me_cmp_init.c 27.5 KB
Newer Older
1
/*
2
 * SIMD-optimized motion estimation
3 4
 * Copyright (c) 2000, 2001 Fabrice Bellard
 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5
 *
6 7
 * MMX optimization by Nick Kurshev <nickols_k@mail.ru>
 *
8 9 10
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
11 12 13 14
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
15
 * FFmpeg is distributed in the hope that it will be useful,
16 17 18 19 20
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
21
 * License along with FFmpeg; if not, write to the Free Software
22 23 24 25 26
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "libavutil/attributes.h"
#include "libavutil/cpu.h"
27
#include "libavutil/x86/asm.h"
28
#include "libavutil/x86/cpu.h"
29
#include "libavcodec/me_cmp.h"
30
#include "libavcodec/mpegvideo.h"
31

32 33 34 35 36
int ff_sum_abs_dctelem_mmx(int16_t *block);
int ff_sum_abs_dctelem_mmxext(int16_t *block);
int ff_sum_abs_dctelem_sse2(int16_t *block);
int ff_sum_abs_dctelem_ssse3(int16_t *block);
int ff_sse8_mmx(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
37
                ptrdiff_t stride, int h);
38
int ff_sse16_mmx(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
39
                 ptrdiff_t stride, int h);
40
int ff_sse16_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
41 42 43
                  ptrdiff_t stride, int h);
int ff_hf_noise8_mmx(uint8_t *pix1, ptrdiff_t stride, int h);
int ff_hf_noise16_mmx(uint8_t *pix1, ptrdiff_t stride, int h);
44
int ff_sad8_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
45
                   ptrdiff_t stride, int h);
46
int ff_sad16_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
47
                    ptrdiff_t stride, int h);
48
int ff_sad16_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
49
                  ptrdiff_t stride, int h);
50
int ff_sad8_x2_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
51
                      ptrdiff_t stride, int h);
52
int ff_sad16_x2_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
53
                       ptrdiff_t stride, int h);
54
int ff_sad16_x2_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
55
                     ptrdiff_t stride, int h);
56
int ff_sad8_y2_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
57
                      ptrdiff_t stride, int h);
58
int ff_sad16_y2_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
59
                       ptrdiff_t stride, int h);
60
int ff_sad16_y2_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
61
                     ptrdiff_t stride, int h);
62
int ff_sad8_approx_xy2_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
63
                              ptrdiff_t stride, int h);
64
int ff_sad16_approx_xy2_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
65
                               ptrdiff_t stride, int h);
66
int ff_sad16_approx_xy2_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
67
                             ptrdiff_t stride, int h);
68
int ff_vsad_intra8_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
69
                          ptrdiff_t stride, int h);
70
int ff_vsad_intra16_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
71
                           ptrdiff_t stride, int h);
72
int ff_vsad_intra16_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
73
                         ptrdiff_t stride, int h);
74
int ff_vsad8_approx_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
75
                    ptrdiff_t stride, int h);
76
int ff_vsad16_approx_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
77
                     ptrdiff_t stride, int h);
78
int ff_vsad16_approx_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
79
                   ptrdiff_t stride, int h);
80

81 82 83 84 85
#define hadamard_func(cpu)                                                    \
    int ff_hadamard8_diff_ ## cpu(MpegEncContext *s, uint8_t *src1,           \
                                  uint8_t *src2, ptrdiff_t stride, int h);    \
    int ff_hadamard8_diff16_ ## cpu(MpegEncContext *s, uint8_t *src1,         \
                                    uint8_t *src2, ptrdiff_t stride, int h);
86

87 88 89 90
hadamard_func(mmx)
hadamard_func(mmxext)
hadamard_func(sse2)
hadamard_func(ssse3)
91

