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

   1 /*
   2  * Copyright (c) 2000, 2001 Fabrice Bellard
   3  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
   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
  22 #include "config.h"
  23 #include "libavutil/attributes.h"
  24 #include "libavutil/cpu.h"
  25 #include "libavutil/internal.h"
  26 #include "libavutil/x86/asm.h"
  27 #include "libavutil/x86/cpu.h"
  28 #include "libavcodec/avcodec.h"
  29 #include "libavcodec/dsputil.h"
  30 #include "libavcodec/pixels.h"
  31 #include "libavcodec/simple_idct.h"
  32 #include "libavcodec/version.h"
  33 #include "dsputil_x86.h"
  34 #include "fpel.h"
  35 #include "idct_xvid.h"
  36
  37 void ff_put_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  38                               int dstStride, int src1Stride, int h);
  39 void ff_put_no_rnd_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1,
  40                                      uint8_t *src2, int dstStride,
  41                                      int src1Stride, int h);
  42 void ff_avg_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  43                               int dstStride, int src1Stride, int h);
  44 void ff_put_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  45                                int dstStride, int src1Stride, int h);
  46 void ff_avg_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  47                                int dstStride, int src1Stride, int h);
  48 void ff_put_no_rnd_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  49                                       int dstStride, int src1Stride, int h);
  50 void ff_put_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  51                                           int dstStride, int srcStride, int h);
  52 void ff_avg_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  53                                           int dstStride, int srcStride, int h);
  54 void ff_put_no_rnd_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  55                                                  int dstStride, int srcStride,
  56                                                  int h);
  57 void ff_put_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  58                                          int dstStride, int srcStride, int h);
  59 void ff_avg_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  60                                          int dstStride, int srcStride, int h);
  61 void ff_put_no_rnd_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  62                                                 int dstStride, int srcStride,
  63                                                 int h);
  64 void ff_put_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  65                                           int dstStride, int srcStride);
  66 void ff_avg_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  67                                           int dstStride, int srcStride);
  68 void ff_put_no_rnd_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  69                                                  int dstStride, int srcStride);
  70 void ff_put_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  71                                          int dstStride, int srcStride);
  72 void ff_avg_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  73                                          int dstStride, int srcStride);
  74 void ff_put_no_rnd_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  75                                                 int dstStride, int srcStride);
  76 #define ff_put_no_rnd_pixels16_mmxext ff_put_pixels16_mmx
  77 #define ff_put_no_rnd_pixels8_mmxext ff_put_pixels8_mmx
  78
  79 int32_t ff_scalarproduct_int16_mmxext(const int16_t *v1, const int16_t *v2,
  80                                       int order);
  81 int32_t ff_scalarproduct_int16_sse2(const int16_t *v1, const int16_t *v2,
  82                                     int order);
  83 int32_t ff_scalarproduct_and_madd_int16_mmxext(int16_t *v1, const int16_t *v2,
  84                                                const int16_t *v3,
  85                                                int order, int mul);
  86 int32_t ff_scalarproduct_and_madd_int16_sse2(int16_t *v1, const int16_t *v2,
  87                                              const int16_t *v3,
  88                                              int order, int mul);
  89 int32_t ff_scalarproduct_and_madd_int16_ssse3(int16_t *v1, const int16_t *v2,
  90                                               const int16_t *v3,
  91                                               int order, int mul);
  92
  93 void ff_bswap32_buf_ssse3(uint32_t *dst, const uint32_t *src, int w);
  94 void ff_bswap32_buf_sse2(uint32_t *dst, const uint32_t *src, int w);
  95
  96 void ff_add_hfyu_median_prediction_mmxext(uint8_t *dst, const uint8_t *top,
  97                                           const uint8_t *diff, int w,
  98                                           int *left, int *left_top);
  99 int ff_add_hfyu_left_prediction_ssse3(uint8_t *dst, const uint8_t *src,
 100                                       int w, int left);
 101 int ff_add_hfyu_left_prediction_sse4(uint8_t *dst, const uint8_t *src,
 102                                      int w, int left);
 103
 104 void ff_vector_clip_int32_mmx(int32_t *dst, const int32_t *src,
 105                               int32_t min, int32_t max, unsigned int len);
 106 void ff_vector_clip_int32_sse2(int32_t *dst, const int32_t *src,
 107                                int32_t min, int32_t max, unsigned int len);
 108 void ff_vector_clip_int32_int_sse2(int32_t *dst, const int32_t *src,
 109                                    int32_t min, int32_t max, unsigned int len);
 110 void ff_vector_clip_int32_sse4(int32_t *dst, const int32_t *src,
 111                                int32_t min, int32_t max, unsigned int len);
 112
 113 #if HAVE_YASM
 114
 115 #define ff_put_pixels16_mmxext ff_put_pixels16_mmx
 116 #define ff_put_pixels8_mmxext  ff_put_pixels8_mmx
 117
 118 void ff_avg_pixels16_mmxext(uint8_t *block, const uint8_t *pixels,
