VP8: unroll MB mode decoding tree
[ffmpeg.git] / libpostproc / postprocess_template.c
index ae16ab2..918d005 100644 (file)
  */
 
 /**
- * @file postprocess_template.c
+ * @file
  * mmx/mmx2/3dnow postprocess code.
  */
 
+#include "libavutil/x86_cpu.h"
 
-#ifdef ARCH_X86_64
-#  define REGa  rax
-#  define REGc  rcx
-#  define REGd  rdx
-#  define REG_a  "rax"
-#  define REG_c  "rcx"
-#  define REG_d  "rdx"
-#  define REG_SP "rsp"
-#  define ALIGN_MASK "$0xFFFFFFFFFFFFFFF8"
-#else
-#  define REGa  eax
-#  define REGc  ecx
-#  define REGd  edx
-#  define REG_a  "eax"
-#  define REG_c  "ecx"
-#  define REG_d  "edx"
-#  define REG_SP "esp"
-#  define ALIGN_MASK "$0xFFFFFFF8"
-#endif
-
+#define ALIGN_MASK "$-8"
 
+#undef REAL_PAVGB
 #undef PAVGB
 #undef PMINUB
 #undef PMAXUB
 
-#ifdef HAVE_MMX2
+#if   HAVE_MMX2
 #define REAL_PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
-#elif defined (HAVE_3DNOW)
+#elif HAVE_AMD3DNOW
 #define REAL_PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
 #endif
 #define PAVGB(a,b)  REAL_PAVGB(a,b)
 
-#ifdef HAVE_MMX2
+#if   HAVE_MMX2
 #define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
-#elif defined (HAVE_MMX)
+#elif HAVE_MMX
 #define PMINUB(b,a,t) \
-        "movq " #a ", " #t " \n\t"\
-        "psubusb " #b ", " #t " \n\t"\
-        "psubb " #t ", " #a " \n\t"
+    "movq " #a ", " #t " \n\t"\
+    "psubusb " #b ", " #t " \n\t"\
+    "psubb " #t ", " #a " \n\t"
 #endif
 
-#ifdef HAVE_MMX2
+#if   HAVE_MMX2
 #define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
-#elif defined (HAVE_MMX)
+#elif HAVE_MMX
 #define PMAXUB(a,b) \
-        "psubusb " #a ", " #b " \n\t"\
-        "paddb " #a ", " #b " \n\t"
+    "psubusb " #a ", " #b " \n\t"\
+    "paddb " #a ", " #b " \n\t"
 #endif
 
 //FIXME? |255-0| = 1 (should not be a problem ...)
-#ifdef HAVE_MMX
+#if HAVE_MMX
 /**
  * Check if the middle 8x8 Block in the given 8x16 block is flat
  */
 static inline int RENAME(vertClassify)(uint8_t src[], int stride, PPContext *c){
-        int numEq= 0, dcOk;
-        src+= stride*4; // src points to begin of the 8x8 Block
-asm volatile(
-                "movq %0, %%mm7                         \n\t"
-                "movq %1, %%mm6                         \n\t"
-                : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
-                );
-
-asm volatile(
-                "lea (%2, %3), %%"REG_a"                \n\t"
+    int numEq= 0, dcOk;
+    src+= stride*4; // src points to begin of the 8x8 Block
+    __asm__ volatile(
+        "movq %0, %%mm7                         \n\t"
+        "movq %1, %%mm6                         \n\t"
+        : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
+        );
+
+    __asm__ volatile(
+        "lea (%2, %3), %%"REG_a"                \n\t"
 //      0       1       2       3       4       5       6       7       8       9
 //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2
 
-                "movq (%2), %%mm0                       \n\t"
-                "movq (%%"REG_a"), %%mm1                \n\t"
-                "movq %%mm0, %%mm3                      \n\t"
-                "movq %%mm0, %%mm4                      \n\t"
-                PMAXUB(%%mm1, %%mm4)
-                PMINUB(%%mm1, %%mm3, %%mm5)
-                "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
-                "paddb %%mm7, %%mm0                     \n\t"
-                "pcmpgtb %%mm6, %%mm0                   \n\t"
-
-                "movq (%%"REG_a",%3), %%mm2             \n\t"
-                PMAXUB(%%mm2, %%mm4)
-                PMINUB(%%mm2, %%mm3, %%mm5)
-                "psubb %%mm2, %%mm1                     \n\t"
-                "paddb %%mm7, %%mm1                     \n\t"
-                "pcmpgtb %%mm6, %%mm1                   \n\t"
-                "paddb %%mm1, %%mm0                     \n\t"
-
-                "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
-                PMAXUB(%%mm1, %%mm4)
-                PMINUB(%%mm1, %%mm3, %%mm5)
-                "psubb %%mm1, %%mm2                     \n\t"
-                "paddb %%mm7, %%mm2                     \n\t"
-                "pcmpgtb %%mm6, %%mm2                   \n\t"
-                "paddb %%mm2, %%mm0                     \n\t"
-
-                "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
-
-                "movq (%2, %3, 4), %%mm2                \n\t"
-                PMAXUB(%%mm2, %%mm4)
-                PMINUB(%%mm2, %%mm3, %%mm5)
-                "psubb %%mm2, %%mm1                     \n\t"
-                "paddb %%mm7, %%mm1                     \n\t"
-                "pcmpgtb %%mm6, %%mm1                   \n\t"
-                "paddb %%mm1, %%mm0                     \n\t"
-
-                "movq (%%"REG_a"), %%mm1                \n\t"
-                PMAXUB(%%mm1, %%mm4)
-                PMINUB(%%mm1, %%mm3, %%mm5)
-                "psubb %%mm1, %%mm2                     \n\t"
-                "paddb %%mm7, %%mm2                     \n\t"
-                "pcmpgtb %%mm6, %%mm2                   \n\t"
-                "paddb %%mm2, %%mm0                     \n\t"
-
-                "movq (%%"REG_a", %3), %%mm2            \n\t"
-                PMAXUB(%%mm2, %%mm4)
-                PMINUB(%%mm2, %%mm3, %%mm5)
-                "psubb %%mm2, %%mm1                     \n\t"
-                "paddb %%mm7, %%mm1                     \n\t"
-                "pcmpgtb %%mm6, %%mm1                   \n\t"
-                "paddb %%mm1, %%mm0                     \n\t"
-
-                "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
-                PMAXUB(%%mm1, %%mm4)
-                PMINUB(%%mm1, %%mm3, %%mm5)
-                "psubb %%mm1, %%mm2                     \n\t"
-                "paddb %%mm7, %%mm2                     \n\t"
-                "pcmpgtb %%mm6, %%mm2                   \n\t"
-                "paddb %%mm2, %%mm0                     \n\t"
-                "psubusb %%mm3, %%mm4                   \n\t"
-
-                "                                       \n\t"
-#ifdef HAVE_MMX2
-                "pxor %%mm7, %%mm7                      \n\t"
-                "psadbw %%mm7, %%mm0                    \n\t"
+        "movq (%2), %%mm0                       \n\t"
+        "movq (%%"REG_a"), %%mm1                \n\t"
+        "movq %%mm0, %%mm3                      \n\t"
+        "movq %%mm0, %%mm4                      \n\t"
+        PMAXUB(%%mm1, %%mm4)
+        PMINUB(%%mm1, %%mm3, %%mm5)
+        "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
+        "paddb %%mm7, %%mm0                     \n\t"
+        "pcmpgtb %%mm6, %%mm0                   \n\t"
+
+        "movq (%%"REG_a",%3), %%mm2             \n\t"
+        PMAXUB(%%mm2, %%mm4)
+        PMINUB(%%mm2, %%mm3, %%mm5)
+        "psubb %%mm2, %%mm1                     \n\t"
+        "paddb %%mm7, %%mm1                     \n\t"
+        "pcmpgtb %%mm6, %%mm1                   \n\t"
+        "paddb %%mm1, %%mm0                     \n\t"
+
+        "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
+        PMAXUB(%%mm1, %%mm4)
+        PMINUB(%%mm1, %%mm3, %%mm5)
+        "psubb %%mm1, %%mm2                     \n\t"
+        "paddb %%mm7, %%mm2                     \n\t"
+        "pcmpgtb %%mm6, %%mm2                   \n\t"
+        "paddb %%mm2, %%mm0                     \n\t"
+
+        "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
+
+        "movq (%2, %3, 4), %%mm2                \n\t"
+        PMAXUB(%%mm2, %%mm4)
+        PMINUB(%%mm2, %%mm3, %%mm5)
+        "psubb %%mm2, %%mm1                     \n\t"
+        "paddb %%mm7, %%mm1                     \n\t"
+        "pcmpgtb %%mm6, %%mm1                   \n\t"
+        "paddb %%mm1, %%mm0                     \n\t"
+
+        "movq (%%"REG_a"), %%mm1                \n\t"
+        PMAXUB(%%mm1, %%mm4)
+        PMINUB(%%mm1, %%mm3, %%mm5)
+        "psubb %%mm1, %%mm2                     \n\t"
+        "paddb %%mm7, %%mm2                     \n\t"
+        "pcmpgtb %%mm6, %%mm2                   \n\t"
+        "paddb %%mm2, %%mm0                     \n\t"
+
+        "movq (%%"REG_a", %3), %%mm2            \n\t"
+        PMAXUB(%%mm2, %%mm4)
+        PMINUB(%%mm2, %%mm3, %%mm5)
+        "psubb %%mm2, %%mm1                     \n\t"
+        "paddb %%mm7, %%mm1                     \n\t"
+        "pcmpgtb %%mm6, %%mm1                   \n\t"
+        "paddb %%mm1, %%mm0                     \n\t"
+
+        "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
+        PMAXUB(%%mm1, %%mm4)
+        PMINUB(%%mm1, %%mm3, %%mm5)
+        "psubb %%mm1, %%mm2                     \n\t"
+        "paddb %%mm7, %%mm2                     \n\t"
+        "pcmpgtb %%mm6, %%mm2                   \n\t"
+        "paddb %%mm2, %%mm0                     \n\t"
+        "psubusb %%mm3, %%mm4                   \n\t"
+
+        "                                       \n\t"
+#if HAVE_MMX2
+        "pxor %%mm7, %%mm7                      \n\t"
+        "psadbw %%mm7, %%mm0                    \n\t"
 #else
-                "movq %%mm0, %%mm1                      \n\t"
-                "psrlw $8, %%mm0                        \n\t"
-                "paddb %%mm1, %%mm0                     \n\t"
-                "movq %%mm0, %%mm1                      \n\t"
-                "psrlq $16, %%mm0                       \n\t"
-                "paddb %%mm1, %%mm0                     \n\t"
-                "movq %%mm0, %%mm1                      \n\t"
-                "psrlq $32, %%mm0                       \n\t"
-                "paddb %%mm1, %%mm0                     \n\t"
+        "movq %%mm0, %%mm1                      \n\t"
+        "psrlw $8, %%mm0                        \n\t"
+        "paddb %%mm1, %%mm0                     \n\t"
+        "movq %%mm0, %%mm1                      \n\t"
+        "psrlq $16, %%mm0                       \n\t"
+        "paddb %%mm1, %%mm0                     \n\t"
+        "movq %%mm0, %%mm1                      \n\t"
+        "psrlq $32, %%mm0                       \n\t"
+        "paddb %%mm1, %%mm0                     \n\t"
 #endif
-                "movq %4, %%mm7                         \n\t" // QP,..., QP
-                "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
-                "psubusb %%mm7, %%mm4                   \n\t" // Diff <= 2QP -> 0
-                "packssdw %%mm4, %%mm4                  \n\t"
-                "movd %%mm0, %0                         \n\t"
-                "movd %%mm4, %1                         \n\t"
-
-                : "=r" (numEq), "=r" (dcOk)
-                : "r" (src), "r" ((long)stride), "m" (c->pQPb)
-                : "%"REG_a
-                );
-
-        numEq= (-numEq) &0xFF;
-        if(numEq > c->ppMode.flatnessThreshold){
-            if(dcOk) return 0;
-            else     return 1;
-        }else{
-            return 2;
-        }
+        "movq %4, %%mm7                         \n\t" // QP,..., QP
+        "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
+        "psubusb %%mm7, %%mm4                   \n\t" // Diff <= 2QP -> 0
+        "packssdw %%mm4, %%mm4                  \n\t"
+        "movd %%mm0, %0                         \n\t"
+        "movd %%mm4, %1                         \n\t"
+
+        : "=r" (numEq), "=r" (dcOk)
+        : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
+        : "%"REG_a
+        );
+
+    numEq= (-numEq) &0xFF;
+    if(numEq > c->ppMode.flatnessThreshold){
+        if(dcOk) return 0;
+        else     return 1;
+    }else{
+        return 2;
+    }
 }
 #endif //HAVE_MMX
 
@@ -194,1028 +177,904 @@ asm volatile(
  * Do a vertical low pass filter on the 8x16 block (only write to the 8x8 block in the middle)
  * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16
  */
-#ifndef HAVE_ALTIVEC
+#if !HAVE_ALTIVEC
 static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
 {
-#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        src+= stride*3;
-        asm volatile(        //"movv %0 %1 %2\n\t"
-                "movq %2, %%mm0                         \n\t"  // QP,..., QP
-                "pxor %%mm4, %%mm4                      \n\t"
-
-                "movq (%0), %%mm6                       \n\t"
-                "movq (%0, %1), %%mm5                   \n\t"
-                "movq %%mm5, %%mm1                      \n\t"
-                "movq %%mm6, %%mm2                      \n\t"
-                "psubusb %%mm6, %%mm5                   \n\t"
-                "psubusb %%mm1, %%mm2                   \n\t"
-                "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
-                "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
-                "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
-
-                "pand %%mm2, %%mm6                      \n\t"
-                "pandn %%mm1, %%mm2                     \n\t"
-                "por %%mm2, %%mm6                       \n\t"// First Line to Filter
-
-                "movq (%0, %1, 8), %%mm5                \n\t"
-                "lea (%0, %1, 4), %%"REG_a"             \n\t"
-                "lea (%0, %1, 8), %%"REG_c"             \n\t"
-                "sub %1, %%"REG_c"                      \n\t"
-                "add %1, %0                             \n\t" // %0 points to line 1 not 0
-                "movq (%0, %1, 8), %%mm7                \n\t"
-                "movq %%mm5, %%mm1                      \n\t"
-                "movq %%mm7, %%mm2                      \n\t"
-                "psubusb %%mm7, %%mm5                   \n\t"
-                "psubusb %%mm1, %%mm2                   \n\t"
-                "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
-                "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
-                "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
-
-                "pand %%mm2, %%mm7                      \n\t"
-                "pandn %%mm1, %%mm2                     \n\t"
-                "por %%mm2, %%mm7                       \n\t" // First Line to Filter
-
-
-                //      1       2       3       4       5       6       7       8
-                //      %0      %0+%1   %0+2%1  eax     %0+4%1  eax+2%1 ecx     eax+4%1
-                // 6 4 2 2 1 1
-                // 6 4 4 2
-                // 6 8 2
-
-                "movq (%0, %1), %%mm0                   \n\t" //  1
-                "movq %%mm0, %%mm1                      \n\t" //  1
-                PAVGB(%%mm6, %%mm0)                           //1 1        /2
-                PAVGB(%%mm6, %%mm0)                           //3 1        /4
-
-                "movq (%0, %1, 4), %%mm2                \n\t" //     1
-                "movq %%mm2, %%mm5                      \n\t" //     1
-                PAVGB((%%REGa), %%mm2)                        //    11        /2
-                PAVGB((%0, %1, 2), %%mm2)                     //   211        /4
-                "movq %%mm2, %%mm3                      \n\t" //   211        /4
-                "movq (%0), %%mm4                       \n\t" // 1
-                PAVGB(%%mm4, %%mm3)                           // 4 211        /8
-                PAVGB(%%mm0, %%mm3)                           //642211        /16
-                "movq %%mm3, (%0)                       \n\t" // X
-                // mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
-                "movq %%mm1, %%mm0                      \n\t" //  1
-                PAVGB(%%mm6, %%mm0)                           //1 1        /2
-                "movq %%mm4, %%mm3                      \n\t" // 1
-                PAVGB((%0,%1,2), %%mm3)                       // 1 1        /2
-                PAVGB((%%REGa,%1,2), %%mm5)                   //     11        /2
-                PAVGB((%%REGa), %%mm5)                        //    211 /4
-                PAVGB(%%mm5, %%mm3)                           // 2 2211 /8
-                PAVGB(%%mm0, %%mm3)                           //4242211 /16
-                "movq %%mm3, (%0,%1)                    \n\t" //  X
-                // mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
-                PAVGB(%%mm4, %%mm6)                                   //11        /2
-                "movq (%%"REG_c"), %%mm0                \n\t" //       1
-                PAVGB((%%REGa, %1, 2), %%mm0)                 //      11/2
-                "movq %%mm0, %%mm3                      \n\t" //      11/2
-                PAVGB(%%mm1, %%mm0)                           //  2   11/4
-                PAVGB(%%mm6, %%mm0)                           //222   11/8
-                PAVGB(%%mm2, %%mm0)                           //22242211/16
-                "movq (%0, %1, 2), %%mm2                \n\t" //   1
-                "movq %%mm0, (%0, %1, 2)                \n\t" //   X
-                // mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
-                "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
-                PAVGB((%%REGc), %%mm0)                        //       11        /2
-                PAVGB(%%mm0, %%mm6)                           //11     11        /4
-                PAVGB(%%mm1, %%mm4)                           // 11                /2
-                PAVGB(%%mm2, %%mm1)                           //  11                /2
-                PAVGB(%%mm1, %%mm6)                           //1122   11        /8
-                PAVGB(%%mm5, %%mm6)                           //112242211        /16
-                "movq (%%"REG_a"), %%mm5                \n\t" //    1
-                "movq %%mm6, (%%"REG_a")                \n\t" //    X
-                // mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
-                "movq (%%"REG_a", %1, 4), %%mm6         \n\t" //        1
-                PAVGB(%%mm7, %%mm6)                           //        11        /2
-                PAVGB(%%mm4, %%mm6)                           // 11     11        /4
-                PAVGB(%%mm3, %%mm6)                           // 11   2211        /8
-                PAVGB(%%mm5, %%mm2)                           //   11                /2
-                "movq (%0, %1, 4), %%mm4                \n\t" //     1
-                PAVGB(%%mm4, %%mm2)                           //   112                /4
-                PAVGB(%%mm2, %%mm6)                           // 112242211        /16
-                "movq %%mm6, (%0, %1, 4)                \n\t" //     X
-                // mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
-                PAVGB(%%mm7, %%mm1)                           //  11     2        /4
-                PAVGB(%%mm4, %%mm5)                           //    11                /2
-                PAVGB(%%mm5, %%mm0)                           //    11 11        /4
-                "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //      1
-                PAVGB(%%mm6, %%mm1)                           //  11  4  2        /8
-                PAVGB(%%mm0, %%mm1)                           //  11224222        /16
-                "movq %%mm1, (%%"REG_a", %1, 2)         \n\t" //      X
-                // mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
-                PAVGB((%%REGc), %%mm2)                        //   112 4        /8
-                "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
-                PAVGB(%%mm0, %%mm6)                           //      1 1        /2
-                PAVGB(%%mm7, %%mm6)                           //      1 12        /4
-                PAVGB(%%mm2, %%mm6)                           //   1122424        /4
-                "movq %%mm6, (%%"REG_c")                \n\t" //       X
-                // mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
-                PAVGB(%%mm7, %%mm5)                           //    11   2        /4
-                PAVGB(%%mm7, %%mm5)                           //    11   6        /8
-
-                PAVGB(%%mm3, %%mm0)                           //      112        /4
-                PAVGB(%%mm0, %%mm5)                           //    112246        /16
-                "movq %%mm5, (%%"REG_a", %1, 4)         \n\t" //        X
-                "sub %1, %0                             \n\t"
-
-                :
-                : "r" (src), "r" ((long)stride), "m" (c->pQPb)
-                : "%"REG_a, "%"REG_c
-        );
-#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        const int l1= stride;
-        const int l2= stride + l1;
-        const int l3= stride + l2;
-        const int l4= stride + l3;
-        const int l5= stride + l4;
-        const int l6= stride + l5;
-        const int l7= stride + l6;
-        const int l8= stride + l7;
-        const int l9= stride + l8;
-        int x;
-        src+= stride*3;
-        for(x=0; x<BLOCK_SIZE; x++)
-        {
-                const int first= FFABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];
-                const int last= FFABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];
-
-                int sums[10];
-                sums[0] = 4*first + src[l1] + src[l2] + src[l3] + 4;
-                sums[1] = sums[0] - first  + src[l4];
-                sums[2] = sums[1] - first  + src[l5];
-                sums[3] = sums[2] - first  + src[l6];
-                sums[4] = sums[3] - first  + src[l7];
-                sums[5] = sums[4] - src[l1] + src[l8];
-                sums[6] = sums[5] - src[l2] + last;
-                sums[7] = sums[6] - src[l3] + last;
-                sums[8] = sums[7] - src[l4] + last;
-                sums[9] = sums[8] - src[l5] + last;
-
-                src[l1]= (sums[0] + sums[2] + 2*src[l1])>>4;
-                src[l2]= (sums[1] + sums[3] + 2*src[l2])>>4;
-                src[l3]= (sums[2] + sums[4] + 2*src[l3])>>4;
-                src[l4]= (sums[3] + sums[5] + 2*src[l4])>>4;
-                src[l5]= (sums[4] + sums[6] + 2*src[l5])>>4;
-                src[l6]= (sums[5] + sums[7] + 2*src[l6])>>4;
-                src[l7]= (sums[6] + sums[8] + 2*src[l7])>>4;
-                src[l8]= (sums[7] + sums[9] + 2*src[l8])>>4;
-
-                src++;
-        }
-#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
+#if HAVE_MMX2 || HAVE_AMD3DNOW
+    src+= stride*3;
+    __asm__ volatile(        //"movv %0 %1 %2\n\t"
+        "movq %2, %%mm0                         \n\t"  // QP,..., QP
+        "pxor %%mm4, %%mm4                      \n\t"
+
+        "movq (%0), %%mm6                       \n\t"
+        "movq (%0, %1), %%mm5                   \n\t"
+        "movq %%mm5, %%mm1                      \n\t"
+        "movq %%mm6, %%mm2                      \n\t"
+        "psubusb %%mm6, %%mm5                   \n\t"
+        "psubusb %%mm1, %%mm2                   \n\t"
+        "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
+        "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
+        "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
+
+        "pand %%mm2, %%mm6                      \n\t"
+        "pandn %%mm1, %%mm2                     \n\t"
+        "por %%mm2, %%mm6                       \n\t"// First Line to Filter
+
+        "movq (%0, %1, 8), %%mm5                \n\t"
+        "lea (%0, %1, 4), %%"REG_a"             \n\t"
+        "lea (%0, %1, 8), %%"REG_c"             \n\t"
+        "sub %1, %%"REG_c"                      \n\t"
+        "add %1, %0                             \n\t" // %0 points to line 1 not 0
+        "movq (%0, %1, 8), %%mm7                \n\t"
+        "movq %%mm5, %%mm1                      \n\t"
+        "movq %%mm7, %%mm2                      \n\t"
+        "psubusb %%mm7, %%mm5                   \n\t"
+        "psubusb %%mm1, %%mm2                   \n\t"
+        "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
+        "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
+        "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
+
+        "pand %%mm2, %%mm7                      \n\t"
+        "pandn %%mm1, %%mm2                     \n\t"
+        "por %%mm2, %%mm7                       \n\t" // First Line to Filter
+
+
+        //      1       2       3       4       5       6       7       8
+        //      %0      %0+%1   %0+2%1  eax     %0+4%1  eax+2%1 ecx     eax+4%1
+        // 6 4 2 2 1 1
+        // 6 4 4 2
+        // 6 8 2
+
+        "movq (%0, %1), %%mm0                   \n\t" //  1
+        "movq %%mm0, %%mm1                      \n\t" //  1
+        PAVGB(%%mm6, %%mm0)                           //1 1        /2
+        PAVGB(%%mm6, %%mm0)                           //3 1        /4
+
+        "movq (%0, %1, 4), %%mm2                \n\t" //     1
+        "movq %%mm2, %%mm5                      \n\t" //     1
+        PAVGB((%%REGa), %%mm2)                        //    11        /2
+        PAVGB((%0, %1, 2), %%mm2)                     //   211        /4
+        "movq %%mm2, %%mm3                      \n\t" //   211        /4
+        "movq (%0), %%mm4                       \n\t" // 1
+        PAVGB(%%mm4, %%mm3)                           // 4 211        /8
+        PAVGB(%%mm0, %%mm3)                           //642211        /16
+        "movq %%mm3, (%0)                       \n\t" // X
+        // mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
+        "movq %%mm1, %%mm0                      \n\t" //  1
+        PAVGB(%%mm6, %%mm0)                           //1 1        /2
+        "movq %%mm4, %%mm3                      \n\t" // 1
+        PAVGB((%0,%1,2), %%mm3)                       // 1 1        /2
+        PAVGB((%%REGa,%1,2), %%mm5)                   //     11        /2
+        PAVGB((%%REGa), %%mm5)                        //    211 /4
+        PAVGB(%%mm5, %%mm3)                           // 2 2211 /8
+        PAVGB(%%mm0, %%mm3)                           //4242211 /16
+        "movq %%mm3, (%0,%1)                    \n\t" //  X
+        // mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
+        PAVGB(%%mm4, %%mm6)                                   //11        /2
+        "movq (%%"REG_c"), %%mm0                \n\t" //       1
+        PAVGB((%%REGa, %1, 2), %%mm0)                 //      11/2
+        "movq %%mm0, %%mm3                      \n\t" //      11/2
+        PAVGB(%%mm1, %%mm0)                           //  2   11/4
+        PAVGB(%%mm6, %%mm0)                           //222   11/8
+        PAVGB(%%mm2, %%mm0)                           //22242211/16
+        "movq (%0, %1, 2), %%mm2                \n\t" //   1
+        "movq %%mm0, (%0, %1, 2)                \n\t" //   X
+        // mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
+        "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
+        PAVGB((%%REGc), %%mm0)                        //       11        /2
+        PAVGB(%%mm0, %%mm6)                           //11     11        /4
+        PAVGB(%%mm1, %%mm4)                           // 11                /2
+        PAVGB(%%mm2, %%mm1)                           //  11                /2
+        PAVGB(%%mm1, %%mm6)                           //1122   11        /8
+        PAVGB(%%mm5, %%mm6)                           //112242211        /16
+        "movq (%%"REG_a"), %%mm5                \n\t" //    1
+        "movq %%mm6, (%%"REG_a")                \n\t" //    X
+        // mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
+        "movq (%%"REG_a", %1, 4), %%mm6         \n\t" //        1
+        PAVGB(%%mm7, %%mm6)                           //        11        /2
+        PAVGB(%%mm4, %%mm6)                           // 11     11        /4
+        PAVGB(%%mm3, %%mm6)                           // 11   2211        /8
+        PAVGB(%%mm5, %%mm2)                           //   11                /2
+        "movq (%0, %1, 4), %%mm4                \n\t" //     1
+        PAVGB(%%mm4, %%mm2)                           //   112                /4
+        PAVGB(%%mm2, %%mm6)                           // 112242211        /16
+        "movq %%mm6, (%0, %1, 4)                \n\t" //     X
+        // mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
+        PAVGB(%%mm7, %%mm1)                           //  11     2        /4
+        PAVGB(%%mm4, %%mm5)                           //    11                /2
+        PAVGB(%%mm5, %%mm0)                           //    11 11        /4
+        "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //      1
+        PAVGB(%%mm6, %%mm1)                           //  11  4  2        /8
+        PAVGB(%%mm0, %%mm1)                           //  11224222        /16
+        "movq %%mm1, (%%"REG_a", %1, 2)         \n\t" //      X
+        // mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
+        PAVGB((%%REGc), %%mm2)                        //   112 4        /8
+        "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
+        PAVGB(%%mm0, %%mm6)                           //      1 1        /2
+        PAVGB(%%mm7, %%mm6)                           //      1 12        /4
+        PAVGB(%%mm2, %%mm6)                           //   1122424        /4
+        "movq %%mm6, (%%"REG_c")                \n\t" //       X
+        // mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
+        PAVGB(%%mm7, %%mm5)                           //    11   2        /4
+        PAVGB(%%mm7, %%mm5)                           //    11   6        /8
+
+        PAVGB(%%mm3, %%mm0)                           //      112        /4
+        PAVGB(%%mm0, %%mm5)                           //    112246        /16
+        "movq %%mm5, (%%"REG_a", %1, 4)         \n\t" //        X
+        "sub %1, %0                             \n\t"
+
+        :
+        : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
+        : "%"REG_a, "%"REG_c
+    );
+#else //HAVE_MMX2 || HAVE_AMD3DNOW
+    const int l1= stride;
+    const int l2= stride + l1;
+    const int l3= stride + l2;
+    const int l4= stride + l3;
+    const int l5= stride + l4;
+    const int l6= stride + l5;
+    const int l7= stride + l6;
+    const int l8= stride + l7;
+    const int l9= stride + l8;
+    int x;
+    src+= stride*3;
+    for(x=0; x<BLOCK_SIZE; x++){
+        const int first= FFABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];
+        const int last= FFABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];
+
+        int sums[10];
+        sums[0] = 4*first + src[l1] + src[l2] + src[l3] + 4;
+        sums[1] = sums[0] - first  + src[l4];
+        sums[2] = sums[1] - first  + src[l5];
+        sums[3] = sums[2] - first  + src[l6];
+        sums[4] = sums[3] - first  + src[l7];
+        sums[5] = sums[4] - src[l1] + src[l8];
+        sums[6] = sums[5] - src[l2] + last;
+        sums[7] = sums[6] - src[l3] + last;
+        sums[8] = sums[7] - src[l4] + last;
+        sums[9] = sums[8] - src[l5] + last;
+
+        src[l1]= (sums[0] + sums[2] + 2*src[l1])>>4;
+        src[l2]= (sums[1] + sums[3] + 2*src[l2])>>4;
+        src[l3]= (sums[2] + sums[4] + 2*src[l3])>>4;
+        src[l4]= (sums[3] + sums[5] + 2*src[l4])>>4;
+        src[l5]= (sums[4] + sums[6] + 2*src[l5])>>4;
+        src[l6]= (sums[5] + sums[7] + 2*src[l6])>>4;
+        src[l7]= (sums[6] + sums[8] + 2*src[l7])>>4;
+        src[l8]= (sums[7] + sums[9] + 2*src[l8])>>4;
+
+        src++;
+    }
+#endif //HAVE_MMX2 || HAVE_AMD3DNOW
 }
 #endif //HAVE_ALTIVEC
 
