qpel in mmx2/3dnow

qpel refinement quality parameter

Originally committed as revision 1393 to svn://svn.ffmpeg.org/ffmpeg/trunk

1 files changed, 362 insertions, 1 deletions

diff --git a/libavcodec/i386/dsputil_mmx_rnd.h b/libavcodec/i386/dsputil_mmx_rnd.h
index 3605e03f9c..f72ded1bcf 100644
--- a/libavcodec/i386/dsputil_mmx_rnd.h
+++ b/libavcodec/i386/dsputil_mmx_rnd.h
@@ -54,6 +54,42 @@ static void DEF(put, pixels8_x2)(UINT8 *block, const UINT8 *pixels, int line_siz
 	:"eax", "memory");
 }
 
+static void DEF(put, pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
+{
+    MOVQ_BFE(mm6);
+    __asm __volatile(
+	".balign 8			\n\t"
+	"1:				\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"movq	(%2), %%mm1		\n\t"
+	"addl	%4, %1			\n\t"
+	"movq	(%1), %%mm2		\n\t"
+	"movq	8(%2), %%mm3		\n\t"
+	"addl	%4, %1			\n\t"
+	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)
+	"movq	%%mm4, (%3)		\n\t"
+	"addl	%5, %3			\n\t"
+	"movq	%%mm5, (%3)		\n\t"
+	"addl	%5, %3			\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"movq	16(%2), %%mm1		\n\t"
+	"addl	%4, %1			\n\t"
+	"movq	(%1), %%mm2		\n\t"
+	"movq	24(%2), %%mm3		\n\t"
+	"addl	%4, %1			\n\t"
+	"addl	$32, %2			\n\t"
+	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)
+	"movq	%%mm4, (%3)		\n\t"
+	"addl	%5, %3			\n\t"
+	"movq	%%mm5, (%3)		\n\t"
+	"addl	%5, %3			\n\t"
+	"subl	$4, %0			\n\t"
+	"jnz	1b			\n\t"
+	:"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
+	:"r"(src1Stride), "r"(dstStride)
+	:"memory");
+}
+
 static void DEF(put, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
 {
     MOVQ_BFE(mm6);
@@ -90,7 +126,7 @@ static void DEF(put, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_si
 	"movq	9(%1, %3), %%mm3	\n\t"
 	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)
 	"movq	%%mm4, 8(%2)		\n\t"
-	"movq	%%mm5, 8(%2, %3)		\n\t"
+	"movq	%%mm5, 8(%2, %3)	\n\t"
 	"addl	%%eax, %1		\n\t"
 	"addl	%%eax, %2		\n\t"
 	"subl	$4, %0			\n\t"
@@ -100,6 +136,38 @@ static void DEF(put, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_si
 	:"eax", "memory");
 }
 
+static void DEF(put, pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
+{
+    MOVQ_BFE(mm6);
+    __asm __volatile(
+	".balign 8			\n\t"
+	"1:				\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"movq	(%2), %%mm1		\n\t"
+	"movq	8(%1), %%mm2		\n\t"
+	"movq	8(%2), %%mm3		\n\t"
+	"addl	%4, %1			\n\t"
+	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)
+	"movq	%%mm4, (%3)		\n\t"
+	"movq	%%mm5, 8(%3)		\n\t"
+	"addl	%5, %3			\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"movq	16(%2), %%mm1		\n\t"
+	"movq	8(%1), %%mm2		\n\t"
+	"movq	24(%2), %%mm3		\n\t"
+	"addl	%4, %1			\n\t"
+	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)
+	"movq	%%mm4, (%3)		\n\t"
+	"movq	%%mm5, 8(%3)		\n\t"
+	"addl	%5, %3			\n\t"
+	"addl	$32, %2			\n\t"
+	"subl	$2, %0			\n\t"
+	"jnz	1b			\n\t"
+	:"+g"(h), "+r"(src1), "+r"(src2), "+r"(dst)
+	:"r"(src1Stride), "r"(dstStride)
+	:"memory");
+}
+
 static void DEF(put, pixels8_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
 {
     MOVQ_BFE(mm6);
@@ -195,6 +263,124 @@ static void DEF(put, pixels8_xy2)(UINT8 *block, const UINT8 *pixels, int line_si
 	:"eax", "memory");
 }
 