 119                             ptrdiff_t line_size, int h);
 120
 121 #define QPEL_OP(OPNAME, RND, MMX)                                       \
 122 static void OPNAME ## qpel8_mc00_ ## MMX(uint8_t *dst, uint8_t *src,    \
 123                                          ptrdiff_t stride)              \
 124 {                                                                       \
 125     ff_ ## OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);              \
 126 }                                                                       \
 127                                                                         \
 128 static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src,    \
 129                                          ptrdiff_t stride)              \
 130 {                                                                       \
 131     uint64_t temp[8];                                                   \
 132     uint8_t *const half = (uint8_t *) temp;                             \
 133     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,        \
 134                                                    stride, 8);          \
 135     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half,                 \
 136                                         stride, stride, 8);             \
 137 }                                                                       \
 138                                                                         \
 139 static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src,    \
 140                                          ptrdiff_t stride)              \
 141 {                                                                       \
 142     ff_ ## OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride,    \
 143                                                    stride, 8);          \
 144 }                                                                       \
 145                                                                         \
 146 static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src,    \
 147                                          ptrdiff_t stride)              \
 148 {                                                                       \
 149     uint64_t temp[8];                                                   \
 150     uint8_t *const half = (uint8_t *) temp;                             \
 151     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,        \
 152                                                    stride, 8);          \
 153     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + 1, half, stride,     \
 154                                         stride, 8);                     \
 155 }                                                                       \
 156                                                                         \
 157 static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, uint8_t *src,    \
 158                                          ptrdiff_t stride)              \
 159 {                                                                       \
 160     uint64_t temp[8];                                                   \
 161     uint8_t *const half = (uint8_t *) temp;                             \
 162     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src,           \
 163                                                    8, stride);          \
 164     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half,                 \
 165                                         stride, stride, 8);             \
 166 }                                                                       \
 167                                                                         \
 168 static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src,    \
 169                                          ptrdiff_t stride)              \
 170 {                                                                       \
 171     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src,            \
 172                                                    stride, stride);     \
 173 }                                                                       \
 174                                                                         \
 175 static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src,    \
 176                                          ptrdiff_t stride)              \
 177 {                                                                       \
 178     uint64_t temp[8];                                                   \
 179     uint8_t *const half = (uint8_t *) temp;                             \
 180     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src,           \
 181                                                    8, stride);          \
 182     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + stride, half, stride,\
 183                                         stride, 8);                     \
 184 }                                                                       \
 185                                                                         \
 186 static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src,    \
 187                                          ptrdiff_t stride)              \
 188 {                                                                       \
 189     uint64_t half[8 + 9];                                               \
 190     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 191     uint8_t *const halfHV = (uint8_t *) half;                           \
 192     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 193                                                    stride, 9);          \
 194     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8,           \
 195                                         stride, 9);                     \
 196     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 197     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
 198                                         stride, 8, 8);                  \
 199 }                                                                       \
