swscale.c

从FFMPEG转换而来的H264解码程序,VC下编译..

		"0:				\n\t"
		"movq (%%"REG_d", %%"REG_a"), %%mm3\n\t"
		"movd (%%"REG_c", %%"REG_S"), %%mm0\n\t"
		"movd 1(%%"REG_c", %%"REG_S"), %%mm1\n\t"
		"punpcklbw %%mm7, %%mm1		\n\t"
		"punpcklbw %%mm7, %%mm0		\n\t"
		"pshufw $0xFF, %%mm1, %%mm1	\n\t"
		"1:				\n\t"
		"pshufw $0xFF, %%mm0, %%mm0	\n\t"
		"2:				\n\t"
		"psubw %%mm1, %%mm0		\n\t"
		"movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
		"pmullw %%mm3, %%mm0		\n\t"
		"psllw $7, %%mm1		\n\t"
		"paddw %%mm1, %%mm0		\n\t"

		"movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"

		"add $8, %%"REG_a"		\n\t"
	// End
		"9:				\n\t"
//		"int $3\n\t"
		"lea 0b, %0			\n\t"
		"lea 1b, %1			\n\t"
		"lea 2b, %2			\n\t"
		"dec %1				\n\t"
		"dec %2				\n\t"
		"sub %0, %1			\n\t"
		"sub %0, %2			\n\t"
		"lea 9b, %3			\n\t"
		"sub %0, %3			\n\t"


		:"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A),
		"=r" (fragmentLengthA)
	);

	asm volatile(
		"jmp 9f				\n\t"
	// Begin
		"0:				\n\t"
		"movq (%%"REG_d", %%"REG_a"), %%mm3\n\t"
		"movd (%%"REG_c", %%"REG_S"), %%mm0\n\t"
		"punpcklbw %%mm7, %%mm0		\n\t"
		"pshufw $0xFF, %%mm0, %%mm1	\n\t"
		"1:				\n\t"
		"pshufw $0xFF, %%mm0, %%mm0	\n\t"
		"2:				\n\t"
		"psubw %%mm1, %%mm0		\n\t"
		"movl 8(%%"REG_b", %%"REG_a"), %%esi\n\t"
		"pmullw %%mm3, %%mm0		\n\t"
		"psllw $7, %%mm1		\n\t"
		"paddw %%mm1, %%mm0		\n\t"

		"movq %%mm0, (%%"REG_D", %%"REG_a")\n\t"

		"add $8, %%"REG_a"		\n\t"
	// End
		"9:				\n\t"
//		"int $3\n\t"
		"lea 0b, %0			\n\t"
		"lea 1b, %1			\n\t"
		"lea 2b, %2			\n\t"
		"dec %1				\n\t"
		"dec %2				\n\t"
		"sub %0, %1			\n\t"
		"sub %0, %2			\n\t"
		"lea 9b, %3			\n\t"
		"sub %0, %3			\n\t"


		:"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B),
		"=r" (fragmentLengthB)
	);

	xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers
	fragmentPos=0;

	for(i=0; i<dstW/numSplits; i++)
	{
		int xx=xpos>>16;

		if((i&3) == 0)
		{
			int a=0;
			int b=((xpos+xInc)>>16) - xx;
			int c=((xpos+xInc*2)>>16) - xx;
			int d=((xpos+xInc*3)>>16) - xx;

			filter[i  ] = (( xpos         & 0xFFFF) ^ 0xFFFF)>>9;
			filter[i+1] = (((xpos+xInc  ) & 0xFFFF) ^ 0xFFFF)>>9;
			filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9;
			filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9;
			filterPos[i/2]= xx;

			if(d+1<4)
			{
				int maxShift= 3-(d+1);
				int shift=0;

				memcpy(funnyCode + fragmentPos, fragmentB, fragmentLengthB);

				funnyCode[fragmentPos + imm8OfPShufW1B]=
					(a+1) | ((b+1)<<2) | ((c+1)<<4) | ((d+1)<<6);
				funnyCode[fragmentPos + imm8OfPShufW2B]=
					a | (b<<2) | (c<<4) | (d<<6);

				if(i+3>=dstW) shift=maxShift; //avoid overread
				else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align

				if(shift && i>=shift)
				{
					funnyCode[fragmentPos + imm8OfPShufW1B]+= 0x55*shift;
					funnyCode[fragmentPos + imm8OfPShufW2B]+= 0x55*shift;
					filterPos[i/2]-=shift;
				}

