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

   1 /*
   2  * quarterpel DSP functions
   3  * Copyright (c) 2000, 2001 Fabrice Bellard
   4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
   5  *
   6  * This file is part of FFmpeg.
   7  *
   8  * FFmpeg is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * FFmpeg is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with FFmpeg; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23 #include <stddef.h>
  24 #include <stdint.h>
  25
  26 #include "config.h"
  27 #include "libavutil/attributes.h"
  28 #include "libavutil/cpu.h"
  29 #include "libavutil/x86/cpu.h"
  30 #include "libavcodec/pixels.h"
  31 #include "libavcodec/qpeldsp.h"
  32 #include "fpel.h"
  33
  34 void ff_put_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  35                               int dstStride, int src1Stride, int h);
  36 void ff_put_no_rnd_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1,
  37                                      uint8_t *src2, int dstStride,
  38                                      int src1Stride, int h);
  39 void ff_avg_pixels8_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  40                               int dstStride, int src1Stride, int h);
  41 void ff_put_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  42                                int dstStride, int src1Stride, int h);
  43 void ff_avg_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  44                                int dstStride, int src1Stride, int h);
  45 void ff_put_no_rnd_pixels16_l2_mmxext(uint8_t *dst, uint8_t *src1, uint8_t *src2,
  46                                       int dstStride, int src1Stride, int h);
  47 void ff_put_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  48                                           int dstStride, int srcStride, int h);
  49 void ff_avg_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  50                                           int dstStride, int srcStride, int h);
  51 void ff_put_no_rnd_mpeg4_qpel16_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  52                                                  int dstStride, int srcStride,
  53                                                  int h);
  54 void ff_put_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  55                                          int dstStride, int srcStride, int h);
  56 void ff_avg_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  57                                          int dstStride, int srcStride, int h);
  58 void ff_put_no_rnd_mpeg4_qpel8_h_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  59                                                 int dstStride, int srcStride,
  60                                                 int h);
  61 void ff_put_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  62                                           int dstStride, int srcStride);
  63 void ff_avg_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  64                                           int dstStride, int srcStride);
  65 void ff_put_no_rnd_mpeg4_qpel16_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  66                                                  int dstStride, int srcStride);
  67 void ff_put_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  68                                          int dstStride, int srcStride);
  69 void ff_avg_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  70                                          int dstStride, int srcStride);
  71 void ff_put_no_rnd_mpeg4_qpel8_v_lowpass_mmxext(uint8_t *dst, uint8_t *src,
  72                                                 int dstStride, int srcStride);
  73 #define ff_put_no_rnd_pixels16_mmxext ff_put_pixels16_mmx
  74 #define ff_put_no_rnd_pixels8_mmxext ff_put_pixels8_mmx
  75
  76 #if HAVE_YASM
  77
  78 #define ff_put_pixels16_mmxext ff_put_pixels16_mmx
  79 #define ff_put_pixels8_mmxext  ff_put_pixels8_mmx
  80
  81 #define QPEL_OP(OPNAME, RND, MMX)                                       \
  82 static void OPNAME ## qpel8_mc00_ ## MMX(uint8_t *dst, uint8_t *src,    \
  83                                          ptrdiff_t stride)              \
  84 {                                                                       \
  85     ff_ ## OPNAME ## pixels8_ ## MMX(dst, src, stride, 8);              \
  86 }                                                                       \
  87                                                                         \
  88 static void OPNAME ## qpel8_mc10_ ## MMX(uint8_t *dst, uint8_t *src,    \
  89                                          ptrdiff_t stride)              \
  90 {                                                                       \
  91     uint64_t temp[8];                                                   \
  92     uint8_t *const half = (uint8_t *) temp;                             \
  93     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,        \
  94                                                    stride, 8);          \
  95     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half,                 \
  96                                         stride, stride, 8);             \
  97 }                                                                       \
  98                                                                         \
  99 static void OPNAME ## qpel8_mc20_ ## MMX(uint8_t *dst, uint8_t *src,    \
 100                                          ptrdiff_t stride)              \
 101 {                                                                       \
 102     ff_ ## OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride,    \
 103                                                    stride, 8);          \
 104 }                                                                       \
 105                                                                         \
 106 static void OPNAME ## qpel8_mc30_ ## MMX(uint8_t *dst, uint8_t *src,    \
 107                                          ptrdiff_t stride)              \
 108 {                                                                       \
 109     uint64_t temp[8];                                                   \
 110     uint8_t *const half = (uint8_t *) temp;                             \
 111     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8,        \
 112                                                    stride, 8);          \
 113     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + 1, half, stride,     \
