git.sesse.net Git - ffmpeg/blob - libavcodec/x86/h264dsp_mmx.c

   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
  21 #include "libavutil/cpu.h"
  22 #include "libavutil/x86_cpu.h"
  23 #include "libavcodec/h264dsp.h"
  24 #include "dsputil_mmx.h"
  25
  26 DECLARE_ALIGNED(8, static const uint64_t, ff_pb_3_1  ) = 0x0103010301030103ULL;
  27
  28 /***********************************/
  29 /* IDCT */
  30 #define IDCT_ADD_FUNC(NUM, DEPTH, OPT) \
  31 void ff_h264_idct ## NUM ## _add_ ## DEPTH ## _ ## OPT (uint8_t *dst, int16_t *block, int stride);
  32
  33 IDCT_ADD_FUNC(, 8, mmx)
  34 IDCT_ADD_FUNC(, 10, sse2)
  35 IDCT_ADD_FUNC(_dc, 8, mmx2)
  36 IDCT_ADD_FUNC(_dc, 10, mmx2)
  37 IDCT_ADD_FUNC(8_dc, 8, mmx2)
  38 IDCT_ADD_FUNC(8_dc, 10, sse2)
  39 IDCT_ADD_FUNC(8, 8, mmx)
  40 IDCT_ADD_FUNC(8, 8, sse2)
  41 IDCT_ADD_FUNC(8, 10, sse2)
  42 #if HAVE_AVX
  43 IDCT_ADD_FUNC(, 10, avx)
  44 IDCT_ADD_FUNC(8_dc, 10, avx)
  45 IDCT_ADD_FUNC(8, 10, avx)
  46 #endif
  47
  48
  49 #define IDCT_ADD_REP_FUNC(NUM, REP, DEPTH, OPT) \
  50 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \
  51                               (uint8_t *dst, const int *block_offset, \
  52                               DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
  53
  54 IDCT_ADD_REP_FUNC(8, 4, 8, mmx)
  55 IDCT_ADD_REP_FUNC(8, 4, 8, mmx2)
  56 IDCT_ADD_REP_FUNC(8, 4, 8, sse2)
  57 IDCT_ADD_REP_FUNC(8, 4, 10, sse2)
  58 IDCT_ADD_REP_FUNC(8, 4, 10, avx)
  59 IDCT_ADD_REP_FUNC(, 16, 8, mmx)
  60 IDCT_ADD_REP_FUNC(, 16, 8, mmx2)
  61 IDCT_ADD_REP_FUNC(, 16, 8, sse2)
  62 IDCT_ADD_REP_FUNC(, 16, 10, sse2)
  63 IDCT_ADD_REP_FUNC(, 16intra, 8, mmx)
  64 IDCT_ADD_REP_FUNC(, 16intra, 8, mmx2)
  65 IDCT_ADD_REP_FUNC(, 16intra, 8, sse2)
  66 IDCT_ADD_REP_FUNC(, 16intra, 10, sse2)
  67 #if HAVE_AVX
  68 IDCT_ADD_REP_FUNC(, 16, 10, avx)
  69 IDCT_ADD_REP_FUNC(, 16intra, 10, avx)
  70 #endif
  71
  72
  73 #define IDCT_ADD_REP_FUNC2(NUM, REP, DEPTH, OPT) \
  74 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT \
  75                               (uint8_t **dst, const int *block_offset, \
  76                               DCTELEM *block, int stride, const uint8_t nnzc[6*8]);
  77 IDCT_ADD_REP_FUNC2(, 8, 8, mmx)
  78 IDCT_ADD_REP_FUNC2(, 8, 8, mmx2)
  79 IDCT_ADD_REP_FUNC2(, 8, 8, sse2)
  80 IDCT_ADD_REP_FUNC2(, 8, 10, sse2)
  81 #if HAVE_AVX
  82 IDCT_ADD_REP_FUNC2(, 8, 10, avx)
  83 #endif
  84
  85 void ff_h264_luma_dc_dequant_idct_mmx (DCTELEM *output, DCTELEM *input, int qmul);
  86 void ff_h264_luma_dc_dequant_idct_sse2(DCTELEM *output, DCTELEM *input, int qmul);
  87
  88 /***********************************/
  89 /* deblocking */
  90
  91 #define h264_loop_filter_strength_iteration_mmx2(bS, nz, ref, mv, bidir, edges, step, mask_mv, dir, d_idx, mask_dir) \
  92     do { \
  93         x86_reg b_idx; \
  94         mask_mv <<= 3; \
  95         for( b_idx=0; b_idx<edges; b_idx+=step ) { \
  96             if (!mask_dir) \
  97             __asm__ volatile( \
  98                     "pxor %%mm0, %%mm0 \n\t" \
  99                     :: \
 100             ); \
 101             if(!(mask_mv & b_idx)) { \
 102                 if(bidir) { \
 103                     __asm__ volatile( \
 104                         "movd         %a3(%0,%2), %%mm2 \n" \
