git.sesse.net Git - ffmpeg/blob - libavcodec/x86/h264dsp_init.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/attributes.h"
  22 #include "libavutil/cpu.h"
  23 #include "libavutil/x86/asm.h"
  24 #include "libavutil/x86/cpu.h"
  25 #include "libavcodec/h264dsp.h"
  26 #include "dsputil_mmx.h"
  27
  28 /***********************************/
  29 /* IDCT */
  30 #define IDCT_ADD_FUNC(NUM, DEPTH, OPT)                                  \
  31 void ff_h264_idct ## NUM ## _add_ ## DEPTH ## _ ## OPT(uint8_t *dst,    \
  32                                                        int16_t *block,  \
  33                                                        int stride);
  34
  35 IDCT_ADD_FUNC(, 8, mmx)
  36 IDCT_ADD_FUNC(, 10, sse2)
  37 IDCT_ADD_FUNC(_dc, 8, mmxext)
  38 IDCT_ADD_FUNC(_dc, 10, mmxext)
  39 IDCT_ADD_FUNC(8_dc, 8, mmxext)
  40 IDCT_ADD_FUNC(8_dc, 10, sse2)
  41 IDCT_ADD_FUNC(8, 8, mmx)
  42 IDCT_ADD_FUNC(8, 8, sse2)
  43 IDCT_ADD_FUNC(8, 10, sse2)
  44 IDCT_ADD_FUNC(, 10, avx)
  45 IDCT_ADD_FUNC(8_dc, 10, avx)
  46 IDCT_ADD_FUNC(8, 10, avx)
  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      int16_t *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, mmxext)
  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, mmxext)
  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, mmxext)
  65 IDCT_ADD_REP_FUNC(, 16intra, 8, sse2)
  66 IDCT_ADD_REP_FUNC(, 16intra, 10, sse2)
  67 IDCT_ADD_REP_FUNC(, 16, 10, avx)
  68 IDCT_ADD_REP_FUNC(, 16intra, 10, avx)
  69
  70
  71 #define IDCT_ADD_REP_FUNC2(NUM, REP, DEPTH, OPT)                      \
  72 void ff_h264_idct ## NUM ## _add ## REP ## _ ## DEPTH ## _ ## OPT     \
  73     (uint8_t **dst, const int *block_offset,                          \
  74      int16_t *block, int stride, const uint8_t nnzc[6 * 8]);
  75
  76 IDCT_ADD_REP_FUNC2(, 8, 8, mmx)
  77 IDCT_ADD_REP_FUNC2(, 8, 8, mmxext)
  78 IDCT_ADD_REP_FUNC2(, 8, 8, sse2)
  79 IDCT_ADD_REP_FUNC2(, 8, 10, sse2)
  80 IDCT_ADD_REP_FUNC2(, 8, 10, avx)
  81
  82 void ff_h264_luma_dc_dequant_idct_mmx(int16_t *output, int16_t *input, int qmul);
  83 void ff_h264_luma_dc_dequant_idct_sse2(int16_t *output, int16_t *input, int qmul);
  84
  85 /***********************************/
  86 /* deblocking */
  87
  88 void ff_h264_loop_filter_strength_mmxext(int16_t bS[2][4][4], uint8_t nnz[40],
  89                                          int8_t ref[2][40],
  90                                          int16_t mv[2][40][2],
  91                                          int bidir, int edges, int step,
  92                                          int mask_mv0, int mask_mv1, int field);
  93
  94 #define LF_FUNC(DIR, TYPE, DEPTH, OPT)                                        \
  95 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix,  \
  96                                                                int stride,    \
  97                                                                int alpha,     \
  98                                                                int beta,      \
  99                                                                int8_t *tc0);
 100 #define LF_IFUNC(DIR, TYPE, DEPTH, OPT) \
 101 void ff_deblock_ ## DIR ## _ ## TYPE ## _ ## DEPTH ## _ ## OPT(uint8_t *pix,  \
 102                                                                int stride,    \
 103                                                                int alpha,     \
 104                                                                int beta);
 105
 106 #define LF_FUNCS(type, depth)                   \
 107 LF_FUNC(h,  chroma,       depth, mmxext)        \
 108 LF_IFUNC(h, chroma_intra, depth, mmxext)        \
 109 LF_FUNC(v,  chroma,       depth, mmxext)        \
 110 LF_IFUNC(v, chroma_intra, depth, mmxext)        \
 111 LF_FUNC(h,  luma,         depth, mmxext)        \
 112 LF_IFUNC(h, luma_intra,   depth, mmxext)        \
 113 LF_FUNC(h,  luma,         depth, sse2)          \
 114 LF_IFUNC(h, luma_intra,   depth, sse2)          \
 115 LF_FUNC(v,  luma,         depth, sse2)          \
 116 LF_IFUNC(v, luma_intra,   depth, sse2)          \
 117 LF_FUNC(h,  chroma,       depth, sse2)          \
 118 LF_IFUNC(h, chroma_intra, depth, sse2)          \
 119 LF_FUNC(v,  chroma,       depth, sse2)          \
 120 LF_IFUNC(v, chroma_intra, depth, sse2)          \
 121 LF_FUNC(h,  luma,         depth, avx)           \
 122 LF_IFUNC(h, luma_intra,   depth, avx)           \
 123 LF_FUNC(v,  luma,         depth, avx)           \
 124 LF_IFUNC(v, luma_intra,   depth, avx)           \
 125 LF_FUNC(h,  chroma,       depth, avx)           \