 200                                                                         \
 201 static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, uint8_t *src,    \
 202                                          ptrdiff_t stride)              \
 203 {                                                                       \
 204     uint64_t half[8 + 9];                                               \
 205     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 206     uint8_t *const halfHV = (uint8_t *) half;                           \
 207     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 208                                                    stride, 9);          \
 209     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
 210                                         stride, 9);                     \
 211     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 212     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
 213                                         stride, 8, 8);                  \
 214 }                                                                       \
 215                                                                         \
 216 static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst, uint8_t *src,    \
 217                                          ptrdiff_t stride)              \
 218 {                                                                       \
 219     uint64_t half[8 + 9];                                               \
 220     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 221     uint8_t *const halfHV = (uint8_t *) half;                           \
 222     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 223                                                    stride, 9);          \
 224     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8,           \
 225                                         stride, 9);                     \
 226     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 227     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
 228                                         stride, 8, 8);                  \
 229 }                                                                       \
 230                                                                         \
 231 static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src,    \
 232                                          ptrdiff_t stride)              \
 233 {                                                                       \
 234     uint64_t half[8 + 9];                                               \
 235     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 236     uint8_t *const halfHV = (uint8_t *) half;                           \
 237     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 238                                                    stride, 9);          \
 239     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
 240                                         stride, 9);                     \
 241     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 242     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
 243                                         stride, 8, 8);                  \
 244 }                                                                       \
 245                                                                         \
 246 static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src,    \
 247                                          ptrdiff_t stride)              \
 248 {                                                                       \
 249     uint64_t half[8 + 9];                                               \
 250     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 251     uint8_t *const halfHV = (uint8_t *) half;                           \
 252     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 253                                                    stride, 9);          \
 254     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 255     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
 256                                         stride, 8, 8);                  \
 257 }                                                                       \
 258                                                                         \
 259 static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src,    \
 260                                          ptrdiff_t stride)              \
 261 {                                                                       \
 262     uint64_t half[8 + 9];                                               \
 263     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 264     uint8_t *const halfHV = (uint8_t *) half;                           \
 265     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 266                                                    stride, 9);          \
 267     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 268     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
 269                                         stride, 8, 8);                  \
 270 }                                                                       \
 271                                                                         \
 272 static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src,    \
 273                                          ptrdiff_t stride)              \
 274 {                                                                       \
 275     uint64_t half[8 + 9];                                               \
 276     uint8_t *const halfH = (uint8_t *) half;                            \
 277     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 278                                                    stride, 9);          \
 279     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH,              \
 280                                         8, stride, 9);                  \
 281     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
 282                                                    stride, 8);          \
 283 }                                                                       \
