comparison i386/h264dsp_mmx.c @ 8031:eebc7209c47f libavcodec

Convert asm keyword into __asm__. Neither the asm() nor the __asm__() keyword is part of the C99 standard, but while GCC accepts the former in C89 syntax, it is not accepted in C99 unless GNU extensions are turned on (with -fasm). The latter form is accepted in any syntax as an extension (without requiring further command-line options). Sun Studio C99 compiler also does not accept asm() while accepting __asm__(), albeit reporting warnings that it's not valid C99 syntax.

8030:a512ac8fa540

    "movd      "#p",    (%0) \n\t"

static void ff_h264_idct_add_mmx(uint8_t *dst, int16_t *block, int stride)
{
    /* Load dct coeffs */
    asm volatile(
        "movq   (%0), %%mm0 \n\t"
        "movq  8(%0), %%mm1 \n\t"
        "movq 16(%0), %%mm2 \n\t"
        "movq 24(%0), %%mm3 \n\t"
    :: "r"(block) );

    asm volatile(
        /* mm1=s02+s13  mm2=s02-s13  mm4=d02+d13  mm0=d02-d13 */
        IDCT4_1D( %%mm2, %%mm1, %%mm0, %%mm3, %%mm4 )

        "movq      %0,    %%mm6 \n\t"
        /* in: 1,4,0,2  out: 1,2,3,0 */

        IDCT4_1D( %%mm4, %%mm2, %%mm3, %%mm0, %%mm1 )

        "pxor %%mm7, %%mm7    \n\t"
    :: "m"(ff_pw_32));

    asm volatile(
    STORE_DIFF_4P( %%mm0, %%mm1, %%mm7)
        "add %1, %0             \n\t"
    STORE_DIFF_4P( %%mm2, %%mm1, %%mm7)
        "add %1, %0             \n\t"
    STORE_DIFF_4P( %%mm3, %%mm1, %%mm7)

    );
}

static inline void h264_idct8_1d(int16_t *block)
{
    asm volatile(
        "movq 112(%0), %%mm7  \n\t"
        "movq  80(%0), %%mm0  \n\t"
        "movq  48(%0), %%mm3  \n\t"
        "movq  16(%0), %%mm5  \n\t"

    for(i=0; i<2; i++){
        DECLARE_ALIGNED_8(uint64_t, tmp);

        h264_idct8_1d(block+4*i);

        asm volatile(
            "movq   %%mm7,    %0   \n\t"
            TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )
            "movq   %%mm0,  8(%1)  \n\t"
            "movq   %%mm6, 24(%1)  \n\t"
            "movq   %%mm7, 40(%1)  \n\t"

    }

    for(i=0; i<2; i++){
        h264_idct8_1d(b2+4*i);

        asm volatile(
            "psraw     $6, %%mm7  \n\t"
            "psraw     $6, %%mm6  \n\t"
            "psraw     $6, %%mm5  \n\t"
            "psraw     $6, %%mm4  \n\t"
            "psraw     $6, %%mm3  \n\t"

        SUMSUB_BA(h, a)\
        SUMSUB_BA(d, f)

static void ff_h264_idct8_add_sse2(uint8_t *dst, int16_t *block, int stride)
{
    asm volatile(
        "movdqa   0x10(%1), %%xmm1 \n"
        "movdqa   0x20(%1), %%xmm2 \n"
        "movdqa   0x30(%1), %%xmm3 \n"
        "movdqa   0x50(%1), %%xmm5 \n"
        "movdqa   0x60(%1), %%xmm6 \n"

}

static void ff_h264_idct_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
{
    int dc = (block[0] + 32) >> 6;
    asm volatile(
        "movd          %0, %%mm0 \n\t"
        "pshufw $0, %%mm0, %%mm0 \n\t"
        "pxor       %%mm1, %%mm1 \n\t"
        "psubw      %%mm0, %%mm1 \n\t"
        "packuswb   %%mm0, %%mm0 \n\t"
        "packuswb   %%mm1, %%mm1 \n\t"
        ::"r"(dc)
    );
    asm volatile(
        "movd          %0, %%mm2 \n\t"
        "movd          %1, %%mm3 \n\t"
        "movd          %2, %%mm4 \n\t"
        "movd          %3, %%mm5 \n\t"
        "paddusb    %%mm0, %%mm2 \n\t"

