qpel fix

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

6 files changed, 344 insertions, 398 deletions

diff --git a/libavcodec/avcodec.h b/libavcodec/avcodec.h
index fd7eafbd4f..06603e3596 100644
--- a/libavcodec/avcodec.h
+++ b/libavcodec/avcodec.h
@@ -5,8 +5,8 @@
 
 #define LIBAVCODEC_VERSION_INT 0x000406
 #define LIBAVCODEC_VERSION     "0.4.6"
-#define LIBAVCODEC_BUILD       4652
-#define LIBAVCODEC_BUILD_STR   "4652"
+#define LIBAVCODEC_BUILD       4653
+#define LIBAVCODEC_BUILD_STR   "4653"
 
 enum CodecID {
     CODEC_ID_NONE, 
@@ -520,6 +520,7 @@ typedef struct AVCodecContext {
 #define FF_BUG_NO_PADDING       16
 #define FF_BUG_AC_VLC           32
 #define FF_BUG_QPEL_CHROMA      64
+#define FF_BUG_STD_QPEL         128
 //#define FF_BUG_FAKE_SCALABILITY 16 //autodetection should work 100%
         
     /**
diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c
index 0d7556f657..fc5afbeb49 100644
--- a/libavcodec/dsputil.c
+++ b/libavcodec/dsputil.c
@@ -801,7 +801,8 @@ static void OPNAME ## mpeg4_qpel8_h_lowpass(UINT8 *dst, UINT8 *src, int dstStrid
     }\
 }\
 \
-static void OPNAME ## mpeg4_qpel8_v_lowpass(UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int w){\
+static void OPNAME ## mpeg4_qpel8_v_lowpass(UINT8 *dst, UINT8 *src, int dstStride, int srcStride){\
+    const int w=8;\
     UINT8 *cm = cropTbl + MAX_NEG_CROP;\
     int i;\
     for(i=0; i<w; i++)\
@@ -923,107 +924,163 @@ static void OPNAME ## qpel8_mc01_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 full[16*9];\
     UINT8 half[64];\
     copy_block9(full, src, 16, stride, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16, 8);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\
     OPNAME ## pixels8_l2(dst, full, half, stride, 16, 8, 8);\
 }\
 \
 static void OPNAME ## qpel8_mc02_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 full[16*9];\
     copy_block9(full, src, 16, stride, 9);\
-    OPNAME ## mpeg4_qpel8_v_lowpass(dst, full, stride, 16, 8);\
+    OPNAME ## mpeg4_qpel8_v_lowpass(dst, full, stride, 16);\
 }\
 \
 static void OPNAME ## qpel8_mc03_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 full[16*9];\
     UINT8 half[64];\
     copy_block9(full, src, 16, stride, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16, 8);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(half, full, 8, 16);\
     OPNAME ## pixels8_l2(dst, full+16, half, stride, 16, 8, 8);\
 }\
-static void OPNAME ## qpel8_mc11_c(UINT8 *dst, UINT8 *src, int stride){\
+void ff_ ## OPNAME ## qpel8_mc11_old_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 full[16*9];\
     UINT8 halfH[72];\
     UINT8 halfV[64];\
     UINT8 halfHV[64];\
     copy_block9(full, src, 16, stride, 9);\
     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16, 8);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8, 8);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
     OPNAME ## pixels8_l4(dst, full, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\
 }\
-static void OPNAME ## qpel8_mc31_c(UINT8 *dst, UINT8 *src, int stride){\
+static void OPNAME ## qpel8_mc11_c(UINT8 *dst, UINT8 *src, int stride){\
+    UINT8 full[16*9];\
+    UINT8 halfH[72];\
+    UINT8 halfHV[64];\
+    copy_block9(full, src, 16, stride, 9);\
+    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
+    put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
+    OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\
+}\
+void ff_ ## OPNAME ## qpel8_mc31_old_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 full[16*9];\
     UINT8 halfH[72];\
     UINT8 halfV[64];\
     UINT8 halfHV[64];\
     copy_block9(full, src, 16, stride, 9);\
     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16, 8);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8, 8);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
     OPNAME ## pixels8_l4(dst, full+1, halfH, halfV, halfHV, stride, 16, 8, 8, 8, 8);\
 }\
-static void OPNAME ## qpel8_mc13_c(UINT8 *dst, UINT8 *src, int stride){\
+static void OPNAME ## qpel8_mc31_c(UINT8 *dst, UINT8 *src, int stride){\
+    UINT8 full[16*9];\
+    UINT8 halfH[72];\
+    UINT8 halfHV[64];\
+    copy_block9(full, src, 16, stride, 9);\
+    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
+    put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
+    OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\
+}\
+void ff_ ## OPNAME ## qpel8_mc13_old_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 full[16*9];\
     UINT8 halfH[72];\
     UINT8 halfV[64];\
     UINT8 halfHV[64];\
     copy_block9(full, src, 16, stride, 9);\
     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16, 8);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8, 8);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
     OPNAME ## pixels8_l4(dst, full+16, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\
 }\
-static void OPNAME ## qpel8_mc33_c(UINT8 *dst, UINT8 *src, int stride){\
+static void OPNAME ## qpel8_mc13_c(UINT8 *dst, UINT8 *src, int stride){\
+    UINT8 full[16*9];\
+    UINT8 halfH[72];\
+    UINT8 halfHV[64];\
+    copy_block9(full, src, 16, stride, 9);\
+    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