				fragmentPos+= fragmentLengthB;
			}
			else
			{
				int maxShift= 3-d;
				int shift=0;

				memcpy(funnyCode + fragmentPos, fragmentA, fragmentLengthA);

				funnyCode[fragmentPos + imm8OfPShufW1A]=
				funnyCode[fragmentPos + imm8OfPShufW2A]=
					a | (b<<2) | (c<<4) | (d<<6);

				if(i+4>=dstW) shift=maxShift; //avoid overread
				else if((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //partial align

				if(shift && i>=shift)
				{
					funnyCode[fragmentPos + imm8OfPShufW1A]+= 0x55*shift;
					funnyCode[fragmentPos + imm8OfPShufW2A]+= 0x55*shift;
					filterPos[i/2]-=shift;
				}

				fragmentPos+= fragmentLengthA;
			}

			funnyCode[fragmentPos]= RET;
		}
		xpos+=xInc;
	}
	filterPos[i/2]= xpos>>16; // needed to jump to the next part
}
#endif // ARCH_X86 || ARCH_X86_64

static void globalInit(void){
    // generating tables:
    int i;
    for(i=0; i<768; i++){
	int c= FFMIN(FFMAX(i-256, 0), 255);
	clip_table[i]=c;
    }
}

static SwsFunc getSwsFunc(int flags){

#ifdef RUNTIME_CPUDETECT
#if defined(ARCH_X86) || defined(ARCH_X86_64)
	// ordered per speed fasterst first
	if(flags & SWS_CPU_CAPS_MMX2)
		return swScale_MMX2;
	else if(flags & SWS_CPU_CAPS_3DNOW)
		return swScale_3DNow;
	else if(flags & SWS_CPU_CAPS_MMX)
		return swScale_MMX;
	else
		return swScale_C;

#else
#ifdef ARCH_POWERPC
	if(flags & SWS_CPU_CAPS_ALTIVEC)
	  return swScale_altivec;
	else
	  return swScale_C;
#endif
	return swScale_C;
#endif /* defined(ARCH_X86) || defined(ARCH_X86_64) */
#else //RUNTIME_CPUDETECT
#ifdef HAVE_MMX2
	return swScale_MMX2;
#elif defined (HAVE_3DNOW)
	return swScale_3DNow;
#elif defined (HAVE_MMX)
	return swScale_MMX;
#elif defined (HAVE_ALTIVEC)
	return swScale_altivec;
#else
	return swScale_C;
#endif
#endif //!RUNTIME_CPUDETECT
}

static int PlanarToNV12Wrapper(SwsContext *c, uint8_t* src[], stride_t srcStride[], int srcSliceY,
             int srcSliceH, uint8_t* dstParam[], stride_t dstStride[]){
	uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;
	/* Copy Y plane */
	if(dstStride[0]==srcStride[0] && srcStride[0] > 0)
		memcpy(dst, src[0], srcSliceH*dstStride[0]);
	else
	{
		int i;
		uint8_t *srcPtr= src[0];
		uint8_t *dstPtr= dst;
		for(i=0; i<srcSliceH; i++)
		{
			memcpy(dstPtr, srcPtr, c->srcW);
			srcPtr+= srcStride[0];
			dstPtr+= dstStride[0];
		}
	}
	dst = dstParam[1]; //dstParam[0] + dstStride[0]*c->srcH;
	if (c->dstFormat == IMGFMT_NV12)
		interleaveBytes( src[1],src[2],dst,c->srcW/2,srcSliceH/2,srcStride[1],srcStride[2],dstStride[0] );
	else
		interleaveBytes( src[2],src[1],dst,c->srcW/2,srcSliceH/2,srcStride[2],srcStride[1],dstStride[0] );

	return srcSliceH;
}

static int NV12ToPlanarWrapper(SwsContext *c, uint8_t* srcParam[], stride_t srcStride[], int srcSliceY,
             int srcSliceH, uint8_t* dst[], stride_t dstStride[]){
	const uint8_t *src=srcParam[0] + srcStride[0]*srcSliceY;
        unsigned char *dst_U =dst[1];
        unsigned char *dst_V =dst[2];
	int swapped=c->srcFormat == IMGFMT_NV12;
	int x,y,idx;

	if(dstStride[0]==srcStride[0] && srcStride[0]>0)
		memcpy(dst[0], src, srcSliceH*dstStride[0]);
	else
	{
		int i;
		uint8_t *dstPtr= dst[0];
		const uint8_t *srcPtr= src;
		for(i=0; i<srcSliceH; i++)
		{
			memcpy(dstPtr, srcPtr, c->srcW);
			srcPtr+= srcStride[0];
			dstPtr+= dstStride[0];
		}
	}

