qpel.c

这是著名的TCPMP播放器在WINDWOWS,和WINCE下编译通过的源程序.笔者对其中的LIBMAD库做了针对ARM MPU的优化. 并增加了词幕功能.

static void ff_ ## OPNAME ## qpel8_mc32_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_l2_aligned(dst, halfV, halfHV, stride, 8, 8, 8);\
}\
static void OPNAME ## qpel8_mc32_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[16*9];\
    uint8_t halfH[72];\
    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);\
    OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\
}\
static void OPNAME ## qpel8_mc22_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t halfH[72];\
    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\
    OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\
}\
static void OPNAME ## qpel16_mc00_c (uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## pixels16_c(dst, src, stride, 16);\
}\
\
static void OPNAME ## qpel16_mc10_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t half[256];\
    put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\
    OPNAME ## pixels16_l2_aligned2(dst, src, half, stride, stride, 16, 16);\
}\
\
static void OPNAME ## qpel16_mc20_c(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## mpeg4_qpel16_h_lowpass(dst, src, stride, stride, 16);\
}\
\
static void OPNAME ## qpel16_mc30_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t half[256];\
    put ## RND ## mpeg4_qpel16_h_lowpass(half, src, 16, stride, 16);\
    OPNAME ## pixels16_l2_aligned2(dst, src+1, half, stride, stride, 16, 16);\
}\
\
static void OPNAME ## qpel16_mc01_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[24*17];\
    uint8_t half[256];\
    copy_block17(full, src, 24, stride, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\
    OPNAME ## pixels16_l2_aligned(dst, full, half, stride, 24, 16, 16);\
}\
\
static void OPNAME ## qpel16_mc02_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[24*17];\
    copy_block17(full, src, 24, stride, 17);\
    OPNAME ## mpeg4_qpel16_v_lowpass(dst, full, stride, 24);\
}\
\
static void OPNAME ## qpel16_mc03_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[24*17];\
    uint8_t half[256];\
    copy_block17(full, src, 24, stride, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\
    OPNAME ## pixels16_l2_aligned(dst, full+24, half, stride, 24, 16, 16);\
}\
static void ff_ ## OPNAME ## qpel16_mc11_old_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[24*17];\
    uint8_t halfH[272];\
    uint8_t halfV[256];\
    uint8_t halfHV[256];\
    copy_block17(full, src, 24, stride, 17);\
    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l4_aligned(dst, full, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc11_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[24*17];\
    uint8_t halfH[272];\
    uint8_t halfHV[256];\
    copy_block17(full, src, 24, stride, 17);\
    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
    put ## RND ## pixels16_l2_aligned(halfH, halfH, full, 16, 16, 24, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l2_aligned(dst, halfH, halfHV, stride, 16, 16, 16);\
}\
static void ff_ ## OPNAME ## qpel16_mc31_old_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[24*17];\
    uint8_t halfH[272];\
    uint8_t halfV[256];\
    uint8_t halfHV[256];\
    copy_block17(full, src, 24, stride, 17);\
    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l4_aligned0(dst, full+1, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc31_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[24*17];\
    uint8_t halfH[272];\
    uint8_t halfHV[256];\
    copy_block17(full, src, 24, stride, 17);\
    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
    put ## RND ## pixels16_l2_aligned1(halfH, halfH, full+1, 16, 16, 24, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l2_aligned(dst, halfH, halfHV, stride, 16, 16, 16);\
}\
static void ff_ ## OPNAME ## qpel16_mc13_old_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[24*17];\