92
#if HAVE_YASM
93
static int nsse16_mmx(MpegEncContext *c, uint8_t *pix1, uint8_t *pix2,
94
                      ptrdiff_t stride, int h)
95 96 97 98
{
    int score1, score2;

    if (c)
99
        score1 = c->mecc.sse[0](c, pix1, pix2, stride, h);
100
    else
101 102 103
        score1 = ff_sse16_mmx(c, pix1, pix2, stride, h);
    score2 = ff_hf_noise16_mmx(pix1, stride, h) + ff_hf_noise8_mmx(pix1+8, stride, h)
           - ff_hf_noise16_mmx(pix2, stride, h) - ff_hf_noise8_mmx(pix2+8, stride, h);
104 105 106 107 108 109 110 111

    if (c)
        return score1 + FFABS(score2) * c->avctx->nsse_weight;
    else
        return score1 + FFABS(score2) * 8;
}

static int nsse8_mmx(MpegEncContext *c, uint8_t *pix1, uint8_t *pix2,
112
                     ptrdiff_t stride, int h)
113
{
114 115 116
    int score1 = ff_sse8_mmx(c, pix1, pix2, stride, h);
    int score2 = ff_hf_noise8_mmx(pix1, stride, h) -
                 ff_hf_noise8_mmx(pix2, stride, h);
117 118 119 120 121 122 123

    if (c)
        return score1 + FFABS(score2) * c->avctx->nsse_weight;
    else
        return score1 + FFABS(score2) * 8;
}

124 125 126 127
#endif /* HAVE_YASM */

#if HAVE_INLINE_ASM

128
static int vsad_intra16_mmx(MpegEncContext *v, uint8_t *pix, uint8_t *dummy,
129
                            ptrdiff_t stride, int h)
130 131 132
{
    int tmp;

133
    av_assert2((((int) pix) & 7) == 0);
134
    av_assert2((stride & 7) == 0);
135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184

#define SUM(in0, in1, out0, out1)               \
    "movq (%0), %%mm2\n"                        \
    "movq 8(%0), %%mm3\n"                       \
    "add %2,%0\n"                               \
    "movq %%mm2, " #out0 "\n"                   \
    "movq %%mm3, " #out1 "\n"                   \
    "psubusb " #in0 ", %%mm2\n"                 \
    "psubusb " #in1 ", %%mm3\n"                 \
    "psubusb " #out0 ", " #in0 "\n"             \
    "psubusb " #out1 ", " #in1 "\n"             \
    "por %%mm2, " #in0 "\n"                     \
    "por %%mm3, " #in1 "\n"                     \
    "movq " #in0 ", %%mm2\n"                    \
    "movq " #in1 ", %%mm3\n"                    \
    "punpcklbw %%mm7, " #in0 "\n"               \
    "punpcklbw %%mm7, " #in1 "\n"               \
    "punpckhbw %%mm7, %%mm2\n"                  \
    "punpckhbw %%mm7, %%mm3\n"                  \
    "paddw " #in1 ", " #in0 "\n"                \
    "paddw %%mm3, %%mm2\n"                      \
    "paddw %%mm2, " #in0 "\n"                   \
    "paddw " #in0 ", %%mm6\n"


    __asm__ volatile (
        "movl    %3, %%ecx\n"
        "pxor %%mm6, %%mm6\n"
        "pxor %%mm7, %%mm7\n"
        "movq  (%0), %%mm0\n"
        "movq 8(%0), %%mm1\n"
        "add %2, %0\n"
        "jmp 2f\n"
        "1:\n"

        SUM(%%mm4, %%mm5, %%mm0, %%mm1)
        "2:\n"
        SUM(%%mm0, %%mm1, %%mm4, %%mm5)

        "subl $2, %%ecx\n"
        "jnz 1b\n"

        "movq  %%mm6, %%mm0\n"
        "psrlq $32,   %%mm6\n"
        "paddw %%mm6, %%mm0\n"
        "movq  %%mm0, %%mm6\n"
        "psrlq $16,   %%mm0\n"
        "paddw %%mm6, %%mm0\n"
        "movd  %%mm0, %1\n"
        : "+r" (pix), "=r" (tmp)
185
        : "r" (stride), "m" (h)
186 187 188 189 190 191 192
        : "%ecx");

    return tmp & 0xFFFF;
}
#undef SUM

static int vsad16_mmx(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
193
                      ptrdiff_t stride, int h)
194 195 196
{
    int tmp;

