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.

        src11 = *s; s+=srcStride;\
        OP(*d , (src8 +src9 )*20 - (src7 +src10)*5 + (src6 +src11));d+=dstStride;\
        src12 = *s; s+=srcStride;\
        OP(*d , (src9 +src10)*20 - (src8 +src11)*5 + (src7 +src12));d+=dstStride;\
        src13 = *s; s+=srcStride;\
        OP(*d, (src10+src11)*20 - (src9 +src12)*5 + (src8 +src13));d+=dstStride;\
        src14 = *s; s+=srcStride;\
        OP(*d, (src11+src12)*20 - (src10+src13)*5 + (src9 +src14));d+=dstStride;\
        src15 = *s; s+=srcStride;\
        OP(*d, (src12+src13)*20 - (src11+src14)*5 + (src10+src15));d+=dstStride;\
        src16 = *s; s+=srcStride;\
        OP(*d, (src13+src14)*20 - (src12+src15)*5 + (src11+src16));d+=dstStride;\
        src17 = *s; s+=srcStride;\
        OP(*d, (src14+src15)*20 - (src13+src16)*5 + (src12+src17));d+=dstStride;\
        src18 = *s; s+=srcStride;\
        OP(*d, (src15+src16)*20 - (src14+src17)*5 + (src13+src18));d+=dstStride;\
       } \
      } \
        dst++;\
        src++;\
    }while(--w);\
}\
\
static inline void OPNAME ## h264_qpel_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride,int w,int h){\
    uint8_t *cm = cropTbl + MAX_NEG_CROP;\
    int i;\
    src -= 2*srcStride;\
    i= h+5; \
    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++;\
        tmp[0] = ((src0+src1)*20 - (srcA+src2)*5 + (srcB+src3));\
        src4 = *s++;\
        tmp[1] = ((src1+src2)*20 - (src0+src3)*5 + (srcA+src4));\
        src5 = *s++;\
        tmp[2] = ((src2+src3)*20 - (src1+src4)*5 + (src0+src5));\
        src6 = *s++;\
        tmp[3] = ((src3+src4)*20 - (src2+src5)*5 + (src1+src6));\
      if (w>4) { /* it optimized */ \
        int src7,src8,src9,src10; \
        src7 = *s++;\
        tmp[4] = ((src4+src5)*20 - (src3+src6)*5 + (src2+src7));\
        src8 = *s++;\
        tmp[5] = ((src5+src6)*20 - (src4+src7)*5 + (src3+src8));\
        src9 = *s++;\
        tmp[6] = ((src6+src7)*20 - (src5+src8)*5 + (src4+src9));\
        src10 = *s++;\
        tmp[7] = ((src7+src8)*20 - (src6+src9)*5 + (src5+src10));\
       if (w>8) { \
        int src11,src12,src13,src14,src15,src16,src17,src18; \
        src11 = *s++;\
        tmp[8] = ((src8 +src9 )*20 - (src7 +src10)*5 + (src6 +src11));\
        src12 = *s++;\
        tmp[9] = ((src9 +src10)*20 - (src8 +src11)*5 + (src7 +src12));\
        src13 = *s++;\
        tmp[10] = ((src10+src11)*20 - (src9 +src12)*5 + (src8 +src13));\
        src14 = *s++;\
        tmp[11] = ((src11+src12)*20 - (src10+src13)*5 + (src9 +src14));\
        src15 = *s++;\
        tmp[12] = ((src12+src13)*20 - (src11+src14)*5 + (src10+src15));\
        src16 = *s++;\
        tmp[13] = ((src13+src14)*20 - (src12+src15)*5 + (src11+src16));\
        src17 = *s++;\
        tmp[14] = ((src14+src15)*20 - (src13+src16)*5 + (src12+src17));\
        src18 = *s++;\
        tmp[15] = ((src15+src16)*20 - (src14+src17)*5 + (src13+src18));\
       } \
      } \
        tmp+=tmpStride;\
        src+=srcStride;\
    }while(--i);\
    tmp -= tmpStride*(h+5-2);\
    i = w; \
    do {\
        int tmpB,tmpA,tmp0,tmp1,tmp2,tmp3,tmp4,tmp5,tmp6;\
        int16_t *s = tmp-2*tmpStride; \
        uint8_t *d=dst;\
        tmpB = *s; s+=tmpStride;\
        tmpA = *s; s+=tmpStride;\
        tmp0 = *s; s+=tmpStride;\
        tmp1 = *s; s+=tmpStride;\
        tmp2 = *s; s+=tmpStride;\