    uint8_t halfH[272];\
    uint8_t halfV[256];\
    uint8_t halfHV[256];\
    copy_block17(full, src, 24, stride, 17);\
    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l4_aligned(dst, full+24, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc13_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[24*17];\
    uint8_t halfH[272];\
    uint8_t halfHV[256];\
    copy_block17(full, src, 24, stride, 17);\
    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
    put ## RND ## pixels16_l2_aligned(halfH, halfH, full, 16, 16, 24, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l2_aligned(dst, halfH+16, halfHV, stride, 16, 16, 16);\
}\
static void ff_ ## OPNAME ## qpel16_mc33_old_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[24*17];\
    uint8_t halfH[272];\
    uint8_t halfV[256];\
    uint8_t halfHV[256];\
    copy_block17(full, src, 24, stride, 17);\
    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full  , 16, 24, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l4_aligned0(dst, full+25, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc33_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[24*17];\
    uint8_t halfH[272];\
    uint8_t halfHV[256];\
    copy_block17(full, src, 24, stride, 17);\
    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
    put ## RND ## pixels16_l2_aligned1(halfH, halfH, full+1, 16, 16, 24, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l2_aligned(dst, halfH+16, halfHV, stride, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc21_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t halfH[272];\
    uint8_t halfHV[256];\
    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l2_aligned(dst, halfH, halfHV, stride, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc23_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t halfH[272];\
    uint8_t halfHV[256];\
    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l2_aligned(dst, halfH+16, halfHV, stride, 16, 16, 16);\
}\
static void ff_ ## OPNAME ## qpel16_mc12_old_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[24*17];\
    uint8_t halfH[272];\
    uint8_t halfV[256];\
    uint8_t halfHV[256];\
    copy_block17(full, src, 24, stride, 17);\
    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full, 16, 24);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l2_aligned(dst, halfV, halfHV, stride, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc12_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[24*17];\
    uint8_t halfH[272];\
    copy_block17(full, src, 24, stride, 17);\
    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
    put ## RND ## pixels16_l2_aligned(halfH, halfH, full, 16, 16, 24, 17);\
    OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\
}\
static void ff_ ## OPNAME ## qpel16_mc32_old_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[24*17];\
    uint8_t halfH[272];\
    uint8_t halfV[256];\
    uint8_t halfHV[256];\
    copy_block17(full, src, 24, stride, 17);\
    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfV, full+1, 16, 24);\
    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
    OPNAME ## pixels16_l2_aligned(dst, halfV, halfHV, stride, 16, 16, 16);\
}\
static void OPNAME ## qpel16_mc32_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[24*17];\
    uint8_t halfH[272];\
    copy_block17(full, src, 24, stride, 17);\
    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
    put ## RND ## pixels16_l2_aligned1(halfH, halfH, full+1, 16, 16, 24, 17);\
    OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\
}\
static void OPNAME ## qpel16_mc22_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t halfH[272];\
    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\
    OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\
}