static void ff_h264_idct8_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
{
    int dc = (block[0] + 32) >> 6;
    int y;
    asm volatile(
        "movd          %0, %%mm0 \n\t"
        "pshufw $0, %%mm0, %%mm0 \n\t"
        "pxor       %%mm1, %%mm1 \n\t"
        "psubw      %%mm0, %%mm1 \n\t"
        "packuswb   %%mm0, %%mm0 \n\t"
        "packuswb   %%mm1, %%mm1 \n\t"
        ::"r"(dc)
    );
    for(y=2; y--; dst += 4*stride){
    asm volatile(
        "movq          %0, %%mm2 \n\t"
        "movq          %1, %%mm3 \n\t"
        "movq          %2, %%mm4 \n\t"
        "movq          %3, %%mm5 \n\t"
        "paddusb    %%mm0, %%mm2 \n\t"

static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
{
    DECLARE_ALIGNED_8(uint64_t, tmp0[2]);

    asm volatile(
        "movq    (%1,%3), %%mm0    \n\t" //p1
        "movq    (%1,%3,2), %%mm1  \n\t" //p0
        "movq    (%2),    %%mm2    \n\t" //q0
        "movq    (%2,%3), %%mm3    \n\t" //q1
        H264_DEBLOCK_MASK(%6, %7)

    }
}

static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
{
    asm volatile(
        "movq    (%0),    %%mm0     \n\t" //p1
        "movq    (%0,%2), %%mm1     \n\t" //p0
        "movq    (%1),    %%mm2     \n\t" //q0
        "movq    (%1,%2), %%mm3     \n\t" //q1
        H264_DEBLOCK_MASK(%4, %5)

    "psubusb %%mm4, "#p0"  \n\t"\
    "pavgb   "#p1", "#p0"  \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\

static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
{
    asm volatile(
        "movq    (%0),    %%mm0     \n\t"
        "movq    (%0,%2), %%mm1     \n\t"
        "movq    (%1),    %%mm2     \n\t"
        "movq    (%1,%2), %%mm3     \n\t"
        H264_DEBLOCK_MASK(%3, %4)

}

static void h264_loop_filter_strength_mmx2( int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
                                            int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field ) {
    int dir;
    asm volatile(
        "pxor %%mm7, %%mm7 \n\t"
        "movq %0, %%mm6 \n\t"
        "movq %1, %%mm5 \n\t"
        "movq %2, %%mm4 \n\t"
        ::"m"(ff_pb_1), "m"(ff_pb_3), "m"(ff_pb_7)
    );
    if(field)
        asm volatile(
            "movq %0, %%mm5 \n\t"
            "movq %1, %%mm4 \n\t"
            ::"m"(ff_pb_3_1), "m"(ff_pb_7_3)
        );

        const int d_idx = dir ? -8 : -1;
        const int mask_mv = dir ? mask_mv1 : mask_mv0;
        DECLARE_ALIGNED_8(const uint64_t, mask_dir) = dir ? 0 : 0xffffffffffffffffULL;
        int b_idx, edge, l;
        for( b_idx=12, edge=0; edge<edges; edge+=step, b_idx+=8*step ) {
            asm volatile(
                "pand %0, %%mm0 \n\t"
                ::"m"(mask_dir)
            );
            if(!(mask_mv & edge)) {
                asm volatile("pxor %%mm0, %%mm0 \n\t":);
                for( l = bidir; l >= 0; l-- ) {
                    asm volatile(
                        "movd %0, %%mm1 \n\t"
                        "punpckldq %1, %%mm1 \n\t"
                        "movq %%mm1, %%mm2 \n\t"
                        "psrlw $7, %%mm2 \n\t"
                        "pand %%mm6, %%mm2 \n\t"

                          "m"(mv[l][b_idx+d_idx][0]),
                          "m"(mv[l][b_idx+d_idx+2][0])
                    );
                }
            }
            asm volatile(
                "movd %0, %%mm1 \n\t"
                "por  %1, %%mm1 \n\t"
                "punpcklbw %%mm7, %%mm1 \n\t"
                "pcmpgtw %%mm7, %%mm1 \n\t" // nnz[b] || nnz[bn]
                ::"m"(nnz[b_idx]),
                  "m"(nnz[b_idx+d_idx])
            );
            asm volatile(
                "pcmpeqw %%mm7, %%mm0 \n\t"
                "pcmpeqw %%mm7, %%mm0 \n\t"
                "psrlw $15, %%mm0 \n\t" // nonzero -> 1
                "psrlw $14, %%mm1 \n\t"
                "movq %%mm0, %%mm2 \n\t"