-#if 0
-/**
- * Experimental implementation of the filter (Algorithm 1) described in a paper from Ramkishor & Karandikar
- * values are correctly clipped (MMX2)
- * values are wraparound (C)
- * Conclusion: It is fast, but introduces ugly horizontal patterns
- * if there is a continuous gradient.
-        0 8 16 24
-        x = 8
-        x/2 = 4
-        x/8 = 1
-        1 12 12 23
- */
-static inline void RENAME(vertRK1Filter)(uint8_t *src, int stride, int QP)
-{
-#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        src+= stride*3;
-// FIXME rounding
-        asm volatile(
-                "pxor %%mm7, %%mm7                      \n\t" // 0
-                "movq "MANGLE(b80)", %%mm6              \n\t" // MIN_SIGNED_BYTE
-                "leal (%0, %1), %%"REG_a"               \n\t"
-                "leal (%%"REG_a", %1, 4), %%"REG_c"     \n\t"
-//      0       1       2       3       4       5       6       7       8       9
-//      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
-                "movq "MANGLE(pQPb)", %%mm0             \n\t" // QP,..., QP
-                "movq %%mm0, %%mm1                      \n\t" // QP,..., QP
-                "paddusb "MANGLE(b02)", %%mm0           \n\t"
-                "psrlw $2, %%mm0                        \n\t"
-                "pand "MANGLE(b3F)", %%mm0              \n\t" // QP/4,..., QP/4
-                "paddusb %%mm1, %%mm0                   \n\t" // QP*1.25 ...
-                "movq (%0, %1, 4), %%mm2                \n\t" // line 4
-                "movq (%%"REG_c"), %%mm3                \n\t" // line 5
-                "movq %%mm2, %%mm4                      \n\t" // line 4
-                "pcmpeqb %%mm5, %%mm5                   \n\t" // -1
-                "pxor %%mm2, %%mm5                      \n\t" // -line 4 - 1
-                PAVGB(%%mm3, %%mm5)
-                "paddb %%mm6, %%mm5                     \n\t" // (l5-l4)/2
-                "psubusb %%mm3, %%mm4                   \n\t"
-                "psubusb %%mm2, %%mm3                   \n\t"
-                "por %%mm3, %%mm4                       \n\t" // |l4 - l5|
-                "psubusb %%mm0, %%mm4                   \n\t"
-                "pcmpeqb %%mm7, %%mm4                   \n\t"
-                "pand %%mm4, %%mm5                      \n\t" // d/2
-
-//                "paddb %%mm6, %%mm2                     \n\t" // line 4 + 0x80
-                "paddb %%mm5, %%mm2                     \n\t"
-//                "psubb %%mm6, %%mm2                     \n\t"
-                "movq %%mm2, (%0,%1, 4)                 \n\t"
-
-                "movq (%%"REG_c"), %%mm2                \n\t"
-//                "paddb %%mm6, %%mm2                     \n\t" // line 5 + 0x80
-                "psubb %%mm5, %%mm2                     \n\t"
-//                "psubb %%mm6, %%mm2                     \n\t"
-                "movq %%mm2, (%%"REG_c")                \n\t"
-
-                "paddb %%mm6, %%mm5                     \n\t"
-                "psrlw $2, %%mm5                        \n\t"
-                "pand "MANGLE(b3F)", %%mm5              \n\t"
-                "psubb "MANGLE(b20)", %%mm5             \n\t" // (l5-l4)/8
-
-                "movq (%%"REG_a", %1, 2), %%mm2         \n\t"
-                "paddb %%mm6, %%mm2                     \n\t" // line 3 + 0x80
-                "paddsb %%mm5, %%mm2                    \n\t"
-                "psubb %%mm6, %%mm2                     \n\t"
-                "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"
-
-                "movq (%%"REG_c", %1), %%mm2            \n\t"
-                "paddb %%mm6, %%mm2                     \n\t" // line 6 + 0x80
-                "psubsb %%mm5, %%mm2                    \n\t"
-                "psubb %%mm6, %%mm2                     \n\t"
-                "movq %%mm2, (%%"REG_c", %1)            \n\t"
-
-                :
-                : "r" (src), "r" ((long)stride)
-                : "%"REG_a, "%"REG_c
-        );
-#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-         const int l1= stride;
-        const int l2= stride + l1;
-        const int l3= stride + l2;
-        const int l4= stride + l3;
-        const int l5= stride + l4;
-        const int l6= stride + l5;
-//        const int l7= stride + l6;
-//        const int l8= stride + l7;
-//        const int l9= stride + l8;
-        int x;
-        const int QP15= QP + (QP>>2);
-        src+= stride*3;
-        for(x=0; x<BLOCK_SIZE; x++)
-        {
-                const int v = (src[x+l5] - src[x+l4]);
-                if(FFABS(v) < QP15)
-                {
-                        src[x+l3] +=v>>3;
-                        src[x+l4] +=v>>1;
-                        src[x+l5] -=v>>1;
-                        src[x+l6] -=v>>3;
-
-                }
-        }
-
-#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-}
-#endif //0
-
 /**
  * Experimental Filter 1
  * will not damage linear gradients
- * Flat blocks should look like they where passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
+ * Flat blocks should look like they were passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
  * can only smooth blocks at the expected locations (it cannot smooth them if they did move)
  * MMX2 version does correct clipping C version does not
  */
 static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
 {
-#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        src+= stride*3;
+#if HAVE_MMX2 || HAVE_AMD3DNOW
+    src+= stride*3;
 
-        asm volatile(
-                "pxor %%mm7, %%mm7                      \n\t" // 0
-                "lea (%0, %1), %%"REG_a"                \n\t"
-                "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
+    __asm__ volatile(
+        "pxor %%mm7, %%mm7                      \n\t" // 0
+        "lea (%0, %1), %%"REG_a"                \n\t"
+        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
 //      0       1       2       3       4       5       6       7       8       9
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
-                "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
-                "movq (%0, %1, 4), %%mm1                \n\t" // line 4
-                "movq %%mm1, %%mm2                      \n\t" // line 4
-                "psubusb %%mm0, %%mm1                   \n\t"
-                "psubusb %%mm2, %%mm0                   \n\t"
-                "por %%mm1, %%mm0                       \n\t" // |l2 - l3|
-                "movq (%%"REG_c"), %%mm3                \n\t" // line 5
-                "movq (%%"REG_c", %1), %%mm4            \n\t" // line 6
-                "movq %%mm3, %%mm5                      \n\t" // line 5
-                "psubusb %%mm4, %%mm3                   \n\t"
-                "psubusb %%mm5, %%mm4                   \n\t"
-                "por %%mm4, %%mm3                       \n\t" // |l5 - l6|
-                PAVGB(%%mm3, %%mm0)                           // (|l2 - l3| + |l5 - l6|)/2
-                "movq %%mm2, %%mm1                      \n\t" // line 4
-                "psubusb %%mm5, %%mm2                   \n\t"
-                "movq %%mm2, %%mm4                      \n\t"
-                "pcmpeqb %%mm7, %%mm2                   \n\t" // (l4 - l5) <= 0 ? -1 : 0
-                "psubusb %%mm1, %%mm5                   \n\t"
-                "por %%mm5, %%mm4                       \n\t" // |l4 - l5|
-                "psubusb %%mm0, %%mm4                   \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
-                "movq %%mm4, %%mm3                      \n\t" // d
-                "movq %2, %%mm0                         \n\t"
-                "paddusb %%mm0, %%mm0                   \n\t"
-                "psubusb %%mm0, %%mm4                   \n\t"
-                "pcmpeqb %%mm7, %%mm4                   \n\t" // d <= QP ? -1 : 0
-                "psubusb "MANGLE(b01)", %%mm3           \n\t"
-                "pand %%mm4, %%mm3                      \n\t" // d <= QP ? d : 0
-
-                PAVGB(%%mm7, %%mm3)                           // d/2
-                "movq %%mm3, %%mm1                      \n\t" // d/2
-                PAVGB(%%mm7, %%mm3)                           // d/4
-                PAVGB(%%mm1, %%mm3)                           // 3*d/8
-
-                "movq (%0, %1, 4), %%mm0                \n\t" // line 4
-                "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
-                "psubusb %%mm3, %%mm0                   \n\t"
-                "pxor %%mm2, %%mm0                      \n\t"
-                "movq %%mm0, (%0, %1, 4)                \n\t" // line 4
-
-                "movq (%%"REG_c"), %%mm0                \n\t" // line 5
-                "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
-                "paddusb %%mm3, %%mm0                   \n\t"
-                "pxor %%mm2, %%mm0                      \n\t"
-                "movq %%mm0, (%%"REG_c")                \n\t" // line 5
-
-                PAVGB(%%mm7, %%mm1)                           // d/4
-
-                "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
-                "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
-                "psubusb %%mm1, %%mm0                   \n\t"
-                "pxor %%mm2, %%mm0                      \n\t"
-                "movq %%mm0, (%%"REG_a", %1, 2)         \n\t" // line 3
-
-                "movq (%%"REG_c", %1), %%mm0            \n\t" // line 6
-                "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
-                "paddusb %%mm1, %%mm0                   \n\t"
-                "pxor %%mm2, %%mm0                      \n\t"
-                "movq %%mm0, (%%"REG_c", %1)            \n\t" // line 6
-
-                PAVGB(%%mm7, %%mm1)                           // d/8
-
-                "movq (%%"REG_a", %1), %%mm0            \n\t" // line 2
-                "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
-                "psubusb %%mm1, %%mm0                   \n\t"
-                "pxor %%mm2, %%mm0                      \n\t"
-                "movq %%mm0, (%%"REG_a", %1)            \n\t" // line 2
-
-                "movq (%%"REG_c", %1, 2), %%mm0         \n\t" // line 7
-                "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
-                "paddusb %%mm1, %%mm0                   \n\t"
-                "pxor %%mm2, %%mm0                      \n\t"
-                "movq %%mm0, (%%"REG_c", %1, 2)         \n\t" // line 7
-
-                :
-                : "r" (src), "r" ((long)stride), "m" (co->pQPb)
-                : "%"REG_a, "%"REG_c
-        );
-#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-
-        const int l1= stride;
-        const int l2= stride + l1;
-        const int l3= stride + l2;
-        const int l4= stride + l3;
-        const int l5= stride + l4;
-        const int l6= stride + l5;
-        const int l7= stride + l6;
-//        const int l8= stride + l7;
-//        const int l9= stride + l8;
-        int x;
-
-        src+= stride*3;
-        for(x=0; x<BLOCK_SIZE; x++)
-        {
-                int a= src[l3] - src[l4];
-                int b= src[l4] - src[l5];
-                int c= src[l5] - src[l6];
-
-                int d= FFABS(b) - ((FFABS(a) + FFABS(c))>>1);
-                d= FFMAX(d, 0);
-
-                if(d < co->QP*2)
-                {
-                        int v = d * FFSIGN(-b);
-
-                        src[l2] +=v>>3;
-                        src[l3] +=v>>2;
-                        src[l4] +=(3*v)>>3;
-                        src[l5] -=(3*v)>>3;
-                        src[l6] -=v>>2;
-                        src[l7] -=v>>3;
-
-                }
-                src++;
+        "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
+        "movq (%0, %1, 4), %%mm1                \n\t" // line 4
+        "movq %%mm1, %%mm2                      \n\t" // line 4
+        "psubusb %%mm0, %%mm1                   \n\t"
+        "psubusb %%mm2, %%mm0                   \n\t"
+        "por %%mm1, %%mm0                       \n\t" // |l2 - l3|
+        "movq (%%"REG_c"), %%mm3                \n\t" // line 5
+        "movq (%%"REG_c", %1), %%mm4            \n\t" // line 6
+        "movq %%mm3, %%mm5                      \n\t" // line 5
+        "psubusb %%mm4, %%mm3                   \n\t"
+        "psubusb %%mm5, %%mm4                   \n\t"
+        "por %%mm4, %%mm3                       \n\t" // |l5 - l6|
+        PAVGB(%%mm3, %%mm0)                           // (|l2 - l3| + |l5 - l6|)/2
+        "movq %%mm2, %%mm1                      \n\t" // line 4
+        "psubusb %%mm5, %%mm2                   \n\t"
+        "movq %%mm2, %%mm4                      \n\t"
+        "pcmpeqb %%mm7, %%mm2                   \n\t" // (l4 - l5) <= 0 ? -1 : 0
+        "psubusb %%mm1, %%mm5                   \n\t"
+        "por %%mm5, %%mm4                       \n\t" // |l4 - l5|
+        "psubusb %%mm0, %%mm4                   \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
+        "movq %%mm4, %%mm3                      \n\t" // d
+        "movq %2, %%mm0                         \n\t"
+        "paddusb %%mm0, %%mm0                   \n\t"
+        "psubusb %%mm0, %%mm4                   \n\t"
+        "pcmpeqb %%mm7, %%mm4                   \n\t" // d <= QP ? -1 : 0
+        "psubusb "MANGLE(b01)", %%mm3           \n\t"
+        "pand %%mm4, %%mm3                      \n\t" // d <= QP ? d : 0
+
+        PAVGB(%%mm7, %%mm3)                           // d/2
+        "movq %%mm3, %%mm1                      \n\t" // d/2
+        PAVGB(%%mm7, %%mm3)                           // d/4
+        PAVGB(%%mm1, %%mm3)                           // 3*d/8
+
+        "movq (%0, %1, 4), %%mm0                \n\t" // line 4
+        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
+        "psubusb %%mm3, %%mm0                   \n\t"
+        "pxor %%mm2, %%mm0                      \n\t"
+        "movq %%mm0, (%0, %1, 4)                \n\t" // line 4
+
+        "movq (%%"REG_c"), %%mm0                \n\t" // line 5
+        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
+        "paddusb %%mm3, %%mm0                   \n\t"
+        "pxor %%mm2, %%mm0                      \n\t"
+        "movq %%mm0, (%%"REG_c")                \n\t" // line 5
+
+        PAVGB(%%mm7, %%mm1)                           // d/4
+
+        "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
+        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
+        "psubusb %%mm1, %%mm0                   \n\t"
+        "pxor %%mm2, %%mm0                      \n\t"
+        "movq %%mm0, (%%"REG_a", %1, 2)         \n\t" // line 3
+
+        "movq (%%"REG_c", %1), %%mm0            \n\t" // line 6
+        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
+        "paddusb %%mm1, %%mm0                   \n\t"
+        "pxor %%mm2, %%mm0                      \n\t"
+        "movq %%mm0, (%%"REG_c", %1)            \n\t" // line 6
+
+        PAVGB(%%mm7, %%mm1)                           // d/8
+
+        "movq (%%"REG_a", %1), %%mm0            \n\t" // line 2
+        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
+        "psubusb %%mm1, %%mm0                   \n\t"
+        "pxor %%mm2, %%mm0                      \n\t"
+        "movq %%mm0, (%%"REG_a", %1)            \n\t" // line 2
+
+        "movq (%%"REG_c", %1, 2), %%mm0         \n\t" // line 7
+        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
+        "paddusb %%mm1, %%mm0                   \n\t"
+        "pxor %%mm2, %%mm0                      \n\t"
+        "movq %%mm0, (%%"REG_c", %1, 2)         \n\t" // line 7
+
+        :
+        : "r" (src), "r" ((x86_reg)stride), "m" (co->pQPb)
+        : "%"REG_a, "%"REG_c
+    );
+#else //HAVE_MMX2 || HAVE_AMD3DNOW
+
+    const int l1= stride;
+    const int l2= stride + l1;
+    const int l3= stride + l2;
+    const int l4= stride + l3;
+    const int l5= stride + l4;
+    const int l6= stride + l5;
+    const int l7= stride + l6;
+//    const int l8= stride + l7;
+//    const int l9= stride + l8;
+    int x;
+
+    src+= stride*3;
+    for(x=0; x<BLOCK_SIZE; x++){
+        int a= src[l3] - src[l4];
+        int b= src[l4] - src[l5];
+        int c= src[l5] - src[l6];
+
+        int d= FFABS(b) - ((FFABS(a) + FFABS(c))>>1);
+        d= FFMAX(d, 0);
+
+        if(d < co->QP*2){
+            int v = d * FFSIGN(-b);
+
+            src[l2] +=v>>3;
+            src[l3] +=v>>2;
+            src[l4] +=(3*v)>>3;
+            src[l5] -=(3*v)>>3;
+            src[l6] -=v>>2;
+            src[l7] -=v>>3;
         }
-#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
+        src++;
+    }
+#endif //HAVE_MMX2 || HAVE_AMD3DNOW
 }
 
-#ifndef HAVE_ALTIVEC
+#if !HAVE_ALTIVEC
 static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext *c)
 {
-#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
+#if HAVE_MMX2 || HAVE_AMD3DNOW
 /*
-        uint8_t tmp[16];
-        const int l1= stride;
-        const int l2= stride + l1;
-        const int l3= stride + l2;
-        const int l4= (int)tmp - (int)src - stride*3;
-        const int l5= (int)tmp - (int)src - stride*3 + 8;
-        const int l6= stride*3 + l3;
-        const int l7= stride + l6;
-        const int l8= stride + l7;
-
-        memcpy(tmp, src+stride*7, 8);
-        memcpy(tmp+8, src+stride*8, 8);
+    uint8_t tmp[16];
+    const int l1= stride;
+    const int l2= stride + l1;
+    const int l3= stride + l2;
+    const int l4= (int)tmp - (int)src - stride*3;
+    const int l5= (int)tmp - (int)src - stride*3 + 8;
+    const int l6= stride*3 + l3;
+    const int l7= stride + l6;
+    const int l8= stride + l7;
+
+    memcpy(tmp, src+stride*7, 8);
+    memcpy(tmp+8, src+stride*8, 8);
 */
-        src+= stride*4;
-        asm volatile(
+    src+= stride*4;
+    __asm__ volatile(
 
-#if 0 //sligtly more accurate and slightly slower
-                "pxor %%mm7, %%mm7                      \n\t" // 0
-                "lea (%0, %1), %%"REG_a"                \n\t"
-                "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
+#if 0 //slightly more accurate and slightly slower
+        "pxor %%mm7, %%mm7                      \n\t" // 0
+        "lea (%0, %1), %%"REG_a"                \n\t"
+        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
 //      0       1       2       3       4       5       6       7
 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1
 
 
-                "movq (%0, %1, 2), %%mm0                \n\t" // l2
-                "movq (%0), %%mm1                       \n\t" // l0
-                "movq %%mm0, %%mm2                      \n\t" // l2
-                PAVGB(%%mm7, %%mm0)                           // ~l2/2
-                PAVGB(%%mm1, %%mm0)                           // ~(l2 + 2l0)/4
-                PAVGB(%%mm2, %%mm0)                           // ~(5l2 + 2l0)/8
-
-                "movq (%%"REG_a"), %%mm1                \n\t" // l1
-                "movq (%%"REG_a", %1, 2), %%mm3         \n\t" // l3
-                "movq %%mm1, %%mm4                      \n\t" // l1
-                PAVGB(%%mm7, %%mm1)                           // ~l1/2
-                PAVGB(%%mm3, %%mm1)                           // ~(l1 + 2l3)/4
-                PAVGB(%%mm4, %%mm1)                           // ~(5l1 + 2l3)/8
-
-                "movq %%mm0, %%mm4                      \n\t" // ~(5l2 + 2l0)/8
-                "psubusb %%mm1, %%mm0                   \n\t"
-                "psubusb %%mm4, %%mm1                   \n\t"
-                "por %%mm0, %%mm1                       \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8
+        "movq (%0, %1, 2), %%mm0                \n\t" // l2
+        "movq (%0), %%mm1                       \n\t" // l0
+        "movq %%mm0, %%mm2                      \n\t" // l2
+        PAVGB(%%mm7, %%mm0)                           // ~l2/2
+        PAVGB(%%mm1, %%mm0)                           // ~(l2 + 2l0)/4
+        PAVGB(%%mm2, %%mm0)                           // ~(5l2 + 2l0)/8
+
+        "movq (%%"REG_a"), %%mm1                \n\t" // l1
+        "movq (%%"REG_a", %1, 2), %%mm3         \n\t" // l3
+        "movq %%mm1, %%mm4                      \n\t" // l1
+        PAVGB(%%mm7, %%mm1)                           // ~l1/2
+        PAVGB(%%mm3, %%mm1)                           // ~(l1 + 2l3)/4
+        PAVGB(%%mm4, %%mm1)                           // ~(5l1 + 2l3)/8
+
+        "movq %%mm0, %%mm4                      \n\t" // ~(5l2 + 2l0)/8
+        "psubusb %%mm1, %%mm0                   \n\t"
+        "psubusb %%mm4, %%mm1                   \n\t"
+        "por %%mm0, %%mm1                       \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8
 // mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0
 
-                "movq (%0, %1, 4), %%mm0                \n\t" // l4
-                "movq %%mm0, %%mm4                      \n\t" // l4
-                PAVGB(%%mm7, %%mm0)                           // ~l4/2
-                PAVGB(%%mm2, %%mm0)                           // ~(l4 + 2l2)/4
-                PAVGB(%%mm4, %%mm0)                           // ~(5l4 + 2l2)/8
-
-                "movq (%%"REG_c"), %%mm2                \n\t" // l5
-                "movq %%mm3, %%mm5                      \n\t" // l3
-                PAVGB(%%mm7, %%mm3)                           // ~l3/2
-                PAVGB(%%mm2, %%mm3)                           // ~(l3 + 2l5)/4
-                PAVGB(%%mm5, %%mm3)                           // ~(5l3 + 2l5)/8
-
-                "movq %%mm0, %%mm6                      \n\t" // ~(5l4 + 2l2)/8
-                "psubusb %%mm3, %%mm0                   \n\t"
-                "psubusb %%mm6, %%mm3                   \n\t"
-                "por %%mm0, %%mm3                       \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8
-                "pcmpeqb %%mm7, %%mm0                   \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)
+        "movq (%0, %1, 4), %%mm0                \n\t" // l4
+        "movq %%mm0, %%mm4                      \n\t" // l4
+        PAVGB(%%mm7, %%mm0)                           // ~l4/2
+        PAVGB(%%mm2, %%mm0)                           // ~(l4 + 2l2)/4
+        PAVGB(%%mm4, %%mm0)                           // ~(5l4 + 2l2)/8
+
+        "movq (%%"REG_c"), %%mm2                \n\t" // l5
+        "movq %%mm3, %%mm5                      \n\t" // l3
+        PAVGB(%%mm7, %%mm3)                           // ~l3/2
+        PAVGB(%%mm2, %%mm3)                           // ~(l3 + 2l5)/4
+        PAVGB(%%mm5, %%mm3)                           // ~(5l3 + 2l5)/8
+
+        "movq %%mm0, %%mm6                      \n\t" // ~(5l4 + 2l2)/8
+        "psubusb %%mm3, %%mm0                   \n\t"
+        "psubusb %%mm6, %%mm3                   \n\t"
+        "por %%mm0, %%mm3                       \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8
+        "pcmpeqb %%mm7, %%mm0                   \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)
 // mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0
 
-                "movq (%%"REG_c", %1), %%mm6            \n\t" // l6
-                "movq %%mm6, %%mm5                      \n\t" // l6
-                PAVGB(%%mm7, %%mm6)                           // ~l6/2
-                PAVGB(%%mm4, %%mm6)                           // ~(l6 + 2l4)/4
-                PAVGB(%%mm5, %%mm6)                           // ~(5l6 + 2l4)/8
-
-                "movq (%%"REG_c", %1, 2), %%mm5         \n\t" // l7
-                "movq %%mm2, %%mm4                      \n\t" // l5
-                PAVGB(%%mm7, %%mm2)                           // ~l5/2
-                PAVGB(%%mm5, %%mm2)                           // ~(l5 + 2l7)/4
-                PAVGB(%%mm4, %%mm2)                           // ~(5l5 + 2l7)/8
-
-                "movq %%mm6, %%mm4                      \n\t" // ~(5l6 + 2l4)/8
-                "psubusb %%mm2, %%mm6                   \n\t"
-                "psubusb %%mm4, %%mm2                   \n\t"
-                "por %%mm6, %%mm2                       \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8
+        "movq (%%"REG_c", %1), %%mm6            \n\t" // l6
+        "movq %%mm6, %%mm5                      \n\t" // l6
+        PAVGB(%%mm7, %%mm6)                           // ~l6/2
+        PAVGB(%%mm4, %%mm6)                           // ~(l6 + 2l4)/4
+        PAVGB(%%mm5, %%mm6)                           // ~(5l6 + 2l4)/8
+
+        "movq (%%"REG_c", %1, 2), %%mm5         \n\t" // l7
+        "movq %%mm2, %%mm4                      \n\t" // l5
+        PAVGB(%%mm7, %%mm2)                           // ~l5/2
+        PAVGB(%%mm5, %%mm2)                           // ~(l5 + 2l7)/4
+        PAVGB(%%mm4, %%mm2)                           // ~(5l5 + 2l7)/8
+
+        "movq %%mm6, %%mm4                      \n\t" // ~(5l6 + 2l4)/8
+        "psubusb %%mm2, %%mm6                   \n\t"
+        "psubusb %%mm4, %%mm2                   \n\t"
+        "por %%mm6, %%mm2                       \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8
 // mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0
 
