1. ffmpeg
  2. Commit 16e0bf7349853d36b604bbc51c8b7f203dafa4e2
Browse code

cosmetics: Fix indentation to be 4 spaces and consistently place {}.

Originally committed as revision 12552 to svn://svn.ffmpeg.org/ffmpeg/trunk

Diego Biurrun authored on 2008/03/23 00:46:34
Showing 5 changed files
libpostproc/postprocess.c
History View file @ 16e0bf7
... ... 
@@ -116,65 +116,65 @@ DECLARE_ASM_CONST(8, int, deringThreshold)= 20;
116 116
117 117 
 static struct PPFilter filters[]=
118 118 
 {
119 
-        {"hb", "hdeblock",              1, 1, 3, H_DEBLOCK},
120 
-        {"vb", "vdeblock",              1, 2, 4, V_DEBLOCK},
121 
-/*      {"hr", "rkhdeblock",            1, 1, 3, H_RK1_FILTER},
122 
-        {"vr", "rkvdeblock",            1, 2, 4, V_RK1_FILTER},*/
123 
-        {"h1", "x1hdeblock",            1, 1, 3, H_X1_FILTER},
124 
-        {"v1", "x1vdeblock",            1, 2, 4, V_X1_FILTER},
125 
-        {"ha", "ahdeblock",             1, 1, 3, H_A_DEBLOCK},
126 
-        {"va", "avdeblock",             1, 2, 4, V_A_DEBLOCK},
127 
-        {"dr", "dering",                1, 5, 6, DERING},
128 
-        {"al", "autolevels",            0, 1, 2, LEVEL_FIX},
129 
-        {"lb", "linblenddeint",         1, 1, 4, LINEAR_BLEND_DEINT_FILTER},
130 
-        {"li", "linipoldeint",          1, 1, 4, LINEAR_IPOL_DEINT_FILTER},
131 
-        {"ci", "cubicipoldeint",        1, 1, 4, CUBIC_IPOL_DEINT_FILTER},
132 
-        {"md", "mediandeint",           1, 1, 4, MEDIAN_DEINT_FILTER},
133 
-        {"fd", "ffmpegdeint",           1, 1, 4, FFMPEG_DEINT_FILTER},
134 
-        {"l5", "lowpass5",              1, 1, 4, LOWPASS5_DEINT_FILTER},
135 
-        {"tn", "tmpnoise",              1, 7, 8, TEMP_NOISE_FILTER},
136 
-        {"fq", "forcequant",            1, 0, 0, FORCE_QUANT},
137 
-        {NULL, NULL,0,0,0,0} //End Marker
119 
+    {"hb", "hdeblock",              1, 1, 3, H_DEBLOCK},
120 
+    {"vb", "vdeblock",              1, 2, 4, V_DEBLOCK},
121 
+/*  {"hr", "rkhdeblock",            1, 1, 3, H_RK1_FILTER},
122 
+    {"vr", "rkvdeblock",            1, 2, 4, V_RK1_FILTER},*/
123 
+    {"h1", "x1hdeblock",            1, 1, 3, H_X1_FILTER},
124 
+    {"v1", "x1vdeblock",            1, 2, 4, V_X1_FILTER},
125 
+    {"ha", "ahdeblock",             1, 1, 3, H_A_DEBLOCK},
126 
+    {"va", "avdeblock",             1, 2, 4, V_A_DEBLOCK},
127 
+    {"dr", "dering",                1, 5, 6, DERING},
128 
+    {"al", "autolevels",            0, 1, 2, LEVEL_FIX},
129 
+    {"lb", "linblenddeint",         1, 1, 4, LINEAR_BLEND_DEINT_FILTER},
130 
+    {"li", "linipoldeint",          1, 1, 4, LINEAR_IPOL_DEINT_FILTER},
131 
+    {"ci", "cubicipoldeint",        1, 1, 4, CUBIC_IPOL_DEINT_FILTER},
132 
+    {"md", "mediandeint",           1, 1, 4, MEDIAN_DEINT_FILTER},
133 
+    {"fd", "ffmpegdeint",           1, 1, 4, FFMPEG_DEINT_FILTER},
134 
+    {"l5", "lowpass5",              1, 1, 4, LOWPASS5_DEINT_FILTER},
135 
+    {"tn", "tmpnoise",              1, 7, 8, TEMP_NOISE_FILTER},
136 
+    {"fq", "forcequant",            1, 0, 0, FORCE_QUANT},
137 
+    {NULL, NULL,0,0,0,0} //End Marker
138 138 
 };
139 139
140 140 
 static const char *replaceTable[]=
141 141 
 {
142 
-        "default",      "hb:a,vb:a,dr:a",
143 
-        "de",           "hb:a,vb:a,dr:a",
144 
-        "fast",         "h1:a,v1:a,dr:a",
145 
-        "fa",           "h1:a,v1:a,dr:a",
146 
-        "ac",           "ha:a:128:7,va:a,dr:a",
147 
-        NULL //End Marker
142 
+    "default",      "hb:a,vb:a,dr:a",
143 
+    "de",           "hb:a,vb:a,dr:a",
144 
+    "fast",         "h1:a,v1:a,dr:a",
145 
+    "fa",           "h1:a,v1:a,dr:a",
146 
+    "ac",           "ha:a:128:7,va:a,dr:a",
147 
+    NULL //End Marker
148 148 
 };
149 149
150 150
151 151 
 #if defined(ARCH_X86)
152 152 
 static inline void prefetchnta(void *p)
153 153 
 {
154 
-        asm volatile(   "prefetchnta (%0)\n\t"
155 
-                : : "r" (p)
156 
-        );
154 
+    asm volatile(   "prefetchnta (%0)\n\t"
155 
+        : : "r" (p)
156 
+    );
157 157 
 }
158 158
159 159 
 static inline void prefetcht0(void *p)
160 160 
 {
161 
-        asm volatile(   "prefetcht0 (%0)\n\t"
162 
-                : : "r" (p)
163 
-        );
161 
+    asm volatile(   "prefetcht0 (%0)\n\t"
162 
+        : : "r" (p)
163 
+    );
164 164 
 }
165 165
166 166 
 static inline void prefetcht1(void *p)
167 167 
 {
168 
-        asm volatile(   "prefetcht1 (%0)\n\t"
169 
-                : : "r" (p)
170 
-        );
168 
+    asm volatile(   "prefetcht1 (%0)\n\t"
169 
+        : : "r" (p)
170 
+    );
171 171 
 }
172 172
173 173 
 static inline void prefetcht2(void *p)
174 174 
 {
175 
-        asm volatile(   "prefetcht2 (%0)\n\t"
176 
-                : : "r" (p)
177 
-        );
175 
+    asm volatile(   "prefetcht2 (%0)\n\t"
176 
+        : : "r" (p)
177 
+    );
178 178 
 }
179 179 
 #endif
180 180
... ... 
@@ -185,171 +185,167 @@ static inline void prefetcht2(void *p)
185 185 
  */
186 186 
 static inline int isHorizDC_C(uint8_t src[], int stride, PPContext *c)
187 187 
 {
188 
-        int numEq= 0;
189 
-        int y;
190 
-        const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
191 
-        const int dcThreshold= dcOffset*2 + 1;
192 
-
193 
-        for(y=0; y<BLOCK_SIZE; y++)
194 
-        {
195 
-                if(((unsigned)(src[0] - src[1] + dcOffset)) < dcThreshold) numEq++;
196 
-                if(((unsigned)(src[1] - src[2] + dcOffset)) < dcThreshold) numEq++;
197 
-                if(((unsigned)(src[2] - src[3] + dcOffset)) < dcThreshold) numEq++;
198 
-                if(((unsigned)(src[3] - src[4] + dcOffset)) < dcThreshold) numEq++;
199 
-                if(((unsigned)(src[4] - src[5] + dcOffset)) < dcThreshold) numEq++;
200 
-                if(((unsigned)(src[5] - src[6] + dcOffset)) < dcThreshold) numEq++;
201 
-                if(((unsigned)(src[6] - src[7] + dcOffset)) < dcThreshold) numEq++;
202 
-                src+= stride;
203 
-        }
204 
-        return numEq > c->ppMode.flatnessThreshold;
188 
+    int numEq= 0;
189 
+    int y;
190 
+    const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
191 
+    const int dcThreshold= dcOffset*2 + 1;
192 
+
193 
+    for(y=0; y<BLOCK_SIZE; y++){
194 
+        if(((unsigned)(src[0] - src[1] + dcOffset)) < dcThreshold) numEq++;
195 
+        if(((unsigned)(src[1] - src[2] + dcOffset)) < dcThreshold) numEq++;
196 
+        if(((unsigned)(src[2] - src[3] + dcOffset)) < dcThreshold) numEq++;
197 
+        if(((unsigned)(src[3] - src[4] + dcOffset)) < dcThreshold) numEq++;
198 
+        if(((unsigned)(src[4] - src[5] + dcOffset)) < dcThreshold) numEq++;
199 
+        if(((unsigned)(src[5] - src[6] + dcOffset)) < dcThreshold) numEq++;
200 
+        if(((unsigned)(src[6] - src[7] + dcOffset)) < dcThreshold) numEq++;
201 
+        src+= stride;
202 
+    }
203 
+    return numEq > c->ppMode.flatnessThreshold;
205 204 
 }
206 205
207 206 
 /**
208 207 
  * Check if the middle 8x8 Block in the given 8x16 block is flat
209 208 
  */
210 
-static inline int isVertDC_C(uint8_t src[], int stride, PPContext *c){
211 
-        int numEq= 0;
212 
-        int y;
213 
-        const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
214 
-        const int dcThreshold= dcOffset*2 + 1;
215 
-
216 
-        src+= stride*4; // src points to begin of the 8x8 Block
217 
-        for(y=0; y<BLOCK_SIZE-1; y++)
218 
-        {
219 
-                if(((unsigned)(src[0] - src[0+stride] + dcOffset)) < dcThreshold) numEq++;
220 
-                if(((unsigned)(src[1] - src[1+stride] + dcOffset)) < dcThreshold) numEq++;
221 
-                if(((unsigned)(src[2] - src[2+stride] + dcOffset)) < dcThreshold) numEq++;
222 
-                if(((unsigned)(src[3] - src[3+stride] + dcOffset)) < dcThreshold) numEq++;
223 
-                if(((unsigned)(src[4] - src[4+stride] + dcOffset)) < dcThreshold) numEq++;
224 
-                if(((unsigned)(src[5] - src[5+stride] + dcOffset)) < dcThreshold) numEq++;
225 
-                if(((unsigned)(src[6] - src[6+stride] + dcOffset)) < dcThreshold) numEq++;
226 
-                if(((unsigned)(src[7] - src[7+stride] + dcOffset)) < dcThreshold) numEq++;
227 
-                src+= stride;
228 
-        }
229 
-        return numEq > c->ppMode.flatnessThreshold;
209 
+static inline int isVertDC_C(uint8_t src[], int stride, PPContext *c)
210 
+{
211 
+    int numEq= 0;
212 
+    int y;
213 
+    const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
214 
+    const int dcThreshold= dcOffset*2 + 1;
215 
+
216 
+    src+= stride*4; // src points to begin of the 8x8 Block
217 
+    for(y=0; y<BLOCK_SIZE-1; y++){
218 
+        if(((unsigned)(src[0] - src[0+stride] + dcOffset)) < dcThreshold) numEq++;
219 
+        if(((unsigned)(src[1] - src[1+stride] + dcOffset)) < dcThreshold) numEq++;
220 
+        if(((unsigned)(src[2] - src[2+stride] + dcOffset)) < dcThreshold) numEq++;
221 
+        if(((unsigned)(src[3] - src[3+stride] + dcOffset)) < dcThreshold) numEq++;
222 
+        if(((unsigned)(src[4] - src[4+stride] + dcOffset)) < dcThreshold) numEq++;
223 
+        if(((unsigned)(src[5] - src[5+stride] + dcOffset)) < dcThreshold) numEq++;
224 
+        if(((unsigned)(src[6] - src[6+stride] + dcOffset)) < dcThreshold) numEq++;
225 
+        if(((unsigned)(src[7] - src[7+stride] + dcOffset)) < dcThreshold) numEq++;
226 
+        src+= stride;
227 
+    }
228 
+    return numEq > c->ppMode.flatnessThreshold;
230 229 
 }
231 230
232 231 
 static inline int isHorizMinMaxOk_C(uint8_t src[], int stride, int QP)
233 232 
 {
234 
-        int i;
233 
+    int i;
235 234 
 #if 1
236 
-        for(i=0; i<2; i++){
237 
-                if((unsigned)(src[0] - src[5] + 2*QP) > 4*QP) return 0;
238 
-                src += stride;
239 
-                if((unsigned)(src[2] - src[7] + 2*QP) > 4*QP) return 0;
240 
-                src += stride;
241 
-                if((unsigned)(src[4] - src[1] + 2*QP) > 4*QP) return 0;
242 
-                src += stride;
243 
-                if((unsigned)(src[6] - src[3] + 2*QP) > 4*QP) return 0;
244 
-                src += stride;
245 
-        }
235 
+    for(i=0; i<2; i++){
236 
+        if((unsigned)(src[0] - src[5] + 2*QP) > 4*QP) return 0;
237 
+        src += stride;
238 
+        if((unsigned)(src[2] - src[7] + 2*QP) > 4*QP) return 0;
239 
+        src += stride;
240 
+        if((unsigned)(src[4] - src[1] + 2*QP) > 4*QP) return 0;
241 
+        src += stride;
242 
+        if((unsigned)(src[6] - src[3] + 2*QP) > 4*QP) return 0;
243 
+        src += stride;
244 
+    }
246 245 
 #else
247 
-        for(i=0; i<8; i++){
248 
-                if((unsigned)(src[0] - src[7] + 2*QP) > 4*QP) return 0;
249 
-                src += stride;
250 
-        }
246 
+    for(i=0; i<8; i++){
247 
+        if((unsigned)(src[0] - src[7] + 2*QP) > 4*QP) return 0;
248 
+        src += stride;
249 
+    }
251 250 
 #endif
252 
-        return 1;
251 
+    return 1;
253 252 
 }
254 253
255 254 
 static inline int isVertMinMaxOk_C(uint8_t src[], int stride, int QP)
256 255 
 {
257 256 
 #if 1
258 257 
 #if 1
259 
-        int x;
260 
-        src+= stride*4;
261 
-        for(x=0; x<BLOCK_SIZE; x+=4)
262 
-        {
263 
-                if((unsigned)(src[  x + 0*stride] - src[  x + 5*stride] + 2*QP) > 4*QP) return 0;
264 
-                if((unsigned)(src[1+x + 2*stride] - src[1+x + 7*stride] + 2*QP) > 4*QP) return 0;
265 
-                if((unsigned)(src[2+x + 4*stride] - src[2+x + 1*stride] + 2*QP) > 4*QP) return 0;
266 
-                if((unsigned)(src[3+x + 6*stride] - src[3+x + 3*stride] + 2*QP) > 4*QP) return 0;
267 
-        }
258 
+    int x;
259 
+    src+= stride*4;
260 
+    for(x=0; x<BLOCK_SIZE; x+=4){
261 
+        if((unsigned)(src[  x + 0*stride] - src[  x + 5*stride] + 2*QP) > 4*QP) return 0;
262 
+        if((unsigned)(src[1+x + 2*stride] - src[1+x + 7*stride] + 2*QP) > 4*QP) return 0;
263 
+        if((unsigned)(src[2+x + 4*stride] - src[2+x + 1*stride] + 2*QP) > 4*QP) return 0;
264 
+        if((unsigned)(src[3+x + 6*stride] - src[3+x + 3*stride] + 2*QP) > 4*QP) return 0;
265 
+    }
268 266 
 #else
269 
-        int x;
270 
-        src+= stride*3;
271 
-        for(x=0; x<BLOCK_SIZE; x++)
272 
-        {
273 
-                if((unsigned)(src[x + stride] - src[x + (stride<<3)] + 2*QP) > 4*QP) return 0;
274 
-        }
267 
+    int x;
268 
+    src+= stride*3;
269 
+    for(x=0; x<BLOCK_SIZE; x++){
270 
+        if((unsigned)(src[x + stride] - src[x + (stride<<3)] + 2*QP) > 4*QP) return 0;
271 
+    }
275 272 
 #endif
276 
-        return 1;
273 
+    return 1;
277 274 
 #else
278 
-        int x;
279 
-        src+= stride*4;
280 
-        for(x=0; x<BLOCK_SIZE; x++)
281 
-        {
282 
-                int min=255;
283 
-                int max=0;
284 
-                int y;
285 
-                for(y=0; y<8; y++){
286 
-                        int v= src[x + y*stride];
287 
-                        if(v>max) max=v;
288 
-                        if(v<min) min=v;
289 
-                }
290 
-                if(max-min > 2*QP) return 0;
275 
+    int x;
276 
+    src+= stride*4;
277 
+    for(x=0; x<BLOCK_SIZE; x++){
278 
+        int min=255;
279 
+        int max=0;
280 
+        int y;
281 
+        for(y=0; y<8; y++){
282 
+            int v= src[x + y*stride];
283 
+            if(v>max) max=v;
284 
+            if(v<min) min=v;
291 285 
         }
292 
-        return 1;
286 
+        if(max-min > 2*QP) return 0;
287 
+    }
288 
+    return 1;
293 289 
 #endif
294 290 
 }
295 291
296 
-static inline int horizClassify_C(uint8_t src[], int stride, PPContext *c){
297 
-        if( isHorizDC_C(src, stride, c) ){
298 
-                if( isHorizMinMaxOk_C(src, stride, c->QP) )
299 
-                        return 1;
300 
-                else
301 
-                        return 0;
302 
-        }else{
303 
-                return 2;
304 
-        }
292 
+static inline int horizClassify_C(uint8_t src[], int stride, PPContext *c)
293 
+{
294 
+    if( isHorizDC_C(src, stride, c) ){
295 
+        if( isHorizMinMaxOk_C(src, stride, c->QP) )
296 
+            return 1;
297 
+        else
298 
+            return 0;
299 
+    }else{
300 
+        return 2;
301 
+    }
305 302 
 }
306 303
307 
-static inline int vertClassify_C(uint8_t src[], int stride, PPContext *c){
308 
-        if( isVertDC_C(src, stride, c) ){
309 
-                if( isVertMinMaxOk_C(src, stride, c->QP) )
310 
-                        return 1;
311 
-                else
312 
-                        return 0;
313 
-        }else{
314 
-                return 2;
315 
-        }
304 
+static inline int vertClassify_C(uint8_t src[], int stride, PPContext *c)
305 
+{
306 
+    if( isVertDC_C(src, stride, c) ){
307 
+        if( isVertMinMaxOk_C(src, stride, c->QP) )
308 
+            return 1;
309 
+        else
310 
+            return 0;
311 
+    }else{
312 
+        return 2;
313 
+    }
316 314 
 }
317 315
318 316 
 static inline void doHorizDefFilter_C(uint8_t dst[], int stride, PPContext *c)
319 317 
 {
320 
-        int y;
321 
-        for(y=0; y<BLOCK_SIZE; y++)
322 
-        {
323 
-                const int middleEnergy= 5*(dst[4] - dst[3]) + 2*(dst[2] - dst[5]);
324 
-
325 
-                if(FFABS(middleEnergy) < 8*c->QP)
326 
-                {
327 
-                        const int q=(dst[3] - dst[4])/2;
328 
-                        const int leftEnergy=  5*(dst[2] - dst[1]) + 2*(dst[0] - dst[3]);
329 
-                        const int rightEnergy= 5*(dst[6] - dst[5]) + 2*(dst[4] - dst[7]);
330 
-
331 
-                        int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
332 
-                        d= FFMAX(d, 0);
333 
-
334 
-                        d= (5*d + 32) >> 6;
335 
-                        d*= FFSIGN(-middleEnergy);
336 
-
337 
-                        if(q>0)
338 
-                        {
339 
-                                d= d<0 ? 0 : d;
340 
-                                d= d>q ? q : d;
341 
-                        }
342 
-                        else
343 
-                        {
344 
-                                d= d>0 ? 0 : d;
345 
-                                d= d<q ? q : d;
346 
-                        }
347 
-
348 
-                        dst[3]-= d;
349 
-                        dst[4]+= d;
350 
-                }
351 
-                dst+= stride;
318 
+    int y;
319 
+    for(y=0; y<BLOCK_SIZE; y++){
320 
+        const int middleEnergy= 5*(dst[4] - dst[3]) + 2*(dst[2] - dst[5]);
321 
+
322 
+        if(FFABS(middleEnergy) < 8*c->QP){
323 
+            const int q=(dst[3] - dst[4])/2;
324 
+            const int leftEnergy=  5*(dst[2] - dst[1]) + 2*(dst[0] - dst[3]);
325 
+            const int rightEnergy= 5*(dst[6] - dst[5]) + 2*(dst[4] - dst[7]);
326 
+
327 
+            int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
328 
+            d= FFMAX(d, 0);
329 
+
330 
+            d= (5*d + 32) >> 6;
331 
+            d*= FFSIGN(-middleEnergy);
332 
+
333 
+            if(q>0)
334 
+            {
335 
+                d= d<0 ? 0 : d;
336 
+                d= d>q ? q : d;
337 
+            }
338 
+            else
339 
+            {
340 
+                d= d>0 ? 0 : d;
341 
+                d= d<q ? q : d;
342 
+            }
343 
+
344 
+            dst[3]-= d;
345 
+            dst[4]+= d;
352 346 
         }
347 
+        dst+= stride;
348 
+    }
353 349 
 }
354 350
355 351 
 /**
... ... 
@@ -358,35 +354,34 @@ static inline void doHorizDefFilter_C(uint8_t dst[], int stride, PPContext *c)
358 358 
  */
359 359 
 static inline void doHorizLowPass_C(uint8_t dst[], int stride, PPContext *c)
360 360 
 {
361 
-        int y;
362 
-        for(y=0; y<BLOCK_SIZE; y++)
363 
-        {
364 
-                const int first= FFABS(dst[-1] - dst[0]) < c->QP ? dst[-1] : dst[0];
365 
-                const int last= FFABS(dst[8] - dst[7]) < c->QP ? dst[8] : dst[7];
366 
-
367 
-                int sums[10];
368 
-                sums[0] = 4*first + dst[0] + dst[1] + dst[2] + 4;
369 
-                sums[1] = sums[0] - first  + dst[3];
370 
-                sums[2] = sums[1] - first  + dst[4];
371 
-                sums[3] = sums[2] - first  + dst[5];
372 
-                sums[4] = sums[3] - first  + dst[6];
373 
-                sums[5] = sums[4] - dst[0] + dst[7];
374 
-                sums[6] = sums[5] - dst[1] + last;
375 
-                sums[7] = sums[6] - dst[2] + last;
376 
-                sums[8] = sums[7] - dst[3] + last;
377 
-                sums[9] = sums[8] - dst[4] + last;
378 
-
379 
-                dst[0]= (sums[0] + sums[2] + 2*dst[0])>>4;
380 
-                dst[1]= (sums[1] + sums[3] + 2*dst[1])>>4;
381 
-                dst[2]= (sums[2] + sums[4] + 2*dst[2])>>4;
382 
-                dst[3]= (sums[3] + sums[5] + 2*dst[3])>>4;
383 
-                dst[4]= (sums[4] + sums[6] + 2*dst[4])>>4;
384 
-                dst[5]= (sums[5] + sums[7] + 2*dst[5])>>4;
385 
-                dst[6]= (sums[6] + sums[8] + 2*dst[6])>>4;
386 
-                dst[7]= (sums[7] + sums[9] + 2*dst[7])>>4;
387 
-
388 
-                dst+= stride;
389 
-        }
361 
+    int y;
362 
+    for(y=0; y<BLOCK_SIZE; y++){
363 
+        const int first= FFABS(dst[-1] - dst[0]) < c->QP ? dst[-1] : dst[0];
364 
+        const int last= FFABS(dst[8] - dst[7]) < c->QP ? dst[8] : dst[7];
365 
+
366 
+        int sums[10];
367 
+        sums[0] = 4*first + dst[0] + dst[1] + dst[2] + 4;
368 
+        sums[1] = sums[0] - first  + dst[3];
369 
+        sums[2] = sums[1] - first  + dst[4];
370 
+        sums[3] = sums[2] - first  + dst[5];
371 
+        sums[4] = sums[3] - first  + dst[6];
372 
+        sums[5] = sums[4] - dst[0] + dst[7];
373 
+        sums[6] = sums[5] - dst[1] + last;
374 
+        sums[7] = sums[6] - dst[2] + last;
375 
+        sums[8] = sums[7] - dst[3] + last;
376 
+        sums[9] = sums[8] - dst[4] + last;
377 
+
378 
+        dst[0]= (sums[0] + sums[2] + 2*dst[0])>>4;
379 
+        dst[1]= (sums[1] + sums[3] + 2*dst[1])>>4;
380 
+        dst[2]= (sums[2] + sums[4] + 2*dst[2])>>4;
381 
+        dst[3]= (sums[3] + sums[5] + 2*dst[3])>>4;
382 
+        dst[4]= (sums[4] + sums[6] + 2*dst[4])>>4;
383 
+        dst[5]= (sums[5] + sums[7] + 2*dst[5])>>4;
384 
+        dst[6]= (sums[6] + sums[8] + 2*dst[6])>>4;
385 
+        dst[7]= (sums[7] + sums[9] + 2*dst[7])>>4;
386 
+
387 
+        dst+= stride;
388 
+    }
390 389 
 }
391 390
392 391 
 /**
... ... 
@@ -399,161 +394,154 @@ static inline void doHorizLowPass_C(uint8_t dst[], int stride, PPContext *c)
399 399 
  */
400 400 
 static inline void horizX1Filter(uint8_t *src, int stride, int QP)
401 401 
 {
402 
-        int y;
403 
-        static uint64_t *lut= NULL;
404 
-        if(lut==NULL)
402 
+    int y;
403 
+    static uint64_t *lut= NULL;
404 
+    if(lut==NULL)
405 
+    {
406 
+        int i;
407 
+        lut = av_malloc(256*8);
408 
+        for(i=0; i<256; i++)
405 409 
         {
406 
-                int i;
407 
-                lut = av_malloc(256*8);
408 
-                for(i=0; i<256; i++)
409 
-                {
410 
-                        int v= i < 128 ? 2*i : 2*(i-256);
410 
+            int v= i < 128 ? 2*i : 2*(i-256);
411 411 
 /*
412 412 
 //Simulate 112242211 9-Tap filter
413 
-                        uint64_t a= (v/16) & 0xFF;
414 
-                        uint64_t b= (v/8) & 0xFF;
415 
-                        uint64_t c= (v/4) & 0xFF;
416 
-                        uint64_t d= (3*v/8) & 0xFF;
413 
+            uint64_t a= (v/16)  & 0xFF;
414 
+            uint64_t b= (v/8)   & 0xFF;
415 
+            uint64_t c= (v/4)   & 0xFF;
416 
+            uint64_t d= (3*v/8) & 0xFF;
417 417 
 */
418 418 
 //Simulate piecewise linear interpolation
419 
-                        uint64_t a= (v/16) & 0xFF;
420 
-                        uint64_t b= (v*3/16) & 0xFF;
421 
-                        uint64_t c= (v*5/16) & 0xFF;
422 
-                        uint64_t d= (7*v/16) & 0xFF;
423 
-                        uint64_t A= (0x100 - a)&0xFF;
424 
-                        uint64_t B= (0x100 - b)&0xFF;
425 
-                        uint64_t C= (0x100 - c)&0xFF;
426 
-                        uint64_t D= (0x100 - c)&0xFF;
427 
-
428 
-                        lut[i]   = (a<<56) | (b<<48) | (c<<40) | (d<<32) |
429 
-                                (D<<24) | (C<<16) | (B<<8) | (A);
430 
-                        //lut[i] = (v<<32) | (v<<24);
431 
-                }
419 
+            uint64_t a= (v/16)   & 0xFF;
420 
+            uint64_t b= (v*3/16) & 0xFF;
421 
+            uint64_t c= (v*5/16) & 0xFF;
422 
+            uint64_t d= (7*v/16) & 0xFF;
423 
+            uint64_t A= (0x100 - a)&0xFF;
424 
+            uint64_t B= (0x100 - b)&0xFF;
425 
+            uint64_t C= (0x100 - c)&0xFF;
426 
+            uint64_t D= (0x100 - c)&0xFF;
427 
+
428 
+            lut[i]   = (a<<56) | (b<<48) | (c<<40) | (d<<32) |
429 
+                       (D<<24) | (C<<16) | (B<<8)  | (A);
430 
+            //lut[i] = (v<<32) | (v<<24);
432 431 
         }
432 
+    }
433 433
434 
-        for(y=0; y<BLOCK_SIZE; y++)
435 
-        {
436 
-                int a= src[1] - src[2];
437 
-                int b= src[3] - src[4];
438 
-                int c= src[5] - src[6];
439 
-
440 
-                int d= FFMAX(FFABS(b) - (FFABS(a) + FFABS(c))/2, 0);
434 
+    for(y=0; y<BLOCK_SIZE; y++){
435 
+        int a= src[1] - src[2];
436 
+        int b= src[3] - src[4];
437 
+        int c= src[5] - src[6];
441 438
442 
-                if(d < QP)
443 
-                {
444 
-                        int v = d * FFSIGN(-b);
439 
+        int d= FFMAX(FFABS(b) - (FFABS(a) + FFABS(c))/2, 0);
445 440
446 
-                        src[1] +=v/8;
447 
-                        src[2] +=v/4;
448 
-                        src[3] +=3*v/8;
449 
-                        src[4] -=3*v/8;
450 
-                        src[5] -=v/4;
451 
-                        src[6] -=v/8;
441 
+        if(d < QP){
442 
+            int v = d * FFSIGN(-b);
452 443
453 
-                }
454 
-                src+=stride;
444 
+            src[1] +=v/8;
445 
+            src[2] +=v/4;
446 
+            src[3] +=3*v/8;
447 
+            src[4] -=3*v/8;
448 
+            src[5] -=v/4;
449 
+            src[6] -=v/8;
455 450 
         }
451 
+        src+=stride;
452 
+    }
456 453 
 }
457 454
458 455 
 /**
459 456 
  * accurate deblock filter
460 457 
  */
461 458 
 static av_always_inline void do_a_deblock_C(uint8_t *src, int step, int stride, PPContext *c){
462 
-        int y;
463 
-        const int QP= c->QP;
464 
-        const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
465 
-        const int dcThreshold= dcOffset*2 + 1;
459 
+    int y;
460 
+    const int QP= c->QP;
461 
+    const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
462 
+    const int dcThreshold= dcOffset*2 + 1;
466 463 
 //START_TIMER
467 
-        src+= step*4; // src points to begin of the 8x8 Block
468 
-        for(y=0; y<8; y++){
469 
-                int numEq= 0;
470 
-
471 
-                if(((unsigned)(src[-1*step] - src[0*step] + dcOffset)) < dcThreshold) numEq++;
472 
-                if(((unsigned)(src[ 0*step] - src[1*step] + dcOffset)) < dcThreshold) numEq++;
473 
-                if(((unsigned)(src[ 1*step] - src[2*step] + dcOffset)) < dcThreshold) numEq++;
474 
-                if(((unsigned)(src[ 2*step] - src[3*step] + dcOffset)) < dcThreshold) numEq++;
475 
-                if(((unsigned)(src[ 3*step] - src[4*step] + dcOffset)) < dcThreshold) numEq++;
476 
-                if(((unsigned)(src[ 4*step] - src[5*step] + dcOffset)) < dcThreshold) numEq++;
477 
-                if(((unsigned)(src[ 5*step] - src[6*step] + dcOffset)) < dcThreshold) numEq++;
478 
-                if(((unsigned)(src[ 6*step] - src[7*step] + dcOffset)) < dcThreshold) numEq++;
479 
-                if(((unsigned)(src[ 7*step] - src[8*step] + dcOffset)) < dcThreshold) numEq++;
480 
-                if(numEq > c->ppMode.flatnessThreshold){
481 
-                        int min, max, x;
482 
-
483 
-                        if(src[0] > src[step]){
484 
-                            max= src[0];
485 
-                            min= src[step];
486 
-                        }else{
487 
-                            max= src[step];
488 
-                            min= src[0];
489 
-                        }
490 
-                        for(x=2; x<8; x+=2){
491 
-                                if(src[x*step] > src[(x+1)*step]){
492 
-                                        if(src[x    *step] > max) max= src[ x   *step];
493 
-                                        if(src[(x+1)*step] < min) min= src[(x+1)*step];
494 
-                                }else{
495 
-                                        if(src[(x+1)*step] > max) max= src[(x+1)*step];
496 
-                                        if(src[ x   *step] < min) min= src[ x   *step];
497 
-                                }
498 
-                        }
499 
-                        if(max-min < 2*QP){
500 
-                                const int first= FFABS(src[-1*step] - src[0]) < QP ? src[-1*step] : src[0];
501 
-                                const int last= FFABS(src[8*step] - src[7*step]) < QP ? src[8*step] : src[7*step];
502 
-
503 
-                                int sums[10];
504 
-                                sums[0] = 4*first + src[0*step] + src[1*step] + src[2*step] + 4;
505 
-                                sums[1] = sums[0] - first       + src[3*step];
506 
-                                sums[2] = sums[1] - first       + src[4*step];
507 
-                                sums[3] = sums[2] - first       + src[5*step];
508 
-                                sums[4] = sums[3] - first       + src[6*step];
509 
-                                sums[5] = sums[4] - src[0*step] + src[7*step];
510 
-                                sums[6] = sums[5] - src[1*step] + last;
511 
-                                sums[7] = sums[6] - src[2*step] + last;
512 
-                                sums[8] = sums[7] - src[3*step] + last;
513 
-                                sums[9] = sums[8] - src[4*step] + last;
514 
-
515 
-                                src[0*step]= (sums[0] + sums[2] + 2*src[0*step])>>4;
516 
-                                src[1*step]= (sums[1] + sums[3] + 2*src[1*step])>>4;
517 
-                                src[2*step]= (sums[2] + sums[4] + 2*src[2*step])>>4;
518 
-                                src[3*step]= (sums[3] + sums[5] + 2*src[3*step])>>4;
519 
-                                src[4*step]= (sums[4] + sums[6] + 2*src[4*step])>>4;
520 
-                                src[5*step]= (sums[5] + sums[7] + 2*src[5*step])>>4;
521 
-                                src[6*step]= (sums[6] + sums[8] + 2*src[6*step])>>4;
522 
-                                src[7*step]= (sums[7] + sums[9] + 2*src[7*step])>>4;
523 
-                        }
464 
+    src+= step*4; // src points to begin of the 8x8 Block
465 
+    for(y=0; y<8; y++){
466 
+        int numEq= 0;
467 
+
468 
+        if(((unsigned)(src[-1*step] - src[0*step] + dcOffset)) < dcThreshold) numEq++;
469 
+        if(((unsigned)(src[ 0*step] - src[1*step] + dcOffset)) < dcThreshold) numEq++;
470 
+        if(((unsigned)(src[ 1*step] - src[2*step] + dcOffset)) < dcThreshold) numEq++;
471 
+        if(((unsigned)(src[ 2*step] - src[3*step] + dcOffset)) < dcThreshold) numEq++;
472 
+        if(((unsigned)(src[ 3*step] - src[4*step] + dcOffset)) < dcThreshold) numEq++;
473 
+        if(((unsigned)(src[ 4*step] - src[5*step] + dcOffset)) < dcThreshold) numEq++;
474 
+        if(((unsigned)(src[ 5*step] - src[6*step] + dcOffset)) < dcThreshold) numEq++;
475 
+        if(((unsigned)(src[ 6*step] - src[7*step] + dcOffset)) < dcThreshold) numEq++;
476 
+        if(((unsigned)(src[ 7*step] - src[8*step] + dcOffset)) < dcThreshold) numEq++;
477 
+        if(numEq > c->ppMode.flatnessThreshold){
478 
+            int min, max, x;
479 
+
480 
+            if(src[0] > src[step]){
481 
+                max= src[0];
482 
+                min= src[step];
483 
+            }else{
484 
+                max= src[step];
485 
+                min= src[0];
486 
+            }
487 
+            for(x=2; x<8; x+=2){
488 
+                if(src[x*step] > src[(x+1)*step]){
489 
+                        if(src[x    *step] > max) max= src[ x   *step];
490 
+                        if(src[(x+1)*step] < min) min= src[(x+1)*step];
524 491 
                 }else{
525 
-                        const int middleEnergy= 5*(src[4*step] - src[3*step]) + 2*(src[2*step] - src[5*step]);
526 
-
527 
-                        if(FFABS(middleEnergy) < 8*QP)
528 
-                        {
529 
-                                const int q=(src[3*step] - src[4*step])/2;
530 
-                                const int leftEnergy=  5*(src[2*step] - src[1*step]) + 2*(src[0*step] - src[3*step]);
531 
-                                const int rightEnergy= 5*(src[6*step] - src[5*step]) + 2*(src[4*step] - src[7*step]);
532 
-
533 
-                                int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
534 
-                                d= FFMAX(d, 0);
535 
-
536 
-                                d= (5*d + 32) >> 6;
537 
-                                d*= FFSIGN(-middleEnergy);
538 
-
539 
-                                if(q>0)
540 
-                                {
541 
-                                        d= d<0 ? 0 : d;
542 
-                                        d= d>q ? q : d;
543 
-                                }
544 
-                                else
545 
-                                {
546 
-                                        d= d>0 ? 0 : d;
547 
-                                        d= d<q ? q : d;
548 
-                                }
549 
-
550 
-                                src[3*step]-= d;
551 
-                                src[4*step]+= d;
552 
-                        }
492 
+                        if(src[(x+1)*step] > max) max= src[(x+1)*step];
493 
+                        if(src[ x   *step] < min) min= src[ x   *step];
494 
+                }
495 
+            }
496 
+            if(max-min < 2*QP){
497 
+                const int first= FFABS(src[-1*step] - src[0]) < QP ? src[-1*step] : src[0];
498 
+                const int last= FFABS(src[8*step] - src[7*step]) < QP ? src[8*step] : src[7*step];
499 
+
500 
+                int sums[10];
501 
+                sums[0] = 4*first + src[0*step] + src[1*step] + src[2*step] + 4;
502 
+                sums[1] = sums[0] - first       + src[3*step];
503 
+                sums[2] = sums[1] - first       + src[4*step];
504 
+                sums[3] = sums[2] - first       + src[5*step];
505 
+                sums[4] = sums[3] - first       + src[6*step];
506 
+                sums[5] = sums[4] - src[0*step] + src[7*step];
507 
+                sums[6] = sums[5] - src[1*step] + last;
508 
+                sums[7] = sums[6] - src[2*step] + last;
509 
+                sums[8] = sums[7] - src[3*step] + last;
510 
+                sums[9] = sums[8] - src[4*step] + last;
511 
+
512 
+                src[0*step]= (sums[0] + sums[2] + 2*src[0*step])>>4;
513 
+                src[1*step]= (sums[1] + sums[3] + 2*src[1*step])>>4;
514 
+                src[2*step]= (sums[2] + sums[4] + 2*src[2*step])>>4;
515 
+                src[3*step]= (sums[3] + sums[5] + 2*src[3*step])>>4;
516 
+                src[4*step]= (sums[4] + sums[6] + 2*src[4*step])>>4;
517 
+                src[5*step]= (sums[5] + sums[7] + 2*src[5*step])>>4;
518 
+                src[6*step]= (sums[6] + sums[8] + 2*src[6*step])>>4;
519 
+                src[7*step]= (sums[7] + sums[9] + 2*src[7*step])>>4;
520 
+            }
521 
+        }else{
522 
+            const int middleEnergy= 5*(src[4*step] - src[3*step]) + 2*(src[2*step] - src[5*step]);
523 
+
524 
+            if(FFABS(middleEnergy) < 8*QP){
525 
+                const int q=(src[3*step] - src[4*step])/2;
526 
+                const int leftEnergy=  5*(src[2*step] - src[1*step]) + 2*(src[0*step] - src[3*step]);
527 
+                const int rightEnergy= 5*(src[6*step] - src[5*step]) + 2*(src[4*step] - src[7*step]);
528 
+
529 
+                int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
530 
+                d= FFMAX(d, 0);
531 
+
532 
+                d= (5*d + 32) >> 6;
533 
+                d*= FFSIGN(-middleEnergy);
534 
+
535 
+                if(q>0){
536 
+                    d= d<0 ? 0 : d;
537 
+                    d= d>q ? q : d;
538 
+                }else{
539 
+                    d= d>0 ? 0 : d;
540 
+                    d= d<q ? q : d;
553 541 
                 }
554 542
555 
-                src += stride;
543 
+                src[3*step]-= d;
544 
+                src[4*step]+= d;
545 
+            }
556 546 
         }
547 
+
548 
+        src += stride;
549 
+    }
557 550 
 /*if(step==16){
558 551 
     STOP_TIMER("step16")
559 552 
 }else{
... ... 
@@ -642,43 +630,43 @@ static av_always_inline void do_a_deblock_C(uint8_t *src, int step, int stride,
642 642 
 static inline void postProcess(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
643 643 
         const QP_STORE_T QPs[], int QPStride, int isColor, pp_mode_t *vm, pp_context_t *vc)
644 644 
 {
645 
-        PPContext *c= (PPContext *)vc;
646 
-        PPMode *ppMode= (PPMode *)vm;
647 
-        c->ppMode= *ppMode; //FIXME
645 
+    PPContext *c= (PPContext *)vc;
646 
+    PPMode *ppMode= (PPMode *)vm;
647 
+    c->ppMode= *ppMode; //FIXME
648 648
649 
-        // Using ifs here as they are faster than function pointers although the
650 
-        // difference would not be measurable here but it is much better because
651 
-        // someone might exchange the CPU whithout restarting MPlayer ;)
649 
+    // Using ifs here as they are faster than function pointers although the
650 
+    // difference would not be measurable here but it is much better because
651 
+    // someone might exchange the CPU whithout restarting MPlayer ;)
652 652 
 #ifdef RUNTIME_CPUDETECT
653 653 
 #if defined(ARCH_X86)
654 
-        // ordered per speed fastest first
655 
-        if(c->cpuCaps & PP_CPU_CAPS_MMX2)
656 
-                postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
657 
-        else if(c->cpuCaps & PP_CPU_CAPS_3DNOW)
658 
-                postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
659 
-        else if(c->cpuCaps & PP_CPU_CAPS_MMX)
660 
-                postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
661 
-        else
662 
-                postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
654 
+    // ordered per speed fastest first
655 
+    if(c->cpuCaps & PP_CPU_CAPS_MMX2)
656 
+        postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
657 
+    else if(c->cpuCaps & PP_CPU_CAPS_3DNOW)
658 
+        postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
659 
+    else if(c->cpuCaps & PP_CPU_CAPS_MMX)
660 
+        postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
661 
+    else
662 
+        postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
663 663 
 #else
664 664 
 #ifdef HAVE_ALTIVEC
665 
-        if(c->cpuCaps & PP_CPU_CAPS_ALTIVEC)
666 
-                postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
667 
-        else
665 
+    if(c->cpuCaps & PP_CPU_CAPS_ALTIVEC)
666 
+            postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
667 
+    else
668 668 
 #endif
669 
-                postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
669 
+            postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
670 670 
 #endif
671 671 
 #else //RUNTIME_CPUDETECT
672 672 
 #ifdef HAVE_MMX2
673 
-                postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
673 
+            postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
674 674 
 #elif defined (HAVE_3DNOW)
675 
-                postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
675 
+            postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
676 676 
 #elif defined (HAVE_MMX)
677 
-                postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
677 
+            postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
678 678 
 #elif defined (HAVE_ALTIVEC)
679 
-                postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
679 
+            postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
680 680 
 #else
681 
-                postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
681 
+            postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
682 682 
 #endif
683 683 
 #endif //!RUNTIME_CPUDETECT
684 684 
 }
... ... 
@@ -738,194 +726,177 @@ const char pp_help[] =
738 738
739 739 
 pp_mode_t *pp_get_mode_by_name_and_quality(const char *name, int quality)
740 740 
 {
741 
-        char temp[GET_MODE_BUFFER_SIZE];
742 
-        char *p= temp;
743 
-        static const char filterDelimiters[] = ",/";
744 
-        static const char optionDelimiters[] = ":";
745 
-        struct PPMode *ppMode;
746 
-        char *filterToken;
747 
-
748 
-        ppMode= av_malloc(sizeof(PPMode));
749 
-
750 
-        ppMode->lumMode= 0;
751 
-        ppMode->chromMode= 0;
752 
-        ppMode->maxTmpNoise[0]= 700;
753 
-        ppMode->maxTmpNoise[1]= 1500;
754 
-        ppMode->maxTmpNoise[2]= 3000;
755 
-        ppMode->maxAllowedY= 234;
756 
-        ppMode->minAllowedY= 16;
757 
-        ppMode->baseDcDiff= 256/8;
758 
-        ppMode->flatnessThreshold= 56-16-1;
759 
-        ppMode->maxClippedThreshold= 0.01;
760 
-        ppMode->error=0;
761 
-
762 
-        strncpy(temp, name, GET_MODE_BUFFER_SIZE);
763 
-
764 
-        av_log(NULL, AV_LOG_DEBUG, "pp: %s\n", name);
765 
-
766 
-        for(;;){
767 
-                char *filterName;
768 
-                int q= 1000000; //PP_QUALITY_MAX;
769 
-                int chrom=-1;
770 
-                int luma=-1;
771 
-                char *option;
772 
-                char *options[OPTIONS_ARRAY_SIZE];
773 
-                int i;
774 
-                int filterNameOk=0;
775 
-                int numOfUnknownOptions=0;
776 
-                int enable=1; //does the user want us to enabled or disabled the filter
777 
-
778 
-                filterToken= strtok(p, filterDelimiters);
779 
-                if(filterToken == NULL) break;
780 
-                p+= strlen(filterToken) + 1; // p points to next filterToken
781 
-                filterName= strtok(filterToken, optionDelimiters);
782 
-                av_log(NULL, AV_LOG_DEBUG, "pp: %s::%s\n", filterToken, filterName);
783 
-
784 
-                if(*filterName == '-')
785 
-                {
786 
-                        enable=0;
787 
-                        filterName++;
788 
-                }
741 
+    char temp[GET_MODE_BUFFER_SIZE];
742 
+    char *p= temp;
743 
+    static const char filterDelimiters[] = ",/";
744 
+    static const char optionDelimiters[] = ":";
745 
+    struct PPMode *ppMode;
746 
+    char *filterToken;
747 
+
748 
+    ppMode= av_malloc(sizeof(PPMode));
749 
+
750 
+    ppMode->lumMode= 0;
751 
+    ppMode->chromMode= 0;
752 
+    ppMode->maxTmpNoise[0]= 700;
753 
+    ppMode->maxTmpNoise[1]= 1500;
754 
+    ppMode->maxTmpNoise[2]= 3000;
755 
+    ppMode->maxAllowedY= 234;
756 
+    ppMode->minAllowedY= 16;
757 
+    ppMode->baseDcDiff= 256/8;
758 
+    ppMode->flatnessThreshold= 56-16-1;
759 
+    ppMode->maxClippedThreshold= 0.01;
760 
+    ppMode->error=0;
761 
+
762 
+    strncpy(temp, name, GET_MODE_BUFFER_SIZE);
763 
+
764 
+    av_log(NULL, AV_LOG_DEBUG, "pp: %s\n", name);
765 
+
766 
+    for(;;){
767 
+        char *filterName;
768 
+        int q= 1000000; //PP_QUALITY_MAX;
769 
+        int chrom=-1;
770 
+        int luma=-1;
771 
+        char *option;
772 
+        char *options[OPTIONS_ARRAY_SIZE];
773 
+        int i;
774 
+        int filterNameOk=0;
775 
+        int numOfUnknownOptions=0;
776 
+        int enable=1; //does the user want us to enabled or disabled the filter
777 
+
778 
+        filterToken= strtok(p, filterDelimiters);
779 
+        if(filterToken == NULL) break;
780 
+        p+= strlen(filterToken) + 1; // p points to next filterToken
781 
+        filterName= strtok(filterToken, optionDelimiters);
782 
+        av_log(NULL, AV_LOG_DEBUG, "pp: %s::%s\n", filterToken, filterName);
783 
+
784 
+        if(*filterName == '-'){
785 
+            enable=0;
786 
+            filterName++;
787 
+        }
789 788
790 
-                for(;;){ //for all options
791 
-                        option= strtok(NULL, optionDelimiters);
792 
-                        if(option == NULL) break;
793 
-
794 
-                        av_log(NULL, AV_LOG_DEBUG, "pp: option: %s\n", option);
795 
-                        if(!strcmp("autoq", option) || !strcmp("a", option)) q= quality;
796 
-                        else if(!strcmp("nochrom", option) || !strcmp("y", option)) chrom=0;
797 
-                        else if(!strcmp("chrom", option) || !strcmp("c", option)) chrom=1;
798 
-                        else if(!strcmp("noluma", option) || !strcmp("n", option)) luma=0;
799 
-                        else
800 
-                        {
801 
-                                options[numOfUnknownOptions] = option;
802 
-                                numOfUnknownOptions++;
803 
-                        }
804 
-                        if(numOfUnknownOptions >= OPTIONS_ARRAY_SIZE-1) break;
789 
+        for(;;){ //for all options
790 
+            option= strtok(NULL, optionDelimiters);
791 
+            if(option == NULL) break;
792 
+
793 
+            av_log(NULL, AV_LOG_DEBUG, "pp: option: %s\n", option);
794 
+            if(!strcmp("autoq", option) || !strcmp("a", option)) q= quality;
795 
+            else if(!strcmp("nochrom", option) || !strcmp("y", option)) chrom=0;
796 
+            else if(!strcmp("chrom", option) || !strcmp("c", option)) chrom=1;
797 
+            else if(!strcmp("noluma", option) || !strcmp("n", option)) luma=0;
798 
+            else{
799 
+                options[numOfUnknownOptions] = option;
800 
+                numOfUnknownOptions++;
801 
+            }
802 
+            if(numOfUnknownOptions >= OPTIONS_ARRAY_SIZE-1) break;
803 
+        }
804 
+        options[numOfUnknownOptions] = NULL;
805 
+
806 
+        /* replace stuff from the replace Table */
807 
+        for(i=0; replaceTable[2*i]!=NULL; i++){
808 
+            if(!strcmp(replaceTable[2*i], filterName)){
809 
+                int newlen= strlen(replaceTable[2*i + 1]);
810 
+                int plen;
811 
+                int spaceLeft;
812 
+
813 
+                if(p==NULL) p= temp, *p=0;      //last filter
814 
+                else p--, *p=',';               //not last filter
815 
+
816 
+                plen= strlen(p);
817 
+                spaceLeft= p - temp + plen;
818 
+                if(spaceLeft + newlen  >= GET_MODE_BUFFER_SIZE){
819 
+                    ppMode->error++;
820 
+                    break;
805 821 
                 }
806 
-                options[numOfUnknownOptions] = NULL;
822 
+                memmove(p + newlen, p, plen+1);
823 
+                memcpy(p, replaceTable[2*i + 1], newlen);
824 
+                filterNameOk=1;
825 
+            }
826 
+        }
807 827
808 
-                /* replace stuff from the replace Table */
809 
-                for(i=0; replaceTable[2*i]!=NULL; i++)
810 
-                {
811 
-                        if(!strcmp(replaceTable[2*i], filterName))
812 
-                        {
813 
-                                int newlen= strlen(replaceTable[2*i + 1]);
814 
-                                int plen;
815 
-                                int spaceLeft;
816 
-
817 
-                                if(p==NULL) p= temp, *p=0;      //last filter
818 
-                                else p--, *p=',';               //not last filter
819 
-
820 
-                                plen= strlen(p);
821 
-                                spaceLeft= p - temp + plen;
822 
-                                if(spaceLeft + newlen  >= GET_MODE_BUFFER_SIZE)
823 
-                                {
824 
-                                        ppMode->error++;
825 
-                                        break;
826 
-                                }
827 
-                                memmove(p + newlen, p, plen+1);
828 
-                                memcpy(p, replaceTable[2*i + 1], newlen);
829 
-                                filterNameOk=1;
828 
+        for(i=0; filters[i].shortName!=NULL; i++){
829 
+            if(   !strcmp(filters[i].longName, filterName)
830 
+               || !strcmp(filters[i].shortName, filterName)){
831 
+                ppMode->lumMode &= ~filters[i].mask;
832 
+                ppMode->chromMode &= ~filters[i].mask;
833 
+
834 
+                filterNameOk=1;
835 
+                if(!enable) break; // user wants to disable it
836 
+
837 
+                if(q >= filters[i].minLumQuality && luma)
838 
+                    ppMode->lumMode|= filters[i].mask;
839 
+                if(chrom==1 || (chrom==-1 && filters[i].chromDefault))
840 
+                    if(q >= filters[i].minChromQuality)
841 
+                            ppMode->chromMode|= filters[i].mask;
842 
+
843 
+                if(filters[i].mask == LEVEL_FIX){
844 
+                    int o;
845 
+                    ppMode->minAllowedY= 16;
846 
+                    ppMode->maxAllowedY= 234;
847 
+                    for(o=0; options[o]!=NULL; o++){
848 
+                        if(  !strcmp(options[o],"fullyrange")
849 
+                           ||!strcmp(options[o],"f")){
850 
+                            ppMode->minAllowedY= 0;
851 
+                            ppMode->maxAllowedY= 255;
852 
+                            numOfUnknownOptions--;
830 853 
                         }
854 
+                    }
831 855 
                 }
832 
-
833 
-                for(i=0; filters[i].shortName!=NULL; i++)
856 
+                else if(filters[i].mask == TEMP_NOISE_FILTER)
834 857 
                 {
835 
-                        if(   !strcmp(filters[i].longName, filterName)
836 
-                           || !strcmp(filters[i].shortName, filterName))
837 
-                        {
838 
-                                ppMode->lumMode &= ~filters[i].mask;
839 
-                                ppMode->chromMode &= ~filters[i].mask;
840 
-
841 
-                                filterNameOk=1;
842 
-                                if(!enable) break; // user wants to disable it
843 
-
844 
-                                if(q >= filters[i].minLumQuality && luma)
845 
-                                        ppMode->lumMode|= filters[i].mask;
846 
-                                if(chrom==1 || (chrom==-1 && filters[i].chromDefault))
847 
-                                        if(q >= filters[i].minChromQuality)
848 
-                                                ppMode->chromMode|= filters[i].mask;
849 
-
850 
-                                if(filters[i].mask == LEVEL_FIX)
851 
-                                {
852 
-                                        int o;
853 
-                                        ppMode->minAllowedY= 16;
854 
-                                        ppMode->maxAllowedY= 234;
855 
-                                        for(o=0; options[o]!=NULL; o++)
856 
-                                        {
857 
-                                                if(  !strcmp(options[o],"fullyrange")
858 
-                                                   ||!strcmp(options[o],"f"))
859 
-                                                {
860 
-                                                        ppMode->minAllowedY= 0;
861 
-                                                        ppMode->maxAllowedY= 255;
862 
-                                                        numOfUnknownOptions--;
863 
-                                                }
864 
-                                        }
865 
-                                }
866 
-                                else if(filters[i].mask == TEMP_NOISE_FILTER)
867 
-                                {
868 
-                                        int o;
869 
-                                        int numOfNoises=0;
870 
-
871 
-                                        for(o=0; options[o]!=NULL; o++)
872 
-                                        {
873 
-                                                char *tail;
874 
-                                                ppMode->maxTmpNoise[numOfNoises]=
875 
-                                                        strtol(options[o], &tail, 0);
876 
-                                                if(tail!=options[o])
877 
-                                                {
878 
-                                                        numOfNoises++;
879 
-                                                        numOfUnknownOptions--;
880 
-                                                        if(numOfNoises >= 3) break;
881 
-                                                }
882 
-                                        }
883 
-                                }
884 
-                                else if(filters[i].mask == V_DEBLOCK   || filters[i].mask == H_DEBLOCK
885 
-                                     || filters[i].mask == V_A_DEBLOCK || filters[i].mask == H_A_DEBLOCK)
886 
-                                {
887 
-                                        int o;
888 
-
889 
-                                        for(o=0; options[o]!=NULL && o<2; o++)
890 
-                                        {
891 
-                                                char *tail;
892 
-                                                int val= strtol(options[o], &tail, 0);
893 
-                                                if(tail==options[o]) break;
894 
-
895 
-                                                numOfUnknownOptions--;
896 
-                                                if(o==0) ppMode->baseDcDiff= val;
897 
-                                                else ppMode->flatnessThreshold= val;
898 
-                                        }
899 
-                                }
900 
-                                else if(filters[i].mask == FORCE_QUANT)
901 
-                                {
902 
-                                        int o;
903 
-                                        ppMode->forcedQuant= 15;
904 
-
905 
-                                        for(o=0; options[o]!=NULL && o<1; o++)
906 
-                                        {
907 
-                                                char *tail;
908 
-                                                int val= strtol(options[o], &tail, 0);
909 
-                                                if(tail==options[o]) break;
910 
-
911 
-                                                numOfUnknownOptions--;
912 
-                                                ppMode->forcedQuant= val;
913 
-                                        }
914 
-                                }
858 
+                    int o;
859 
+                    int numOfNoises=0;
860 
+
861 
+                    for(o=0; options[o]!=NULL; o++){
862 
+                        char *tail;
863 
+                        ppMode->maxTmpNoise[numOfNoises]=
864 
+                            strtol(options[o], &tail, 0);
865 
+                        if(tail!=options[o]){
866 
+                            numOfNoises++;
867 
+                            numOfUnknownOptions--;
868 
+                            if(numOfNoises >= 3) break;
915 869 
                         }
870 
+                    }
916 871 
                 }
917 
-                if(!filterNameOk) ppMode->error++;
918 
-                ppMode->error += numOfUnknownOptions;
919 
-        }
920 
-
921 
-        av_log(NULL, AV_LOG_DEBUG, "pp: lumMode=%X, chromMode=%X\n", ppMode->lumMode, ppMode->chromMode);
922 
-        if(ppMode->error)
923 
-        {
924 
-                av_log(NULL, AV_LOG_ERROR, "%d errors in postprocess string \"%s\"\n", ppMode->error, name);
925 
-                av_free(ppMode);
926 
-                return NULL;
872 
+                else if(filters[i].mask == V_DEBLOCK   || filters[i].mask == H_DEBLOCK
873 
+                     || filters[i].mask == V_A_DEBLOCK || filters[i].mask == H_A_DEBLOCK){
874 
+                    int o;
875 
+
876 
+                    for(o=0; options[o]!=NULL && o<2; o++){
877 
+                        char *tail;
878 
+                        int val= strtol(options[o], &tail, 0);
879 
+                        if(tail==options[o]) break;
880 
+
881 
+                        numOfUnknownOptions--;
882 
+                        if(o==0) ppMode->baseDcDiff= val;
883 
+                        else ppMode->flatnessThreshold= val;
884 
+                    }
885 
+                }
886 
+                else if(filters[i].mask == FORCE_QUANT){
887 
+                    int o;
888 
+                    ppMode->forcedQuant= 15;
889 
+
890 
+                    for(o=0; options[o]!=NULL && o<1; o++){
891 
+                        char *tail;
892 
+                        int val= strtol(options[o], &tail, 0);
893 
+                        if(tail==options[o]) break;
894 
+
895 
+                        numOfUnknownOptions--;
896 
+                        ppMode->forcedQuant= val;
897 
+                    }
898 
+                }
899 
+            }
927 900 
         }
928 
-        return ppMode;
901 
+        if(!filterNameOk) ppMode->error++;
902 
+        ppMode->error += numOfUnknownOptions;
903 
+    }
904 
+
905 
+    av_log(NULL, AV_LOG_DEBUG, "pp: lumMode=%X, chromMode=%X\n", ppMode->lumMode, ppMode->chromMode);
906 
+    if(ppMode->error){
907 
+        av_log(NULL, AV_LOG_ERROR, "%d errors in postprocess string \"%s\"\n", ppMode->error, name);
908 
+        av_free(ppMode);
909 
+        return NULL;
910 
+    }
911 
+    return ppMode;
929 912 
 }
930 913
931 914 
 void pp_free_mode(pp_mode_t *mode){
... ... 
@@ -933,36 +904,35 @@ void pp_free_mode(pp_mode_t *mode){
933 933 
 }
934 934
935 935 
 static void reallocAlign(void **p, int alignment, int size){
936 
-        av_free(*p);
937 
-        *p= av_mallocz(size);
936 
+    av_free(*p);
937 
+    *p= av_mallocz(size);
938 938 
 }
939 939
940 940 
 static void reallocBuffers(PPContext *c, int width, int height, int stride, int qpStride){
941 
-        int mbWidth = (width+15)>>4;
942 
-        int mbHeight= (height+15)>>4;
943 
-        int i;
944 
-
945 
-        c->stride= stride;
946 
-        c->qpStride= qpStride;
947 
-
948 
-        reallocAlign((void **)&c->tempDst, 8, stride*24);
949 
-        reallocAlign((void **)&c->tempSrc, 8, stride*24);
950 
-        reallocAlign((void **)&c->tempBlocks, 8, 2*16*8);
951 
-        reallocAlign((void **)&c->yHistogram, 8, 256*sizeof(uint64_t));
952 
-        for(i=0; i<256; i++)
953 
-                c->yHistogram[i]= width*height/64*15/256;
954 
-
955 
-        for(i=0; i<3; i++)
956 
-        {
957 
-                //Note: The +17*1024 is just there so i do not have to worry about r/w over the end.
