libavcodec/x86/vp8dsp_init.c

   1 /*
   2  * VP8 DSP functions x86-optimized
   3  * Copyright (c) 2010 Ronald S. Bultje <rsbultje@gmail.com>
   4  * Copyright (c) 2010 Fiona Glaser <fiona@x264.com>
   5  *
   6  * This file is part of Libav.
   7  *
   8  * Libav 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  * Libav 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 Libav; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23 #include "libavutil/attributes.h"
  24 #include "libavutil/cpu.h"
  25 #include "libavutil/mem.h"
  26 #include "libavutil/x86/cpu.h"
  27 #include "libavcodec/vp8dsp.h"
  28
  29 #if HAVE_X86ASM
  30
  31 /*
  32  * MC functions
  33  */
  34 void ff_put_vp8_epel4_h4_mmxext(uint8_t *dst, ptrdiff_t dststride,
  35                                 uint8_t *src, ptrdiff_t srcstride,
  36                                 int height, int mx, int my);
  37 void ff_put_vp8_epel4_h6_mmxext(uint8_t *dst, ptrdiff_t dststride,
  38                                 uint8_t *src, ptrdiff_t srcstride,
  39                                 int height, int mx, int my);
  40 void ff_put_vp8_epel4_v4_mmxext(uint8_t *dst, ptrdiff_t dststride,
  41                                 uint8_t *src, ptrdiff_t srcstride,
  42                                 int height, int mx, int my);
  43 void ff_put_vp8_epel4_v6_mmxext(uint8_t *dst, ptrdiff_t dststride,
  44                                 uint8_t *src, ptrdiff_t srcstride,
  45                                 int height, int mx, int my);
  46
  47 void ff_put_vp8_epel8_h4_sse2  (uint8_t *dst, ptrdiff_t dststride,
  48                                 uint8_t *src, ptrdiff_t srcstride,
  49                                 int height, int mx, int my);
  50 void ff_put_vp8_epel8_h6_sse2  (uint8_t *dst, ptrdiff_t dststride,
  51                                 uint8_t *src, ptrdiff_t srcstride,
  52                                 int height, int mx, int my);
  53 void ff_put_vp8_epel8_v4_sse2  (uint8_t *dst, ptrdiff_t dststride,
  54                                 uint8_t *src, ptrdiff_t srcstride,
  55                                 int height, int mx, int my);
  56 void ff_put_vp8_epel8_v6_sse2  (uint8_t *dst, ptrdiff_t dststride,
  57                                 uint8_t *src, ptrdiff_t srcstride,
  58                                 int height, int mx, int my);
  59
  60 void ff_put_vp8_epel4_h4_ssse3 (uint8_t *dst, ptrdiff_t dststride,
  61                                 uint8_t *src, ptrdiff_t srcstride,
  62                                 int height, int mx, int my);
  63 void ff_put_vp8_epel4_h6_ssse3 (uint8_t *dst, ptrdiff_t dststride,
  64                                 uint8_t *src, ptrdiff_t srcstride,
  65                                 int height, int mx, int my);
  66 void ff_put_vp8_epel4_v4_ssse3 (uint8_t *dst, ptrdiff_t dststride,
  67                                 uint8_t *src, ptrdiff_t srcstride,
  68                                 int height, int mx, int my);
  69 void ff_put_vp8_epel4_v6_ssse3 (uint8_t *dst, ptrdiff_t dststride,
  70                                 uint8_t *src, ptrdiff_t srcstride,
  71                                 int height, int mx, int my);
  72 void ff_put_vp8_epel8_h4_ssse3 (uint8_t *dst, ptrdiff_t dststride,