197 198
    av_assert2((((int) pix1) & 7) == 0);
    av_assert2((((int) pix2) & 7) == 0);
199
    av_assert2((stride & 7) == 0);
200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265

#define SUM(in0, in1, out0, out1)       \
    "movq (%0), %%mm2\n"                \
    "movq (%1), " #out0 "\n"            \
    "movq 8(%0), %%mm3\n"               \
    "movq 8(%1), " #out1 "\n"           \
    "add %3, %0\n"                      \
    "add %3, %1\n"                      \
    "psubb " #out0 ", %%mm2\n"          \
    "psubb " #out1 ", %%mm3\n"          \
    "pxor %%mm7, %%mm2\n"               \
    "pxor %%mm7, %%mm3\n"               \
    "movq %%mm2, " #out0 "\n"           \
    "movq %%mm3, " #out1 "\n"           \
    "psubusb " #in0 ", %%mm2\n"         \
    "psubusb " #in1 ", %%mm3\n"         \
    "psubusb " #out0 ", " #in0 "\n"     \
    "psubusb " #out1 ", " #in1 "\n"     \
    "por %%mm2, " #in0 "\n"             \
    "por %%mm3, " #in1 "\n"             \
    "movq " #in0 ", %%mm2\n"            \
    "movq " #in1 ", %%mm3\n"            \
    "punpcklbw %%mm7, " #in0 "\n"       \
    "punpcklbw %%mm7, " #in1 "\n"       \
    "punpckhbw %%mm7, %%mm2\n"          \
    "punpckhbw %%mm7, %%mm3\n"          \
    "paddw " #in1 ", " #in0 "\n"        \
    "paddw %%mm3, %%mm2\n"              \
    "paddw %%mm2, " #in0 "\n"           \
    "paddw " #in0 ", %%mm6\n"


    __asm__ volatile (
        "movl %4, %%ecx\n"
        "pxor %%mm6, %%mm6\n"
        "pcmpeqw %%mm7, %%mm7\n"
        "psllw $15, %%mm7\n"
        "packsswb %%mm7, %%mm7\n"
        "movq (%0), %%mm0\n"
        "movq (%1), %%mm2\n"
        "movq 8(%0), %%mm1\n"
        "movq 8(%1), %%mm3\n"
        "add %3, %0\n"
        "add %3, %1\n"
        "psubb %%mm2, %%mm0\n"
        "psubb %%mm3, %%mm1\n"
        "pxor %%mm7, %%mm0\n"
        "pxor %%mm7, %%mm1\n"
        "jmp 2f\n"
        "1:\n"

        SUM(%%mm4, %%mm5, %%mm0, %%mm1)
        "2:\n"
        SUM(%%mm0, %%mm1, %%mm4, %%mm5)

        "subl $2, %%ecx\n"
        "jnz 1b\n"

        "movq %%mm6, %%mm0\n"
        "psrlq $32, %%mm6\n"
        "paddw %%mm6, %%mm0\n"
        "movq %%mm0, %%mm6\n"
        "psrlq $16, %%mm0\n"
        "paddw %%mm6, %%mm0\n"
        "movd %%mm0, %2\n"
        : "+r" (pix1), "+r" (pix2), "=r" (tmp)
266
        : "r" (stride), "m" (h)
267 268 269 270 271 272
        : "%ecx");

    return tmp & 0x7FFF;
}
#undef SUM

273 274 275 276 277 278
DECLARE_ASM_CONST(8, uint64_t, round_tab)[3] = {
    0x0000000000000000ULL,
    0x0001000100010001ULL,
    0x0002000200020002ULL,
};