            );
        }
        edges = 4;
        step = 1;
    }
    asm volatile(
        "movq   (%0), %%mm0 \n\t"
        "movq  8(%0), %%mm1 \n\t"
        "movq 16(%0), %%mm2 \n\t"
        "movq 24(%0), %%mm3 \n\t"
        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4)

#define QPEL_H264(OPNAME, OP, MMX)\
static av_noinline void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    int h=4;\
\
    asm volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
        "movq %5, %%mm4             \n\t"\
        "movq %6, %%mm5             \n\t"\
        "1:                         \n\t"\
        "movd  -1(%0), %%mm1        \n\t"\

        : "memory"\
    );\
}\
static av_noinline void OPNAME ## h264_qpel4_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
    int h=4;\
    asm volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
        "movq %0, %%mm4             \n\t"\
        "movq %1, %%mm5             \n\t"\
        :: "m"(ff_pw_5), "m"(ff_pw_16)\
    );\
    do{\
    asm volatile(\
        "movd  -1(%0), %%mm1        \n\t"\
        "movd    (%0), %%mm2        \n\t"\
        "movd   1(%0), %%mm3        \n\t"\
        "movd   2(%0), %%mm0        \n\t"\
        "punpcklbw %%mm7, %%mm1     \n\t"\

    );\
    }while(--h);\
}\
static av_noinline void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    src -= 2*srcStride;\
    asm volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
        "movd (%0), %%mm0           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm1           \n\t"\
        "add %2, %0                 \n\t"\

static av_noinline void OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
    int h=4;\
    int w=3;\
    src -= 2*srcStride+2;\
    while(w--){\
        asm volatile(\
            "pxor %%mm7, %%mm7      \n\t"\
            "movd (%0), %%mm0       \n\t"\
            "add %2, %0             \n\t"\
            "movd (%0), %%mm1       \n\t"\
            "add %2, %0             \n\t"\

        );\
        tmp += 4;\
        src += 4 - 9*srcStride;\
    }\
    tmp -= 3*4;\
    asm volatile(\
        "1:                         \n\t"\
        "movq     (%0), %%mm0       \n\t"\
        "paddw  10(%0), %%mm0       \n\t"\
        "movq    2(%0), %%mm1       \n\t"\
        "paddw   8(%0), %%mm1       \n\t"\

    );\
}\
\
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    int h=8;\
    asm volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
        "movq %5, %%mm6             \n\t"\
        "1:                         \n\t"\
        "movq    (%0), %%mm0        \n\t"\
        "movq   1(%0), %%mm2        \n\t"\

    );\
}\
\
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
    int h=8;\
    asm volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
        "movq %0, %%mm6             \n\t"\
        :: "m"(ff_pw_5)\
    );\
    do{\
    asm volatile(\
        "movq    (%0), %%mm0        \n\t"\
        "movq   1(%0), %%mm2        \n\t"\
        "movq %%mm0, %%mm1          \n\t"\
        "movq %%mm2, %%mm3          \n\t"\
        "punpcklbw %%mm7, %%mm0     \n\t"\

static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
    int w= 2;\
    src -= 2*srcStride;\
    \
    while(w--){\
      asm volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
        "movd (%0), %%mm0           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm1           \n\t"\
        "add %2, %0                 \n\t"\

        : "+a"(src), "+c"(dst)\
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
        : "memory"\
     );\
     if(h==16){\
        asm volatile(\
            QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
            QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
            QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
            QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
            QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\

}\
static av_always_inline void OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){\
    int w = (size+8)>>2;\
    src -= 2*srcStride+2;\
    while(w--){\
        asm volatile(\
            "pxor %%mm7, %%mm7      \n\t"\
            "movd (%0), %%mm0       \n\t"\
            "add %2, %0             \n\t"\
            "movd (%0), %%mm1       \n\t"\
            "add %2, %0             \n\t"\

            : "+a"(src)\
            : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
            : "memory"\
        );\
        if(size==16){\
            asm volatile(\
                QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1,  8*48)\
                QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2,  9*48)\
                QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\
                QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\
                QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\

}\
static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
    int w = size>>4;\
    do{\
    int h = size;\
    asm volatile(\
        "1:                         \n\t"\
        "movq     (%0), %%mm0       \n\t"\
        "movq    8(%0), %%mm3       \n\t"\
        "movq    2(%0), %%mm1       \n\t"\
        "movq   10(%0), %%mm4       \n\t"\

    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      (%1), %%mm0          \n\t"\
        "movq    24(%1), %%mm1          \n\t"\
        "psraw      $5,  %%mm0          \n\t"\
        "psraw      $5,  %%mm1          \n\t"\
        "packuswb %%mm0, %%mm0          \n\t"\