    // chroma data is interlaced UVUV... so deinterlace it
    idx=0;
    src=srcParam[0]+c->srcW*c->srcH;
    for(y=0; y<srcSliceH/2;y++,dst_U+=dstStride[1],dst_V+=dstStride[2] ) {
        for(x=0;x<c->srcW/2;x++,idx++){
            *(dst_U + x) = *(src + (idx<<1) + (swapped ? 1 : 0));
            *(dst_V + x) = *(src + (idx<<1) + (swapped ? 0 : 1));
        }
    }
    return srcSliceH;
}

static int PlanarToYuy2Wrapper(SwsContext *c, uint8_t* src[], stride_t srcStride[], int srcSliceY,
             int srcSliceH, uint8_t* dstParam[], stride_t dstStride[]){
	uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;

	yv12toyuy2( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );

	return srcSliceH;
}

static int PlanarToUyvyWrapper(SwsContext *c, uint8_t* src[], stride_t srcStride[], int srcSliceY,
             int srcSliceH, uint8_t* dstParam[], stride_t dstStride[]){
	uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;

	yv12touyvy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );

	return srcSliceH;
}

static int PlanarToYvyuWrapper(SwsContext *c, uint8_t* src[], stride_t srcStride[], int srcSliceY,
             int srcSliceH, uint8_t* dstParam[], stride_t dstStride[]){
	uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;

	yv12toyvyu( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );

	return srcSliceH;
}
static int PlanarToVyuyWrapper(SwsContext *c, uint8_t* src[], stride_t srcStride[], int srcSliceY,
             int srcSliceH, uint8_t* dstParam[], stride_t dstStride[]){
	uint8_t *dst=dstParam[0] + dstStride[0]*srcSliceY;

	yv12tovyuy( src[0],src[1],src[2],dst,c->srcW,srcSliceH,srcStride[0],srcStride[1],dstStride[0] );

	return srcSliceH;
}

/* {RGB,BGR}{15,16,24,32} -> {RGB,BGR}{15,16,24,32} */
static int rgb2rgbWrapper(SwsContext *c, uint8_t* src[], stride_t srcStride[], int srcSliceY,
			   int srcSliceH, uint8_t* dst[], stride_t dstStride[]){
	const int srcFormat= c->srcFormat;
	const int dstFormat= c->dstFormat;
	const int srcBpp= ((srcFormat&0xFF) + 7)>>3;
	const int dstBpp= ((dstFormat&0xFF) + 7)>>3;
	const int srcId= (srcFormat&0xFF)>>2; // 1:0, 4:1, 8:2, 15:3, 16:4, 24:6, 32:8
	const int dstId= (dstFormat&0xFF)>>2;
	void (*conv)(const uint8_t *src, uint8_t *dst, stride_t src_size)=NULL;

	/* BGR -> BGR */
	if(   (isBGR(srcFormat) && isBGR(dstFormat))
	   || (isRGB(srcFormat) && isRGB(dstFormat))){
		switch(srcId | (dstId<<4)){
		case 0x34: conv= rgb16to15; break;
		case 0x36: conv= rgb24to15; break;
		case 0x38: conv= rgb32to15; break;
		case 0x43: conv= rgb15to16; break;
		case 0x46: conv= rgb24to16; break;
		case 0x48: conv= rgb32to16; break;
		case 0x63: conv= rgb15to24; break;
		case 0x64: conv= rgb16to24; break;
		case 0x68: conv= rgb32to24; break;
		case 0x83: conv= rgb15to32; break;
		case 0x84: conv= rgb16to32; break;
		case 0x86: conv= rgb24to32; break;
		default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
				 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
		}
	}else if(   (isBGR(srcFormat) && isRGB(dstFormat))
		 || (isRGB(srcFormat) && isBGR(dstFormat))){
		switch(srcId | (dstId<<4)){
		case 0x33: conv= rgb15tobgr15; break;
		case 0x34: conv= rgb16tobgr15; break;
		case 0x36: conv= rgb24tobgr15; break;
		case 0x38: conv= rgb32tobgr15; break;
		case 0x43: conv= rgb15tobgr16; break;
		case 0x44: conv= rgb16tobgr16; break;
		case 0x46: conv= rgb24tobgr16; break;
		case 0x48: conv= rgb32tobgr16; break;
		case 0x63: conv= rgb15tobgr24; break;
		case 0x64: conv= rgb16tobgr24; break;
		case 0x66: conv= rgb24tobgr24; break;
		case 0x68: conv= rgb32tobgr24; break;
		case 0x83: conv= rgb15tobgr32; break;
		case 0x84: conv= rgb16tobgr32; break;
		case 0x86: conv= rgb24tobgr32; break;
		case 0x88: conv= rgb32tobgr32; break;
		default: MSG_ERR("swScaler: internal error %s -> %s converter\n",
				 sws_format_name(srcFormat), sws_format_name(dstFormat)); break;
		}
	}else{
		MSG_ERR("swScaler: internal error %s -> %s converter\n",
			 sws_format_name(srcFormat), sws_format_name(dstFormat));
	}

