From 6f1c8d4eafabd914b87e9171bf4d04f4ef9160ea Mon Sep 17 00:00:00 2001
From: Miguel Freitas <miguelfreitas@users.sourceforge.net>
Date: Wed, 8 Jan 2003 13:18:42 +0000
Subject: syncing ffmpeg (with some compilation fixes) - fixes wma bugs - mace,
 huffyuv and mp3 decoders imported (but not enabled) tested: wma (v1 and v2),
 mpeg4, msmpeg4 v1, v2 and v3, divx3, divx4, divx5, xvid and dv decoders.
 everything looks fine.

CVS patchset: 3828
CVS date: 2003/01/08 13:18:42
---
 src/libffmpeg/libavcodec/i386/Makefile.am       |   2 +-
 src/libffmpeg/libavcodec/i386/dsputil_mmx.c     | 934 +++++++++++++++++++++++-
 src/libffmpeg/libavcodec/i386/dsputil_mmx_avg.h |  60 ++
 src/libffmpeg/libavcodec/i386/dsputil_mmx_rnd.h | 371 +++++++++-
 src/libffmpeg/libavcodec/i386/fft_sse.c         |  12 +
 src/libffmpeg/libavcodec/i386/motion_est_mmx.c  |  19 +
 src/libffmpeg/libavcodec/i386/mpegvideo_mmx.c   |   1 -
 7 files changed, 1393 insertions(+), 6 deletions(-)

(limited to 'src/libffmpeg/libavcodec/i386')

diff --git a/src/libffmpeg/libavcodec/i386/Makefile.am b/src/libffmpeg/libavcodec/i386/Makefile.am
index 0ef6bb0eb..6386800f6 100644
--- a/src/libffmpeg/libavcodec/i386/Makefile.am
+++ b/src/libffmpeg/libavcodec/i386/Makefile.am
@@ -16,11 +16,11 @@ libavcodec_mmx_src = \
 	cputest.c \
 	dsputil_mmx.c \
 	fdct_mmx.c \
+	fft_sse.c \
 	idct_mmx.c \
 	motion_est_mmx.c \
 	mpegvideo_mmx.c \
 	simple_idct_mmx.c
-#	fft_sse.c - needs new header from gcc 3.1
 
 libavcodec_mmx_dummy = libavcodec_mmx_dummy.c
 
diff --git a/src/libffmpeg/libavcodec/i386/dsputil_mmx.c b/src/libffmpeg/libavcodec/i386/dsputil_mmx.c
index 12a360154..5fce7f914 100644
--- a/src/libffmpeg/libavcodec/i386/dsputil_mmx.c
+++ b/src/libffmpeg/libavcodec/i386/dsputil_mmx.c
@@ -43,11 +43,21 @@ int pix_abs8x8_x2_mmx2(UINT8 *blk1, UINT8 *blk2, int lx);
 int pix_abs8x8_y2_mmx2(UINT8 *blk1, UINT8 *blk2, int lx);
 int pix_abs8x8_xy2_mmx2(UINT8 *blk1, UINT8 *blk2, int lx);
 
+int sad16x16_mmx(void *s, UINT8 *blk1, UINT8 *blk2, int lx);
+int sad8x8_mmx(void *s, UINT8 *blk1, UINT8 *blk2, int lx);
+int sad16x16_mmx2(void *s, UINT8 *blk1, UINT8 *blk2, int lx);
+int sad8x8_mmx2(void *s, UINT8 *blk1, UINT8 *blk2, int lx);
+
 /* pixel operations */
 static const uint64_t mm_bone __attribute__ ((aligned(8))) = 0x0101010101010101ULL;
 static const uint64_t mm_wone __attribute__ ((aligned(8))) = 0x0001000100010001ULL;
 static const uint64_t mm_wtwo __attribute__ ((aligned(8))) = 0x0002000200020002ULL;
 
+static const uint64_t ff_pw_20 __attribute__ ((aligned(8))) = 0x0014001400140014ULL;
+static const uint64_t ff_pw_3  __attribute__ ((aligned(8))) = 0x0003000300030003ULL;
+static const uint64_t ff_pw_16 __attribute__ ((aligned(8))) = 0x0010001000100010ULL;
+static const uint64_t ff_pw_15 __attribute__ ((aligned(8))) = 0x000F000F000F000FULL;
+
 #define JUMPALIGN() __asm __volatile (".balign 8"::)
 #define MOVQ_ZERO(regd)  __asm __volatile ("pxor %%" #regd ", %%" #regd ::)
 
@@ -213,7 +223,7 @@ static void get_pixels_mmx(DCTELEM *block, const UINT8 *pixels, int line_size)
     );
 }
 
-static void diff_pixels_mmx(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride)
+static inline void diff_pixels_mmx(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride)
 {
     asm volatile(
         "pxor %%mm7, %%mm7	\n\t"
@@ -496,12 +506,853 @@ static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w){
     for(; i<w; i++)
         dst[i+0] = src1[i+0]-src2[i+0];
 }
+#define LBUTTERFLY(a,b)\
+    "paddw " #b ", " #a "		\n\t"\
+    "paddw " #b ", " #b "		\n\t"\
+    "psubw " #a ", " #b "		\n\t"
+
+#define HADAMARD48\
+        LBUTTERFLY(%%mm0, %%mm1)\
+        LBUTTERFLY(%%mm2, %%mm3)\
+        LBUTTERFLY(%%mm4, %%mm5)\
+        LBUTTERFLY(%%mm6, %%mm7)\
+        \
+        LBUTTERFLY(%%mm0, %%mm2)\
+        LBUTTERFLY(%%mm1, %%mm3)\
+        LBUTTERFLY(%%mm4, %%mm6)\
+        LBUTTERFLY(%%mm5, %%mm7)\
+        \
+        LBUTTERFLY(%%mm0, %%mm4)\
+        LBUTTERFLY(%%mm1, %%mm5)\
+        LBUTTERFLY(%%mm2, %%mm6)\
+        LBUTTERFLY(%%mm3, %%mm7)
+
+#define MMABS(a,z)\
+    "pxor " #z ", " #z "		\n\t"\
+    "pcmpgtw " #a ", " #z "		\n\t"\
+    "pxor " #z ", " #a "		\n\t"\
+    "psubw " #z ", " #a "		\n\t"
+
+#define MMABS_SUM(a,z, sum)\
+    "pxor " #z ", " #z "		\n\t"\
+    "pcmpgtw " #a ", " #z "		\n\t"\
+    "pxor " #z ", " #a "		\n\t"\
+    "psubw " #z ", " #a "		\n\t"\
+    "paddusw " #a ", " #sum "		\n\t"
+
+    
+#define SBUTTERFLY(a,b,t,n)\
+    "movq " #a ", " #t "		\n\t" /* abcd */\
+    "punpckl" #n " " #b ", " #a "	\n\t" /* aebf */\
+    "punpckh" #n " " #b ", " #t "	\n\t" /* cgdh */\
+    
+#define TRANSPOSE4(a,b,c,d,t)\
+    SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\
+    SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\
+    SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\
+    SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */
+
+#define LOAD4(o, a, b, c, d)\
+        "movq "#o"(%1), " #a "		\n\t"\
+        "movq "#o"+16(%1), " #b "	\n\t"\
+        "movq "#o"+32(%1), " #c "	\n\t"\
+        "movq "#o"+48(%1), " #d "	\n\t"
+
+#define STORE4(o, a, b, c, d)\
+        "movq "#a", "#o"(%1)		\n\t"\
+        "movq "#b", "#o"+16(%1)		\n\t"\
+        "movq "#c", "#o"+32(%1)		\n\t"\
+        "movq "#d", "#o"+48(%1)		\n\t"\
+
+static int hadamard8_diff_mmx(void *s, uint8_t *src1, uint8_t *src2, int stride){
+    uint64_t temp[16] __align8;
+    int sum=0;
+
+    diff_pixels_mmx((DCTELEM*)temp, src1, src2, stride);
 
