cosmetics: Fix indentation to be 4 spaces and consistently place {}.
authorDiego Biurrun <diego@biurrun.de>
Sat, 22 Mar 2008 15:46:34 +0000 (15:46 +0000)
committerDiego Biurrun <diego@biurrun.de>
Sat, 22 Mar 2008 15:46:34 +0000 (15:46 +0000)
Originally committed as revision 12552 to svn://svn.ffmpeg.org/ffmpeg/trunk

libpostproc/postprocess.c
libpostproc/postprocess.h
libpostproc/postprocess_altivec_template.c
libpostproc/postprocess_internal.h
libpostproc/postprocess_template.c

index ec39e1c..7ee3c0e 100644 (file)
@@ -116,65 +116,65 @@ DECLARE_ASM_CONST(8, int, deringThreshold)= 20;
 
 static struct PPFilter filters[]=
 {
-        {"hb", "hdeblock",              1, 1, 3, H_DEBLOCK},
-        {"vb", "vdeblock",              1, 2, 4, V_DEBLOCK},
-/*      {"hr", "rkhdeblock",            1, 1, 3, H_RK1_FILTER},
-        {"vr", "rkvdeblock",            1, 2, 4, V_RK1_FILTER},*/
-        {"h1", "x1hdeblock",            1, 1, 3, H_X1_FILTER},
-        {"v1", "x1vdeblock",            1, 2, 4, V_X1_FILTER},
-        {"ha", "ahdeblock",             1, 1, 3, H_A_DEBLOCK},
-        {"va", "avdeblock",             1, 2, 4, V_A_DEBLOCK},
-        {"dr", "dering",                1, 5, 6, DERING},
-        {"al", "autolevels",            0, 1, 2, LEVEL_FIX},
-        {"lb", "linblenddeint",         1, 1, 4, LINEAR_BLEND_DEINT_FILTER},
-        {"li", "linipoldeint",          1, 1, 4, LINEAR_IPOL_DEINT_FILTER},
-        {"ci", "cubicipoldeint",        1, 1, 4, CUBIC_IPOL_DEINT_FILTER},
-        {"md", "mediandeint",           1, 1, 4, MEDIAN_DEINT_FILTER},
-        {"fd", "ffmpegdeint",           1, 1, 4, FFMPEG_DEINT_FILTER},
-        {"l5", "lowpass5",              1, 1, 4, LOWPASS5_DEINT_FILTER},
-        {"tn", "tmpnoise",              1, 7, 8, TEMP_NOISE_FILTER},
-        {"fq", "forcequant",            1, 0, 0, FORCE_QUANT},
-        {NULL, NULL,0,0,0,0} //End Marker
+    {"hb", "hdeblock",              1, 1, 3, H_DEBLOCK},
+    {"vb", "vdeblock",              1, 2, 4, V_DEBLOCK},
+/*  {"hr", "rkhdeblock",            1, 1, 3, H_RK1_FILTER},
+    {"vr", "rkvdeblock",            1, 2, 4, V_RK1_FILTER},*/
+    {"h1", "x1hdeblock",            1, 1, 3, H_X1_FILTER},
+    {"v1", "x1vdeblock",            1, 2, 4, V_X1_FILTER},
+    {"ha", "ahdeblock",             1, 1, 3, H_A_DEBLOCK},
+    {"va", "avdeblock",             1, 2, 4, V_A_DEBLOCK},
+    {"dr", "dering",                1, 5, 6, DERING},
+    {"al", "autolevels",            0, 1, 2, LEVEL_FIX},
+    {"lb", "linblenddeint",         1, 1, 4, LINEAR_BLEND_DEINT_FILTER},
+    {"li", "linipoldeint",          1, 1, 4, LINEAR_IPOL_DEINT_FILTER},
+    {"ci", "cubicipoldeint",        1, 1, 4, CUBIC_IPOL_DEINT_FILTER},
+    {"md", "mediandeint",           1, 1, 4, MEDIAN_DEINT_FILTER},
+    {"fd", "ffmpegdeint",           1, 1, 4, FFMPEG_DEINT_FILTER},
+    {"l5", "lowpass5",              1, 1, 4, LOWPASS5_DEINT_FILTER},
+    {"tn", "tmpnoise",              1, 7, 8, TEMP_NOISE_FILTER},
+    {"fq", "forcequant",            1, 0, 0, FORCE_QUANT},
+    {NULL, NULL,0,0,0,0} //End Marker
 };
 
 static const char *replaceTable[]=
 {
-        "default",      "hb:a,vb:a,dr:a",
-        "de",           "hb:a,vb:a,dr:a",
-        "fast",         "h1:a,v1:a,dr:a",
-        "fa",           "h1:a,v1:a,dr:a",
-        "ac",           "ha:a:128:7,va:a,dr:a",
-        NULL //End Marker
+    "default",      "hb:a,vb:a,dr:a",
+    "de",           "hb:a,vb:a,dr:a",
+    "fast",         "h1:a,v1:a,dr:a",
+    "fa",           "h1:a,v1:a,dr:a",
+    "ac",           "ha:a:128:7,va:a,dr:a",
+    NULL //End Marker
 };
 
 
 #if defined(ARCH_X86)
 static inline void prefetchnta(void *p)
 {
-        asm volatile(   "prefetchnta (%0)\n\t"
-                : : "r" (p)
-        );
+    asm volatile(   "prefetchnta (%0)\n\t"
+        : : "r" (p)
+    );
 }
 
 static inline void prefetcht0(void *p)
 {
-        asm volatile(   "prefetcht0 (%0)\n\t"
-                : : "r" (p)
-        );
+    asm volatile(   "prefetcht0 (%0)\n\t"
+        : : "r" (p)
+    );
 }
 
 static inline void prefetcht1(void *p)
 {
-        asm volatile(   "prefetcht1 (%0)\n\t"
-                : : "r" (p)
-        );
+    asm volatile(   "prefetcht1 (%0)\n\t"
+        : : "r" (p)
+    );
 }
 
 static inline void prefetcht2(void *p)
 {
-        asm volatile(   "prefetcht2 (%0)\n\t"
-                : : "r" (p)
-        );
+    asm volatile(   "prefetcht2 (%0)\n\t"
+        : : "r" (p)
+    );
 }
 #endif
 
@@ -185,171 +185,167 @@ static inline void prefetcht2(void *p)
  */
 static inline int isHorizDC_C(uint8_t src[], int stride, PPContext *c)
 {
-        int numEq= 0;
-        int y;
-        const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
-        const int dcThreshold= dcOffset*2 + 1;
-
-        for(y=0; y<BLOCK_SIZE; y++)
-        {
-                if(((unsigned)(src[0] - src[1] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[1] - src[2] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[2] - src[3] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[3] - src[4] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[4] - src[5] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[5] - src[6] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[6] - src[7] + dcOffset)) < dcThreshold) numEq++;
-                src+= stride;
-        }
-        return numEq > c->ppMode.flatnessThreshold;
+    int numEq= 0;
+    int y;
+    const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
+    const int dcThreshold= dcOffset*2 + 1;
+
+    for(y=0; y<BLOCK_SIZE; y++){
+        if(((unsigned)(src[0] - src[1] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[1] - src[2] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[2] - src[3] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[3] - src[4] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[4] - src[5] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[5] - src[6] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[6] - src[7] + dcOffset)) < dcThreshold) numEq++;
+        src+= stride;
+    }
+    return numEq > c->ppMode.flatnessThreshold;
 }
 
 /**
  * Check if the middle 8x8 Block in the given 8x16 block is flat
  */
-static inline int isVertDC_C(uint8_t src[], int stride, PPContext *c){
-        int numEq= 0;
-        int y;
-        const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
-        const int dcThreshold= dcOffset*2 + 1;
-
-        src+= stride*4; // src points to begin of the 8x8 Block
-        for(y=0; y<BLOCK_SIZE-1; y++)
-        {
-                if(((unsigned)(src[0] - src[0+stride] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[1] - src[1+stride] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[2] - src[2+stride] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[3] - src[3+stride] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[4] - src[4+stride] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[5] - src[5+stride] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[6] - src[6+stride] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[7] - src[7+stride] + dcOffset)) < dcThreshold) numEq++;
-                src+= stride;
-        }
-        return numEq > c->ppMode.flatnessThreshold;
+static inline int isVertDC_C(uint8_t src[], int stride, PPContext *c)
+{
+    int numEq= 0;
+    int y;
+    const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
+    const int dcThreshold= dcOffset*2 + 1;
+
+    src+= stride*4; // src points to begin of the 8x8 Block
+    for(y=0; y<BLOCK_SIZE-1; y++){
+        if(((unsigned)(src[0] - src[0+stride] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[1] - src[1+stride] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[2] - src[2+stride] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[3] - src[3+stride] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[4] - src[4+stride] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[5] - src[5+stride] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[6] - src[6+stride] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[7] - src[7+stride] + dcOffset)) < dcThreshold) numEq++;
+        src+= stride;
+    }
+    return numEq > c->ppMode.flatnessThreshold;
 }
 
 static inline int isHorizMinMaxOk_C(uint8_t src[], int stride, int QP)
 {
-        int i;
+    int i;
 #if 1
-        for(i=0; i<2; i++){
-                if((unsigned)(src[0] - src[5] + 2*QP) > 4*QP) return 0;
-                src += stride;
-                if((unsigned)(src[2] - src[7] + 2*QP) > 4*QP) return 0;
-                src += stride;
-                if((unsigned)(src[4] - src[1] + 2*QP) > 4*QP) return 0;
-                src += stride;
-                if((unsigned)(src[6] - src[3] + 2*QP) > 4*QP) return 0;
-                src += stride;
-        }
+    for(i=0; i<2; i++){
+        if((unsigned)(src[0] - src[5] + 2*QP) > 4*QP) return 0;
+        src += stride;
+        if((unsigned)(src[2] - src[7] + 2*QP) > 4*QP) return 0;
+        src += stride;
+        if((unsigned)(src[4] - src[1] + 2*QP) > 4*QP) return 0;
+        src += stride;
+        if((unsigned)(src[6] - src[3] + 2*QP) > 4*QP) return 0;
+        src += stride;
+    }
 #else
-        for(i=0; i<8; i++){
-                if((unsigned)(src[0] - src[7] + 2*QP) > 4*QP) return 0;
-                src += stride;
-        }
+    for(i=0; i<8; i++){
+        if((unsigned)(src[0] - src[7] + 2*QP) > 4*QP) return 0;
+        src += stride;
+    }
 #endif
-        return 1;
+    return 1;
 }
 
 static inline int isVertMinMaxOk_C(uint8_t src[], int stride, int QP)
 {
 #if 1
 #if 1
-        int x;
-        src+= stride*4;
-        for(x=0; x<BLOCK_SIZE; x+=4)
-        {
-                if((unsigned)(src[  x + 0*stride] - src[  x + 5*stride] + 2*QP) > 4*QP) return 0;
-                if((unsigned)(src[1+x + 2*stride] - src[1+x + 7*stride] + 2*QP) > 4*QP) return 0;
-                if((unsigned)(src[2+x + 4*stride] - src[2+x + 1*stride] + 2*QP) > 4*QP) return 0;
-                if((unsigned)(src[3+x + 6*stride] - src[3+x + 3*stride] + 2*QP) > 4*QP) return 0;
-        }
+    int x;
+    src+= stride*4;
+    for(x=0; x<BLOCK_SIZE; x+=4){
+        if((unsigned)(src[  x + 0*stride] - src[  x + 5*stride] + 2*QP) > 4*QP) return 0;
+        if((unsigned)(src[1+x + 2*stride] - src[1+x + 7*stride] + 2*QP) > 4*QP) return 0;
+        if((unsigned)(src[2+x + 4*stride] - src[2+x + 1*stride] + 2*QP) > 4*QP) return 0;
+        if((unsigned)(src[3+x + 6*stride] - src[3+x + 3*stride] + 2*QP) > 4*QP) return 0;
+    }
 #else
-        int x;
-        src+= stride*3;
-        for(x=0; x<BLOCK_SIZE; x++)
-        {
-                if((unsigned)(src[x + stride] - src[x + (stride<<3)] + 2*QP) > 4*QP) return 0;
-        }
+    int x;
+    src+= stride*3;
+    for(x=0; x<BLOCK_SIZE; x++){
+        if((unsigned)(src[x + stride] - src[x + (stride<<3)] + 2*QP) > 4*QP) return 0;
+    }
 #endif
-        return 1;
+    return 1;
 #else
-        int x;
-        src+= stride*4;
-        for(x=0; x<BLOCK_SIZE; x++)
-        {
-                int min=255;
-                int max=0;
-                int y;
-                for(y=0; y<8; y++){
-                        int v= src[x + y*stride];
-                        if(v>max) max=v;
-                        if(v<min) min=v;
-                }
-                if(max-min > 2*QP) return 0;
+    int x;
+    src+= stride*4;
+    for(x=0; x<BLOCK_SIZE; x++){
+        int min=255;
+        int max=0;
+        int y;
+        for(y=0; y<8; y++){
+            int v= src[x + y*stride];
+            if(v>max) max=v;
+            if(v<min) min=v;
         }
-        return 1;
+        if(max-min > 2*QP) return 0;
+    }
+    return 1;
 #endif
 }
 
-static inline int horizClassify_C(uint8_t src[], int stride, PPContext *c){
-        if( isHorizDC_C(src, stride, c) ){
-                if( isHorizMinMaxOk_C(src, stride, c->QP) )
-                        return 1;
-                else
-                        return 0;
-        }else{
-                return 2;
-        }
+static inline int horizClassify_C(uint8_t src[], int stride, PPContext *c)
+{
+    if( isHorizDC_C(src, stride, c) ){
+        if( isHorizMinMaxOk_C(src, stride, c->QP) )
+            return 1;
+        else
+            return 0;
+    }else{
+        return 2;
+    }
 }
 
-static inline int vertClassify_C(uint8_t src[], int stride, PPContext *c){
-        if( isVertDC_C(src, stride, c) ){
-                if( isVertMinMaxOk_C(src, stride, c->QP) )
-                        return 1;
-                else
-                        return 0;
-        }else{
-                return 2;
-        }
+static inline int vertClassify_C(uint8_t src[], int stride, PPContext *c)
+{
+    if( isVertDC_C(src, stride, c) ){
+        if( isVertMinMaxOk_C(src, stride, c->QP) )
+            return 1;
+        else
+            return 0;
+    }else{
+        return 2;
+    }
 }
 
 static inline void doHorizDefFilter_C(uint8_t dst[], int stride, PPContext *c)
 {
-        int y;
-        for(y=0; y<BLOCK_SIZE; y++)
-        {
-                const int middleEnergy= 5*(dst[4] - dst[3]) + 2*(dst[2] - dst[5]);
-
-                if(FFABS(middleEnergy) < 8*c->QP)
-                {
-                        const int q=(dst[3] - dst[4])/2;
-                        const int leftEnergy=  5*(dst[2] - dst[1]) + 2*(dst[0] - dst[3]);
-                        const int rightEnergy= 5*(dst[6] - dst[5]) + 2*(dst[4] - dst[7]);
-
-                        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;
-                        }
-
-                        dst[3]-= d;
-                        dst[4]+= d;
-                }
-                dst+= stride;
+    int y;
+    for(y=0; y<BLOCK_SIZE; y++){
+        const int middleEnergy= 5*(dst[4] - dst[3]) + 2*(dst[2] - dst[5]);
+
+        if(FFABS(middleEnergy) < 8*c->QP){
+            const int q=(dst[3] - dst[4])/2;
+            const int leftEnergy=  5*(dst[2] - dst[1]) + 2*(dst[0] - dst[3]);
+            const int rightEnergy= 5*(dst[6] - dst[5]) + 2*(dst[4] - dst[7]);
+
+            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;
+            }
+
+            dst[3]-= d;
+            dst[4]+= d;
         }
+        dst+= stride;
+    }
 }
 
 /**
@@ -358,35 +354,34 @@ static inline void doHorizDefFilter_C(uint8_t dst[], int stride, PPContext *c)
  */
 static inline void doHorizLowPass_C(uint8_t dst[], int stride, PPContext *c)
 {
-        int y;
-        for(y=0; y<BLOCK_SIZE; y++)
-        {
-                const int first= FFABS(dst[-1] - dst[0]) < c->QP ? dst[-1] : dst[0];
-                const int last= FFABS(dst[8] - dst[7]) < c->QP ? dst[8] : dst[7];
-
-                int sums[10];
-                sums[0] = 4*first + dst[0] + dst[1] + dst[2] + 4;
-                sums[1] = sums[0] - first  + dst[3];
-                sums[2] = sums[1] - first  + dst[4];
-                sums[3] = sums[2] - first  + dst[5];
-                sums[4] = sums[3] - first  + dst[6];
-                sums[5] = sums[4] - dst[0] + dst[7];
-                sums[6] = sums[5] - dst[1] + last;
-                sums[7] = sums[6] - dst[2] + last;
-                sums[8] = sums[7] - dst[3] + last;
-                sums[9] = sums[8] - dst[4] + last;
-
-                dst[0]= (sums[0] + sums[2] + 2*dst[0])>>4;
-                dst[1]= (sums[1] + sums[3] + 2*dst[1])>>4;
-                dst[2]= (sums[2] + sums[4] + 2*dst[2])>>4;
-                dst[3]= (sums[3] + sums[5] + 2*dst[3])>>4;
-                dst[4]= (sums[4] + sums[6] + 2*dst[4])>>4;
-                dst[5]= (sums[5] + sums[7] + 2*dst[5])>>4;
-                dst[6]= (sums[6] + sums[8] + 2*dst[6])>>4;
-                dst[7]= (sums[7] + sums[9] + 2*dst[7])>>4;
-
-                dst+= stride;
-        }
+    int y;
+    for(y=0; y<BLOCK_SIZE; y++){
+        const int first= FFABS(dst[-1] - dst[0]) < c->QP ? dst[-1] : dst[0];
+        const int last= FFABS(dst[8] - dst[7]) < c->QP ? dst[8] : dst[7];
+
+        int sums[10];
+        sums[0] = 4*first + dst[0] + dst[1] + dst[2] + 4;
+        sums[1] = sums[0] - first  + dst[3];
+        sums[2] = sums[1] - first  + dst[4];
+        sums[3] = sums[2] - first  + dst[5];
+        sums[4] = sums[3] - first  + dst[6];
+        sums[5] = sums[4] - dst[0] + dst[7];
+        sums[6] = sums[5] - dst[1] + last;
+        sums[7] = sums[6] - dst[2] + last;
+        sums[8] = sums[7] - dst[3] + last;
+        sums[9] = sums[8] - dst[4] + last;
+
+        dst[0]= (sums[0] + sums[2] + 2*dst[0])>>4;
+        dst[1]= (sums[1] + sums[3] + 2*dst[1])>>4;
+        dst[2]= (sums[2] + sums[4] + 2*dst[2])>>4;
+        dst[3]= (sums[3] + sums[5] + 2*dst[3])>>4;
+        dst[4]= (sums[4] + sums[6] + 2*dst[4])>>4;
+        dst[5]= (sums[5] + sums[7] + 2*dst[5])>>4;
+        dst[6]= (sums[6] + sums[8] + 2*dst[6])>>4;
+        dst[7]= (sums[7] + sums[9] + 2*dst[7])>>4;
+
+        dst+= stride;
+    }
 }
 
 /**
@@ -399,161 +394,154 @@ static inline void doHorizLowPass_C(uint8_t dst[], int stride, PPContext *c)
  */
 static inline void horizX1Filter(uint8_t *src, int stride, int QP)
 {
-        int y;
-        static uint64_t *lut= NULL;
-        if(lut==NULL)
+    int y;
+    static uint64_t *lut= NULL;
+    if(lut==NULL)
+    {
+        int i;
+        lut = av_malloc(256*8);
+        for(i=0; i<256; i++)
         {
-                int i;
-                lut = av_malloc(256*8);
-                for(i=0; i<256; i++)
-                {
-                        int v= i < 128 ? 2*i : 2*(i-256);
+            int v= i < 128 ? 2*i : 2*(i-256);
 /*
 //Simulate 112242211 9-Tap filter
-                        uint64_t a= (v/16) & 0xFF;
-                        uint64_t b= (v/8) & 0xFF;
-                        uint64_t c= (v/4) & 0xFF;
-                        uint64_t d= (3*v/8) & 0xFF;
+            uint64_t a= (v/16)  & 0xFF;
+            uint64_t b= (v/8)   & 0xFF;
+            uint64_t c= (v/4)   & 0xFF;
+            uint64_t d= (3*v/8) & 0xFF;
 */
 //Simulate piecewise linear interpolation
-                        uint64_t a= (v/16) & 0xFF;
-                        uint64_t b= (v*3/16) & 0xFF;
-                        uint64_t c= (v*5/16) & 0xFF;
-                        uint64_t d= (7*v/16) & 0xFF;
-                        uint64_t A= (0x100 - a)&0xFF;
-                        uint64_t B= (0x100 - b)&0xFF;
-                        uint64_t C= (0x100 - c)&0xFF;
-                        uint64_t D= (0x100 - c)&0xFF;
-
-                        lut[i]   = (a<<56) | (b<<48) | (c<<40) | (d<<32) |
-                                (D<<24) | (C<<16) | (B<<8) | (A);
-                        //lut[i] = (v<<32) | (v<<24);
-                }
+            uint64_t a= (v/16)   & 0xFF;
+            uint64_t b= (v*3/16) & 0xFF;
+            uint64_t c= (v*5/16) & 0xFF;
+            uint64_t d= (7*v/16) & 0xFF;
+            uint64_t A= (0x100 - a)&0xFF;
+            uint64_t B= (0x100 - b)&0xFF;
+            uint64_t C= (0x100 - c)&0xFF;
+            uint64_t D= (0x100 - c)&0xFF;
+
+            lut[i]   = (a<<56) | (b<<48) | (c<<40) | (d<<32) |
+                       (D<<24) | (C<<16) | (B<<8)  | (A);
+            //lut[i] = (v<<32) | (v<<24);
         }
+    }
 
-        for(y=0; y<BLOCK_SIZE; y++)
-        {
-                int a= src[1] - src[2];
-                int b= src[3] - src[4];
-                int c= src[5] - src[6];
-
-                int d= FFMAX(FFABS(b) - (FFABS(a) + FFABS(c))/2, 0);
+    for(y=0; y<BLOCK_SIZE; y++){
+        int a= src[1] - src[2];
+        int b= src[3] - src[4];
+        int c= src[5] - src[6];
 
-                if(d < QP)
-                {
-                        int v = d * FFSIGN(-b);
+        int d= FFMAX(FFABS(b) - (FFABS(a) + FFABS(c))/2, 0);
 