958 
-                reallocAlign((void **)&c->tempBlured[i], 8, stride*mbHeight*16 + 17*1024);
959 
-                reallocAlign((void **)&c->tempBluredPast[i], 8, 256*((height+7)&(~7))/2 + 17*1024);//FIXME size
960 
-        }
961 
-
962 
-        reallocAlign((void **)&c->deintTemp, 8, 2*width+32);
963 
-        reallocAlign((void **)&c->nonBQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T));
964 
-        reallocAlign((void **)&c->stdQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T));
965 
-        reallocAlign((void **)&c->forcedQPTable, 8, mbWidth*sizeof(QP_STORE_T));
941 
+    int mbWidth = (width+15)>>4;
942 
+    int mbHeight= (height+15)>>4;
943 
+    int i;
944 
+
945 
+    c->stride= stride;
946 
+    c->qpStride= qpStride;
947 
+
948 
+    reallocAlign((void **)&c->tempDst, 8, stride*24);
949 
+    reallocAlign((void **)&c->tempSrc, 8, stride*24);
950 
+    reallocAlign((void **)&c->tempBlocks, 8, 2*16*8);
951 
+    reallocAlign((void **)&c->yHistogram, 8, 256*sizeof(uint64_t));
952 
+    for(i=0; i<256; i++)
953 
+            c->yHistogram[i]= width*height/64*15/256;
954 
+
955 
+    for(i=0; i<3; i++){
956 
+        //Note: The +17*1024 is just there so i do not have to worry about r/w over the end.
957 
+        reallocAlign((void **)&c->tempBlured[i], 8, stride*mbHeight*16 + 17*1024);
958 
+        reallocAlign((void **)&c->tempBluredPast[i], 8, 256*((height+7)&(~7))/2 + 17*1024);//FIXME size
959 
+    }
960 
+
961 
+    reallocAlign((void **)&c->deintTemp, 8, 2*width+32);
962 
+    reallocAlign((void **)&c->nonBQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T));
963 
+    reallocAlign((void **)&c->stdQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T));
964 
+    reallocAlign((void **)&c->forcedQPTable, 8, mbWidth*sizeof(QP_STORE_T));
966 965 
 }
967 966
968 967 
 static const char * context_to_name(void * ptr) {
... ... 
@@ -972,153 +942,146 @@ static const char * context_to_name(void * ptr) {
972 972 
 static const AVClass av_codec_context_class = { "Postproc", context_to_name, NULL };
973 973
974 974 
 pp_context_t *pp_get_context(int width, int height, int cpuCaps){
975 
-        PPContext *c= av_malloc(sizeof(PPContext));
976 
-        int stride= (width+15)&(~15);    //assumed / will realloc if needed
977 
-        int qpStride= (width+15)/16 + 2; //assumed / will realloc if needed
978 
-
979 
-        memset(c, 0, sizeof(PPContext));
980 
-        c->av_class = &av_codec_context_class;
981 
-        c->cpuCaps= cpuCaps;
982 
-        if(cpuCaps&PP_FORMAT){
983 
-                c->hChromaSubSample= cpuCaps&0x3;
984 
-                c->vChromaSubSample= (cpuCaps>>4)&0x3;
985 
-        }else{
986 
-                c->hChromaSubSample= 1;
987 
-                c->vChromaSubSample= 1;
988 
-        }
989 
-
990 
-        reallocBuffers(c, width, height, stride, qpStride);
991 
-
992 
-        c->frameNum=-1;
993 
-
994 
-        return c;
975 
+    PPContext *c= av_malloc(sizeof(PPContext));
976 
+    int stride= (width+15)&(~15);    //assumed / will realloc if needed
977 
+    int qpStride= (width+15)/16 + 2; //assumed / will realloc if needed
978 
+
979 
+    memset(c, 0, sizeof(PPContext));
980 
+    c->av_class = &av_codec_context_class;
981 
+    c->cpuCaps= cpuCaps;
982 
+    if(cpuCaps&PP_FORMAT){
983 
+        c->hChromaSubSample= cpuCaps&0x3;
984 
+        c->vChromaSubSample= (cpuCaps>>4)&0x3;
985 
+    }else{
986 
+        c->hChromaSubSample= 1;
987 
+        c->vChromaSubSample= 1;
988 
+    }
989 
+
990 
+    reallocBuffers(c, width, height, stride, qpStride);
991 
+
992 
+    c->frameNum=-1;
993 
+
994 
+    return c;
995 995 
 }
996 996
997 997 
 void pp_free_context(void *vc){
998 
-        PPContext *c = (PPContext*)vc;
999 
-        int i;
998 
+    PPContext *c = (PPContext*)vc;
999 
+    int i;
1000 1000
1001 
-        for(i=0; i<3; i++) av_free(c->tempBlured[i]);
1002 
-        for(i=0; i<3; i++) av_free(c->tempBluredPast[i]);
1001 
+    for(i=0; i<3; i++) av_free(c->tempBlured[i]);
1002 
+    for(i=0; i<3; i++) av_free(c->tempBluredPast[i]);
1003 1003
1004 
-        av_free(c->tempBlocks);
1005 
-        av_free(c->yHistogram);
1006 
-        av_free(c->tempDst);
1007 
-        av_free(c->tempSrc);
1008 
-        av_free(c->deintTemp);
1009 
-        av_free(c->stdQPTable);
1010 
-        av_free(c->nonBQPTable);
1011 
-        av_free(c->forcedQPTable);
1004 
+    av_free(c->tempBlocks);
1005 
+    av_free(c->yHistogram);
1006 
+    av_free(c->tempDst);
1007 
+    av_free(c->tempSrc);
1008 
+    av_free(c->deintTemp);
1009 
+    av_free(c->stdQPTable);
1010 
+    av_free(c->nonBQPTable);
1011 
+    av_free(c->forcedQPTable);
1012 1012
1013 
-        memset(c, 0, sizeof(PPContext));
1013 
+    memset(c, 0, sizeof(PPContext));
1014 1014
1015 
-        av_free(c);
1015 
+    av_free(c);
1016 1016 
 }
1017 1017
1018 1018 
 void  pp_postprocess(const uint8_t * src[3], const int srcStride[3],
1019 
-                 uint8_t * dst[3], const int dstStride[3],
1020 
-                 int width, int height,
1021 
-                 const QP_STORE_T *QP_store,  int QPStride,
1022 
-                 pp_mode_t *vm,  void *vc, int pict_type)
1019 
+                     uint8_t * dst[3], const int dstStride[3],
1020 
+                     int width, int height,
1021 
+                     const QP_STORE_T *QP_store,  int QPStride,
1022 
+                     pp_mode_t *vm,  void *vc, int pict_type)
1023 1023 
 {
1024 
-        int mbWidth = (width+15)>>4;
1025 
-        int mbHeight= (height+15)>>4;
1026 
-        PPMode *mode = (PPMode*)vm;
1027 
-        PPContext *c = (PPContext*)vc;
1028 
-        int minStride= FFMAX(FFABS(srcStride[0]), FFABS(dstStride[0]));
1029 
-        int absQPStride = FFABS(QPStride);
1030 
-
1031 
-        // c->stride and c->QPStride are always positive
1032 
-        if(c->stride < minStride || c->qpStride < absQPStride)
1033 
-                reallocBuffers(c, width, height,
1034 
-                                FFMAX(minStride, c->stride),
1035 
-                                FFMAX(c->qpStride, absQPStride));
1036 
-
1037 
-        if(QP_store==NULL || (mode->lumMode & FORCE_QUANT))
1038 
-        {
1039 
-                int i;
1040 
-                QP_store= c->forcedQPTable;
1041 
-                absQPStride = QPStride = 0;
1042 
-                if(mode->lumMode & FORCE_QUANT)
1043 
-                        for(i=0; i<mbWidth; i++) c->forcedQPTable[i]= mode->forcedQuant;
1044 
-                else
1045 
-                        for(i=0; i<mbWidth; i++) c->forcedQPTable[i]= 1;
1046 
-        }
1024 
+    int mbWidth = (width+15)>>4;
1025 
+    int mbHeight= (height+15)>>4;
1026 
+    PPMode *mode = (PPMode*)vm;
1027 
+    PPContext *c = (PPContext*)vc;
1028 
+    int minStride= FFMAX(FFABS(srcStride[0]), FFABS(dstStride[0]));
1029 
+    int absQPStride = FFABS(QPStride);
1030 
+
1031 
+    // c->stride and c->QPStride are always positive
1032 
+    if(c->stride < minStride || c->qpStride < absQPStride)
1033 
+        reallocBuffers(c, width, height,
1034 
+                       FFMAX(minStride, c->stride),
1035 
+                       FFMAX(c->qpStride, absQPStride));
1036 
+
1037 
+    if(QP_store==NULL || (mode->lumMode & FORCE_QUANT)){
1038 
+        int i;
1039 
+        QP_store= c->forcedQPTable;
1040 
+        absQPStride = QPStride = 0;
1041 
+        if(mode->lumMode & FORCE_QUANT)
1042 
+            for(i=0; i<mbWidth; i++) c->forcedQPTable[i]= mode->forcedQuant;
1043 
+        else
1044 
+            for(i=0; i<mbWidth; i++) c->forcedQPTable[i]= 1;
1045 
+    }
1047 1046
1048 
-        if(pict_type & PP_PICT_TYPE_QP2){
1049 
-                int i;
1050 
-                const int count= mbHeight * absQPStride;
1051 
-                for(i=0; i<(count>>2); i++){
1052 
-                        ((uint32_t*)c->stdQPTable)[i] = (((const uint32_t*)QP_store)[i]>>1) & 0x7F7F7F7F;
1053 
-                }
1054 
-                for(i<<=2; i<count; i++){
1055 
-                        c->stdQPTable[i] = QP_store[i]>>1;
1056 
-                }
1057 
-                QP_store= c->stdQPTable;
1058 
-                QPStride= absQPStride;
1047 
+    if(pict_type & PP_PICT_TYPE_QP2){
1048 
+        int i;
1049 
+        const int count= mbHeight * absQPStride;
1050 
+        for(i=0; i<(count>>2); i++){
1051 
+            ((uint32_t*)c->stdQPTable)[i] = (((const uint32_t*)QP_store)[i]>>1) & 0x7F7F7F7F;
1059 1052 
         }
1060 
-
1061 
-if(0){
1062 
-int x,y;
1063 
-for(y=0; y<mbHeight; y++){
1064 
-        for(x=0; x<mbWidth; x++){
1053 
+        for(i<<=2; i<count; i++){
1054 
+            c->stdQPTable[i] = QP_store[i]>>1;
1055 
+        }
1056 
+        QP_store= c->stdQPTable;
1057 
+        QPStride= absQPStride;
1058 
+    }
1059 
+
1060 
+    if(0){
1061 
+        int x,y;
1062 
+        for(y=0; y<mbHeight; y++){
1063 
+            for(x=0; x<mbWidth; x++){
1065 1064 
                 av_log(c, AV_LOG_INFO, "%2d ", QP_store[x + y*QPStride]);
1065 
+            }
1066 
+            av_log(c, AV_LOG_INFO, "\n");
1066 1067 
         }
1067 1068 
         av_log(c, AV_LOG_INFO, "\n");
1068 
-}
1069 
-        av_log(c, AV_LOG_INFO, "\n");
1070 
-}
1071 
-
1072 
-        if((pict_type&7)!=3)
1073 
-        {
1074 
-                if (QPStride >= 0) {
1075 
-                        int i;
1076 
-                        const int count= mbHeight * QPStride;
1077 
-                        for(i=0; i<(count>>2); i++){
1078 
-                                ((uint32_t*)c->nonBQPTable)[i] = ((const uint32_t*)QP_store)[i] & 0x3F3F3F3F;
1079 
-                        }
1080 
-                        for(i<<=2; i<count; i++){
1081 
-                                c->nonBQPTable[i] = QP_store[i] & 0x3F;
1082 
-                        }
1083 
-                } else {
1084 
-                        int i,j;
1085 
-                        for(i=0; i<mbHeight; i++) {
1086 
-                                    for(j=0; j<absQPStride; j++) {
1087 
-                                        c->nonBQPTable[i*absQPStride+j] = QP_store[i*QPStride+j] & 0x3F;
1088 
-                                }
1089 
-                        }
1069 
+    }
1070 
+
1071 
+    if((pict_type&7)!=3){
1072 
+        if (QPStride >= 0){
1073 
+            int i;
1074 
+            const int count= mbHeight * QPStride;
1075 
+            for(i=0; i<(count>>2); i++){
1076 
+                ((uint32_t*)c->nonBQPTable)[i] = ((const uint32_t*)QP_store)[i] & 0x3F3F3F3F;
1077 
+            }
1078 
+            for(i<<=2; i<count; i++){
1079 
+                c->nonBQPTable[i] = QP_store[i] & 0x3F;
1080 
+            }
1081 
+        } else {
1082 
+            int i,j;
1083 
+            for(i=0; i<mbHeight; i++) {
1084 
+                for(j=0; j<absQPStride; j++) {
1085 
+                    c->nonBQPTable[i*absQPStride+j] = QP_store[i*QPStride+j] & 0x3F;
1090 1086 
                 }
1087 
+            }
1091 1088 
         }
1089 
+    }
1092 1090
1093 
-        av_log(c, AV_LOG_DEBUG, "using npp filters 0x%X/0x%X\n",
1094 
-               mode->lumMode, mode->chromMode);
1091 
+    av_log(c, AV_LOG_DEBUG, "using npp filters 0x%X/0x%X\n",
1092 
+           mode->lumMode, mode->chromMode);
1095 1093
1096 
-        postProcess(src[0], srcStride[0], dst[0], dstStride[0],
1094 
+    postProcess(src[0], srcStride[0], dst[0], dstStride[0],
1097 1095 
                 width, height, QP_store, QPStride, 0, mode, c);
1098 1096
1099 
-        width  = (width )>>c->hChromaSubSample;
1100 
-        height = (height)>>c->vChromaSubSample;
1101 
-
1102 
-        if(mode->chromMode)
1103 
-        {
1104 
-                postProcess(src[1], srcStride[1], dst[1], dstStride[1],
1105 
-                        width, height, QP_store, QPStride, 1, mode, c);
1106 
-                postProcess(src[2], srcStride[2], dst[2], dstStride[2],
1107 
-                        width, height, QP_store, QPStride, 2, mode, c);
1108 
-        }
1109 
-        else if(srcStride[1] == dstStride[1] && srcStride[2] == dstStride[2])
1110 
-        {
1111 
-                linecpy(dst[1], src[1], height, srcStride[1]);
1112 
-                linecpy(dst[2], src[2], height, srcStride[2]);
1113 
-        }
1114 
-        else
1115 
-        {
1116 
-                int y;
1117 
-                for(y=0; y<height; y++)
1118 
-                {
1119 
-                        memcpy(&(dst[1][y*dstStride[1]]), &(src[1][y*srcStride[1]]), width);
1120 
-                        memcpy(&(dst[2][y*dstStride[2]]), &(src[2][y*srcStride[2]]), width);
1121 
-                }
1097 
+    width  = (width )>>c->hChromaSubSample;
1098 
+    height = (height)>>c->vChromaSubSample;
1099 
+
1100 
+    if(mode->chromMode){
1101 
+        postProcess(src[1], srcStride[1], dst[1], dstStride[1],
1102 
+                    width, height, QP_store, QPStride, 1, mode, c);
1103 
+        postProcess(src[2], srcStride[2], dst[2], dstStride[2],
1104 
+                    width, height, QP_store, QPStride, 2, mode, c);
1105 
+    }
1106 
+    else if(srcStride[1] == dstStride[1] && srcStride[2] == dstStride[2]){
1107 
+        linecpy(dst[1], src[1], height, srcStride[1]);
1108 
+        linecpy(dst[2], src[2], height, srcStride[2]);
1109 
+    }else{
1110 
+        int y;
1111 
+        for(y=0; y<height; y++){
1112 
+            memcpy(&(dst[1][y*dstStride[1]]), &(src[1][y*srcStride[1]]), width);
1113 
+            memcpy(&(dst[2][y*dstStride[2]]), &(src[2][y*srcStride[2]]), width);
1122 1114 
         }
1115 
+    }
1123 1116 
 }
1124 1117
libpostproc/postprocess.h
History View file @ 16e0bf7
... ... 
@@ -59,10 +59,10 @@ extern const char pp_help[]; ///< a simple help text
59 59 
 #endif
60 60
61 61 
 void  pp_postprocess(const uint8_t * src[3], const int srcStride[3],
62 
-                 uint8_t * dst[3], const int dstStride[3],
63 
-                 int horizontalSize, int verticalSize,
64 
-                 const QP_STORE_T *QP_store,  int QP_stride,
65 
-                 pp_mode_t *mode, pp_context_t *ppContext, int pict_type);
62 
+                     uint8_t * dst[3], const int dstStride[3],
63 
+                     int horizontalSize, int verticalSize,
64 
+                     const QP_STORE_T *QP_store,  int QP_stride,
65 
+                     pp_mode_t *mode, pp_context_t *ppContext, int pict_type);
66 66
67 67
68 68 
 /**
libpostproc/postprocess_altivec_template.c
History View file @ 16e0bf7
... ... 
@@ -23,353 +23,352 @@
23 23 
 #include "avutil.h"
24 24
25 25 
 #define ALTIVEC_TRANSPOSE_8x8_SHORT(src_a,src_b,src_c,src_d,src_e,src_f,src_g,src_h) \
26 
-  do {                                                                  \
27 
-    __typeof__(src_a) tempA1, tempB1, tempC1, tempD1;                   \
28 
-    __typeof__(src_a) tempE1, tempF1, tempG1, tempH1;                   \
29 
-    __typeof__(src_a) tempA2, tempB2, tempC2, tempD2;                   \
30 
-    __typeof__(src_a) tempE2, tempF2, tempG2, tempH2;                   \
31 
-    tempA1 = vec_mergeh (src_a, src_e);                                 \
32 
-    tempB1 = vec_mergel (src_a, src_e);                                 \
33 
-    tempC1 = vec_mergeh (src_b, src_f);                                 \
34 
-    tempD1 = vec_mergel (src_b, src_f);                                 \
35 
-    tempE1 = vec_mergeh (src_c, src_g);                                 \
36 
-    tempF1 = vec_mergel (src_c, src_g);                                 \
37 
-    tempG1 = vec_mergeh (src_d, src_h);                                 \
38 
-    tempH1 = vec_mergel (src_d, src_h);                                 \
39 
-    tempA2 = vec_mergeh (tempA1, tempE1);                               \
40 
-    tempB2 = vec_mergel (tempA1, tempE1);                               \
41 
-    tempC2 = vec_mergeh (tempB1, tempF1);                               \
42 
-    tempD2 = vec_mergel (tempB1, tempF1);                               \
43 
-    tempE2 = vec_mergeh (tempC1, tempG1);                               \
44 
-    tempF2 = vec_mergel (tempC1, tempG1);                               \
45 
-    tempG2 = vec_mergeh (tempD1, tempH1);                               \
46 
-    tempH2 = vec_mergel (tempD1, tempH1);                               \
47 
-    src_a = vec_mergeh (tempA2, tempE2);                                \
48 
-    src_b = vec_mergel (tempA2, tempE2);                                \
49 
-    src_c = vec_mergeh (tempB2, tempF2);                                \
50 
-    src_d = vec_mergel (tempB2, tempF2);                                \
51 
-    src_e = vec_mergeh (tempC2, tempG2);                                \
52 
-    src_f = vec_mergel (tempC2, tempG2);                                \
53 
-    src_g = vec_mergeh (tempD2, tempH2);                                \
54 
-    src_h = vec_mergel (tempD2, tempH2);                                \
55 
-  } while (0)
26 
+    do {                                                          \
27 
+        __typeof__(src_a) tempA1, tempB1, tempC1, tempD1;         \
28 
+        __typeof__(src_a) tempE1, tempF1, tempG1, tempH1;         \
29 
+        __typeof__(src_a) tempA2, tempB2, tempC2, tempD2;         \
30 
+        __typeof__(src_a) tempE2, tempF2, tempG2, tempH2;         \
31 
+        tempA1 = vec_mergeh (src_a, src_e);                       \
32 
+        tempB1 = vec_mergel (src_a, src_e);                       \
33 
+        tempC1 = vec_mergeh (src_b, src_f);                       \
34 
+        tempD1 = vec_mergel (src_b, src_f);                       \
35 
+        tempE1 = vec_mergeh (src_c, src_g);                       \
36 
+        tempF1 = vec_mergel (src_c, src_g);                       \
37 
+        tempG1 = vec_mergeh (src_d, src_h);                       \
38 
+        tempH1 = vec_mergel (src_d, src_h);                       \
39 
+        tempA2 = vec_mergeh (tempA1, tempE1);                     \
40 
+        tempB2 = vec_mergel (tempA1, tempE1);                     \
41 
+        tempC2 = vec_mergeh (tempB1, tempF1);                     \
42 
+        tempD2 = vec_mergel (tempB1, tempF1);                     \
43 
+        tempE2 = vec_mergeh (tempC1, tempG1);                     \
44 
+        tempF2 = vec_mergel (tempC1, tempG1);                     \
45 
+        tempG2 = vec_mergeh (tempD1, tempH1);                     \
46 
+        tempH2 = vec_mergel (tempD1, tempH1);                     \
47 
+        src_a = vec_mergeh (tempA2, tempE2);                      \
48 
+        src_b = vec_mergel (tempA2, tempE2);                      \
49 
+        src_c = vec_mergeh (tempB2, tempF2);                      \
50 
+        src_d = vec_mergel (tempB2, tempF2);                      \
51 
+        src_e = vec_mergeh (tempC2, tempG2);                      \
52 
+        src_f = vec_mergel (tempC2, tempG2);                      \
53 
+        src_g = vec_mergeh (tempD2, tempH2);                      \
54 
+        src_h = vec_mergel (tempD2, tempH2);                      \
55 
+    } while (0)
56 56
57 57
58 58 
 static inline int vertClassify_altivec(uint8_t src[], int stride, PPContext *c) {
59 
-  /*
59 
+    /*
60 60 
     this code makes no assumption on src or stride.
61 61 
     One could remove the recomputation of the perm
62 62 
     vector by assuming (stride % 16) == 0, unfortunately
63 63 
     this is not always true.
64 
-  */
65 
-  DECLARE_ALIGNED(16, short, data[8]);
66 
-  int numEq;
67 
-  uint8_t *src2 = src;
68 
-  vector signed short v_dcOffset;
69 
-  vector signed short v2QP;
70 
-  vector unsigned short v4QP;
71 
-  vector unsigned short v_dcThreshold;
72 
-  const int properStride = (stride % 16);
73 
-  const int srcAlign = ((unsigned long)src2 % 16);
74 
-  const int two_vectors = ((srcAlign > 8) || properStride) ? 1 : 0;
75 
-  const vector signed int zero = vec_splat_s32(0);
76 
-  const vector signed short mask = vec_splat_s16(1);
77 
-  vector signed int v_numEq = vec_splat_s32(0);
78 
-
79 
-  data[0] = ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
80 
-  data[1] = data[0] * 2 + 1;
81 
-  data[2] = c->QP * 2;
82 
-  data[3] = c->QP * 4;
83 
-  vector signed short v_data = vec_ld(0, data);
84 
-  v_dcOffset = vec_splat(v_data, 0);
85 
-  v_dcThreshold = (vector unsigned short)vec_splat(v_data, 1);
86 
-  v2QP = vec_splat(v_data, 2);
87 
-  v4QP = (vector unsigned short)vec_splat(v_data, 3);
88 
-
89 
-  src2 += stride * 4;
90 
-
91 
-  vector signed short v_srcAss0, v_srcAss1, v_srcAss2, v_srcAss3, v_srcAss4, v_srcAss5, v_srcAss6, v_srcAss7;
64 
+    */
65 
+    DECLARE_ALIGNED(16, short, data[8]);
66 
+    int numEq;
67 
+    uint8_t *src2 = src;
68 
+    vector signed short v_dcOffset;
69 
+    vector signed short v2QP;
70 
+    vector unsigned short v4QP;
71 
+    vector unsigned short v_dcThreshold;
72 
+    const int properStride = (stride % 16);
73 
+    const int srcAlign = ((unsigned long)src2 % 16);
74 
+    const int two_vectors = ((srcAlign > 8) || properStride) ? 1 : 0;
75 
+    const vector signed int zero = vec_splat_s32(0);
76 
+    const vector signed short mask = vec_splat_s16(1);
77 
+    vector signed int v_numEq = vec_splat_s32(0);
78 
+
79 
+    data[0] = ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
80 
+    data[1] = data[0] * 2 + 1;
81 
+    data[2] = c->QP * 2;
82 
+    data[3] = c->QP * 4;
83 
+    vector signed short v_data = vec_ld(0, data);
84 
+    v_dcOffset = vec_splat(v_data, 0);
85 
+    v_dcThreshold = (vector unsigned short)vec_splat(v_data, 1);
86 
+    v2QP = vec_splat(v_data, 2);
87 
+    v4QP = (vector unsigned short)vec_splat(v_data, 3);
88 
+
89 
+    src2 += stride * 4;
90 
+
91 
+    vector signed short v_srcAss0, v_srcAss1, v_srcAss2, v_srcAss3, v_srcAss4, v_srcAss5, v_srcAss6, v_srcAss7;
92 92
93 93 
 #define LOAD_LINE(i)                                                    \
94 
-  register int j##i = i * stride;                                       \
95 
-  vector unsigned char perm##i = vec_lvsl(j##i, src2);                  \
96 
-  const vector unsigned char v_srcA1##i = vec_ld(j##i, src2);           \
97 
-  vector unsigned char v_srcA2##i;                                      \
98 
-  if (two_vectors)                                                      \
99 
-    v_srcA2##i = vec_ld(j##i + 16, src2);                               \
100 
-  const vector unsigned char v_srcA##i =                                \
101 
-    vec_perm(v_srcA1##i, v_srcA2##i, perm##i);                          \
102 
-  v_srcAss##i =                                                         \
103 
-    (vector signed short)vec_mergeh((vector signed char)zero,           \
104 
-                                    (vector signed char)v_srcA##i)
94 
+    register int j##i = i * stride;                                     \
95 
+    vector unsigned char perm##i = vec_lvsl(j##i, src2);                \
96 
+    const vector unsigned char v_srcA1##i = vec_ld(j##i, src2);         \
97 
+    vector unsigned char v_srcA2##i;                                    \
98 
+    if (two_vectors)                                                    \
99 
+        v_srcA2##i = vec_ld(j##i + 16, src2);                           \
100 
+    const vector unsigned char v_srcA##i =                              \
101 
+        vec_perm(v_srcA1##i, v_srcA2##i, perm##i);                      \
102 
+    v_srcAss##i =                                                       \
103 
+        (vector signed short)vec_mergeh((vector signed char)zero,       \
104 
+                                        (vector signed char)v_srcA##i)
105 105
106 106 
 #define LOAD_LINE_ALIGNED(i)                                            \
107 
-  register int j##i = i * stride;                                       \
108 
-  const vector unsigned char v_srcA##i = vec_ld(j##i, src2);            \
109 
-  v_srcAss##i =                                                         \
110 
-    (vector signed short)vec_mergeh((vector signed char)zero,           \
111 
-                                    (vector signed char)v_srcA##i)
107 
+    register int j##i = i * stride;                                     \
108 
+    const vector unsigned char v_srcA##i = vec_ld(j##i, src2);          \
109 
+    v_srcAss##i =                                                       \
110 
+        (vector signed short)vec_mergeh((vector signed char)zero,       \
111 
+                                        (vector signed char)v_srcA##i)
112 112
113 113 
     /* Special-casing the aligned case is worthwhile, as all calls from
114 114 
      * the (transposed) horizontable deblocks will be aligned, in addition
115 115 
      * to the naturally aligned vertical deblocks. */
116 116 
     if (properStride && srcAlign) {
117 
-      LOAD_LINE_ALIGNED(0);
118 
-      LOAD_LINE_ALIGNED(1);
119 
-      LOAD_LINE_ALIGNED(2);
120 
-      LOAD_LINE_ALIGNED(3);
121 
-      LOAD_LINE_ALIGNED(4);
122 
-      LOAD_LINE_ALIGNED(5);
123 
-      LOAD_LINE_ALIGNED(6);
124 
-      LOAD_LINE_ALIGNED(7);
117 
+        LOAD_LINE_ALIGNED(0);
118 
+        LOAD_LINE_ALIGNED(1);
119 
+        LOAD_LINE_ALIGNED(2);
120 
+        LOAD_LINE_ALIGNED(3);
121 
+        LOAD_LINE_ALIGNED(4);
122 
+        LOAD_LINE_ALIGNED(5);
123 
+        LOAD_LINE_ALIGNED(6);
124 
+        LOAD_LINE_ALIGNED(7);
125 125 
     } else {
126 
-      LOAD_LINE(0);
127 
-      LOAD_LINE(1);
128 
-      LOAD_LINE(2);
129 
-      LOAD_LINE(3);
130 
-      LOAD_LINE(4);
131 
-      LOAD_LINE(5);
132 
-      LOAD_LINE(6);
133 
-      LOAD_LINE(7);
126 
+        LOAD_LINE(0);
127 
+        LOAD_LINE(1);
128 
+        LOAD_LINE(2);
129 
+        LOAD_LINE(3);
130 
+        LOAD_LINE(4);
131 
+        LOAD_LINE(5);
132 
+        LOAD_LINE(6);
133 
+        LOAD_LINE(7);
134 134 
     }
135 135 
 #undef LOAD_LINE
136 136 
 #undef LOAD_LINE_ALIGNED
137 137
138 138 
 #define ITER(i, j)                                                      \
139 
-  const vector signed short v_diff##i =                                 \
140 
-    vec_sub(v_srcAss##i, v_srcAss##j);                                  \
141 
-  const vector signed short v_sum##i =                                  \
142 
-    vec_add(v_diff##i, v_dcOffset);                                     \
143 
-  const vector signed short v_comp##i =                                 \
144 
-    (vector signed short)vec_cmplt((vector unsigned short)v_sum##i,     \
145 
-                                   v_dcThreshold);                      \
146 
-  const vector signed short v_part##i = vec_and(mask, v_comp##i);       \
147 
-  v_numEq = vec_sum4s(v_part##i, v_numEq);
148 
-
149 
-  ITER(0, 1);
150 
-  ITER(1, 2);
151 
-  ITER(2, 3);
152 
-  ITER(3, 4);
153 
-  ITER(4, 5);
154 
-  ITER(5, 6);
155 
-  ITER(6, 7);
139 
+    const vector signed short v_diff##i =                               \
140 
+        vec_sub(v_srcAss##i, v_srcAss##j);                              \
141 
+    const vector signed short v_sum##i =                                \
142 
+        vec_add(v_diff##i, v_dcOffset);                                 \
143 
+    const vector signed short v_comp##i =                               \
144 
+        (vector signed short)vec_cmplt((vector unsigned short)v_sum##i, \
145 
+                                       v_dcThreshold);                  \
146 
+    const vector signed short v_part##i = vec_and(mask, v_comp##i);     \
147 
+    v_numEq = vec_sum4s(v_part##i, v_numEq);
148 
+
149 
+    ITER(0, 1);
150 
+    ITER(1, 2);
151 
+    ITER(2, 3);
152 
+    ITER(3, 4);
153 
+    ITER(4, 5);
154 
+    ITER(5, 6);
155 
+    ITER(6, 7);
156 156 
 #undef ITER
157 157
158 
-  v_numEq = vec_sums(v_numEq, zero);
159 
-
160 
-  v_numEq = vec_splat(v_numEq, 3);
161 
-  vec_ste(v_numEq, 0, &numEq);
162 
-
163 
-  if (numEq > c->ppMode.flatnessThreshold)
164 
-    {
165 
-      const vector unsigned char mmoP1 = (const vector unsigned char)
166 
-        AVV(0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f,
167 
-            0x00, 0x01, 0x12, 0x13, 0x08, 0x09, 0x1A, 0x1B);
168 
-      const vector unsigned char mmoP2 = (const vector unsigned char)
169 
-        AVV(0x04, 0x05, 0x16, 0x17, 0x0C, 0x0D, 0x1E, 0x1F,
170 
-            0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f);
171 
-      const vector unsigned char mmoP = (const vector unsigned char)
172 
-        vec_lvsl(8, (unsigned char*)0);
173 
-
174 
-      vector signed short mmoL1 = vec_perm(v_srcAss0, v_srcAss2, mmoP1);
175 
-      vector signed short mmoL2 = vec_perm(v_srcAss4, v_srcAss6, mmoP2);
176 
-      vector signed short mmoL = vec_perm(mmoL1, mmoL2, mmoP);
177 
-      vector signed short mmoR1 = vec_perm(v_srcAss5, v_srcAss7, mmoP1);
178 
-      vector signed short mmoR2 = vec_perm(v_srcAss1, v_srcAss3, mmoP2);
179 
-      vector signed short mmoR = vec_perm(mmoR1, mmoR2, mmoP);
180 
-      vector signed short mmoDiff = vec_sub(mmoL, mmoR);
181 
-      vector unsigned short mmoSum = (vector unsigned short)vec_add(mmoDiff, v2QP);
182 
-
183 
-      if (vec_any_gt(mmoSum, v4QP))
184 
-        return 0;
185 
-      else
186 
-        return 1;
158 
+    v_numEq = vec_sums(v_numEq, zero);
159 
+
160 
+    v_numEq = vec_splat(v_numEq, 3);
161 
+    vec_ste(v_numEq, 0, &numEq);
162 
+
163 
+    if (numEq > c->ppMode.flatnessThreshold){
164 
+        const vector unsigned char mmoP1 = (const vector unsigned char)
165 
+            AVV(0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f,
166 
+                0x00, 0x01, 0x12, 0x13, 0x08, 0x09, 0x1A, 0x1B);
167 
+        const vector unsigned char mmoP2 = (const vector unsigned char)
168 
+            AVV(0x04, 0x05, 0x16, 0x17, 0x0C, 0x0D, 0x1E, 0x1F,
169 
+                0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f, 0x1f);
170 
+        const vector unsigned char mmoP = (const vector unsigned char)
171 
+            vec_lvsl(8, (unsigned char*)0);
172 
+
173 
+        vector signed short mmoL1 = vec_perm(v_srcAss0, v_srcAss2, mmoP1);
174 
+        vector signed short mmoL2 = vec_perm(v_srcAss4, v_srcAss6, mmoP2);
175 
+        vector signed short mmoL = vec_perm(mmoL1, mmoL2, mmoP);
176 
+        vector signed short mmoR1 = vec_perm(v_srcAss5, v_srcAss7, mmoP1);
177 
+        vector signed short mmoR2 = vec_perm(v_srcAss1, v_srcAss3, mmoP2);
178 
+        vector signed short mmoR = vec_perm(mmoR1, mmoR2, mmoP);
179 
+        vector signed short mmoDiff = vec_sub(mmoL, mmoR);
180 
+        vector unsigned short mmoSum = (vector unsigned short)vec_add(mmoDiff, v2QP);
181 
+
182 
+        if (vec_any_gt(mmoSum, v4QP))
183 
+            return 0;
184 
+        else
185 
+            return 1;
187 186 
     }
188 
-  else return 2;
187 
+    else return 2;
189 188 
 }
190 189
191 190 
 static inline void doVertLowPass_altivec(uint8_t *src, int stride, PPContext *c) {
192 
-  /*
191 
+    /*
193 192 
     this code makes no assumption on src or stride.
194 193 
     One could remove the recomputation of the perm
195 194 
     vector by assuming (stride % 16) == 0, unfortunately
196 195 
     this is not always true. Quite a lot of load/stores
197 196 
     can be removed by assuming proper alignment of
198 197 
     src & stride :-(
199 
-  */
200 
-  uint8_t *src2 = src;
201 
-  const vector signed int zero = vec_splat_s32(0);
202 
-  const int properStride = (stride % 16);
203 
-  const int srcAlign = ((unsigned long)src2 % 16);
204 
-  DECLARE_ALIGNED(16, short, qp[8]);
205 
-  qp[0] = c->QP;
206 
-  vector signed short vqp = vec_ld(0, qp);
207 
-  vqp = vec_splat(vqp, 0);
208 
-
209 
-  src2 += stride*3;
210 
-
211 
-  vector signed short vb0, vb1, vb2, vb3, vb4, vb5, vb6, vb7, vb8, vb9;
212 
-  vector unsigned char vbA0, vbA1, vbA2, vbA3, vbA4, vbA5, vbA6, vbA7, vbA8, vbA9;
213 
-  vector unsigned char vbB0, vbB1, vbB2, vbB3, vbB4, vbB5, vbB6, vbB7, vbB8, vbB9;
214 
-  vector unsigned char vbT0, vbT1, vbT2, vbT3, vbT4, vbT5, vbT6, vbT7, vbT8, vbT9;
198 
+    */
199 
+    uint8_t *src2 = src;
200 
+    const vector signed int zero = vec_splat_s32(0);
201 
+    const int properStride = (stride % 16);
202 
+    const int srcAlign = ((unsigned long)src2 % 16);
203 
+    DECLARE_ALIGNED(16, short, qp[8]);
204 
+    qp[0] = c->QP;
205 
+    vector signed short vqp = vec_ld(0, qp);
206 
+    vqp = vec_splat(vqp, 0);
207 
+
208 
+    src2 += stride*3;
209 
+
210 
+    vector signed short vb0, vb1, vb2, vb3, vb4, vb5, vb6, vb7, vb8, vb9;
211 
+    vector unsigned char vbA0, vbA1, vbA2, vbA3, vbA4, vbA5, vbA6, vbA7, vbA8, vbA9;
212 
+    vector unsigned char vbB0, vbB1, vbB2, vbB3, vbB4, vbB5, vbB6, vbB7, vbB8, vbB9;
213 
+    vector unsigned char vbT0, vbT1, vbT2, vbT3, vbT4, vbT5, vbT6, vbT7, vbT8, vbT9;
215 214
216 215 
 #define LOAD_LINE(i)                                                    \
217 
-  const vector unsigned char perml##i =                                 \
218 
-    vec_lvsl(i * stride, src2);                                         \
219 
-  vbA##i = vec_ld(i * stride, src2);                                    \
220 
-  vbB##i = vec_ld(i * stride + 16, src2);                               \
221 
-  vbT##i = vec_perm(vbA##i, vbB##i, perml##i);                          \
222 
-  vb##i =                                                               \
223 
-    (vector signed short)vec_mergeh((vector unsigned char)zero,         \
224 
-                                    (vector unsigned char)vbT##i)
216 
+    const vector unsigned char perml##i =                               \
217 
+        vec_lvsl(i * stride, src2);                                     \
218 
+    vbA##i = vec_ld(i * stride, src2);                                  \
219 
+    vbB##i = vec_ld(i * stride + 16, src2);                             \
220 
+    vbT##i = vec_perm(vbA##i, vbB##i, perml##i);                        \
221 
+    vb##i =                                                             \
222 
+        (vector signed short)vec_mergeh((vector unsigned char)zero,     \
223 
+                                        (vector unsigned char)vbT##i)
225 224
226 225 
 #define LOAD_LINE_ALIGNED(i)                                            \
227 
-  register int j##i = i * stride;                                       \
228 
-  vbT##i = vec_ld(j##i, src2);                                          \
229 
-  vb##i =                                                               \
230 
-    (vector signed short)vec_mergeh((vector signed char)zero,           \
231 
-                                    (vector signed char)vbT##i)
232 
-
233 
-    /* Special-casing the aligned case is worthwhile, as all calls from
234 
-     * the (transposed) horizontable deblocks will be aligned, in addition
235 
-     * to the naturally aligned vertical deblocks. */
236 
-    if (properStride && srcAlign) {
237 
-      LOAD_LINE_ALIGNED(0);
238 
-      LOAD_LINE_ALIGNED(1);
239 
-      LOAD_LINE_ALIGNED(2);
240 
-      LOAD_LINE_ALIGNED(3);
241 
-      LOAD_LINE_ALIGNED(4);
242 
-      LOAD_LINE_ALIGNED(5);
243 
-      LOAD_LINE_ALIGNED(6);
244 
-      LOAD_LINE_ALIGNED(7);
245 
-      LOAD_LINE_ALIGNED(8);
246 
-      LOAD_LINE_ALIGNED(9);
247 
-    } else {
248 
-      LOAD_LINE(0);
249 
-      LOAD_LINE(1);
250 
-      LOAD_LINE(2);
251 
-      LOAD_LINE(3);
252 
-      LOAD_LINE(4);
253 
-      LOAD_LINE(5);
254 
-      LOAD_LINE(6);
255 
-      LOAD_LINE(7);
256 
-      LOAD_LINE(8);
257 
-      LOAD_LINE(9);
258 
-    }
226 
+    register int j##i = i * stride;                                     \
227 
+    vbT##i = vec_ld(j##i, src2);                                        \
228 
+    vb##i =                                                             \
229 
+        (vector signed short)vec_mergeh((vector signed char)zero,       \
230 
+                                        (vector signed char)vbT##i)
231 
+
232 
+      /* Special-casing the aligned case is worthwhile, as all calls from
233 
+       * the (transposed) horizontable deblocks will be aligned, in addition
234 
+       * to the naturally aligned vertical deblocks. */
235 
+      if (properStride && srcAlign) {
236 
+          LOAD_LINE_ALIGNED(0);
237 
+          LOAD_LINE_ALIGNED(1);
238 
+          LOAD_LINE_ALIGNED(2);
239 
+          LOAD_LINE_ALIGNED(3);
240 
+          LOAD_LINE_ALIGNED(4);
241 
+          LOAD_LINE_ALIGNED(5);
242 
+          LOAD_LINE_ALIGNED(6);
243 
+          LOAD_LINE_ALIGNED(7);
244 
+          LOAD_LINE_ALIGNED(8);
245 
+          LOAD_LINE_ALIGNED(9);
246 
+      } else {
247 
+          LOAD_LINE(0);
248 
+          LOAD_LINE(1);
249 
+          LOAD_LINE(2);
250 
+          LOAD_LINE(3);
251 
+          LOAD_LINE(4);
252 
+          LOAD_LINE(5);
253 
+          LOAD_LINE(6);
254 
+          LOAD_LINE(7);
255 
+          LOAD_LINE(8);
256 
+          LOAD_LINE(9);
257 
+      }
259 258 
 #undef LOAD_LINE
260 259 
 #undef LOAD_LINE_ALIGNED
261 260
262 
-  const vector unsigned short v_2 = vec_splat_u16(2);
263 
-  const vector unsigned short v_4 = vec_splat_u16(4);
261 
+    const vector unsigned short v_2 = vec_splat_u16(2);
262 
+    const vector unsigned short v_4 = vec_splat_u16(4);
264 263
265 
-  const vector signed short v_diff01 = vec_sub(vb0, vb1);
266 
-  const vector unsigned short v_cmp01 =
267 
-    (const vector unsigned short) vec_cmplt(vec_abs(v_diff01), vqp);
268 
-  const vector signed short v_first = vec_sel(vb1, vb0, v_cmp01);
269 
-  const vector signed short v_diff89 = vec_sub(vb8, vb9);
270 
-  const vector unsigned short v_cmp89 =
271 
-    (const vector unsigned short) vec_cmplt(vec_abs(v_diff89), vqp);
272 
-  const vector signed short v_last = vec_sel(vb8, vb9, v_cmp89);
264 
+    const vector signed short v_diff01 = vec_sub(vb0, vb1);
265 
+    const vector unsigned short v_cmp01 =
266 
+        (const vector unsigned short) vec_cmplt(vec_abs(v_diff01), vqp);
267 
+    const vector signed short v_first = vec_sel(vb1, vb0, v_cmp01);
268 
+    const vector signed short v_diff89 = vec_sub(vb8, vb9);
269 
+    const vector unsigned short v_cmp89 =
270 
+        (const vector unsigned short) vec_cmplt(vec_abs(v_diff89), vqp);
271 
+    const vector signed short v_last = vec_sel(vb8, vb9, v_cmp89);
273 272
274 
-  const vector signed short temp01 = vec_mladd(v_first, (vector signed short)v_4, vb1);
275 
-  const vector signed short temp02 = vec_add(vb2, vb3);
276 
-  const vector signed short temp03 = vec_add(temp01, (vector signed short)v_4);
277 
-  const vector signed short v_sumsB0 = vec_add(temp02, temp03);
273 
+    const vector signed short temp01 = vec_mladd(v_first, (vector signed short)v_4, vb1);
274 
+    const vector signed short temp02 = vec_add(vb2, vb3);
275 
+    const vector signed short temp03 = vec_add(temp01, (vector signed short)v_4);
276 
+    const vector signed short v_sumsB0 = vec_add(temp02, temp03);
278 277
279 
-  const vector signed short temp11 = vec_sub(v_sumsB0, v_first);
280 
-  const vector signed short v_sumsB1 = vec_add(temp11, vb4);
278 
+    const vector signed short temp11 = vec_sub(v_sumsB0, v_first);
279 
+    const vector signed short v_sumsB1 = vec_add(temp11, vb4);
281 280
282 
-  const vector signed short temp21 = vec_sub(v_sumsB1, v_first);
283 
-  const vector signed short v_sumsB2 = vec_add(temp21, vb5);
281 
+    const vector signed short temp21 = vec_sub(v_sumsB1, v_first);
282 
+    const vector signed short v_sumsB2 = vec_add(temp21, vb5);
284 283
285 
-  const vector signed short temp31 = vec_sub(v_sumsB2, v_first);
286 
-  const vector signed short v_sumsB3 = vec_add(temp31, vb6);
284 
+    const vector signed short temp31 = vec_sub(v_sumsB2, v_first);
285 
+    const vector signed short v_sumsB3 = vec_add(temp31, vb6);
287 286
288 
-  const vector signed short temp41 = vec_sub(v_sumsB3, v_first);
289 
-  const vector signed short v_sumsB4 = vec_add(temp41, vb7);
287 
+    const vector signed short temp41 = vec_sub(v_sumsB3, v_first);
288 
+    const vector signed short v_sumsB4 = vec_add(temp41, vb7);
290 289
291 
-  const vector signed short temp51 = vec_sub(v_sumsB4, vb1);
292 
-  const vector signed short v_sumsB5 = vec_add(temp51, vb8);
290 
+    const vector signed short temp51 = vec_sub(v_sumsB4, vb1);
291 
+    const vector signed short v_sumsB5 = vec_add(temp51, vb8);
293 292
294 
-  const vector signed short temp61 = vec_sub(v_sumsB5, vb2);
295 
-  const vector signed short v_sumsB6 = vec_add(temp61, v_last);
293 
+    const vector signed short temp61 = vec_sub(v_sumsB5, vb2);
294 
+    const vector signed short v_sumsB6 = vec_add(temp61, v_last);
296 295
297 
-  const vector signed short temp71 = vec_sub(v_sumsB6, vb3);
298 
-  const vector signed short v_sumsB7 = vec_add(temp71, v_last);
296 
+    const vector signed short temp71 = vec_sub(v_sumsB6, vb3);
297 
+    const vector signed short v_sumsB7 = vec_add(temp71, v_last);
299 298
300 
-  const vector signed short temp81 = vec_sub(v_sumsB7, vb4);
301 
-  const vector signed short v_sumsB8 = vec_add(temp81, v_last);
299 
+    const vector signed short temp81 = vec_sub(v_sumsB7, vb4);
300 
+    const vector signed short v_sumsB8 = vec_add(temp81, v_last);
302 301
303 
-  const vector signed short temp91 = vec_sub(v_sumsB8, vb5);
304 
-  const vector signed short v_sumsB9 = vec_add(temp91, v_last);
302 
+    const vector signed short temp91 = vec_sub(v_sumsB8, vb5);
303 
+    const vector signed short v_sumsB9 = vec_add(temp91, v_last);
305 304
306 305 
 #define COMPUTE_VR(i, j, k)                                             \
307 
-  const vector signed short temps1##i =                                 \
308 
-    vec_add(v_sumsB##i, v_sumsB##k);                                    \
309 
-  const vector signed short temps2##i =                                 \
310 
-    vec_mladd(vb##j, (vector signed short)v_2, temps1##i);              \
311 
-  const vector signed short  vr##j = vec_sra(temps2##i, v_4)
312 
-
313 
-  COMPUTE_VR(0, 1, 2);
314 
-  COMPUTE_VR(1, 2, 3);
315 
-  COMPUTE_VR(2, 3, 4);
316 
-  COMPUTE_VR(3, 4, 5);
317 
-  COMPUTE_VR(4, 5, 6);
318 
-  COMPUTE_VR(5, 6, 7);
319 
-  COMPUTE_VR(6, 7, 8);
320 
-  COMPUTE_VR(7, 8, 9);
321 
-
322 
-  const vector signed char neg1 = vec_splat_s8(-1);
323 
-  const vector unsigned char permHH = (const vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
324 
-                                                                      0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
306 
+    const vector signed short temps1##i =                               \
307 
+        vec_add(v_sumsB##i, v_sumsB##k);                                \
308 
+    const vector signed short temps2##i =                               \
309 
+        vec_mladd(vb##j, (vector signed short)v_2, temps1##i);          \
310 
+    const vector signed short  vr##j = vec_sra(temps2##i, v_4)
311 
+
312 
+    COMPUTE_VR(0, 1, 2);
313 
+    COMPUTE_VR(1, 2, 3);
314 
+    COMPUTE_VR(2, 3, 4);
315 
+    COMPUTE_VR(3, 4, 5);
316 
+    COMPUTE_VR(4, 5, 6);
317 
+    COMPUTE_VR(5, 6, 7);
318 
+    COMPUTE_VR(6, 7, 8);
319 
+    COMPUTE_VR(7, 8, 9);
320 
+
321 
+    const vector signed char neg1 = vec_splat_s8(-1);
322 
+    const vector unsigned char permHH = (const vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
323 
+                                                                        0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
325 324
326 325 
 #define PACK_AND_STORE(i)                                       \
327 
-  const vector unsigned char perms##i =                         \
328 
-    vec_lvsr(i * stride, src2);                                 \
329 
-  const vector unsigned char vf##i =                            \
330 
-    vec_packsu(vr##i, (vector signed short)zero);               \
331 
-  const vector unsigned char vg##i =                            \
332 
-    vec_perm(vf##i, vbT##i, permHH);                            \
333 
-  const vector unsigned char mask##i =                          \
334 
-    vec_perm((vector unsigned char)zero, (vector unsigned char)neg1, perms##i); \
335 
-  const vector unsigned char vg2##i =                           \
336 
-    vec_perm(vg##i, vg##i, perms##i);                           \
337 
-  const vector unsigned char svA##i =                           \
338 
-    vec_sel(vbA##i, vg2##i, mask##i);                           \
339 
-  const vector unsigned char svB##i =                           \
340 
-    vec_sel(vg2##i, vbB##i, mask##i);                           \
341 
-  vec_st(svA##i, i * stride, src2);                             \
342 
-  vec_st(svB##i, i * stride + 16, src2)
326 
+    const vector unsigned char perms##i =                       \
327 
+        vec_lvsr(i * stride, src2);                             \
328 
+    const vector unsigned char vf##i =                          \
329 
+        vec_packsu(vr##i, (vector signed short)zero);           \
330 
+    const vector unsigned char vg##i =                          \
331 
+        vec_perm(vf##i, vbT##i, permHH);                        \
332 
+    const vector unsigned char mask##i =                        \
333 
+        vec_perm((vector unsigned char)zero, (vector unsigned char)neg1, perms##i); \
334 
+    const vector unsigned char vg2##i =                         \
335 
+        vec_perm(vg##i, vg##i, perms##i);                       \
336 
+    const vector unsigned char svA##i =                         \
337 
+        vec_sel(vbA##i, vg2##i, mask##i);                       \
338 
+    const vector unsigned char svB##i =                         \
339 
+        vec_sel(vg2##i, vbB##i, mask##i);                       \
340 
+    vec_st(svA##i, i * stride, src2);                           \
341 
+    vec_st(svB##i, i * stride + 16, src2)
343 342
344 343 
 #define PACK_AND_STORE_ALIGNED(i)                               \
345 
-  const vector unsigned char vf##i =                            \
346 
-    vec_packsu(vr##i, (vector signed short)zero);               \
347 
-  const vector unsigned char vg##i =                            \
348 
-    vec_perm(vf##i, vbT##i, permHH);                            \
349 
-  vec_st(vg##i, i * stride, src2)
350 
-
351 
-  /* Special-casing the aligned case is worthwhile, as all calls from
352 
-   * the (transposed) horizontable deblocks will be aligned, in addition
353 
-   * to the naturally aligned vertical deblocks. */
354 
-  if (properStride && srcAlign) {
355 
-    PACK_AND_STORE_ALIGNED(1);
356 
-    PACK_AND_STORE_ALIGNED(2);
357 
-    PACK_AND_STORE_ALIGNED(3);
358 
-    PACK_AND_STORE_ALIGNED(4);
359 
-    PACK_AND_STORE_ALIGNED(5);
360 
-    PACK_AND_STORE_ALIGNED(6);
361 
-    PACK_AND_STORE_ALIGNED(7);
362 
-    PACK_AND_STORE_ALIGNED(8);
363 
-  } else {
364 
-    PACK_AND_STORE(1);
365 
-    PACK_AND_STORE(2);
366 
-    PACK_AND_STORE(3);
367 
-    PACK_AND_STORE(4);
368 
-    PACK_AND_STORE(5);
369 
-    PACK_AND_STORE(6);
370 
-    PACK_AND_STORE(7);
371 
-    PACK_AND_STORE(8);
372 
-  }
344 
+    const vector unsigned char vf##i =                          \
345 
+        vec_packsu(vr##i, (vector signed short)zero);           \
346 
+    const vector unsigned char vg##i =                          \
347 
+        vec_perm(vf##i, vbT##i, permHH);                        \
348 
+    vec_st(vg##i, i * stride, src2)
349 
+
350 
+    /* Special-casing the aligned case is worthwhile, as all calls from
351 
+     * the (transposed) horizontable deblocks will be aligned, in addition
352 
+     * to the naturally aligned vertical deblocks. */
353 
+    if (properStride && srcAlign) {
354 
+        PACK_AND_STORE_ALIGNED(1);
355 
+        PACK_AND_STORE_ALIGNED(2);
356 
+        PACK_AND_STORE_ALIGNED(3);
357 
+        PACK_AND_STORE_ALIGNED(4);
358 
+        PACK_AND_STORE_ALIGNED(5);
359 
+        PACK_AND_STORE_ALIGNED(6);
360 
+        PACK_AND_STORE_ALIGNED(7);
361 
+        PACK_AND_STORE_ALIGNED(8);
362 
+    } else {
363 
+        PACK_AND_STORE(1);
364 
+        PACK_AND_STORE(2);
365 
+        PACK_AND_STORE(3);
366 
+        PACK_AND_STORE(4);
367 
+        PACK_AND_STORE(5);
368 
+        PACK_AND_STORE(6);
369 
+        PACK_AND_STORE(7);
370 
+        PACK_AND_STORE(8);
371 
+    }
373 372 
 #undef PACK_AND_STORE
374 373 
 #undef PACK_AND_STORE_ALIGNED
375 374 
 }
... ... 
@@ -377,190 +376,190 @@ static inline void doVertLowPass_altivec(uint8_t *src, int stride, PPContext *c)
377 377
378 378
379 379 
 static inline void doVertDefFilter_altivec(uint8_t src[], int stride, PPContext *c) {
380 
-  /*
380 
+    /*
381 381 
     this code makes no assumption on src or stride.
382 382 
     One could remove the recomputation of the perm
383 383 
     vector by assuming (stride % 16) == 0, unfortunately
384 384 
     this is not always true. Quite a lot of load/stores
385 385 
     can be removed by assuming proper alignment of
386 386 
     src & stride :-(
387 
-  */
388 
-  uint8_t *src2 = src;
389 
-  const vector signed int zero = vec_splat_s32(0);
390 
-  DECLARE_ALIGNED(16, short, qp[8]);
391 
-  qp[0] = 8*c->QP;
392 
-  vector signed short vqp = vec_ld(0, qp);
393 
-  vqp = vec_splat(vqp, 0);
387 
+    */
388 
+    uint8_t *src2 = src;
389 
+    const vector signed int zero = vec_splat_s32(0);
390 
+    DECLARE_ALIGNED(16, short, qp[8]);
391 
+    qp[0] = 8*c->QP;
392 
+    vector signed short vqp = vec_ld(0, qp);
393 
+    vqp = vec_splat(vqp, 0);
394 394
395 395 
 #define LOAD_LINE(i)                                                    \
396 
-  const vector unsigned char perm##i =                                  \
397 
-    vec_lvsl(i * stride, src2);                                         \
398 
-  const vector unsigned char vbA##i =                                   \
399 
-    vec_ld(i * stride, src2);                                           \
400 
-  const vector unsigned char vbB##i =                                   \
401 
-    vec_ld(i * stride + 16, src2);                                      \
402 
-  const vector unsigned char vbT##i =                                   \
403 
-    vec_perm(vbA##i, vbB##i, perm##i);                                  \
404 
-  const vector signed short vb##i =                                     \
405 
-    (vector signed short)vec_mergeh((vector unsigned char)zero,         \
406 
-                                    (vector unsigned char)vbT##i)
407 
-
408 
-  src2 += stride*3;
409 
-
410 
-  LOAD_LINE(1);
411 
-  LOAD_LINE(2);
412 
-  LOAD_LINE(3);
413 
-  LOAD_LINE(4);
414 
-  LOAD_LINE(5);
415 
-  LOAD_LINE(6);
416 
-  LOAD_LINE(7);
417 
-  LOAD_LINE(8);
396 
+    const vector unsigned char perm##i =                                \
397 
+        vec_lvsl(i * stride, src2);                                     \
398 
+    const vector unsigned char vbA##i =                                 \
399 
+        vec_ld(i * stride, src2);                                       \
400 
+    const vector unsigned char vbB##i =                                 \
401 
+        vec_ld(i * stride + 16, src2);                                  \
402 
+    const vector unsigned char vbT##i =                                 \
403 
+        vec_perm(vbA##i, vbB##i, perm##i);                              \
404 
+    const vector signed short vb##i =                                   \
405 
+        (vector signed short)vec_mergeh((vector unsigned char)zero,     \
406 
+                                        (vector unsigned char)vbT##i)
407 
+
408 
+    src2 += stride*3;
409 
+
410 
+     LOAD_LINE(1);
411 
+     LOAD_LINE(2);
412 
+     LOAD_LINE(3);
413 
+     LOAD_LINE(4);
414 
+     LOAD_LINE(5);
415 
+     LOAD_LINE(6);
416 
+     LOAD_LINE(7);
417 
+     LOAD_LINE(8);
418 418 
 #undef LOAD_LINE
419 419
420 
-  const vector signed short v_1 = vec_splat_s16(1);
421 
-  const vector signed short v_2 = vec_splat_s16(2);
422 
-  const vector signed short v_5 = vec_splat_s16(5);
423 
-  const vector signed short v_32 = vec_sl(v_1,
424 
-                                          (vector unsigned short)v_5);
425 
-  /* middle energy */
426 
-  const vector signed short l3minusl6 = vec_sub(vb3, vb6);
427 
-  const vector signed short l5minusl4 = vec_sub(vb5, vb4);
428 
-  const vector signed short twotimes_l3minusl6 = vec_mladd(v_2, l3minusl6, (vector signed short)zero);
429 
-  const vector signed short mE = vec_mladd(v_5, l5minusl4, twotimes_l3minusl6);
430 
-  const vector signed short absmE = vec_abs(mE);
431 
-  /* left & right energy */
432 
-  const vector signed short l1minusl4 = vec_sub(vb1, vb4);
433 
-  const vector signed short l3minusl2 = vec_sub(vb3, vb2);
434 
-  const vector signed short l5minusl8 = vec_sub(vb5, vb8);
435 
-  const vector signed short l7minusl6 = vec_sub(vb7, vb6);
436 
-  const vector signed short twotimes_l1minusl4 = vec_mladd(v_2, l1minusl4, (vector signed short)zero);
437 
-  const vector signed short twotimes_l5minusl8 = vec_mladd(v_2, l5minusl8, (vector signed short)zero);
438 
-  const vector signed short lE = vec_mladd(v_5, l3minusl2, twotimes_l1minusl4);
439 
-  const vector signed short rE = vec_mladd(v_5, l7minusl6, twotimes_l5minusl8);
440 
-  /* d */
441 
-  const vector signed short ddiff = vec_sub(absmE,
442 
-                                            vec_min(vec_abs(lE),
443 
-                                                    vec_abs(rE)));
444 
-  const vector signed short ddiffclamp = vec_max(ddiff, (vector signed short)zero);
445 
-  const vector signed short dtimes64 = vec_mladd(v_5, ddiffclamp, v_32);
446 
-  const vector signed short d = vec_sra(dtimes64, vec_splat_u16(6));
447 
-  const vector signed short minusd = vec_sub((vector signed short)zero, d);
448 
-  const vector signed short finald = vec_sel(minusd,
449 
-                                             d,
450 
-                                             vec_cmpgt(vec_sub((vector signed short)zero, mE),
451 
-                                                       (vector signed short)zero));
452 
-  /* q */
453 
-  const vector signed short qtimes2 = vec_sub(vb4, vb5);
454 
-  /* for a shift right to behave like /2, we need to add one
455 
-     to all negative integer */
456 
-  const vector signed short rounddown = vec_sel((vector signed short)zero,
457 
-                                                v_1,
458 
-                                                vec_cmplt(qtimes2, (vector signed short)zero));
459 
-  const vector signed short q = vec_sra(vec_add(qtimes2, rounddown), vec_splat_u16(1));
460 
-  /* clamp */
461 
-  const vector signed short dclamp_P1 = vec_max((vector signed short)zero, finald);
462 
-  const vector signed short dclamp_P = vec_min(dclamp_P1, q);
463 
-  const vector signed short dclamp_N1 = vec_min((vector signed short)zero, finald);
464 
-  const vector signed short dclamp_N = vec_max(dclamp_N1, q);
465 
-
466 
-  const vector signed short dclampedfinal = vec_sel(dclamp_N,
467 
-                                                    dclamp_P,
468 
-                                                    vec_cmpgt(q, (vector signed short)zero));
469 
-  const vector signed short dornotd = vec_sel((vector signed short)zero,
470 
-                                              dclampedfinal,
471 
-                                              vec_cmplt(absmE, vqp));
472 
-  /* add/subtract to l4 and l5 */
473 
-  const vector signed short vb4minusd = vec_sub(vb4, dornotd);
474 
-  const vector signed short vb5plusd = vec_add(vb5, dornotd);
475 
-  /* finally, stores */
476 
-  const vector unsigned char st4 = vec_packsu(vb4minusd, (vector signed short)zero);
477 
-  const vector unsigned char st5 = vec_packsu(vb5plusd, (vector signed short)zero);
478 
-
479 
-  const vector signed char neg1 = vec_splat_s8(-1);
480 
-  const vector unsigned char permHH = (const vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
481 
-                                                                      0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
420 
+     const vector signed short v_1 = vec_splat_s16(1);
421 
+     const vector signed short v_2 = vec_splat_s16(2);
422 
+     const vector signed short v_5 = vec_splat_s16(5);
423 
+     const vector signed short v_32 = vec_sl(v_1,
424 
+                                             (vector unsigned short)v_5);
425 
+     /* middle energy */
426 
+     const vector signed short l3minusl6 = vec_sub(vb3, vb6);
427 
+     const vector signed short l5minusl4 = vec_sub(vb5, vb4);
428 
+     const vector signed short twotimes_l3minusl6 = vec_mladd(v_2, l3minusl6, (vector signed short)zero);
429 
+     const vector signed short mE = vec_mladd(v_5, l5minusl4, twotimes_l3minusl6);
430 
+     const vector signed short absmE = vec_abs(mE);
431 
+     /* left & right energy */
432 
+     const vector signed short l1minusl4 = vec_sub(vb1, vb4);
433 
+     const vector signed short l3minusl2 = vec_sub(vb3, vb2);
434 
+     const vector signed short l5minusl8 = vec_sub(vb5, vb8);
435 
+     const vector signed short l7minusl6 = vec_sub(vb7, vb6);
436 
+     const vector signed short twotimes_l1minusl4 = vec_mladd(v_2, l1minusl4, (vector signed short)zero);
437 
+     const vector signed short twotimes_l5minusl8 = vec_mladd(v_2, l5minusl8, (vector signed short)zero);
438 
+     const vector signed short lE = vec_mladd(v_5, l3minusl2, twotimes_l1minusl4);
439 
+     const vector signed short rE = vec_mladd(v_5, l7minusl6, twotimes_l5minusl8);
440 
+     /* d */
441 
+     const vector signed short ddiff = vec_sub(absmE,
442 
+                                               vec_min(vec_abs(lE),
443 
+                                                       vec_abs(rE)));
444 
+     const vector signed short ddiffclamp = vec_max(ddiff, (vector signed short)zero);
445 
+     const vector signed short dtimes64 = vec_mladd(v_5, ddiffclamp, v_32);
446 
+     const vector signed short d = vec_sra(dtimes64, vec_splat_u16(6));
447 
+     const vector signed short minusd = vec_sub((vector signed short)zero, d);
448 
+     const vector signed short finald = vec_sel(minusd,
449 
+                                                d,
450 
+                                                vec_cmpgt(vec_sub((vector signed short)zero, mE),
451 
+                                                          (vector signed short)zero));
452 
+     /* q */
453 
+     const vector signed short qtimes2 = vec_sub(vb4, vb5);
454 
+     /* for a shift right to behave like /2, we need to add one
455 
+        to all negative integer */
456 
+     const vector signed short rounddown = vec_sel((vector signed short)zero,
457 
+                                                   v_1,
458 
+                                                   vec_cmplt(qtimes2, (vector signed short)zero));
459 
+     const vector signed short q = vec_sra(vec_add(qtimes2, rounddown), vec_splat_u16(1));
460 
+     /* clamp */
461 
+     const vector signed short dclamp_P1 = vec_max((vector signed short)zero, finald);
462 
+     const vector signed short dclamp_P = vec_min(dclamp_P1, q);
463 
+     const vector signed short dclamp_N1 = vec_min((vector signed short)zero, finald);
464 
+     const vector signed short dclamp_N = vec_max(dclamp_N1, q);
465 
+
466 
+     const vector signed short dclampedfinal = vec_sel(dclamp_N,
467 
+                                                       dclamp_P,
468 
+                                                       vec_cmpgt(q, (vector signed short)zero));
469 
+     const vector signed short dornotd = vec_sel((vector signed short)zero,
470 
+                                                 dclampedfinal,
471 
+                                                 vec_cmplt(absmE, vqp));
472 
+     /* add/subtract to l4 and l5 */
473 
+     const vector signed short vb4minusd = vec_sub(vb4, dornotd);
474 
+     const vector signed short vb5plusd  = vec_add(vb5, dornotd);
475 
+     /* finally, stores */
476 
+     const vector unsigned char st4 = vec_packsu(vb4minusd, (vector signed short)zero);
477 
+     const vector unsigned char st5 = vec_packsu(vb5plusd,  (vector signed short)zero);
478 
+
479 
+     const vector signed char neg1 = vec_splat_s8(-1);
480 
+     const vector unsigned char permHH = (const vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
481 
+                                                                         0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
482 482
483 483 
 #define STORE(i)                                                \
484 
-  const vector unsigned char perms##i =                         \
485 
-    vec_lvsr(i * stride, src2);                                 \
486 
-  const vector unsigned char vg##i =                            \
487 
-    vec_perm(st##i, vbT##i, permHH);                            \
488 
-  const vector unsigned char mask##i =                          \
489 
-    vec_perm((vector unsigned char)zero, (vector unsigned char)neg1, perms##i); \
490 
-  const vector unsigned char vg2##i =                           \
491 
-    vec_perm(vg##i, vg##i, perms##i);                           \
492 
-  const vector unsigned char svA##i =                           \
493 
-    vec_sel(vbA##i, vg2##i, mask##i);                           \
494 
-  const vector unsigned char svB##i =                           \
495 
-    vec_sel(vg2##i, vbB##i, mask##i);                           \
496 
-  vec_st(svA##i, i * stride, src2);                             \
497 
-  vec_st(svB##i, i * stride + 16, src2)
498 
-
499 
-  STORE(4);
500 
-  STORE(5);
484 
+     const vector unsigned char perms##i =                      \
485 
+         vec_lvsr(i * stride, src2);                            \
486 
+     const vector unsigned char vg##i =                         \
487 
+         vec_perm(st##i, vbT##i, permHH);                       \
488 
+     const vector unsigned char mask##i =                       \
489 
+         vec_perm((vector unsigned char)zero, (vector unsigned char)neg1, perms##i); \
490 
+     const vector unsigned char vg2##i =                        \
491 
+         vec_perm(vg##i, vg##i, perms##i);                      \
492 
+     const vector unsigned char svA##i =                        \
493 
+         vec_sel(vbA##i, vg2##i, mask##i);                      \
494 
+     const vector unsigned char svB##i =                        \
495 
+         vec_sel(vg2##i, vbB##i, mask##i);                      \
496 
+     vec_st(svA##i, i * stride, src2);                          \
497 
+     vec_st(svB##i, i * stride + 16, src2)
498 
+
499 
+     STORE(4);
500 
+     STORE(5);
501 501 
 }
502 502
503 503 
 static inline void dering_altivec(uint8_t src[], int stride, PPContext *c) {
504 
-  /*
504 
+    /*
505 505 
     this code makes no assumption on src or stride.
