h264dsp_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 void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\
}\
static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\
    OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\
}\
\
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
    src += 8*srcStride;\
    dst += 8*dstStride;\
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\
    OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\
}\
\
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
    src += 8*dstStride;\
    dst += 8*dstStride;\
    src2 += 8*src2Stride;\
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\
    OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\
}\
\
static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 8);\
}\
\
static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
    OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 16);\
}\
\
static av_noinline void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
{\
    asm volatile(\
        "movq       %5,  %%mm6          \n\t"\
        "movq      (%1), %%mm0          \n\t"\
        "movq    24(%1), %%mm1          \n\t"\
        "paddw    %%mm6, %%mm0          \n\t"\
        "paddw    %%mm6, %%mm1          \n\t"\
        "psraw      $5,  %%mm0          \n\t"\
        "psraw      $5,  %%mm1          \n\t"\
        "packuswb %%mm0, %%mm0          \n\t"\
        "packuswb %%mm1, %%mm1          \n\t"\
        PAVGB"     (%0), %%mm0          \n\t"\
        PAVGB"  (%0,%3), %%mm1          \n\t"\
        OP(%%mm0, (%2),    %%mm4, d)\
        OP(%%mm1, (%2,%4), %%mm5, d)\
        "lea  (%0,%3,2), %0             \n\t"\
        "lea  (%2,%4,2), %2             \n\t"\
        "movq    48(%1), %%mm0          \n\t"\
        "movq    72(%1), %%mm1          \n\t"\
        "paddw    %%mm6, %%mm0          \n\t"\
        "paddw    %%mm6, %%mm1          \n\t"\
        "psraw      $5,  %%mm0          \n\t"\
        "psraw      $5,  %%mm1          \n\t"\
        "packuswb %%mm0, %%mm0          \n\t"\
        "packuswb %%mm1, %%mm1          \n\t"\
        PAVGB"     (%0), %%mm0          \n\t"\
        PAVGB"  (%0,%3), %%mm1          \n\t"\
        OP(%%mm0, (%2),    %%mm4, d)\
        OP(%%mm1, (%2,%4), %%mm5, d)\
        :"+a"(src8), "+c"(src16), "+d"(dst)\
        :"S"((long)src8Stride), "D"((long)dstStride), "m"(ff_pw_16)\
        :"memory");\
}\
static av_noinline void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
{\
    asm volatile(\
        "movq       %0,  %%mm6          \n\t"\
        ::"m"(ff_pw_16)\
        );\
    while(h--){\
    asm volatile(\
        "movq      (%1), %%mm0          \n\t"\
        "movq     8(%1), %%mm1          \n\t"\
        "paddw    %%mm6, %%mm0          \n\t"\
        "paddw    %%mm6, %%mm1          \n\t"\
        "psraw      $5,  %%mm0          \n\t"\
        "psraw      $5,  %%mm1          \n\t"\
        "packuswb %%mm1, %%mm0          \n\t"\
        PAVGB"     (%0), %%mm0          \n\t"\
        OP(%%mm0, (%2), %%mm5, q)\
        ::"a"(src8), "c"(src16), "d"(dst)\
        :"memory");\
        src8 += src8Stride;\
        src16 += 24;\
        dst += dstStride;\
    }\
}\
static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\
{\
    OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\
    OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\
}\