+    asm volatile(
+        LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)
+        LOAD4(64, %%mm4, %%mm5, %%mm6, %%mm7)
+        
+        HADAMARD48
+        
+        "movq %%mm7, 112(%1)		\n\t"
+        
+        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)
+        STORE4(0 , %%mm0, %%mm3, %%mm7, %%mm2)
+        
+        "movq 112(%1), %%mm7 		\n\t"
+        TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)
+        STORE4(64, %%mm4, %%mm7, %%mm0, %%mm6)
+
+        LOAD4(8 , %%mm0, %%mm1, %%mm2, %%mm3)
+        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)
+        
+        HADAMARD48
+        
+        "movq %%mm7, 120(%1)		\n\t"
+        
+        TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm7)
+        STORE4(8 , %%mm0, %%mm3, %%mm7, %%mm2)
+        
+        "movq 120(%1), %%mm7 		\n\t"
+        TRANSPOSE4(%%mm4, %%mm5, %%mm6, %%mm7, %%mm0)
+        "movq %%mm7, %%mm5		\n\t"//FIXME remove
+        "movq %%mm6, %%mm7		\n\t"
+        "movq %%mm0, %%mm6		\n\t"
+//        STORE4(72, %%mm4, %%mm7, %%mm0, %%mm6) //FIXME remove
+        
+        LOAD4(64, %%mm0, %%mm1, %%mm2, %%mm3)
+//        LOAD4(72, %%mm4, %%mm5, %%mm6, %%mm7)
+        
+        HADAMARD48
+        "movq %%mm7, 64(%1)		\n\t"
+        MMABS(%%mm0, %%mm7)
+        MMABS_SUM(%%mm1, %%mm7, %%mm0)
+        MMABS_SUM(%%mm2, %%mm7, %%mm0)
+        MMABS_SUM(%%mm3, %%mm7, %%mm0)
+        MMABS_SUM(%%mm4, %%mm7, %%mm0)
+        MMABS_SUM(%%mm5, %%mm7, %%mm0)
+        MMABS_SUM(%%mm6, %%mm7, %%mm0)
+        "movq 64(%1), %%mm1		\n\t"
+        MMABS_SUM(%%mm1, %%mm7, %%mm0)
+        "movq %%mm0, 64(%1)		\n\t"
+        
+        LOAD4(0 , %%mm0, %%mm1, %%mm2, %%mm3)
+        LOAD4(8 , %%mm4, %%mm5, %%mm6, %%mm7)
+        
+        HADAMARD48
+        "movq %%mm7, (%1)		\n\t"
+        MMABS(%%mm0, %%mm7)
+        MMABS_SUM(%%mm1, %%mm7, %%mm0)
+        MMABS_SUM(%%mm2, %%mm7, %%mm0)
+        MMABS_SUM(%%mm3, %%mm7, %%mm0)
+        MMABS_SUM(%%mm4, %%mm7, %%mm0)
+        MMABS_SUM(%%mm5, %%mm7, %%mm0)
+        MMABS_SUM(%%mm6, %%mm7, %%mm0)
+        "movq (%1), %%mm1		\n\t"
+        MMABS_SUM(%%mm1, %%mm7, %%mm0)
+        "movq 64(%1), %%mm1		\n\t"
+        MMABS_SUM(%%mm1, %%mm7, %%mm0)
+        
+        "movq %%mm0, %%mm1		\n\t"
+        "psrlq $32, %%mm0		\n\t"
+        "paddusw %%mm1, %%mm0		\n\t"
+        "movq %%mm0, %%mm1		\n\t"
+        "psrlq $16, %%mm0		\n\t"
+        "paddusw %%mm1, %%mm0		\n\t"
+        "movd %%mm0, %0			\n\t"
+                
+        : "=r" (sum)
+        : "r"(temp)
+    );
+    return sum&0xFFFF;
+}
+
+WARPER88_1616(hadamard8_diff_mmx, hadamard8_diff16_mmx)
+
+#define put_no_rnd_pixels8_mmx(a,b,c,d) put_pixels8_mmx(a,b,c,d)
+#define put_no_rnd_pixels16_mmx(a,b,c,d) put_pixels16_mmx(a,b,c,d)
+
+#define QPEL_V_LOW(m3,m4,m5,m6, pw_20, pw_3, rnd, in0, in1, in2, in7, out, OP)\
+        "paddw " #m4 ", " #m3 "		\n\t" /* x1 */\
+        "movq "MANGLE(ff_pw_20)", %%mm4		\n\t" /* 20 */\
+        "pmullw " #m3 ", %%mm4		\n\t" /* 20x1 */\
+        "movq "#in7", " #m3 "		\n\t" /* d */\
+        "movq "#in0", %%mm5		\n\t" /* D */\
+        "paddw " #m3 ", %%mm5		\n\t" /* x4 */\
+        "psubw %%mm5, %%mm4		\n\t" /* 20x1 - x4 */\
+        "movq "#in1", %%mm5		\n\t" /* C */\
+        "movq "#in2", %%mm6		\n\t" /* B */\
+        "paddw " #m6 ", %%mm5		\n\t" /* x3 */\
+        "paddw " #m5 ", %%mm6		\n\t" /* x2 */\
+        "paddw %%mm6, %%mm6		\n\t" /* 2x2 */\
+        "psubw %%mm6, %%mm5		\n\t" /* -2x2 + x3 */\
+        "pmullw "MANGLE(ff_pw_3)", %%mm5	\n\t" /* -6x2 + 3x3 */\
+        "paddw " #rnd ", %%mm4		\n\t" /* x2 */\
+        "paddw %%mm4, %%mm5		\n\t" /* 20x1 - 6x2 + 3x3 - x4 */\
+        "psraw $5, %%mm5		\n\t"\
+        "packuswb %%mm5, %%mm5		\n\t"\
+        OP(%%mm5, out, %%mm7, d)
+
+#define QPEL_BASE(OPNAME, ROUNDER, RND, OP_MMX2, OP_3DNOW)\
+void OPNAME ## mpeg4_qpel16_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
+    uint64_t temp;\
+\
+    asm volatile(\
+        "pxor %%mm7, %%mm7		\n\t"\
+        "1:				\n\t"\
+        "movq  (%0), %%mm0		\n\t" /* ABCDEFGH */\
+        "movq %%mm0, %%mm1		\n\t" /* ABCDEFGH */\
+        "movq %%mm0, %%mm2		\n\t" /* ABCDEFGH */\
+        "punpcklbw %%mm7, %%mm0		\n\t" /* 0A0B0C0D */\
+        "punpckhbw %%mm7, %%mm1		\n\t" /* 0E0F0G0H */\
+        "pshufw $0x90, %%mm0, %%mm5	\n\t" /* 0A0A0B0C */\
+        "pshufw $0x41, %%mm0, %%mm6	\n\t" /* 0B0A0A0B */\
+        "movq %%mm2, %%mm3		\n\t" /* ABCDEFGH */\
+        "movq %%mm2, %%mm4		\n\t" /* ABCDEFGH */\
+        "psllq $8, %%mm2		\n\t" /* 0ABCDEFG */\
+        "psllq $16, %%mm3		\n\t" /* 00ABCDEF */\
+        "psllq $24, %%mm4		\n\t" /* 000ABCDE */\
+        "punpckhbw %%mm7, %%mm2		\n\t" /* 0D0E0F0G */\
+        "punpckhbw %%mm7, %%mm3		\n\t" /* 0C0D0E0F */\
+        "punpckhbw %%mm7, %%mm4		\n\t" /* 0B0C0D0E */\
+        "paddw %%mm3, %%mm5		\n\t" /* b */\
+        "paddw %%mm2, %%mm6		\n\t" /* c */\
+        "paddw %%mm5, %%mm5		\n\t" /* 2b */\
+        "psubw %%mm5, %%mm6		\n\t" /* c - 2b */\
+        "pshufw $0x06, %%mm0, %%mm5	\n\t" /* 0C0B0A0A */\
+        "pmullw "MANGLE(ff_pw_3)", %%mm6		\n\t" /* 3c - 6b */\
+        "paddw %%mm4, %%mm0		\n\t" /* a */\
+        "paddw %%mm1, %%mm5		\n\t" /* d */\
+        "pmullw "MANGLE(ff_pw_20)", %%mm0		\n\t" /* 20a */\
+        "psubw %%mm5, %%mm0		\n\t" /* 20a - d */\
+        "paddw %6, %%mm6		\n\t"\
+        "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\
+        "psraw $5, %%mm0		\n\t"\
+        "movq %%mm0, %5			\n\t"\
+        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
+        \
+        "movq 5(%0), %%mm0		\n\t" /* FGHIJKLM */\
+        "movq %%mm0, %%mm5		\n\t" /* FGHIJKLM */\
+        "movq %%mm0, %%mm6		\n\t" /* FGHIJKLM */\
+        "psrlq $8, %%mm0		\n\t" /* GHIJKLM0 */\
