qpel fix

Originally committed as revision 1426 to svn://svn.ffmpeg.org/ffmpeg/trunk

2 files changed, 89 insertions, 331 deletions

diff --git a/libavcodec/i386/dsputil_mmx.c b/libavcodec/i386/dsputil_mmx.c
index 5fce7f914c..595881f784 100644
--- a/libavcodec/i386/dsputil_mmx.c
+++ b/libavcodec/i386/dsputil_mmx.c
@@ -1085,7 +1085,7 @@ static void OPNAME ## qpel8_mc00_ ## MMX (UINT8 *dst, UINT8 *src, int stride){\
 }\
 \
 static void OPNAME ## qpel8_mc10_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t temp[32];\
+    uint64_t temp[8];\
     uint8_t * const half= (uint8_t*)temp;\
     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
     OPNAME ## pixels8_l2_mmx(dst, src, half, stride, stride, 8);\
@@ -1096,14 +1096,14 @@ static void OPNAME ## qpel8_mc20_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
 }\
 \
 static void OPNAME ## qpel8_mc30_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t temp[32];\
+    uint64_t temp[8];\
     uint8_t * const half= (uint8_t*)temp;\
     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(half, src, 8, stride, 8);\
     OPNAME ## pixels8_l2_mmx(dst, src+1, half, stride, stride, 8);\
 }\
 \
 static void OPNAME ## qpel8_mc01_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t temp[32];\
+    uint64_t temp[8];\
     uint8_t * const half= (uint8_t*)temp;\
     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
     OPNAME ## pixels8_l2_mmx(dst, src, half, stride, stride, 8);\
@@ -1114,53 +1114,49 @@ static void OPNAME ## qpel8_mc02_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
 }\
 \
 static void OPNAME ## qpel8_mc03_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t temp[32];\
+    uint64_t temp[8];\
     uint8_t * const half= (uint8_t*)temp;\
     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(half, src, 8, stride);\
     OPNAME ## pixels8_l2_mmx(dst, src+stride, half, stride, stride, 8);\
 }\
 static void OPNAME ## qpel8_mc11_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t half[8*2 + 8*2 + 18*2];\
-    uint8_t * const halfH= ((uint8_t*)half) + 2*64 + 8;\
-    uint8_t * const halfV= ((uint8_t*)half);\
-    uint8_t * const halfHV= ((uint8_t*)half) + 64;\
+    uint64_t half[8 + 9];\
+    uint8_t * const halfH= ((uint8_t*)half) + 64;\
+    uint8_t * const halfHV= ((uint8_t*)half);\
     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfV, src, 8, stride);\
+    put ## RND ## pixels8_l2_mmx(halfH, src, halfH, 8, stride, 9);\
     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
-    OPNAME ## pixels8_l4_mmx(dst, src, (uint8_t*)half, stride, 8);\
+    OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\
 }\
 static void OPNAME ## qpel8_mc31_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t half[8*2 + 8*2 + 18*2];\
-    uint8_t * const halfH= ((uint8_t*)half) + 2*64 + 8;\
-    uint8_t * const halfV= ((uint8_t*)half);\
-    uint8_t * const halfHV= ((uint8_t*)half) + 64;\
+    uint64_t half[8 + 9];\
+    uint8_t * const halfH= ((uint8_t*)half) + 64;\
+    uint8_t * const halfHV= ((uint8_t*)half);\
     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfV, src+1, 8, stride);\
+    put ## RND ## pixels8_l2_mmx(halfH, src+1, halfH, 8, stride, 9);\
     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
-    OPNAME ## pixels8_l4_mmx(dst, src+1, (uint8_t*)half, stride, 8);\
+    OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\
 }\
 static void OPNAME ## qpel8_mc13_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t half[8*2 + 8*2 + 9*2];\
-    uint8_t * const halfH= ((uint8_t*)half) + 2*64;\
-    uint8_t * const halfV= ((uint8_t*)half);\
-    uint8_t * const halfHV= ((uint8_t*)half) + 64;\
+    uint64_t half[8 + 9];\
+    uint8_t * const halfH= ((uint8_t*)half) + 64;\
+    uint8_t * const halfHV= ((uint8_t*)half);\
     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfV, src, 8, stride);\
