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 inline void OPNAME ## _pixels16_l4_aligned0(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\
	do { /* src1 is unaligned */\
		uint32_t a0,a1,a2,a3; \
		UNPACK(a0,a1,LD32(src1),LP(src2)); \
		UNPACK(a2,a3,LP(src3),LP(src4)); \
		OP(LP(dst),rnd_PACK(a0,a1,a2,a3)); \
		UNPACK(a0,a1,LD32(src1+4),LP(src2+4)); \
		UNPACK(a2,a3,LP(src3+4),LP(src4+4)); \
		OP(LP(dst+8),rnd_PACK(a0,a1,a2,a3)); \
		UNPACK(a0,a1,LD32(src1+8),LP(src2+8)); \
		UNPACK(a2,a3,LP(src3+8),LP(src4+8)); \
		OP(LP(dst+8),rnd_PACK(a0,a1,a2,a3)); \
		UNPACK(a0,a1,LD32(src1+12),LP(src2+12)); \
		UNPACK(a2,a3,LP(src3+12),LP(src4+12)); \
		OP(LP(dst+12),rnd_PACK(a0,a1,a2,a3)); \
		src1+=src_stride1;\
		src2+=src_stride2;\
		src3+=src_stride3;\
		src4+=src_stride4;\
		dst+=dst_stride;\
	} while(--h); \
} \
\
static inline void OPNAME ## _no_rnd_pixels16_l4_aligned0(uint8_t *dst, const uint8_t *src1, uint8_t *src2, uint8_t *src3, uint8_t *src4,int dst_stride, int src_stride1, int src_stride2,int src_stride3,int src_stride4, int h){\
	do { \
		uint32_t a0,a1,a2,a3; \
		UNPACK(a0,a1,LD32(src1),LP(src2)); \
		UNPACK(a2,a3,LP(src3),LP(src4)); \
		OP(LP(dst),no_rnd_PACK(a0,a1,a2,a3)); \
		UNPACK(a0,a1,LD32(src1+4),LP(src2+4)); \
		UNPACK(a2,a3,LP(src3+4),LP(src4+4)); \
		OP(LP(dst+4),no_rnd_PACK(a0,a1,a2,a3)); \
		UNPACK(a0,a1,LD32(src1+8),LP(src2+8)); \
		UNPACK(a2,a3,LP(src3+8),LP(src4+8)); \
		OP(LP(dst+8),no_rnd_PACK(a0,a1,a2,a3)); \
		UNPACK(a0,a1,LD32(src1+12),LP(src2+12)); \
		UNPACK(a2,a3,LP(src3+12),LP(src4+12)); \
		OP(LP(dst+12),no_rnd_PACK(a0,a1,a2,a3)); \
		src1+=src_stride1;\
		src2+=src_stride2;\
		src3+=src_stride3;\
		src4+=src_stride4;\
		dst+=dst_stride;\
	} while(--h); \
} \
\

#define op_avg(a, b) a = rnd_avg32(a,b)
#define op_put(a, b) a = b

PIXOP2(avg, op_avg)
PIXOP2(put, op_put)
#undef op_avg
#undef op_put

#define avg2(a,b) ((a+b+1)>>1)
#define avg4(a,b,c,d) ((a+b+c+d+2)>>2)


static void gmc1_c(uint8_t *dst, uint8_t *src, int stride, int h, int x16, int y16, int rounder)
{
    const int A=(16-x16)*(16-y16);
    const int B=(   x16)*(16-y16);
    const int C=(16-x16)*(   y16);
    const int D=(   x16)*(   y16);