-                        src[1] +=v/8;
-                        src[2] +=v/4;
-                        src[3] +=3*v/8;
-                        src[4] -=3*v/8;
-                        src[5] -=v/4;
-                        src[6] -=v/8;
+        if(d < QP){
+            int v = d * FFSIGN(-b);
 
-                }
-                src+=stride;
+            src[1] +=v/8;
+            src[2] +=v/4;
+            src[3] +=3*v/8;
+            src[4] -=3*v/8;
+            src[5] -=v/4;
+            src[6] -=v/8;
         }
+        src+=stride;
+    }
 }
 
 /**
  * accurate deblock filter
  */
 static av_always_inline void do_a_deblock_C(uint8_t *src, int step, int stride, PPContext *c){
-        int y;
-        const int QP= c->QP;
-        const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
-        const int dcThreshold= dcOffset*2 + 1;
+    int y;
+    const int QP= c->QP;
+    const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
+    const int dcThreshold= dcOffset*2 + 1;
 //START_TIMER
-        src+= step*4; // src points to begin of the 8x8 Block
-        for(y=0; y<8; y++){
-                int numEq= 0;
-
-                if(((unsigned)(src[-1*step] - src[0*step] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[ 0*step] - src[1*step] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[ 1*step] - src[2*step] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[ 2*step] - src[3*step] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[ 3*step] - src[4*step] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[ 4*step] - src[5*step] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[ 5*step] - src[6*step] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[ 6*step] - src[7*step] + dcOffset)) < dcThreshold) numEq++;
-                if(((unsigned)(src[ 7*step] - src[8*step] + dcOffset)) < dcThreshold) numEq++;
-                if(numEq > c->ppMode.flatnessThreshold){
-                        int min, max, x;
-
-                        if(src[0] > src[step]){
-                            max= src[0];
-                            min= src[step];
-                        }else{
-                            max= src[step];
-                            min= src[0];
-                        }
-                        for(x=2; x<8; x+=2){
-                                if(src[x*step] > src[(x+1)*step]){
-                                        if(src[x    *step] > max) max= src[ x   *step];
-                                        if(src[(x+1)*step] < min) min= src[(x+1)*step];
-                                }else{
-                                        if(src[(x+1)*step] > max) max= src[(x+1)*step];
-                                        if(src[ x   *step] < min) min= src[ x   *step];
-                                }
-                        }
-                        if(max-min < 2*QP){
-                                const int first= FFABS(src[-1*step] - src[0]) < QP ? src[-1*step] : src[0];
-                                const int last= FFABS(src[8*step] - src[7*step]) < QP ? src[8*step] : src[7*step];
-
-                                int sums[10];
-                                sums[0] = 4*first + src[0*step] + src[1*step] + src[2*step] + 4;
-                                sums[1] = sums[0] - first       + src[3*step];
-                                sums[2] = sums[1] - first       + src[4*step];
-                                sums[3] = sums[2] - first       + src[5*step];
-                                sums[4] = sums[3] - first       + src[6*step];
-                                sums[5] = sums[4] - src[0*step] + src[7*step];
-                                sums[6] = sums[5] - src[1*step] + last;
-                                sums[7] = sums[6] - src[2*step] + last;
-                                sums[8] = sums[7] - src[3*step] + last;
-                                sums[9] = sums[8] - src[4*step] + last;
-
-                                src[0*step]= (sums[0] + sums[2] + 2*src[0*step])>>4;
-                                src[1*step]= (sums[1] + sums[3] + 2*src[1*step])>>4;
-                                src[2*step]= (sums[2] + sums[4] + 2*src[2*step])>>4;
-                                src[3*step]= (sums[3] + sums[5] + 2*src[3*step])>>4;
-                                src[4*step]= (sums[4] + sums[6] + 2*src[4*step])>>4;
-                                src[5*step]= (sums[5] + sums[7] + 2*src[5*step])>>4;
-                                src[6*step]= (sums[6] + sums[8] + 2*src[6*step])>>4;
-                                src[7*step]= (sums[7] + sums[9] + 2*src[7*step])>>4;
-                        }
+    src+= step*4; // src points to begin of the 8x8 Block
+    for(y=0; y<8; y++){
+        int numEq= 0;
+
+        if(((unsigned)(src[-1*step] - src[0*step] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[ 0*step] - src[1*step] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[ 1*step] - src[2*step] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[ 2*step] - src[3*step] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[ 3*step] - src[4*step] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[ 4*step] - src[5*step] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[ 5*step] - src[6*step] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[ 6*step] - src[7*step] + dcOffset)) < dcThreshold) numEq++;
+        if(((unsigned)(src[ 7*step] - src[8*step] + dcOffset)) < dcThreshold) numEq++;
+        if(numEq > c->ppMode.flatnessThreshold){
+            int min, max, x;
+
+            if(src[0] > src[step]){
+                max= src[0];
+                min= src[step];
+            }else{
+                max= src[step];
+                min= src[0];
+            }
+            for(x=2; x<8; x+=2){
+                if(src[x*step] > src[(x+1)*step]){
+                        if(src[x    *step] > max) max= src[ x   *step];
+                        if(src[(x+1)*step] < min) min= src[(x+1)*step];
                 }else{
-                        const int middleEnergy= 5*(src[4*step] - src[3*step]) + 2*(src[2*step] - src[5*step]);
-
-                        if(FFABS(middleEnergy) < 8*QP)
-                        {
-                                const int q=(src[3*step] - src[4*step])/2;
-                                const int leftEnergy=  5*(src[2*step] - src[1*step]) + 2*(src[0*step] - src[3*step]);
-                                const int rightEnergy= 5*(src[6*step] - src[5*step]) + 2*(src[4*step] - src[7*step]);
-
-                                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[3*step]-= d;
-                                src[4*step]+= d;
-                        }
+                        if(src[(x+1)*step] > max) max= src[(x+1)*step];
+                        if(src[ x   *step] < min) min= src[ x   *step];
+                }
+            }
+            if(max-min < 2*QP){
+                const int first= FFABS(src[-1*step] - src[0]) < QP ? src[-1*step] : src[0];
+                const int last= FFABS(src[8*step] - src[7*step]) < QP ? src[8*step] : src[7*step];
+
+                int sums[10];
+                sums[0] = 4*first + src[0*step] + src[1*step] + src[2*step] + 4;
+                sums[1] = sums[0] - first       + src[3*step];
+                sums[2] = sums[1] - first       + src[4*step];
+                sums[3] = sums[2] - first       + src[5*step];
+                sums[4] = sums[3] - first       + src[6*step];
+                sums[5] = sums[4] - src[0*step] + src[7*step];
+                sums[6] = sums[5] - src[1*step] + last;
+                sums[7] = sums[6] - src[2*step] + last;
+                sums[8] = sums[7] - src[3*step] + last;
+                sums[9] = sums[8] - src[4*step] + last;
+
+                src[0*step]= (sums[0] + sums[2] + 2*src[0*step])>>4;
+                src[1*step]= (sums[1] + sums[3] + 2*src[1*step])>>4;
+                src[2*step]= (sums[2] + sums[4] + 2*src[2*step])>>4;
+                src[3*step]= (sums[3] + sums[5] + 2*src[3*step])>>4;
+                src[4*step]= (sums[4] + sums[6] + 2*src[4*step])>>4;
+                src[5*step]= (sums[5] + sums[7] + 2*src[5*step])>>4;
+                src[6*step]= (sums[6] + sums[8] + 2*src[6*step])>>4;
+                src[7*step]= (sums[7] + sums[9] + 2*src[7*step])>>4;
+            }
+        }else{
+            const int middleEnergy= 5*(src[4*step] - src[3*step]) + 2*(src[2*step] - src[5*step]);
+
+            if(FFABS(middleEnergy) < 8*QP){
+                const int q=(src[3*step] - src[4*step])/2;
+                const int leftEnergy=  5*(src[2*step] - src[1*step]) + 2*(src[0*step] - src[3*step]);
+                const int rightEnergy= 5*(src[6*step] - src[5*step]) + 2*(src[4*step] - src[7*step]);
+
+                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 += stride;
+                src[3*step]-= d;
+                src[4*step]+= d;
+            }
         }
+
+        src += stride;
+    }
 /*if(step==16){
     STOP_TIMER("step16")
 }else{
@@ -642,43 +630,43 @@ static av_always_inline void do_a_deblock_C(uint8_t *src, int step, int stride,
 static inline void postProcess(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
         const QP_STORE_T QPs[], int QPStride, int isColor, pp_mode_t *vm, pp_context_t *vc)
 {
-        PPContext *c= (PPContext *)vc;
-        PPMode *ppMode= (PPMode *)vm;
-        c->ppMode= *ppMode; //FIXME
+    PPContext *c= (PPContext *)vc;
+    PPMode *ppMode= (PPMode *)vm;
+    c->ppMode= *ppMode; //FIXME
 
-        // Using ifs here as they are faster than function pointers although the
-        // difference would not be measurable here but it is much better because
-        // someone might exchange the CPU whithout restarting MPlayer ;)
+    // Using ifs here as they are faster than function pointers although the
+    // difference would not be measurable here but it is much better because
+    // someone might exchange the CPU whithout restarting MPlayer ;)
 #ifdef RUNTIME_CPUDETECT
 #if defined(ARCH_X86)
-        // ordered per speed fastest first
-        if(c->cpuCaps & PP_CPU_CAPS_MMX2)
-                postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
-        else if(c->cpuCaps & PP_CPU_CAPS_3DNOW)
-                postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
-        else if(c->cpuCaps & PP_CPU_CAPS_MMX)
-                postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
-        else
-                postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
+    // ordered per speed fastest first
+    if(c->cpuCaps & PP_CPU_CAPS_MMX2)
+        postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
+    else if(c->cpuCaps & PP_CPU_CAPS_3DNOW)
+        postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
+    else if(c->cpuCaps & PP_CPU_CAPS_MMX)
+        postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
+    else
+        postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
 #else
 #ifdef HAVE_ALTIVEC
-        if(c->cpuCaps & PP_CPU_CAPS_ALTIVEC)
-                postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
-        else
+    if(c->cpuCaps & PP_CPU_CAPS_ALTIVEC)
+            postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
+    else
 #endif
-                postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
+            postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
 #endif
 #else //RUNTIME_CPUDETECT
 #ifdef HAVE_MMX2
-                postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
+            postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
 #elif defined (HAVE_3DNOW)
-                postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
+            postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
 #elif defined (HAVE_MMX)
-                postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
+            postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
 #elif defined (HAVE_ALTIVEC)
-                postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
+            postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
 #else
-                postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
+            postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
 #endif
 #endif //!RUNTIME_CPUDETECT
 }
@@ -738,194 +726,177 @@ const char pp_help[] =
 
 pp_mode_t *pp_get_mode_by_name_and_quality(const char *name, int quality)
 {
-        char temp[GET_MODE_BUFFER_SIZE];
-        char *p= temp;
-        static const char filterDelimiters[] = ",/";
-        static const char optionDelimiters[] = ":";
-        struct PPMode *ppMode;
-        char *filterToken;
-
-        ppMode= av_malloc(sizeof(PPMode));
-
-        ppMode->lumMode= 0;
-        ppMode->chromMode= 0;
-        ppMode->maxTmpNoise[0]= 700;
-        ppMode->maxTmpNoise[1]= 1500;
-        ppMode->maxTmpNoise[2]= 3000;
-        ppMode->maxAllowedY= 234;
-        ppMode->minAllowedY= 16;
-        ppMode->baseDcDiff= 256/8;
-        ppMode->flatnessThreshold= 56-16-1;
-        ppMode->maxClippedThreshold= 0.01;
-        ppMode->error=0;
-
-        strncpy(temp, name, GET_MODE_BUFFER_SIZE);
-
-        av_log(NULL, AV_LOG_DEBUG, "pp: %s\n", name);
-
-        for(;;){
-                char *filterName;
-                int q= 1000000; //PP_QUALITY_MAX;
-                int chrom=-1;
-                int luma=-1;
-                char *option;
-                char *options[OPTIONS_ARRAY_SIZE];
-                int i;
-                int filterNameOk=0;
-                int numOfUnknownOptions=0;
-                int enable=1; //does the user want us to enabled or disabled the filter
-
-                filterToken= strtok(p, filterDelimiters);
-                if(filterToken == NULL) break;
-                p+= strlen(filterToken) + 1; // p points to next filterToken
-                filterName= strtok(filterToken, optionDelimiters);
-                av_log(NULL, AV_LOG_DEBUG, "pp: %s::%s\n", filterToken, filterName);
-
-                if(*filterName == '-')
-                {
-                        enable=0;
-                        filterName++;
-                }
+    char temp[GET_MODE_BUFFER_SIZE];
+    char *p= temp;
+    static const char filterDelimiters[] = ",/";
+    static const char optionDelimiters[] = ":";
+    struct PPMode *ppMode;
+    char *filterToken;
+
+    ppMode= av_malloc(sizeof(PPMode));
+
+    ppMode->lumMode= 0;
+    ppMode->chromMode= 0;
+    ppMode->maxTmpNoise[0]= 700;
+    ppMode->maxTmpNoise[1]= 1500;
+    ppMode->maxTmpNoise[2]= 3000;
+    ppMode->maxAllowedY= 234;
+    ppMode->minAllowedY= 16;
+    ppMode->baseDcDiff= 256/8;
+    ppMode->flatnessThreshold= 56-16-1;
+    ppMode->maxClippedThreshold= 0.01;
+    ppMode->error=0;
+
+    strncpy(temp, name, GET_MODE_BUFFER_SIZE);
+
+    av_log(NULL, AV_LOG_DEBUG, "pp: %s\n", name);
+
+    for(;;){
+        char *filterName;
+        int q= 1000000; //PP_QUALITY_MAX;
+        int chrom=-1;
+        int luma=-1;
+        char *option;
+        char *options[OPTIONS_ARRAY_SIZE];
+        int i;
+        int filterNameOk=0;
+        int numOfUnknownOptions=0;
+        int enable=1; //does the user want us to enabled or disabled the filter
+
+        filterToken= strtok(p, filterDelimiters);
+        if(filterToken == NULL) break;
+        p+= strlen(filterToken) + 1; // p points to next filterToken
+        filterName= strtok(filterToken, optionDelimiters);
+        av_log(NULL, AV_LOG_DEBUG, "pp: %s::%s\n", filterToken, filterName);
+
+        if(*filterName == '-'){
+            enable=0;
+            filterName++;
+        }
 
-                for(;;){ //for all options
-                        option= strtok(NULL, optionDelimiters);
-                        if(option == NULL) break;
-
-                        av_log(NULL, AV_LOG_DEBUG, "pp: option: %s\n", option);
-                        if(!strcmp("autoq", option) || !strcmp("a", option)) q= quality;
-                        else if(!strcmp("nochrom", option) || !strcmp("y", option)) chrom=0;
-                        else if(!strcmp("chrom", option) || !strcmp("c", option)) chrom=1;
-                        else if(!strcmp("noluma", option) || !strcmp("n", option)) luma=0;
-                        else
-                        {
-                                options[numOfUnknownOptions] = option;
-                                numOfUnknownOptions++;
-                        }
-                        if(numOfUnknownOptions >= OPTIONS_ARRAY_SIZE-1) break;
+        for(;;){ //for all options
+            option= strtok(NULL, optionDelimiters);
+            if(option == NULL) break;
+
+            av_log(NULL, AV_LOG_DEBUG, "pp: option: %s\n", option);
+            if(!strcmp("autoq", option) || !strcmp("a", option)) q= quality;
+            else if(!strcmp("nochrom", option) || !strcmp("y", option)) chrom=0;
+            else if(!strcmp("chrom", option) || !strcmp("c", option)) chrom=1;
+            else if(!strcmp("noluma", option) || !strcmp("n", option)) luma=0;
+            else{
+                options[numOfUnknownOptions] = option;
+                numOfUnknownOptions++;
+            }
+            if(numOfUnknownOptions >= OPTIONS_ARRAY_SIZE-1) break;
+        }
+        options[numOfUnknownOptions] = NULL;
+
+        /* replace stuff from the replace Table */
+        for(i=0; replaceTable[2*i]!=NULL; i++){
+            if(!strcmp(replaceTable[2*i], filterName)){
+                int newlen= strlen(replaceTable[2*i + 1]);
+                int plen;
+                int spaceLeft;
+
+                if(p==NULL) p= temp, *p=0;      //last filter
+                else p--, *p=',';               //not last filter
+
+                plen= strlen(p);
+                spaceLeft= p - temp + plen;
+                if(spaceLeft + newlen  >= GET_MODE_BUFFER_SIZE){
+                    ppMode->error++;
+                    break;
                 }
-                options[numOfUnknownOptions] = NULL;
+                memmove(p + newlen, p, plen+1);
+                memcpy(p, replaceTable[2*i + 1], newlen);
+                filterNameOk=1;
+            }
+        }
 
-                /* replace stuff from the replace Table */
-                for(i=0; replaceTable[2*i]!=NULL; i++)
-                {
-                        if(!strcmp(replaceTable[2*i], filterName))
-                        {
-                                int newlen= strlen(replaceTable[2*i + 1]);
-                                int plen;
-                                int spaceLeft;
-
-                                if(p==NULL) p= temp, *p=0;      //last filter
-                                else p--, *p=',';               //not last filter
-
-                                plen= strlen(p);
-                                spaceLeft= p - temp + plen;
-                                if(spaceLeft + newlen  >= GET_MODE_BUFFER_SIZE)
-                                {
-                                        ppMode->error++;
-                                        break;
-                                }
-                                memmove(p + newlen, p, plen+1);
-                                memcpy(p, replaceTable[2*i + 1], newlen);
-                                filterNameOk=1;
+        for(i=0; filters[i].shortName!=NULL; i++){
+            if(   !strcmp(filters[i].longName, filterName)
+               || !strcmp(filters[i].shortName, filterName)){
+                ppMode->lumMode &= ~filters[i].mask;
+                ppMode->chromMode &= ~filters[i].mask;
+
+                filterNameOk=1;
+                if(!enable) break; // user wants to disable it
+
+                if(q >= filters[i].minLumQuality && luma)
+                    ppMode->lumMode|= filters[i].mask;
+                if(chrom==1 || (chrom==-1 && filters[i].chromDefault))
+                    if(q >= filters[i].minChromQuality)
+                            ppMode->chromMode|= filters[i].mask;
+
+                if(filters[i].mask == LEVEL_FIX){
+                    int o;
+                    ppMode->minAllowedY= 16;
+                    ppMode->maxAllowedY= 234;
+                    for(o=0; options[o]!=NULL; o++){
+                        if(  !strcmp(options[o],"fullyrange")
+                           ||!strcmp(options[o],"f")){
+                            ppMode->minAllowedY= 0;
+                            ppMode->maxAllowedY= 255;
+                            numOfUnknownOptions--;
                         }
+                    }
                 }
-
-                for(i=0; filters[i].shortName!=NULL; i++)
+                else if(filters[i].mask == TEMP_NOISE_FILTER)
                 {
-                        if(   !strcmp(filters[i].longName, filterName)
-                           || !strcmp(filters[i].shortName, filterName))
-                        {
-                                ppMode->lumMode &= ~filters[i].mask;
-                                ppMode->chromMode &= ~filters[i].mask;
-
-                                filterNameOk=1;
-                                if(!enable) break; // user wants to disable it
-
-                                if(q >= filters[i].minLumQuality && luma)
-                                        ppMode->lumMode|= filters[i].mask;
-                                if(chrom==1 || (chrom==-1 && filters[i].chromDefault))
-                                        if(q >= filters[i].minChromQuality)
-                                                ppMode->chromMode|= filters[i].mask;
-
-                                if(filters[i].mask == LEVEL_FIX)
-                                {
-                                        int o;
-                                        ppMode->minAllowedY= 16;
-                                        ppMode->maxAllowedY= 234;
-                                        for(o=0; options[o]!=NULL; o++)
-                                        {
-                                                if(  !strcmp(options[o],"fullyrange")
-                                                   ||!strcmp(options[o],"f"))
-                                                {
-                                                        ppMode->minAllowedY= 0;
-                                                        ppMode->maxAllowedY= 255;
-                                                        numOfUnknownOptions--;
-                                                }
-                                        }
-                                }
-                                else if(filters[i].mask == TEMP_NOISE_FILTER)
-                                {
-                                        int o;
-                                        int numOfNoises=0;
-
-                                        for(o=0; options[o]!=NULL; o++)
-                                        {
-                                                char *tail;
-                                                ppMode->maxTmpNoise[numOfNoises]=
-                                                        strtol(options[o], &tail, 0);
-                                                if(tail!=options[o])
-                                                {
-                                                        numOfNoises++;
-                                                        numOfUnknownOptions--;
-                                                        if(numOfNoises >= 3) break;
-                                                }
-                                        }
-                                }
-                                else if(filters[i].mask == V_DEBLOCK   || filters[i].mask == H_DEBLOCK
-                                     || filters[i].mask == V_A_DEBLOCK || filters[i].mask == H_A_DEBLOCK)
-                                {
-                                        int o;
-
-                                        for(o=0; options[o]!=NULL && o<2; o++)
-                                        {
-                                                char *tail;
-                                                int val= strtol(options[o], &tail, 0);
-                                                if(tail==options[o]) break;
-
-                                                numOfUnknownOptions--;
-                                                if(o==0) ppMode->baseDcDiff= val;
-                                                else ppMode->flatnessThreshold= val;
-                                        }
-                                }
-                                else if(filters[i].mask == FORCE_QUANT)
-                                {
-                                        int o;
-                                        ppMode->forcedQuant= 15;
-
-                                        for(o=0; options[o]!=NULL && o<1; o++)
-                                        {
-                                                char *tail;
-                                                int val= strtol(options[o], &tail, 0);
-                                                if(tail==options[o]) break;
-
-                                                numOfUnknownOptions--;
-                                                ppMode->forcedQuant= val;
-                                        }
-                                }
+                    int o;
+                    int numOfNoises=0;
+
+                    for(o=0; options[o]!=NULL; o++){
+                        char *tail;
+                        ppMode->maxTmpNoise[numOfNoises]=
+                            strtol(options[o], &tail, 0);
+                        if(tail!=options[o]){
+                            numOfNoises++;
+                            numOfUnknownOptions--;
+                            if(numOfNoises >= 3) break;
                         }
+                    }
                 }
-                if(!filterNameOk) ppMode->error++;
-                ppMode->error += numOfUnknownOptions;
-        }
-
-        av_log(NULL, AV_LOG_DEBUG, "pp: lumMode=%X, chromMode=%X\n", ppMode->lumMode, ppMode->chromMode);
-        if(ppMode->error)
-        {
-                av_log(NULL, AV_LOG_ERROR, "%d errors in postprocess string \"%s\"\n", ppMode->error, name);
-                av_free(ppMode);
-                return NULL;
+                else if(filters[i].mask == V_DEBLOCK   || filters[i].mask == H_DEBLOCK
+                     || filters[i].mask == V_A_DEBLOCK || filters[i].mask == H_A_DEBLOCK){
+                    int o;
+
+                    for(o=0; options[o]!=NULL && o<2; o++){
+                        char *tail;
+                        int val= strtol(options[o], &tail, 0);
+                        if(tail==options[o]) break;
+
+                        numOfUnknownOptions--;
+                        if(o==0) ppMode->baseDcDiff= val;
+                        else ppMode->flatnessThreshold= val;
+                    }
+                }
+                else if(filters[i].mask == FORCE_QUANT){
+                    int o;
+                    ppMode->forcedQuant= 15;
+
+                    for(o=0; options[o]!=NULL && o<1; o++){
+                        char *tail;
+                        int val= strtol(options[o], &tail, 0);
+                        if(tail==options[o]) break;
+
+                        numOfUnknownOptions--;
+                        ppMode->forcedQuant= val;
+                    }
+                }
+            }
         }
-        return ppMode;
+        if(!filterNameOk) ppMode->error++;
+        ppMode->error += numOfUnknownOptions;
+    }
+
+    av_log(NULL, AV_LOG_DEBUG, "pp: lumMode=%X, chromMode=%X\n", ppMode->lumMode, ppMode->chromMode);
+    if(ppMode->error){
+        av_log(NULL, AV_LOG_ERROR, "%d errors in postprocess string \"%s\"\n", ppMode->error, name);
+        av_free(ppMode);
+        return NULL;
+    }
+    return ppMode;
 }
 