        tmp3 = *s; s+=tmpStride;\
        OP2(*d, (tmp0+tmp1)*20 - (tmpA+tmp2)*5 + (tmpB+tmp3));d+=dstStride;\
        tmp4 = *s; s+=tmpStride;\
        OP2(*d, (tmp1+tmp2)*20 - (tmp0+tmp3)*5 + (tmpA+tmp4));d+=dstStride;\
        tmp5 = *s; s+=tmpStride;\
        OP2(*d, (tmp2+tmp3)*20 - (tmp1+tmp4)*5 + (tmp0+tmp5));d+=dstStride;\
        tmp6 = *s; s+=tmpStride;\
        OP2(*d, (tmp3+tmp4)*20 - (tmp2+tmp5)*5 + (tmp1+tmp6));d+=dstStride;\
      if (h>4) { \
        int tmp7,tmp8,tmp9,tmp10; \
        tmp7 = *s; s+=tmpStride;\
        OP2(*d, (tmp4+tmp5)*20 - (tmp3+tmp6)*5 + (tmp2+tmp7));d+=dstStride;\
        tmp8 = *s; s+=tmpStride;\
        OP2(*d, (tmp5+tmp6)*20 - (tmp4+tmp7)*5 + (tmp3+tmp8));d+=dstStride;\
        tmp9 = *s; s+=tmpStride;\
        OP2(*d, (tmp6+tmp7)*20 - (tmp5+tmp8)*5 + (tmp4+tmp9));d+=dstStride;\
        tmp10 = *s; s+=tmpStride;\
        OP2(*d, (tmp7+tmp8)*20 - (tmp6+tmp9)*5 + (tmp5+tmp10));d+=dstStride;\
       if (h>8) { \
        int tmp11,tmp12,tmp13,tmp14,tmp15,tmp16,tmp17,tmp18; \
        tmp11 = *s; s+=tmpStride;\
        OP2(*d , (tmp8 +tmp9 )*20 - (tmp7 +tmp10)*5 + (tmp6 +tmp11));d+=dstStride;\
        tmp12 = *s; s+=tmpStride;\
        OP2(*d , (tmp9 +tmp10)*20 - (tmp8 +tmp11)*5 + (tmp7 +tmp12));d+=dstStride;\
        tmp13 = *s; s+=tmpStride;\
        OP2(*d, (tmp10+tmp11)*20 - (tmp9 +tmp12)*5 + (tmp8 +tmp13));d+=dstStride;\
        tmp14 = *s; s+=tmpStride;\
        OP2(*d, (tmp11+tmp12)*20 - (tmp10+tmp13)*5 + (tmp9 +tmp14));d+=dstStride;\
        tmp15 = *s; s+=tmpStride;\
        OP2(*d, (tmp12+tmp13)*20 - (tmp11+tmp14)*5 + (tmp10+tmp15));d+=dstStride;\
        tmp16 = *s; s+=tmpStride;\
        OP2(*d, (tmp13+tmp14)*20 - (tmp12+tmp15)*5 + (tmp11+tmp16));d+=dstStride;\
        tmp17 = *s; s+=tmpStride;\
        OP2(*d, (tmp14+tmp15)*20 - (tmp13+tmp16)*5 + (tmp12+tmp17));d+=dstStride;\
        tmp18 = *s; s+=tmpStride;\
        OP2(*d, (tmp15+tmp16)*20 - (tmp14+tmp17)*5 + (tmp13+tmp18));d+=dstStride;\
       } \
      } \
        dst++;\
        tmp++;\
    }while(--i);\
}\
\
static void OPNAME ## h264_qpel4_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
    OPNAME ## h264_qpel_h_lowpass(dst,src,dstStride,srcStride,4,4); \
}\
static void OPNAME ## h264_qpel8_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
   OPNAME ## h264_qpel_h_lowpass(dst,src,dstStride,srcStride,8,8); \
}\
static void OPNAME ## h264_qpel16_h_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
   OPNAME ## h264_qpel_h_lowpass(dst,src,dstStride,srcStride,16,16); \
}\
\
static void OPNAME ## h264_qpel4_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
   OPNAME ## h264_qpel_v_lowpass(dst,src,dstStride,srcStride,4,4); \
}\
static void OPNAME ## h264_qpel8_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
   OPNAME ## h264_qpel_v_lowpass(dst,src,dstStride,srcStride,8,8); \
}\
static void OPNAME ## h264_qpel16_v_lowpass(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
   OPNAME ## h264_qpel_v_lowpass(dst,src,dstStride,srcStride,16,16); \
}\
static void OPNAME ## h264_qpel4_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
   OPNAME ## h264_qpel_hv_lowpass(dst,tmp,src,dstStride,tmpStride,srcStride,4,4); \
}\
static void OPNAME ## h264_qpel8_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
   OPNAME ## h264_qpel_hv_lowpass(dst,tmp,src,dstStride,tmpStride,srcStride,8,8); \
}\