506 506 
     One could remove the recomputation of the perm
507 507 
     vector by assuming (stride % 16) == 0, unfortunately
508 508 
     this is not always true. Quite a lot of load/stores
509 509 
     can be removed by assuming proper alignment of
510 510 
     src & stride :-(
511 
-  */
512 
-  uint8_t *srcCopy = src;
513 
-  DECLARE_ALIGNED(16, uint8_t, dt[16]);
514 
-  const vector signed int zero = vec_splat_s32(0);
515 
-  vector unsigned char v_dt;
516 
-  dt[0] = deringThreshold;
517 
-  v_dt = vec_splat(vec_ld(0, dt), 0);
511 
+    */
512 
+    uint8_t *srcCopy = src;
513 
+    DECLARE_ALIGNED(16, uint8_t, dt[16]);
514 
+    const vector signed int zero = vec_splat_s32(0);
515 
+    vector unsigned char v_dt;
516 
+    dt[0] = deringThreshold;
517 
+    v_dt = vec_splat(vec_ld(0, dt), 0);
518 518
519 519 
 #define LOAD_LINE(i)                                                    \
520 
-  const vector unsigned char perm##i =                                  \
521 
-    vec_lvsl(i * stride, srcCopy);                                      \
522 
-  vector unsigned char sA##i = vec_ld(i * stride, srcCopy);             \
523 
-  vector unsigned char sB##i = vec_ld(i * stride + 16, srcCopy);        \
524 
-  vector unsigned char src##i = vec_perm(sA##i, sB##i, perm##i)
525 
-
526 
-  LOAD_LINE(0);
527 
-  LOAD_LINE(1);
528 
-  LOAD_LINE(2);
529 
-  LOAD_LINE(3);
530 
-  LOAD_LINE(4);
531 
-  LOAD_LINE(5);
532 
-  LOAD_LINE(6);
533 
-  LOAD_LINE(7);
534 
-  LOAD_LINE(8);
535 
-  LOAD_LINE(9);
520 
+    const vector unsigned char perm##i =                               \
521 
+        vec_lvsl(i * stride, srcCopy);                                 \
522 
+    vector unsigned char sA##i = vec_ld(i * stride, srcCopy);          \
523 
+    vector unsigned char sB##i = vec_ld(i * stride + 16, srcCopy);     \
524 
+    vector unsigned char src##i = vec_perm(sA##i, sB##i, perm##i)
525 
+
526 
+    LOAD_LINE(0);
527 
+    LOAD_LINE(1);
528 
+    LOAD_LINE(2);
529 
+    LOAD_LINE(3);
530 
+    LOAD_LINE(4);
531 
+    LOAD_LINE(5);
532 
+    LOAD_LINE(6);
533 
+    LOAD_LINE(7);
534 
+    LOAD_LINE(8);
535 
+    LOAD_LINE(9);
536 536 
 #undef LOAD_LINE
537 537
538 
-  vector unsigned char v_avg;
539 
-  {
538 
+    vector unsigned char v_avg;
539 
+    {
540 540 
     const vector unsigned char trunc_perm = (vector unsigned char)
541 
-      AVV(0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,
542 
-          0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18);
541 
+        AVV(0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08,
542 
+            0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17, 0x18);
543 543 
     const vector unsigned char trunc_src12 = vec_perm(src1, src2, trunc_perm);
544 544 
     const vector unsigned char trunc_src34 = vec_perm(src3, src4, trunc_perm);
545 545 
     const vector unsigned char trunc_src56 = vec_perm(src5, src6, trunc_perm);
546 546 
     const vector unsigned char trunc_src78 = vec_perm(src7, src8, trunc_perm);
547 547
548 548 
 #define EXTRACT(op) do {                                                \
549 
-      const vector unsigned char s##op##_1 = vec_##op(trunc_src12, trunc_src34); \
550 
-      const vector unsigned char s##op##_2 = vec_##op(trunc_src56, trunc_src78); \
551 
-      const vector unsigned char s##op##_6 = vec_##op(s##op##_1, s##op##_2); \
552 
-      const vector unsigned char s##op##_8h = vec_mergeh(s##op##_6, s##op##_6); \
553 
-      const vector unsigned char s##op##_8l = vec_mergel(s##op##_6, s##op##_6); \
554 
-      const vector unsigned char s##op##_9 = vec_##op(s##op##_8h, s##op##_8l); \
555 
-      const vector unsigned char s##op##_9h = vec_mergeh(s##op##_9, s##op##_9); \
556 
-      const vector unsigned char s##op##_9l = vec_mergel(s##op##_9, s##op##_9); \
557 
-      const vector unsigned char s##op##_10 = vec_##op(s##op##_9h, s##op##_9l); \
558 
-      const vector unsigned char s##op##_10h = vec_mergeh(s##op##_10, s##op##_10); \
559 
-      const vector unsigned char s##op##_10l = vec_mergel(s##op##_10, s##op##_10); \
560 
-      const vector unsigned char s##op##_11 = vec_##op(s##op##_10h, s##op##_10l); \
561 
-      const vector unsigned char s##op##_11h = vec_mergeh(s##op##_11, s##op##_11); \
562 
-      const vector unsigned char s##op##_11l = vec_mergel(s##op##_11, s##op##_11); \
563 
-      v_##op = vec_##op(s##op##_11h, s##op##_11l); } while (0)
549 
+    const vector unsigned char s##op##_1   = vec_##op(trunc_src12, trunc_src34); \
550 
+    const vector unsigned char s##op##_2   = vec_##op(trunc_src56, trunc_src78); \
551 
+    const vector unsigned char s##op##_6   = vec_##op(s##op##_1, s##op##_2);     \
552 
+    const vector unsigned char s##op##_8h  = vec_mergeh(s##op##_6, s##op##_6);   \
553 
+    const vector unsigned char s##op##_8l  = vec_mergel(s##op##_6, s##op##_6);   \
554 
+    const vector unsigned char s##op##_9   = vec_##op(s##op##_8h, s##op##_8l);   \
555 
+    const vector unsigned char s##op##_9h  = vec_mergeh(s##op##_9, s##op##_9);   \
556 
+    const vector unsigned char s##op##_9l  = vec_mergel(s##op##_9, s##op##_9);   \
557 
+    const vector unsigned char s##op##_10  = vec_##op(s##op##_9h, s##op##_9l);   \
558 
+    const vector unsigned char s##op##_10h = vec_mergeh(s##op##_10, s##op##_10); \
559 
+    const vector unsigned char s##op##_10l = vec_mergel(s##op##_10, s##op##_10); \
560 
+    const vector unsigned char s##op##_11  = vec_##op(s##op##_10h, s##op##_10l); \
561 
+    const vector unsigned char s##op##_11h = vec_mergeh(s##op##_11, s##op##_11); \
562 
+    const vector unsigned char s##op##_11l = vec_mergel(s##op##_11, s##op##_11); \
563 
+    v_##op = vec_##op(s##op##_11h, s##op##_11l); } while (0)
564 564
565 565 
     vector unsigned char v_min;
566 566 
     vector unsigned char v_max;
... ... 
@@ -569,19 +568,19 @@ static inline void dering_altivec(uint8_t src[], int stride, PPContext *c) {
569 569 
 #undef EXTRACT
570 570
571 571 
     if (vec_all_lt(vec_sub(v_max, v_min), v_dt))
572 
-      return;
572 
+        return;
573 573
574 574 
     v_avg = vec_avg(v_min, v_max);
575 
-  }
575 
+    }
576 576
577 
-  DECLARE_ALIGNED(16, signed int, S[8]);
578 
-  {
577 
+    DECLARE_ALIGNED(16, signed int, S[8]);
578 
+    {
579 579 
     const vector unsigned short mask1 = (vector unsigned short)
580 
-      AVV(0x0001, 0x0002, 0x0004, 0x0008,
581 
-          0x0010, 0x0020, 0x0040, 0x0080);
580 
+        AVV(0x0001, 0x0002, 0x0004, 0x0008,
581 
+            0x0010, 0x0020, 0x0040, 0x0080);
582 582 
     const vector unsigned short mask2 = (vector unsigned short)
583 
-      AVV(0x0100, 0x0200, 0x0000, 0x0000,
584 
-          0x0000, 0x0000, 0x0000, 0x0000);
583 
+        AVV(0x0100, 0x0200, 0x0000, 0x0000,
584 
+            0x0000, 0x0000, 0x0000, 0x0000);
585 585
586 586 
     const vector unsigned int vuint32_16 = vec_sl(vec_splat_u32(1), vec_splat_u32(4));
587 587 
     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) {
589 589 
 #define COMPARE(i)                                                      \
590 590 
     vector signed int sum##i;                                           \
591 591 
     do {                                                                \
592 
-      const vector unsigned char cmp##i =                               \
593 
-        (vector unsigned char)vec_cmpgt(src##i, v_avg);                 \
594 
-      const vector unsigned short cmpHi##i =                            \
595 
-        (vector unsigned short)vec_mergeh(cmp##i, cmp##i);              \
596 
-      const vector unsigned short cmpLi##i =                            \
597 
-        (vector unsigned short)vec_mergel(cmp##i, cmp##i);              \
598 
-      const vector signed short cmpHf##i =                              \
599 
-        (vector signed short)vec_and(cmpHi##i, mask1);                  \
600 
-      const vector signed short cmpLf##i =                              \
601 
-        (vector signed short)vec_and(cmpLi##i, mask2);                  \
602 
-      const vector signed int sump##i = vec_sum4s(cmpHf##i, zero);      \
603 
-      const vector signed int sumq##i = vec_sum4s(cmpLf##i, sump##i);   \
604 
-      sum##i  = vec_sums(sumq##i, zero); } while (0)
592 
+        const vector unsigned char cmp##i =                             \
593 
+            (vector unsigned char)vec_cmpgt(src##i, v_avg);             \
594 
+        const vector unsigned short cmpHi##i =                          \
595 
+            (vector unsigned short)vec_mergeh(cmp##i, cmp##i);          \
596 
+        const vector unsigned short cmpLi##i =                          \
597 
+            (vector unsigned short)vec_mergel(cmp##i, cmp##i);          \
598 
+        const vector signed short cmpHf##i =                            \
599 
+            (vector signed short)vec_and(cmpHi##i, mask1);              \
600 
+        const vector signed short cmpLf##i =                            \
601 
+            (vector signed short)vec_and(cmpLi##i, mask2);              \
602 
+        const vector signed int sump##i = vec_sum4s(cmpHf##i, zero);    \
603 
+        const vector signed int sumq##i = vec_sum4s(cmpLf##i, sump##i); \
604 
+        sum##i  = vec_sums(sumq##i, zero); } while (0)
605 605
606 606 
     COMPARE(0);
607 607 
     COMPARE(1);
... ... 
@@ -618,178 +617,178 @@ static inline void dering_altivec(uint8_t src[], int stride, PPContext *c) {
618 618 
     vector signed int sumA2;
619 619 
     vector signed int sumB2;
620 620 
     {
621 
-      const vector signed int sump02 = vec_mergel(sum0, sum2);
622 
-      const vector signed int sump13 = vec_mergel(sum1, sum3);
623 
-      const vector signed int sumA = vec_mergel(sump02, sump13);
624 
-
625 
-      const vector signed int sump46 = vec_mergel(sum4, sum6);
626 
-      const vector signed int sump57 = vec_mergel(sum5, sum7);
627 
-      const vector signed int sumB = vec_mergel(sump46, sump57);
628 
-
629 
-      const vector signed int sump8A = vec_mergel(sum8, zero);
630 
-      const vector signed int sump9B = vec_mergel(sum9, zero);
631 
-      const vector signed int sumC = vec_mergel(sump8A, sump9B);
632 
-
633 
-      const vector signed int tA = vec_sl(vec_nor(zero, sumA), vuint32_16);
634 
-      const vector signed int tB = vec_sl(vec_nor(zero, sumB), vuint32_16);
635 
-      const vector signed int tC = vec_sl(vec_nor(zero, sumC), vuint32_16);
636 
-      const vector signed int t2A = vec_or(sumA, tA);
637 
-      const vector signed int t2B = vec_or(sumB, tB);
638 
-      const vector signed int t2C = vec_or(sumC, tC);
639 
-      const vector signed int t3A = vec_and(vec_sra(t2A, vuint32_1),
640 
-                                            vec_sl(t2A, vuint32_1));
641 
-      const vector signed int t3B = vec_and(vec_sra(t2B, vuint32_1),
642 
-                                            vec_sl(t2B, vuint32_1));
643 
-      const vector signed int t3C = vec_and(vec_sra(t2C, vuint32_1),
644 
-                                            vec_sl(t2C, vuint32_1));
645 
-      const vector signed int yA = vec_and(t2A, t3A);
646 
-      const vector signed int yB = vec_and(t2B, t3B);
647 
-      const vector signed int yC = vec_and(t2C, t3C);
648 
-
649 
-      const vector unsigned char strangeperm1 = vec_lvsl(4, (unsigned char*)0);
650 
-      const vector unsigned char strangeperm2 = vec_lvsl(8, (unsigned char*)0);
651 
-      const vector signed int sumAd4 = vec_perm(yA, yB, strangeperm1);
652 
-      const vector signed int sumAd8 = vec_perm(yA, yB, strangeperm2);
653 
-      const vector signed int sumBd4 = vec_perm(yB, yC, strangeperm1);
654 
-      const vector signed int sumBd8 = vec_perm(yB, yC, strangeperm2);
655 
-      const vector signed int sumAp = vec_and(yA,
656 
-                                              vec_and(sumAd4,sumAd8));
657 
-      const vector signed int sumBp = vec_and(yB,
658 
-                                              vec_and(sumBd4,sumBd8));
659 
-      sumA2 = vec_or(sumAp,
660 
-                     vec_sra(sumAp,
661 
-                             vuint32_16));
662 
-      sumB2  = vec_or(sumBp,
663 
-                      vec_sra(sumBp,
664 
-                              vuint32_16));
621 
+    const vector signed int sump02 = vec_mergel(sum0, sum2);
622 
+    const vector signed int sump13 = vec_mergel(sum1, sum3);
623 
+    const vector signed int sumA = vec_mergel(sump02, sump13);
624 
+
625 
+    const vector signed int sump46 = vec_mergel(sum4, sum6);
626 
+    const vector signed int sump57 = vec_mergel(sum5, sum7);
627 
+    const vector signed int sumB = vec_mergel(sump46, sump57);
628 
+
629 
+    const vector signed int sump8A = vec_mergel(sum8, zero);
630 
+    const vector signed int sump9B = vec_mergel(sum9, zero);
631 
+    const vector signed int sumC = vec_mergel(sump8A, sump9B);
632 
+
633 
+    const vector signed int tA = vec_sl(vec_nor(zero, sumA), vuint32_16);
634 
+    const vector signed int tB = vec_sl(vec_nor(zero, sumB), vuint32_16);
635 
+    const vector signed int tC = vec_sl(vec_nor(zero, sumC), vuint32_16);
636 
+    const vector signed int t2A = vec_or(sumA, tA);
637 
+    const vector signed int t2B = vec_or(sumB, tB);
638 
+    const vector signed int t2C = vec_or(sumC, tC);
639 
+    const vector signed int t3A = vec_and(vec_sra(t2A, vuint32_1),
640 
+                                          vec_sl(t2A, vuint32_1));
641 
+    const vector signed int t3B = vec_and(vec_sra(t2B, vuint32_1),
642 
+                                          vec_sl(t2B, vuint32_1));
643 
+    const vector signed int t3C = vec_and(vec_sra(t2C, vuint32_1),
644 
+                                          vec_sl(t2C, vuint32_1));
645 
+    const vector signed int yA = vec_and(t2A, t3A);
646 
+    const vector signed int yB = vec_and(t2B, t3B);
647 
+    const vector signed int yC = vec_and(t2C, t3C);
648 
+
649 
+    const vector unsigned char strangeperm1 = vec_lvsl(4, (unsigned char*)0);
650 
+    const vector unsigned char strangeperm2 = vec_lvsl(8, (unsigned char*)0);
651 
+    const vector signed int sumAd4 = vec_perm(yA, yB, strangeperm1);
652 
+    const vector signed int sumAd8 = vec_perm(yA, yB, strangeperm2);
653 
+    const vector signed int sumBd4 = vec_perm(yB, yC, strangeperm1);
654 
+    const vector signed int sumBd8 = vec_perm(yB, yC, strangeperm2);
655 
+    const vector signed int sumAp = vec_and(yA,
656 
+                                            vec_and(sumAd4,sumAd8));
657 
+    const vector signed int sumBp = vec_and(yB,
658 
+                                            vec_and(sumBd4,sumBd8));
659 
+    sumA2 = vec_or(sumAp,
660 
+                   vec_sra(sumAp,
661 
+                           vuint32_16));
662 
+    sumB2  = vec_or(sumBp,
663 
+                    vec_sra(sumBp,
664 
+                            vuint32_16));
665 665 
     }
666 666 
     vec_st(sumA2, 0, S);
667 667 
     vec_st(sumB2, 16, S);
668 
-  }
669 
-
670 
-  /* I'm not sure the following is actually faster
671 
-     than straight, unvectorized C code :-( */
672 
-
673 
-  DECLARE_ALIGNED(16, int, tQP2[4]);
674 
-  tQP2[0]= c->QP/2 + 1;
675 
-  vector signed int vQP2 = vec_ld(0, tQP2);
676 
-  vQP2 = vec_splat(vQP2, 0);
677 
-  const vector signed int vsint32_8 = vec_splat_s32(8);
678 
-  const vector unsigned int vuint32_4 = vec_splat_u32(4);
679 
-
680 
-  const vector unsigned char permA1 = (vector unsigned char)
681 
-    AVV(0x00, 0x01, 0x02, 0x10, 0x11, 0x12, 0x1F, 0x1F,
682 
-        0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F);
683 
-  const vector unsigned char permA2 = (vector unsigned char)
684 
-    AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x10, 0x11,
685 
-        0x12, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F);
686 
-  const vector unsigned char permA1inc = (vector unsigned char)
687 
-    AVV(0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00,
688 
-        0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
689 
-  const vector unsigned char permA2inc = (vector unsigned char)
690 
-    AVV(0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x01,
691 
-        0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
692 
-  const vector unsigned char magic = (vector unsigned char)
693 
-    AVV(0x01, 0x02, 0x01, 0x02, 0x04, 0x02, 0x01, 0x02,
694 
-        0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
695 
-  const vector unsigned char extractPerm = (vector unsigned char)
696 
-    AVV(0x10, 0x10, 0x10, 0x01, 0x10, 0x10, 0x10, 0x01,
697 
-        0x10, 0x10, 0x10, 0x01, 0x10, 0x10, 0x10, 0x01);
698 
-  const vector unsigned char extractPermInc = (vector unsigned char)
699 
-    AVV(0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01,
700 
-        0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01);
701 
-  const vector unsigned char identity = vec_lvsl(0,(unsigned char *)0);
702 
-  const vector unsigned char tenRight = (vector unsigned char)
703 
-    AVV(0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
704 
-        0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
705 
-  const vector unsigned char eightLeft = (vector unsigned char)
706 
-    AVV(0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
707 
-        0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x08);
668 
+    }
669 
+
670 
+    /* I'm not sure the following is actually faster
671 
+       than straight, unvectorized C code :-( */
672 
+
673 
+    DECLARE_ALIGNED(16, int, tQP2[4]);
674 
+    tQP2[0]= c->QP/2 + 1;
675 
+    vector signed int vQP2 = vec_ld(0, tQP2);
676 
+    vQP2 = vec_splat(vQP2, 0);
677 
+    const vector signed int vsint32_8 = vec_splat_s32(8);
678 
+    const vector unsigned int vuint32_4 = vec_splat_u32(4);
679 
+
680 
+    const vector unsigned char permA1 = (vector unsigned char)
681 
+        AVV(0x00, 0x01, 0x02, 0x10, 0x11, 0x12, 0x1F, 0x1F,
682 
+            0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F);
683 
+    const vector unsigned char permA2 = (vector unsigned char)
684 
+        AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x10, 0x11,
685 
+            0x12, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F, 0x1F);
686 
+    const vector unsigned char permA1inc = (vector unsigned char)
687 
+        AVV(0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00,
688 
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
689 
+    const vector unsigned char permA2inc = (vector unsigned char)
690 
+        AVV(0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x01,
691 
+            0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
692 
+    const vector unsigned char magic = (vector unsigned char)
693 
+        AVV(0x01, 0x02, 0x01, 0x02, 0x04, 0x02, 0x01, 0x02,
694 
+            0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
695 
+    const vector unsigned char extractPerm = (vector unsigned char)
696 
+        AVV(0x10, 0x10, 0x10, 0x01, 0x10, 0x10, 0x10, 0x01,
697 
+            0x10, 0x10, 0x10, 0x01, 0x10, 0x10, 0x10, 0x01);
698 
+    const vector unsigned char extractPermInc = (vector unsigned char)
699 
+        AVV(0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01,
700 
+            0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01);
701 
+    const vector unsigned char identity = vec_lvsl(0,(unsigned char *)0);
702 
+    const vector unsigned char tenRight = (vector unsigned char)
703 
+        AVV(0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
704 
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00);
705 
+    const vector unsigned char eightLeft = (vector unsigned char)
706 
+        AVV(0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
707 
+            0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x08);
708 708
709 709
710 710 
 #define F_INIT(i)                                       \
711 
-  vector unsigned char tenRightM##i = tenRight;         \
712 
-  vector unsigned char permA1M##i = permA1;             \
713 
-  vector unsigned char permA2M##i = permA2;             \
714 
-  vector unsigned char extractPermM##i = extractPerm
711 
+    vector unsigned char tenRightM##i = tenRight;       \
712 
+    vector unsigned char permA1M##i = permA1;           \
713 
+    vector unsigned char permA2M##i = permA2;           \
714 
+    vector unsigned char extractPermM##i = extractPerm
715 715
716 716 
 #define F2(i, j, k, l)                                                  \
717 
-  if (S[i] & (1 << (l+1))) {                                            \
718 
-    const vector unsigned char a_##j##_A##l =                           \
719 
-      vec_perm(src##i, src##j, permA1M##i);                             \
720 
-    const vector unsigned char a_##j##_B##l =                           \
721 
-      vec_perm(a_##j##_A##l, src##k, permA2M##i);                       \
722 
-    const vector signed int a_##j##_sump##l =                           \
723 
-      (vector signed int)vec_msum(a_##j##_B##l, magic,                  \
724 
-                                  (vector unsigned int)zero);           \
725 
-    vector signed int F_##j##_##l =                                     \
726 
-      vec_sr(vec_sums(a_##j##_sump##l, vsint32_8), vuint32_4);          \
727 
-    F_##j##_##l = vec_splat(F_##j##_##l, 3);                            \
728 
-    const vector signed int p_##j##_##l =                               \
729 
-      (vector signed int)vec_perm(src##j,                               \
730 
-                                  (vector unsigned char)zero,           \
731 
-                                  extractPermM##i);                     \
732 
-    const vector signed int sum_##j##_##l = vec_add( p_##j##_##l, vQP2);\
733 
-    const vector signed int diff_##j##_##l = vec_sub( p_##j##_##l, vQP2);\
734 
-    vector signed int newpm_##j##_##l;                                  \
735 
-    if (vec_all_lt(sum_##j##_##l, F_##j##_##l))                         \
736 
-      newpm_##j##_##l = sum_##j##_##l;                                  \
737 
-    else if (vec_all_gt(diff_##j##_##l, F_##j##_##l))                   \
738 
-      newpm_##j##_##l = diff_##j##_##l;                                 \
739 
-    else newpm_##j##_##l = F_##j##_##l;                                 \
740 
-    const vector unsigned char newpm2_##j##_##l =                       \
741 
-      vec_splat((vector unsigned char)newpm_##j##_##l, 15);             \
742 
-    const vector unsigned char mask##j##l = vec_add(identity,           \
743 
-                                                    tenRightM##i);      \
744 
-    src##j = vec_perm(src##j, newpm2_##j##_##l, mask##j##l);            \
745 
-  }                                                                     \
746 
-  permA1M##i = vec_add(permA1M##i, permA1inc);                          \
747 
-  permA2M##i = vec_add(permA2M##i, permA2inc);                          \
748 
-  tenRightM##i = vec_sro(tenRightM##i, eightLeft);                      \
749 
-  extractPermM##i = vec_add(extractPermM##i, extractPermInc)
717 
+    if (S[i] & (1 << (l+1))) {                                          \
718 
+        const vector unsigned char a_##j##_A##l =                       \
719 
+            vec_perm(src##i, src##j, permA1M##i);                       \
720 
+        const vector unsigned char a_##j##_B##l =                       \
721 
+            vec_perm(a_##j##_A##l, src##k, permA2M##i);                 \
722 
+        const vector signed int a_##j##_sump##l =                       \
723 
+            (vector signed int)vec_msum(a_##j##_B##l, magic,            \
724 
+                                        (vector unsigned int)zero);     \
725 
+        vector signed int F_##j##_##l =                                 \
726 
+            vec_sr(vec_sums(a_##j##_sump##l, vsint32_8), vuint32_4);    \
727 
+        F_##j##_##l = vec_splat(F_##j##_##l, 3);                        \
728 
+        const vector signed int p_##j##_##l =                           \
729 
+            (vector signed int)vec_perm(src##j,                         \
730 
+                                        (vector unsigned char)zero,     \
731 
+                                        extractPermM##i);               \
732 
+        const vector signed int sum_##j##_##l  = vec_add( p_##j##_##l, vQP2);\
733 
+        const vector signed int diff_##j##_##l = vec_sub( p_##j##_##l, vQP2);\
734 
+        vector signed int newpm_##j##_##l;                              \
735 
+        if (vec_all_lt(sum_##j##_##l, F_##j##_##l))                     \
736 
+            newpm_##j##_##l = sum_##j##_##l;                            \
737 
+        else if (vec_all_gt(diff_##j##_##l, F_##j##_##l))               \
738 
+            newpm_##j##_##l = diff_##j##_##l;                           \
739 
+        else newpm_##j##_##l = F_##j##_##l;                             \
740 
+        const vector unsigned char newpm2_##j##_##l =                   \
741 
+            vec_splat((vector unsigned char)newpm_##j##_##l, 15);       \
742 
+        const vector unsigned char mask##j##l = vec_add(identity,       \
743 
+                                                        tenRightM##i);  \
744 
+        src##j = vec_perm(src##j, newpm2_##j##_##l, mask##j##l);        \
745 
+    }                                                                   \
746 
+    permA1M##i = vec_add(permA1M##i, permA1inc);                        \
747 
+    permA2M##i = vec_add(permA2M##i, permA2inc);                        \
748 
+    tenRightM##i = vec_sro(tenRightM##i, eightLeft);                    \
749 
+    extractPermM##i = vec_add(extractPermM##i, extractPermInc)
750 750
751 751 
 #define ITER(i, j, k)                           \
752 
-  F_INIT(i);                                    \
753 
-  F2(i, j, k, 0);                               \
754 
-  F2(i, j, k, 1);                               \
755 
-  F2(i, j, k, 2);                               \
756 
-  F2(i, j, k, 3);                               \
757 
-  F2(i, j, k, 4);                               \
758 
-  F2(i, j, k, 5);                               \
759 
-  F2(i, j, k, 6);                               \
760 
-  F2(i, j, k, 7)
761 
-
762 
-  ITER(0, 1, 2);
763 
-  ITER(1, 2, 3);
764 
-  ITER(2, 3, 4);
765 
-  ITER(3, 4, 5);
766 
-  ITER(4, 5, 6);
767 
-  ITER(5, 6, 7);
768 
-  ITER(6, 7, 8);
769 
-  ITER(7, 8, 9);
770 
-
771 
-  const vector signed char neg1 = vec_splat_s8(-1);
752 
+    F_INIT(i);                                  \
753 
+    F2(i, j, k, 0);                             \
754 
+    F2(i, j, k, 1);                             \
755 
+    F2(i, j, k, 2);                             \
756 
+    F2(i, j, k, 3);                             \
757 
+    F2(i, j, k, 4);                             \
758 
+    F2(i, j, k, 5);                             \
759 
+    F2(i, j, k, 6);                             \
760 
+    F2(i, j, k, 7)
761 
+
762 
+    ITER(0, 1, 2);
763 
+    ITER(1, 2, 3);
764 
+    ITER(2, 3, 4);
765 
+    ITER(3, 4, 5);
766 
+    ITER(4, 5, 6);
767 
+    ITER(5, 6, 7);
768 
+    ITER(6, 7, 8);
769 
+    ITER(7, 8, 9);
770 
+
771 
+    const vector signed char neg1 = vec_splat_s8(-1);
772 772
773 773 
 #define STORE_LINE(i)                                   \
774 
-  const vector unsigned char permST##i =                \
775 
-    vec_lvsr(i * stride, srcCopy);                      \
776 
-  const vector unsigned char maskST##i =                \
777 
-    vec_perm((vector unsigned char)zero,                \
778 
-             (vector unsigned char)neg1, permST##i);    \
779 
-  src##i = vec_perm(src##i ,src##i, permST##i);         \
780 
-  sA##i= vec_sel(sA##i, src##i, maskST##i);             \
781 
-  sB##i= vec_sel(src##i, sB##i, maskST##i);             \
782 
-  vec_st(sA##i, i * stride, srcCopy);                   \
783 
-  vec_st(sB##i, i * stride + 16, srcCopy)
784 
-
785 
-  STORE_LINE(1);
786 
-  STORE_LINE(2);
787 
-  STORE_LINE(3);
788 
-  STORE_LINE(4);
789 
-  STORE_LINE(5);
790 
-  STORE_LINE(6);
791 
-  STORE_LINE(7);
792 
-  STORE_LINE(8);
774 
+    const vector unsigned char permST##i =              \
775 
+        vec_lvsr(i * stride, srcCopy);                  \
776 
+    const vector unsigned char maskST##i =              \
777 
+        vec_perm((vector unsigned char)zero,            \
778 
+                 (vector unsigned char)neg1, permST##i);\
779 
+    src##i = vec_perm(src##i ,src##i, permST##i);       \
780 
+    sA##i= vec_sel(sA##i, src##i, maskST##i);           \
781 
+    sB##i= vec_sel(src##i, sB##i, maskST##i);           \
782 
+    vec_st(sA##i, i * stride, srcCopy);                 \
783 
+    vec_st(sB##i, i * stride + 16, srcCopy)
784 
+
785 
+    STORE_LINE(1);
786 
+    STORE_LINE(2);
787 
+    STORE_LINE(3);
788 
+    STORE_LINE(4);
789 
+    STORE_LINE(5);
790 
+    STORE_LINE(6);
791 
+    STORE_LINE(7);
792 
+    STORE_LINE(8);
793 793
794 794 
 #undef STORE_LINE
795 795 
 #undef ITER
... ... 
@@ -801,386 +800,386 @@ static inline void dering_altivec(uint8_t src[], int stride, PPContext *c) {
801 801 
 #define do_a_deblock_altivec(a...) do_a_deblock_C(a)
802 802
803 803 
 static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
804 
-                                    uint8_t *tempBlured, uint32_t *tempBluredPast, int *maxNoise)
804 
+                                            uint8_t *tempBlured, uint32_t *tempBluredPast, int *maxNoise)
805 805 
 {
806 
-  const vector signed int zero = vec_splat_s32(0);
807 
-  const vector signed short vsint16_1 = vec_splat_s16(1);
808 
-  vector signed int v_dp = zero;
809 
-  vector signed int v_sysdp = zero;
810 
-  int d, sysd, i;
806 
+    const vector signed int zero = vec_splat_s32(0);
807 
+    const vector signed short vsint16_1 = vec_splat_s16(1);
808 
+    vector signed int v_dp = zero;
809 
+    vector signed int v_sysdp = zero;
810 
+    int d, sysd, i;
811 811
812 
-  tempBluredPast[127]= maxNoise[0];
813 
-  tempBluredPast[128]= maxNoise[1];
814 
-  tempBluredPast[129]= maxNoise[2];
812 
+    tempBluredPast[127]= maxNoise[0];
813 
+    tempBluredPast[128]= maxNoise[1];
814 
+    tempBluredPast[129]= maxNoise[2];
815 815
816 816 
 #define LOAD_LINE(src, i)                                               \
817 
-  register int j##src##i = i * stride;                                  \
818 
-  vector unsigned char perm##src##i = vec_lvsl(j##src##i, src);         \
819 
-  const vector unsigned char v_##src##A1##i = vec_ld(j##src##i, src);   \
820 
-  const vector unsigned char v_##src##A2##i = vec_ld(j##src##i + 16, src); \
821 
-  const vector unsigned char v_##src##A##i =                            \
822 
-    vec_perm(v_##src##A1##i, v_##src##A2##i, perm##src##i);             \
823 
-  vector signed short v_##src##Ass##i =                                 \
824 
-    (vector signed short)vec_mergeh((vector signed char)zero,           \
825 
-                                    (vector signed char)v_##src##A##i)
826 
-
827 
-  LOAD_LINE(src, 0);
828 
-  LOAD_LINE(src, 1);
829 
-  LOAD_LINE(src, 2);
830 
-  LOAD_LINE(src, 3);
831 
-  LOAD_LINE(src, 4);
832 
-  LOAD_LINE(src, 5);
833 
-  LOAD_LINE(src, 6);
834 
-  LOAD_LINE(src, 7);
835 
-
836 
-  LOAD_LINE(tempBlured, 0);
837 
-  LOAD_LINE(tempBlured, 1);
838 
-  LOAD_LINE(tempBlured, 2);
839 
-  LOAD_LINE(tempBlured, 3);
840 
-  LOAD_LINE(tempBlured, 4);
841 
-  LOAD_LINE(tempBlured, 5);
842 
-  LOAD_LINE(tempBlured, 6);
843 
-  LOAD_LINE(tempBlured, 7);
817 
+    register int j##src##i = i * stride;                                \
818 
+    vector unsigned char perm##src##i = vec_lvsl(j##src##i, src);       \
819 
+    const vector unsigned char v_##src##A1##i = vec_ld(j##src##i, src); \
820 
+    const vector unsigned char v_##src##A2##i = vec_ld(j##src##i + 16, src); \
821 
+    const vector unsigned char v_##src##A##i =                          \
822 
+        vec_perm(v_##src##A1##i, v_##src##A2##i, perm##src##i);         \
823 
+    vector signed short v_##src##Ass##i =                               \
824 
+        (vector signed short)vec_mergeh((vector signed char)zero,       \
825 
+                                        (vector signed char)v_##src##A##i)
826 
+
827 
+    LOAD_LINE(src, 0);
828 
+    LOAD_LINE(src, 1);
829 
+    LOAD_LINE(src, 2);
830 
+    LOAD_LINE(src, 3);
831 
+    LOAD_LINE(src, 4);
832 
+    LOAD_LINE(src, 5);
833 
+    LOAD_LINE(src, 6);
834 
+    LOAD_LINE(src, 7);
835 
+
836 
+    LOAD_LINE(tempBlured, 0);
837 
+    LOAD_LINE(tempBlured, 1);
838 
+    LOAD_LINE(tempBlured, 2);
839 
+    LOAD_LINE(tempBlured, 3);
840 
+    LOAD_LINE(tempBlured, 4);
841 
+    LOAD_LINE(tempBlured, 5);
842 
+    LOAD_LINE(tempBlured, 6);
843 
+    LOAD_LINE(tempBlured, 7);
844 844 
 #undef LOAD_LINE
845 845
846 846 
 #define ACCUMULATE_DIFFS(i)                                     \
847 
-  vector signed short v_d##i = vec_sub(v_tempBluredAss##i,      \
848 
-                                       v_srcAss##i);            \
849 
-  v_dp = vec_msums(v_d##i, v_d##i, v_dp);                       \
850 
-  v_sysdp = vec_msums(v_d##i, vsint16_1, v_sysdp)
851 
-
852 
-  ACCUMULATE_DIFFS(0);
853 
-  ACCUMULATE_DIFFS(1);
854 
-  ACCUMULATE_DIFFS(2);
855 
-  ACCUMULATE_DIFFS(3);
856 
-  ACCUMULATE_DIFFS(4);
857 
-  ACCUMULATE_DIFFS(5);
858 
-  ACCUMULATE_DIFFS(6);
859 
-  ACCUMULATE_DIFFS(7);
847 
+    vector signed short v_d##i = vec_sub(v_tempBluredAss##i,    \
848 
+                                         v_srcAss##i);          \
849 
+    v_dp = vec_msums(v_d##i, v_d##i, v_dp);                     \
850 
+    v_sysdp = vec_msums(v_d##i, vsint16_1, v_sysdp)
851 
+
852 
+    ACCUMULATE_DIFFS(0);
853 
+    ACCUMULATE_DIFFS(1);
854 
+    ACCUMULATE_DIFFS(2);
855 
+    ACCUMULATE_DIFFS(3);
856 
+    ACCUMULATE_DIFFS(4);
857 
+    ACCUMULATE_DIFFS(5);
858 
+    ACCUMULATE_DIFFS(6);
859 
+    ACCUMULATE_DIFFS(7);
860 860 
 #undef ACCUMULATE_DIFFS
861 861
862 
-  v_dp = vec_sums(v_dp, zero);
863 
-  v_sysdp = vec_sums(v_sysdp, zero);
862 
+    v_dp = vec_sums(v_dp, zero);
863 
+    v_sysdp = vec_sums(v_sysdp, zero);
864 864
865 
-  v_dp = vec_splat(v_dp, 3);
866 
-  v_sysdp = vec_splat(v_sysdp, 3);
865 
+    v_dp = vec_splat(v_dp, 3);
866 
+    v_sysdp = vec_splat(v_sysdp, 3);
867 867
868 
-  vec_ste(v_dp, 0, &d);
869 
-  vec_ste(v_sysdp, 0, &sysd);
868 
+    vec_ste(v_dp, 0, &d);
869 
+    vec_ste(v_sysdp, 0, &sysd);
870 870
871 
-  i = d;
872 
-  d = (4*d
873 
-       +(*(tempBluredPast-256))
874 
-       +(*(tempBluredPast-1))+ (*(tempBluredPast+1))
875 
-       +(*(tempBluredPast+256))
876 
-       +4)>>3;
871 
+    i = d;
872 
+    d = (4*d
873 
+         +(*(tempBluredPast-256))
874 
+         +(*(tempBluredPast-1))+ (*(tempBluredPast+1))
875 
+         +(*(tempBluredPast+256))
876 
+         +4)>>3;
877 877
878 
-  *tempBluredPast=i;
878 
+    *tempBluredPast=i;
879 879
880 
-  if (d > maxNoise[1]) {
881 
-    if (d < maxNoise[2]) {
880 
+    if (d > maxNoise[1]) {
881 
+        if (d < maxNoise[2]) {
882 882 
 #define OP(i) v_tempBluredAss##i = vec_avg(v_tempBluredAss##i, v_srcAss##i);
883 883
884 
-      OP(0);
885 
-      OP(1);
886 
-      OP(2);
887 
-      OP(3);
888 
-      OP(4);
889 
-      OP(5);
890 
-      OP(6);
891 
-      OP(7);
884 
+            OP(0);
885 
+            OP(1);
886 
+            OP(2);
887 
+            OP(3);
888 
+            OP(4);
889 
+            OP(5);
890 
+            OP(6);
891 
+            OP(7);
892 892 
 #undef OP
893 
-    } else {
893 
+        } else {
894 894 
 #define OP(i) v_tempBluredAss##i = v_srcAss##i;
895 895
896 
-      OP(0);
897 
-      OP(1);
898 
-      OP(2);
899 
-      OP(3);
900 
-      OP(4);
901 
-      OP(5);
902 
-      OP(6);
903 
-      OP(7);
896 
+            OP(0);
897 
+            OP(1);
898 
+            OP(2);
899 
+            OP(3);
900 
+            OP(4);
901 
+            OP(5);
902 
+            OP(6);
903 
+            OP(7);
904 904 
 #undef OP
905 
-    }
906 
-  } else {
907 
-    if (d < maxNoise[0]) {
908 
-      const vector signed short vsint16_7 = vec_splat_s16(7);
909 
-      const vector signed short vsint16_4 = vec_splat_s16(4);
910 
-      const vector unsigned short vuint16_3 = vec_splat_u16(3);
905 
+        }
906 
+    } else {
907 
+        if (d < maxNoise[0]) {
908 
+            const vector signed short vsint16_7 = vec_splat_s16(7);
909 
+            const vector signed short vsint16_4 = vec_splat_s16(4);
910 
+            const vector unsigned short vuint16_3 = vec_splat_u16(3);
911 911
912 912 
 #define OP(i)                                                   \
913 
-      const vector signed short v_temp##i =                     \
914 
-        vec_mladd(v_tempBluredAss##i,                           \
915 
-                  vsint16_7, v_srcAss##i);                      \
916 
-      const vector signed short v_temp2##i =                    \
917 
-        vec_add(v_temp##i, vsint16_4);                          \
918 
-      v_tempBluredAss##i = vec_sr(v_temp2##i, vuint16_3)
919 
-
920 
-      OP(0);
921 
-      OP(1);
922 
-      OP(2);
923 
-      OP(3);
924 
-      OP(4);
925 
-      OP(5);
926 
-      OP(6);
927 
-      OP(7);
913 
+            const vector signed short v_temp##i =               \
914 
+                vec_mladd(v_tempBluredAss##i,                   \
915 
+                          vsint16_7, v_srcAss##i);              \
916 
+            const vector signed short v_temp2##i =              \
917 
+                vec_add(v_temp##i, vsint16_4);                  \
918 
+            v_tempBluredAss##i = vec_sr(v_temp2##i, vuint16_3)
919 
+
920 
+            OP(0);
921 
+            OP(1);
922 
+            OP(2);
923 
+            OP(3);
924 
+            OP(4);
925 
+            OP(5);
926 
+            OP(6);
927 
+            OP(7);
928 928 
 #undef OP
929 
-    } else {
930 
-      const vector signed short vsint16_3 = vec_splat_s16(3);
931 
-      const vector signed short vsint16_2 = vec_splat_s16(2);
929 
+        } else {
930 
+            const vector signed short vsint16_3 = vec_splat_s16(3);
931 
+            const vector signed short vsint16_2 = vec_splat_s16(2);
932 932
933 933 
 #define OP(i)                                                   \
934 
-      const vector signed short v_temp##i =                     \
935 
-        vec_mladd(v_tempBluredAss##i,                           \
936 
-                  vsint16_3, v_srcAss##i);                      \
937 
-      const vector signed short v_temp2##i =                    \
938 
-        vec_add(v_temp##i, vsint16_2);                          \
939 
-      v_tempBluredAss##i = vec_sr(v_temp2##i, (vector unsigned short)vsint16_2)
940 
-
941 
-      OP(0);
942 
-      OP(1);
943 
-      OP(2);
944 
-      OP(3);
945 
-      OP(4);
946 
-      OP(5);
947 
-      OP(6);
948 
-      OP(7);
934 
+            const vector signed short v_temp##i =               \
935 
+                vec_mladd(v_tempBluredAss##i,                   \
936 
+                          vsint16_3, v_srcAss##i);              \
937 
+            const vector signed short v_temp2##i =              \
938 
+                vec_add(v_temp##i, vsint16_2);                  \
939 
+            v_tempBluredAss##i = vec_sr(v_temp2##i, (vector unsigned short)vsint16_2)
940 
+
941 
+            OP(0);
942 
+            OP(1);
943 
+            OP(2);
944 
+            OP(3);
945 
+            OP(4);
946 
+            OP(5);
947 
+            OP(6);
948 
+            OP(7);
949 949 
 #undef OP
950 
+        }
950 951 
     }
951 
-  }
952 952
953 
-  const vector signed char neg1 = vec_splat_s8(-1);
954 
-  const vector unsigned char permHH = (const vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
955 
-                                                                      0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
953 
+    const vector signed char neg1 = vec_splat_s8(-1);
954 
+    const vector unsigned char permHH = (const vector unsigned char)AVV(0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
955 
+                                                                        0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F);
956 956
957 957 
 #define PACK_AND_STORE(src, i)                                  \
958 
-  const vector unsigned char perms##src##i =                    \
959 
-    vec_lvsr(i * stride, src);                                  \
960 
-  const vector unsigned char vf##src##i =                       \
961 
-    vec_packsu(v_tempBluredAss##i, (vector signed short)zero);  \
962 
-  const vector unsigned char vg##src##i =                       \
963 
-    vec_perm(vf##src##i, v_##src##A##i, permHH);                \
964 
-  const vector unsigned char mask##src##i =                     \
965 
-    vec_perm((vector unsigned char)zero, (vector unsigned char)neg1, perms##src##i); \
966 
-  const vector unsigned char vg2##src##i =                      \
967 
-    vec_perm(vg##src##i, vg##src##i, perms##src##i);            \
968 
-  const vector unsigned char svA##src##i =                      \
969 
-    vec_sel(v_##src##A1##i, vg2##src##i, mask##src##i);         \
970 
-  const vector unsigned char svB##src##i =                      \
971 
-    vec_sel(vg2##src##i, v_##src##A2##i, mask##src##i);         \
972 
-  vec_st(svA##src##i, i * stride, src);                         \
973 
-  vec_st(svB##src##i, i * stride + 16, src)
974 
-
975 
-  PACK_AND_STORE(src, 0);
976 
-  PACK_AND_STORE(src, 1);
977 
-  PACK_AND_STORE(src, 2);
978 
-  PACK_AND_STORE(src, 3);
979 
-  PACK_AND_STORE(src, 4);
980 
-  PACK_AND_STORE(src, 5);
981 
-  PACK_AND_STORE(src, 6);
982 
-  PACK_AND_STORE(src, 7);
983 
-  PACK_AND_STORE(tempBlured, 0);
984 
-  PACK_AND_STORE(tempBlured, 1);
985 
-  PACK_AND_STORE(tempBlured, 2);
986 
-  PACK_AND_STORE(tempBlured, 3);
987 
-  PACK_AND_STORE(tempBlured, 4);
988 
-  PACK_AND_STORE(tempBlured, 5);
989 
-  PACK_AND_STORE(tempBlured, 6);
990 
-  PACK_AND_STORE(tempBlured, 7);
958 
+    const vector unsigned char perms##src##i =                  \
959 
+        vec_lvsr(i * stride, src);                              \
960 
+    const vector unsigned char vf##src##i =                     \
961 
+        vec_packsu(v_tempBluredAss##i, (vector signed short)zero); \
962 
+    const vector unsigned char vg##src##i =                     \
963 
+        vec_perm(vf##src##i, v_##src##A##i, permHH);            \
964 
+    const vector unsigned char mask##src##i =                   \
965 
+        vec_perm((vector unsigned char)zero, (vector unsigned char)neg1, perms##src##i); \
966 
+    const vector unsigned char vg2##src##i =                    \
967 
+        vec_perm(vg##src##i, vg##src##i, perms##src##i);        \
968 
+    const vector unsigned char svA##src##i =                    \
969 
+        vec_sel(v_##src##A1##i, vg2##src##i, mask##src##i);     \
970 
+    const vector unsigned char svB##src##i =                    \
971 
+        vec_sel(vg2##src##i, v_##src##A2##i, mask##src##i);     \
972 
+    vec_st(svA##src##i, i * stride, src);                       \
973 
+    vec_st(svB##src##i, i * stride + 16, src)
974 
+
975 
+    PACK_AND_STORE(src, 0);
976 
+    PACK_AND_STORE(src, 1);
977 
+    PACK_AND_STORE(src, 2);
978 
+    PACK_AND_STORE(src, 3);
979 
+    PACK_AND_STORE(src, 4);
980 
+    PACK_AND_STORE(src, 5);
981 
+    PACK_AND_STORE(src, 6);
982 
+    PACK_AND_STORE(src, 7);
983 
+    PACK_AND_STORE(tempBlured, 0);
984 
+    PACK_AND_STORE(tempBlured, 1);
985 
+    PACK_AND_STORE(tempBlured, 2);
986 
+    PACK_AND_STORE(tempBlured, 3);
987 
+    PACK_AND_STORE(tempBlured, 4);
988 
+    PACK_AND_STORE(tempBlured, 5);
989 
+    PACK_AND_STORE(tempBlured, 6);
990 
+    PACK_AND_STORE(tempBlured, 7);
991 991 
 #undef PACK_AND_STORE
992 992 
 }
993 993
994 994 
 static inline void transpose_16x8_char_toPackedAlign_altivec(unsigned char* dst, unsigned char* src, int stride) {
995 
-  const vector unsigned char zero = vec_splat_u8(0);
995 
+    const vector unsigned char zero = vec_splat_u8(0);
996 996
997 997 
 #define LOAD_DOUBLE_LINE(i, j)                                          \
998 
-  vector unsigned char perm1##i = vec_lvsl(i * stride, src);            \
999 
-  vector unsigned char perm2##i = vec_lvsl(j * stride, src);            \
1000 
-  vector unsigned char srcA##i = vec_ld(i * stride, src);               \
1001 
-  vector unsigned char srcB##i = vec_ld(i * stride + 16, src);          \
1002 
-  vector unsigned char srcC##i = vec_ld(j * stride, src);               \
1003 
-  vector unsigned char srcD##i = vec_ld(j * stride+ 16, src);           \
1004 
-  vector unsigned char src##i = vec_perm(srcA##i, srcB##i, perm1##i);   \
1005 
-  vector unsigned char src##j = vec_perm(srcC##i, srcD##i, perm2##i)
1006 
-
1007 
-  LOAD_DOUBLE_LINE(0, 1);
1008 
-  LOAD_DOUBLE_LINE(2, 3);
1009 
-  LOAD_DOUBLE_LINE(4, 5);
1010 
-  LOAD_DOUBLE_LINE(6, 7);
998 
+    vector unsigned char perm1##i = vec_lvsl(i * stride, src);          \
999 
+    vector unsigned char perm2##i = vec_lvsl(j * stride, src);          \
1000 
+    vector unsigned char srcA##i = vec_ld(i * stride, src);             \
1001 
+    vector unsigned char srcB##i = vec_ld(i * stride + 16, src);        \
1002 
+    vector unsigned char srcC##i = vec_ld(j * stride, src);             \
1003 
+    vector unsigned char srcD##i = vec_ld(j * stride+ 16, src);         \
1004 
+    vector unsigned char src##i = vec_perm(srcA##i, srcB##i, perm1##i); \
1005 
+    vector unsigned char src##j = vec_perm(srcC##i, srcD##i, perm2##i)
1006 
+
1007 
+    LOAD_DOUBLE_LINE(0, 1);
1008 
+    LOAD_DOUBLE_LINE(2, 3);
1009 
+    LOAD_DOUBLE_LINE(4, 5);
1010 
+    LOAD_DOUBLE_LINE(6, 7);
1011 1011 
 #undef LOAD_DOUBLE_LINE
1012 1012
1013 
-  vector unsigned char tempA = vec_mergeh(src0, zero);
1014 
-  vector unsigned char tempB = vec_mergel(src0, zero);
1015 
-  vector unsigned char tempC = vec_mergeh(src1, zero);
1016 
-  vector unsigned char tempD = vec_mergel(src1, zero);
1017 
-  vector unsigned char tempE = vec_mergeh(src2, zero);
1018 
-  vector unsigned char tempF = vec_mergel(src2, zero);
1019 
-  vector unsigned char tempG = vec_mergeh(src3, zero);
1020 
-  vector unsigned char tempH = vec_mergel(src3, zero);
1021 
-  vector unsigned char tempI = vec_mergeh(src4, zero);
1022 
-  vector unsigned char tempJ = vec_mergel(src4, zero);
1023 
-  vector unsigned char tempK = vec_mergeh(src5, zero);
1024 
-  vector unsigned char tempL = vec_mergel(src5, zero);
1025 
-  vector unsigned char tempM = vec_mergeh(src6, zero);
1026 
-  vector unsigned char tempN = vec_mergel(src6, zero);
1027 
-  vector unsigned char tempO = vec_mergeh(src7, zero);
1028 
-  vector unsigned char tempP = vec_mergel(src7, zero);
1029 
-
1030 
-  vector unsigned char temp0 = vec_mergeh(tempA, tempI);
1031 
-  vector unsigned char temp1 = vec_mergel(tempA, tempI);
1032 
-  vector unsigned char temp2 = vec_mergeh(tempB, tempJ);
1033 
-  vector unsigned char temp3 = vec_mergel(tempB, tempJ);
1034 
-  vector unsigned char temp4 = vec_mergeh(tempC, tempK);
1035 
-  vector unsigned char temp5 = vec_mergel(tempC, tempK);
1036 
-  vector unsigned char temp6 = vec_mergeh(tempD, tempL);
1037 
-  vector unsigned char temp7 = vec_mergel(tempD, tempL);
1038 
-  vector unsigned char temp8 = vec_mergeh(tempE, tempM);
1039 
-  vector unsigned char temp9 = vec_mergel(tempE, tempM);
1040 
-  vector unsigned char temp10 = vec_mergeh(tempF, tempN);
1041 
-  vector unsigned char temp11 = vec_mergel(tempF, tempN);
1042 
-  vector unsigned char temp12 = vec_mergeh(tempG, tempO);
1043 
-  vector unsigned char temp13 = vec_mergel(tempG, tempO);
1044 
-  vector unsigned char temp14 = vec_mergeh(tempH, tempP);
1045 
-  vector unsigned char temp15 = vec_mergel(tempH, tempP);
1046 
-
1047 
-  tempA = vec_mergeh(temp0, temp8);
1048 
-  tempB = vec_mergel(temp0, temp8);
1049 
-  tempC = vec_mergeh(temp1, temp9);
1050 
-  tempD = vec_mergel(temp1, temp9);
1051 
-  tempE = vec_mergeh(temp2, temp10);
1052 
-  tempF = vec_mergel(temp2, temp10);
1053 
-  tempG = vec_mergeh(temp3, temp11);
1054 
-  tempH = vec_mergel(temp3, temp11);
1055 
-  tempI = vec_mergeh(temp4, temp12);
1056 
-  tempJ = vec_mergel(temp4, temp12);
1057 
-  tempK = vec_mergeh(temp5, temp13);
1058 
-  tempL = vec_mergel(temp5, temp13);
1059 
-  tempM = vec_mergeh(temp6, temp14);
1060 
-  tempN = vec_mergel(temp6, temp14);
1061 
-  tempO = vec_mergeh(temp7, temp15);
1062 
-  tempP = vec_mergel(temp7, temp15);
1063 
-
1064 
-  temp0 = vec_mergeh(tempA, tempI);
1065 
-  temp1 = vec_mergel(tempA, tempI);
1066 
-  temp2 = vec_mergeh(tempB, tempJ);
1067 
-  temp3 = vec_mergel(tempB, tempJ);
1068 
-  temp4 = vec_mergeh(tempC, tempK);
1069 
-  temp5 = vec_mergel(tempC, tempK);
1070 
-  temp6 = vec_mergeh(tempD, tempL);
1071 
-  temp7 = vec_mergel(tempD, tempL);
1072 
-  temp8 = vec_mergeh(tempE, tempM);
1073 
-  temp9 = vec_mergel(tempE, tempM);
1074 
-  temp10 = vec_mergeh(tempF, tempN);
1075 
-  temp11 = vec_mergel(tempF, tempN);
1076 
-  temp12 = vec_mergeh(tempG, tempO);
1077 
-  temp13 = vec_mergel(tempG, tempO);
1078 
-  temp14 = vec_mergeh(tempH, tempP);
1079 
-  temp15 = vec_mergel(tempH, tempP);
1080 
-
1081 
-  vec_st(temp0, 0, dst);
1082 
-  vec_st(temp1, 16, dst);
1083 
-  vec_st(temp2, 32, dst);
1084 
-  vec_st(temp3, 48, dst);
1085 
-  vec_st(temp4, 64, dst);
1086 
-  vec_st(temp5, 80, dst);
1087 
-  vec_st(temp6, 96, dst);
1088 
-  vec_st(temp7, 112, dst);
1089 
-  vec_st(temp8, 128, dst);
1090 
-  vec_st(temp9, 144, dst);
1091 
-  vec_st(temp10, 160, dst);
1092 
-  vec_st(temp11, 176, dst);
1093 
-  vec_st(temp12, 192, dst);
1094 
-  vec_st(temp13, 208, dst);
1095 
-  vec_st(temp14, 224, dst);
1096 
-  vec_st(temp15, 240, dst);
1013 
+    vector unsigned char tempA = vec_mergeh(src0, zero);
1014 
+    vector unsigned char tempB = vec_mergel(src0, zero);
1015 
+    vector unsigned char tempC = vec_mergeh(src1, zero);
1016 
+    vector unsigned char tempD = vec_mergel(src1, zero);
1017 
+    vector unsigned char tempE = vec_mergeh(src2, zero);
1018 
+    vector unsigned char tempF = vec_mergel(src2, zero);
1019 
+    vector unsigned char tempG = vec_mergeh(src3, zero);
1020 
+    vector unsigned char tempH = vec_mergel(src3, zero);
1021 
+    vector unsigned char tempI = vec_mergeh(src4, zero);
1022 
+    vector unsigned char tempJ = vec_mergel(src4, zero);
1023 
+    vector unsigned char tempK = vec_mergeh(src5, zero);
1024 
+    vector unsigned char tempL = vec_mergel(src5, zero);
1025 
+    vector unsigned char tempM = vec_mergeh(src6, zero);
1026 
+    vector unsigned char tempN = vec_mergel(src6, zero);
1027 
+    vector unsigned char tempO = vec_mergeh(src7, zero);
1028 
+    vector unsigned char tempP = vec_mergel(src7, zero);
1029 
+
1030 
+    vector unsigned char temp0 = vec_mergeh(tempA, tempI);
1031 
+    vector unsigned char temp1 = vec_mergel(tempA, tempI);
1032 
+    vector unsigned char temp2 = vec_mergeh(tempB, tempJ);
1033 
+    vector unsigned char temp3 = vec_mergel(tempB, tempJ);
1034 
+    vector unsigned char temp4 = vec_mergeh(tempC, tempK);
1035 
+    vector unsigned char temp5 = vec_mergel(tempC, tempK);
1036 
+    vector unsigned char temp6 = vec_mergeh(tempD, tempL);
1037 
+    vector unsigned char temp7 = vec_mergel(tempD, tempL);
1038 
+    vector unsigned char temp8 = vec_mergeh(tempE, tempM);
1039 
+    vector unsigned char temp9 = vec_mergel(tempE, tempM);
1040 
+    vector unsigned char temp10 = vec_mergeh(tempF, tempN);
1041 
+    vector unsigned char temp11 = vec_mergel(tempF, tempN);
1042 
+    vector unsigned char temp12 = vec_mergeh(tempG, tempO);
1043 
+    vector unsigned char temp13 = vec_mergel(tempG, tempO);
1044 
+    vector unsigned char temp14 = vec_mergeh(tempH, tempP);
1045 
+    vector unsigned char temp15 = vec_mergel(tempH, tempP);
1046 
+
1047 
+    tempA = vec_mergeh(temp0, temp8);
1048 
+    tempB = vec_mergel(temp0, temp8);
1049 
+    tempC = vec_mergeh(temp1, temp9);
1050 
+    tempD = vec_mergel(temp1, temp9);
1051 
+    tempE = vec_mergeh(temp2, temp10);
1052 
+    tempF = vec_mergel(temp2, temp10);
1053 
+    tempG = vec_mergeh(temp3, temp11);
1054 
+    tempH = vec_mergel(temp3, temp11);
1055 
+    tempI = vec_mergeh(temp4, temp12);
1056 
+    tempJ = vec_mergel(temp4, temp12);
1057 
+    tempK = vec_mergeh(temp5, temp13);
1058 
+    tempL = vec_mergel(temp5, temp13);
1059 
+    tempM = vec_mergeh(temp6, temp14);
1060 
+    tempN = vec_mergel(temp6, temp14);
1061 
+    tempO = vec_mergeh(temp7, temp15);
1062 
+    tempP = vec_mergel(temp7, temp15);
1063 
+
1064 
+    temp0 = vec_mergeh(tempA, tempI);
1065 
+    temp1 = vec_mergel(tempA, tempI);
1066 
+    temp2 = vec_mergeh(tempB, tempJ);
1067 
+    temp3 = vec_mergel(tempB, tempJ);
1068 
+    temp4 = vec_mergeh(tempC, tempK);
1069 
+    temp5 = vec_mergel(tempC, tempK);
1070 
+    temp6 = vec_mergeh(tempD, tempL);
1071 
+    temp7 = vec_mergel(tempD, tempL);
1072 
+    temp8 = vec_mergeh(tempE, tempM);
1073 
+    temp9 = vec_mergel(tempE, tempM);
1074 
+    temp10 = vec_mergeh(tempF, tempN);
1075 
+    temp11 = vec_mergel(tempF, tempN);
1076 
+    temp12 = vec_mergeh(tempG, tempO);
1077 
+    temp13 = vec_mergel(tempG, tempO);
1078 
+    temp14 = vec_mergeh(tempH, tempP);
1079 
+    temp15 = vec_mergel(tempH, tempP);
1080 
+
1081 
+    vec_st(temp0, 0, dst);
1082 
+    vec_st(temp1, 16, dst);
1083 
+    vec_st(temp2, 32, dst);
1084 
+    vec_st(temp3, 48, dst);
1085 
+    vec_st(temp4, 64, dst);
1086 
+    vec_st(temp5, 80, dst);
1087 
+    vec_st(temp6, 96, dst);
1088 
+    vec_st(temp7, 112, dst);
1089 
+    vec_st(temp8, 128, dst);
1090 
+    vec_st(temp9, 144, dst);
1091 
+    vec_st(temp10, 160, dst);
1092 
+    vec_st(temp11, 176, dst);
1093 
+    vec_st(temp12, 192, dst);
1094 
+    vec_st(temp13, 208, dst);
1095 
+    vec_st(temp14, 224, dst);
1096 
+    vec_st(temp15, 240, dst);
1097 1097 
 }
1098 1098
1099 1099 
 static inline void transpose_8x16_char_fromPackedAlign_altivec(unsigned char* dst, unsigned char* src, int stride) {
1100 
-  const vector unsigned char zero = vec_splat_u8(0);
1100 
+    const vector unsigned char zero = vec_splat_u8(0);
1101 1101
1102 1102 
 #define LOAD_DOUBLE_LINE(i, j)                                  \
1103 
-  vector unsigned char src##i = vec_ld(i * 16, src);            \
1104 
-  vector unsigned char src##j = vec_ld(j * 16, src)
1105 
-
1106 
-  LOAD_DOUBLE_LINE(0, 1);
1107 
-  LOAD_DOUBLE_LINE(2, 3);
1108 
-  LOAD_DOUBLE_LINE(4, 5);
1109 
-  LOAD_DOUBLE_LINE(6, 7);
1110 
-  LOAD_DOUBLE_LINE(8, 9);
1111 
-  LOAD_DOUBLE_LINE(10, 11);
1112 
-  LOAD_DOUBLE_LINE(12, 13);
1113 
-  LOAD_DOUBLE_LINE(14, 15);
1103 
+    vector unsigned char src##i = vec_ld(i * 16, src);            \
1104 
+    vector unsigned char src##j = vec_ld(j * 16, src)
1105 
+
1106 
+    LOAD_DOUBLE_LINE(0, 1);
1107 
+    LOAD_DOUBLE_LINE(2, 3);
1108 
+    LOAD_DOUBLE_LINE(4, 5);
1109 
+    LOAD_DOUBLE_LINE(6, 7);
1110 
+    LOAD_DOUBLE_LINE(8, 9);
1111 
+    LOAD_DOUBLE_LINE(10, 11);
1112 
+    LOAD_DOUBLE_LINE(12, 13);
1113 
+    LOAD_DOUBLE_LINE(14, 15);
1114 1114 
 #undef LOAD_DOUBLE_LINE
1115 1115
1116 
-  vector unsigned char tempA = vec_mergeh(src0, src8);
1117 
-  vector unsigned char tempB;
1118 
-  vector unsigned char tempC = vec_mergeh(src1, src9);
1119 
-  vector unsigned char tempD;
1120 
-  vector unsigned char tempE = vec_mergeh(src2, src10);
1121 
-  vector unsigned char tempG = vec_mergeh(src3, src11);
1122 
-  vector unsigned char tempI = vec_mergeh(src4, src12);
1123 
-  vector unsigned char tempJ;
1124 
-  vector unsigned char tempK = vec_mergeh(src5, src13);
1125 
-  vector unsigned char tempL;
1126 
-  vector unsigned char tempM = vec_mergeh(src6, src14);
1127 
-  vector unsigned char tempO = vec_mergeh(src7, src15);
1128 
-
1129 
-  vector unsigned char temp0 = vec_mergeh(tempA, tempI);
1130 
-  vector unsigned char temp1 = vec_mergel(tempA, tempI);
1131 
-  vector unsigned char temp2;
1132 
-  vector unsigned char temp3;
1133 
-  vector unsigned char temp4 = vec_mergeh(tempC, tempK);
1134 
-  vector unsigned char temp5 = vec_mergel(tempC, tempK);
1135 
-  vector unsigned char temp6;
1136 
-  vector unsigned char temp7;
1137 
-  vector unsigned char temp8 = vec_mergeh(tempE, tempM);
1138 
-  vector unsigned char temp9 = vec_mergel(tempE, tempM);
1139 
-  vector unsigned char temp12 = vec_mergeh(tempG, tempO);
1140 
-  vector unsigned char temp13 = vec_mergel(tempG, tempO);
1141 
-
1142 
-  tempA = vec_mergeh(temp0, temp8);
1143 
-  tempB = vec_mergel(temp0, temp8);
1144 
-  tempC = vec_mergeh(temp1, temp9);
1145 
-  tempD = vec_mergel(temp1, temp9);
1146 
-  tempI = vec_mergeh(temp4, temp12);
1147 
-  tempJ = vec_mergel(temp4, temp12);
1148 
-  tempK = vec_mergeh(temp5, temp13);
1149 
-  tempL = vec_mergel(temp5, temp13);
1150 
-
1151 
-  temp0 = vec_mergeh(tempA, tempI);
1152 
-  temp1 = vec_mergel(tempA, tempI);
1153 
-  temp2 = vec_mergeh(tempB, tempJ);
1154 
-  temp3 = vec_mergel(tempB, tempJ);
1155 
-  temp4 = vec_mergeh(tempC, tempK);
1156 
-  temp5 = vec_mergel(tempC, tempK);
1157 
-  temp6 = vec_mergeh(tempD, tempL);
1158 
-  temp7 = vec_mergel(tempD, tempL);
1159 
-
1160 
-
1161 
-  const vector signed char neg1 = vec_splat_s8(-1);
1116 
+    vector unsigned char tempA = vec_mergeh(src0, src8);
1117 
+    vector unsigned char tempB;
1118 
+    vector unsigned char tempC = vec_mergeh(src1, src9);
1119 
+    vector unsigned char tempD;
1120 
+    vector unsigned char tempE = vec_mergeh(src2, src10);
1121 
+    vector unsigned char tempG = vec_mergeh(src3, src11);
1122 
+    vector unsigned char tempI = vec_mergeh(src4, src12);
1123 
+    vector unsigned char tempJ;
1124 
+    vector unsigned char tempK = vec_mergeh(src5, src13);
1125 
+    vector unsigned char tempL;
1126 
+    vector unsigned char tempM = vec_mergeh(src6, src14);
1127 
+    vector unsigned char tempO = vec_mergeh(src7, src15);
1128 
+
1129 
+    vector unsigned char temp0 = vec_mergeh(tempA, tempI);
1130 
+    vector unsigned char temp1 = vec_mergel(tempA, tempI);
1131 
+    vector unsigned char temp2;
1132 
+    vector unsigned char temp3;
1133 
+    vector unsigned char temp4 = vec_mergeh(tempC, tempK);
1134 
+    vector unsigned char temp5 = vec_mergel(tempC, tempK);
1135 
+    vector unsigned char temp6;
1136 
+    vector unsigned char temp7;
1137 
+    vector unsigned char temp8 = vec_mergeh(tempE, tempM);
1138 
+    vector unsigned char temp9 = vec_mergel(tempE, tempM);
1139 
+    vector unsigned char temp12 = vec_mergeh(tempG, tempO);
1140 
+    vector unsigned char temp13 = vec_mergel(tempG, tempO);
1141 
+
1142 
+    tempA = vec_mergeh(temp0, temp8);
1143 
+    tempB = vec_mergel(temp0, temp8);
1144 
+    tempC = vec_mergeh(temp1, temp9);
1145 
+    tempD = vec_mergel(temp1, temp9);
1146 
+    tempI = vec_mergeh(temp4, temp12);
1147 
+    tempJ = vec_mergel(temp4, temp12);
1148 
+    tempK = vec_mergeh(temp5, temp13);
1149 
+    tempL = vec_mergel(temp5, temp13);
1150 
+
1151 
+    temp0 = vec_mergeh(tempA, tempI);
1152 
+    temp1 = vec_mergel(tempA, tempI);
1153 
+    temp2 = vec_mergeh(tempB, tempJ);
1154 
+    temp3 = vec_mergel(tempB, tempJ);
1155 
+    temp4 = vec_mergeh(tempC, tempK);
1156 
+    temp5 = vec_mergel(tempC, tempK);
1157 
+    temp6 = vec_mergeh(tempD, tempL);
1158 
+    temp7 = vec_mergel(tempD, tempL);
1159 
+
1160 
+
1161 
+    const vector signed char neg1 = vec_splat_s8(-1);
1162 1162 
 #define STORE_DOUBLE_LINE(i, j)                                         \
1163 
-  vector unsigned char dstA##i = vec_ld(i * stride, dst);               \
1164 
-  vector unsigned char dstB##i = vec_ld(i * stride + 16, dst);          \
1165 
-  vector unsigned char dstA##j = vec_ld(j * stride, dst);               \
1166 
-  vector unsigned char dstB##j = vec_ld(j * stride+ 16, dst);           \
1167 
-  vector unsigned char align##i = vec_lvsr(i * stride, dst);            \
1168 
-  vector unsigned char align##j = vec_lvsr(j * stride, dst);            \
1169 
-  vector unsigned char mask##i = vec_perm(zero, (vector unsigned char)neg1, align##i); \
1170 
-  vector unsigned char mask##j = vec_perm(zero, (vector unsigned char)neg1, align##j); \
1171 
-  vector unsigned char dstR##i = vec_perm(temp##i, temp##i, align##i);  \
1172 
-  vector unsigned char dstR##j = vec_perm(temp##j, temp##j, align##j);  \
1173 
-  vector unsigned char dstAF##i = vec_sel(dstA##i, dstR##i, mask##i);   \
1174 
-  vector unsigned char dstBF##i = vec_sel(dstR##i, dstB##i, mask##i);   \
1175 
-  vector unsigned char dstAF##j = vec_sel(dstA##j, dstR##j, mask##j);   \
1176 
-  vector unsigned char dstBF##j = vec_sel(dstR##j, dstB##j, mask##j);   \
1177 
-  vec_st(dstAF##i, i * stride, dst);                                    \
1178 
-  vec_st(dstBF##i, i * stride + 16, dst);                               \
1179 
-  vec_st(dstAF##j, j * stride, dst);                                    \
1180 
-  vec_st(dstBF##j, j * stride + 16, dst)
1181 
-
1182 
-  STORE_DOUBLE_LINE(0,1);
1183 
-  STORE_DOUBLE_LINE(2,3);
1184 
-  STORE_DOUBLE_LINE(4,5);
1185 
-  STORE_DOUBLE_LINE(6,7);
1163 
+    vector unsigned char dstA##i = vec_ld(i * stride, dst);             \
1164 
+    vector unsigned char dstB##i = vec_ld(i * stride + 16, dst);        \
1165 
+    vector unsigned char dstA##j = vec_ld(j * stride, dst);             \
1166 
+    vector unsigned char dstB##j = vec_ld(j * stride+ 16, dst);         \
1167 
+    vector unsigned char align##i = vec_lvsr(i * stride, dst);          \
1168 
+    vector unsigned char align##j = vec_lvsr(j * stride, dst);          \
1169 
+    vector unsigned char mask##i = vec_perm(zero, (vector unsigned char)neg1, align##i); \
1170 
+    vector unsigned char mask##j = vec_perm(zero, (vector unsigned char)neg1, align##j); \
1171 
+    vector unsigned char dstR##i = vec_perm(temp##i, temp##i, align##i);\
1172 
+    vector unsigned char dstR##j = vec_perm(temp##j, temp##j, align##j);\
1173 
+    vector unsigned char dstAF##i = vec_sel(dstA##i, dstR##i, mask##i); \
1174 
+    vector unsigned char dstBF##i = vec_sel(dstR##i, dstB##i, mask##i); \
1175 
+    vector unsigned char dstAF##j = vec_sel(dstA##j, dstR##j, mask##j); \
1176 
+    vector unsigned char dstBF##j = vec_sel(dstR##j, dstB##j, mask##j); \
1177 
+    vec_st(dstAF##i, i * stride, dst);                                  \
1178 
+    vec_st(dstBF##i, i * stride + 16, dst);                             \
1179 
+    vec_st(dstAF##j, j * stride, dst);                                  \
1180 
+    vec_st(dstBF##j, j * stride + 16, dst)
1181 
+
1182 
+    STORE_DOUBLE_LINE(0,1);
1183 
+    STORE_DOUBLE_LINE(2,3);
1184 
+    STORE_DOUBLE_LINE(4,5);
1185 
+    STORE_DOUBLE_LINE(6,7);
1186 1186 
 }
libpostproc/postprocess_internal.h
History View file @ 16e0bf7
... ... 
