qpel.c

ffmpeg源码分析

    }
}

static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
{
    int i;
    for(i=0; i<h; i++)
    {
        ST32(dst   , LD32(src   ));
        ST32(dst+4 , LD32(src+4 ));
        dst[8]= src[8];
        dst+=dstStride;
        src+=srcStride;
    }
}
/* end not optimized */

#define QPEL_MC(r, OPNAME, RND, OP) \
static void OPNAME ## mpeg4_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
    uint8_t *cm = cropTbl + MAX_NEG_CROP;\
    do {\
        uint8_t *s = src; \
        int src0,src1,src2,src3,src4,src5,src6,src7,src8;\
        src0= *s++;\
        src1= *s++;\
        src2= *s++;\
        src3= *s++;\
        src4= *s++;\
        OP(dst[0], (src0+src1)*20 - (src0+src2)*6 + (src1+src3)*3 - (src2+src4));\
        src5= *s++;\
        OP(dst[1], (src1+src2)*20 - (src0+src3)*6 + (src0+src4)*3 - (src1+src5));\
        src6= *s++;\
        OP(dst[2], (src2+src3)*20 - (src1+src4)*6 + (src0+src5)*3 - (src0+src6));\
        src7= *s++;\
        OP(dst[3], (src3+src4)*20 - (src2+src5)*6 + (src1+src6)*3 - (src0+src7));\
        src8= *s++;\
        OP(dst[4], (src4+src5)*20 - (src3+src6)*6 + (src2+src7)*3 - (src1+src8));\
        OP(dst[5], (src5+src6)*20 - (src4+src7)*6 + (src3+src8)*3 - (src2+src8));\
        OP(dst[6], (src6+src7)*20 - (src5+src8)*6 + (src4+src8)*3 - (src3+src7));\
        OP(dst[7], (src7+src8)*20 - (src6+src8)*6 + (src5+src7)*3 - (src4+src6));\
        dst+=dstStride;\
        src+=srcStride;\
    }while(--h);\
}\
\
static void OPNAME ## mpeg4_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    uint8_t *cm = cropTbl + MAX_NEG_CROP;\
    int w=8;\
    do{\
        uint8_t *s = src, *d=dst;\
        int src0,src1,src2,src3,src4,src5,src6,src7,src8;\
        src0 = *s; s+=srcStride; \
        src1 = *s; s+=srcStride; \
        src2 = *s; s+=srcStride; \
        src3 = *s; s+=srcStride; \
        src4 = *s; s+=srcStride; \
        OP(*d, (src0+src1)*20 - (src0+src2)*6 + (src1+src3)*3 - (src2+src4));d+=dstStride;\
        src5 = *s; s+=srcStride; \
        OP(*d, (src1+src2)*20 - (src0+src3)*6 + (src0+src4)*3 - (src1+src5));d+=dstStride;\
        src6 = *s; s+=srcStride; \
        OP(*d, (src2+src3)*20 - (src1+src4)*6 + (src0+src5)*3 - (src0+src6));d+=dstStride;\
        src7 = *s; s+=srcStride; \
        OP(*d, (src3+src4)*20 - (src2+src5)*6 + (src1+src6)*3 - (src0+src7));d+=dstStride;\
        src8 = *s; \
        OP(*d, (src4+src5)*20 - (src3+src6)*6 + (src2+src7)*3 - (src1+src8));d+=dstStride;\
        OP(*d, (src5+src6)*20 - (src4+src7)*6 + (src3+src8)*3 - (src2+src8));d+=dstStride;\
        OP(*d, (src6+src7)*20 - (src5+src8)*6 + (src4+src8)*3 - (src3+src7));d+=dstStride;\
        OP(*d, (src7+src8)*20 - (src6+src8)*6 + (src5+src7)*3 - (src4+src6));\
        dst++;\
        src++;\
    }while(--w);\
}\
\
static void OPNAME ## mpeg4_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
    uint8_t *cm = cropTbl + MAX_NEG_CROP;\
    do {\
        uint8_t *s = src;\
        int src0,src1,src2,src3,src4,src5,src6,src7,src8;\
        int src9,src10,src11,src12,src13,src14,src15,src16;\
        src0= *s++;\
        src1= *s++;\
        src2= *s++;\
        src3= *s++;\
        src4= *s++;\
        OP(dst[ 0], (src0 +src1 )*20 - (src0 +src2 )*6 + (src1 +src3 )*3 - (src2 +src4 ));\