 105                         "punpckldq    %a4(%0,%2), %%mm2 \n" /* { ref0[bn], ref1[bn] } */ \
 106                         "pshufw $0x44, 12(%0,%2), %%mm0 \n" /* { ref0[b], ref0[b] } */ \
 107                         "pshufw $0x44, 52(%0,%2), %%mm1 \n" /* { ref1[b], ref1[b] } */ \
 108                         "pshufw $0x4E, %%mm2, %%mm3 \n" \
 109                         "psubb         %%mm2, %%mm0 \n" /* { ref0[b]!=ref0[bn], ref0[b]!=ref1[bn] } */ \
 110                         "psubb         %%mm3, %%mm1 \n" /* { ref1[b]!=ref1[bn], ref1[b]!=ref0[bn] } */ \
 111  \
 112                         "por           %%mm1, %%mm0 \n" \
 113                         "movq   %a5(%1,%2,4), %%mm1 \n" \
 114                         "movq   %a6(%1,%2,4), %%mm2 \n" \
 115                         "movq          %%mm1, %%mm3 \n" \
 116                         "movq          %%mm2, %%mm4 \n" \
 117                         "psubw   48(%1,%2,4), %%mm1 \n" \
 118                         "psubw   56(%1,%2,4), %%mm2 \n" \
 119                         "psubw  208(%1,%2,4), %%mm3 \n" \
 120                         "psubw  216(%1,%2,4), %%mm4 \n" \
 121                         "packsswb      %%mm2, %%mm1 \n" \
 122                         "packsswb      %%mm4, %%mm3 \n" \
 123                         "paddb         %%mm6, %%mm1 \n" \
 124                         "paddb         %%mm6, %%mm3 \n" \
 125                         "psubusb       %%mm5, %%mm1 \n" /* abs(mv[b] - mv[bn]) >= limit */ \
 126                         "psubusb       %%mm5, %%mm3 \n" \
 127                         "packsswb      %%mm3, %%mm1 \n" \
 128  \
 129                         "por           %%mm1, %%mm0 \n" \
 130                         "movq   %a7(%1,%2,4), %%mm1 \n" \
 131                         "movq   %a8(%1,%2,4), %%mm2 \n" \
 132                         "movq          %%mm1, %%mm3 \n" \
 133                         "movq          %%mm2, %%mm4 \n" \
 134                         "psubw   48(%1,%2,4), %%mm1 \n" \
 135                         "psubw   56(%1,%2,4), %%mm2 \n" \
 136                         "psubw  208(%1,%2,4), %%mm3 \n" \
 137                         "psubw  216(%1,%2,4), %%mm4 \n" \
 138                         "packsswb      %%mm2, %%mm1 \n" \
 139                         "packsswb      %%mm4, %%mm3 \n" \
 140                         "paddb         %%mm6, %%mm1 \n" \
 141                         "paddb         %%mm6, %%mm3 \n" \
 142                         "psubusb       %%mm5, %%mm1 \n" /* abs(mv[b] - mv[bn]) >= limit */ \
 143                         "psubusb       %%mm5, %%mm3 \n" \
 144                         "packsswb      %%mm3, %%mm1 \n" \
 145  \
 146                         "pshufw $0x4E, %%mm1, %%mm1 \n" \
 147                         "por           %%mm1, %%mm0 \n" \
 148                         "pshufw $0x4E, %%mm0, %%mm1 \n" \
 149                         "pminub        %%mm1, %%mm0 \n" \
 150                         ::"r"(ref), \
 151                           "r"(mv), \
 152                           "r"(b_idx), \
 153                           "i"(d_idx+12), \
 154                           "i"(d_idx+52), \
 155                           "i"(d_idx*4+48), \
 156                           "i"(d_idx*4+56), \
 157                           "i"(d_idx*4+208), \
 158                           "i"(d_idx*4+216) \