@@ -55,13 +55,13 @@
55 55 
 #define FULL_Y_RANGE    0x8000                  // 32768
56 56
57 57 
 //Deinterlacing Filters
58 
-#define        LINEAR_IPOL_DEINT_FILTER         0x10000 // 65536
59 
-#define        LINEAR_BLEND_DEINT_FILTER        0x20000 // 131072
60 
-#define        CUBIC_BLEND_DEINT_FILTER         0x8000  // (not implemented yet)
61 
-#define        CUBIC_IPOL_DEINT_FILTER          0x40000 // 262144
62 
-#define        MEDIAN_DEINT_FILTER              0x80000 // 524288
63 
-#define        FFMPEG_DEINT_FILTER              0x400000
64 
-#define        LOWPASS5_DEINT_FILTER            0x800000
58 
+#define LINEAR_IPOL_DEINT_FILTER        0x10000 // 65536
59 
+#define LINEAR_BLEND_DEINT_FILTER       0x20000 // 131072
60 
+#define CUBIC_BLEND_DEINT_FILTER        0x8000  // (not implemented yet)
61 
+#define CUBIC_IPOL_DEINT_FILTER         0x40000 // 262144
62 
+#define MEDIAN_DEINT_FILTER             0x80000 // 524288
63 
+#define FFMPEG_DEINT_FILTER             0x400000
64 
+#define LOWPASS5_DEINT_FILTER           0x800000
65 65
66 66 
 #define TEMP_NOISE_FILTER               0x100000
67 67 
 #define FORCE_QUANT                     0x200000
... ... 
@@ -73,106 +73,105 @@
73 73 
 //#define COMPILE_TIME_MODE 0x77
74 74
75 75 
 static inline int CLIP(int a){
76 
-        if(a&256) return ((a)>>31)^(-1);
77 
-        else      return a;
76 
+    if(a&256) return ((a)>>31)^(-1);
77 
+    else      return a;
78 78 
 }
79 79 
 /**
80 80 
  * Postprocessng filter.
81 81 
  */
82 82 
 struct PPFilter{
83 
-        const char *shortName;
84 
-        const char *longName;
85 
-        int chromDefault;       ///< is chrominance filtering on by default if this filter is manually activated
86 
-        int minLumQuality;      ///< minimum quality to turn luminance filtering on
87 
-        int minChromQuality;    ///< minimum quality to turn chrominance filtering on
88 
-        int mask;               ///< Bitmask to turn this filter on
83 
+    const char *shortName;
84 
+    const char *longName;
85 
+    int chromDefault;       ///< is chrominance filtering on by default if this filter is manually activated
86 
+    int minLumQuality;      ///< minimum quality to turn luminance filtering on
87 
+    int minChromQuality;    ///< minimum quality to turn chrominance filtering on
88 
+    int mask;               ///< Bitmask to turn this filter on
89 89 
 };
90 90
91 91 
 /**
92 92 
  * Postprocessng mode.
93 93 
  */
94 94 
 typedef struct PPMode{
95 
-        int lumMode;                    ///< acivates filters for luminance
96 
-        int chromMode;                  ///< acivates filters for chrominance
97 
-        int error;                      ///< non zero on error
95 
+    int lumMode;                    ///< acivates filters for luminance
96 
+    int chromMode;                  ///< acivates filters for chrominance
97 
+    int error;                      ///< non zero on error
98 98
99 
-        int minAllowedY;                ///< for brigtness correction
100 
-        int maxAllowedY;                ///< for brihtness correction
101 
-        float maxClippedThreshold;      ///< amount of "black" u r willing to loose to get a brightness corrected picture
99 
+    int minAllowedY;                ///< for brigtness correction
100 
+    int maxAllowedY;                ///< for brihtness correction
101 
+    float maxClippedThreshold;      ///< amount of "black" u r willing to loose to get a brightness corrected picture
102 102
103 
-        int maxTmpNoise[3];             ///< for Temporal Noise Reducing filter (Maximal sum of abs differences)
103 
+    int maxTmpNoise[3];             ///< for Temporal Noise Reducing filter (Maximal sum of abs differences)
104 104
105 
-        int baseDcDiff;
106 
-        int flatnessThreshold;
105 
+    int baseDcDiff;
106 
+    int flatnessThreshold;
107 107
108 
-        int forcedQuant;                ///< quantizer if FORCE_QUANT is used
108 
+    int forcedQuant;                ///< quantizer if FORCE_QUANT is used
109 109 
 } PPMode;
110 110
111 111 
 /**
112 112 
  * postprocess context.
113 113 
  */
114 114 
 typedef struct PPContext{
115 
-        /**
116 
-         * info on struct for av_log
117 
-         */
118 
-        const AVClass *av_class;
115 
+    /**
116 
+     * info on struct for av_log
117 
+     */
118 
+    const AVClass *av_class;
119 119
120 
-        uint8_t *tempBlocks; ///<used for the horizontal code
120 
+    uint8_t *tempBlocks; ///<used for the horizontal code
121 121
122 
-        /**
123 
-         * luma histogram.
124 
-         * we need 64bit here otherwise we'll going to have a problem
125 
-         * after watching a black picture for 5 hours
126 
-         */
127 
-        uint64_t *yHistogram;
122 
+    /**
123 
+     * luma histogram.
124 
+     * we need 64bit here otherwise we'll going to have a problem
125 
+     * after watching a black picture for 5 hours
126 
+     */
127 
+    uint64_t *yHistogram;
128 128
129 
-        DECLARE_ALIGNED(8, uint64_t, packedYOffset);
130 
-        DECLARE_ALIGNED(8, uint64_t, packedYScale);
129 
+    DECLARE_ALIGNED(8, uint64_t, packedYOffset);
130 
+    DECLARE_ALIGNED(8, uint64_t, packedYScale);
131 131
132 
-        /** Temporal noise reducing buffers */
133 
-        uint8_t *tempBlured[3];
134 
-        int32_t *tempBluredPast[3];
132 
+    /** Temporal noise reducing buffers */
133 
+    uint8_t *tempBlured[3];
134 
+    int32_t *tempBluredPast[3];
135 135
136 
-        /** Temporary buffers for handling the last row(s) */
137 
-        uint8_t *tempDst;
138 
-        uint8_t *tempSrc;
136 
+    /** Temporary buffers for handling the last row(s) */
137 
+    uint8_t *tempDst;
138 
+    uint8_t *tempSrc;
139 139
140 
-        uint8_t *deintTemp;
140 
+    uint8_t *deintTemp;
141 141
142 
-        DECLARE_ALIGNED(8, uint64_t, pQPb);
143 
-        DECLARE_ALIGNED(8, uint64_t, pQPb2);
142 
+    DECLARE_ALIGNED(8, uint64_t, pQPb);
143 
+    DECLARE_ALIGNED(8, uint64_t, pQPb2);
144 144
145 
-        DECLARE_ALIGNED(8, uint64_t, mmxDcOffset[64]);
146 
-        DECLARE_ALIGNED(8, uint64_t, mmxDcThreshold[64]);
145 
+    DECLARE_ALIGNED(8, uint64_t, mmxDcOffset[64]);
146 
+    DECLARE_ALIGNED(8, uint64_t, mmxDcThreshold[64]);
147 147
148 
-        QP_STORE_T *stdQPTable;       ///< used to fix MPEG2 style qscale
149 
-        QP_STORE_T *nonBQPTable;
150 
-        QP_STORE_T *forcedQPTable;
148 
+    QP_STORE_T *stdQPTable;       ///< used to fix MPEG2 style qscale
149 
+    QP_STORE_T *nonBQPTable;
150 
+    QP_STORE_T *forcedQPTable;
151 151
152 
-        int QP;
153 
-        int nonBQP;
152 
+    int QP;
153 
+    int nonBQP;
154 154
155 
-        int frameNum;
155 
+    int frameNum;
156 156
157 
-        int cpuCaps;
157 
+    int cpuCaps;
158 158
159 
-        int qpStride; ///<size of qp buffers (needed to realloc them if needed)
160 
-        int stride;   ///<size of some buffers (needed to realloc them if needed)
159 
+    int qpStride; ///<size of qp buffers (needed to realloc them if needed)
160 
+    int stride;   ///<size of some buffers (needed to realloc them if needed)
161 161
162 
-        int hChromaSubSample;
163 
-        int vChromaSubSample;
162 
+    int hChromaSubSample;
163 
+    int vChromaSubSample;
164 164
165 
-        PPMode ppMode;
165 
+    PPMode ppMode;
166 166 
 } PPContext;
167 167
168 168
169 
-static inline void linecpy(void *dest, const void *src, int lines, int stride)
170 
-{
171 
-        if (stride > 0) {
172 
-                memcpy(dest, src, lines*stride);
173 
-        } else {
174 
-                memcpy((uint8_t*)dest+(lines-1)*stride, (const uint8_t*)src+(lines-1)*stride, -lines*stride);
175 
-        }
169 
+static inline void linecpy(void *dest, const void *src, int lines, int stride) {
170 
+    if (stride > 0) {
171 
+        memcpy(dest, src, lines*stride);
172 
+    } else {
173 
+        memcpy((uint8_t*)dest+(lines-1)*stride, (const uint8_t*)src+(lines-1)*stride, -lines*stride);
174 
+    }
176 175 
 }
177 176
178 177 
 #endif /* FFMPEG_POSTPROCESS_INTERNAL_H */
libpostproc/postprocess_template.c
History View file @ 16e0bf7
... ... 
@@ -42,17 +42,17 @@
42 42 
 #define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
43 43 
 #elif defined (HAVE_MMX)
44 44 
 #define PMINUB(b,a,t) \
45 
-        "movq " #a ", " #t " \n\t"\
46 
-        "psubusb " #b ", " #t " \n\t"\
47 
-        "psubb " #t ", " #a " \n\t"
45 
+    "movq " #a ", " #t " \n\t"\
46 
+    "psubusb " #b ", " #t " \n\t"\
47 
+    "psubb " #t ", " #a " \n\t"
48 48 
 #endif
49 49
50 50 
 #ifdef HAVE_MMX2
51 51 
 #define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
52 52 
 #elif defined (HAVE_MMX)
53 53 
 #define PMAXUB(a,b) \
54 
-        "psubusb " #a ", " #b " \n\t"\
55 
-        "paddb " #a ", " #b " \n\t"
54 
+    "psubusb " #a ", " #b " \n\t"\
55 
+    "paddb " #a ", " #b " \n\t"
56 56 
 #endif
57 57
58 58 
 //FIXME? |255-0| = 1 (should not be a problem ...)
... ... 
@@ -61,114 +61,114 @@
61 61 
  * Check if the middle 8x8 Block in the given 8x16 block is flat
62 62 
  */
63 63 
 static inline int RENAME(vertClassify)(uint8_t src[], int stride, PPContext *c){
64 
-        int numEq= 0, dcOk;
65 
-        src+= stride*4; // src points to begin of the 8x8 Block
66 
-asm volatile(
67 
-                "movq %0, %%mm7                         \n\t"
68 
-                "movq %1, %%mm6                         \n\t"
69 
-                : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
70 
-                );
71 
-
72 
-asm volatile(
73 
-                "lea (%2, %3), %%"REG_a"                \n\t"
64 
+    int numEq= 0, dcOk;
65 
+    src+= stride*4; // src points to begin of the 8x8 Block
66 
+    asm volatile(
67 
+        "movq %0, %%mm7                         \n\t"
68 
+        "movq %1, %%mm6                         \n\t"
69 
+        : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
70 
+        );
71 
+
72 
+    asm volatile(
73 
+        "lea (%2, %3), %%"REG_a"                \n\t"
74 74 
 //      0       1       2       3       4       5       6       7       8       9
75 75 
 //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2
76 76
77 
-                "movq (%2), %%mm0                       \n\t"
78 
-                "movq (%%"REG_a"), %%mm1                \n\t"
79 
-                "movq %%mm0, %%mm3                      \n\t"
80 
-                "movq %%mm0, %%mm4                      \n\t"
81 
-                PMAXUB(%%mm1, %%mm4)
82 
-                PMINUB(%%mm1, %%mm3, %%mm5)
83 
-                "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
84 
-                "paddb %%mm7, %%mm0                     \n\t"
85 
-                "pcmpgtb %%mm6, %%mm0                   \n\t"
86 
-
87 
-                "movq (%%"REG_a",%3), %%mm2             \n\t"
88 
-                PMAXUB(%%mm2, %%mm4)
89 
-                PMINUB(%%mm2, %%mm3, %%mm5)
90 
-                "psubb %%mm2, %%mm1                     \n\t"
91 
-                "paddb %%mm7, %%mm1                     \n\t"
92 
-                "pcmpgtb %%mm6, %%mm1                   \n\t"
93 
-                "paddb %%mm1, %%mm0                     \n\t"
94 
-
95 
-                "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
96 
-                PMAXUB(%%mm1, %%mm4)
97 
-                PMINUB(%%mm1, %%mm3, %%mm5)
98 
-                "psubb %%mm1, %%mm2                     \n\t"
99 
-                "paddb %%mm7, %%mm2                     \n\t"
100 
-                "pcmpgtb %%mm6, %%mm2                   \n\t"
101 
-                "paddb %%mm2, %%mm0                     \n\t"
102 
-
103 
-                "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
104 
-
105 
-                "movq (%2, %3, 4), %%mm2                \n\t"
106 
-                PMAXUB(%%mm2, %%mm4)
107 
-                PMINUB(%%mm2, %%mm3, %%mm5)
108 
-                "psubb %%mm2, %%mm1                     \n\t"
109 
-                "paddb %%mm7, %%mm1                     \n\t"
110 
-                "pcmpgtb %%mm6, %%mm1                   \n\t"
111 
-                "paddb %%mm1, %%mm0                     \n\t"
112 
-
113 
-                "movq (%%"REG_a"), %%mm1                \n\t"
114 
-                PMAXUB(%%mm1, %%mm4)
115 
-                PMINUB(%%mm1, %%mm3, %%mm5)
116 
-                "psubb %%mm1, %%mm2                     \n\t"
117 
-                "paddb %%mm7, %%mm2                     \n\t"
118 
-                "pcmpgtb %%mm6, %%mm2                   \n\t"
119 
-                "paddb %%mm2, %%mm0                     \n\t"
120 
-
121 
-                "movq (%%"REG_a", %3), %%mm2            \n\t"
122 
-                PMAXUB(%%mm2, %%mm4)
123 
-                PMINUB(%%mm2, %%mm3, %%mm5)
124 
-                "psubb %%mm2, %%mm1                     \n\t"
125 
-                "paddb %%mm7, %%mm1                     \n\t"
126 
-                "pcmpgtb %%mm6, %%mm1                   \n\t"
127 
-                "paddb %%mm1, %%mm0                     \n\t"
128 
-
129 
-                "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
130 
-                PMAXUB(%%mm1, %%mm4)
131 
-                PMINUB(%%mm1, %%mm3, %%mm5)
132 
-                "psubb %%mm1, %%mm2                     \n\t"
133 
-                "paddb %%mm7, %%mm2                     \n\t"
134 
-                "pcmpgtb %%mm6, %%mm2                   \n\t"
135 
-                "paddb %%mm2, %%mm0                     \n\t"
136 
-                "psubusb %%mm3, %%mm4                   \n\t"
137 
-
138 
-                "                                       \n\t"
77 
+        "movq (%2), %%mm0                       \n\t"
78 
+        "movq (%%"REG_a"), %%mm1                \n\t"
79 
+        "movq %%mm0, %%mm3                      \n\t"
80 
+        "movq %%mm0, %%mm4                      \n\t"
81 
+        PMAXUB(%%mm1, %%mm4)
82 
+        PMINUB(%%mm1, %%mm3, %%mm5)
83 
+        "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
84 
+        "paddb %%mm7, %%mm0                     \n\t"
85 
+        "pcmpgtb %%mm6, %%mm0                   \n\t"
86 
+
87 
+        "movq (%%"REG_a",%3), %%mm2             \n\t"
88 
+        PMAXUB(%%mm2, %%mm4)
89 
+        PMINUB(%%mm2, %%mm3, %%mm5)
90 
+        "psubb %%mm2, %%mm1                     \n\t"
91 
+        "paddb %%mm7, %%mm1                     \n\t"
92 
+        "pcmpgtb %%mm6, %%mm1                   \n\t"
93 
+        "paddb %%mm1, %%mm0                     \n\t"
94 
+
95 
+        "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
96 
+        PMAXUB(%%mm1, %%mm4)
97 
+        PMINUB(%%mm1, %%mm3, %%mm5)
98 
+        "psubb %%mm1, %%mm2                     \n\t"
99 
+        "paddb %%mm7, %%mm2                     \n\t"
100 
+        "pcmpgtb %%mm6, %%mm2                   \n\t"
101 
+        "paddb %%mm2, %%mm0                     \n\t"
102 
+
103 
+        "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
104 
+
105 
+        "movq (%2, %3, 4), %%mm2                \n\t"
106 
+        PMAXUB(%%mm2, %%mm4)
107 
+        PMINUB(%%mm2, %%mm3, %%mm5)
108 
+        "psubb %%mm2, %%mm1                     \n\t"
109 
+        "paddb %%mm7, %%mm1                     \n\t"
110 
+        "pcmpgtb %%mm6, %%mm1                   \n\t"
111 
+        "paddb %%mm1, %%mm0                     \n\t"
112 
+
113 
+        "movq (%%"REG_a"), %%mm1                \n\t"
114 
+        PMAXUB(%%mm1, %%mm4)
115 
+        PMINUB(%%mm1, %%mm3, %%mm5)
116 
+        "psubb %%mm1, %%mm2                     \n\t"
117 
+        "paddb %%mm7, %%mm2                     \n\t"
118 
+        "pcmpgtb %%mm6, %%mm2                   \n\t"
119 
+        "paddb %%mm2, %%mm0                     \n\t"
120 
+
121 
+        "movq (%%"REG_a", %3), %%mm2            \n\t"
122 
+        PMAXUB(%%mm2, %%mm4)
123 
+        PMINUB(%%mm2, %%mm3, %%mm5)
124 
+        "psubb %%mm2, %%mm1                     \n\t"
125 
+        "paddb %%mm7, %%mm1                     \n\t"
126 
+        "pcmpgtb %%mm6, %%mm1                   \n\t"
127 
+        "paddb %%mm1, %%mm0                     \n\t"
128 
+
129 
+        "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
130 
+        PMAXUB(%%mm1, %%mm4)
131 
+        PMINUB(%%mm1, %%mm3, %%mm5)
132 
+        "psubb %%mm1, %%mm2                     \n\t"
133 
+        "paddb %%mm7, %%mm2                     \n\t"
134 
+        "pcmpgtb %%mm6, %%mm2                   \n\t"
135 
+        "paddb %%mm2, %%mm0                     \n\t"
136 
+        "psubusb %%mm3, %%mm4                   \n\t"
137 
+
138 
+        "                                       \n\t"
139 139 
 #ifdef HAVE_MMX2
140 
-                "pxor %%mm7, %%mm7                      \n\t"
141 
-                "psadbw %%mm7, %%mm0                    \n\t"
140 
+        "pxor %%mm7, %%mm7                      \n\t"
141 
+        "psadbw %%mm7, %%mm0                    \n\t"
142 142 
 #else
143 
-                "movq %%mm0, %%mm1                      \n\t"
144 
-                "psrlw $8, %%mm0                        \n\t"
145 
-                "paddb %%mm1, %%mm0                     \n\t"
146 
-                "movq %%mm0, %%mm1                      \n\t"
147 
-                "psrlq $16, %%mm0                       \n\t"
148 
-                "paddb %%mm1, %%mm0                     \n\t"
149 
-                "movq %%mm0, %%mm1                      \n\t"
150 
-                "psrlq $32, %%mm0                       \n\t"
151 
-                "paddb %%mm1, %%mm0                     \n\t"
143 
+        "movq %%mm0, %%mm1                      \n\t"
144 
+        "psrlw $8, %%mm0                        \n\t"
145 
+        "paddb %%mm1, %%mm0                     \n\t"
146 
+        "movq %%mm0, %%mm1                      \n\t"
147 
+        "psrlq $16, %%mm0                       \n\t"
148 
+        "paddb %%mm1, %%mm0                     \n\t"
149 
+        "movq %%mm0, %%mm1                      \n\t"
150 
+        "psrlq $32, %%mm0                       \n\t"
151 
+        "paddb %%mm1, %%mm0                     \n\t"
152 152 
 #endif
153 
-                "movq %4, %%mm7                         \n\t" // QP,..., QP
154 
-                "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
155 
-                "psubusb %%mm7, %%mm4                   \n\t" // Diff <= 2QP -> 0
156 
-                "packssdw %%mm4, %%mm4                  \n\t"
157 
-                "movd %%mm0, %0                         \n\t"
158 
-                "movd %%mm4, %1                         \n\t"
159 
-
160 
-                : "=r" (numEq), "=r" (dcOk)
161 
-                : "r" (src), "r" ((long)stride), "m" (c->pQPb)
162 
-                : "%"REG_a
163 
-                );
164 
-
165 
-        numEq= (-numEq) &0xFF;
166 
-        if(numEq > c->ppMode.flatnessThreshold){
167 
-            if(dcOk) return 0;
168 
-            else     return 1;
169 
-        }else{
170 
-            return 2;
171 
-        }
153 
+        "movq %4, %%mm7                         \n\t" // QP,..., QP
154 
+        "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
155 
+        "psubusb %%mm7, %%mm4                   \n\t" // Diff <= 2QP -> 0
156 
+        "packssdw %%mm4, %%mm4                  \n\t"
157 
+        "movd %%mm0, %0                         \n\t"
158 
+        "movd %%mm4, %1                         \n\t"
159 
+
160 
+        : "=r" (numEq), "=r" (dcOk)
161 
+        : "r" (src), "r" ((long)stride), "m" (c->pQPb)
162 
+        : "%"REG_a
163 
+        );
164 
+
165 
+    numEq= (-numEq) &0xFF;
166 
+    if(numEq > c->ppMode.flatnessThreshold){
167 
+        if(dcOk) return 0;
168 
+        else     return 1;
169 
+    }else{
170 
+        return 2;
171 
+    }
172 172 
 }
173 173 
 #endif //HAVE_MMX
174 174
... ... 
@@ -180,172 +180,171 @@ asm volatile(
180 180 
 static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
181 181 
 {
182 182 
 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
183 
-        src+= stride*3;
184 
-        asm volatile(        //"movv %0 %1 %2\n\t"
185 
-                "movq %2, %%mm0                         \n\t"  // QP,..., QP
186 
-                "pxor %%mm4, %%mm4                      \n\t"
187 
-
188 
-                "movq (%0), %%mm6                       \n\t"
189 
-                "movq (%0, %1), %%mm5                   \n\t"
190 
-                "movq %%mm5, %%mm1                      \n\t"
191 
-                "movq %%mm6, %%mm2                      \n\t"
192 
-                "psubusb %%mm6, %%mm5                   \n\t"
193 
-                "psubusb %%mm1, %%mm2                   \n\t"
194 
-                "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
195 
-                "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
196 
-                "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
197 
-
198 
-                "pand %%mm2, %%mm6                      \n\t"
199 
-                "pandn %%mm1, %%mm2                     \n\t"
200 
-                "por %%mm2, %%mm6                       \n\t"// First Line to Filter
201 
-
202 
-                "movq (%0, %1, 8), %%mm5                \n\t"
203 
-                "lea (%0, %1, 4), %%"REG_a"             \n\t"
204 
-                "lea (%0, %1, 8), %%"REG_c"             \n\t"
205 
-                "sub %1, %%"REG_c"                      \n\t"
206 
-                "add %1, %0                             \n\t" // %0 points to line 1 not 0
207 
-                "movq (%0, %1, 8), %%mm7                \n\t"
208 
-                "movq %%mm5, %%mm1                      \n\t"
209 
-                "movq %%mm7, %%mm2                      \n\t"
210 
-                "psubusb %%mm7, %%mm5                   \n\t"
211 
-                "psubusb %%mm1, %%mm2                   \n\t"
212 
-                "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
213 
-                "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
214 
-                "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
215 
-
216 
-                "pand %%mm2, %%mm7                      \n\t"
217 
-                "pandn %%mm1, %%mm2                     \n\t"
218 
-                "por %%mm2, %%mm7                       \n\t" // First Line to Filter
219 
-
220 
-
221 
-                //      1       2       3       4       5       6       7       8
222 
-                //      %0      %0+%1   %0+2%1  eax     %0+4%1  eax+2%1 ecx     eax+4%1
223 
-                // 6 4 2 2 1 1
224 
-                // 6 4 4 2
225 
-                // 6 8 2
226 
-
227 
-                "movq (%0, %1), %%mm0                   \n\t" //  1
228 
-                "movq %%mm0, %%mm1                      \n\t" //  1
229 
-                PAVGB(%%mm6, %%mm0)                           //1 1        /2
230 
-                PAVGB(%%mm6, %%mm0)                           //3 1        /4
231 
-
232 
-                "movq (%0, %1, 4), %%mm2                \n\t" //     1
233 
-                "movq %%mm2, %%mm5                      \n\t" //     1
234 
-                PAVGB((%%REGa), %%mm2)                        //    11        /2
235 
-                PAVGB((%0, %1, 2), %%mm2)                     //   211        /4
236 
-                "movq %%mm2, %%mm3                      \n\t" //   211        /4
237 
-                "movq (%0), %%mm4                       \n\t" // 1
238 
-                PAVGB(%%mm4, %%mm3)                           // 4 211        /8
239 
-                PAVGB(%%mm0, %%mm3)                           //642211        /16
240 
-                "movq %%mm3, (%0)                       \n\t" // X
241 
-                // mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
242 
-                "movq %%mm1, %%mm0                      \n\t" //  1
243 
-                PAVGB(%%mm6, %%mm0)                           //1 1        /2
244 
-                "movq %%mm4, %%mm3                      \n\t" // 1
245 
-                PAVGB((%0,%1,2), %%mm3)                       // 1 1        /2
246 
-                PAVGB((%%REGa,%1,2), %%mm5)                   //     11        /2
247 
-                PAVGB((%%REGa), %%mm5)                        //    211 /4
248 
-                PAVGB(%%mm5, %%mm3)                           // 2 2211 /8
249 
-                PAVGB(%%mm0, %%mm3)                           //4242211 /16
250 
-                "movq %%mm3, (%0,%1)                    \n\t" //  X
251 
-                // mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
252 
-                PAVGB(%%mm4, %%mm6)                                   //11        /2
253 
-                "movq (%%"REG_c"), %%mm0                \n\t" //       1
254 
-                PAVGB((%%REGa, %1, 2), %%mm0)                 //      11/2
255 
-                "movq %%mm0, %%mm3                      \n\t" //      11/2
256 
-                PAVGB(%%mm1, %%mm0)                           //  2   11/4
257 
-                PAVGB(%%mm6, %%mm0)                           //222   11/8
258 
-                PAVGB(%%mm2, %%mm0)                           //22242211/16
259 
-                "movq (%0, %1, 2), %%mm2                \n\t" //   1
260 
-                "movq %%mm0, (%0, %1, 2)                \n\t" //   X
261 
-                // mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
262 
-                "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
263 
-                PAVGB((%%REGc), %%mm0)                        //       11        /2
264 
-                PAVGB(%%mm0, %%mm6)                           //11     11        /4
265 
-                PAVGB(%%mm1, %%mm4)                           // 11                /2
266 
-                PAVGB(%%mm2, %%mm1)                           //  11                /2
267 
-                PAVGB(%%mm1, %%mm6)                           //1122   11        /8
268 
-                PAVGB(%%mm5, %%mm6)                           //112242211        /16
269 
-                "movq (%%"REG_a"), %%mm5                \n\t" //    1
270 
-                "movq %%mm6, (%%"REG_a")                \n\t" //    X
271 
-                // mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
272 
-                "movq (%%"REG_a", %1, 4), %%mm6         \n\t" //        1
273 
-                PAVGB(%%mm7, %%mm6)                           //        11        /2
274 
-                PAVGB(%%mm4, %%mm6)                           // 11     11        /4
275 
-                PAVGB(%%mm3, %%mm6)                           // 11   2211        /8
276 
-                PAVGB(%%mm5, %%mm2)                           //   11                /2
277 
-                "movq (%0, %1, 4), %%mm4                \n\t" //     1
278 
-                PAVGB(%%mm4, %%mm2)                           //   112                /4
279 
-                PAVGB(%%mm2, %%mm6)                           // 112242211        /16
280 
-                "movq %%mm6, (%0, %1, 4)                \n\t" //     X
281 
-                // mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
282 
-                PAVGB(%%mm7, %%mm1)                           //  11     2        /4
283 
-                PAVGB(%%mm4, %%mm5)                           //    11                /2
284 
-                PAVGB(%%mm5, %%mm0)                           //    11 11        /4
285 
-                "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //      1
286 
-                PAVGB(%%mm6, %%mm1)                           //  11  4  2        /8
287 
-                PAVGB(%%mm0, %%mm1)                           //  11224222        /16
288 
-                "movq %%mm1, (%%"REG_a", %1, 2)         \n\t" //      X
289 
-                // mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
290 
-                PAVGB((%%REGc), %%mm2)                        //   112 4        /8
291 
-                "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
292 
-                PAVGB(%%mm0, %%mm6)                           //      1 1        /2
293 
-                PAVGB(%%mm7, %%mm6)                           //      1 12        /4
294 
-                PAVGB(%%mm2, %%mm6)                           //   1122424        /4
295 
-                "movq %%mm6, (%%"REG_c")                \n\t" //       X
296 
-                // mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
297 
-                PAVGB(%%mm7, %%mm5)                           //    11   2        /4
298 
-                PAVGB(%%mm7, %%mm5)                           //    11   6        /8
299 
-
300 
-                PAVGB(%%mm3, %%mm0)                           //      112        /4
301 
-                PAVGB(%%mm0, %%mm5)                           //    112246        /16
302 
-                "movq %%mm5, (%%"REG_a", %1, 4)         \n\t" //        X
303 
-                "sub %1, %0                             \n\t"
304 
-
305 
-                :
306 
-                : "r" (src), "r" ((long)stride), "m" (c->pQPb)
307 
-                : "%"REG_a, "%"REG_c
308 
-        );
183 
+    src+= stride*3;
184 
+    asm volatile(        //"movv %0 %1 %2\n\t"
185 
+        "movq %2, %%mm0                         \n\t"  // QP,..., QP
186 
+        "pxor %%mm4, %%mm4                      \n\t"
187 
+
188 
+        "movq (%0), %%mm6                       \n\t"
189 
+        "movq (%0, %1), %%mm5                   \n\t"
190 
+        "movq %%mm5, %%mm1                      \n\t"
191 
+        "movq %%mm6, %%mm2                      \n\t"
192 
+        "psubusb %%mm6, %%mm5                   \n\t"
193 
+        "psubusb %%mm1, %%mm2                   \n\t"
194 
+        "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
195 
+        "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
196 
+        "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
197 
+
198 
+        "pand %%mm2, %%mm6                      \n\t"
199 
+        "pandn %%mm1, %%mm2                     \n\t"
200 
+        "por %%mm2, %%mm6                       \n\t"// First Line to Filter
201 
+
202 
+        "movq (%0, %1, 8), %%mm5                \n\t"
203 
+        "lea (%0, %1, 4), %%"REG_a"             \n\t"
204 
+        "lea (%0, %1, 8), %%"REG_c"             \n\t"
205 
+        "sub %1, %%"REG_c"                      \n\t"
206 
+        "add %1, %0                             \n\t" // %0 points to line 1 not 0
207 
+        "movq (%0, %1, 8), %%mm7                \n\t"
208 
+        "movq %%mm5, %%mm1                      \n\t"
209 
+        "movq %%mm7, %%mm2                      \n\t"
210 
+        "psubusb %%mm7, %%mm5                   \n\t"
211 
+        "psubusb %%mm1, %%mm2                   \n\t"
212 
+        "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
213 
+        "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0
214 
+        "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF
215 
+
216 
+        "pand %%mm2, %%mm7                      \n\t"
217 
+        "pandn %%mm1, %%mm2                     \n\t"
218 
+        "por %%mm2, %%mm7                       \n\t" // First Line to Filter
219 
+
220 
+
221 
+        //      1       2       3       4       5       6       7       8
222 
+        //      %0      %0+%1   %0+2%1  eax     %0+4%1  eax+2%1 ecx     eax+4%1
223 
+        // 6 4 2 2 1 1
224 
+        // 6 4 4 2
225 
+        // 6 8 2
226 
+
227 
+        "movq (%0, %1), %%mm0                   \n\t" //  1
228 
+        "movq %%mm0, %%mm1                      \n\t" //  1
229 
+        PAVGB(%%mm6, %%mm0)                           //1 1        /2
230 
+        PAVGB(%%mm6, %%mm0)                           //3 1        /4
231 
+
232 
+        "movq (%0, %1, 4), %%mm2                \n\t" //     1
233 
+        "movq %%mm2, %%mm5                      \n\t" //     1
234 
+        PAVGB((%%REGa), %%mm2)                        //    11        /2
235 
+        PAVGB((%0, %1, 2), %%mm2)                     //   211        /4
236 
+        "movq %%mm2, %%mm3                      \n\t" //   211        /4
237 
+        "movq (%0), %%mm4                       \n\t" // 1
238 
+        PAVGB(%%mm4, %%mm3)                           // 4 211        /8
239 
+        PAVGB(%%mm0, %%mm3)                           //642211        /16
240 
+        "movq %%mm3, (%0)                       \n\t" // X
241 
+        // mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
242 
+        "movq %%mm1, %%mm0                      \n\t" //  1
243 
+        PAVGB(%%mm6, %%mm0)                           //1 1        /2
244 
+        "movq %%mm4, %%mm3                      \n\t" // 1
245 
+        PAVGB((%0,%1,2), %%mm3)                       // 1 1        /2
246 
+        PAVGB((%%REGa,%1,2), %%mm5)                   //     11        /2
247 
+        PAVGB((%%REGa), %%mm5)                        //    211 /4
248 
+        PAVGB(%%mm5, %%mm3)                           // 2 2211 /8
249 
+        PAVGB(%%mm0, %%mm3)                           //4242211 /16
250 
+        "movq %%mm3, (%0,%1)                    \n\t" //  X
251 
+        // mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
252 
+        PAVGB(%%mm4, %%mm6)                                   //11        /2
253 
+        "movq (%%"REG_c"), %%mm0                \n\t" //       1
254 
+        PAVGB((%%REGa, %1, 2), %%mm0)                 //      11/2
255 
+        "movq %%mm0, %%mm3                      \n\t" //      11/2
256 
+        PAVGB(%%mm1, %%mm0)                           //  2   11/4
257 
+        PAVGB(%%mm6, %%mm0)                           //222   11/8
258 
+        PAVGB(%%mm2, %%mm0)                           //22242211/16
259 
+        "movq (%0, %1, 2), %%mm2                \n\t" //   1
260 
+        "movq %%mm0, (%0, %1, 2)                \n\t" //   X
261 
+        // mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
262 
+        "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
263 
+        PAVGB((%%REGc), %%mm0)                        //       11        /2
264 
+        PAVGB(%%mm0, %%mm6)                           //11     11        /4
265 
+        PAVGB(%%mm1, %%mm4)                           // 11                /2
266 
+        PAVGB(%%mm2, %%mm1)                           //  11                /2
267 
+        PAVGB(%%mm1, %%mm6)                           //1122   11        /8
268 
+        PAVGB(%%mm5, %%mm6)                           //112242211        /16
269 
+        "movq (%%"REG_a"), %%mm5                \n\t" //    1
270 
+        "movq %%mm6, (%%"REG_a")                \n\t" //    X
271 
+        // mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
272 
+        "movq (%%"REG_a", %1, 4), %%mm6         \n\t" //        1
273 
+        PAVGB(%%mm7, %%mm6)                           //        11        /2
274 
+        PAVGB(%%mm4, %%mm6)                           // 11     11        /4
275 
+        PAVGB(%%mm3, %%mm6)                           // 11   2211        /8
276 
+        PAVGB(%%mm5, %%mm2)                           //   11                /2
277 
+        "movq (%0, %1, 4), %%mm4                \n\t" //     1
278 
+        PAVGB(%%mm4, %%mm2)                           //   112                /4
279 
+        PAVGB(%%mm2, %%mm6)                           // 112242211        /16
280 
+        "movq %%mm6, (%0, %1, 4)                \n\t" //     X
281 
+        // mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
282 
+        PAVGB(%%mm7, %%mm1)                           //  11     2        /4
283 
+        PAVGB(%%mm4, %%mm5)                           //    11                /2
284 
+        PAVGB(%%mm5, %%mm0)                           //    11 11        /4
285 
+        "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //      1
286 
+        PAVGB(%%mm6, %%mm1)                           //  11  4  2        /8
287 
+        PAVGB(%%mm0, %%mm1)                           //  11224222        /16
288 
+        "movq %%mm1, (%%"REG_a", %1, 2)         \n\t" //      X
289 
+        // mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
290 
+        PAVGB((%%REGc), %%mm2)                        //   112 4        /8
291 
+        "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1
292 
+        PAVGB(%%mm0, %%mm6)                           //      1 1        /2
293 
+        PAVGB(%%mm7, %%mm6)                           //      1 12        /4
294 
+        PAVGB(%%mm2, %%mm6)                           //   1122424        /4
295 
+        "movq %%mm6, (%%"REG_c")                \n\t" //       X
296 
+        // mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
297 
+        PAVGB(%%mm7, %%mm5)                           //    11   2        /4
298 
+        PAVGB(%%mm7, %%mm5)                           //    11   6        /8
299 
+
300 
+        PAVGB(%%mm3, %%mm0)                           //      112        /4
301 
+        PAVGB(%%mm0, %%mm5)                           //    112246        /16
302 
+        "movq %%mm5, (%%"REG_a", %1, 4)         \n\t" //        X
303 
+        "sub %1, %0                             \n\t"
304 
+
305 
+        :
306 
+        : "r" (src), "r" ((long)stride), "m" (c->pQPb)
307 
+        : "%"REG_a, "%"REG_c
308 
+    );
309 309 
 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
310 
-        const int l1= stride;
311 
-        const int l2= stride + l1;
312 
-        const int l3= stride + l2;
313 
-        const int l4= stride + l3;
314 
-        const int l5= stride + l4;
315 
-        const int l6= stride + l5;
316 
-        const int l7= stride + l6;
317 
-        const int l8= stride + l7;
318 
-        const int l9= stride + l8;
319 
-        int x;
320 
-        src+= stride*3;
321 
-        for(x=0; x<BLOCK_SIZE; x++)
322 
-        {
323 
-                const int first= FFABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];
324 
-                const int last= FFABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];
325 
-
326 
-                int sums[10];
327 
-                sums[0] = 4*first + src[l1] + src[l2] + src[l3] + 4;
328 
-                sums[1] = sums[0] - first  + src[l4];
329 
-                sums[2] = sums[1] - first  + src[l5];
330 
-                sums[3] = sums[2] - first  + src[l6];
331 
-                sums[4] = sums[3] - first  + src[l7];
332 
-                sums[5] = sums[4] - src[l1] + src[l8];
333 
-                sums[6] = sums[5] - src[l2] + last;
334 
-                sums[7] = sums[6] - src[l3] + last;
335 
-                sums[8] = sums[7] - src[l4] + last;
336 
-                sums[9] = sums[8] - src[l5] + last;
337 
-
338 
-                src[l1]= (sums[0] + sums[2] + 2*src[l1])>>4;
339 
-                src[l2]= (sums[1] + sums[3] + 2*src[l2])>>4;
340 
-                src[l3]= (sums[2] + sums[4] + 2*src[l3])>>4;
341 
-                src[l4]= (sums[3] + sums[5] + 2*src[l4])>>4;
342 
-                src[l5]= (sums[4] + sums[6] + 2*src[l5])>>4;
343 
-                src[l6]= (sums[5] + sums[7] + 2*src[l6])>>4;
344 
-                src[l7]= (sums[6] + sums[8] + 2*src[l7])>>4;
345 
-                src[l8]= (sums[7] + sums[9] + 2*src[l8])>>4;
346 
-
347 
-                src++;
348 
-        }
310 
+    const int l1= stride;
311 
+    const int l2= stride + l1;
312 
+    const int l3= stride + l2;
313 
+    const int l4= stride + l3;
314 
+    const int l5= stride + l4;
315 
+    const int l6= stride + l5;
316 
+    const int l7= stride + l6;
317 
+    const int l8= stride + l7;
318 
+    const int l9= stride + l8;
319 
+    int x;
320 
+    src+= stride*3;
321 
+    for(x=0; x<BLOCK_SIZE; x++){
322 
+        const int first= FFABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];
323 
+        const int last= FFABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];
324 
+
325 
+        int sums[10];
326 
+        sums[0] = 4*first + src[l1] + src[l2] + src[l3] + 4;
327 
+        sums[1] = sums[0] - first  + src[l4];
328 
+        sums[2] = sums[1] - first  + src[l5];
329 
+        sums[3] = sums[2] - first  + src[l6];
330 
+        sums[4] = sums[3] - first  + src[l7];
331 
+        sums[5] = sums[4] - src[l1] + src[l8];
332 
+        sums[6] = sums[5] - src[l2] + last;
333 
+        sums[7] = sums[6] - src[l3] + last;
334 
+        sums[8] = sums[7] - src[l4] + last;
335 
+        sums[9] = sums[8] - src[l5] + last;
336 
+
337 
+        src[l1]= (sums[0] + sums[2] + 2*src[l1])>>4;
338 
+        src[l2]= (sums[1] + sums[3] + 2*src[l2])>>4;
339 
+        src[l3]= (sums[2] + sums[4] + 2*src[l3])>>4;
340 
+        src[l4]= (sums[3] + sums[5] + 2*src[l4])>>4;
341 
+        src[l5]= (sums[4] + sums[6] + 2*src[l5])>>4;
342 
+        src[l6]= (sums[5] + sums[7] + 2*src[l6])>>4;
343 
+        src[l7]= (sums[6] + sums[8] + 2*src[l7])>>4;
344 
+        src[l8]= (sums[7] + sums[9] + 2*src[l8])>>4;
345 
+
346 
+        src++;
347 
+    }
349 348 
 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
350 349 
 }
351 350 
 #endif //HAVE_ALTIVEC
... ... 
@@ -366,92 +365,89 @@ static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
366 366 
 static inline void RENAME(vertRK1Filter)(uint8_t *src, int stride, int QP)
367 367 
 {
368 368 
 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
369 
-        src+= stride*3;
369 
+    src+= stride*3;
370 370 
 // FIXME rounding
371 
-        asm volatile(
372 
-                "pxor %%mm7, %%mm7                      \n\t" // 0
373 
-                "movq "MANGLE(b80)", %%mm6              \n\t" // MIN_SIGNED_BYTE
374 
-                "leal (%0, %1), %%"REG_a"               \n\t"
375 
-                "leal (%%"REG_a", %1, 4), %%"REG_c"     \n\t"
371 
+    asm volatile(
372 
+        "pxor %%mm7, %%mm7                      \n\t" // 0
373 
+        "movq "MANGLE(b80)", %%mm6              \n\t" // MIN_SIGNED_BYTE
374 
+        "leal (%0, %1), %%"REG_a"               \n\t"
375 
+        "leal (%%"REG_a", %1, 4), %%"REG_c"     \n\t"
376 376 
 //      0       1       2       3       4       5       6       7       8       9
377 377 
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
378 
-                "movq "MANGLE(pQPb)", %%mm0             \n\t" // QP,..., QP
379 
-                "movq %%mm0, %%mm1                      \n\t" // QP,..., QP
380 
-                "paddusb "MANGLE(b02)", %%mm0           \n\t"
381 
-                "psrlw $2, %%mm0                        \n\t"
382 
-                "pand "MANGLE(b3F)", %%mm0              \n\t" // QP/4,..., QP/4
383 
-                "paddusb %%mm1, %%mm0                   \n\t" // QP*1.25 ...