+        "psrlq $16, %%mm5		\n\t" /* HIJKLM00 */\
+        "punpcklbw %%mm7, %%mm0		\n\t" /* 0G0H0I0J */\
+        "punpcklbw %%mm7, %%mm5		\n\t" /* 0H0I0J0K */\
+        "paddw %%mm0, %%mm2		\n\t" /* b */\
+        "paddw %%mm5, %%mm3		\n\t" /* c */\
+        "paddw %%mm2, %%mm2		\n\t" /* 2b */\
+        "psubw %%mm2, %%mm3		\n\t" /* c - 2b */\
+        "movq %%mm6, %%mm2		\n\t" /* FGHIJKLM */\
+        "psrlq $24, %%mm6		\n\t" /* IJKLM000 */\
+        "punpcklbw %%mm7, %%mm2		\n\t" /* 0F0G0H0I */\
+        "punpcklbw %%mm7, %%mm6		\n\t" /* 0I0J0K0L */\
+        "pmullw "MANGLE(ff_pw_3)", %%mm3		\n\t" /* 3c - 6b */\
+        "paddw %%mm2, %%mm1		\n\t" /* a */\
+        "paddw %%mm6, %%mm4		\n\t" /* d */\
+        "pmullw "MANGLE(ff_pw_20)", %%mm1		\n\t" /* 20a */\
+        "psubw %%mm4, %%mm3		\n\t" /* - 6b +3c - d */\
+        "paddw %6, %%mm1		\n\t"\
+        "paddw %%mm1, %%mm3		\n\t" /* 20a - 6b +3c - d */\
+        "psraw $5, %%mm3		\n\t"\
+        "movq %5, %%mm1			\n\t"\
+        "packuswb %%mm3, %%mm1		\n\t"\
+        OP_MMX2(%%mm1, (%1),%%mm4, q)\
+        /* mm0= GHIJ, mm2=FGHI, mm5=HIJK, mm6=IJKL, mm7=0 */\
+        \
+        "movq 9(%0), %%mm1		\n\t" /* JKLMNOPQ */\
+        "movq %%mm1, %%mm4		\n\t" /* JKLMNOPQ */\
+        "movq %%mm1, %%mm3		\n\t" /* JKLMNOPQ */\
+        "psrlq $8, %%mm1		\n\t" /* KLMNOPQ0 */\
+        "psrlq $16, %%mm4		\n\t" /* LMNOPQ00 */\
+        "punpcklbw %%mm7, %%mm1		\n\t" /* 0K0L0M0N */\
+        "punpcklbw %%mm7, %%mm4		\n\t" /* 0L0M0N0O */\
+        "paddw %%mm1, %%mm5		\n\t" /* b */\
+        "paddw %%mm4, %%mm0		\n\t" /* c */\
+        "paddw %%mm5, %%mm5		\n\t" /* 2b */\
+        "psubw %%mm5, %%mm0		\n\t" /* c - 2b */\
+        "movq %%mm3, %%mm5		\n\t" /* JKLMNOPQ */\
+        "psrlq $24, %%mm3		\n\t" /* MNOPQ000 */\
+        "pmullw "MANGLE(ff_pw_3)", %%mm0		\n\t" /* 3c - 6b */\
+        "punpcklbw %%mm7, %%mm3		\n\t" /* 0M0N0O0P */\
+        "paddw %%mm3, %%mm2		\n\t" /* d */\
+        "psubw %%mm2, %%mm0		\n\t" /* -6b + 3c - d */\
+        "movq %%mm5, %%mm2		\n\t" /* JKLMNOPQ */\
+        "punpcklbw %%mm7, %%mm2		\n\t" /* 0J0K0L0M */\
+        "punpckhbw %%mm7, %%mm5		\n\t" /* 0N0O0P0Q */\
+        "paddw %%mm2, %%mm6		\n\t" /* a */\
+        "pmullw "MANGLE(ff_pw_20)", %%mm6		\n\t" /* 20a */\
+        "paddw %6, %%mm0		\n\t"\
+        "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\
+        "psraw $5, %%mm0		\n\t"\
+        /* mm1=KLMN, mm2=JKLM, mm3=MNOP, mm4=LMNO, mm5=NOPQ mm7=0 */\
+        \
+        "paddw %%mm5, %%mm3		\n\t" /* a */\
+        "pshufw $0xF9, %%mm5, %%mm6	\n\t" /* 0O0P0Q0Q */\
+        "paddw %%mm4, %%mm6		\n\t" /* b */\
+        "pshufw $0xBE, %%mm5, %%mm4	\n\t" /* 0P0Q0Q0P */\
+        "pshufw $0x6F, %%mm5, %%mm5	\n\t" /* 0Q0Q0P0O */\
+        "paddw %%mm1, %%mm4		\n\t" /* c */\
+        "paddw %%mm2, %%mm5		\n\t" /* d */\
+        "paddw %%mm6, %%mm6		\n\t" /* 2b */\
+        "psubw %%mm6, %%mm4		\n\t" /* c - 2b */\
+        "pmullw "MANGLE(ff_pw_20)", %%mm3		\n\t" /* 20a */\
+        "pmullw "MANGLE(ff_pw_3)", %%mm4		\n\t" /* 3c - 6b */\
+        "psubw %%mm5, %%mm3		\n\t" /* -6b + 3c - d */\
+        "paddw %6, %%mm4		\n\t"\
+        "paddw %%mm3, %%mm4		\n\t" /* 20a - 6b + 3c - d */\
+        "psraw $5, %%mm4		\n\t"\
+        "packuswb %%mm4, %%mm0		\n\t"\
+        OP_MMX2(%%mm0, 8(%1), %%mm4, q)\
+        \
+        "addl %3, %0			\n\t"\
+        "addl %4, %1			\n\t"\
+        "decl %2			\n\t"\
+        " jnz 1b				\n\t"\
+        : "+a"(src), "+c"(dst), "+m"(h)\
+        : "d"(srcStride), "S"(dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
+        : "memory"\
+    );\
+}\
+\
+static void OPNAME ## mpeg4_qpel16_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
+    int i;\
+    int16_t temp[16];\
+    /* quick HACK, XXX FIXME MUST be optimized */\
+    for(i=0; i<h; i++)\
+    {\
+        temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
+        temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
+        temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
+        temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
+        temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
+        temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 9]);\
+        temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 9])*3 - (src[ 3]+src[10]);\
+        temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 9])*6 + (src[ 5]+src[10])*3 - (src[ 4]+src[11]);\
+        temp[ 8]= (src[ 8]+src[ 9])*20 - (src[ 7]+src[10])*6 + (src[ 6]+src[11])*3 - (src[ 5]+src[12]);\
+        temp[ 9]= (src[ 9]+src[10])*20 - (src[ 8]+src[11])*6 + (src[ 7]+src[12])*3 - (src[ 6]+src[13]);\
+        temp[10]= (src[10]+src[11])*20 - (src[ 9]+src[12])*6 + (src[ 8]+src[13])*3 - (src[ 7]+src[14]);\
+        temp[11]= (src[11]+src[12])*20 - (src[10]+src[13])*6 + (src[ 9]+src[14])*3 - (src[ 8]+src[15]);\
+        temp[12]= (src[12]+src[13])*20 - (src[11]+src[14])*6 + (src[10]+src[15])*3 - (src[ 9]+src[16]);\
+        temp[13]= (src[13]+src[14])*20 - (src[12]+src[15])*6 + (src[11]+src[16])*3 - (src[10]+src[16]);\
+        temp[14]= (src[14]+src[15])*20 - (src[13]+src[16])*6 + (src[12]+src[16])*3 - (src[11]+src[15]);\
+        temp[15]= (src[15]+src[16])*20 - (src[14]+src[16])*6 + (src[13]+src[15])*3 - (src[12]+src[14]);\
+        asm volatile(\
+            "movq (%0), %%mm0		\n\t"\
+            "movq 8(%0), %%mm1		\n\t"\
+            "paddw %2, %%mm0		\n\t"\
+            "paddw %2, %%mm1		\n\t"\
+            "psraw $5, %%mm0		\n\t"\
+            "psraw $5, %%mm1		\n\t"\
+            "packuswb %%mm1, %%mm0	\n\t"\
+            OP_3DNOW(%%mm0, (%1), %%mm1, q)\
+            "movq 16(%0), %%mm0		\n\t"\
+            "movq 24(%0), %%mm1		\n\t"\
+            "paddw %2, %%mm0		\n\t"\