+static void DEF(put, pixels8_l4)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int stride, int h)
+{
+    MOVQ_ZERO(mm7);
+    SET_RND(mm6); // =2 for rnd  and  =1 for no_rnd version
+    __asm __volatile(
+	".balign 8      		\n\t"
+	"1:				\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"movq	(%2), %%mm1		\n\t"
+	"movq	64(%2), %%mm2		\n\t"
+	"movq	136(%2), %%mm3		\n\t"
+	"punpcklbw %%mm7, %%mm0		\n\t"
+	"punpcklbw %%mm7, %%mm1		\n\t"
+	"punpcklbw %%mm7, %%mm2		\n\t"
+	"punpcklbw %%mm7, %%mm3		\n\t"
+	"paddusw %%mm6, %%mm0		\n\t"
+	"paddusw %%mm0, %%mm1		\n\t"
+	"paddusw %%mm2, %%mm3		\n\t"
+	"paddusw %%mm1, %%mm3		\n\t"
+	"psrlw	$2, %%mm3		\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"movq	(%2), %%mm1		\n\t"
+	"movq	64(%2), %%mm2		\n\t"
+	"movq	136(%2), %%mm4		\n\t"
+	"punpckhbw %%mm7, %%mm0		\n\t"
+	"punpckhbw %%mm7, %%mm1		\n\t"
+	"punpckhbw %%mm7, %%mm2		\n\t"
+	"punpckhbw %%mm7, %%mm4		\n\t"
+	"paddusw %%mm6, %%mm0		\n\t"
+	"paddusw %%mm0, %%mm1		\n\t"
+	"paddusw %%mm2, %%mm4		\n\t"
+	"paddusw %%mm1, %%mm4		\n\t"
+	"psrlw	$2, %%mm4		\n\t"
+	"packuswb  %%mm4, %%mm3		\n\t"
+	"movq	%%mm3, (%0)		\n\t"
+        "addl	%4, %0			\n\t"
+        "addl	%4, %1			\n\t"
+        "addl	$8, %2			\n\t" 
+        "decl	%3			\n\t"
+	"jnz	1b			\n\t"
+	:"+r"(dst), "+r"(src1), "+r"(src2), "+r"(h)
+	:"r"(stride)
+	:"memory");
+}
+
+static void DEF(put, pixels16_l4)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int stride, int h)
+{
+    MOVQ_ZERO(mm7);
+    SET_RND(mm6); // =2 for rnd  and  =1 for no_rnd version
+    __asm __volatile(
+	".balign 8      		\n\t"
+	"1:				\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"movq	(%2), %%mm1		\n\t"
+	"movq	256(%2), %%mm2		\n\t"
+	"movq	528(%2), %%mm3		\n\t"
+	"punpcklbw %%mm7, %%mm0		\n\t"
+	"punpcklbw %%mm7, %%mm1		\n\t"
+	"punpcklbw %%mm7, %%mm2		\n\t"
+	"punpcklbw %%mm7, %%mm3		\n\t"
+	"paddusw %%mm6, %%mm0		\n\t"
+	"paddusw %%mm0, %%mm1		\n\t"
+	"paddusw %%mm2, %%mm3		\n\t"
+	"paddusw %%mm1, %%mm3		\n\t"
+	"psrlw	$2, %%mm3		\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"movq	(%2), %%mm1		\n\t"
+	"movq	256(%2), %%mm2		\n\t"
+	"movq	528(%2), %%mm4		\n\t"
+	"punpckhbw %%mm7, %%mm0		\n\t"
+	"punpckhbw %%mm7, %%mm1		\n\t"
+	"punpckhbw %%mm7, %%mm2		\n\t"
+	"punpckhbw %%mm7, %%mm4		\n\t"
+	"paddusw %%mm6, %%mm0		\n\t"
+	"paddusw %%mm0, %%mm1		\n\t"
+	"paddusw %%mm2, %%mm4		\n\t"
+	"paddusw %%mm1, %%mm4		\n\t"
+	"psrlw	$2, %%mm4		\n\t"
+	"packuswb  %%mm4, %%mm3		\n\t"
+	"movq	%%mm3, (%0)		\n\t"
+	"movq	8(%1), %%mm0		\n\t"
+	"movq	8(%2), %%mm1		\n\t"
+	"movq	264(%2), %%mm2		\n\t"
+	"movq	536(%2), %%mm3		\n\t"
+	"punpcklbw %%mm7, %%mm0		\n\t"
+	"punpcklbw %%mm7, %%mm1		\n\t"
+	"punpcklbw %%mm7, %%mm2		\n\t"
+	"punpcklbw %%mm7, %%mm3		\n\t"
+	"paddusw %%mm6, %%mm0		\n\t"
+	"paddusw %%mm0, %%mm1		\n\t"
+	"paddusw %%mm2, %%mm3		\n\t"
+	"paddusw %%mm1, %%mm3		\n\t"
+	"psrlw	$2, %%mm3		\n\t"
+	"movq	8(%1), %%mm0		\n\t"
+	"movq	8(%2), %%mm1		\n\t"
+	"movq	264(%2), %%mm2		\n\t"
+	"movq	536(%2), %%mm4		\n\t"
+	"punpckhbw %%mm7, %%mm0		\n\t"
+	"punpckhbw %%mm7, %%mm1		\n\t"
+	"punpckhbw %%mm7, %%mm2		\n\t"
+	"punpckhbw %%mm7, %%mm4		\n\t"
+	"paddusw %%mm6, %%mm0		\n\t"
+	"paddusw %%mm0, %%mm1		\n\t"
+	"paddusw %%mm2, %%mm4		\n\t"
+	"paddusw %%mm1, %%mm4		\n\t"
+	"psrlw	$2, %%mm4		\n\t"
+	"packuswb  %%mm4, %%mm3		\n\t"
+	"movq	%%mm3, 8(%0)		\n\t"
+        "addl	%4, %0			\n\t"
+        "addl	%4, %1			\n\t"
+        "addl	$16, %2			\n\t" 
+        "decl	%3			\n\t"
+	"jnz	1b			\n\t"
+	:"+r"(dst), "+r"(src1), "+r"(src2), "+r"(h)
+	:"r"(stride)
+	:"memory");
+}
+
 // avg_pixels
 // in case more speed is needed - unroling would certainly help
 static void DEF(avg, pixels8)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
@@ -259,6 +445,27 @@ static void DEF(avg, pixels8_x2)(UINT8 *block, const UINT8 *pixels, int line_siz
     } while (--h);
 }
 