 
-                PMINUB(%%mm2, %%mm1, %%mm4)                   // MIN(|lenergy|,|renergy|)/8
-                "movq %2, %%mm4                         \n\t" // QP //FIXME QP+1 ?
-                "paddusb "MANGLE(b01)", %%mm4           \n\t"
-                "pcmpgtb %%mm3, %%mm4                   \n\t" // |menergy|/8 < QP
-                "psubusb %%mm1, %%mm3                   \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8
-                "pand %%mm4, %%mm3                      \n\t"
-
-                "movq %%mm3, %%mm1                      \n\t"
-//                "psubusb "MANGLE(b01)", %%mm3           \n\t"
-                PAVGB(%%mm7, %%mm3)
-                PAVGB(%%mm7, %%mm3)
-                "paddusb %%mm1, %%mm3                   \n\t"
-//                "paddusb "MANGLE(b01)", %%mm3           \n\t"
-
-                "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //l3
-                "movq (%0, %1, 4), %%mm5                \n\t" //l4
-                "movq (%0, %1, 4), %%mm4                \n\t" //l4
-                "psubusb %%mm6, %%mm5                   \n\t"
-                "psubusb %%mm4, %%mm6                   \n\t"
-                "por %%mm6, %%mm5                       \n\t" // |l3-l4|
-                "pcmpeqb %%mm7, %%mm6                   \n\t" // SIGN(l3-l4)
-                "pxor %%mm6, %%mm0                      \n\t"
-                "pand %%mm0, %%mm3                      \n\t"
-                PMINUB(%%mm5, %%mm3, %%mm0)
-
-                "psubusb "MANGLE(b01)", %%mm3           \n\t"
-                PAVGB(%%mm7, %%mm3)
-
-                "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
-                "movq (%0, %1, 4), %%mm2                \n\t"
-                "pxor %%mm6, %%mm0                      \n\t"
-                "pxor %%mm6, %%mm2                      \n\t"
-                "psubb %%mm3, %%mm0                     \n\t"
-                "paddb %%mm3, %%mm2                     \n\t"
-                "pxor %%mm6, %%mm0                      \n\t"
-                "pxor %%mm6, %%mm2                      \n\t"
-                "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
-                "movq %%mm2, (%0, %1, 4)                \n\t"
+        PMINUB(%%mm2, %%mm1, %%mm4)                   // MIN(|lenergy|,|renergy|)/8
+        "movq %2, %%mm4                         \n\t" // QP //FIXME QP+1 ?
+        "paddusb "MANGLE(b01)", %%mm4           \n\t"
+        "pcmpgtb %%mm3, %%mm4                   \n\t" // |menergy|/8 < QP
+        "psubusb %%mm1, %%mm3                   \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8
+        "pand %%mm4, %%mm3                      \n\t"
+
+        "movq %%mm3, %%mm1                      \n\t"
+//        "psubusb "MANGLE(b01)", %%mm3           \n\t"
+        PAVGB(%%mm7, %%mm3)
+        PAVGB(%%mm7, %%mm3)
+        "paddusb %%mm1, %%mm3                   \n\t"
+//        "paddusb "MANGLE(b01)", %%mm3           \n\t"
+
+        "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //l3
+        "movq (%0, %1, 4), %%mm5                \n\t" //l4
+        "movq (%0, %1, 4), %%mm4                \n\t" //l4
+        "psubusb %%mm6, %%mm5                   \n\t"
+        "psubusb %%mm4, %%mm6                   \n\t"
+        "por %%mm6, %%mm5                       \n\t" // |l3-l4|
+        "pcmpeqb %%mm7, %%mm6                   \n\t" // SIGN(l3-l4)
+        "pxor %%mm6, %%mm0                      \n\t"
+        "pand %%mm0, %%mm3                      \n\t"
+        PMINUB(%%mm5, %%mm3, %%mm0)
+
+        "psubusb "MANGLE(b01)", %%mm3           \n\t"
+        PAVGB(%%mm7, %%mm3)
+
+        "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
+        "movq (%0, %1, 4), %%mm2                \n\t"
+        "pxor %%mm6, %%mm0                      \n\t"
+        "pxor %%mm6, %%mm2                      \n\t"
+        "psubb %%mm3, %%mm0                     \n\t"
+        "paddb %%mm3, %%mm2                     \n\t"
+        "pxor %%mm6, %%mm0                      \n\t"
+        "pxor %%mm6, %%mm2                      \n\t"
+        "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
+        "movq %%mm2, (%0, %1, 4)                \n\t"
 #endif //0
 
-                "lea (%0, %1), %%"REG_a"                \n\t"
-                "pcmpeqb %%mm6, %%mm6                   \n\t" // -1
+        "lea (%0, %1), %%"REG_a"                \n\t"
+        "pcmpeqb %%mm6, %%mm6                   \n\t" // -1
 //      0       1       2       3       4       5       6       7
 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1
 
 
-                "movq (%%"REG_a", %1, 2), %%mm1         \n\t" // l3
-                "movq (%0, %1, 4), %%mm0                \n\t" // l4
-                "pxor %%mm6, %%mm1                      \n\t" // -l3-1
-                PAVGB(%%mm1, %%mm0)                           // -q+128 = (l4-l3+256)/2
+        "movq (%%"REG_a", %1, 2), %%mm1         \n\t" // l3
+        "movq (%0, %1, 4), %%mm0                \n\t" // l4
+        "pxor %%mm6, %%mm1                      \n\t" // -l3-1
+        PAVGB(%%mm1, %%mm0)                           // -q+128 = (l4-l3+256)/2
 // mm1=-l3-1, mm0=128-q
 
-                "movq (%%"REG_a", %1, 4), %%mm2         \n\t" // l5
-                "movq (%%"REG_a", %1), %%mm3            \n\t" // l2
-                "pxor %%mm6, %%mm2                      \n\t" // -l5-1
-                "movq %%mm2, %%mm5                      \n\t" // -l5-1
-                "movq "MANGLE(b80)", %%mm4              \n\t" // 128
-                "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
-                PAVGB(%%mm3, %%mm2)                           // (l2-l5+256)/2
-                PAVGB(%%mm0, %%mm4)                           // ~(l4-l3)/4 + 128
-                PAVGB(%%mm2, %%mm4)                           // ~(l2-l5)/4 +(l4-l3)/8 + 128
-                PAVGB(%%mm0, %%mm4)                           // ~(l2-l5)/8 +5(l4-l3)/16 + 128
+        "movq (%%"REG_a", %1, 4), %%mm2         \n\t" // l5
+        "movq (%%"REG_a", %1), %%mm3            \n\t" // l2
+        "pxor %%mm6, %%mm2                      \n\t" // -l5-1
+        "movq %%mm2, %%mm5                      \n\t" // -l5-1
+        "movq "MANGLE(b80)", %%mm4              \n\t" // 128
+        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
+        PAVGB(%%mm3, %%mm2)                           // (l2-l5+256)/2
+        PAVGB(%%mm0, %%mm4)                           // ~(l4-l3)/4 + 128
+        PAVGB(%%mm2, %%mm4)                           // ~(l2-l5)/4 +(l4-l3)/8 + 128
+        PAVGB(%%mm0, %%mm4)                           // ~(l2-l5)/8 +5(l4-l3)/16 + 128
 // mm1=-l3-1, mm0=128-q, mm3=l2, mm4=menergy/16 + 128, mm5= -l5-1
 
-                "movq (%%"REG_a"), %%mm2                \n\t" // l1
-                "pxor %%mm6, %%mm2                      \n\t" // -l1-1
-                PAVGB(%%mm3, %%mm2)                           // (l2-l1+256)/2
-                PAVGB((%0), %%mm1)                            // (l0-l3+256)/2
-                "movq "MANGLE(b80)", %%mm3              \n\t" // 128
-                PAVGB(%%mm2, %%mm3)                           // ~(l2-l1)/4 + 128
-                PAVGB(%%mm1, %%mm3)                           // ~(l0-l3)/4 +(l2-l1)/8 + 128
-                PAVGB(%%mm2, %%mm3)                           // ~(l0-l3)/8 +5(l2-l1)/16 + 128
+        "movq (%%"REG_a"), %%mm2                \n\t" // l1
+        "pxor %%mm6, %%mm2                      \n\t" // -l1-1
+        PAVGB(%%mm3, %%mm2)                           // (l2-l1+256)/2
+        PAVGB((%0), %%mm1)                            // (l0-l3+256)/2
+        "movq "MANGLE(b80)", %%mm3              \n\t" // 128
+        PAVGB(%%mm2, %%mm3)                           // ~(l2-l1)/4 + 128
+        PAVGB(%%mm1, %%mm3)                           // ~(l0-l3)/4 +(l2-l1)/8 + 128
+        PAVGB(%%mm2, %%mm3)                           // ~(l0-l3)/8 +5(l2-l1)/16 + 128
 // mm0=128-q, mm3=lenergy/16 + 128, mm4= menergy/16 + 128, mm5= -l5-1
 
-                PAVGB((%%REGc, %1), %%mm5)                    // (l6-l5+256)/2
-                "movq (%%"REG_c", %1, 2), %%mm1         \n\t" // l7
-                "pxor %%mm6, %%mm1                      \n\t" // -l7-1
-                PAVGB((%0, %1, 4), %%mm1)                     // (l4-l7+256)/2
-                "movq "MANGLE(b80)", %%mm2              \n\t" // 128
-                PAVGB(%%mm5, %%mm2)                           // ~(l6-l5)/4 + 128
-                PAVGB(%%mm1, %%mm2)                           // ~(l4-l7)/4 +(l6-l5)/8 + 128
-                PAVGB(%%mm5, %%mm2)                           // ~(l4-l7)/8 +5(l6-l5)/16 + 128
+        PAVGB((%%REGc, %1), %%mm5)                    // (l6-l5+256)/2
+        "movq (%%"REG_c", %1, 2), %%mm1         \n\t" // l7
+        "pxor %%mm6, %%mm1                      \n\t" // -l7-1
+        PAVGB((%0, %1, 4), %%mm1)                     // (l4-l7+256)/2
+        "movq "MANGLE(b80)", %%mm2              \n\t" // 128
+        PAVGB(%%mm5, %%mm2)                           // ~(l6-l5)/4 + 128
+        PAVGB(%%mm1, %%mm2)                           // ~(l4-l7)/4 +(l6-l5)/8 + 128
+        PAVGB(%%mm5, %%mm2)                           // ~(l4-l7)/8 +5(l6-l5)/16 + 128
 // mm0=128-q, mm2=renergy/16 + 128, mm3=lenergy/16 + 128, mm4= menergy/16 + 128
 
-                "movq "MANGLE(b00)", %%mm1              \n\t" // 0
-                "movq "MANGLE(b00)", %%mm5              \n\t" // 0
-                "psubb %%mm2, %%mm1                     \n\t" // 128 - renergy/16
-                "psubb %%mm3, %%mm5                     \n\t" // 128 - lenergy/16
-                PMAXUB(%%mm1, %%mm2)                          // 128 + |renergy/16|
-                 PMAXUB(%%mm5, %%mm3)                         // 128 + |lenergy/16|
-                PMINUB(%%mm2, %%mm3, %%mm1)                   // 128 + MIN(|lenergy|,|renergy|)/16
+        "movq "MANGLE(b00)", %%mm1              \n\t" // 0
+        "movq "MANGLE(b00)", %%mm5              \n\t" // 0
+        "psubb %%mm2, %%mm1                     \n\t" // 128 - renergy/16
+        "psubb %%mm3, %%mm5                     \n\t" // 128 - lenergy/16
+        PMAXUB(%%mm1, %%mm2)                          // 128 + |renergy/16|
+        PMAXUB(%%mm5, %%mm3)                          // 128 + |lenergy/16|
+        PMINUB(%%mm2, %%mm3, %%mm1)                   // 128 + MIN(|lenergy|,|renergy|)/16
 
 // mm0=128-q, mm3=128 + MIN(|lenergy|,|renergy|)/16, mm4= menergy/16 + 128
 
-                "movq "MANGLE(b00)", %%mm7              \n\t" // 0
-                "movq %2, %%mm2                         \n\t" // QP
-                PAVGB(%%mm6, %%mm2)                           // 128 + QP/2
-                "psubb %%mm6, %%mm2                     \n\t"
-
-                "movq %%mm4, %%mm1                      \n\t"
-                "pcmpgtb %%mm7, %%mm1                   \n\t" // SIGN(menergy)
-                "pxor %%mm1, %%mm4                      \n\t"
-                "psubb %%mm1, %%mm4                     \n\t" // 128 + |menergy|/16
-                "pcmpgtb %%mm4, %%mm2                   \n\t" // |menergy|/16 < QP/2
-                "psubusb %%mm3, %%mm4                   \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16
+        "movq "MANGLE(b00)", %%mm7              \n\t" // 0
+        "movq %2, %%mm2                         \n\t" // QP
+        PAVGB(%%mm6, %%mm2)                           // 128 + QP/2
+        "psubb %%mm6, %%mm2                     \n\t"
+
+        "movq %%mm4, %%mm1                      \n\t"
+        "pcmpgtb %%mm7, %%mm1                   \n\t" // SIGN(menergy)
+        "pxor %%mm1, %%mm4                      \n\t"
+        "psubb %%mm1, %%mm4                     \n\t" // 128 + |menergy|/16
+        "pcmpgtb %%mm4, %%mm2                   \n\t" // |menergy|/16 < QP/2
+        "psubusb %%mm3, %%mm4                   \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16
 // mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16
 
-                "movq %%mm4, %%mm3                      \n\t" // d
-                "psubusb "MANGLE(b01)", %%mm4           \n\t"
-                PAVGB(%%mm7, %%mm4)                           // d/32
-                PAVGB(%%mm7, %%mm4)                           // (d + 32)/64
-                "paddb %%mm3, %%mm4                     \n\t" // 5d/64
-                "pand %%mm2, %%mm4                      \n\t"
-
-                "movq "MANGLE(b80)", %%mm5              \n\t" // 128
-                "psubb %%mm0, %%mm5                     \n\t" // q
-                "paddsb %%mm6, %%mm5                    \n\t" // fix bad rounding
-                "pcmpgtb %%mm5, %%mm7                   \n\t" // SIGN(q)
-                "pxor %%mm7, %%mm5                      \n\t"
-
-                PMINUB(%%mm5, %%mm4, %%mm3)                   // MIN(|q|, 5d/64)
-                "pxor %%mm1, %%mm7                      \n\t" // SIGN(d*q)
-
-                "pand %%mm7, %%mm4                      \n\t"
-                "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
-                "movq (%0, %1, 4), %%mm2                \n\t"
-                "pxor %%mm1, %%mm0                      \n\t"
-                "pxor %%mm1, %%mm2                      \n\t"
-                "paddb %%mm4, %%mm0                     \n\t"
-                "psubb %%mm4, %%mm2                     \n\t"
-                "pxor %%mm1, %%mm0                      \n\t"
-                "pxor %%mm1, %%mm2                      \n\t"
-                "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
-                "movq %%mm2, (%0, %1, 4)                \n\t"
-
-                :
-                : "r" (src), "r" ((long)stride), "m" (c->pQPb)
-                : "%"REG_a, "%"REG_c
-        );
+        "movq %%mm4, %%mm3                      \n\t" // d
+        "psubusb "MANGLE(b01)", %%mm4           \n\t"
+        PAVGB(%%mm7, %%mm4)                           // d/32
+        PAVGB(%%mm7, %%mm4)                           // (d + 32)/64
+        "paddb %%mm3, %%mm4                     \n\t" // 5d/64
+        "pand %%mm2, %%mm4                      \n\t"
+
+        "movq "MANGLE(b80)", %%mm5              \n\t" // 128
+        "psubb %%mm0, %%mm5                     \n\t" // q
+        "paddsb %%mm6, %%mm5                    \n\t" // fix bad rounding
+        "pcmpgtb %%mm5, %%mm7                   \n\t" // SIGN(q)
+        "pxor %%mm7, %%mm5                      \n\t"
+
+        PMINUB(%%mm5, %%mm4, %%mm3)                   // MIN(|q|, 5d/64)
+        "pxor %%mm1, %%mm7                      \n\t" // SIGN(d*q)
+
+        "pand %%mm7, %%mm4                      \n\t"
+        "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
+        "movq (%0, %1, 4), %%mm2                \n\t"
+        "pxor %%mm1, %%mm0                      \n\t"
+        "pxor %%mm1, %%mm2                      \n\t"
+        "paddb %%mm4, %%mm0                     \n\t"
+        "psubb %%mm4, %%mm2                     \n\t"
+        "pxor %%mm1, %%mm0                      \n\t"
+        "pxor %%mm1, %%mm2                      \n\t"
+        "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
+        "movq %%mm2, (%0, %1, 4)                \n\t"
+
+        :
+        : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
+        : "%"REG_a, "%"REG_c
+    );
 
 /*
-        {
-        int x;
-        src-= stride;
-        for(x=0; x<BLOCK_SIZE; x++)
-        {
-                const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
-                if(FFABS(middleEnergy)< 8*QP)
-                {
-                        const int q=(src[l4] - src[l5])/2;
-                        const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
-                        const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
-
-                        int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
-                        d= FFMAX(d, 0);
-
-                        d= (5*d + 32) >> 6;
-                        d*= FFSIGN(-middleEnergy);
-
-                        if(q>0)
-                        {
-                                d= d<0 ? 0 : d;
-                                d= d>q ? q : d;
-                        }
-                        else
-                        {
-                                d= d>0 ? 0 : d;
-                                d= d<q ? q : d;
-                        }
-
-                        src[l4]-= d;
-                        src[l5]+= d;
-                }
-                src++;
+    {
+    int x;
+    src-= stride;
+    for(x=0; x<BLOCK_SIZE; x++){
+        const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
+        if(FFABS(middleEnergy)< 8*QP){
+            const int q=(src[l4] - src[l5])/2;
+            const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
+            const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
+
+            int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
+            d= FFMAX(d, 0);
+
+            d= (5*d + 32) >> 6;
+            d*= FFSIGN(-middleEnergy);
+
+            if(q>0){
+                d= d<0 ? 0 : d;
+                d= d>q ? q : d;
+            }else{
+                d= d>0 ? 0 : d;
+                d= d<q ? q : d;
+            }
+
+            src[l4]-= d;
+            src[l5]+= d;
         }
-src-=8;
-        for(x=0; x<8; x++)
-        {
-                int y;
-                for(y=4; y<6; y++)
-                {
-                        int d= src[x+y*stride] - tmp[x+(y-4)*8];
-                        int ad= FFABS(d);
-                        static int max=0;
-                        static int sum=0;
-                        static int num=0;
-                        static int bias=0;
-
-                        if(max<ad) max=ad;
-                        sum+= ad>3 ? 1 : 0;
-                        if(ad>3)
-                        {
-                                src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
-                        }
-                        if(y==4) bias+=d;
-                        num++;
-                        if(num%1000000 == 0)
-                        {
-                                av_log(c, AV_LOG_INFO, " %d %d %d %d\n", num, sum, max, bias);
-                        }
-                }
+        src++;
+    }
+    src-=8;
+    for(x=0; x<8; x++){
+        int y;
+        for(y=4; y<6; y++){
+            int d= src[x+y*stride] - tmp[x+(y-4)*8];
+            int ad= FFABS(d);
+            static int max=0;
+            static int sum=0;
+            static int num=0;
+            static int bias=0;
+
+            if(max<ad) max=ad;
+            sum+= ad>3 ? 1 : 0;
+            if(ad>3){
+                src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
+            }
+            if(y==4) bias+=d;
+            num++;
+            if(num%1000000 == 0){
+                av_log(c, AV_LOG_INFO, " %d %d %d %d\n", num, sum, max, bias);
+            }
         }
+    }
 }
 */
-#elif defined (HAVE_MMX)
-        src+= stride*4;
-        asm volatile(
-                "pxor %%mm7, %%mm7                      \n\t"
-                "lea -40(%%"REG_SP"), %%"REG_c"         \n\t" // make space for 4 8-byte vars
-                "and "ALIGN_MASK", %%"REG_c"            \n\t" // align
+#elif HAVE_MMX
+    src+= stride*4;
+    __asm__ volatile(
+        "pxor %%mm7, %%mm7                      \n\t"
+        "lea -40(%%"REG_SP"), %%"REG_c"         \n\t" // make space for 4 8-byte vars
+        "and "ALIGN_MASK", %%"REG_c"            \n\t" // align
 //      0       1       2       3       4       5       6       7
 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 edx+%1  edx+2%1
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1
 
-                "movq (%0), %%mm0                       \n\t"
-                "movq %%mm0, %%mm1                      \n\t"
-                "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
-                "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
-
-                "movq (%0, %1), %%mm2                   \n\t"
-                "lea (%0, %1, 2), %%"REG_a"             \n\t"
-                "movq %%mm2, %%mm3                      \n\t"
-                "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
-                "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
-
-                "movq (%%"REG_a"), %%mm4                \n\t"
-                "movq %%mm4, %%mm5                      \n\t"
-                "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
-                "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
-
-                "paddw %%mm0, %%mm0                     \n\t" // 2L0
-                "paddw %%mm1, %%mm1                     \n\t" // 2H0
-                "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
-                "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
-                "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
-                "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
-
-                "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
-                "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
-                "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
-                "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
-
-                "movq (%%"REG_a", %1), %%mm2            \n\t"
-                "movq %%mm2, %%mm3                      \n\t"
-                "punpcklbw %%mm7, %%mm2                 \n\t" // L3
-                "punpckhbw %%mm7, %%mm3                 \n\t" // H3
-
-                "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
-                "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
-                "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
-                "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
-                "movq %%mm0, (%%"REG_c")                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
-                "movq %%mm1, 8(%%"REG_c")               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
-
-                "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
-                "movq %%mm0, %%mm1                      \n\t"
-                "punpcklbw %%mm7, %%mm0                 \n\t" // L4
-                "punpckhbw %%mm7, %%mm1                 \n\t" // H4
-
-                "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
-                "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
-                "movq %%mm2, 16(%%"REG_c")              \n\t" // L3 - L4
-                "movq %%mm3, 24(%%"REG_c")              \n\t" // H3 - H4
-                "paddw %%mm4, %%mm4                     \n\t" // 2L2
-                "paddw %%mm5, %%mm5                     \n\t" // 2H2
-                "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
-                "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
-
-                "lea (%%"REG_a", %1), %0                \n\t"
-                "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
-                "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
-                "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
-                "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
+        "movq (%0), %%mm0                       \n\t"
+        "movq %%mm0, %%mm1                      \n\t"
+        "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
+        "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
+
+        "movq (%0, %1), %%mm2                   \n\t"
+        "lea (%0, %1, 2), %%"REG_a"             \n\t"
+        "movq %%mm2, %%mm3                      \n\t"
+        "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
+        "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
+
+        "movq (%%"REG_a"), %%mm4                \n\t"
+        "movq %%mm4, %%mm5                      \n\t"
+        "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
+        "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
+
+        "paddw %%mm0, %%mm0                     \n\t" // 2L0
+        "paddw %%mm1, %%mm1                     \n\t" // 2H0
+        "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
+        "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
+        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
+        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
+
+        "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
+        "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
+        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
+        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
+
+        "movq (%%"REG_a", %1), %%mm2            \n\t"
+        "movq %%mm2, %%mm3                      \n\t"
+        "punpcklbw %%mm7, %%mm2                 \n\t" // L3
+        "punpckhbw %%mm7, %%mm3                 \n\t" // H3
+
+        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
+        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
+        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
+        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
+        "movq %%mm0, (%%"REG_c")                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
+        "movq %%mm1, 8(%%"REG_c")               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
+
+        "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
+        "movq %%mm0, %%mm1                      \n\t"
+        "punpcklbw %%mm7, %%mm0                 \n\t" // L4
+        "punpckhbw %%mm7, %%mm1                 \n\t" // H4
+
+        "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
+        "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
+        "movq %%mm2, 16(%%"REG_c")              \n\t" // L3 - L4
+        "movq %%mm3, 24(%%"REG_c")              \n\t" // H3 - H4
+        "paddw %%mm4, %%mm4                     \n\t" // 2L2
+        "paddw %%mm5, %%mm5                     \n\t" // 2H2
+        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
+        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
+
+        "lea (%%"REG_a", %1), %0                \n\t"
+        "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
+        "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
+        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
+        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
 //50 opcodes so far
-                "movq (%0, %1, 2), %%mm2                \n\t"
-                "movq %%mm2, %%mm3                      \n\t"
-                "punpcklbw %%mm7, %%mm2                 \n\t" // L5
-                "punpckhbw %%mm7, %%mm3                 \n\t" // H5
-                "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
-                "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
-                "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
-                "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
-
-                "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
-                "punpcklbw %%mm7, %%mm6                 \n\t" // L6
-                "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
-                "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
-                "punpckhbw %%mm7, %%mm6                 \n\t" // H6
-                "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
-
-                "paddw %%mm0, %%mm0                     \n\t" // 2L4
-                "paddw %%mm1, %%mm1                     \n\t" // 2H4
-                "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
-                "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
-
-                "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
-                "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
-                "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
-                "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
-
-                "movq (%0, %1, 4), %%mm2                \n\t"
-                "movq %%mm2, %%mm3                      \n\t"
-                "punpcklbw %%mm7, %%mm2                 \n\t" // L7
-                "punpckhbw %%mm7, %%mm3                 \n\t" // H7
-
-                "paddw %%mm2, %%mm2                     \n\t" // 2L7
-                "paddw %%mm3, %%mm3                     \n\t" // 2H7
-                "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
-                "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
-
-                "movq (%%"REG_c"), %%mm2                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
-                "movq 8(%%"REG_c"), %%mm3               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
-
-#ifdef HAVE_MMX2
-                "movq %%mm7, %%mm6                      \n\t" // 0
-                "psubw %%mm0, %%mm6                     \n\t"
-                "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
-                "movq %%mm7, %%mm6                      \n\t" // 0
-                "psubw %%mm1, %%mm6                     \n\t"
-                "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
-                "movq %%mm7, %%mm6                      \n\t" // 0
-                "psubw %%mm2, %%mm6                     \n\t"
-                "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
-                "movq %%mm7, %%mm6                      \n\t" // 0
-                "psubw %%mm3, %%mm6                     \n\t"
-                "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
+        "movq (%0, %1, 2), %%mm2                \n\t"
+        "movq %%mm2, %%mm3                      \n\t"
+        "punpcklbw %%mm7, %%mm2                 \n\t" // L5
+        "punpckhbw %%mm7, %%mm3                 \n\t" // H5
+        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
+        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
+        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
+        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
+
+        "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
+        "punpcklbw %%mm7, %%mm6                 \n\t" // L6
+        "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
+        "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
+        "punpckhbw %%mm7, %%mm6                 \n\t" // H6
+        "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
+
+        "paddw %%mm0, %%mm0                     \n\t" // 2L4
+        "paddw %%mm1, %%mm1                     \n\t" // 2H4
+        "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
+        "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
+
+        "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
+        "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
+        "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
+        "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
+
+        "movq (%0, %1, 4), %%mm2                \n\t"
+        "movq %%mm2, %%mm3                      \n\t"
+        "punpcklbw %%mm7, %%mm2                 \n\t" // L7
+        "punpckhbw %%mm7, %%mm3                 \n\t" // H7
+
+        "paddw %%mm2, %%mm2                     \n\t" // 2L7
+        "paddw %%mm3, %%mm3                     \n\t" // 2H7
+        "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
+        "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
+
+        "movq (%%"REG_c"), %%mm2                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
+        "movq 8(%%"REG_c"), %%mm3               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
+
+#if HAVE_MMX2
+        "movq %%mm7, %%mm6                      \n\t" // 0
+        "psubw %%mm0, %%mm6                     \n\t"
+        "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
+        "movq %%mm7, %%mm6                      \n\t" // 0
+        "psubw %%mm1, %%mm6                     \n\t"
+        "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
+        "movq %%mm7, %%mm6                      \n\t" // 0
+        "psubw %%mm2, %%mm6                     \n\t"
+        "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
+        "movq %%mm7, %%mm6                      \n\t" // 0
+        "psubw %%mm3, %%mm6                     \n\t"
+        "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
 #else
-                "movq %%mm7, %%mm6                      \n\t" // 0
-                "pcmpgtw %%mm0, %%mm6                   \n\t"
-                "pxor %%mm6, %%mm0                      \n\t"
-                "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
-                "movq %%mm7, %%mm6                      \n\t" // 0
-                "pcmpgtw %%mm1, %%mm6                   \n\t"
-                "pxor %%mm6, %%mm1                      \n\t"
-                "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
-                "movq %%mm7, %%mm6                      \n\t" // 0
-                "pcmpgtw %%mm2, %%mm6                   \n\t"
-                "pxor %%mm6, %%mm2                      \n\t"
-                "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
-                "movq %%mm7, %%mm6                      \n\t" // 0
-                "pcmpgtw %%mm3, %%mm6                   \n\t"
-                "pxor %%mm6, %%mm3                      \n\t"
-                "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
+        "movq %%mm7, %%mm6                      \n\t" // 0
+        "pcmpgtw %%mm0, %%mm6                   \n\t"
+        "pxor %%mm6, %%mm0                      \n\t"
+        "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
+        "movq %%mm7, %%mm6                      \n\t" // 0
+        "pcmpgtw %%mm1, %%mm6                   \n\t"
+        "pxor %%mm6, %%mm1                      \n\t"
+        "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
+        "movq %%mm7, %%mm6                      \n\t" // 0
+        "pcmpgtw %%mm2, %%mm6                   \n\t"
+        "pxor %%mm6, %%mm2                      \n\t"
+        "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
+        "movq %%mm7, %%mm6                      \n\t" // 0
+        "pcmpgtw %%mm3, %%mm6                   \n\t"
+        "pxor %%mm6, %%mm3                      \n\t"
+        "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
 #endif
 