 284                                                                         \
 285 static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src,    \
 286                                          ptrdiff_t stride)              \
 287 {                                                                       \
 288     uint64_t half[8 + 9];                                               \
 289     uint8_t *const halfH = (uint8_t *) half;                            \
 290     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 291                                                    stride, 9);          \
 292     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
 293                                         stride, 9);                     \
 294     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
 295                                                    stride, 8);          \
 296 }                                                                       \
 297                                                                         \
 298 static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src,    \
 299                                          ptrdiff_t stride)              \
 300 {                                                                       \
 301     uint64_t half[9];                                                   \
 302     uint8_t *const halfH = (uint8_t *) half;                            \
 303     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 304                                                    stride, 9);          \
 305     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
 306                                                    stride, 8);          \
 307 }                                                                       \
 308                                                                         \
 309 static void OPNAME ## qpel16_mc00_ ## MMX(uint8_t *dst, uint8_t *src,   \
 310                                           ptrdiff_t stride)             \
 311 {                                                                       \
 312     ff_ ## OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);            \
 313 }                                                                       \
 314                                                                         \
 315 static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src,   \
 316                                           ptrdiff_t stride)             \
 317 {                                                                       \
 318     uint64_t temp[32];                                                  \
 319     uint8_t *const half = (uint8_t *) temp;                             \
 320     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16,      \
 321                                                     stride, 16);        \
 322     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride,        \
 323                                          stride, 16);                   \
 324 }                                                                       \
 325                                                                         \
 326 static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src,   \
 327                                           ptrdiff_t stride)             \
 328 {                                                                       \
 329     ff_ ## OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src,           \
 330                                                     stride, stride, 16);\
 331 }                                                                       \
 332                                                                         \
 333 static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src,   \
 334                                           ptrdiff_t stride)             \
 335 {                                                                       \
 336     uint64_t temp[32];                                                  \
 337     uint8_t *const half = (uint8_t*) temp;                              \
 338     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16,      \
 339                                                     stride, 16);        \
 340     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src + 1, half,            \
 341                                          stride, stride, 16);           \
 342 }                                                                       \
 343                                                                         \
 344 static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src,   \
 345                                           ptrdiff_t stride)             \
 346 {                                                                       \
 347     uint64_t temp[32];                                                  \
 348     uint8_t *const half = (uint8_t *) temp;                             \
 349     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16,      \
 350                                                     stride);            \
 351     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride,        \
 352                                          stride, 16);                   \
 353 }                                                                       \
 354                                                                         \
 355 static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src,   \
 356                                           ptrdiff_t stride)             \
 357 {                                                                       \
 358     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src,           \
 359                                                     stride, stride);    \
 360 }                                                                       \
 361                                                                         \
 362 static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src,   \
 363                                           ptrdiff_t stride)             \
