_subpel1(3, idx, idxh, idxv,  8, dir, type, opt); \

  216.     init_subpel1(4, idx, idxh, idxv,  4, dir, type, opt)

  217. 

  218. #define init_subpel3(idx, type, opt) \

  219.     init_subpel2(idx, 1, 1, hv, type, opt); \

  220.     init_subpel2(idx, 0, 1, v, type, opt); \

  221.     init_subpel2(idx, 1, 0, h, type, opt)

  222. 

  223.     if (EXTERNAL_MMX(cpu_flags)) {

  224.         init_fpel(4, 0,  4, put, mmx);

  225.         init_fpel(3, 0,  8, put, mmx);

  226.     }

  227. 

  228.     if (EXTERNAL_MMXEXT(cpu_flags)) {

  229.         init_fpel(4, 1,  4, avg, mmxext);

  230.         init_fpel(3, 1,  8, avg, mmxext);

  231.     }

  232. 

  233.     if (EXTERNAL_SSE(cpu_flags)) {

  234.         init_fpel(2, 0, 16, put, sse);

  235.         init_fpel(1, 0, 32, put, sse);

  236.         init_fpel(0, 0, 64, put, sse);

  237.     }

  238. 

  239.     if (EXTERNAL_SSE2(cpu_flags)) {

  240.         init_fpel(2, 1, 16, avg, sse2);

  241.         init_fpel(1, 1, 32, avg, sse2);

  242.         init_fpel(0, 1, 64, avg, sse2);

  243.         if (ARCH_X86_64) {

  244.             dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_sse2;

  245.             dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_sse2;

  246.         }

  247.     }

  248. 

  249.     if (EXTERNAL_SSSE3(cpu_flags)) {

  250.         init_subpel3(0, put, ssse3);

  251.         init_subpel3(1, avg, ssse3);

  252.         dsp->itxfm_add[TX_4X4][DCT_DCT] = ff_vp9_idct_idct_4x4_add_ssse3;

  253.         dsp->itxfm_add[TX_4X4][ADST_DCT]  = ff_vp9_idct_iadst_4x4_add_ssse3;

  254.         dsp->itxfm_add[TX_4X4][DCT_ADST]  = ff_vp9_iadst_idct_4x4_add_ssse3;

  255.         dsp->itxfm_add[TX_4X4][ADST_ADST] = ff_vp9_iadst_iadst_4x4_add_ssse3;

  256.         if (ARCH_X86_64) {

  257.             dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_ssse3;

  258.             dsp->itxfm_add[TX_8X8][ADST_DCT]  = ff_vp9_idct_iadst_8x8_add_ssse3;

  259.             dsp->itxfm_add[TX_8X8][DCT_ADST]  = ff_vp9_iadst_idct_8x8_add_ssse3;

  260.             dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_ssse3;

  261.             dsp->itxfm_add[TX_16X16][DCT_DCT]   = ff_vp9_idct_idct_16x16_add_ssse3;

  262.             dsp->itxfm_add[TX_16X16][ADST_DCT]  = ff_vp9_idct_iadst_16x16_add_ssse3;

  263.             dsp->itxfm_add[TX_16X16][DCT_ADST]  = ff_vp9_iadst_idct_16x16_add_ssse3;

  264.             dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_ssse3;

  265.             dsp->itxfm_add[TX_32X32][ADST_ADST] =

  266.             dsp->itxfm_add[TX_32X32][ADST_DCT] =

  267.             dsp->itxfm_add[TX_32X32][DCT_ADST] =

  268.             dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_ssse3;

  269.             dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_ssse3;

  270.             dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_ssse3;

  271.         }

  272.     }

  273. 

  274.     if (EXTERNAL_AVX(cpu_flags)) {

  275.         if (ARCH_X86_64) {

  276.             dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_avx;

  277.             dsp->itxfm_add[TX_8X8][ADST_DCT]  = ff_vp9_idct_iadst_8x8_add_avx;

  278.             dsp->itxfm_add[TX_8X8][DCT_ADST]  = ff_vp9_iadst_idct_8x8_add_avx;

  279.             dsp->itxfm_add[TX_8X8][ADST_ADST] = ff_vp9_iadst_iadst_8x8_add_avx;

  280.             dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_avx;

  281.             dsp->itxfm_add[TX_16X16][ADST_DCT]  = ff_vp9_idct_iadst_16x16_add_avx;

  282.             dsp->itxfm_add[TX_16X16][DCT_ADST]  = ff_vp9_iadst_idct_16x16_add_avx;

  283.             dsp->itxfm_add[TX_16X16][ADST_ADST] = ff_vp9_iadst_iadst_16x16_add_avx;

  284.             dsp->itxfm_add[TX_32X32][ADST_ADST] =

  285.             dsp->itxfm_add[TX_32X32][ADST_DCT] =

  286.             dsp->itxfm_add[TX_32X32][DCT_ADST] =

  287.             dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_avx;

  288.             dsp->loop_filter_16[0] = ff_vp9_loop_filter_h_16_16_avx;

  289.             dsp->loop_filter_16[1] = ff_vp9_loop_filter_v_16_16_avx;

  290.         }

  291.     }

  292. 

  293. #undef init_fpel

  294. #undef init_subpel1

  295. #undef init_subpel2

  296. #undef init_subpel3

  297. 

  298. #endif /* HAVE_YASM */

  299. }