+            "paddw %2, %%mm1		\n\t"\
+            "psraw $5, %%mm0		\n\t"\
+            "psraw $5, %%mm1		\n\t"\
+            "packuswb %%mm1, %%mm0	\n\t"\
+            OP_3DNOW(%%mm0, 8(%1), %%mm1, q)\
+            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
+            : "memory"\
+        );\
+        dst+=dstStride;\
+        src+=srcStride;\
+    }\
+}\
+\
+void OPNAME ## mpeg4_qpel8_h_lowpass_mmx2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
+    uint64_t temp;\
+\
+    asm volatile(\
+        "pxor %%mm7, %%mm7		\n\t"\
+        "1:				\n\t"\
+        "movq  (%0), %%mm0		\n\t" /* ABCDEFGH */\
+        "movq %%mm0, %%mm1		\n\t" /* ABCDEFGH */\
+        "movq %%mm0, %%mm2		\n\t" /* ABCDEFGH */\
+        "punpcklbw %%mm7, %%mm0		\n\t" /* 0A0B0C0D */\
+        "punpckhbw %%mm7, %%mm1		\n\t" /* 0E0F0G0H */\
+        "pshufw $0x90, %%mm0, %%mm5	\n\t" /* 0A0A0B0C */\
+        "pshufw $0x41, %%mm0, %%mm6	\n\t" /* 0B0A0A0B */\
+        "movq %%mm2, %%mm3		\n\t" /* ABCDEFGH */\
+        "movq %%mm2, %%mm4		\n\t" /* ABCDEFGH */\
+        "psllq $8, %%mm2		\n\t" /* 0ABCDEFG */\
+        "psllq $16, %%mm3		\n\t" /* 00ABCDEF */\
+        "psllq $24, %%mm4		\n\t" /* 000ABCDE */\
+        "punpckhbw %%mm7, %%mm2		\n\t" /* 0D0E0F0G */\
+        "punpckhbw %%mm7, %%mm3		\n\t" /* 0C0D0E0F */\
+        "punpckhbw %%mm7, %%mm4		\n\t" /* 0B0C0D0E */\
+        "paddw %%mm3, %%mm5		\n\t" /* b */\
+        "paddw %%mm2, %%mm6		\n\t" /* c */\
+        "paddw %%mm5, %%mm5		\n\t" /* 2b */\
+        "psubw %%mm5, %%mm6		\n\t" /* c - 2b */\
+        "pshufw $0x06, %%mm0, %%mm5	\n\t" /* 0C0B0A0A */\
+        "pmullw "MANGLE(ff_pw_3)", %%mm6		\n\t" /* 3c - 6b */\
+        "paddw %%mm4, %%mm0		\n\t" /* a */\
+        "paddw %%mm1, %%mm5		\n\t" /* d */\
+        "pmullw "MANGLE(ff_pw_20)", %%mm0		\n\t" /* 20a */\
+        "psubw %%mm5, %%mm0		\n\t" /* 20a - d */\
+        "paddw %6, %%mm6		\n\t"\
+        "paddw %%mm6, %%mm0		\n\t" /* 20a - 6b + 3c - d */\
+        "psraw $5, %%mm0		\n\t"\
+        /* mm1=EFGH, mm2=DEFG, mm3=CDEF, mm4=BCDE, mm7=0 */\
+        \
+        "movd 5(%0), %%mm5		\n\t" /* FGHI */\
+        "punpcklbw %%mm7, %%mm5		\n\t" /* 0F0G0H0I */\
+        "pshufw $0xF9, %%mm5, %%mm6	\n\t" /* 0G0H0I0I */\
+        "paddw %%mm5, %%mm1		\n\t" /* a */\
+        "paddw %%mm6, %%mm2		\n\t" /* b */\
+        "pshufw $0xBE, %%mm5, %%mm6	\n\t" /* 0H0I0I0H */\
+        "pshufw $0x6F, %%mm5, %%mm5	\n\t" /* 0I0I0H0G */\
+        "paddw %%mm6, %%mm3		\n\t" /* c */\
+        "paddw %%mm5, %%mm4		\n\t" /* d */\
+        "paddw %%mm2, %%mm2		\n\t" /* 2b */\
+        "psubw %%mm2, %%mm3		\n\t" /* c - 2b */\
+        "pmullw "MANGLE(ff_pw_20)", %%mm1		\n\t" /* 20a */\
+        "pmullw "MANGLE(ff_pw_3)", %%mm3		\n\t" /* 3c - 6b */\
+        "psubw %%mm4, %%mm3		\n\t" /* -6b + 3c - d */\
+        "paddw %6, %%mm1		\n\t"\
+        "paddw %%mm1, %%mm3		\n\t" /* 20a - 6b + 3c - d */\
+        "psraw $5, %%mm3		\n\t"\
+        "packuswb %%mm3, %%mm0		\n\t"\
+        OP_MMX2(%%mm0, (%1), %%mm4, q)\
+        \
+        "addl %3, %0			\n\t"\
+        "addl %4, %1			\n\t"\
+        "decl %2			\n\t"\
+        " jnz 1b			\n\t"\
+        : "+a"(src), "+c"(dst), "+m"(h)\
+        : "S"(srcStride), "D"(dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(temp), "m"(ROUNDER)\
+        : "memory"\
+    );\
+}\
+\
+static void OPNAME ## mpeg4_qpel8_h_lowpass_3dnow(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\
+    int i;\
+    int16_t temp[8];\
+    /* quick HACK, XXX FIXME MUST be optimized */\
+    for(i=0; i<h; i++)\
+    {\
+        temp[ 0]= (src[ 0]+src[ 1])*20 - (src[ 0]+src[ 2])*6 + (src[ 1]+src[ 3])*3 - (src[ 2]+src[ 4]);\
+        temp[ 1]= (src[ 1]+src[ 2])*20 - (src[ 0]+src[ 3])*6 + (src[ 0]+src[ 4])*3 - (src[ 1]+src[ 5]);\
+        temp[ 2]= (src[ 2]+src[ 3])*20 - (src[ 1]+src[ 4])*6 + (src[ 0]+src[ 5])*3 - (src[ 0]+src[ 6]);\
+        temp[ 3]= (src[ 3]+src[ 4])*20 - (src[ 2]+src[ 5])*6 + (src[ 1]+src[ 6])*3 - (src[ 0]+src[ 7]);\
+        temp[ 4]= (src[ 4]+src[ 5])*20 - (src[ 3]+src[ 6])*6 + (src[ 2]+src[ 7])*3 - (src[ 1]+src[ 8]);\
+        temp[ 5]= (src[ 5]+src[ 6])*20 - (src[ 4]+src[ 7])*6 + (src[ 3]+src[ 8])*3 - (src[ 2]+src[ 8]);\
+        temp[ 6]= (src[ 6]+src[ 7])*20 - (src[ 5]+src[ 8])*6 + (src[ 4]+src[ 8])*3 - (src[ 3]+src[ 7]);\
+        temp[ 7]= (src[ 7]+src[ 8])*20 - (src[ 6]+src[ 8])*6 + (src[ 5]+src[ 7])*3 - (src[ 4]+src[ 6]);\
+        asm volatile(\
+            "movq (%0), %%mm0		\n\t"\
+            "movq 8(%0), %%mm1		\n\t"\
+            "paddw %2, %%mm0		\n\t"\
+            "paddw %2, %%mm1		\n\t"\
+            "psraw $5, %%mm0		\n\t"\
+            "psraw $5, %%mm1		\n\t"\
+            "packuswb %%mm1, %%mm0	\n\t"\
+            OP_3DNOW(%%mm0, (%1), %%mm1, q)\
+            :: "r"(temp), "r"(dst), "m"(ROUNDER)\
+            :"memory"\
+        );\
+        dst+=dstStride;\
+        src+=srcStride;\
+    }\
+}
+
+#define QPEL_OP(OPNAME, ROUNDER, RND, OP, MMX)\
+\
+static void OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
+    uint64_t temp[17*4];\
+    uint64_t *temp_ptr= temp;\
+    int count= 17;\
+\
+    /*FIXME unroll */\
+    asm volatile(\
+        "pxor %%mm7, %%mm7		\n\t"\
+        "1:				\n\t"\
+        "movq (%0), %%mm0		\n\t"\
+        "movq (%0), %%mm1		\n\t"\
+        "movq 8(%0), %%mm2		\n\t"\
+        "movq 8(%0), %%mm3		\n\t"\
+        "punpcklbw %%mm7, %%mm0		\n\t"\
+        "punpckhbw %%mm7, %%mm1		\n\t"\
+        "punpcklbw %%mm7, %%mm2		\n\t"\
+        "punpckhbw %%mm7, %%mm3		\n\t"\
+        "movq %%mm0, (%1)		\n\t"\
+        "movq %%mm1, 17*8(%1)		\n\t"\
+        "movq %%mm2, 2*17*8(%1)		\n\t"\
+        "movq %%mm3, 3*17*8(%1)		\n\t"\
+        "addl $8, %1			\n\t"\
+        "addl %3, %0			\n\t"\
+        "decl %2			\n\t"\