        :"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)\
{\
    do{\
    asm volatile(\
        "movq      (%1), %%mm0          \n\t"\
        "movq     8(%1), %%mm1          \n\t"\
        "movq    48(%1), %%mm2          \n\t"\
        "movq  8+48(%1), %%mm3          \n\t"\
        "psraw      $5,  %%mm0          \n\t"\

#ifdef ARCH_X86_64
#define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
    int h=16;\
    asm volatile(\
        "pxor %%xmm15, %%xmm15      \n\t"\
        "movdqa %6, %%xmm14         \n\t"\
        "movdqa %7, %%xmm13         \n\t"\
        "1:                         \n\t"\
        "lddqu    3(%0), %%xmm1     \n\t"\

#endif // ARCH_X86_64

#define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
    int h=8;\
    asm volatile(\
        "pxor %%xmm7, %%xmm7        \n\t"\
        "movdqa %0, %%xmm6          \n\t"\
        :: "m"(ff_pw_5)\
    );\
    do{\
    asm volatile(\
        "lddqu   -5(%0), %%xmm1     \n\t"\
        "movdqa  %%xmm1, %%xmm0     \n\t"\
        "punpckhbw %%xmm7, %%xmm1   \n\t"\
        "punpcklbw %%xmm7, %%xmm0   \n\t"\
        "movdqa  %%xmm1, %%xmm2     \n\t"\

}\
QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
\
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    int h=8;\
    asm volatile(\
        "pxor %%xmm7, %%xmm7        \n\t"\
        "movdqa %5, %%xmm6          \n\t"\
        "1:                         \n\t"\
        "lddqu   -5(%0), %%xmm1     \n\t"\
        "movdqa  %%xmm1, %%xmm0     \n\t"\

#define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
    src -= 2*srcStride;\
    \
    asm volatile(\
        "pxor %%xmm7, %%xmm7        \n\t"\
        "movq (%0), %%xmm0          \n\t"\
        "add %2, %0                 \n\t"\
        "movq (%0), %%xmm1          \n\t"\
        "add %2, %0                 \n\t"\

        : "+a"(src), "+c"(dst)\
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
        : "memory"\
    );\
    if(h==16){\
        asm volatile(\
            QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
            QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
            QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
            QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
            QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\

static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
    int w = (size+8)>>3;
    src -= 2*srcStride+2;
    while(w--){
        asm volatile(
            "pxor %%xmm7, %%xmm7        \n\t"
            "movq (%0), %%xmm0          \n\t"
            "add %2, %0                 \n\t"
            "movq (%0), %%xmm1          \n\t"
            "add %2, %0                 \n\t"

            : "+a"(src)
            : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
            : "memory"
        );
        if(size==16){
            asm volatile(
                QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1,  8*48)
                QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2,  9*48)
                QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 10*48)
                QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 11*48)
                QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 12*48)

#define QPEL_H264_HV2_XMM(OPNAME, OP, MMX)\
static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
    int h = size;\
    if(size == 16){\
        asm volatile(\
            "1:                         \n\t"\
            "movdqa 32(%0), %%xmm4      \n\t"\
            "movdqa 16(%0), %%xmm5      \n\t"\
            "movdqa   (%0), %%xmm7      \n\t"\
            "movdqa %%xmm4, %%xmm3      \n\t"\

            : "+a"(tmp), "+c"(dst), "+g"(h)\
            : "S"((x86_reg)dstStride)\
            : "memory"\
        );\
    }else{\
        asm volatile(\
            "1:                         \n\t"\
            "movdqa 16(%0), %%xmm1      \n\t"\
            "movdqa   (%0), %%xmm0      \n\t"\
            "movdqa %%xmm1, %%xmm2      \n\t"\
            "movdqa %%xmm1, %%xmm3      \n\t"\

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"

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"

        "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"

8031:eebc7209c47f

    "movd      "#p",    (%0) \n\t"

static void ff_h264_idct_add_mmx(uint8_t *dst, int16_t *block, int stride)
{
    /* Load dct coeffs */
    __asm__ volatile(
        "movq   (%0), %%mm0 \n\t"
        "movq  8(%0), %%mm1 \n\t"
        "movq 16(%0), %%mm2 \n\t"
        "movq 24(%0), %%mm3 \n\t"
    :: "r"(block) );

    __asm__ volatile(
        /* mm1=s02+s13  mm2=s02-s13  mm4=d02+d13  mm0=d02-d13 */
        IDCT4_1D( %%mm2, %%mm1, %%mm0, %%mm3, %%mm4 )

