sync to libavcodec build 4755

CVS patchset: 7567
CVS date: 2005/05/28 01:40:40

3 files changed, 306 insertions, 437 deletions

diff --git a/src/libffmpeg/libavcodec/i386/dsputil_mmx.c b/src/libffmpeg/libavcodec/i386/dsputil_mmx.c
index 1d1f2b213..32565c3d3 100644
--- a/src/libffmpeg/libavcodec/i386/dsputil_mmx.c
+++ b/src/libffmpeg/libavcodec/i386/dsputil_mmx.c
@@ -45,6 +45,7 @@ static const uint64_t ff_pw_16 attribute_used __attribute__ ((aligned(8))) = 0x0
 static const uint64_t ff_pw_32 attribute_used __attribute__ ((aligned(8))) = 0x0020002000200020ULL;
 static const uint64_t ff_pw_15 attribute_used __attribute__ ((aligned(8))) = 0x000F000F000F000FULL;
 
+static const uint64_t ff_pb_3F attribute_used __attribute__ ((aligned(8))) = 0x3F3F3F3F3F3F3F3FULL;
 static const uint64_t ff_pb_FC attribute_used __attribute__ ((aligned(8))) = 0xFCFCFCFCFCFCFCFCULL;
 
 #define JUMPALIGN() __asm __volatile (".balign 8"::)
@@ -692,204 +693,265 @@ static void h263_h_loop_filter_mmx(uint8_t *src, int stride, int qscale){
     );
 }
 
