yvu9 -> yv12 unscaled converter with linear chroma scaling

Originally committed as revision 6583 to svn://svn.mplayerhq.hu/mplayer/trunk/postproc

postproc/rgb2rgb.c

@@ -512,6 +512,19 @@ void yvu9toyv12(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc,
 #endif
 }
 
+void planar2x(const uint8_t *src, uint8_t *dst, int width, int height, int srcStride, int dstStride)
+{
+#ifdef CAN_COMPILE_X86_ASM
+	// ordered per speed fasterst first
+	if(gCpuCaps.hasMMX2)
+		planar2x_MMX2(src, dst, width, height, srcStride, dstStride);
+	else if(gCpuCaps.has3DNow)
+		planar2x_3DNow(src, dst, width, height, srcStride, dstStride);
+	else
+#endif
+		planar2x_C(src, dst, width, height, srcStride, dstStride);
+}
+
 /**
  *
  * height should be a multiple of 2 and width should be a multiple of 2 (if this is a

postproc/rgb2rgb.h

@@ -41,6 +41,7 @@ extern void yuy2toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t
 extern void rgb24toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,
 	unsigned int width, unsigned int height,
 	unsigned int lumStride, unsigned int chromStride, unsigned int srcStride);
+extern void planar2x(const uint8_t *src, uint8_t *dst, int width, int height, int srcStride, int dstStride);
 
 extern void interleaveBytes(uint8_t *src1, uint8_t *src2, uint8_t *dst,
 			    unsigned width, unsigned height, unsigned src1Stride,

postproc/rgb2rgb_template.c

@@ -1295,6 +1295,95 @@ static inline void RENAME(yvu9toyv12)(const uint8_t *ysrc, const uint8_t *usrc,
 	/* XXX: implement upscaling for U,V */
 }
 
+static inline void RENAME(planar2x)(const uint8_t *src, uint8_t *dst, int srcWidth, int srcHeight, int srcStride, int dstStride)
+{
+	int x,y;
+	
+	// first line
+	for(x=0; x<srcWidth; x++){
+		dst[2*x+0]=
+		dst[2*x+1]= src[x];
+	}
+	dst+= dstStride;
+
+	for(y=1; y<srcHeight; y++){
+#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
+		const int mmxSize= srcWidth;
+		asm volatile(
+			"movl %4, %%eax			\n\t"
+			"1:				\n\t"
+			"movq (%0, %%eax), %%mm0	\n\t"
+			"movq (%1, %%eax), %%mm1	\n\t"
+			"movq 1(%0, %%eax), %%mm2	\n\t"
+			"movq 1(%1, %%eax), %%mm3	\n\t"
+			"movq %%mm0, %%mm4		\n\t"
+			"movq %%mm1, %%mm5		\n\t"
+			PAVGB" %%mm3, %%mm0		\n\t"
+			PAVGB" %%mm3, %%mm0		\n\t"
+			PAVGB" %%mm4, %%mm3		\n\t"
+			PAVGB" %%mm4, %%mm3		\n\t"
+			PAVGB" %%mm2, %%mm1		\n\t"
+			PAVGB" %%mm2, %%mm1		\n\t"
+			PAVGB" %%mm5, %%mm2		\n\t"
+			PAVGB" %%mm5, %%mm2		\n\t"
+			"movq %%mm3, %%mm4		\n\t"
+			"movq %%mm2, %%mm5		\n\t"
+			"punpcklbw %%mm1, %%mm3		\n\t"
+			"punpckhbw %%mm1, %%mm4		\n\t"
+			"punpcklbw %%mm0, %%mm2		\n\t"
+			"punpckhbw %%mm0, %%mm5		\n\t"
+#if 1
+			MOVNTQ" %%mm3, (%2, %%eax, 2)	\n\t"
+			MOVNTQ" %%mm4, 8(%2, %%eax, 2)	\n\t"
+			MOVNTQ" %%mm2, (%3, %%eax, 2)	\n\t"
+			MOVNTQ" %%mm5, 8(%3, %%eax, 2)	\n\t"
+#else
+			"movq %%mm3, (%2, %%eax, 2)	\n\t"
+			"movq %%mm4, 8(%2, %%eax, 2)	\n\t"
+			"movq %%mm2, (%3, %%eax, 2)	\n\t"
+			"movq %%mm5, 8(%3, %%eax, 2)	\n\t"
+#endif
+			"addl $8, %%eax			\n\t"
+			" js 1b				\n\t"
+			:: "r" (src + mmxSize-1), "r" (src + srcStride + mmxSize-1),
+			   "r" (dst + mmxSize*2), "r" (dst + dstStride + mmxSize*2),
+			   "g" (-mmxSize)
+			: "%eax"
+
+		);
+		dst[0]= 
+		dst[dstStride]= src[0];
+#else
+		dst[0]= 
+		dst[dstStride]= src[0];
+
+		for(x=0; x<srcWidth-1; x++){
+			dst[2*x          +1]= (3*src[x+0] +   src[x+srcStride+1])>>2;
+			dst[2*x+dstStride+2]= (  src[x+0] + 3*src[x+srcStride+1])>>2;
+			dst[2*x+dstStride+1]= (  src[x+1] + 3*src[x+srcStride  ])>>2;
+			dst[2*x          +2]= (3*src[x+1] +   src[x+srcStride  ])>>2;
+		}
+#endif
+		dst[srcWidth*2 -1]= 
+		dst[srcWidth*2 -1 + dstStride]= src[srcWidth-1];
+
+		dst+=dstStride*2;
+		src+=srcStride;
+	}
+	src-=srcStride;
+	
+	// last line
+	for(x=0; x<srcWidth; x++){
+		dst[2*x+0]=
+		dst[2*x+1]= src[x];
+	}
+#ifdef HAVE_MMX
+asm volatile(   EMMS" \n\t"
+        	SFENCE" \n\t"
+        	:::"memory");
+#endif
+}
+
 /**
  *
  * height should be a multiple of 2 and width should be a multiple of 16 (if this is a