        src5= *s++;\
        OP(dst[ 1], (src1 +src2 )*20 - (src0 +src3 )*6 + (src0 +src4 )*3 - (src1 +src5 ));\
        src6= *s++;\
        OP(dst[ 2], (src2 +src3 )*20 - (src1 +src4 )*6 + (src0 +src5 )*3 - (src0 +src6 ));\
        src7= *s++;\
        OP(dst[ 3], (src3 +src4 )*20 - (src2 +src5 )*6 + (src1 +src6 )*3 - (src0 +src7 ));\
        src8= *s++;\
        OP(dst[ 4], (src4 +src5 )*20 - (src3 +src6 )*6 + (src2 +src7 )*3 - (src1 +src8 ));\
        src9= *s++;\
        OP(dst[ 5], (src5 +src6 )*20 - (src4 +src7 )*6 + (src3 +src8 )*3 - (src2 +src9 ));\
        src10= *s++;\
        OP(dst[ 6], (src6 +src7 )*20 - (src5 +src8 )*6 + (src4 +src9 )*3 - (src3 +src10));\
        src11= *s++;\
        OP(dst[ 7], (src7 +src8 )*20 - (src6 +src9 )*6 + (src5 +src10)*3 - (src4 +src11));\
        src12= *s++;\
        OP(dst[ 8], (src8 +src9 )*20 - (src7 +src10)*6 + (src6 +src11)*3 - (src5 +src12));\
        src13= *s++;\
        OP(dst[ 9], (src9 +src10)*20 - (src8 +src11)*6 + (src7 +src12)*3 - (src6 +src13));\
        src14= *s++;\
        OP(dst[10], (src10+src11)*20 - (src9 +src12)*6 + (src8 +src13)*3 - (src7 +src14));\
        src15= *s++;\
        OP(dst[11], (src11+src12)*20 - (src10+src13)*6 + (src9 +src14)*3 - (src8 +src15));\
        src16= *s++;\
        OP(dst[12], (src12+src13)*20 - (src11+src14)*6 + (src10+src15)*3 - (src9 +src16));\
        OP(dst[13], (src13+src14)*20 - (src12+src15)*6 + (src11+src16)*3 - (src10+src16));\
        OP(dst[14], (src14+src15)*20 - (src13+src16)*6 + (src12+src16)*3 - (src11+src15));\
        OP(dst[15], (src15+src16)*20 - (src14+src16)*6 + (src13+src15)*3 - (src12+src14));\
        dst+=dstStride;\
        src+=srcStride;\
    }while(--h);\
}\
\
static void OPNAME ## mpeg4_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    uint8_t *cm = cropTbl + MAX_NEG_CROP;\
    int w=16;\
    do {\
        uint8_t *s = src, *d=dst;\
        int src0,src1,src2,src3,src4,src5,src6,src7,src8;\
        int src9,src10,src11,src12,src13,src14,src15,src16;\
        src0 = *s; s+=srcStride; \
        src1 = *s; s+=srcStride; \
        src2 = *s; s+=srcStride; \
        src3 = *s; s+=srcStride; \
        src4 = *s; s+=srcStride; \
        OP(*d, (src0 +src1 )*20 - (src0 +src2 )*6 + (src1 +src3 )*3 - (src2 +src4 ));d+=dstStride;\
        src5 = *s; s+=srcStride; \
        OP(*d, (src1 +src2 )*20 - (src0 +src3 )*6 + (src0 +src4 )*3 - (src1 +src5 ));d+=dstStride;\
        src6 = *s; s+=srcStride; \
        OP(*d, (src2 +src3 )*20 - (src1 +src4 )*6 + (src0 +src5 )*3 - (src0 +src6 ));d+=dstStride;\
        src7 = *s; s+=srcStride; \
        OP(*d, (src3 +src4 )*20 - (src2 +src5 )*6 + (src1 +src6 )*3 - (src0 +src7 ));d+=dstStride;\
        src8 = *s; s+=srcStride; \
        OP(*d, (src4 +src5 )*20 - (src3 +src6 )*6 + (src2 +src7 )*3 - (src1 +src8 ));d+=dstStride;\
        src9 = *s; s+=srcStride; \
        OP(*d, (src5 +src6 )*20 - (src4 +src7 )*6 + (src3 +src8 )*3 - (src2 +src9 ));d+=dstStride;\
        src10 = *s; s+=srcStride; \
        OP(*d, (src6 +src7 )*20 - (src5 +src8 )*6 + (src4 +src9 )*3 - (src3 +src10));d+=dstStride;\
        src11 = *s; s+=srcStride; \
        OP(*d, (src7 +src8 )*20 - (src6 +src9 )*6 + (src5 +src10)*3 - (src4 +src11));d+=dstStride;\
        src12 = *s; s+=srcStride; \