 364 {                                                                       \
 365     uint64_t temp[32];                                                  \
 366     uint8_t *const half = (uint8_t *) temp;                             \
 367     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16,      \
 368                                                     stride);            \
 369     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half,         \
 370                                          stride, stride, 16);           \
 371 }                                                                       \
 372                                                                         \
 373 static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src,   \
 374                                           ptrdiff_t stride)             \
 375 {                                                                       \
 376     uint64_t half[16 * 2 + 17 * 2];                                     \
 377     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 378     uint8_t *const halfHV = (uint8_t *) half;                           \
 379     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 380                                                     stride, 17);        \
 381     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
 382                                          stride, 17);                   \
 383     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 384                                                     16, 16);            \
 385     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
 386                                          stride, 16, 16);               \
 387 }                                                                       \
 388                                                                         \
 389 static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src,   \
 390                                           ptrdiff_t stride)             \
 391 {                                                                       \
 392     uint64_t half[16 * 2 + 17 * 2];                                     \
 393     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 394     uint8_t *const halfHV = (uint8_t *) half;                           \
 395     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 396                                                     stride, 17);        \
 397     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
 398                                          stride, 17);                   \
 399     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 400                                                     16, 16);            \
 401     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
 402                                          stride, 16, 16);               \
 403 }                                                                       \
 404                                                                         \
 405 static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src,   \
 406                                           ptrdiff_t stride)             \
 407 {                                                                       \
 408     uint64_t half[16 * 2 + 17 * 2];                                     \
 409     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 410     uint8_t *const halfHV = (uint8_t *) half;                           \
 411     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 412                                                     stride, 17);        \
 413     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
 414                                          stride, 17);                   \
 415     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 416                                                     16, 16);            \
 417     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
 418                                          stride, 16, 16);               \
 419 }                                                                       \
 420                                                                         \
 421 static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, uint8_t *src,   \
 422                                           ptrdiff_t stride)             \
 423 {                                                                       \
 424     uint64_t half[16 * 2 + 17 * 2];                                     \
 425     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 426     uint8_t *const halfHV = (uint8_t *) half;                           \
 427     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 428                                                     stride, 17);        \
 429     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
 430                                          stride, 17);                   \
 431     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 432                                                     16, 16);            \
 433     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
 434                                          stride, 16, 16);               \
 435 }                                                                       \
 436                                                                         \
 437 static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src,   \
 438                                           ptrdiff_t stride)             \
 439 {                                                                       \
 440     uint64_t half[16 * 2 + 17 * 2];                                     \
 441     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 442     uint8_t *const halfHV = (uint8_t *) half;                           \
 443     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 444                                                     stride, 17);        \
 445     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 446                                                     16, 16);            \
 447     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