  73                                 uint8_t *src, ptrdiff_t srcstride,
  74                                 int height, int mx, int my);
  75 void ff_put_vp8_epel8_h6_ssse3 (uint8_t *dst, ptrdiff_t dststride,
  76                                 uint8_t *src, ptrdiff_t srcstride,
  77                                 int height, int mx, int my);
  78 void ff_put_vp8_epel8_v4_ssse3 (uint8_t *dst, ptrdiff_t dststride,
  79                                 uint8_t *src, ptrdiff_t srcstride,
  80                                 int height, int mx, int my);
  81 void ff_put_vp8_epel8_v6_ssse3 (uint8_t *dst, ptrdiff_t dststride,
  82                                 uint8_t *src, ptrdiff_t srcstride,
  83                                 int height, int mx, int my);
  84
  85 void ff_put_vp8_bilinear4_h_mmxext(uint8_t *dst, ptrdiff_t dststride,
  86                                    uint8_t *src, ptrdiff_t srcstride,
  87                                    int height, int mx, int my);
  88 void ff_put_vp8_bilinear8_h_sse2  (uint8_t *dst, ptrdiff_t dststride,
  89                                    uint8_t *src, ptrdiff_t srcstride,
  90                                    int height, int mx, int my);
  91 void ff_put_vp8_bilinear4_h_ssse3 (uint8_t *dst, ptrdiff_t dststride,
  92                                    uint8_t *src, ptrdiff_t srcstride,
  93                                    int height, int mx, int my);
  94 void ff_put_vp8_bilinear8_h_ssse3 (uint8_t *dst, ptrdiff_t dststride,
  95                                    uint8_t *src, ptrdiff_t srcstride,
  96                                    int height, int mx, int my);
  97
  98 void ff_put_vp8_bilinear4_v_mmxext(uint8_t *dst, ptrdiff_t dststride,
  99                                    uint8_t *src, ptrdiff_t srcstride,
 100                                    int height, int mx, int my);
 101 void ff_put_vp8_bilinear8_v_sse2  (uint8_t *dst, ptrdiff_t dststride,
 102                                    uint8_t *src, ptrdiff_t srcstride,
 103                                    int height, int mx, int my);
 104 void ff_put_vp8_bilinear4_v_ssse3 (uint8_t *dst, ptrdiff_t dststride,
 105                                    uint8_t *src, ptrdiff_t srcstride,
 106                                    int height, int mx, int my);
 107 void ff_put_vp8_bilinear8_v_ssse3 (uint8_t *dst, ptrdiff_t dststride,
 108                                    uint8_t *src, ptrdiff_t srcstride,
 109                                    int height, int mx, int my);
 110
 111
 112 void ff_put_vp8_pixels8_mmx (uint8_t *dst, ptrdiff_t dststride,
 113                              uint8_t *src, ptrdiff_t srcstride,
 114                              int height, int mx, int my);
 115 void ff_put_vp8_pixels16_mmx(uint8_t *dst, ptrdiff_t dststride,
 116                              uint8_t *src, ptrdiff_t srcstride,
 117                              int height, int mx, int my);
 118 void ff_put_vp8_pixels16_sse(uint8_t *dst, ptrdiff_t dststride,
 119                              uint8_t *src, ptrdiff_t srcstride,
 120                              int height, int mx, int my);
 121
 122 #define TAP_W16(OPT, FILTERTYPE, TAPTYPE) \