    do {
        int t0,t1,t2,t3;
        uint8_t *s0 = src;
        uint8_t *s1 = src+stride;
        t0 = *s0++; t2 = *s1++;
        t1 = *s0++; t3 = *s1++;
        dst[0]= (A*t0 + B*t1 + C*t2 + D*t3 + rounder)>>8;
        t0 = *s0++; t2 = *s1++;
        dst[1]= (A*t1 + B*t0 + C*t3 + D*t2 + rounder)>>8;
        t1 = *s0++; t3 = *s1++;
        dst[2]= (A*t0 + B*t1 + C*t2 + D*t3 + rounder)>>8;
        t0 = *s0++; t2 = *s1++;
        dst[3]= (A*t1 + B*t0 + C*t3 + D*t2 + rounder)>>8;
        t1 = *s0++; t3 = *s1++;
        dst[4]= (A*t0 + B*t1 + C*t2 + D*t3 + rounder)>>8;
        t0 = *s0++; t2 = *s1++;
        dst[5]= (A*t1 + B*t0 + C*t3 + D*t2 + rounder)>>8;
        t1 = *s0++; t3 = *s1++;
        dst[6]= (A*t0 + B*t1 + C*t2 + D*t3 + rounder)>>8;
        t0 = *s0++; t2 = *s1++;
        dst[7]= (A*t1 + B*t0 + C*t3 + D*t2 + rounder)>>8;
        dst+= stride;
        src+= stride;
    }while(--h);
}

static void gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, 
                  int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height)
{
    int y, vx, vy;
    const int s= 1<<shift;
    
    width--;
    height--;

    for(y=0; y<h; y++){
        int x;

        vx= ox;
        vy= oy;
        for(x=0; x<8; x++){ //XXX FIXME optimize
            int src_x, src_y, frac_x, frac_y, index;

            src_x= vx>>16;
            src_y= vy>>16;
            frac_x= src_x&(s-1);
            frac_y= src_y&(s-1);
            src_x>>=shift;
            src_y>>=shift;
  
            if((unsigned)src_x < width){
                if((unsigned)src_y < height){
                    index= src_x + src_y*stride;
                    dst[y*stride + x]= (  (  src[index         ]*(s-frac_x)
                                           + src[index       +1]*   frac_x )*(s-frac_y)
                                        + (  src[index+stride  ]*(s-frac_x)
                                           + src[index+stride+1]*   frac_x )*   frac_y
                                        + r)>>(shift*2);
                }else{
                    index= src_x + clip(src_y, 0, height)*stride;                    
                    dst[y*stride + x]= ( (  src[index         ]*(s-frac_x) 
                                          + src[index       +1]*   frac_x )*s
                                        + r)>>(shift*2);
                }
            }else{
                if((unsigned)src_y < height){
                    index= clip(src_x, 0, width) + src_y*stride;                    
                    dst[y*stride + x]= (  (  src[index         ]*(s-frac_y) 
                                           + src[index+stride  ]*   frac_y )*s
                                        + r)>>(shift*2);
                }else{
                    index= clip(src_x, 0, width) + clip(src_y, 0, height)*stride;                    
                    dst[y*stride + x]=    src[index         ];
                }
            }
            