 114                                         stride, 8);                     \
 115 }                                                                       \
 116                                                                         \
 117 static void OPNAME ## qpel8_mc01_ ## MMX(uint8_t *dst, 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_v_lowpass_ ## MMX(half, src,           \
 123                                                    8, stride);          \
 124     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src, half,                 \
 125                                         stride, stride, 8);             \
 126 }                                                                       \
 127                                                                         \
 128 static void OPNAME ## qpel8_mc02_ ## MMX(uint8_t *dst, uint8_t *src,    \
 129                                          ptrdiff_t stride)              \
 130 {                                                                       \
 131     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src,            \
 132                                                    stride, stride);     \
 133 }                                                                       \
 134                                                                         \
 135 static void OPNAME ## qpel8_mc03_ ## MMX(uint8_t *dst, uint8_t *src,    \
 136                                          ptrdiff_t stride)              \
 137 {                                                                       \
 138     uint64_t temp[8];                                                   \
 139     uint8_t *const half = (uint8_t *) temp;                             \
 140     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src,           \
 141                                                    8, stride);          \
 142     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, src + stride, half, stride,\
 143                                         stride, 8);                     \
 144 }                                                                       \
 145                                                                         \
 146 static void OPNAME ## qpel8_mc11_ ## MMX(uint8_t *dst, uint8_t *src,    \
 147                                          ptrdiff_t stride)              \
 148 {                                                                       \
 149     uint64_t half[8 + 9];                                               \
 150     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 151     uint8_t *const halfHV = (uint8_t *) half;                           \
 152     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 153                                                    stride, 9);          \
 154     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8,           \
 155                                         stride, 9);                     \
 156     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 157     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
 158                                         stride, 8, 8);                  \
 159 }                                                                       \
 160                                                                         \
 161 static void OPNAME ## qpel8_mc31_ ## MMX(uint8_t *dst, 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 + 1, 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_mc13_ ## MMX(uint8_t *dst, uint8_t *src,    \
 177                                          ptrdiff_t stride)              \
 178 {                                                                       \
 179     uint64_t half[8 + 9];                                               \
 180     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 181     uint8_t *const halfHV = (uint8_t *) half;                           \
 182     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 183                                                    stride, 9);          \
 184     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src, halfH, 8,           \
 185                                         stride, 9);                     \
 186     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 187     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
 188                                         stride, 8, 8);                  \
 189 }                                                                       \
 190                                                                         \
 191 static void OPNAME ## qpel8_mc33_ ## MMX(uint8_t *dst, uint8_t *src,    \
 192                                          ptrdiff_t stride)              \
 193 {                                                                       \
 194     uint64_t half[8 + 9];                                               \
 195     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 196     uint8_t *const halfHV = (uint8_t *) half;                           \
 197     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 198                                                    stride, 9);          \
 199     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
 200                                         stride, 9);                     \
 201     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 202     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
 203                                         stride, 8, 8);                  \
 204 }                                                                       \
 205                                                                         \
 206 static void OPNAME ## qpel8_mc21_ ## MMX(uint8_t *dst, uint8_t *src,    \
 207                                          ptrdiff_t stride)              \
 208 {                                                                       \
 209     uint64_t half[8 + 9];                                               \
 210     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 211     uint8_t *const halfHV = (uint8_t *) half;                           \
 212     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 213                                                    stride, 9);          \
 214     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 215     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH, halfHV,             \
 216                                         stride, 8, 8);                  \
 217 }                                                                       \