        "movq      %0,    %%mm6 \n\t"
        /* in: 1,4,0,2  out: 1,2,3,0 */

        IDCT4_1D( %%mm4, %%mm2, %%mm3, %%mm0, %%mm1 )

        "pxor %%mm7, %%mm7    \n\t"
    :: "m"(ff_pw_32));

    __asm__ volatile(
    STORE_DIFF_4P( %%mm0, %%mm1, %%mm7)
        "add %1, %0             \n\t"
    STORE_DIFF_4P( %%mm2, %%mm1, %%mm7)
        "add %1, %0             \n\t"
    STORE_DIFF_4P( %%mm3, %%mm1, %%mm7)

    );
}

static inline void h264_idct8_1d(int16_t *block)
{
    __asm__ volatile(
        "movq 112(%0), %%mm7  \n\t"
        "movq  80(%0), %%mm0  \n\t"
        "movq  48(%0), %%mm3  \n\t"
        "movq  16(%0), %%mm5  \n\t"

    for(i=0; i<2; i++){
        DECLARE_ALIGNED_8(uint64_t, tmp);

        h264_idct8_1d(block+4*i);

        __asm__ volatile(
            "movq   %%mm7,    %0   \n\t"
            TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )
            "movq   %%mm0,  8(%1)  \n\t"
            "movq   %%mm6, 24(%1)  \n\t"
            "movq   %%mm7, 40(%1)  \n\t"

    }

    for(i=0; i<2; i++){
        h264_idct8_1d(b2+4*i);

        __asm__ volatile(
            "psraw     $6, %%mm7  \n\t"
            "psraw     $6, %%mm6  \n\t"
            "psraw     $6, %%mm5  \n\t"
            "psraw     $6, %%mm4  \n\t"
            "psraw     $6, %%mm3  \n\t"

        SUMSUB_BA(h, a)\
        SUMSUB_BA(d, f)

static void ff_h264_idct8_add_sse2(uint8_t *dst, int16_t *block, int stride)
{
    __asm__ volatile(
        "movdqa   0x10(%1), %%xmm1 \n"
        "movdqa   0x20(%1), %%xmm2 \n"
        "movdqa   0x30(%1), %%xmm3 \n"
        "movdqa   0x50(%1), %%xmm5 \n"
        "movdqa   0x60(%1), %%xmm6 \n"

}

static void ff_h264_idct_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
{
    int dc = (block[0] + 32) >> 6;
    __asm__ volatile(
        "movd          %0, %%mm0 \n\t"
        "pshufw $0, %%mm0, %%mm0 \n\t"
        "pxor       %%mm1, %%mm1 \n\t"
        "psubw      %%mm0, %%mm1 \n\t"
        "packuswb   %%mm0, %%mm0 \n\t"
        "packuswb   %%mm1, %%mm1 \n\t"
        ::"r"(dc)
    );
    __asm__ volatile(
        "movd          %0, %%mm2 \n\t"
        "movd          %1, %%mm3 \n\t"
        "movd          %2, %%mm4 \n\t"
        "movd          %3, %%mm5 \n\t"
        "paddusb    %%mm0, %%mm2 \n\t"

static void ff_h264_idct8_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
{
    int dc = (block[0] + 32) >> 6;
    int y;
    __asm__ volatile(
        "movd          %0, %%mm0 \n\t"
        "pshufw $0, %%mm0, %%mm0 \n\t"
        "pxor       %%mm1, %%mm1 \n\t"
        "psubw      %%mm0, %%mm1 \n\t"
        "packuswb   %%mm0, %%mm0 \n\t"
        "packuswb   %%mm1, %%mm1 \n\t"
        ::"r"(dc)
    );
    for(y=2; y--; dst += 4*stride){
    __asm__ volatile(
        "movq          %0, %%mm2 \n\t"
        "movq          %1, %%mm3 \n\t"
        "movq          %2, %%mm4 \n\t"
        "movq          %3, %%mm5 \n\t"
        "paddusb    %%mm0, %%mm2 \n\t"

static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
{
    DECLARE_ALIGNED_8(uint64_t, tmp0[2]);

    __asm__ volatile(
        "movq    (%1,%3), %%mm0    \n\t" //p1
        "movq    (%1,%3,2), %%mm1  \n\t" //p0
        "movq    (%2),    %%mm2    \n\t" //q0
        "movq    (%2,%3), %%mm3    \n\t" //q1
        H264_DEBLOCK_MASK(%6, %7)