 123 static void ff_put_vp8_ ## FILTERTYPE ## 16_ ## TAPTYPE ## _ ## OPT( \
 124     uint8_t *dst,  ptrdiff_t dststride, uint8_t *src, \
 125     ptrdiff_t srcstride, int height, int mx, int my) \
 126 { \
 127     ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \
 128         dst,     dststride, src,     srcstride, height, mx, my); \
 129     ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \
 130         dst + 8, dststride, src + 8, srcstride, height, mx, my); \
 131 }
 132 #define TAP_W8(OPT, FILTERTYPE, TAPTYPE) \
 133 static void ff_put_vp8_ ## FILTERTYPE ## 8_ ## TAPTYPE ## _ ## OPT( \
 134     uint8_t *dst,  ptrdiff_t dststride, uint8_t *src, \
 135     ptrdiff_t srcstride, int height, int mx, int my) \
 136 { \
 137     ff_put_vp8_ ## FILTERTYPE ## 4_ ## TAPTYPE ## _ ## OPT( \
 138         dst,     dststride, src,     srcstride, height, mx, my); \
 139     ff_put_vp8_ ## FILTERTYPE ## 4_ ## TAPTYPE ## _ ## OPT( \
 140         dst + 4, dststride, src + 4, srcstride, height, mx, my); \
 141 }
 142
 143 #if ARCH_X86_32
 144 TAP_W8 (mmxext, epel, h4)
 145 TAP_W8 (mmxext, epel, h6)
 146 TAP_W16(mmxext, epel, h6)
 147 TAP_W8 (mmxext, epel, v4)
 148 TAP_W8 (mmxext, epel, v6)
 149 TAP_W16(mmxext, epel, v6)
 150 TAP_W8 (mmxext, bilinear, h)
 151 TAP_W16(mmxext, bilinear, h)
 152 TAP_W8 (mmxext, bilinear, v)
 153 TAP_W16(mmxext, bilinear, v)
 154 #endif
 155
 156 TAP_W16(sse2,  epel, h6)
 157 TAP_W16(sse2,  epel, v6)
 158 TAP_W16(sse2,  bilinear, h)
 159 TAP_W16(sse2,  bilinear, v)
 160
 161 TAP_W16(ssse3, epel, h6)
 162 TAP_W16(ssse3, epel, v6)
 163 TAP_W16(ssse3, bilinear, h)
 164 TAP_W16(ssse3, bilinear, v)
 165
 166 #define HVTAP(OPT, ALIGN, TAPNUMX, TAPNUMY, SIZE, MAXHEIGHT) \
 167 static void ff_put_vp8_epel ## SIZE ## _h ## TAPNUMX ## v ## TAPNUMY ## _ ## OPT( \
 168     uint8_t *dst, ptrdiff_t dststride, uint8_t *src, \
 169     ptrdiff_t srcstride, int height, int mx, int my) \
 170 { \
 171     DECLARE_ALIGNED(ALIGN, uint8_t, tmp)[SIZE * (MAXHEIGHT + TAPNUMY - 1)]; \
 172     uint8_t *tmpptr = tmp + SIZE * (TAPNUMY / 2 - 1); \
 173     src -= srcstride * (TAPNUMY / 2 - 1); \
 174     ff_put_vp8_epel ## SIZE ## _h ## TAPNUMX ## _ ## OPT( \
 175         tmp, SIZE,      src,    srcstride, height + TAPNUMY - 1, mx, my); \
 176     ff_put_vp8_epel ## SIZE ## _v ## TAPNUMY ## _ ## OPT( \
 177         dst, dststride, tmpptr, SIZE,      height,               mx, my); \
 178 }
 179
 180 #if ARCH_X86_32
 181 #define HVTAPMMX(x, y) \
 182 HVTAP(mmxext, 8, x, y,  4,  8) \
 183 HVTAP(mmxext, 8, x, y,  8, 16)
 184
 185 HVTAP(mmxext, 8, 6, 6, 16, 16)
 186 #else
 187 #define HVTAPMMX(x, y) \
 188 HVTAP(mmxext, 8, x, y,  4,  8)
 189 #endif
 190
 191 HVTAPMMX(4, 4)
 192 HVTAPMMX(4, 6)
 193 HVTAPMMX(6, 4)
 194 HVTAPMMX(6, 6)
 195
 196 #define HVTAPSSE2(x, y, w) \
 197 HVTAP(sse2,  16, x, y, w, 16) \
 198 HVTAP(ssse3, 16, x, y, w, 16)
 199
 200 HVTAPSSE2(4, 4, 8)
 201 HVTAPSSE2(4, 6, 8)
 202 HVTAPSSE2(6, 4, 8)
 203 HVTAPSSE2(6, 6, 8)
 204 HVTAPSSE2(6, 6, 16)
 205
 206 HVTAP(ssse3, 16, 4, 4, 4, 8)
 207 HVTAP(ssse3, 16, 4, 6, 4, 8)
 208 HVTAP(ssse3, 16, 6, 4, 4, 8)
 209 HVTAP(ssse3, 16, 6, 6, 4, 8)
 210
 211 #define HVBILIN(OPT, ALIGN, SIZE, MAXHEIGHT) \
 212 static void ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT( \