 126 LF_IFUNC(h, chroma_intra, depth, avx)           \
 127 LF_FUNC(v,  chroma,       depth, avx)           \
 128 LF_IFUNC(v, chroma_intra, depth, avx)
 129
 130 LF_FUNCS(uint8_t,   8)
 131 LF_FUNCS(uint16_t, 10)
 132
 133 #if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL
 134 LF_FUNC(v8, luma, 8, mmxext)
 135 static void ff_deblock_v_luma_8_mmxext(uint8_t *pix, int stride, int alpha,
 136                                        int beta, int8_t *tc0)
 137 {
 138     if ((tc0[0] & tc0[1]) >= 0)
 139         ff_deblock_v8_luma_8_mmxext(pix + 0, stride, alpha, beta, tc0);
 140     if ((tc0[2] & tc0[3]) >= 0)
 141         ff_deblock_v8_luma_8_mmxext(pix + 8, stride, alpha, beta, tc0 + 2);
 142 }
 143 LF_IFUNC(v8, luma_intra, 8, mmxext)
 144 static void ff_deblock_v_luma_intra_8_mmxext(uint8_t *pix, int stride,
 145                                              int alpha, int beta)
 146 {
 147     ff_deblock_v8_luma_intra_8_mmxext(pix + 0, stride, alpha, beta);
 148     ff_deblock_v8_luma_intra_8_mmxext(pix + 8, stride, alpha, beta);
 149 }
 150 #endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */
 151
 152 LF_FUNC(v,  luma,       10, mmxext)
 153 LF_IFUNC(v, luma_intra, 10, mmxext)
 154
 155 /***********************************/
 156 /* weighted prediction */
 157
 158 #define H264_WEIGHT(W, OPT)                                             \
 159 void ff_h264_weight_ ## W ## _ ## OPT(uint8_t *dst, int stride,         \
 160                                       int height, int log2_denom,       \
 161                                       int weight, int offset);
 162
 163 #define H264_BIWEIGHT(W, OPT)                                           \
 164 void ff_h264_biweight_ ## W ## _ ## OPT(uint8_t *dst, uint8_t *src,     \
 165                                         int stride, int height,         \
 166                                         int log2_denom, int weightd,    \
 167                                         int weights, int offset);
 168
 169 #define H264_BIWEIGHT_MMX(W)                    \
 170     H264_WEIGHT(W, mmxext)                      \
 171     H264_BIWEIGHT(W, mmxext)
 172
 173 #define H264_BIWEIGHT_MMX_SSE(W)                \
 174     H264_BIWEIGHT_MMX(W)                        \
 175     H264_WEIGHT(W, sse2)                        \
 176     H264_BIWEIGHT(W, sse2)                      \
 177     H264_BIWEIGHT(W, ssse3)
 178
 179 H264_BIWEIGHT_MMX_SSE(16)
 180 H264_BIWEIGHT_MMX_SSE(8)
 181 H264_BIWEIGHT_MMX(4)
 182
 183 #define H264_WEIGHT_10(W, DEPTH, OPT)                                   \
 184 void ff_h264_weight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst,       \
 185                                                     int stride,         \
 186                                                     int height,         \
 187                                                     int log2_denom,     \
 188                                                     int weight,         \
 189                                                     int offset);
 190
 191 #define H264_BIWEIGHT_10(W, DEPTH, OPT)                                 \
 192 void ff_h264_biweight_ ## W ## _ ## DEPTH ## _ ## OPT(uint8_t *dst,     \
 193                                                       uint8_t *src,     \
 194                                                       int stride,       \
 195                                                       int height,       \
 196                                                       int log2_denom,   \
 197                                                       int weightd,      \
 198                                                       int weights,      \
 199                                                       int offset);
 200
 201 #define H264_BIWEIGHT_10_SSE(W, DEPTH)          \
 202     H264_WEIGHT_10(W, DEPTH, sse2)              \
 203     H264_WEIGHT_10(W, DEPTH, sse4)              \
 204     H264_BIWEIGHT_10(W, DEPTH, sse2)            \
 205     H264_BIWEIGHT_10(W, DEPTH, sse4)
 206
 207 H264_BIWEIGHT_10_SSE(16, 10)
 208 H264_BIWEIGHT_10_SSE(8,  10)
 209 H264_BIWEIGHT_10_SSE(4,  10)
 210
 211 av_cold void ff_h264dsp_init_x86(H264DSPContext *c, const int bit_depth,
 212                                  const int chroma_format_idc)
 213 {
 214 #if HAVE_YASM
 215     int mm_flags = av_get_cpu_flags();
 216
 217     if (chroma_format_idc == 1 && EXTERNAL_MMXEXT(mm_flags))
 218         c->h264_loop_filter_strength = ff_h264_loop_filter_strength_mmxext;
 219
 220     if (bit_depth == 8) {
 221         if (EXTERNAL_MMX(mm_flags)) {