+    put ## RND ## pixels8_l2_mmx(halfH, src, halfH, 8, stride, 9);\
     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
-    OPNAME ## pixels8_l4_mmx(dst, src+stride, (uint8_t*)half, stride, 8);\
+    OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\
 }\
 static void OPNAME ## qpel8_mc33_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t half[8*2 + 8*2 + 9*2];\
-    uint8_t * const halfH= ((uint8_t*)half) + 2*64;\
-    uint8_t * const halfV= ((uint8_t*)half);\
-    uint8_t * const halfHV= ((uint8_t*)half) + 64;\
-    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src  , 8, stride, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfV, src+1, 8, stride);\
+    uint64_t half[8 + 9];\
+    uint8_t * const halfH= ((uint8_t*)half) + 64;\
+    uint8_t * const halfHV= ((uint8_t*)half);\
+    put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
+    put ## RND ## pixels8_l2_mmx(halfH, src+1, halfH, 8, stride, 9);\
     put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
-    OPNAME ## pixels8_l4_mmx(dst, src+stride+1, (uint8_t*)half, stride, 8);\
+    OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\
 }\
 static void OPNAME ## qpel8_mc21_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t half[8*2 + 9*2];\
+    uint64_t half[8 + 9];\
     uint8_t * const halfH= ((uint8_t*)half) + 64;\
     uint8_t * const halfHV= ((uint8_t*)half);\
     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
@@ -1168,7 +1164,7 @@ static void OPNAME ## qpel8_mc21_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
     OPNAME ## pixels8_l2_mmx(dst, halfH, halfHV, stride, 8, 8);\
 }\
 static void OPNAME ## qpel8_mc23_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t half[8*2 + 9*2];\
+    uint64_t half[8 + 9];\
     uint8_t * const halfH= ((uint8_t*)half) + 64;\
     uint8_t * const halfHV= ((uint8_t*)half);\
     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
@@ -1176,27 +1172,21 @@ static void OPNAME ## qpel8_mc23_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
     OPNAME ## pixels8_l2_mmx(dst, halfH+8, halfHV, stride, 8, 8);\
 }\
 static void OPNAME ## qpel8_mc12_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t half[8*2 + 8*2 + 9*2];\
-    uint8_t * const halfH= ((uint8_t*)half) + 2*64;\
-    uint8_t * const halfV= ((uint8_t*)half);\
-    uint8_t * const halfHV= ((uint8_t*)half) + 64;\
+    uint64_t half[8 + 9];\
+    uint8_t * const halfH= ((uint8_t*)half);\
     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfV, src, 8, stride);\
-    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
-    OPNAME ## pixels8_l2_mmx(dst, halfV, halfHV, stride, 8, 8);\
+    put ## RND ## pixels8_l2_mmx(halfH, src, halfH, 8, stride, 9);\
+    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
 }\
 static void OPNAME ## qpel8_mc32_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t half[8*2 + 8*2 + 9*2];\
-    uint8_t * const halfH= ((uint8_t*)half) + 2*64;\
-    uint8_t * const halfV= ((uint8_t*)half);\
-    uint8_t * const halfHV= ((uint8_t*)half) + 64;\
+    uint64_t half[8 + 9];\
+    uint8_t * const halfH= ((uint8_t*)half);\
     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfV, src+1, 8, stride);\
-    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
-    OPNAME ## pixels8_l2_mmx(dst, halfV, halfHV, stride, 8, 8);\
+    put ## RND ## pixels8_l2_mmx(halfH, src+1, halfH, 8, stride, 9);\
+    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
 }\
 static void OPNAME ## qpel8_mc22_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t half[9*2];\
+    uint64_t half[9];\
     uint8_t * const halfH= ((uint8_t*)half);\
     put ## RND ## mpeg4_qpel8_h_lowpass_ ## MMX(halfH, src, 8, stride, 9);\
     OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, halfH, stride, 8);\