    }
}

static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
{
    __asm__ volatile(
        "movq    (%0),    %%mm0     \n\t" //p1
        "movq    (%0,%2), %%mm1     \n\t" //p0
        "movq    (%1),    %%mm2     \n\t" //q0
        "movq    (%1,%2), %%mm3     \n\t" //q1
        H264_DEBLOCK_MASK(%4, %5)

    "psubusb %%mm4, "#p0"  \n\t"\
    "pavgb   "#p1", "#p0"  \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\

static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
{
    __asm__ volatile(
        "movq    (%0),    %%mm0     \n\t"
        "movq    (%0,%2), %%mm1     \n\t"
        "movq    (%1),    %%mm2     \n\t"
        "movq    (%1,%2), %%mm3     \n\t"
        H264_DEBLOCK_MASK(%3, %4)

}

static void h264_loop_filter_strength_mmx2( int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
                                            int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field ) {
    int dir;
    __asm__ volatile(
        "pxor %%mm7, %%mm7 \n\t"
        "movq %0, %%mm6 \n\t"
        "movq %1, %%mm5 \n\t"
        "movq %2, %%mm4 \n\t"
        ::"m"(ff_pb_1), "m"(ff_pb_3), "m"(ff_pb_7)
    );
    if(field)
        __asm__ volatile(
            "movq %0, %%mm5 \n\t"
            "movq %1, %%mm4 \n\t"
            ::"m"(ff_pb_3_1), "m"(ff_pb_7_3)
        );

        const int d_idx = dir ? -8 : -1;
        const int mask_mv = dir ? mask_mv1 : mask_mv0;
        DECLARE_ALIGNED_8(const uint64_t, mask_dir) = dir ? 0 : 0xffffffffffffffffULL;
        int b_idx, edge, l;
        for( b_idx=12, edge=0; edge<edges; edge+=step, b_idx+=8*step ) {
            __asm__ volatile(
                "pand %0, %%mm0 \n\t"
                ::"m"(mask_dir)
            );
            if(!(mask_mv & edge)) {
                __asm__ volatile("pxor %%mm0, %%mm0 \n\t":);
                for( l = bidir; l >= 0; l-- ) {
                    __asm__ volatile(
                        "movd %0, %%mm1 \n\t"
                        "punpckldq %1, %%mm1 \n\t"
                        "movq %%mm1, %%mm2 \n\t"
                        "psrlw $7, %%mm2 \n\t"
                        "pand %%mm6, %%mm2 \n\t"

                          "m"(mv[l][b_idx+d_idx][0]),
                          "m"(mv[l][b_idx+d_idx+2][0])
                    );
                }
            }
            __asm__ volatile(
                "movd %0, %%mm1 \n\t"
                "por  %1, %%mm1 \n\t"
                "punpcklbw %%mm7, %%mm1 \n\t"
                "pcmpgtw %%mm7, %%mm1 \n\t" // nnz[b] || nnz[bn]
                ::"m"(nnz[b_idx]),
                  "m"(nnz[b_idx+d_idx])
            );
            __asm__ volatile(
                "pcmpeqw %%mm7, %%mm0 \n\t"
                "pcmpeqw %%mm7, %%mm0 \n\t"
                "psrlw $15, %%mm0 \n\t" // nonzero -> 1
                "psrlw $14, %%mm1 \n\t"
                "movq %%mm0, %%mm2 \n\t"

            );
        }
        edges = 4;
        step = 1;
    }
    __asm__ volatile(
        "movq   (%0), %%mm0 \n\t"
        "movq  8(%0), %%mm1 \n\t"
        "movq 16(%0), %%mm2 \n\t"
        "movq 24(%0), %%mm3 \n\t"
        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4)

#define QPEL_H264(OPNAME, OP, MMX)\
static av_noinline void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    int h=4;\
\
    __asm__ volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
        "movq %5, %%mm4             \n\t"\
        "movq %6, %%mm5             \n\t"\
        "1:                         \n\t"\
        "movd  -1(%0), %%mm1        \n\t"\

        : "memory"\
    );\
}\
static av_noinline void OPNAME ## h264_qpel4_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
    int h=4;\
    __asm__ volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
        "movq %0, %%mm4             \n\t"\
        "movq %1, %%mm5             \n\t"\
        :: "m"(ff_pw_5), "m"(ff_pw_16)\
    );\
    do{\
    __asm__ volatile(\
        "movd  -1(%0), %%mm1        \n\t"\
        "movd    (%0), %%mm2        \n\t"\
        "movd   1(%0), %%mm3        \n\t"\
        "movd   2(%0), %%mm0        \n\t"\
        "punpcklbw %%mm7, %%mm1     \n\t"\

