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

   1 /*
   2  * quarterpel DSP functions
   3  *
   4  * This file is part of Libav.
   5  *
   6  * Libav 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  * Libav 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 Libav; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 #include <stddef.h>
  22 #include <stdint.h>
  23
  24 #include "config.h"
  25 #include "libavutil/attributes.h"
  26 #include "libavutil/cpu.h"
  27 #include "libavutil/x86/cpu.h"
  28 #include "libavcodec/pixels.h"
  29 #include "libavcodec/qpeldsp.h"
  30 #include "fpel.h"
  31
  32 void ff_put_pixels8_l2_mmxext(uint8_t *dst,
  33                               const uint8_t *src1, const uint8_t *src2,
  34                               int dstStride, int src1Stride, int h);
  35 void ff_put_no_rnd_pixels8_l2_mmxext(uint8_t *dst,
  36                                      const uint8_t *src1, const uint8_t *src2,
  37                                      int dstStride, int src1Stride, int h);
  38 void ff_avg_pixels8_l2_mmxext(uint8_t *dst,
  39                               const uint8_t *src1, const uint8_t *src2,
  40                               int dstStride, int src1Stride, int h);
  41 void ff_put_pixels16_l2_mmxext(uint8_t *dst,
  42                                const uint8_t *src1, const uint8_t *src2,
  43                                int dstStride, int src1Stride, int h);
  44 void ff_avg_pixels16_l2_mmxext(uint8_t *dst,
  45                                const uint8_t *src1, const uint8_t *src2,
  46                                int dstStride, int src1Stride, int h);
  47 void ff_put_no_rnd_pixels16_l2_mmxext(uint8_t *dst,
  48                                       const uint8_t *src1, const uint8_t *src2,
  49                                       int dstStride, int src1Stride, int h);
  50 void ff_put_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, const uint8_t *src,
  51                                           int dstStride, int srcStride, int h);
  52 void ff_avg_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, const uint8_t *src,
  53                                           int dstStride, int srcStride, int h);
  54 void ff_put_no_rnd_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst,
  55                                                  const uint8_t *src,
  56                                                  int dstStride, int srcStride,
  57                                                  int h);
  58 void ff_put_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, const uint8_t *src,
  59                                          int dstStride, int srcStride, int h);
  60 void ff_avg_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, const uint8_t *src,
  61                                          int dstStride, int srcStride, int h);
  62 void ff_put_no_rnd_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst,
  63                                                 const uint8_t *src,
  64                                                 int dstStride, int srcStride,
  65                                                 int h);
  66 void ff_put_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, const uint8_t *src,
  67                                           int dstStride, int srcStride);
  68 void ff_avg_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, const uint8_t *src,
  69                                           int dstStride, int srcStride);
  70 void ff_put_no_rnd_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst,
  71                                                  const uint8_t *src,
  72                                                  int dstStride, int srcStride);
  73 void ff_put_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, const uint8_t *src,
  74                                          int dstStride, int srcStride);
  75 void ff_avg_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, const uint8_t *src,
  76                                          int dstStride, int srcStride);
  77 void ff_put_no_rnd_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst,
  78                                                 const uint8_t *src,
  79                                                 int dstStride, int srcStride);
  80 #define ff_put_no_rnd_pixels16_mmxext ff_put_pixels16_mmxext
  81 #define ff_put_no_rnd_pixels8_mmxext ff_put_pixels8_mmxext
  82
  83 #if HAVE_YASM
  84
  85 CALL_2X_PIXELS(ff_avg_pixels16_mmxext, ff_avg_pixels8_mmxext, 8)
  86 CALL_2X_PIXELS(ff_put_pixels16_mmxext, ff_put_pixels8_mmxext, 8)
  87
  88 #define QPEL_OP(OPNAME, RND, MMX)                                       \
  89 static void OPNAME ## qpel8_mc00_ ## MMX(uint8_t *dst,                  \
  90                                          const uint8_t *src,            \
  91                                          ptrdiff_t stride)              \
  92 {                                                                       \
  93     ff_ ## OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);              \
  94 }                                                                       \
  95                                                                         \
  96 static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst,                  \
  97                                          const uint8_t *src,            \
  98                                          ptrdiff_t stride)              \
  99 {                                                                       \
 100     uint64_t temp[8];                                                   \
 101     uint8_t *const half = (uint8_t *) temp;                             \
 102     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,        \
 103                                                    stride, 8);          \
 104     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half,                 \
 105                                         stride, stride, 8);             \
 106 }                                                                       \
 107                                                                         \
 108 static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst,                  \
 109                                          const uint8_t *src,            \
 110                                          ptrdiff_t stride)              \
 111 {                                                                       \
 112     ff_ ## OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride,    \
 113                                                    stride, 8);          \
 114 }                                                                       \
 115                                                                         \
 116 static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst,                  \
 117                                          const uint8_t *src,            \
 118                                          ptrdiff_t stride)              \
 119 {                                                                       \
 120     uint64_t temp[8];                                                   \
 121     uint8_t *const half = (uint8_t *) temp;                             \
 122     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,        \
 123                                                    stride, 8);          \
 124     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + 1, half, stride,     \
 125                                         stride, 8);                     \
