motion_est_mmx.c

ffmpeg的完整源代码和作者自己写的文档。不但有在Linux的工程哦

    );
}

static inline void sad8_4_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
{
    long len= -(stride*h);
    asm volatile(
        "movq (%1, %%"REG_a"), %%mm0    \n\t"
        "movq 1(%1, %%"REG_a"), %%mm2   \n\t"
        "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"
        ASMALIGN(4)
        "1:                             \n\t"
        "movq (%2, %%"REG_a"), %%mm2    \n\t"
        "movq 1(%2, %%"REG_a"), %%mm4   \n\t"
        "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"
        "movq 16+"MANGLE(round_tab)", %%mm5 \n\t"
        "paddw %%mm2, %%mm0             \n\t"
        "paddw %%mm3, %%mm1             \n\t"
        "paddw %%mm5, %%mm0             \n\t"
        "paddw %%mm5, %%mm1             \n\t"
        "movq (%3, %%"REG_a"), %%mm4    \n\t"
        "movq (%3, %%"REG_a"), %%mm5    \n\t"
        "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"
        "add %4, %%"REG_a"              \n\t"
        " js 1b                         \n\t"
        : "+a" (len)
        : "r" (blk1 - len), "r" (blk1 -len + stride), "r" (blk2 - len), "r" ((long)stride)
    );
}

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;
}

static inline int sum_mmx2(void)
{
    int ret;
    asm volatile(
        "movd %%mm6, %0                 \n\t"
        : "=r" (ret)
    );
    return ret;
}

static inline void sad8_x2a_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
{
    sad8_2_mmx(blk1, blk1+1, blk2, stride, h);
}
static inline void sad8_y2a_mmx(uint8_t *blk1, uint8_t *blk2, int stride, int h)
{
    sad8_2_mmx(blk1, blk1+stride, blk2, stride, h);
}


#define PIX_SAD(suf)\
static int sad8_ ## suf(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
{\
    assert(h==8);\
    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(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
{\
    assert(h==8);\
    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(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
{\
    assert(h==8);\
    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(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
{\
    assert(h==8);\
    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(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
{\
    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(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
{\
    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(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
{\
    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(void *v, uint8_t *blk2, uint8_t *blk1, int stride, int h)\
{\
    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)
PIX_SAD(mmx2)

void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx)
{
    if (mm_flags & MM_MMX) {
        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;
    }
    if (mm_flags & MM_MMXEXT) {
        c->pix_abs[0][0] = sad16_mmx2;
        c->pix_abs[1][0] = sad8_mmx2;

        c->sad[0]= sad16_mmx2;
        c->sad[1]= sad8_mmx2;

        if(!(avctx->flags & CODEC_FLAG_BITEXACT)){
            c->pix_abs[0][1] = sad16_x2_mmx2;
            c->pix_abs[0][2] = sad16_y2_mmx2;
            c->pix_abs[0][3] = sad16_xy2_mmx2;
            c->pix_abs[1][1] = sad8_x2_mmx2;
            c->pix_abs[1][2] = sad8_y2_mmx2;
            c->pix_abs[1][3] = sad8_xy2_mmx2;
        }
    }
    if ((mm_flags & MM_SSE2) && !(mm_flags & MM_3DNOW)) {
        c->sad[0]= sad16_sse2;
    }
}