 159                     ); \
 160                 } else { \
 161                     __asm__ volatile( \
 162                         "movd   12(%0,%2), %%mm0 \n" \
 163                         "psubb %a3(%0,%2), %%mm0 \n" /* ref[b] != ref[bn] */ \
 164                         "movq   48(%1,%2,4), %%mm1 \n" \
 165                         "movq   56(%1,%2,4), %%mm2 \n" \
 166                         "psubw %a4(%1,%2,4), %%mm1 \n" \
 167                         "psubw %a5(%1,%2,4), %%mm2 \n" \
 168                         "packsswb   %%mm2, %%mm1 \n" \
 169                         "paddb      %%mm6, %%mm1 \n" \
 170                         "psubusb    %%mm5, %%mm1 \n" /* abs(mv[b] - mv[bn]) >= limit */ \
 171                         "packsswb   %%mm1, %%mm1 \n" \
 172                         "por        %%mm1, %%mm0 \n" \
 173                         ::"r"(ref), \
 174                           "r"(mv), \
 175                           "r"(b_idx), \
 176                           "i"(d_idx+12), \
 177                           "i"(d_idx*4+48), \
 178                           "i"(d_idx*4+56) \
 179                     ); \
 180                 } \
 181             } \
 182             __asm__ volatile( \
 183                 "movd 12(%0,%1), %%mm1 \n" \
 184                 "por %a2(%0,%1), %%mm1 \n" /* nnz[b] || nnz[bn] */ \
 185                 ::"r"(nnz), \
 186                   "r"(b_idx), \
 187                   "i"(d_idx+12) \
 188             ); \
 189             __asm__ volatile( \
 190                 "pminub    %%mm7, %%mm1 \n" \
 191                 "pminub    %%mm7, %%mm0 \n" \
 192                 "psllw        $1, %%mm1 \n" \
 193                 "pxor      %%mm2, %%mm2 \n" \
 194                 "pmaxub    %%mm0, %%mm1 \n" \
 195                 "punpcklbw %%mm2, %%mm1 \n" \
 196                 "movq      %%mm1, %a1(%0,%2) \n" \
 197                 ::"r"(bS), \
 198                   "i"(32*dir), \
 199                   "r"(b_idx) \
 200                 :"memory" \
 201             ); \
 202         } \
 203     } while (0)
 204
 205 static void h264_loop_filter_strength_mmx2( int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
 206                                             int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field ) {
 207     __asm__ volatile(
 208         "movq %0, %%mm7 \n"
 209         "movq %1, %%mm6 \n"
 210         ::"m"(ff_pb_1), "m"(ff_pb_3)
 211     );
 212     if(field)
 213         __asm__ volatile(
 214             "movq %0, %%mm6 \n"
 215             ::"m"(ff_pb_3_1)
 216         );
 217     __asm__ volatile(
 218         "movq  %%mm6, %%mm5 \n"
 219         "paddb %%mm5, %%mm5 \n"
 220     :);
 221
 222     // could do a special case for dir==0 && edges==1, but it only reduces the
 223     // average filter time by 1.2%
 224     step  <<= 3;
 225     edges <<= 3;
 226     h264_loop_filter_strength_iteration_mmx2(bS, nnz, ref, mv, bidir, edges, step, mask_mv1, 1, -8,  0);
 227     h264_loop_filter_strength_iteration_mmx2(bS, nnz, ref, mv, bidir,    32,    8, mask_mv0, 0, -1, -1);
 228
 229     __asm__ volatile(
 230         "movq   (%0), %%mm0 \n\t"
 231         "movq  8(%0), %%mm1 \n\t"
 232         "movq 16(%0), %%mm2 \n\t"
 233         "movq 24(%0), %%mm3 \n\t"