postproc/swscale.c

@@ -1777,6 +1777,34 @@ static void bgr24toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], i
 		dstStride[0], dstStride[1], srcStride[0]);
 }
 
+static void yvu9toyv12Wrapper(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY,
+             int srcSliceH, uint8_t* dst[], int dstStride[]){
+	int i;
+
+	/* copy Y */
+	if(srcStride[0]==dstStride[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]);
+	}
+}
+
 /**
  * bring pointers in YUV order instead of YVU
  */
@@ -2051,7 +2079,7 @@ SwsContext *getSwsContext(int srcW, int srcH, int srcFormat, int dstW, int dstH,
 					vo_format_name(srcFormat), vo_format_name(dstFormat));
 			return c;
 		}
-#if 1
+		
 		/* simple copy */
 		if(   srcFormat == dstFormat
 		   || (srcFormat==IMGFMT_YV12 && dstFormat==IMGFMT_I420)
@@ -2067,7 +2095,17 @@ SwsContext *getSwsContext(int srcW, int srcH, int srcFormat, int dstW, int dstH,
 					vo_format_name(srcFormat), vo_format_name(dstFormat));
 			return c;
 		}
-#endif
+		
+		if( srcFormat==IMGFMT_YVU9 && (dstFormat==IMGFMT_YV12 || dstFormat==IMGFMT_I420) )
+		{
+			c->swScale= yvu9toyv12Wrapper;
+
+			if(flags&SWS_PRINT_INFO)
+				MSG_INFO("SwScaler: using unscaled %s -> %s special converter\n", 
+					vo_format_name(srcFormat), vo_format_name(dstFormat));
+			return c;
+		}
+
 		/* bgr32to24 & rgb32to24*/
 		if((srcFormat==IMGFMT_BGR32 && dstFormat==IMGFMT_BGR24)
 		 ||(srcFormat==IMGFMT_RGB32 && dstFormat==IMGFMT_RGB24))