    );\
    }while(--h);\
}\
static av_noinline void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    src -= 2*srcStride;\
    __asm__ volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
        "movd (%0), %%mm0           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm1           \n\t"\
        "add %2, %0                 \n\t"\

static av_noinline void OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
    int h=4;\
    int w=3;\
    src -= 2*srcStride+2;\
    while(w--){\
        __asm__ volatile(\
            "pxor %%mm7, %%mm7      \n\t"\
            "movd (%0), %%mm0       \n\t"\
            "add %2, %0             \n\t"\
            "movd (%0), %%mm1       \n\t"\
            "add %2, %0             \n\t"\

        );\
        tmp += 4;\
        src += 4 - 9*srcStride;\
    }\
    tmp -= 3*4;\
    __asm__ volatile(\
        "1:                         \n\t"\
        "movq     (%0), %%mm0       \n\t"\
        "paddw  10(%0), %%mm0       \n\t"\
        "movq    2(%0), %%mm1       \n\t"\
        "paddw   8(%0), %%mm1       \n\t"\

    );\
}\
\
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    int h=8;\
    __asm__ volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
        "movq %5, %%mm6             \n\t"\
        "1:                         \n\t"\
        "movq    (%0), %%mm0        \n\t"\
        "movq   1(%0), %%mm2        \n\t"\

    );\
}\
\
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
    int h=8;\
    __asm__ volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
        "movq %0, %%mm6             \n\t"\
        :: "m"(ff_pw_5)\
    );\
    do{\
    __asm__ volatile(\
        "movq    (%0), %%mm0        \n\t"\
        "movq   1(%0), %%mm2        \n\t"\
        "movq %%mm0, %%mm1          \n\t"\
        "movq %%mm2, %%mm3          \n\t"\
        "punpcklbw %%mm7, %%mm0     \n\t"\

static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
    int w= 2;\
    src -= 2*srcStride;\
    \
    while(w--){\
      __asm__ volatile(\
        "pxor %%mm7, %%mm7          \n\t"\
        "movd (%0), %%mm0           \n\t"\
        "add %2, %0                 \n\t"\
        "movd (%0), %%mm1           \n\t"\
        "add %2, %0                 \n\t"\

        : "+a"(src), "+c"(dst)\
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
        : "memory"\
     );\
     if(h==16){\
        __asm__ volatile(\
            QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\
            QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\
            QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\
            QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\
            QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\

}\
static av_always_inline void OPNAME ## h264_qpel8or16_hv1_lowpass_ ## MMX(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){\
    int w = (size+8)>>2;\
    src -= 2*srcStride+2;\
    while(w--){\
        __asm__ volatile(\
            "pxor %%mm7, %%mm7      \n\t"\
            "movd (%0), %%mm0       \n\t"\
            "add %2, %0             \n\t"\
            "movd (%0), %%mm1       \n\t"\
            "add %2, %0             \n\t"\

            : "+a"(src)\
            : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)\
            : "memory"\
        );\
        if(size==16){\
            __asm__ volatile(\
                QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1,  8*48)\
                QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2,  9*48)\
                QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\
                QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\
                QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\

}\
static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
    int w = size>>4;\
    do{\
    int h = size;\
    __asm__ volatile(\
        "1:                         \n\t"\
        "movq     (%0), %%mm0       \n\t"\
        "movq    8(%0), %%mm3       \n\t"\
        "movq    2(%0), %%mm1       \n\t"\
        "movq   10(%0), %%mm4       \n\t"\

    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      (%1), %%mm0          \n\t"\
        "movq    24(%1), %%mm1          \n\t"\
        "psraw      $5,  %%mm0          \n\t"\
        "psraw      $5,  %%mm1          \n\t"\
        "packuswb %%mm0, %%mm0          \n\t"\

        :"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)\
{\
    do{\
    __asm__ volatile(\
        "movq      (%1), %%mm0          \n\t"\
        "movq     8(%1), %%mm1          \n\t"\
        "movq    48(%1), %%mm2          \n\t"\
        "movq  8+48(%1), %%mm3          \n\t"\
        "psraw      $5,  %%mm0          \n\t"\

#ifdef ARCH_X86_64
#define QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
    int h=16;\
    __asm__ volatile(\
        "pxor %%xmm15, %%xmm15      \n\t"\
        "movdqa %6, %%xmm14         \n\t"\
        "movdqa %7, %%xmm13         \n\t"\
        "1:                         \n\t"\
        "lddqu    3(%0), %%xmm1     \n\t"\