+    put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
+    OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\
+}\
+void ff_ ## OPNAME ## qpel8_mc33_old_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 full[16*9];\
     UINT8 halfH[72];\
     UINT8 halfV[64];\
     UINT8 halfHV[64];\
     copy_block9(full, src, 16, stride, 9);\
     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full  , 8, 16, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16, 8);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8, 8);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
     OPNAME ## pixels8_l4(dst, full+17, halfH+8, halfV, halfHV, stride, 16, 8, 8, 8, 8);\
 }\
+static void OPNAME ## qpel8_mc33_c(UINT8 *dst, UINT8 *src, int stride){\
+    UINT8 full[16*9];\
+    UINT8 halfH[72];\
+    UINT8 halfHV[64];\
+    copy_block9(full, src, 16, stride, 9);\
+    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
+    put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
+    OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\
+}\
 static void OPNAME ## qpel8_mc21_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 halfH[72];\
     UINT8 halfHV[64];\
     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8, 8);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
     OPNAME ## pixels8_l2(dst, halfH, halfHV, stride, 8, 8, 8);\
 }\
 static void OPNAME ## qpel8_mc23_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 halfH[72];\
     UINT8 halfHV[64];\
     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8, 8);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
     OPNAME ## pixels8_l2(dst, halfH+8, halfHV, stride, 8, 8, 8);\
 }\
-static void OPNAME ## qpel8_mc12_c(UINT8 *dst, UINT8 *src, int stride){\
+void ff_ ## OPNAME ## qpel8_mc12_old_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 full[16*9];\
     UINT8 halfH[72];\
     UINT8 halfV[64];\
     UINT8 halfHV[64];\
     copy_block9(full, src, 16, stride, 9);\
     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16, 8);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8, 8);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full, 8, 16);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
     OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\
 }\
-static void OPNAME ## qpel8_mc32_c(UINT8 *dst, UINT8 *src, int stride){\
+static void OPNAME ## qpel8_mc12_c(UINT8 *dst, UINT8 *src, int stride){\
+    UINT8 full[16*9];\
+    UINT8 halfH[72];\
+    copy_block9(full, src, 16, stride, 9);\
+    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
+    put ## RND ## pixels8_l2(halfH, halfH, full, 8, 8, 16, 9);\
+    OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\
+}\
+void ff_ ## OPNAME ## qpel8_mc32_old_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 full[16*9];\
     UINT8 halfH[72];\
     UINT8 halfV[64];\
     UINT8 halfHV[64];\
     copy_block9(full, src, 16, stride, 9);\
     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16, 8);\
-    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8, 8);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfV, full+1, 8, 16);\
+    put ## RND ## mpeg4_qpel8_v_lowpass(halfHV, halfH, 8, 8);\
     OPNAME ## pixels8_l2(dst, halfV, halfHV, stride, 8, 8, 8);\
 }\
+static void OPNAME ## qpel8_mc32_c(UINT8 *dst, UINT8 *src, int stride){\
+    UINT8 full[16*9];\
+    UINT8 halfH[72];\
+    copy_block9(full, src, 16, stride, 9);\
+    put ## RND ## mpeg4_qpel8_h_lowpass(halfH, full, 8, 16, 9);\
+    put ## RND ## pixels8_l2(halfH, halfH, full+1, 8, 8, 16, 9);\
+    OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\
+}\
 static void OPNAME ## qpel8_mc22_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 halfH[72];\
     put ## RND ## mpeg4_qpel8_h_lowpass(halfH, src, 8, stride, 9);\
-    OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8, 8);\
+    OPNAME ## mpeg4_qpel8_v_lowpass(dst, halfH, stride, 8);\
 }\
 static void OPNAME ## qpel16_mc00_c (UINT8 *dst, UINT8 *src, int stride){\
     OPNAME ## pixels16_c(dst, src, stride, 16);\
@@ -1066,7 +1123,7 @@ static void OPNAME ## qpel16_mc03_c(UINT8 *dst, UINT8 *src, int stride){\
     put ## RND ## mpeg4_qpel16_v_lowpass(half, full, 16, 24);\
     OPNAME ## pixels16_l2(dst, full+24, half, stride, 24, 16, 16);\
 }\
-static void OPNAME ## qpel16_mc11_c(UINT8 *dst, UINT8 *src, int stride){\
+void ff_ ## OPNAME ## qpel16_mc11_old_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 full[24*17];\
     UINT8 halfH[272];\
     UINT8 halfV[256];\
@@ -1077,7 +1134,17 @@ static void OPNAME ## qpel16_mc11_c(UINT8 *dst, UINT8 *src, int stride){\
     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
     OPNAME ## pixels16_l4(dst, full, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
 }\
-static void OPNAME ## qpel16_mc31_c(UINT8 *dst, UINT8 *src, int stride){\
+static void OPNAME ## qpel16_mc11_c(UINT8 *dst, UINT8 *src, int stride){\
+    UINT8 full[24*17];\
+    UINT8 halfH[272];\
+    UINT8 halfHV[256];\
+    copy_block17(full, src, 24, stride, 17);\
+    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
+    put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
+    OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\