 213     uint8_t *dst, ptrdiff_t dststride, uint8_t *src, \
 214     ptrdiff_t srcstride, int height, int mx, int my) \
 215 { \
 216     DECLARE_ALIGNED(ALIGN, uint8_t, tmp)[SIZE * (MAXHEIGHT + 2)]; \
 217     ff_put_vp8_bilinear ## SIZE ## _h_ ## OPT( \
 218         tmp, SIZE,      src, srcstride, height + 1, mx, my); \
 219     ff_put_vp8_bilinear ## SIZE ## _v_ ## OPT( \
 220         dst, dststride, tmp, SIZE,      height,     mx, my); \
 221 }
 222
 223 HVBILIN(mmxext,  8,  4,  8)
 224 #if ARCH_X86_32
 225 HVBILIN(mmxext,  8,  8, 16)
 226 HVBILIN(mmxext,  8, 16, 16)
 227 #endif
 228 HVBILIN(sse2,  8,  8, 16)
 229 HVBILIN(sse2,  8, 16, 16)
 230 HVBILIN(ssse3, 8,  4,  8)
 231 HVBILIN(ssse3, 8,  8, 16)
 232 HVBILIN(ssse3, 8, 16, 16)
 233
 234 void ff_vp8_idct_dc_add_mmx(uint8_t *dst, int16_t block[16],
 235                             ptrdiff_t stride);
 236 void ff_vp8_idct_dc_add_sse4(uint8_t *dst, int16_t block[16],
 237                              ptrdiff_t stride);
 238 void ff_vp8_idct_dc_add4y_mmx(uint8_t *dst, int16_t block[4][16],
 239                                ptrdiff_t stride);
 240 void ff_vp8_idct_dc_add4y_sse2(uint8_t *dst, int16_t block[4][16],
 241                                ptrdiff_t stride);
 242 void ff_vp8_idct_dc_add4uv_mmx(uint8_t *dst, int16_t block[2][16],
 243                                ptrdiff_t stride);
 244 void ff_vp8_luma_dc_wht_mmx(int16_t block[4][4][16], int16_t dc[16]);
 245 void ff_vp8_luma_dc_wht_sse(int16_t block[4][4][16], int16_t dc[16]);
 246 void ff_vp8_idct_add_mmx(uint8_t *dst, int16_t block[16], ptrdiff_t stride);
 247 void ff_vp8_idct_add_sse(uint8_t *dst, int16_t block[16], ptrdiff_t stride);
 248
 249 #define DECLARE_LOOP_FILTER(NAME)                                       \
 250 void ff_vp8_v_loop_filter_simple_ ## NAME(uint8_t *dst,                 \
 251                                           ptrdiff_t stride,             \
 252                                           int flim);                    \
 253 void ff_vp8_h_loop_filter_simple_ ## NAME(uint8_t *dst,                 \
 254                                           ptrdiff_t stride,             \
 255                                           int flim);                    \
 256 void ff_vp8_v_loop_filter16y_inner_ ## NAME (uint8_t *dst,              \
 257                                              ptrdiff_t stride,          \
 258                                              int e, int i, int hvt);    \
 259 void ff_vp8_h_loop_filter16y_inner_ ## NAME (uint8_t *dst,              \
 260                                              ptrdiff_t stride,          \
 261                                              int e, int i, int hvt);    \
 262 void ff_vp8_v_loop_filter8uv_inner_ ## NAME (uint8_t *dstU,             \
 263                                              uint8_t *dstV,             \
 264                                              ptrdiff_t s,               \
 265                                              int e, int i, int hvt);    \
 266 void ff_vp8_h_loop_filter8uv_inner_ ## NAME (uint8_t *dstU,             \
 267                                              uint8_t *dstV,             \
 268                                              ptrdiff_t s,               \