 218                                                                         \
 219 static void OPNAME ## qpel8_mc23_ ## MMX(uint8_t *dst, uint8_t *src,    \
 220                                          ptrdiff_t stride)              \
 221 {                                                                       \
 222     uint64_t half[8 + 9];                                               \
 223     uint8_t *const halfH  = (uint8_t *) half + 64;                      \
 224     uint8_t *const halfHV = (uint8_t *) half;                           \
 225     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 226                                                    stride, 9);          \
 227     ff_put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
 228     ff_ ## OPNAME ## pixels8_l2_ ## MMX(dst, halfH + 8, halfHV,         \
 229                                         stride, 8, 8);                  \
 230 }                                                                       \
 231                                                                         \
 232 static void OPNAME ## qpel8_mc12_ ## MMX(uint8_t *dst, uint8_t *src,    \
 233                                          ptrdiff_t stride)              \
 234 {                                                                       \
 235     uint64_t half[8 + 9];                                               \
 236     uint8_t *const halfH = (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, halfH,              \
 240                                         8, stride, 9);                  \
 241     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
 242                                                    stride, 8);          \
 243 }                                                                       \
 244                                                                         \
 245 static void OPNAME ## qpel8_mc32_ ## MMX(uint8_t *dst, uint8_t *src,    \
 246                                          ptrdiff_t stride)              \
 247 {                                                                       \
 248     uint64_t half[8 + 9];                                               \
 249     uint8_t *const halfH = (uint8_t *) half;                            \
 250     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 251                                                    stride, 9);          \
 252     ff_put ## RND ## pixels8_l2_ ## MMX(halfH, src + 1, halfH, 8,       \
 253                                         stride, 9);                     \
 254     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
 255                                                    stride, 8);          \
 256 }                                                                       \
 257                                                                         \
 258 static void OPNAME ## qpel8_mc22_ ## MMX(uint8_t *dst, uint8_t *src,    \
 259                                          ptrdiff_t stride)              \
 260 {                                                                       \
 261     uint64_t half[9];                                                   \
 262     uint8_t *const halfH = (uint8_t *) half;                            \
 263     ff_put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8,       \
 264                                                    stride, 9);          \
 265     ff_ ## OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH,          \
 266                                                    stride, 8);          \
 267 }                                                                       \
 268                                                                         \
 269 static void OPNAME ## qpel16_mc00_ ## MMX(uint8_t *dst, uint8_t *src,   \
 270                                           ptrdiff_t stride)             \
 271 {                                                                       \
 272     ff_ ## OPNAME ## pixels16_ ## MMX(dst, src, stride, 16);            \
 273 }                                                                       \
 274                                                                         \
 275 static void OPNAME ## qpel16_mc10_ ## MMX(uint8_t *dst, uint8_t *src,   \
 276                                           ptrdiff_t stride)             \
 277 {                                                                       \
 278     uint64_t temp[32];                                                  \
 279     uint8_t *const half = (uint8_t *) temp;                             \
 280     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16,      \
 281                                                     stride, 16);        \
 282     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride,        \
 283                                          stride, 16);                   \
 284 }                                                                       \
 285                                                                         \
 286 static void OPNAME ## qpel16_mc20_ ## MMX(uint8_t *dst, uint8_t *src,   \
 287                                           ptrdiff_t stride)             \
 288 {                                                                       \
 289     ff_ ## OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src,           \
 290                                                     stride, stride, 16);\
 291 }                                                                       \
 292                                                                         \
 293 static void OPNAME ## qpel16_mc30_ ## MMX(uint8_t *dst, uint8_t *src,   \
 294                                           ptrdiff_t stride)             \
 295 {                                                                       \
 296     uint64_t temp[32];                                                  \
 297     uint8_t *const half = (uint8_t*) temp;                              \
 298     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16,      \
 299                                                     stride, 16);        \
 300     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src + 1, half,            \
 301                                          stride, stride, 16);           \
 302 }                                                                       \
 303                                                                         \
 304 static void OPNAME ## qpel16_mc01_ ## MMX(uint8_t *dst, uint8_t *src,   \
 305                                           ptrdiff_t stride)             \