-// dst = ABS( a - b )
-#define MMABS_DIFF_MMX2(a,b,dst,z)\
-    "movq    " #b ", " #dst " \n\t"\
-    "movq    " #a ", " #z   " \n\t"\
-    "psubusw " #b ", " #z   " \n\t"\
-    "psubusw " #a ", " #dst " \n\t"\
-    "pmaxsw  " #z ", " #dst " \n\t"
-
-// a = clip( a, -tc, tc )
-#define CLIP_MMX2(a,tc,z)\
-    "pxor    " #z  ", " #z "  \n\t"\
-    "psubw   " #tc ", " #z "  \n\t"\
-    "pmaxsw  " #z  ", " #a "  \n\t"\
-    "pminsw  " #tc ", " #a "  \n\t"
-
-// in: mm0=p1, mm1=p0, mm2=q0, mm3=q1
-// out: mm7 = do we filter this pixel?
-#define H264_DEBLOCK_THRESH(alpha,beta)\
-    "pxor      %%mm7, %%mm7     \n\t"\
-    "punpcklbw %%mm7, %%mm0     \n\t"\
-    "punpcklbw %%mm7, %%mm1     \n\t"\
-    "punpcklbw %%mm7, %%mm2     \n\t"\
-    "punpcklbw %%mm7, %%mm3     \n\t"\
-    MMABS_DIFF_MMX2(%%mm1, %%mm2, %%mm5, %%mm4)\
-    "movd " #alpha ", %%mm6     \n\t"\
-    "pshufw    $0, %%mm6, %%mm6 \n\t"\
-    "pcmpgtw   %%mm5, %%mm6     \n\t" /* ABS(p0-q0) < alpha */\
-    MMABS_DIFF_MMX2(%%mm0, %%mm1, %%mm5, %%mm4)\
-    MMABS_DIFF_MMX2(%%mm3, %%mm2, %%mm7, %%mm4)\
-    "pmaxsw    %%mm7, %%mm5     \n\t"\
-    "movd  " #beta ", %%mm7     \n\t"\
-    "pshufw    $0, %%mm7, %%mm7 \n\t"\
-    "movq      %%mm7, %%mm4     \n\t"\
-    "pcmpgtw   %%mm5, %%mm7     \n\t" /* ABS(p1-p0) < beta && ABS(q1-q0) < beta */\
-    "pand      %%mm6, %%mm7     \n\t"
-
-// in: mm0=p1, mm1=p0, mm2=q0, mm3=q1, mm6=tc
-// out: mm1=p0', mm2=q0'
-#define H264_DEBLOCK_P0_Q0(pw4)\
-    "movq   " #pw4 ", %%mm4     \n\t"\
-    "movq      %%mm2, %%mm5     \n\t"\
-    "paddw     %%mm4, %%mm0     \n\t"\
-    "psubw     %%mm1, %%mm5     \n\t"\
-    "psubw     %%mm3, %%mm0     \n\t"\
-    "psllw     $2,    %%mm5     \n\t"\
-    "paddw     %%mm0, %%mm5     \n\t"\
-    "psraw     $3,    %%mm5     \n\t" /* mm5 = (((q0 - p0) << 2) + (p1 - q1) + 4) >> 3 */\
-    CLIP_MMX2(%%mm5, %%mm6, %%mm4)    /* delta = clip( mm5, -tc, tc ) */\
-    "paddw     %%mm5, %%mm1     \n\t" /* p0 += delta */\
-    "psubw     %%mm5, %%mm2     \n\t" /* q0 -= delta */
-
-// in: mm1=p0, mm2=q0, mm6=tc0
-// out: mm5=delta
-#define H264_DEBLOCK_DELTA_PQ1(p1,p2,z)\
-    "movq      %%mm1, %%mm5     \n\t"\
-    "pavgb     %%mm2, %%mm5     \n\t"\
-    "paddw   " #p2 ", %%mm5     \n\t"\
-    "psraw     $1, %%mm5        \n\t"\
-    "psubw   " #p1 ", %%mm5     \n\t" /* ( ( q2 + ((p0+q0+1)>>1) ) >> 1 ) - q1 */\
-    CLIP_MMX2(%%mm5, %%mm6, z)
-
-static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int tc0)
+
+// out: o = |x-y|>a
+// clobbers: t
+#define DIFF_GT_MMX(x,y,a,o,t)\
+    "movq     "#y", "#t"  \n\t"\
+    "movq     "#x", "#o"  \n\t"\
+    "psubusb  "#x", "#t"  \n\t"\
+    "psubusb  "#y", "#o"  \n\t"\
+    "por      "#t", "#o"  \n\t"\
+    "psubusb  "#a", "#o"  \n\t"
+
+// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1
+// out: mm5=beta-1, mm7=mask
+// clobbers: mm4,mm6
+#define H264_DEBLOCK_MASK(alpha1, beta1) \
+    "pshufw $0, "#alpha1", %%mm4 \n\t"\
+    "pshufw $0, "#beta1 ", %%mm5 \n\t"\
+    "packuswb  %%mm4, %%mm4      \n\t"\
+    "packuswb  %%mm5, %%mm5      \n\t"\
+    DIFF_GT_MMX(%%mm1, %%mm2, %%mm4, %%mm7, %%mm6) /* |p0-q0| > alpha-1 */\
+    DIFF_GT_MMX(%%mm0, %%mm1, %%mm5, %%mm4, %%mm6) /* |p1-p0| > beta-1 */\
+    "por       %%mm4, %%mm7      \n\t"\
+    DIFF_GT_MMX(%%mm3, %%mm2, %%mm5, %%mm4, %%mm6) /* |q1-q0| > beta-1 */\
+    "por       %%mm4, %%mm7      \n\t"\
+    "pxor      %%mm6, %%mm6      \n\t"\
+    "pcmpeqb   %%mm6, %%mm7      \n\t"
+
+// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask)
+// out: mm1=p0' mm2=q0'
+// clobbers: mm0,3-6
+#define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\
+        /* a = q0^p0^((p1-q1)>>2) */\
+        "movq    %%mm0, %%mm4  \n\t"\
+        "psubb   %%mm3, %%mm4  \n\t"\
+        "psrlw   $2,    %%mm4  \n\t"\
+        "pxor    %%mm1, %%mm4  \n\t"\
+        "pxor    %%mm2, %%mm4  \n\t"\
+        /* b = p0^(q1>>2) */\
+        "psrlw   $2,    %%mm3  \n\t"\
+        "pand "#pb_3f", %%mm3  \n\t"\
+        "movq    %%mm1, %%mm5  \n\t"\
+        "pxor    %%mm3, %%mm5  \n\t"\
+        /* c = q0^(p1>>2) */\
+        "psrlw   $2,    %%mm0  \n\t"\
+        "pand "#pb_3f", %%mm0  \n\t"\
+        "movq    %%mm2, %%mm6  \n\t"\
+        "pxor    %%mm0, %%mm6  \n\t"\
+        /* d = (c^b) & ~(b^a) & 1 */\
+        "pxor    %%mm5, %%mm6  \n\t"\
+        "pxor    %%mm4, %%mm5  \n\t"\
+        "pandn   %%mm6, %%mm5  \n\t"\
+        "pand "#pb_01", %%mm5  \n\t"\
+        /* delta = (avg(q0, p1>>2) + (d&a))
+         *       - (avg(p0, q1>>2) + (d&~a)) */\
+        "pavgb   %%mm2, %%mm0  \n\t"\
+        "movq    %%mm5, %%mm6  \n\t"\
+        "pand    %%mm4, %%mm6  \n\t"\
+        "paddusb %%mm6, %%mm0  \n\t"\
+        "pavgb   %%mm1, %%mm3  \n\t"\
+        "pandn   %%mm5, %%mm4  \n\t"\
+        "paddusb %%mm4, %%mm3  \n\t"\
+        /* p0 += clip(delta, -tc0, tc0)
+         * q0 -= clip(delta, -tc0, tc0) */\
+        "movq    %%mm0, %%mm4  \n\t"\
+        "psubusb %%mm3, %%mm0  \n\t"\
+        "psubusb %%mm4, %%mm3  \n\t"\
+        "pminub  %%mm7, %%mm0  \n\t"\
+        "pminub  %%mm7, %%mm3  \n\t"\
+        "paddusb %%mm0, %%mm1  \n\t"\
+        "paddusb %%mm3, %%mm2  \n\t"\
+        "psubusb %%mm3, %%mm1  \n\t"\
+        "psubusb %%mm0, %%mm2  \n\t"
+
+// in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=mm_bone
+// out: (q1addr) = clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )
+// clobbers: q2, tmp, tc0
+#define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\
+        "movq     %%mm1,  "#tmp"   \n\t"\
+        "pavgb    %%mm2,  "#tmp"   \n\t"\
+        "pavgb    "#tmp", "#q2"    \n\t" /* avg(p2,avg(p0,q0)) */\
+        "pxor   "q2addr", "#tmp"   \n\t"\
+        "pand     %8,     "#tmp"   \n\t" /* (p2^avg(p0,q0))&1 */\
+        "psubusb  "#tmp", "#q2"    \n\t" /* (p2+((p0+q0+1)>>1))>>1 */\
+        "movq     "#p1",  "#tmp"   \n\t"\
+        "psubusb  "#tc0", "#tmp"   \n\t"\
+        "paddusb  "#p1",  "#tc0"   \n\t"\
+        "pmaxub   "#tmp", "#q2"    \n\t"\
+        "pminub   "#tc0", "#q2"    \n\t"\
+        "movq     "#q2",  "q1addr" \n\t"
+
+static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
 {
-    uint64_t tmp0, tmp1;
+    uint64_t tmp0;
+    uint64_t tc = (uint8_t)tc0[1]*0x01010000 | (uint8_t)tc0[0]*0x0101;
+    // with luma, tc0=0 doesn't mean no filtering, so we need a separate input mask
+    uint32_t mask[2] = { (tc0[0]>=0)*0xffffffff, (tc0[1]>=0)*0xffffffff };
+
     asm volatile(
-        "movd  (%2,%4),   %%mm0        \n\t" //p1
-        "movd  (%2,%4,2), %%mm1        \n\t" //p0
-        "movd  (%3),      %%mm2        \n\t" //q0
-        "movd  (%3,%4),   %%mm3        \n\t" //q1
-        H264_DEBLOCK_THRESH(%6,%7)
-        "movq      %%mm7, %0           \n\t"
-
-// filter p1 if ABS(p2-p0) < beta
-        "movd      (%2),  %%mm3        \n\t"
-        "pxor      %%mm6, %%mm6        \n\t"
-        "punpcklbw %%mm6, %%mm3        \n\t" //p2
-        MMABS_DIFF_MMX2(%%mm1, %%mm3, %%mm5, %%mm6)
-        "pcmpgtw   %%mm5, %%mm4        \n\t"
-        "pand      %%mm7, %%mm4        \n\t" // mm4 = ( ABS( p2 - p0 ) < beta && filterp )
-        "movd      %5,    %%mm6        \n\t"
-        "pshufw    $0, %%mm6, %%mm6    \n\t" //tc
-
-        H264_DEBLOCK_DELTA_PQ1(%%mm0, %%mm3, %%mm7) // delta = clip( ( p2 + ((p0+q0+1)>>1) ) >> 1 ) - p1 )
-        "pand      %%mm4, %%mm5        \n\t"
-        "paddw     %%mm0, %%mm5        \n\t"
-        "packuswb  %%mm5, %%mm5        \n\t"
-        "movd      %%mm5, (%2,%4)      \n\t" // *p1 += delta
-        "psrlw     $15, %%mm4          \n\t"
-        "paddw     %%mm6, %%mm4        \n\t" // tc++
-        "movq      %%mm4, %1           \n\t"
-
-// filter q1 if ABS(q2-q0) < beta
-        "pxor      %%mm7, %%mm7        \n\t"
-        "movd  (%3,%4),   %%mm3        \n\t" //q1
-        "movd  (%3,%4,2), %%mm4        \n\t" //q2
-        "punpcklbw %%mm7, %%mm3        \n\t"
-        "punpcklbw %%mm7, %%mm4        \n\t"
-        MMABS_DIFF_MMX2(%%mm2, %%mm4, %%mm5, %%mm7)
-        "movd      %7,    %%mm7        \n\t"
-        "pshufw    $0, %%mm7, %%mm7    \n\t"
-        "pcmpgtw   %%mm5, %%mm7        \n\t"
-
-        H264_DEBLOCK_DELTA_PQ1(%%mm3, %%mm4, %%mm4) // delta = clip( ( q2 + ((p0+q0+1)>>1) ) >> 1 ) - q1 )
-        "movq      %0,    %%mm4        \n\t"
-        "pand      %%mm4, %%mm7        \n\t" // mm7 = ( ABS( q2 - q0 ) < beta && filterp )
-        "pand      %%mm7, %%mm5        \n\t"
-        "paddw     %%mm3, %%mm5        \n\t"
-        "packuswb  %%mm5, %%mm5        \n\t"
-        "movd      %%mm5, (%3,%4)      \n\t" // *q1 += delta
-        "movq      %1, %%mm6           \n\t"
-        "psrlw     $15, %%mm7          \n\t"
-        "paddw     %%mm7, %%mm6        \n\t" // tc++
-        "movq      %0,    %%mm4        \n\t"
-        "pand      %%mm4, %%mm6        \n\t"
-
-        H264_DEBLOCK_P0_Q0(%8)
-        "packuswb  %%mm1, %%mm1        \n\t"
-        "packuswb  %%mm2, %%mm2        \n\t"
-        "movd      %%mm1, (%2,%4,2)    \n\t"
-        "movd      %%mm2, (%3)         \n\t"
-
-        : "=m"(tmp0), "=m"(tmp1)
+        "movq    (%1,%3), %%mm0    \n\t" //p1
+        "movq    (%1,%3,2), %%mm1  \n\t" //p0
+        "movq    (%2),    %%mm2    \n\t" //q0
+        "movq    (%2,%3), %%mm3    \n\t" //q1
+        H264_DEBLOCK_MASK(%6, %7)
+        "pand     %5,     %%mm7    \n\t"
+        "movq     %%mm7,  %0       \n\t"
+
+        /* filter p1 */
+        "movq     (%1),   %%mm3    \n\t" //p2
+        DIFF_GT_MMX(%%mm1, %%mm3, %%mm5, %%mm6, %%mm4) // |p2-p0|>beta-1
+        "pandn    %%mm7,  %%mm6    \n\t"
+        "pcmpeqb  %%mm7,  %%mm6    \n\t"
+        "pand     %%mm7,  %%mm6    \n\t" // mask & |p2-p0|<beta
+        "pshufw  $80, %4, %%mm4    \n\t"
+        "pand     %%mm7,  %%mm4    \n\t" // mask & tc0
+        "movq     %8,     %%mm7    \n\t"
+        "pand     %%mm6,  %%mm7    \n\t" // mask & |p2-p0|<beta & 1
+        "pand     %%mm4,  %%mm6    \n\t" // mask & |p2-p0|<beta & tc0
+        "paddb    %%mm4,  %%mm7    \n\t" // tc++
+        H264_DEBLOCK_Q1(%%mm0, %%mm3, "(%1)", "(%1,%3)", %%mm6, %%mm4)
+
+        /* filter q1 */
+        "movq    (%2,%3,2), %%mm4  \n\t" //q2
+        DIFF_GT_MMX(%%mm2, %%mm4, %%mm5, %%mm6, %%mm3) // |q2-q0|>beta-1
+        "pandn    %0,     %%mm6    \n\t"
+        "pcmpeqb  %0,     %%mm6    \n\t"
+        "pand     %0,     %%mm6    \n\t"
+        "pshufw  $80, %4, %%mm5    \n\t"
+        "pand     %%mm6,  %%mm5    \n\t"
+        "pand     %8,     %%mm6    \n\t"
+        "paddb    %%mm6,  %%mm7    \n\t"
+        "movq    (%2,%3), %%mm3    \n\t"
+        H264_DEBLOCK_Q1(%%mm3, %%mm4, "(%2,%3,2)", "(%2,%3)", %%mm5, %%mm6)
+
+        /* filter p0, q0 */
+        H264_DEBLOCK_P0_Q0(%8, %9)
+        "movq      %%mm1, (%1,%3,2) \n\t"
+        "movq      %%mm2, (%2)      \n\t"
+
+        : "=m"(tmp0)
         : "r"(pix-3*stride), "r"(pix), "r"((long)stride),
-          "r"(tc0), "r"(alpha), "r"(beta), "m"(ff_pw_4)
+          "m"(tc), "m"(*(uint64_t*)mask), "m"(alpha1), "m"(beta1),
+          "m"(mm_bone), "m"(ff_pb_3F)
     );
 }
 