 126 }                                                                       \
 127                                                                         \
 128 static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst,                  \
 129                                          const uint8_t *src,            \
 130                                          ptrdiff_t stride)              \
 131 {                                                                       \
 132     uint64_t temp[8];                                                   \
 133     uint8_t *const half = (uint8_t *) temp;                             \
 134     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src,           \
 135                                                    8, stride);          \
 136     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half,                 \
 137                                         stride, stride, 8);             \
 138 }                                                                       \
 139                                                                         \
 140 static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst,                  \
 141                                          const uint8_t *src,            \
 142                                          ptrdiff_t stride)              \
 143 {                                                                       \
 144     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src,            \
 145                                                    stride, stride);     \
 146 }                                                                       \
 147                                                                         \
 148 static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst,                  \
 149                                          const uint8_t *src,            \
 150                                          ptrdiff_t stride)              \
 151 {                                                                       \
 152     uint64_t temp[8];                                                   \
 153     uint8_t *const half = (uint8_t *) temp;                             \
 154     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src,           \
 155                                                    8, stride);          \
 156     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + stride, half, stride,\
 157                                         stride, 8);                     \
 158 }                                                                       \
 159                                                                         \
 160 static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst,                  \
 161                                          const uint8_t *src,            \
 162                                          ptrdiff_t stride)              \
 163 {                                                                       \
 164     uint64_t half[8 + 9];                                               \
 165     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 166     uint8_t *const halfHV = (uint8_t *) half;                           \
 167     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 168                                                    stride, 9);          \
 169     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8,           \
 170                                         stride, 9);                     \
 171     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 172     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
 173                                         stride, 8, 8);                  \
 174 }                                                                       \
 175                                                                         \
 176 static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst,                  \
 177                                          const uint8_t *src,            \
 178                                          ptrdiff_t stride)              \
 179 {                                                                       \
 180     uint64_t half[8 + 9];                                               \
 181     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 182     uint8_t *const halfHV = (uint8_t *) half;                           \
 183     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 184                                                    stride, 9);          \
 185     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
 186                                         stride, 9);                     \
 187     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 188     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
 189                                         stride, 8, 8);                  \
 190 }                                                                       \
 191                                                                         \
 192 static void OPNAME ## qpel8_mc13_ ## MMX(uint8_t *dst,                  \
 193                                          const uint8_t *src,            \
 194                                          ptrdiff_t stride)              \
 195 {                                                                       \
 196     uint64_t half[8 + 9];                                               \
 197     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 198     uint8_t *const halfHV = (uint8_t *) half;                           \
 199     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 200                                                    stride, 9);          \
 201     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8,           \
 202                                         stride, 9);                     \
 203     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 204     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
 205                                         stride, 8, 8);                  \
 206 }                                                                       \
 207                                                                         \
 208 static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst,                  \
 209                                          const uint8_t *src,            \
 210                                          ptrdiff_t stride)              \
 211 {                                                                       \
 212     uint64_t half[8 + 9];                                               \
 213     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 214     uint8_t *const halfHV = (uint8_t *) half;                           \
 215     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 216                                                    stride, 9);          \
 217     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
 218                                         stride, 9);                     \
 219     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 220     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
 221                                         stride, 8, 8);                  \
 222 }                                                                       \
 223                                                                         \
 224 static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst,                  \