+}\
+void ff_ ## OPNAME ## qpel16_mc31_old_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 full[24*17];\
     UINT8 halfH[272];\
     UINT8 halfV[256];\
@@ -1088,7 +1155,17 @@ static void OPNAME ## qpel16_mc31_c(UINT8 *dst, UINT8 *src, int stride){\
     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
     OPNAME ## pixels16_l4(dst, full+1, halfH, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
 }\
-static void OPNAME ## qpel16_mc13_c(UINT8 *dst, UINT8 *src, int stride){\
+static void OPNAME ## qpel16_mc31_c(UINT8 *dst, UINT8 *src, int stride){\
+    UINT8 full[24*17];\
+    UINT8 halfH[272];\
+    UINT8 halfHV[256];\
+    copy_block17(full, src, 24, stride, 17);\
+    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
+    put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
+    OPNAME ## pixels16_l2(dst, halfH, halfHV, stride, 16, 16, 16);\
+}\
+void ff_ ## OPNAME ## qpel16_mc13_old_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 full[24*17];\
     UINT8 halfH[272];\
     UINT8 halfV[256];\
@@ -1099,7 +1176,17 @@ static void OPNAME ## qpel16_mc13_c(UINT8 *dst, UINT8 *src, int stride){\
     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
     OPNAME ## pixels16_l4(dst, full+24, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
 }\
-static void OPNAME ## qpel16_mc33_c(UINT8 *dst, UINT8 *src, int stride){\
+static void OPNAME ## qpel16_mc13_c(UINT8 *dst, UINT8 *src, int stride){\
+    UINT8 full[24*17];\
+    UINT8 halfH[272];\
+    UINT8 halfHV[256];\
+    copy_block17(full, src, 24, stride, 17);\
+    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
+    put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
+    OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\
+}\
+void ff_ ## OPNAME ## qpel16_mc33_old_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 full[24*17];\
     UINT8 halfH[272];\
     UINT8 halfV[256];\
@@ -1110,6 +1197,16 @@ static void OPNAME ## qpel16_mc33_c(UINT8 *dst, UINT8 *src, int stride){\
     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
     OPNAME ## pixels16_l4(dst, full+25, halfH+16, halfV, halfHV, stride, 24, 16, 16, 16, 16);\
 }\
+static void OPNAME ## qpel16_mc33_c(UINT8 *dst, UINT8 *src, int stride){\
+    UINT8 full[24*17];\
+    UINT8 halfH[272];\
+    UINT8 halfHV[256];\
+    copy_block17(full, src, 24, stride, 17);\
+    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
+    put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\
+    put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
+    OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\
+}\
 static void OPNAME ## qpel16_mc21_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 halfH[272];\
     UINT8 halfHV[256];\
@@ -1124,7 +1221,7 @@ static void OPNAME ## qpel16_mc23_c(UINT8 *dst, UINT8 *src, int stride){\
     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
     OPNAME ## pixels16_l2(dst, halfH+16, halfHV, stride, 16, 16, 16);\
 }\
-static void OPNAME ## qpel16_mc12_c(UINT8 *dst, UINT8 *src, int stride){\
+void ff_ ## OPNAME ## qpel16_mc12_old_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 full[24*17];\
     UINT8 halfH[272];\
     UINT8 halfV[256];\
@@ -1135,7 +1232,15 @@ static void OPNAME ## qpel16_mc12_c(UINT8 *dst, UINT8 *src, int stride){\
     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
     OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\
 }\
-static void OPNAME ## qpel16_mc32_c(UINT8 *dst, UINT8 *src, int stride){\
+static void OPNAME ## qpel16_mc12_c(UINT8 *dst, UINT8 *src, int stride){\
+    UINT8 full[24*17];\
+    UINT8 halfH[272];\
+    copy_block17(full, src, 24, stride, 17);\
+    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
+    put ## RND ## pixels16_l2(halfH, halfH, full, 16, 16, 24, 17);\
+    OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\
+}\
+void ff_ ## OPNAME ## qpel16_mc32_old_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 full[24*17];\
     UINT8 halfH[272];\
     UINT8 halfV[256];\
@@ -1146,6 +1251,14 @@ static void OPNAME ## qpel16_mc32_c(UINT8 *dst, UINT8 *src, int stride){\
     put ## RND ## mpeg4_qpel16_v_lowpass(halfHV, halfH, 16, 16);\
     OPNAME ## pixels16_l2(dst, halfV, halfHV, stride, 16, 16, 16);\
 }\
+static void OPNAME ## qpel16_mc32_c(UINT8 *dst, UINT8 *src, int stride){\
+    UINT8 full[24*17];\
+    UINT8 halfH[272];\
+    copy_block17(full, src, 24, stride, 17);\
+    put ## RND ## mpeg4_qpel16_h_lowpass(halfH, full, 16, 24, 17);\
+    put ## RND ## pixels16_l2(halfH, halfH, full+1, 16, 16, 24, 17);\
+    OPNAME ## mpeg4_qpel16_v_lowpass(dst, halfH, stride, 16);\
+}\
 static void OPNAME ## qpel16_mc22_c(UINT8 *dst, UINT8 *src, int stride){\
     UINT8 halfH[272];\
     put ## RND ## mpeg4_qpel16_h_lowpass(halfH, src, 16, stride, 17);\
diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h
index c7df750c6a..08c679cc1c 100644
--- a/libavcodec/dsputil.h
+++ b/libavcodec/dsputil.h
@@ -69,7 +69,23 @@ void clear_blocks_c(DCTELEM *blocks);
 typedef void (*op_pixels_func)(UINT8 *block/*align width (8 or 16)*/, const UINT8 *pixels/*align 1*/, int line_size, int h);
 typedef void (*qpel_mc_func)(UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);
 