 234         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4)
 235         "movq %%mm0,   (%0) \n\t"
 236         "movq %%mm3,  8(%0) \n\t"
 237         "movq %%mm4, 16(%0) \n\t"
 238         "movq %%mm2, 24(%0) \n\t"
 239         ::"r"(bS[0])
 240         :"memory"
 241     );
 242 }
 243
 244 #define LF_FUNC(DIR, TYPE, DEPTH, OPT) \
 245 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *pix, int stride, \
 246                                                                 int alpha, int beta, int8_t *tc0);
 247 #define LF_IFUNC(DIR, TYPE, DEPTH, OPT) \
 248 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT (uint8_t *pix, int stride, \
 249                                                                 int alpha, int beta);
 250
 251 #define LF_FUNCS(type, depth)\
 252 LF_FUNC (h,  chroma,       depth, mmxext)\
 253 LF_IFUNC(h,  chroma_intra, depth, mmxext)\
 254 LF_FUNC (v,  chroma,       depth, mmxext)\
 255 LF_IFUNC(v,  chroma_intra, depth, mmxext)\
 256 LF_FUNC (h,  luma,         depth, mmxext)\
 257 LF_IFUNC(h,  luma_intra,   depth, mmxext)\
 258 LF_FUNC (h,  luma,         depth, sse2)\
 259 LF_IFUNC(h,  luma_intra,   depth, sse2)\
 260 LF_FUNC (v,  luma,         depth, sse2)\
 261 LF_IFUNC(v,  luma_intra,   depth, sse2)\
 262 LF_FUNC (h,  chroma,       depth, sse2)\
 263 LF_IFUNC(h,  chroma_intra, depth, sse2)\
 264 LF_FUNC (v,  chroma,       depth, sse2)\
 265 LF_IFUNC(v,  chroma_intra, depth, sse2)\
 266 LF_FUNC (h,  luma,         depth,  avx)\
 267 LF_IFUNC(h,  luma_intra,   depth,  avx)\
 268 LF_FUNC (v,  luma,         depth,  avx)\
 269 LF_IFUNC(v,  luma_intra,   depth,  avx)\
 270 LF_FUNC (h,  chroma,       depth,  avx)\
 271 LF_IFUNC(h,  chroma_intra, depth,  avx)\
 272 LF_FUNC (v,  chroma,       depth,  avx)\
 273 LF_IFUNC(v,  chroma_intra, depth,  avx)
 274
 275 LF_FUNCS( uint8_t,  8)
 276 LF_FUNCS(uint16_t, 10)
 277
 278 #if ARCH_X86_32
 279 LF_FUNC (v8, luma,             8, mmxext)
 280 static void ff_deblock_v_luma_8_mmxext(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
 281 {
 282     if((tc0[0] & tc0[1]) >= 0)
 283         ff_deblock_v8_luma_8_mmxext(pix+0, stride, alpha, beta, tc0);
 284     if((tc0[2] & tc0[3]) >= 0)
 285         ff_deblock_v8_luma_8_mmxext(pix+8, stride, alpha, beta, tc0+2);
 286 }
 287 LF_IFUNC(v8, luma_intra,        8, mmxext)
 288 static void ff_deblock_v_luma_intra_8_mmxext(uint8_t *pix, int stride, int alpha, int beta)
 289 {
 290     ff_deblock_v8_luma_intra_8_mmxext(pix+0, stride, alpha, beta);
 291     ff_deblock_v8_luma_intra_8_mmxext(pix+8, stride, alpha, beta);
 292 }
 293 #endif /* ARCH_X86_32 */
 294
 295 LF_FUNC (v,  luma,            10, mmxext)
 296 LF_IFUNC(v,  luma_intra,      10, mmxext)
 297
 298 /***********************************/
 299 /* weighted prediction */
 300
 301 #define H264_WEIGHT(W, H, OPT) \
 302 void ff_h264_weight_ ## W ## x ## H ## _ ## OPT(uint8_t *dst, \
 303     int stride, int log2_denom, int weight, int offset);
 304
 305 #define H264_BIWEIGHT(W, H, OPT) \
 306 void ff_h264_biweight_ ## W ## x ## H ## _ ## OPT(uint8_t *dst, \
 307     uint8_t *src, int stride, int log2_denom, int weightd, \
 308     int weights, int offset);
 309
 310 #define H264_BIWEIGHT_MMX(W,H) \
 311 H264_WEIGHT  (W, H, mmx2) \
 312 H264_BIWEIGHT(W, H, mmx2)
 313
 314 #define H264_BIWEIGHT_MMX_SSE(W,H) \