 void pp_free_mode(pp_mode_t *mode){
@@ -933,36 +904,35 @@ void pp_free_mode(pp_mode_t *mode){
 }
 
 static void reallocAlign(void **p, int alignment, int size){
-        av_free(*p);
-        *p= av_mallocz(size);
+    av_free(*p);
+    *p= av_mallocz(size);
 }
 
 static void reallocBuffers(PPContext *c, int width, int height, int stride, int qpStride){
-        int mbWidth = (width+15)>>4;
-        int mbHeight= (height+15)>>4;
-        int i;
-
-        c->stride= stride;
-        c->qpStride= qpStride;
-
-        reallocAlign((void **)&c->tempDst, 8, stride*24);
-        reallocAlign((void **)&c->tempSrc, 8, stride*24);
-        reallocAlign((void **)&c->tempBlocks, 8, 2*16*8);
-        reallocAlign((void **)&c->yHistogram, 8, 256*sizeof(uint64_t));
-        for(i=0; i<256; i++)
-                c->yHistogram[i]= width*height/64*15/256;
-
-        for(i=0; i<3; i++)
-        {
-                //Note: The +17*1024 is just there so i do not have to worry about r/w over the end.
-                reallocAlign((void **)&c->tempBlured[i], 8, stride*mbHeight*16 + 17*1024);
-                reallocAlign((void **)&c->tempBluredPast[i], 8, 256*((height+7)&(~7))/2 + 17*1024);//FIXME size
-        }
-
-        reallocAlign((void **)&c->deintTemp, 8, 2*width+32);
-        reallocAlign((void **)&c->nonBQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T));
-        reallocAlign((void **)&c->stdQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T));
-        reallocAlign((void **)&c->forcedQPTable, 8, mbWidth*sizeof(QP_STORE_T));
+    int mbWidth = (width+15)>>4;
+    int mbHeight= (height+15)>>4;
+    int i;
+
+    c->stride= stride;
+    c->qpStride= qpStride;
+
+    reallocAlign((void **)&c->tempDst, 8, stride*24);
+    reallocAlign((void **)&c->tempSrc, 8, stride*24);
+    reallocAlign((void **)&c->tempBlocks, 8, 2*16*8);
+    reallocAlign((void **)&c->yHistogram, 8, 256*sizeof(uint64_t));
+    for(i=0; i<256; i++)
+            c->yHistogram[i]= width*height/64*15/256;
+
+    for(i=0; i<3; i++){
+        //Note: The +17*1024 is just there so i do not have to worry about r/w over the end.
+        reallocAlign((void **)&c->tempBlured[i], 8, stride*mbHeight*16 + 17*1024);
+        reallocAlign((void **)&c->tempBluredPast[i], 8, 256*((height+7)&(~7))/2 + 17*1024);//FIXME size
+    }
+
+    reallocAlign((void **)&c->deintTemp, 8, 2*width+32);
+    reallocAlign((void **)&c->nonBQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T));
+    reallocAlign((void **)&c->stdQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T));
+    reallocAlign((void **)&c->forcedQPTable, 8, mbWidth*sizeof(QP_STORE_T));
 }
 
 static const char * context_to_name(void * ptr) {
@@ -972,153 +942,146 @@ static const char * context_to_name(void * ptr) {
 static const AVClass av_codec_context_class = { "Postproc", context_to_name, NULL };
 
 pp_context_t *pp_get_context(int width, int height, int cpuCaps){
-        PPContext *c= av_malloc(sizeof(PPContext));
-        int stride= (width+15)&(~15);    //assumed / will realloc if needed
-        int qpStride= (width+15)/16 + 2; //assumed / will realloc if needed
-
-        memset(c, 0, sizeof(PPContext));
-        c->av_class = &av_codec_context_class;
-        c->cpuCaps= cpuCaps;
-        if(cpuCaps&PP_FORMAT){
-                c->hChromaSubSample= cpuCaps&0x3;
-                c->vChromaSubSample= (cpuCaps>>4)&0x3;
-        }else{
-                c->hChromaSubSample= 1;
-                c->vChromaSubSample= 1;
-        }
-
-        reallocBuffers(c, width, height, stride, qpStride);
-
-        c->frameNum=-1;
-
-        return c;
+    PPContext *c= av_malloc(sizeof(PPContext));
+    int stride= (width+15)&(~15);    //assumed / will realloc if needed
+    int qpStride= (width+15)/16 + 2; //assumed / will realloc if needed
+
+    memset(c, 0, sizeof(PPContext));
+    c->av_class = &av_codec_context_class;
+    c->cpuCaps= cpuCaps;
+    if(cpuCaps&PP_FORMAT){
+        c->hChromaSubSample= cpuCaps&0x3;
+        c->vChromaSubSample= (cpuCaps>>4)&0x3;
+    }else{
+        c->hChromaSubSample= 1;
+        c->vChromaSubSample= 1;
+    }
+
+    reallocBuffers(c, width, height, stride, qpStride);
+
+    c->frameNum=-1;
+
+    return c;
 }
 
 void pp_free_context(void *vc){
-        PPContext *c = (PPContext*)vc;
-        int i;
+    PPContext *c = (PPContext*)vc;
+    int i;
 
-        for(i=0; i<3; i++) av_free(c->tempBlured[i]);
-        for(i=0; i<3; i++) av_free(c->tempBluredPast[i]);
+    for(i=0; i<3; i++) av_free(c->tempBlured[i]);
+    for(i=0; i<3; i++) av_free(c->tempBluredPast[i]);
 
-        av_free(c->tempBlocks);
-        av_free(c->yHistogram);
-        av_free(c->tempDst);
-        av_free(c->tempSrc);
-        av_free(c->deintTemp);
-        av_free(c->stdQPTable);
-        av_free(c->nonBQPTable);
-        av_free(c->forcedQPTable);
+    av_free(c->tempBlocks);
+    av_free(c->yHistogram);
+    av_free(c->tempDst);
+    av_free(c->tempSrc);
+    av_free(c->deintTemp);
+    av_free(c->stdQPTable);
+    av_free(c->nonBQPTable);
+    av_free(c->forcedQPTable);
 
-        memset(c, 0, sizeof(PPContext));
+    memset(c, 0, sizeof(PPContext));
 
-        av_free(c);
+    av_free(c);
 }
 
 void  pp_postprocess(const uint8_t * src[3], const int srcStride[3],
-                 uint8_t * dst[3], const int dstStride[3],
-                 int width, int height,
-                 const QP_STORE_T *QP_store,  int QPStride,
-                 pp_mode_t *vm,  void *vc, int pict_type)
+                     uint8_t * dst[3], const int dstStride[3],
+                     int width, int height,
+                     const QP_STORE_T *QP_store,  int QPStride,
+                     pp_mode_t *vm,  void *vc, int pict_type)
 {
-        int mbWidth = (width+15)>>4;
-        int mbHeight= (height+15)>>4;
-        PPMode *mode = (PPMode*)vm;
-        PPContext *c = (PPContext*)vc;
-        int minStride= FFMAX(FFABS(srcStride[0]), FFABS(dstStride[0]));
-        int absQPStride = FFABS(QPStride);
-
-        // c->stride and c->QPStride are always positive
-        if(c->stride < minStride || c->qpStride < absQPStride)
-                reallocBuffers(c, width, height,
-                                FFMAX(minStride, c->stride),
-                                FFMAX(c->qpStride, absQPStride));
-
-        if(QP_store==NULL || (mode->lumMode & FORCE_QUANT))
-        {
-                int i;
-                QP_store= c->forcedQPTable;
-                absQPStride = QPStride = 0;
-                if(mode->lumMode & FORCE_QUANT)
-                        for(i=0; i<mbWidth; i++) c->forcedQPTable[i]= mode->forcedQuant;
-                else
-                        for(i=0; i<mbWidth; i++) c->forcedQPTable[i]= 1;
-        }
+    int mbWidth = (width+15)>>4;
+    int mbHeight= (height+15)>>4;
+    PPMode *mode = (PPMode*)vm;
+    PPContext *c = (PPContext*)vc;
+    int minStride= FFMAX(FFABS(srcStride[0]), FFABS(dstStride[0]));
+    int absQPStride = FFABS(QPStride);
+
+    // c->stride and c->QPStride are always positive
+    if(c->stride < minStride || c->qpStride < absQPStride)
+        reallocBuffers(c, width, height,
+                       FFMAX(minStride, c->stride),
+                       FFMAX(c->qpStride, absQPStride));
+
+    if(QP_store==NULL || (mode->lumMode & FORCE_QUANT)){
+        int i;
+        QP_store= c->forcedQPTable;
+        absQPStride = QPStride = 0;
+        if(mode->lumMode & FORCE_QUANT)
+            for(i=0; i<mbWidth; i++) c->forcedQPTable[i]= mode->forcedQuant;
+        else
+            for(i=0; i<mbWidth; i++) c->forcedQPTable[i]= 1;
+    }
 
-        if(pict_type & PP_PICT_TYPE_QP2){
-                int i;
-                const int count= mbHeight * absQPStride;
-                for(i=0; i<(count>>2); i++){
-                        ((uint32_t*)c->stdQPTable)[i] = (((const uint32_t*)QP_store)[i]>>1) & 0x7F7F7F7F;
-                }
-                for(i<<=2; i<count; i++){
-                        c->stdQPTable[i] = QP_store[i]>>1;
-                }
-                QP_store= c->stdQPTable;
-                QPStride= absQPStride;
+    if(pict_type & PP_PICT_TYPE_QP2){
+        int i;
+        const int count= mbHeight * absQPStride;
+        for(i=0; i<(count>>2); i++){
+            ((uint32_t*)c->stdQPTable)[i] = (((const uint32_t*)QP_store)[i]>>1) & 0x7F7F7F7F;
         }
-
-if(0){
-int x,y;
-for(y=0; y<mbHeight; y++){
-        for(x=0; x<mbWidth; x++){
+        for(i<<=2; i<count; i++){
+            c->stdQPTable[i] = QP_store[i]>>1;
+        }
+        QP_store= c->stdQPTable;
+        QPStride= absQPStride;
+    }
+
+    if(0){
+        int x,y;
+        for(y=0; y<mbHeight; y++){
+            for(x=0; x<mbWidth; x++){
                 av_log(c, AV_LOG_INFO, "%2d ", QP_store[x + y*QPStride]);
+            }
+            av_log(c, AV_LOG_INFO, "\n");
         }
         av_log(c, AV_LOG_INFO, "\n");
-}
-        av_log(c, AV_LOG_INFO, "\n");
-}
-
-        if((pict_type&7)!=3)
-        {
-                if (QPStride >= 0) {
-                        int i;
-                        const int count= mbHeight * QPStride;
-                        for(i=0; i<(count>>2); i++){
-                                ((uint32_t*)c->nonBQPTable)[i] = ((const uint32_t*)QP_store)[i] & 0x3F3F3F3F;
-                        }
-                        for(i<<=2; i<count; i++){
-                                c->nonBQPTable[i] = QP_store[i] & 0x3F;
-                        }
-                } else {
-                        int i,j;
-                        for(i=0; i<mbHeight; i++) {
-                                    for(j=0; j<absQPStride; j++) {
-                                        c->nonBQPTable[i*absQPStride+j] = QP_store[i*QPStride+j] & 0x3F;
-                                }
-                        }
+    }
+
+    if((pict_type&7)!=3){
+        if (QPStride >= 0){
+            int i;
+            const int count= mbHeight * QPStride;
+            for(i=0; i<(count>>2); i++){
+                ((uint32_t*)c->nonBQPTable)[i] = ((const uint32_t*)QP_store)[i] & 0x3F3F3F3F;
+            }
+            for(i<<=2; i<count; i++){
+                c->nonBQPTable[i] = QP_store[i] & 0x3F;
+            }
+        } else {
+            int i,j;
+            for(i=0; i<mbHeight; i++) {
+                for(j=0; j<absQPStride; j++) {
+                    c->nonBQPTable[i*absQPStride+j] = QP_store[i*QPStride+j] & 0x3F;
                 }
+            }
         }
+    }
 
-        av_log(c, AV_LOG_DEBUG, "using npp filters 0x%X/0x%X\n",
-               mode->lumMode, mode->chromMode);
+    av_log(c, AV_LOG_DEBUG, "using npp filters 0x%X/0x%X\n",
+           mode->lumMode, mode->chromMode);
 
-        postProcess(src[0], srcStride[0], dst[0], dstStride[0],
+    postProcess(src[0], srcStride[0], dst[0], dstStride[0],
                 width, height, QP_store, QPStride, 0, mode, c);
 
-        width  = (width )>>c->hChromaSubSample;
-        height = (height)>>c->vChromaSubSample;
-
-        if(mode->chromMode)
-        {
-                postProcess(src[1], srcStride[1], dst[1], dstStride[1],
-                        width, height, QP_store, QPStride, 1, mode, c);
-                postProcess(src[2], srcStride[2], dst[2], dstStride[2],
-                        width, height, QP_store, QPStride, 2, mode, c);
-        }
-        else if(srcStride[1] == dstStride[1] && srcStride[2] == dstStride[2])
-        {
-                linecpy(dst[1], src[1], height, srcStride[1]);
-                linecpy(dst[2], src[2], height, srcStride[2]);
-        }
-        else
-        {
-                int y;
-                for(y=0; y<height; y++)
-                {
-                        memcpy(&(dst[1][y*dstStride[1]]), &(src[1][y*srcStride[1]]), width);
-                        memcpy(&(dst[2][y*dstStride[2]]), &(src[2][y*srcStride[2]]), width);
-                }
+    width  = (width )>>c->hChromaSubSample;
+    height = (height)>>c->vChromaSubSample;
+
+    if(mode->chromMode){
+        postProcess(src[1], srcStride[1], dst[1], dstStride[1],
+                    width, height, QP_store, QPStride, 1, mode, c);
+        postProcess(src[2], srcStride[2], dst[2], dstStride[2],
+                    width, height, QP_store, QPStride, 2, mode, c);
+    }
+    else if(srcStride[1] == dstStride[1] && srcStride[2] == dstStride[2]){
+        linecpy(dst[1], src[1], height, srcStride[1]);
+        linecpy(dst[2], src[2], height, srcStride[2]);
+    }else{
+        int y;
+        for(y=0; y<height; y++){
+            memcpy(&(dst[1][y*dstStride[1]]), &(src[1][y*srcStride[1]]), width);
+            memcpy(&(dst[2][y*dstStride[2]]), &(src[2][y*srcStride[2]]), width);
         }
+    }
 }
 
index 7c7abb3..38baeec 100644 (file)
@@ -59,10 +59,10 @@ extern const char pp_help[]; ///< a simple help text
 #endif
 
 void  pp_postprocess(const uint8_t * src[3], const int srcStride[3],
-                 uint8_t * dst[3], const int dstStride[3],
-                 int horizontalSize, int verticalSize,
-                 const QP_STORE_T *QP_store,  int QP_stride,
-                 pp_mode_t *mode, pp_context_t *ppContext, int pict_type);
+                     uint8_t * dst[3], const int dstStride[3],
+                     int horizontalSize, int verticalSize,
+                     const QP_STORE_T *QP_store,  int QP_stride,
+                     pp_mode_t *mode, pp_context_t *ppContext, int pict_type);
 
 
 /**
index 3c704bb..c87ada5 100644 (file)
 #include "avutil.h"
 
 #define ALTIVEC_TRANSPOSE_8x8_SHORT(src_a,src_b,src_c,src_d,src_e,src_f,src_g,src_h) \
-  do {                                                                  \
-    __typeof__(src_a) tempA1, tempB1, tempC1, tempD1;                   \
-    __typeof__(src_a) tempE1, tempF1, tempG1, tempH1;                   \
-    __typeof__(src_a) tempA2, tempB2, tempC2, tempD2;                   \
-    __typeof__(src_a) tempE2, tempF2, tempG2, tempH2;                   \
-    tempA1 = vec_mergeh (src_a, src_e);                                 \
-    tempB1 = vec_mergel (src_a, src_e);                                 \
-    tempC1 = vec_mergeh (src_b, src_f);                                 \
-    tempD1 = vec_mergel (src_b, src_f);                                 \
-    tempE1 = vec_mergeh (src_c, src_g);                                 \
-    tempF1 = vec_mergel (src_c, src_g);                                 \
-    tempG1 = vec_mergeh (src_d, src_h);                                 \
-    tempH1 = vec_mergel (src_d, src_h);                                 \
-    tempA2 = vec_mergeh (tempA1, tempE1);                               \
-    tempB2 = vec_mergel (tempA1, tempE1);                               \
-    tempC2 = vec_mergeh (tempB1, tempF1);                               \
-    tempD2 = vec_mergel (tempB1, tempF1);                               \
-    tempE2 = vec_mergeh (tempC1, tempG1);                               \
-    tempF2 = vec_mergel (tempC1, tempG1);                               \
-    tempG2 = vec_mergeh (tempD1, tempH1);                               \
-    tempH2 = vec_mergel (tempD1, tempH1);                               \
-    src_a = vec_mergeh (tempA2, tempE2);                                \
-    src_b = vec_mergel (tempA2, tempE2);                                \
-    src_c = vec_mergeh (tempB2, tempF2);                                \
-    src_d = vec_mergel (tempB2, tempF2);                                \
-    src_e = vec_mergeh (tempC2, tempG2);                                \
-    src_f = vec_mergel (tempC2, tempG2);                                \
-    src_g = vec_mergeh (tempD2, tempH2);                                \
-    src_h = vec_mergel (tempD2, tempH2);                                \
-  } while (0)
+    do {                                                          \
+        __typeof__(src_a) tempA1, tempB1, tempC1, tempD1;         \
+        __typeof__(src_a) tempE1, tempF1, tempG1, tempH1;         \
+        __typeof__(src_a) tempA2, tempB2, tempC2, tempD2;         \
+        __typeof__(src_a) tempE2, tempF2, tempG2, tempH2;         \
+        tempA1 = vec_mergeh (src_a, src_e);                       \
+        tempB1 = vec_mergel (src_a, src_e);                       \
+        tempC1 = vec_mergeh (src_b, src_f);                       \
+        tempD1 = vec_mergel (src_b, src_f);                       \
+        tempE1 = vec_mergeh (src_c, src_g);                       \
+        tempF1 = vec_mergel (src_c, src_g);                       \
+        tempG1 = vec_mergeh (src_d, src_h);                       \
+        tempH1 = vec_mergel (src_d, src_h);                       \
+        tempA2 = vec_mergeh (tempA1, tempE1);                     \
+        tempB2 = vec_mergel (tempA1, tempE1);                     \
+        tempC2 = vec_mergeh (tempB1, tempF1);                     \
+        tempD2 = vec_mergel (tempB1, tempF1);                     \
+        tempE2 = vec_mergeh (tempC1, tempG1);                     \
+        tempF2 = vec_mergel (tempC1, tempG1);                     \
+        tempG2 = vec_mergeh (tempD1, tempH1);                     \
+        tempH2 = vec_mergel (tempD1, tempH1);                     \
+        src_a = vec_mergeh (tempA2, tempE2);                      \
+        src_b = vec_mergel (tempA2, tempE2);                      \
+        src_c = vec_mergeh (tempB2, tempF2);                      \
+        src_d = vec_mergel (tempB2, tempF2);                      \
+        src_e = vec_mergeh (tempC2, tempG2);                      \
+        src_f = vec_mergel (tempC2, tempG2);                      \
+        src_g = vec_mergeh (tempD2, tempH2);                      \
+        src_h = vec_mergel (tempD2, tempH2);                      \
+    } while (0)
 
 
 static inline int vertClassify_altivec(uint8_t src[], int stride, PPContext *c) {
-  /*
+    /*
     this code makes no assumption on src or stride.
     One could remove the recomputation of the perm
     vector by assuming (stride % 16) == 0, unfortunately
     this is not always true.
-  */
-  DECLARE_ALIGNED(16, short, data[8]);
-  int numEq;
-  uint8_t *src2 = src;
-  vector signed short v_dcOffset;
-  vector signed short v2QP;
-  vector unsigned short v4QP;
-  vector unsigned short v_dcThreshold;
-  const int properStride = (stride % 16);
-  const int srcAlign = ((unsigned long)src2 % 16);
-  const int two_vectors = ((srcAlign > 8) || properStride) ? 1 : 0;
-  const vector signed int zero = vec_splat_s32(0);
-  const vector signed short mask = vec_splat_s16(1);
-  vector signed int v_numEq = vec_splat_s32(0);
-
-  data[0] = ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
-  data[1] = data[0] * 2 + 1;
-  data[2] = c->QP * 2;
-  data[3] = c->QP * 4;
-  vector signed short v_data = vec_ld(0, data);
-  v_dcOffset = vec_splat(v_data, 0);
-  v_dcThreshold = (vector unsigned short)vec_splat(v_data, 1);
-  v2QP = vec_splat(v_data, 2);
-  v4QP = (vector unsigned short)vec_splat(v_data, 3);
-
-  src2 += stride * 4;
-
-  vector signed short v_srcAss0, v_srcAss1, v_srcAss2, v_srcAss3, v_srcAss4, v_srcAss5, v_srcAss6, v_srcAss7;
+    */
+    DECLARE_ALIGNED(16, short, data[8]);
+    int numEq;
+    uint8_t *src2 = src;
+    vector signed short v_dcOffset;
+    vector signed short v2QP;
+    vector unsigned short v4QP;
+    vector unsigned short v_dcThreshold;
+    const int properStride = (stride % 16);
+    const int srcAlign = ((unsigned long)src2 % 16);
+    const int two_vectors = ((srcAlign > 8) || properStride) ? 1 : 0;
+    const vector signed int zero = vec_splat_s32(0);
+    const vector signed short mask = vec_splat_s16(1);
+    vector signed int v_numEq = vec_splat_s32(0);
+
+    data[0] = ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
+    data[1] = data[0] * 2 + 1;
+    data[2] = c->QP * 2;
+    data[3] = c->QP * 4;
+    vector signed short v_data = vec_ld(0, data);
+    v_dcOffset = vec_splat(v_data, 0);
+    v_dcThreshold = (vector unsigned short)vec_splat(v_data, 1);
+    v2QP = vec_splat(v_data, 2);
+    v4QP = (vector unsigned short)vec_splat(v_data, 3);
+
+    src2 += stride * 4;
+
+    vector signed short v_srcAss0, v_srcAss1, v_srcAss2, v_srcAss3, v_srcAss4, v_srcAss5, v_srcAss6, v_srcAss7;
 