279 280
static inline void sad8_1_mmx(uint8_t *blk1, uint8_t *blk2,
                              ptrdiff_t stride, int h)
281
{
282
    x86_reg len = -stride * h;
283 284 285
    __asm__ volatile (
        ".p2align 4                     \n\t"
        "1:                             \n\t"
286 287 288 289
        "movq (%1, %%"FF_REG_a"), %%mm0 \n\t"
        "movq (%2, %%"FF_REG_a"), %%mm2 \n\t"
        "movq (%2, %%"FF_REG_a"), %%mm4 \n\t"
        "add %3, %%"FF_REG_a"           \n\t"
290 291
        "psubusb %%mm0, %%mm2           \n\t"
        "psubusb %%mm4, %%mm0           \n\t"
292 293 294
        "movq (%1, %%"FF_REG_a"), %%mm1 \n\t"
        "movq (%2, %%"FF_REG_a"), %%mm3 \n\t"
        "movq (%2, %%"FF_REG_a"), %%mm5 \n\t"
295 296 297 298 299 300 301 302 303 304 305 306 307 308
        "psubusb %%mm1, %%mm3           \n\t"
        "psubusb %%mm5, %%mm1           \n\t"
        "por %%mm2, %%mm0               \n\t"
        "por %%mm1, %%mm3               \n\t"
        "movq %%mm0, %%mm1              \n\t"
        "movq %%mm3, %%mm2              \n\t"
        "punpcklbw %%mm7, %%mm0         \n\t"
        "punpckhbw %%mm7, %%mm1         \n\t"
        "punpcklbw %%mm7, %%mm3         \n\t"
        "punpckhbw %%mm7, %%mm2         \n\t"
        "paddw %%mm1, %%mm0             \n\t"
        "paddw %%mm3, %%mm2             \n\t"
        "paddw %%mm2, %%mm0             \n\t"
        "paddw %%mm0, %%mm6             \n\t"
309
        "add %3, %%"FF_REG_a"           \n\t"
310 311
        " js 1b                         \n\t"
        : "+a" (len)
312
        : "r" (blk1 - len), "r" (blk2 - len), "r" (stride));
313 314 315
}

static inline void sad8_2_mmx(uint8_t *blk1a, uint8_t *blk1b, uint8_t *blk2,
316
                              ptrdiff_t stride, int h)
317
{
318
    x86_reg len = -stride * h;
319 320 321
    __asm__ volatile (
        ".p2align 4                     \n\t"
        "1:                             \n\t"
322 323 324 325
        "movq (%1, %%"FF_REG_a"), %%mm0 \n\t"
        "movq (%2, %%"FF_REG_a"), %%mm1 \n\t"
        "movq (%1, %%"FF_REG_a"), %%mm2 \n\t"
        "movq (%2, %%"FF_REG_a"), %%mm3 \n\t"
326 327 328 329 330 331
        "punpcklbw %%mm7, %%mm0         \n\t"
        "punpcklbw %%mm7, %%mm1         \n\t"
        "punpckhbw %%mm7, %%mm2         \n\t"
        "punpckhbw %%mm7, %%mm3         \n\t"
        "paddw %%mm0, %%mm1             \n\t"
        "paddw %%mm2, %%mm3             \n\t"
332 333
        "movq (%3, %%"FF_REG_a"), %%mm4 \n\t"
        "movq (%3, %%"FF_REG_a"), %%mm2 \n\t"
334 335 336 337 338 339 340 341 342 343 344 345 346
        "paddw %%mm5, %%mm1             \n\t"
        "paddw %%mm5, %%mm3             \n\t"
        "psrlw $1, %%mm1                \n\t"
        "psrlw $1, %%mm3                \n\t"
        "packuswb %%mm3, %%mm1          \n\t"
        "psubusb %%mm1, %%mm4           \n\t"
        "psubusb %%mm2, %%mm1           \n\t"
        "por %%mm4, %%mm1               \n\t"
        "movq %%mm1, %%mm0              \n\t"
        "punpcklbw %%mm7, %%mm0         \n\t"
        "punpckhbw %%mm7, %%mm1         \n\t"
        "paddw %%mm1, %%mm0             \n\t"
        "paddw %%mm0, %%mm6             \n\t"
347
        "add %4, %%"FF_REG_a"           \n\t"
348 349 350
        " js 1b                         \n\t"
        : "+a" (len)
        : "r" (blk1a - len), "r" (blk1b - len), "r" (blk2 - len),
351
          "r" (stride));
352 353
}