+        " jnz 1b			\n\t"\
+        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
+        : "r" (srcStride)\
+        : "memory"\
+    );\
+    \
+    temp_ptr= temp;\
+    count=4;\
+    \
+/*FIXME reorder for speed */\
+    asm volatile(\
+        /*"pxor %%mm7, %%mm7		\n\t"*/\
+        "1:				\n\t"\
+        "movq (%0), %%mm0		\n\t"\
+        "movq 8(%0), %%mm1		\n\t"\
+        "movq 16(%0), %%mm2		\n\t"\
+        "movq 24(%0), %%mm3		\n\t"\
+        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
+        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
+        "addl %4, %1			\n\t"\
+        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
+        \
+        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
+        "addl %4, %1			\n\t"\
+        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
+        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 72(%0), (%1, %3), OP)\
+        "addl %4, %1			\n\t"\
+        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 80(%0), (%1), OP)\
+        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 88(%0), (%1, %3), OP)\
+        "addl %4, %1			\n\t"\
+        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 40(%0), 48(%0), 56(%0), 96(%0), (%1), OP)\
+        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 48(%0), 56(%0), 64(%0),104(%0), (%1, %3), OP)\
+        "addl %4, %1			\n\t"\
+        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 56(%0), 64(%0), 72(%0),112(%0), (%1), OP)\
+        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 64(%0), 72(%0), 80(%0),120(%0), (%1, %3), OP)\
+        "addl %4, %1			\n\t"\
+        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 72(%0), 80(%0), 88(%0),128(%0), (%1), OP)\
+        \
+        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 80(%0), 88(%0), 96(%0),128(%0), (%1, %3), OP)\
+        "addl %4, %1			\n\t"  \
+        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 88(%0), 96(%0),104(%0),120(%0), (%1), OP)\
+        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 96(%0),104(%0),112(%0),112(%0), (%1, %3), OP)\
+        \
+        "addl $136, %0			\n\t"\
+        "addl %6, %1			\n\t"\
+        "decl %2			\n\t"\
+        " jnz 1b			\n\t"\
+        \
+        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
+        : "r"(dstStride), "r"(2*dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-14*dstStride)\
+        :"memory"\
+    );\
+}\
+\
+void OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\
+    uint64_t temp[9*4];\
+    uint64_t *temp_ptr= temp;\
+    int count= 9;\
+\
+    /*FIXME unroll */\
+    asm volatile(\
+        "pxor %%mm7, %%mm7		\n\t"\
+        "1:				\n\t"\
+        "movq (%0), %%mm0		\n\t"\
+        "movq (%0), %%mm1		\n\t"\
+        "punpcklbw %%mm7, %%mm0		\n\t"\
+        "punpckhbw %%mm7, %%mm1		\n\t"\
+        "movq %%mm0, (%1)		\n\t"\
+        "movq %%mm1, 9*8(%1)		\n\t"\
+        "addl $8, %1			\n\t"\
+        "addl %3, %0			\n\t"\
+        "decl %2			\n\t"\
+        " jnz 1b			\n\t"\
+        : "+r" (src), "+r" (temp_ptr), "+r"(count)\
+        : "r" (srcStride)\
+        : "memory"\
+    );\
+    \
+    temp_ptr= temp;\
+    count=2;\
+    \
+/*FIXME reorder for speed */\
+    asm volatile(\
+        /*"pxor %%mm7, %%mm7		\n\t"*/\
+        "1:				\n\t"\
+        "movq (%0), %%mm0		\n\t"\
+        "movq 8(%0), %%mm1		\n\t"\
+        "movq 16(%0), %%mm2		\n\t"\
+        "movq 24(%0), %%mm3		\n\t"\
+        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5, 16(%0),  8(%0),   (%0), 32(%0), (%1), OP)\
+        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5,  8(%0),   (%0),   (%0), 40(%0), (%1, %3), OP)\
+        "addl %4, %1			\n\t"\
+        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5,   (%0),   (%0),  8(%0), 48(%0), (%1), OP)\
+        \
+        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5,   (%0),  8(%0), 16(%0), 56(%0), (%1, %3), OP)\
+        "addl %4, %1			\n\t"\
+        QPEL_V_LOW(%%mm0, %%mm1, %%mm2, %%mm3, %5, %6, %5,  8(%0), 16(%0), 24(%0), 64(%0), (%1), OP)\
+        \
+        QPEL_V_LOW(%%mm1, %%mm2, %%mm3, %%mm0, %5, %6, %5, 16(%0), 24(%0), 32(%0), 64(%0), (%1, %3), OP)\
+        "addl %4, %1			\n\t"\
+        QPEL_V_LOW(%%mm2, %%mm3, %%mm0, %%mm1, %5, %6, %5, 24(%0), 32(%0), 40(%0), 56(%0), (%1), OP)\
+        QPEL_V_LOW(%%mm3, %%mm0, %%mm1, %%mm2, %5, %6, %5, 32(%0), 40(%0), 48(%0), 48(%0), (%1, %3), OP)\
+                \
+        "addl $72, %0			\n\t"\
+        "addl %6, %1			\n\t"\
+        "decl %2			\n\t"\
+        " jnz 1b			\n\t"\
+         \
+        : "+r"(temp_ptr), "+r"(dst), "+g"(count)\
+        : "r"(dstStride), "r"(2*dstStride), /*"m"(ff_pw_20), "m"(ff_pw_3),*/ "m"(ROUNDER), "g"(4-6*dstStride)\
+        : "memory"\
+   );\
+}\
+\
+static void OPNAME ## qpel8_mc00_ ## MMX (UINT8 *dst, UINT8 *src, int stride){\
+    OPNAME ## pixels8_mmx(dst, src, stride, 8);\
+}\
+\
+static void OPNAME ## qpel8_mc10_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
+    uint64_t temp[32];\
+    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);\
+}\
+\
+static void OPNAME ## qpel8_mc20_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
+    OPNAME ## mpeg4_qpel8_h_lowpass_ ## MMX(dst, src, stride, stride, 8);\
+}\
+\
+static void OPNAME ## qpel8_mc30_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
+    uint64_t temp[32];\
+    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];\
+    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);\
+}\
+\
+static void OPNAME ## qpel8_mc02_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
+    OPNAME ## mpeg4_qpel8_v_lowpass_ ## MMX(dst, src, stride, stride);\
+}\
+\
+static void OPNAME ## qpel8_mc03_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