 269                                              int e, int i, int hvt);    \
 270 void ff_vp8_v_loop_filter16y_mbedge_ ## NAME(uint8_t *dst,              \
 271                                              ptrdiff_t stride,          \
 272                                              int e, int i, int hvt);    \
 273 void ff_vp8_h_loop_filter16y_mbedge_ ## NAME(uint8_t *dst,              \
 274                                              ptrdiff_t stride,          \
 275                                              int e, int i, int hvt);    \
 276 void ff_vp8_v_loop_filter8uv_mbedge_ ## NAME(uint8_t *dstU,             \
 277                                              uint8_t *dstV,             \
 278                                              ptrdiff_t s,               \
 279                                              int e, int i, int hvt);    \
 280 void ff_vp8_h_loop_filter8uv_mbedge_ ## NAME(uint8_t *dstU,             \
 281                                              uint8_t *dstV,             \
 282                                              ptrdiff_t s,               \
 283                                              int e, int i, int hvt);
 284
 285 DECLARE_LOOP_FILTER(mmx)
 286 DECLARE_LOOP_FILTER(mmxext)
 287 DECLARE_LOOP_FILTER(sse2)
 288 DECLARE_LOOP_FILTER(ssse3)
 289 DECLARE_LOOP_FILTER(sse4)
 290
 291 #endif /* HAVE_X86ASM */
 292
 293 #define VP8_LUMA_MC_FUNC(IDX, SIZE, OPT) \
 294     c->put_vp8_epel_pixels_tab[IDX][0][2] = ff_put_vp8_epel ## SIZE ## _h6_ ## OPT; \
 295     c->put_vp8_epel_pixels_tab[IDX][2][0] = ff_put_vp8_epel ## SIZE ## _v6_ ## OPT; \
 296     c->put_vp8_epel_pixels_tab[IDX][2][2] = ff_put_vp8_epel ## SIZE ## _h6v6_ ## OPT
 297
 298 #define VP8_MC_FUNC(IDX, SIZE, OPT) \
 299     c->put_vp8_epel_pixels_tab[IDX][0][1] = ff_put_vp8_epel ## SIZE ## _h4_ ## OPT; \
 300     c->put_vp8_epel_pixels_tab[IDX][1][0] = ff_put_vp8_epel ## SIZE ## _v4_ ## OPT; \
 301     c->put_vp8_epel_pixels_tab[IDX][1][1] = ff_put_vp8_epel ## SIZE ## _h4v4_ ## OPT; \
 302     c->put_vp8_epel_pixels_tab[IDX][1][2] = ff_put_vp8_epel ## SIZE ## _h6v4_ ## OPT; \
 303     c->put_vp8_epel_pixels_tab[IDX][2][1] = ff_put_vp8_epel ## SIZE ## _h4v6_ ## OPT; \
 304     VP8_LUMA_MC_FUNC(IDX, SIZE, OPT)
 305
 306 #define VP8_BILINEAR_MC_FUNC(IDX, SIZE, OPT) \
 307     c->put_vp8_bilinear_pixels_tab[IDX][0][1] = ff_put_vp8_bilinear ## SIZE ## _h_ ## OPT; \
 308     c->put_vp8_bilinear_pixels_tab[IDX][0][2] = ff_put_vp8_bilinear ## SIZE ## _h_ ## OPT; \
 309     c->put_vp8_bilinear_pixels_tab[IDX][1][0] = ff_put_vp8_bilinear ## SIZE ## _v_ ## OPT; \
 310     c->put_vp8_bilinear_pixels_tab[IDX][1][1] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT; \
 311     c->put_vp8_bilinear_pixels_tab[IDX][1][2] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT; \
 312     c->put_vp8_bilinear_pixels_tab[IDX][2][0] = ff_put_vp8_bilinear ## SIZE ## _v_ ## OPT; \
 313     c->put_vp8_bilinear_pixels_tab[IDX][2][1] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT; \
 314     c->put_vp8_bilinear_pixels_tab[IDX][2][2] = ff_put_vp8_bilinear ## SIZE ## _hv_ ## OPT
 315
 316
 317 av_cold void ff_vp78dsp_init_x86(VP8DSPContext *c)
 318 {
 319 #if HAVE_X86ASM
 320     int cpu_flags = av_get_cpu_flags();
