fixed yv12toyuy2

Originally committed as revision 2724 to svn://svn.mplayerhq.hu/mplayer/trunk/postproc

postproc/rgb2rgb.c

@@ -291,11 +291,17 @@ void palette8torgb15(const uint8_t *src, uint8_t *dst, unsigned num_pixels, cons
 }
 /**
  *
- * num_pixels must be a multiple of 16 for the MMX version
+ * width must be a multiple of 16 for the MMX version
  */
-void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst, unsigned num_pixels)
+void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
+	int width, int height, int lumStride, int chromStride, int dstStride)
 {
+	int y;
+	const int chromWidth= width>>1;
+	for(y=0; y<height; y++)
+	{
 #ifdef HAVE_MMX
+//FIXME handle 2 lines a once (fewer prefetch, reuse some chrom, but very likely limited by mem anyway)
 		asm volatile(
 			"xorl %%eax, %%eax		\n\t"
 			"1:				\n\t"
@@ -325,22 +331,31 @@ void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, u
 			"addl $8, %%eax			\n\t"
 			"cmpl %4, %%eax			\n\t"
 			" jb 1b				\n\t"
-		EMMS" \n\t"
-		SFENCE
-		::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "r" (num_pixels>>1)
-		: "memory", "%eax"
+			::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "r" (chromWidth)
+			: "%eax"
 		);
-
 #else
 		int i;
-	num_pixels>>=1;
-	for(i=0; i<num_pixels; i++)
+		for(i=0; i<chromWidth; i++)
 		{
 			dst[4*i+0] = ysrc[2*i+0];
 			dst[4*i+1] = usrc[i];
 			dst[4*i+2] = ysrc[2*i+1];
 			dst[4*i+3] = vsrc[i];
 		}
+#endif
+		if(y&1)
+		{
+			usrc += chromStride;
+			vsrc += chromStride;
+		}
+		ysrc += lumStride;
+		dst += dstStride;
+	}
+#ifdef HAVE_MMX
+asm(    EMMS" \n\t"
+        SFENCE" \n\t"
+        :::"memory");
 #endif
 }
 

postproc/rgb2rgb.h

@@ -21,8 +21,8 @@ extern void palette8torgb16(const uint8_t *src, uint8_t *dst, unsigned num_pixel
 extern void palette8torgb15(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette);
 extern void palette8torgb24(const uint8_t *src, uint8_t *dst, unsigned num_pixels, const uint8_t *palette);
 
-extern void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst, unsigned num_pixels);
+extern void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
+	int width, int height, int lumStride, int chromStride, int dstStride);
 extern void yuy2toyv12(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst, unsigned num_pixels);
 
-
 #endif

postproc/rgb2rgb_template.c

@@ -291,11 +291,17 @@ void palette8torgb15(const uint8_t *src, uint8_t *dst, unsigned num_pixels, cons
 }
 /**
  *
- * num_pixels must be a multiple of 16 for the MMX version
+ * width must be a multiple of 16 for the MMX version
  */
-void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst, unsigned num_pixels)
+void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,
+	int width, int height, int lumStride, int chromStride, int dstStride)
 {
+	int y;
+	const int chromWidth= width>>1;
+	for(y=0; y<height; y++)
+	{
 #ifdef HAVE_MMX
+//FIXME handle 2 lines a once (fewer prefetch, reuse some chrom, but very likely limited by mem anyway)
 		asm volatile(
 			"xorl %%eax, %%eax		\n\t"
 			"1:				\n\t"
@@ -325,22 +331,31 @@ void yv12toyuy2(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, u
 			"addl $8, %%eax			\n\t"
 			"cmpl %4, %%eax			\n\t"
 			" jb 1b				\n\t"
-		EMMS" \n\t"
-		SFENCE
-		::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "r" (num_pixels>>1)
-		: "memory", "%eax"
+			::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "r" (chromWidth)
+			: "%eax"
 		);
-
 #else
 		int i;
-	num_pixels>>=1;
-	for(i=0; i<num_pixels; i++)
+		for(i=0; i<chromWidth; i++)
 		{
 			dst[4*i+0] = ysrc[2*i+0];
 			dst[4*i+1] = usrc[i];
 			dst[4*i+2] = ysrc[2*i+1];
 			dst[4*i+3] = vsrc[i];
 		}
+#endif
+		if(y&1)
+		{
+			usrc += chromStride;
+			vsrc += chromStride;
+		}
+		ysrc += lumStride;
+		dst += dstStride;
+	}
+#ifdef HAVE_MMX
+asm(    EMMS" \n\t"
+        SFENCE" \n\t"
+        :::"memory");
 #endif
 }