 315 H264_BIWEIGHT_MMX(W, H) \
 316 H264_WEIGHT      (W, H, sse2) \
 317 H264_BIWEIGHT    (W, H, sse2) \
 318 H264_BIWEIGHT    (W, H, ssse3)
 319
 320 H264_BIWEIGHT_MMX_SSE(16, 16)
 321 H264_BIWEIGHT_MMX_SSE(16,  8)
 322 H264_BIWEIGHT_MMX_SSE( 8, 16)
 323 H264_BIWEIGHT_MMX_SSE( 8,  8)
 324 H264_BIWEIGHT_MMX_SSE( 8,  4)
 325 H264_BIWEIGHT_MMX    ( 4,  8)
 326 H264_BIWEIGHT_MMX    ( 4,  4)
 327 H264_BIWEIGHT_MMX    ( 4,  2)
 328
 329 #define H264_WEIGHT_10(W, H, DEPTH, OPT) \
 330 void ff_h264_weight_ ## W ## x ## H ## _ ## DEPTH ## _ ## OPT(uint8_t *dst, \
 331     int stride, int log2_denom, int weight, int offset);
 332
 333 #define H264_BIWEIGHT_10(W, H, DEPTH, OPT) \
 334 void ff_h264_biweight_ ## W ## x ## H ## _ ## DEPTH ## _ ## OPT \
 335     (uint8_t *dst, uint8_t *src, int stride, int log2_denom, \
 336      int weightd, int weights, int offset);
 337
 338 #define H264_BIWEIGHT_10_SSE(W, H, DEPTH) \
 339 H264_WEIGHT_10  (W, H, DEPTH, sse2) \
 340 H264_WEIGHT_10  (W, H, DEPTH, sse4) \
 341 H264_BIWEIGHT_10(W, H, DEPTH, sse2) \
 342 H264_BIWEIGHT_10(W, H, DEPTH, sse4)
 343
 344 H264_BIWEIGHT_10_SSE(16, 16, 10)
 345 H264_BIWEIGHT_10_SSE(16,  8, 10)
 346 H264_BIWEIGHT_10_SSE( 8, 16, 10)
 347 H264_BIWEIGHT_10_SSE( 8,  8, 10)
 348 H264_BIWEIGHT_10_SSE( 8,  4, 10)
 349 H264_BIWEIGHT_10_SSE( 4,  8, 10)
 350 H264_BIWEIGHT_10_SSE( 4,  4, 10)
 351 H264_BIWEIGHT_10_SSE( 4,  2, 10)
 352
 353 void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth)
 354 {
 355     int mm_flags = av_get_cpu_flags();
 356
 357     if (mm_flags & AV_CPU_FLAG_MMX2) {
 358         c->h264_loop_filter_strength= h264_loop_filter_strength_mmx2;
 359     }
 360
 361     if (bit_depth == 8) {
 362 #if HAVE_YASM
 363     if (mm_flags & AV_CPU_FLAG_MMX) {
 364         c->h264_idct_dc_add         =
 365         c->h264_idct_add            = ff_h264_idct_add_8_mmx;
 366         c->h264_idct8_dc_add        =
 367         c->h264_idct8_add           = ff_h264_idct8_add_8_mmx;
 368
 369         c->h264_idct_add16          = ff_h264_idct_add16_8_mmx;
 370         c->h264_idct8_add4          = ff_h264_idct8_add4_8_mmx;
 371         c->h264_idct_add8           = ff_h264_idct_add8_8_mmx;
 372         c->h264_idct_add16intra     = ff_h264_idct_add16intra_8_mmx;
 373         c->h264_luma_dc_dequant_idct= ff_h264_luma_dc_dequant_idct_mmx;
 374
 375         if (mm_flags & AV_CPU_FLAG_MMX2) {
 376             c->h264_idct_dc_add    = ff_h264_idct_dc_add_8_mmx2;
 377             c->h264_idct8_dc_add   = ff_h264_idct8_dc_add_8_mmx2;
 378             c->h264_idct_add16     = ff_h264_idct_add16_8_mmx2;
 379             c->h264_idct8_add4     = ff_h264_idct8_add4_8_mmx2;
 380             c->h264_idct_add8      = ff_h264_idct_add8_8_mmx2;
 381             c->h264_idct_add16intra= ff_h264_idct_add16intra_8_mmx2;
 382
 383             c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_8_mmxext;
 384             c->h264_h_loop_filter_chroma= ff_deblock_h_chroma_8_mmxext;
 385             c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_8_mmxext;
 386             c->h264_h_loop_filter_chroma_intra= ff_deblock_h_chroma_intra_8_mmxext;
 387 #if ARCH_X86_32
 388             c->h264_v_loop_filter_luma= ff_deblock_v_luma_8_mmxext;
 389             c->h264_h_loop_filter_luma= ff_deblock_h_luma_8_mmxext;