            vx+= dxx;
            vy+= dyx;
        }
        ox += dxy;
        oy += dyy;
    }
}
#define H264_CHROMA_MC(OPNAME, OP)\
static void OPNAME ## h264_chroma_mc2_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){\
    const int A=(8-x)*(8-y);\
    const int B=(  x)*(8-y);\
    const int C=(8-x)*(  y);\
    const int D=(  x)*(  y);\
    \
    assert(x<8 && y<8 && x>=0 && y>=0);\
\
    do {\
        int t0,t1,t2,t3; \
        uint8_t *s0 = src; \
        uint8_t *s1 = src+stride; \
        t0 = *s0++; t2 = *s1++; \
        t1 = *s0++; t3 = *s1++; \
        OP(dst[0], (A*t0 + B*t1 + C*t2 + D*t3));\
        t0 = *s0++; t2 = *s1++; \
        OP(dst[1], (A*t1 + B*t0 + C*t3 + D*t2));\
        dst+= stride;\
        src+= stride;\
    }while(--h);\
}\
\
static void OPNAME ## h264_chroma_mc4_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){\
    const int A=(8-x)*(8-y);\
    const int B=(  x)*(8-y);\
    const int C=(8-x)*(  y);\
    const int D=(  x)*(  y);\
    \
    assert(x<8 && y<8 && x>=0 && y>=0);\
\
    do {\
        int t0,t1,t2,t3; \
        uint8_t *s0 = src; \
        uint8_t *s1 = src+stride; \
        t0 = *s0++; t2 = *s1++; \
        t1 = *s0++; t3 = *s1++; \
        OP(dst[0], (A*t0 + B*t1 + C*t2 + D*t3));\
        t0 = *s0++; t2 = *s1++; \
        OP(dst[1], (A*t1 + B*t0 + C*t3 + D*t2));\
        t1 = *s0++; t3 = *s1++; \
        OP(dst[2], (A*t0 + B*t1 + C*t2 + D*t3));\
        t0 = *s0++; t2 = *s1++; \
        OP(dst[3], (A*t1 + B*t0 + C*t3 + D*t2));\
        dst+= stride;\
        src+= stride;\
    }while(--h);\
}\
\
static void OPNAME ## h264_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){\
    const int A=(8-x)*(8-y);\
    const int B=(  x)*(8-y);\
    const int C=(8-x)*(  y);\
    const int D=(  x)*(  y);\
    \
    assert(x<8 && y<8 && x>=0 && y>=0);\
\
    do {\
        int t0,t1,t2,t3; \
        uint8_t *s0 = src; \
        uint8_t *s1 = src+stride; \
        t0 = *s0++; t2 = *s1++; \
        t1 = *s0++; t3 = *s1++; \
        OP(dst[0], (A*t0 + B*t1 + C*t2 + D*t3));\
        t0 = *s0++; t2 = *s1++; \
        OP(dst[1], (A*t1 + B*t0 + C*t3 + D*t2));\
        t1 = *s0++; t3 = *s1++; \
        OP(dst[2], (A*t0 + B*t1 + C*t2 + D*t3));\
        t0 = *s0++; t2 = *s1++; \
        OP(dst[3], (A*t1 + B*t0 + C*t3 + D*t2));\
        t1 = *s0++; t3 = *s1++; \
        OP(dst[4], (A*t0 + B*t1 + C*t2 + D*t3));\
        t0 = *s0++; t2 = *s1++; \
        OP(dst[5], (A*t1 + B*t0 + C*t3 + D*t2));\
        t1 = *s0++; t3 = *s1++; \
        OP(dst[6], (A*t0 + B*t1 + C*t2 + D*t3));\
        t0 = *s0++; t2 = *s1++; \
        OP(dst[7], (A*t1 + B*t0 + C*t3 + D*t2));\
        dst+= stride;\
        src+= stride;\
    }while(--h);\
}

#define op_avg(a, b) a = (((a)+(((b) + 32)>>6)+1)>>1)
#define op_put(a, b) a = (((b) + 32)>>6)

H264_CHROMA_MC(put_       , op_put)
H264_CHROMA_MC(avg_       , op_avg)
#undef op_avg
#undef op_put

/* not yet optimized */
static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
{
    int i;
    for(i=0; i<h; i++)
    {
        ST32(dst   , LD32(src   ));
        dst+=dstStride;
        src+=srcStride;
    }
}

static inline void copy_block8(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+=dstStride;
        src+=srcStride;
    }
}

static inline void copy_block16(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 ));
        ST32(dst+8 , LD32(src+8 ));
        ST32(dst+12, LD32(src+12));
        dst+=dstStride;
        src+=srcStride;
    }
}

static inline void copy_block17(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 ));
        ST32(dst+8 , LD32(src+8 ));
        ST32(dst+12, LD32(src+12));
        dst[16]= src[16];
        dst+=dstStride;
        src+=srcStride;
    }
}

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;
    }