	if(dstStride[0]*srcBpp == srcStride[0]*dstBpp && dstStride[0]>0)
		conv(src[0], dst[0] + dstStride[0]*srcSliceY, srcSliceH*srcStride[0]);
	else
	{
		int i;
		uint8_t *srcPtr= src[0];
		uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;

		for(i=0; i<srcSliceH; i++)
		{
			conv(srcPtr, dstPtr, c->srcW*srcBpp);
			srcPtr+= srcStride[0];
			dstPtr+= dstStride[0];
		}
	}
	return srcSliceH;
}

static int bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], stride_t srcStride[], int srcSliceY,
             int srcSliceH, uint8_t* dst[], stride_t dstStride[]){

	rgb24toyv12(
		src[0],
		dst[0]+ srcSliceY    *dstStride[0],
		dst[1]+(srcSliceY>>1)*dstStride[1],
		dst[2]+(srcSliceY>>1)*dstStride[2],
		c->srcW, srcSliceH,
		dstStride[0], dstStride[1], srcStride[0]);
	return srcSliceH;
}

static int yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], stride_t srcStride[], int srcSliceY,
             int srcSliceH, uint8_t* dst[], stride_t dstStride[]){
	int i;

	/* copy Y */
	if(srcStride[0]==dstStride[0] && srcStride[0] > 0)
		memcpy(dst[0]+ srcSliceY*dstStride[0], src[0], srcStride[0]*srcSliceH);
	else{
		uint8_t *srcPtr= src[0];
		uint8_t *dstPtr= dst[0] + dstStride[0]*srcSliceY;

		for(i=0; i<srcSliceH; i++)
		{
			memcpy(dstPtr, srcPtr, c->srcW);
			srcPtr+= srcStride[0];
			dstPtr+= dstStride[0];
		}
	}

	if(c->dstFormat==IMGFMT_YV12){
		planar2x(src[1], dst[1], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[1]);
		planar2x(src[2], dst[2], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[2]);
	}else{
		planar2x(src[1], dst[2], c->chrSrcW, c->chrSrcH, srcStride[1], dstStride[2]);
		planar2x(src[2], dst[1], c->chrSrcW, c->chrSrcH, srcStride[2], dstStride[1]);
	}
	return srcSliceH;
}

/**
 * bring pointers in YUV order instead of YVU
 */
static inline void sws_orderYUV(int format, uint8_t * sortedP[], stride_t sortedStride[], uint8_t * p[], stride_t stride[]){
	if(format == IMGFMT_YV12 || format == IMGFMT_YVU9
           || format == IMGFMT_444P || format == IMGFMT_422P || format == IMGFMT_411P){
		sortedP[0]= p[0];
		sortedP[1]= p[2];
		sortedP[2]= p[1];
		sortedStride[0]= stride[0];
		sortedStride[1]= stride[2];
		sortedStride[2]= stride[1];
	}
	else if(isPacked(format) || isGray(format) || format == IMGFMT_Y8)
	{
		sortedP[0]= p[0];
		sortedP[1]=
		sortedP[2]= NULL;
		sortedStride[0]= stride[0];
		sortedStride[1]=
		sortedStride[2]= 0;
	}
	else if(format == IMGFMT_I420 || format == IMGFMT_IYUV)
	{
		sortedP[0]= p[0];
		sortedP[1]= p[1];