@@ -1241,44 +1231,40 @@ static void OPNAME ## qpel16_mc03_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
     OPNAME ## pixels16_l2_mmx(dst, src+stride, half, stride, stride, 16);\
 }\
 static void OPNAME ## qpel16_mc11_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t half[16*2 + 16*2 + 18*2];\
-    uint8_t * const halfH= ((uint8_t*)half) + 2*256 + 16;\
-    uint8_t * const halfV= ((uint8_t*)half);\
-    uint8_t * const halfHV= ((uint8_t*)half) + 256;\
+    uint64_t half[16*2 + 17*2];\
+    uint8_t * const halfH= ((uint8_t*)half) + 256;\
+    uint8_t * const halfHV= ((uint8_t*)half);\
     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
-    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfV, src, 16, stride);\
+    put ## RND ## pixels16_l2_mmx(halfH, src, halfH, 16, stride, 17);\
     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
-    OPNAME ## pixels16_l4_mmx(dst, src, (uint8_t*)half, stride, 16);\
+    OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\
 }\
 static void OPNAME ## qpel16_mc31_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t half[16*2 + 16*2 + 18*2];\
-    uint8_t * const halfH= ((uint8_t*)half) + 2*256 + 16;\
-    uint8_t * const halfV= ((uint8_t*)half);\
-    uint8_t * const halfHV= ((uint8_t*)half) + 256;\
+    uint64_t half[16*2 + 17*2];\
+    uint8_t * const halfH= ((uint8_t*)half) + 256;\
+    uint8_t * const halfHV= ((uint8_t*)half);\
     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
-    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfV, src+1, 16, stride);\
+    put ## RND ## pixels16_l2_mmx(halfH, src+1, halfH, 16, stride, 17);\
     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
-    OPNAME ## pixels16_l4_mmx(dst, src+1, (uint8_t*)half, stride, 16);\
+    OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\
 }\
 static void OPNAME ## qpel16_mc13_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t half[16*2 + 16*2 + 17*2];\
-    uint8_t * const halfH= ((uint8_t*)half) + 2*256;\
-    uint8_t * const halfV= ((uint8_t*)half);\
-    uint8_t * const halfHV= ((uint8_t*)half) + 256;\
+    uint64_t half[16*2 + 17*2];\
+    uint8_t * const halfH= ((uint8_t*)half) + 256;\
+    uint8_t * const halfHV= ((uint8_t*)half);\
     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
-    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfV, src, 16, stride);\
+    put ## RND ## pixels16_l2_mmx(halfH, src, halfH, 16, stride, 17);\
     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
-    OPNAME ## pixels16_l4_mmx(dst, src+stride, (uint8_t*)half, stride, 16);\
+    OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\
 }\
 static void OPNAME ## qpel16_mc33_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t half[16*2 + 16*2 + 17*2];\
-    uint8_t * const halfH= ((uint8_t*)half) + 2*256;\
-    uint8_t * const halfV= ((uint8_t*)half);\
-    uint8_t * const halfHV= ((uint8_t*)half) + 256;\
-    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src  , 16, stride, 17);\
-    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfV, src+1, 16, stride);\
+    uint64_t half[16*2 + 17*2];\
+    uint8_t * const halfH= ((uint8_t*)half) + 256;\
+    uint8_t * const halfHV= ((uint8_t*)half);\
+    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
+    put ## RND ## pixels16_l2_mmx(halfH, src+1, halfH, 16, stride, 17);\
     put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
-    OPNAME ## pixels16_l4_mmx(dst, src+stride+1, (uint8_t*)half, stride, 16);\
+    OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\
 }\
 static void OPNAME ## qpel16_mc21_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
     uint64_t half[16*2 + 17*2];\
@@ -1297,24 +1283,18 @@ static void OPNAME ## qpel16_mc23_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
     OPNAME ## pixels16_l2_mmx(dst, halfH+16, halfHV, stride, 16, 16);\
 }\