-#ifdef HAVE_MMX2
-                "pminsw %%mm2, %%mm0                    \n\t"
-                "pminsw %%mm3, %%mm1                    \n\t"
+#if HAVE_MMX2
+        "pminsw %%mm2, %%mm0                    \n\t"
+        "pminsw %%mm3, %%mm1                    \n\t"
 #else
-                "movq %%mm0, %%mm6                      \n\t"
-                "psubusw %%mm2, %%mm6                   \n\t"
-                "psubw %%mm6, %%mm0                     \n\t"
-                "movq %%mm1, %%mm6                      \n\t"
-                "psubusw %%mm3, %%mm6                   \n\t"
-                "psubw %%mm6, %%mm1                     \n\t"
+        "movq %%mm0, %%mm6                      \n\t"
+        "psubusw %%mm2, %%mm6                   \n\t"
+        "psubw %%mm6, %%mm0                     \n\t"
+        "movq %%mm1, %%mm6                      \n\t"
+        "psubusw %%mm3, %%mm6                   \n\t"
+        "psubw %%mm6, %%mm1                     \n\t"
 #endif
 
-                "movd %2, %%mm2                         \n\t" // QP
-                "punpcklbw %%mm7, %%mm2                 \n\t"
+        "movd %2, %%mm2                         \n\t" // QP
+        "punpcklbw %%mm7, %%mm2                 \n\t"
 
-                "movq %%mm7, %%mm6                      \n\t" // 0
-                "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
-                "pxor %%mm6, %%mm4                      \n\t"
-                "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
-                "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
-                "pxor %%mm7, %%mm5                      \n\t"
-                "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
+        "movq %%mm7, %%mm6                      \n\t" // 0
+        "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
+        "pxor %%mm6, %%mm4                      \n\t"
+        "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
+        "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
+        "pxor %%mm7, %%mm5                      \n\t"
+        "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
 // 100 opcodes
-                "psllw $3, %%mm2                        \n\t" // 8QP
-                "movq %%mm2, %%mm3                      \n\t" // 8QP
-                "pcmpgtw %%mm4, %%mm2                   \n\t"
-                "pcmpgtw %%mm5, %%mm3                   \n\t"
-                "pand %%mm2, %%mm4                      \n\t"
-                "pand %%mm3, %%mm5                      \n\t"
-
-
-                "psubusw %%mm0, %%mm4                   \n\t" // hd
-                "psubusw %%mm1, %%mm5                   \n\t" // ld
-
-
-                "movq "MANGLE(w05)", %%mm2              \n\t" // 5
-                "pmullw %%mm2, %%mm4                    \n\t"
-                "pmullw %%mm2, %%mm5                    \n\t"
-                "movq "MANGLE(w20)", %%mm2              \n\t" // 32
-                "paddw %%mm2, %%mm4                     \n\t"
-                "paddw %%mm2, %%mm5                     \n\t"
-                "psrlw $6, %%mm4                        \n\t"
-                "psrlw $6, %%mm5                        \n\t"
-
-                "movq 16(%%"REG_c"), %%mm0              \n\t" // L3 - L4
-                "movq 24(%%"REG_c"), %%mm1              \n\t" // H3 - H4
-
-                "pxor %%mm2, %%mm2                      \n\t"
-                "pxor %%mm3, %%mm3                      \n\t"
-
-                "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
-                "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
-                "pxor %%mm2, %%mm0                      \n\t"
-                "pxor %%mm3, %%mm1                      \n\t"
-                "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
-                "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
-                "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
-                "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
-
-                "pxor %%mm6, %%mm2                      \n\t"
-                "pxor %%mm7, %%mm3                      \n\t"
-                "pand %%mm2, %%mm4                      \n\t"
-                "pand %%mm3, %%mm5                      \n\t"
-
-#ifdef HAVE_MMX2
-                "pminsw %%mm0, %%mm4                    \n\t"
-                "pminsw %%mm1, %%mm5                    \n\t"
+        "psllw $3, %%mm2                        \n\t" // 8QP
+        "movq %%mm2, %%mm3                      \n\t" // 8QP
+        "pcmpgtw %%mm4, %%mm2                   \n\t"
+        "pcmpgtw %%mm5, %%mm3                   \n\t"
+        "pand %%mm2, %%mm4                      \n\t"
+        "pand %%mm3, %%mm5                      \n\t"
+
+
+        "psubusw %%mm0, %%mm4                   \n\t" // hd
+        "psubusw %%mm1, %%mm5                   \n\t" // ld
+
+
+        "movq "MANGLE(w05)", %%mm2              \n\t" // 5
+        "pmullw %%mm2, %%mm4                    \n\t"
+        "pmullw %%mm2, %%mm5                    \n\t"
+        "movq "MANGLE(w20)", %%mm2              \n\t" // 32
+        "paddw %%mm2, %%mm4                     \n\t"
+        "paddw %%mm2, %%mm5                     \n\t"
+        "psrlw $6, %%mm4                        \n\t"
+        "psrlw $6, %%mm5                        \n\t"
+
+        "movq 16(%%"REG_c"), %%mm0              \n\t" // L3 - L4
+        "movq 24(%%"REG_c"), %%mm1              \n\t" // H3 - H4
+
+        "pxor %%mm2, %%mm2                      \n\t"
+        "pxor %%mm3, %%mm3                      \n\t"
+
+        "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
+        "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
+        "pxor %%mm2, %%mm0                      \n\t"
+        "pxor %%mm3, %%mm1                      \n\t"
+        "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
+        "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
+        "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
+        "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
+
+        "pxor %%mm6, %%mm2                      \n\t"
+        "pxor %%mm7, %%mm3                      \n\t"
+        "pand %%mm2, %%mm4                      \n\t"
+        "pand %%mm3, %%mm5                      \n\t"
+
+#if HAVE_MMX2
+        "pminsw %%mm0, %%mm4                    \n\t"
+        "pminsw %%mm1, %%mm5                    \n\t"
 #else
-                "movq %%mm4, %%mm2                      \n\t"
-                "psubusw %%mm0, %%mm2                   \n\t"
-                "psubw %%mm2, %%mm4                     \n\t"
-                "movq %%mm5, %%mm2                      \n\t"
-                "psubusw %%mm1, %%mm2                   \n\t"
-                "psubw %%mm2, %%mm5                     \n\t"
+        "movq %%mm4, %%mm2                      \n\t"
+        "psubusw %%mm0, %%mm2                   \n\t"
+        "psubw %%mm2, %%mm4                     \n\t"
+        "movq %%mm5, %%mm2                      \n\t"
+        "psubusw %%mm1, %%mm2                   \n\t"
+        "psubw %%mm2, %%mm5                     \n\t"
 #endif
-                "pxor %%mm6, %%mm4                      \n\t"
-                "pxor %%mm7, %%mm5                      \n\t"
-                "psubw %%mm6, %%mm4                     \n\t"
-                "psubw %%mm7, %%mm5                     \n\t"
-                "packsswb %%mm5, %%mm4                  \n\t"
-                "movq (%0), %%mm0                       \n\t"
-                "paddb   %%mm4, %%mm0                   \n\t"
-                "movq %%mm0, (%0)                       \n\t"
-                "movq (%0, %1), %%mm0                   \n\t"
-                "psubb %%mm4, %%mm0                     \n\t"
-                "movq %%mm0, (%0, %1)                   \n\t"
-
-                : "+r" (src)
-                : "r" ((long)stride), "m" (c->pQPb)
-                : "%"REG_a, "%"REG_c
-        );
-#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        const int l1= stride;
-        const int l2= stride + l1;
-        const int l3= stride + l2;
-        const int l4= stride + l3;
-        const int l5= stride + l4;
-        const int l6= stride + l5;
-        const int l7= stride + l6;
-        const int l8= stride + l7;
-//        const int l9= stride + l8;
-        int x;
-        src+= stride*3;
-        for(x=0; x<BLOCK_SIZE; x++)
-        {
-                const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
-                if(FFABS(middleEnergy) < 8*c->QP)
-                {
-                        const int q=(src[l4] - src[l5])/2;
-                        const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
-                        const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
-
-                        int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
-                        d= FFMAX(d, 0);
-
-                        d= (5*d + 32) >> 6;
-                        d*= FFSIGN(-middleEnergy);
-
-                        if(q>0)
-                        {
-                                d= d<0 ? 0 : d;
-                                d= d>q ? q : d;
-                        }
-                        else
-                        {
-                                d= d>0 ? 0 : d;
-                                d= d<q ? q : d;
-                        }
-
-                        src[l4]-= d;
-                        src[l5]+= d;
-                }
-                src++;
+        "pxor %%mm6, %%mm4                      \n\t"
+        "pxor %%mm7, %%mm5                      \n\t"
+        "psubw %%mm6, %%mm4                     \n\t"
+        "psubw %%mm7, %%mm5                     \n\t"
+        "packsswb %%mm5, %%mm4                  \n\t"
+        "movq (%0), %%mm0                       \n\t"
+        "paddb   %%mm4, %%mm0                   \n\t"
+        "movq %%mm0, (%0)                       \n\t"
+        "movq (%0, %1), %%mm0                   \n\t"
+        "psubb %%mm4, %%mm0                     \n\t"
+        "movq %%mm0, (%0, %1)                   \n\t"
+
+        : "+r" (src)
+        : "r" ((x86_reg)stride), "m" (c->pQPb)
+        : "%"REG_a, "%"REG_c
+    );
+#else //HAVE_MMX2 || HAVE_AMD3DNOW
+    const int l1= stride;
+    const int l2= stride + l1;
+    const int l3= stride + l2;
+    const int l4= stride + l3;
+    const int l5= stride + l4;
+    const int l6= stride + l5;
+    const int l7= stride + l6;
+    const int l8= stride + l7;
+//    const int l9= stride + l8;
+    int x;
+    src+= stride*3;
+    for(x=0; x<BLOCK_SIZE; x++){
+        const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
+        if(FFABS(middleEnergy) < 8*c->QP){
+            const int q=(src[l4] - src[l5])/2;
+            const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
+            const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
+
+            int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
+            d= FFMAX(d, 0);
+
+            d= (5*d + 32) >> 6;
+            d*= FFSIGN(-middleEnergy);
+
+            if(q>0){
+                d= d<0 ? 0 : d;
+                d= d>q ? q : d;
+            }else{
+                d= d>0 ? 0 : d;
+                d= d<q ? q : d;
+            }
+
+            src[l4]-= d;
+            src[l5]+= d;
         }
-#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
+        src++;
+    }
+#endif //HAVE_MMX2 || HAVE_AMD3DNOW
 }
 #endif //HAVE_ALTIVEC
 
-#ifndef HAVE_ALTIVEC
+#if !HAVE_ALTIVEC
 static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
 {
-#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        asm volatile(
-                "pxor %%mm6, %%mm6                      \n\t"
-                "pcmpeqb %%mm7, %%mm7                   \n\t"
-                "movq %2, %%mm0                         \n\t"
-                "punpcklbw %%mm6, %%mm0                 \n\t"
-                "psrlw $1, %%mm0                        \n\t"
-                "psubw %%mm7, %%mm0                     \n\t"
-                "packuswb %%mm0, %%mm0                  \n\t"
-                "movq %%mm0, %3                         \n\t"
-
-                "lea (%0, %1), %%"REG_a"                \n\t"
-                "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
+#if HAVE_MMX2 || HAVE_AMD3DNOW
+    __asm__ volatile(
+        "pxor %%mm6, %%mm6                      \n\t"
+        "pcmpeqb %%mm7, %%mm7                   \n\t"
+        "movq %2, %%mm0                         \n\t"
+        "punpcklbw %%mm6, %%mm0                 \n\t"
+        "psrlw $1, %%mm0                        \n\t"
+        "psubw %%mm7, %%mm0                     \n\t"
+        "packuswb %%mm0, %%mm0                  \n\t"
+        "movq %%mm0, %3                         \n\t"
+
+        "lea (%0, %1), %%"REG_a"                \n\t"
+        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
 
 //        0        1        2        3        4        5        6        7        8        9
 //        %0        eax        eax+%1        eax+2%1        %0+4%1        edx        edx+%1        edx+2%1        %0+8%1        edx+4%1
 
+#undef REAL_FIND_MIN_MAX
 #undef FIND_MIN_MAX
-#ifdef HAVE_MMX2
+#if HAVE_MMX2
 #define REAL_FIND_MIN_MAX(addr)\
-                "movq " #addr ", %%mm0                  \n\t"\
-                "pminub %%mm0, %%mm7                    \n\t"\
-                "pmaxub %%mm0, %%mm6                    \n\t"
+        "movq " #addr ", %%mm0                  \n\t"\
+        "pminub %%mm0, %%mm7                    \n\t"\
+        "pmaxub %%mm0, %%mm6                    \n\t"
 #else
 #define REAL_FIND_MIN_MAX(addr)\
-                "movq " #addr ", %%mm0                  \n\t"\
-                "movq %%mm7, %%mm1                      \n\t"\
-                "psubusb %%mm0, %%mm6                   \n\t"\
-                "paddb %%mm0, %%mm6                     \n\t"\
-                "psubusb %%mm0, %%mm1                   \n\t"\
-                "psubb %%mm1, %%mm7                     \n\t"
+        "movq " #addr ", %%mm0                  \n\t"\
+        "movq %%mm7, %%mm1                      \n\t"\
+        "psubusb %%mm0, %%mm6                   \n\t"\
+        "paddb %%mm0, %%mm6                     \n\t"\
+        "psubusb %%mm0, %%mm1                   \n\t"\
+        "psubb %%mm1, %%mm7                     \n\t"
 #endif
 #define FIND_MIN_MAX(addr)  REAL_FIND_MIN_MAX(addr)
 
@@ -1228,155 +1087,155 @@ FIND_MIN_MAX((%%REGd, %1))
 FIND_MIN_MAX((%%REGd, %1, 2))
 FIND_MIN_MAX((%0, %1, 8))
 
-                "movq %%mm7, %%mm4                      \n\t"
-                "psrlq $8, %%mm7                        \n\t"
-#ifdef HAVE_MMX2
-                "pminub %%mm4, %%mm7                    \n\t" // min of pixels
-                "pshufw $0xF9, %%mm7, %%mm4             \n\t"
-                "pminub %%mm4, %%mm7                    \n\t" // min of pixels
-                "pshufw $0xFE, %%mm7, %%mm4             \n\t"
-                "pminub %%mm4, %%mm7                    \n\t"
+        "movq %%mm7, %%mm4                      \n\t"
+        "psrlq $8, %%mm7                        \n\t"
+#if HAVE_MMX2
+        "pminub %%mm4, %%mm7                    \n\t" // min of pixels
+        "pshufw $0xF9, %%mm7, %%mm4             \n\t"
+        "pminub %%mm4, %%mm7                    \n\t" // min of pixels
+        "pshufw $0xFE, %%mm7, %%mm4             \n\t"
+        "pminub %%mm4, %%mm7                    \n\t"
 #else
-                "movq %%mm7, %%mm1                      \n\t"
-                "psubusb %%mm4, %%mm1                   \n\t"
-                "psubb %%mm1, %%mm7                     \n\t"
-                "movq %%mm7, %%mm4                      \n\t"
-                "psrlq $16, %%mm7                       \n\t"
-                "movq %%mm7, %%mm1                      \n\t"
-                "psubusb %%mm4, %%mm1                   \n\t"
-                "psubb %%mm1, %%mm7                     \n\t"
-                "movq %%mm7, %%mm4                      \n\t"
-                "psrlq $32, %%mm7                       \n\t"
-                "movq %%mm7, %%mm1                      \n\t"
-                "psubusb %%mm4, %%mm1                   \n\t"
-                "psubb %%mm1, %%mm7                     \n\t"
+        "movq %%mm7, %%mm1                      \n\t"
+        "psubusb %%mm4, %%mm1                   \n\t"
+        "psubb %%mm1, %%mm7                     \n\t"
+        "movq %%mm7, %%mm4                      \n\t"
+        "psrlq $16, %%mm7                       \n\t"
+        "movq %%mm7, %%mm1                      \n\t"
+        "psubusb %%mm4, %%mm1                   \n\t"
+        "psubb %%mm1, %%mm7                     \n\t"
+        "movq %%mm7, %%mm4                      \n\t"
+        "psrlq $32, %%mm7                       \n\t"
+        "movq %%mm7, %%mm1                      \n\t"
+        "psubusb %%mm4, %%mm1                   \n\t"
+        "psubb %%mm1, %%mm7                     \n\t"
 #endif
 
 
-                "movq %%mm6, %%mm4                      \n\t"
-                "psrlq $8, %%mm6                        \n\t"
-#ifdef HAVE_MMX2
-                "pmaxub %%mm4, %%mm6                    \n\t" // max of pixels
-                "pshufw $0xF9, %%mm6, %%mm4             \n\t"
-                "pmaxub %%mm4, %%mm6                    \n\t"
-                "pshufw $0xFE, %%mm6, %%mm4             \n\t"
-                "pmaxub %%mm4, %%mm6                    \n\t"
+        "movq %%mm6, %%mm4                      \n\t"
+        "psrlq $8, %%mm6                        \n\t"
+#if HAVE_MMX2
+        "pmaxub %%mm4, %%mm6                    \n\t" // max of pixels
+        "pshufw $0xF9, %%mm6, %%mm4             \n\t"
+        "pmaxub %%mm4, %%mm6                    \n\t"
+        "pshufw $0xFE, %%mm6, %%mm4             \n\t"
+        "pmaxub %%mm4, %%mm6                    \n\t"
 #else
-                "psubusb %%mm4, %%mm6                   \n\t"
-                "paddb %%mm4, %%mm6                     \n\t"
-                "movq %%mm6, %%mm4                      \n\t"
-                "psrlq $16, %%mm6                       \n\t"
-                "psubusb %%mm4, %%mm6                   \n\t"
-                "paddb %%mm4, %%mm6                     \n\t"
-                "movq %%mm6, %%mm4                      \n\t"
-                "psrlq $32, %%mm6                       \n\t"
-                "psubusb %%mm4, %%mm6                   \n\t"
-                "paddb %%mm4, %%mm6                     \n\t"
+        "psubusb %%mm4, %%mm6                   \n\t"
+        "paddb %%mm4, %%mm6                     \n\t"
+        "movq %%mm6, %%mm4                      \n\t"
+        "psrlq $16, %%mm6                       \n\t"
+        "psubusb %%mm4, %%mm6                   \n\t"
+        "paddb %%mm4, %%mm6                     \n\t"
+        "movq %%mm6, %%mm4                      \n\t"
+        "psrlq $32, %%mm6                       \n\t"
+        "psubusb %%mm4, %%mm6                   \n\t"
+        "paddb %%mm4, %%mm6                     \n\t"
 #endif
-                "movq %%mm6, %%mm0                      \n\t" // max
-                "psubb %%mm7, %%mm6                     \n\t" // max - min
-                "movd %%mm6, %%ecx                      \n\t"
-                "cmpb "MANGLE(deringThreshold)", %%cl   \n\t"
-                " jb 1f                                 \n\t"
-                "lea -24(%%"REG_SP"), %%"REG_c"         \n\t"
-                "and "ALIGN_MASK", %%"REG_c"            \n\t"
-                PAVGB(%%mm0, %%mm7)                           // a=(max + min)/2
-                "punpcklbw %%mm7, %%mm7                 \n\t"
-                "punpcklbw %%mm7, %%mm7                 \n\t"
-                "punpcklbw %%mm7, %%mm7                 \n\t"
-                "movq %%mm7, (%%"REG_c")                \n\t"
-
-                "movq (%0), %%mm0                       \n\t" // L10
-                "movq %%mm0, %%mm1                      \n\t" // L10
-                "movq %%mm0, %%mm2                      \n\t" // L10
-                "psllq $8, %%mm1                        \n\t"
-                "psrlq $8, %%mm2                        \n\t"
-                "movd -4(%0), %%mm3                     \n\t"
-                "movd 8(%0), %%mm4                      \n\t"
-                "psrlq $24, %%mm3                       \n\t"
-                "psllq $56, %%mm4                       \n\t"
-                "por %%mm3, %%mm1                       \n\t" // L00
-                "por %%mm4, %%mm2                       \n\t" // L20
-                "movq %%mm1, %%mm3                      \n\t" // L00
-                PAVGB(%%mm2, %%mm1)                           // (L20 + L00)/2
-                PAVGB(%%mm0, %%mm1)                           // (L20 + L00 + 2L10)/4
-                "psubusb %%mm7, %%mm0                   \n\t"
-                "psubusb %%mm7, %%mm2                   \n\t"
-                "psubusb %%mm7, %%mm3                   \n\t"
-                "pcmpeqb "MANGLE(b00)", %%mm0           \n\t" // L10 > a ? 0 : -1
-                "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L20 > a ? 0 : -1
-                "pcmpeqb "MANGLE(b00)", %%mm3           \n\t" // L00 > a ? 0 : -1
-                "paddb %%mm2, %%mm0                     \n\t"
-                "paddb %%mm3, %%mm0                     \n\t"
-
-                "movq (%%"REG_a"), %%mm2                \n\t" // L11
-                "movq %%mm2, %%mm3                      \n\t" // L11
-                "movq %%mm2, %%mm4                      \n\t" // L11
-                "psllq $8, %%mm3                        \n\t"
-                "psrlq $8, %%mm4                        \n\t"
-                "movd -4(%%"REG_a"), %%mm5              \n\t"
-                "movd 8(%%"REG_a"), %%mm6               \n\t"
-                "psrlq $24, %%mm5                       \n\t"
-                "psllq $56, %%mm6                       \n\t"
-                "por %%mm5, %%mm3                       \n\t" // L01
-                "por %%mm6, %%mm4                       \n\t" // L21
-                "movq %%mm3, %%mm5                      \n\t" // L01
-                PAVGB(%%mm4, %%mm3)                           // (L21 + L01)/2
-                PAVGB(%%mm2, %%mm3)                           // (L21 + L01 + 2L11)/4
-                "psubusb %%mm7, %%mm2                   \n\t"
-                "psubusb %%mm7, %%mm4                   \n\t"
-                "psubusb %%mm7, %%mm5                   \n\t"
-                "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L11 > a ? 0 : -1
-                "pcmpeqb "MANGLE(b00)", %%mm4           \n\t" // L21 > a ? 0 : -1
-                "pcmpeqb "MANGLE(b00)", %%mm5           \n\t" // L01 > a ? 0 : -1
-                "paddb %%mm4, %%mm2                     \n\t"
-                "paddb %%mm5, %%mm2                     \n\t"
+        "movq %%mm6, %%mm0                      \n\t" // max
+        "psubb %%mm7, %%mm6                     \n\t" // max - min
+        "movd %%mm6, %%ecx                      \n\t"
+        "cmpb "MANGLE(deringThreshold)", %%cl   \n\t"
+        " jb 1f                                 \n\t"
+        "lea -24(%%"REG_SP"), %%"REG_c"         \n\t"
+        "and "ALIGN_MASK", %%"REG_c"            \n\t"
+        PAVGB(%%mm0, %%mm7)                           // a=(max + min)/2
+        "punpcklbw %%mm7, %%mm7                 \n\t"
+        "punpcklbw %%mm7, %%mm7                 \n\t"
+        "punpcklbw %%mm7, %%mm7                 \n\t"
+        "movq %%mm7, (%%"REG_c")                \n\t"
+
+        "movq (%0), %%mm0                       \n\t" // L10
+        "movq %%mm0, %%mm1                      \n\t" // L10
+        "movq %%mm0, %%mm2                      \n\t" // L10
+        "psllq $8, %%mm1                        \n\t"
+        "psrlq $8, %%mm2                        \n\t"
+        "movd -4(%0), %%mm3                     \n\t"
+        "movd 8(%0), %%mm4                      \n\t"
+        "psrlq $24, %%mm3                       \n\t"
+        "psllq $56, %%mm4                       \n\t"
+        "por %%mm3, %%mm1                       \n\t" // L00
+        "por %%mm4, %%mm2                       \n\t" // L20
+        "movq %%mm1, %%mm3                      \n\t" // L00
+        PAVGB(%%mm2, %%mm1)                           // (L20 + L00)/2
+        PAVGB(%%mm0, %%mm1)                           // (L20 + L00 + 2L10)/4
+        "psubusb %%mm7, %%mm0                   \n\t"
+        "psubusb %%mm7, %%mm2                   \n\t"
+        "psubusb %%mm7, %%mm3                   \n\t"
+        "pcmpeqb "MANGLE(b00)", %%mm0           \n\t" // L10 > a ? 0 : -1
+        "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L20 > a ? 0 : -1
+        "pcmpeqb "MANGLE(b00)", %%mm3           \n\t" // L00 > a ? 0 : -1
+        "paddb %%mm2, %%mm0                     \n\t"
+        "paddb %%mm3, %%mm0                     \n\t"
+
+        "movq (%%"REG_a"), %%mm2                \n\t" // L11
+        "movq %%mm2, %%mm3                      \n\t" // L11
+        "movq %%mm2, %%mm4                      \n\t" // L11
+        "psllq $8, %%mm3                        \n\t"
+        "psrlq $8, %%mm4                        \n\t"
+        "movd -4(%%"REG_a"), %%mm5              \n\t"
+        "movd 8(%%"REG_a"), %%mm6               \n\t"
+        "psrlq $24, %%mm5                       \n\t"
+        "psllq $56, %%mm6                       \n\t"
+        "por %%mm5, %%mm3                       \n\t" // L01
+        "por %%mm6, %%mm4                       \n\t" // L21
+        "movq %%mm3, %%mm5                      \n\t" // L01
+        PAVGB(%%mm4, %%mm3)                           // (L21 + L01)/2
+        PAVGB(%%mm2, %%mm3)                           // (L21 + L01 + 2L11)/4
+        "psubusb %%mm7, %%mm2                   \n\t"
+        "psubusb %%mm7, %%mm4                   \n\t"
+        "psubusb %%mm7, %%mm5                   \n\t"
+        "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L11 > a ? 0 : -1
+        "pcmpeqb "MANGLE(b00)", %%mm4           \n\t" // L21 > a ? 0 : -1
+        "pcmpeqb "MANGLE(b00)", %%mm5           \n\t" // L01 > a ? 0 : -1
+        "paddb %%mm4, %%mm2                     \n\t"
+        "paddb %%mm5, %%mm2                     \n\t"
 // 0, 2, 3, 1
 #define REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
-                "movq " #src ", " #sx "                 \n\t" /* src[0] */\
-                "movq " #sx ", " #lx "                  \n\t" /* src[0] */\
-                "movq " #sx ", " #t0 "                  \n\t" /* src[0] */\
-                "psllq $8, " #lx "                      \n\t"\
-                "psrlq $8, " #t0 "                      \n\t"\
-                "movd -4" #src ", " #t1 "               \n\t"\
-                "psrlq $24, " #t1 "                     \n\t"\
-                "por " #t1 ", " #lx "                   \n\t" /* src[-1] */\
-                "movd 8" #src ", " #t1 "                \n\t"\
-                "psllq $56, " #t1 "                     \n\t"\
-                "por " #t1 ", " #t0 "                   \n\t" /* src[+1] */\
-                "movq " #lx ", " #t1 "                  \n\t" /* src[-1] */\
-                PAVGB(t0, lx)                                 /* (src[-1] + src[+1])/2 */\
-                PAVGB(sx, lx)                                 /* (src[-1] + 2src[0] + src[+1])/4 */\
-                PAVGB(lx, pplx)                                     \
-                "movq " #lx ", 8(%%"REG_c")             \n\t"\
-                "movq (%%"REG_c"), " #lx "              \n\t"\
-                "psubusb " #lx ", " #t1 "               \n\t"\
-                "psubusb " #lx ", " #t0 "               \n\t"\
-                "psubusb " #lx ", " #sx "               \n\t"\
-                "movq "MANGLE(b00)", " #lx "            \n\t"\
-                "pcmpeqb " #lx ", " #t1 "               \n\t" /* src[-1] > a ? 0 : -1*/\
-                "pcmpeqb " #lx ", " #t0 "               \n\t" /* src[+1] > a ? 0 : -1*/\
-                "pcmpeqb " #lx ", " #sx "               \n\t" /* src[0]  > a ? 0 : -1*/\
-                "paddb " #t1 ", " #t0 "                 \n\t"\
-                "paddb " #t0 ", " #sx "                 \n\t"\
+        "movq " #src ", " #sx "                 \n\t" /* src[0] */\
+        "movq " #sx ", " #lx "                  \n\t" /* src[0] */\
+        "movq " #sx ", " #t0 "                  \n\t" /* src[0] */\
+        "psllq $8, " #lx "                      \n\t"\
+        "psrlq $8, " #t0 "                      \n\t"\
+        "movd -4" #src ", " #t1 "               \n\t"\
+        "psrlq $24, " #t1 "                     \n\t"\
+        "por " #t1 ", " #lx "                   \n\t" /* src[-1] */\
+        "movd 8" #src ", " #t1 "                \n\t"\
+        "psllq $56, " #t1 "                     \n\t"\
+        "por " #t1 ", " #t0 "                   \n\t" /* src[+1] */\
+        "movq " #lx ", " #t1 "                  \n\t" /* src[-1] */\
+        PAVGB(t0, lx)                                 /* (src[-1] + src[+1])/2 */\
+        PAVGB(sx, lx)                                 /* (src[-1] + 2src[0] + src[+1])/4 */\
+        PAVGB(lx, pplx)                                     \
+        "movq " #lx ", 8(%%"REG_c")             \n\t"\
+        "movq (%%"REG_c"), " #lx "              \n\t"\
+        "psubusb " #lx ", " #t1 "               \n\t"\
+        "psubusb " #lx ", " #t0 "               \n\t"\
+        "psubusb " #lx ", " #sx "               \n\t"\
+        "movq "MANGLE(b00)", " #lx "            \n\t"\
+        "pcmpeqb " #lx ", " #t1 "               \n\t" /* src[-1] > a ? 0 : -1*/\
+        "pcmpeqb " #lx ", " #t0 "               \n\t" /* src[+1] > a ? 0 : -1*/\
+        "pcmpeqb " #lx ", " #sx "               \n\t" /* src[0]  > a ? 0 : -1*/\
+        "paddb " #t1 ", " #t0 "                 \n\t"\
+        "paddb " #t0 ", " #sx "                 \n\t"\
 \
-                PAVGB(plx, pplx)                              /* filtered */\
-                "movq " #dst ", " #t0 "                 \n\t" /* dst */\
-                "movq " #t0 ", " #t1 "                  \n\t" /* dst */\
-                "psubusb %3, " #t0 "                    \n\t"\
-                "paddusb %3, " #t1 "                    \n\t"\
-                PMAXUB(t0, pplx)\
-                PMINUB(t1, pplx, t0)\
-                "paddb " #sx ", " #ppsx "               \n\t"\
-                "paddb " #psx ", " #ppsx "              \n\t"\
-                "#paddb "MANGLE(b02)", " #ppsx "        \n\t"\
-                "pand "MANGLE(b08)", " #ppsx "          \n\t"\
-                "pcmpeqb " #lx ", " #ppsx "             \n\t"\
-                "pand " #ppsx ", " #pplx "              \n\t"\
-                "pandn " #dst ", " #ppsx "              \n\t"\
-                "por " #pplx ", " #ppsx "               \n\t"\
-                "movq " #ppsx ", " #dst "               \n\t"\
-                "movq 8(%%"REG_c"), " #lx "             \n\t"
+        PAVGB(plx, pplx)                              /* filtered */\
+        "movq " #dst ", " #t0 "                 \n\t" /* dst */\
+        "movq " #t0 ", " #t1 "                  \n\t" /* dst */\
+        "psubusb %3, " #t0 "                    \n\t"\
+        "paddusb %3, " #t1 "                    \n\t"\
+        PMAXUB(t0, pplx)\
+        PMINUB(t1, pplx, t0)\
+        "paddb " #sx ", " #ppsx "               \n\t"\
+        "paddb " #psx ", " #ppsx "              \n\t"\
+        "#paddb "MANGLE(b02)", " #ppsx "        \n\t"\
+        "pand "MANGLE(b08)", " #ppsx "          \n\t"\
+        "pcmpeqb " #lx ", " #ppsx "             \n\t"\
+        "pand " #ppsx ", " #pplx "              \n\t"\
+        "pandn " #dst ", " #ppsx "              \n\t"\
+        "por " #pplx ", " #ppsx "               \n\t"\
+        "movq " #ppsx ", " #dst "               \n\t"\
+        "movq 8(%%"REG_c"), " #lx "             \n\t"
 