+static void DEF(avg, pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
+{
+    MOVQ_BFE(mm6);
+    JUMPALIGN();
+    do {
+	__asm __volatile(
+	    "movq  %1, %%mm0		\n\t"
+	    "movq  %2, %%mm1		\n\t"
+	    "movq  %0, %%mm3		\n\t"
+	    PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
+	    PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
+	    "movq  %%mm0, %0		\n\t"
+	    :"+m"(*dst)
+	    :"m"(*src1), "m"(*src2)
+	    :"memory");
+	dst += dstStride;
+        src1 += src1Stride;
+        src2 += 8;
+    } while (--h);
+}
+
 static void DEF(avg, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
 {
     MOVQ_BFE(mm6);
@@ -285,6 +492,33 @@ static void DEF(avg, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_si
     } while (--h);
 }
 
+static void DEF(avg, pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
+{
+    MOVQ_BFE(mm6);
+    JUMPALIGN();
+    do {
+	__asm __volatile(
+	    "movq  %1, %%mm0		\n\t"
+	    "movq  %2, %%mm1		\n\t"
+	    "movq  %0, %%mm3		\n\t"
+	    PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
+	    PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
+	    "movq  %%mm0, %0		\n\t"
+	    "movq  8%1, %%mm0		\n\t"
+	    "movq  8%2, %%mm1		\n\t"
+	    "movq  8%0, %%mm3		\n\t"
+	    PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)
+	    PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)
+	    "movq  %%mm0, 8%0		\n\t"
+	    :"+m"(*dst)
+	    :"m"(*src1), "m"(*src2)
+	    :"memory");
+	dst += dstStride;
+        src1 += src1Stride;
+        src2 += 16;
+    } while (--h);
+}
+
 static void DEF(avg, pixels8_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
 {
     MOVQ_BFE(mm6);
@@ -399,6 +633,133 @@ static void DEF(avg, pixels8_xy2)(UINT8 *block, const UINT8 *pixels, int line_si
 	:"eax", "memory");
 }
 