static void OPNAME ## h264_qpel16_hv_lowpass(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\
   OPNAME ## h264_qpel_hv_lowpass(dst,tmp,src,dstStride,tmpStride,srcStride,16,16); \
}\

#define H264_MC(OPNAME, SIZE) \
static void OPNAME ## h264_qpel ## SIZE ## _mc00_c (uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## pixels ## SIZE ## _c(dst, src, stride, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc10_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t half[SIZE*SIZE];\
    put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\
    OPNAME ## pixels ## SIZE ## _l2_aligned2(dst, src, half, stride, stride, SIZE, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc20_c(uint8_t *dst, uint8_t *src, int stride){\
    OPNAME ## h264_qpel ## SIZE ## _h_lowpass(dst, src, stride, stride);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc30_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t half[SIZE*SIZE];\
    put_h264_qpel ## SIZE ## _h_lowpass(half, src, SIZE, stride);\
    OPNAME ## pixels ## SIZE ## _l2_aligned2(dst, src+1, half, stride, stride, SIZE, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc01_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[SIZE*(SIZE+5)];\
    uint8_t * const full_mid= full + SIZE*2;\
    uint8_t half[SIZE*SIZE];\
    copy_block ## SIZE (full, src - stride*2, SIZE,  stride, SIZE + 5);\
    put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\
    OPNAME ## pixels ## SIZE ## _l2_aligned(dst, full_mid, half, stride, SIZE, SIZE, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc02_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[SIZE*(SIZE+5)];\
    uint8_t * const full_mid= full + SIZE*2;\
    copy_block ## SIZE (full, src - stride*2, SIZE,  stride, SIZE + 5);\
    OPNAME ## h264_qpel ## SIZE ## _v_lowpass(dst, full_mid, stride, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc03_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[SIZE*(SIZE+5)];\
    uint8_t * const full_mid= full + SIZE*2;\
    uint8_t half[SIZE*SIZE];\
    copy_block ## SIZE (full, src - stride*2, SIZE,  stride, SIZE + 5);\
    put_h264_qpel ## SIZE ## _v_lowpass(half, full_mid, SIZE, SIZE);\
    OPNAME ## pixels ## SIZE ## _l2_aligned(dst, full_mid+SIZE, half, stride, SIZE, SIZE, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc11_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[SIZE*(SIZE+5)];\
    uint8_t * const full_mid= full + SIZE*2;\
    uint8_t halfH[SIZE*SIZE];\
    uint8_t halfV[SIZE*SIZE];\
    put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\
    copy_block ## SIZE (full, src - stride*2, SIZE,  stride, SIZE + 5);\
    put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\
    OPNAME ## pixels ## SIZE ## _l2_aligned(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc31_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[SIZE*(SIZE+5)];\
    uint8_t * const full_mid= full + SIZE*2;\
    uint8_t halfH[SIZE*SIZE];\
    uint8_t halfV[SIZE*SIZE];\
    put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\
    copy_block ## SIZE (full, src - stride*2 + 1, SIZE,  stride, SIZE + 5);\
    put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\
    OPNAME ## pixels ## SIZE ## _l2_aligned(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc13_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[SIZE*(SIZE+5)];\
    uint8_t * const full_mid= full + SIZE*2;\