 225                                          const uint8_t *src,            \
 226                                          ptrdiff_t stride)              \
 227 {                                                                       \
 228     uint64_t half[8 + 9];                                               \
 229     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 230     uint8_t *const halfHV = (uint8_t *) half;                           \
 231     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 232                                                    stride, 9);          \
 233     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 234     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
 235                                         stride, 8, 8);                  \
 236 }                                                                       \
 237                                                                         \
 238 static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst,                  \
 239                                          const uint8_t *src,            \
 240                                          ptrdiff_t stride)              \
 241 {                                                                       \
 242     uint64_t half[8 + 9];                                               \
 243     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 244     uint8_t *const halfHV = (uint8_t *) half;                           \
 245     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 246                                                    stride, 9);          \
 247     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 248     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
 249                                         stride, 8, 8);                  \
 250 }                                                                       \
 251                                                                         \
 252 static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst,                  \
 253                                          const uint8_t *src,            \
 254                                          ptrdiff_t stride)              \
 255 {                                                                       \
 256     uint64_t half[8 + 9];                                               \
 257     uint8_t *const halfH = (uint8_t *) half;                            \
 258     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 259                                                    stride, 9);          \
 260     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH,              \
 261                                         8, stride, 9);                  \
 262     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
 263                                                    stride, 8);          \
 264 }                                                                       \
 265                                                                         \
 266 static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst,                  \
 267                                          const uint8_t *src,            \
 268                                          ptrdiff_t stride)              \
 269 {                                                                       \
 270     uint64_t half[8 + 9];                                               \
 271     uint8_t *const halfH = (uint8_t *) half;                            \
 272     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 273                                                    stride, 9);          \
 274     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
 275                                         stride, 9);                     \
 276     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
 277                                                    stride, 8);          \
 278 }                                                                       \
 279                                                                         \
 280 static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst,                  \
 281                                          const uint8_t *src,            \
 282                                          ptrdiff_t stride)              \
 283 {                                                                       \
 284     uint64_t half[9];                                                   \
 285     uint8_t *const halfH = (uint8_t *) half;                            \
 286     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 287                                                    stride, 9);          \
 288     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
 289                                                    stride, 8);          \
 290 }                                                                       \
 291                                                                         \
 292 static void OPNAME ## qpel16_mc00_ ## MMX(uint8_t *dst,                 \
 293                                           const uint8_t *src,           \
 294                                           ptrdiff_t stride)             \
 295 {                                                                       \
 296     ff_ ## OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);            \
 297 }                                                                       \
 298                                                                         \
 299 static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst,                 \
 300                                           const uint8_t *src,           \
 301                                           ptrdiff_t stride)             \
 302 {                                                                       \
 303     uint64_t temp[32];                                                  \
 304     uint8_t *const half = (uint8_t *) temp;                             \
 305     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16,      \
 306                                                     stride, 16);        \
 307     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride,        \
 308                                          stride, 16);                   \
 309 }                                                                       \
 310                                                                         \
 311 static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst,                 \
 312                                           const uint8_t *src,           \
 313                                           ptrdiff_t stride)             \
 314 {                                                                       \
 315     ff_ ## OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src,           \
 316                                                     stride, stride, 16);\
 317 }                                                                       \
 318                                                                         \
 319 static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst,                 \
 320                                           const uint8_t *src,           \
 321                                           ptrdiff_t stride)             \
 322 {                                                                       \
 323     uint64_t temp[32];                                                  \
 324     uint8_t *const half = (uint8_t*) temp;                              \