 390             c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_mmxext;
 391             c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmxext;
 392 #endif
 393             c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_mmx2;
 394             c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_mmx2;
 395             c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_mmx2;
 396             c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_mmx2;
 397             c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_mmx2;
 398             c->weight_h264_pixels_tab[5]= ff_h264_weight_4x8_mmx2;
 399             c->weight_h264_pixels_tab[6]= ff_h264_weight_4x4_mmx2;
 400             c->weight_h264_pixels_tab[7]= ff_h264_weight_4x2_mmx2;
 401
 402             c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_mmx2;
 403             c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_mmx2;
 404             c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_mmx2;
 405             c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_mmx2;
 406             c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_mmx2;
 407             c->biweight_h264_pixels_tab[5]= ff_h264_biweight_4x8_mmx2;
 408             c->biweight_h264_pixels_tab[6]= ff_h264_biweight_4x4_mmx2;
 409             c->biweight_h264_pixels_tab[7]= ff_h264_biweight_4x2_mmx2;
 410
 411             if (mm_flags&AV_CPU_FLAG_SSE2) {
 412                 c->h264_idct8_add           = ff_h264_idct8_add_8_sse2;
 413
 414                 c->h264_idct_add16          = ff_h264_idct_add16_8_sse2;
 415                 c->h264_idct8_add4          = ff_h264_idct8_add4_8_sse2;
 416                 c->h264_idct_add8           = ff_h264_idct_add8_8_sse2;
 417                 c->h264_idct_add16intra     = ff_h264_idct_add16intra_8_sse2;
 418                 c->h264_luma_dc_dequant_idct= ff_h264_luma_dc_dequant_idct_sse2;
 419
 420                 c->weight_h264_pixels_tab[0]= ff_h264_weight_16x16_sse2;
 421                 c->weight_h264_pixels_tab[1]= ff_h264_weight_16x8_sse2;
 422                 c->weight_h264_pixels_tab[2]= ff_h264_weight_8x16_sse2;
 423                 c->weight_h264_pixels_tab[3]= ff_h264_weight_8x8_sse2;
 424                 c->weight_h264_pixels_tab[4]= ff_h264_weight_8x4_sse2;
 425
 426                 c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_sse2;
 427                 c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_sse2;
 428                 c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_sse2;
 429                 c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_sse2;
 430                 c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_sse2;
 431
 432 #if HAVE_ALIGNED_STACK
 433                 c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_sse2;
 434                 c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_sse2;
 435                 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
 436                 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
 437 #endif
 438             }
 439             if (mm_flags&AV_CPU_FLAG_SSSE3) {
 440                 c->biweight_h264_pixels_tab[0]= ff_h264_biweight_16x16_ssse3;
 441                 c->biweight_h264_pixels_tab[1]= ff_h264_biweight_16x8_ssse3;