        OP(*d, (src8 +src9 )*20 - (src7 +src10)*6 + (src6 +src11)*3 - (src5 +src12));d+=dstStride;\
        src13 = *s; s+=srcStride; \
        OP(*d, (src9 +src10)*20 - (src8 +src11)*6 + (src7 +src12)*3 - (src6 +src13));d+=dstStride;\
        src14 = *s; s+=srcStride; \
        OP(*d, (src10+src11)*20 - (src9 +src12)*6 + (src8 +src13)*3 - (src7 +src14));d+=dstStride;\
        src15 = *s; s+=srcStride; \
        OP(*d, (src11+src12)*20 - (src10+src13)*6 + (src9 +src14)*3 - (src8 +src15));d+=dstStride;\
        src16 = *s; \
        OP(*d, (src12+src13)*20 - (src11+src14)*6 + (src10+src15)*3 - (src9 +src16));d+=dstStride;\
        OP(*d, (src13+src14)*20 - (src12+src15)*6 + (src11+src16)*3 - (src10+src16));d+=dstStride;\
        OP(*d, (src14+src15)*20 - (src13+src16)*6 + (src12+src16)*3 - (src11+src15));d+=dstStride;\
        OP(*d, (src15+src16)*20 - (src14+src16)*6 + (src13+src15)*3 - (src12+src14));\
        dst++;\
        src++;\
    }while(--w);\
}\
\
static void OPNAME ## qpel8_mc00_c (uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## pixels8_c(dst, src, stride, 8);\
}\
\
static void OPNAME ## qpel8_mc10_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t half[64];\
    put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\
    OPNAME ## pixels8_l2_aligned2(dst, src, half, stride, stride, 8, 8);\
}\
\
static void OPNAME ## qpel8_mc20_c(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## mpeg4_qpel8_h_lowpass(dst, src, stride, stride, 8);\
}\
\
static void OPNAME ## qpel8_mc30_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t half[64];\
    put ## RND ## mpeg4_qpel8_h_lowpass(half, src, 8, stride, 8);\
    OPNAME ## pixels8_l2_aligned2(dst, src+1, half, stride, stride, 8, 8);\
}\
\
static void OPNAME ## qpel8_mc01_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[16*9];\
    uint8_t half[64];\
    copy_block9(full, src, 16, stride, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\
    OPNAME ## pixels8_l2_aligned(dst, full, half, stride, 16, 8, 8);\
}\
\
static void OPNAME ## qpel8_mc02_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[16*9];\
    copy_block9(full, src, 16, stride, 9);\
    OPNAME ## mpeg4_qpel8_v_lowpass(dst, full, stride, 16);\
}\
\
static void OPNAME ## qpel8_mc03_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[16*9];\
    uint8_t half[64];\
    copy_block9(full, src, 16, stride, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\
    OPNAME ## pixels8_l2_aligned(dst, full+16, half, stride, 16, 8, 8);\
}\
static void ff_ ## OPNAME ## qpel8_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[16*9];\
    uint8_t halfH[72];\
    uint8_t halfV[64];\
    uint8_t halfHV[64];\
    copy_block9(full, src, 16, stride, 9);\
    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\
    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
    OPNAME ## pixels8_l4_aligned(dst, full, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\
}\
static void OPNAME ## qpel8_mc11_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[16*9];\
    uint8_t halfH[72];\
    uint8_t halfHV[64];\
    copy_block9(full, src, 16, stride, 9);\
    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
    put ## RND ## pixels8_l2_aligned(halfH, halfH, full, 8, 8, 16, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