-
+#define DEF_OLD_QPEL(name)\
+void ff_put_        ## name (UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);\
+void ff_put_no_rnd_ ## name (UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);\
+void ff_avg_        ## name (UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);
+
+DEF_OLD_QPEL(qpel16_mc11_old_c)
+DEF_OLD_QPEL(qpel16_mc31_old_c)
+DEF_OLD_QPEL(qpel16_mc12_old_c)
+DEF_OLD_QPEL(qpel16_mc32_old_c)
+DEF_OLD_QPEL(qpel16_mc13_old_c)
+DEF_OLD_QPEL(qpel16_mc33_old_c)
+DEF_OLD_QPEL(qpel8_mc11_old_c)
+DEF_OLD_QPEL(qpel8_mc31_old_c)
+DEF_OLD_QPEL(qpel8_mc12_old_c)
+DEF_OLD_QPEL(qpel8_mc32_old_c)
+DEF_OLD_QPEL(qpel8_mc13_old_c)
+DEF_OLD_QPEL(qpel8_mc33_old_c)
 
 #define CALL_2X_PIXELS(a, b, n)\
 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
diff --git a/libavcodec/h263dec.c b/libavcodec/h263dec.c
index cb473ea87b..d88464fe1d 100644
--- a/libavcodec/h263dec.c
+++ b/libavcodec/h263dec.c
@@ -344,6 +344,42 @@ static int mpeg4_find_frame_end(MpegEncContext *s, UINT8 *buf, int buf_size){
     return -1;
 }
 