 306 {                                                                       \
 307     uint64_t temp[32];                                                  \
 308     uint8_t *const half = (uint8_t *) temp;                             \
 309     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16,      \
 310                                                     stride);            \
 311     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src, half, stride,        \
 312                                          stride, 16);                   \
 313 }                                                                       \
 314                                                                         \
 315 static void OPNAME ## qpel16_mc02_ ## MMX(uint8_t *dst, uint8_t *src,   \
 316                                           ptrdiff_t stride)             \
 317 {                                                                       \
 318     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src,           \
 319                                                     stride, stride);    \
 320 }                                                                       \
 321                                                                         \
 322 static void OPNAME ## qpel16_mc03_ ## MMX(uint8_t *dst, uint8_t *src,   \
 323                                           ptrdiff_t stride)             \
 324 {                                                                       \
 325     uint64_t temp[32];                                                  \
 326     uint8_t *const half = (uint8_t *) temp;                             \
 327     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16,      \
 328                                                     stride);            \
 329     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, src+stride, half,         \
 330                                          stride, stride, 16);           \
 331 }                                                                       \
 332                                                                         \
 333 static void OPNAME ## qpel16_mc11_ ## MMX(uint8_t *dst, uint8_t *src,   \
 334                                           ptrdiff_t stride)             \
 335 {                                                                       \
 336     uint64_t half[16 * 2 + 17 * 2];                                     \
 337     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 338     uint8_t *const halfHV = (uint8_t *) half;                           \
 339     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 340                                                     stride, 17);        \
 341     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
 342                                          stride, 17);                   \
 343     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 344                                                     16, 16);            \
 345     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
 346                                          stride, 16, 16);               \
 347 }                                                                       \
 348                                                                         \
 349 static void OPNAME ## qpel16_mc31_ ## MMX(uint8_t *dst, uint8_t *src,   \
 350                                           ptrdiff_t stride)             \
 351 {                                                                       \
 352     uint64_t half[16 * 2 + 17 * 2];                                     \
 353     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 354     uint8_t *const halfHV = (uint8_t *) half;                           \
 355     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 356                                                     stride, 17);        \
 357     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
 358                                          stride, 17);                   \
 359     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 360                                                     16, 16);            \
 361     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
 362                                          stride, 16, 16);               \
 363 }                                                                       \
 364                                                                         \
 365 static void OPNAME ## qpel16_mc13_ ## MMX(uint8_t *dst, uint8_t *src,   \
 366                                           ptrdiff_t stride)             \
 367 {                                                                       \
 368     uint64_t half[16 * 2 + 17 * 2];                                     \
 369     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 370     uint8_t *const halfHV = (uint8_t *) half;                           \
 371     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 372                                                     stride, 17);        \
 373     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
 374                                          stride, 17);                   \
 375     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 376                                                     16, 16);            \
 377     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
 378                                          stride, 16, 16);               \
 379 }                                                                       \
 380                                                                         \
 381 static void OPNAME ## qpel16_mc33_ ## MMX(uint8_t *dst, 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 + 16, halfHV,       \
 394                                          stride, 16, 16);               \
 395 }                                                                       \
 396                                                                         \
 397 static void OPNAME ## qpel16_mc21_ ## MMX(uint8_t *dst, uint8_t *src,   \
 398                                           ptrdiff_t stride)             \
 399 {                                                                       \
 400     uint64_t half[16 * 2 + 17 * 2];                                     \
 401     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 402     uint8_t *const halfHV = (uint8_t *) half;                           \
 403     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 404                                                     stride, 17);        \