+static void DEF(avg, pixels8_l4)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int stride, int h)
+{
+    MOVQ_ZERO(mm7);
+    SET_RND(mm6); // =2 for rnd  and  =1 for no_rnd version
+    MOVQ_BFE(mm5);
+    __asm __volatile(
+	".balign 8      		\n\t"
+	"1:				\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"movq	(%2), %%mm1		\n\t"
+	"movq	64(%2), %%mm2		\n\t"
+	"movq	136(%2), %%mm3		\n\t"
+	"punpcklbw %%mm7, %%mm0		\n\t"
+	"punpcklbw %%mm7, %%mm1		\n\t"
+	"punpcklbw %%mm7, %%mm2		\n\t"
+	"punpcklbw %%mm7, %%mm3		\n\t"
+	"paddusw %%mm6, %%mm0		\n\t"
+	"paddusw %%mm0, %%mm1		\n\t"
+	"paddusw %%mm2, %%mm3		\n\t"
+	"paddusw %%mm1, %%mm3		\n\t"
+	"psrlw	$2, %%mm3		\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"movq	(%2), %%mm1		\n\t"
+	"movq	64(%2), %%mm2		\n\t"
+	"movq	136(%4), %%mm4		\n\t"
+	"punpckhbw %%mm7, %%mm0		\n\t"
+	"punpckhbw %%mm7, %%mm1		\n\t"
+	"punpckhbw %%mm7, %%mm2		\n\t"
+	"punpckhbw %%mm7, %%mm4		\n\t"
+	"paddusw %%mm6, %%mm0		\n\t"
+	"paddusw %%mm0, %%mm1		\n\t"
+	"paddusw %%mm2, %%mm4		\n\t"
+	"paddusw %%mm1, %%mm4		\n\t"
+	"psrlw	$2, %%mm4		\n\t"
+	"packuswb  %%mm4, %%mm3		\n\t"
+	"movq	(%0), %%mm4		\n\t"
+        PAVGB(%%mm3, %%mm4, %%mm0, %%mm5)
+	"movq	%%mm3, (%0)		\n\t"
+        "addl	%4, %0			\n\t"
+        "addl	%4, %1			\n\t"
+        "addl	$8, %2			\n\t" 
+        "decl	%3			\n\t"
+	"jnz	1b			\n\t"
+	:"+r"(dst), "+r"(src1), "+r"(src2), "+r"(h)
+	:"r"(stride)
+	:"memory");
+}
+
+static void DEF(avg, pixels16_l4)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int stride, int h)
+{
+    MOVQ_ZERO(mm7);
+    SET_RND(mm6); // =2 for rnd  and  =1 for no_rnd version
+    MOVQ_BFE(mm5);
+    __asm __volatile(
+	".balign 8      		\n\t"
+	"1:				\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"movq	(%2), %%mm1		\n\t"
+	"movq	256(%2), %%mm2		\n\t"
+	"movq	528(%2), %%mm3		\n\t"
+	"punpcklbw %%mm7, %%mm0		\n\t"
+	"punpcklbw %%mm7, %%mm1		\n\t"
+	"punpcklbw %%mm7, %%mm2		\n\t"
+	"punpcklbw %%mm7, %%mm3		\n\t"