 321
 322     if (EXTERNAL_MMX(cpu_flags)) {
 323 #if ARCH_X86_32
 324         c->put_vp8_epel_pixels_tab[0][0][0]     =
 325         c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_mmx;
 326 #endif
 327         c->put_vp8_epel_pixels_tab[1][0][0]     =
 328         c->put_vp8_bilinear_pixels_tab[1][0][0] = ff_put_vp8_pixels8_mmx;
 329     }
 330
 331     /* note that 4-tap width=16 functions are missing because w=16
 332      * is only used for luma, and luma is always a copy or sixtap. */
 333     if (EXTERNAL_MMXEXT(cpu_flags)) {
 334         VP8_MC_FUNC(2, 4, mmxext);
 335         VP8_BILINEAR_MC_FUNC(2, 4, mmxext);
 336 #if ARCH_X86_32
 337         VP8_LUMA_MC_FUNC(0, 16, mmxext);
 338         VP8_MC_FUNC(1, 8, mmxext);
 339         VP8_BILINEAR_MC_FUNC(0, 16, mmxext);
 340         VP8_BILINEAR_MC_FUNC(1,  8, mmxext);
 341 #endif
 342     }
 343
 344     if (EXTERNAL_SSE(cpu_flags)) {
 345         c->put_vp8_epel_pixels_tab[0][0][0]     =
 346         c->put_vp8_bilinear_pixels_tab[0][0][0] = ff_put_vp8_pixels16_sse;
 347     }
 348
 349     if (EXTERNAL_SSE2_SLOW(cpu_flags)) {
 350         VP8_LUMA_MC_FUNC(0, 16, sse2);
 351         VP8_MC_FUNC(1, 8, sse2);
 352         VP8_BILINEAR_MC_FUNC(0, 16, sse2);
 353         VP8_BILINEAR_MC_FUNC(1, 8, sse2);
 354     }
 355
 356     if (EXTERNAL_SSSE3(cpu_flags)) {
 357         VP8_LUMA_MC_FUNC(0, 16, ssse3);
 358         VP8_MC_FUNC(1, 8, ssse3);
 359         VP8_MC_FUNC(2, 4, ssse3);
 360         VP8_BILINEAR_MC_FUNC(0, 16, ssse3);
 361         VP8_BILINEAR_MC_FUNC(1, 8, ssse3);
 362         VP8_BILINEAR_MC_FUNC(2, 4, ssse3);
 363     }
 364 #endif /* HAVE_X86ASM */
 365 }
 366
 367 av_cold void ff_vp8dsp_init_x86(VP8DSPContext *c)
 368 {
 369 #if HAVE_X86ASM
 370     int cpu_flags = av_get_cpu_flags();
 371
 372     if (EXTERNAL_MMX(cpu_flags)) {
 373         c->vp8_idct_dc_add    = ff_vp8_idct_dc_add_mmx;
 374         c->vp8_idct_dc_add4uv = ff_vp8_idct_dc_add4uv_mmx;
 375 #if ARCH_X86_32
 376         c->vp8_idct_dc_add4y  = ff_vp8_idct_dc_add4y_mmx;
 377         c->vp8_idct_add       = ff_vp8_idct_add_mmx;
 378         c->vp8_luma_dc_wht    = ff_vp8_luma_dc_wht_mmx;
 379
 380         c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_mmx;
 381         c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_mmx;
 382
 383         c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_mmx;
 384         c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_mmx;
 385         c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_mmx;
 386         c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_mmx;
 387
 388         c->vp8_v_loop_filter16y       = ff_vp8_v_loop_filter16y_mbedge_mmx;
 389         c->vp8_h_loop_filter16y       = ff_vp8_h_loop_filter16y_mbedge_mmx;
 390         c->vp8_v_loop_filter8uv       = ff_vp8_v_loop_filter8uv_mbedge_mmx;
 391         c->vp8_h_loop_filter8uv       = ff_vp8_h_loop_filter8uv_mbedge_mmx;
 392 #endif
 393     }
 394
 395     /* note that 4-tap width=16 functions are missing because w=16
 396      * is only used for luma, and luma is always a copy or sixtap. */
 397     if (EXTERNAL_MMXEXT(cpu_flags)) {
 398 #if ARCH_X86_32
 399         c->vp8_v_loop_filter_simple   = ff_vp8_v_loop_filter_simple_mmxext;