 static void OPNAME ## qpel16_mc12_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t half[16*2 + 16*2 + 17*2];\
-    uint8_t * const halfH= ((uint8_t*)half) + 2*256;\
-    uint8_t * const halfV= ((uint8_t*)half);\
-    uint8_t * const halfHV= ((uint8_t*)half) + 256;\
+    uint64_t half[17*2];\
+    uint8_t * const halfH= ((uint8_t*)half);\
     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
-    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfV, src, 16, stride);\
-    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
-    OPNAME ## pixels16_l2_mmx(dst, halfV, halfHV, stride, 16, 16);\
+    put ## RND ## pixels16_l2_mmx(halfH, src, halfH, 16, stride, 17);\
+    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
 }\
 static void OPNAME ## qpel16_mc32_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
-    uint64_t half[16*2 + 16*2 + 17*2];\
-    uint8_t * const halfH= ((uint8_t*)half) + 2*256;\
-    uint8_t * const halfV= ((uint8_t*)half);\
-    uint8_t * const halfHV= ((uint8_t*)half) + 256;\
+    uint64_t half[17*2];\
+    uint8_t * const halfH= ((uint8_t*)half);\
     put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
-    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfV, src+1, 16, stride);\
-    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
-    OPNAME ## pixels16_l2_mmx(dst, halfV, halfHV, stride, 16, 16);\
+    put ## RND ## pixels16_l2_mmx(halfH, src+1, halfH, 16, stride, 17);\
+    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
 }\
 static void OPNAME ## qpel16_mc22_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
     uint64_t half[17*2];\
@@ -1525,7 +1505,7 @@ void dsputil_init_mmx(DSPContext* c, unsigned mask)
             c->avg_pixels_tab[1][1] = avg_pixels8_x2_3dnow;
             c->avg_pixels_tab[1][2] = avg_pixels8_y2_3dnow;
             c->avg_pixels_tab[1][3] = avg_pixels8_xy2_3dnow;
-        
+
             SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_3dnow)
             SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_3dnow)
             SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_3dnow)
diff --git a/libavcodec/i386/dsputil_mmx_rnd.h b/libavcodec/i386/dsputil_mmx_rnd.h
index 0ae1cd99de..924972fc3a 100644
--- a/libavcodec/i386/dsputil_mmx_rnd.h
+++ b/libavcodec/i386/dsputil_mmx_rnd.h
@@ -58,6 +58,16 @@ static void DEF(put, pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int
 {
     MOVQ_BFE(mm6);
     __asm __volatile(
+	"test $1, %0			\n\t"
+        " jz 1f				\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"movq	(%2), %%mm1		\n\t"
+	"addl	%4, %1			\n\t"
+        "addl	$8, %2			\n\t"
+	PAVGB(%%mm0, %%mm1, %%mm4, %%mm6)
+	"movq	%%mm4, (%3)		\n\t"
+	"addl	%5, %3			\n\t"
+        "decl	%0			\n\t"
 	".balign 8			\n\t"
 	"1:				\n\t"
 	"movq	(%1), %%mm0		\n\t"
@@ -144,6 +154,19 @@ static void DEF(put, pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, in
 {
     MOVQ_BFE(mm6);
     __asm __volatile(
+	"test $1, %0			\n\t"
+        " jz 1f				\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"
+	"addl	$16, %2			\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"
+	"decl	%0			\n\t"
 	".balign 8			\n\t"
 	"1:				\n\t"
 	"movq	(%1), %%mm0		\n\t"
@@ -271,124 +294,6 @@ 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)
@@ -641,133 +546,6 @@ 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(%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	(%0), %%mm4		\n\t"
-        PAVGB(%%mm3, %%mm4, %%mm0, %%mm5)
-	"movq	%%mm0, (%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(%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	(%0), %%mm4		\n\t"
-        PAVGB(%%mm3, %%mm4, %%mm0, %%mm5)
-	"movq	%%mm0, (%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	8(%0), %%mm4		\n\t"
-        PAVGB(%%mm3, %%mm4, %%mm0, %%mm5)
-	"movq	%%mm0, 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);