354 355
static inline void sad8_4_mmx(uint8_t *blk1, uint8_t *blk2,
                              ptrdiff_t stride, int h)
356
{
357
    x86_reg len = -stride * h;
358
    __asm__ volatile (
359 360
        "movq  (%1, %%"FF_REG_a"), %%mm0\n\t"
        "movq 1(%1, %%"FF_REG_a"), %%mm2\n\t"
361 362 363 364 365 366 367 368 369 370
        "movq %%mm0, %%mm1              \n\t"
        "movq %%mm2, %%mm3              \n\t"
        "punpcklbw %%mm7, %%mm0         \n\t"
        "punpckhbw %%mm7, %%mm1         \n\t"
        "punpcklbw %%mm7, %%mm2         \n\t"
        "punpckhbw %%mm7, %%mm3         \n\t"
        "paddw %%mm2, %%mm0             \n\t"
        "paddw %%mm3, %%mm1             \n\t"
        ".p2align 4                     \n\t"
        "1:                             \n\t"
371 372
        "movq  (%2, %%"FF_REG_a"), %%mm2\n\t"
        "movq 1(%2, %%"FF_REG_a"), %%mm4\n\t"
373 374 375 376 377 378 379 380
        "movq %%mm2, %%mm3              \n\t"
        "movq %%mm4, %%mm5              \n\t"
        "punpcklbw %%mm7, %%mm2         \n\t"
        "punpckhbw %%mm7, %%mm3         \n\t"
        "punpcklbw %%mm7, %%mm4         \n\t"
        "punpckhbw %%mm7, %%mm5         \n\t"
        "paddw %%mm4, %%mm2             \n\t"
        "paddw %%mm5, %%mm3             \n\t"
381
        "movq %5, %%mm5                 \n\t"
382 383 384 385
        "paddw %%mm2, %%mm0             \n\t"
        "paddw %%mm3, %%mm1             \n\t"
        "paddw %%mm5, %%mm0             \n\t"
        "paddw %%mm5, %%mm1             \n\t"
386 387
        "movq (%3, %%"FF_REG_a"), %%mm4 \n\t"
        "movq (%3, %%"FF_REG_a"), %%mm5 \n\t"
388 389 390 391 392 393 394 395 396 397 398 399 400
        "psrlw $2, %%mm0                \n\t"
        "psrlw $2, %%mm1                \n\t"
        "packuswb %%mm1, %%mm0          \n\t"
        "psubusb %%mm0, %%mm4           \n\t"
        "psubusb %%mm5, %%mm0           \n\t"
        "por %%mm4, %%mm0               \n\t"
        "movq %%mm0, %%mm4              \n\t"
        "punpcklbw %%mm7, %%mm0         \n\t"
        "punpckhbw %%mm7, %%mm4         \n\t"
        "paddw %%mm0, %%mm6             \n\t"
        "paddw %%mm4, %%mm6             \n\t"
        "movq  %%mm2, %%mm0             \n\t"
        "movq  %%mm3, %%mm1             \n\t"
401
        "add %4, %%"FF_REG_a"           \n\t"
402 403 404
        " js 1b                         \n\t"
        : "+a" (len)
        : "r" (blk1 - len), "r" (blk1 - len + stride), "r" (blk2 - len),
405
          "r" (stride), "m" (round_tab[2]));
406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422
}

static inline int sum_mmx(void)
{
    int ret;
    __asm__ volatile (
        "movq %%mm6, %%mm0              \n\t"
        "psrlq $32, %%mm6               \n\t"
        "paddw %%mm0, %%mm6             \n\t"
        "movq %%mm6, %%mm0              \n\t"
        "psrlq $16, %%mm6               \n\t"
        "paddw %%mm0, %%mm6             \n\t"
        "movd %%mm6, %0                 \n\t"
        : "=r" (ret));
    return ret & 0xFFFF;
}