 442                 c->biweight_h264_pixels_tab[2]= ff_h264_biweight_8x16_ssse3;
 443                 c->biweight_h264_pixels_tab[3]= ff_h264_biweight_8x8_ssse3;
 444                 c->biweight_h264_pixels_tab[4]= ff_h264_biweight_8x4_ssse3;
 445             }
 446             if (HAVE_AVX && mm_flags&AV_CPU_FLAG_AVX) {
 447 #if HAVE_ALIGNED_STACK
 448                 c->h264_v_loop_filter_luma = ff_deblock_v_luma_8_avx;
 449                 c->h264_h_loop_filter_luma = ff_deblock_h_luma_8_avx;
 450                 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
 451                 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
 452 #endif
 453             }
 454         }
 455     }
 456 #endif
 457     } else if (bit_depth == 10) {
 458 #if HAVE_YASM
 459     if (mm_flags & AV_CPU_FLAG_MMX) {
 460         if (mm_flags & AV_CPU_FLAG_MMX2) {
 461 #if ARCH_X86_32
 462             c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_10_mmxext;
 463             c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_10_mmxext;
 464             c->h264_v_loop_filter_luma= ff_deblock_v_luma_10_mmxext;
 465             c->h264_h_loop_filter_luma= ff_deblock_h_luma_10_mmxext;
 466             c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_mmxext;
 467             c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_mmxext;
 468 #endif
 469             c->h264_idct_dc_add= ff_h264_idct_dc_add_10_mmx2;
 470             if (mm_flags&AV_CPU_FLAG_SSE2) {
 471                 c->h264_idct_add       = ff_h264_idct_add_10_sse2;
 472                 c->h264_idct8_dc_add   = ff_h264_idct8_dc_add_10_sse2;
 473
 474                 c->h264_idct_add16     = ff_h264_idct_add16_10_sse2;
 475                 c->h264_idct_add8      = ff_h264_idct_add8_10_sse2;
 476                 c->h264_idct_add16intra= ff_h264_idct_add16intra_10_sse2;
 477 #if HAVE_ALIGNED_STACK
 478                 c->h264_idct8_add      = ff_h264_idct8_add_10_sse2;
 479                 c->h264_idct8_add4     = ff_h264_idct8_add4_10_sse2;
 480 #endif
 481
 482                 c->weight_h264_pixels_tab[0] = ff_h264_weight_16x16_10_sse2;
 483                 c->weight_h264_pixels_tab[1] = ff_h264_weight_16x8_10_sse2;
 484                 c->weight_h264_pixels_tab[2] = ff_h264_weight_8x16_10_sse2;
 485                 c->weight_h264_pixels_tab[3] = ff_h264_weight_8x8_10_sse2;
 486                 c->weight_h264_pixels_tab[4] = ff_h264_weight_8x4_10_sse2;
 487                 c->weight_h264_pixels_tab[5] = ff_h264_weight_4x8_10_sse2;
 488                 c->weight_h264_pixels_tab[6] = ff_h264_weight_4x4_10_sse2;
 489                 c->weight_h264_pixels_tab[7] = ff_h264_weight_4x2_10_sse2;
 490
 491                 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16x16_10_sse2;
 492                 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_16x8_10_sse2;
 493                 c->biweight_h264_pixels_tab[2] = ff_h264_biweight_8x16_10_sse2;
 494                 c->biweight_h264_pixels_tab[3] = ff_h264_biweight_8x8_10_sse2;
 495                 c->biweight_h264_pixels_tab[4] = ff_h264_biweight_8x4_10_sse2;
 496                 c->biweight_h264_pixels_tab[5] = ff_h264_biweight_4x8_10_sse2;
 497                 c->biweight_h264_pixels_tab[6] = ff_h264_biweight_4x4_10_sse2;