#define H264_MC(OPNAME, SIZE, MMX) \
static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## pixels ## SIZE ## _mmx(dst, src, stride, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[SIZE*SIZE/8];\
    uint8_t * const half= (uint8_t*)temp;\
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
    OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[SIZE*SIZE/8];\
    uint8_t * const half= (uint8_t*)temp;\
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\
    OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, half, stride, stride, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[SIZE*SIZE/8];\
    uint8_t * const halfV= (uint8_t*)temp;\
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfV, stride, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[SIZE*SIZE/8];\
    uint8_t * const halfV= (uint8_t*)temp;\
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfV, stride, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[SIZE*SIZE/8];\
    uint8_t * const halfV= (uint8_t*)temp;\
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfV, stride, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[SIZE*SIZE/8];\
    uint8_t * const halfV= (uint8_t*)temp;\
    put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfV, stride, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[SIZE*(SIZE<8?12:24)/4];\
    int16_t * const tmp= (int16_t*)temp;\
    OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, tmp, src, stride, SIZE, stride);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
    uint8_t * const halfHV= (uint8_t*)temp;\
    int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE/2;\
    assert((int)temp & 7 == 0);\
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
    uint8_t * const halfHV= (uint8_t*)temp;\
    int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE/2;\
    assert((int)temp & 7 == 0);\
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
    int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\
    uint8_t * const halfHV= ((uint8_t*)temp);\
    assert((int)temp & 7 == 0);\
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
    OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\
    uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\
    int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\
    uint8_t * const halfHV= ((uint8_t*)temp);\
    assert((int)temp & 7 == 0);\
    put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\
    OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\
}\


#define AVG_3DNOW_OP(a,b,temp, size) \
"mov" #size " " #b ", " #temp "   \n\t"\
"pavgusb " #temp ", " #a "        \n\t"\
"mov" #size " " #a ", " #b "      \n\t"
#define AVG_MMX2_OP(a,b,temp, size) \
"mov" #size " " #b ", " #temp "   \n\t"\
"pavgb " #temp ", " #a "          \n\t"\
"mov" #size " " #a ", " #b "      \n\t"

#define PAVGB "pavgusb"
QPEL_H264(put_,       PUT_OP, 3dnow)
QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)
#undef PAVGB
#define PAVGB "pavgb"
QPEL_H264(put_,       PUT_OP, mmx2)
QPEL_H264(avg_,  AVG_MMX2_OP, mmx2)
#undef PAVGB

H264_MC(put_, 4, 3dnow)
H264_MC(put_, 8, 3dnow)
H264_MC(put_, 16,3dnow)
H264_MC(avg_, 4, 3dnow)
H264_MC(avg_, 8, 3dnow)
H264_MC(avg_, 16,3dnow)
H264_MC(put_, 4, mmx2)
H264_MC(put_, 8, mmx2)
H264_MC(put_, 16,mmx2)
H264_MC(avg_, 4, mmx2)
H264_MC(avg_, 8, mmx2)
H264_MC(avg_, 16,mmx2)


#define H264_CHROMA_OP(S,D)
#define H264_CHROMA_OP4(S,D,T)
#define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx
#define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_mmx
#define H264_CHROMA_MC2_TMPL put_h264_chroma_mc2_mmx2
#define H264_CHROMA_MC8_MV0 put_pixels8_mmx
#include "dsputil_h264_template_mmx.c"
#undef H264_CHROMA_OP
#undef H264_CHROMA_OP4
#undef H264_CHROMA_MC8_TMPL
#undef H264_CHROMA_MC4_TMPL
#undef H264_CHROMA_MC2_TMPL
#undef H264_CHROMA_MC8_MV0

#define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"
#define H264_CHROMA_OP4(S,D,T) "movd  " #S ", " #T " \n\t"\
                               "pavgb " #T ", " #D " \n\t"
#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2
#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_mmx2
#define H264_CHROMA_MC2_TMPL avg_h264_chroma_mc2_mmx2
#define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2
#include "dsputil_h264_template_mmx.c"
#undef H264_CHROMA_OP
#undef H264_CHROMA_OP4
#undef H264_CHROMA_MC8_TMPL
#undef H264_CHROMA_MC4_TMPL
#undef H264_CHROMA_MC2_TMPL
#undef H264_CHROMA_MC8_MV0

#define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"
#define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\
                               "pavgusb " #T ", " #D " \n\t"
#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow
#define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_3dnow
#define H264_CHROMA_MC8_MV0 avg_pixels8_3dnow
#include "dsputil_h264_template_mmx.c"
#undef H264_CHROMA_OP
#undef H264_CHROMA_OP4
#undef H264_CHROMA_MC8_TMPL
#undef H264_CHROMA_MC4_TMPL
#undef H264_CHROMA_MC8_MV0