+static void draw_line(uint8_t *buf, int sx, int sy, int ex, int ey, int w, int h, int stride, int color){
+    int t, x, y, f;
+    
+    ex= clip(ex, 0, w-1);
+    ey= clip(ey, 0, h-1);
+    
+    buf[sy*stride + sx]+= color;
+    
+    if(ABS(ex - sx) > ABS(ey - sy)){
+        if(sx > ex){
+            t=sx; sx=ex; ex=t;
+            t=sy; sy=ey; ey=t;
+        }
+        buf+= sx + sy*stride;
+        ex-= sx;
+        f= ((ey-sy)<<16)/ex;
+        for(x= 0; x <= ex; x++){
+            y= ((x*f) + (1<<15))>>16;
+            buf[y*stride + x]+= color;
+        }
+    }else{
+        if(sy > ey){
+            t=sx; sx=ex; ex=t;
+            t=sy; sy=ey; ey=t;
+        }
+        buf+= sx + sy*stride;
+        ey-= sy;
+        if(ey) f= ((ex-sx)<<16)/ey;
+        else   f= 0;
+        for(y= 0; y <= ey; y++){
+            x= ((y*f) + (1<<15))>>16;
+            buf[y*stride + x]+= color;
+        }
+    }
+}
+
 int ff_h263_decode_frame(AVCodecContext *avctx, 
                              void *data, int *data_size,
                              UINT8 *buf, int buf_size)
@@ -472,6 +508,14 @@ retry:
         if(s->xvid_build && s->xvid_build<=1)
             s->workaround_bugs|= FF_BUG_QPEL_CHROMA;
 
+#define SET_QPEL_FUNC(postfix1, postfix2) \
+    s->dsp.put_ ## postfix1 = ff_put_ ## postfix2;\
+    s->dsp.put_no_rnd_ ## postfix1 = ff_put_no_rnd_ ## postfix2;\
+    s->dsp.avg_ ## postfix1 = ff_avg_ ## postfix2;
+
+        if(s->lavc_build && s->lavc_build<4653)
+            s->workaround_bugs|= FF_BUG_STD_QPEL;
+        
 //printf("padding_bug_score: %d\n", s->padding_bug_score);
 #if 0
         if(s->divx_version==500)
@@ -489,6 +533,21 @@ retry:
 #endif
     }
     