384 
-                "movq (%0, %1, 4), %%mm2                \n\t" // line 4
385 
-                "movq (%%"REG_c"), %%mm3                \n\t" // line 5
386 
-                "movq %%mm2, %%mm4                      \n\t" // line 4
387 
-                "pcmpeqb %%mm5, %%mm5                   \n\t" // -1
388 
-                "pxor %%mm2, %%mm5                      \n\t" // -line 4 - 1
389 
-                PAVGB(%%mm3, %%mm5)
390 
-                "paddb %%mm6, %%mm5                     \n\t" // (l5-l4)/2
391 
-                "psubusb %%mm3, %%mm4                   \n\t"
392 
-                "psubusb %%mm2, %%mm3                   \n\t"
393 
-                "por %%mm3, %%mm4                       \n\t" // |l4 - l5|
394 
-                "psubusb %%mm0, %%mm4                   \n\t"
395 
-                "pcmpeqb %%mm7, %%mm4                   \n\t"
396 
-                "pand %%mm4, %%mm5                      \n\t" // d/2
397 
-
398 
-//                "paddb %%mm6, %%mm2                     \n\t" // line 4 + 0x80
399 
-                "paddb %%mm5, %%mm2                     \n\t"
400 
-//                "psubb %%mm6, %%mm2                     \n\t"
401 
-                "movq %%mm2, (%0,%1, 4)                 \n\t"
402 
-
403 
-                "movq (%%"REG_c"), %%mm2                \n\t"
404 
-//                "paddb %%mm6, %%mm2                     \n\t" // line 5 + 0x80
405 
-                "psubb %%mm5, %%mm2                     \n\t"
406 
-//                "psubb %%mm6, %%mm2                     \n\t"
407 
-                "movq %%mm2, (%%"REG_c")                \n\t"
408 
-
409 
-                "paddb %%mm6, %%mm5                     \n\t"
410 
-                "psrlw $2, %%mm5                        \n\t"
411 
-                "pand "MANGLE(b3F)", %%mm5              \n\t"
412 
-                "psubb "MANGLE(b20)", %%mm5             \n\t" // (l5-l4)/8
413 
-
414 
-                "movq (%%"REG_a", %1, 2), %%mm2         \n\t"
415 
-                "paddb %%mm6, %%mm2                     \n\t" // line 3 + 0x80
416 
-                "paddsb %%mm5, %%mm2                    \n\t"
417 
-                "psubb %%mm6, %%mm2                     \n\t"
418 
-                "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"
419 
-
420 
-                "movq (%%"REG_c", %1), %%mm2            \n\t"
421 
-                "paddb %%mm6, %%mm2                     \n\t" // line 6 + 0x80
422 
-                "psubsb %%mm5, %%mm2                    \n\t"
423 
-                "psubb %%mm6, %%mm2                     \n\t"
424 
-                "movq %%mm2, (%%"REG_c", %1)            \n\t"
425 
-
426 
-                :
427 
-                : "r" (src), "r" ((long)stride)
428 
-                : "%"REG_a, "%"REG_c
429 
-        );
378 
+        "movq "MANGLE(pQPb)", %%mm0             \n\t" // QP,..., QP
379 
+        "movq %%mm0, %%mm1                      \n\t" // QP,..., QP
380 
+        "paddusb "MANGLE(b02)", %%mm0           \n\t"
381 
+        "psrlw $2, %%mm0                        \n\t"
382 
+        "pand "MANGLE(b3F)", %%mm0              \n\t" // QP/4,..., QP/4
383 
+        "paddusb %%mm1, %%mm0                   \n\t" // QP*1.25 ...
384 
+        "movq (%0, %1, 4), %%mm2                \n\t" // line 4
385 
+        "movq (%%"REG_c"), %%mm3                \n\t" // line 5
386 
+        "movq %%mm2, %%mm4                      \n\t" // line 4
387 
+        "pcmpeqb %%mm5, %%mm5                   \n\t" // -1
388 
+        "pxor %%mm2, %%mm5                      \n\t" // -line 4 - 1
389 
+        PAVGB(%%mm3, %%mm5)
390 
+        "paddb %%mm6, %%mm5                     \n\t" // (l5-l4)/2
391 
+        "psubusb %%mm3, %%mm4                   \n\t"
392 
+        "psubusb %%mm2, %%mm3                   \n\t"
393 
+        "por %%mm3, %%mm4                       \n\t" // |l4 - l5|
394 
+        "psubusb %%mm0, %%mm4                   \n\t"
395 
+        "pcmpeqb %%mm7, %%mm4                   \n\t"
396 
+        "pand %%mm4, %%mm5                      \n\t" // d/2
397 
+
398 
+//        "paddb %%mm6, %%mm2                     \n\t" // line 4 + 0x80
399 
+        "paddb %%mm5, %%mm2                     \n\t"
400 
+//        "psubb %%mm6, %%mm2                     \n\t"
401 
+        "movq %%mm2, (%0,%1, 4)                 \n\t"
402 
+
403 
+        "movq (%%"REG_c"), %%mm2                \n\t"
404 
+//        "paddb %%mm6, %%mm2                     \n\t" // line 5 + 0x80
405 
+        "psubb %%mm5, %%mm2                     \n\t"
406 
+//        "psubb %%mm6, %%mm2                     \n\t"
407 
+        "movq %%mm2, (%%"REG_c")                \n\t"
408 
+
409 
+        "paddb %%mm6, %%mm5                     \n\t"
410 
+        "psrlw $2, %%mm5                        \n\t"
411 
+        "pand "MANGLE(b3F)", %%mm5              \n\t"
412 
+        "psubb "MANGLE(b20)", %%mm5             \n\t" // (l5-l4)/8
413 
+
414 
+        "movq (%%"REG_a", %1, 2), %%mm2         \n\t"
415 
+        "paddb %%mm6, %%mm2                     \n\t" // line 3 + 0x80
416 
+        "paddsb %%mm5, %%mm2                    \n\t"
417 
+        "psubb %%mm6, %%mm2                     \n\t"
418 
+        "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"
419 
+
420 
+        "movq (%%"REG_c", %1), %%mm2            \n\t"
421 
+        "paddb %%mm6, %%mm2                     \n\t" // line 6 + 0x80
422 
+        "psubsb %%mm5, %%mm2                    \n\t"
423 
+        "psubb %%mm6, %%mm2                     \n\t"
424 
+        "movq %%mm2, (%%"REG_c", %1)            \n\t"
425 
+
426 
+        :
427 
+        : "r" (src), "r" ((long)stride)
428 
+        : "%"REG_a, "%"REG_c
429 
+    );
430 430 
 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
431 
-         const int l1= stride;
432 
-        const int l2= stride + l1;
433 
-        const int l3= stride + l2;
434 
-        const int l4= stride + l3;
435 
-        const int l5= stride + l4;
436 
-        const int l6= stride + l5;
437 
-//        const int l7= stride + l6;
438 
-//        const int l8= stride + l7;
439 
-//        const int l9= stride + l8;
440 
-        int x;
441 
-        const int QP15= QP + (QP>>2);
442 
-        src+= stride*3;
443 
-        for(x=0; x<BLOCK_SIZE; x++)
444 
-        {
445 
-                const int v = (src[x+l5] - src[x+l4]);
446 
-                if(FFABS(v) < QP15)
447 
-                {
448 
-                        src[x+l3] +=v>>3;
449 
-                        src[x+l4] +=v>>1;
450 
-                        src[x+l5] -=v>>1;
451 
-                        src[x+l6] -=v>>3;
452 
-
453 
-                }
431 
+    const int l1= stride;
432 
+    const int l2= stride + l1;
433 
+    const int l3= stride + l2;
434 
+    const int l4= stride + l3;
435 
+    const int l5= stride + l4;
436 
+    const int l6= stride + l5;
437 
+//    const int l7= stride + l6;
438 
+//    const int l8= stride + l7;
439 
+//    const int l9= stride + l8;
440 
+    int x;
441 
+    const int QP15= QP + (QP>>2);
442 
+    src+= stride*3;
443 
+    for(x=0; x<BLOCK_SIZE; x++){
444 
+        const int v = (src[x+l5] - src[x+l4]);
445 
+        if(FFABS(v) < QP15){
446 
+            src[x+l3] +=v>>3;
447 
+            src[x+l4] +=v>>1;
448 
+            src[x+l5] -=v>>1;
449 
+            src[x+l6] -=v>>3;
454 450 
         }
451 
+    }
455 452
456 453 
 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
457 454 
 }
... ... 
@@ -467,128 +463,125 @@ static inline void RENAME(vertRK1Filter)(uint8_t *src, int stride, int QP)
467 467 
 static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
468 468 
 {
469 469 
 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
470 
-        src+= stride*3;
470 
+    src+= stride*3;
471 471
472 
-        asm volatile(
473 
-                "pxor %%mm7, %%mm7                      \n\t" // 0
474 
-                "lea (%0, %1), %%"REG_a"                \n\t"
475 
-                "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
472 
+    asm volatile(
473 
+        "pxor %%mm7, %%mm7                      \n\t" // 0
474 
+        "lea (%0, %1), %%"REG_a"                \n\t"
475 
+        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
476 476 
 //      0       1       2       3       4       5       6       7       8       9
477 477 
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
478 
-                "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
479 
-                "movq (%0, %1, 4), %%mm1                \n\t" // line 4
480 
-                "movq %%mm1, %%mm2                      \n\t" // line 4
481 
-                "psubusb %%mm0, %%mm1                   \n\t"
482 
-                "psubusb %%mm2, %%mm0                   \n\t"
483 
-                "por %%mm1, %%mm0                       \n\t" // |l2 - l3|
484 
-                "movq (%%"REG_c"), %%mm3                \n\t" // line 5
485 
-                "movq (%%"REG_c", %1), %%mm4            \n\t" // line 6
486 
-                "movq %%mm3, %%mm5                      \n\t" // line 5
487 
-                "psubusb %%mm4, %%mm3                   \n\t"
488 
-                "psubusb %%mm5, %%mm4                   \n\t"
489 
-                "por %%mm4, %%mm3                       \n\t" // |l5 - l6|
490 
-                PAVGB(%%mm3, %%mm0)                           // (|l2 - l3| + |l5 - l6|)/2
491 
-                "movq %%mm2, %%mm1                      \n\t" // line 4
492 
-                "psubusb %%mm5, %%mm2                   \n\t"
493 
-                "movq %%mm2, %%mm4                      \n\t"
494 
-                "pcmpeqb %%mm7, %%mm2                   \n\t" // (l4 - l5) <= 0 ? -1 : 0
495 
-                "psubusb %%mm1, %%mm5                   \n\t"
496 
-                "por %%mm5, %%mm4                       \n\t" // |l4 - l5|
497 
-                "psubusb %%mm0, %%mm4                   \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
498 
-                "movq %%mm4, %%mm3                      \n\t" // d
499 
-                "movq %2, %%mm0                         \n\t"
500 
-                "paddusb %%mm0, %%mm0                   \n\t"
501 
-                "psubusb %%mm0, %%mm4                   \n\t"
502 
-                "pcmpeqb %%mm7, %%mm4                   \n\t" // d <= QP ? -1 : 0
503 
-                "psubusb "MANGLE(b01)", %%mm3           \n\t"
504 
-                "pand %%mm4, %%mm3                      \n\t" // d <= QP ? d : 0
505 
-
506 
-                PAVGB(%%mm7, %%mm3)                           // d/2
507 
-                "movq %%mm3, %%mm1                      \n\t" // d/2
508 
-                PAVGB(%%mm7, %%mm3)                           // d/4
509 
-                PAVGB(%%mm1, %%mm3)                           // 3*d/8
510 
-
511 
-                "movq (%0, %1, 4), %%mm0                \n\t" // line 4
512 
-                "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
513 
-                "psubusb %%mm3, %%mm0                   \n\t"
514 
-                "pxor %%mm2, %%mm0                      \n\t"
515 
-                "movq %%mm0, (%0, %1, 4)                \n\t" // line 4
516 
-
517 
-                "movq (%%"REG_c"), %%mm0                \n\t" // line 5
518 
-                "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
519 
-                "paddusb %%mm3, %%mm0                   \n\t"
520 
-                "pxor %%mm2, %%mm0                      \n\t"
521 
-                "movq %%mm0, (%%"REG_c")                \n\t" // line 5
522 
-
523 
-                PAVGB(%%mm7, %%mm1)                           // d/4
524 
-
525 
-                "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
526 
-                "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
527 
-                "psubusb %%mm1, %%mm0                   \n\t"
528 
-                "pxor %%mm2, %%mm0                      \n\t"
529 
-                "movq %%mm0, (%%"REG_a", %1, 2)         \n\t" // line 3
530 
-
531 
-                "movq (%%"REG_c", %1), %%mm0            \n\t" // line 6
532 
-                "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
533 
-                "paddusb %%mm1, %%mm0                   \n\t"
534 
-                "pxor %%mm2, %%mm0                      \n\t"
535 
-                "movq %%mm0, (%%"REG_c", %1)            \n\t" // line 6
536 
-
537 
-                PAVGB(%%mm7, %%mm1)                           // d/8
538 
-
539 
-                "movq (%%"REG_a", %1), %%mm0            \n\t" // line 2
540 
-                "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
541 
-                "psubusb %%mm1, %%mm0                   \n\t"
542 
-                "pxor %%mm2, %%mm0                      \n\t"
543 
-                "movq %%mm0, (%%"REG_a", %1)            \n\t" // line 2
544 
-
545 
-                "movq (%%"REG_c", %1, 2), %%mm0         \n\t" // line 7
546 
-                "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
547 
-                "paddusb %%mm1, %%mm0                   \n\t"
548 
-                "pxor %%mm2, %%mm0                      \n\t"
549 
-                "movq %%mm0, (%%"REG_c", %1, 2)         \n\t" // line 7
550 
-
551 
-                :
552 
-                : "r" (src), "r" ((long)stride), "m" (co->pQPb)
553 
-                : "%"REG_a, "%"REG_c
554 
-        );
478 
+        "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
479 
+        "movq (%0, %1, 4), %%mm1                \n\t" // line 4
480 
+        "movq %%mm1, %%mm2                      \n\t" // line 4
481 
+        "psubusb %%mm0, %%mm1                   \n\t"
482 
+        "psubusb %%mm2, %%mm0                   \n\t"
483 
+        "por %%mm1, %%mm0                       \n\t" // |l2 - l3|
484 
+        "movq (%%"REG_c"), %%mm3                \n\t" // line 5
485 
+        "movq (%%"REG_c", %1), %%mm4            \n\t" // line 6
486 
+        "movq %%mm3, %%mm5                      \n\t" // line 5
487 
+        "psubusb %%mm4, %%mm3                   \n\t"
488 
+        "psubusb %%mm5, %%mm4                   \n\t"
489 
+        "por %%mm4, %%mm3                       \n\t" // |l5 - l6|
490 
+        PAVGB(%%mm3, %%mm0)                           // (|l2 - l3| + |l5 - l6|)/2
491 
+        "movq %%mm2, %%mm1                      \n\t" // line 4
492 
+        "psubusb %%mm5, %%mm2                   \n\t"
493 
+        "movq %%mm2, %%mm4                      \n\t"
494 
+        "pcmpeqb %%mm7, %%mm2                   \n\t" // (l4 - l5) <= 0 ? -1 : 0
495 
+        "psubusb %%mm1, %%mm5                   \n\t"
496 
+        "por %%mm5, %%mm4                       \n\t" // |l4 - l5|
497 
+        "psubusb %%mm0, %%mm4                   \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
498 
+        "movq %%mm4, %%mm3                      \n\t" // d
499 
+        "movq %2, %%mm0                         \n\t"
500 
+        "paddusb %%mm0, %%mm0                   \n\t"
501 
+        "psubusb %%mm0, %%mm4                   \n\t"
502 
+        "pcmpeqb %%mm7, %%mm4                   \n\t" // d <= QP ? -1 : 0
503 
+        "psubusb "MANGLE(b01)", %%mm3           \n\t"
504 
+        "pand %%mm4, %%mm3                      \n\t" // d <= QP ? d : 0
505 
+
506 
+        PAVGB(%%mm7, %%mm3)                           // d/2
507 
+        "movq %%mm3, %%mm1                      \n\t" // d/2
508 
+        PAVGB(%%mm7, %%mm3)                           // d/4
509 
+        PAVGB(%%mm1, %%mm3)                           // 3*d/8
510 
+
511 
+        "movq (%0, %1, 4), %%mm0                \n\t" // line 4
512 
+        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
513 
+        "psubusb %%mm3, %%mm0                   \n\t"
514 
+        "pxor %%mm2, %%mm0                      \n\t"
515 
+        "movq %%mm0, (%0, %1, 4)                \n\t" // line 4
516 
+
517 
+        "movq (%%"REG_c"), %%mm0                \n\t" // line 5
518 
+        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
519 
+        "paddusb %%mm3, %%mm0                   \n\t"
520 
+        "pxor %%mm2, %%mm0                      \n\t"
521 
+        "movq %%mm0, (%%"REG_c")                \n\t" // line 5
522 
+
523 
+        PAVGB(%%mm7, %%mm1)                           // d/4
524 
+
525 
+        "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3
526 
+        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
527 
+        "psubusb %%mm1, %%mm0                   \n\t"
528 
+        "pxor %%mm2, %%mm0                      \n\t"
529 
+        "movq %%mm0, (%%"REG_a", %1, 2)         \n\t" // line 3
530 
+
531 
+        "movq (%%"REG_c", %1), %%mm0            \n\t" // line 6
532 
+        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
533 
+        "paddusb %%mm1, %%mm0                   \n\t"
534 
+        "pxor %%mm2, %%mm0                      \n\t"
535 
+        "movq %%mm0, (%%"REG_c", %1)            \n\t" // line 6
536 
+
537 
+        PAVGB(%%mm7, %%mm1)                           // d/8
538 
+
539 
+        "movq (%%"REG_a", %1), %%mm0            \n\t" // line 2
540 
+        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
541 
+        "psubusb %%mm1, %%mm0                   \n\t"
542 
+        "pxor %%mm2, %%mm0                      \n\t"
543 
+        "movq %%mm0, (%%"REG_a", %1)            \n\t" // line 2
544 
+
545 
+        "movq (%%"REG_c", %1, 2), %%mm0         \n\t" // line 7
546 
+        "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
547 
+        "paddusb %%mm1, %%mm0                   \n\t"
548 
+        "pxor %%mm2, %%mm0                      \n\t"
549 
+        "movq %%mm0, (%%"REG_c", %1, 2)         \n\t" // line 7
550 
+
551 
+        :
552 
+        : "r" (src), "r" ((long)stride), "m" (co->pQPb)
553 
+        : "%"REG_a, "%"REG_c
554 
+    );
555 555 
 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
556 556
557 
-        const int l1= stride;
558 
-        const int l2= stride + l1;
559 
-        const int l3= stride + l2;
560 
-        const int l4= stride + l3;
561 
-        const int l5= stride + l4;
562 
-        const int l6= stride + l5;
563 
-        const int l7= stride + l6;
564 
-//        const int l8= stride + l7;
565 
-//        const int l9= stride + l8;
566 
-        int x;
567 
-
568 
-        src+= stride*3;
569 
-        for(x=0; x<BLOCK_SIZE; x++)
570 
-        {
571 
-                int a= src[l3] - src[l4];
572 
-                int b= src[l4] - src[l5];
573 
-                int c= src[l5] - src[l6];
574 
-
575 
-                int d= FFABS(b) - ((FFABS(a) + FFABS(c))>>1);
576 
-                d= FFMAX(d, 0);
577 
-
578 
-                if(d < co->QP*2)
579 
-                {
580 
-                        int v = d * FFSIGN(-b);
581 
-
582 
-                        src[l2] +=v>>3;
583 
-                        src[l3] +=v>>2;
584 
-                        src[l4] +=(3*v)>>3;
585 
-                        src[l5] -=(3*v)>>3;
586 
-                        src[l6] -=v>>2;
587 
-                        src[l7] -=v>>3;
588 
-
589 
-                }
590 
-                src++;
557 
+    const int l1= stride;
558 
+    const int l2= stride + l1;
559 
+    const int l3= stride + l2;
560 
+    const int l4= stride + l3;
561 
+    const int l5= stride + l4;
562 
+    const int l6= stride + l5;
563 
+    const int l7= stride + l6;
564 
+//    const int l8= stride + l7;
565 
+//    const int l9= stride + l8;
566 
+    int x;
567 
+
568 
+    src+= stride*3;
569 
+    for(x=0; x<BLOCK_SIZE; x++){
570 
+        int a= src[l3] - src[l4];
571 
+        int b= src[l4] - src[l5];
572 
+        int c= src[l5] - src[l6];
573 
+
574 
+        int d= FFABS(b) - ((FFABS(a) + FFABS(c))>>1);
575 
+        d= FFMAX(d, 0);
576 
+
577 
+        if(d < co->QP*2){
578 
+            int v = d * FFSIGN(-b);
579 
+
580 
+            src[l2] +=v>>3;
581 
+            src[l3] +=v>>2;
582 
+            src[l4] +=(3*v)>>3;
583 
+            src[l5] -=(3*v)>>3;
584 
+            src[l6] -=v>>2;
585 
+            src[l7] -=v>>3;
591 586 
         }
587 
+        src++;
588 
+    }
592 589 
 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
593 590 
 }
594 591
... ... 
@@ -597,569 +590,555 @@ static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext
597 597 
 {
598 598 
 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
599 599 
 /*
600 
-        uint8_t tmp[16];
601 
-        const int l1= stride;
602 
-        const int l2= stride + l1;
603 
-        const int l3= stride + l2;
604 
-        const int l4= (int)tmp - (int)src - stride*3;
605 
-        const int l5= (int)tmp - (int)src - stride*3 + 8;
606 
-        const int l6= stride*3 + l3;
607 
-        const int l7= stride + l6;
608 
-        const int l8= stride + l7;
609 
-
610 
-        memcpy(tmp, src+stride*7, 8);
611 
-        memcpy(tmp+8, src+stride*8, 8);
600 
+    uint8_t tmp[16];
601 
+    const int l1= stride;
602 
+    const int l2= stride + l1;
603 
+    const int l3= stride + l2;
604 
+    const int l4= (int)tmp - (int)src - stride*3;
605 
+    const int l5= (int)tmp - (int)src - stride*3 + 8;
606 
+    const int l6= stride*3 + l3;
607 
+    const int l7= stride + l6;
608 
+    const int l8= stride + l7;
609 
+
610 
+    memcpy(tmp, src+stride*7, 8);
611 
+    memcpy(tmp+8, src+stride*8, 8);
612 612 
 */
613 
-        src+= stride*4;
614 
-        asm volatile(
613 
+    src+= stride*4;
614 
+    asm volatile(
615 615
616 616 
 #if 0 //sligtly more accurate and slightly slower
617 
-                "pxor %%mm7, %%mm7                      \n\t" // 0
618 
-                "lea (%0, %1), %%"REG_a"                \n\t"
619 
-                "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
617 
+        "pxor %%mm7, %%mm7                      \n\t" // 0
618 
+        "lea (%0, %1), %%"REG_a"                \n\t"
619 
+        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
620 620 
 //      0       1       2       3       4       5       6       7
621 621 
 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1
622 622 
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1
623 623
624 624
625 
-                "movq (%0, %1, 2), %%mm0                \n\t" // l2
626 
-                "movq (%0), %%mm1                       \n\t" // l0
627 
-                "movq %%mm0, %%mm2                      \n\t" // l2
628 
-                PAVGB(%%mm7, %%mm0)                           // ~l2/2
629 
-                PAVGB(%%mm1, %%mm0)                           // ~(l2 + 2l0)/4
630 
-                PAVGB(%%mm2, %%mm0)                           // ~(5l2 + 2l0)/8
631 
-
632 
-                "movq (%%"REG_a"), %%mm1                \n\t" // l1
633 
-                "movq (%%"REG_a", %1, 2), %%mm3         \n\t" // l3
634 
-                "movq %%mm1, %%mm4                      \n\t" // l1
635 
-                PAVGB(%%mm7, %%mm1)                           // ~l1/2
636 
-                PAVGB(%%mm3, %%mm1)                           // ~(l1 + 2l3)/4
637 
-                PAVGB(%%mm4, %%mm1)                           // ~(5l1 + 2l3)/8
638 
-
639 
-                "movq %%mm0, %%mm4                      \n\t" // ~(5l2 + 2l0)/8
640 
-                "psubusb %%mm1, %%mm0                   \n\t"
641 
-                "psubusb %%mm4, %%mm1                   \n\t"
642 
-                "por %%mm0, %%mm1                       \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8
625 
+        "movq (%0, %1, 2), %%mm0                \n\t" // l2
626 
+        "movq (%0), %%mm1                       \n\t" // l0
627 
+        "movq %%mm0, %%mm2                      \n\t" // l2
628 
+        PAVGB(%%mm7, %%mm0)                           // ~l2/2
629 
+        PAVGB(%%mm1, %%mm0)                           // ~(l2 + 2l0)/4
630 
+        PAVGB(%%mm2, %%mm0)                           // ~(5l2 + 2l0)/8
631 
+
632 
+        "movq (%%"REG_a"), %%mm1                \n\t" // l1
633 
+        "movq (%%"REG_a", %1, 2), %%mm3         \n\t" // l3
634 
+        "movq %%mm1, %%mm4                      \n\t" // l1
635 
+        PAVGB(%%mm7, %%mm1)                           // ~l1/2
636 
+        PAVGB(%%mm3, %%mm1)                           // ~(l1 + 2l3)/4
637 
+        PAVGB(%%mm4, %%mm1)                           // ~(5l1 + 2l3)/8
638 
+
639 
+        "movq %%mm0, %%mm4                      \n\t" // ~(5l2 + 2l0)/8
640 
+        "psubusb %%mm1, %%mm0                   \n\t"
641 
+        "psubusb %%mm4, %%mm1                   \n\t"
642 
+        "por %%mm0, %%mm1                       \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8
643 643 
 // mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0
644 644
645 
-                "movq (%0, %1, 4), %%mm0                \n\t" // l4
646 
-                "movq %%mm0, %%mm4                      \n\t" // l4
647 
-                PAVGB(%%mm7, %%mm0)                           // ~l4/2
648 
-                PAVGB(%%mm2, %%mm0)                           // ~(l4 + 2l2)/4
649 
-                PAVGB(%%mm4, %%mm0)                           // ~(5l4 + 2l2)/8
650 
-
651 
-                "movq (%%"REG_c"), %%mm2                \n\t" // l5
652 
-                "movq %%mm3, %%mm5                      \n\t" // l3
653 
-                PAVGB(%%mm7, %%mm3)                           // ~l3/2
654 
-                PAVGB(%%mm2, %%mm3)                           // ~(l3 + 2l5)/4
655 
-                PAVGB(%%mm5, %%mm3)                           // ~(5l3 + 2l5)/8
656 
-
657 
-                "movq %%mm0, %%mm6                      \n\t" // ~(5l4 + 2l2)/8
658 
-                "psubusb %%mm3, %%mm0                   \n\t"
659 
-                "psubusb %%mm6, %%mm3                   \n\t"
660 
-                "por %%mm0, %%mm3                       \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8
661 
-                "pcmpeqb %%mm7, %%mm0                   \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)
645 
+        "movq (%0, %1, 4), %%mm0                \n\t" // l4
646 
+        "movq %%mm0, %%mm4                      \n\t" // l4
647 
+        PAVGB(%%mm7, %%mm0)                           // ~l4/2
648 
+        PAVGB(%%mm2, %%mm0)                           // ~(l4 + 2l2)/4
649 
+        PAVGB(%%mm4, %%mm0)                           // ~(5l4 + 2l2)/8
650 
+
651 
+        "movq (%%"REG_c"), %%mm2                \n\t" // l5
652 
+        "movq %%mm3, %%mm5                      \n\t" // l3
653 
+        PAVGB(%%mm7, %%mm3)                           // ~l3/2
654 
+        PAVGB(%%mm2, %%mm3)                           // ~(l3 + 2l5)/4
655 
+        PAVGB(%%mm5, %%mm3)                           // ~(5l3 + 2l5)/8
656 
+
657 
+        "movq %%mm0, %%mm6                      \n\t" // ~(5l4 + 2l2)/8
658 
+        "psubusb %%mm3, %%mm0                   \n\t"
659 
+        "psubusb %%mm6, %%mm3                   \n\t"
660 
+        "por %%mm0, %%mm3                       \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8
661 
+        "pcmpeqb %%mm7, %%mm0                   \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)
662 662 
 // mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0
663 663
664 
-                "movq (%%"REG_c", %1), %%mm6            \n\t" // l6
665 
-                "movq %%mm6, %%mm5                      \n\t" // l6
666 
-                PAVGB(%%mm7, %%mm6)                           // ~l6/2
667 
-                PAVGB(%%mm4, %%mm6)                           // ~(l6 + 2l4)/4
668 
-                PAVGB(%%mm5, %%mm6)                           // ~(5l6 + 2l4)/8
669 
-
670 
-                "movq (%%"REG_c", %1, 2), %%mm5         \n\t" // l7
671 
-                "movq %%mm2, %%mm4                      \n\t" // l5
672 
-                PAVGB(%%mm7, %%mm2)                           // ~l5/2
673 
-                PAVGB(%%mm5, %%mm2)                           // ~(l5 + 2l7)/4
674 
-                PAVGB(%%mm4, %%mm2)                           // ~(5l5 + 2l7)/8
675 
-
676 
-                "movq %%mm6, %%mm4                      \n\t" // ~(5l6 + 2l4)/8
677 
-                "psubusb %%mm2, %%mm6                   \n\t"
678 
-                "psubusb %%mm4, %%mm2                   \n\t"
679 
-                "por %%mm6, %%mm2                       \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8
664 
+        "movq (%%"REG_c", %1), %%mm6            \n\t" // l6
665 
+        "movq %%mm6, %%mm5                      \n\t" // l6
666 
+        PAVGB(%%mm7, %%mm6)                           // ~l6/2
667 
+        PAVGB(%%mm4, %%mm6)                           // ~(l6 + 2l4)/4
668 
+        PAVGB(%%mm5, %%mm6)                           // ~(5l6 + 2l4)/8
669 
+
670 
+        "movq (%%"REG_c", %1, 2), %%mm5         \n\t" // l7
671 
+        "movq %%mm2, %%mm4                      \n\t" // l5
672 
+        PAVGB(%%mm7, %%mm2)                           // ~l5/2
673 
+        PAVGB(%%mm5, %%mm2)                           // ~(l5 + 2l7)/4
674 
+        PAVGB(%%mm4, %%mm2)                           // ~(5l5 + 2l7)/8
675 
+
676 
+        "movq %%mm6, %%mm4                      \n\t" // ~(5l6 + 2l4)/8
677 
+        "psubusb %%mm2, %%mm6                   \n\t"
678 
+        "psubusb %%mm4, %%mm2                   \n\t"
679 
+        "por %%mm6, %%mm2                       \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8
680 680 
 // mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0
681 681
682 682
683 
-                PMINUB(%%mm2, %%mm1, %%mm4)                   // MIN(|lenergy|,|renergy|)/8
684 
-                "movq %2, %%mm4                         \n\t" // QP //FIXME QP+1 ?
685 
-                "paddusb "MANGLE(b01)", %%mm4           \n\t"
686 
-                "pcmpgtb %%mm3, %%mm4                   \n\t" // |menergy|/8 < QP
687 
-                "psubusb %%mm1, %%mm3                   \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8
688 
-                "pand %%mm4, %%mm3                      \n\t"
689 
-
690 
-                "movq %%mm3, %%mm1                      \n\t"
691 
-//                "psubusb "MANGLE(b01)", %%mm3           \n\t"
692 
-                PAVGB(%%mm7, %%mm3)
693 
-                PAVGB(%%mm7, %%mm3)
694 
-                "paddusb %%mm1, %%mm3                   \n\t"
695 
-//                "paddusb "MANGLE(b01)", %%mm3           \n\t"
696 
-
697 
-                "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //l3
698 
-                "movq (%0, %1, 4), %%mm5                \n\t" //l4
699 
-                "movq (%0, %1, 4), %%mm4                \n\t" //l4
700 
-                "psubusb %%mm6, %%mm5                   \n\t"
701 
-                "psubusb %%mm4, %%mm6                   \n\t"
702 
-                "por %%mm6, %%mm5                       \n\t" // |l3-l4|
703 
-                "pcmpeqb %%mm7, %%mm6                   \n\t" // SIGN(l3-l4)
704 
-                "pxor %%mm6, %%mm0                      \n\t"
705 
-                "pand %%mm0, %%mm3                      \n\t"
706 
-                PMINUB(%%mm5, %%mm3, %%mm0)
707 
-
708 
-                "psubusb "MANGLE(b01)", %%mm3           \n\t"
709 
-                PAVGB(%%mm7, %%mm3)
710 
-
711 
-                "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
712 
-                "movq (%0, %1, 4), %%mm2                \n\t"
713 
-                "pxor %%mm6, %%mm0                      \n\t"
714 
-                "pxor %%mm6, %%mm2                      \n\t"
715 
-                "psubb %%mm3, %%mm0                     \n\t"
716 
-                "paddb %%mm3, %%mm2                     \n\t"
717 
-                "pxor %%mm6, %%mm0                      \n\t"
718 
-                "pxor %%mm6, %%mm2                      \n\t"
719 
-                "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
720 
-                "movq %%mm2, (%0, %1, 4)                \n\t"
683 
+        PMINUB(%%mm2, %%mm1, %%mm4)                   // MIN(|lenergy|,|renergy|)/8
684 
+        "movq %2, %%mm4                         \n\t" // QP //FIXME QP+1 ?
685 
+        "paddusb "MANGLE(b01)", %%mm4           \n\t"
686 
+        "pcmpgtb %%mm3, %%mm4                   \n\t" // |menergy|/8 < QP
687 
+        "psubusb %%mm1, %%mm3                   \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8
688 
+        "pand %%mm4, %%mm3                      \n\t"
689 
+
690 
+        "movq %%mm3, %%mm1                      \n\t"
691 
+//        "psubusb "MANGLE(b01)", %%mm3           \n\t"
692 
+        PAVGB(%%mm7, %%mm3)
693 
+        PAVGB(%%mm7, %%mm3)
694 
+        "paddusb %%mm1, %%mm3                   \n\t"
695 
+//        "paddusb "MANGLE(b01)", %%mm3           \n\t"
696 
+
697 
+        "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //l3
698 
+        "movq (%0, %1, 4), %%mm5                \n\t" //l4
699 
+        "movq (%0, %1, 4), %%mm4                \n\t" //l4
700 
+        "psubusb %%mm6, %%mm5                   \n\t"
701 
+        "psubusb %%mm4, %%mm6                   \n\t"
702 
+        "por %%mm6, %%mm5                       \n\t" // |l3-l4|
703 
+        "pcmpeqb %%mm7, %%mm6                   \n\t" // SIGN(l3-l4)
704 
+        "pxor %%mm6, %%mm0                      \n\t"
705 
+        "pand %%mm0, %%mm3                      \n\t"
706 
+        PMINUB(%%mm5, %%mm3, %%mm0)
707 
+
708 
+        "psubusb "MANGLE(b01)", %%mm3           \n\t"
709 
+        PAVGB(%%mm7, %%mm3)
710 
+
711 
+        "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
712 
+        "movq (%0, %1, 4), %%mm2                \n\t"
713 
+        "pxor %%mm6, %%mm0                      \n\t"
714 
+        "pxor %%mm6, %%mm2                      \n\t"
715 
+        "psubb %%mm3, %%mm0                     \n\t"
716 
+        "paddb %%mm3, %%mm2                     \n\t"
717 
+        "pxor %%mm6, %%mm0                      \n\t"
718 
+        "pxor %%mm6, %%mm2                      \n\t"
719 
+        "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
720 
+        "movq %%mm2, (%0, %1, 4)                \n\t"
721 721 
 #endif //0
722 722
723 
-                "lea (%0, %1), %%"REG_a"                \n\t"
724 
-                "pcmpeqb %%mm6, %%mm6                   \n\t" // -1
723 
+        "lea (%0, %1), %%"REG_a"                \n\t"
724 
+        "pcmpeqb %%mm6, %%mm6                   \n\t" // -1
725 725 
 //      0       1       2       3       4       5       6       7
726 726 
 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1
727 727 
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1
728 728
729 729
730 
-                "movq (%%"REG_a", %1, 2), %%mm1         \n\t" // l3
731 
-                "movq (%0, %1, 4), %%mm0                \n\t" // l4
732 
-                "pxor %%mm6, %%mm1                      \n\t" // -l3-1
733 
-                PAVGB(%%mm1, %%mm0)                           // -q+128 = (l4-l3+256)/2
730 
+        "movq (%%"REG_a", %1, 2), %%mm1         \n\t" // l3
731 
+        "movq (%0, %1, 4), %%mm0                \n\t" // l4
732 
+        "pxor %%mm6, %%mm1                      \n\t" // -l3-1
733 
+        PAVGB(%%mm1, %%mm0)                           // -q+128 = (l4-l3+256)/2
734 734 
 // mm1=-l3-1, mm0=128-q
735 735
736 
-                "movq (%%"REG_a", %1, 4), %%mm2         \n\t" // l5
737 
-                "movq (%%"REG_a", %1), %%mm3            \n\t" // l2
738 
-                "pxor %%mm6, %%mm2                      \n\t" // -l5-1
739 
-                "movq %%mm2, %%mm5                      \n\t" // -l5-1
740 
-                "movq "MANGLE(b80)", %%mm4              \n\t" // 128
741 
-                "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
742 
-                PAVGB(%%mm3, %%mm2)                           // (l2-l5+256)/2
743 
-                PAVGB(%%mm0, %%mm4)                           // ~(l4-l3)/4 + 128
744 
-                PAVGB(%%mm2, %%mm4)                           // ~(l2-l5)/4 +(l4-l3)/8 + 128
745 
-                PAVGB(%%mm0, %%mm4)                           // ~(l2-l5)/8 +5(l4-l3)/16 + 128
736 
+        "movq (%%"REG_a", %1, 4), %%mm2         \n\t" // l5
737 
+        "movq (%%"REG_a", %1), %%mm3            \n\t" // l2
738 
+        "pxor %%mm6, %%mm2                      \n\t" // -l5-1
739 
+        "movq %%mm2, %%mm5                      \n\t" // -l5-1
740 
+        "movq "MANGLE(b80)", %%mm4              \n\t" // 128
741 
+        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
742 
+        PAVGB(%%mm3, %%mm2)                           // (l2-l5+256)/2
743 
+        PAVGB(%%mm0, %%mm4)                           // ~(l4-l3)/4 + 128
744 
+        PAVGB(%%mm2, %%mm4)                           // ~(l2-l5)/4 +(l4-l3)/8 + 128
745 
+        PAVGB(%%mm0, %%mm4)                           // ~(l2-l5)/8 +5(l4-l3)/16 + 128
746 746 
 // mm1=-l3-1, mm0=128-q, mm3=l2, mm4=menergy/16 + 128, mm5= -l5-1
747 747
748 
-                "movq (%%"REG_a"), %%mm2                \n\t" // l1
749 
-                "pxor %%mm6, %%mm2                      \n\t" // -l1-1
750 
-                PAVGB(%%mm3, %%mm2)                           // (l2-l1+256)/2
751 
-                PAVGB((%0), %%mm1)                            // (l0-l3+256)/2
752 
-                "movq "MANGLE(b80)", %%mm3              \n\t" // 128
753 
-                PAVGB(%%mm2, %%mm3)                           // ~(l2-l1)/4 + 128
754 
-                PAVGB(%%mm1, %%mm3)                           // ~(l0-l3)/4 +(l2-l1)/8 + 128
755 
-                PAVGB(%%mm2, %%mm3)                           // ~(l0-l3)/8 +5(l2-l1)/16 + 128
748 
+        "movq (%%"REG_a"), %%mm2                \n\t" // l1
749 
+        "pxor %%mm6, %%mm2                      \n\t" // -l1-1
750 
+        PAVGB(%%mm3, %%mm2)                           // (l2-l1+256)/2
751 
+        PAVGB((%0), %%mm1)                            // (l0-l3+256)/2
752 
+        "movq "MANGLE(b80)", %%mm3              \n\t" // 128
753 
+        PAVGB(%%mm2, %%mm3)                           // ~(l2-l1)/4 + 128
754 
+        PAVGB(%%mm1, %%mm3)                           // ~(l0-l3)/4 +(l2-l1)/8 + 128
755 
+        PAVGB(%%mm2, %%mm3)                           // ~(l0-l3)/8 +5(l2-l1)/16 + 128
756 756 
 // mm0=128-q, mm3=lenergy/16 + 128, mm4= menergy/16 + 128, mm5= -l5-1
757 757
758 
-                PAVGB((%%REGc, %1), %%mm5)                    // (l6-l5+256)/2
759 
-                "movq (%%"REG_c", %1, 2), %%mm1         \n\t" // l7
760 
-                "pxor %%mm6, %%mm1                      \n\t" // -l7-1
761 
-                PAVGB((%0, %1, 4), %%mm1)                     // (l4-l7+256)/2
762 
-                "movq "MANGLE(b80)", %%mm2              \n\t" // 128
763 
-                PAVGB(%%mm5, %%mm2)                           // ~(l6-l5)/4 + 128
764 
-                PAVGB(%%mm1, %%mm2)                           // ~(l4-l7)/4 +(l6-l5)/8 + 128
765 
-                PAVGB(%%mm5, %%mm2)                           // ~(l4-l7)/8 +5(l6-l5)/16 + 128
758 
+        PAVGB((%%REGc, %1), %%mm5)                    // (l6-l5+256)/2
759 
+        "movq (%%"REG_c", %1, 2), %%mm1         \n\t" // l7
760 
+        "pxor %%mm6, %%mm1                      \n\t" // -l7-1
761 
+        PAVGB((%0, %1, 4), %%mm1)                     // (l4-l7+256)/2
762 
+        "movq "MANGLE(b80)", %%mm2              \n\t" // 128
763 
+        PAVGB(%%mm5, %%mm2)                           // ~(l6-l5)/4 + 128
764 
+        PAVGB(%%mm1, %%mm2)                           // ~(l4-l7)/4 +(l6-l5)/8 + 128
765 
+        PAVGB(%%mm5, %%mm2)                           // ~(l4-l7)/8 +5(l6-l5)/16 + 128
766 766 
 // mm0=128-q, mm2=renergy/16 + 128, mm3=lenergy/16 + 128, mm4= menergy/16 + 128
767 767
768 
-                "movq "MANGLE(b00)", %%mm1              \n\t" // 0
769 
-                "movq "MANGLE(b00)", %%mm5              \n\t" // 0
770 
-                "psubb %%mm2, %%mm1                     \n\t" // 128 - renergy/16
771 
-                "psubb %%mm3, %%mm5                     \n\t" // 128 - lenergy/16
772 
-                PMAXUB(%%mm1, %%mm2)                          // 128 + |renergy/16|
773 
-                 PMAXUB(%%mm5, %%mm3)                         // 128 + |lenergy/16|
774 
-                PMINUB(%%mm2, %%mm3, %%mm1)                   // 128 + MIN(|lenergy|,|renergy|)/16
768 
+        "movq "MANGLE(b00)", %%mm1              \n\t" // 0
769 
+        "movq "MANGLE(b00)", %%mm5              \n\t" // 0
770 
+        "psubb %%mm2, %%mm1                     \n\t" // 128 - renergy/16
771 
+        "psubb %%mm3, %%mm5                     \n\t" // 128 - lenergy/16
772 
+        PMAXUB(%%mm1, %%mm2)                          // 128 + |renergy/16|
773 
+        PMAXUB(%%mm5, %%mm3)                          // 128 + |lenergy/16|
774 
+        PMINUB(%%mm2, %%mm3, %%mm1)                   // 128 + MIN(|lenergy|,|renergy|)/16
775 775
776 776 
 // mm0=128-q, mm3=128 + MIN(|lenergy|,|renergy|)/16, mm4= menergy/16 + 128
777 777
778 
-                "movq "MANGLE(b00)", %%mm7              \n\t" // 0
779 
-                "movq %2, %%mm2                         \n\t" // QP
780 
-                PAVGB(%%mm6, %%mm2)                           // 128 + QP/2
781 
-                "psubb %%mm6, %%mm2                     \n\t"
782 
-
783 
-                "movq %%mm4, %%mm1                      \n\t"
784 
-                "pcmpgtb %%mm7, %%mm1                   \n\t" // SIGN(menergy)
785 
-                "pxor %%mm1, %%mm4                      \n\t"
786 
-                "psubb %%mm1, %%mm4                     \n\t" // 128 + |menergy|/16
787 
-                "pcmpgtb %%mm4, %%mm2                   \n\t" // |menergy|/16 < QP/2
788 
-                "psubusb %%mm3, %%mm4                   \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16
778 
+        "movq "MANGLE(b00)", %%mm7              \n\t" // 0
779 
+        "movq %2, %%mm2                         \n\t" // QP
780 
+        PAVGB(%%mm6, %%mm2)                           // 128 + QP/2
781 
+        "psubb %%mm6, %%mm2                     \n\t"
782 
+
783 
+        "movq %%mm4, %%mm1                      \n\t"
784 
+        "pcmpgtb %%mm7, %%mm1                   \n\t" // SIGN(menergy)
785 
+        "pxor %%mm1, %%mm4                      \n\t"
786 
+        "psubb %%mm1, %%mm4                     \n\t" // 128 + |menergy|/16
787 
+        "pcmpgtb %%mm4, %%mm2                   \n\t" // |menergy|/16 < QP/2
788 
+        "psubusb %%mm3, %%mm4                   \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16
789 789 
 // mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16
790 790
791 
-                "movq %%mm4, %%mm3                      \n\t" // d
792 
-                "psubusb "MANGLE(b01)", %%mm4           \n\t"
793 
-                PAVGB(%%mm7, %%mm4)                           // d/32
794 
-                PAVGB(%%mm7, %%mm4)                           // (d + 32)/64
795 
-                "paddb %%mm3, %%mm4                     \n\t" // 5d/64
796 
-                "pand %%mm2, %%mm4                      \n\t"
797 
-
798 
-                "movq "MANGLE(b80)", %%mm5              \n\t" // 128
799 
-                "psubb %%mm0, %%mm5                     \n\t" // q
800 
-                "paddsb %%mm6, %%mm5                    \n\t" // fix bad rounding
801 
-                "pcmpgtb %%mm5, %%mm7                   \n\t" // SIGN(q)
802 
-                "pxor %%mm7, %%mm5                      \n\t"
803 
-
804 
-                PMINUB(%%mm5, %%mm4, %%mm3)                   // MIN(|q|, 5d/64)
805 
-                "pxor %%mm1, %%mm7                      \n\t" // SIGN(d*q)
806 
-
807 
-                "pand %%mm7, %%mm4                      \n\t"
808 
-                "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
809 
-                "movq (%0, %1, 4), %%mm2                \n\t"
810 
-                "pxor %%mm1, %%mm0                      \n\t"
811 
-                "pxor %%mm1, %%mm2                      \n\t"
812 
-                "paddb %%mm4, %%mm0                     \n\t"
813 
-                "psubb %%mm4, %%mm2                     \n\t"
814 
-                "pxor %%mm1, %%mm0                      \n\t"
815 
-                "pxor %%mm1, %%mm2                      \n\t"
816 
-                "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
817 
-                "movq %%mm2, (%0, %1, 4)                \n\t"
818 
-
819 
-                :
820 
-                : "r" (src), "r" ((long)stride), "m" (c->pQPb)
821 
-                : "%"REG_a, "%"REG_c
822 
-        );
791 
+        "movq %%mm4, %%mm3                      \n\t" // d
792 
+        "psubusb "MANGLE(b01)", %%mm4           \n\t"
793 
+        PAVGB(%%mm7, %%mm4)                           // d/32
794 
+        PAVGB(%%mm7, %%mm4)                           // (d + 32)/64
795 
+        "paddb %%mm3, %%mm4                     \n\t" // 5d/64
796 
+        "pand %%mm2, %%mm4                      \n\t"
797 
+
798 
+        "movq "MANGLE(b80)", %%mm5              \n\t" // 128
799 
+        "psubb %%mm0, %%mm5                     \n\t" // q
800 
+        "paddsb %%mm6, %%mm5                    \n\t" // fix bad rounding
801 
+        "pcmpgtb %%mm5, %%mm7                   \n\t" // SIGN(q)
802 
+        "pxor %%mm7, %%mm5                      \n\t"
803 
+
804 
+        PMINUB(%%mm5, %%mm4, %%mm3)                   // MIN(|q|, 5d/64)
805 
+        "pxor %%mm1, %%mm7                      \n\t" // SIGN(d*q)
806 
+
807 
+        "pand %%mm7, %%mm4                      \n\t"
808 
+        "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
809 
+        "movq (%0, %1, 4), %%mm2                \n\t"
810 
+        "pxor %%mm1, %%mm0                      \n\t"
811 
+        "pxor %%mm1, %%mm2                      \n\t"
812 
+        "paddb %%mm4, %%mm0                     \n\t"
813 
+        "psubb %%mm4, %%mm2                     \n\t"
814 
+        "pxor %%mm1, %%mm0                      \n\t"
815 
+        "pxor %%mm1, %%mm2                      \n\t"
816 
+        "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
817 
+        "movq %%mm2, (%0, %1, 4)                \n\t"
818 
+
819 
+        :
820 
+        : "r" (src), "r" ((long)stride), "m" (c->pQPb)
821 
+        : "%"REG_a, "%"REG_c
822 
+    );
823 823
824 824 
 /*
825 
-        {
826 
-        int x;
827 
-        src-= stride;
828 
-        for(x=0; x<BLOCK_SIZE; x++)
829 
-        {
830 
-                const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
831 
-                if(FFABS(middleEnergy)< 8*QP)
832 
-                {
833 
-                        const int q=(src[l4] - src[l5])/2;
834 
-                        const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
835 
-                        const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
836 
-
837 
-                        int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
838 
-                        d= FFMAX(d, 0);
839 
-
840 
-                        d= (5*d + 32) >> 6;
841 
-                        d*= FFSIGN(-middleEnergy);
842 
-
843 
-                        if(q>0)
844 
-                        {
845 
-                                d= d<0 ? 0 : d;
846 
-                                d= d>q ? q : d;
847 
-                        }
848 
-                        else
849 
-                        {
850 
-                                d= d>0 ? 0 : d;
851 
-                                d= d<q ? q : d;
852 
-                        }
853 
-
854 
-                        src[l4]-= d;
855 
-                        src[l5]+= d;
856 
-                }
857 
-                src++;
825 
+    {
826 
+    int x;
827 
+    src-= stride;
828 
+    for(x=0; x<BLOCK_SIZE; x++){
829 
+        const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
830 
+        if(FFABS(middleEnergy)< 8*QP){
831 
+            const int q=(src[l4] - src[l5])/2;
832 
+            const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
833 
+            const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
834 
+
835 
+            int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
836 
+            d= FFMAX(d, 0);
837 
+
838 
+            d= (5*d + 32) >> 6;
839 
+            d*= FFSIGN(-middleEnergy);
840 
+
841 
+            if(q>0){
842 
+                d= d<0 ? 0 : d;
843 
+                d= d>q ? q : d;
844 
+            }else{
845 
+                d= d>0 ? 0 : d;
846 
+                d= d<q ? q : d;
847 
+            }
848 
+
849 
+            src[l4]-= d;
850 
+            src[l5]+= d;
858 851 
         }
859 
-src-=8;
860 
-        for(x=0; x<8; x++)
861 
-        {
862 
-                int y;
863 
-                for(y=4; y<6; y++)
864 
-                {
865 
-                        int d= src[x+y*stride] - tmp[x+(y-4)*8];
866 
-                        int ad= FFABS(d);
867 
-                        static int max=0;
868 
-                        static int sum=0;
869 
-                        static int num=0;
870 
-                        static int bias=0;
871 
-
872 
-                        if(max<ad) max=ad;
873 
-                        sum+= ad>3 ? 1 : 0;
874 
-                        if(ad>3)
875 
-                        {
876 
-                                src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
877 
-                        }
878 
-                        if(y==4) bias+=d;
879 
-                        num++;
880 
-                        if(num%1000000 == 0)
881 
-                        {
882 
-                                av_log(c, AV_LOG_INFO, " %d %d %d %d\n", num, sum, max, bias);
883 
-                        }
884 
-                }
852 
+        src++;
853 
+    }
854 
+    src-=8;
855 
+    for(x=0; x<8; x++){
856 
+        int y;
857 
+        for(y=4; y<6; y++){
858 
+            int d= src[x+y*stride] - tmp[x+(y-4)*8];
859 
+            int ad= FFABS(d);
860 
+            static int max=0;
861 
+            static int sum=0;
862 
+            static int num=0;
863 
+            static int bias=0;
864 
+
865 
+            if(max<ad) max=ad;
866 
+            sum+= ad>3 ? 1 : 0;
867 
+            if(ad>3){
868 
+                src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
869 
+            }
870 
+            if(y==4) bias+=d;
871 
+            num++;
872 
+            if(num%1000000 == 0){
873 
+                av_log(c, AV_LOG_INFO, " %d %d %d %d\n", num, sum, max, bias);
874 
+            }
885 875 
         }
876 
+    }
886 877 
 }
887 878 
 */
888 879 
 #elif defined (HAVE_MMX)
889 
-        src+= stride*4;
890 
-        asm volatile(
891 
-                "pxor %%mm7, %%mm7                      \n\t"
892 
-                "lea -40(%%"REG_SP"), %%"REG_c"         \n\t" // make space for 4 8-byte vars
893 
-                "and "ALIGN_MASK", %%"REG_c"            \n\t" // align
880 
+    src+= stride*4;
881 
+    asm volatile(
882 
+        "pxor %%mm7, %%mm7                      \n\t"
883 
+        "lea -40(%%"REG_SP"), %%"REG_c"         \n\t" // make space for 4 8-byte vars
884 
+        "and "ALIGN_MASK", %%"REG_c"            \n\t" // align
894 885 
 //      0       1       2       3       4       5       6       7
895 886 
 //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 edx+%1  edx+2%1
896 887 
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1
897 888
898 
-                "movq (%0), %%mm0                       \n\t"
899 
-                "movq %%mm0, %%mm1                      \n\t"
900 
-                "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
901 
-                "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
902 
-
903 
-                "movq (%0, %1), %%mm2                   \n\t"
904 
-                "lea (%0, %1, 2), %%"REG_a"             \n\t"
905 
-                "movq %%mm2, %%mm3                      \n\t"
906 
-                "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
907 
-                "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
908 
-
909 
-                "movq (%%"REG_a"), %%mm4                \n\t"
910 
-                "movq %%mm4, %%mm5                      \n\t"
911 
-                "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
912 
-                "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
913 
-
914 
-                "paddw %%mm0, %%mm0                     \n\t" // 2L0
915 
-                "paddw %%mm1, %%mm1                     \n\t" // 2H0
916 
-                "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
917 
-                "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
918 
-                "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
919 
-                "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
920 
-
921 
-                "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
922 
-                "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
923 
-                "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
924 
-                "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
925 
-
926 
-                "movq (%%"REG_a", %1), %%mm2            \n\t"
927 
-                "movq %%mm2, %%mm3                      \n\t"
928 
-                "punpcklbw %%mm7, %%mm2                 \n\t" // L3
929 
-                "punpckhbw %%mm7, %%mm3                 \n\t" // H3
930 
-
931 
-                "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
932 
-                "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
933 
-                "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
934 
-                "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
935 
-                "movq %%mm0, (%%"REG_c")                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
936 
-                "movq %%mm1, 8(%%"REG_c")               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
937 
-
938 
-                "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
939 
-                "movq %%mm0, %%mm1                      \n\t"
940 
-                "punpcklbw %%mm7, %%mm0                 \n\t" // L4
941 
-                "punpckhbw %%mm7, %%mm1                 \n\t" // H4
942 
-
943 
-                "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
944 
-                "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
945 
-                "movq %%mm2, 16(%%"REG_c")              \n\t" // L3 - L4
946 
-                "movq %%mm3, 24(%%"REG_c")              \n\t" // H3 - H4
947 
-                "paddw %%mm4, %%mm4                     \n\t" // 2L2
948 
-                "paddw %%mm5, %%mm5                     \n\t" // 2H2
949 
-                "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
950 
-                "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
951 
-
952 
-                "lea (%%"REG_a", %1), %0                \n\t"
953 
-                "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
954 
-                "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
955 
-                "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
956 
-                "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
889 
+        "movq (%0), %%mm0                       \n\t"
890 
+        "movq %%mm0, %%mm1                      \n\t"
891 
+        "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
892 
+        "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
893 
+
894 
+        "movq (%0, %1), %%mm2                   \n\t"
895 
+        "lea (%0, %1, 2), %%"REG_a"             \n\t"
896 
+        "movq %%mm2, %%mm3                      \n\t"
897 
+        "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
898 
+        "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
899 
+
900 
+        "movq (%%"REG_a"), %%mm4                \n\t"
901 
+        "movq %%mm4, %%mm5                      \n\t"
902 
+        "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
903 
+        "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
904 
+
905 
+        "paddw %%mm0, %%mm0                     \n\t" // 2L0
906 
+        "paddw %%mm1, %%mm1                     \n\t" // 2H0
907 
+        "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
908 
+        "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
909 
+        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
910 
+        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
911 
+
912 
+        "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
913 
+        "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
914 
+        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
915 
+        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
916 
+
917 
+        "movq (%%"REG_a", %1), %%mm2            \n\t"
918 
+        "movq %%mm2, %%mm3                      \n\t"
919 
+        "punpcklbw %%mm7, %%mm2                 \n\t" // L3
920 
+        "punpckhbw %%mm7, %%mm3                 \n\t" // H3
921 
+
922 
+        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
923 
+        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
924 
+        "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
925 
+        "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
926 
+        "movq %%mm0, (%%"REG_c")                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
927 
+        "movq %%mm1, 8(%%"REG_c")               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
928 
+
929 
+        "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
930 
+        "movq %%mm0, %%mm1                      \n\t"
931 
+        "punpcklbw %%mm7, %%mm0                 \n\t" // L4
932 
+        "punpckhbw %%mm7, %%mm1                 \n\t" // H4
933 
+
934 
+        "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
935 
+        "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
936 
+        "movq %%mm2, 16(%%"REG_c")              \n\t" // L3 - L4
937 
+        "movq %%mm3, 24(%%"REG_c")              \n\t" // H3 - H4
938 
+        "paddw %%mm4, %%mm4                     \n\t" // 2L2
939 
+        "paddw %%mm5, %%mm5                     \n\t" // 2H2
940 
+        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
941 
+        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
942 
+
943 
+        "lea (%%"REG_a", %1), %0                \n\t"
944 
+        "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
945 
+        "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
946 
+        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
947 
+        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
957 948 
 //50 opcodes so far
958 
-                "movq (%0, %1, 2), %%mm2                \n\t"
959 
-                "movq %%mm2, %%mm3                      \n\t"
960 
-                "punpcklbw %%mm7, %%mm2                 \n\t" // L5
961 
-                "punpckhbw %%mm7, %%mm3                 \n\t" // H5
962 
-                "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
963 
-                "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
964 
-                "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
965 
-                "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
966 
-
967 
-                "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
968 
-                "punpcklbw %%mm7, %%mm6                 \n\t" // L6
969 
-                "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
970 
-                "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
971 
-                "punpckhbw %%mm7, %%mm6                 \n\t" // H6
972 
-                "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
973 
-
974 
-                "paddw %%mm0, %%mm0                     \n\t" // 2L4
975 
-                "paddw %%mm1, %%mm1                     \n\t" // 2H4
976 
-                "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
977 
-                "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
978 
-
979 
-                "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
980 
-                "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
981 
-                "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
982 
-                "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
983 
-
984 
-                "movq (%0, %1, 4), %%mm2                \n\t"
985 
-                "movq %%mm2, %%mm3                      \n\t"
986 
-                "punpcklbw %%mm7, %%mm2                 \n\t" // L7
987 
-                "punpckhbw %%mm7, %%mm3                 \n\t" // H7
988 
-
989 
-                "paddw %%mm2, %%mm2                     \n\t" // 2L7
990 
-                "paddw %%mm3, %%mm3                     \n\t" // 2H7
991 
-                "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
992 
-                "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
993 
-
994 
-                "movq (%%"REG_c"), %%mm2                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
995 
-                "movq 8(%%"REG_c"), %%mm3               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
949 
+        "movq (%0, %1, 2), %%mm2                \n\t"
950 
+        "movq %%mm2, %%mm3                      \n\t"
951 
+        "punpcklbw %%mm7, %%mm2                 \n\t" // L5
952 
+        "punpckhbw %%mm7, %%mm3                 \n\t" // H5
953 
+        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
954 
+        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
955 
+        "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
956 
+        "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
957 
+
958 
+        "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
959 
+        "punpcklbw %%mm7, %%mm6                 \n\t" // L6
960 
+        "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
961 
+        "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
962 
+        "punpckhbw %%mm7, %%mm6                 \n\t" // H6
963 
+        "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
964 
+
965 
+        "paddw %%mm0, %%mm0                     \n\t" // 2L4
966 
+        "paddw %%mm1, %%mm1                     \n\t" // 2H4
967 
+        "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
968 
+        "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
969 
+
970 
+        "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
971 
+        "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
972 
+        "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
973 
+        "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
974 
+
975 
+        "movq (%0, %1, 4), %%mm2                \n\t"
976 
+        "movq %%mm2, %%mm3                      \n\t"
977 
+        "punpcklbw %%mm7, %%mm2                 \n\t" // L7
978 
+        "punpckhbw %%mm7, %%mm3                 \n\t" // H7
979 
+
980 
+        "paddw %%mm2, %%mm2                     \n\t" // 2L7
981 
+        "paddw %%mm3, %%mm3                     \n\t" // 2H7
982 
+        "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
983 
+        "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
984 
+
985 
+        "movq (%%"REG_c"), %%mm2                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
986 
+        "movq 8(%%"REG_c"), %%mm3               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
996 987
997 988 
 #ifdef HAVE_MMX2
998 
-                "movq %%mm7, %%mm6                      \n\t" // 0
999 
-                "psubw %%mm0, %%mm6                     \n\t"
1000 
-                "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
1001 
-                "movq %%mm7, %%mm6                      \n\t" // 0
1002 
-                "psubw %%mm1, %%mm6                     \n\t"
1003 
-                "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
1004 
-                "movq %%mm7, %%mm6                      \n\t" // 0
1005 
-                "psubw %%mm2, %%mm6                     \n\t"
1006 
-                "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
1007 
-                "movq %%mm7, %%mm6                      \n\t" // 0
1008 
-                "psubw %%mm3, %%mm6                     \n\t"
1009 
-                "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
989 
+        "movq %%mm7, %%mm6                      \n\t" // 0
990 
+        "psubw %%mm0, %%mm6                     \n\t"
991 
+        "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
992 
+        "movq %%mm7, %%mm6                      \n\t" // 0
993 
+        "psubw %%mm1, %%mm6                     \n\t"
994 
+        "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
995 
+        "movq %%mm7, %%mm6                      \n\t" // 0
996 
+        "psubw %%mm2, %%mm6                     \n\t"
997 
+        "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
998 
+        "movq %%mm7, %%mm6                      \n\t" // 0
999 
+        "psubw %%mm3, %%mm6                     \n\t"
1000 
+        "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
1010 1001 
 #else
1011 
-                "movq %%mm7, %%mm6                      \n\t" // 0
1012 
-                "pcmpgtw %%mm0, %%mm6                   \n\t"
1013 
-                "pxor %%mm6, %%mm0                      \n\t"
1014 
-                "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
1015 
-                "movq %%mm7, %%mm6                      \n\t" // 0
1016 
-                "pcmpgtw %%mm1, %%mm6                   \n\t"
1017 
-                "pxor %%mm6, %%mm1                      \n\t"
1018 
-                "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
1019 
-                "movq %%mm7, %%mm6                      \n\t" // 0
1020 
-                "pcmpgtw %%mm2, %%mm6                   \n\t"
1021 
-                "pxor %%mm6, %%mm2                      \n\t"
1022 
-                "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
1023 
-                "movq %%mm7, %%mm6                      \n\t" // 0
1024 
-                "pcmpgtw %%mm3, %%mm6                   \n\t"
1025 
-                "pxor %%mm6, %%mm3                      \n\t"
1026 
-                "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
1002 
+        "movq %%mm7, %%mm6                      \n\t" // 0
1003 
+        "pcmpgtw %%mm0, %%mm6                   \n\t"
1004 
+        "pxor %%mm6, %%mm0                      \n\t"
1005 
+        "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
1006 
+        "movq %%mm7, %%mm6                      \n\t" // 0
1007 
+        "pcmpgtw %%mm1, %%mm6                   \n\t"
1008 
+        "pxor %%mm6, %%mm1                      \n\t"
1009 
+        "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
1010 
+        "movq %%mm7, %%mm6                      \n\t" // 0
1011 
+        "pcmpgtw %%mm2, %%mm6                   \n\t"
1012 
+        "pxor %%mm6, %%mm2                      \n\t"
1013 
+        "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
1014 
+        "movq %%mm7, %%mm6                      \n\t" // 0
1015 
+        "pcmpgtw %%mm3, %%mm6                   \n\t"
1016 
+        "pxor %%mm6, %%mm3                      \n\t"
1017 
+        "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
1027 1018 
 #endif
1028 1019
1029 1020 
 #ifdef HAVE_MMX2
1030 
-                "pminsw %%mm2, %%mm0                    \n\t"
1031 
-                "pminsw %%mm3, %%mm1                    \n\t"
1021 
+        "pminsw %%mm2, %%mm0                    \n\t"
1022 
+        "pminsw %%mm3, %%mm1                    \n\t"
1032 1023 
 #else
1033 
-                "movq %%mm0, %%mm6                      \n\t"
1034 
-                "psubusw %%mm2, %%mm6                   \n\t"
1035 
-                "psubw %%mm6, %%mm0                     \n\t"
1036 
-                "movq %%mm1, %%mm6                      \n\t"
1037 
-                "psubusw %%mm3, %%mm6                   \n\t"
1038 
-                "psubw %%mm6, %%mm1                     \n\t"
1024 
+        "movq %%mm0, %%mm6                      \n\t"
1025 
+        "psubusw %%mm2, %%mm6                   \n\t"
1026 
+        "psubw %%mm6, %%mm0                     \n\t"
1027 
+        "movq %%mm1, %%mm6                      \n\t"
1028 
+        "psubusw %%mm3, %%mm6                   \n\t"
1029 
+        "psubw %%mm6, %%mm1                     \n\t"
1039 1030 
 #endif
1040 1031
1041 
-                "movd %2, %%mm2                         \n\t" // QP
1042 
-                "punpcklbw %%mm7, %%mm2                 \n\t"
1032 
+        "movd %2, %%mm2                         \n\t" // QP
1033 
+        "punpcklbw %%mm7, %%mm2                 \n\t"
1043 1034
1044 
-                "movq %%mm7, %%mm6                      \n\t" // 0
1045 
-                "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
1046 
-                "pxor %%mm6, %%mm4                      \n\t"
1047 
-                "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
1048 
-                "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
1049 
-                "pxor %%mm7, %%mm5                      \n\t"
1050 
-                "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
1035 
+        "movq %%mm7, %%mm6                      \n\t" // 0
1036 
+        "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
1037 
+        "pxor %%mm6, %%mm4                      \n\t"
1038 
+        "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
1039 
+        "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
1040 
+        "pxor %%mm7, %%mm5                      \n\t"
1041 
+        "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
1051 1042 
 // 100 opcodes
1052 
-                "psllw $3, %%mm2                        \n\t" // 8QP
1053 
-                "movq %%mm2, %%mm3                      \n\t" // 8QP
1054 
-                "pcmpgtw %%mm4, %%mm2                   \n\t"
1055 
-                "pcmpgtw %%mm5, %%mm3                   \n\t"
1056 
-                "pand %%mm2, %%mm4                      \n\t"
1057 
-                "pand %%mm3, %%mm5                      \n\t"
1058 
-
1059 
-
1060 
-                "psubusw %%mm0, %%mm4                   \n\t" // hd
1061 
-                "psubusw %%mm1, %%mm5                   \n\t" // ld
1062 
-
1063 
-
1064 
-                "movq "MANGLE(w05)", %%mm2              \n\t" // 5
1065 
-                "pmullw %%mm2, %%mm4                    \n\t"
1066 
-                "pmullw %%mm2, %%mm5                    \n\t"
1067 
-                "movq "MANGLE(w20)", %%mm2              \n\t" // 32
1068 
-                "paddw %%mm2, %%mm4                     \n\t"
1069 
-                "paddw %%mm2, %%mm5                     \n\t"
1070 
-                "psrlw $6, %%mm4                        \n\t"
1071 
-                "psrlw $6, %%mm5                        \n\t"
1072 
-
1073 
-                "movq 16(%%"REG_c"), %%mm0              \n\t" // L3 - L4
1074 
-                "movq 24(%%"REG_c"), %%mm1              \n\t" // H3 - H4
1075 
-
1076 
-                "pxor %%mm2, %%mm2                      \n\t"
1077 
-                "pxor %%mm3, %%mm3                      \n\t"
1078 
-
1079 
-                "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
1080 
-                "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
1081 
-                "pxor %%mm2, %%mm0                      \n\t"
1082 
-                "pxor %%mm3, %%mm1                      \n\t"
1083 
-                "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
1084 
-                "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
1085 
-                "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
1086 
-                "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
1087 
-
1088 
-                "pxor %%mm6, %%mm2                      \n\t"
1089 
-                "pxor %%mm7, %%mm3                      \n\t"
1090 
-                "pand %%mm2, %%mm4                      \n\t"
1091 
-                "pand %%mm3, %%mm5                      \n\t"
1043 
+        "psllw $3, %%mm2                        \n\t" // 8QP
1044 
+        "movq %%mm2, %%mm3                      \n\t" // 8QP
1045 
+        "pcmpgtw %%mm4, %%mm2                   \n\t"
1046 
+        "pcmpgtw %%mm5, %%mm3                   \n\t"
1047 
+        "pand %%mm2, %%mm4                      \n\t"
1048 
+        "pand %%mm3, %%mm5                      \n\t"
1049 
+
1050 
+
1051 
+        "psubusw %%mm0, %%mm4                   \n\t" // hd
1052 
+        "psubusw %%mm1, %%mm5                   \n\t" // ld
1053 
+
1054 
+
1055 
+        "movq "MANGLE(w05)", %%mm2              \n\t" // 5
1056 
+        "pmullw %%mm2, %%mm4                    \n\t"
1057 
+        "pmullw %%mm2, %%mm5                    \n\t"
1058 
+        "movq "MANGLE(w20)", %%mm2              \n\t" // 32
1059 
+        "paddw %%mm2, %%mm4                     \n\t"
1060 
+        "paddw %%mm2, %%mm5                     \n\t"
1061 
+        "psrlw $6, %%mm4                        \n\t"
1062 
+        "psrlw $6, %%mm5                        \n\t"
1063 
+
1064 
+        "movq 16(%%"REG_c"), %%mm0              \n\t" // L3 - L4
1065 
+        "movq 24(%%"REG_c"), %%mm1              \n\t" // H3 - H4
1066 
+
1067 
+        "pxor %%mm2, %%mm2                      \n\t"
1068 
+        "pxor %%mm3, %%mm3                      \n\t"
1069 
+
1070 
+        "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
1071 
+        "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
1072 
+        "pxor %%mm2, %%mm0                      \n\t"
1073 
+        "pxor %%mm3, %%mm1                      \n\t"
1074 
+        "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
1075 
+        "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
1076 
+        "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
1077 
+        "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
1078 
+
1079 
+        "pxor %%mm6, %%mm2                      \n\t"
1080 
+        "pxor %%mm7, %%mm3                      \n\t"
1081 
+        "pand %%mm2, %%mm4                      \n\t"
1082 
+        "pand %%mm3, %%mm5                      \n\t"
1092 1083
1093 1084 
 #ifdef HAVE_MMX2
1094 
-                "pminsw %%mm0, %%mm4                    \n\t"
1095 
-                "pminsw %%mm1, %%mm5                    \n\t"
1085 
+        "pminsw %%mm0, %%mm4                    \n\t"
1086 
+        "pminsw %%mm1, %%mm5                    \n\t"
1096 1087 
 #else
1097 
-                "movq %%mm4, %%mm2                      \n\t"
1098 
-                "psubusw %%mm0, %%mm2                   \n\t"
1099 
-                "psubw %%mm2, %%mm4                     \n\t"
1100 
-                "movq %%mm5, %%mm2                      \n\t"
1101 
-                "psubusw %%mm1, %%mm2                   \n\t"
1102 
-                "psubw %%mm2, %%mm5                     \n\t"
1088 
+        "movq %%mm4, %%mm2                      \n\t"
1089 
+        "psubusw %%mm0, %%mm2                   \n\t"
1090 
+        "psubw %%mm2, %%mm4                     \n\t"
1091 
+        "movq %%mm5, %%mm2                      \n\t"
1092 
+        "psubusw %%mm1, %%mm2                   \n\t"
1093 
+        "psubw %%mm2, %%mm5                     \n\t"
1103 1094 
 #endif
1104 
-                "pxor %%mm6, %%mm4                      \n\t"
1105 
-                "pxor %%mm7, %%mm5                      \n\t"
1106 
-                "psubw %%mm6, %%mm4                     \n\t"
1107 
-                "psubw %%mm7, %%mm5                     \n\t"
1108 
-                "packsswb %%mm5, %%mm4                  \n\t"
1109 
-                "movq (%0), %%mm0                       \n\t"
1110 
-                "paddb   %%mm4, %%mm0                   \n\t"
1111 
-                "movq %%mm0, (%0)                       \n\t"
1112 
-                "movq (%0, %1), %%mm0                   \n\t"
1113 
-                "psubb %%mm4, %%mm0                     \n\t"
1114 
-                "movq %%mm0, (%0, %1)                   \n\t"
1115 
-
1116 
-                : "+r" (src)
1117 
-                : "r" ((long)stride), "m" (c->pQPb)
1118 
-                : "%"REG_a, "%"REG_c
1119 
-        );
1095 
+        "pxor %%mm6, %%mm4                      \n\t"
1096 
+        "pxor %%mm7, %%mm5                      \n\t"
1097 
+        "psubw %%mm6, %%mm4                     \n\t"
1098 
+        "psubw %%mm7, %%mm5                     \n\t"
1099 
+        "packsswb %%mm5, %%mm4                  \n\t"
1100 
+        "movq (%0), %%mm0                       \n\t"
1101 
+        "paddb   %%mm4, %%mm0                   \n\t"
1102 
+        "movq %%mm0, (%0)                       \n\t"
1103 
+        "movq (%0, %1), %%mm0                   \n\t"
1104 
+        "psubb %%mm4, %%mm0                     \n\t"
1105 
+        "movq %%mm0, (%0, %1)                   \n\t"
1106 
+
1107 
+        : "+r" (src)
1108 
+        : "r" ((long)stride), "m" (c->pQPb)
1109 
+        : "%"REG_a, "%"REG_c
1110 
+    );
1120 1111 
 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1121 
-        const int l1= stride;
1122 
-        const int l2= stride + l1;
1123 
-        const int l3= stride + l2;
1124 
-        const int l4= stride + l3;
1125 
-        const int l5= stride + l4;
1126 
-        const int l6= stride + l5;
1127 
-        const int l7= stride + l6;
1128 
-        const int l8= stride + l7;
1129 
-//        const int l9= stride + l8;
1130 
-        int x;
1131 
-        src+= stride*3;
1132 
-        for(x=0; x<BLOCK_SIZE; x++)
1133 
-        {
1134 
-                const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
1135 
-                if(FFABS(middleEnergy) < 8*c->QP)
1136 
-                {
1137 
-                        const int q=(src[l4] - src[l5])/2;
1138 
-                        const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
1139 
-                        const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
1140 
-
1141 
-                        int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
1142 
-                        d= FFMAX(d, 0);
1143 
-
1144 
-                        d= (5*d + 32) >> 6;
1145 
-                        d*= FFSIGN(-middleEnergy);
1146 
-
1147 
-                        if(q>0)
1148 
-                        {
1149 
-                                d= d<0 ? 0 : d;
1150 
-                                d= d>q ? q : d;
1151 
-                        }
1152 
-                        else
1153 
-                        {
1154 
-                                d= d>0 ? 0 : d;
1155 
-                                d= d<q ? q : d;
1156 
-                        }
1157 
-
1158 
-                        src[l4]-= d;
1159 
-                        src[l5]+= d;
1160 
-                }
1161 
-                src++;
1112 
+    const int l1= stride;
1113 
+    const int l2= stride + l1;
1114 
+    const int l3= stride + l2;
1115 
+    const int l4= stride + l3;
1116 
+    const int l5= stride + l4;
1117 
+    const int l6= stride + l5;
1118 
+    const int l7= stride + l6;
1119 
+    const int l8= stride + l7;
1120 
+//    const int l9= stride + l8;
1121 
+    int x;
1122 
+    src+= stride*3;
1123 
+    for(x=0; x<BLOCK_SIZE; x++){
1124 
+        const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
1125 
+        if(FFABS(middleEnergy) < 8*c->QP){
1126 
+            const int q=(src[l4] - src[l5])/2;
1127 
+            const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
1128 
+            const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
1129 
+
1130 
+            int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
1131 
+            d= FFMAX(d, 0);
1132 
+
1133 
+            d= (5*d + 32) >> 6;
1134 
+            d*= FFSIGN(-middleEnergy);
1135 
+
1136 
+            if(q>0){
1137 
+                d= d<0 ? 0 : d;
1138 
+                d= d>q ? q : d;
1139 
+            }else{
1140 
+                d= d>0 ? 0 : d;
1141 
+                d= d<q ? q : d;
1142 
+            }
1143 
+
1144 
+            src[l4]-= d;
1145 
+            src[l5]+= d;
1162 1146 
         }
1147 
+        src++;
1148 
+    }
1163 1149 
 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1164 1150 
 }
1165 1151 
 #endif //HAVE_ALTIVEC
... ... 