 448                                          stride, 16, 16);               \
 449 }                                                                       \
 450                                                                         \
 451 static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src,   \
 452                                           ptrdiff_t stride)             \
 453 {                                                                       \
 454     uint64_t half[16 * 2 + 17 * 2];                                     \
 455     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 456     uint8_t *const halfHV = (uint8_t *) half;                           \
 457     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 458                                                     stride, 17);        \
 459     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 460                                                     16, 16);            \
 461     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
 462                                          stride, 16, 16);               \
 463 }                                                                       \
 464                                                                         \
 465 static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src,   \
 466                                           ptrdiff_t stride)             \
 467 {                                                                       \
 468     uint64_t half[17 * 2];                                              \
 469     uint8_t *const halfH = (uint8_t *) half;                            \
 470     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 471                                                     stride, 17);        \
 472     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
 473                                          stride, 17);                   \
 474     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
 475                                                     stride, 16);        \
 476 }                                                                       \
 477                                                                         \
 478 static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src,   \
 479                                           ptrdiff_t stride)             \
 480 {                                                                       \
 481     uint64_t half[17 * 2];                                              \
 482     uint8_t *const halfH = (uint8_t *) half;                            \
 483     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 484                                                     stride, 17);        \
 485     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
 486                                          stride, 17);                   \
 487     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
 488                                                     stride, 16);        \
 489 }                                                                       \
 490                                                                         \
 491 static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src,   \
 492                                           ptrdiff_t stride)             \
 493 {                                                                       \
 494     uint64_t half[17 * 2];                                              \
 495     uint8_t *const halfH = (uint8_t *) half;                            \
 496     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 497                                                     stride, 17);        \
 498     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
 499                                                     stride, 16);        \
 500 }
 501
 502 QPEL_OP(put_,        _,        mmxext)
 503 QPEL_OP(avg_,        _,        mmxext)
 504 QPEL_OP(put_no_rnd_, _no_rnd_, mmxext)
 505
 506 #endif /* HAVE_YASM */
 507
 508 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX)                          \
 509 do {                                                                         \
 510     c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
 511     c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
 512     c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
 513     c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
 514     c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
 515     c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
 516     c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
 517     c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
 518     c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
 519     c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
 520     c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
 521     c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
 522     c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
 523     c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
 524     c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
 525     c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
 526 } while (0)
 527
 528 static av_cold void dsputil_init_mmx(DSPContext *c, AVCodecContext *avctx,
 529                                      int cpu_flags, unsigned high_bit_depth)
 530 {
 531 #if HAVE_MMX_INLINE
 532     c->put_pixels_clamped        = ff_put_pixels_clamped_mmx;
 533     c->add_pixels_clamped        = ff_add_pixels_clamped_mmx;
 534
 535     if (!high_bit_depth) {
 536         c->draw_edges   = ff_draw_edges_mmx;
 537     }
 538
 539 #if CONFIG_VIDEODSP && (ARCH_X86_32 || !HAVE_YASM)
 540     c->gmc = ff_gmc_mmx;
 541 #endif
 542
 543     c->add_bytes = ff_add_bytes_mmx;
 544 #endif /* HAVE_MMX_INLINE */
 545
 546 #if HAVE_MMX_EXTERNAL
 547     if (!high_bit_depth) {
 548         c->clear_block  = ff_clear_block_mmx;
 549         c->clear_blocks = ff_clear_blocks_mmx;
 550     }
 551     c->vector_clip_int32 = ff_vector_clip_int32_mmx;
 552     c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_mmx;
 553 #endif /* HAVE_MMX_EXTERNAL */