 #define LOAD_LINE(i)                                                    \
-  register int j##i = i * stride;                                       \
-  vector unsigned char perm##i = vec_lvsl(j##i, src2);                  \
-  const vector unsigned char v_srcA1##i = vec_ld(j##i, src2);           \
-  vector unsigned char v_srcA2##i;                                      \
-  if (two_vectors)                                                      \
-    v_srcA2##i = vec_ld(j##i + 16, src2);                               \
-  const vector unsigned char v_srcA##i =                                \
-    vec_perm(v_srcA1##i, v_srcA2##i, perm##i);                          \
-  v_srcAss##i =                                                         \
-    (vector signed short)vec_mergeh((vector signed char)zero,           \
-                                    (vector signed char)v_srcA##i)
+    register int j##i = i * stride;                                     \
+    vector unsigned char perm##i = vec_lvsl(j##i, src2);                \
+    const vector unsigned char v_srcA1##i = vec_ld(j##i, src2);         \
+    vector unsigned char v_srcA2##i;                                    \
+    if (two_vectors)                                                    \
+        v_srcA2##i = vec_ld(j##i + 16, src2);                           \
+    const vector unsigned char v_srcA##i =                              \
+        vec_perm(v_srcA1##i, v_srcA2##i, perm##i);                      \
+    v_srcAss##i =                                                       \
+        (vector signed short)vec_mergeh((vector signed char)zero,       \
+                                        (vector signed char)v_srcA##i)
 
 #define LOAD_LINE_ALIGNED(i)                                            \
-  register int j##i = i * stride;                                       \
-  const vector unsigned char v_srcA##i = vec_ld(j##i, src2);            \
-  v_srcAss##i =                                                         \
-    (vector signed short)vec_mergeh((vector signed char)zero,           \
-                                    (vector signed char)v_srcA##i)
+    register int j##i = i * stride;                                     \
+    const vector unsigned char v_srcA##i = vec_ld(j##i, src2);          \
+    v_srcAss##i =                                                       \
+        (vector signed short)vec_mergeh((vector signed char)zero,       \
+                                        (vector signed char)v_srcA##i)
 
     /* Special-casing the aligned case is worthwhile, as all calls from
      * the (transposed) horizontable deblocks will be aligned, in addition
      * to the naturally aligned vertical deblocks. */
     if (properStride && srcAlign) {
-      LOAD_LINE_ALIGNED(0);
-      LOAD_LINE_ALIGNED(1);
-      LOAD_LINE_ALIGNED(2);
-      LOAD_LINE_ALIGNED(3);
-      LOAD_LINE_ALIGNED(4);
-      LOAD_LINE_ALIGNED(5);
-      LOAD_LINE_ALIGNED(6);
-      LOAD_LINE_ALIGNED(7);
+        LOAD_LINE_ALIGNED(0);
+        LOAD_LINE_ALIGNED(1);
+        LOAD_LINE_ALIGNED(2);
+        LOAD_LINE_ALIGNED(3);
+        LOAD_LINE_ALIGNED(4);
+        LOAD_LINE_ALIGNED(5);
+        LOAD_LINE_ALIGNED(6);
+        LOAD_LINE_ALIGNED(7);
     } else {
-      LOAD_LINE(0);
-      LOAD_LINE(1);
-      LOAD_LINE(2);
-      LOAD_LINE(3);
-      LOAD_LINE(4);
-      LOAD_LINE(5);
-      LOAD_LINE(6);
-      LOAD_LINE(7);
+        LOAD_LINE(0);
+        LOAD_LINE(1);
+        LOAD_LINE(2);
+        LOAD_LINE(3);
+        LOAD_LINE(4);
+        LOAD_LINE(5);
+        LOAD_LINE(6);
+        LOAD_LINE(7);
     }
 #undef LOAD_LINE
 #undef LOAD_LINE_ALIGNED
 
 #define ITER(i, j)                                                      \
-  const vector signed short v_diff##i =                                 \
-    vec_sub(v_srcAss##i, v_srcAss##j);                                  \
-  const vector signed short v_sum##i =                                  \
-    vec_add(v_diff##i, v_dcOffset);                                     \
-  const vector signed short v_comp##i =                                 \
-    (vector signed short)vec_cmplt((vector unsigned short)v_sum##i,     \
-                                   v_dcThreshold);                      \
-  const vector signed short v_part##i = vec_and(mask, v_comp##i);       \
-  v_numEq = vec_sum4s(v_part##i, v_numEq);
-
-  ITER(0, 1);
-  ITER(1, 2);
-  ITER(2, 3);
-  ITER(3, 4);
-  ITER(4, 5);
-  ITER(5, 6);
-  ITER(6, 7);
+    const vector signed short v_diff##i =                               \
+        vec_sub(v_srcAss##i, v_srcAss##j);                              \
+    const vector signed short v_sum##i =                                \
+        vec_add(v_diff##i, v_dcOffset);                                 \
+    const vector signed short v_comp##i =                               \
+        (vector signed short)vec_cmplt((vector unsigned short)v_sum##i, \
+                                       v_dcThreshold);                  \
+    const vector signed short v_part##i = vec_and(mask, v_comp##i);     \
+    v_numEq = vec_sum4s(v_part##i, v_numEq);
+
+    ITER(0, 1);
+    ITER(1, 2);
+    ITER(2, 3);
+    ITER(3, 4);
+    ITER(4, 5);
+    ITER(5, 6);
+    ITER(6, 7);
 #undef ITER
 
-  v_numEq = vec_sums(v_numEq, zero);
-
-  v_numEq = vec_splat(v_numEq, 3);
-  vec_ste(v_numEq, 0, &numEq);
-
-  if (numEq > c->ppMode.flatnessThreshold)
-    {
-      const vector unsigned char mmoP1 = (const vector unsigned char)
-        AVV(0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f,
-            0x00, 0x01, 0x12, 0x13, 0x08, 0x09, 0x1A, 0x1B);
-      const vector unsigned char mmoP2 = (const vector unsigned char)
-        AVV(0x04, 0x05, 0x16, 0x17, 0x0C, 0x0D, 0x1E, 0x1F,
-            0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f);
-      const vector unsigned char mmoP = (const vector unsigned char)
-        vec_lvsl(8, (unsigned char*)0);
-
-      vector signed short mmoL1 = vec_perm(v_srcAss0, v_srcAss2, mmoP1);
-      vector signed short mmoL2 = vec_perm(v_srcAss4, v_srcAss6, mmoP2);
-      vector signed short mmoL = vec_perm(mmoL1, mmoL2, mmoP);
-      vector signed short mmoR1 = vec_perm(v_srcAss5, v_srcAss7, mmoP1);
-      vector signed short mmoR2 = vec_perm(v_srcAss1, v_srcAss3, mmoP2);
-      vector signed short mmoR = vec_perm(mmoR1, mmoR2, mmoP);
-      vector signed short mmoDiff = vec_sub(mmoL, mmoR);
-      vector unsigned short mmoSum = (vector unsigned short)vec_add(mmoDiff, v2QP);
-
-      if (vec_any_gt(mmoSum, v4QP))
-        return 0;
-      else
-        return 1;
+    v_numEq = vec_sums(v_numEq, zero);
+
+    v_numEq = vec_splat(v_numEq, 3);
+    vec_ste(v_numEq, 0, &numEq);
+
+    if (numEq > c->ppMode.flatnessThreshold){
+        const vector unsigned char mmoP1 = (const vector unsigned char)
+            AVV(0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f,
+                0x00, 0x01, 0x12, 0x13, 0x08, 0x09, 0x1A, 0x1B);
+        const vector unsigned char mmoP2 = (const vector unsigned char)
+            AVV(0x04, 0x05, 0x16, 0x17, 0x0C, 0x0D, 0x1E, 0x1F,
+                0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f);
+        const vector unsigned char mmoP = (const vector unsigned char)
+            vec_lvsl(8, (unsigned char*)0);
+
+        vector signed short mmoL1 = vec_perm(v_srcAss0, v_srcAss2, mmoP1);
+        vector signed short mmoL2 = vec_perm(v_srcAss4, v_srcAss6, mmoP2);
+        vector signed short mmoL = vec_perm(mmoL1, mmoL2, mmoP);
+        vector signed short mmoR1 = vec_perm(v_srcAss5, v_srcAss7, mmoP1);
+        vector signed short mmoR2 = vec_perm(v_srcAss1, v_srcAss3, mmoP2);
+        vector signed short mmoR = vec_perm(mmoR1, mmoR2, mmoP);
+        vector signed short mmoDiff = vec_sub(mmoL, mmoR);
+        vector unsigned short mmoSum = (vector unsigned short)vec_add(mmoDiff, v2QP);
+
+        if (vec_any_gt(mmoSum, v4QP))
+            return 0;
+        else
+            return 1;
     }
-  else return 2;
+    else return 2;
 }
 
 static inline void doVertLowPass_altivec(uint8_t *src, int stride, PPContext *c) {
-  /*
+    /*
     this code makes no assumption on src or stride.
     One could remove the recomputation of the perm
     vector by assuming (stride % 16) == 0, unfortunately
     this is not always true. Quite a lot of load/stores
     can be removed by assuming proper alignment of
     src & stride :-(
-  */
-  uint8_t *src2 = src;
-  const vector signed int zero = vec_splat_s32(0);
-  const int properStride = (stride % 16);
-  const int srcAlign = ((unsigned long)src2 % 16);
-  DECLARE_ALIGNED(16, short, qp[8]);
-  qp[0] = c->QP;
-  vector signed short vqp = vec_ld(0, qp);
-  vqp = vec_splat(vqp, 0);
-
-  src2 += stride*3;
-
-  vector signed short vb0, vb1, vb2, vb3, vb4, vb5, vb6, vb7, vb8, vb9;
-  vector unsigned char vbA0, vbA1, vbA2, vbA3, vbA4, vbA5, vbA6, vbA7, vbA8, vbA9;
-  vector unsigned char vbB0, vbB1, vbB2, vbB3, vbB4, vbB5, vbB6, vbB7, vbB8, vbB9;
-  vector unsigned char vbT0, vbT1, vbT2, vbT3, vbT4, vbT5, vbT6, vbT7, vbT8, vbT9;
+    */
+    uint8_t *src2 = src;
+    const vector signed int zero = vec_splat_s32(0);
+    const int properStride = (stride % 16);
+    const int srcAlign = ((unsigned long)src2 % 16);
+    DECLARE_ALIGNED(16, short, qp[8]);
+    qp[0] = c->QP;
+    vector signed short vqp = vec_ld(0, qp);
+    vqp = vec_splat(vqp, 0);
+
+    src2 += stride*3;
+
+    vector signed short vb0, vb1, vb2, vb3, vb4, vb5, vb6, vb7, vb8, vb9;
+    vector unsigned char vbA0, vbA1, vbA2, vbA3, vbA4, vbA5, vbA6, vbA7, vbA8, vbA9;
+    vector unsigned char vbB0, vbB1, vbB2, vbB3, vbB4, vbB5, vbB6, vbB7, vbB8, vbB9;
+    vector unsigned char vbT0, vbT1, vbT2, vbT3, vbT4, vbT5, vbT6, vbT7, vbT8, vbT9;
 
 #define LOAD_LINE(i)                                                    \
-  const vector unsigned char perml##i =                                 \
-    vec_lvsl(i * stride, src2);                                         \
-  vbA##i = vec_ld(i * stride, src2);                                    \
-  vbB##i = vec_ld(i * stride + 16, src2);                               \
-  vbT##i = vec_perm(vbA##i, vbB##i, perml##i);                          \
-  vb##i =                                                               \
-    (vector signed short)vec_mergeh((vector unsigned char)zero,         \
-                                    (vector unsigned char)vbT##i)
+    const vector unsigned char perml##i =                               \
+        vec_lvsl(i * stride, src2);                                     \
+    vbA##i = vec_ld(i * stride, src2);                                  \
+    vbB##i = vec_ld(i * stride + 16, src2);                             \
+    vbT##i = vec_perm(vbA##i, vbB##i, perml##i);                        \
+    vb##i =                                                             \
+        (vector signed short)vec_mergeh((vector unsigned char)zero,     \
+                                        (vector unsigned char)vbT##i)
 
 #define LOAD_LINE_ALIGNED(i)                                            \
-  register int j##i = i * stride;                                       \
-  vbT##i = vec_ld(j##i, src2);                                          \
-  vb##i =                                                               \
-    (vector signed short)vec_mergeh((vector signed char)zero,           \
-                                    (vector signed char)vbT##i)
-
-    /* Special-casing the aligned case is worthwhile, as all calls from
-     * the (transposed) horizontable deblocks will be aligned, in addition
-     * to the naturally aligned vertical deblocks. */
-    if (properStride && srcAlign) {
-      LOAD_LINE_ALIGNED(0);
-      LOAD_LINE_ALIGNED(1);
-      LOAD_LINE_ALIGNED(2);
-      LOAD_LINE_ALIGNED(3);
-      LOAD_LINE_ALIGNED(4);
-      LOAD_LINE_ALIGNED(5);
-      LOAD_LINE_ALIGNED(6);
-      LOAD_LINE_ALIGNED(7);
-      LOAD_LINE_ALIGNED(8);
-      LOAD_LINE_ALIGNED(9);
-    } else {
-      LOAD_LINE(0);
-      LOAD_LINE(1);
-      LOAD_LINE(2);
-      LOAD_LINE(3);
-      LOAD_LINE(4);
-      LOAD_LINE(5);
-      LOAD_LINE(6);
-      LOAD_LINE(7);
-      LOAD_LINE(8);
-      LOAD_LINE(9);
-    }
+    register int j##i = i * stride;                                     \
+    vbT##i = vec_ld(j##i, src2);                                        \
+    vb##i =                                                             \
+        (vector signed short)vec_mergeh((vector signed char)zero,       \
+                                        (vector signed char)vbT##i)
+
+      /* Special-casing the aligned case is worthwhile, as all calls from
+       * the (transposed) horizontable deblocks will be aligned, in addition
+       * to the naturally aligned vertical deblocks. */
+      if (properStride && srcAlign) {
+          LOAD_LINE_ALIGNED(0);
+          LOAD_LINE_ALIGNED(1);
+          LOAD_LINE_ALIGNED(2);
+          LOAD_LINE_ALIGNED(3);
+          LOAD_LINE_ALIGNED(4);
+          LOAD_LINE_ALIGNED(5);
+          LOAD_LINE_ALIGNED(6);
+          LOAD_LINE_ALIGNED(7);
+          LOAD_LINE_ALIGNED(8);
+          LOAD_LINE_ALIGNED(9);
+      } else {
+          LOAD_LINE(0);
+          LOAD_LINE(1);
+          LOAD_LINE(2);
+          LOAD_LINE(3);
+          LOAD_LINE(4);
+          LOAD_LINE(5);
+          LOAD_LINE(6);
+          LOAD_LINE(7);
+          LOAD_LINE(8);
+          LOAD_LINE(9);
+      }
 #undef LOAD_LINE
 #undef LOAD_LINE_ALIGNED
 
-  const vector unsigned short v_2 = vec_splat_u16(2);
-  const vector unsigned short v_4 = vec_splat_u16(4);
+    const vector unsigned short v_2 = vec_splat_u16(2);
+    const vector unsigned short v_4 = vec_splat_u16(4);
 
-  const vector signed short v_diff01 = vec_sub(vb0, vb1);
-  const vector unsigned short v_cmp01 =
-    (const vector unsigned short) vec_cmplt(vec_abs(v_diff01), vqp);
-  const vector signed short v_first = vec_sel(vb1, vb0, v_cmp01);
-  const vector signed short v_diff89 = vec_sub(vb8, vb9);
-  const vector unsigned short v_cmp89 =
-    (const vector unsigned short) vec_cmplt(vec_abs(v_diff89), vqp);
-  const vector signed short v_last = vec_sel(vb8, vb9, v_cmp89);
+    const vector signed short v_diff01 = vec_sub(vb0, vb1);
+    const vector unsigned short v_cmp01 =
+        (const vector unsigned short) vec_cmplt(vec_abs(v_diff01), vqp);
+    const vector signed short v_first = vec_sel(vb1, vb0, v_cmp01);
+    const vector signed short v_diff89 = vec_sub(vb8, vb9);
+    const vector unsigned short v_cmp89 =
+        (const vector unsigned short) vec_cmplt(vec_abs(v_diff89), vqp);
+    const vector signed short v_last = vec_sel(vb8, vb9, v_cmp89);
 
-  const vector signed short temp01 = vec_mladd(v_first, (vector signed short)v_4, vb1);
-  const vector signed short temp02 = vec_add(vb2, vb3);
-  const vector signed short temp03 = vec_add(temp01, (vector signed short)v_4);
-  const vector signed short v_sumsB0 = vec_add(temp02, temp03);
+    const vector signed short temp01 = vec_mladd(v_first, (vector signed short)v_4, vb1);
+    const vector signed short temp02 = vec_add(vb2, vb3);
+    const vector signed short temp03 = vec_add(temp01, (vector signed short)v_4);
+    const vector signed short v_sumsB0 = vec_add(temp02, temp03);
 
-  const vector signed short temp11 = vec_sub(v_sumsB0, v_first);
-  const vector signed short v_sumsB1 = vec_add(temp11, vb4);
+    const vector signed short temp11 = vec_sub(v_sumsB0, v_first);
+    const vector signed short v_sumsB1 = vec_add(temp11, vb4);
 
-  const vector signed short temp21 = vec_sub(v_sumsB1, v_first);
-  const vector signed short v_sumsB2 = vec_add(temp21, vb5);
+    const vector signed short temp21 = vec_sub(v_sumsB1, v_first);
+    const vector signed short v_sumsB2 = vec_add(temp21, vb5);
 
-  const vector signed short temp31 = vec_sub(v_sumsB2, v_first);
-  const vector signed short v_sumsB3 = vec_add(temp31, vb6);
+    const vector signed short temp31 = vec_sub(v_sumsB2, v_first);
+    const vector signed short v_sumsB3 = vec_add(temp31, vb6);
 
-  const vector signed short temp41 = vec_sub(v_sumsB3, v_first);
-  const vector signed short v_sumsB4 = vec_add(temp41, vb7);
+    const vector signed short temp41 = vec_sub(v_sumsB3, v_first);
+    const vector signed short v_sumsB4 = vec_add(temp41, vb7);
 
-  const vector signed short temp51 = vec_sub(v_sumsB4, vb1);
-  const vector signed short v_sumsB5 = vec_add(temp51, vb8);
+    const vector signed short temp51 = vec_sub(v_sumsB4, vb1);
+    const vector signed short v_sumsB5 = vec_add(temp51, vb8);
 
-  const vector signed short temp61 = vec_sub(v_sumsB5, vb2);
-  const vector signed short v_sumsB6 = vec_add(temp61, v_last);
+    const vector signed short temp61 = vec_sub(v_sumsB5, vb2);
+    const vector signed short v_sumsB6 = vec_add(temp61, v_last);
 
-  const vector signed short temp71 = vec_sub(v_sumsB6, vb3);
-  const vector signed short v_sumsB7 = vec_add(temp71, v_last);
+    const vector signed short temp71 = vec_sub(v_sumsB6, vb3);
+    const vector signed short v_sumsB7 = vec_add(temp71, v_last);
 
-  const vector signed short temp81 = vec_sub(v_sumsB7, vb4);
-  const vector signed short v_sumsB8 = vec_add(temp81, v_last);
+    const vector signed short temp81 = vec_sub(v_sumsB7, vb4);
+    const vector signed short v_sumsB8 = vec_add(temp81, v_last);
 
-  const vector signed short temp91 = vec_sub(v_sumsB8, vb5);
-  const vector signed short v_sumsB9 = vec_add(temp91, v_last);
+    const vector signed short temp91 = vec_sub(v_sumsB8, vb5);
+    const vector signed short v_sumsB9 = vec_add(temp91, v_last);
 
 #define COMPUTE_VR(i, j, k)                                             \
-  const vector signed short temps1##i =                                 \
-    vec_add(v_sumsB##i, v_sumsB##k);                                    \
-  const vector signed short temps2##i =                                 \
-    vec_mladd(vb##j, (vector signed short)v_2, temps1##i);              \
-  const vector signed short  vr##j = vec_sra(temps2##i, v_4)
-
-  COMPUTE_VR(0, 1, 2);
-  COMPUTE_VR(1, 2, 3);
-  COMPUTE_VR(2, 3, 4);
-  COMPUTE_VR(3, 4, 5);
-  COMPUTE_VR(4, 5, 6);
-  COMPUTE_VR(5, 6, 7);
-  COMPUTE_VR(6, 7, 8);
-  COMPUTE_VR(7, 8, 9);
-
-  const vector signed char neg1 = vec_splat_s8(-1);
-  const vector unsigned char permHH = (const vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
-                                                                      0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
+    const vector signed short temps1##i =                               \
+        vec_add(v_sumsB##i, v_sumsB##k);                                \
+    const vector signed short temps2##i =                               \
+        vec_mladd(vb##j, (vector signed short)v_2, temps1##i);          \
+    const vector signed short  vr##j = vec_sra(temps2##i, v_4)
+
+    COMPUTE_VR(0, 1, 2);
+    COMPUTE_VR(1, 2, 3);
+    COMPUTE_VR(2, 3, 4);
+    COMPUTE_VR(3, 4, 5);
+    COMPUTE_VR(4, 5, 6);
+    COMPUTE_VR(5, 6, 7);
+    COMPUTE_VR(6, 7, 8);
+    COMPUTE_VR(7, 8, 9);
+
+    const vector signed char neg1 = vec_splat_s8(-1);
+    const vector unsigned char permHH = (const vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
+                                                                        0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
 