@@ -1168,18 +1147,18 @@ src-=8;
1168 1168 
 static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
1169 1169 
 {
1170 1170 
 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1171 
-        asm volatile(
1172 
-                "pxor %%mm6, %%mm6                      \n\t"
1173 
-                "pcmpeqb %%mm7, %%mm7                   \n\t"
1174 
-                "movq %2, %%mm0                         \n\t"
1175 
-                "punpcklbw %%mm6, %%mm0                 \n\t"
1176 
-                "psrlw $1, %%mm0                        \n\t"
1177 
-                "psubw %%mm7, %%mm0                     \n\t"
1178 
-                "packuswb %%mm0, %%mm0                  \n\t"
1179 
-                "movq %%mm0, %3                         \n\t"
1180 
-
1181 
-                "lea (%0, %1), %%"REG_a"                \n\t"
1182 
-                "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1171 
+    asm volatile(
1172 
+        "pxor %%mm6, %%mm6                      \n\t"
1173 
+        "pcmpeqb %%mm7, %%mm7                   \n\t"
1174 
+        "movq %2, %%mm0                         \n\t"
1175 
+        "punpcklbw %%mm6, %%mm0                 \n\t"
1176 
+        "psrlw $1, %%mm0                        \n\t"
1177 
+        "psubw %%mm7, %%mm0                     \n\t"
1178 
+        "packuswb %%mm0, %%mm0                  \n\t"
1179 
+        "movq %%mm0, %3                         \n\t"
1180 
+
1181 
+        "lea (%0, %1), %%"REG_a"                \n\t"
1182 
+        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1183 1183
1184 1184 
 //        0        1        2        3        4        5        6        7        8        9
1185 1185 
 //        %0        eax        eax+%1        eax+2%1        %0+4%1        edx        edx+%1        edx+2%1        %0+8%1        edx+4%1
... ... 
@@ -1187,17 +1166,17 @@ static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
1187 1187 
 #undef FIND_MIN_MAX
1188 1188 
 #ifdef HAVE_MMX2
1189 1189 
 #define REAL_FIND_MIN_MAX(addr)\
1190 
-                "movq " #addr ", %%mm0                  \n\t"\
1191 
-                "pminub %%mm0, %%mm7                    \n\t"\
1192 
-                "pmaxub %%mm0, %%mm6                    \n\t"
1190 
+        "movq " #addr ", %%mm0                  \n\t"\
1191 
+        "pminub %%mm0, %%mm7                    \n\t"\
1192 
+        "pmaxub %%mm0, %%mm6                    \n\t"
1193 1193 
 #else
1194 1194 
 #define REAL_FIND_MIN_MAX(addr)\
1195 
-                "movq " #addr ", %%mm0                  \n\t"\
1196 
-                "movq %%mm7, %%mm1                      \n\t"\
1197 
-                "psubusb %%mm0, %%mm6                   \n\t"\
1198 
-                "paddb %%mm0, %%mm6                     \n\t"\
1199 
-                "psubusb %%mm0, %%mm1                   \n\t"\
1200 
-                "psubb %%mm1, %%mm7                     \n\t"
1195 
+        "movq " #addr ", %%mm0                  \n\t"\
1196 
+        "movq %%mm7, %%mm1                      \n\t"\
1197 
+        "psubusb %%mm0, %%mm6                   \n\t"\
1198 
+        "paddb %%mm0, %%mm6                     \n\t"\
1199 
+        "psubusb %%mm0, %%mm1                   \n\t"\
1200 
+        "psubb %%mm1, %%mm7                     \n\t"
1201 1201 
 #endif
1202 1202 
 #define FIND_MIN_MAX(addr)  REAL_FIND_MIN_MAX(addr)
1203 1203
... ... 
@@ -1210,155 +1189,155 @@ FIND_MIN_MAX((%%REGd, %1))
1210 1210 
 FIND_MIN_MAX((%%REGd, %1, 2))
1211 1211 
 FIND_MIN_MAX((%0, %1, 8))
1212 1212
1213 
-                "movq %%mm7, %%mm4                      \n\t"
1214 
-                "psrlq $8, %%mm7                        \n\t"
1213 
+        "movq %%mm7, %%mm4                      \n\t"
1214 
+        "psrlq $8, %%mm7                        \n\t"
1215 1215 
 #ifdef HAVE_MMX2
1216 
-                "pminub %%mm4, %%mm7                    \n\t" // min of pixels
1217 
-                "pshufw $0xF9, %%mm7, %%mm4             \n\t"
1218 
-                "pminub %%mm4, %%mm7                    \n\t" // min of pixels
1219 
-                "pshufw $0xFE, %%mm7, %%mm4             \n\t"
1220 
-                "pminub %%mm4, %%mm7                    \n\t"
1216 
+        "pminub %%mm4, %%mm7                    \n\t" // min of pixels
1217 
+        "pshufw $0xF9, %%mm7, %%mm4             \n\t"
1218 
+        "pminub %%mm4, %%mm7                    \n\t" // min of pixels
1219 
+        "pshufw $0xFE, %%mm7, %%mm4             \n\t"
1220 
+        "pminub %%mm4, %%mm7                    \n\t"
1221 1221 
 #else
1222 
-                "movq %%mm7, %%mm1                      \n\t"
1223 
-                "psubusb %%mm4, %%mm1                   \n\t"
1224 
-                "psubb %%mm1, %%mm7                     \n\t"
1225 
-                "movq %%mm7, %%mm4                      \n\t"
1226 
-                "psrlq $16, %%mm7                       \n\t"
1227 
-                "movq %%mm7, %%mm1                      \n\t"
1228 
-                "psubusb %%mm4, %%mm1                   \n\t"
1229 
-                "psubb %%mm1, %%mm7                     \n\t"
1230 
-                "movq %%mm7, %%mm4                      \n\t"
1231 
-                "psrlq $32, %%mm7                       \n\t"
1232 
-                "movq %%mm7, %%mm1                      \n\t"
1233 
-                "psubusb %%mm4, %%mm1                   \n\t"
1234 
-                "psubb %%mm1, %%mm7                     \n\t"
1222 
+        "movq %%mm7, %%mm1                      \n\t"
1223 
+        "psubusb %%mm4, %%mm1                   \n\t"
1224 
+        "psubb %%mm1, %%mm7                     \n\t"
1225 
+        "movq %%mm7, %%mm4                      \n\t"
1226 
+        "psrlq $16, %%mm7                       \n\t"
1227 
+        "movq %%mm7, %%mm1                      \n\t"
1228 
+        "psubusb %%mm4, %%mm1                   \n\t"
1229 
+        "psubb %%mm1, %%mm7                     \n\t"
1230 
+        "movq %%mm7, %%mm4                      \n\t"
1231 
+        "psrlq $32, %%mm7                       \n\t"
1232 
+        "movq %%mm7, %%mm1                      \n\t"
1233 
+        "psubusb %%mm4, %%mm1                   \n\t"
1234 
+        "psubb %%mm1, %%mm7                     \n\t"
1235 1235 
 #endif
1236 1236
1237 1237
1238 
-                "movq %%mm6, %%mm4                      \n\t"
1239 
-                "psrlq $8, %%mm6                        \n\t"
1238 
+        "movq %%mm6, %%mm4                      \n\t"
1239 
+        "psrlq $8, %%mm6                        \n\t"
1240 1240 
 #ifdef HAVE_MMX2
1241 
-                "pmaxub %%mm4, %%mm6                    \n\t" // max of pixels
1242 
-                "pshufw $0xF9, %%mm6, %%mm4             \n\t"
1243 
-                "pmaxub %%mm4, %%mm6                    \n\t"
1244 
-                "pshufw $0xFE, %%mm6, %%mm4             \n\t"
1245 
-                "pmaxub %%mm4, %%mm6                    \n\t"
1241 
+        "pmaxub %%mm4, %%mm6                    \n\t" // max of pixels
1242 
+        "pshufw $0xF9, %%mm6, %%mm4             \n\t"
1243 
+        "pmaxub %%mm4, %%mm6                    \n\t"
1244 
+        "pshufw $0xFE, %%mm6, %%mm4             \n\t"
1245 
+        "pmaxub %%mm4, %%mm6                    \n\t"
1246 1246 
 #else
1247 
-                "psubusb %%mm4, %%mm6                   \n\t"
1248 
-                "paddb %%mm4, %%mm6                     \n\t"
1249 
-                "movq %%mm6, %%mm4                      \n\t"
1250 
-                "psrlq $16, %%mm6                       \n\t"
1251 
-                "psubusb %%mm4, %%mm6                   \n\t"
1252 
-                "paddb %%mm4, %%mm6                     \n\t"
1253 
-                "movq %%mm6, %%mm4                      \n\t"
1254 
-                "psrlq $32, %%mm6                       \n\t"
1255 
-                "psubusb %%mm4, %%mm6                   \n\t"
1256 
-                "paddb %%mm4, %%mm6                     \n\t"
1247 
+        "psubusb %%mm4, %%mm6                   \n\t"
1248 
+        "paddb %%mm4, %%mm6                     \n\t"
1249 
+        "movq %%mm6, %%mm4                      \n\t"
1250 
+        "psrlq $16, %%mm6                       \n\t"
1251 
+        "psubusb %%mm4, %%mm6                   \n\t"
1252 
+        "paddb %%mm4, %%mm6                     \n\t"
1253 
+        "movq %%mm6, %%mm4                      \n\t"
1254 
+        "psrlq $32, %%mm6                       \n\t"
1255 
+        "psubusb %%mm4, %%mm6                   \n\t"
1256 
+        "paddb %%mm4, %%mm6                     \n\t"
1257 1257 
 #endif
1258 
-                "movq %%mm6, %%mm0                      \n\t" // max
1259 
-                "psubb %%mm7, %%mm6                     \n\t" // max - min
1260 
-                "movd %%mm6, %%ecx                      \n\t"
1261 
-                "cmpb "MANGLE(deringThreshold)", %%cl   \n\t"
1262 
-                " jb 1f                                 \n\t"
1263 
-                "lea -24(%%"REG_SP"), %%"REG_c"         \n\t"
1264 
-                "and "ALIGN_MASK", %%"REG_c"            \n\t"
1265 
-                PAVGB(%%mm0, %%mm7)                           // a=(max + min)/2
1266 
-                "punpcklbw %%mm7, %%mm7                 \n\t"
1267 
-                "punpcklbw %%mm7, %%mm7                 \n\t"
1268 
-                "punpcklbw %%mm7, %%mm7                 \n\t"
1269 
-                "movq %%mm7, (%%"REG_c")                \n\t"
1270 
-
1271 
-                "movq (%0), %%mm0                       \n\t" // L10
1272 
-                "movq %%mm0, %%mm1                      \n\t" // L10
1273 
-                "movq %%mm0, %%mm2                      \n\t" // L10
1274 
-                "psllq $8, %%mm1                        \n\t"
1275 
-                "psrlq $8, %%mm2                        \n\t"
1276 
-                "movd -4(%0), %%mm3                     \n\t"
1277 
-                "movd 8(%0), %%mm4                      \n\t"
1278 
-                "psrlq $24, %%mm3                       \n\t"
1279 
-                "psllq $56, %%mm4                       \n\t"
1280 
-                "por %%mm3, %%mm1                       \n\t" // L00
1281 
-                "por %%mm4, %%mm2                       \n\t" // L20
1282 
-                "movq %%mm1, %%mm3                      \n\t" // L00
1283 
-                PAVGB(%%mm2, %%mm1)                           // (L20 + L00)/2
1284 
-                PAVGB(%%mm0, %%mm1)                           // (L20 + L00 + 2L10)/4
1285 
-                "psubusb %%mm7, %%mm0                   \n\t"
1286 
-                "psubusb %%mm7, %%mm2                   \n\t"
1287 
-                "psubusb %%mm7, %%mm3                   \n\t"
1288 
-                "pcmpeqb "MANGLE(b00)", %%mm0           \n\t" // L10 > a ? 0 : -1
1289 
-                "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L20 > a ? 0 : -1
1290 
-                "pcmpeqb "MANGLE(b00)", %%mm3           \n\t" // L00 > a ? 0 : -1
1291 
-                "paddb %%mm2, %%mm0                     \n\t"
1292 
-                "paddb %%mm3, %%mm0                     \n\t"
1293 
-
1294 
-                "movq (%%"REG_a"), %%mm2                \n\t" // L11
1295 
-                "movq %%mm2, %%mm3                      \n\t" // L11
1296 
-                "movq %%mm2, %%mm4                      \n\t" // L11
1297 
-                "psllq $8, %%mm3                        \n\t"
1298 
-                "psrlq $8, %%mm4                        \n\t"
1299 
-                "movd -4(%%"REG_a"), %%mm5              \n\t"
1300 
-                "movd 8(%%"REG_a"), %%mm6               \n\t"
1301 
-                "psrlq $24, %%mm5                       \n\t"
1302 
-                "psllq $56, %%mm6                       \n\t"
1303 
-                "por %%mm5, %%mm3                       \n\t" // L01
1304 
-                "por %%mm6, %%mm4                       \n\t" // L21
1305 
-                "movq %%mm3, %%mm5                      \n\t" // L01
1306 
-                PAVGB(%%mm4, %%mm3)                           // (L21 + L01)/2
1307 
-                PAVGB(%%mm2, %%mm3)                           // (L21 + L01 + 2L11)/4
1308 
-                "psubusb %%mm7, %%mm2                   \n\t"
1309 
-                "psubusb %%mm7, %%mm4                   \n\t"
1310 
-                "psubusb %%mm7, %%mm5                   \n\t"
1311 
-                "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L11 > a ? 0 : -1
1312 
-                "pcmpeqb "MANGLE(b00)", %%mm4           \n\t" // L21 > a ? 0 : -1
1313 
-                "pcmpeqb "MANGLE(b00)", %%mm5           \n\t" // L01 > a ? 0 : -1
1314 
-                "paddb %%mm4, %%mm2                     \n\t"
1315 
-                "paddb %%mm5, %%mm2                     \n\t"
1258 
+        "movq %%mm6, %%mm0                      \n\t" // max
1259 
+        "psubb %%mm7, %%mm6                     \n\t" // max - min
1260 
+        "movd %%mm6, %%ecx                      \n\t"
1261 
+        "cmpb "MANGLE(deringThreshold)", %%cl   \n\t"
1262 
+        " jb 1f                                 \n\t"
1263 
+        "lea -24(%%"REG_SP"), %%"REG_c"         \n\t"
1264 
+        "and "ALIGN_MASK", %%"REG_c"            \n\t"
1265 
+        PAVGB(%%mm0, %%mm7)                           // a=(max + min)/2
1266 
+        "punpcklbw %%mm7, %%mm7                 \n\t"
1267 
+        "punpcklbw %%mm7, %%mm7                 \n\t"
1268 
+        "punpcklbw %%mm7, %%mm7                 \n\t"
1269 
+        "movq %%mm7, (%%"REG_c")                \n\t"
1270 
+
1271 
+        "movq (%0), %%mm0                       \n\t" // L10
1272 
+        "movq %%mm0, %%mm1                      \n\t" // L10
1273 
+        "movq %%mm0, %%mm2                      \n\t" // L10
1274 
+        "psllq $8, %%mm1                        \n\t"
1275 
+        "psrlq $8, %%mm2                        \n\t"
1276 
+        "movd -4(%0), %%mm3                     \n\t"
1277 
+        "movd 8(%0), %%mm4                      \n\t"
1278 
+        "psrlq $24, %%mm3                       \n\t"
1279 
+        "psllq $56, %%mm4                       \n\t"
1280 
+        "por %%mm3, %%mm1                       \n\t" // L00
1281 
+        "por %%mm4, %%mm2                       \n\t" // L20
1282 
+        "movq %%mm1, %%mm3                      \n\t" // L00
1283 
+        PAVGB(%%mm2, %%mm1)                           // (L20 + L00)/2
1284 
+        PAVGB(%%mm0, %%mm1)                           // (L20 + L00 + 2L10)/4
1285 
+        "psubusb %%mm7, %%mm0                   \n\t"
1286 
+        "psubusb %%mm7, %%mm2                   \n\t"
1287 
+        "psubusb %%mm7, %%mm3                   \n\t"
1288 
+        "pcmpeqb "MANGLE(b00)", %%mm0           \n\t" // L10 > a ? 0 : -1
1289 
+        "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L20 > a ? 0 : -1
1290 
+        "pcmpeqb "MANGLE(b00)", %%mm3           \n\t" // L00 > a ? 0 : -1
1291 
+        "paddb %%mm2, %%mm0                     \n\t"
1292 
+        "paddb %%mm3, %%mm0                     \n\t"
1293 
+
1294 
+        "movq (%%"REG_a"), %%mm2                \n\t" // L11
1295 
+        "movq %%mm2, %%mm3                      \n\t" // L11
1296 
+        "movq %%mm2, %%mm4                      \n\t" // L11
1297 
+        "psllq $8, %%mm3                        \n\t"
1298 
+        "psrlq $8, %%mm4                        \n\t"
1299 
+        "movd -4(%%"REG_a"), %%mm5              \n\t"
1300 
+        "movd 8(%%"REG_a"), %%mm6               \n\t"
1301 
+        "psrlq $24, %%mm5                       \n\t"
1302 
+        "psllq $56, %%mm6                       \n\t"
1303 
+        "por %%mm5, %%mm3                       \n\t" // L01
1304 
+        "por %%mm6, %%mm4                       \n\t" // L21
1305 
+        "movq %%mm3, %%mm5                      \n\t" // L01
1306 
+        PAVGB(%%mm4, %%mm3)                           // (L21 + L01)/2
1307 
+        PAVGB(%%mm2, %%mm3)                           // (L21 + L01 + 2L11)/4
1308 
+        "psubusb %%mm7, %%mm2                   \n\t"
1309 
+        "psubusb %%mm7, %%mm4                   \n\t"
1310 
+        "psubusb %%mm7, %%mm5                   \n\t"
1311 
+        "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L11 > a ? 0 : -1
1312 
+        "pcmpeqb "MANGLE(b00)", %%mm4           \n\t" // L21 > a ? 0 : -1
1313 
+        "pcmpeqb "MANGLE(b00)", %%mm5           \n\t" // L01 > a ? 0 : -1
1314 
+        "paddb %%mm4, %%mm2                     \n\t"
1315 
+        "paddb %%mm5, %%mm2                     \n\t"
1316 1316 
 // 0, 2, 3, 1
1317 1317 
 #define REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
1318 
-                "movq " #src ", " #sx "                 \n\t" /* src[0] */\
1319 
-                "movq " #sx ", " #lx "                  \n\t" /* src[0] */\
1320 
-                "movq " #sx ", " #t0 "                  \n\t" /* src[0] */\
1321 
-                "psllq $8, " #lx "                      \n\t"\
1322 
-                "psrlq $8, " #t0 "                      \n\t"\
1323 
-                "movd -4" #src ", " #t1 "               \n\t"\
1324 
-                "psrlq $24, " #t1 "                     \n\t"\
1325 
-                "por " #t1 ", " #lx "                   \n\t" /* src[-1] */\
1326 
-                "movd 8" #src ", " #t1 "                \n\t"\
1327 
-                "psllq $56, " #t1 "                     \n\t"\
1328 
-                "por " #t1 ", " #t0 "                   \n\t" /* src[+1] */\
1329 
-                "movq " #lx ", " #t1 "                  \n\t" /* src[-1] */\
1330 
-                PAVGB(t0, lx)                                 /* (src[-1] + src[+1])/2 */\
1331 
-                PAVGB(sx, lx)                                 /* (src[-1] + 2src[0] + src[+1])/4 */\
1332 
-                PAVGB(lx, pplx)                                     \
1333 
-                "movq " #lx ", 8(%%"REG_c")             \n\t"\
1334 
-                "movq (%%"REG_c"), " #lx "              \n\t"\
1335 
-                "psubusb " #lx ", " #t1 "               \n\t"\
1336 
-                "psubusb " #lx ", " #t0 "               \n\t"\
1337 
-                "psubusb " #lx ", " #sx "               \n\t"\
1338 
-                "movq "MANGLE(b00)", " #lx "            \n\t"\
1339 
-                "pcmpeqb " #lx ", " #t1 "               \n\t" /* src[-1] > a ? 0 : -1*/\
1340 
-                "pcmpeqb " #lx ", " #t0 "               \n\t" /* src[+1] > a ? 0 : -1*/\
1341 
-                "pcmpeqb " #lx ", " #sx "               \n\t" /* src[0]  > a ? 0 : -1*/\
1342 
-                "paddb " #t1 ", " #t0 "                 \n\t"\
1343 
-                "paddb " #t0 ", " #sx "                 \n\t"\
1318 
+        "movq " #src ", " #sx "                 \n\t" /* src[0] */\
1319 
+        "movq " #sx ", " #lx "                  \n\t" /* src[0] */\
1320 
+        "movq " #sx ", " #t0 "                  \n\t" /* src[0] */\
1321 
+        "psllq $8, " #lx "                      \n\t"\
1322 
+        "psrlq $8, " #t0 "                      \n\t"\
1323 
+        "movd -4" #src ", " #t1 "               \n\t"\
1324 
+        "psrlq $24, " #t1 "                     \n\t"\
1325 
+        "por " #t1 ", " #lx "                   \n\t" /* src[-1] */\
1326 
+        "movd 8" #src ", " #t1 "                \n\t"\
1327 
+        "psllq $56, " #t1 "                     \n\t"\
1328 
+        "por " #t1 ", " #t0 "                   \n\t" /* src[+1] */\
1329 
+        "movq " #lx ", " #t1 "                  \n\t" /* src[-1] */\
1330 
+        PAVGB(t0, lx)                                 /* (src[-1] + src[+1])/2 */\
1331 
+        PAVGB(sx, lx)                                 /* (src[-1] + 2src[0] + src[+1])/4 */\
1332 
+        PAVGB(lx, pplx)                                     \
1333 
+        "movq " #lx ", 8(%%"REG_c")             \n\t"\
1334 
+        "movq (%%"REG_c"), " #lx "              \n\t"\
1335 
+        "psubusb " #lx ", " #t1 "               \n\t"\
1336 
+        "psubusb " #lx ", " #t0 "               \n\t"\
1337 
+        "psubusb " #lx ", " #sx "               \n\t"\
1338 
+        "movq "MANGLE(b00)", " #lx "            \n\t"\
1339 
+        "pcmpeqb " #lx ", " #t1 "               \n\t" /* src[-1] > a ? 0 : -1*/\
1340 
+        "pcmpeqb " #lx ", " #t0 "               \n\t" /* src[+1] > a ? 0 : -1*/\
1341 
+        "pcmpeqb " #lx ", " #sx "               \n\t" /* src[0]  > a ? 0 : -1*/\
1342 
+        "paddb " #t1 ", " #t0 "                 \n\t"\
1343 
+        "paddb " #t0 ", " #sx "                 \n\t"\
1344 1344 
 \
1345 
-                PAVGB(plx, pplx)                              /* filtered */\
1346 
-                "movq " #dst ", " #t0 "                 \n\t" /* dst */\
1347 
-                "movq " #t0 ", " #t1 "                  \n\t" /* dst */\
1348 
-                "psubusb %3, " #t0 "                    \n\t"\
1349 
-                "paddusb %3, " #t1 "                    \n\t"\
1350 
-                PMAXUB(t0, pplx)\
1351 
-                PMINUB(t1, pplx, t0)\
1352 
-                "paddb " #sx ", " #ppsx "               \n\t"\
1353 
-                "paddb " #psx ", " #ppsx "              \n\t"\
1354 
-                "#paddb "MANGLE(b02)", " #ppsx "        \n\t"\
1355 
-                "pand "MANGLE(b08)", " #ppsx "          \n\t"\
1356 
-                "pcmpeqb " #lx ", " #ppsx "             \n\t"\
1357 
-                "pand " #ppsx ", " #pplx "              \n\t"\
1358 
-                "pandn " #dst ", " #ppsx "              \n\t"\
1359 
-                "por " #pplx ", " #ppsx "               \n\t"\
1360 
-                "movq " #ppsx ", " #dst "               \n\t"\
1361 
-                "movq 8(%%"REG_c"), " #lx "             \n\t"
1345 
+        PAVGB(plx, pplx)                              /* filtered */\
1346 
+        "movq " #dst ", " #t0 "                 \n\t" /* dst */\
1347 
+        "movq " #t0 ", " #t1 "                  \n\t" /* dst */\
1348 
+        "psubusb %3, " #t0 "                    \n\t"\
1349 
+        "paddusb %3, " #t1 "                    \n\t"\
1350 
+        PMAXUB(t0, pplx)\
1351 
+        PMINUB(t1, pplx, t0)\
1352 
+        "paddb " #sx ", " #ppsx "               \n\t"\
1353 
+        "paddb " #psx ", " #ppsx "              \n\t"\
1354 
+        "#paddb "MANGLE(b02)", " #ppsx "        \n\t"\
1355 
+        "pand "MANGLE(b08)", " #ppsx "          \n\t"\
1356 
+        "pcmpeqb " #lx ", " #ppsx "             \n\t"\
1357 
+        "pand " #ppsx ", " #pplx "              \n\t"\
1358 
+        "pandn " #dst ", " #ppsx "              \n\t"\
1359 
+        "por " #pplx ", " #ppsx "               \n\t"\
1360 
+        "movq " #ppsx ", " #dst "               \n\t"\
1361 
+        "movq 8(%%"REG_c"), " #lx "             \n\t"
1362 1362
1363 1363 
 #define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
1364 1364 
    REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1)
... ... 
@@ -1387,139 +1366,126 @@ DERING_CORE((%%REGd, %1)   ,(%%REGd, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,
1387 1387 
 DERING_CORE((%%REGd, %1, 2),(%0, %1, 8)    ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1388 1388 
 DERING_CORE((%0, %1, 8)    ,(%%REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1389 1389
1390 
-                "1:                        \n\t"
1391 
-                : : "r" (src), "r" ((long)stride), "m" (c->pQPb), "m"(c->pQPb2)
1392 
-                : "%"REG_a, "%"REG_d, "%"REG_c
1393 
-        );
1390 
+        "1:                        \n\t"
1391 
+        : : "r" (src), "r" ((long)stride), "m" (c->pQPb), "m"(c->pQPb2)
1392 
+        : "%"REG_a, "%"REG_d, "%"REG_c
1393 
+    );
1394 1394 
 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1395 
-        int y;
1396 
-        int min=255;
1397 
-        int max=0;
1398 
-        int avg;
1399 
-        uint8_t *p;
1400 
-        int s[10];
1401 
-        const int QP2= c->QP/2 + 1;
1402 
-
1403 
-        for(y=1; y<9; y++)
1404 
-        {
1405 
-                int x;
1406 
-                p= src + stride*y;
1407 
-                for(x=1; x<9; x++)
1408 
-                {
1409 
-                        p++;
1410 
-                        if(*p > max) max= *p;
1411 
-                        if(*p < min) min= *p;
1412 
-                }
1413 
-        }
1414 
-        avg= (min + max + 1)>>1;
1415 
-
1416 
-        if(max - min <deringThreshold) return;
1417 
-
1418 
-        for(y=0; y<10; y++)
1419 
-        {
1420 
-                int t = 0;
1421 
-
1422 
-                if(src[stride*y + 0] > avg) t+= 1;
1423 
-                if(src[stride*y + 1] > avg) t+= 2;
1424 
-                if(src[stride*y + 2] > avg) t+= 4;
1425 
-                if(src[stride*y + 3] > avg) t+= 8;
1426 
-                if(src[stride*y + 4] > avg) t+= 16;
1427 
-                if(src[stride*y + 5] > avg) t+= 32;
1428 
-                if(src[stride*y + 6] > avg) t+= 64;
1429 
-                if(src[stride*y + 7] > avg) t+= 128;
1430 
-                if(src[stride*y + 8] > avg) t+= 256;
1431 
-                if(src[stride*y + 9] > avg) t+= 512;
1432 
-
1433 
-                t |= (~t)<<16;
1434 
-                t &= (t<<1) & (t>>1);
1435 
-                s[y] = t;
1436 
-        }
1437 
-
1438 
-        for(y=1; y<9; y++)
1439 
-        {
1440 
-                int t = s[y-1] & s[y] & s[y+1];
1441 
-                t|= t>>16;
1442 
-                s[y-1]= t;
1395 
+    int y;
1396 
+    int min=255;
1397 
+    int max=0;
1398 
+    int avg;
1399 
+    uint8_t *p;
1400 
+    int s[10];
1401 
+    const int QP2= c->QP/2 + 1;
1402 
+
1403 
+    for(y=1; y<9; y++){
1404 
+        int x;
1405 
+        p= src + stride*y;
1406 
+        for(x=1; x<9; x++){
1407 
+            p++;
1408 
+            if(*p > max) max= *p;
1409 
+            if(*p < min) min= *p;
1443 1410 
         }
1411 
+    }
1412 
+    avg= (min + max + 1)>>1;
1413 
+
1414 
+    if(max - min <deringThreshold) return;
1415 
+
1416 
+    for(y=0; y<10; y++){
1417 
+        int t = 0;
1418 
+
1419 
+        if(src[stride*y + 0] > avg) t+= 1;
1420 
+        if(src[stride*y + 1] > avg) t+= 2;
1421 
+        if(src[stride*y + 2] > avg) t+= 4;
1422 
+        if(src[stride*y + 3] > avg) t+= 8;
1423 
+        if(src[stride*y + 4] > avg) t+= 16;
1424 
+        if(src[stride*y + 5] > avg) t+= 32;
1425 
+        if(src[stride*y + 6] > avg) t+= 64;
1426 
+        if(src[stride*y + 7] > avg) t+= 128;
1427 
+        if(src[stride*y + 8] > avg) t+= 256;
1428 
+        if(src[stride*y + 9] > avg) t+= 512;
1429 
+
1430 
+        t |= (~t)<<16;
1431 
+        t &= (t<<1) & (t>>1);
1432 
+        s[y] = t;
1433 
+    }
1434 
+
1435 
+    for(y=1; y<9; y++){
1436 
+        int t = s[y-1] & s[y] & s[y+1];
1437 
+        t|= t>>16;
1438 
+        s[y-1]= t;
1439 
+    }
1440 
+
1441 
+    for(y=1; y<9; y++){
1442 
+        int x;
1443 
+        int t = s[y-1];
1444 1444
1445 
-        for(y=1; y<9; y++)
1446 
-        {
1447 
-                int x;
1448 
-                int t = s[y-1];
1449 
-
1450 
-                p= src + stride*y;
1451 
-                for(x=1; x<9; x++)
1452 
-                {
1453 
-                        p++;
1454 
-                        if(t & (1<<x))
1455 
-                        {
1456 
-                                int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))
1457 
-                                      +2*(*(p     -1)) + 4*(*p         ) + 2*(*(p     +1))
1458 
-                                      +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));
1459 
-                                f= (f + 8)>>4;
1445 
+        p= src + stride*y;
1446 
+        for(x=1; x<9; x++){
1447 
+            p++;
1448 
+            if(t & (1<<x)){
1449 
+                int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))
1450 
+                      +2*(*(p     -1)) + 4*(*p         ) + 2*(*(p     +1))
1451 
+                      +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));
1452 
+                f= (f + 8)>>4;
1460 1453
1461 1454 
 #ifdef DEBUG_DERING_THRESHOLD
1462 
-                                asm volatile("emms\n\t":);
1463 
-                                {
1464 
-                                static long long numPixels=0;
1465 
-                                if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;
1466 
-//                                if((max-min)<20 || (max-min)*QP<200)
1467 
-//                                if((max-min)*QP < 500)
1468 
-//                                if(max-min<QP/2)
1469 
-                                if(max-min < 20)
1470 
-                                {
1471 
-                                        static int numSkiped=0;
1472 
-                                        static int errorSum=0;
1473 
-                                        static int worstQP=0;
1474 
-                                        static int worstRange=0;
1475 
-                                        static int worstDiff=0;
1476 
-                                        int diff= (f - *p);
1477 
-                                        int absDiff= FFABS(diff);
1478 
-                                        int error= diff*diff;
1479 
-
1480 
-                                        if(x==1 || x==8 || y==1 || y==8) continue;
1481 
-
1482 
-                                        numSkiped++;
1483 
-                                        if(absDiff > worstDiff)
1484 
-                                        {
1485 
-                                                worstDiff= absDiff;
1486 
-                                                worstQP= QP;
1487 
-                                                worstRange= max-min;
1488 
-                                        }
1489 
-                                        errorSum+= error;
1490 
-
1491 
-                                        if(1024LL*1024LL*1024LL % numSkiped == 0)
1492 
-                                        {
1493 
-                                                av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "
1494 
-                                                        "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
1495 
-                                                        (float)errorSum/numSkiped, numSkiped, worstQP, worstRange,
1496 
-                                                        worstDiff, (float)numSkiped/numPixels);
1497 
-                                        }
1498 
-                                }
1499 
-                                }
1500 
-#endif
1501 
-                                if     (*p + QP2 < f) *p= *p + QP2;
1502 
-                                else if(*p - QP2 > f) *p= *p - QP2;
1503 
-                                else *p=f;
1455 
+                    asm volatile("emms\n\t":);
1456 
+                    {
1457 
+                    static long long numPixels=0;
1458 
+                    if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;
1459 
+//                    if((max-min)<20 || (max-min)*QP<200)
1460 
+//                    if((max-min)*QP < 500)
1461 
+//                    if(max-min<QP/2)
1462 
+                    if(max-min < 20){
1463 
+                        static int numSkiped=0;
1464 
+                        static int errorSum=0;
1465 
+                        static int worstQP=0;
1466 
+                        static int worstRange=0;
1467 
+                        static int worstDiff=0;
1468 
+                        int diff= (f - *p);
1469 
+                        int absDiff= FFABS(diff);
1470 
+                        int error= diff*diff;
1471 
+
1472 
+                        if(x==1 || x==8 || y==1 || y==8) continue;
1473 
+
1474 
+                        numSkiped++;
1475 
+                        if(absDiff > worstDiff){
1476 
+                            worstDiff= absDiff;
1477 
+                            worstQP= QP;
1478 
+                            worstRange= max-min;
1504 1479 
                         }
1505 
-                }
1480 
+                        errorSum+= error;
1481 
+
1482 
+                        if(1024LL*1024LL*1024LL % numSkiped == 0){
1483 
+                            av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "
1484 
+                                   "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
1485 
+                                   (float)errorSum/numSkiped, numSkiped, worstQP, worstRange,
1486 
+                                   worstDiff, (float)numSkiped/numPixels);
1487 
+                        }
1488 
+                    }
1489 
+                    }
1490 
+#endif
1491 
+                    if     (*p + QP2 < f) *p= *p + QP2;
1492 
+                    else if(*p - QP2 > f) *p= *p - QP2;
1493 
+                    else *p=f;
1494 
+            }
1506 1495 
         }
1496 
+    }
1507 1497 
 #ifdef DEBUG_DERING_THRESHOLD
1508 
-        if(max-min < 20)
1509 
-        {
1510 
-                for(y=1; y<9; y++)
1511 
-                {
1512 
-                        int x;
1513 
-                        int t = 0;
1514 
-                        p= src + stride*y;
1515 
-                        for(x=1; x<9; x++)
1516 
-                        {
1517 
-                                p++;
1518 
-                                *p = FFMIN(*p + 20, 255);
1519 
-                        }
1520 
-                }
1521 
-//                src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
1498 
+    if(max-min < 20){
1499 
+        for(y=1; y<9; y++){
1500 
+            int x;
1501 
+            int t = 0;
1502 
+            p= src + stride*y;
1503 
+            for(x=1; x<9; x++){
1504 
+                p++;
1505 
+                *p = FFMIN(*p + 20, 255);
1506 
+            }
1522 1507 
         }
1508 
+//        src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
1509 
+    }
1523 1510 
 #endif
1524 1511 
 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1525 1512 
 }
... ... 
@@ -1534,46 +1500,46 @@ DERING_CORE((%0, %1, 8)    ,(%%REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,
1534 1534 
 static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)
1535 1535 
 {
1536 1536 
 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1537 
-        src+= 4*stride;
1538 
-        asm volatile(
1539 
-                "lea (%0, %1), %%"REG_a"                \n\t"
1540 
-                "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
1537 
+    src+= 4*stride;
1538 
+    asm volatile(
1539 
+        "lea (%0, %1), %%"REG_a"                \n\t"
1540 
+        "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"
1541 1541 
 //      0       1       2       3       4       5       6       7       8       9
1542 1542 
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1
1543 1543
1544 
-                "movq (%0), %%mm0                       \n\t"
1545 
-                "movq (%%"REG_a", %1), %%mm1            \n\t"
1546 
-                PAVGB(%%mm1, %%mm0)
1547 
-                "movq %%mm0, (%%"REG_a")                \n\t"
1548 
-                "movq (%0, %1, 4), %%mm0                \n\t"
1549 
-                PAVGB(%%mm0, %%mm1)
1550 
-                "movq %%mm1, (%%"REG_a", %1, 2)         \n\t"
1551 
-                "movq (%%"REG_c", %1), %%mm1            \n\t"
1552 
-                PAVGB(%%mm1, %%mm0)
1553 
-                "movq %%mm0, (%%"REG_c")                \n\t"
1554 
-                "movq (%0, %1, 8), %%mm0                \n\t"
1555 
-                PAVGB(%%mm0, %%mm1)
1556 
-                "movq %%mm1, (%%"REG_c", %1, 2)         \n\t"
1557 
-
1558 
-                : : "r" (src), "r" ((long)stride)
1559 
-                : "%"REG_a, "%"REG_c
1560 
-        );
1544 
+        "movq (%0), %%mm0                       \n\t"
1545 
+        "movq (%%"REG_a", %1), %%mm1            \n\t"
1546 
+        PAVGB(%%mm1, %%mm0)
1547 
+        "movq %%mm0, (%%"REG_a")                \n\t"
1548 
+        "movq (%0, %1, 4), %%mm0                \n\t"
1549 
+        PAVGB(%%mm0, %%mm1)
1550 
+        "movq %%mm1, (%%"REG_a", %1, 2)         \n\t"
1551 
+        "movq (%%"REG_c", %1), %%mm1            \n\t"
1552 
+        PAVGB(%%mm1, %%mm0)
1553 
+        "movq %%mm0, (%%"REG_c")                \n\t"
1554 
+        "movq (%0, %1, 8), %%mm0                \n\t"
1555 
+        PAVGB(%%mm0, %%mm1)
1556 
+        "movq %%mm1, (%%"REG_c", %1, 2)         \n\t"
1557 
+
1558 
+        : : "r" (src), "r" ((long)stride)
1559 
+        : "%"REG_a, "%"REG_c
1560 
+    );
1561 1561 
 #else
1562 
-        int a, b, x;
1563 
-        src+= 4*stride;
1564 
-
1565 
-        for(x=0; x<2; x++){
1566 
-                a= *(uint32_t*)&src[stride*0];
1567 
-                b= *(uint32_t*)&src[stride*2];
1568 
-                *(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1569 
-                a= *(uint32_t*)&src[stride*4];
1570 
-                *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1571 
-                b= *(uint32_t*)&src[stride*6];
1572 
-                *(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1573 
-                a= *(uint32_t*)&src[stride*8];
1574 
-                *(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1575 
-                src += 4;
1576 
-        }
1562 
+    int a, b, x;
1563 
+    src+= 4*stride;
1564 
+
1565 
+    for(x=0; x<2; x++){
1566 
+        a= *(uint32_t*)&src[stride*0];
1567 
+        b= *(uint32_t*)&src[stride*2];
1568 
+        *(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1569 
+        a= *(uint32_t*)&src[stride*4];
1570 
+        *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1571 
+        b= *(uint32_t*)&src[stride*6];
1572 
+        *(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1573 
+        a= *(uint32_t*)&src[stride*8];
1574 
+        *(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1575 
+        src += 4;
1576 
+    }
1577 1577 
 #endif
1578 1578 
 }
1579 1579
... ... 
@@ -1587,37 +1553,37 @@ static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int strid
1587 1587 
 static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)
1588 1588 
 {
1589 1589 
 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1590 
-        src+= stride*3;
1591 
-        asm volatile(
1592 
-                "lea (%0, %1), %%"REG_a"                \n\t"
1593 
-                "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1594 
-                "lea (%%"REG_d", %1, 4), %%"REG_c"      \n\t"
1595 
-                "add %1, %%"REG_c"                      \n\t"
1596 
-                "pxor %%mm7, %%mm7                      \n\t"
1590 
+    src+= stride*3;
1591 
+    asm volatile(
1592 
+        "lea (%0, %1), %%"REG_a"                \n\t"
1593 
+        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1594 
+        "lea (%%"REG_d", %1, 4), %%"REG_c"      \n\t"
1595 
+        "add %1, %%"REG_c"                      \n\t"
1596 
+        "pxor %%mm7, %%mm7                      \n\t"
1597 1597 
 //      0       1       2       3       4       5       6       7       8       9       10
1598 1598 
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
1599 1599
1600 1600 
 #define REAL_DEINT_CUBIC(a,b,c,d,e)\
1601 
-                "movq " #a ", %%mm0                     \n\t"\
1602 
-                "movq " #b ", %%mm1                     \n\t"\
1603 
-                "movq " #d ", %%mm2                     \n\t"\
1604 
-                "movq " #e ", %%mm3                     \n\t"\
1605 
-                PAVGB(%%mm2, %%mm1)                             /* (b+d) /2 */\
1606 
-                PAVGB(%%mm3, %%mm0)                             /* a(a+e) /2 */\
1607 
-                "movq %%mm0, %%mm2                      \n\t"\
1608 
-                "punpcklbw %%mm7, %%mm0                 \n\t"\
1609 
-                "punpckhbw %%mm7, %%mm2                 \n\t"\
1610 
-                "movq %%mm1, %%mm3                      \n\t"\
1611 
-                "punpcklbw %%mm7, %%mm1                 \n\t"\
1612 
-                "punpckhbw %%mm7, %%mm3                 \n\t"\
1613 
-                "psubw %%mm1, %%mm0                     \n\t"   /* L(a+e - (b+d))/2 */\
1614 
-                "psubw %%mm3, %%mm2                     \n\t"   /* H(a+e - (b+d))/2 */\
1615 
-                "psraw $3, %%mm0                        \n\t"   /* L(a+e - (b+d))/16 */\
1616 
-                "psraw $3, %%mm2                        \n\t"   /* H(a+e - (b+d))/16 */\
1617 
-                "psubw %%mm0, %%mm1                     \n\t"   /* L(9b + 9d - a - e)/16 */\
1618 
-                "psubw %%mm2, %%mm3                     \n\t"   /* H(9b + 9d - a - e)/16 */\
1619 
-                "packuswb %%mm3, %%mm1                  \n\t"\
1620 
-                "movq %%mm1, " #c "                     \n\t"
1601 
+        "movq " #a ", %%mm0                     \n\t"\
1602 
+        "movq " #b ", %%mm1                     \n\t"\
1603 
+        "movq " #d ", %%mm2                     \n\t"\
1604 
+        "movq " #e ", %%mm3                     \n\t"\
1605 
+        PAVGB(%%mm2, %%mm1)                             /* (b+d) /2 */\
1606 
+        PAVGB(%%mm3, %%mm0)                             /* a(a+e) /2 */\
1607 
+        "movq %%mm0, %%mm2                      \n\t"\
1608 
+        "punpcklbw %%mm7, %%mm0                 \n\t"\
1609 
+        "punpckhbw %%mm7, %%mm2                 \n\t"\
1610 
+        "movq %%mm1, %%mm3                      \n\t"\
1611 
+        "punpcklbw %%mm7, %%mm1                 \n\t"\
1612 
+        "punpckhbw %%mm7, %%mm3                 \n\t"\
1613 
+        "psubw %%mm1, %%mm0                     \n\t"   /* L(a+e - (b+d))/2 */\
1614 
+        "psubw %%mm3, %%mm2                     \n\t"   /* H(a+e - (b+d))/2 */\
1615 
+        "psraw $3, %%mm0                        \n\t"   /* L(a+e - (b+d))/16 */\
1616 
+        "psraw $3, %%mm2                        \n\t"   /* H(a+e - (b+d))/16 */\
1617 
+        "psubw %%mm0, %%mm1                     \n\t"   /* L(9b + 9d - a - e)/16 */\
1618 
+        "psubw %%mm2, %%mm3                     \n\t"   /* H(9b + 9d - a - e)/16 */\
1619 
+        "packuswb %%mm3, %%mm1                  \n\t"\
1620 
+        "movq %%mm1, " #c "                     \n\t"
1621 1621 
 #define DEINT_CUBIC(a,b,c,d,e)  REAL_DEINT_CUBIC(a,b,c,d,e)
1622 1622
1623 1623 
 DEINT_CUBIC((%0)        , (%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd, %1))
... ... 
@@ -1625,20 +1591,19 @@ DEINT_CUBIC((%%REGa, %1), (%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%0, %1,
1625 1625 
 DEINT_CUBIC((%0, %1, 4) , (%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGc))
1626 1626 
 DEINT_CUBIC((%%REGd, %1), (%0, %1, 8) , (%%REGd, %1, 4), (%%REGc)    , (%%REGc, %1, 2))
1627 1627
1628 
-                : : "r" (src), "r" ((long)stride)
1629 
-                : "%"REG_a, "%"REG_d, "%"REG_c
1630 
-        );
1628 
+        : : "r" (src), "r" ((long)stride)
1629 
+        : "%"REG_a, "%"REG_d, "%"REG_c
1630 
+    );
1631 1631 
 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1632 
-        int x;
1633 
-        src+= stride*3;
1634 
-        for(x=0; x<8; x++)
1635 
-        {
1636 
-                src[stride*3] = CLIP((-src[0]        + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);
1637 
-                src[stride*5] = CLIP((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);
1638 
-                src[stride*7] = CLIP((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);
1639 
-                src[stride*9] = CLIP((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);
1640 
-                src++;
1641 
-        }
1632 
+    int x;
1633 
+    src+= stride*3;
1634 
+    for(x=0; x<8; x++){
1635 
+        src[stride*3] = CLIP((-src[0]        + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);
1636 
+        src[stride*5] = CLIP((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);
1637 
+        src[stride*7] = CLIP((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);
1638 
+        src[stride*9] = CLIP((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);
1639 
+        src++;
1640 
+    }
1642 1641 
 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1643 1642 
 }
1644 1643
... ... 
