comparison i386/h264dsp_mmx.c @ 5947:37a03989871b libavcodec

use ff_ prefix for extern vars

5946:55251379b5b1

        "psubusb %%mm6              , %%mm1 \n\t"\
        "psubusb %%mm3              , %%mm2 \n\t"\
        "paddusb %%mm3              , %%mm1 \n\t"\
        "paddusb %%mm6              , %%mm2 \n\t"

// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=mm_bone
// out: (q1addr) = av_clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
// clobbers: q2, tmp, tc0
#define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\
        "movq     %%mm1,  "#tmp"   \n\t"\
        "pavgb    %%mm2,  "#tmp"   \n\t"\

        "movq      %%mm2, (%2)      \n\t"

        : "=m"(*tmp0)
        : "r"(pix-3*stride), "r"(pix), "r"((long)stride),
          "m"(*tmp0/*unused*/), "m"(*(uint32_t*)tc0), "m"(alpha1), "m"(beta1),
          "m"(mm_bone)
    );
}

static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
{

        "movq      %%mm1, (%0,%2)   \n\t"
        "movq      %%mm2, (%1)      \n\t"

        :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
           "r"(*(uint32_t*)tc0),
           "m"(alpha1), "m"(beta1), "m"(mm_bone), "m"(ff_pb_3F)
    );
}

static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
{

        "paddb   %%mm5,   %%mm1     \n\t"
        "paddb   %%mm6,   %%mm2     \n\t"
        "movq    %%mm1,   (%0,%2)   \n\t"
        "movq    %%mm2,   (%1)      \n\t"
        :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
           "m"(alpha1), "m"(beta1), "m"(mm_bone)
    );
}

static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
{

5947:37a03989871b

        "psubusb %%mm6              , %%mm1 \n\t"\
        "psubusb %%mm3              , %%mm2 \n\t"\
        "paddusb %%mm3              , %%mm1 \n\t"\
        "paddusb %%mm6              , %%mm2 \n\t"

// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=ff_bone
// out: (q1addr) = av_clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
// clobbers: q2, tmp, tc0
#define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\
        "movq     %%mm1,  "#tmp"   \n\t"\
        "pavgb    %%mm2,  "#tmp"   \n\t"\

        "movq      %%mm2, (%2)      \n\t"

        : "=m"(*tmp0)
        : "r"(pix-3*stride), "r"(pix), "r"((long)stride),
          "m"(*tmp0/*unused*/), "m"(*(uint32_t*)tc0), "m"(alpha1), "m"(beta1),
          "m"(ff_bone)
    );
}

static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
{

        "movq      %%mm1, (%0,%2)   \n\t"
        "movq      %%mm2, (%1)      \n\t"

        :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
           "r"(*(uint32_t*)tc0),
           "m"(alpha1), "m"(beta1), "m"(ff_bone), "m"(ff_pb_3F)
    );
}

static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
{

        "paddb   %%mm5,   %%mm1     \n\t"
        "paddb   %%mm6,   %%mm2     \n\t"
        "movq    %%mm1,   (%0,%2)   \n\t"
        "movq    %%mm2,   (%1)      \n\t"
        :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
           "m"(alpha1), "m"(beta1), "m"(ff_bone)
    );
}

static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
{