 498                 c->biweight_h264_pixels_tab[7] = ff_h264_biweight_4x2_10_sse2;
 499
 500                 c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_10_sse2;
 501                 c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_10_sse2;
 502 #if HAVE_ALIGNED_STACK
 503                 c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_sse2;
 504                 c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_sse2;
 505                 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
 506                 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
 507 #endif
 508             }
 509             if (mm_flags&AV_CPU_FLAG_SSE4) {
 510                 c->weight_h264_pixels_tab[0] = ff_h264_weight_16x16_10_sse4;
 511                 c->weight_h264_pixels_tab[1] = ff_h264_weight_16x8_10_sse4;
 512                 c->weight_h264_pixels_tab[2] = ff_h264_weight_8x16_10_sse4;
 513                 c->weight_h264_pixels_tab[3] = ff_h264_weight_8x8_10_sse4;
 514                 c->weight_h264_pixels_tab[4] = ff_h264_weight_8x4_10_sse4;
 515                 c->weight_h264_pixels_tab[5] = ff_h264_weight_4x8_10_sse4;
 516                 c->weight_h264_pixels_tab[6] = ff_h264_weight_4x4_10_sse4;
 517                 c->weight_h264_pixels_tab[7] = ff_h264_weight_4x2_10_sse4;
 518
 519                 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16x16_10_sse4;
 520                 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_16x8_10_sse4;
 521                 c->biweight_h264_pixels_tab[2] = ff_h264_biweight_8x16_10_sse4;
 522                 c->biweight_h264_pixels_tab[3] = ff_h264_biweight_8x8_10_sse4;
 523                 c->biweight_h264_pixels_tab[4] = ff_h264_biweight_8x4_10_sse4;
 524                 c->biweight_h264_pixels_tab[5] = ff_h264_biweight_4x8_10_sse4;
 525                 c->biweight_h264_pixels_tab[6] = ff_h264_biweight_4x4_10_sse4;
 526                 c->biweight_h264_pixels_tab[7] = ff_h264_biweight_4x2_10_sse4;
 527             }
 528 #if HAVE_AVX
 529             if (mm_flags&AV_CPU_FLAG_AVX) {
 530                 c->h264_idct_dc_add    =
 531                 c->h264_idct_add       = ff_h264_idct_add_10_avx;
 532                 c->h264_idct8_dc_add   = ff_h264_idct8_dc_add_10_avx;
 533
 534                 c->h264_idct_add16     = ff_h264_idct_add16_10_avx;
 535                 c->h264_idct_add8      = ff_h264_idct_add8_10_avx;
 536                 c->h264_idct_add16intra= ff_h264_idct_add16intra_10_avx;
 537 #if HAVE_ALIGNED_STACK
 538                 c->h264_idct8_add      = ff_h264_idct8_add_10_avx;
 539                 c->h264_idct8_add4     = ff_h264_idct8_add4_10_avx;
 540 #endif
 541
 542                 c->h264_v_loop_filter_chroma= ff_deblock_v_chroma_10_avx;
 543                 c->h264_v_loop_filter_chroma_intra= ff_deblock_v_chroma_intra_10_avx;
 544 #if HAVE_ALIGNED_STACK
 545                 c->h264_v_loop_filter_luma = ff_deblock_v_luma_10_avx;
 546                 c->h264_h_loop_filter_luma = ff_deblock_h_luma_10_avx;
 547                 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_avx;
 548                 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_avx;
 549 #endif
 550             }
 551 #endif /* HAVE_AVX */
 552         }
 553     }
 554 #endif
 555     }
 556 }