 554 }
 555
 556 static av_cold void dsputil_init_mmxext(DSPContext *c, AVCodecContext *avctx,
 557                                         int cpu_flags, unsigned high_bit_depth)
 558 {
 559 #if HAVE_MMXEXT_INLINE
 560     if (!high_bit_depth && avctx->idct_algo == FF_IDCT_XVIDMMX && avctx->lowres == 0) {
 561         c->idct_put = ff_idct_xvid_mmxext_put;
 562         c->idct_add = ff_idct_xvid_mmxext_add;
 563         c->idct     = ff_idct_xvid_mmxext;
 564     }
 565 #endif /* HAVE_MMXEXT_INLINE */
 566
 567 #if HAVE_MMXEXT_EXTERNAL
 568     SET_QPEL_FUNCS(avg_qpel,        0, 16, mmxext, );
 569     SET_QPEL_FUNCS(avg_qpel,        1,  8, mmxext, );
 570
 571     SET_QPEL_FUNCS(put_qpel,        0, 16, mmxext, );
 572     SET_QPEL_FUNCS(put_qpel,        1,  8, mmxext, );
 573     SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmxext, );
 574     SET_QPEL_FUNCS(put_no_rnd_qpel, 1,  8, mmxext, );
 575
 576     /* slower than cmov version on AMD */
 577     if (!(cpu_flags & AV_CPU_FLAG_3DNOW))
 578         c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_mmxext;
 579
 580     c->scalarproduct_int16          = ff_scalarproduct_int16_mmxext;
 581     c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_mmxext;
 582 #endif /* HAVE_MMXEXT_EXTERNAL */
 583 }
 584
 585 static av_cold void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx,
 586                                      int cpu_flags, unsigned high_bit_depth)
 587 {
 588 #if HAVE_SSE_INLINE
 589     c->vector_clipf = ff_vector_clipf_sse;
 590 #endif /* HAVE_SSE_INLINE */
 591
 592 #if HAVE_YASM
 593 #if HAVE_SSE_EXTERNAL
 594     /* XvMCCreateBlocks() may not allocate 16-byte aligned blocks */
 595     if (CONFIG_XVMC && avctx->hwaccel && avctx->hwaccel->decode_mb)
 596         return;
 597
 598     if (!high_bit_depth) {
 599         c->clear_block  = ff_clear_block_sse;
 600         c->clear_blocks = ff_clear_blocks_sse;
 601     }
 602 #endif
 603 #if HAVE_INLINE_ASM && CONFIG_VIDEODSP
 604     c->gmc = ff_gmc_sse;
 605 #endif
 606 #endif /* HAVE_YASM */
 607 }
 608
 609 static av_cold void dsputil_init_sse2(DSPContext *c, AVCodecContext *avctx,
 610                                       int cpu_flags, unsigned high_bit_depth)
 611 {
 612 #if HAVE_SSE2_INLINE
 613     if (!high_bit_depth && avctx->idct_algo == FF_IDCT_XVIDMMX && avctx->lowres == 0) {
 614         c->idct_put              = ff_idct_xvid_sse2_put;
 615         c->idct_add              = ff_idct_xvid_sse2_add;
 616         c->idct                  = ff_idct_xvid_sse2;
 617         c->idct_permutation_type = FF_SSE2_IDCT_PERM;
 618     }
 619 #endif /* HAVE_SSE2_INLINE */
 620
 621 #if HAVE_SSE2_EXTERNAL
 622     c->scalarproduct_int16          = ff_scalarproduct_int16_sse2;
 623     c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_sse2;
 624     if (cpu_flags & AV_CPU_FLAG_ATOM) {
 625         c->vector_clip_int32 = ff_vector_clip_int32_int_sse2;
 626     } else {
 627         c->vector_clip_int32 = ff_vector_clip_int32_sse2;
 628     }
 629     c->bswap_buf = ff_bswap32_buf_sse2;
 630     c->put_signed_pixels_clamped = ff_put_signed_pixels_clamped_sse2;
 631 #endif /* HAVE_SSE2_EXTERNAL */
 632 }
 633
 634 static av_cold void dsputil_init_ssse3(DSPContext *c, AVCodecContext *avctx,
 635                                        int cpu_flags, unsigned high_bit_depth)
 636 {
 637 #if HAVE_SSSE3_EXTERNAL
 638     c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_ssse3;
 639     if (cpu_flags & AV_CPU_FLAG_SSE4) // not really SSE4, just slow on Conroe
 640         c->add_hfyu_left_prediction = ff_add_hfyu_left_prediction_sse4;
 641
 642     if (!(cpu_flags & (AV_CPU_FLAG_SSE42 | AV_CPU_FLAG_3DNOW))) // cachesplit
 643         c->scalarproduct_and_madd_int16 = ff_scalarproduct_and_madd_int16_ssse3;
 644     c->bswap_buf = ff_bswap32_buf_ssse3;
 645 #endif /* HAVE_SSSE3_EXTERNAL */
 646 }
 647
 648 static av_cold void dsputil_init_sse4(DSPContext *c, AVCodecContext *avctx,
 649                                       int cpu_flags, unsigned high_bit_depth)
 650 {
 651 #if HAVE_SSE4_EXTERNAL
 652     c->vector_clip_int32 = ff_vector_clip_int32_sse4;
 653 #endif /* HAVE_SSE4_EXTERNAL */
 654 }
 655
 656 av_cold void ff_dsputil_init_x86(DSPContext *c, AVCodecContext *avctx,
 657                                  unsigned high_bit_depth)
 658 {
 659     int cpu_flags = av_get_cpu_flags();
 660
 661 #if HAVE_7REGS && HAVE_INLINE_ASM
 662     if (HAVE_MMX && cpu_flags & AV_CPU_FLAG_CMOV)
 663         c->add_hfyu_median_prediction = ff_add_hfyu_median_prediction_cmov;
 664 #endif
 665
 666     if (X86_MMX(cpu_flags)) {
 667 #if HAVE_INLINE_ASM
 668         const int idct_algo = avctx->idct_algo;
 669
 670         if (avctx->lowres == 0 && !high_bit_depth) {
 671             if (idct_algo == FF_IDCT_AUTO || idct_algo == FF_IDCT_SIMPLEMMX) {
 672                 c->idct_put              = ff_simple_idct_put_mmx;
 673                 c->idct_add              = ff_simple_idct_add_mmx;
 674                 c->idct                  = ff_simple_idct_mmx;
 675                 c->idct_permutation_type = FF_SIMPLE_IDCT_PERM;
 676             } else if (idct_algo == FF_IDCT_XVIDMMX) {
 677                 c->idct_put              = ff_idct_xvid_mmx_put;
 678                 c->idct_add              = ff_idct_xvid_mmx_add;
 679                 c->idct                  = ff_idct_xvid_mmx;
 680             }
 681         }
 682 #endif /* HAVE_INLINE_ASM */
 683
 684         dsputil_init_mmx(c, avctx, cpu_flags, high_bit_depth);
 685     }
 686
 687     if (X86_MMXEXT(cpu_flags))
 688         dsputil_init_mmxext(c, avctx, cpu_flags, high_bit_depth);
 689
 690     if (X86_SSE(cpu_flags))
 691         dsputil_init_sse(c, avctx, cpu_flags, high_bit_depth);
 692
 693     if (X86_SSE2(cpu_flags))
 694         dsputil_init_sse2(c, avctx, cpu_flags, high_bit_depth);
 695
 696     if (EXTERNAL_SSSE3(cpu_flags))
 697         dsputil_init_ssse3(c, avctx, cpu_flags, high_bit_depth);
 698
 699     if (EXTERNAL_SSE4(cpu_flags))
 700         dsputil_init_sse4(c, avctx, cpu_flags, high_bit_depth);
 701
 702     if (CONFIG_ENCODERS)
 703         ff_dsputilenc_init_mmx(c, avctx, high_bit_depth);
 704 }