-static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int *tc0)
+static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
 {
-    int i;
-    for(i=0; i<4; i++, pix+=4) {
-        if(tc0[i] < 0)
-            continue;
-        h264_loop_filter_luma_mmx2(pix, stride, alpha, beta, tc0[i]);
-    }
+    if((tc0[0] & tc0[1]) >= 0)
+        h264_loop_filter_luma_mmx2(pix, stride, alpha-1, beta-1, tc0);
+    if((tc0[2] & tc0[3]) >= 0)
+        h264_loop_filter_luma_mmx2(pix+8, stride, alpha-1, beta-1, tc0+2);
 }
-
-static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int *tc0)
+static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
 {
-    uint8_t trans[4*8];
+    //FIXME: could cut some load/stores by merging transpose with filter
+    // also, it only needs to transpose 6x8
+    uint8_t trans[8*8];
     int i;
-    for(i=0; i<4; i++, pix+=4*stride) {
-        if(tc0[i] < 0)
+    for(i=0; i<2; i++, pix+=8*stride, tc0+=2) {
+        if((tc0[0] & tc0[1]) < 0)
             continue;
-        //FIXME: could cut some load/stores by merging transpose with filter
-        transpose4x4(trans, pix-4, 4, stride);
-        transpose4x4(trans+4*4, pix, 4, stride);
-        h264_loop_filter_luma_mmx2(trans+4*4, 4, alpha, beta, tc0[i]);
-        transpose4x4(pix-2, trans+2*4, stride, 4);
+        transpose4x4(trans,       pix-4,          8, stride);
+        transpose4x4(trans  +4*8, pix,            8, stride);
+        transpose4x4(trans+4,     pix-4+4*stride, 8, stride);
+        transpose4x4(trans+4+4*8, pix  +4*stride, 8, stride);
+        h264_loop_filter_luma_mmx2(trans+4*8, 8, alpha-1, beta-1, tc0);
+        transpose4x4(pix-2,          trans  +2*8, stride, 8);
+        transpose4x4(pix-2+4*stride, trans+4+2*8, stride, 8);
     }
 }
 