423 424
static inline void sad8_x2a_mmx(uint8_t *blk1, uint8_t *blk2,
                                ptrdiff_t stride, int h)
425 426 427 428
{
    sad8_2_mmx(blk1, blk1 + 1, blk2, stride, h);
}

429 430
static inline void sad8_y2a_mmx(uint8_t *blk1, uint8_t *blk2,
                                ptrdiff_t stride, int h)
431 432 433 434 435 436
{
    sad8_2_mmx(blk1, blk1 + stride, blk2, stride, h);
}

#define PIX_SAD(suf)                                                    \
static int sad8_ ## suf(MpegEncContext *v, uint8_t *blk2,               \
437
                        uint8_t *blk1, ptrdiff_t stride, int h)         \
438
{                                                                       \
439
    av_assert2(h == 8);                                                     \
440 441 442 443 444 445 446 447 448 449 450
    __asm__ volatile (                                                  \
        "pxor %%mm7, %%mm7     \n\t"                                    \
        "pxor %%mm6, %%mm6     \n\t"                                    \
        :);                                                             \
                                                                        \
    sad8_1_ ## suf(blk1, blk2, stride, 8);                              \
                                                                        \
    return sum_ ## suf();                                               \
}                                                                       \
                                                                        \
static int sad8_x2_ ## suf(MpegEncContext *v, uint8_t *blk2,            \
451
                           uint8_t *blk1, ptrdiff_t stride, int h)      \
452
{                                                                       \
453
    av_assert2(h == 8);                                                     \
454 455 456 457 458 459 460 461 462 463 464 465
    __asm__ volatile (                                                  \
        "pxor %%mm7, %%mm7     \n\t"                                    \
        "pxor %%mm6, %%mm6     \n\t"                                    \
        "movq %0, %%mm5        \n\t"                                    \
        :: "m" (round_tab[1]));                                         \
                                                                        \
    sad8_x2a_ ## suf(blk1, blk2, stride, 8);                            \
                                                                        \
    return sum_ ## suf();                                               \
}                                                                       \
                                                                        \
static int sad8_y2_ ## suf(MpegEncContext *v, uint8_t *blk2,            \
466
                           uint8_t *blk1, ptrdiff_t stride, int h)      \
467
{                                                                       \
468
    av_assert2(h == 8);                                                     \
469 470 471 472 473 474 475 476 477 478 479 480
    __asm__ volatile (                                                  \
        "pxor %%mm7, %%mm7     \n\t"                                    \
        "pxor %%mm6, %%mm6     \n\t"                                    \
        "movq %0, %%mm5        \n\t"                                    \
        :: "m" (round_tab[1]));                                         \
                                                                        \
    sad8_y2a_ ## suf(blk1, blk2, stride, 8);                            \
                                                                        \
    return sum_ ## suf();                                               \
}                                                                       \
                                                                        \
static int sad8_xy2_ ## suf(MpegEncContext *v, uint8_t *blk2,           \
481
                            uint8_t *blk1, ptrdiff_t stride, int h)     \
482
{                                                                       \
483
    av_assert2(h == 8);                                                     \
484 485 486 487 488 489 490 491 492 493 494
    __asm__ volatile (                                                  \
        "pxor %%mm7, %%mm7     \n\t"                                    \
        "pxor %%mm6, %%mm6     \n\t"                                    \
        ::);                                                            \
                                                                        \
    sad8_4_ ## suf(blk1, blk2, stride, 8);                              \
                                                                        \
    return sum_ ## suf();                                               \
}                                                                       \
                                                                        \
static int sad16_ ## suf(MpegEncContext *v, uint8_t *blk2,              \
495
                         uint8_t *blk1, ptrdiff_t stride, int h)        \
496 497 498 499 500 501 502 503 504 505 506 507 508
{                                                                       \
    __asm__ volatile (                                                  \
        "pxor %%mm7, %%mm7     \n\t"                                    \
        "pxor %%mm6, %%mm6     \n\t"                                    \
        :);                                                             \
                                                                        \
    sad8_1_ ## suf(blk1,     blk2,     stride, h);                      \
    sad8_1_ ## suf(blk1 + 8, blk2 + 8, stride, h);                      \
                                                                        \
    return sum_ ## suf();                                               \
}                                                                       \
                                                                        \
static int sad16_x2_ ## suf(MpegEncContext *v, uint8_t *blk2,           \
509
                            uint8_t *blk1, ptrdiff_t stride, int h)     \
510 511 512 513 514 515 516 517 518 519 520 521 522 523
{                                                                       \
    __asm__ volatile (                                                  \
        "pxor %%mm7, %%mm7     \n\t"                                    \
        "pxor %%mm6, %%mm6     \n\t"                                    \
        "movq %0, %%mm5        \n\t"                                    \
        :: "m" (round_tab[1]));                                         \
                                                                        \
    sad8_x2a_ ## suf(blk1,     blk2,     stride, h);                    \
    sad8_x2a_ ## suf(blk1 + 8, blk2 + 8, stride, h);                    \
                                                                        \
    return sum_ ## suf();                                               \
}                                                                       \
                                                                        \
static int sad16_y2_ ## suf(MpegEncContext *v, uint8_t *blk2,           \
524
                            uint8_t *blk1, ptrdiff_t stride, int h)     \
525 526 527 528 529 530 531 532 533 534 535 536 537 538
{                                                                       \
    __asm__ volatile (                                                  \
        "pxor %%mm7, %%mm7     \n\t"                                    \
        "pxor %%mm6, %%mm6     \n\t"                                    \
        "movq %0, %%mm5        \n\t"                                    \
        :: "m" (round_tab[1]));                                         \
                                                                        \
    sad8_y2a_ ## suf(blk1,     blk2,     stride, h);                    \
    sad8_y2a_ ## suf(blk1 + 8, blk2 + 8, stride, h);                    \
                                                                        \
    return sum_ ## suf();                                               \
}                                                                       \
                                                                        \
static int sad16_xy2_ ## suf(MpegEncContext *v, uint8_t *blk2,          \
539
                             uint8_t *blk1, ptrdiff_t stride, int h)    \
540 541 542 543 544 545 546 547 548 549 550 551 552 553
{                                                                       \
    __asm__ volatile (                                                  \
        "pxor %%mm7, %%mm7     \n\t"                                    \
        "pxor %%mm6, %%mm6     \n\t"                                    \
        ::);                                                            \
                                                                        \
    sad8_4_ ## suf(blk1,     blk2,     stride, h);                      \
    sad8_4_ ## suf(blk1 + 8, blk2 + 8, stride, h);                      \
                                                                        \
    return sum_ ## suf();                                               \
}                                                                       \