#define op_avg(a, b) a = (((a)+cm[((b) + 16)>>5]+1)>>1)
#define op_avg_no_rnd(a, b) a = (((a)+cm[((b) + 15)>>5])>>1)
#define op_put(a, b) a = cm[((b) + 16)>>5]
#define op_put_no_rnd(a, b) a = cm[((b) + 15)>>5]

QPEL_MC(0, put_       , _       , op_put)
QPEL_MC(1, put_no_rnd_, _no_rnd_, op_put_no_rnd)
QPEL_MC(0, avg_       , _       , op_avg)
//QPEL_MC(1, avg_no_rnd , _       , op_avg)
#undef op_avg
#undef op_avg_no_rnd
#undef op_put
#undef op_put_no_rnd

#if 1
#define H264_LOWPASS(OPNAME, OP, OP2) \
static inline void OPNAME ## h264_qpel_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride,int w,int h){\
    uint8_t *cm = cropTbl + MAX_NEG_CROP;\
    do {\
        int srcB,srcA,src0,src1,src2,src3,src4,src5,src6;\
        uint8_t *s = src-2;\
        srcB = *s++;\
        srcA = *s++;\
        src0 = *s++;\
        src1 = *s++;\
        src2 = *s++;\
        src3 = *s++;\
        OP(dst[0], (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\
        src4 = *s++;\
        OP(dst[1], (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\
        src5 = *s++;\
        OP(dst[2], (src2+src3)*20 - (src1+src4)*5 + (src0+src5));\
        src6 = *s++;\
        OP(dst[3], (src3+src4)*20 - (src2+src5)*5 + (src1+src6));\
      if (w>4) { /* it optimized */ \
        int src7,src8,src9,src10; \
        src7 = *s++;\
        OP(dst[4], (src4+src5)*20 - (src3+src6)*5 + (src2+src7));\
        src8 = *s++;\
        OP(dst[5], (src5+src6)*20 - (src4+src7)*5 + (src3+src8));\
        src9 = *s++;\
        OP(dst[6], (src6+src7)*20 - (src5+src8)*5 + (src4+src9));\
        src10 = *s++;\
        OP(dst[7], (src7+src8)*20 - (src6+src9)*5 + (src5+src10));\
       if (w>8) { \
        int src11,src12,src13,src14,src15,src16,src17,src18; \
        src11 = *s++;\
        OP(dst[8] , (src8 +src9 )*20 - (src7 +src10)*5 + (src6 +src11));\
        src12 = *s++;\
        OP(dst[9] , (src9 +src10)*20 - (src8 +src11)*5 + (src7 +src12));\
        src13 = *s++;\
        OP(dst[10], (src10+src11)*20 - (src9 +src12)*5 + (src8 +src13));\
        src14 = *s++;\
        OP(dst[11], (src11+src12)*20 - (src10+src13)*5 + (src9 +src14));\
        src15 = *s++;\
        OP(dst[12], (src12+src13)*20 - (src11+src14)*5 + (src10+src15));\
        src16 = *s++;\
        OP(dst[13], (src13+src14)*20 - (src12+src15)*5 + (src11+src16));\
        src17 = *s++;\
        OP(dst[14], (src14+src15)*20 - (src13+src16)*5 + (src12+src17));\
        src18 = *s++;\
        OP(dst[15], (src15+src16)*20 - (src14+src17)*5 + (src13+src18));\
       } \
      } \
        dst+=dstStride;\
        src+=srcStride;\
    }while(--h);\
}\
\
static inline void OPNAME ## h264_qpel_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride,int w,int h){\
    uint8_t *cm = cropTbl + MAX_NEG_CROP;\
    do{\
        int srcB,srcA,src0,src1,src2,src3,src4,src5,src6;\
        uint8_t *s = src-2*srcStride,*d=dst;\
        srcB = *s; s+=srcStride;\
        srcA = *s; s+=srcStride;\
        src0 = *s; s+=srcStride;\
        src1 = *s; s+=srcStride;\
        src2 = *s; s+=srcStride;\
        src3 = *s; s+=srcStride;\
        OP(*d, (src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));d+=dstStride;\
        src4 = *s; s+=srcStride;\
        OP(*d, (src1+src2)*20 - (src0+src3)*5 + (srcA+src4));d+=dstStride;\
        src5 = *s; s+=srcStride;\
        OP(*d, (src2+src3)*20 - (src1+src4)*5 + (src0+src5));d+=dstStride;\
        src6 = *s; s+=srcStride;\
        OP(*d, (src3+src4)*20 - (src2+src5)*5 + (src1+src6));d+=dstStride;\
      if (h>4) { \
        int src7,src8,src9,src10; \
        src7 = *s; s+=srcStride;\
        OP(*d, (src4+src5)*20 - (src3+src6)*5 + (src2+src7));d+=dstStride;\
        src8 = *s; s+=srcStride;\
        OP(*d, (src5+src6)*20 - (src4+src7)*5 + (src3+src8));d+=dstStride;\
        src9 = *s; s+=srcStride;\
        OP(*d, (src6+src7)*20 - (src5+src8)*5 + (src4+src9));d+=dstStride;\
        src10 = *s; s+=srcStride;\
        OP(*d, (src7+src8)*20 - (src6+src9)*5 + (src5+src10));d+=dstStride;\
       if (h>8) { \
        int src11,src12,src13,src14,src15,src16,src17,src18; \