 #define PACK_AND_STORE(i)                                       \
-  const vector unsigned char perms##i =                         \
-    vec_lvsr(i * stride, src2);                                 \
-  const vector unsigned char vf##i =                            \
-    vec_packsu(vr##i, (vector signed short)zero);               \
-  const vector unsigned char vg##i =                            \
-    vec_perm(vf##i, vbT##i, permHH);                            \
-  const vector unsigned char mask##i =                          \
-    vec_perm((vector unsigned char)zero, (vector unsigned char)neg1, perms##i); \
-  const vector unsigned char vg2##i =                           \
-    vec_perm(vg##i, vg##i, perms##i);                           \
-  const vector unsigned char svA##i =                           \
-    vec_sel(vbA##i, vg2##i, mask##i);                           \
-  const vector unsigned char svB##i =                           \
-    vec_sel(vg2##i, vbB##i, mask##i);                           \
-  vec_st(svA##i, i * stride, src2);                             \
-  vec_st(svB##i, i * stride + 16, src2)
+    const vector unsigned char perms##i =                       \
+        vec_lvsr(i * stride, src2);                             \
+    const vector unsigned char vf##i =                          \
+        vec_packsu(vr##i, (vector signed short)zero);           \
+    const vector unsigned char vg##i =                          \
+        vec_perm(vf##i, vbT##i, permHH);                        \
+    const vector unsigned char mask##i =                        \
+        vec_perm((vector unsigned char)zero, (vector unsigned char)neg1, perms##i); \
+    const vector unsigned char vg2##i =                         \
+        vec_perm(vg##i, vg##i, perms##i);                       \
+    const vector unsigned char svA##i =                         \
+        vec_sel(vbA##i, vg2##i, mask##i);                       \
+    const vector unsigned char svB##i =                         \
+        vec_sel(vg2##i, vbB##i, mask##i);                       \
+    vec_st(svA##i, i * stride, src2);                           \
+    vec_st(svB##i, i * stride + 16, src2)
 
 #define PACK_AND_STORE_ALIGNED(i)                               \
-  const vector unsigned char vf##i =                            \
-    vec_packsu(vr##i, (vector signed short)zero);               \
-  const vector unsigned char vg##i =                            \
-    vec_perm(vf##i, vbT##i, permHH);                            \
-  vec_st(vg##i, i * stride, src2)
-
-  /* Special-casing the aligned case is worthwhile, as all calls from
-   * the (transposed) horizontable deblocks will be aligned, in addition
-   * to the naturally aligned vertical deblocks. */
-  if (properStride && srcAlign) {
-    PACK_AND_STORE_ALIGNED(1);
-    PACK_AND_STORE_ALIGNED(2);
-    PACK_AND_STORE_ALIGNED(3);
-    PACK_AND_STORE_ALIGNED(4);
-    PACK_AND_STORE_ALIGNED(5);
-    PACK_AND_STORE_ALIGNED(6);
-    PACK_AND_STORE_ALIGNED(7);
-    PACK_AND_STORE_ALIGNED(8);
-  } else {
-    PACK_AND_STORE(1);
-    PACK_AND_STORE(2);
-    PACK_AND_STORE(3);
-    PACK_AND_STORE(4);
-    PACK_AND_STORE(5);
-    PACK_AND_STORE(6);
-    PACK_AND_STORE(7);
-    PACK_AND_STORE(8);
-  }
+    const vector unsigned char vf##i =                          \
+        vec_packsu(vr##i, (vector signed short)zero);           \
+    const vector unsigned char vg##i =                          \
+        vec_perm(vf##i, vbT##i, permHH);                        \
+    vec_st(vg##i, i * stride, src2)
+
+    /* Special-casing the aligned case is worthwhile, as all calls from
+     * the (transposed) horizontable deblocks will be aligned, in addition
+     * to the naturally aligned vertical deblocks. */
+    if (properStride && srcAlign) {
+        PACK_AND_STORE_ALIGNED(1);
+        PACK_AND_STORE_ALIGNED(2);
+        PACK_AND_STORE_ALIGNED(3);
+        PACK_AND_STORE_ALIGNED(4);
+        PACK_AND_STORE_ALIGNED(5);
+        PACK_AND_STORE_ALIGNED(6);
+        PACK_AND_STORE_ALIGNED(7);
+        PACK_AND_STORE_ALIGNED(8);
+    } else {
+        PACK_AND_STORE(1);
+        PACK_AND_STORE(2);
+        PACK_AND_STORE(3);
+        PACK_AND_STORE(4);
+        PACK_AND_STORE(5);
+        PACK_AND_STORE(6);
+        PACK_AND_STORE(7);
+        PACK_AND_STORE(8);
+    }
 #undef PACK_AND_STORE
 #undef PACK_AND_STORE_ALIGNED
 }
@@ -377,190 +376,190 @@ static inline void doVertLowPass_altivec(uint8_t *src, int stride, PPContext *c)
 
 
 static inline void doVertDefFilter_altivec(uint8_t src[], int stride, PPContext *c) {
-  /*
+    /*
     this code makes no assumption on src or stride.
     One could remove the recomputation of the perm
     vector by assuming (stride % 16) == 0, unfortunately
     this is not always true. Quite a lot of load/stores
     can be removed by assuming proper alignment of
     src & stride :-(
-  */
-  uint8_t *src2 = src;
-  const vector signed int zero = vec_splat_s32(0);
-  DECLARE_ALIGNED(16, short, qp[8]);
-  qp[0] = 8*c->QP;
-  vector signed short vqp = vec_ld(0, qp);
-  vqp = vec_splat(vqp, 0);
+    */
+    uint8_t *src2 = src;
+    const vector signed int zero = vec_splat_s32(0);
+    DECLARE_ALIGNED(16, short, qp[8]);
+    qp[0] = 8*c->QP;
+    vector signed short vqp = vec_ld(0, qp);
+    vqp = vec_splat(vqp, 0);
 
 #define LOAD_LINE(i)                                                    \
-  const vector unsigned char perm##i =                                  \
-    vec_lvsl(i * stride, src2);                                         \
-  const vector unsigned char vbA##i =                                   \
-    vec_ld(i * stride, src2);                                           \
-  const vector unsigned char vbB##i =                                   \
-    vec_ld(i * stride + 16, src2);                                      \
-  const vector unsigned char vbT##i =                                   \
-    vec_perm(vbA##i, vbB##i, perm##i);                                  \
-  const vector signed short vb##i =                                     \
-    (vector signed short)vec_mergeh((vector unsigned char)zero,         \
-                                    (vector unsigned char)vbT##i)
-
-  src2 += stride*3;
-
-  LOAD_LINE(1);
-  LOAD_LINE(2);
-  LOAD_LINE(3);
-  LOAD_LINE(4);
-  LOAD_LINE(5);
-  LOAD_LINE(6);
-  LOAD_LINE(7);
-  LOAD_LINE(8);
+    const vector unsigned char perm##i =                                \
+        vec_lvsl(i * stride, src2);                                     \
+    const vector unsigned char vbA##i =                                 \
+        vec_ld(i * stride, src2);                                       \
+    const vector unsigned char vbB##i =                                 \
+        vec_ld(i * stride + 16, src2);                                  \
+    const vector unsigned char vbT##i =                                 \
+        vec_perm(vbA##i, vbB##i, perm##i);                              \
+    const vector signed short vb##i =                                   \
+        (vector signed short)vec_mergeh((vector unsigned char)zero,     \
+                                        (vector unsigned char)vbT##i)
+
+    src2 += stride*3;
+
+     LOAD_LINE(1);
+     LOAD_LINE(2);
+     LOAD_LINE(3);
+     LOAD_LINE(4);
+     LOAD_LINE(5);
+     LOAD_LINE(6);
+     LOAD_LINE(7);
+     LOAD_LINE(8);
 #undef LOAD_LINE
 
-  const vector signed short v_1 = vec_splat_s16(1);
-  const vector signed short v_2 = vec_splat_s16(2);
-  const vector signed short v_5 = vec_splat_s16(5);
-  const vector signed short v_32 = vec_sl(v_1,
-                                          (vector unsigned short)v_5);
-  /* middle energy */
-  const vector signed short l3minusl6 = vec_sub(vb3, vb6);
-  const vector signed short l5minusl4 = vec_sub(vb5, vb4);
-  const vector signed short twotimes_l3minusl6 = vec_mladd(v_2, l3minusl6, (vector signed short)zero);
-  const vector signed short mE = vec_mladd(v_5, l5minusl4, twotimes_l3minusl6);
-  const vector signed short absmE = vec_abs(mE);
-  /* left & right energy */
-  const vector signed short l1minusl4 = vec_sub(vb1, vb4);
-  const vector signed short l3minusl2 = vec_sub(vb3, vb2);
-  const vector signed short l5minusl8 = vec_sub(vb5, vb8);
-  const vector signed short l7minusl6 = vec_sub(vb7, vb6);
-  const vector signed short twotimes_l1minusl4 = vec_mladd(v_2, l1minusl4, (vector signed short)zero);
-  const vector signed short twotimes_l5minusl8 = vec_mladd(v_2, l5minusl8, (vector signed short)zero);
-  const vector signed short lE = vec_mladd(v_5, l3minusl2, twotimes_l1minusl4);
-  const vector signed short rE = vec_mladd(v_5, l7minusl6, twotimes_l5minusl8);
-  /* d */
-  const vector signed short ddiff = vec_sub(absmE,
-                                            vec_min(vec_abs(lE),
-                                                    vec_abs(rE)));
-  const vector signed short ddiffclamp = vec_max(ddiff, (vector signed short)zero);
-  const vector signed short dtimes64 = vec_mladd(v_5, ddiffclamp, v_32);
-  const vector signed short d = vec_sra(dtimes64, vec_splat_u16(6));
-  const vector signed short minusd = vec_sub((vector signed short)zero, d);
-  const vector signed short finald = vec_sel(minusd,
-                                             d,
-                                             vec_cmpgt(vec_sub((vector signed short)zero, mE),
-                                                       (vector signed short)zero));
-  /* q */
-  const vector signed short qtimes2 = vec_sub(vb4, vb5);
-  /* for a shift right to behave like /2, we need to add one
-     to all negative integer */
-  const vector signed short rounddown = vec_sel((vector signed short)zero,
-                                                v_1,
-                                                vec_cmplt(qtimes2, (vector signed short)zero));
-  const vector signed short q = vec_sra(vec_add(qtimes2, rounddown), vec_splat_u16(1));
-  /* clamp */
-  const vector signed short dclamp_P1 = vec_max((vector signed short)zero, finald);
-  const vector signed short dclamp_P = vec_min(dclamp_P1, q);
-  const vector signed short dclamp_N1 = vec_min((vector signed short)zero, finald);
-  const vector signed short dclamp_N = vec_max(dclamp_N1, q);
-
-  const vector signed short dclampedfinal = vec_sel(dclamp_N,
-                                                    dclamp_P,
-                                                    vec_cmpgt(q, (vector signed short)zero));
-  const vector signed short dornotd = vec_sel((vector signed short)zero,
-                                              dclampedfinal,
-                                              vec_cmplt(absmE, vqp));
-  /* add/subtract to l4 and l5 */
-  const vector signed short vb4minusd = vec_sub(vb4, dornotd);
-  const vector signed short vb5plusd = vec_add(vb5, dornotd);
-  /* finally, stores */
-  const vector unsigned char st4 = vec_packsu(vb4minusd, (vector signed short)zero);
-  const vector unsigned char st5 = vec_packsu(vb5plusd, (vector signed short)zero);
-
-  const vector signed char neg1 = vec_splat_s8(-1);
-  const vector unsigned char permHH = (const vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
-                                                                      0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
+     const vector signed short v_1 = vec_splat_s16(1);
+     const vector signed short v_2 = vec_splat_s16(2);
+     const vector signed short v_5 = vec_splat_s16(5);
+     const vector signed short v_32 = vec_sl(v_1,
+                                             (vector unsigned short)v_5);
+     /* middle energy */
+     const vector signed short l3minusl6 = vec_sub(vb3, vb6);
+     const vector signed short l5minusl4 = vec_sub(vb5, vb4);
+     const vector signed short twotimes_l3minusl6 = vec_mladd(v_2, l3minusl6, (vector signed short)zero);
+     const vector signed short mE = vec_mladd(v_5, l5minusl4, twotimes_l3minusl6);
+     const vector signed short absmE = vec_abs(mE);
+     /* left & right energy */
+     const vector signed short l1minusl4 = vec_sub(vb1, vb4);
+     const vector signed short l3minusl2 = vec_sub(vb3, vb2);
+     const vector signed short l5minusl8 = vec_sub(vb5, vb8);
+     const vector signed short l7minusl6 = vec_sub(vb7, vb6);
+     const vector signed short twotimes_l1minusl4 = vec_mladd(v_2, l1minusl4, (vector signed short)zero);
+     const vector signed short twotimes_l5minusl8 = vec_mladd(v_2, l5minusl8, (vector signed short)zero);
+     const vector signed short lE = vec_mladd(v_5, l3minusl2, twotimes_l1minusl4);
+     const vector signed short rE = vec_mladd(v_5, l7minusl6, twotimes_l5minusl8);
+     /* d */
+     const vector signed short ddiff = vec_sub(absmE,
+                                               vec_min(vec_abs(lE),
+                                                       vec_abs(rE)));
+     const vector signed short ddiffclamp = vec_max(ddiff, (vector signed short)zero);
+     const vector signed short dtimes64 = vec_mladd(v_5, ddiffclamp, v_32);
+     const vector signed short d = vec_sra(dtimes64, vec_splat_u16(6));
+     const vector signed short minusd = vec_sub((vector signed short)zero, d);
+     const vector signed short finald = vec_sel(minusd,
+                                                d,
+                                                vec_cmpgt(vec_sub((vector signed short)zero, mE),
+                                                          (vector signed short)zero));
+     /* q */
+     const vector signed short qtimes2 = vec_sub(vb4, vb5);
+     /* for a shift right to behave like /2, we need to add one
+        to all negative integer */
+     const vector signed short rounddown = vec_sel((vector signed short)zero,
+                                                   v_1,
+                                                   vec_cmplt(qtimes2, (vector signed short)zero));
+     const vector signed short q = vec_sra(vec_add(qtimes2, rounddown), vec_splat_u16(1));
+     /* clamp */
+     const vector signed short dclamp_P1 = vec_max((vector signed short)zero, finald);
+     const vector signed short dclamp_P = vec_min(dclamp_P1, q);
+     const vector signed short dclamp_N1 = vec_min((vector signed short)zero, finald);
+     const vector signed short dclamp_N = vec_max(dclamp_N1, q);
+
+     const vector signed short dclampedfinal = vec_sel(dclamp_N,
+                                                       dclamp_P,
+                                                       vec_cmpgt(q, (vector signed short)zero));
+     const vector signed short dornotd = vec_sel((vector signed short)zero,
+                                                 dclampedfinal,
+                                                 vec_cmplt(absmE, vqp));
+     /* add/subtract to l4 and l5 */
+     const vector signed short vb4minusd = vec_sub(vb4, dornotd);
+     const vector signed short vb5plusd  = vec_add(vb5, dornotd);
+     /* finally, stores */
+     const vector unsigned char st4 = vec_packsu(vb4minusd, (vector signed short)zero);
+     const vector unsigned char st5 = vec_packsu(vb5plusd,  (vector signed short)zero);
+
+     const vector signed char neg1 = vec_splat_s8(-1);
+     const vector unsigned char permHH = (const vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
+                                                                         0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
 
 #define STORE(i)                                                \
-  const vector unsigned char perms##i =                         \
-    vec_lvsr(i * stride, src2);                                 \
-  const vector unsigned char vg##i =                            \
-    vec_perm(st##i, vbT##i, permHH);                            \
-  const vector unsigned char mask##i =                          \
-    vec_perm((vector unsigned char)zero, (vector unsigned char)neg1, perms##i); \
-  const vector unsigned char vg2##i =                           \
-    vec_perm(vg##i, vg##i, perms##i);                           \
-  const vector unsigned char svA##i =                           \
-    vec_sel(vbA##i, vg2##i, mask##i);                           \
-  const vector unsigned char svB##i =                           \
-    vec_sel(vg2##i, vbB##i, mask##i);                           \
-  vec_st(svA##i, i * stride, src2);                             \
-  vec_st(svB##i, i * stride + 16, src2)
-
-  STORE(4);
-  STORE(5);
+     const vector unsigned char perms##i =                      \
+         vec_lvsr(i * stride, src2);                            \
+     const vector unsigned char vg##i =                         \
+         vec_perm(st##i, vbT##i, permHH);                       \
+     const vector unsigned char mask##i =                       \
+         vec_perm((vector unsigned char)zero, (vector unsigned char)neg1, perms##i); \
+     const vector unsigned char vg2##i =                        \
+         vec_perm(vg##i, vg##i, perms##i);                      \
+     const vector unsigned char svA##i =                        \
+         vec_sel(vbA##i, vg2##i, mask##i);                      \
+     const vector unsigned char svB##i =                        \
+         vec_sel(vg2##i, vbB##i, mask##i);                      \
+     vec_st(svA##i, i * stride, src2);                          \
+     vec_st(svB##i, i * stride + 16, src2)
+
+     STORE(4);
+     STORE(5);
 }
 
 static inline void dering_altivec(uint8_t src[], int stride, PPContext *c) {
-  /*
+    /*
     this code makes no assumption on src or stride.
     One could remove the recomputation of the perm
     vector by assuming (stride % 16) == 0, unfortunately
     this is not always true. Quite a lot of load/stores
     can be removed by assuming proper alignment of
     src & stride :-(
-  */
-  uint8_t *srcCopy = src;
-  DECLARE_ALIGNED(16, uint8_t, dt[16]);
-  const vector signed int zero = vec_splat_s32(0);
-  vector unsigned char v_dt;
-  dt[0] = deringThreshold;
-  v_dt = vec_splat(vec_ld(0, dt), 0);
+    */
+    uint8_t *srcCopy = src;
+    DECLARE_ALIGNED(16, uint8_t, dt[16]);
+    const vector signed int zero = vec_splat_s32(0);
+    vector unsigned char v_dt;
+    dt[0] = deringThreshold;
+    v_dt = vec_splat(vec_ld(0, dt), 0);
 
 #define LOAD_LINE(i)                                                    \
-  const vector unsigned char perm##i =                                  \
-    vec_lvsl(i * stride, srcCopy);                                      \
-  vector unsigned char sA##i = vec_ld(i * stride, srcCopy);             \
-  vector unsigned char sB##i = vec_ld(i * stride + 16, srcCopy);        \
-  vector unsigned char src##i = vec_perm(sA##i, sB##i, perm##i)
-
-  LOAD_LINE(0);
-  LOAD_LINE(1);
-  LOAD_LINE(2);
-  LOAD_LINE(3);
-  LOAD_LINE(4);
-  LOAD_LINE(5);
-  LOAD_LINE(6);
-  LOAD_LINE(7);
-  LOAD_LINE(8);
-  LOAD_LINE(9);
+    const vector unsigned char perm##i =                               \
+        vec_lvsl(i * stride, srcCopy);                                 \
+    vector unsigned char sA##i = vec_ld(i * stride, srcCopy);          \
+    vector unsigned char sB##i = vec_ld(i * stride + 16, srcCopy);     \
+    vector unsigned char src##i = vec_perm(sA##i, sB##i, perm##i)
+
+    LOAD_LINE(0);
+    LOAD_LINE(1);
+    LOAD_LINE(2);
+    LOAD_LINE(3);
+    LOAD_LINE(4);
+    LOAD_LINE(5);
+    LOAD_LINE(6);
+    LOAD_LINE(7);
+    LOAD_LINE(8);
+    LOAD_LINE(9);
 #undef LOAD_LINE
 
-  vector unsigned char v_avg;
-  {
+    vector unsigned char v_avg;
+    {
     const vector unsigned char trunc_perm = (vector unsigned char)
-      AVV(0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,
-          0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18);
+        AVV(0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,
+            0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18);
     const vector unsigned char trunc_src12 = vec_perm(src1, src2, trunc_perm);
     const vector unsigned char trunc_src34 = vec_perm(src3, src4, trunc_perm);
     const vector unsigned char trunc_src56 = vec_perm(src5, src6, trunc_perm);
     const vector unsigned char trunc_src78 = vec_perm(src7, src8, trunc_perm);
 