 #define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
    REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1)
@@ -1405,237 +1264,224 @@ DERING_CORE((%%REGd, %1)   ,(%%REGd, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,
 DERING_CORE((%%REGd, %1, 2),(%0, %1, 8)    ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
 DERING_CORE((%0, %1, 8)    ,(%%REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
 
-                "1:                        \n\t"
-                : : "r" (src), "r" ((long)stride), "m" (c->pQPb), "m"(c->pQPb2)
-                : "%"REG_a, "%"REG_d, "%"REG_c
-        );
-#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        int y;
-        int min=255;
-        int max=0;
-        int avg;
-        uint8_t *p;
-        int s[10];
-        const int QP2= c->QP/2 + 1;
-
-        for(y=1; y<9; y++)
-        {
-                int x;
-                p= src + stride*y;
-                for(x=1; x<9; x++)
-                {
-                        p++;
-                        if(*p > max) max= *p;
-                        if(*p < min) min= *p;
-                }
-        }
-        avg= (min + max + 1)>>1;
-
-        if(max - min <deringThreshold) return;
-
-        for(y=0; y<10; y++)
-        {
-                int t = 0;
-
-                if(src[stride*y + 0] > avg) t+= 1;
-                if(src[stride*y + 1] > avg) t+= 2;
-                if(src[stride*y + 2] > avg) t+= 4;
-                if(src[stride*y + 3] > avg) t+= 8;
-                if(src[stride*y + 4] > avg) t+= 16;
-                if(src[stride*y + 5] > avg) t+= 32;
-                if(src[stride*y + 6] > avg) t+= 64;
-                if(src[stride*y + 7] > avg) t+= 128;
-                if(src[stride*y + 8] > avg) t+= 256;
-                if(src[stride*y + 9] > avg) t+= 512;
-
-                t |= (~t)<<16;
-                t &= (t<<1) & (t>>1);
-                s[y] = t;
-        }
-
-        for(y=1; y<9; y++)
-        {
-                int t = s[y-1] & s[y] & s[y+1];
-                t|= t>>16;
-                s[y-1]= t;
+        "1:                        \n\t"
+        : : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb), "m"(c->pQPb2)
+        : "%"REG_a, "%"REG_d, "%"REG_c
+    );
+#else //HAVE_MMX2 || HAVE_AMD3DNOW
+    int y;
+    int min=255;
+    int max=0;
+    int avg;
+    uint8_t *p;
+    int s[10];
+    const int QP2= c->QP/2 + 1;
+
+    for(y=1; y<9; y++){
+        int x;
+        p= src + stride*y;
+        for(x=1; x<9; x++){
+            p++;
+            if(*p > max) max= *p;
+            if(*p < min) min= *p;
         }
+    }
+    avg= (min + max + 1)>>1;
+
+    if(max - min <deringThreshold) return;
+
+    for(y=0; y<10; y++){
+        int t = 0;
+
+        if(src[stride*y + 0] > avg) t+= 1;
+        if(src[stride*y + 1] > avg) t+= 2;
+        if(src[stride*y + 2] > avg) t+= 4;
+        if(src[stride*y + 3] > avg) t+= 8;
+        if(src[stride*y + 4] > avg) t+= 16;
+        if(src[stride*y + 5] > avg) t+= 32;
+        if(src[stride*y + 6] > avg) t+= 64;
+        if(src[stride*y + 7] > avg) t+= 128;
+        if(src[stride*y + 8] > avg) t+= 256;
+        if(src[stride*y + 9] > avg) t+= 512;
+
+        t |= (~t)<<16;
+        t &= (t<<1) & (t>>1);
+        s[y] = t;
+    }
+
+    for(y=1; y<9; y++){
+        int t = s[y-1] & s[y] & s[y+1];
+        t|= t>>16;
+        s[y-1]= t;
+    }
+
+    for(y=1; y<9; y++){
+        int x;
+        int t = s[y-1];
 
-        for(y=1; y<9; y++)
-        {
-                int x;
-                int t = s[y-1];
-
-                p= src + stride*y;
-                for(x=1; x<9; x++)
-                {
-                        p++;
-                        if(t & (1<<x))
-                        {
-                                int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))
-                                      +2*(*(p     -1)) + 4*(*p         ) + 2*(*(p     +1))
-                                      +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));
-                                f= (f + 8)>>4;
+        p= src + stride*y;
+        for(x=1; x<9; x++){
+            p++;
+            if(t & (1<<x)){
+                int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))
+                      +2*(*(p     -1)) + 4*(*p         ) + 2*(*(p     +1))
+                      +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));
+                f= (f + 8)>>4;
 
 #ifdef DEBUG_DERING_THRESHOLD
-                                asm volatile("emms\n\t":);
-                                {
-                                static long long numPixels=0;
-                                if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;
-//                                if((max-min)<20 || (max-min)*QP<200)
-//                                if((max-min)*QP < 500)
-//                                if(max-min<QP/2)
-                                if(max-min < 20)
-                                {
-                                        static int numSkiped=0;
-                                        static int errorSum=0;
-                                        static int worstQP=0;
-                                        static int worstRange=0;
-                                        static int worstDiff=0;
-                                        int diff= (f - *p);
-                                        int absDiff= FFABS(diff);
-                                        int error= diff*diff;
-
-                                        if(x==1 || x==8 || y==1 || y==8) continue;
-
-                                        numSkiped++;
-                                        if(absDiff > worstDiff)
-                                        {
-                                                worstDiff= absDiff;
-                                                worstQP= QP;
-                                                worstRange= max-min;
-                                        }
-                                        errorSum+= error;
-
-                                        if(1024LL*1024LL*1024LL % numSkiped == 0)
-                                        {
-                                                av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "
-                                                        "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
-                                                        (float)errorSum/numSkiped, numSkiped, worstQP, worstRange,
-                                                        worstDiff, (float)numSkiped/numPixels);
-                                        }
-                                }
-                                }
-#endif
-                                if     (*p + QP2 < f) *p= *p + QP2;
-                                else if(*p - QP2 > f) *p= *p - QP2;
-                                else *p=f;
+                    __asm__ volatile("emms\n\t":);
+                    {
+                    static long long numPixels=0;
+                    if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;
+//                    if((max-min)<20 || (max-min)*QP<200)
+//                    if((max-min)*QP < 500)
+//                    if(max-min<QP/2)
+                    if(max-min < 20){
+                        static int numSkipped=0;
+                        static int errorSum=0;
+                        static int worstQP=0;
+                        static int worstRange=0;
+                        static int worstDiff=0;
+                        int diff= (f - *p);
+                        int absDiff= FFABS(diff);
+                        int error= diff*diff;
+
+                        if(x==1 || x==8 || y==1 || y==8) continue;
+
+                        numSkipped++;
+                        if(absDiff > worstDiff){
+                            worstDiff= absDiff;
+                            worstQP= QP;
+                            worstRange= max-min;
                         }
-                }
+                        errorSum+= error;
+
+                        if(1024LL*1024LL*1024LL % numSkipped == 0){
+                            av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "
+                                   "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
+                                   (float)errorSum/numSkipped, numSkipped, worstQP, worstRange,
+                                   worstDiff, (float)numSkipped/numPixels);
+                        }
+                    }
+                    }
+#endif
+                    if     (*p + QP2 < f) *p= *p + QP2;
+                    else if(*p - QP2 > f) *p= *p - QP2;
+                    else *p=f;
+            }
         }
+    }
 #ifdef DEBUG_DERING_THRESHOLD
-        if(max-min < 20)
-        {
-                for(y=1; y<9; y++)
-                {
-                        int x;
-                        int t = 0;
-                        p= src + stride*y;
-                        for(x=1; x<9; x++)
-                        {
-                                p++;
-                                *p = FFMIN(*p + 20, 255);
-                        }
-                }
-//                src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
+    if(max-min < 20){
+        for(y=1; y<9; y++){
+            int x;
+            int t = 0;
+            p= src + stride*y;
+            for(x=1; x<9; x++){
+                p++;
+                *p = FFMIN(*p + 20, 255);
+            }
         }
+//        src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
+    }
 #endif
-#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
+#endif //HAVE_MMX2 || HAVE_AMD3DNOW
 }
 #endif //HAVE_ALTIVEC
 
 /**
- * Deinterlaces the given block by linearly interpolating every second line.
+ * Deinterlace the given block by linearly interpolating every second line.
  * will be called for every 8x8 block and can read & write from line 4-15
- * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
+ * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
  * lines 4-12 will be read into the deblocking filter and should be deinterlaced
  */
 static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)
 {
-#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        src+= 4*stride;
-        asm volatile(
-                "lea (%0, %1), %%"REG_a"                \n\t"
-                "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
+#if HAVE_MMX2 || HAVE_AMD3DNOW
+    src+= 4*stride;
+    __asm__ volatile(
+        "lea (%0, %1), %%"REG_a"                \n\t"
+        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
 //      0       1       2       3       4       5       6       7       8       9
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
 
-                "movq (%0), %%mm0                       \n\t"
-                "movq (%%"REG_a", %1), %%mm1            \n\t"
-                PAVGB(%%mm1, %%mm0)
-                "movq %%mm0, (%%"REG_a")                \n\t"
-                "movq (%0, %1, 4), %%mm0                \n\t"
-                PAVGB(%%mm0, %%mm1)
-                "movq %%mm1, (%%"REG_a", %1, 2)         \n\t"
-                "movq (%%"REG_c", %1), %%mm1            \n\t"
-                PAVGB(%%mm1, %%mm0)
-                "movq %%mm0, (%%"REG_c")                \n\t"
-                "movq (%0, %1, 8), %%mm0                \n\t"
-                PAVGB(%%mm0, %%mm1)
-                "movq %%mm1, (%%"REG_c", %1, 2)         \n\t"
-
-                : : "r" (src), "r" ((long)stride)
-                : "%"REG_a, "%"REG_c
-        );
+        "movq (%0), %%mm0                       \n\t"
+        "movq (%%"REG_a", %1), %%mm1            \n\t"
+        PAVGB(%%mm1, %%mm0)
+        "movq %%mm0, (%%"REG_a")                \n\t"
+        "movq (%0, %1, 4), %%mm0                \n\t"
+        PAVGB(%%mm0, %%mm1)
+        "movq %%mm1, (%%"REG_a", %1, 2)         \n\t"
+        "movq (%%"REG_c", %1), %%mm1            \n\t"
+        PAVGB(%%mm1, %%mm0)
+        "movq %%mm0, (%%"REG_c")                \n\t"
+        "movq (%0, %1, 8), %%mm0                \n\t"
+        PAVGB(%%mm0, %%mm1)
+        "movq %%mm1, (%%"REG_c", %1, 2)         \n\t"
+
+        : : "r" (src), "r" ((x86_reg)stride)
+        : "%"REG_a, "%"REG_c
+    );
 #else
-        int a, b, x;
-        src+= 4*stride;
-
-        for(x=0; x<2; x++){
-                a= *(uint32_t*)&src[stride*0];
-                b= *(uint32_t*)&src[stride*2];
-                *(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
-                a= *(uint32_t*)&src[stride*4];
-                *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
-                b= *(uint32_t*)&src[stride*6];
-                *(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
-                a= *(uint32_t*)&src[stride*8];
-                *(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
-                src += 4;
-        }
+    int a, b, x;
+    src+= 4*stride;
+
+    for(x=0; x<2; x++){
+        a= *(uint32_t*)&src[stride*0];
+        b= *(uint32_t*)&src[stride*2];
+        *(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
+        a= *(uint32_t*)&src[stride*4];
+        *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
+        b= *(uint32_t*)&src[stride*6];
+        *(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
+        a= *(uint32_t*)&src[stride*8];
+        *(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
+        src += 4;
+    }
 #endif
 }
 
 /**
- * Deinterlaces the given block by cubic interpolating every second line.
+ * Deinterlace the given block by cubic interpolating every second line.
  * will be called for every 8x8 block and can read & write from line 4-15
- * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
+ * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
  * lines 4-12 will be read into the deblocking filter and should be deinterlaced
  * this filter will read lines 3-15 and write 7-13
  */
 static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)
 {
-#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        src+= stride*3;
-        asm volatile(
-                "lea (%0, %1), %%"REG_a"                \n\t"
-                "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
-                "lea (%%"REG_d", %1, 4), %%"REG_c"      \n\t"
-                "add %1, %%"REG_c"                      \n\t"
-                "pxor %%mm7, %%mm7                      \n\t"
+#if HAVE_MMX2 || HAVE_AMD3DNOW
+    src+= stride*3;
+    __asm__ volatile(
+        "lea (%0, %1), %%"REG_a"                \n\t"
+        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
+        "lea (%%"REG_d", %1, 4), %%"REG_c"      \n\t"
+        "add %1, %%"REG_c"                      \n\t"
+        "pxor %%mm7, %%mm7                      \n\t"
 //      0       1       2       3       4       5       6       7       8       9       10
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
 
 #define REAL_DEINT_CUBIC(a,b,c,d,e)\
-                "movq " #a ", %%mm0                     \n\t"\
-                "movq " #b ", %%mm1                     \n\t"\
-                "movq " #d ", %%mm2                     \n\t"\
-                "movq " #e ", %%mm3                     \n\t"\
-                PAVGB(%%mm2, %%mm1)                             /* (b+d) /2 */\
-                PAVGB(%%mm3, %%mm0)                             /* a(a+e) /2 */\
-                "movq %%mm0, %%mm2                      \n\t"\
-                "punpcklbw %%mm7, %%mm0                 \n\t"\
-                "punpckhbw %%mm7, %%mm2                 \n\t"\
-                "movq %%mm1, %%mm3                      \n\t"\
-                "punpcklbw %%mm7, %%mm1                 \n\t"\
-                "punpckhbw %%mm7, %%mm3                 \n\t"\
-                "psubw %%mm1, %%mm0                     \n\t"   /* L(a+e - (b+d))/2 */\
-                "psubw %%mm3, %%mm2                     \n\t"   /* H(a+e - (b+d))/2 */\
-                "psraw $3, %%mm0                        \n\t"   /* L(a+e - (b+d))/16 */\
-                "psraw $3, %%mm2                        \n\t"   /* H(a+e - (b+d))/16 */\
-                "psubw %%mm0, %%mm1                     \n\t"   /* L(9b + 9d - a - e)/16 */\
-                "psubw %%mm2, %%mm3                     \n\t"   /* H(9b + 9d - a - e)/16 */\
-                "packuswb %%mm3, %%mm1                  \n\t"\
-                "movq %%mm1, " #c "                     \n\t"
+        "movq " #a ", %%mm0                     \n\t"\
+        "movq " #b ", %%mm1                     \n\t"\
+        "movq " #d ", %%mm2                     \n\t"\
+        "movq " #e ", %%mm3                     \n\t"\
+        PAVGB(%%mm2, %%mm1)                             /* (b+d) /2 */\
+        PAVGB(%%mm3, %%mm0)                             /* a(a+e) /2 */\
+        "movq %%mm0, %%mm2                      \n\t"\
+        "punpcklbw %%mm7, %%mm0                 \n\t"\
+        "punpckhbw %%mm7, %%mm2                 \n\t"\
+        "movq %%mm1, %%mm3                      \n\t"\
+        "punpcklbw %%mm7, %%mm1                 \n\t"\
+        "punpckhbw %%mm7, %%mm3                 \n\t"\
+        "psubw %%mm1, %%mm0                     \n\t"   /* L(a+e - (b+d))/2 */\
+        "psubw %%mm3, %%mm2                     \n\t"   /* H(a+e - (b+d))/2 */\
+        "psraw $3, %%mm0                        \n\t"   /* L(a+e - (b+d))/16 */\
+        "psraw $3, %%mm2                        \n\t"   /* H(a+e - (b+d))/16 */\
+        "psubw %%mm0, %%mm1                     \n\t"   /* L(9b + 9d - a - e)/16 */\
+        "psubw %%mm2, %%mm3                     \n\t"   /* H(9b + 9d - a - e)/16 */\
+        "packuswb %%mm3, %%mm1                  \n\t"\
+        "movq %%mm1, " #c "                     \n\t"
 #define DEINT_CUBIC(a,b,c,d,e)  REAL_DEINT_CUBIC(a,b,c,d,e)
 
 DEINT_CUBIC((%0)        , (%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd, %1))
@@ -1643,69 +1489,68 @@ DEINT_CUBIC((%%REGa, %1), (%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%0, %1,
 DEINT_CUBIC((%0, %1, 4) , (%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGc))
 DEINT_CUBIC((%%REGd, %1), (%0, %1, 8) , (%%REGd, %1, 4), (%%REGc)    , (%%REGc, %1, 2))
 
-                : : "r" (src), "r" ((long)stride)
-                : "%"REG_a, "%"REG_d, "%"REG_c
-        );
-#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        int x;
-        src+= stride*3;
-        for(x=0; x<8; x++)
-        {
-                src[stride*3] = CLIP((-src[0]        + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);
-                src[stride*5] = CLIP((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);
-                src[stride*7] = CLIP((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);
-                src[stride*9] = CLIP((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);
-                src++;
-        }
-#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
+        : : "r" (src), "r" ((x86_reg)stride)
+        : "%"REG_a, "%"REG_d, "%"REG_c
+    );
+#else //HAVE_MMX2 || HAVE_AMD3DNOW
+    int x;
+    src+= stride*3;
+    for(x=0; x<8; x++){
+        src[stride*3] = CLIP((-src[0]        + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);
+        src[stride*5] = CLIP((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);
+        src[stride*7] = CLIP((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);
+        src[stride*9] = CLIP((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);
+        src++;
+    }
+#endif //HAVE_MMX2 || HAVE_AMD3DNOW
 }
 
 /**
- * Deinterlaces the given block by filtering every second line with a (-1 4 2 4 -1) filter.
+ * Deinterlace the given block by filtering every second line with a (-1 4 2 4 -1) filter.
  * will be called for every 8x8 block and can read & write from line 4-15
- * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
+ * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
  * lines 4-12 will be read into the deblocking filter and should be deinterlaced
  * this filter will read lines 4-13 and write 5-11
  */
 static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)
 {
-#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        src+= stride*4;
-        asm volatile(
-                "lea (%0, %1), %%"REG_a"                \n\t"
-                "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
-                "pxor %%mm7, %%mm7                      \n\t"
-                "movq (%2), %%mm0                       \n\t"
+#if HAVE_MMX2 || HAVE_AMD3DNOW
+    src+= stride*4;
+    __asm__ volatile(
+        "lea (%0, %1), %%"REG_a"                \n\t"
+        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
+        "pxor %%mm7, %%mm7                      \n\t"
+        "movq (%2), %%mm0                       \n\t"
 //      0       1       2       3       4       5       6       7       8       9       10
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
 
 #define REAL_DEINT_FF(a,b,c,d)\
-                "movq " #a ", %%mm1                     \n\t"\
-                "movq " #b ", %%mm2                     \n\t"\
-                "movq " #c ", %%mm3                     \n\t"\
-                "movq " #d ", %%mm4                     \n\t"\
-                PAVGB(%%mm3, %%mm1)                          \
-                PAVGB(%%mm4, %%mm0)                          \
-                "movq %%mm0, %%mm3                      \n\t"\
-                "punpcklbw %%mm7, %%mm0                 \n\t"\
-                "punpckhbw %%mm7, %%mm3                 \n\t"\
-                "movq %%mm1, %%mm4                      \n\t"\
-                "punpcklbw %%mm7, %%mm1                 \n\t"\
-                "punpckhbw %%mm7, %%mm4                 \n\t"\
-                "psllw $2, %%mm1                        \n\t"\
-                "psllw $2, %%mm4                        \n\t"\
-                "psubw %%mm0, %%mm1                     \n\t"\
-                "psubw %%mm3, %%mm4                     \n\t"\
-                "movq %%mm2, %%mm5                      \n\t"\
-                "movq %%mm2, %%mm0                      \n\t"\
-                "punpcklbw %%mm7, %%mm2                 \n\t"\
-                "punpckhbw %%mm7, %%mm5                 \n\t"\
-                "paddw %%mm2, %%mm1                     \n\t"\
-                "paddw %%mm5, %%mm4                     \n\t"\
-                "psraw $2, %%mm1                        \n\t"\
-                "psraw $2, %%mm4                        \n\t"\
-                "packuswb %%mm4, %%mm1                  \n\t"\
-                "movq %%mm1, " #b "                     \n\t"\
+        "movq " #a ", %%mm1                     \n\t"\
+        "movq " #b ", %%mm2                     \n\t"\
+        "movq " #c ", %%mm3                     \n\t"\
+        "movq " #d ", %%mm4                     \n\t"\
+        PAVGB(%%mm3, %%mm1)                          \
+        PAVGB(%%mm4, %%mm0)                          \
+        "movq %%mm0, %%mm3                      \n\t"\
+        "punpcklbw %%mm7, %%mm0                 \n\t"\
+        "punpckhbw %%mm7, %%mm3                 \n\t"\
+        "movq %%mm1, %%mm4                      \n\t"\
+        "punpcklbw %%mm7, %%mm1                 \n\t"\
+        "punpckhbw %%mm7, %%mm4                 \n\t"\
+        "psllw $2, %%mm1                        \n\t"\
+        "psllw $2, %%mm4                        \n\t"\
+        "psubw %%mm0, %%mm1                     \n\t"\
+        "psubw %%mm3, %%mm4                     \n\t"\
+        "movq %%mm2, %%mm5                      \n\t"\
+        "movq %%mm2, %%mm0                      \n\t"\
+        "punpcklbw %%mm7, %%mm2                 \n\t"\
+        "punpckhbw %%mm7, %%mm5                 \n\t"\
+        "paddw %%mm2, %%mm1                     \n\t"\
+        "paddw %%mm5, %%mm4                     \n\t"\
+        "psraw $2, %%mm1                        \n\t"\
+        "psraw $2, %%mm4                        \n\t"\
+        "packuswb %%mm4, %%mm1                  \n\t"\
+        "movq %%mm1, " #b "                     \n\t"\
 