 400         c->vp8_h_loop_filter_simple   = ff_vp8_h_loop_filter_simple_mmxext;
 401
 402         c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_mmxext;
 403         c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_mmxext;
 404         c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_mmxext;
 405         c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_mmxext;
 406
 407         c->vp8_v_loop_filter16y       = ff_vp8_v_loop_filter16y_mbedge_mmxext;
 408         c->vp8_h_loop_filter16y       = ff_vp8_h_loop_filter16y_mbedge_mmxext;
 409         c->vp8_v_loop_filter8uv       = ff_vp8_v_loop_filter8uv_mbedge_mmxext;
 410         c->vp8_h_loop_filter8uv       = ff_vp8_h_loop_filter8uv_mbedge_mmxext;
 411 #endif
 412     }
 413
 414     if (EXTERNAL_SSE(cpu_flags)) {
 415         c->vp8_idct_add                         = ff_vp8_idct_add_sse;
 416         c->vp8_luma_dc_wht                      = ff_vp8_luma_dc_wht_sse;
 417     }
 418
 419     if (EXTERNAL_SSE2_SLOW(cpu_flags)) {
 420         c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_sse2;
 421
 422         c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_sse2;
 423         c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_sse2;
 424
 425         c->vp8_v_loop_filter16y       = ff_vp8_v_loop_filter16y_mbedge_sse2;
 426         c->vp8_v_loop_filter8uv       = ff_vp8_v_loop_filter8uv_mbedge_sse2;
 427     }
 428
 429     if (EXTERNAL_SSE2(cpu_flags)) {
 430         c->vp8_idct_dc_add4y          = ff_vp8_idct_dc_add4y_sse2;
 431
 432         c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_sse2;
 433
 434         c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_sse2;
 435         c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_sse2;
 436
 437         c->vp8_h_loop_filter16y       = ff_vp8_h_loop_filter16y_mbedge_sse2;
 438         c->vp8_h_loop_filter8uv       = ff_vp8_h_loop_filter8uv_mbedge_sse2;
 439     }
 440
 441     if (EXTERNAL_SSSE3(cpu_flags)) {
 442         c->vp8_v_loop_filter_simple = ff_vp8_v_loop_filter_simple_ssse3;
 443         c->vp8_h_loop_filter_simple = ff_vp8_h_loop_filter_simple_ssse3;
 444
 445         c->vp8_v_loop_filter16y_inner = ff_vp8_v_loop_filter16y_inner_ssse3;
 446         c->vp8_h_loop_filter16y_inner = ff_vp8_h_loop_filter16y_inner_ssse3;
 447         c->vp8_v_loop_filter8uv_inner = ff_vp8_v_loop_filter8uv_inner_ssse3;
 448         c->vp8_h_loop_filter8uv_inner = ff_vp8_h_loop_filter8uv_inner_ssse3;
 449
 450         c->vp8_v_loop_filter16y       = ff_vp8_v_loop_filter16y_mbedge_ssse3;
 451         c->vp8_h_loop_filter16y       = ff_vp8_h_loop_filter16y_mbedge_ssse3;
 452         c->vp8_v_loop_filter8uv       = ff_vp8_v_loop_filter8uv_mbedge_ssse3;
 453         c->vp8_h_loop_filter8uv       = ff_vp8_h_loop_filter8uv_mbedge_ssse3;
 454     }
 455
 456     if (EXTERNAL_SSE4(cpu_flags)) {
 457         c->vp8_idct_dc_add                  = ff_vp8_idct_dc_add_sse4;
 458
 459         c->vp8_h_loop_filter_simple   = ff_vp8_h_loop_filter_simple_sse4;
 460         c->vp8_h_loop_filter16y       = ff_vp8_h_loop_filter16y_mbedge_sse4;
 461         c->vp8_h_loop_filter8uv       = ff_vp8_h_loop_filter8uv_mbedge_sse4;
 462     }
 463 #endif /* HAVE_X86ASM */
 464 }