 #define EXTRACT(op) do {                                                \
-      const vector unsigned char s##op##_1 = vec_##op(trunc_src12, trunc_src34); \
-      const vector unsigned char s##op##_2 = vec_##op(trunc_src56, trunc_src78); \
-      const vector unsigned char s##op##_6 = vec_##op(s##op##_1, s##op##_2); \
-      const vector unsigned char s##op##_8h = vec_mergeh(s##op##_6, s##op##_6); \
-      const vector unsigned char s##op##_8l = vec_mergel(s##op##_6, s##op##_6); \
-      const vector unsigned char s##op##_9 = vec_##op(s##op##_8h, s##op##_8l); \
-      const vector unsigned char s##op##_9h = vec_mergeh(s##op##_9, s##op##_9); \
-      const vector unsigned char s##op##_9l = vec_mergel(s##op##_9, s##op##_9); \
-      const vector unsigned char s##op##_10 = vec_##op(s##op##_9h, s##op##_9l); \
-      const vector unsigned char s##op##_10h = vec_mergeh(s##op##_10, s##op##_10); \
-      const vector unsigned char s##op##_10l = vec_mergel(s##op##_10, s##op##_10); \
-      const vector unsigned char s##op##_11 = vec_##op(s##op##_10h, s##op##_10l); \
-      const vector unsigned char s##op##_11h = vec_mergeh(s##op##_11, s##op##_11); \
-      const vector unsigned char s##op##_11l = vec_mergel(s##op##_11, s##op##_11); \
-      v_##op = vec_##op(s##op##_11h, s##op##_11l); } while (0)
+    const vector unsigned char s##op##_1   = vec_##op(trunc_src12, trunc_src34); \
+    const vector unsigned char s##op##_2   = vec_##op(trunc_src56, trunc_src78); \
+    const vector unsigned char s##op##_6   = vec_##op(s##op##_1, s##op##_2);     \
+    const vector unsigned char s##op##_8h  = vec_mergeh(s##op##_6, s##op##_6);   \
+    const vector unsigned char s##op##_8l  = vec_mergel(s##op##_6, s##op##_6);   \
+    const vector unsigned char s##op##_9   = vec_##op(s##op##_8h, s##op##_8l);   \
+    const vector unsigned char s##op##_9h  = vec_mergeh(s##op##_9, s##op##_9);   \
+    const vector unsigned char s##op##_9l  = vec_mergel(s##op##_9, s##op##_9);   \
+    const vector unsigned char s##op##_10  = vec_##op(s##op##_9h, s##op##_9l);   \
+    const vector unsigned char s##op##_10h = vec_mergeh(s##op##_10, s##op##_10); \
+    const vector unsigned char s##op##_10l = vec_mergel(s##op##_10, s##op##_10); \
+    const vector unsigned char s##op##_11  = vec_##op(s##op##_10h, s##op##_10l); \
+    const vector unsigned char s##op##_11h = vec_mergeh(s##op##_11, s##op##_11); \
+    const vector unsigned char s##op##_11l = vec_mergel(s##op##_11, s##op##_11); \
+    v_##op = vec_##op(s##op##_11h, s##op##_11l); } while (0)
 
     vector unsigned char v_min;
     vector unsigned char v_max;
@@ -569,19 +568,19 @@ static inline void dering_altivec(uint8_t src[], int stride, PPContext *c) {
 #undef EXTRACT
 
     if (vec_all_lt(vec_sub(v_max, v_min), v_dt))
-      return;
+        return;
 
     v_avg = vec_avg(v_min, v_max);
-  }
+    }
 
-  DECLARE_ALIGNED(16, signed int, S[8]);
-  {
+    DECLARE_ALIGNED(16, signed int, S[8]);
+    {
     const vector unsigned short mask1 = (vector unsigned short)
-      AVV(0x0001, 0x0002, 0x0004, 0x0008,
-          0x0010, 0x0020, 0x0040, 0x0080);
+        AVV(0x0001, 0x0002, 0x0004, 0x0008,
+            0x0010, 0x0020, 0x0040, 0x0080);
     const vector unsigned short mask2 = (vector unsigned short)
-      AVV(0x0100, 0x0200, 0x0000, 0x0000,
-          0x0000, 0x0000, 0x0000, 0x0000);
+        AVV(0x0100, 0x0200, 0x0000, 0x0000,
+            0x0000, 0x0000, 0x0000, 0x0000);
 
     const vector unsigned int vuint32_16 = vec_sl(vec_splat_u32(1), vec_splat_u32(4));
     const vector unsigned int vuint32_1 = vec_splat_u32(1);
@@ -589,19 +588,19 @@ static inline void dering_altivec(uint8_t src[], int stride, PPContext *c) {
 #define COMPARE(i)                                                      \
     vector signed int sum##i;                                           \
     do {                                                                \
-      const vector unsigned char cmp##i =                               \
-        (vector unsigned char)vec_cmpgt(src##i, v_avg);                 \
-      const vector unsigned short cmpHi##i =                            \
-        (vector unsigned short)vec_mergeh(cmp##i, cmp##i);              \
-      const vector unsigned short cmpLi##i =                            \
-        (vector unsigned short)vec_mergel(cmp##i, cmp##i);              \
-      const vector signed short cmpHf##i =                              \
-        (vector signed short)vec_and(cmpHi##i, mask1);                  \
-      const vector signed short cmpLf##i =                              \
-        (vector signed short)vec_and(cmpLi##i, mask2);                  \
-      const vector signed int sump##i = vec_sum4s(cmpHf##i, zero);      \
-      const vector signed int sumq##i = vec_sum4s(cmpLf##i, sump##i);   \
-      sum##i  = vec_sums(sumq##i, zero); } while (0)
+        const vector unsigned char cmp##i =                             \
+            (vector unsigned char)vec_cmpgt(src##i, v_avg);             \
+        const vector unsigned short cmpHi##i =                          \
+            (vector unsigned short)vec_mergeh(cmp##i, cmp##i);          \
+        const vector unsigned short cmpLi##i =                          \
+            (vector unsigned short)vec_mergel(cmp##i, cmp##i);          \
+        const vector signed short cmpHf##i =                            \
+            (vector signed short)vec_and(cmpHi##i, mask1);              \
+        const vector signed short cmpLf##i =                            \
+            (vector signed short)vec_and(cmpLi##i, mask2);              \
+        const vector signed int sump##i = vec_sum4s(cmpHf##i, zero);    \
+        const vector signed int sumq##i = vec_sum4s(cmpLf##i, sump##i); \
+        sum##i  = vec_sums(sumq##i, zero); } while (0)
 
     COMPARE(0);
     COMPARE(1);
@@ -618,178 +617,178 @@ static inline void dering_altivec(uint8_t src[], int stride, PPContext *c) {
     vector signed int sumA2;
     vector signed int sumB2;
     {
-      const vector signed int sump02 = vec_mergel(sum0, sum2);
-      const vector signed int sump13 = vec_mergel(sum1, sum3);
-      const vector signed int sumA = vec_mergel(sump02, sump13);
-
-      const vector signed int sump46 = vec_mergel(sum4, sum6);
-      const vector signed int sump57 = vec_mergel(sum5, sum7);
-      const vector signed int sumB = vec_mergel(sump46, sump57);
-
-      const vector signed int sump8A = vec_mergel(sum8, zero);
-      const vector signed int sump9B = vec_mergel(sum9, zero);
-      const vector signed int sumC = vec_mergel(sump8A, sump9B);
-
-      const vector signed int tA = vec_sl(vec_nor(zero, sumA), vuint32_16);
-      const vector signed int tB = vec_sl(vec_nor(zero, sumB), vuint32_16);
-      const vector signed int tC = vec_sl(vec_nor(zero, sumC), vuint32_16);
-      const vector signed int t2A = vec_or(sumA, tA);
-      const vector signed int t2B = vec_or(sumB, tB);
-      const vector signed int t2C = vec_or(sumC, tC);
-      const vector signed int t3A = vec_and(vec_sra(t2A, vuint32_1),
-                                            vec_sl(t2A, vuint32_1));
-      const vector signed int t3B = vec_and(vec_sra(t2B, vuint32_1),
-                                            vec_sl(t2B, vuint32_1));
-      const vector signed int t3C = vec_and(vec_sra(t2C, vuint32_1),
-                                            vec_sl(t2C, vuint32_1));
-      const vector signed int yA = vec_and(t2A, t3A);
-      const vector signed int yB = vec_and(t2B, t3B);
-      const vector signed int yC = vec_and(t2C, t3C);
-
-      const vector unsigned char strangeperm1 = vec_lvsl(4, (unsigned char*)0);
-      const vector unsigned char strangeperm2 = vec_lvsl(8, (unsigned char*)0);
-      const vector signed int sumAd4 = vec_perm(yA, yB, strangeperm1);
-      const vector signed int sumAd8 = vec_perm(yA, yB, strangeperm2);
-      const vector signed int sumBd4 = vec_perm(yB, yC, strangeperm1);
-      const vector signed int sumBd8 = vec_perm(yB, yC, strangeperm2);
-      const vector signed int sumAp = vec_and(yA,
-                                              vec_and(sumAd4,sumAd8));
-      const vector signed int sumBp = vec_and(yB,
-                                              vec_and(sumBd4,sumBd8));
-      sumA2 = vec_or(sumAp,
-                     vec_sra(sumAp,
-                             vuint32_16));
-      sumB2  = vec_or(sumBp,
-                      vec_sra(sumBp,
-                              vuint32_16));
+    const vector signed int sump02 = vec_mergel(sum0, sum2);
+    const vector signed int sump13 = vec_mergel(sum1, sum3);
+    const vector signed int sumA = vec_mergel(sump02, sump13);
+
+    const vector signed int sump46 = vec_mergel(sum4, sum6);
+    const vector signed int sump57 = vec_mergel(sum5, sum7);
+    const vector signed int sumB = vec_mergel(sump46, sump57);
+
+    const vector signed int sump8A = vec_mergel(sum8, zero);
+    const vector signed int sump9B = vec_mergel(sum9, zero);
+    const vector signed int sumC = vec_mergel(sump8A, sump9B);
+
+    const vector signed int tA = vec_sl(vec_nor(zero, sumA), vuint32_16);
+    const vector signed int tB = vec_sl(vec_nor(zero, sumB), vuint32_16);
+    const vector signed int tC = vec_sl(vec_nor(zero, sumC), vuint32_16);
+    const vector signed int t2A = vec_or(sumA, tA);
+    const vector signed int t2B = vec_or(sumB, tB);
+    const vector signed int t2C = vec_or(sumC, tC);
+    const vector signed int t3A = vec_and(vec_sra(t2A, vuint32_1),
+                                          vec_sl(t2A, vuint32_1));
+    const vector signed int t3B = vec_and(vec_sra(t2B, vuint32_1),
+                                          vec_sl(t2B, vuint32_1));
+    const vector signed int t3C = vec_and(vec_sra(t2C, vuint32_1),
+                                          vec_sl(t2C, vuint32_1));
+    const vector signed int yA = vec_and(t2A, t3A);
+    const vector signed int yB = vec_and(t2B, t3B);
+    const vector signed int yC = vec_and(t2C, t3C);
+
+    const vector unsigned char strangeperm1 = vec_lvsl(4, (unsigned char*)0);
+    const vector unsigned char strangeperm2 = vec_lvsl(8, (unsigned char*)0);
+    const vector signed int sumAd4 = vec_perm(yA, yB, strangeperm1);
+    const vector signed int sumAd8 = vec_perm(yA, yB, strangeperm2);
+    const vector signed int sumBd4 = vec_perm(yB, yC, strangeperm1);
+    const vector signed int sumBd8 = vec_perm(yB, yC, strangeperm2);
+    const vector signed int sumAp = vec_and(yA,
+                                            vec_and(sumAd4,sumAd8));
+    const vector signed int sumBp = vec_and(yB,
+                                            vec_and(sumBd4,sumBd8));
+    sumA2 = vec_or(sumAp,
+                   vec_sra(sumAp,
+                           vuint32_16));
+    sumB2  = vec_or(sumBp,
+                    vec_sra(sumBp,
+                            vuint32_16));
     }
     vec_st(sumA2, 0, S);
     vec_st(sumB2, 16, S);
-  }
-
-  /* I'm not sure the following is actually faster
-     than straight, unvectorized C code :-( */
-
-  DECLARE_ALIGNED(16, int, tQP2[4]);
-  tQP2[0]= c->QP/2 + 1;
-  vector signed int vQP2 = vec_ld(0, tQP2);
-  vQP2 = vec_splat(vQP2, 0);
-  const vector signed int vsint32_8 = vec_splat_s32(8);
-  const vector unsigned int vuint32_4 = vec_splat_u32(4);
-
-  const vector unsigned char permA1 = (vector unsigned char)
-    AVV(0x00, 0x01, 0x02, 0x10, 0x11, 0x12, 0x1F, 0x1F,
-        0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F);
-  const vector unsigned char permA2 = (vector unsigned char)
-    AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x10, 0x11,
-        0x12, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F);
-  const vector unsigned char permA1inc = (vector unsigned char)
-    AVV(0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00,
-        0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
-  const vector unsigned char permA2inc = (vector unsigned char)
-    AVV(0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x01,
-        0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
-  const vector unsigned char magic = (vector unsigned char)
-    AVV(0x01, 0x02, 0x01, 0x02, 0x04, 0x02, 0x01, 0x02,
-        0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
-  const vector unsigned char extractPerm = (vector unsigned char)
-    AVV(0x10, 0x10, 0x10, 0x01, 0x10, 0x10, 0x10, 0x01,
-        0x10, 0x10, 0x10, 0x01, 0x10, 0x10, 0x10, 0x01);
-  const vector unsigned char extractPermInc = (vector unsigned char)
-    AVV(0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01,
-        0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01);
-  const vector unsigned char identity = vec_lvsl(0,(unsigned char *)0);
-  const vector unsigned char tenRight = (vector unsigned char)
-    AVV(0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-        0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
-  const vector unsigned char eightLeft = (vector unsigned char)
-    AVV(0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-        0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x08);
+    }
+
+    /* I'm not sure the following is actually faster
+       than straight, unvectorized C code :-( */
+
+    DECLARE_ALIGNED(16, int, tQP2[4]);
+    tQP2[0]= c->QP/2 + 1;
+    vector signed int vQP2 = vec_ld(0, tQP2);
+    vQP2 = vec_splat(vQP2, 0);
+    const vector signed int vsint32_8 = vec_splat_s32(8);
+    const vector unsigned int vuint32_4 = vec_splat_u32(4);
+
+    const vector unsigned char permA1 = (vector unsigned char)
+        AVV(0x00, 0x01, 0x02, 0x10, 0x11, 0x12, 0x1F, 0x1F,
+            0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F);
+    const vector unsigned char permA2 = (vector unsigned char)
+        AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x10, 0x11,
+            0x12, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F);
+    const vector unsigned char permA1inc = (vector unsigned char)
+        AVV(0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
+    const vector unsigned char permA2inc = (vector unsigned char)
+        AVV(0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x01,
+            0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
+    const vector unsigned char magic = (vector unsigned char)
+        AVV(0x01, 0x02, 0x01, 0x02, 0x04, 0x02, 0x01, 0x02,
+            0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
+    const vector unsigned char extractPerm = (vector unsigned char)
+        AVV(0x10, 0x10, 0x10, 0x01, 0x10, 0x10, 0x10, 0x01,
+            0x10, 0x10, 0x10, 0x01, 0x10, 0x10, 0x10, 0x01);
+    const vector unsigned char extractPermInc = (vector unsigned char)
+        AVV(0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01,
+            0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01);
+    const vector unsigned char identity = vec_lvsl(0,(unsigned char *)0);
+    const vector unsigned char tenRight = (vector unsigned char)
+        AVV(0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
+    const vector unsigned char eightLeft = (vector unsigned char)
+        AVV(0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x08);
 
 
 #define F_INIT(i)                                       \
-  vector unsigned char tenRightM##i = tenRight;         \
-  vector unsigned char permA1M##i = permA1;             \
-  vector unsigned char permA2M##i = permA2;             \
-  vector unsigned char extractPermM##i = extractPerm
+    vector unsigned char tenRightM##i = tenRight;       \
+    vector unsigned char permA1M##i = permA1;           \
+    vector unsigned char permA2M##i = permA2;           \
+    vector unsigned char extractPermM##i = extractPerm
 
 #define F2(i, j, k, l)                                                  \
-  if (S[i] & (1 << (l+1))) {                                            \
-    const vector unsigned char a_##j##_A##l =                           \
-      vec_perm(src##i, src##j, permA1M##i);                             \
-    const vector unsigned char a_##j##_B##l =                           \
-      vec_perm(a_##j##_A##l, src##k, permA2M##i);                       \
-    const vector signed int a_##j##_sump##l =                           \
-      (vector signed int)vec_msum(a_##j##_B##l, magic,                  \
-                                  (vector unsigned int)zero);           \
-    vector signed int F_##j##_##l =                                     \
-      vec_sr(vec_sums(a_##j##_sump##l, vsint32_8), vuint32_4);          \
-    F_##j##_##l = vec_splat(F_##j##_##l, 3);                            \
-    const vector signed int p_##j##_##l =                               \
-      (vector signed int)vec_perm(src##j,                               \
-                                  (vector unsigned char)zero,           \
-                                  extractPermM##i);                     \
-    const vector signed int sum_##j##_##l = vec_add( p_##j##_##l, vQP2);\
-    const vector signed int diff_##j##_##l = vec_sub( p_##j##_##l, vQP2);\
-    vector signed int newpm_##j##_##l;                                  \
-    if (vec_all_lt(sum_##j##_##l, F_##j##_##l))                         \
-      newpm_##j##_##l = sum_##j##_##l;                                  \
-    else if (vec_all_gt(diff_##j##_##l, F_##j##_##l))                   \
-      newpm_##j##_##l = diff_##j##_##l;                                 \
-    else newpm_##j##_##l = F_##j##_##l;                                 \
-    const vector unsigned char newpm2_##j##_##l =                       \
-      vec_splat((vector unsigned char)newpm_##j##_##l, 15);             \
-    const vector unsigned char mask##j##l = vec_add(identity,           \
-                                                    tenRightM##i);      \
-    src##j = vec_perm(src##j, newpm2_##j##_##l, mask##j##l);            \
-  }                                                                     \
-  permA1M##i = vec_add(permA1M##i, permA1inc);                          \
-  permA2M##i = vec_add(permA2M##i, permA2inc);                          \
-  tenRightM##i = vec_sro(tenRightM##i, eightLeft);                      \
-  extractPermM##i = vec_add(extractPermM##i, extractPermInc)
+    if (S[i] & (1 << (l+1))) {                                          \
+        const vector unsigned char a_##j##_A##l =                       \
+            vec_perm(src##i, src##j, permA1M##i);                       \
+        const vector unsigned char a_##j##_B##l =                       \
+            vec_perm(a_##j##_A##l, src##k, permA2M##i);                 \
+        const vector signed int a_##j##_sump##l =                       \
+            (vector signed int)vec_msum(a_##j##_B##l, magic,            \
+                                        (vector unsigned int)zero);     \
+        vector signed int F_##j##_##l =                                 \
+            vec_sr(vec_sums(a_##j##_sump##l, vsint32_8), vuint32_4);    \
+        F_##j##_##l = vec_splat(F_##j##_##l, 3);                        \
+        const vector signed int p_##j##_##l =                           \
+            (vector signed int)vec_perm(src##j,                         \
+                                        (vector unsigned char)zero,     \
+                                        extractPermM##i);               \
+        const vector signed int sum_##j##_##l  = vec_add( p_##j##_##l, vQP2);\
+        const vector signed int diff_##j##_##l = vec_sub( p_##j##_##l, vQP2);\
+        vector signed int newpm_##j##_##l;                              \
+        if (vec_all_lt(sum_##j##_##l, F_##j##_##l))                     \
+            newpm_##j##_##l = sum_##j##_##l;                            \
+        else if (vec_all_gt(diff_##j##_##l, F_##j##_##l))               \
+            newpm_##j##_##l = diff_##j##_##l;                           \
+        else newpm_##j##_##l = F_##j##_##l;                             \
+        const vector unsigned char newpm2_##j##_##l =                   \
+            vec_splat((vector unsigned char)newpm_##j##_##l, 15);       \
+        const vector unsigned char mask##j##l = vec_add(identity,       \
+                                                        tenRightM##i);  \
+        src##j = vec_perm(src##j, newpm2_##j##_##l, mask##j##l);        \
+    }                                                                   \
+    permA1M##i = vec_add(permA1M##i, permA1inc);                        \
+    permA2M##i = vec_add(permA2M##i, permA2inc);                        \
+    tenRightM##i = vec_sro(tenRightM##i, eightLeft);                    \
+    extractPermM##i = vec_add(extractPermM##i, extractPermInc)
 
 #define ITER(i, j, k)                           \
-  F_INIT(i);                                    \
-  F2(i, j, k, 0);                               \
-  F2(i, j, k, 1);                               \
-  F2(i, j, k, 2);                               \
-  F2(i, j, k, 3);                               \
-  F2(i, j, k, 4);                               \
-  F2(i, j, k, 5);                               \
-  F2(i, j, k, 6);                               \
-  F2(i, j, k, 7)
-
-  ITER(0, 1, 2);
-  ITER(1, 2, 3);
-  ITER(2, 3, 4);
-  ITER(3, 4, 5);
-  ITER(4, 5, 6);
-  ITER(5, 6, 7);
-  ITER(6, 7, 8);
-  ITER(7, 8, 9);
-
-  const vector signed char neg1 = vec_splat_s8(-1);
+    F_INIT(i);                                  \
+    F2(i, j, k, 0);                             \
+    F2(i, j, k, 1);                             \
+    F2(i, j, k, 2);                             \
+    F2(i, j, k, 3);                             \
+    F2(i, j, k, 4);                             \
+    F2(i, j, k, 5);                             \
+    F2(i, j, k, 6);                             \
+    F2(i, j, k, 7)
+
+    ITER(0, 1, 2);
+    ITER(1, 2, 3);
+    ITER(2, 3, 4);
+    ITER(3, 4, 5);
+    ITER(4, 5, 6);
+    ITER(5, 6, 7);
+    ITER(6, 7, 8);
+    ITER(7, 8, 9);
+
+    const vector signed char neg1 = vec_splat_s8(-1);
 
 #define STORE_LINE(i)                                   \
-  const vector unsigned char permST##i =                \
-    vec_lvsr(i * stride, srcCopy);                      \
-  const vector unsigned char maskST##i =                \
-    vec_perm((vector unsigned char)zero,                \
-             (vector unsigned char)neg1, permST##i);    \
-  src##i = vec_perm(src##i ,src##i, permST##i);         \
-  sA##i= vec_sel(sA##i, src##i, maskST##i);             \
-  sB##i= vec_sel(src##i, sB##i, maskST##i);             \
-  vec_st(sA##i, i * stride, srcCopy);                   \
-  vec_st(sB##i, i * stride + 16, srcCopy)
-
-  STORE_LINE(1);
-  STORE_LINE(2);
-  STORE_LINE(3);
-  STORE_LINE(4);
-  STORE_LINE(5);
-  STORE_LINE(6);
-  STORE_LINE(7);
-  STORE_LINE(8);
+    const vector unsigned char permST##i =              \
+        vec_lvsr(i * stride, srcCopy);                  \
+    const vector unsigned char maskST##i =              \
+        vec_perm((vector unsigned char)zero,            \
+                 (vector unsigned char)neg1, permST##i);\
+    src##i = vec_perm(src##i ,src##i, permST##i);       \
+    sA##i= vec_sel(sA##i, src##i, maskST##i);           \
+    sB##i= vec_sel(src##i, sB##i, maskST##i);           \
+    vec_st(sA##i, i * stride, srcCopy);                 \
+    vec_st(sB##i, i * stride + 16, srcCopy)
+
+    STORE_LINE(1);
+    STORE_LINE(2);
+    STORE_LINE(3);
+    STORE_LINE(4);
+    STORE_LINE(5);
+    STORE_LINE(6);
+    STORE_LINE(7);
+    STORE_LINE(8);
 