-static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int *tc0)
+static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)
 {
     asm volatile(
-        "movd    (%0),    %%mm0     \n\t"
-        "movd    (%0,%2), %%mm1     \n\t"
-        "movd    (%1),    %%mm2     \n\t"
-        "movd    (%1,%2), %%mm3     \n\t"
-        H264_DEBLOCK_THRESH(%4,%5)
+        "movq    (%0),    %%mm0     \n\t" //p1
+        "movq    (%0,%2), %%mm1     \n\t" //p0
+        "movq    (%1),    %%mm2     \n\t" //q0
+        "movq    (%1,%2), %%mm3     \n\t" //q1
+        H264_DEBLOCK_MASK(%4, %5)
         "movd      %3,    %%mm6     \n\t"
-        "pshufw $0x50, %%mm6, %%mm6 \n\t" // mm6 = tc[1], tc[1], tc[0], tc[0]
-        "pand      %%mm7, %%mm6     \n\t"
-        H264_DEBLOCK_P0_Q0(%6)
-        "packuswb  %%mm1, %%mm1     \n\t"
-        "packuswb  %%mm2, %%mm2     \n\t"
-        "movd      %%mm1, (%0,%2)   \n\t"
-        "movd      %%mm2, (%1)      \n\t"
+        "punpcklbw %%mm6, %%mm6     \n\t"
+        "pand      %%mm6, %%mm7     \n\t" // mm7 = tc&mask
+        H264_DEBLOCK_P0_Q0(%6, %7)
+        "movq      %%mm1, (%0,%2)   \n\t"
+        "movq      %%mm2, (%1)      \n\t"
+
         :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
-           "r"(tc0[1]<<16 | tc0[0]),
-           "r"(alpha), "r"(beta), "m"(ff_pw_4)
+           "r"(*(uint32_t*)tc0),
+           "m"(alpha1), "m"(beta1), "m"(mm_bone), "m"(ff_pb_3F)
     );
 }
 
-static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int *tc0)
+static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
 {
-    int i;
-    for(i=0; i<2; i++) {
-        h264_loop_filter_chroma_mmx2(pix, stride, alpha, beta, tc0);
-        pix += 4;
-        tc0 += 2;
-    }
+    h264_loop_filter_chroma_mmx2(pix, stride, alpha-1, beta-1, tc0);
 }
 
-static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int *tc0)
+static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)
 {
-    uint8_t trans[4*4];
-    int i;
-    for(i=0; i<2; i++) {
-        //FIXME: could cut some load/stores by merging transpose with filter
-        transpose4x4(trans, pix-2, 4, stride);
-        h264_loop_filter_chroma_mmx2(trans+2*4, 4, alpha, beta, tc0);
-        transpose4x4(pix-2, trans, stride, 4);
-        pix += 4*stride;
-        tc0 += 2;
-    }
+    //FIXME: could cut some load/stores by merging transpose with filter
+    uint8_t trans[8*4];
+    transpose4x4(trans, pix-2, 8, stride);
+    transpose4x4(trans+4, pix-2+4*stride, 8, stride);
+    h264_loop_filter_chroma_mmx2(trans+2*8, 8, alpha-1, beta-1, tc0);
+    transpose4x4(pix-2, trans, stride, 8);
+    transpose4x4(pix-2+4*stride, trans+4, stride, 8);
+}
+
+// p0 = (p0 + q1 + 2*p1 + 2) >> 2
+#define H264_FILTER_CHROMA4(p0, p1, q1, one) \
+    "movq    "#p0", %%mm4  \n\t"\
+    "pxor    "#q1", %%mm4  \n\t"\
+    "pand   "#one", %%mm4  \n\t" /* mm4 = (p0^q1)&1 */\
+    "pavgb   "#q1", "#p0"  \n\t"\
+    "psubusb %%mm4, "#p0"  \n\t"\
+    "pavgb   "#p1", "#p0"  \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\
+
+static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)
+{
+    asm volatile(
+        "movq    (%0),    %%mm0     \n\t"
+        "movq    (%0,%2), %%mm1     \n\t"
+        "movq    (%1),    %%mm2     \n\t"
+        "movq    (%1,%2), %%mm3     \n\t"
+        H264_DEBLOCK_MASK(%3, %4)
+        "movq    %%mm1,   %%mm5     \n\t"
+        "movq    %%mm2,   %%mm6     \n\t"
+        H264_FILTER_CHROMA4(%%mm1, %%mm0, %%mm3, %5) //p0'
+        H264_FILTER_CHROMA4(%%mm2, %%mm3, %%mm0, %5) //q0'
+        "psubb   %%mm5,   %%mm1     \n\t"
+        "psubb   %%mm6,   %%mm2     \n\t"
+        "pand    %%mm7,   %%mm1     \n\t"
+        "pand    %%mm7,   %%mm2     \n\t"
+        "paddb   %%mm5,   %%mm1     \n\t"
+        "paddb   %%mm6,   %%mm2     \n\t"
+        "movq    %%mm1,   (%0,%2)   \n\t"
+        "movq    %%mm2,   (%1)      \n\t"
+        :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),
+           "m"(alpha1), "m"(beta1), "m"(mm_bone)
+    );
+}
+
+static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
+{
+    h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);
+}
+
+static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)
+{
+    //FIXME: could cut some load/stores by merging transpose with filter
+    uint8_t trans[8*4];
+    transpose4x4(trans, pix-2, 8, stride);
+    transpose4x4(trans+4, pix-2+4*stride, 8, stride);
+    h264_loop_filter_chroma_intra_mmx2(trans+2*8, 8, alpha-1, beta-1);
+    transpose4x4(pix-2, trans, stride, 8);
+    transpose4x4(pix-2+4*stride, trans+4, stride, 8);
 }
 
+
 #ifdef CONFIG_ENCODERS
 static int pix_norm1_mmx(uint8_t *pix, int line_size) {
     int tmp;
@@ -3016,6 +3078,30 @@ H264_MC(avg_, 4, mmx2)
 H264_MC(avg_, 8, mmx2)
 H264_MC(avg_, 16,mmx2)
 
+
+/** These are used by *_h264_chroma_mc8_* */
+static const uint64_t thirtytwo __align8 = 0x0020002000200020ULL;
+static const uint64_t sixtyfour __align8 = 0x0040004000400040ULL;
+
+#define H264_CHROMA_OP(S,D)
+#define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx
+#include "dsputil_h264_template_mmx.c"
+#undef H264_CHROMA_OP
+#undef H264_CHROMA_MC8_TMPL
+
+#define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"
+#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2
+#include "dsputil_h264_template_mmx.c"
+#undef H264_CHROMA_OP
+#undef H264_CHROMA_MC8_TMPL
+
+#define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"
+#define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow
+#include "dsputil_h264_template_mmx.c"
+#undef H264_CHROMA_OP
+#undef H264_CHROMA_MC8_TMPL
+
+
 #if 0
 static void just_return() { return; }
 #endif
@@ -3115,6 +3201,10 @@ static void add_8x8basis_mmx(int16_t rem[64], int16_t basis[64], int scale){
 void ff_mmx_idct(DCTELEM *block);
 void ff_mmxext_idct(DCTELEM *block);
 