 222             c->h264_idct_dc_add   =
 223             c->h264_idct_add      = ff_h264_idct_add_8_mmx;
 224             c->h264_idct8_dc_add  =
 225             c->h264_idct8_add     = ff_h264_idct8_add_8_mmx;
 226
 227             c->h264_idct_add16 = ff_h264_idct_add16_8_mmx;
 228             c->h264_idct8_add4 = ff_h264_idct8_add4_8_mmx;
 229             if (chroma_format_idc == 1)
 230                 c->h264_idct_add8 = ff_h264_idct_add8_8_mmx;
 231             c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmx;
 232             if (mm_flags & AV_CPU_FLAG_CMOV)
 233                 c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_mmx;
 234
 235             if (EXTERNAL_MMXEXT(mm_flags)) {
 236                 c->h264_idct_dc_add  = ff_h264_idct_dc_add_8_mmxext;
 237                 c->h264_idct8_dc_add = ff_h264_idct8_dc_add_8_mmxext;
 238                 c->h264_idct_add16   = ff_h264_idct_add16_8_mmxext;
 239                 c->h264_idct8_add4   = ff_h264_idct8_add4_8_mmxext;
 240                 if (chroma_format_idc == 1)
 241                     c->h264_idct_add8 = ff_h264_idct_add8_8_mmxext;
 242                 c->h264_idct_add16intra = ff_h264_idct_add16intra_8_mmxext;
 243
 244                 c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_8_mmxext;
 245                 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_8_mmxext;
 246                 if (chroma_format_idc == 1) {
 247                     c->h264_h_loop_filter_chroma       = ff_deblock_h_chroma_8_mmxext;
 248                     c->h264_h_loop_filter_chroma_intra = ff_deblock_h_chroma_intra_8_mmxext;
 249                 }
 250 #if ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL
 251                 c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_mmxext;
 252                 c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_mmxext;
 253                 c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_mmxext;
 254                 c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_mmxext;
 255 #endif /* ARCH_X86_32 && HAVE_MMXEXT_EXTERNAL */
 256                 c->weight_h264_pixels_tab[0] = ff_h264_weight_16_mmxext;
 257                 c->weight_h264_pixels_tab[1] = ff_h264_weight_8_mmxext;
 258                 c->weight_h264_pixels_tab[2] = ff_h264_weight_4_mmxext;
 259
 260                 c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_mmxext;
 261                 c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_mmxext;
 262                 c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_mmxext;
 263
 264                 if (EXTERNAL_SSE2(mm_flags)) {
 265                     c->h264_idct8_add  = ff_h264_idct8_add_8_sse2;
 266
 267                     c->h264_idct_add16 = ff_h264_idct_add16_8_sse2;
 268                     c->h264_idct8_add4 = ff_h264_idct8_add4_8_sse2;
 269                     if (chroma_format_idc == 1)
 270                         c->h264_idct_add8 = ff_h264_idct_add8_8_sse2;
 271                     c->h264_idct_add16intra      = ff_h264_idct_add16intra_8_sse2;
 272                     c->h264_luma_dc_dequant_idct = ff_h264_luma_dc_dequant_idct_sse2;
 273
 274                     c->weight_h264_pixels_tab[0] = ff_h264_weight_16_sse2;
 275                     c->weight_h264_pixels_tab[1] = ff_h264_weight_8_sse2;
 276
 277                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_sse2;
 278                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_sse2;
 279
 280                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_sse2;
 281                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_sse2;
 282                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_sse2;
 283                     c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_sse2;
 284                 }
 285                 if (EXTERNAL_SSSE3(mm_flags)) {
 286                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_ssse3;
 287                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_ssse3;
 288                 }
 289                 if (EXTERNAL_AVX(mm_flags)) {
 290                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_8_avx;
 291                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_8_avx;
 292                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_8_avx;
 293                     c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_8_avx;
 294                 }
 295             }
 296         }
 297     } else if (bit_depth == 10) {
 298         if (EXTERNAL_MMX(mm_flags)) {