 #undef STORE_LINE
 #undef ITER
@@ -801,386 +800,386 @@ static inline void dering_altivec(uint8_t src[], int stride, PPContext *c) {
 #define do_a_deblock_altivec(a...) do_a_deblock_C(a)
 
 static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
-                                    uint8_t *tempBlured, uint32_t *tempBluredPast, int *maxNoise)
+                                            uint8_t *tempBlured, uint32_t *tempBluredPast, int *maxNoise)
 {
-  const vector signed int zero = vec_splat_s32(0);
-  const vector signed short vsint16_1 = vec_splat_s16(1);
-  vector signed int v_dp = zero;
-  vector signed int v_sysdp = zero;
-  int d, sysd, i;
+    const vector signed int zero = vec_splat_s32(0);
+    const vector signed short vsint16_1 = vec_splat_s16(1);
+    vector signed int v_dp = zero;
+    vector signed int v_sysdp = zero;
+    int d, sysd, i;
 
-  tempBluredPast[127]= maxNoise[0];
-  tempBluredPast[128]= maxNoise[1];
-  tempBluredPast[129]= maxNoise[2];
+    tempBluredPast[127]= maxNoise[0];
+    tempBluredPast[128]= maxNoise[1];
+    tempBluredPast[129]= maxNoise[2];
 
 #define LOAD_LINE(src, i)                                               \
-  register int j##src##i = i * stride;                                  \
-  vector unsigned char perm##src##i = vec_lvsl(j##src##i, src);         \
-  const vector unsigned char v_##src##A1##i = vec_ld(j##src##i, src);   \
-  const vector unsigned char v_##src##A2##i = vec_ld(j##src##i + 16, src); \
-  const vector unsigned char v_##src##A##i =                            \
-    vec_perm(v_##src##A1##i, v_##src##A2##i, perm##src##i);             \
-  vector signed short v_##src##Ass##i =                                 \
-    (vector signed short)vec_mergeh((vector signed char)zero,           \
-                                    (vector signed char)v_##src##A##i)
-
-  LOAD_LINE(src, 0);
-  LOAD_LINE(src, 1);
-  LOAD_LINE(src, 2);
-  LOAD_LINE(src, 3);
-  LOAD_LINE(src, 4);
-  LOAD_LINE(src, 5);
-  LOAD_LINE(src, 6);
-  LOAD_LINE(src, 7);
-
-  LOAD_LINE(tempBlured, 0);
-  LOAD_LINE(tempBlured, 1);
-  LOAD_LINE(tempBlured, 2);
-  LOAD_LINE(tempBlured, 3);
-  LOAD_LINE(tempBlured, 4);
-  LOAD_LINE(tempBlured, 5);
-  LOAD_LINE(tempBlured, 6);
-  LOAD_LINE(tempBlured, 7);
+    register int j##src##i = i * stride;                                \
+    vector unsigned char perm##src##i = vec_lvsl(j##src##i, src);       \
+    const vector unsigned char v_##src##A1##i = vec_ld(j##src##i, src); \
+    const vector unsigned char v_##src##A2##i = vec_ld(j##src##i + 16, src); \
+    const vector unsigned char v_##src##A##i =                          \
+        vec_perm(v_##src##A1##i, v_##src##A2##i, perm##src##i);         \
+    vector signed short v_##src##Ass##i =                               \
+        (vector signed short)vec_mergeh((vector signed char)zero,       \
+                                        (vector signed char)v_##src##A##i)
+
+    LOAD_LINE(src, 0);
+    LOAD_LINE(src, 1);
+    LOAD_LINE(src, 2);
+    LOAD_LINE(src, 3);
+    LOAD_LINE(src, 4);
+    LOAD_LINE(src, 5);
+    LOAD_LINE(src, 6);
+    LOAD_LINE(src, 7);
+
+    LOAD_LINE(tempBlured, 0);
+    LOAD_LINE(tempBlured, 1);
+    LOAD_LINE(tempBlured, 2);
+    LOAD_LINE(tempBlured, 3);
+    LOAD_LINE(tempBlured, 4);
+    LOAD_LINE(tempBlured, 5);
+    LOAD_LINE(tempBlured, 6);
+    LOAD_LINE(tempBlured, 7);
 #undef LOAD_LINE
 
 #define ACCUMULATE_DIFFS(i)                                     \
-  vector signed short v_d##i = vec_sub(v_tempBluredAss##i,      \
-                                       v_srcAss##i);            \
-  v_dp = vec_msums(v_d##i, v_d##i, v_dp);                       \
-  v_sysdp = vec_msums(v_d##i, vsint16_1, v_sysdp)
-
-  ACCUMULATE_DIFFS(0);
-  ACCUMULATE_DIFFS(1);
-  ACCUMULATE_DIFFS(2);
-  ACCUMULATE_DIFFS(3);
-  ACCUMULATE_DIFFS(4);
-  ACCUMULATE_DIFFS(5);
-  ACCUMULATE_DIFFS(6);
-  ACCUMULATE_DIFFS(7);
+    vector signed short v_d##i = vec_sub(v_tempBluredAss##i,    \
+                                         v_srcAss##i);          \
+    v_dp = vec_msums(v_d##i, v_d##i, v_dp);                     \
+    v_sysdp = vec_msums(v_d##i, vsint16_1, v_sysdp)
+
+    ACCUMULATE_DIFFS(0);
+    ACCUMULATE_DIFFS(1);
+    ACCUMULATE_DIFFS(2);
+    ACCUMULATE_DIFFS(3);
+    ACCUMULATE_DIFFS(4);
+    ACCUMULATE_DIFFS(5);
+    ACCUMULATE_DIFFS(6);
+    ACCUMULATE_DIFFS(7);
 #undef ACCUMULATE_DIFFS
 
-  v_dp = vec_sums(v_dp, zero);
-  v_sysdp = vec_sums(v_sysdp, zero);
+    v_dp = vec_sums(v_dp, zero);
+    v_sysdp = vec_sums(v_sysdp, zero);
 
-  v_dp = vec_splat(v_dp, 3);
-  v_sysdp = vec_splat(v_sysdp, 3);
+    v_dp = vec_splat(v_dp, 3);
+    v_sysdp = vec_splat(v_sysdp, 3);
 
-  vec_ste(v_dp, 0, &d);
-  vec_ste(v_sysdp, 0, &sysd);
+    vec_ste(v_dp, 0, &d);
+    vec_ste(v_sysdp, 0, &sysd);
 
-  i = d;
-  d = (4*d
-       +(*(tempBluredPast-256))
-       +(*(tempBluredPast-1))+ (*(tempBluredPast+1))
-       +(*(tempBluredPast+256))
-       +4)>>3;
+    i = d;
+    d = (4*d
+         +(*(tempBluredPast-256))
+         +(*(tempBluredPast-1))+ (*(tempBluredPast+1))
+         +(*(tempBluredPast+256))
+         +4)>>3;
 
-  *tempBluredPast=i;
+    *tempBluredPast=i;
 
-  if (d > maxNoise[1]) {
-    if (d < maxNoise[2]) {
+    if (d > maxNoise[1]) {
+        if (d < maxNoise[2]) {
 #define OP(i) v_tempBluredAss##i = vec_avg(v_tempBluredAss##i, v_srcAss##i);
 
-      OP(0);
-      OP(1);
-      OP(2);
-      OP(3);
-      OP(4);
-      OP(5);
-      OP(6);
-      OP(7);
+            OP(0);
+            OP(1);
+            OP(2);
+            OP(3);
+            OP(4);
+            OP(5);
+            OP(6);
+            OP(7);
 #undef OP
-    } else {
+        } else {
 #define OP(i) v_tempBluredAss##i = v_srcAss##i;
 
-      OP(0);
-      OP(1);
-      OP(2);
-      OP(3);
-      OP(4);
-      OP(5);
-      OP(6);
-      OP(7);
+            OP(0);
+            OP(1);
+            OP(2);
+            OP(3);
+            OP(4);
+            OP(5);
+            OP(6);
+            OP(7);
 #undef OP
-    }
-  } else {
-    if (d < maxNoise[0]) {
-      const vector signed short vsint16_7 = vec_splat_s16(7);
-      const vector signed short vsint16_4 = vec_splat_s16(4);
-      const vector unsigned short vuint16_3 = vec_splat_u16(3);
+        }
+    } else {
+        if (d < maxNoise[0]) {
+            const vector signed short vsint16_7 = vec_splat_s16(7);
+            const vector signed short vsint16_4 = vec_splat_s16(4);
+            const vector unsigned short vuint16_3 = vec_splat_u16(3);
 
 #define OP(i)                                                   \
-      const vector signed short v_temp##i =                     \
-        vec_mladd(v_tempBluredAss##i,                           \
-                  vsint16_7, v_srcAss##i);                      \
-      const vector signed short v_temp2##i =                    \
-        vec_add(v_temp##i, vsint16_4);                          \
-      v_tempBluredAss##i = vec_sr(v_temp2##i, vuint16_3)
-
-      OP(0);
-      OP(1);
-      OP(2);
-      OP(3);
-      OP(4);
-      OP(5);
-      OP(6);
-      OP(7);
+            const vector signed short v_temp##i =               \
+                vec_mladd(v_tempBluredAss##i,                   \
+                          vsint16_7, v_srcAss##i);              \
+            const vector signed short v_temp2##i =              \
+                vec_add(v_temp##i, vsint16_4);                  \
+            v_tempBluredAss##i = vec_sr(v_temp2##i, vuint16_3)
+
+            OP(0);
+            OP(1);
+            OP(2);
+            OP(3);
+            OP(4);
+            OP(5);
+            OP(6);
+            OP(7);
 #undef OP
-    } else {
-      const vector signed short vsint16_3 = vec_splat_s16(3);
-      const vector signed short vsint16_2 = vec_splat_s16(2);
+        } else {
+            const vector signed short vsint16_3 = vec_splat_s16(3);
+            const vector signed short vsint16_2 = vec_splat_s16(2);
 
 #define OP(i)                                                   \
-      const vector signed short v_temp##i =                     \
-        vec_mladd(v_tempBluredAss##i,                           \
-                  vsint16_3, v_srcAss##i);                      \
-      const vector signed short v_temp2##i =                    \
-        vec_add(v_temp##i, vsint16_2);                          \
-      v_tempBluredAss##i = vec_sr(v_temp2##i, (vector unsigned short)vsint16_2)
-
-      OP(0);
-      OP(1);
-      OP(2);
-      OP(3);
-      OP(4);
-      OP(5);
-      OP(6);
-      OP(7);
+            const vector signed short v_temp##i =               \
+                vec_mladd(v_tempBluredAss##i,                   \
+                          vsint16_3, v_srcAss##i);              \
+            const vector signed short v_temp2##i =              \
+                vec_add(v_temp##i, vsint16_2);                  \
+            v_tempBluredAss##i = vec_sr(v_temp2##i, (vector unsigned short)vsint16_2)
+
+            OP(0);
+            OP(1);
+            OP(2);
+            OP(3);
+            OP(4);
+            OP(5);
+            OP(6);
+            OP(7);
 #undef OP
+        }
     }
-  }
 
-  const vector signed char neg1 = vec_splat_s8(-1);
-  const vector unsigned char permHH = (const vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
-                                                                      0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
+    const vector signed char neg1 = vec_splat_s8(-1);
+    const vector unsigned char permHH = (const vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
+                                                                        0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
 
 #define PACK_AND_STORE(src, i)                                  \
-  const vector unsigned char perms##src##i =                    \
-    vec_lvsr(i * stride, src);                                  \
-  const vector unsigned char vf##src##i =                       \
-    vec_packsu(v_tempBluredAss##i, (vector signed short)zero);  \
-  const vector unsigned char vg##src##i =                       \
-    vec_perm(vf##src##i, v_##src##A##i, permHH);                \
-  const vector unsigned char mask##src##i =                     \
-    vec_perm((vector unsigned char)zero, (vector unsigned char)neg1, perms##src##i); \
-  const vector unsigned char vg2##src##i =                      \
-    vec_perm(vg##src##i, vg##src##i, perms##src##i);            \
-  const vector unsigned char svA##src##i =                      \
-    vec_sel(v_##src##A1##i, vg2##src##i, mask##src##i);         \
-  const vector unsigned char svB##src##i =                      \
-    vec_sel(vg2##src##i, v_##src##A2##i, mask##src##i);         \
-  vec_st(svA##src##i, i * stride, src);                         \
-  vec_st(svB##src##i, i * stride + 16, src)
-
-  PACK_AND_STORE(src, 0);
-  PACK_AND_STORE(src, 1);
-  PACK_AND_STORE(src, 2);
-  PACK_AND_STORE(src, 3);
-  PACK_AND_STORE(src, 4);
-  PACK_AND_STORE(src, 5);
-  PACK_AND_STORE(src, 6);
-  PACK_AND_STORE(src, 7);
-  PACK_AND_STORE(tempBlured, 0);
-  PACK_AND_STORE(tempBlured, 1);
-  PACK_AND_STORE(tempBlured, 2);
-  PACK_AND_STORE(tempBlured, 3);
-  PACK_AND_STORE(tempBlured, 4);
-  PACK_AND_STORE(tempBlured, 5);
-  PACK_AND_STORE(tempBlured, 6);
-  PACK_AND_STORE(tempBlured, 7);
+    const vector unsigned char perms##src##i =                  \
+        vec_lvsr(i * stride, src);                              \
+    const vector unsigned char vf##src##i =                     \
+        vec_packsu(v_tempBluredAss##i, (vector signed short)zero); \
+    const vector unsigned char vg##src##i =                     \
+        vec_perm(vf##src##i, v_##src##A##i, permHH);            \
+    const vector unsigned char mask##src##i =                   \
+        vec_perm((vector unsigned char)zero, (vector unsigned char)neg1, perms##src##i); \
+    const vector unsigned char vg2##src##i =                    \
+        vec_perm(vg##src##i, vg##src##i, perms##src##i);        \
+    const vector unsigned char svA##src##i =                    \
+        vec_sel(v_##src##A1##i, vg2##src##i, mask##src##i);     \
+    const vector unsigned char svB##src##i =                    \
+        vec_sel(vg2##src##i, v_##src##A2##i, mask##src##i);     \
+    vec_st(svA##src##i, i * stride, src);                       \
+    vec_st(svB##src##i, i * stride + 16, src)
+
+    PACK_AND_STORE(src, 0);
+    PACK_AND_STORE(src, 1);
+    PACK_AND_STORE(src, 2);
+    PACK_AND_STORE(src, 3);
+    PACK_AND_STORE(src, 4);
+    PACK_AND_STORE(src, 5);
+    PACK_AND_STORE(src, 6);
+    PACK_AND_STORE(src, 7);
+    PACK_AND_STORE(tempBlured, 0);
+    PACK_AND_STORE(tempBlured, 1);
+    PACK_AND_STORE(tempBlured, 2);
+    PACK_AND_STORE(tempBlured, 3);
+    PACK_AND_STORE(tempBlured, 4);
+    PACK_AND_STORE(tempBlured, 5);
+    PACK_AND_STORE(tempBlured, 6);
+    PACK_AND_STORE(tempBlured, 7);
 #undef PACK_AND_STORE
 }
 
 static inline void transpose_16x8_char_toPackedAlign_altivec(unsigned char* dst, unsigned char* src, int stride) {
-  const vector unsigned char zero = vec_splat_u8(0);
+    const vector unsigned char zero = vec_splat_u8(0);
 
 #define LOAD_DOUBLE_LINE(i, j)                                          \
-  vector unsigned char perm1##i = vec_lvsl(i * stride, src);            \
-  vector unsigned char perm2##i = vec_lvsl(j * stride, src);            \
-  vector unsigned char srcA##i = vec_ld(i * stride, src);               \
-  vector unsigned char srcB##i = vec_ld(i * stride + 16, src);          \
-  vector unsigned char srcC##i = vec_ld(j * stride, src);               \
-  vector unsigned char srcD##i = vec_ld(j * stride+ 16, src);           \
-  vector unsigned char src##i = vec_perm(srcA##i, srcB##i, perm1##i);   \
-  vector unsigned char src##j = vec_perm(srcC##i, srcD##i, perm2##i)
-
-  LOAD_DOUBLE_LINE(0, 1);
-  LOAD_DOUBLE_LINE(2, 3);
-  LOAD_DOUBLE_LINE(4, 5);
-  LOAD_DOUBLE_LINE(6, 7);
+    vector unsigned char perm1##i = vec_lvsl(i * stride, src);          \
+    vector unsigned char perm2##i = vec_lvsl(j * stride, src);          \
+    vector unsigned char srcA##i = vec_ld(i * stride, src);             \
+    vector unsigned char srcB##i = vec_ld(i * stride + 16, src);        \
+    vector unsigned char srcC##i = vec_ld(j * stride, src);             \
+    vector unsigned char srcD##i = vec_ld(j * stride+ 16, src);         \
+    vector unsigned char src##i = vec_perm(srcA##i, srcB##i, perm1##i); \
+    vector unsigned char src##j = vec_perm(srcC##i, srcD##i, perm2##i)
+
+    LOAD_DOUBLE_LINE(0, 1);
+    LOAD_DOUBLE_LINE(2, 3);
+    LOAD_DOUBLE_LINE(4, 5);
+    LOAD_DOUBLE_LINE(6, 7);
 #undef LOAD_DOUBLE_LINE
 
-  vector unsigned char tempA = vec_mergeh(src0, zero);
-  vector unsigned char tempB = vec_mergel(src0, zero);
-  vector unsigned char tempC = vec_mergeh(src1, zero);
-  vector unsigned char tempD = vec_mergel(src1, zero);
-  vector unsigned char tempE = vec_mergeh(src2, zero);
-  vector unsigned char tempF = vec_mergel(src2, zero);
-  vector unsigned char tempG = vec_mergeh(src3, zero);
-  vector unsigned char tempH = vec_mergel(src3, zero);
-  vector unsigned char tempI = vec_mergeh(src4, zero);
-  vector unsigned char tempJ = vec_mergel(src4, zero);
-  vector unsigned char tempK = vec_mergeh(src5, zero);
-  vector unsigned char tempL = vec_mergel(src5, zero);
-  vector unsigned char tempM = vec_mergeh(src6, zero);
-  vector unsigned char tempN = vec_mergel(src6, zero);
-  vector unsigned char tempO = vec_mergeh(src7, zero);
-  vector unsigned char tempP = vec_mergel(src7, zero);
-
-  vector unsigned char temp0 = vec_mergeh(tempA, tempI);
-  vector unsigned char temp1 = vec_mergel(tempA, tempI);
-  vector unsigned char temp2 = vec_mergeh(tempB, tempJ);
-  vector unsigned char temp3 = vec_mergel(tempB, tempJ);
-  vector unsigned char temp4 = vec_mergeh(tempC, tempK);
-  vector unsigned char temp5 = vec_mergel(tempC, tempK);
-  vector unsigned char temp6 = vec_mergeh(tempD, tempL);
-  vector unsigned char temp7 = vec_mergel(tempD, tempL);
-  vector unsigned char temp8 = vec_mergeh(tempE, tempM);
-  vector unsigned char temp9 = vec_mergel(tempE, tempM);
-  vector unsigned char temp10 = vec_mergeh(tempF, tempN);
-  vector unsigned char temp11 = vec_mergel(tempF, tempN);
-  vector unsigned char temp12 = vec_mergeh(tempG, tempO);
-  vector unsigned char temp13 = vec_mergel(tempG, tempO);
-  vector unsigned char temp14 = vec_mergeh(tempH, tempP);
-  vector unsigned char temp15 = vec_mergel(tempH, tempP);
-
-  tempA = vec_mergeh(temp0, temp8);
-  tempB = vec_mergel(temp0, temp8);
-  tempC = vec_mergeh(temp1, temp9);
-  tempD = vec_mergel(temp1, temp9);
-  tempE = vec_mergeh(temp2, temp10);
-  tempF = vec_mergel(temp2, temp10);
-  tempG = vec_mergeh(temp3, temp11);
-  tempH = vec_mergel(temp3, temp11);
-  tempI = vec_mergeh(temp4, temp12);
-  tempJ = vec_mergel(temp4, temp12);
-  tempK = vec_mergeh(temp5, temp13);
-  tempL = vec_mergel(temp5, temp13);
-  tempM = vec_mergeh(temp6, temp14);
-  tempN = vec_mergel(temp6, temp14);
-  tempO = vec_mergeh(temp7, temp15);
-  tempP = vec_mergel(temp7, temp15);
-
-  temp0 = vec_mergeh(tempA, tempI);
-  temp1 = vec_mergel(tempA, tempI);
-  temp2 = vec_mergeh(tempB, tempJ);
-  temp3 = vec_mergel(tempB, tempJ);
-  temp4 = vec_mergeh(tempC, tempK);
-  temp5 = vec_mergel(tempC, tempK);
-  temp6 = vec_mergeh(tempD, tempL);
-  temp7 = vec_mergel(tempD, tempL);
-  temp8 = vec_mergeh(tempE, tempM);
-  temp9 = vec_mergel(tempE, tempM);
-  temp10 = vec_mergeh(tempF, tempN);
-  temp11 = vec_mergel(tempF, tempN);
-  temp12 = vec_mergeh(tempG, tempO);
-  temp13 = vec_mergel(tempG, tempO);
-  temp14 = vec_mergeh(tempH, tempP);
-  temp15 = vec_mergel(tempH, tempP);
-
-  vec_st(temp0, 0, dst);
-  vec_st(temp1, 16, dst);
-  vec_st(temp2, 32, dst);
-  vec_st(temp3, 48, dst);
-  vec_st(temp4, 64, dst);
-  vec_st(temp5, 80, dst);
-  vec_st(temp6, 96, dst);
-  vec_st(temp7, 112, dst);
-  vec_st(temp8, 128, dst);
-  vec_st(temp9, 144, dst);
-  vec_st(temp10, 160, dst);
-  vec_st(temp11, 176, dst);
-  vec_st(temp12, 192, dst);
-  vec_st(temp13, 208, dst);
-  vec_st(temp14, 224, dst);
-  vec_st(temp15, 240, dst);
+    vector unsigned char tempA = vec_mergeh(src0, zero);
+    vector unsigned char tempB = vec_mergel(src0, zero);
+    vector unsigned char tempC = vec_mergeh(src1, zero);
+    vector unsigned char tempD = vec_mergel(src1, zero);
+    vector unsigned char tempE = vec_mergeh(src2, zero);
+    vector unsigned char tempF = vec_mergel(src2, zero);
+    vector unsigned char tempG = vec_mergeh(src3, zero);
+    vector unsigned char tempH = vec_mergel(src3, zero);
+    vector unsigned char tempI = vec_mergeh(src4, zero);
+    vector unsigned char tempJ = vec_mergel(src4, zero);
+    vector unsigned char tempK = vec_mergeh(src5, zero);
+    vector unsigned char tempL = vec_mergel(src5, zero);
+    vector unsigned char tempM = vec_mergeh(src6, zero);
+    vector unsigned char tempN = vec_mergel(src6, zero);
+    vector unsigned char tempO = vec_mergeh(src7, zero);
+    vector unsigned char tempP = vec_mergel(src7, zero);
+
+    vector unsigned char temp0 = vec_mergeh(tempA, tempI);
+    vector unsigned char temp1 = vec_mergel(tempA, tempI);
+    vector unsigned char temp2 = vec_mergeh(tempB, tempJ);
+    vector unsigned char temp3 = vec_mergel(tempB, tempJ);
+    vector unsigned char temp4 = vec_mergeh(tempC, tempK);
+    vector unsigned char temp5 = vec_mergel(tempC, tempK);
+    vector unsigned char temp6 = vec_mergeh(tempD, tempL);
+    vector unsigned char temp7 = vec_mergel(tempD, tempL);
+    vector unsigned char temp8 = vec_mergeh(tempE, tempM);
+    vector unsigned char temp9 = vec_mergel(tempE, tempM);
+    vector unsigned char temp10 = vec_mergeh(tempF, tempN);
+    vector unsigned char temp11 = vec_mergel(tempF, tempN);
+    vector unsigned char temp12 = vec_mergeh(tempG, tempO);
+    vector unsigned char temp13 = vec_mergel(tempG, tempO);
+    vector unsigned char temp14 = vec_mergeh(tempH, tempP);
+    vector unsigned char temp15 = vec_mergel(tempH, tempP);
+
+    tempA = vec_mergeh(temp0, temp8);
+    tempB = vec_mergel(temp0, temp8);
+    tempC = vec_mergeh(temp1, temp9);
+    tempD = vec_mergel(temp1, temp9);
+    tempE = vec_mergeh(temp2, temp10);
+    tempF = vec_mergel(temp2, temp10);
+    tempG = vec_mergeh(temp3, temp11);
+    tempH = vec_mergel(temp3, temp11);
+    tempI = vec_mergeh(temp4, temp12);
+    tempJ = vec_mergel(temp4, temp12);
+    tempK = vec_mergeh(temp5, temp13);
+    tempL = vec_mergel(temp5, temp13);
+    tempM = vec_mergeh(temp6, temp14);
+    tempN = vec_mergel(temp6, temp14);
+    tempO = vec_mergeh(temp7, temp15);
+    tempP = vec_mergel(temp7, temp15);
+
+    temp0 = vec_mergeh(tempA, tempI);
+    temp1 = vec_mergel(tempA, tempI);
+    temp2 = vec_mergeh(tempB, tempJ);
+    temp3 = vec_mergel(tempB, tempJ);
+    temp4 = vec_mergeh(tempC, tempK);
+    temp5 = vec_mergel(tempC, tempK);
+    temp6 = vec_mergeh(tempD, tempL);
+    temp7 = vec_mergel(tempD, tempL);
+    temp8 = vec_mergeh(tempE, tempM);
+    temp9 = vec_mergel(tempE, tempM);
+    temp10 = vec_mergeh(tempF, tempN);
+    temp11 = vec_mergel(tempF, tempN);
+    temp12 = vec_mergeh(tempG, tempO);
+    temp13 = vec_mergel(tempG, tempO);
+    temp14 = vec_mergeh(tempH, tempP);
+    temp15 = vec_mergel(tempH, tempP);
+
+    vec_st(temp0, 0, dst);
+    vec_st(temp1, 16, dst);
+    vec_st(temp2, 32, dst);
+    vec_st(temp3, 48, dst);
+    vec_st(temp4, 64, dst);
+    vec_st(temp5, 80, dst);
+    vec_st(temp6, 96, dst);
+    vec_st(temp7, 112, dst);
+    vec_st(temp8, 128, dst);
+    vec_st(temp9, 144, dst);
+    vec_st(temp10, 160, dst);
+    vec_st(temp11, 176, dst);
+    vec_st(temp12, 192, dst);
+    vec_st(temp13, 208, dst);
+    vec_st(temp14, 224, dst);
+    vec_st(temp15, 240, dst);
 }
 
