[Mplayer-cvslog] CVS: main/postproc swscale.c,1.77,1.78 swscale_template.c,1.69,1.70

[Michael Niedermayer]

Update of /cvsroot/mplayer/main/postproc
In directory mplayer:/var/tmp.root/cvs-serv17461

Modified Files:
	swscale.c swscale_template.c 
Log Message:
bgr24toY in MMX


Index: swscale.c
===================================================================
RCS file: /cvsroot/mplayer/main/postproc/swscale.c,v
retrieving revision 1.77
retrieving revision 1.78
diff -u -r1.77 -r1.78
--- swscale.c	8 Feb 2002 15:54:56 -0000	1.77
+++ swscale.c	9 Feb 2002 17:03:53 -0000	1.78
@@ -68,6 +68,8 @@
 //#undef ARCH_X86
 #define DITHER1XBPP
 
+#define FAST_BGR2YV12 // use 7 bit coeffs instead of 15bit
+
 #define RET 0xC3 //near return opcode for X86
 
 #ifdef MP_DEBUG
@@ -177,6 +179,14 @@
 static uint64_t __attribute__((aligned(8))) M24A=   0x00FF0000FF0000FFLL;
 static uint64_t __attribute__((aligned(8))) M24B=   0xFF0000FF0000FF00LL;
 static uint64_t __attribute__((aligned(8))) M24C=   0x0000FF0000FF0000LL;
+
+#ifdef FAST_BGR2YV12
+static const uint64_t bgr2YCoeff  __attribute__((aligned(8))) = 0x000000210041000DULL;
+#else
+static const uint64_t bgr2YCoeff  __attribute__((aligned(8))) = 0x000020E540830C8BULL;
+#endif
+static const uint64_t bgr2YOffset __attribute__((aligned(8))) = 0x1010101010101010ULL;
+static const uint64_t w1111       __attribute__((aligned(8))) = 0x0001000100010001ULL;
 
 // FIXME remove
 static uint64_t __attribute__((aligned(8))) asm_yalpha1;

Index: swscale_template.c
===================================================================
RCS file: /cvsroot/mplayer/main/postproc/swscale_template.c,v
retrieving revision 1.69
retrieving revision 1.70
diff -u -r1.69 -r1.70
--- swscale_template.c	8 Feb 2002 03:09:46 -0000	1.69
+++ swscale_template.c	9 Feb 2002 17:03:53 -0000	1.70
@@ -1635,7 +1635,76 @@
 
 static inline void RENAME(bgr24ToY)(uint8_t *dst, uint8_t *src, int width)
 {
-#ifdef HAVE_MMXFIXME
+#ifdef HAVE_MMX
+	asm volatile(
+		"movl %2, %%eax			\n\t"
+		"movq bgr2YCoeff, %%mm6		\n\t"
+		"movq w1111, %%mm5		\n\t"
+		"pxor %%mm7, %%mm7		\n\t"
+		"leal (%%eax, %%eax, 2), %%ebx	\n\t"
+		".balign 16			\n\t"
+		"1:				\n\t"
+		PREFETCH" 64(%0, %%ebx)		\n\t"
+		"movd (%0, %%ebx), %%mm0	\n\t"
+		"movd 3(%0, %%ebx), %%mm1	\n\t"
+		"punpcklbw %%mm7, %%mm0		\n\t"
+		"punpcklbw %%mm7, %%mm1		\n\t"
+		"movd 6(%0, %%ebx), %%mm2	\n\t"
+		"movd 9(%0, %%ebx), %%mm3	\n\t"
+		"punpcklbw %%mm7, %%mm2		\n\t"
+		"punpcklbw %%mm7, %%mm3		\n\t"
+		"pmaddwd %%mm6, %%mm0		\n\t"
+		"pmaddwd %%mm6, %%mm1		\n\t"
+		"pmaddwd %%mm6, %%mm2		\n\t"
+		"pmaddwd %%mm6, %%mm3		\n\t"
+#ifndef FAST_BGR2YV12
+		"psrad $8, %%mm0		\n\t"
+		"psrad $8, %%mm1		\n\t"
+		"psrad $8, %%mm2		\n\t"
+		"psrad $8, %%mm3		\n\t"
+#endif
+		"packssdw %%mm1, %%mm0		\n\t"
+		"packssdw %%mm3, %%mm2		\n\t"
+		"pmaddwd %%mm5, %%mm0		\n\t"
+		"pmaddwd %%mm5, %%mm2		\n\t"
+		"packssdw %%mm2, %%mm0		\n\t"
+		"psraw $7, %%mm0		\n\t"
+
+		"movd 12(%0, %%ebx), %%mm4	\n\t"
+		"movd 15(%0, %%ebx), %%mm1	\n\t"
+		"punpcklbw %%mm7, %%mm4		\n\t"
+		"punpcklbw %%mm7, %%mm1		\n\t"
+		"movd 18(%0, %%ebx), %%mm2	\n\t"
+		"movd 21(%0, %%ebx), %%mm3	\n\t"
+		"punpcklbw %%mm7, %%mm2		\n\t"
+		"punpcklbw %%mm7, %%mm3		\n\t"
+		"pmaddwd %%mm6, %%mm4		\n\t"
+		"pmaddwd %%mm6, %%mm1		\n\t"
+		"pmaddwd %%mm6, %%mm2		\n\t"
+		"pmaddwd %%mm6, %%mm3		\n\t"
+#ifndef FAST_BGR2YV12
+		"psrad $8, %%mm4		\n\t"
+		"psrad $8, %%mm1		\n\t"
+		"psrad $8, %%mm2		\n\t"
+		"psrad $8, %%mm3		\n\t"
+#endif
+		"packssdw %%mm1, %%mm4		\n\t"
+		"packssdw %%mm3, %%mm2		\n\t"
+		"pmaddwd %%mm5, %%mm4		\n\t"
+		"pmaddwd %%mm5, %%mm2		\n\t"
+		"addl $24, %%ebx		\n\t"
+		"packssdw %%mm2, %%mm4		\n\t"
+		"psraw $7, %%mm4		\n\t"
+
+		"packuswb %%mm4, %%mm0		\n\t"
+		"paddusb bgr2YOffset, %%mm0	\n\t"
+
+		MOVNTQ(%%mm0, (%1, %%eax))
+		"addl $8, %%eax			\n\t"
+		" js 1b				\n\t"
+		: : "r" (src+width*3), "r" (dst+width), "g" (-width)
+		: "%eax", "%ebx"
+	);
 #else
 	int i;
 	for(i=0; i<width; i++)