 #define DEINT_FF(a,b,c,d)  REAL_DEINT_FF(a,b,c,d)
 
@@ -1714,84 +1559,83 @@ DEINT_FF((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd)       )
 DEINT_FF((%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%%REGd, %1, 2))
 DEINT_FF((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))
 
-                "movq %%mm0, (%2)                       \n\t"
-                : : "r" (src), "r" ((long)stride), "r"(tmp)
-                : "%"REG_a, "%"REG_d
-        );
-#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        int x;
-        src+= stride*4;
-        for(x=0; x<8; x++)
-        {
-                int t1= tmp[x];
-                int t2= src[stride*1];
-
-                src[stride*1]= CLIP((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);
-                t1= src[stride*4];
-                src[stride*3]= CLIP((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);
-                t2= src[stride*6];
-                src[stride*5]= CLIP((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);
-                t1= src[stride*8];
-                src[stride*7]= CLIP((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);
-                tmp[x]= t1;
-
-                src++;
-        }
-#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
+        "movq %%mm0, (%2)                       \n\t"
+        : : "r" (src), "r" ((x86_reg)stride), "r"(tmp)
+        : "%"REG_a, "%"REG_d
+    );
+#else //HAVE_MMX2 || HAVE_AMD3DNOW
+    int x;
+    src+= stride*4;
+    for(x=0; x<8; x++){
+        int t1= tmp[x];
+        int t2= src[stride*1];
+
+        src[stride*1]= CLIP((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);
+        t1= src[stride*4];
+        src[stride*3]= CLIP((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);
+        t2= src[stride*6];
+        src[stride*5]= CLIP((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);
+        t1= src[stride*8];
+        src[stride*7]= CLIP((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);
+        tmp[x]= t1;
+
+        src++;
+    }
+#endif //HAVE_MMX2 || HAVE_AMD3DNOW
 }
 
 /**
- * Deinterlaces the given block by filtering every line with a (-1 2 6 2 -1) filter.
+ * Deinterlace the given block by filtering every line with a (-1 2 6 2 -1) filter.
  * will be called for every 8x8 block and can read & write from line 4-15
- * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
+ * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
  * lines 4-12 will be read into the deblocking filter and should be deinterlaced
  * this filter will read lines 4-13 and write 4-11
  */
 static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)
 {
-#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        src+= stride*4;
-        asm volatile(
-                "lea (%0, %1), %%"REG_a"                \n\t"
-                "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
-                "pxor %%mm7, %%mm7                      \n\t"
-                "movq (%2), %%mm0                       \n\t"
-                "movq (%3), %%mm1                       \n\t"
+#if HAVE_MMX2 || HAVE_AMD3DNOW
+    src+= stride*4;
+    __asm__ volatile(
+        "lea (%0, %1), %%"REG_a"                \n\t"
+        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
+        "pxor %%mm7, %%mm7                      \n\t"
+        "movq (%2), %%mm0                       \n\t"
+        "movq (%3), %%mm1                       \n\t"
 //      0       1       2       3       4       5       6       7       8       9       10
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
 
 #define REAL_DEINT_L5(t1,t2,a,b,c)\
-                "movq " #a ", %%mm2                     \n\t"\
-                "movq " #b ", %%mm3                     \n\t"\
-                "movq " #c ", %%mm4                     \n\t"\
-                PAVGB(t2, %%mm3)                             \
-                PAVGB(t1, %%mm4)                             \
-                "movq %%mm2, %%mm5                      \n\t"\
-                "movq %%mm2, " #t1 "                    \n\t"\
-                "punpcklbw %%mm7, %%mm2                 \n\t"\
-                "punpckhbw %%mm7, %%mm5                 \n\t"\
-                "movq %%mm2, %%mm6                      \n\t"\
-                "paddw %%mm2, %%mm2                     \n\t"\
-                "paddw %%mm6, %%mm2                     \n\t"\
-                "movq %%mm5, %%mm6                      \n\t"\
-                "paddw %%mm5, %%mm5                     \n\t"\
-                "paddw %%mm6, %%mm5                     \n\t"\
-                "movq %%mm3, %%mm6                      \n\t"\
-                "punpcklbw %%mm7, %%mm3                 \n\t"\
-                "punpckhbw %%mm7, %%mm6                 \n\t"\
-                "paddw %%mm3, %%mm3                     \n\t"\
-                "paddw %%mm6, %%mm6                     \n\t"\
-                "paddw %%mm3, %%mm2                     \n\t"\
-                "paddw %%mm6, %%mm5                     \n\t"\
-                "movq %%mm4, %%mm6                      \n\t"\
-                "punpcklbw %%mm7, %%mm4                 \n\t"\
-                "punpckhbw %%mm7, %%mm6                 \n\t"\
-                "psubw %%mm4, %%mm2                     \n\t"\
-                "psubw %%mm6, %%mm5                     \n\t"\
-                "psraw $2, %%mm2                        \n\t"\
-                "psraw $2, %%mm5                        \n\t"\
-                "packuswb %%mm5, %%mm2                  \n\t"\
-                "movq %%mm2, " #a "                     \n\t"\
+        "movq " #a ", %%mm2                     \n\t"\
+        "movq " #b ", %%mm3                     \n\t"\
+        "movq " #c ", %%mm4                     \n\t"\
+        PAVGB(t2, %%mm3)                             \
+        PAVGB(t1, %%mm4)                             \
+        "movq %%mm2, %%mm5                      \n\t"\
+        "movq %%mm2, " #t1 "                    \n\t"\
+        "punpcklbw %%mm7, %%mm2                 \n\t"\
+        "punpckhbw %%mm7, %%mm5                 \n\t"\
+        "movq %%mm2, %%mm6                      \n\t"\
+        "paddw %%mm2, %%mm2                     \n\t"\
+        "paddw %%mm6, %%mm2                     \n\t"\
+        "movq %%mm5, %%mm6                      \n\t"\
+        "paddw %%mm5, %%mm5                     \n\t"\
+        "paddw %%mm6, %%mm5                     \n\t"\
+        "movq %%mm3, %%mm6                      \n\t"\
+        "punpcklbw %%mm7, %%mm3                 \n\t"\
+        "punpckhbw %%mm7, %%mm6                 \n\t"\
+        "paddw %%mm3, %%mm3                     \n\t"\
+        "paddw %%mm6, %%mm6                     \n\t"\
+        "paddw %%mm3, %%mm2                     \n\t"\
+        "paddw %%mm6, %%mm5                     \n\t"\
+        "movq %%mm4, %%mm6                      \n\t"\
+        "punpcklbw %%mm7, %%mm4                 \n\t"\
+        "punpckhbw %%mm7, %%mm6                 \n\t"\
+        "psubw %%mm4, %%mm2                     \n\t"\
+        "psubw %%mm6, %%mm5                     \n\t"\
+        "psraw $2, %%mm2                        \n\t"\
+        "psraw $2, %%mm5                        \n\t"\
+        "packuswb %%mm5, %%mm2                  \n\t"\
+        "movq %%mm2, " #a "                     \n\t"\
 
 #define DEINT_L5(t1,t2,a,b,c)  REAL_DEINT_L5(t1,t2,a,b,c)
 
@@ -1804,236 +1648,235 @@ DEINT_L5(%%mm1, %%mm0, (%%REGd)       , (%%REGd, %1)   , (%%REGd, %1, 2))
 DEINT_L5(%%mm0, %%mm1, (%%REGd, %1)   , (%%REGd, %1, 2), (%0, %1, 8)   )
 DEINT_L5(%%mm1, %%mm0, (%%REGd, %1, 2), (%0, %1, 8)    , (%%REGd, %1, 4))
 
-                "movq %%mm0, (%2)                       \n\t"
-                "movq %%mm1, (%3)                       \n\t"
-                : : "r" (src), "r" ((long)stride), "r"(tmp), "r"(tmp2)
-                : "%"REG_a, "%"REG_d
-        );
-#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        int x;
-        src+= stride*4;
-        for(x=0; x<8; x++)
-        {
-                int t1= tmp[x];
-                int t2= tmp2[x];
-                int t3= src[0];
-
-                src[stride*0]= CLIP((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);
-                t1= src[stride*1];
-                src[stride*1]= CLIP((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);
-                t2= src[stride*2];
-                src[stride*2]= CLIP((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);
-                t3= src[stride*3];
-                src[stride*3]= CLIP((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);
-                t1= src[stride*4];
-                src[stride*4]= CLIP((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);
-                t2= src[stride*5];
-                src[stride*5]= CLIP((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);
-                t3= src[stride*6];
-                src[stride*6]= CLIP((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);
-                t1= src[stride*7];
-                src[stride*7]= CLIP((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);
-
-                tmp[x]= t3;
-                tmp2[x]= t1;
-
-                src++;
-        }
-#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
+        "movq %%mm0, (%2)                       \n\t"
+        "movq %%mm1, (%3)                       \n\t"
+        : : "r" (src), "r" ((x86_reg)stride), "r"(tmp), "r"(tmp2)
+        : "%"REG_a, "%"REG_d
+    );
+#else //HAVE_MMX2 || HAVE_AMD3DNOW
+    int x;
+    src+= stride*4;
+    for(x=0; x<8; x++){
+        int t1= tmp[x];
+        int t2= tmp2[x];
+        int t3= src[0];
+
+        src[stride*0]= CLIP((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);
+        t1= src[stride*1];
+        src[stride*1]= CLIP((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);
+        t2= src[stride*2];
+        src[stride*2]= CLIP((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);
+        t3= src[stride*3];
+        src[stride*3]= CLIP((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);
+        t1= src[stride*4];
+        src[stride*4]= CLIP((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);
+        t2= src[stride*5];
+        src[stride*5]= CLIP((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);
+        t3= src[stride*6];
+        src[stride*6]= CLIP((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);
+        t1= src[stride*7];
+        src[stride*7]= CLIP((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);
+
+        tmp[x]= t3;
+        tmp2[x]= t1;
+
+        src++;
+    }
+#endif //HAVE_MMX2 || HAVE_AMD3DNOW
 }
 
 /**
- * Deinterlaces the given block by filtering all lines with a (1 2 1) filter.
+ * Deinterlace the given block by filtering all lines with a (1 2 1) filter.
  * will be called for every 8x8 block and can read & write from line 4-15
- * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
+ * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
  * lines 4-12 will be read into the deblocking filter and should be deinterlaced
  * this filter will read lines 4-13 and write 4-11
  */
 static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uint8_t *tmp)
 {
-#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        src+= 4*stride;
-        asm volatile(
-                "lea (%0, %1), %%"REG_a"                \n\t"
-                "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
+#if HAVE_MMX2 || HAVE_AMD3DNOW
+    src+= 4*stride;
+    __asm__ volatile(
+        "lea (%0, %1), %%"REG_a"                \n\t"
+        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
 //      0       1       2       3       4       5       6       7       8       9
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
 
-                "movq (%2), %%mm0                       \n\t" // L0
-                "movq (%%"REG_a"), %%mm1                \n\t" // L2
-                PAVGB(%%mm1, %%mm0)                           // L0+L2
-                "movq (%0), %%mm2                       \n\t" // L1
-                PAVGB(%%mm2, %%mm0)
-                "movq %%mm0, (%0)                       \n\t"
-                "movq (%%"REG_a", %1), %%mm0            \n\t" // L3
-                PAVGB(%%mm0, %%mm2)                           // L1+L3
-                PAVGB(%%mm1, %%mm2)                           // 2L2 + L1 + L3
-                "movq %%mm2, (%%"REG_a")                \n\t"
-                "movq (%%"REG_a", %1, 2), %%mm2         \n\t" // L4
-                PAVGB(%%mm2, %%mm1)                           // L2+L4
-                PAVGB(%%mm0, %%mm1)                           // 2L3 + L2 + L4
-                "movq %%mm1, (%%"REG_a", %1)            \n\t"
-                "movq (%0, %1, 4), %%mm1                \n\t" // L5
-                PAVGB(%%mm1, %%mm0)                           // L3+L5
-                PAVGB(%%mm2, %%mm0)                           // 2L4 + L3 + L5
-                "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
-                "movq (%%"REG_d"), %%mm0                \n\t" // L6
-                PAVGB(%%mm0, %%mm2)                           // L4+L6
-                PAVGB(%%mm1, %%mm2)                           // 2L5 + L4 + L6
-                "movq %%mm2, (%0, %1, 4)                \n\t"
-                "movq (%%"REG_d", %1), %%mm2            \n\t" // L7
-                PAVGB(%%mm2, %%mm1)                           // L5+L7
-                PAVGB(%%mm0, %%mm1)                           // 2L6 + L5 + L7
-                "movq %%mm1, (%%"REG_d")                \n\t"
-                "movq (%%"REG_d", %1, 2), %%mm1         \n\t" // L8
-                PAVGB(%%mm1, %%mm0)                           // L6+L8
-                PAVGB(%%mm2, %%mm0)                           // 2L7 + L6 + L8
-                "movq %%mm0, (%%"REG_d", %1)            \n\t"
-                "movq (%0, %1, 8), %%mm0                \n\t" // L9
-                PAVGB(%%mm0, %%mm2)                           // L7+L9
-                PAVGB(%%mm1, %%mm2)                           // 2L8 + L7 + L9
-                "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
-                "movq %%mm1, (%2)                       \n\t"
-
-                : : "r" (src), "r" ((long)stride), "r" (tmp)
-                : "%"REG_a, "%"REG_d
-        );
-#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        int a, b, c, x;
-        src+= 4*stride;
-
-        for(x=0; x<2; x++){
-                a= *(uint32_t*)&tmp[stride*0];
-                b= *(uint32_t*)&src[stride*0];
-                c= *(uint32_t*)&src[stride*1];
-                a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
-                *(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
-
-                a= *(uint32_t*)&src[stride*2];
-                b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
-                *(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
-
-                b= *(uint32_t*)&src[stride*3];
-                c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
-                *(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
-
-                c= *(uint32_t*)&src[stride*4];
-                a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
-                *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
-
-                a= *(uint32_t*)&src[stride*5];
-                b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
-                *(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
-
-                b= *(uint32_t*)&src[stride*6];
-                c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
-                *(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
-
-                c= *(uint32_t*)&src[stride*7];
-                a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
-                *(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
-
-                a= *(uint32_t*)&src[stride*8];
-                b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
-                *(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
-
-                *(uint32_t*)&tmp[stride*0]= c;
-                src += 4;
-                tmp += 4;
-        }
-#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
+        "movq (%2), %%mm0                       \n\t" // L0
+        "movq (%%"REG_a"), %%mm1                \n\t" // L2
+        PAVGB(%%mm1, %%mm0)                           // L0+L2
+        "movq (%0), %%mm2                       \n\t" // L1
+        PAVGB(%%mm2, %%mm0)
+        "movq %%mm0, (%0)                       \n\t"
+        "movq (%%"REG_a", %1), %%mm0            \n\t" // L3
+        PAVGB(%%mm0, %%mm2)                           // L1+L3
+        PAVGB(%%mm1, %%mm2)                           // 2L2 + L1 + L3
+        "movq %%mm2, (%%"REG_a")                \n\t"
+        "movq (%%"REG_a", %1, 2), %%mm2         \n\t" // L4
+        PAVGB(%%mm2, %%mm1)                           // L2+L4
+        PAVGB(%%mm0, %%mm1)                           // 2L3 + L2 + L4
+        "movq %%mm1, (%%"REG_a", %1)            \n\t"
+        "movq (%0, %1, 4), %%mm1                \n\t" // L5
+        PAVGB(%%mm1, %%mm0)                           // L3+L5
+        PAVGB(%%mm2, %%mm0)                           // 2L4 + L3 + L5
+        "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
+        "movq (%%"REG_d"), %%mm0                \n\t" // L6
+        PAVGB(%%mm0, %%mm2)                           // L4+L6
+        PAVGB(%%mm1, %%mm2)                           // 2L5 + L4 + L6
+        "movq %%mm2, (%0, %1, 4)                \n\t"
+        "movq (%%"REG_d", %1), %%mm2            \n\t" // L7
+        PAVGB(%%mm2, %%mm1)                           // L5+L7
+        PAVGB(%%mm0, %%mm1)                           // 2L6 + L5 + L7
+        "movq %%mm1, (%%"REG_d")                \n\t"
+        "movq (%%"REG_d", %1, 2), %%mm1         \n\t" // L8
+        PAVGB(%%mm1, %%mm0)                           // L6+L8
+        PAVGB(%%mm2, %%mm0)                           // 2L7 + L6 + L8
+        "movq %%mm0, (%%"REG_d", %1)            \n\t"
+        "movq (%0, %1, 8), %%mm0                \n\t" // L9
+        PAVGB(%%mm0, %%mm2)                           // L7+L9
+        PAVGB(%%mm1, %%mm2)                           // 2L8 + L7 + L9
+        "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
+        "movq %%mm1, (%2)                       \n\t"
+
+        : : "r" (src), "r" ((x86_reg)stride), "r" (tmp)
+        : "%"REG_a, "%"REG_d
+    );
+#else //HAVE_MMX2 || HAVE_AMD3DNOW
+    int a, b, c, x;
+    src+= 4*stride;
+
+    for(x=0; x<2; x++){
+        a= *(uint32_t*)&tmp[stride*0];
+        b= *(uint32_t*)&src[stride*0];
+        c= *(uint32_t*)&src[stride*1];
+        a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
+        *(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
+
+        a= *(uint32_t*)&src[stride*2];
+        b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
+        *(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
+
+        b= *(uint32_t*)&src[stride*3];
+        c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
+        *(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
+
+        c= *(uint32_t*)&src[stride*4];
+        a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
+        *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
+
+        a= *(uint32_t*)&src[stride*5];
+        b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
+        *(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
+
+        b= *(uint32_t*)&src[stride*6];
+        c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
+        *(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
+
+        c= *(uint32_t*)&src[stride*7];
+        a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
+        *(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
+
+        a= *(uint32_t*)&src[stride*8];
+        b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
+        *(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
+
+        *(uint32_t*)&tmp[stride*0]= c;
+        src += 4;
+        tmp += 4;
+    }
+#endif //HAVE_MMX2 || HAVE_AMD3DNOW
 }
 
 /**
- * Deinterlaces the given block by applying a median filter to every second line.
+ * Deinterlace the given block by applying a median filter to every second line.
  * will be called for every 8x8 block and can read & write from line 4-15,
- * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
+ * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
  * lines 4-12 will be read into the deblocking filter and should be deinterlaced
  */
 static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)
 {
-#ifdef HAVE_MMX
-        src+= 4*stride;
-#ifdef HAVE_MMX2
-        asm volatile(
-                "lea (%0, %1), %%"REG_a"                \n\t"
-                "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
+#if HAVE_MMX
+    src+= 4*stride;
+#if HAVE_MMX2
+    __asm__ volatile(
+        "lea (%0, %1), %%"REG_a"                \n\t"
+        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
 //      0       1       2       3       4       5       6       7       8       9
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
 
-                "movq (%0), %%mm0                       \n\t" //
-                "movq (%%"REG_a", %1), %%mm2            \n\t" //
-                "movq (%%"REG_a"), %%mm1                \n\t" //
-                "movq %%mm0, %%mm3                      \n\t"
-                "pmaxub %%mm1, %%mm0                    \n\t" //
-                "pminub %%mm3, %%mm1                    \n\t" //
-                "pmaxub %%mm2, %%mm1                    \n\t" //
-                "pminub %%mm1, %%mm0                    \n\t"
-                "movq %%mm0, (%%"REG_a")                \n\t"
-
-                "movq (%0, %1, 4), %%mm0                \n\t" //
-                "movq (%%"REG_a", %1, 2), %%mm1         \n\t" //
-                "movq %%mm2, %%mm3                      \n\t"
-                "pmaxub %%mm1, %%mm2                    \n\t" //
-                "pminub %%mm3, %%mm1                    \n\t" //
-                "pmaxub %%mm0, %%mm1                    \n\t" //
-                "pminub %%mm1, %%mm2                    \n\t"
-                "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"
-
-                "movq (%%"REG_d"), %%mm2                \n\t" //
-                "movq (%%"REG_d", %1), %%mm1            \n\t" //
-                "movq %%mm2, %%mm3                      \n\t"
-                "pmaxub %%mm0, %%mm2                    \n\t" //
-                "pminub %%mm3, %%mm0                    \n\t" //
-                "pmaxub %%mm1, %%mm0                    \n\t" //
-                "pminub %%mm0, %%mm2                    \n\t"
-                "movq %%mm2, (%%"REG_d")                \n\t"
-
-                "movq (%%"REG_d", %1, 2), %%mm2         \n\t" //
-                "movq (%0, %1, 8), %%mm0                \n\t" //
-                "movq %%mm2, %%mm3                      \n\t"
-                "pmaxub %%mm0, %%mm2                    \n\t" //
-                "pminub %%mm3, %%mm0                    \n\t" //
-                "pmaxub %%mm1, %%mm0                    \n\t" //
-                "pminub %%mm0, %%mm2                    \n\t"
-                "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
-
-
-                : : "r" (src), "r" ((long)stride)
-                : "%"REG_a, "%"REG_d
-        );
+        "movq (%0), %%mm0                       \n\t" //
+        "movq (%%"REG_a", %1), %%mm2            \n\t" //
+        "movq (%%"REG_a"), %%mm1                \n\t" //
+        "movq %%mm0, %%mm3                      \n\t"
+        "pmaxub %%mm1, %%mm0                    \n\t" //
+        "pminub %%mm3, %%mm1                    \n\t" //
+        "pmaxub %%mm2, %%mm1                    \n\t" //
+        "pminub %%mm1, %%mm0                    \n\t"
+        "movq %%mm0, (%%"REG_a")                \n\t"
+
+        "movq (%0, %1, 4), %%mm0                \n\t" //
+        "movq (%%"REG_a", %1, 2), %%mm1         \n\t" //
+        "movq %%mm2, %%mm3                      \n\t"
+        "pmaxub %%mm1, %%mm2                    \n\t" //
+        "pminub %%mm3, %%mm1                    \n\t" //
+        "pmaxub %%mm0, %%mm1                    \n\t" //
+        "pminub %%mm1, %%mm2                    \n\t"
+        "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"
+
+        "movq (%%"REG_d"), %%mm2                \n\t" //
+        "movq (%%"REG_d", %1), %%mm1            \n\t" //
+        "movq %%mm2, %%mm3                      \n\t"
+        "pmaxub %%mm0, %%mm2                    \n\t" //
+        "pminub %%mm3, %%mm0                    \n\t" //
+        "pmaxub %%mm1, %%mm0                    \n\t" //
+        "pminub %%mm0, %%mm2                    \n\t"
+        "movq %%mm2, (%%"REG_d")                \n\t"
+
+        "movq (%%"REG_d", %1, 2), %%mm2         \n\t" //
+        "movq (%0, %1, 8), %%mm0                \n\t" //
+        "movq %%mm2, %%mm3                      \n\t"
+        "pmaxub %%mm0, %%mm2                    \n\t" //
+        "pminub %%mm3, %%mm0                    \n\t" //
+        "pmaxub %%mm1, %%mm0                    \n\t" //
+        "pminub %%mm0, %%mm2                    \n\t"
+        "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
+
+
+        : : "r" (src), "r" ((x86_reg)stride)
+        : "%"REG_a, "%"REG_d
+    );
 
 #else // MMX without MMX2
-        asm volatile(
-                "lea (%0, %1), %%"REG_a"                \n\t"
-                "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
+    __asm__ volatile(
+        "lea (%0, %1), %%"REG_a"                \n\t"
+        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
 //      0       1       2       3       4       5       6       7       8       9
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
-                "pxor %%mm7, %%mm7                      \n\t"
+        "pxor %%mm7, %%mm7                      \n\t"
 
 #define REAL_MEDIAN(a,b,c)\
-                "movq " #a ", %%mm0                     \n\t"\
-                "movq " #b ", %%mm2                     \n\t"\
-                "movq " #c ", %%mm1                     \n\t"\
-                "movq %%mm0, %%mm3                      \n\t"\
-                "movq %%mm1, %%mm4                      \n\t"\
-                "movq %%mm2, %%mm5                      \n\t"\
-                "psubusb %%mm1, %%mm3                   \n\t"\
-                "psubusb %%mm2, %%mm4                   \n\t"\
-                "psubusb %%mm0, %%mm5                   \n\t"\
-                "pcmpeqb %%mm7, %%mm3                   \n\t"\
-                "pcmpeqb %%mm7, %%mm4                   \n\t"\
-                "pcmpeqb %%mm7, %%mm5                   \n\t"\
-                "movq %%mm3, %%mm6                      \n\t"\
-                "pxor %%mm4, %%mm3                      \n\t"\
-                "pxor %%mm5, %%mm4                      \n\t"\
-                "pxor %%mm6, %%mm5                      \n\t"\
-                "por %%mm3, %%mm1                       \n\t"\
-                "por %%mm4, %%mm2                       \n\t"\
-                "por %%mm5, %%mm0                       \n\t"\
-                "pand %%mm2, %%mm0                      \n\t"\
-                "pand %%mm1, %%mm0                      \n\t"\
-                "movq %%mm0, " #b "                     \n\t"
+        "movq " #a ", %%mm0                     \n\t"\
+        "movq " #b ", %%mm2                     \n\t"\
+        "movq " #c ", %%mm1                     \n\t"\
+        "movq %%mm0, %%mm3                      \n\t"\
+        "movq %%mm1, %%mm4                      \n\t"\
+        "movq %%mm2, %%mm5                      \n\t"\
+        "psubusb %%mm1, %%mm3                   \n\t"\
+        "psubusb %%mm2, %%mm4                   \n\t"\
+        "psubusb %%mm0, %%mm5                   \n\t"\
+        "pcmpeqb %%mm7, %%mm3                   \n\t"\
+        "pcmpeqb %%mm7, %%mm4                   \n\t"\
+        "pcmpeqb %%mm7, %%mm5                   \n\t"\
+        "movq %%mm3, %%mm6                      \n\t"\
+        "pxor %%mm4, %%mm3                      \n\t"\
+        "pxor %%mm5, %%mm4                      \n\t"\
+        "pxor %%mm6, %%mm5                      \n\t"\
+        "por %%mm3, %%mm1                       \n\t"\
+        "por %%mm4, %%mm2                       \n\t"\
+        "por %%mm5, %%mm0                       \n\t"\
+        "pand %%mm2, %%mm0                      \n\t"\
+        "pand %%mm1, %%mm0                      \n\t"\
+        "movq %%mm0, " #b "                     \n\t"
 #define MEDIAN(a,b,c)  REAL_MEDIAN(a,b,c)
 
 MEDIAN((%0)        , (%%REGa)       , (%%REGa, %1))
@@ -2041,118 +1884,116 @@ MEDIAN((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4))
 MEDIAN((%0, %1, 4) , (%%REGd)       , (%%REGd, %1))
 MEDIAN((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8))
 