+void ff_vp3_idct_sse2(int16_t *input_data);
+void ff_vp3_idct_mmx(int16_t *data);
+void ff_vp3_dsp_init_mmx(void);
+
 /* XXX: those functions should be suppressed ASAP when all IDCTs are
    converted */
 static void ff_libmpeg2mmx_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
@@ -3137,6 +3227,26 @@ static void ff_libmpeg2mmx2_idct_add(uint8_t *dest, int line_size, DCTELEM *bloc
     ff_mmxext_idct (block);
     add_pixels_clamped_mmx(block, dest, line_size);
 }
+static void ff_vp3_idct_put_sse2(uint8_t *dest, int line_size, DCTELEM *block)
+{
+    ff_vp3_idct_sse2(block);
+    put_signed_pixels_clamped_mmx(block, dest, line_size);
+}
+static void ff_vp3_idct_add_sse2(uint8_t *dest, int line_size, DCTELEM *block)
+{
+    ff_vp3_idct_sse2(block);
+    add_pixels_clamped_mmx(block, dest, line_size);
+}
+static void ff_vp3_idct_put_mmx(uint8_t *dest, int line_size, DCTELEM *block)
+{
+    ff_vp3_idct_mmx(block);
+    put_signed_pixels_clamped_mmx(block, dest, line_size);
+}
+static void ff_vp3_idct_add_mmx(uint8_t *dest, int line_size, DCTELEM *block)
+{
+    ff_vp3_idct_mmx(block);
+    add_pixels_clamped_mmx(block, dest, line_size);
+}
     
 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
 {
@@ -3196,18 +3306,22 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
                     c->idct    = ff_mmx_idct;
                 }
                 c->idct_permutation_type= FF_LIBMPEG2_IDCT_PERM;
+            }else if(idct_algo==FF_IDCT_VP3){
+                if(mm_flags & MM_SSE2){
+                    c->idct_put= ff_vp3_idct_put_sse2;
+                    c->idct_add= ff_vp3_idct_add_sse2;
+                    c->idct    = ff_vp3_idct_sse2;
+                    c->idct_permutation_type= FF_TRANSPOSE_IDCT_PERM;
+                }else{
+                    ff_vp3_dsp_init_mmx();
+                    c->idct_put= ff_vp3_idct_put_mmx;
+                    c->idct_add= ff_vp3_idct_add_mmx;
+                    c->idct    = ff_vp3_idct_mmx;
+                    c->idct_permutation_type= FF_PARTTRANS_IDCT_PERM;
+                }
             }
         }
 
-        /* VP3 optimized DSP functions */
-        if (mm_flags & MM_SSE2) {
-            c->vp3_dsp_init = vp3_dsp_init_sse2;
-            c->vp3_idct = vp3_idct_sse2;
-        } else {
-            c->vp3_dsp_init = vp3_dsp_init_mmx;
-            c->vp3_idct = vp3_idct_mmx;
-        }
-
 #ifdef CONFIG_ENCODERS
         c->get_pixels = get_pixels_mmx;
         c->diff_pixels = diff_pixels_mmx;
@@ -3287,6 +3401,7 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
 
         c->h263_v_loop_filter= h263_v_loop_filter_mmx;
         c->h263_h_loop_filter= h263_h_loop_filter_mmx;        
+	c->put_h264_chroma_pixels_tab[0]= put_h264_chroma_mc8_mmx;
         
         if (mm_flags & MM_MMXEXT) {
             c->put_pixels_tab[0][1] = put_pixels16_x2_mmx2;
@@ -3383,10 +3498,13 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
             dspfunc(avg_h264_qpel, 2, 4);
 #undef dspfunc
 
+	    c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_mmx2;
             c->h264_v_loop_filter_luma= h264_v_loop_filter_luma_mmx2;
             c->h264_h_loop_filter_luma= h264_h_loop_filter_luma_mmx2;
             c->h264_v_loop_filter_chroma= h264_v_loop_filter_chroma_mmx2;
             c->h264_h_loop_filter_chroma= h264_h_loop_filter_chroma_mmx2;
+            c->h264_v_loop_filter_chroma_intra= h264_v_loop_filter_chroma_intra_mmx2;
+            c->h264_h_loop_filter_chroma_intra= h264_h_loop_filter_chroma_intra_mmx2;
 
 #ifdef CONFIG_ENCODERS
             c->sub_hfyu_median_prediction= sub_hfyu_median_prediction_mmx2;
@@ -3472,6 +3590,8 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
             dspfunc(avg_h264_qpel, 0, 16);
             dspfunc(avg_h264_qpel, 1, 8);
             dspfunc(avg_h264_qpel, 2, 4);
+
+	    c->avg_h264_chroma_pixels_tab[0]= avg_h264_chroma_mc8_3dnow;
         }
     }
         
diff --git a/src/libffmpeg/libavcodec/i386/vp3dsp_mmx.c b/src/libffmpeg/libavcodec/i386/vp3dsp_mmx.c
index 319e57f1b..ea0405717 100644
--- a/src/libffmpeg/libavcodec/i386/vp3dsp_mmx.c
+++ b/src/libffmpeg/libavcodec/i386/vp3dsp_mmx.c
@@ -257,18 +257,11 @@ static uint16_t idct_cosine_table[7] = {
     movq_r2m(r2, *I(2)); \
 }
 
-void vp3_dsp_init_mmx(void)
+void ff_vp3_dsp_init_mmx(void)
 {
     int j = 16;
     uint16_t *p;
 
-    do {
-        idct_constants[--j] = 0;
-    } while (j);
-
-    idct_constants[0]  = idct_constants[5] =
-    idct_constants[10] = idct_constants[15] = 65535;
-
     j = 1;
     do {
         p = idct_constants + ((j + 3) << 2);
@@ -279,8 +272,7 @@ void vp3_dsp_init_mmx(void)
     idct_constants[46] = idct_constants[47] = IdctAdjustBeforeShift;
 }
 
-void vp3_idct_mmx(int16_t *input_data, int16_t *dequant_matrix,
-    int coeff_count, int16_t *output_data)
+void ff_vp3_idct_mmx(int16_t *output_data)
 {
     /* eax = quantized input
      * ebx = dequantizer matrix
@@ -291,246 +283,11 @@ void vp3_idct_mmx(int16_t *input_data, int16_t *dequant_matrix,
      * r0..r7 = mm0..mm7
      */
 
-#define M(x) (idct_constants + x * 4)
 #define C(x) (idct_constants + 16 + (x - 1) * 4)
 #define Eight (idct_constants + 44)
 