+    uint64_t temp[32];\
+    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;\
+    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_l4_mmx(dst, src, (uint8_t*)half, stride, 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;\
+    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_l4_mmx(dst, src+1, (uint8_t*)half, stride, 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;\
+    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_l4_mmx(dst, src+stride, (uint8_t*)half, stride, 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);\
+    put ## RND ## mpeg4_qpel8_v_lowpass_ ## MMX(halfHV, halfH, 8, 8);\
+    OPNAME ## pixels8_l4_mmx(dst, src+stride+1, (uint8_t*)half, stride, 8);\
+}\
+static void OPNAME ## qpel8_mc21_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
+    uint64_t half[8*2 + 9*2];\
+    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(halfHV, halfH, 8, 8);\
+    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];\
+    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(halfHV, halfH, 8, 8);\
+    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;\
+    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);\
+}\
+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;\
+    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);\
+}\
+static void OPNAME ## qpel8_mc22_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
+    uint64_t half[9*2];\
+    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);\
+}\
+static void OPNAME ## qpel16_mc00_ ## MMX (UINT8 *dst, UINT8 *src, int stride){\
+    OPNAME ## pixels16_mmx(dst, src, stride, 16);\
+}\
+\
+static void OPNAME ## qpel16_mc10_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
+    uint64_t temp[32];\
+    uint8_t * const half= (uint8_t*)temp;\
+    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
+    OPNAME ## pixels16_l2_mmx(dst, src, half, stride, stride, 16);\
+}\
+\
+static void OPNAME ## qpel16_mc20_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
+    OPNAME ## mpeg4_qpel16_h_lowpass_ ## MMX(dst, src, stride, stride, 16);\
+}\
+\
+static void OPNAME ## qpel16_mc30_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
+    uint64_t temp[32];\
+    uint8_t * const half= (uint8_t*)temp;\
+    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(half, src, 16, stride, 16);\
+    OPNAME ## pixels16_l2_mmx(dst, src+1, half, stride, stride, 16);\
+}\
+\
+static void OPNAME ## qpel16_mc01_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
+    uint64_t temp[32];\
+    uint8_t * const half= (uint8_t*)temp;\
+    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, stride);\
+    OPNAME ## pixels16_l2_mmx(dst, src, half, stride, stride, 16);\
+}\
+\
+static void OPNAME ## qpel16_mc02_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
+    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, src, stride, stride);\
+}\
+\
+static void OPNAME ## qpel16_mc03_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
+    uint64_t temp[32];\
+    uint8_t * const half= (uint8_t*)temp;\
+    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(half, src, 16, 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;\
+    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_l4_mmx(dst, src, (uint8_t*)half, stride, 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;\
+    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_l4_mmx(dst, src+1, (uint8_t*)half, stride, 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;\
+    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_l4_mmx(dst, src+stride, (uint8_t*)half, stride, 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);\
+    put ## RND ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
+    OPNAME ## pixels16_l4_mmx(dst, src+stride+1, (uint8_t*)half, stride, 16);\
+}\
+static void OPNAME ## qpel16_mc21_ ## MMX(UINT8 *dst, UINT8 *src, int 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 ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
+    OPNAME ## pixels16_l2_mmx(dst, halfH, halfHV, stride, 16, 16);\
+}\
+static void OPNAME ## qpel16_mc23_ ## MMX(UINT8 *dst, UINT8 *src, int 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 ## mpeg4_qpel16_v_lowpass_ ## MMX(halfHV, halfH, 16, 16);\
+    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;\
+    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);\
+}\
+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;\
+    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);\
+}\
+static void OPNAME ## qpel16_mc22_ ## MMX(UINT8 *dst, UINT8 *src, int stride){\
+    uint64_t half[17*2];\
+    uint8_t * const halfH= ((uint8_t*)half);\
+    put ## RND ## mpeg4_qpel16_h_lowpass_ ## MMX(halfH, src, 16, stride, 17);\
+    OPNAME ## mpeg4_qpel16_v_lowpass_ ## MMX(dst, halfH, stride, 16);\
+}
+
+
+#define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "	\n\t"
+#define AVG_3DNOW_OP(a,b,temp, size) \
+"mov" #size " " #b ", " #temp "	\n\t"\
+"pavgusb " #temp ", " #a "	\n\t"\
+"mov" #size " " #a ", " #b "	\n\t"
+#define AVG_MMX2_OP(a,b,temp, size) \
+"mov" #size " " #b ", " #temp "	\n\t"\
+"pavgb " #temp ", " #a "	\n\t"\
+"mov" #size " " #a ", " #b "	\n\t"
+
+QPEL_BASE(put_       , ff_pw_16, _       , PUT_OP, PUT_OP)
+QPEL_BASE(avg_       , ff_pw_16, _       , AVG_MMX2_OP, AVG_3DNOW_OP)
+QPEL_BASE(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, PUT_OP)
+QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, 3dnow)
+QPEL_OP(avg_       , ff_pw_16, _       , AVG_3DNOW_OP, 3dnow)
+QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, 3dnow)
+QPEL_OP(put_       , ff_pw_16, _       , PUT_OP, mmx2)
+QPEL_OP(avg_       , ff_pw_16, _       , AVG_MMX2_OP, mmx2)
+QPEL_OP(put_no_rnd_, ff_pw_15, _no_rnd_, PUT_OP, mmx2)
 