-                : : "r" (src), "r" ((long)stride)
-                : "%"REG_a, "%"REG_d
-        );
+        : : "r" (src), "r" ((x86_reg)stride)
+        : "%"REG_a, "%"REG_d
+    );
 #endif //HAVE_MMX2
 #else //HAVE_MMX
-        int x, y;
-        src+= 4*stride;
-        // FIXME - there should be a way to do a few columns in parallel like w/mmx
-        for(x=0; x<8; x++)
-        {
-                uint8_t *colsrc = src;
-                for (y=0; y<4; y++)
-                {
-                        int a, b, c, d, e, f;
-                        a = colsrc[0       ];
-                        b = colsrc[stride  ];
-                        c = colsrc[stride*2];
-                        d = (a-b)>>31;
-                        e = (b-c)>>31;
-                        f = (c-a)>>31;
-                        colsrc[stride  ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));
-                        colsrc += stride*2;
-                }
-                src++;
+    int x, y;
+    src+= 4*stride;
+    // FIXME - there should be a way to do a few columns in parallel like w/mmx
+    for(x=0; x<8; x++){
+        uint8_t *colsrc = src;
+        for (y=0; y<4; y++){
+            int a, b, c, d, e, f;
+            a = colsrc[0       ];
+            b = colsrc[stride  ];
+            c = colsrc[stride*2];
+            d = (a-b)>>31;
+            e = (b-c)>>31;
+            f = (c-a)>>31;
+            colsrc[stride  ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));
+            colsrc += stride*2;
         }
+        src++;
+    }
 #endif //HAVE_MMX
 }
 
-#ifdef HAVE_MMX
+#if HAVE_MMX
 /**
  * transposes and shift the given 8x8 Block into dst1 and dst2
  */
 static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)
 {
-        asm(
-                "lea (%0, %1), %%"REG_a"                \n\t"
+    __asm__(
+        "lea (%0, %1), %%"REG_a"                \n\t"
 //      0       1       2       3       4       5       6       7       8       9
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
-                "movq (%0), %%mm0                       \n\t" // 12345678
-                "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
-                "movq %%mm0, %%mm2                      \n\t" // 12345678
-                "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
-                "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
-
-                "movq (%%"REG_a", %1), %%mm1            \n\t"
-                "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
-                "movq %%mm1, %%mm4                      \n\t"
-                "punpcklbw %%mm3, %%mm1                 \n\t"
-                "punpckhbw %%mm3, %%mm4                 \n\t"
-
-                "movq %%mm0, %%mm3                      \n\t"
-                "punpcklwd %%mm1, %%mm0                 \n\t"
-                "punpckhwd %%mm1, %%mm3                 \n\t"
-                "movq %%mm2, %%mm1                      \n\t"
-                "punpcklwd %%mm4, %%mm2                 \n\t"
-                "punpckhwd %%mm4, %%mm1                 \n\t"
-
-                "movd %%mm0, 128(%2)                    \n\t"
-                "psrlq $32, %%mm0                       \n\t"
-                "movd %%mm0, 144(%2)                    \n\t"
-                "movd %%mm3, 160(%2)                    \n\t"
-                "psrlq $32, %%mm3                       \n\t"
-                "movd %%mm3, 176(%2)                    \n\t"
-                "movd %%mm3, 48(%3)                     \n\t"
-                "movd %%mm2, 192(%2)                    \n\t"
-                "movd %%mm2, 64(%3)                     \n\t"
-                "psrlq $32, %%mm2                       \n\t"
-                "movd %%mm2, 80(%3)                     \n\t"
-                "movd %%mm1, 96(%3)                     \n\t"
-                "psrlq $32, %%mm1                       \n\t"
-                "movd %%mm1, 112(%3)                    \n\t"
-
-                "lea (%%"REG_a", %1, 4), %%"REG_a"      \n\t"
-
-                "movq (%0, %1, 4), %%mm0                \n\t" // 12345678
-                "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
-                "movq %%mm0, %%mm2                      \n\t" // 12345678
-                "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
-                "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
-
-                "movq (%%"REG_a", %1), %%mm1            \n\t"
-                "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
-                "movq %%mm1, %%mm4                      \n\t"
-                "punpcklbw %%mm3, %%mm1                 \n\t"
-                "punpckhbw %%mm3, %%mm4                 \n\t"
-
-                "movq %%mm0, %%mm3                      \n\t"
-                "punpcklwd %%mm1, %%mm0                 \n\t"
-                "punpckhwd %%mm1, %%mm3                 \n\t"
-                "movq %%mm2, %%mm1                      \n\t"
-                "punpcklwd %%mm4, %%mm2                 \n\t"
-                "punpckhwd %%mm4, %%mm1                 \n\t"
-
-                "movd %%mm0, 132(%2)                    \n\t"
-                "psrlq $32, %%mm0                       \n\t"
-                "movd %%mm0, 148(%2)                    \n\t"
-                "movd %%mm3, 164(%2)                    \n\t"
-                "psrlq $32, %%mm3                       \n\t"
-                "movd %%mm3, 180(%2)                    \n\t"
-                "movd %%mm3, 52(%3)                     \n\t"
-                "movd %%mm2, 196(%2)                    \n\t"
-                "movd %%mm2, 68(%3)                     \n\t"
-                "psrlq $32, %%mm2                       \n\t"
-                "movd %%mm2, 84(%3)                     \n\t"
-                "movd %%mm1, 100(%3)                    \n\t"
-                "psrlq $32, %%mm1                       \n\t"
-                "movd %%mm1, 116(%3)                    \n\t"
-
-
-        :: "r" (src), "r" ((long)srcStride), "r" (dst1), "r" (dst2)
+        "movq (%0), %%mm0                       \n\t" // 12345678
+        "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
+        "movq %%mm0, %%mm2                      \n\t" // 12345678
+        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
+        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
+
+        "movq (%%"REG_a", %1), %%mm1            \n\t"
+        "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
+        "movq %%mm1, %%mm4                      \n\t"
+        "punpcklbw %%mm3, %%mm1                 \n\t"
+        "punpckhbw %%mm3, %%mm4                 \n\t"
+
+        "movq %%mm0, %%mm3                      \n\t"
+        "punpcklwd %%mm1, %%mm0                 \n\t"
+        "punpckhwd %%mm1, %%mm3                 \n\t"
+        "movq %%mm2, %%mm1                      \n\t"
+        "punpcklwd %%mm4, %%mm2                 \n\t"
+        "punpckhwd %%mm4, %%mm1                 \n\t"
+
+        "movd %%mm0, 128(%2)                    \n\t"
+        "psrlq $32, %%mm0                       \n\t"
+        "movd %%mm0, 144(%2)                    \n\t"
+        "movd %%mm3, 160(%2)                    \n\t"
+        "psrlq $32, %%mm3                       \n\t"
+        "movd %%mm3, 176(%2)                    \n\t"
+        "movd %%mm3, 48(%3)                     \n\t"
+        "movd %%mm2, 192(%2)                    \n\t"
+        "movd %%mm2, 64(%3)                     \n\t"
+        "psrlq $32, %%mm2                       \n\t"
+        "movd %%mm2, 80(%3)                     \n\t"
+        "movd %%mm1, 96(%3)                     \n\t"
+        "psrlq $32, %%mm1                       \n\t"
+        "movd %%mm1, 112(%3)                    \n\t"
+
+        "lea (%%"REG_a", %1, 4), %%"REG_a"      \n\t"
+
+        "movq (%0, %1, 4), %%mm0                \n\t" // 12345678
+        "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
+        "movq %%mm0, %%mm2                      \n\t" // 12345678
+        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
+        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
+
+        "movq (%%"REG_a", %1), %%mm1            \n\t"
+        "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
+        "movq %%mm1, %%mm4                      \n\t"
+        "punpcklbw %%mm3, %%mm1                 \n\t"
+        "punpckhbw %%mm3, %%mm4                 \n\t"
+
+        "movq %%mm0, %%mm3                      \n\t"
+        "punpcklwd %%mm1, %%mm0                 \n\t"
+        "punpckhwd %%mm1, %%mm3                 \n\t"
+        "movq %%mm2, %%mm1                      \n\t"
+        "punpcklwd %%mm4, %%mm2                 \n\t"
+        "punpckhwd %%mm4, %%mm1                 \n\t"
+
+        "movd %%mm0, 132(%2)                    \n\t"
+        "psrlq $32, %%mm0                       \n\t"
+        "movd %%mm0, 148(%2)                    \n\t"
+        "movd %%mm3, 164(%2)                    \n\t"
+        "psrlq $32, %%mm3                       \n\t"
+        "movd %%mm3, 180(%2)                    \n\t"
+        "movd %%mm3, 52(%3)                     \n\t"
+        "movd %%mm2, 196(%2)                    \n\t"
+        "movd %%mm2, 68(%3)                     \n\t"
+        "psrlq $32, %%mm2                       \n\t"
+        "movd %%mm2, 84(%3)                     \n\t"
+        "movd %%mm1, 100(%3)                    \n\t"
+        "psrlq $32, %%mm1                       \n\t"
+        "movd %%mm1, 116(%3)                    \n\t"
+
+
+        :: "r" (src), "r" ((x86_reg)srcStride), "r" (dst1), "r" (dst2)
         : "%"REG_a
-        );
+    );
 }
 
 /**
@@ -2160,164 +2001,164 @@ static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src
  */
 static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)
 {
-        asm(
-                "lea (%0, %1), %%"REG_a"                \n\t"
-                "lea (%%"REG_a",%1,4), %%"REG_d"        \n\t"
+    __asm__(
+        "lea (%0, %1), %%"REG_a"                \n\t"
+        "lea (%%"REG_a",%1,4), %%"REG_d"        \n\t"
 //      0       1       2       3       4       5       6       7       8       9
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
-                "movq (%2), %%mm0                       \n\t" // 12345678
-                "movq 16(%2), %%mm1                     \n\t" // abcdefgh
-                "movq %%mm0, %%mm2                      \n\t" // 12345678
-                "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
-                "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
-
-                "movq 32(%2), %%mm1                     \n\t"
-                "movq 48(%2), %%mm3                     \n\t"
-                "movq %%mm1, %%mm4                      \n\t"
-                "punpcklbw %%mm3, %%mm1                 \n\t"
-                "punpckhbw %%mm3, %%mm4                 \n\t"
-
-                "movq %%mm0, %%mm3                      \n\t"
-                "punpcklwd %%mm1, %%mm0                 \n\t"
-                "punpckhwd %%mm1, %%mm3                 \n\t"
-                "movq %%mm2, %%mm1                      \n\t"
-                "punpcklwd %%mm4, %%mm2                 \n\t"
-                "punpckhwd %%mm4, %%mm1                 \n\t"
-
-                "movd %%mm0, (%0)                       \n\t"
-                "psrlq $32, %%mm0                       \n\t"
-                "movd %%mm0, (%%"REG_a")                \n\t"
-                "movd %%mm3, (%%"REG_a", %1)            \n\t"
-                "psrlq $32, %%mm3                       \n\t"
-                "movd %%mm3, (%%"REG_a", %1, 2)         \n\t"
-                "movd %%mm2, (%0, %1, 4)                \n\t"
-                "psrlq $32, %%mm2                       \n\t"
-                "movd %%mm2, (%%"REG_d")                \n\t"
-                "movd %%mm1, (%%"REG_d", %1)            \n\t"
-                "psrlq $32, %%mm1                       \n\t"
-                "movd %%mm1, (%%"REG_d", %1, 2)         \n\t"
-
-
-                "movq 64(%2), %%mm0                     \n\t" // 12345678
-                "movq 80(%2), %%mm1                     \n\t" // abcdefgh
-                "movq %%mm0, %%mm2                      \n\t" // 12345678
-                "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
-                "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
-
-                "movq 96(%2), %%mm1                     \n\t"
-                "movq 112(%2), %%mm3                    \n\t"
-                "movq %%mm1, %%mm4                      \n\t"
-                "punpcklbw %%mm3, %%mm1                 \n\t"
-                "punpckhbw %%mm3, %%mm4                 \n\t"
-
-                "movq %%mm0, %%mm3                      \n\t"
-                "punpcklwd %%mm1, %%mm0                 \n\t"
-                "punpckhwd %%mm1, %%mm3                 \n\t"
-                "movq %%mm2, %%mm1                      \n\t"
-                "punpcklwd %%mm4, %%mm2                 \n\t"
-                "punpckhwd %%mm4, %%mm1                 \n\t"
-
-                "movd %%mm0, 4(%0)                      \n\t"
-                "psrlq $32, %%mm0                       \n\t"
-                "movd %%mm0, 4(%%"REG_a")               \n\t"
-                "movd %%mm3, 4(%%"REG_a", %1)           \n\t"
-                "psrlq $32, %%mm3                       \n\t"
-                "movd %%mm3, 4(%%"REG_a", %1, 2)        \n\t"
-                "movd %%mm2, 4(%0, %1, 4)               \n\t"
-                "psrlq $32, %%mm2                       \n\t"
-                "movd %%mm2, 4(%%"REG_d")               \n\t"
-                "movd %%mm1, 4(%%"REG_d", %1)           \n\t"
-                "psrlq $32, %%mm1                       \n\t"
-                "movd %%mm1, 4(%%"REG_d", %1, 2)        \n\t"
-
-        :: "r" (dst), "r" ((long)dstStride), "r" (src)
+        "movq (%2), %%mm0                       \n\t" // 12345678
+        "movq 16(%2), %%mm1                     \n\t" // abcdefgh
+        "movq %%mm0, %%mm2                      \n\t" // 12345678
+        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
+        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
+
+        "movq 32(%2), %%mm1                     \n\t"
+        "movq 48(%2), %%mm3                     \n\t"
+        "movq %%mm1, %%mm4                      \n\t"
+        "punpcklbw %%mm3, %%mm1                 \n\t"
+        "punpckhbw %%mm3, %%mm4                 \n\t"
+
+        "movq %%mm0, %%mm3                      \n\t"
+        "punpcklwd %%mm1, %%mm0                 \n\t"
+        "punpckhwd %%mm1, %%mm3                 \n\t"
+        "movq %%mm2, %%mm1                      \n\t"
+        "punpcklwd %%mm4, %%mm2                 \n\t"
+        "punpckhwd %%mm4, %%mm1                 \n\t"
+
+        "movd %%mm0, (%0)                       \n\t"
+        "psrlq $32, %%mm0                       \n\t"
+        "movd %%mm0, (%%"REG_a")                \n\t"
+        "movd %%mm3, (%%"REG_a", %1)            \n\t"
+        "psrlq $32, %%mm3                       \n\t"
+        "movd %%mm3, (%%"REG_a", %1, 2)         \n\t"
+        "movd %%mm2, (%0, %1, 4)                \n\t"
+        "psrlq $32, %%mm2                       \n\t"
+        "movd %%mm2, (%%"REG_d")                \n\t"
+        "movd %%mm1, (%%"REG_d", %1)            \n\t"
+        "psrlq $32, %%mm1                       \n\t"
+        "movd %%mm1, (%%"REG_d", %1, 2)         \n\t"
+
+
+        "movq 64(%2), %%mm0                     \n\t" // 12345678
+        "movq 80(%2), %%mm1                     \n\t" // abcdefgh
+        "movq %%mm0, %%mm2                      \n\t" // 12345678
+        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
+        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
+
+        "movq 96(%2), %%mm1                     \n\t"
+        "movq 112(%2), %%mm3                    \n\t"
+        "movq %%mm1, %%mm4                      \n\t"
+        "punpcklbw %%mm3, %%mm1                 \n\t"
+        "punpckhbw %%mm3, %%mm4                 \n\t"
+
+        "movq %%mm0, %%mm3                      \n\t"
+        "punpcklwd %%mm1, %%mm0                 \n\t"
+        "punpckhwd %%mm1, %%mm3                 \n\t"
+        "movq %%mm2, %%mm1                      \n\t"
+        "punpcklwd %%mm4, %%mm2                 \n\t"
+        "punpckhwd %%mm4, %%mm1                 \n\t"
+
+        "movd %%mm0, 4(%0)                      \n\t"
+        "psrlq $32, %%mm0                       \n\t"
+        "movd %%mm0, 4(%%"REG_a")               \n\t"
+        "movd %%mm3, 4(%%"REG_a", %1)           \n\t"
+        "psrlq $32, %%mm3                       \n\t"
+        "movd %%mm3, 4(%%"REG_a", %1, 2)        \n\t"
+        "movd %%mm2, 4(%0, %1, 4)               \n\t"
+        "psrlq $32, %%mm2                       \n\t"
+        "movd %%mm2, 4(%%"REG_d")               \n\t"
+        "movd %%mm1, 4(%%"REG_d", %1)           \n\t"
+        "psrlq $32, %%mm1                       \n\t"
+        "movd %%mm1, 4(%%"REG_d", %1, 2)        \n\t"
+
+        :: "r" (dst), "r" ((x86_reg)dstStride), "r" (src)
         : "%"REG_a, "%"REG_d
-        );
+    );
 }
 #endif //HAVE_MMX
 //static long test=0;
 
-#ifndef HAVE_ALTIVEC
+#if !HAVE_ALTIVEC
 static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
-                                    uint8_t *tempBlured, uint32_t *tempBluredPast, int *maxNoise)
+                                    uint8_t *tempBlurred, uint32_t *tempBlurredPast, int *maxNoise)
 {
-        // to save a register (FIXME do this outside of the loops)
-        tempBluredPast[127]= maxNoise[0];
-        tempBluredPast[128]= maxNoise[1];
-        tempBluredPast[129]= maxNoise[2];
+    // to save a register (FIXME do this outside of the loops)
+    tempBlurredPast[127]= maxNoise[0];
+    tempBlurredPast[128]= maxNoise[1];
+    tempBlurredPast[129]= maxNoise[2];
 
 #define FAST_L2_DIFF
 //#define L1_DIFF //u should change the thresholds too if u try that one
-#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        asm volatile(
-                "lea (%2, %2, 2), %%"REG_a"             \n\t" // 3*stride
-                "lea (%2, %2, 4), %%"REG_d"             \n\t" // 5*stride
-                "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
+#if HAVE_MMX2 || HAVE_AMD3DNOW
+    __asm__ volatile(
+        "lea (%2, %2, 2), %%"REG_a"             \n\t" // 3*stride
+        "lea (%2, %2, 4), %%"REG_d"             \n\t" // 5*stride
+        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
 //      0       1       2       3       4       5       6       7       8       9
 //      %x      %x+%2   %x+2%2  %x+eax  %x+4%2  %x+edx  %x+2eax %x+ecx  %x+8%2
 //FIXME reorder?
 #ifdef L1_DIFF //needs mmx2
-                "movq (%0), %%mm0                       \n\t" // L0
-                "psadbw (%1), %%mm0                     \n\t" // |L0-R0|
-                "movq (%0, %2), %%mm1                   \n\t" // L1
-                "psadbw (%1, %2), %%mm1                 \n\t" // |L1-R1|
-                "movq (%0, %2, 2), %%mm2                \n\t" // L2
-                "psadbw (%1, %2, 2), %%mm2              \n\t" // |L2-R2|
-                "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
-                "psadbw (%1, %%"REG_a"), %%mm3          \n\t" // |L3-R3|
-
-                "movq (%0, %2, 4), %%mm4                \n\t" // L4
-                "paddw %%mm1, %%mm0                     \n\t"
-                "psadbw (%1, %2, 4), %%mm4              \n\t" // |L4-R4|
-                "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
-                "paddw %%mm2, %%mm0                     \n\t"
-                "psadbw (%1, %%"REG_d"), %%mm5          \n\t" // |L5-R5|
-                "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
-                "paddw %%mm3, %%mm0                     \n\t"
-                "psadbw (%1, %%"REG_a", 2), %%mm6       \n\t" // |L6-R6|
-                "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
-                "paddw %%mm4, %%mm0                     \n\t"
-                "psadbw (%1, %%"REG_c"), %%mm7          \n\t" // |L7-R7|
-                "paddw %%mm5, %%mm6                     \n\t"
-                "paddw %%mm7, %%mm6                     \n\t"
-                "paddw %%mm6, %%mm0                     \n\t"
+        "movq (%0), %%mm0                       \n\t" // L0
+        "psadbw (%1), %%mm0                     \n\t" // |L0-R0|
+        "movq (%0, %2), %%mm1                   \n\t" // L1
+        "psadbw (%1, %2), %%mm1                 \n\t" // |L1-R1|
+        "movq (%0, %2, 2), %%mm2                \n\t" // L2
+        "psadbw (%1, %2, 2), %%mm2              \n\t" // |L2-R2|
+        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
+        "psadbw (%1, %%"REG_a"), %%mm3          \n\t" // |L3-R3|
+
+        "movq (%0, %2, 4), %%mm4                \n\t" // L4
+        "paddw %%mm1, %%mm0                     \n\t"
+        "psadbw (%1, %2, 4), %%mm4              \n\t" // |L4-R4|
+        "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
+        "paddw %%mm2, %%mm0                     \n\t"
+        "psadbw (%1, %%"REG_d"), %%mm5          \n\t" // |L5-R5|
+        "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
+        "paddw %%mm3, %%mm0                     \n\t"
+        "psadbw (%1, %%"REG_a", 2), %%mm6       \n\t" // |L6-R6|
+        "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
+        "paddw %%mm4, %%mm0                     \n\t"
+        "psadbw (%1, %%"REG_c"), %%mm7          \n\t" // |L7-R7|
+        "paddw %%mm5, %%mm6                     \n\t"
+        "paddw %%mm7, %%mm6                     \n\t"
+        "paddw %%mm6, %%mm0                     \n\t"
 #else //L1_DIFF
 #if defined (FAST_L2_DIFF)
-                "pcmpeqb %%mm7, %%mm7                   \n\t"
-                "movq "MANGLE(b80)", %%mm6              \n\t"
-                "pxor %%mm0, %%mm0                      \n\t"
+        "pcmpeqb %%mm7, %%mm7                   \n\t"
+        "movq "MANGLE(b80)", %%mm6              \n\t"
+        "pxor %%mm0, %%mm0                      \n\t"
 #define REAL_L2_DIFF_CORE(a, b)\
-                "movq " #a ", %%mm5                     \n\t"\
-                "movq " #b ", %%mm2                     \n\t"\
-                "pxor %%mm7, %%mm2                      \n\t"\
-                PAVGB(%%mm2, %%mm5)\
-                "paddb %%mm6, %%mm5                     \n\t"\
-                "movq %%mm5, %%mm2                      \n\t"\
-                "psllw $8, %%mm5                        \n\t"\
-                "pmaddwd %%mm5, %%mm5                   \n\t"\
-                "pmaddwd %%mm2, %%mm2                   \n\t"\
-                "paddd %%mm2, %%mm5                     \n\t"\
-                "psrld $14, %%mm5                       \n\t"\
-                "paddd %%mm5, %%mm0                     \n\t"
+        "movq " #a ", %%mm5                     \n\t"\
+        "movq " #b ", %%mm2                     \n\t"\
+        "pxor %%mm7, %%mm2                      \n\t"\
+        PAVGB(%%mm2, %%mm5)\
+        "paddb %%mm6, %%mm5                     \n\t"\
+        "movq %%mm5, %%mm2                      \n\t"\
+        "psllw $8, %%mm5                        \n\t"\
+        "pmaddwd %%mm5, %%mm5                   \n\t"\
+        "pmaddwd %%mm2, %%mm2                   \n\t"\
+        "paddd %%mm2, %%mm5                     \n\t"\
+        "psrld $14, %%mm5                       \n\t"\
+        "paddd %%mm5, %%mm0                     \n\t"
 
 #else //defined (FAST_L2_DIFF)
-                "pxor %%mm7, %%mm7                      \n\t"
-                "pxor %%mm0, %%mm0                      \n\t"
+        "pxor %%mm7, %%mm7                      \n\t"
+        "pxor %%mm0, %%mm0                      \n\t"
 #define REAL_L2_DIFF_CORE(a, b)\
-                "movq " #a ", %%mm5                     \n\t"\
-                "movq " #b ", %%mm2                     \n\t"\
-                "movq %%mm5, %%mm1                      \n\t"\
-                "movq %%mm2, %%mm3                      \n\t"\
-                "punpcklbw %%mm7, %%mm5                 \n\t"\
-                "punpckhbw %%mm7, %%mm1                 \n\t"\
-                "punpcklbw %%mm7, %%mm2                 \n\t"\
-                "punpckhbw %%mm7, %%mm3                 \n\t"\
-                "psubw %%mm2, %%mm5                     \n\t"\
-                "psubw %%mm3, %%mm1                     \n\t"\
-                "pmaddwd %%mm5, %%mm5                   \n\t"\
-                "pmaddwd %%mm1, %%mm1                   \n\t"\
-                "paddd %%mm1, %%mm5                     \n\t"\
-                "paddd %%mm5, %%mm0                     \n\t"
+        "movq " #a ", %%mm5                     \n\t"\
+        "movq " #b ", %%mm2                     \n\t"\
+        "movq %%mm5, %%mm1                      \n\t"\
+        "movq %%mm2, %%mm3                      \n\t"\
+        "punpcklbw %%mm7, %%mm5                 \n\t"\
+        "punpckhbw %%mm7, %%mm1                 \n\t"\
+        "punpcklbw %%mm7, %%mm2                 \n\t"\
+        "punpckhbw %%mm7, %%mm3                 \n\t"\
+        "psubw %%mm2, %%mm5                     \n\t"\
+        "psubw %%mm3, %%mm1                     \n\t"\
+        "pmaddwd %%mm5, %%mm5                   \n\t"\
+        "pmaddwd %%mm1, %%mm1                   \n\t"\
+        "paddd %%mm1, %%mm5                     \n\t"\
+        "paddd %%mm5, %%mm0                     \n\t"
 
 #endif //defined (FAST_L2_DIFF)
 
@@ -2334,239 +2175,237 @@ L2_DIFF_CORE((%0, %%REGc)  , (%1, %%REGc))
 