 299             if (EXTERNAL_MMXEXT(mm_flags)) {
 300 #if ARCH_X86_32
 301                 c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_mmxext;
 302                 c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_mmxext;
 303                 c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_mmxext;
 304                 c->h264_h_loop_filter_luma         = ff_deblock_h_luma_10_mmxext;
 305                 c->h264_v_loop_filter_luma_intra   = ff_deblock_v_luma_intra_10_mmxext;
 306                 c->h264_h_loop_filter_luma_intra   = ff_deblock_h_luma_intra_10_mmxext;
 307 #endif /* ARCH_X86_32 */
 308                 c->h264_idct_dc_add = ff_h264_idct_dc_add_10_mmxext;
 309                 if (EXTERNAL_SSE2(mm_flags)) {
 310                     c->h264_idct_add     = ff_h264_idct_add_10_sse2;
 311                     c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_sse2;
 312
 313                     c->h264_idct_add16 = ff_h264_idct_add16_10_sse2;
 314                     if (chroma_format_idc == 1)
 315                         c->h264_idct_add8 = ff_h264_idct_add8_10_sse2;
 316                     c->h264_idct_add16intra = ff_h264_idct_add16intra_10_sse2;
 317 #if HAVE_ALIGNED_STACK
 318                     c->h264_idct8_add  = ff_h264_idct8_add_10_sse2;
 319                     c->h264_idct8_add4 = ff_h264_idct8_add4_10_sse2;
 320 #endif /* HAVE_ALIGNED_STACK */
 321
 322                     c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse2;
 323                     c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse2;
 324                     c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse2;
 325
 326                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse2;
 327                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse2;
 328                     c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse2;
 329
 330                     c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_sse2;
 331                     c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_sse2;
 332 #if HAVE_ALIGNED_STACK
 333                     c->h264_v_loop_filter_luma       = ff_deblock_v_luma_10_sse2;
 334                     c->h264_h_loop_filter_luma       = ff_deblock_h_luma_10_sse2;
 335                     c->h264_v_loop_filter_luma_intra = ff_deblock_v_luma_intra_10_sse2;
 336                     c->h264_h_loop_filter_luma_intra = ff_deblock_h_luma_intra_10_sse2;
 337 #endif /* HAVE_ALIGNED_STACK */
 338                 }
 339                 if (EXTERNAL_SSE4(mm_flags)) {
 340                     c->weight_h264_pixels_tab[0] = ff_h264_weight_16_10_sse4;
 341                     c->weight_h264_pixels_tab[1] = ff_h264_weight_8_10_sse4;
 342                     c->weight_h264_pixels_tab[2] = ff_h264_weight_4_10_sse4;
 343
 344                     c->biweight_h264_pixels_tab[0] = ff_h264_biweight_16_10_sse4;
 345                     c->biweight_h264_pixels_tab[1] = ff_h264_biweight_8_10_sse4;
 346                     c->biweight_h264_pixels_tab[2] = ff_h264_biweight_4_10_sse4;
 347                 }
 348                 if (EXTERNAL_AVX(mm_flags)) {
 349                     c->h264_idct_dc_add  =
 350                     c->h264_idct_add     = ff_h264_idct_add_10_avx;
 351                     c->h264_idct8_dc_add = ff_h264_idct8_dc_add_10_avx;
 352
 353                     c->h264_idct_add16 = ff_h264_idct_add16_10_avx;
 354                     if (chroma_format_idc == 1)
 355                         c->h264_idct_add8 = ff_h264_idct_add8_10_avx;
 356                     c->h264_idct_add16intra = ff_h264_idct_add16intra_10_avx;
 357 #if HAVE_ALIGNED_STACK
 358                     c->h264_idct8_add  = ff_h264_idct8_add_10_avx;
 359                     c->h264_idct8_add4 = ff_h264_idct8_add4_10_avx;
 360 #endif /* HAVE_ALIGNED_STACK */
 361
 362                     c->h264_v_loop_filter_chroma       = ff_deblock_v_chroma_10_avx;
 363                     c->h264_v_loop_filter_chroma_intra = ff_deblock_v_chroma_intra_10_avx;
 364 #if HAVE_ALIGNED_STACK
 365                     c->h264_v_loop_filter_luma         = ff_deblock_v_luma_10_avx;
 366                     c->h264_h_loop_filter_luma         = ff_deblock_h_luma_10_avx;
 367                     c->h264_v_loop_filter_luma_intra   = ff_deblock_v_luma_intra_10_avx;
 368                     c->h264_h_loop_filter_luma_intra   = ff_deblock_h_luma_intra_10_avx;
 369 #endif /* HAVE_ALIGNED_STACK */
 370                 }
 371             }
 372         }
 373     }
 374 #endif
 375 }