 325     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16,      \
 326                                                     stride, 16);        \
 327     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src + 1, half,            \
 328                                          stride, stride, 16);           \
 329 }                                                                       \
 330                                                                         \
 331 static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst,                 \
 332                                           const uint8_t *src,           \
 333                                           ptrdiff_t stride)             \
 334 {                                                                       \
 335     uint64_t temp[32];                                                  \
 336     uint8_t *const half = (uint8_t *) temp;                             \
 337     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16,      \
 338                                                     stride);            \
 339     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride,        \
 340                                          stride, 16);                   \
 341 }                                                                       \
 342                                                                         \
 343 static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst,                 \
 344                                           const uint8_t *src,           \
 345                                           ptrdiff_t stride)             \
 346 {                                                                       \
 347     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src,           \
 348                                                     stride, stride);    \
 349 }                                                                       \
 350                                                                         \
 351 static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst,                 \
 352                                           const uint8_t *src,           \
 353                                           ptrdiff_t stride)             \
 354 {                                                                       \
 355     uint64_t temp[32];                                                  \
 356     uint8_t *const half = (uint8_t *) temp;                             \
 357     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16,      \
 358                                                     stride);            \
 359     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half,         \
 360                                          stride, stride, 16);           \
 361 }                                                                       \
 362                                                                         \
 363 static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst,                 \
 364                                           const uint8_t *src,           \
 365                                           ptrdiff_t stride)             \
 366 {                                                                       \
 367     uint64_t half[16 * 2 + 17 * 2];                                     \
 368     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 369     uint8_t *const halfHV = (uint8_t *) half;                           \
 370     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 371                                                     stride, 17);        \
 372     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
 373                                          stride, 17);                   \
 374     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 375                                                     16, 16);            \
 376     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
 377                                          stride, 16, 16);               \
 378 }                                                                       \
 379                                                                         \
 380 static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst,                 \
 381                                           const uint8_t *src,           \
 382                                           ptrdiff_t stride)             \
 383 {                                                                       \
 384     uint64_t half[16 * 2 + 17 * 2];                                     \
 385     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 386     uint8_t *const halfHV = (uint8_t *) half;                           \
 387     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 388                                                     stride, 17);        \
 389     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
 390                                          stride, 17);                   \
 391     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 392                                                     16, 16);            \
 393     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
 394                                          stride, 16, 16);               \
 395 }                                                                       \
 396                                                                         \
 397 static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst,                 \
 398                                           const uint8_t *src,           \
 399                                           ptrdiff_t stride)             \
 400 {                                                                       \
 401     uint64_t half[16 * 2 + 17 * 2];                                     \
 402     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 403     uint8_t *const halfHV = (uint8_t *) half;                           \
 404     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 405                                                     stride, 17);        \
 406     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
 407                                          stride, 17);                   \
 408     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 409                                                     16, 16);            \
 410     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
 411                                          stride, 16, 16);               \
 412 }                                                                       \
 413                                                                         \
 414 static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst,                 \
 415                                           const uint8_t *src,           \
 416                                           ptrdiff_t stride)             \
 417 {                                                                       \
 418     uint64_t half[16 * 2 + 17 * 2];                                     \
 419     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 420     uint8_t *const halfHV = (uint8_t *) half;                           \
 421     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 422                                                     stride, 17);        \
 423     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
 424                                          stride, 17);                   \