 static inline void transpose_8x16_char_fromPackedAlign_altivec(unsigned char* dst, unsigned char* src, int stride) {
-  const vector unsigned char zero = vec_splat_u8(0);
+    const vector unsigned char zero = vec_splat_u8(0);
 
 #define LOAD_DOUBLE_LINE(i, j)                                  \
-  vector unsigned char src##i = vec_ld(i * 16, src);            \
-  vector unsigned char src##j = vec_ld(j * 16, src)
-
-  LOAD_DOUBLE_LINE(0, 1);
-  LOAD_DOUBLE_LINE(2, 3);
-  LOAD_DOUBLE_LINE(4, 5);
-  LOAD_DOUBLE_LINE(6, 7);
-  LOAD_DOUBLE_LINE(8, 9);
-  LOAD_DOUBLE_LINE(10, 11);
-  LOAD_DOUBLE_LINE(12, 13);
-  LOAD_DOUBLE_LINE(14, 15);
+    vector unsigned char src##i = vec_ld(i * 16, src);            \
+    vector unsigned char src##j = vec_ld(j * 16, src)
+
+    LOAD_DOUBLE_LINE(0, 1);
+    LOAD_DOUBLE_LINE(2, 3);
+    LOAD_DOUBLE_LINE(4, 5);
+    LOAD_DOUBLE_LINE(6, 7);
+    LOAD_DOUBLE_LINE(8, 9);
+    LOAD_DOUBLE_LINE(10, 11);
+    LOAD_DOUBLE_LINE(12, 13);
+    LOAD_DOUBLE_LINE(14, 15);
 #undef LOAD_DOUBLE_LINE
 
-  vector unsigned char tempA = vec_mergeh(src0, src8);
-  vector unsigned char tempB;
-  vector unsigned char tempC = vec_mergeh(src1, src9);
-  vector unsigned char tempD;
-  vector unsigned char tempE = vec_mergeh(src2, src10);
-  vector unsigned char tempG = vec_mergeh(src3, src11);
-  vector unsigned char tempI = vec_mergeh(src4, src12);
-  vector unsigned char tempJ;
-  vector unsigned char tempK = vec_mergeh(src5, src13);
-  vector unsigned char tempL;
-  vector unsigned char tempM = vec_mergeh(src6, src14);
-  vector unsigned char tempO = vec_mergeh(src7, src15);
-
-  vector unsigned char temp0 = vec_mergeh(tempA, tempI);
-  vector unsigned char temp1 = vec_mergel(tempA, tempI);
-  vector unsigned char temp2;
-  vector unsigned char temp3;
-  vector unsigned char temp4 = vec_mergeh(tempC, tempK);
-  vector unsigned char temp5 = vec_mergel(tempC, tempK);
-  vector unsigned char temp6;
-  vector unsigned char temp7;
-  vector unsigned char temp8 = vec_mergeh(tempE, tempM);
-  vector unsigned char temp9 = vec_mergel(tempE, tempM);
-  vector unsigned char temp12 = vec_mergeh(tempG, tempO);
-  vector unsigned char temp13 = vec_mergel(tempG, tempO);
-
-  tempA = vec_mergeh(temp0, temp8);
-  tempB = vec_mergel(temp0, temp8);
-  tempC = vec_mergeh(temp1, temp9);
-  tempD = vec_mergel(temp1, temp9);
-  tempI = vec_mergeh(temp4, temp12);
-  tempJ = vec_mergel(temp4, temp12);
-  tempK = vec_mergeh(temp5, temp13);
-  tempL = vec_mergel(temp5, temp13);
-
-  temp0 = vec_mergeh(tempA, tempI);
-  temp1 = vec_mergel(tempA, tempI);
-  temp2 = vec_mergeh(tempB, tempJ);
-  temp3 = vec_mergel(tempB, tempJ);
-  temp4 = vec_mergeh(tempC, tempK);
-  temp5 = vec_mergel(tempC, tempK);
-  temp6 = vec_mergeh(tempD, tempL);
-  temp7 = vec_mergel(tempD, tempL);
-
-
-  const vector signed char neg1 = vec_splat_s8(-1);
+    vector unsigned char tempA = vec_mergeh(src0, src8);
+    vector unsigned char tempB;
+    vector unsigned char tempC = vec_mergeh(src1, src9);
+    vector unsigned char tempD;
+    vector unsigned char tempE = vec_mergeh(src2, src10);
+    vector unsigned char tempG = vec_mergeh(src3, src11);
+    vector unsigned char tempI = vec_mergeh(src4, src12);
+    vector unsigned char tempJ;
+    vector unsigned char tempK = vec_mergeh(src5, src13);
+    vector unsigned char tempL;
+    vector unsigned char tempM = vec_mergeh(src6, src14);
+    vector unsigned char tempO = vec_mergeh(src7, src15);
+
+    vector unsigned char temp0 = vec_mergeh(tempA, tempI);
+    vector unsigned char temp1 = vec_mergel(tempA, tempI);
+    vector unsigned char temp2;
+    vector unsigned char temp3;
+    vector unsigned char temp4 = vec_mergeh(tempC, tempK);
+    vector unsigned char temp5 = vec_mergel(tempC, tempK);
+    vector unsigned char temp6;
+    vector unsigned char temp7;
+    vector unsigned char temp8 = vec_mergeh(tempE, tempM);
+    vector unsigned char temp9 = vec_mergel(tempE, tempM);
+    vector unsigned char temp12 = vec_mergeh(tempG, tempO);
+    vector unsigned char temp13 = vec_mergel(tempG, tempO);
+
+    tempA = vec_mergeh(temp0, temp8);
+    tempB = vec_mergel(temp0, temp8);
+    tempC = vec_mergeh(temp1, temp9);
+    tempD = vec_mergel(temp1, temp9);
+    tempI = vec_mergeh(temp4, temp12);
+    tempJ = vec_mergel(temp4, temp12);
+    tempK = vec_mergeh(temp5, temp13);
+    tempL = vec_mergel(temp5, temp13);
+
+    temp0 = vec_mergeh(tempA, tempI);
+    temp1 = vec_mergel(tempA, tempI);
+    temp2 = vec_mergeh(tempB, tempJ);
+    temp3 = vec_mergel(tempB, tempJ);
+    temp4 = vec_mergeh(tempC, tempK);
+    temp5 = vec_mergel(tempC, tempK);
+    temp6 = vec_mergeh(tempD, tempL);
+    temp7 = vec_mergel(tempD, tempL);
+
+
+    const vector signed char neg1 = vec_splat_s8(-1);
 #define STORE_DOUBLE_LINE(i, j)                                         \
-  vector unsigned char dstA##i = vec_ld(i * stride, dst);               \
-  vector unsigned char dstB##i = vec_ld(i * stride + 16, dst);          \
-  vector unsigned char dstA##j = vec_ld(j * stride, dst);               \
-  vector unsigned char dstB##j = vec_ld(j * stride+ 16, dst);           \
-  vector unsigned char align##i = vec_lvsr(i * stride, dst);            \
-  vector unsigned char align##j = vec_lvsr(j * stride, dst);            \
-  vector unsigned char mask##i = vec_perm(zero, (vector unsigned char)neg1, align##i); \
-  vector unsigned char mask##j = vec_perm(zero, (vector unsigned char)neg1, align##j); \
-  vector unsigned char dstR##i = vec_perm(temp##i, temp##i, align##i);  \
-  vector unsigned char dstR##j = vec_perm(temp##j, temp##j, align##j);  \
-  vector unsigned char dstAF##i = vec_sel(dstA##i, dstR##i, mask##i);   \
-  vector unsigned char dstBF##i = vec_sel(dstR##i, dstB##i, mask##i);   \
-  vector unsigned char dstAF##j = vec_sel(dstA##j, dstR##j, mask##j);   \
-  vector unsigned char dstBF##j = vec_sel(dstR##j, dstB##j, mask##j);   \
-  vec_st(dstAF##i, i * stride, dst);                                    \
-  vec_st(dstBF##i, i * stride + 16, dst);                               \
-  vec_st(dstAF##j, j * stride, dst);                                    \
-  vec_st(dstBF##j, j * stride + 16, dst)
-
-  STORE_DOUBLE_LINE(0,1);
-  STORE_DOUBLE_LINE(2,3);
-  STORE_DOUBLE_LINE(4,5);
-  STORE_DOUBLE_LINE(6,7);
+    vector unsigned char dstA##i = vec_ld(i * stride, dst);             \
+    vector unsigned char dstB##i = vec_ld(i * stride + 16, dst);        \
+    vector unsigned char dstA##j = vec_ld(j * stride, dst);             \
+    vector unsigned char dstB##j = vec_ld(j * stride+ 16, dst);         \
+    vector unsigned char align##i = vec_lvsr(i * stride, dst);          \
+    vector unsigned char align##j = vec_lvsr(j * stride, dst);          \
+    vector unsigned char mask##i = vec_perm(zero, (vector unsigned char)neg1, align##i); \
+    vector unsigned char mask##j = vec_perm(zero, (vector unsigned char)neg1, align##j); \
+    vector unsigned char dstR##i = vec_perm(temp##i, temp##i, align##i);\
+    vector unsigned char dstR##j = vec_perm(temp##j, temp##j, align##j);\
+    vector unsigned char dstAF##i = vec_sel(dstA##i, dstR##i, mask##i); \
+    vector unsigned char dstBF##i = vec_sel(dstR##i, dstB##i, mask##i); \
+    vector unsigned char dstAF##j = vec_sel(dstA##j, dstR##j, mask##j); \
+    vector unsigned char dstBF##j = vec_sel(dstR##j, dstB##j, mask##j); \
+    vec_st(dstAF##i, i * stride, dst);                                  \
+    vec_st(dstBF##i, i * stride + 16, dst);                             \
+    vec_st(dstAF##j, j * stride, dst);                                  \
+    vec_st(dstBF##j, j * stride + 16, dst)
+
+    STORE_DOUBLE_LINE(0,1);
+    STORE_DOUBLE_LINE(2,3);
+    STORE_DOUBLE_LINE(4,5);
+    STORE_DOUBLE_LINE(6,7);
 }
index f2d40e1..c011687 100644 (file)
 #define FULL_Y_RANGE    0x8000                  // 32768
 
 //Deinterlacing Filters
-#define        LINEAR_IPOL_DEINT_FILTER         0x10000 // 65536
-#define        LINEAR_BLEND_DEINT_FILTER        0x20000 // 131072
-#define        CUBIC_BLEND_DEINT_FILTER         0x8000  // (not implemented yet)
-#define        CUBIC_IPOL_DEINT_FILTER          0x40000 // 262144
-#define        MEDIAN_DEINT_FILTER              0x80000 // 524288
-#define        FFMPEG_DEINT_FILTER              0x400000
-#define        LOWPASS5_DEINT_FILTER            0x800000
+#define LINEAR_IPOL_DEINT_FILTER        0x10000 // 65536
+#define LINEAR_BLEND_DEINT_FILTER       0x20000 // 131072
+#define CUBIC_BLEND_DEINT_FILTER        0x8000  // (not implemented yet)
+#define CUBIC_IPOL_DEINT_FILTER         0x40000 // 262144
+#define MEDIAN_DEINT_FILTER             0x80000 // 524288
+#define FFMPEG_DEINT_FILTER             0x400000
+#define LOWPASS5_DEINT_FILTER           0x800000
 
 #define TEMP_NOISE_FILTER               0x100000
 #define FORCE_QUANT                     0x200000
 //#define COMPILE_TIME_MODE 0x77
 
 static inline int CLIP(int a){
-        if(a&256) return ((a)>>31)^(-1);
-        else      return a;
+    if(a&256) return ((a)>>31)^(-1);
+    else      return a;
 }
 /**
  * Postprocessng filter.
  */
 struct PPFilter{
-        const char *shortName;
-        const char *longName;
-        int chromDefault;       ///< is chrominance filtering on by default if this filter is manually activated
-        int minLumQuality;      ///< minimum quality to turn luminance filtering on
-        int minChromQuality;    ///< minimum quality to turn chrominance filtering on
-        int mask;               ///< Bitmask to turn this filter on
+    const char *shortName;
+    const char *longName;
+    int chromDefault;       ///< is chrominance filtering on by default if this filter is manually activated
+    int minLumQuality;      ///< minimum quality to turn luminance filtering on
+    int minChromQuality;    ///< minimum quality to turn chrominance filtering on
+    int mask;               ///< Bitmask to turn this filter on
 };
 
 /**
  * Postprocessng mode.
  */
 typedef struct PPMode{
-        int lumMode;                    ///< acivates filters for luminance
-        int chromMode;                  ///< acivates filters for chrominance
-        int error;                      ///< non zero on error
+    int lumMode;                    ///< acivates filters for luminance
+    int chromMode;                  ///< acivates filters for chrominance
+    int error;                      ///< non zero on error
 
-        int minAllowedY;                ///< for brigtness correction
-        int maxAllowedY;                ///< for brihtness correction
-        float maxClippedThreshold;      ///< amount of "black" u r willing to loose to get a brightness corrected picture
+    int minAllowedY;                ///< for brigtness correction
+    int maxAllowedY;                ///< for brihtness correction
+    float maxClippedThreshold;      ///< amount of "black" u r willing to loose to get a brightness corrected picture
 
-        int maxTmpNoise[3];             ///< for Temporal Noise Reducing filter (Maximal sum of abs differences)
+    int maxTmpNoise[3];             ///< for Temporal Noise Reducing filter (Maximal sum of abs differences)
 
-        int baseDcDiff;
-        int flatnessThreshold;
+    int baseDcDiff;
+    int flatnessThreshold;
 
-        int forcedQuant;                ///< quantizer if FORCE_QUANT is used
+    int forcedQuant;                ///< quantizer if FORCE_QUANT is used
 } PPMode;
 
 /**
  * postprocess context.
  */
 typedef struct PPContext{
-        /**
-         * info on struct for av_log
-         */
-        const AVClass *av_class;
+    /**
+     * info on struct for av_log
+     */
+    const AVClass *av_class;
 
-        uint8_t *tempBlocks; ///<used for the horizontal code
+    uint8_t *tempBlocks; ///<used for the horizontal code
 
-        /**
-         * luma histogram.
-         * we need 64bit here otherwise we'll going to have a problem
-         * after watching a black picture for 5 hours
-         */
-        uint64_t *yHistogram;
+    /**
+     * luma histogram.
+     * we need 64bit here otherwise we'll going to have a problem
+     * after watching a black picture for 5 hours
+     */
+    uint64_t *yHistogram;
 
-        DECLARE_ALIGNED(8, uint64_t, packedYOffset);
-        DECLARE_ALIGNED(8, uint64_t, packedYScale);
+    DECLARE_ALIGNED(8, uint64_t, packedYOffset);
+    DECLARE_ALIGNED(8, uint64_t, packedYScale);
 
-        /** Temporal noise reducing buffers */
-        uint8_t *tempBlured[3];
-        int32_t *tempBluredPast[3];
+    /** Temporal noise reducing buffers */
+    uint8_t *tempBlured[3];
+    int32_t *tempBluredPast[3];
 
-        /** Temporary buffers for handling the last row(s) */
-        uint8_t *tempDst;
-        uint8_t *tempSrc;
+    /** Temporary buffers for handling the last row(s) */
+    uint8_t *tempDst;
+    uint8_t *tempSrc;
 
-        uint8_t *deintTemp;
+    uint8_t *deintTemp;
 
-        DECLARE_ALIGNED(8, uint64_t, pQPb);
-        DECLARE_ALIGNED(8, uint64_t, pQPb2);
+    DECLARE_ALIGNED(8, uint64_t, pQPb);
+    DECLARE_ALIGNED(8, uint64_t, pQPb2);
 
-        DECLARE_ALIGNED(8, uint64_t, mmxDcOffset[64]);
-        DECLARE_ALIGNED(8, uint64_t, mmxDcThreshold[64]);
+    DECLARE_ALIGNED(8, uint64_t, mmxDcOffset[64]);
+    DECLARE_ALIGNED(8, uint64_t, mmxDcThreshold[64]);
 
-        QP_STORE_T *stdQPTable;       ///< used to fix MPEG2 style qscale
-        QP_STORE_T *nonBQPTable;
-        QP_STORE_T *forcedQPTable;
+    QP_STORE_T *stdQPTable;       ///< used to fix MPEG2 style qscale
+    QP_STORE_T *nonBQPTable;
+    QP_STORE_T *forcedQPTable;
 
-        int QP;
-        int nonBQP;
+    int QP;
+    int nonBQP;
 
-        int frameNum;
+    int frameNum;
 
-        int cpuCaps;
+    int cpuCaps;
 
-        int qpStride; ///<size of qp buffers (needed to realloc them if needed)
-        int stride;   ///<size of some buffers (needed to realloc them if needed)
+    int qpStride; ///<size of qp buffers (needed to realloc them if needed)
+    int stride;   ///<size of some buffers (needed to realloc them if needed)
 
-        int hChromaSubSample;
-        int vChromaSubSample;
+    int hChromaSubSample;
+    int vChromaSubSample;
 
-        PPMode ppMode;
+    PPMode ppMode;
 } PPContext;
 
 
-static inline void linecpy(void *dest, const void *src, int lines, int stride)
-{
-        if (stride > 0) {
-                memcpy(dest, src, lines*stride);
-        } else {
-                memcpy((uint8_t*)dest+(lines-1)*stride, (const uint8_t*)src+(lines-1)*stride, -lines*stride);
-        }
+static inline void linecpy(void *dest, const void *src, int lines, int stride) {
+    if (stride > 0) {
+        memcpy(dest, src, lines*stride);
+    } else {
+        memcpy((uint8_t*)dest+(lines-1)*stride, (const uint8_t*)src+(lines-1)*stride, -lines*stride);
+    }
 }
 
 #endif /* FFMPEG_POSTPROCESS_INTERNAL_H */
index 74d8a5f..dffdf45 100644 (file)
 #define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
 #elif defined (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
 #define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
 #elif defined (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 ...)
  * 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"
+        "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"
+        "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" ((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;
+    }
 }
 #endif //HAVE_MMX
 
@@ -180,172 +180,171 @@ asm volatile(
 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
-        );
+    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++;
-        }
+    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)
 }
 #endif //HAVE_ALTIVEC
@@ -366,92 +365,89 @@ static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
 static inline void RENAME(vertRK1Filter)(uint8_t *src, int stride, int QP)
 {
 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        src+= stride*3;
+    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"
+    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
-        );
+        "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;
-
-                }
+    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)
 }
@@ -467,128 +463,125 @@ static inline void RENAME(vertRK1Filter)(uint8_t *src, int stride, int QP)
 static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
 {
 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
-        src+= stride*3;
+    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
-        );
+        "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++;
+    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++;
+    }
 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
 }
 
@@ -597,569 +590,555 @@ static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext
 {
 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
 /*
-        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"
+        "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" ((long)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
+    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
+        "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 %%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"
+        "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)<