@@ -1652,42 +1617,42 @@ DEINT_CUBIC((%%REGd, %1), (%0, %1, 8) , (%%REGd, %1, 4), (%%REGc)    , (%%REGc,
1652 1652 
 static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)
1653 1653 
 {
1654 1654 
 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1655 
-        src+= stride*4;
1656 
-        asm volatile(
1657 
-                "lea (%0, %1), %%"REG_a"                \n\t"
1658 
-                "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1659 
-                "pxor %%mm7, %%mm7                      \n\t"
1660 
-                "movq (%2), %%mm0                       \n\t"
1655 
+    src+= stride*4;
1656 
+    asm volatile(
1657 
+        "lea (%0, %1), %%"REG_a"                \n\t"
1658 
+        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1659 
+        "pxor %%mm7, %%mm7                      \n\t"
1660 
+        "movq (%2), %%mm0                       \n\t"
1661 1661 
 //      0       1       2       3       4       5       6       7       8       9       10
1662 1662 
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
1663 1663
1664 1664 
 #define REAL_DEINT_FF(a,b,c,d)\
1665 
-                "movq " #a ", %%mm1                     \n\t"\
1666 
-                "movq " #b ", %%mm2                     \n\t"\
1667 
-                "movq " #c ", %%mm3                     \n\t"\
1668 
-                "movq " #d ", %%mm4                     \n\t"\
1669 
-                PAVGB(%%mm3, %%mm1)                          \
1670 
-                PAVGB(%%mm4, %%mm0)                          \
1671 
-                "movq %%mm0, %%mm3                      \n\t"\
1672 
-                "punpcklbw %%mm7, %%mm0                 \n\t"\
1673 
-                "punpckhbw %%mm7, %%mm3                 \n\t"\
1674 
-                "movq %%mm1, %%mm4                      \n\t"\
1675 
-                "punpcklbw %%mm7, %%mm1                 \n\t"\
1676 
-                "punpckhbw %%mm7, %%mm4                 \n\t"\
1677 
-                "psllw $2, %%mm1                        \n\t"\
1678 
-                "psllw $2, %%mm4                        \n\t"\
1679 
-                "psubw %%mm0, %%mm1                     \n\t"\
1680 
-                "psubw %%mm3, %%mm4                     \n\t"\
1681 
-                "movq %%mm2, %%mm5                      \n\t"\
1682 
-                "movq %%mm2, %%mm0                      \n\t"\
1683 
-                "punpcklbw %%mm7, %%mm2                 \n\t"\
1684 
-                "punpckhbw %%mm7, %%mm5                 \n\t"\
1685 
-                "paddw %%mm2, %%mm1                     \n\t"\
1686 
-                "paddw %%mm5, %%mm4                     \n\t"\
1687 
-                "psraw $2, %%mm1                        \n\t"\
1688 
-                "psraw $2, %%mm4                        \n\t"\
1689 
-                "packuswb %%mm4, %%mm1                  \n\t"\
1690 
-                "movq %%mm1, " #b "                     \n\t"\
1665 
+        "movq " #a ", %%mm1                     \n\t"\
1666 
+        "movq " #b ", %%mm2                     \n\t"\
1667 
+        "movq " #c ", %%mm3                     \n\t"\
1668 
+        "movq " #d ", %%mm4                     \n\t"\
1669 
+        PAVGB(%%mm3, %%mm1)                          \
1670 
+        PAVGB(%%mm4, %%mm0)                          \
1671 
+        "movq %%mm0, %%mm3                      \n\t"\
1672 
+        "punpcklbw %%mm7, %%mm0                 \n\t"\
1673 
+        "punpckhbw %%mm7, %%mm3                 \n\t"\
1674 
+        "movq %%mm1, %%mm4                      \n\t"\
1675 
+        "punpcklbw %%mm7, %%mm1                 \n\t"\
1676 
+        "punpckhbw %%mm7, %%mm4                 \n\t"\
1677 
+        "psllw $2, %%mm1                        \n\t"\
1678 
+        "psllw $2, %%mm4                        \n\t"\
1679 
+        "psubw %%mm0, %%mm1                     \n\t"\
1680 
+        "psubw %%mm3, %%mm4                     \n\t"\
1681 
+        "movq %%mm2, %%mm5                      \n\t"\
1682 
+        "movq %%mm2, %%mm0                      \n\t"\
1683 
+        "punpcklbw %%mm7, %%mm2                 \n\t"\
1684 
+        "punpckhbw %%mm7, %%mm5                 \n\t"\
1685 
+        "paddw %%mm2, %%mm1                     \n\t"\
1686 
+        "paddw %%mm5, %%mm4                     \n\t"\
1687 
+        "psraw $2, %%mm1                        \n\t"\
1688 
+        "psraw $2, %%mm4                        \n\t"\
1689 
+        "packuswb %%mm4, %%mm1                  \n\t"\
1690 
+        "movq %%mm1, " #b "                     \n\t"\
1691 1691
1692 1692 
 #define DEINT_FF(a,b,c,d)  REAL_DEINT_FF(a,b,c,d)
1693 1693
... ... 
@@ -1696,29 +1661,28 @@ DEINT_FF((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd)       )
1696 1696 
 DEINT_FF((%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%%REGd, %1, 2))
1697 1697 
 DEINT_FF((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))
1698 1698
1699 
-                "movq %%mm0, (%2)                       \n\t"
1700 
-                : : "r" (src), "r" ((long)stride), "r"(tmp)
1701 
-                : "%"REG_a, "%"REG_d
1702 
-        );
1699 
+        "movq %%mm0, (%2)                       \n\t"
1700 
+        : : "r" (src), "r" ((long)stride), "r"(tmp)
1701 
+        : "%"REG_a, "%"REG_d
1702 
+    );
1703 1703 
 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1704 
-        int x;
1705 
-        src+= stride*4;
1706 
-        for(x=0; x<8; x++)
1707 
-        {
1708 
-                int t1= tmp[x];
1709 
-                int t2= src[stride*1];
1710 
-
1711 
-                src[stride*1]= CLIP((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);
1712 
-                t1= src[stride*4];
1713 
-                src[stride*3]= CLIP((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);
1714 
-                t2= src[stride*6];
1715 
-                src[stride*5]= CLIP((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);
1716 
-                t1= src[stride*8];
1717 
-                src[stride*7]= CLIP((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);
1718 
-                tmp[x]= t1;
1719 
-
1720 
-                src++;
1721 
-        }
1704 
+    int x;
1705 
+    src+= stride*4;
1706 
+    for(x=0; x<8; x++){
1707 
+        int t1= tmp[x];
1708 
+        int t2= src[stride*1];
1709 
+
1710 
+        src[stride*1]= CLIP((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);
1711 
+        t1= src[stride*4];
1712 
+        src[stride*3]= CLIP((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);
1713 
+        t2= src[stride*6];
1714 
+        src[stride*5]= CLIP((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);
1715 
+        t1= src[stride*8];
1716 
+        src[stride*7]= CLIP((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);
1717 
+        tmp[x]= t1;
1718 
+
1719 
+        src++;
1720 
+    }
1722 1721 
 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1723 1722 
 }
1724 1723
... ... 
@@ -1732,48 +1696,48 @@ DEINT_FF((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))
1732 1732 
 static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)
1733 1733 
 {
1734 1734 
 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1735 
-        src+= stride*4;
1736 
-        asm volatile(
1737 
-                "lea (%0, %1), %%"REG_a"                \n\t"
1738 
-                "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1739 
-                "pxor %%mm7, %%mm7                      \n\t"
1740 
-                "movq (%2), %%mm0                       \n\t"
1741 
-                "movq (%3), %%mm1                       \n\t"
1735 
+    src+= stride*4;
1736 
+    asm volatile(
1737 
+        "lea (%0, %1), %%"REG_a"                \n\t"
1738 
+        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1739 
+        "pxor %%mm7, %%mm7                      \n\t"
1740 
+        "movq (%2), %%mm0                       \n\t"
1741 
+        "movq (%3), %%mm1                       \n\t"
1742 1742 
 //      0       1       2       3       4       5       6       7       8       9       10
1743 1743 
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx
1744 1744
1745 1745 
 #define REAL_DEINT_L5(t1,t2,a,b,c)\
1746 
-                "movq " #a ", %%mm2                     \n\t"\
1747 
-                "movq " #b ", %%mm3                     \n\t"\
1748 
-                "movq " #c ", %%mm4                     \n\t"\
1749 
-                PAVGB(t2, %%mm3)                             \
1750 
-                PAVGB(t1, %%mm4)                             \
1751 
-                "movq %%mm2, %%mm5                      \n\t"\
1752 
-                "movq %%mm2, " #t1 "                    \n\t"\
1753 
-                "punpcklbw %%mm7, %%mm2                 \n\t"\
1754 
-                "punpckhbw %%mm7, %%mm5                 \n\t"\
1755 
-                "movq %%mm2, %%mm6                      \n\t"\
1756 
-                "paddw %%mm2, %%mm2                     \n\t"\
1757 
-                "paddw %%mm6, %%mm2                     \n\t"\
1758 
-                "movq %%mm5, %%mm6                      \n\t"\
1759 
-                "paddw %%mm5, %%mm5                     \n\t"\
1760 
-                "paddw %%mm6, %%mm5                     \n\t"\
1761 
-                "movq %%mm3, %%mm6                      \n\t"\
1762 
-                "punpcklbw %%mm7, %%mm3                 \n\t"\
1763 
-                "punpckhbw %%mm7, %%mm6                 \n\t"\
1764 
-                "paddw %%mm3, %%mm3                     \n\t"\
1765 
-                "paddw %%mm6, %%mm6                     \n\t"\
1766 
-                "paddw %%mm3, %%mm2                     \n\t"\
1767 
-                "paddw %%mm6, %%mm5                     \n\t"\
1768 
-                "movq %%mm4, %%mm6                      \n\t"\
1769 
-                "punpcklbw %%mm7, %%mm4                 \n\t"\
1770 
-                "punpckhbw %%mm7, %%mm6                 \n\t"\
1771 
-                "psubw %%mm4, %%mm2                     \n\t"\
1772 
-                "psubw %%mm6, %%mm5                     \n\t"\
1773 
-                "psraw $2, %%mm2                        \n\t"\
1774 
-                "psraw $2, %%mm5                        \n\t"\
1775 
-                "packuswb %%mm5, %%mm2                  \n\t"\
1776 
-                "movq %%mm2, " #a "                     \n\t"\
1746 
+        "movq " #a ", %%mm2                     \n\t"\
1747 
+        "movq " #b ", %%mm3                     \n\t"\
1748 
+        "movq " #c ", %%mm4                     \n\t"\
1749 
+        PAVGB(t2, %%mm3)                             \
1750 
+        PAVGB(t1, %%mm4)                             \
1751 
+        "movq %%mm2, %%mm5                      \n\t"\
1752 
+        "movq %%mm2, " #t1 "                    \n\t"\
1753 
+        "punpcklbw %%mm7, %%mm2                 \n\t"\
1754 
+        "punpckhbw %%mm7, %%mm5                 \n\t"\
1755 
+        "movq %%mm2, %%mm6                      \n\t"\
1756 
+        "paddw %%mm2, %%mm2                     \n\t"\
1757 
+        "paddw %%mm6, %%mm2                     \n\t"\
1758 
+        "movq %%mm5, %%mm6                      \n\t"\
1759 
+        "paddw %%mm5, %%mm5                     \n\t"\
1760 
+        "paddw %%mm6, %%mm5                     \n\t"\
1761 
+        "movq %%mm3, %%mm6                      \n\t"\
1762 
+        "punpcklbw %%mm7, %%mm3                 \n\t"\
1763 
+        "punpckhbw %%mm7, %%mm6                 \n\t"\
1764 
+        "paddw %%mm3, %%mm3                     \n\t"\
1765 
+        "paddw %%mm6, %%mm6                     \n\t"\
1766 
+        "paddw %%mm3, %%mm2                     \n\t"\
1767 
+        "paddw %%mm6, %%mm5                     \n\t"\
1768 
+        "movq %%mm4, %%mm6                      \n\t"\
1769 
+        "punpcklbw %%mm7, %%mm4                 \n\t"\
1770 
+        "punpckhbw %%mm7, %%mm6                 \n\t"\
1771 
+        "psubw %%mm4, %%mm2                     \n\t"\
1772 
+        "psubw %%mm6, %%mm5                     \n\t"\
1773 
+        "psraw $2, %%mm2                        \n\t"\
1774 
+        "psraw $2, %%mm5                        \n\t"\
1775 
+        "packuswb %%mm5, %%mm2                  \n\t"\
1776 
+        "movq %%mm2, " #a "                     \n\t"\
1777 1777
1778 1778 
 #define DEINT_L5(t1,t2,a,b,c)  REAL_DEINT_L5(t1,t2,a,b,c)
1779 1779
... ... 
@@ -1786,41 +1750,40 @@ DEINT_L5(%%mm1, %%mm0, (%%REGd)       , (%%REGd, %1)   , (%%REGd, %1, 2))
1786 1786 
 DEINT_L5(%%mm0, %%mm1, (%%REGd, %1)   , (%%REGd, %1, 2), (%0, %1, 8)   )
1787 1787 
 DEINT_L5(%%mm1, %%mm0, (%%REGd, %1, 2), (%0, %1, 8)    , (%%REGd, %1, 4))
1788 1788
1789 
-                "movq %%mm0, (%2)                       \n\t"
1790 
-                "movq %%mm1, (%3)                       \n\t"
1791 
-                : : "r" (src), "r" ((long)stride), "r"(tmp), "r"(tmp2)
1792 
-                : "%"REG_a, "%"REG_d
1793 
-        );
1789 
+        "movq %%mm0, (%2)                       \n\t"
1790 
+        "movq %%mm1, (%3)                       \n\t"
1791 
+        : : "r" (src), "r" ((long)stride), "r"(tmp), "r"(tmp2)
1792 
+        : "%"REG_a, "%"REG_d
1793 
+    );
1794 1794 
 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1795 
-        int x;
1796 
-        src+= stride*4;
1797 
-        for(x=0; x<8; x++)
1798 
-        {
1799 
-                int t1= tmp[x];
1800 
-                int t2= tmp2[x];
1801 
-                int t3= src[0];
1802 
-
1803 
-                src[stride*0]= CLIP((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);
1804 
-                t1= src[stride*1];
1805 
-                src[stride*1]= CLIP((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);
1806 
-                t2= src[stride*2];
1807 
-                src[stride*2]= CLIP((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);
1808 
-                t3= src[stride*3];
1809 
-                src[stride*3]= CLIP((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);
1810 
-                t1= src[stride*4];
1811 
-                src[stride*4]= CLIP((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);
1812 
-                t2= src[stride*5];
1813 
-                src[stride*5]= CLIP((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);
1814 
-                t3= src[stride*6];
1815 
-                src[stride*6]= CLIP((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);
1816 
-                t1= src[stride*7];
1817 
-                src[stride*7]= CLIP((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);
1818 
-
1819 
-                tmp[x]= t3;
1820 
-                tmp2[x]= t1;
1821 
-
1822 
-                src++;
1823 
-        }
1795 
+    int x;
1796 
+    src+= stride*4;
1797 
+    for(x=0; x<8; x++){
1798 
+        int t1= tmp[x];
1799 
+        int t2= tmp2[x];
1800 
+        int t3= src[0];
1801 
+
1802 
+        src[stride*0]= CLIP((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);
1803 
+        t1= src[stride*1];
1804 
+        src[stride*1]= CLIP((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);
1805 
+        t2= src[stride*2];
1806 
+        src[stride*2]= CLIP((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);
1807 
+        t3= src[stride*3];
1808 
+        src[stride*3]= CLIP((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);
1809 
+        t1= src[stride*4];
1810 
+        src[stride*4]= CLIP((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);
1811 
+        t2= src[stride*5];
1812 
+        src[stride*5]= CLIP((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);
1813 
+        t3= src[stride*6];
1814 
+        src[stride*6]= CLIP((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);
1815 
+        t1= src[stride*7];
1816 
+        src[stride*7]= CLIP((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);
1817 
+
1818 
+        tmp[x]= t3;
1819 
+        tmp2[x]= t1;
1820 
+
1821 
+        src++;
1822 
+    }
1824 1823 
 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1825 1824 
 }
1826 1825
... ... 
@@ -1834,95 +1797,95 @@ DEINT_L5(%%mm1, %%mm0, (%%REGd, %1, 2), (%0, %1, 8)    , (%%REGd, %1, 4))
1834 1834 
 static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uint8_t *tmp)
1835 1835 
 {
1836 1836 
 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1837 
-        src+= 4*stride;
1838 
-        asm volatile(
1839 
-                "lea (%0, %1), %%"REG_a"                \n\t"
1840 
-                "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1837 
+    src+= 4*stride;
1838 
+    asm volatile(
1839 
+        "lea (%0, %1), %%"REG_a"                \n\t"
1840 
+        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1841 1841 
 //      0       1       2       3       4       5       6       7       8       9
1842 1842 
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
1843 1843
1844 
-                "movq (%2), %%mm0                       \n\t" // L0
1845 
-                "movq (%%"REG_a"), %%mm1                \n\t" // L2
1846 
-                PAVGB(%%mm1, %%mm0)                           // L0+L2
1847 
-                "movq (%0), %%mm2                       \n\t" // L1
1848 
-                PAVGB(%%mm2, %%mm0)
1849 
-                "movq %%mm0, (%0)                       \n\t"
1850 
-                "movq (%%"REG_a", %1), %%mm0            \n\t" // L3
1851 
-                PAVGB(%%mm0, %%mm2)                           // L1+L3
1852 
-                PAVGB(%%mm1, %%mm2)                           // 2L2 + L1 + L3
1853 
-                "movq %%mm2, (%%"REG_a")                \n\t"
1854 
-                "movq (%%"REG_a", %1, 2), %%mm2         \n\t" // L4
1855 
-                PAVGB(%%mm2, %%mm1)                           // L2+L4
1856 
-                PAVGB(%%mm0, %%mm1)                           // 2L3 + L2 + L4
1857 
-                "movq %%mm1, (%%"REG_a", %1)            \n\t"
1858 
-                "movq (%0, %1, 4), %%mm1                \n\t" // L5
1859 
-                PAVGB(%%mm1, %%mm0)                           // L3+L5
1860 
-                PAVGB(%%mm2, %%mm0)                           // 2L4 + L3 + L5
1861 
-                "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
1862 
-                "movq (%%"REG_d"), %%mm0                \n\t" // L6
1863 
-                PAVGB(%%mm0, %%mm2)                           // L4+L6
1864 
-                PAVGB(%%mm1, %%mm2)                           // 2L5 + L4 + L6
1865 
-                "movq %%mm2, (%0, %1, 4)                \n\t"
1866 
-                "movq (%%"REG_d", %1), %%mm2            \n\t" // L7
1867 
-                PAVGB(%%mm2, %%mm1)                           // L5+L7
1868 
-                PAVGB(%%mm0, %%mm1)                           // 2L6 + L5 + L7
1869 
-                "movq %%mm1, (%%"REG_d")                \n\t"
1870 
-                "movq (%%"REG_d", %1, 2), %%mm1         \n\t" // L8
1871 
-                PAVGB(%%mm1, %%mm0)                           // L6+L8
1872 
-                PAVGB(%%mm2, %%mm0)                           // 2L7 + L6 + L8
1873 
-                "movq %%mm0, (%%"REG_d", %1)            \n\t"
1874 
-                "movq (%0, %1, 8), %%mm0                \n\t" // L9
1875 
-                PAVGB(%%mm0, %%mm2)                           // L7+L9
1876 
-                PAVGB(%%mm1, %%mm2)                           // 2L8 + L7 + L9
1877 
-                "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
1878 
-                "movq %%mm1, (%2)                       \n\t"
1879 
-
1880 
-                : : "r" (src), "r" ((long)stride), "r" (tmp)
1881 
-                : "%"REG_a, "%"REG_d
1882 
-        );
1844 
+        "movq (%2), %%mm0                       \n\t" // L0
1845 
+        "movq (%%"REG_a"), %%mm1                \n\t" // L2
1846 
+        PAVGB(%%mm1, %%mm0)                           // L0+L2
1847 
+        "movq (%0), %%mm2                       \n\t" // L1
1848 
+        PAVGB(%%mm2, %%mm0)
1849 
+        "movq %%mm0, (%0)                       \n\t"
1850 
+        "movq (%%"REG_a", %1), %%mm0            \n\t" // L3
1851 
+        PAVGB(%%mm0, %%mm2)                           // L1+L3
1852 
+        PAVGB(%%mm1, %%mm2)                           // 2L2 + L1 + L3
1853 
+        "movq %%mm2, (%%"REG_a")                \n\t"
1854 
+        "movq (%%"REG_a", %1, 2), %%mm2         \n\t" // L4
1855 
+        PAVGB(%%mm2, %%mm1)                           // L2+L4
1856 
+        PAVGB(%%mm0, %%mm1)                           // 2L3 + L2 + L4
1857 
+        "movq %%mm1, (%%"REG_a", %1)            \n\t"
1858 
+        "movq (%0, %1, 4), %%mm1                \n\t" // L5
1859 
+        PAVGB(%%mm1, %%mm0)                           // L3+L5
1860 
+        PAVGB(%%mm2, %%mm0)                           // 2L4 + L3 + L5
1861 
+        "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"
1862 
+        "movq (%%"REG_d"), %%mm0                \n\t" // L6
1863 
+        PAVGB(%%mm0, %%mm2)                           // L4+L6
1864 
+        PAVGB(%%mm1, %%mm2)                           // 2L5 + L4 + L6
1865 
+        "movq %%mm2, (%0, %1, 4)                \n\t"
1866 
+        "movq (%%"REG_d", %1), %%mm2            \n\t" // L7
1867 
+        PAVGB(%%mm2, %%mm1)                           // L5+L7
1868 
+        PAVGB(%%mm0, %%mm1)                           // 2L6 + L5 + L7
1869 
+        "movq %%mm1, (%%"REG_d")                \n\t"
1870 
+        "movq (%%"REG_d", %1, 2), %%mm1         \n\t" // L8
1871 
+        PAVGB(%%mm1, %%mm0)                           // L6+L8
1872 
+        PAVGB(%%mm2, %%mm0)                           // 2L7 + L6 + L8
1873 
+        "movq %%mm0, (%%"REG_d", %1)            \n\t"
1874 
+        "movq (%0, %1, 8), %%mm0                \n\t" // L9
1875 
+        PAVGB(%%mm0, %%mm2)                           // L7+L9
1876 
+        PAVGB(%%mm1, %%mm2)                           // 2L8 + L7 + L9
1877 
+        "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
1878 
+        "movq %%mm1, (%2)                       \n\t"
1879 
+
1880 
+        : : "r" (src), "r" ((long)stride), "r" (tmp)
1881 
+        : "%"REG_a, "%"REG_d
1882 
+    );
1883 1883 
 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1884 
-        int a, b, c, x;
1885 
-        src+= 4*stride;
1886 
-
1887 
-        for(x=0; x<2; x++){
1888 
-                a= *(uint32_t*)&tmp[stride*0];
1889 
-                b= *(uint32_t*)&src[stride*0];
1890 
-                c= *(uint32_t*)&src[stride*1];
1891 
-                a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1892 
-                *(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1893 
-
1894 
-                a= *(uint32_t*)&src[stride*2];
1895 
-                b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1896 
-                *(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1897 
-
1898 
-                b= *(uint32_t*)&src[stride*3];
1899 
-                c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1900 
-                *(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1901 
-
1902 
-                c= *(uint32_t*)&src[stride*4];
1903 
-                a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1904 
-                *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1905 
-
1906 
-                a= *(uint32_t*)&src[stride*5];
1907 
-                b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1908 
-                *(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1909 
-
1910 
-                b= *(uint32_t*)&src[stride*6];
1911 
-                c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1912 
-                *(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1913 
-
1914 
-                c= *(uint32_t*)&src[stride*7];
1915 
-                a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1916 
-                *(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1917 
-
1918 
-                a= *(uint32_t*)&src[stride*8];
1919 
-                b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1920 
-                *(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1921 
-
1922 
-                *(uint32_t*)&tmp[stride*0]= c;
1923 
-                src += 4;
1924 
-                tmp += 4;
1925 
-        }
1884 
+    int a, b, c, x;
1885 
+    src+= 4*stride;
1886 
+
1887 
+    for(x=0; x<2; x++){
1888 
+        a= *(uint32_t*)&tmp[stride*0];
1889 
+        b= *(uint32_t*)&src[stride*0];
1890 
+        c= *(uint32_t*)&src[stride*1];
1891 
+        a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1892 
+        *(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1893 
+
1894 
+        a= *(uint32_t*)&src[stride*2];
1895 
+        b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1896 
+        *(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1897 
+
1898 
+        b= *(uint32_t*)&src[stride*3];
1899 
+        c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1900 
+        *(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1901 
+
1902 
+        c= *(uint32_t*)&src[stride*4];
1903 
+        a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1904 
+        *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1905 
+
1906 
+        a= *(uint32_t*)&src[stride*5];
1907 
+        b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1908 
+        *(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1909 
+
1910 
+        b= *(uint32_t*)&src[stride*6];
1911 
+        c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1912 
+        *(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1913 
+
1914 
+        c= *(uint32_t*)&src[stride*7];
1915 
+        a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1916 
+        *(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1917 
+
1918 
+        a= *(uint32_t*)&src[stride*8];
1919 
+        b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1920 
+        *(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1921 
+
1922 
+        *(uint32_t*)&tmp[stride*0]= c;
1923 
+        src += 4;
1924 
+        tmp += 4;
1925 
+    }
1926 1926 
 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1927 1927 
 }
1928 1928
... ... 
@@ -1935,87 +1898,87 @@ static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uin
1935 1935 
 static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)
1936 1936 
 {
1937 1937 
 #ifdef HAVE_MMX
1938 
-        src+= 4*stride;
1938 
+    src+= 4*stride;
1939 1939 
 #ifdef HAVE_MMX2
1940 
-        asm volatile(
1941 
-                "lea (%0, %1), %%"REG_a"                \n\t"
1942 
-                "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1940 
+    asm volatile(
1941 
+        "lea (%0, %1), %%"REG_a"                \n\t"
1942 
+        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1943 1943 
 //      0       1       2       3       4       5       6       7       8       9
1944 1944 
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
1945 1945
1946 
-                "movq (%0), %%mm0                       \n\t" //
1947 
-                "movq (%%"REG_a", %1), %%mm2            \n\t" //
1948 
-                "movq (%%"REG_a"), %%mm1                \n\t" //
1949 
-                "movq %%mm0, %%mm3                      \n\t"
1950 
-                "pmaxub %%mm1, %%mm0                    \n\t" //
1951 
-                "pminub %%mm3, %%mm1                    \n\t" //
1952 
-                "pmaxub %%mm2, %%mm1                    \n\t" //
1953 
-                "pminub %%mm1, %%mm0                    \n\t"
1954 
-                "movq %%mm0, (%%"REG_a")                \n\t"
1955 
-
1956 
-                "movq (%0, %1, 4), %%mm0                \n\t" //
1957 
-                "movq (%%"REG_a", %1, 2), %%mm1         \n\t" //
1958 
-                "movq %%mm2, %%mm3                      \n\t"
1959 
-                "pmaxub %%mm1, %%mm2                    \n\t" //
1960 
-                "pminub %%mm3, %%mm1                    \n\t" //
1961 
-                "pmaxub %%mm0, %%mm1                    \n\t" //
1962 
-                "pminub %%mm1, %%mm2                    \n\t"
1963 
-                "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"
1964 
-
1965 
-                "movq (%%"REG_d"), %%mm2                \n\t" //
1966 
-                "movq (%%"REG_d", %1), %%mm1            \n\t" //
1967 
-                "movq %%mm2, %%mm3                      \n\t"
1968 
-                "pmaxub %%mm0, %%mm2                    \n\t" //
1969 
-                "pminub %%mm3, %%mm0                    \n\t" //
1970 
-                "pmaxub %%mm1, %%mm0                    \n\t" //
1971 
-                "pminub %%mm0, %%mm2                    \n\t"
1972 
-                "movq %%mm2, (%%"REG_d")                \n\t"
1973 
-
1974 
-                "movq (%%"REG_d", %1, 2), %%mm2         \n\t" //
1975 
-                "movq (%0, %1, 8), %%mm0                \n\t" //
1976 
-                "movq %%mm2, %%mm3                      \n\t"
1977 
-                "pmaxub %%mm0, %%mm2                    \n\t" //
1978 
-                "pminub %%mm3, %%mm0                    \n\t" //
1979 
-                "pmaxub %%mm1, %%mm0                    \n\t" //
1980 
-                "pminub %%mm0, %%mm2                    \n\t"
1981 
-                "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
1982 
-
1983 
-
1984 
-                : : "r" (src), "r" ((long)stride)
1985 
-                : "%"REG_a, "%"REG_d
1986 
-        );
1946 
+        "movq (%0), %%mm0                       \n\t" //
1947 
+        "movq (%%"REG_a", %1), %%mm2            \n\t" //
1948 
+        "movq (%%"REG_a"), %%mm1                \n\t" //
1949 
+        "movq %%mm0, %%mm3                      \n\t"
1950 
+        "pmaxub %%mm1, %%mm0                    \n\t" //
1951 
+        "pminub %%mm3, %%mm1                    \n\t" //
1952 
+        "pmaxub %%mm2, %%mm1                    \n\t" //
1953 
+        "pminub %%mm1, %%mm0                    \n\t"
1954 
+        "movq %%mm0, (%%"REG_a")                \n\t"
1955 
+
1956 
+        "movq (%0, %1, 4), %%mm0                \n\t" //
1957 
+        "movq (%%"REG_a", %1, 2), %%mm1         \n\t" //
1958 
+        "movq %%mm2, %%mm3                      \n\t"
1959 
+        "pmaxub %%mm1, %%mm2                    \n\t" //
1960 
+        "pminub %%mm3, %%mm1                    \n\t" //
1961 
+        "pmaxub %%mm0, %%mm1                    \n\t" //
1962 
+        "pminub %%mm1, %%mm2                    \n\t"
1963 
+        "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"
1964 
+
1965 
+        "movq (%%"REG_d"), %%mm2                \n\t" //
1966 
+        "movq (%%"REG_d", %1), %%mm1            \n\t" //
1967 
+        "movq %%mm2, %%mm3                      \n\t"
1968 
+        "pmaxub %%mm0, %%mm2                    \n\t" //
1969 
+        "pminub %%mm3, %%mm0                    \n\t" //
1970 
+        "pmaxub %%mm1, %%mm0                    \n\t" //
1971 
+        "pminub %%mm0, %%mm2                    \n\t"
1972 
+        "movq %%mm2, (%%"REG_d")                \n\t"
1973 
+
1974 
+        "movq (%%"REG_d", %1, 2), %%mm2         \n\t" //
1975 
+        "movq (%0, %1, 8), %%mm0                \n\t" //
1976 
+        "movq %%mm2, %%mm3                      \n\t"
1977 
+        "pmaxub %%mm0, %%mm2                    \n\t" //
1978 
+        "pminub %%mm3, %%mm0                    \n\t" //
1979 
+        "pmaxub %%mm1, %%mm0                    \n\t" //
1980 
+        "pminub %%mm0, %%mm2                    \n\t"
1981 
+        "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"
1982 
+
1983 
+
1984 
+        : : "r" (src), "r" ((long)stride)
1985 
+        : "%"REG_a, "%"REG_d
1986 
+    );
1987 1987
1988 1988 
 #else // MMX without MMX2
1989 
-        asm volatile(
1990 
-                "lea (%0, %1), %%"REG_a"                \n\t"
1991 
-                "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1989 
+    asm volatile(
1990 
+        "lea (%0, %1), %%"REG_a"                \n\t"
1991 
+        "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"
1992 1992 
 //      0       1       2       3       4       5       6       7       8       9
1993 1993 
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
1994 
-                "pxor %%mm7, %%mm7                      \n\t"
1994 
+        "pxor %%mm7, %%mm7                      \n\t"
1995 1995
1996 1996 
 #define REAL_MEDIAN(a,b,c)\
1997 
-                "movq " #a ", %%mm0                     \n\t"\
1998 
-                "movq " #b ", %%mm2                     \n\t"\
1999 
-                "movq " #c ", %%mm1                     \n\t"\
2000 
-                "movq %%mm0, %%mm3                      \n\t"\
2001 
-                "movq %%mm1, %%mm4                      \n\t"\
2002 
-                "movq %%mm2, %%mm5                      \n\t"\
2003 
-                "psubusb %%mm1, %%mm3                   \n\t"\
2004 
-                "psubusb %%mm2, %%mm4                   \n\t"\
2005 
-                "psubusb %%mm0, %%mm5                   \n\t"\
2006 
-                "pcmpeqb %%mm7, %%mm3                   \n\t"\
2007 
-                "pcmpeqb %%mm7, %%mm4                   \n\t"\
2008 
-                "pcmpeqb %%mm7, %%mm5                   \n\t"\
2009 
-                "movq %%mm3, %%mm6                      \n\t"\
2010 
-                "pxor %%mm4, %%mm3                      \n\t"\
2011 
-                "pxor %%mm5, %%mm4                      \n\t"\
2012 
-                "pxor %%mm6, %%mm5                      \n\t"\
2013 
-                "por %%mm3, %%mm1                       \n\t"\
2014 
-                "por %%mm4, %%mm2                       \n\t"\
2015 
-                "por %%mm5, %%mm0                       \n\t"\
2016 
-                "pand %%mm2, %%mm0                      \n\t"\
2017 
-                "pand %%mm1, %%mm0                      \n\t"\
2018 
-                "movq %%mm0, " #b "                     \n\t"
1997 
+        "movq " #a ", %%mm0                     \n\t"\
1998 
+        "movq " #b ", %%mm2                     \n\t"\
1999 
+        "movq " #c ", %%mm1                     \n\t"\
2000 
+        "movq %%mm0, %%mm3                      \n\t"\
2001 
+        "movq %%mm1, %%mm4                      \n\t"\
2002 
+        "movq %%mm2, %%mm5                      \n\t"\
2003 
+        "psubusb %%mm1, %%mm3                   \n\t"\
2004 
+        "psubusb %%mm2, %%mm4                   \n\t"\
2005 
+        "psubusb %%mm0, %%mm5                   \n\t"\
2006 
+        "pcmpeqb %%mm7, %%mm3                   \n\t"\
2007 
+        "pcmpeqb %%mm7, %%mm4                   \n\t"\
2008 
+        "pcmpeqb %%mm7, %%mm5                   \n\t"\
2009 
+        "movq %%mm3, %%mm6                      \n\t"\
2010 
+        "pxor %%mm4, %%mm3                      \n\t"\
2011 
+        "pxor %%mm5, %%mm4                      \n\t"\
2012 
+        "pxor %%mm6, %%mm5                      \n\t"\
2013 
+        "por %%mm3, %%mm1                       \n\t"\
2014 
+        "por %%mm4, %%mm2                       \n\t"\
2015 
+        "por %%mm5, %%mm0                       \n\t"\
2016 
+        "pand %%mm2, %%mm0                      \n\t"\
2017 
+        "pand %%mm1, %%mm0                      \n\t"\
2018 
+        "movq %%mm0, " #b "                     \n\t"
2019 2019 
 #define MEDIAN(a,b,c)  REAL_MEDIAN(a,b,c)
2020 2020
2021 2021 
 MEDIAN((%0)        , (%%REGa)       , (%%REGa, %1))
... ... 
@@ -2023,31 +1986,29 @@ MEDIAN((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4))
2023 2023 
 MEDIAN((%0, %1, 4) , (%%REGd)       , (%%REGd, %1))
2024 2024 
 MEDIAN((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8))
2025 2025
2026 
-                : : "r" (src), "r" ((long)stride)
2027 
-                : "%"REG_a, "%"REG_d
2028 
-        );
2026 
+        : : "r" (src), "r" ((long)stride)
2027 
+        : "%"REG_a, "%"REG_d
2028 
+    );
2029 2029 
 #endif //HAVE_MMX2
2030 2030 
 #else //HAVE_MMX
2031 
-        int x, y;
2032 
-        src+= 4*stride;
2033 
-        // FIXME - there should be a way to do a few columns in parallel like w/mmx
2034 
-        for(x=0; x<8; x++)
2035 
-        {
2036 
-                uint8_t *colsrc = src;
2037 
-                for (y=0; y<4; y++)
2038 
-                {
2039 
-                        int a, b, c, d, e, f;
2040 
-                        a = colsrc[0       ];
2041 
-                        b = colsrc[stride  ];
2042 
-                        c = colsrc[stride*2];
2043 
-                        d = (a-b)>>31;
2044 
-                        e = (b-c)>>31;
2045 
-                        f = (c-a)>>31;
2046 
-                        colsrc[stride  ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));
2047 
-                        colsrc += stride*2;
2048 
-                }
2049 
-                src++;
2031 
+    int x, y;
2032 
+    src+= 4*stride;
2033 
+    // FIXME - there should be a way to do a few columns in parallel like w/mmx
2034 
+    for(x=0; x<8; x++){
2035 
+        uint8_t *colsrc = src;
2036 
+        for (y=0; y<4; y++){
2037 
+            int a, b, c, d, e, f;
2038 
+            a = colsrc[0       ];
2039 
+            b = colsrc[stride  ];
2040 
+            c = colsrc[stride*2];
2041 
+            d = (a-b)>>31;
2042 
+            e = (b-c)>>31;
2043 
+            f = (c-a)>>31;
2044 
+            colsrc[stride  ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));
2045 
+            colsrc += stride*2;
2050 2046 
         }
2047 
+        src++;
2048 
+    }
2051 2049 
 #endif //HAVE_MMX
2052 2050 
 }
2053 2051
... ... 
@@ -2057,84 +2018,84 @@ MEDIAN((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8))
2057 2057 
  */
2058 2058 
 static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)
2059 2059 
 {
2060 
-        asm(
2061 
-                "lea (%0, %1), %%"REG_a"                \n\t"
2060 
+    asm(
2061 
+        "lea (%0, %1), %%"REG_a"                \n\t"
2062 2062 
 //      0       1       2       3       4       5       6       7       8       9
2063 2063 
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
2064 
-                "movq (%0), %%mm0                       \n\t" // 12345678
2065 
-                "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
2066 
-                "movq %%mm0, %%mm2                      \n\t" // 12345678
2067 
-                "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2068 
-                "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2069 
-
2070 
-                "movq (%%"REG_a", %1), %%mm1            \n\t"
2071 
-                "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
2072 
-                "movq %%mm1, %%mm4                      \n\t"
2073 
-                "punpcklbw %%mm3, %%mm1                 \n\t"
2074 
-                "punpckhbw %%mm3, %%mm4                 \n\t"
2075 
-
2076 
-                "movq %%mm0, %%mm3                      \n\t"
2077 
-                "punpcklwd %%mm1, %%mm0                 \n\t"
2078 
-                "punpckhwd %%mm1, %%mm3                 \n\t"
2079 
-                "movq %%mm2, %%mm1                      \n\t"
2080 
-                "punpcklwd %%mm4, %%mm2                 \n\t"
2081 
-                "punpckhwd %%mm4, %%mm1                 \n\t"
2082 
-
2083 
-                "movd %%mm0, 128(%2)                    \n\t"
2084 
-                "psrlq $32, %%mm0                       \n\t"
2085 
-                "movd %%mm0, 144(%2)                    \n\t"
2086 
-                "movd %%mm3, 160(%2)                    \n\t"
2087 
-                "psrlq $32, %%mm3                       \n\t"
2088 
-                "movd %%mm3, 176(%2)                    \n\t"
2089 
-                "movd %%mm3, 48(%3)                     \n\t"
2090 
-                "movd %%mm2, 192(%2)                    \n\t"
2091 
-                "movd %%mm2, 64(%3)                     \n\t"
2092 
-                "psrlq $32, %%mm2                       \n\t"
2093 
-                "movd %%mm2, 80(%3)                     \n\t"
2094 
-                "movd %%mm1, 96(%3)                     \n\t"
2095 
-                "psrlq $32, %%mm1                       \n\t"
2096 
-                "movd %%mm1, 112(%3)                    \n\t"
2097 
-
2098 
-                "lea (%%"REG_a", %1, 4), %%"REG_a"      \n\t"
2099 
-
2100 
-                "movq (%0, %1, 4), %%mm0                \n\t" // 12345678
2101 
-                "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
2102 
-                "movq %%mm0, %%mm2                      \n\t" // 12345678
2103 
-                "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2104 
-                "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2105 
-
2106 
-                "movq (%%"REG_a", %1), %%mm1            \n\t"
2107 
-                "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
2108 
-                "movq %%mm1, %%mm4                      \n\t"
2109 
-                "punpcklbw %%mm3, %%mm1                 \n\t"
2110 
-                "punpckhbw %%mm3, %%mm4                 \n\t"
2111 
-
2112 
-                "movq %%mm0, %%mm3                      \n\t"
2113 
-                "punpcklwd %%mm1, %%mm0                 \n\t"
2114 
-                "punpckhwd %%mm1, %%mm3                 \n\t"
2115 
-                "movq %%mm2, %%mm1                      \n\t"
2116 
-                "punpcklwd %%mm4, %%mm2                 \n\t"
2117 
-                "punpckhwd %%mm4, %%mm1                 \n\t"
2118 
-
2119 
-                "movd %%mm0, 132(%2)                    \n\t"
2120 
-                "psrlq $32, %%mm0                       \n\t"
2121 
-                "movd %%mm0, 148(%2)                    \n\t"
2122 
-                "movd %%mm3, 164(%2)                    \n\t"
2123 
-                "psrlq $32, %%mm3                       \n\t"
2124 
-                "movd %%mm3, 180(%2)                    \n\t"
2125 
-                "movd %%mm3, 52(%3)                     \n\t"
2126 
-                "movd %%mm2, 196(%2)                    \n\t"
2127 
-                "movd %%mm2, 68(%3)                     \n\t"
2128 
-                "psrlq $32, %%mm2                       \n\t"
2129 
-                "movd %%mm2, 84(%3)                     \n\t"
2130 
-                "movd %%mm1, 100(%3)                    \n\t"
2131 
-                "psrlq $32, %%mm1                       \n\t"
2132 
-                "movd %%mm1, 116(%3)                    \n\t"
2064 
+        "movq (%0), %%mm0                       \n\t" // 12345678
2065 
+        "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
2066 
+        "movq %%mm0, %%mm2                      \n\t" // 12345678
2067 
+        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2068 
+        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2069 
+
2070 
+        "movq (%%"REG_a", %1), %%mm1            \n\t"
2071 
+        "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
2072 
+        "movq %%mm1, %%mm4                      \n\t"
2073 
+        "punpcklbw %%mm3, %%mm1                 \n\t"
2074 
+        "punpckhbw %%mm3, %%mm4                 \n\t"
2075 
+
2076 
+        "movq %%mm0, %%mm3                      \n\t"
2077 
+        "punpcklwd %%mm1, %%mm0                 \n\t"
2078 
+        "punpckhwd %%mm1, %%mm3                 \n\t"
2079 
+        "movq %%mm2, %%mm1                      \n\t"
2080 
+        "punpcklwd %%mm4, %%mm2                 \n\t"
2081 
+        "punpckhwd %%mm4, %%mm1                 \n\t"
2082 
+
2083 
+        "movd %%mm0, 128(%2)                    \n\t"
2084 
+        "psrlq $32, %%mm0                       \n\t"
2085 
+        "movd %%mm0, 144(%2)                    \n\t"
2086 
+        "movd %%mm3, 160(%2)                    \n\t"
2087 
+        "psrlq $32, %%mm3                       \n\t"
2088 
+        "movd %%mm3, 176(%2)                    \n\t"
2089 
+        "movd %%mm3, 48(%3)                     \n\t"
2090 
+        "movd %%mm2, 192(%2)                    \n\t"
2091 
+        "movd %%mm2, 64(%3)                     \n\t"
2092 
+        "psrlq $32, %%mm2                       \n\t"
2093 
+        "movd %%mm2, 80(%3)                     \n\t"
2094 
+        "movd %%mm1, 96(%3)                     \n\t"
2095 
+        "psrlq $32, %%mm1                       \n\t"
2096 
+        "movd %%mm1, 112(%3)                    \n\t"
2097 
+
2098 
+        "lea (%%"REG_a", %1, 4), %%"REG_a"      \n\t"
2099 
+
2100 
+        "movq (%0, %1, 4), %%mm0                \n\t" // 12345678
2101 
+        "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh
2102 
+        "movq %%mm0, %%mm2                      \n\t" // 12345678
2103 
+        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2104 
+        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2105 
+
2106 
+        "movq (%%"REG_a", %1), %%mm1            \n\t"
2107 
+        "movq (%%"REG_a", %1, 2), %%mm3         \n\t"
2108 
+        "movq %%mm1, %%mm4                      \n\t"
2109 
+        "punpcklbw %%mm3, %%mm1                 \n\t"
2110 
+        "punpckhbw %%mm3, %%mm4                 \n\t"
2111 
+
2112 
+        "movq %%mm0, %%mm3                      \n\t"
2113 
+        "punpcklwd %%mm1, %%mm0                 \n\t"
2114 
+        "punpckhwd %%mm1, %%mm3                 \n\t"
2115 
+        "movq %%mm2, %%mm1                      \n\t"
2116 
+        "punpcklwd %%mm4, %%mm2                 \n\t"
2117 
+        "punpckhwd %%mm4, %%mm1                 \n\t"
2118 
+
2119 
+        "movd %%mm0, 132(%2)                    \n\t"
2120 
+        "psrlq $32, %%mm0                       \n\t"
2121 
+        "movd %%mm0, 148(%2)                    \n\t"
2122 
+        "movd %%mm3, 164(%2)                    \n\t"
2123 
+        "psrlq $32, %%mm3                       \n\t"
2124 
+        "movd %%mm3, 180(%2)                    \n\t"
2125 
+        "movd %%mm3, 52(%3)                     \n\t"
2126 
+        "movd %%mm2, 196(%2)                    \n\t"
2127 
+        "movd %%mm2, 68(%3)                     \n\t"
2128 
+        "psrlq $32, %%mm2                       \n\t"
2129 
+        "movd %%mm2, 84(%3)                     \n\t"
2130 
+        "movd %%mm1, 100(%3)                    \n\t"
2131 
+        "psrlq $32, %%mm1                       \n\t"
2132 
+        "movd %%mm1, 116(%3)                    \n\t"
2133 2133
2134 2134
2135 2135 
         :: "r" (src), "r" ((long)srcStride), "r" (dst1), "r" (dst2)
2136 2136 
         : "%"REG_a
2137 
-        );
2137 
+    );
2138 2138 
 }
2139 2139
2140 2140 
 /**
... ... 
@@ -2142,79 +2103,79 @@ static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src
2142 2142 
  */
2143 2143 
 static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)
2144 2144 
 {
2145 
-        asm(
2146 
-                "lea (%0, %1), %%"REG_a"                \n\t"
2147 
-                "lea (%%"REG_a",%1,4), %%"REG_d"        \n\t"
2145 
+    asm(
2146 
+        "lea (%0, %1), %%"REG_a"                \n\t"
2147 
+        "lea (%%"REG_a",%1,4), %%"REG_d"        \n\t"
2148 2148 
 //      0       1       2       3       4       5       6       7       8       9
2149 2149 
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1
2150 
-                "movq (%2), %%mm0                       \n\t" // 12345678
2151 
-                "movq 16(%2), %%mm1                     \n\t" // abcdefgh
2152 
-                "movq %%mm0, %%mm2                      \n\t" // 12345678
2153 
-                "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2154 
-                "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2155 
-
2156 
-                "movq 32(%2), %%mm1                     \n\t"
2157 
-                "movq 48(%2), %%mm3                     \n\t"
2158 
-                "movq %%mm1, %%mm4                      \n\t"
2159 
-                "punpcklbw %%mm3, %%mm1                 \n\t"
2160 
-                "punpckhbw %%mm3, %%mm4                 \n\t"
2161 
-
2162 
-                "movq %%mm0, %%mm3                      \n\t"
2163 
-                "punpcklwd %%mm1, %%mm0                 \n\t"
2164 
-                "punpckhwd %%mm1, %%mm3                 \n\t"
2165 
-                "movq %%mm2, %%mm1                      \n\t"
2166 
-                "punpcklwd %%mm4, %%mm2                 \n\t"
2167 
-                "punpckhwd %%mm4, %%mm1                 \n\t"
2168 
-
2169 
-                "movd %%mm0, (%0)                       \n\t"
2170 
-                "psrlq $32, %%mm0                       \n\t"
2171 
-                "movd %%mm0, (%%"REG_a")                \n\t"
2172 
-                "movd %%mm3, (%%"REG_a", %1)            \n\t"
2173 
-                "psrlq $32, %%mm3                       \n\t"
2174 
-                "movd %%mm3, (%%"REG_a", %1, 2)         \n\t"
2175 
-                "movd %%mm2, (%0, %1, 4)                \n\t"
2176 
-                "psrlq $32, %%mm2                       \n\t"
2177 
-                "movd %%mm2, (%%"REG_d")                \n\t"
2178 
-                "movd %%mm1, (%%"REG_d", %1)            \n\t"
2179 
-                "psrlq $32, %%mm1                       \n\t"
2180 
-                "movd %%mm1, (%%"REG_d", %1, 2)         \n\t"
2181 
-
2182 
-
2183 
-                "movq 64(%2), %%mm0                     \n\t" // 12345678
2184 
-                "movq 80(%2), %%mm1                     \n\t" // abcdefgh
2185 
-                "movq %%mm0, %%mm2                      \n\t" // 12345678
2186 
-                "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2187 
-                "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2188 
-
2189 
-                "movq 96(%2), %%mm1                     \n\t"
2190 
-                "movq 112(%2), %%mm3                    \n\t"
2191 
-                "movq %%mm1, %%mm4                      \n\t"
2192 
-                "punpcklbw %%mm3, %%mm1                 \n\t"
2193 
-                "punpckhbw %%mm3, %%mm4                 \n\t"
2194 
-
2195 
-                "movq %%mm0, %%mm3                      \n\t"
2196 
-                "punpcklwd %%mm1, %%mm0                 \n\t"
2197 
-                "punpckhwd %%mm1, %%mm3                 \n\t"
2198 
-                "movq %%mm2, %%mm1                      \n\t"
2199 
-                "punpcklwd %%mm4, %%mm2                 \n\t"
2200 
-                "punpckhwd %%mm4, %%mm1                 \n\t"
2201 
-
2202 
-                "movd %%mm0, 4(%0)                      \n\t"
2203 
-                "psrlq $32, %%mm0                       \n\t"
2204 
-                "movd %%mm0, 4(%%"REG_a")               \n\t"
2205 
-                "movd %%mm3, 4(%%"REG_a", %1)           \n\t"
2206 
-                "psrlq $32, %%mm3                       \n\t"
2207 
-                "movd %%mm3, 4(%%"REG_a", %1, 2)        \n\t"
2208 
-                "movd %%mm2, 4(%0, %1, 4)               \n\t"
2209 
-                "psrlq $32, %%mm2                       \n\t"
2210 
-                "movd %%mm2, 4(%%"REG_d")               \n\t"
2211 
-                "movd %%mm1, 4(%%"REG_d", %1)           \n\t"
2212 
-                "psrlq $32, %%mm1                       \n\t"
2213 
-                "movd %%mm1, 4(%%"REG_d", %1, 2)        \n\t"
2150 
+        "movq (%2), %%mm0                       \n\t" // 12345678
2151 
+        "movq 16(%2), %%mm1                     \n\t" // abcdefgh
2152 
+        "movq %%mm0, %%mm2                      \n\t" // 12345678
2153 
+        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2154 
+        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2155 
+
2156 
+        "movq 32(%2), %%mm1                     \n\t"
2157 
+        "movq 48(%2), %%mm3                     \n\t"
2158 
+        "movq %%mm1, %%mm4                      \n\t"
2159 
+        "punpcklbw %%mm3, %%mm1                 \n\t"
2160 
+        "punpckhbw %%mm3, %%mm4                 \n\t"
2161 
+
2162 
+        "movq %%mm0, %%mm3                      \n\t"
2163 
+        "punpcklwd %%mm1, %%mm0                 \n\t"
2164 
+        "punpckhwd %%mm1, %%mm3                 \n\t"
2165 
+        "movq %%mm2, %%mm1                      \n\t"
2166 
+        "punpcklwd %%mm4, %%mm2                 \n\t"
2167 
+        "punpckhwd %%mm4, %%mm1                 \n\t"
2168 
+
2169 
+        "movd %%mm0, (%0)                       \n\t"
2170 
+        "psrlq $32, %%mm0                       \n\t"
2171 
+        "movd %%mm0, (%%"REG_a")                \n\t"
2172 
+        "movd %%mm3, (%%"REG_a", %1)            \n\t"
2173 
+        "psrlq $32, %%mm3                       \n\t"
2174 
+        "movd %%mm3, (%%"REG_a", %1, 2)         \n\t"
2175 
+        "movd %%mm2, (%0, %1, 4)                \n\t"
2176 
+        "psrlq $32, %%mm2                       \n\t"
2177 
+        "movd %%mm2, (%%"REG_d")                \n\t"
2178 
+        "movd %%mm1, (%%"REG_d", %1)            \n\t"
2179 
+        "psrlq $32, %%mm1                       \n\t"
2180 
+        "movd %%mm1, (%%"REG_d", %1, 2)         \n\t"
2181 
+
2182 
+
2183 
+        "movq 64(%2), %%mm0                     \n\t" // 12345678
2184 
+        "movq 80(%2), %%mm1                     \n\t" // abcdefgh
2185 
+        "movq %%mm0, %%mm2                      \n\t" // 12345678
2186 
+        "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d
2187 
+        "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h
2188 
+
2189 
+        "movq 96(%2), %%mm1                     \n\t"
2190 
+        "movq 112(%2), %%mm3                    \n\t"
2191 
+        "movq %%mm1, %%mm4                      \n\t"
2192 
+        "punpcklbw %%mm3, %%mm1                 \n\t"
2193 
+        "punpckhbw %%mm3, %%mm4                 \n\t"
2194 
+
2195 
+        "movq %%mm0, %%mm3                      \n\t"
2196 
+        "punpcklwd %%mm1, %%mm0                 \n\t"
2197 
+        "punpckhwd %%mm1, %%mm3                 \n\t"
2198 
+        "movq %%mm2, %%mm1                      \n\t"
2199 
+        "punpcklwd %%mm4, %%mm2                 \n\t"
2200 
+        "punpckhwd %%mm4, %%mm1                 \n\t"
2201 
+
2202 
+        "movd %%mm0, 4(%0)                      \n\t"
2203 
+        "psrlq $32, %%mm0                       \n\t"
2204 
+        "movd %%mm0, 4(%%"REG_a")               \n\t"
2205 
+        "movd %%mm3, 4(%%"REG_a", %1)           \n\t"
2206 
+        "psrlq $32, %%mm3                       \n\t"
2207 
+        "movd %%mm3, 4(%%"REG_a", %1, 2)        \n\t"
2208 
+        "movd %%mm2, 4(%0, %1, 4)               \n\t"
2209 
+        "psrlq $32, %%mm2                       \n\t"
2210 
+        "movd %%mm2, 4(%%"REG_d")               \n\t"
2211 
+        "movd %%mm1, 4(%%"REG_d", %1)           \n\t"
2212 
+        "psrlq $32, %%mm1                       \n\t"
2213 
+        "movd %%mm1, 4(%%"REG_d", %1, 2)        \n\t"
2214 2214
2215 2215 
         :: "r" (dst), "r" ((long)dstStride), "r" (src)
2216 2216 
         : "%"REG_a, "%"REG_d
2217 
-        );
2217 
+    );
2218 2218 
 }
2219 2219 
 #endif //HAVE_MMX
2220 2220 
 //static long test=0;
... ... 
@@ -2223,83 +2184,83 @@ static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)
2223 2223 
 static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
2224 2224 
                                     uint8_t *tempBlured, uint32_t *tempBluredPast, int *maxNoise)
2225 2225 
 {
2226 
-        // to save a register (FIXME do this outside of the loops)
2227 
-        tempBluredPast[127]= maxNoise[0];
2228 
-        tempBluredPast[128]= maxNoise[1];
2229 
-        tempBluredPast[129]= maxNoise[2];
2226 
+    // to save a register (FIXME do this outside of the loops)
2227 
+    tempBluredPast[127]= maxNoise[0];
2228 
+    tempBluredPast[128]= maxNoise[1];
2229 
+    tempBluredPast[129]= maxNoise[2];
2230 2230
2231 2231 
 #define FAST_L2_DIFF
2232 2232 
 //#define L1_DIFF //u should change the thresholds too if u try that one
2233 2233 
 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
2234 
-        asm volatile(
2235 
-                "lea (%2, %2, 2), %%"REG_a"             \n\t" // 3*stride
2236 
-                "lea (%2, %2, 4), %%"REG_d"             \n\t" // 5*stride
2237 
-                "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2234 
+    asm volatile(
2235 
+        "lea (%2, %2, 2), %%"REG_a"             \n\t" // 3*stride
2236 
+        "lea (%2, %2, 4), %%"REG_d"             \n\t" // 5*stride
2237 
+        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2238 2238 
 //      0       1       2       3       4       5       6       7       8       9
2239 2239 
 //      %x      %x+%2   %x+2%2  %x+eax  %x+4%2  %x+edx  %x+2eax %x+ecx  %x+8%2
2240 2240 
 //FIXME reorder?
2241 2241 
 #ifdef L1_DIFF //needs mmx2
2242 
-                "movq (%0), %%mm0                       \n\t" // L0
2243 
-                "psadbw (%1), %%mm0                     \n\t" // |L0-R0|
2244 
-                "movq (%0, %2), %%mm1                   \n\t" // L1
2245 
-                "psadbw (%1, %2), %%mm1                 \n\t" // |L1-R1|
2246 
-                "movq (%0, %2, 2), %%mm2                \n\t" // L2
2247 
-                "psadbw (%1, %2, 2), %%mm2              \n\t" // |L2-R2|
2248 
-                "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2249 
-                "psadbw (%1, %%"REG_a"), %%mm3          \n\t" // |L3-R3|
2250 
-
2251 
-                "movq (%0, %2, 4), %%mm4                \n\t" // L4
2252 
-                "paddw %%mm1, %%mm0                     \n\t"
2253 
-                "psadbw (%1, %2, 4), %%mm4              \n\t" // |L4-R4|
2254 
-                "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
2255 
-                "paddw %%mm2, %%mm0                     \n\t"
2256 
-                "psadbw (%1, %%"REG_d"), %%mm5          \n\t" // |L5-R5|
2257 
-                "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
2258 
-                "paddw %%mm3, %%mm0                     \n\t"
2259 
-                "psadbw (%1, %%"REG_a", 2), %%mm6       \n\t" // |L6-R6|
2260 
-                "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
2261 
-                "paddw %%mm4, %%mm0                     \n\t"
2262 
-                "psadbw (%1, %%"REG_c"), %%mm7          \n\t" // |L7-R7|
2263 
-                "paddw %%mm5, %%mm6                     \n\t"
2264 
-                "paddw %%mm7, %%mm6                     \n\t"
2265 
-                "paddw %%mm6, %%mm0                     \n\t"
2242 
+        "movq (%0), %%mm0                       \n\t" // L0
2243 
+        "psadbw (%1), %%mm0                     \n\t" // |L0-R0|
2244 
+        "movq (%0, %2), %%mm1                   \n\t" // L1
2245 
+        "psadbw (%1, %2), %%mm1                 \n\t" // |L1-R1|
2246 
+        "movq (%0, %2, 2), %%mm2                \n\t" // L2
2247 
+        "psadbw (%1, %2, 2), %%mm2              \n\t" // |L2-R2|
2248 
+        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2249 
+        "psadbw (%1, %%"REG_a"), %%mm3          \n\t" // |L3-R3|
2250 
+
2251 
+        "movq (%0, %2, 4), %%mm4                \n\t" // L4
2252 
+        "paddw %%mm1, %%mm0                     \n\t"
2253 
+        "psadbw (%1, %2, 4), %%mm4              \n\t" // |L4-R4|
2254 
+        "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
2255 
+        "paddw %%mm2, %%mm0                     \n\t"
2256 
+        "psadbw (%1, %%"REG_d"), %%mm5          \n\t" // |L5-R5|
2257 
+        "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
2258 
+        "paddw %%mm3, %%mm0                     \n\t"
2259 
+        "psadbw (%1, %%"REG_a", 2), %%mm6       \n\t" // |L6-R6|
2260 
+        "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
2261 
+        "paddw %%mm4, %%mm0                     \n\t"
2262 
+        "psadbw (%1, %%"REG_c"), %%mm7          \n\t" // |L7-R7|
2263 
+        "paddw %%mm5, %%mm6                     \n\t"
2264 
+        "paddw %%mm7, %%mm6                     \n\t"
2265 
+        "paddw %%mm6, %%mm0                     \n\t"
2266 2266 
 #else //L1_DIFF
2267 2267 
 #if defined (FAST_L2_DIFF)
2268 
-                "pcmpeqb %%mm7, %%mm7                   \n\t"
2269 
-                "movq "MANGLE(b80)", %%mm6              \n\t"
2270 
-                "pxor %%mm0, %%mm0                      \n\t"
2268 
+        "pcmpeqb %%mm7, %%mm7                   \n\t"
2269 
+        "movq "MANGLE(b80)", %%mm6              \n\t"
2270 
+        "pxor %%mm0, %%mm0                      \n\t"
2271 2271 
 #define REAL_L2_DIFF_CORE(a, b)\
2272 
-                "movq " #a ", %%mm5                     \n\t"\
2273 
-                "movq " #b ", %%mm2                     \n\t"\
2274 
-                "pxor %%mm7, %%mm2                      \n\t"\
2275 
-                PAVGB(%%mm2, %%mm5)\
2276 
-                "paddb %%mm6, %%mm5                     \n\t"\
2277 
-                "movq %%mm5, %%mm2                      \n\t"\
2278 
-                "psllw $8, %%mm5                        \n\t"\
2279 
-                "pmaddwd %%mm5, %%mm5                   \n\t"\
2280 
-                "pmaddwd %%mm2, %%mm2                   \n\t"\
2281 
-                "paddd %%mm2, %%mm5                     \n\t"\
2282 
-                "psrld $14, %%mm5                       \n\t"\
2283 
-                "paddd %%mm5, %%mm0                     \n\t"
2272 
+        "movq " #a ", %%mm5                     \n\t"\
2273 
+        "movq " #b ", %%mm2                     \n\t"\
2274 
+        "pxor %%mm7, %%mm2                      \n\t"\
2275 
+        PAVGB(%%mm2, %%mm5)\
2276 
+        "paddb %%mm6, %%mm5                     \n\t"\
2277 
+        "movq %%mm5, %%mm2                      \n\t"\
2278 
+        "psllw $8, %%mm5                        \n\t"\
2279 
+        "pmaddwd %%mm5, %%mm5                   \n\t"\
2280 
+        "pmaddwd %%mm2, %%mm2                   \n\t"\
2281 
+        "paddd %%mm2, %%mm5                     \n\t"\
2282 
+        "psrld $14, %%mm5                       \n\t"\
2283 
+        "paddd %%mm5, %%mm0                     \n\t"
2284 2284
2285 2285 
 #else //defined (FAST_L2_DIFF)
2286 
-                "pxor %%mm7, %%mm7                      \n\t"
2287 
-                "pxor %%mm0, %%mm0                      \n\t"
2286 
+        "pxor %%mm7, %%mm7                      \n\t"
2287 
+        "pxor %%mm0, %%mm0                      \n\t"
2288 2288 
 #define REAL_L2_DIFF_CORE(a, b)\
2289 
-                "movq " #a ", %%mm5                     \n\t"\
2290 
-                "movq " #b ", %%mm2                     \n\t"\
2291 
-                "movq %%mm5, %%mm1                      \n\t"\
2292 
-                "movq %%mm2, %%mm3                      \n\t"\
2293 
-                "punpcklbw %%mm7, %%mm5                 \n\t"\
2294 
-                "punpckhbw %%mm7, %%mm1                 \n\t"\
2295 
-                "punpcklbw %%mm7, %%mm2                 \n\t"\
2296 
-                "punpckhbw %%mm7, %%mm3                 \n\t"\
2297 
-                "psubw %%mm2, %%mm5                     \n\t"\
2298 
-                "psubw %%mm3, %%mm1                     \n\t"\
2299 
-                "pmaddwd %%mm5, %%mm5                   \n\t"\
2300 
-                "pmaddwd %%mm1, %%mm1                   \n\t"\
2301 
-                "paddd %%mm1, %%mm5                     \n\t"\
2302 
-                "paddd %%mm5, %%mm0                     \n\t"
2289 
+        "movq " #a ", %%mm5                     \n\t"\
2290 
+        "movq " #b ", %%mm2                     \n\t"\
2291 
+        "movq %%mm5, %%mm1                      \n\t"\
2292 
+        "movq %%mm2, %%mm3                      \n\t"\
2293 
+        "punpcklbw %%mm7, %%mm5                 \n\t"\
2294 
+        "punpckhbw %%mm7, %%mm1                 \n\t"\
2295 
+        "punpcklbw %%mm7, %%mm2                 \n\t"\
2296 
+        "punpckhbw %%mm7, %%mm3                 \n\t"\
2297 
+        "psubw %%mm2, %%mm5                     \n\t"\
2298 
+        "psubw %%mm3, %%mm1                     \n\t"\
2299 
+        "pmaddwd %%mm5, %%mm5                   \n\t"\
2300 
+        "pmaddwd %%mm1, %%mm1                   \n\t"\
2301 
+        "paddd %%mm1, %%mm5                     \n\t"\
2302 
+        "paddd %%mm5, %%mm0                     \n\t"
2303 2303
2304 2304 
 #endif //defined (FAST_L2_DIFF)
2305 2305
... ... 