    OPNAME ## pixels8_l2_aligned(dst, halfH, halfHV, stride, 8, 8, 8);\
}\
static void ff_ ## OPNAME ## qpel8_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[16*9];\
    uint8_t halfH[72];\
    uint8_t halfV[64];\
    uint8_t halfHV[64];\
    copy_block9(full, src, 16, stride, 9);\
    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\
    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
    OPNAME ## pixels8_l4_aligned0(dst, full+1, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\
}\
static void OPNAME ## qpel8_mc31_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[16*9];\
    uint8_t halfH[72];\
    uint8_t halfHV[64];\
    copy_block9(full, src, 16, stride, 9);\
    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
    put ## RND ## pixels8_l2_aligned1(halfH, halfH, full+1, 8, 8, 16, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
    OPNAME ## pixels8_l2_aligned(dst, halfH, halfHV, stride, 8, 8, 8);\
}\
static void ff_ ## OPNAME ## qpel8_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[16*9];\
    uint8_t halfH[72];\
    uint8_t halfV[64];\
    uint8_t halfHV[64];\
    copy_block9(full, src, 16, stride, 9);\
    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\
    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
    OPNAME ## pixels8_l4_aligned(dst, full+16, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\
}\
static void OPNAME ## qpel8_mc13_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[16*9];\
    uint8_t halfH[72];\
    uint8_t halfHV[64];\
    copy_block9(full, src, 16, stride, 9);\
    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
    put ## RND ## pixels8_l2_aligned(halfH, halfH, full, 8, 8, 16, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
    OPNAME ## pixels8_l2_aligned(dst, halfH+8, halfHV, stride, 8, 8, 8);\
}\
static void ff_ ## OPNAME ## qpel8_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[16*9];\
    uint8_t halfH[72];\
    uint8_t halfV[64];\
    uint8_t halfHV[64];\
    copy_block9(full, src, 16, stride, 9);\
    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full  , 8, 16, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\
    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
    OPNAME ## pixels8_l4_aligned0(dst, full+17, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\
}\
static void OPNAME ## qpel8_mc33_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[16*9];\
    uint8_t halfH[72];\
    uint8_t halfHV[64];\
    copy_block9(full, src, 16, stride, 9);\
    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
    put ## RND ## pixels8_l2_aligned1(halfH, halfH, full+1, 8, 8, 16, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
    OPNAME ## pixels8_l2_aligned(dst, halfH+8, halfHV, stride, 8, 8, 8);\
}\
static void OPNAME ## qpel8_mc21_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t halfH[72];\
    uint8_t halfHV[64];\
    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
    OPNAME ## pixels8_l2_aligned(dst, halfH, halfHV, stride, 8, 8, 8);\
}\
static void OPNAME ## qpel8_mc23_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t halfH[72];\
    uint8_t halfHV[64];\
    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\
    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\