 405     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 406                                                     16, 16);            \
 407     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH, halfHV,            \
 408                                          stride, 16, 16);               \
 409 }                                                                       \
 410                                                                         \
 411 static void OPNAME ## qpel16_mc23_ ## MMX(uint8_t *dst, uint8_t *src,   \
 412                                           ptrdiff_t stride)             \
 413 {                                                                       \
 414     uint64_t half[16 * 2 + 17 * 2];                                     \
 415     uint8_t *const halfH  = (uint8_t *) half + 256;                     \
 416     uint8_t *const halfHV = (uint8_t *) half;                           \
 417     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 418                                                     stride, 17);        \
 419     ff_put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH,      \
 420                                                     16, 16);            \
 421     ff_ ## OPNAME ## pixels16_l2_ ## MMX(dst, halfH + 16, halfHV,       \
 422                                          stride, 16, 16);               \
 423 }                                                                       \
 424                                                                         \
 425 static void OPNAME ## qpel16_mc12_ ## MMX(uint8_t *dst, uint8_t *src,   \
 426                                           ptrdiff_t stride)             \
 427 {                                                                       \
 428     uint64_t half[17 * 2];                                              \
 429     uint8_t *const halfH = (uint8_t *) half;                            \
 430     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 431                                                     stride, 17);        \
 432     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src, halfH, 16,         \
 433                                          stride, 17);                   \
 434     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
 435                                                     stride, 16);        \
 436 }                                                                       \
 437                                                                         \
 438 static void OPNAME ## qpel16_mc32_ ## MMX(uint8_t *dst, uint8_t *src,   \
 439                                           ptrdiff_t stride)             \
 440 {                                                                       \
 441     uint64_t half[17 * 2];                                              \
 442     uint8_t *const halfH = (uint8_t *) half;                            \
 443     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 444                                                     stride, 17);        \
 445     ff_put ## RND ## pixels16_l2_ ## MMX(halfH, src + 1, halfH, 16,     \
 446                                          stride, 17);                   \
 447     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
 448                                                     stride, 16);        \
 449 }                                                                       \
 450                                                                         \
 451 static void OPNAME ## qpel16_mc22_ ## MMX(uint8_t *dst, uint8_t *src,   \
 452                                           ptrdiff_t stride)             \
 453 {                                                                       \
 454     uint64_t half[17 * 2];                                              \
 455     uint8_t *const halfH = (uint8_t *) half;                            \
 456     ff_put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16,     \
 457                                                     stride, 17);        \
 458     ff_ ## OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH,         \
 459                                                     stride, 16);        \
 460 }
 461
 462 QPEL_OP(put_,        _,        mmxext)
 463 QPEL_OP(avg_,        _,        mmxext)
 464 QPEL_OP(put_no_rnd_, _no_rnd_, mmxext)
 465
 466 #endif /* HAVE_YASM */
 467
 468 #define SET_QPEL_FUNCS(PFX, IDX, SIZE, CPU, PREFIX)                          \
 469 do {                                                                         \
 470     c->PFX ## _pixels_tab[IDX][ 0] = PREFIX ## PFX ## SIZE ## _mc00_ ## CPU; \
 471     c->PFX ## _pixels_tab[IDX][ 1] = PREFIX ## PFX ## SIZE ## _mc10_ ## CPU; \
 472     c->PFX ## _pixels_tab[IDX][ 2] = PREFIX ## PFX ## SIZE ## _mc20_ ## CPU; \
 473     c->PFX ## _pixels_tab[IDX][ 3] = PREFIX ## PFX ## SIZE ## _mc30_ ## CPU; \
 474     c->PFX ## _pixels_tab[IDX][ 4] = PREFIX ## PFX ## SIZE ## _mc01_ ## CPU; \
 475     c->PFX ## _pixels_tab[IDX][ 5] = PREFIX ## PFX ## SIZE ## _mc11_ ## CPU; \
 476     c->PFX ## _pixels_tab[IDX][ 6] = PREFIX ## PFX ## SIZE ## _mc21_ ## CPU; \
 477     c->PFX ## _pixels_tab[IDX][ 7] = PREFIX ## PFX ## SIZE ## _mc31_ ## CPU; \
 478     c->PFX ## _pixels_tab[IDX][ 8] = PREFIX ## PFX ## SIZE ## _mc02_ ## CPU; \
 479     c->PFX ## _pixels_tab[IDX][ 9] = PREFIX ## PFX ## SIZE ## _mc12_ ## CPU; \
 480     c->PFX ## _pixels_tab[IDX][10] = PREFIX ## PFX ## SIZE ## _mc22_ ## CPU; \
 481     c->PFX ## _pixels_tab[IDX][11] = PREFIX ## PFX ## SIZE ## _mc32_ ## CPU; \
 482     c->PFX ## _pixels_tab[IDX][12] = PREFIX ## PFX ## SIZE ## _mc03_ ## CPU; \
 483     c->PFX ## _pixels_tab[IDX][13] = PREFIX ## PFX ## SIZE ## _mc13_ ## CPU; \
 484     c->PFX ## _pixels_tab[IDX][14] = PREFIX ## PFX ## SIZE ## _mc23_ ## CPU; \
 485     c->PFX ## _pixels_tab[IDX][15] = PREFIX ## PFX ## SIZE ## _mc33_ ## CPU; \
 486 } while (0)
 487
 488 av_cold void ff_qpeldsp_init_x86(QpelDSPContext *c)
 489 {
 490     int cpu_flags = av_get_cpu_flags();
 491
 492     if (X86_MMXEXT(cpu_flags)) {
 493 #if HAVE_MMXEXT_EXTERNAL
 494         SET_QPEL_FUNCS(avg_qpel,        0, 16, mmxext, );
 495         SET_QPEL_FUNCS(avg_qpel,        1,  8, mmxext, );
 496
 497         SET_QPEL_FUNCS(put_qpel,        0, 16, mmxext, );
 498         SET_QPEL_FUNCS(put_qpel,        1,  8, mmxext, );
 499         SET_QPEL_FUNCS(put_no_rnd_qpel, 0, 16, mmxext, );
 500         SET_QPEL_FUNCS(put_no_rnd_qpel, 1,  8, mmxext, );
 501 #endif /* HAVE_MMXEXT_EXTERNAL */
 502     }
 503 }