-    unsigned char *input_bytes = (unsigned char *)input_data;
-    unsigned char *dequant_matrix_bytes = (unsigned char *)dequant_matrix;
-    unsigned char *output_data_bytes = (unsigned char *)output_data;
-
-    movq_m2r(*(input_bytes), r0);
-    pmullw_m2r(*(dequant_matrix_bytes), r0);       /* r0 = 03 02 01 00 */
-    movq_m2r(*(input_bytes+16), r1);
-    pmullw_m2r(*(dequant_matrix_bytes+16), r1);    /* r1 = 13 12 11 10 */
-    movq_m2r(*M(0), r2);                           /* r2 = __ __ __ FF */
-    movq_r2r(r0, r3);                              /* r3 = 03 02 01 00 */
-    movq_m2r(*(input_bytes+8), r4);
-    psrlq_i2r(16, r0);                             /* r0 = __ 03 02 01 */
-    pmullw_m2r(*(dequant_matrix_bytes+8), r4);     /* r4 = 07 06 05 04 */
-    pand_r2r(r2, r3);                              /* r3 = __ __ __ 00 */
-    movq_r2r(r0, r5);                              /* r5 = __ 03 02 01 */
-    movq_r2r(r1, r6);                              /* r6 = 13 12 11 10 */
-    pand_r2r(r2, r5);                              /* r5 = __ __ __ 01 */
-    psllq_i2r(32, r6);                             /* r6 = 11 10 __ __ */
-    movq_m2r(*M(3), r7);                           /* r7 = FF __ __ __ */
-    pxor_r2r(r5, r0);                              /* r0 = __ 03 02 __ */
-    pand_r2r(r6, r7);                              /* r7 = 11 __ __ __ */
-    por_r2r(r3, r0);                               /* r0 = __ 03 02 00 */
-    pxor_r2r(r7, r6);                              /* r6 = __ 10 __ __ */
-    por_r2r(r7, r0);                               /* r0 = 11 03 02 00 = R0 */
-    movq_m2r(*M(3), r7);                           /* r7 = FF __ __ __ */
-    movq_r2r(r4, r3);                              /* r3 = 07 06 05 04 */
-    movq_r2m(r0, *(output_data_bytes));            /* write R0 = r0 */
-    pand_r2r(r2, r3);                              /* r3 = __ __ __ 04 */
-    movq_m2r(*(input_bytes+32), r0);
-    psllq_i2r(16, r3);                             /* r3 = __ __ 04 __ */
-    pmullw_m2r(*(dequant_matrix_bytes+32), r0);    /* r0 = 23 22 21 20 */
-    pand_r2r(r1, r7);                              /* r7 = 13 __ __ __ */
-    por_r2r(r3, r5);                               /* r5 = __ __ 04 01 */
-    por_r2r(r6, r7);                               /* r7 = 13 10 __ __ */
-    movq_m2r(*(input_bytes+24), r3);
-    por_r2r(r5, r7);                               /* r7 = 13 10 04 01 = R1 */
-    pmullw_m2r(*(dequant_matrix_bytes+24), r3);    /* r3 = 17 16 15 14 */
-    psrlq_i2r(16, r4);                             /* r4 = __ 07 06 05 */
-    movq_r2m(r7, *(output_data_bytes+16));         /* write R1 = r7 */
-    movq_r2r(r4, r5);                              /* r5 = __ 07 06 05 */
-    movq_r2r(r0, r7);                              /* r7 = 23 22 21 20 */
-    psrlq_i2r(16, r4);                             /* r4 = __ __ 07 06 */
-    psrlq_i2r(48, r7);                             /* r7 = __ __ __ 23 */
-    movq_r2r(r2, r6);                              /* r6 = __ __ __ FF */
-    pand_r2r(r2, r5);                              /* r5 = __ __ __ 05 */
-    pand_r2r(r4, r6);                              /* r6 = __ __ __ 06 */
-    movq_r2m(r7, *(output_data_bytes+80));      /* partial R9 = __ __ __ 23 */
-    pxor_r2r(r6, r4);                              /* r4 = __ __ 07 __ */
-    psrlq_i2r(32, r1);                             /* r1 = __ __ 13 12 */
-    por_r2r(r5, r4);                               /* r4 = __ __ 07 05 */
-    movq_m2r(*M(3), r7);                           /* r7 = FF __ __ __ */
-    pand_r2r(r2, r1);                              /* r1 = __ __ __ 12 */
-    movq_m2r(*(input_bytes+48), r5);
-    psllq_i2r(16, r0);                             /* r0 = 22 21 20 __ */
-    pmullw_m2r(*(dequant_matrix_bytes+48), r5);    /* r5 = 33 32 31 30 */
-    pand_r2r(r0, r7);                              /* r7 = 22 __ __ __ */
-    movq_r2m(r1, *(output_data_bytes+64));      /* partial R8 = __ __ __ 12 */
-    por_r2r(r4, r7);                               /* r7 = 22 __ 07 05 */
-    movq_r2r(r3, r4);                              /* r4 = 17 16 15 14 */
-    pand_r2r(r2, r3);                              /* r3 = __ __ __ 14 */
-    movq_m2r(*M(2), r1);                           /* r1 = __ FF __ __ */
-    psllq_i2r(32, r3);                             /* r3 = __ 14 __ __ */
-    por_r2r(r3, r7);                               /* r7 = 22 14 07 05 = R2 */
-    movq_r2r(r5, r3);                              /* r3 = 33 32 31 30 */
-    psllq_i2r(48, r3);                             /* r3 = 30 __ __ __ */
-    pand_r2r(r0, r1);                              /* r1 = __ 21 __ __ */
-    movq_r2m(r7, *(output_data_bytes+32));         /* write R2 = r7 */
-    por_r2r(r3, r6);                               /* r6 = 30 __ __ 06 */
-    movq_m2r(*M(1), r7);                           /* r7 = __ __ FF __ */
-    por_r2r(r1, r6);                               /* r6 = 30 21 __ 06 */
-    movq_m2r(*(input_bytes+56), r1);
-    pand_r2r(r4, r7);                              /* r7 = __ __ 15 __ */
-    pmullw_m2r(*(dequant_matrix_bytes+56), r1);    /* r1 = 37 36 35 34 */
-    por_r2r(r6, r7);                               /* r7 = 30 21 15 06 = R3 */
-    pand_m2r(*M(1), r0);                           /* r0 = __ __ 20 __ */
-    psrlq_i2r(32, r4);                             /* r4 = __ __ 17 16 */
-    movq_r2m(r7, *(output_data_bytes+48));         /* write R3 = r7 */
-    movq_r2r(r4, r6);                              /* r6 = __ __ 17 16 */
-    movq_m2r(*M(3), r7);                           /* r7 = FF __ __ __ */
-    pand_r2r(r2, r4);                              /* r4 = __ __ __ 16 */
-    movq_m2r(*M(1), r3);                           /* r3 = __ __ FF __ */
-    pand_r2r(r1, r7);                              /* r7 = 37 __ __ __ */
-    pand_r2r(r5, r3);                              /* r3 = __ __ 31 __ */
-    por_r2r(r4, r0);                               /* r0 = __ __ 20 16 */
-    psllq_i2r(16, r3);                             /* r3 = __ 31 __ __ */
-    por_r2r(r0, r7);                               /* r7 = 37 __ 20 16 */
-    movq_m2r(*M(2), r4);                           /* r4 = __ FF __ __ */
-    por_r2r(r3, r7);                               /* r7 = 37 31 20 16 = R4 */
-    movq_m2r(*(input_bytes+80), r0);
-    movq_r2r(r4, r3);                              /* r3 = __ __ FF __ */
-    pmullw_m2r(*(dequant_matrix_bytes+80), r0);    /* r0 = 53 52 51 50 */
-    pand_r2r(r5, r4);                              /* r4 = __ 32 __ __ */
-    movq_r2m(r7, *(output_data_bytes+8));          /* write R4 = r7 */
-    por_r2r(r4, r6);                               /* r6 = __ 32 17 16 */
-    movq_r2r(r3, r4);                              /* r4 = __ FF __ __ */
-    psrlq_i2r(16, r6);                             /* r6 = __ __ 32 17 */
-    movq_r2r(r0, r7);                              /* r7 = 53 52 51 50 */
-    pand_r2r(r1, r4);                              /* r4 = __ 36 __ __ */
-    psllq_i2r(48, r7);                             /* r7 = 50 __ __ __ */
-    por_r2r(r4, r6);                               /* r6 = __ 36 32 17 */
-    movq_m2r(*(input_bytes+88), r4);
-    por_r2r(r6, r7);                               /* r7 = 50 36 32 17 = R5 */
-    pmullw_m2r(*(dequant_matrix_bytes+88), r4);    /* r4 = 57 56 55 54 */
-    psrlq_i2r(16, r3);                             /* r3 = __ __ FF __ */
-    movq_r2m(r7, *(output_data_bytes+24));         /* write R5 = r7 */
-    pand_r2r(r1, r3);                              /* r3 = __ __ 35 __ */
-    psrlq_i2r(48, r5);                             /* r5 = __ __ __ 33 */
-    pand_r2r(r2, r1);                              /* r1 = __ __ __ 34 */
-    movq_m2r(*(input_bytes+104), r6);
-    por_r2r(r3, r5);                               /* r5 = __ __ 35 33 */
-    pmullw_m2r(*(dequant_matrix_bytes+104), r6);   /* r6 = 67 66 65 64 */
-    psrlq_i2r(16, r0);                             /* r0 = __ 53 52 51 */
-    movq_r2r(r4, r7);                              /* r7 = 57 56 55 54 */
-    movq_r2r(r2, r3);                              /* r3 = __ __ __ FF */
-    psllq_i2r(48, r7);                             /* r7 = 54 __ __ __ */
-    pand_r2r(r0, r3);                              /* r3 = __ __ __ 51 */
-    pxor_r2r(r3, r0);                              /* r0 = __ 53 52 __ */
-    psllq_i2r(32, r3);                             /* r3 = __ 51 __ __ */