 #if 0
 static void just_return() { return; }
 #endif
 
+#define SET_QPEL_FUNC(postfix1, postfix2) \
+    c->put_ ## postfix1 = put_ ## postfix2;\
+    c->put_no_rnd_ ## postfix1 = put_no_rnd_ ## postfix2;\
+    c->avg_ ## postfix1 = avg_ ## postfix2;
+    
 void dsputil_init_mmx(DSPContext* c, unsigned mask)
 {
     mm_flags = mm_support();
@@ -576,10 +1427,16 @@ void dsputil_init_mmx(DSPContext* c, unsigned mask)
         c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_mmx;
         c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_mmx;
         c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_mmx;
-        
+                
         c->add_bytes= add_bytes_mmx;
         c->diff_bytes= diff_bytes_mmx;
-
+        
+        c->hadamard8_diff[0]= hadamard8_diff16_mmx;
+        c->hadamard8_diff[1]= hadamard8_diff_mmx;
+        
+        c->sad[0]= sad16x16_mmx;
+        c->sad[1]= sad8x8_mmx;
+        
         if (mm_flags & MM_MMXEXT) {
             c->pix_abs16x16     = pix_abs16x16_mmx2;
             c->pix_abs16x16_x2  = pix_abs16x16_x2_mmx2;
@@ -591,6 +1448,9 @@ void dsputil_init_mmx(DSPContext* c, unsigned mask)
             c->pix_abs8x8_y2  = pix_abs8x8_y2_mmx2;
             c->pix_abs8x8_xy2 = pix_abs8x8_xy2_mmx2;
 
+            c->sad[0]= sad16x16_mmx2;
+            c->sad[1]= sad8x8_mmx2;
+            
             c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
             c->put_pixels_tab[0][2] = put_pixels16_y2_mmx2;
             c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_mmx2;
@@ -610,6 +1470,41 @@ void dsputil_init_mmx(DSPContext* c, unsigned mask)
             c->avg_pixels_tab[1][1] = avg_pixels8_x2_mmx2;
             c->avg_pixels_tab[1][2] = avg_pixels8_y2_mmx2;
             c->avg_pixels_tab[1][3] = avg_pixels8_xy2_mmx2;
+
+#if 1
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 0], qpel16_mc00_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 1], qpel16_mc10_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 2], qpel16_mc20_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_mmx2)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_mmx2)
+#endif
         } else if (mm_flags & MM_3DNOW) {
             c->put_pixels_tab[0][1] = put_pixels16_x2_3dnow;
             c->put_pixels_tab[0][2] = put_pixels16_y2_3dnow;
@@ -630,6 +1525,39 @@ 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)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 3], qpel16_mc30_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 4], qpel16_mc01_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 5], qpel16_mc11_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 6], qpel16_mc21_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 7], qpel16_mc31_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 8], qpel16_mc02_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][ 9], qpel16_mc12_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][10], qpel16_mc22_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][11], qpel16_mc32_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][12], qpel16_mc03_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][13], qpel16_mc13_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][14], qpel16_mc23_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[0][15], qpel16_mc33_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 0], qpel8_mc00_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 1], qpel8_mc10_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 2], qpel8_mc20_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 3], qpel8_mc30_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 4], qpel8_mc01_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 5], qpel8_mc11_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 6], qpel8_mc21_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 7], qpel8_mc31_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 8], qpel8_mc02_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][ 9], qpel8_mc12_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][10], qpel8_mc22_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][11], qpel8_mc32_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][12], qpel8_mc03_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][13], qpel8_mc13_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][14], qpel8_mc23_3dnow)
+            SET_QPEL_FUNC(qpel_pixels_tab[1][15], qpel8_mc33_3dnow)
         }
     }
 
diff --git a/src/libffmpeg/libavcodec/i386/dsputil_mmx_avg.h b/src/libffmpeg/libavcodec/i386/dsputil_mmx_avg.h
index 6873432ce..4a8841156 100644
--- a/src/libffmpeg/libavcodec/i386/dsputil_mmx_avg.h
+++ b/src/libffmpeg/libavcodec/i386/dsputil_mmx_avg.h
@@ -53,6 +53,38 @@ static void DEF(put_pixels8_x2)(UINT8 *block, const UINT8 *pixels, int line_size
 	:"%eax", "memory");
 }
 