PIX_SAD(mmx)

554 555
#endif /* HAVE_INLINE_ASM */

556
av_cold void ff_me_cmp_init_x86(MECmpContext *c, AVCodecContext *avctx)
557
{
558 559 560 561
    int cpu_flags = av_get_cpu_flags();

#if HAVE_INLINE_ASM
    if (INLINE_MMX(cpu_flags)) {
562 563 564 565 566 567 568 569 570 571 572 573
        c->pix_abs[0][0] = sad16_mmx;
        c->pix_abs[0][1] = sad16_x2_mmx;
        c->pix_abs[0][2] = sad16_y2_mmx;
        c->pix_abs[0][3] = sad16_xy2_mmx;
        c->pix_abs[1][0] = sad8_mmx;
        c->pix_abs[1][1] = sad8_x2_mmx;
        c->pix_abs[1][2] = sad8_y2_mmx;
        c->pix_abs[1][3] = sad8_xy2_mmx;

        c->sad[0] = sad16_mmx;
        c->sad[1] = sad8_mmx;

574 575
        c->vsad[4] = vsad_intra16_mmx;

576
        if (!(avctx->flags & AV_CODEC_FLAG_BITEXACT)) {
577
            c->vsad[0] = vsad16_mmx;
578 579 580 581 582 583 584 585
        }
    }