    uint8_t halfH[SIZE*SIZE];\
    uint8_t halfV[SIZE*SIZE];\
    put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\
    copy_block ## SIZE (full, src - stride*2, SIZE,  stride, SIZE + 5);\
    put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\
    OPNAME ## pixels ## SIZE ## _l2_aligned(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc33_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[SIZE*(SIZE+5)];\
    uint8_t * const full_mid= full + SIZE*2;\
    uint8_t halfH[SIZE*SIZE];\
    uint8_t halfV[SIZE*SIZE];\
    put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\
    copy_block ## SIZE (full, src - stride*2 + 1, SIZE,  stride, SIZE + 5);\
    put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\
    OPNAME ## pixels ## SIZE ## _l2_aligned(dst, halfH, halfV, stride, SIZE, SIZE, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc22_c(uint8_t *dst, uint8_t *src, int stride){\
    int16_t tmp[SIZE*(SIZE+5)];\
    OPNAME ## h264_qpel ## SIZE ## _hv_lowpass(dst, tmp, src, stride, SIZE, stride);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc21_c(uint8_t *dst, uint8_t *src, int stride){\
    int16_t tmp[SIZE*(SIZE+5)];\
    uint8_t halfH[SIZE*SIZE];\
    uint8_t halfHV[SIZE*SIZE];\
    put_h264_qpel ## SIZE ## _h_lowpass(halfH, src, SIZE, stride);\
    put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\
    OPNAME ## pixels ## SIZE ## _l2_aligned(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc23_c(uint8_t *dst, uint8_t *src, int stride){\
    int16_t tmp[SIZE*(SIZE+5)];\
    uint8_t halfH[SIZE*SIZE];\
    uint8_t halfHV[SIZE*SIZE];\
    put_h264_qpel ## SIZE ## _h_lowpass(halfH, src + stride, SIZE, stride);\
    put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\
    OPNAME ## pixels ## SIZE ## _l2_aligned(dst, halfH, halfHV, stride, SIZE, SIZE, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc12_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[SIZE*(SIZE+5)];\
    uint8_t * const full_mid= full + SIZE*2;\
    int16_t tmp[SIZE*(SIZE+5)];\
    uint8_t halfV[SIZE*SIZE];\
    uint8_t halfHV[SIZE*SIZE];\
    copy_block ## SIZE (full, src - stride*2, SIZE,  stride, SIZE + 5);\
    put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\
    put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\
    OPNAME ## pixels ## SIZE ## _l2_aligned(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\
}\
\
static void OPNAME ## h264_qpel ## SIZE ## _mc32_c(uint8_t *dst, uint8_t *src, int stride){\
    uint8_t full[SIZE*(SIZE+5)];\
    uint8_t * const full_mid= full + SIZE*2;\
    int16_t tmp[SIZE*(SIZE+5)];\
    uint8_t halfV[SIZE*SIZE];\
    uint8_t halfHV[SIZE*SIZE];\
    copy_block ## SIZE (full, src - stride*2 + 1, SIZE,  stride, SIZE + 5);\
    put_h264_qpel ## SIZE ## _v_lowpass(halfV, full_mid, SIZE, SIZE);\
    put_h264_qpel ## SIZE ## _hv_lowpass(halfHV, tmp, src, SIZE, SIZE, stride);\
    OPNAME ## pixels ## SIZE ## _l2_aligned(dst, halfV, halfHV, stride, SIZE, SIZE, SIZE);\
}\

#define op_avg(a, b)  a = (((a)+cm[((b) + 16)>>5]+1)>>1)
//#define op_avg2(a, b) a = (((a)*w1+cm[((b) + 16)>>5]*w2 + o + 64)>>7)
#define op_put(a, b)  a = cm[((b) + 16)>>5]
#define op2_avg(a, b)  a = (((a)+cm[((b) + 512)>>10]+1)>>1)
#define op2_put(a, b)  a = cm[((b) + 512)>>10]

H264_LOWPASS(put_       , op_put, op2_put)
H264_LOWPASS(avg_       , op_avg, op2_avg)
H264