 425     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 426                                                     16, 16);            \
 427     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
 428                                          stride, 16, 16);               \
 429 }                                                                       \
 430                                                                         \
 431 static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst,                 \
 432                                           const uint8_t *src,           \
 433                                           ptrdiff_t stride)             \
 434 {                                                                       \
 435     uint64_t half[16 * 2 + 17 * 2];                                     \
 436     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 437     uint8_t *const halfHV = (uint8_t *) half;                           \
 438     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 439                                                     stride, 17);        \
 440     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 441                                                     16, 16);            \
 442     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
 443                                          stride, 16, 16);               \
 444 }                                                                       \
 445                                                                         \
 446 static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst,                 \
 447                                           const uint8_t *src,           \
 448                                           ptrdiff_t stride)             \
 449 {                                                                       \
 450     uint64_t half[16 * 2 + 17 * 2];                                     \
 451     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 452     uint8_t *const halfHV = (uint8_t *) half;                           \
 453     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 454                                                     stride, 17);        \
 455     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 456                                                     16, 16);            \
 457     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
 458                                          stride, 16, 16);               \
 459 }                                                                       \
 460                                                                         \
 461 static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst,                 \
 462                                           const uint8_t *src,           \
 463                                           ptrdiff_t stride)             \
 464 {                                                                       \
 465     uint64_t half[17 * 2];                                              \
 466     uint8_t *const halfH = (uint8_t *) half;                            \
 467     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 468                                                     stride, 17);        \
 469     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
 470                                          stride, 17);                   \
 471     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
 472                                                     stride, 16);        \
 473 }                                                                       \
 474                                                                         \
 475 static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst,                 \
 476                                           const uint8_t *src,           \
 477                                           ptrdiff_t stride)             \
 478 {                                                                       \
 479     uint64_t half[17 * 2];                                              \
 480     uint8_t *const halfH = (uint8_t *) half;                            \
 481     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 482                                                     stride, 17);        \
 483     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
 484                                          stride, 17);                   \
 485     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
 486                                                     stride, 16);        \
 487 }                                                                       \
 488                                                                         \
 489 static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst,                 \
 490                                           const uint8_t *src,           \
 491                                           ptrdiff_t stride)             \
 492 {                                                                       \
 493     uint64_t half[17 * 2];                                              \
 494     uint8_t *const halfH = (uint8_t *) half;                            \
 495     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 496                                                     stride, 17);        \
 497     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
 498                                                     stride, 16);        \
 499 }
 500
 501 QPEL_OP(put_,        _,        mmxext)
 502 QPEL_OP(avg_,        _,        mmxext)
 503 QPEL_OP(put_no_rnd_, _no_rnd_, mmxext)
 504
 505 #endif /* HAVE_YASM */
 506
 507 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX)                          \
 508 do {                                                                         \
 509     c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
 510     c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
 511     c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
 512     c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
 513     c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
 514     c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
 515     c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
 516     c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
 517     c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
 518     c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
 519     c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
 520     c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
 521     c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
 522     c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
 523     c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
 524     c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
 525 } while (0)
 526
 527 av_cold void ff_qpeldsp_init_x86(QpelDSPContext *c)
 528 {
 529     int cpu_flags = av_get_cpu_flags();
 530
 531     if (X86_MMXEXT(cpu_flags)) {
 532 #if HAVE_MMXEXT_EXTERNAL
 533         SET_QPEL_FUNCS(avg_qpel,        0, 16, mmxext, );
 534         SET_QPEL_FUNCS(avg_qpel,        1,  8, mmxext, );
 535
 536         SET_QPEL_FUNCS(put_qpel,        0, 16, mmxext, );
 537         SET_QPEL_FUNCS(put_qpel,        1,  8, mmxext, );
 538         SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmxext, );
 539         SET_QPEL_FUNCS(put_no_rnd_qpel, 1,  8, mmxext, );
 540 #endif /* HAVE_MMXEXT_EXTERNAL */
 541     }
 542 }