#endif // ARCH_X86_64

#define QPEL_H264_H_XMM(OPNAME, OP, MMX)\
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\
    int h=8;\
    __asm__ volatile(\
        "pxor %%xmm7, %%xmm7        \n\t"\
        "movdqa %0, %%xmm6          \n\t"\
        :: "m"(ff_pw_5)\
    );\
    do{\
    __asm__ volatile(\
        "lddqu   -5(%0), %%xmm1     \n\t"\
        "movdqa  %%xmm1, %%xmm0     \n\t"\
        "punpckhbw %%xmm7, %%xmm1   \n\t"\
        "punpcklbw %%xmm7, %%xmm0   \n\t"\
        "movdqa  %%xmm1, %%xmm2     \n\t"\

}\
QPEL_H264_H16_XMM(OPNAME, OP, MMX)\
\
static av_noinline void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    int h=8;\
    __asm__ volatile(\
        "pxor %%xmm7, %%xmm7        \n\t"\
        "movdqa %5, %%xmm6          \n\t"\
        "1:                         \n\t"\
        "lddqu   -5(%0), %%xmm1     \n\t"\
        "movdqa  %%xmm1, %%xmm0     \n\t"\

#define QPEL_H264_V_XMM(OPNAME, OP, MMX)\
static av_noinline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
    src -= 2*srcStride;\
    \
    __asm__ volatile(\
        "pxor %%xmm7, %%xmm7        \n\t"\
        "movq (%0), %%xmm0          \n\t"\
        "add %2, %0                 \n\t"\
        "movq (%0), %%xmm1          \n\t"\
        "add %2, %0                 \n\t"\

        : "+a"(src), "+c"(dst)\
        : "S"((x86_reg)srcStride), "D"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\
        : "memory"\
    );\
    if(h==16){\
        __asm__ volatile(\
            QPEL_H264V_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, OP)\
            QPEL_H264V_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, OP)\
            QPEL_H264V_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, OP)\
            QPEL_H264V_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, OP)\
            QPEL_H264V_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, OP)\

static av_always_inline void put_h264_qpel8or16_hv1_lowpass_sse2(int16_t *tmp, uint8_t *src, int tmpStride, int srcStride, int size){
    int w = (size+8)>>3;
    src -= 2*srcStride+2;
    while(w--){
        __asm__ volatile(
            "pxor %%xmm7, %%xmm7        \n\t"
            "movq (%0), %%xmm0          \n\t"
            "add %2, %0                 \n\t"
            "movq (%0), %%xmm1          \n\t"
            "add %2, %0                 \n\t"

            : "+a"(src)
            : "c"(tmp), "S"((x86_reg)srcStride), "m"(ff_pw_5), "m"(ff_pw_16)
            : "memory"
        );
        if(size==16){
            __asm__ volatile(
                QPEL_H264HV_XMM(%%xmm2, %%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1,  8*48)
                QPEL_H264HV_XMM(%%xmm3, %%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2,  9*48)
                QPEL_H264HV_XMM(%%xmm4, %%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, 10*48)
                QPEL_H264HV_XMM(%%xmm5, %%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, 11*48)
                QPEL_H264HV_XMM(%%xmm0, %%xmm1, %%xmm2, %%xmm3, %%xmm4, %%xmm5, 12*48)

#define QPEL_H264_HV2_XMM(OPNAME, OP, MMX)\
static av_always_inline void OPNAME ## h264_qpel8or16_hv2_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, int dstStride, int tmpStride, int size){\
    int h = size;\
    if(size == 16){\
        __asm__ volatile(\
            "1:                         \n\t"\
            "movdqa 32(%0), %%xmm4      \n\t"\
            "movdqa 16(%0), %%xmm5      \n\t"\
            "movdqa   (%0), %%xmm7      \n\t"\
            "movdqa %%xmm4, %%xmm3      \n\t"\

            : "+a"(tmp), "+c"(dst), "+g"(h)\
            : "S"((x86_reg)dstStride)\
            : "memory"\
        );\
    }else{\
        __asm__ volatile(\
            "1:                         \n\t"\
            "movdqa 16(%0), %%xmm1      \n\t"\
            "movdqa   (%0), %%xmm0      \n\t"\
            "movdqa %%xmm1, %%xmm2      \n\t"\
            "movdqa %%xmm1, %%xmm3      \n\t"\

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"

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"

        "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"