+	"paddusw %%mm6, %%mm0		\n\t"
+	"paddusw %%mm0, %%mm1		\n\t"
+	"paddusw %%mm2, %%mm3		\n\t"
+	"paddusw %%mm1, %%mm3		\n\t"
+	"psrlw	$2, %%mm3		\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"movq	(%2), %%mm1		\n\t"
+	"movq	256(%2), %%mm2		\n\t"
+	"movq	528(%4), %%mm4		\n\t"
+	"punpckhbw %%mm7, %%mm0		\n\t"
+	"punpckhbw %%mm7, %%mm1		\n\t"
+	"punpckhbw %%mm7, %%mm2		\n\t"
+	"punpckhbw %%mm7, %%mm4		\n\t"
+	"paddusw %%mm6, %%mm0		\n\t"
+	"paddusw %%mm0, %%mm1		\n\t"
+	"paddusw %%mm2, %%mm4		\n\t"
+	"paddusw %%mm1, %%mm4		\n\t"
+	"psrlw	$2, %%mm4		\n\t"
+	"packuswb  %%mm4, %%mm3		\n\t"
+	"movq	(%0), %%mm4		\n\t"
+        PAVGB(%%mm3, %%mm4, %%mm0, %%mm5)
+	"movq	%%mm3, (%0)		\n\t"
+	"movq	8(%1), %%mm0		\n\t"
+	"movq	8(%2), %%mm1		\n\t"
+	"movq	264(%2), %%mm2		\n\t"
+	"movq	536(%2), %%mm3		\n\t"
+	"punpcklbw %%mm7, %%mm0		\n\t"
+	"punpcklbw %%mm7, %%mm1		\n\t"
+	"punpcklbw %%mm7, %%mm2		\n\t"
+	"punpcklbw %%mm7, %%mm3		\n\t"
+	"paddusw %%mm6, %%mm0		\n\t"
+	"paddusw %%mm0, %%mm1		\n\t"
+	"paddusw %%mm2, %%mm3		\n\t"
+	"paddusw %%mm1, %%mm3		\n\t"
+	"psrlw	$2, %%mm3		\n\t"
+	"movq	8(%1), %%mm0		\n\t"
+	"movq	8(%2), %%mm1		\n\t"
+	"movq	264(%2), %%mm2		\n\t"
+	"movq	536(%4), %%mm4		\n\t"
+	"punpckhbw %%mm7, %%mm0		\n\t"
+	"punpckhbw %%mm7, %%mm1		\n\t"
+	"punpckhbw %%mm7, %%mm2		\n\t"
+	"punpckhbw %%mm7, %%mm4		\n\t"
+	"paddusw %%mm6, %%mm0		\n\t"
+	"paddusw %%mm0, %%mm1		\n\t"
+	"paddusw %%mm2, %%mm4		\n\t"
+	"paddusw %%mm1, %%mm4		\n\t"
+	"psrlw	$2, %%mm4		\n\t"
+	"packuswb  %%mm4, %%mm3		\n\t"
+	"movq	8(%0), %%mm4		\n\t"
+        PAVGB(%%mm3, %%mm4, %%mm0, %%mm5)
+	"movq	%%mm3, 8(%0)		\n\t"
+        "addl	%4, %0			\n\t"
+        "addl	%4, %1			\n\t"
+        "addl	$16, %2			\n\t" 
+        "decl	%3			\n\t"
+	"jnz	1b			\n\t"
+	:"+r"(dst), "+r"(src1), "+r"(src2), "+r"(h)
+	:"r"(stride)
+	:"memory");
+}
+
+
 //FIXME optimize
 static void DEF(put, pixels16_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h){
     DEF(put, pixels8_y2)(block  , pixels  , line_size, h);