-    por_r2r(r5, r7);                               /* r7 = 54 __ 35 33 */
-    movq_r2r(r6, r5);                              /* r5 = 67 66 65 64 */
-    pand_m2r(*M(1), r6);                           /* r6 = __ __ 65 __ */
-    por_r2r(r3, r7);                               /* r7 = 54 51 35 33 = R6 */
-    psllq_i2r(32, r6);                             /* r6 = 65 __ __ __ */
-    por_r2r(r1, r0);                               /* r0 = __ 53 52 34 */
-    movq_r2m(r7, *(output_data_bytes+40));         /* write R6 = r7 */
-    por_r2r(r6, r0);                               /* r0 = 65 53 52 34 = R7 */
-    movq_m2r(*(input_bytes+120), r7);
-    movq_r2r(r5, r6);                              /* r6 = 67 66 65 64 */
-    pmullw_m2r(*(dequant_matrix_bytes+120), r7);   /* r7 = 77 76 75 74 */
-    psrlq_i2r(32, r5);                             /* r5 = __ __ 67 66 */
-    pand_r2r(r2, r6);                              /* r6 = __ __ __ 64 */
-    movq_r2r(r5, r1);                              /* r1 = __ __ 67 66 */
-    movq_r2m(r0, *(output_data_bytes+56));         /* write R7 = r0 */
-    pand_r2r(r2, r1);                              /* r1 = __ __ __ 66 */
-    movq_m2r(*(input_bytes+112), r0);
-    movq_r2r(r7, r3);                              /* r3 = 77 76 75 74 */
-    pmullw_m2r(*(dequant_matrix_bytes+112), r0);   /* r0 = 73 72 71 70 */
-    psllq_i2r(16, r3);                             /* r3 = 76 75 74 __ */
-    pand_m2r(*M(3), r7);                           /* r7 = 77 __ __ __ */
-    pxor_r2r(r1, r5);                              /* r5 = __ __ 67 __ */
-    por_r2r(r5, r6);                               /* r6 = __ __ 67 64 */
-    movq_r2r(r3, r5);                              /* r5 = 76 75 74 __ */
-    pand_m2r(*M(3), r5);                           /* r5 = 76 __ __ __ */
-    por_r2r(r1, r7);                               /* r7 = 77 __ __ 66 */
-    movq_m2r(*(input_bytes+96), r1);
-    pxor_r2r(r5, r3);                              /* r3 = __ 75 74 __ */
-    pmullw_m2r(*(dequant_matrix_bytes+96), r1);    /* r1 = 63 62 61 60 */
-    por_r2r(r3, r7);                               /* r7 = 77 75 74 66 = R15 */
-    por_r2r(r5, r6);                               /* r6 = 76 __ 67 64 */
-    movq_r2r(r0, r5);                              /* r5 = 73 72 71 70 */
-    movq_r2m(r7, *(output_data_bytes+120));        /* store R15 = r7 */
-    psrlq_i2r(16, r5);                             /* r5 = __ 73 72 71 */
-    pand_m2r(*M(2), r5);                           /* r5 = __ 73 __ __ */
-    movq_r2r(r0, r7);                              /* r7 = 73 72 71 70 */
-    por_r2r(r5, r6);                               /* r6 = 76 73 67 64 = R14 */
-    pand_r2r(r2, r0);                              /* r0 = __ __ __ 70 */
-    pxor_r2r(r0, r7);                              /* r7 = 73 72 71 __ */
-    psllq_i2r(32, r0);                             /* r0 = __ 70 __ __ */
-    movq_r2m(r6, *(output_data_bytes+104));        /* write R14 = r6 */
-    psrlq_i2r(16, r4);                             /* r4 = __ 57 56 55 */
-    movq_m2r(*(input_bytes+72), r5);
-    psllq_i2r(16, r7);                             /* r7 = 72 71 __ __ */
-    pmullw_m2r(*(dequant_matrix_bytes+72), r5);    /* r5 = 47 46 45 44 */
-    movq_r2r(r7, r6);                              /* r6 = 72 71 __ __ */
-    movq_m2r(*M(2), r3);                           /* r3 = __ FF __ __ */
-    psllq_i2r(16, r6);                             /* r6 = 71 __ __ __ */
-    pand_m2r(*M(3), r7);                           /* r7 = 72 __ __ __ */
-    pand_r2r(r1, r3);                              /* r3 = __ 62 __ __ */
-    por_r2r(r0, r7);                               /* r7 = 72 70 __ __ */
-    movq_r2r(r1, r0);                              /* r0 = 63 62 61 60 */
-    pand_m2r(*M(3), r1);                           /* r1 = 63 __ __ __ */
-    por_r2r(r3, r6);                               /* r6 = 71 62 __ __ */
-    movq_r2r(r4, r3);                              /* r3 = __ 57 56 55 */
-    psrlq_i2r(32, r1);                             /* r1 = __ __ 63 __ */
-    pand_r2r(r2, r3);                              /* r3 = __ __ __ 55 */
-    por_r2r(r1, r7);                               /* r7 = 72 70 63 __ */
-    por_r2r(r3, r7);                               /* r7 = 72 70 63 55 = R13 */
-    movq_r2r(r4, r3);                              /* r3 = __ 57 56 55 */
-    pand_m2r(*M(1), r3);                           /* r3 = __ __ 56 __ */
-    movq_r2r(r5, r1);                              /* r1 = 47 46 45 44 */
-    movq_r2m(r7, *(output_data_bytes+88));         /* write R13 = r7 */
-    psrlq_i2r(48, r5);                             /* r5 = __ __ __ 47 */
-    movq_m2r(*(input_bytes+64), r7);
-    por_r2r(r3, r6);                               /* r6 = 71 62 56 __ */
-    pmullw_m2r(*(dequant_matrix_bytes+64), r7);    /* r7 = 43 42 41 40 */
-    por_r2r(r5, r6);                               /* r6 = 71 62 56 47 = R12 */
-    pand_m2r(*M(2), r4);                           /* r4 = __ 57 __ __ */
-    psllq_i2r(32, r0);                             /* r0 = 61 60 __ __ */
-    movq_r2m(r6, *(output_data_bytes+72));         /* write R12 = r6 */
-    movq_r2r(r0, r6);                              /* r6 = 61 60 __ __ */
-    pand_m2r(*M(3), r0);                           /* r0 = 61 __ __ __ */
-    psllq_i2r(16, r6);                             /* r6 = 60 __ __ __ */
-    movq_m2r(*(input_bytes+40), r5);
-    movq_r2r(r1, r3);                              /* r3 = 47 46 45 44 */
-    pmullw_m2r(*(dequant_matrix_bytes+40), r5);    /* r5 = 27 26 25 24 */
-    psrlq_i2r(16, r1);                             /* r1 = __ 47 46 45 */
-    pand_m2r(*M(1), r1);                           /* r1 = __ __ 46 __ */
-    por_r2r(r4, r0);                               /* r0 = 61 57 __ __ */
-    pand_r2r(r7, r2);                              /* r2 = __ __ __ 40 */
-    por_r2r(r1, r0);                               /* r0 = 61 57 46 __ */
-    por_r2r(r2, r0);                               /* r0 = 61 57 46 40 = R11 */
-    psllq_i2r(16, r3);                             /* r3 = 46 45 44 __ */
-    movq_r2r(r3, r4);                              /* r4 = 46 45 44 __ */
-    movq_r2r(r5, r2);                              /* r2 = 27 26 25 24 */
-    movq_r2m(r0, *(output_data_bytes+112));        /* write R11 = r0 */
-    psrlq_i2r(48, r2);                             /* r2 = __ __ __ 27 */
-    pand_m2r(*M(2), r4);                           /* r4 = __ 45 __ __ */
-    por_r2r(r2, r6);                               /* r6 = 60 __ __ 27 */
-    movq_m2r(*M(1), r2);                           /* r2 = __ __ FF __ */
-    por_r2r(r4, r6);                               /* r6 = 60 45 __ 27 */
-    pand_r2r(r7, r2);                              /* r2 = __ __ 41 __ */
-    psllq_i2r(32, r3);                             /* r3 = 44 __ __ __ */
-    por_m2r(*(output_data_bytes+80), r3);          /* r3 = 44 __ __ 23 */
-    por_r2r(r2, r6);                               /* r6 = 60 45 41 27 = R10 */
-    movq_m2r(*M(3), r2);                           /* r2 = FF __ __ __ */
-    psllq_i2r(16, r5);                             /* r5 = 26 25 24 __ */
-    movq_r2m(r6, *(output_data_bytes+96));         /* store R10 = r6 */
-    pand_r2r(r5, r2);                              /* r2 = 26 __ __ __ */
-    movq_m2r(*M(2), r6);                           /* r6 = __ FF __ __ */
-    pxor_r2r(r2, r5);                              /* r5 = __ 25 24 __ */
-    pand_r2r(r7, r6);                              /* r6 = __ 42 __ __ */
-    psrlq_i2r(32, r2);                             /* r2 = __ __ 26 __ */
-    pand_m2r(*M(3), r7);                           /* r7 = 43 __ __ __ */
-    por_r2r(r2, r3);                               /* r3 = 44 __ 26 23 */
-    por_m2r(*(output_data_bytes+64), r7);          /* r7 = 43 __ __ 12 */
-    por_r2r(r3, r6);                               /* r6 = 44 42 26 23 = R9 */
-    por_r2r(r5, r7);                               /* r7 = 43 25 24 12 = R8 */
-    movq_r2m(r6, *(output_data_bytes+80));         /* store R9 = r6 */
-    movq_r2m(r7, *(output_data_bytes+64));         /* store R8 = r7 */
-
-
-#undef M
-
     /* at this point, function has completed dequantization + dezigzag +
      * partial transposition; now do the idct itself */
-
 #define I(K) (output_data + K * 8)
 #define J(K) (output_data + ((K - 4) * 8) + 4)
 