#endif /* HAVE_INLINE_ASM */

    if (EXTERNAL_MMX(cpu_flags)) {
        c->hadamard8_diff[0] = ff_hadamard8_diff16_mmx;
        c->hadamard8_diff[1] = ff_hadamard8_diff_mmx;
586 587 588 589 590 591 592
        c->sum_abs_dctelem   = ff_sum_abs_dctelem_mmx;
        c->sse[0]            = ff_sse16_mmx;
        c->sse[1]            = ff_sse8_mmx;
#if HAVE_YASM
        c->nsse[0]           = nsse16_mmx;
        c->nsse[1]           = nsse8_mmx;
#endif
593 594 595 596 597
    }

    if (EXTERNAL_MMXEXT(cpu_flags)) {
        c->hadamard8_diff[0] = ff_hadamard8_diff16_mmxext;
        c->hadamard8_diff[1] = ff_hadamard8_diff_mmxext;
598
        c->sum_abs_dctelem   = ff_sum_abs_dctelem_mmxext;
599 600 601 602 603 604 605 606 607 608 609

        c->sad[0] = ff_sad16_mmxext;
        c->sad[1] = ff_sad8_mmxext;

        c->pix_abs[0][0] = ff_sad16_mmxext;
        c->pix_abs[0][1] = ff_sad16_x2_mmxext;
        c->pix_abs[0][2] = ff_sad16_y2_mmxext;
        c->pix_abs[1][0] = ff_sad8_mmxext;
        c->pix_abs[1][1] = ff_sad8_x2_mmxext;
        c->pix_abs[1][2] = ff_sad8_y2_mmxext;

610 611 612
        c->vsad[4] = ff_vsad_intra16_mmxext;
        c->vsad[5] = ff_vsad_intra8_mmxext;

613
        if (!(avctx->flags & AV_CODEC_FLAG_BITEXACT)) {
614 615
            c->pix_abs[0][3] = ff_sad16_approx_xy2_mmxext;
            c->pix_abs[1][3] = ff_sad8_approx_xy2_mmxext;
616 617 618

            c->vsad[0] = ff_vsad16_approx_mmxext;
            c->vsad[1] = ff_vsad8_approx_mmxext;
619
        }
620 621 622 623
    }

    if (EXTERNAL_SSE2(cpu_flags)) {
        c->sse[0] = ff_sse16_sse2;
624
        c->sum_abs_dctelem   = ff_sum_abs_dctelem_sse2;
625 626 627 628 629

#if HAVE_ALIGNED_STACK
        c->hadamard8_diff[0] = ff_hadamard8_diff16_sse2;
        c->hadamard8_diff[1] = ff_hadamard8_diff_sse2;
#endif
630 631 632 633 634 635
        if (!(cpu_flags & AV_CPU_FLAG_SSE2SLOW) && avctx->codec_id != AV_CODEC_ID_SNOW) {
            c->sad[0]        = ff_sad16_sse2;
            c->pix_abs[0][0] = ff_sad16_sse2;
            c->pix_abs[0][1] = ff_sad16_x2_sse2;
            c->pix_abs[0][2] = ff_sad16_y2_sse2;

636
            c->vsad[4]       = ff_vsad_intra16_sse2;
637
            if (!(avctx->flags & AV_CODEC_FLAG_BITEXACT)) {
638
                c->pix_abs[0][3] = ff_sad16_approx_xy2_sse2;
639
                c->vsad[0]       = ff_vsad16_approx_sse2;
640 641
            }
        }
642 643
    }

644 645 646
    if (EXTERNAL_SSSE3(cpu_flags)) {
        c->sum_abs_dctelem   = ff_sum_abs_dctelem_ssse3;
#if HAVE_ALIGNED_STACK
647 648
        c->hadamard8_diff[0] = ff_hadamard8_diff16_ssse3;
        c->hadamard8_diff[1] = ff_hadamard8_diff_ssse3;
649
#endif
650
    }
651
}