/***********************************/
/* weighted prediction */

static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)
{
    int x, y;
    offset <<= log2_denom;
    offset += (1 << log2_denom) >> 1;
    asm volatile(
        "movd    %0, %%mm4        \n\t"
        "movd    %1, %%mm5        \n\t"
        "movd    %2, %%mm6        \n\t"
        "pshufw  $0, %%mm4, %%mm4 \n\t"
        "pshufw  $0, %%mm5, %%mm5 \n\t"
        "pxor    %%mm7, %%mm7     \n\t"
        :: "g"(weight), "g"(offset), "g"(log2_denom)
    );
    for(y=0; y<h; y+=2){
        for(x=0; x<w; x+=4){
            asm volatile(
                "movd      %0,    %%mm0 \n\t"
                "movd      %1,    %%mm1 \n\t"
                "punpcklbw %%mm7, %%mm0 \n\t"
                "punpcklbw %%mm7, %%mm1 \n\t"
                "pmullw    %%mm4, %%mm0 \n\t"
                "pmullw    %%mm4, %%mm1 \n\t"
                "paddsw    %%mm5, %%mm0 \n\t"
                "paddsw    %%mm5, %%mm1 \n\t"
                "psraw     %%mm6, %%mm0 \n\t"
                "psraw     %%mm6, %%mm1 \n\t"
                "packuswb  %%mm7, %%mm0 \n\t"
                "packuswb  %%mm7, %%mm1 \n\t"
                "movd      %%mm0, %0    \n\t"
                "movd      %%mm1, %1    \n\t"
                : "+m"(*(uint32_t*)(dst+x)),
                  "+m"(*(uint32_t*)(dst+x+stride))
            );
        }
        dst += 2*stride;
    }
}

static inline void ff_h264_biweight_WxH_mmx2(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset, int w, int h)
{
    int x, y;
    offset = ((offset + 1) | 1) << log2_denom;
    asm volatile(
        "movd    %0, %%mm3        \n\t"
        "movd    %1, %%mm4        \n\t"
        "movd    %2, %%mm5        \n\t"
        "movd    %3, %%mm6        \n\t"
        "pshufw  $0, %%mm3, %%mm3 \n\t"
        "pshufw  $0, %%mm4, %%mm4 \n\t"
        "pshufw  $0, %%mm5, %%mm5 \n\t"
        "pxor    %%mm7, %%mm7     \n\t"
        :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)
    );
    for(y=0; y<h; y++){
        for(x=0; x<w; x+=4){
            asm volatile(
                "movd      %0,    %%mm0 \n\t"
                "movd      %1,    %%mm1 \n\t"
                "punpcklbw %%mm7, %%mm0 \n\t"
                "punpcklbw %%mm7, %%mm1 \n\t"
                "pmullw    %%mm3, %%mm0 \n\t"
                "pmullw    %%mm4, %%mm1 \n\t"
                "paddsw    %%mm1, %%mm0 \n\t"
                "paddsw    %%mm5, %%mm0 \n\t"
                "psraw     %%mm6, %%mm0 \n\t"
                "packuswb  %%mm0, %%mm0 \n\t"
                "movd      %%mm0, %0    \n\t"
                : "+m"(*(uint32_t*)(dst+x))
                :  "m"(*(uint32_t*)(src+x))
            );
        }
        src += stride;
        dst += stride;
    }
}

#define H264_WEIGHT(W,H) \
static void ff_h264_biweight_ ## W ## x ## H ## _mmx2(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset){ \
    ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \
} \
static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \
    ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \
}

H264_WEIGHT(16,16)
H264_WEIGHT(16, 8)
H264_WEIGHT( 8,16)
H264_WEIGHT( 8, 8)
H264_WEIGHT( 8, 4)
H264_WEIGHT( 4, 8)
H264_WEIGHT( 4, 4)
H264_WEIGHT( 4, 2)