diff --git a/src/libffmpeg/libavcodec/i386/vp3dsp_sse2.c b/src/libffmpeg/libavcodec/i386/vp3dsp_sse2.c
index 60c6bf80e..9c69ddb21 100644
--- a/src/libffmpeg/libavcodec/i386/vp3dsp_sse2.c
+++ b/src/libffmpeg/libavcodec/i386/vp3dsp_sse2.c
@@ -796,24 +796,16 @@ static unsigned short __align16 SSE2_idct_data[7 * 8] =
 } /* end of SSE2_Dequantize Macro */
 
 
-void vp3_dsp_init_sse2(void)
-{
-    /* nop */
-}
-
-
-void vp3_idct_sse2(int16_t *input_data, int16_t *dequant_matrix,
-    int coeff_count, int16_t *output_data)
+void ff_vp3_idct_sse2(int16_t *input_data)
 {
     unsigned char *input_bytes = (unsigned char *)input_data;
-    unsigned char *dequant_matrix_bytes = (unsigned char *)dequant_matrix;
     unsigned char *dequant_const_bytes = (unsigned char *)SSE2_dequant_const;
-    unsigned char *output_data_bytes = (unsigned char *)output_data;
+    unsigned char *output_data_bytes = (unsigned char *)input_data;
     unsigned char *idct_data_bytes = (unsigned char *)SSE2_idct_data;
     unsigned char *Eight = (unsigned char *)eight_data;
 
 #define eax input_bytes
-#define ebx dequant_matrix_bytes
+//#define ebx dequant_matrix_bytes
 #define ecx dequant_const_bytes
 #define edx idct_data_bytes
 
@@ -821,7 +813,7 @@ void vp3_idct_sse2(int16_t *input_data, int16_t *dequant_matrix,
 #define O(i) (ebx + 16 * i)
 #define C(i) (edx + 16 * (i-1))
 
-    SSE2_Dequantize();
+ //   SSE2_Dequantize();
 
 #undef ebx
 #define ebx output_data_bytes