+static void DEF(put_pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
+{
+    __asm __volatile(
+	"1:				\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"addl	%4, %1			\n\t"
+	"movq	(%1), %%mm1		\n\t"
+	"addl	%4, %1			\n\t"
+	PAVGB" (%2), %%mm0		\n\t"
+	PAVGB" 8(%2), %%mm1		\n\t"
+	"movq	%%mm0, (%3)		\n\t"
+	"addl	%5, %3			\n\t"
+	"movq	%%mm1, (%3)		\n\t"
+	"addl	%5, %3			\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"addl	%4, %1			\n\t"
+	"movq	(%1), %%mm1		\n\t"
+	"addl	%4, %1			\n\t"
+	PAVGB" 16(%2), %%mm0		\n\t"
+	PAVGB" 24(%2), %%mm1		\n\t"
+	"movq	%%mm0, (%3)		\n\t"
+	"addl	%5, %3			\n\t"
+	"movq	%%mm1, (%3)		\n\t"
+	"addl	%5, %3			\n\t"
+        "addl	$32, %2			\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)
 {
     __asm __volatile(
@@ -92,6 +124,34 @@ static void DEF(put_pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_siz
 	:"r" (line_size)
 	:"%eax", "memory");
 }
+
+static void DEF(put_pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)
+{
+    __asm __volatile(
+	"1:				\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"movq	8(%1), %%mm1		\n\t"
+	"addl	%4, %1			\n\t"
+	PAVGB" (%2), %%mm0		\n\t"
+	PAVGB" 8(%2), %%mm1		\n\t"
+	"movq	%%mm0, (%3)		\n\t"
+	"movq	%%mm1, 8(%3)		\n\t"
+	"addl	%5, %3			\n\t"
+	"movq	(%1), %%mm0		\n\t"
+	"movq	8(%1), %%mm1		\n\t"
+	"addl	%4, %1			\n\t"
+	PAVGB" 16(%2), %%mm0		\n\t"
+	PAVGB" 24(%2), %%mm1		\n\t"
+	"movq	%%mm0, (%3)		\n\t"
+	"movq	%%mm1, 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");
+}
  
 /* GL: this function does incorrect rounding if overflow */
 static void DEF(put_no_rnd_pixels8_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
diff --git a/src/libffmpeg/libavcodec/i386/dsputil_mmx_rnd.h b/src/libffmpeg/libavcodec/i386/dsputil_mmx_rnd.h
index 3605e03f9..0ae1cd99d 100644
--- a/src/libffmpeg/libavcodec/i386/dsputil_mmx_rnd.h
+++ b/src/libffmpeg/libavcodec/i386/dsputil_mmx_rnd.h
@@ -54,6 +54,46 @@ 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"
+#ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cant be used
+        :"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
+#else
+        :"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
+#endif
+	:"S"(src1Stride), "D"(dstStride)
+	:"memory");
+}
+
 static void DEF(put, pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
 {
     MOVQ_BFE(mm6);
@@ -90,7 +130,7 @@ 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 +140,42 @@ 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"
+#ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cant be used
+	:"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)
+#else
+	:"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)
+#endif
+	:"S"(src1Stride), "D"(dstStride)
+	:"memory"); 
+}
+
 static void DEF(put, pixels8_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)
 {
     MOVQ_BFE(mm6);
@@ -195,6 +271,124 @@ 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 +453,27 @@ 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 +500,33 @@ 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 +641,133 @@ 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(%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);
diff --git a/src/libffmpeg/libavcodec/i386/fft_sse.c b/src/libffmpeg/libavcodec/i386/fft_sse.c
index 8e8e36b0f..175cea506 100644
--- a/src/libffmpeg/libavcodec/i386/fft_sse.c
+++ b/src/libffmpeg/libavcodec/i386/fft_sse.c
@@ -19,11 +19,16 @@
 #include "../dsputil.h"
 #include <math.h>
 
+#ifdef HAVE_BUILTIN_VECTOR
+
 #include <xmmintrin.h>
 
 static const float p1p1p1m1[4] __attribute__((aligned(16))) = 
     { 1.0, 1.0, 1.0, -1.0 };
 
+static const float p1p1m1p1[4] __attribute__((aligned(16))) = 
+    { 1.0, 1.0, -1.0, 1.0 };
+
 static const float p1p1m1m1[4] __attribute__((aligned(16))) = 
     { 1.0, 1.0, -1.0, -1.0 };
 
@@ -54,6 +59,11 @@ void fft_calc_sse(FFTContext *s, FFTComplex *z)
         r = (__m128 *)&z[0];
         c1 = *(__m128 *)p1p1m1m1;
         c2 = *(__m128 *)p1p1p1m1;
+        if (s->inverse)
+            c2 = *(__m128 *)p1p1m1p1;
+        else
+            c2 = *(__m128 *)p1p1p1m1;
+
         j = (np >> 2);
         do {
             a = r[0];
@@ -126,3 +136,5 @@ void fft_calc_sse(FFTContext *s, FFTComplex *z)
         nloops = nloops << 1;
     } while (nblocks != 0);
 }
+
+#endif
diff --git a/src/libffmpeg/libavcodec/i386/motion_est_mmx.c b/src/libffmpeg/libavcodec/i386/motion_est_mmx.c
index 3368e7333..fa85db67b 100644
--- a/src/libffmpeg/libavcodec/i386/motion_est_mmx.c
+++ b/src/libffmpeg/libavcodec/i386/motion_est_mmx.c
@@ -274,6 +274,15 @@ int pix_abs8x8_ ## suf(UINT8 *blk2, UINT8 *blk1, int stride)\
 \
     return sum_ ## suf();\
 }\
+int sad8x8_ ## suf(void *s, UINT8 *blk2, UINT8 *blk1, int stride)\
+{\
+    asm volatile("pxor %%mm7, %%mm7		\n\t"\
+                 "pxor %%mm6, %%mm6		\n\t":);\
+\
+    sad8_ ## suf(blk1, blk2, stride, 3);\
+\
+    return sum_ ## suf();\
+}\
 \
 int pix_abs8x8_x2_ ## suf(UINT8 *blk2, UINT8 *blk1, int stride)\
 {\
@@ -324,6 +333,16 @@ int pix_abs16x16_ ## suf(UINT8 *blk2, UINT8 *blk1, int stride)\
 \
     return sum_ ## suf();\
 }\
+int sad16x16_ ## suf(void *s, UINT8 *blk2, UINT8 *blk1, int stride)\
+{\
+    asm volatile("pxor %%mm7, %%mm7		\n\t"\
+                 "pxor %%mm6, %%mm6		\n\t":);\
+\
+    sad8_ ## suf(blk1  , blk2  , stride, 4);\
+    sad8_ ## suf(blk1+8, blk2+8, stride, 4);\
+\
+    return sum_ ## suf();\
+}\
 int pix_abs16x16_x2_ ## suf(UINT8 *blk2, UINT8 *blk1, int stride)\
 {\
     asm volatile("pxor %%mm7, %%mm7		\n\t"\
diff --git a/src/libffmpeg/libavcodec/i386/mpegvideo_mmx.c b/src/libffmpeg/libavcodec/i386/mpegvideo_mmx.c
index cb7af3e62..d936abfd5 100644
--- a/src/libffmpeg/libavcodec/i386/mpegvideo_mmx.c
+++ b/src/libffmpeg/libavcodec/i386/mpegvideo_mmx.c
@@ -529,7 +529,6 @@ static void ff_libmpeg2mmx2_idct_add(UINT8 *dest, int line_size, DCTELEM *block)
 void MPV_common_init_mmx(MpegEncContext *s)
 {
     if (mm_flags & MM_MMX) {
-        int i;
         const int dct_algo = s->avctx->dct_algo;
         const int idct_algo= s->avctx->idct_algo;
         
-- 
cgit v1.2.3