+    if(s->workaround_bugs& FF_BUG_STD_QPEL){
+        SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_old_c)
+        SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_old_c)
+        SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_old_c)
+        SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_old_c)
+        SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_old_c)
+        SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_old_c)
+
+        SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_old_c)
+        SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_old_c)
+        SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_old_c)
+        SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_old_c)
+        SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_old_c)
+        SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_old_c)
+    }
 
 #if 0 // dump bits per frame / qp / complexity
 {
@@ -645,41 +704,40 @@ retry:
     }
 
     MPV_frame_end(s);
-#if 0 //dirty show MVs, we should export the MV tables and write a filter to show them
-{
-  int mb_y;
-  s->has_b_frames=1;
-  for(mb_y=0; mb_y<s->mb_height; mb_y++){
-    int mb_x;
-    int y= mb_y*16 + 8;
-    for(mb_x=0; mb_x<s->mb_width; mb_x++){
-      int x= mb_x*16 + 8;
-      uint8_t *ptr= s->last_picture.data[0];
-      int xy= 1 + mb_x*2 + (mb_y*2 + 1)*(s->mb_width*2 + 2);
-      int mx= (s->motion_val[xy][0]>>1) + x;
-      int my= (s->motion_val[xy][1]>>1) + y;
-      int i;
-      int max;
-
-      if(mx<0) mx=0;
-      if(my<0) my=0;
-      if(mx>=s->width)  mx= s->width -1;
-      if(my>=s->height) my= s->height-1;
-      max= ABS(mx-x);
-      if(ABS(my-y) > max) max= ABS(my-y);
-      /* the ugliest linedrawing routine ... */
-      for(i=0; i<max; i++){
-        int x1= x + (mx-x)*i/max;
-        int y1= y + (my-y)*i/max;
-        ptr[y1*s->linesize + x1]+=100;
-      }
-      ptr[y*s->linesize + x]+=100;
-      s->mbskip_table[mb_x + mb_y*s->mb_width]=0;
+
+    if((avctx->debug&FF_DEBUG_VIS_MV) && s->last_picture.data[0]){
+        const int shift= 1 + s->quarter_sample;
+        int mb_y;
+        uint8_t *ptr= s->last_picture.data[0];
+        s->low_delay=0; //needed to see the vectors without trashing the buffers
+
+        for(mb_y=0; mb_y<s->mb_height; mb_y++){
+            int mb_x;
+            for(mb_x=0; mb_x<s->mb_width; mb_x++){
+                const int mb_index= mb_x + mb_y*s->mb_width;
+                if(s->co_located_type_table[mb_index] == MV_TYPE_8X8){
+                    int i;
+                    for(i=0; i<4; i++){
+                        int sx= mb_x*16 + 4 + 8*(i&1);
+                        int sy= mb_y*16 + 4 + 8*(i>>1);
+                        int xy= 1 + mb_x*2 + (i&1) + (mb_y*2 + 1 + (i>>1))*(s->mb_width*2 + 2);
+                        int mx= (s->motion_val[xy][0]>>shift) + sx;
+                        int my= (s->motion_val[xy][1]>>shift) + sy;
+                        draw_line(ptr, sx, sy, mx, my, s->width, s->height, s->linesize, 100);
+                    }
+                }else{
+                    int sx= mb_x*16 + 8;
+                    int sy= mb_y*16 + 8;
+                    int xy= 1 + mb_x*2 + (mb_y*2 + 1)*(s->mb_width*2 + 2);
+                    int mx= (s->motion_val[xy][0]>>shift) + sx;
+                    int my= (s->motion_val[xy][1]>>shift) + sy;
+                    draw_line(ptr, sx, sy, mx, my, s->width, s->height, s->linesize, 100);
+                }
+                s->mbskip_table[mb_index]=0;
+            }
+        }
     }
-  }
 
-}
-#endif
 
     if(s->pict_type==B_TYPE || s->low_delay){
         *pict= *(AVFrame*)&s->current_picture;
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);