 #endif //L1_DIFF
 
-                "movq %%mm0, %%mm4                      \n\t"
-                "psrlq $32, %%mm0                       \n\t"
-                "paddd %%mm0, %%mm4                     \n\t"
-                "movd %%mm4, %%ecx                      \n\t"
-                "shll $2, %%ecx                         \n\t"
-                "mov %3, %%"REG_d"                      \n\t"
-                "addl -4(%%"REG_d"), %%ecx              \n\t"
-                "addl 4(%%"REG_d"), %%ecx               \n\t"
-                "addl -1024(%%"REG_d"), %%ecx           \n\t"
-                "addl $4, %%ecx                         \n\t"
-                "addl 1024(%%"REG_d"), %%ecx            \n\t"
-                "shrl $3, %%ecx                         \n\t"
-                "movl %%ecx, (%%"REG_d")                \n\t"
-
-//                "mov %3, %%"REG_c"                      \n\t"
-//                "mov %%"REG_c", test                    \n\t"
-//                "jmp 4f                                 \n\t"
-                "cmpl 512(%%"REG_d"), %%ecx             \n\t"
-                " jb 2f                                 \n\t"
-                "cmpl 516(%%"REG_d"), %%ecx             \n\t"
-                " jb 1f                                 \n\t"
-
-                "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
-                "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
-                "movq (%0), %%mm0                       \n\t" // L0
-                "movq (%0, %2), %%mm1                   \n\t" // L1
-                "movq (%0, %2, 2), %%mm2                \n\t" // L2
-                "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
-                "movq (%0, %2, 4), %%mm4                \n\t" // L4
-                "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
-                "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
-                "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
-                "movq %%mm0, (%1)                       \n\t" // L0
-                "movq %%mm1, (%1, %2)                   \n\t" // L1
-                "movq %%mm2, (%1, %2, 2)                \n\t" // L2
-                "movq %%mm3, (%1, %%"REG_a")            \n\t" // L3
-                "movq %%mm4, (%1, %2, 4)                \n\t" // L4
-                "movq %%mm5, (%1, %%"REG_d")            \n\t" // L5
-                "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // L6
-                "movq %%mm7, (%1, %%"REG_c")            \n\t" // L7
-                "jmp 4f                                 \n\t"
-
-                "1:                                     \n\t"
-                "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
-                "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
-                "movq (%0), %%mm0                       \n\t" // L0
-                PAVGB((%1), %%mm0)                            // L0
-                "movq (%0, %2), %%mm1                   \n\t" // L1
-                PAVGB((%1, %2), %%mm1)                        // L1
-                "movq (%0, %2, 2), %%mm2                \n\t" // L2
-                PAVGB((%1, %2, 2), %%mm2)                     // L2
-                "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
-                PAVGB((%1, %%REGa), %%mm3)                    // L3
-                "movq (%0, %2, 4), %%mm4                \n\t" // L4
-                PAVGB((%1, %2, 4), %%mm4)                     // L4
-                "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
-                PAVGB((%1, %%REGd), %%mm5)                    // L5
-                "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
-                PAVGB((%1, %%REGa, 2), %%mm6)                 // L6
-                "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
-                PAVGB((%1, %%REGc), %%mm7)                    // L7
-                "movq %%mm0, (%1)                       \n\t" // R0
-                "movq %%mm1, (%1, %2)                   \n\t" // R1
-                "movq %%mm2, (%1, %2, 2)                \n\t" // R2
-                "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
-                "movq %%mm4, (%1, %2, 4)                \n\t" // R4
-                "movq %%mm5, (%1, %%"REG_d")            \n\t" // R5
-                "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // R6
-                "movq %%mm7, (%1, %%"REG_c")            \n\t" // R7
-                "movq %%mm0, (%0)                       \n\t" // L0
-                "movq %%mm1, (%0, %2)                   \n\t" // L1
-                "movq %%mm2, (%0, %2, 2)                \n\t" // L2
-                "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
-                "movq %%mm4, (%0, %2, 4)                \n\t" // L4
-                "movq %%mm5, (%0, %%"REG_d")            \n\t" // L5
-                "movq %%mm6, (%0, %%"REG_a", 2)         \n\t" // L6
-                "movq %%mm7, (%0, %%"REG_c")            \n\t" // L7
-                "jmp 4f                                 \n\t"
-
-                "2:                                     \n\t"
-                "cmpl 508(%%"REG_d"), %%ecx             \n\t"
-                " jb 3f                                 \n\t"
-
-                "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
-                "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
-                "movq (%0), %%mm0                       \n\t" // L0
-                "movq (%0, %2), %%mm1                   \n\t" // L1
-                "movq (%0, %2, 2), %%mm2                \n\t" // L2
-                "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
-                "movq (%1), %%mm4                       \n\t" // R0
-                "movq (%1, %2), %%mm5                   \n\t" // R1
-                "movq (%1, %2, 2), %%mm6                \n\t" // R2
-                "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
-                PAVGB(%%mm4, %%mm0)
-                PAVGB(%%mm5, %%mm1)
-                PAVGB(%%mm6, %%mm2)
-                PAVGB(%%mm7, %%mm3)
-                PAVGB(%%mm4, %%mm0)
-                PAVGB(%%mm5, %%mm1)
-                PAVGB(%%mm6, %%mm2)
-                PAVGB(%%mm7, %%mm3)
-                "movq %%mm0, (%1)                       \n\t" // R0
-                "movq %%mm1, (%1, %2)                   \n\t" // R1
-                "movq %%mm2, (%1, %2, 2)                \n\t" // R2
-                "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
-                "movq %%mm0, (%0)                       \n\t" // L0
-                "movq %%mm1, (%0, %2)                   \n\t" // L1
-                "movq %%mm2, (%0, %2, 2)                \n\t" // L2
-                "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
-
-                "movq (%0, %2, 4), %%mm0                \n\t" // L4
-                "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
-                "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
-                "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
-                "movq (%1, %2, 4), %%mm4                \n\t" // R4
-                "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
-                "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
-                "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
-                PAVGB(%%mm4, %%mm0)
-                PAVGB(%%mm5, %%mm1)
-                PAVGB(%%mm6, %%mm2)
-                PAVGB(%%mm7, %%mm3)
-                PAVGB(%%mm4, %%mm0)
-                PAVGB(%%mm5, %%mm1)
-                PAVGB(%%mm6, %%mm2)
-                PAVGB(%%mm7, %%mm3)
-                "movq %%mm0, (%1, %2, 4)                \n\t" // R4
-                "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
-                "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
-                "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
-                "movq %%mm0, (%0, %2, 4)                \n\t" // L4
-                "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
-                "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
-                "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
-                "jmp 4f                                 \n\t"
-
-                "3:                                     \n\t"
-                "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
-                "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
-                "movq (%0), %%mm0                       \n\t" // L0
-                "movq (%0, %2), %%mm1                   \n\t" // L1
-                "movq (%0, %2, 2), %%mm2                \n\t" // L2
-                "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
-                "movq (%1), %%mm4                       \n\t" // R0
-                "movq (%1, %2), %%mm5                   \n\t" // R1
-                "movq (%1, %2, 2), %%mm6                \n\t" // R2
-                "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
-                PAVGB(%%mm4, %%mm0)
-                PAVGB(%%mm5, %%mm1)
-                PAVGB(%%mm6, %%mm2)
-                PAVGB(%%mm7, %%mm3)
-                PAVGB(%%mm4, %%mm0)
-                PAVGB(%%mm5, %%mm1)
-                PAVGB(%%mm6, %%mm2)
-                PAVGB(%%mm7, %%mm3)
-                PAVGB(%%mm4, %%mm0)
-                PAVGB(%%mm5, %%mm1)
-                PAVGB(%%mm6, %%mm2)
-                PAVGB(%%mm7, %%mm3)
-                "movq %%mm0, (%1)                       \n\t" // R0
-                "movq %%mm1, (%1, %2)                   \n\t" // R1
-                "movq %%mm2, (%1, %2, 2)                \n\t" // R2
-                "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
-                "movq %%mm0, (%0)                       \n\t" // L0
-                "movq %%mm1, (%0, %2)                   \n\t" // L1
-                "movq %%mm2, (%0, %2, 2)                \n\t" // L2
-                "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
-
-                "movq (%0, %2, 4), %%mm0                \n\t" // L4
-                "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
-                "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
-                "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
-                "movq (%1, %2, 4), %%mm4                \n\t" // R4
-                "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
-                "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
-                "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
-                PAVGB(%%mm4, %%mm0)
-                PAVGB(%%mm5, %%mm1)
-                PAVGB(%%mm6, %%mm2)
-                PAVGB(%%mm7, %%mm3)
-                PAVGB(%%mm4, %%mm0)
-                PAVGB(%%mm5, %%mm1)
-                PAVGB(%%mm6, %%mm2)
-                PAVGB(%%mm7, %%mm3)
-                PAVGB(%%mm4, %%mm0)
-                PAVGB(%%mm5, %%mm1)
-                PAVGB(%%mm6, %%mm2)
-                PAVGB(%%mm7, %%mm3)
-                "movq %%mm0, (%1, %2, 4)                \n\t" // R4
-                "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
-                "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
-                "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
-                "movq %%mm0, (%0, %2, 4)                \n\t" // L4
-                "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
-                "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
-                "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
-
-                "4:                                     \n\t"
-
-                :: "r" (src), "r" (tempBlured), "r"((long)stride), "m" (tempBluredPast)
-                : "%"REG_a, "%"REG_d, "%"REG_c, "memory"
-                );
-#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
+        "movq %%mm0, %%mm4                      \n\t"
+        "psrlq $32, %%mm0                       \n\t"
+        "paddd %%mm0, %%mm4                     \n\t"
+        "movd %%mm4, %%ecx                      \n\t"
+        "shll $2, %%ecx                         \n\t"
+        "mov %3, %%"REG_d"                      \n\t"
+        "addl -4(%%"REG_d"), %%ecx              \n\t"
+        "addl 4(%%"REG_d"), %%ecx               \n\t"
+        "addl -1024(%%"REG_d"), %%ecx           \n\t"
+        "addl $4, %%ecx                         \n\t"
+        "addl 1024(%%"REG_d"), %%ecx            \n\t"
+        "shrl $3, %%ecx                         \n\t"
+        "movl %%ecx, (%%"REG_d")                \n\t"
+
+//        "mov %3, %%"REG_c"                      \n\t"
+//        "mov %%"REG_c", test                    \n\t"
+//        "jmp 4f                                 \n\t"
+        "cmpl 512(%%"REG_d"), %%ecx             \n\t"
+        " jb 2f                                 \n\t"
+        "cmpl 516(%%"REG_d"), %%ecx             \n\t"
+        " jb 1f                                 \n\t"
+
+        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
+        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
+        "movq (%0), %%mm0                       \n\t" // L0
+        "movq (%0, %2), %%mm1                   \n\t" // L1
+        "movq (%0, %2, 2), %%mm2                \n\t" // L2
+        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
+        "movq (%0, %2, 4), %%mm4                \n\t" // L4
+        "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
+        "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
+        "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
+        "movq %%mm0, (%1)                       \n\t" // L0
+        "movq %%mm1, (%1, %2)                   \n\t" // L1
+        "movq %%mm2, (%1, %2, 2)                \n\t" // L2
+        "movq %%mm3, (%1, %%"REG_a")            \n\t" // L3
+        "movq %%mm4, (%1, %2, 4)                \n\t" // L4
+        "movq %%mm5, (%1, %%"REG_d")            \n\t" // L5
+        "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // L6
+        "movq %%mm7, (%1, %%"REG_c")            \n\t" // L7
+        "jmp 4f                                 \n\t"
+
+        "1:                                     \n\t"
+        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
+        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
+        "movq (%0), %%mm0                       \n\t" // L0
+        PAVGB((%1), %%mm0)                            // L0
+        "movq (%0, %2), %%mm1                   \n\t" // L1
+        PAVGB((%1, %2), %%mm1)                        // L1
+        "movq (%0, %2, 2), %%mm2                \n\t" // L2
+        PAVGB((%1, %2, 2), %%mm2)                     // L2
+        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
+        PAVGB((%1, %%REGa), %%mm3)                    // L3
+        "movq (%0, %2, 4), %%mm4                \n\t" // L4
+        PAVGB((%1, %2, 4), %%mm4)                     // L4
+        "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
+        PAVGB((%1, %%REGd), %%mm5)                    // L5
+        "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
+        PAVGB((%1, %%REGa, 2), %%mm6)                 // L6
+        "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
+        PAVGB((%1, %%REGc), %%mm7)                    // L7
+        "movq %%mm0, (%1)                       \n\t" // R0
+        "movq %%mm1, (%1, %2)                   \n\t" // R1
+        "movq %%mm2, (%1, %2, 2)                \n\t" // R2
+        "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
+        "movq %%mm4, (%1, %2, 4)                \n\t" // R4
+        "movq %%mm5, (%1, %%"REG_d")            \n\t" // R5
+        "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // R6
+        "movq %%mm7, (%1, %%"REG_c")            \n\t" // R7
+        "movq %%mm0, (%0)                       \n\t" // L0
+        "movq %%mm1, (%0, %2)                   \n\t" // L1
+        "movq %%mm2, (%0, %2, 2)                \n\t" // L2
+        "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
+        "movq %%mm4, (%0, %2, 4)                \n\t" // L4
+        "movq %%mm5, (%0, %%"REG_d")            \n\t" // L5
+        "movq %%mm6, (%0, %%"REG_a", 2)         \n\t" // L6
+        "movq %%mm7, (%0, %%"REG_c")            \n\t" // L7
+        "jmp 4f                                 \n\t"
+
+        "2:                                     \n\t"
+        "cmpl 508(%%"REG_d"), %%ecx             \n\t"
+        " jb 3f                                 \n\t"
+
+        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
+        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
+        "movq (%0), %%mm0                       \n\t" // L0
+        "movq (%0, %2), %%mm1                   \n\t" // L1
+        "movq (%0, %2, 2), %%mm2                \n\t" // L2
+        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
+        "movq (%1), %%mm4                       \n\t" // R0
+        "movq (%1, %2), %%mm5                   \n\t" // R1
+        "movq (%1, %2, 2), %%mm6                \n\t" // R2
+        "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
+        PAVGB(%%mm4, %%mm0)
+        PAVGB(%%mm5, %%mm1)
+        PAVGB(%%mm6, %%mm2)
+        PAVGB(%%mm7, %%mm3)
+        PAVGB(%%mm4, %%mm0)
+        PAVGB(%%mm5, %%mm1)
+        PAVGB(%%mm6, %%mm2)
+        PAVGB(%%mm7, %%mm3)
+        "movq %%mm0, (%1)                       \n\t" // R0
+        "movq %%mm1, (%1, %2)                   \n\t" // R1
+        "movq %%mm2, (%1, %2, 2)                \n\t" // R2
+        "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
+        "movq %%mm0, (%0)                       \n\t" // L0
+        "movq %%mm1, (%0, %2)                   \n\t" // L1
+        "movq %%mm2, (%0, %2, 2)                \n\t" // L2
+        "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
+
+        "movq (%0, %2, 4), %%mm0                \n\t" // L4
+        "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
+        "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
+        "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
+        "movq (%1, %2, 4), %%mm4                \n\t" // R4
+        "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
+        "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
+        "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
+        PAVGB(%%mm4, %%mm0)
+        PAVGB(%%mm5, %%mm1)
+        PAVGB(%%mm6, %%mm2)
+        PAVGB(%%mm7, %%mm3)
+        PAVGB(%%mm4, %%mm0)
+        PAVGB(%%mm5, %%mm1)
+        PAVGB(%%mm6, %%mm2)
+        PAVGB(%%mm7, %%mm3)
+        "movq %%mm0, (%1, %2, 4)                \n\t" // R4
+        "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
+        "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
+        "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
+        "movq %%mm0, (%0, %2, 4)                \n\t" // L4
+        "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
+        "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
+        "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
+        "jmp 4f                                 \n\t"
+
+        "3:                                     \n\t"
+        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
+        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
+        "movq (%0), %%mm0                       \n\t" // L0
+        "movq (%0, %2), %%mm1                   \n\t" // L1
+        "movq (%0, %2, 2), %%mm2                \n\t" // L2
+        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
+        "movq (%1), %%mm4                       \n\t" // R0
+        "movq (%1, %2), %%mm5                   \n\t" // R1
+        "movq (%1, %2, 2), %%mm6                \n\t" // R2
+        "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
+        PAVGB(%%mm4, %%mm0)
+        PAVGB(%%mm5, %%mm1)
+        PAVGB(%%mm6, %%mm2)
+        PAVGB(%%mm7, %%mm3)
+        PAVGB(%%mm4, %%mm0)
+        PAVGB(%%mm5, %%mm1)
+        PAVGB(%%mm6, %%mm2)
+        PAVGB(%%mm7, %%mm3)
+        PAVGB(%%mm4, %%mm0)
+        PAVGB(%%mm5, %%mm1)
+        PAVGB(%%mm6, %%mm2)
+        PAVGB(%%mm7, %%mm3)
+        "movq %%mm0, (%1)                       \n\t" // R0
+        "movq %%mm1, (%1, %2)                   \n\t" // R1
+        "movq %%mm2, (%1, %2, 2)                \n\t" // R2
+        "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
+        "movq %%mm0, (%0)                       \n\t" // L0
+        "movq %%mm1, (%0, %2)                   \n\t" // L1
+        "movq %%mm2, (%0, %2, 2)                \n\t" // L2
+        "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
+
+        "movq (%0, %2, 4), %%mm0                \n\t" // L4
+        "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
+        "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
+        "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
+        "movq (%1, %2, 4), %%mm4                \n\t" // R4
+        "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
+        "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
+        "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
+        PAVGB(%%mm4, %%mm0)
+        PAVGB(%%mm5, %%mm1)
+        PAVGB(%%mm6, %%mm2)
+        PAVGB(%%mm7, %%mm3)
+        PAVGB(%%mm4, %%mm0)
+        PAVGB(%%mm5, %%mm1)
+        PAVGB(%%mm6, %%mm2)
+        PAVGB(%%mm7, %%mm3)
+        PAVGB(%%mm4, %%mm0)
+        PAVGB(%%mm5, %%mm1)
+        PAVGB(%%mm6, %%mm2)
+        PAVGB(%%mm7, %%mm3)
+        "movq %%mm0, (%1, %2, 4)                \n\t" // R4
+        "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
+        "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
+        "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
+        "movq %%mm0, (%0, %2, 4)                \n\t" // L4
+        "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
+        "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
+        "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
+
+        "4:                                     \n\t"
+
+        :: "r" (src), "r" (tempBlurred), "r"((x86_reg)stride), "m" (tempBlurredPast)
+        : "%"REG_a, "%"REG_d, "%"REG_c, "memory"
+    );
+#else //HAVE_MMX2 || HAVE_AMD3DNOW
 {
-        int y;
-        int d=0;
-//        int sysd=0;
-        int i;
+    int y;
+    int d=0;
+//    int sysd=0;
+    int i;
 
-        for(y=0; y<8; y++)
-        {
-                int x;
-                for(x=0; x<8; x++)
-                {
-                        int ref= tempBlured[ x + y*stride ];
-                        int cur= src[ x + y*stride ];
-                        int d1=ref - cur;
-//                        if(x==0 || x==7) d1+= d1>>1;
-//                        if(y==0 || y==7) d1+= d1>>1;
-//                        d+= FFABS(d1);
-                        d+= d1*d1;
-//                        sysd+= d1;
-                }
+    for(y=0; y<8; y++){
+        int x;
+        for(x=0; x<8; x++){
+            int ref= tempBlurred[ x + y*stride ];
+            int cur= src[ x + y*stride ];
+            int d1=ref - cur;
+//            if(x==0 || x==7) d1+= d1>>1;
+//            if(y==0 || y==7) d1+= d1>>1;
+//            d+= FFABS(d1);
+            d+= d1*d1;
+//            sysd+= d1;
         }
-        i=d;
-        d=         (
-                4*d
-                +(*(tempBluredPast-256))
-                +(*(tempBluredPast-1))+ (*(tempBluredPast+1))
-                +(*(tempBluredPast+256))
-                +4)>>3;
-        *tempBluredPast=i;
-//        ((*tempBluredPast)*3 + d + 2)>>2;
+    }
+    i=d;
+    d=  (
+        4*d
+        +(*(tempBlurredPast-256))
+        +(*(tempBlurredPast-1))+ (*(tempBlurredPast+1))
+        +(*(tempBlurredPast+256))
+        +4)>>3;
+    *tempBlurredPast=i;
+//    ((*tempBlurredPast)*3 + d + 2)>>2;
 
 /*
 Switch between
@@ -2575,603 +2414,586 @@ Switch between
 64 48 36 27 20 15 11 (33) (approx)
 64 56 49 43 37 33 29 (200) (approx)
 */
-        if(d > maxNoise[1])
-        {
-                if(d < maxNoise[2])
-                {
-                        for(y=0; y<8; y++)
-                        {
-                                int x;
-                                for(x=0; x<8; x++)
-                                {
-                                        int ref= tempBlured[ x + y*stride ];
-                                        int cur= src[ x + y*stride ];
-                                        tempBlured[ x + y*stride ]=
-                                        src[ x + y*stride ]=
-                                                (ref + cur + 1)>>1;
-                                }
-                        }
+    if(d > maxNoise[1]){
+        if(d < maxNoise[2]){
+            for(y=0; y<8; y++){
+                int x;
+                for(x=0; x<8; x++){
+                    int ref= tempBlurred[ x + y*stride ];
+                    int cur= src[ x + y*stride ];
+                    tempBlurred[ x + y*stride ]=
+                    src[ x + y*stride ]=
+                        (ref + cur + 1)>>1;
                 }
-                else
-                {
-                        for(y=0; y<8; y++)
-                        {
-                                int x;
-                                for(x=0; x<8; x++)
-                                {
-                                        tempBlured[ x + y*stride ]= src[ x + y*stride ];
-                                }
-                        }
+            }
+        }else{
+            for(y=0; y<8; y++){
+                int x;
+                for(x=0; x<8; x++){
+                    tempBlurred[ x + y*stride ]= src[ x + y*stride ];
                 }
+            }
         }
-        else
-        {
-                if(d < maxNoise[0])
-                {
-                        for(y=0; y<8; y++)
-                        {
-                                int x;
-                                for(x=0; x<8; x++)
-                                {
-                                        int ref= tempBlured[ x + y*stride ];
-                                        int cur= src[ x + y*stride ];
-                                        tempBlured[ x + y*stride ]=
-                                        src[ x + y*stride ]=
-                                                (ref*7 + cur + 4)>>3;
-                                }
-                        }
+    }else{
+        if(d < maxNoise[0]){
+            for(y=0; y<8; y++){
+                int x;
+                for(x=0; x<8; x++){
+                    int ref= tempBlurred[ x + y*stride ];
+                    int cur= src[ x + y*stride ];
+                    tempBlurred[ x + y*stride ]=
+                    src[ x + y*stride ]=
+                        (ref*7 + cur + 4)>>3;
                 }
-                else
-                {
-                        for(y=0; y<8; y++)
-                        {
-                                int x;
-                                for(x=0; x<8; x++)
-                                {
-                                        int ref= tempBlured[ x + y*stride ];
-                                        int cur= src[ x + y*stride ];
-                                        tempBlured[ x + y*stride ]=
-                                        src[ x + y*stride ]=
-                                                (ref*3 + cur + 2)>>2;
-                                }
-                        }
+            }
+        }else{
+            for(y=0; y<8; y++){
+                int x;
+                for(x=0; x<8; x++){
+                    int ref= tempBlurred[ x + y*stride ];
+                    int cur= src[ x + y*stride ];
+                    tempBlurred[ x + y*stride ]=
+                    src[ x + y*stride ]=
+                        (ref*3 + cur + 2)>>2;
                 }
+            }
         }
+    }
 }
-#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
+#endif //HAVE_MMX2 || HAVE_AMD3DNOW
 }
 #endif //HAVE_ALTIVEC
 
-#ifdef HAVE_MMX
+#if HAVE_MMX
 /**
  * accurate deblock filter
  */
 static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, PPContext *c){
-        int64_t dc_mask, eq_mask, both_masks;
-        int64_t sums[10*8*2];
-        src+= step*3; // src points to begin of the 8x8 Block
+    int64_t dc_mask, eq_mask, both_masks;
+    int64_t sums[10*8*2];
+    src+= step*3; // src points to begin of the 8x8 Block
 //START_TIMER
-asm volatile(
-                "movq %0, %%mm7                         \n\t"
-                "movq %1, %%mm6                         \n\t"
-                : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
-                );
-
-asm volatile(
-                "lea (%2, %3), %%"REG_a"                \n\t"
+    __asm__ volatile(
+        "movq %0, %%mm7                         \n\t"
+        "movq %1, %%mm6                         \n\t"
+        : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
+        );
+
+    __asm__ volatile(
+        "lea (%2, %3), %%"REG_a"                \n\t"
 //      0       1       2       3       4       5       6       7       8       9
 //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2
 
-                "movq (%2), %%mm0                       \n\t"
-                "movq (%%"REG_a"), %%mm1                \n\t"
-                "movq %%mm1, %%mm3                      \n\t"
-                "movq %%mm1, %%mm4                      \n\t"
-                "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
-                "paddb %%mm7, %%mm0                     \n\t"
-                "pcmpgtb %%mm6, %%mm0                   \n\t"
-
-                "movq (%%"REG_a",%3), %%mm2             \n\t"
-                PMAXUB(%%mm2, %%mm4)
-                PMINUB(%%mm2, %%mm3, %%mm5)
-                "psubb %%mm2, %%mm1                     \n\t"
-                "paddb %%mm7, %%mm1                     \n\t"
-                "pcmpgtb %%mm6, %%mm1                   \n\t"
-                "paddb %%mm1, %%mm0                     \n\t"
-
-                "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
-                PMAXUB(%%mm1, %%mm4)
-                PMINUB(%%mm1, %%mm3, %%mm5)
-                "psubb %%mm1, %%mm2                     \n\t"
-                "paddb %%mm7, %%mm2                     \n\t"
-                "pcmpgtb %%mm6, %%mm2                   \n\t"
-                "paddb %%mm2, %%mm0                     \n\t"
-
-                "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
-
-                "movq (%2, %3, 4), %%mm2                \n\t"
-                PMAXUB(%%mm2, %%mm4)
-                PMINUB(%%mm2, %%mm3, %%mm5)
-                "psubb %%mm2, %%mm1                     \n\t"
-                "paddb %%mm7, %%mm1                     \n\t"
-                "pcmpgtb %%mm6, %%mm1                   \n\t"
-                "paddb %%mm1, %%mm0                     \n\t"
-
-                "movq (%%"REG_a"), %%mm1                \n\t"
-                PMAXUB(%%mm1, %%mm4)
-                PMINUB(%%mm1, %%mm3, %%mm5)
-                "psubb %%mm1, %%mm2                     \n\t"
-                "paddb %%mm7, %%mm2                     \n\t"
-                "pcmpgtb %%mm6, %%mm2                   \n\t"
-                "paddb %%mm2, %%mm0                     \n\t"
-
-                "movq (%%"REG_a", %3), %%mm2            \n\t"
-                PMAXUB(%%mm2, %%mm4)
-                PMINUB(%%mm2, %%mm3, %%mm5)
-                "psubb %%mm2, %%mm1                     \n\t"
-                "paddb %%mm7, %%mm1                     \n\t"
-                "pcmpgtb %%mm6, %%mm1                   \n\t"
-                "paddb %%mm1, %%mm0                     \n\t"
-
-                "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
-                PMAXUB(%%mm1, %%mm4)
-                PMINUB(%%mm1, %%mm3, %%mm5)
-                "psubb %%mm1, %%mm2                     \n\t"
-                "paddb %%mm7, %%mm2                     \n\t"
-                "pcmpgtb %%mm6, %%mm2                   \n\t"
-                "paddb %%mm2, %%mm0                     \n\t"
-
-                "movq (%2, %3, 8), %%mm2                \n\t"
-                PMAXUB(%%mm2, %%mm4)
-                PMINUB(%%mm2, %%mm3, %%mm5)
-                "psubb %%mm2, %%mm1                     \n\t"
-                "paddb %%mm7, %%mm1                     \n\t"
-                "pcmpgtb %%mm6, %%mm1                   \n\t"
-                "paddb %%mm1, %%mm0                     \n\t"
-
-                "movq (%%"REG_a", %3, 4), %%mm1         \n\t"
-                "psubb %%mm1, %%mm2                     \n\t"
-                "paddb %%mm7, %%mm2                     \n\t"
-                "pcmpgtb %%mm6, %%mm2                   \n\t"
-                "paddb %%mm2, %%mm0                     \n\t"
-                "psubusb %%mm3, %%mm4                   \n\t"
-
-                "pxor %%mm6, %%mm6                      \n\t"
-                "movq %4, %%mm7                         \n\t" // QP,..., QP
-                "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
-                "psubusb %%mm4, %%mm7                   \n\t" // Diff >=2QP -> 0
-                "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
-                "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
-                "movq %%mm7, %1                         \n\t"
-
-                "movq %5, %%mm7                         \n\t"
-                "punpcklbw %%mm7, %%mm7                 \n\t"
-                "punpcklbw %%mm7, %%mm7                 \n\t"
-                "punpcklbw %%mm7, %%mm7                 \n\t"
-                "psubb %%mm0, %%mm6                     \n\t"
-                "pcmpgtb %%mm7, %%mm6                   \n\t"
-                "movq %%mm6, %0                         \n\t"
-
-                : "=m" (eq_mask), "=m" (dc_mas