@@ -2316,239 +2277,237 @@ L2_DIFF_CORE((%0, %%REGc)  , (%1, %%REGc))
2316 2316
2317 2317 
 #endif //L1_DIFF
2318 2318
2319 
-                "movq %%mm0, %%mm4                      \n\t"
2320 
-                "psrlq $32, %%mm0                       \n\t"
2321 
-                "paddd %%mm0, %%mm4                     \n\t"
2322 
-                "movd %%mm4, %%ecx                      \n\t"
2323 
-                "shll $2, %%ecx                         \n\t"
2324 
-                "mov %3, %%"REG_d"                      \n\t"
2325 
-                "addl -4(%%"REG_d"), %%ecx              \n\t"
2326 
-                "addl 4(%%"REG_d"), %%ecx               \n\t"
2327 
-                "addl -1024(%%"REG_d"), %%ecx           \n\t"
2328 
-                "addl $4, %%ecx                         \n\t"
2329 
-                "addl 1024(%%"REG_d"), %%ecx            \n\t"
2330 
-                "shrl $3, %%ecx                         \n\t"
2331 
-                "movl %%ecx, (%%"REG_d")                \n\t"
2332 
-
2333 
-//                "mov %3, %%"REG_c"                      \n\t"
2334 
-//                "mov %%"REG_c", test                    \n\t"
2335 
-//                "jmp 4f                                 \n\t"
2336 
-                "cmpl 512(%%"REG_d"), %%ecx             \n\t"
2337 
-                " jb 2f                                 \n\t"
2338 
-                "cmpl 516(%%"REG_d"), %%ecx             \n\t"
2339 
-                " jb 1f                                 \n\t"
2340 
-
2341 
-                "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2342 
-                "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2343 
-                "movq (%0), %%mm0                       \n\t" // L0
2344 
-                "movq (%0, %2), %%mm1                   \n\t" // L1
2345 
-                "movq (%0, %2, 2), %%mm2                \n\t" // L2
2346 
-                "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2347 
-                "movq (%0, %2, 4), %%mm4                \n\t" // L4
2348 
-                "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
2349 
-                "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
2350 
-                "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
2351 
-                "movq %%mm0, (%1)                       \n\t" // L0
2352 
-                "movq %%mm1, (%1, %2)                   \n\t" // L1
2353 
-                "movq %%mm2, (%1, %2, 2)                \n\t" // L2
2354 
-                "movq %%mm3, (%1, %%"REG_a")            \n\t" // L3
2355 
-                "movq %%mm4, (%1, %2, 4)                \n\t" // L4
2356 
-                "movq %%mm5, (%1, %%"REG_d")            \n\t" // L5
2357 
-                "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // L6
2358 
-                "movq %%mm7, (%1, %%"REG_c")            \n\t" // L7
2359 
-                "jmp 4f                                 \n\t"
2360 
-
2361 
-                "1:                                     \n\t"
2362 
-                "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2363 
-                "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2364 
-                "movq (%0), %%mm0                       \n\t" // L0
2365 
-                PAVGB((%1), %%mm0)                            // L0
2366 
-                "movq (%0, %2), %%mm1                   \n\t" // L1
2367 
-                PAVGB((%1, %2), %%mm1)                        // L1
2368 
-                "movq (%0, %2, 2), %%mm2                \n\t" // L2
2369 
-                PAVGB((%1, %2, 2), %%mm2)                     // L2
2370 
-                "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2371 
-                PAVGB((%1, %%REGa), %%mm3)                    // L3
2372 
-                "movq (%0, %2, 4), %%mm4                \n\t" // L4
2373 
-                PAVGB((%1, %2, 4), %%mm4)                     // L4
2374 
-                "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
2375 
-                PAVGB((%1, %%REGd), %%mm5)                    // L5
2376 
-                "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
2377 
-                PAVGB((%1, %%REGa, 2), %%mm6)                 // L6
2378 
-                "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
2379 
-                PAVGB((%1, %%REGc), %%mm7)                    // L7
2380 
-                "movq %%mm0, (%1)                       \n\t" // R0
2381 
-                "movq %%mm1, (%1, %2)                   \n\t" // R1
2382 
-                "movq %%mm2, (%1, %2, 2)                \n\t" // R2
2383 
-                "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
2384 
-                "movq %%mm4, (%1, %2, 4)                \n\t" // R4
2385 
-                "movq %%mm5, (%1, %%"REG_d")            \n\t" // R5
2386 
-                "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // R6
2387 
-                "movq %%mm7, (%1, %%"REG_c")            \n\t" // R7
2388 
-                "movq %%mm0, (%0)                       \n\t" // L0
2389 
-                "movq %%mm1, (%0, %2)                   \n\t" // L1
2390 
-                "movq %%mm2, (%0, %2, 2)                \n\t" // L2
2391 
-                "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
2392 
-                "movq %%mm4, (%0, %2, 4)                \n\t" // L4
2393 
-                "movq %%mm5, (%0, %%"REG_d")            \n\t" // L5
2394 
-                "movq %%mm6, (%0, %%"REG_a", 2)         \n\t" // L6
2395 
-                "movq %%mm7, (%0, %%"REG_c")            \n\t" // L7
2396 
-                "jmp 4f                                 \n\t"
2397 
-
2398 
-                "2:                                     \n\t"
2399 
-                "cmpl 508(%%"REG_d"), %%ecx             \n\t"
2400 
-                " jb 3f                                 \n\t"
2401 
-
2402 
-                "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2403 
-                "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2404 
-                "movq (%0), %%mm0                       \n\t" // L0
2405 
-                "movq (%0, %2), %%mm1                   \n\t" // L1
2406 
-                "movq (%0, %2, 2), %%mm2                \n\t" // L2
2407 
-                "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2408 
-                "movq (%1), %%mm4                       \n\t" // R0
2409 
-                "movq (%1, %2), %%mm5                   \n\t" // R1
2410 
-                "movq (%1, %2, 2), %%mm6                \n\t" // R2
2411 
-                "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
2412 
-                PAVGB(%%mm4, %%mm0)
2413 
-                PAVGB(%%mm5, %%mm1)
2414 
-                PAVGB(%%mm6, %%mm2)
2415 
-                PAVGB(%%mm7, %%mm3)
2416 
-                PAVGB(%%mm4, %%mm0)
2417 
-                PAVGB(%%mm5, %%mm1)
2418 
-                PAVGB(%%mm6, %%mm2)
2419 
-                PAVGB(%%mm7, %%mm3)
2420 
-                "movq %%mm0, (%1)                       \n\t" // R0
2421 
-                "movq %%mm1, (%1, %2)                   \n\t" // R1
2422 
-                "movq %%mm2, (%1, %2, 2)                \n\t" // R2
2423 
-                "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
2424 
-                "movq %%mm0, (%0)                       \n\t" // L0
2425 
-                "movq %%mm1, (%0, %2)                   \n\t" // L1
2426 
-                "movq %%mm2, (%0, %2, 2)                \n\t" // L2
2427 
-                "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
2428 
-
2429 
-                "movq (%0, %2, 4), %%mm0                \n\t" // L4
2430 
-                "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
2431 
-                "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
2432 
-                "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
2433 
-                "movq (%1, %2, 4), %%mm4                \n\t" // R4
2434 
-                "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
2435 
-                "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
2436 
-                "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
2437 
-                PAVGB(%%mm4, %%mm0)
2438 
-                PAVGB(%%mm5, %%mm1)
2439 
-                PAVGB(%%mm6, %%mm2)
2440 
-                PAVGB(%%mm7, %%mm3)
2441 
-                PAVGB(%%mm4, %%mm0)
2442 
-                PAVGB(%%mm5, %%mm1)
2443 
-                PAVGB(%%mm6, %%mm2)
2444 
-                PAVGB(%%mm7, %%mm3)
2445 
-                "movq %%mm0, (%1, %2, 4)                \n\t" // R4
2446 
-                "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
2447 
-                "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
2448 
-                "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
2449 
-                "movq %%mm0, (%0, %2, 4)                \n\t" // L4
2450 
-                "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
2451 
-                "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
2452 
-                "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
2453 
-                "jmp 4f                                 \n\t"
2454 
-
2455 
-                "3:                                     \n\t"
2456 
-                "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2457 
-                "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2458 
-                "movq (%0), %%mm0                       \n\t" // L0
2459 
-                "movq (%0, %2), %%mm1                   \n\t" // L1
2460 
-                "movq (%0, %2, 2), %%mm2                \n\t" // L2
2461 
-                "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2462 
-                "movq (%1), %%mm4                       \n\t" // R0
2463 
-                "movq (%1, %2), %%mm5                   \n\t" // R1
2464 
-                "movq (%1, %2, 2), %%mm6                \n\t" // R2
2465 
-                "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
2466 
-                PAVGB(%%mm4, %%mm0)
2467 
-                PAVGB(%%mm5, %%mm1)
2468 
-                PAVGB(%%mm6, %%mm2)
2469 
-                PAVGB(%%mm7, %%mm3)
2470 
-                PAVGB(%%mm4, %%mm0)
2471 
-                PAVGB(%%mm5, %%mm1)
2472 
-                PAVGB(%%mm6, %%mm2)
2473 
-                PAVGB(%%mm7, %%mm3)
2474 
-                PAVGB(%%mm4, %%mm0)
2475 
-                PAVGB(%%mm5, %%mm1)
2476 
-                PAVGB(%%mm6, %%mm2)
2477 
-                PAVGB(%%mm7, %%mm3)
2478 
-                "movq %%mm0, (%1)                       \n\t" // R0
2479 
-                "movq %%mm1, (%1, %2)                   \n\t" // R1
2480 
-                "movq %%mm2, (%1, %2, 2)                \n\t" // R2
2481 
-                "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
2482 
-                "movq %%mm0, (%0)                       \n\t" // L0
2483 
-                "movq %%mm1, (%0, %2)                   \n\t" // L1
2484 
-                "movq %%mm2, (%0, %2, 2)                \n\t" // L2
2485 
-                "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
2486 
-
2487 
-                "movq (%0, %2, 4), %%mm0                \n\t" // L4
2488 
-                "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
2489 
-                "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
2490 
-                "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
2491 
-                "movq (%1, %2, 4), %%mm4                \n\t" // R4
2492 
-                "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
2493 
-                "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
2494 
-                "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
2495 
-                PAVGB(%%mm4, %%mm0)
2496 
-                PAVGB(%%mm5, %%mm1)
2497 
-                PAVGB(%%mm6, %%mm2)
2498 
-                PAVGB(%%mm7, %%mm3)
2499 
-                PAVGB(%%mm4, %%mm0)
2500 
-                PAVGB(%%mm5, %%mm1)
2501 
-                PAVGB(%%mm6, %%mm2)
2502 
-                PAVGB(%%mm7, %%mm3)
2503 
-                PAVGB(%%mm4, %%mm0)
2504 
-                PAVGB(%%mm5, %%mm1)
2505 
-                PAVGB(%%mm6, %%mm2)
2506 
-                PAVGB(%%mm7, %%mm3)
2507 
-                "movq %%mm0, (%1, %2, 4)                \n\t" // R4
2508 
-                "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
2509 
-                "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
2510 
-                "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
2511 
-                "movq %%mm0, (%0, %2, 4)                \n\t" // L4
2512 
-                "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
2513 
-                "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
2514 
-                "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
2515 
-
2516 
-                "4:                                     \n\t"
2517 
-
2518 
-                :: "r" (src), "r" (tempBlured), "r"((long)stride), "m" (tempBluredPast)
2519 
-                : "%"REG_a, "%"REG_d, "%"REG_c, "memory"
2520 
-                );
2319 
+        "movq %%mm0, %%mm4                      \n\t"
2320 
+        "psrlq $32, %%mm0                       \n\t"
2321 
+        "paddd %%mm0, %%mm4                     \n\t"
2322 
+        "movd %%mm4, %%ecx                      \n\t"
2323 
+        "shll $2, %%ecx                         \n\t"
2324 
+        "mov %3, %%"REG_d"                      \n\t"
2325 
+        "addl -4(%%"REG_d"), %%ecx              \n\t"
2326 
+        "addl 4(%%"REG_d"), %%ecx               \n\t"
2327 
+        "addl -1024(%%"REG_d"), %%ecx           \n\t"
2328 
+        "addl $4, %%ecx                         \n\t"
2329 
+        "addl 1024(%%"REG_d"), %%ecx            \n\t"
2330 
+        "shrl $3, %%ecx                         \n\t"
2331 
+        "movl %%ecx, (%%"REG_d")                \n\t"
2332 
+
2333 
+//        "mov %3, %%"REG_c"                      \n\t"
2334 
+//        "mov %%"REG_c", test                    \n\t"
2335 
+//        "jmp 4f                                 \n\t"
2336 
+        "cmpl 512(%%"REG_d"), %%ecx             \n\t"
2337 
+        " jb 2f                                 \n\t"
2338 
+        "cmpl 516(%%"REG_d"), %%ecx             \n\t"
2339 
+        " jb 1f                                 \n\t"
2340 
+
2341 
+        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2342 
+        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2343 
+        "movq (%0), %%mm0                       \n\t" // L0
2344 
+        "movq (%0, %2), %%mm1                   \n\t" // L1
2345 
+        "movq (%0, %2, 2), %%mm2                \n\t" // L2
2346 
+        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2347 
+        "movq (%0, %2, 4), %%mm4                \n\t" // L4
2348 
+        "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
2349 
+        "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
2350 
+        "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
2351 
+        "movq %%mm0, (%1)                       \n\t" // L0
2352 
+        "movq %%mm1, (%1, %2)                   \n\t" // L1
2353 
+        "movq %%mm2, (%1, %2, 2)                \n\t" // L2
2354 
+        "movq %%mm3, (%1, %%"REG_a")            \n\t" // L3
2355 
+        "movq %%mm4, (%1, %2, 4)                \n\t" // L4
2356 
+        "movq %%mm5, (%1, %%"REG_d")            \n\t" // L5
2357 
+        "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // L6
2358 
+        "movq %%mm7, (%1, %%"REG_c")            \n\t" // L7
2359 
+        "jmp 4f                                 \n\t"
2360 
+
2361 
+        "1:                                     \n\t"
2362 
+        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2363 
+        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2364 
+        "movq (%0), %%mm0                       \n\t" // L0
2365 
+        PAVGB((%1), %%mm0)                            // L0
2366 
+        "movq (%0, %2), %%mm1                   \n\t" // L1
2367 
+        PAVGB((%1, %2), %%mm1)                        // L1
2368 
+        "movq (%0, %2, 2), %%mm2                \n\t" // L2
2369 
+        PAVGB((%1, %2, 2), %%mm2)                     // L2
2370 
+        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2371 
+        PAVGB((%1, %%REGa), %%mm3)                    // L3
2372 
+        "movq (%0, %2, 4), %%mm4                \n\t" // L4
2373 
+        PAVGB((%1, %2, 4), %%mm4)                     // L4
2374 
+        "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5
2375 
+        PAVGB((%1, %%REGd), %%mm5)                    // L5
2376 
+        "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6
2377 
+        PAVGB((%1, %%REGa, 2), %%mm6)                 // L6
2378 
+        "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7
2379 
+        PAVGB((%1, %%REGc), %%mm7)                    // L7
2380 
+        "movq %%mm0, (%1)                       \n\t" // R0
2381 
+        "movq %%mm1, (%1, %2)                   \n\t" // R1
2382 
+        "movq %%mm2, (%1, %2, 2)                \n\t" // R2
2383 
+        "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
2384 
+        "movq %%mm4, (%1, %2, 4)                \n\t" // R4
2385 
+        "movq %%mm5, (%1, %%"REG_d")            \n\t" // R5
2386 
+        "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // R6
2387 
+        "movq %%mm7, (%1, %%"REG_c")            \n\t" // R7
2388 
+        "movq %%mm0, (%0)                       \n\t" // L0
2389 
+        "movq %%mm1, (%0, %2)                   \n\t" // L1
2390 
+        "movq %%mm2, (%0, %2, 2)                \n\t" // L2
2391 
+        "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
2392 
+        "movq %%mm4, (%0, %2, 4)                \n\t" // L4
2393 
+        "movq %%mm5, (%0, %%"REG_d")            \n\t" // L5
2394 
+        "movq %%mm6, (%0, %%"REG_a", 2)         \n\t" // L6
2395 
+        "movq %%mm7, (%0, %%"REG_c")            \n\t" // L7
2396 
+        "jmp 4f                                 \n\t"
2397 
+
2398 
+        "2:                                     \n\t"
2399 
+        "cmpl 508(%%"REG_d"), %%ecx             \n\t"
2400 
+        " jb 3f                                 \n\t"
2401 
+
2402 
+        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2403 
+        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2404 
+        "movq (%0), %%mm0                       \n\t" // L0
2405 
+        "movq (%0, %2), %%mm1                   \n\t" // L1
2406 
+        "movq (%0, %2, 2), %%mm2                \n\t" // L2
2407 
+        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2408 
+        "movq (%1), %%mm4                       \n\t" // R0
2409 
+        "movq (%1, %2), %%mm5                   \n\t" // R1
2410 
+        "movq (%1, %2, 2), %%mm6                \n\t" // R2
2411 
+        "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
2412 
+        PAVGB(%%mm4, %%mm0)
2413 
+        PAVGB(%%mm5, %%mm1)
2414 
+        PAVGB(%%mm6, %%mm2)
2415 
+        PAVGB(%%mm7, %%mm3)
2416 
+        PAVGB(%%mm4, %%mm0)
2417 
+        PAVGB(%%mm5, %%mm1)
2418 
+        PAVGB(%%mm6, %%mm2)
2419 
+        PAVGB(%%mm7, %%mm3)
2420 
+        "movq %%mm0, (%1)                       \n\t" // R0
2421 
+        "movq %%mm1, (%1, %2)                   \n\t" // R1
2422 
+        "movq %%mm2, (%1, %2, 2)                \n\t" // R2
2423 
+        "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
2424 
+        "movq %%mm0, (%0)                       \n\t" // L0
2425 
+        "movq %%mm1, (%0, %2)                   \n\t" // L1
2426 
+        "movq %%mm2, (%0, %2, 2)                \n\t" // L2
2427 
+        "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
2428 
+
2429 
+        "movq (%0, %2, 4), %%mm0                \n\t" // L4
2430 
+        "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
2431 
+        "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
2432 
+        "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
2433 
+        "movq (%1, %2, 4), %%mm4                \n\t" // R4
2434 
+        "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
2435 
+        "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
2436 
+        "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
2437 
+        PAVGB(%%mm4, %%mm0)
2438 
+        PAVGB(%%mm5, %%mm1)
2439 
+        PAVGB(%%mm6, %%mm2)
2440 
+        PAVGB(%%mm7, %%mm3)
2441 
+        PAVGB(%%mm4, %%mm0)
2442 
+        PAVGB(%%mm5, %%mm1)
2443 
+        PAVGB(%%mm6, %%mm2)
2444 
+        PAVGB(%%mm7, %%mm3)
2445 
+        "movq %%mm0, (%1, %2, 4)                \n\t" // R4
2446 
+        "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
2447 
+        "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
2448 
+        "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
2449 
+        "movq %%mm0, (%0, %2, 4)                \n\t" // L4
2450 
+        "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
2451 
+        "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
2452 
+        "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
2453 
+        "jmp 4f                                 \n\t"
2454 
+
2455 
+        "3:                                     \n\t"
2456 
+        "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride
2457 
+        "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride
2458 
+        "movq (%0), %%mm0                       \n\t" // L0
2459 
+        "movq (%0, %2), %%mm1                   \n\t" // L1
2460 
+        "movq (%0, %2, 2), %%mm2                \n\t" // L2
2461 
+        "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3
2462 
+        "movq (%1), %%mm4                       \n\t" // R0
2463 
+        "movq (%1, %2), %%mm5                   \n\t" // R1
2464 
+        "movq (%1, %2, 2), %%mm6                \n\t" // R2
2465 
+        "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3
2466 
+        PAVGB(%%mm4, %%mm0)
2467 
+        PAVGB(%%mm5, %%mm1)
2468 
+        PAVGB(%%mm6, %%mm2)
2469 
+        PAVGB(%%mm7, %%mm3)
2470 
+        PAVGB(%%mm4, %%mm0)
2471 
+        PAVGB(%%mm5, %%mm1)
2472 
+        PAVGB(%%mm6, %%mm2)
2473 
+        PAVGB(%%mm7, %%mm3)
2474 
+        PAVGB(%%mm4, %%mm0)
2475 
+        PAVGB(%%mm5, %%mm1)
2476 
+        PAVGB(%%mm6, %%mm2)
2477 
+        PAVGB(%%mm7, %%mm3)
2478 
+        "movq %%mm0, (%1)                       \n\t" // R0
2479 
+        "movq %%mm1, (%1, %2)                   \n\t" // R1
2480 
+        "movq %%mm2, (%1, %2, 2)                \n\t" // R2
2481 
+        "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3
2482 
+        "movq %%mm0, (%0)                       \n\t" // L0
2483 
+        "movq %%mm1, (%0, %2)                   \n\t" // L1
2484 
+        "movq %%mm2, (%0, %2, 2)                \n\t" // L2
2485 
+        "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3
2486 
+
2487 
+        "movq (%0, %2, 4), %%mm0                \n\t" // L4
2488 
+        "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5
2489 
+        "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6
2490 
+        "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7
2491 
+        "movq (%1, %2, 4), %%mm4                \n\t" // R4
2492 
+        "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5
2493 
+        "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6
2494 
+        "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7
2495 
+        PAVGB(%%mm4, %%mm0)
2496 
+        PAVGB(%%mm5, %%mm1)
2497 
+        PAVGB(%%mm6, %%mm2)
2498 
+        PAVGB(%%mm7, %%mm3)
2499 
+        PAVGB(%%mm4, %%mm0)
2500 
+        PAVGB(%%mm5, %%mm1)
2501 
+        PAVGB(%%mm6, %%mm2)
2502 
+        PAVGB(%%mm7, %%mm3)
2503 
+        PAVGB(%%mm4, %%mm0)
2504 
+        PAVGB(%%mm5, %%mm1)
2505 
+        PAVGB(%%mm6, %%mm2)
2506 
+        PAVGB(%%mm7, %%mm3)
2507 
+        "movq %%mm0, (%1, %2, 4)                \n\t" // R4
2508 
+        "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5
2509 
+        "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6
2510 
+        "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7
2511 
+        "movq %%mm0, (%0, %2, 4)                \n\t" // L4
2512 
+        "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5
2513 
+        "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6
2514 
+        "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7
2515 
+
2516 
+        "4:                                     \n\t"
2517 
+
2518 
+        :: "r" (src), "r" (tempBlured), "r"((long)stride), "m" (tempBluredPast)
2519 
+        : "%"REG_a, "%"REG_d, "%"REG_c, "memory"
2520 
+    );
2521 2521 
 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
2522 2522 
 {
2523 
-        int y;
2524 
-        int d=0;
2525 
-//        int sysd=0;
2526 
-        int i;
2523 
+    int y;
2524 
+    int d=0;
2525 
+//    int sysd=0;
2526 
+    int i;
2527 2527
2528 
-        for(y=0; y<8; y++)
2529 
-        {
2530 
-                int x;
2531 
-                for(x=0; x<8; x++)
2532 
-                {
2533 
-                        int ref= tempBlured[ x + y*stride ];
2534 
-                        int cur= src[ x + y*stride ];
2535 
-                        int d1=ref - cur;
2536 
-//                        if(x==0 || x==7) d1+= d1>>1;
2537 
-//                        if(y==0 || y==7) d1+= d1>>1;
2538 
-//                        d+= FFABS(d1);
2539 
-                        d+= d1*d1;
2540 
-//                        sysd+= d1;
2541 
-                }
2528 
+    for(y=0; y<8; y++){
2529 
+        int x;
2530 
+        for(x=0; x<8; x++){
2531 
+            int ref= tempBlured[ x + y*stride ];
2532 
+            int cur= src[ x + y*stride ];
2533 
+            int d1=ref - cur;
2534 
+//            if(x==0 || x==7) d1+= d1>>1;
2535 
+//            if(y==0 || y==7) d1+= d1>>1;
2536 
+//            d+= FFABS(d1);
2537 
+            d+= d1*d1;
2538 
+//            sysd+= d1;
2542 2539 
         }
2543 
-        i=d;
2544 
-        d=         (
2545 
-                4*d
2546 
-                +(*(tempBluredPast-256))
2547 
-                +(*(tempBluredPast-1))+ (*(tempBluredPast+1))
2548 
-                +(*(tempBluredPast+256))
2549 
-                +4)>>3;
2550 
-        *tempBluredPast=i;
2551 
-//        ((*tempBluredPast)*3 + d + 2)>>2;
2540 
+    }
2541 
+    i=d;
2542 
+    d=  (
2543 
+        4*d
2544 
+        +(*(tempBluredPast-256))
2545 
+        +(*(tempBluredPast-1))+ (*(tempBluredPast+1))
2546 
+        +(*(tempBluredPast+256))
2547 
+        +4)>>3;
2548 
+    *tempBluredPast=i;
2549 
+//    ((*tempBluredPast)*3 + d + 2)>>2;
2552 2550
2553 2551 
 /*
2554 2552 
 Switch between
... ... 
@@ -2557,68 +2516,51 @@ Switch between
2557 2557 
 64 48 36 27 20 15 11 (33) (approx)
2558 2558 
 64 56 49 43 37 33 29 (200) (approx)
2559 2559 
 */
2560 
-        if(d > maxNoise[1])
2561 
-        {
2562 
-                if(d < maxNoise[2])
2563 
-                {
2564 
-                        for(y=0; y<8; y++)
2565 
-                        {
2566 
-                                int x;
2567 
-                                for(x=0; x<8; x++)
2568 
-                                {
2569 
-                                        int ref= tempBlured[ x + y*stride ];
2570 
-                                        int cur= src[ x + y*stride ];
2571 
-                                        tempBlured[ x + y*stride ]=
2572 
-                                        src[ x + y*stride ]=
2573 
-                                                (ref + cur + 1)>>1;
2574 
-                                }
2575 
-                        }
2560 
+    if(d > maxNoise[1]){
2561 
+        if(d < maxNoise[2]){
2562 
+            for(y=0; y<8; y++){
2563 
+                int x;
2564 
+                for(x=0; x<8; x++){
2565 
+                    int ref= tempBlured[ x + y*stride ];
2566 
+                    int cur= src[ x + y*stride ];
2567 
+                    tempBlured[ x + y*stride ]=
2568 
+                    src[ x + y*stride ]=
2569 
+                        (ref + cur + 1)>>1;
2576 2570 
                 }
2577 
-                else
2578 
-                {
2579 
-                        for(y=0; y<8; y++)
2580 
-                        {
2581 
-                                int x;
2582 
-                                for(x=0; x<8; x++)
2583 
-                                {
2584 
-                                        tempBlured[ x + y*stride ]= src[ x + y*stride ];
2585 
-                                }
2586 
-                        }
2571 
+            }
2572 
+        }else{
2573 
+            for(y=0; y<8; y++){
2574 
+                int x;
2575 
+                for(x=0; x<8; x++){
2576 
+                    tempBlured[ x + y*stride ]= src[ x + y*stride ];
2587 2577 
                 }
2578 
+            }
2588 2579 
         }
2589 
-        else
2590 
-        {
2591 
-                if(d < maxNoise[0])
2592 
-                {
2593 
-                        for(y=0; y<8; y++)
2594 
-                        {
2595 
-                                int x;
2596 
-                                for(x=0; x<8; x++)
2597 
-                                {
2598 
-                                        int ref= tempBlured[ x + y*stride ];
2599 
-                                        int cur= src[ x + y*stride ];
2600 
-                                        tempBlured[ x + y*stride ]=
2601 
-                                        src[ x + y*stride ]=
2602 
-                                                (ref*7 + cur + 4)>>3;
2603 
-                                }
2604 
-                        }
2580 
+    }else{
2581 
+        if(d < maxNoise[0]){
2582 
+            for(y=0; y<8; y++){
2583 
+                int x;
2584 
+                for(x=0; x<8; x++){
2585 
+                    int ref= tempBlured[ x + y*stride ];
2586 
+                    int cur= src[ x + y*stride ];
2587 
+                    tempBlured[ x + y*stride ]=
2588 
+                    src[ x + y*stride ]=
2589 
+                        (ref*7 + cur + 4)>>3;
2605 2590 
                 }
2606 
-                else
2607 
-                {
2608 
-                        for(y=0; y<8; y++)
2609 
-                        {
2610 
-                                int x;
2611 
-                                for(x=0; x<8; x++)
2612 
-                                {
2613 
-                                        int ref= tempBlured[ x + y*stride ];
2614 
-                                        int cur= src[ x + y*stride ];
2615 
-                                        tempBlured[ x + y*stride ]=
2616 
-                                        src[ x + y*stride ]=
2617 
-                                                (ref*3 + cur + 2)>>2;
2618 
-                                }
2619 
-                        }
2591 
+            }
2592 
+        }else{
2593 
+            for(y=0; y<8; y++){
2594 
+                int x;
2595 
+                for(x=0; x<8; x++){
2596 
+                    int ref= tempBlured[ x + y*stride ];
2597 
+                    int cur= src[ x + y*stride ];
2598 
+                    tempBlured[ x + y*stride ]=
2599 
+                    src[ x + y*stride ]=
2600 
+                        (ref*3 + cur + 2)>>2;
2620 2601 
                 }
2602 
+            }
2621 2603 
         }
2604 
+    }
2622 2605 
 }
2623 2606 
 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
2624 2607 
 }
... ... 
@@ -2629,531 +2571,531 @@ Switch between
2629 2629 
  * accurate deblock filter
2630 2630 
  */
2631 2631 
 static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, PPContext *c){
2632 
-        int64_t dc_mask, eq_mask, both_masks;
2633 
-        int64_t sums[10*8*2];
2634 
-        src+= step*3; // src points to begin of the 8x8 Block
2632 
+    int64_t dc_mask, eq_mask, both_masks;
2633 
+    int64_t sums[10*8*2];
2634 
+    src+= step*3; // src points to begin of the 8x8 Block
2635 2635 
 //START_TIMER
2636 
-asm volatile(
2637 
-                "movq %0, %%mm7                         \n\t"
2638 
-                "movq %1, %%mm6                         \n\t"
2639 
-                : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
2640 
-                );
2641 
-
2642 
-asm volatile(
2643 
-                "lea (%2, %3), %%"REG_a"                \n\t"
2636 
+    asm volatile(
2637 
+        "movq %0, %%mm7                         \n\t"
2638 
+        "movq %1, %%mm6                         \n\t"
2639 
+        : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])
2640 
+        );
2641 
+
2642 
+    asm volatile(
2643 
+        "lea (%2, %3), %%"REG_a"                \n\t"
2644 2644 
 //      0       1       2       3       4       5       6       7       8       9
2645 2645 
 //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2
2646 2646
2647 
-                "movq (%2), %%mm0                       \n\t"
2648 
-                "movq (%%"REG_a"), %%mm1                \n\t"
2649 
-                "movq %%mm1, %%mm3                      \n\t"
2650 
-                "movq %%mm1, %%mm4                      \n\t"
2651 
-                "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
2652 
-                "paddb %%mm7, %%mm0                     \n\t"
2653 
-                "pcmpgtb %%mm6, %%mm0                   \n\t"
2654 
-
2655 
-                "movq (%%"REG_a",%3), %%mm2             \n\t"
2656 
-                PMAXUB(%%mm2, %%mm4)
2657 
-                PMINUB(%%mm2, %%mm3, %%mm5)
2658 
-                "psubb %%mm2, %%mm1                     \n\t"
2659 
-                "paddb %%mm7, %%mm1                     \n\t"
2660 
-                "pcmpgtb %%mm6, %%mm1                   \n\t"
2661 
-                "paddb %%mm1, %%mm0                     \n\t"
2662 
-
2663 
-                "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
2664 
-                PMAXUB(%%mm1, %%mm4)
2665 
-                PMINUB(%%mm1, %%mm3, %%mm5)
2666 
-                "psubb %%mm1, %%mm2                     \n\t"
2667 
-                "paddb %%mm7, %%mm2                     \n\t"
2668 
-                "pcmpgtb %%mm6, %%mm2                   \n\t"
2669 
-                "paddb %%mm2, %%mm0                     \n\t"
2670 
-
2671 
-                "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
2672 
-
2673 
-                "movq (%2, %3, 4), %%mm2                \n\t"
2674 
-                PMAXUB(%%mm2, %%mm4)
2675 
-                PMINUB(%%mm2, %%mm3, %%mm5)
2676 
-                "psubb %%mm2, %%mm1                     \n\t"
2677 
-                "paddb %%mm7, %%mm1                     \n\t"
2678 
-                "pcmpgtb %%mm6, %%mm1                   \n\t"
2679 
-                "paddb %%mm1, %%mm0                     \n\t"
2680 
-
2681 
-                "movq (%%"REG_a"), %%mm1                \n\t"
2682 
-                PMAXUB(%%mm1, %%mm4)
2683 
-                PMINUB(%%mm1, %%mm3, %%mm5)
2684 
-                "psubb %%mm1, %%mm2                     \n\t"
2685 
-                "paddb %%mm7, %%mm2                     \n\t"
2686 
-                "pcmpgtb %%mm6, %%mm2                   \n\t"
2687 
-                "paddb %%mm2, %%mm0                     \n\t"
2688 
-
2689 
-                "movq (%%"REG_a", %3), %%mm2            \n\t"
2690 
-                PMAXUB(%%mm2, %%mm4)
2691 
-                PMINUB(%%mm2, %%mm3, %%mm5)
2692 
-                "psubb %%mm2, %%mm1                     \n\t"
2693 
-                "paddb %%mm7, %%mm1                     \n\t"
2694 
-                "pcmpgtb %%mm6, %%mm1                   \n\t"
2695 
-                "paddb %%mm1, %%mm0                     \n\t"
2696 
-
2697 
-                "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
2698 
-                PMAXUB(%%mm1, %%mm4)
2699 
-                PMINUB(%%mm1, %%mm3, %%mm5)
2700 
-                "psubb %%mm1, %%mm2                     \n\t"
2701 
-                "paddb %%mm7, %%mm2                     \n\t"
2702 
-                "pcmpgtb %%mm6, %%mm2                   \n\t"
2703 
-                "paddb %%mm2, %%mm0                     \n\t"
2704 
-
2705 
-                "movq (%2, %3, 8), %%mm2                \n\t"
2706 
-                PMAXUB(%%mm2, %%mm4)
2707 
-                PMINUB(%%mm2, %%mm3, %%mm5)
2708 
-                "psubb %%mm2, %%mm1                     \n\t"
2709 
-                "paddb %%mm7, %%mm1                     \n\t"
2710 
-                "pcmpgtb %%mm6, %%mm1                   \n\t"
2711 
-                "paddb %%mm1, %%mm0                     \n\t"
2712 
-
2713 
-                "movq (%%"REG_a", %3, 4), %%mm1         \n\t"
2714 
-                "psubb %%mm1, %%mm2                     \n\t"
2715 
-                "paddb %%mm7, %%mm2                     \n\t"
2716 
-                "pcmpgtb %%mm6, %%mm2                   \n\t"
2717 
-                "paddb %%mm2, %%mm0                     \n\t"
2718 
-                "psubusb %%mm3, %%mm4                   \n\t"
2719 
-
2720 
-                "pxor %%mm6, %%mm6                      \n\t"
2721 
-                "movq %4, %%mm7                         \n\t" // QP,..., QP
2722 
-                "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
2723 
-                "psubusb %%mm4, %%mm7                   \n\t" // Diff >=2QP -> 0
2724 
-                "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
2725 
-                "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
2726 
-                "movq %%mm7, %1                         \n\t"
2727 
-
2728 
-                "movq %5, %%mm7                         \n\t"
2729 
-                "punpcklbw %%mm7, %%mm7                 \n\t"
2730 
-                "punpcklbw %%mm7, %%mm7                 \n\t"
2731 
-                "punpcklbw %%mm7, %%mm7                 \n\t"
2732 
-                "psubb %%mm0, %%mm6                     \n\t"
2733 
-                "pcmpgtb %%mm7, %%mm6                   \n\t"
2734 
-                "movq %%mm6, %0                         \n\t"
2735 
-
2736 
-                : "=m" (eq_mask), "=m" (dc_mask)
2737 
-                : "r" (src), "r" ((long)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)
2738 
-                : "%"REG_a
2739 
-                );
2740 
-
2741 
-        both_masks = dc_mask & eq_mask;
2742 
-
2743 
-        if(both_masks){
2744 
-                long offset= -8*step;
2745 
-                int64_t *temp_sums= sums;
2746 
-
2747 
-                asm volatile(
2748 
-                "movq %2, %%mm0                         \n\t"  // QP,..., QP
2749 
-                "pxor %%mm4, %%mm4                      \n\t"
2750 
-
2751 
-                "movq (%0), %%mm6                       \n\t"
2752 
-                "movq (%0, %1), %%mm5                   \n\t"
2753 
-                "movq %%mm5, %%mm1                      \n\t"
2754 
-                "movq %%mm6, %%mm2                      \n\t"
2755 
-                "psubusb %%mm6, %%mm5                   \n\t"
2756 
-                "psubusb %%mm1, %%mm2                   \n\t"
2757 
-                "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
2758 
-                "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
2759 
-                "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF
2760 
-
2761 
-                "pxor %%mm6, %%mm1                      \n\t"
2762 
-                "pand %%mm0, %%mm1                      \n\t"
2763 
-                "pxor %%mm1, %%mm6                      \n\t"
2764 
-                // 0:QP  6:First
2765 
-
2766 
-                "movq (%0, %1, 8), %%mm5                \n\t"
2767 
-                "add %1, %0                             \n\t" // %0 points to line 1 not 0
2768 
-                "movq (%0, %1, 8), %%mm7                \n\t"
2769 
-                "movq %%mm5, %%mm1                      \n\t"
2770 
-                "movq %%mm7, %%mm2                      \n\t"
2771 
-                "psubusb %%mm7, %%mm5                   \n\t"
2772 
-                "psubusb %%mm1, %%mm2                   \n\t"
2773 
-                "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
2774 
-                "movq %2, %%mm0                         \n\t"  // QP,..., QP
2775 
-                "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
2776 
-                "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF
2777 
-
2778 
-                "pxor %%mm7, %%mm1                      \n\t"
2779 
-                "pand %%mm0, %%mm1                      \n\t"
2780 
-                "pxor %%mm1, %%mm7                      \n\t"
2781 
-
2782 
-                "movq %%mm6, %%mm5                      \n\t"
2783 
-                "punpckhbw %%mm4, %%mm6                 \n\t"
2784 
-                "punpcklbw %%mm4, %%mm5                 \n\t"
2785 
-                // 4:0 5/6:First 7:Last
2786 
-
2787 
-                "movq %%mm5, %%mm0                      \n\t"
2788 
-                "movq %%mm6, %%mm1                      \n\t"
2789 
-                "psllw $2, %%mm0                        \n\t"
2790 
-                "psllw $2, %%mm1                        \n\t"
2791 
-                "paddw "MANGLE(w04)", %%mm0             \n\t"
2792 
-                "paddw "MANGLE(w04)", %%mm1             \n\t"
2647 
+        "movq (%2), %%mm0                       \n\t"
2648 
+        "movq (%%"REG_a"), %%mm1                \n\t"
2649 
+        "movq %%mm1, %%mm3                      \n\t"
2650 
+        "movq %%mm1, %%mm4                      \n\t"
2651 
+        "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece
2652 
+        "paddb %%mm7, %%mm0                     \n\t"
2653 
+        "pcmpgtb %%mm6, %%mm0                   \n\t"
2654 
+
2655 
+        "movq (%%"REG_a",%3), %%mm2             \n\t"
2656 
+        PMAXUB(%%mm2, %%mm4)
2657 
+        PMINUB(%%mm2, %%mm3, %%mm5)
2658 
+        "psubb %%mm2, %%mm1                     \n\t"
2659 
+        "paddb %%mm7, %%mm1                     \n\t"
2660 
+        "pcmpgtb %%mm6, %%mm1                   \n\t"
2661 
+        "paddb %%mm1, %%mm0                     \n\t"
2662 
+
2663 
+        "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
2664 
+        PMAXUB(%%mm1, %%mm4)
2665 
+        PMINUB(%%mm1, %%mm3, %%mm5)
2666 
+        "psubb %%mm1, %%mm2                     \n\t"
2667 
+        "paddb %%mm7, %%mm2                     \n\t"
2668 
+        "pcmpgtb %%mm6, %%mm2                   \n\t"
2669 
+        "paddb %%mm2, %%mm0                     \n\t"
2670 
+
2671 
+        "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"
2672 
+
2673 
+        "movq (%2, %3, 4), %%mm2                \n\t"
2674 
+        PMAXUB(%%mm2, %%mm4)
2675 
+        PMINUB(%%mm2, %%mm3, %%mm5)
2676 
+        "psubb %%mm2, %%mm1                     \n\t"
2677 
+        "paddb %%mm7, %%mm1                     \n\t"
2678 
+        "pcmpgtb %%mm6, %%mm1                   \n\t"
2679 
+        "paddb %%mm1, %%mm0                     \n\t"
2680 
+
2681 
+        "movq (%%"REG_a"), %%mm1                \n\t"
2682 
+        PMAXUB(%%mm1, %%mm4)
2683 
+        PMINUB(%%mm1, %%mm3, %%mm5)
2684 
+        "psubb %%mm1, %%mm2                     \n\t"
2685 
+        "paddb %%mm7, %%mm2                     \n\t"
2686 
+        "pcmpgtb %%mm6, %%mm2                   \n\t"
2687 
+        "paddb %%mm2, %%mm0                     \n\t"
2688 
+
2689 
+        "movq (%%"REG_a", %3), %%mm2            \n\t"
2690 
+        PMAXUB(%%mm2, %%mm4)
2691 
+        PMINUB(%%mm2, %%mm3, %%mm5)
2692 
+        "psubb %%mm2, %%mm1                     \n\t"
2693 
+        "paddb %%mm7, %%mm1                     \n\t"
2694 
+        "pcmpgtb %%mm6, %%mm1                   \n\t"
2695 
+        "paddb %%mm1, %%mm0                     \n\t"
2696 
+
2697 
+        "movq (%%"REG_a", %3, 2), %%mm1         \n\t"
2698 
+        PMAXUB(%%mm1, %%mm4)
2699 
+        PMINUB(%%mm1, %%mm3, %%mm5)
2700 
+        "psubb %%mm1, %%mm2                     \n\t"
2701 
+        "paddb %%mm7, %%mm2                     \n\t"
2702 
+        "pcmpgtb %%mm6, %%mm2                   \n\t"
2703 
+        "paddb %%mm2, %%mm0                     \n\t"
2704 
+
2705 
+        "movq (%2, %3, 8), %%mm2                \n\t"
2706 
+        PMAXUB(%%mm2, %%mm4)
2707 
+        PMINUB(%%mm2, %%mm3, %%mm5)
2708 
+        "psubb %%mm2, %%mm1                     \n\t"
2709 
+        "paddb %%mm7, %%mm1                     \n\t"
2710 
+        "pcmpgtb %%mm6, %%mm1                   \n\t"
2711 
+        "paddb %%mm1, %%mm0                     \n\t"
2712 
+
2713 
+        "movq (%%"REG_a", %3, 4), %%mm1         \n\t"
2714 
+        "psubb %%mm1, %%mm2                     \n\t"
2715 
+        "paddb %%mm7, %%mm2                     \n\t"
2716 
+        "pcmpgtb %%mm6, %%mm2                   \n\t"
2717 
+        "paddb %%mm2, %%mm0                     \n\t"
2718 
+        "psubusb %%mm3, %%mm4                   \n\t"
2719 
+
2720 
+        "pxor %%mm6, %%mm6                      \n\t"
2721 
+        "movq %4, %%mm7                         \n\t" // QP,..., QP
2722 
+        "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP
2723 
+        "psubusb %%mm4, %%mm7                   \n\t" // Diff >=2QP -> 0
2724 
+        "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
2725 
+        "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0
2726 
+        "movq %%mm7, %1                         \n\t"
2727 
+
2728 
+        "movq %5, %%mm7                         \n\t"
2729 
+        "punpcklbw %%mm7, %%mm7                 \n\t"
2730 
+        "punpcklbw %%mm7, %%mm7                 \n\t"
2731 
+        "punpcklbw %%mm7, %%mm7                 \n\t"
2732 
+        "psubb %%mm0, %%mm6                     \n\t"
2733 
+        "pcmpgtb %%mm7, %%mm6                   \n\t"
2734 
+        "movq %%mm6, %0                         \n\t"
2735 
+
2736 
+        : "=m" (eq_mask), "=m" (dc_mask)
2737 
+        : "r" (src), "r" ((long)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)
2738 
+        : "%"REG_a
2739 
+    );
2740 
+
2741 
+    both_masks = dc_mask & eq_mask;
2742 
+
2743 
+    if(both_masks){
2744 
+        long offset= -8*step;
2745 
+        int64_t *temp_sums= sums;
2746 
+
2747 
+        asm volatile(
2748 
+            "movq %2, %%mm0                         \n\t"  // QP,..., QP
2749 
+            "pxor %%mm4, %%mm4                      \n\t"
2750 
+
2751 
+            "movq (%0), %%mm6                       \n\t"
2752 
+            "movq (%0, %1), %%mm5                   \n\t"
2753 
+            "movq %%mm5, %%mm1                      \n\t"
2754 
+            "movq %%mm6, %%mm2                      \n\t"
2755 
+            "psubusb %%mm6, %%mm5                   \n\t"
2756 
+            "psubusb %%mm1, %%mm2                   \n\t"
2757 
+            "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
2758 
+            "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
2759 
+            "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF
2760 
+
2761 
+            "pxor %%mm6, %%mm1                      \n\t"
2762 
+            "pand %%mm0, %%mm1                      \n\t"
2763 
+            "pxor %%mm1, %%mm6                      \n\t"
2764 
+            // 0:QP  6:First
2765 
+
2766 
+            "movq (%0, %1, 8), %%mm5                \n\t"
2767 
+            "add %1, %0                             \n\t" // %0 points to line 1 not 0
2768 
+            "movq (%0, %1, 8), %%mm7                \n\t"
2769 
+            "movq %%mm5, %%mm1                      \n\t"
2770 
+            "movq %%mm7, %%mm2                      \n\t"
2771 
+            "psubusb %%mm7, %%mm5                   \n\t"
2772 
+            "psubusb %%mm1, %%mm2                   \n\t"
2773 
+            "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines
2774 
+            "movq %2, %%mm0                         \n\t"  // QP,..., QP
2775 
+            "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0
2776 
+            "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF
2777 
+
2778 
+            "pxor %%mm7, %%mm1                      \n\t"
2779 
+            "pand %%mm0, %%mm1                      \n\t"
2780 
+            "pxor %%mm1, %%mm7                      \n\t"
2781 
+
2782 
+            "movq %%mm6, %%mm5                      \n\t"
2783 
+            "punpckhbw %%mm4, %%mm6                 \n\t"
2784 
+            "punpcklbw %%mm4, %%mm5                 \n\t"
2785 
+            // 4:0 5/6:First 7:Last
2786 
+
2787 
+            "movq %%mm5, %%mm0                      \n\t"
2788 
+            "movq %%mm6, %%mm1                      \n\t"
2789 
+            "psllw $2, %%mm0                        \n\t"
2790 
+            "psllw $2, %%mm1                        \n\t"
2791 
+            "paddw "MANGLE(w04)", %%mm0             \n\t"
2792 
+            "paddw "MANGLE(w04)", %%mm1             \n\t"
2793 2793
2794 2794 
 #define NEXT\
2795 
-                "movq (%0), %%mm2                       \n\t"\
2796 
-                "movq (%0), %%mm3                       \n\t"\
2797 
-                "add %1, %0                             \n\t"\
2798 
-                "punpcklbw %%mm4, %%mm2                 \n\t"\
2799 
-                "punpckhbw %%mm4, %%mm3                 \n\t"\
2800 
-                "paddw %%mm2, %%mm0                     \n\t"\
2801 
-                "paddw %%mm3, %%mm1                     \n\t"
2795 
+            "movq (%0), %%mm2                       \n\t"\
2796 
+            "movq (%0), %%mm3                       \n\t"\
2797 
+            "add %1, %0                             \n\t"\
2798 
+            "punpcklbw %%mm4, %%mm2                 \n\t"\
2799 
+            "punpckhbw %%mm4, %%mm3                 \n\t"\
2800 
+            "paddw %%mm2, %%mm0                     \n\t"\
2801 
+            "paddw %%mm3, %%mm1                     \n\t"
2802 2802
2803 2803 
 #define PREV\
2804 
-                "movq (%0), %%mm2                       \n\t"\
2805 
-                "movq (%0), %%mm3                       \n\t"\
2806 
-                "add %1, %0                             \n\t"\
2807 
-                "punpcklbw %%mm4, %%mm2                 \n\t"\
2808 
-                "punpckhbw %%mm4, %%mm3                 \n\t"\
2809 
-                "psubw %%mm2, %%mm0                     \n\t"\
2810 
-                "psubw %%mm3, %%mm1                     \n\t"
2811 
-
2812 
-
2813 
-                NEXT //0
2814 
-                NEXT //1
2815 
-                NEXT //2
2816 
-                "movq %%mm0, (%3)                       \n\t"
2817 
-                "movq %%mm1, 8(%3)                      \n\t"
2818 
-
2819 
-                NEXT //3
2820 
-                "psubw %%mm5, %%mm0                     \n\t"
2821 
-                "psubw %%mm6, %%mm1                     \n\t"
2822 
-                "movq %%mm0, 16(%3)                     \n\t"
2823 
-                "movq %%mm1, 24(%3)                     \n\t"
2824 
-
2825 
-                NEXT //4
2826 
-                "psubw %%mm5, %%mm0                     \n\t"
2827 
-                "psubw %%mm6, %%mm1                     \n\t"
2828 
-                "movq %%mm0, 32(%3)                     \n\t"
2829 
-                "movq %%mm1, 40(%3)                     \n\t"
2830 
-
2831 
-                NEXT //5
2832 
-                "psubw %%mm5, %%mm0                     \n\t"
2833 
-                "psubw %%mm6, %%mm1                     \n\t"
2834 
-                "movq %%mm0, 48(%3)                     \n\t"
2835 
-                "movq %%mm1, 56(%3)                     \n\t"
2836 
-
2837 
-                NEXT //6
2838 
-                "psubw %%mm5, %%mm0                     \n\t"
2839 
-                "psubw %%mm6, %%mm1                     \n\t"
2840 
-                "movq %%mm0, 64(%3)                     \n\t"
2841 
-                "movq %%mm1, 72(%3)                     \n\t"
2842 
-
2843 
-                "movq %%mm7, %%mm6                      \n\t"
2844 
-                "punpckhbw %%mm4, %%mm7                 \n\t"
2845 
-                "punpcklbw %%mm4, %%mm6                 \n\t"
2846 
-
2847 
-                NEXT //7
2848 
-                "mov %4, %0                             \n\t"
2849 
-                "add %1, %0                             \n\t"
2850 
-                PREV //0
2851 
-                "movq %%mm0, 80(%3)                     \n\t"
2852 
-                "movq %%mm1, 88(%3)                     \n\t"
2853 
-
2854 
-                PREV //1
2855 
-                "paddw %%mm6, %%mm0                     \n\t"
2856 
-                "paddw %%mm7, %%mm1                     \n\t"
2857 
-                "movq %%mm0, 96(%3)                     \n\t"
2858 
-                "movq %%mm1, 104(%3)                    \n\t"
2859 
-
2860 
-                PREV //2
2861 
-                "paddw %%mm6, %%mm0                     \n\t"
2862 
-                "paddw %%mm7, %%mm1                     \n\t"
2863 
-                "movq %%mm0, 112(%3)                    \n\t"
2864 
-                "movq %%mm1, 120(%3)                    \n\t"
2865 
-
2866 
-                PREV //3
2867 
-                "paddw %%mm6, %%mm0                     \n\t"
2868 
-                "paddw %%mm7, %%mm1                     \n\t"
2869 
-                "movq %%mm0, 128(%3)                    \n\t"
2870 
-                "movq %%mm1, 136(%3)                    \n\t"
2871 
-
2872 
-                PREV //4
2873 
-                "paddw %%mm6, %%mm0                     \n\t"
2874 
-                "paddw %%mm7, %%mm1                     \n\t"
2875 
-                "movq %%mm0, 144(%3)                    \n\t"
2876 
-                "movq %%mm1, 152(%3)                    \n\t"
2877 
-
2878 
-                "mov %4, %0                             \n\t" //FIXME
2879 
-
2880 
-                : "+&r"(src)
2881 
-                : "r" ((long)step), "m" (c->pQPb), "r"(sums), "g"(src)
2882 
-                );
2883 
-
2884 
-                src+= step; // src points to begin of the 8x8 Block
2885 
-
2886 
-                asm volatile(
2887 
-                "movq %4, %%mm6                         \n\t"
2888 
-                "pcmpeqb %%mm5, %%mm5                   \n\t"
2889 
-                "pxor %%mm6, %%mm5                      \n\t"
2890 
-                "pxor %%mm7, %%mm7                      \n\t"
2891 
-
2892 
-                "1:                                     \n\t"
2893 
-                "movq (%1), %%mm0                       \n\t"
2894 
-                "movq 8(%1), %%mm1                      \n\t"
2895 
-                "paddw 32(%1), %%mm0                    \n\t"
2896 
-                "paddw 40(%1), %%mm1                    \n\t"
2897 
-                "movq (%0, %3), %%mm2                   \n\t"
2898 
-                "movq %%mm2, %%mm3                      \n\t"
2899 
-                "movq %%mm2, %%mm4                      \n\t"
2900 
-                "punpcklbw %%mm7, %%mm2                 \n\t"
2901 
-                "punpckhbw %%mm7, %%mm3                 \n\t"
2902 
-                "paddw %%mm2, %%mm0                     \n\t"
2903 
-                "paddw %%mm3, %%mm1                     \n\t"
2904 
-                "paddw %%mm2, %%mm0                     \n\t"
2905 
-                "paddw %%mm3, %%mm1                     \n\t"
2906 
-                "psrlw $4, %%mm0                        \n\t"
2907 
-                "psrlw $4, %%mm1                        \n\t"
2908 
-                "packuswb %%mm1, %%mm0                  \n\t"
2909 
-                "pand %%mm6, %%mm0                      \n\t"
2910 
-                "pand %%mm5, %%mm4                      \n\t"
2911 
-                "por %%mm4, %%mm0                       \n\t"
2912 
-                "movq %%mm0, (%0, %3)                   \n\t"
2913 
-                "add $16, %1                            \n\t"
2914 
-                "add %2, %0                             \n\t"
2915 
-                " js 1b                                 \n\t"
2916 
-
2917 
-                : "+r"(offset), "+r"(temp_sums)
2918 
-                : "r" ((long)step), "r"(src - offset), "m"(both_masks)
2919 
-                );
2920 
-        }else
2921 
-                src+= step; // src points to begin of the 8x8 Block
2922 
-
2923 
-        if(eq_mask != -1LL){
2924 
-                uint8_t *temp_src= src;
2925 
-                asm volatile(
2926 
-                "pxor %%mm7, %%mm7                      \n\t"
2927 
-                "lea -40(%%"REG_SP"), %%"REG_c"         \n\t" // make space for 4 8-byte vars
2928 
-                "and "ALIGN_MASK", %%"REG_c"            \n\t" // align
2804 
+            "movq (%0), %%mm2                       \n\t"\
2805 
+            "movq (%0), %%mm3                       \n\t"\
2806 
+            "add %1, %0                             \n\t"\
2807 
+            "punpcklbw %%mm4, %%mm2                 \n\t"\
2808 
+            "punpckhbw %%mm4, %%mm3                 \n\t"\
2809 
+            "psubw %%mm2, %%mm0                     \n\t"\
2810 
+            "psubw %%mm3, %%mm1                     \n\t"
2811 
+
2812 
+
2813 
+            NEXT //0
2814 
+            NEXT //1
2815 
+            NEXT //2
2816 
+            "movq %%mm0, (%3)                       \n\t"
2817 
+            "movq %%mm1, 8(%3)                      \n\t"
2818 
+
2819 
+            NEXT //3
2820 
+            "psubw %%mm5, %%mm0                     \n\t"
2821 
+            "psubw %%mm6, %%mm1                     \n\t"
2822 
+            "movq %%mm0, 16(%3)                     \n\t"
2823 
+            "movq %%mm1, 24(%3)                     \n\t"
2824 
+
2825 
+            NEXT //4
2826 
+            "psubw %%mm5, %%mm0                     \n\t"
2827 
+            "psubw %%mm6, %%mm1                     \n\t"
2828 
+            "movq %%mm0, 32(%3)                     \n\t"
2829 
+            "movq %%mm1, 40(%3)                     \n\t"
2830 
+
2831 
+            NEXT //5
2832 
+            "psubw %%mm5, %%mm0                     \n\t"
2833 
+            "psubw %%mm6, %%mm1                     \n\t"
2834 
+            "movq %%mm0, 48(%3)                     \n\t"
2835 
+            "movq %%mm1, 56(%3)                     \n\t"
2836 
+
2837 
+            NEXT //6
2838 
+            "psubw %%mm5, %%mm0                     \n\t"
2839 
+            "psubw %%mm6, %%mm1                     \n\t"
2840 
+            "movq %%mm0, 64(%3)                     \n\t"
2841 
+            "movq %%mm1, 72(%3)                     \n\t"
2842 
+
2843 
+            "movq %%mm7, %%mm6                      \n\t"
2844 
+            "punpckhbw %%mm4, %%mm7                 \n\t"
2845 
+            "punpcklbw %%mm4, %%mm6                 \n\t"
2846 
+
2847 
+            NEXT //7
2848 
+            "mov %4, %0                             \n\t"
2849 
+            "add %1, %0                             \n\t"
2850 
+            PREV //0
2851 
+            "movq %%mm0, 80(%3)                     \n\t"
2852 
+            "movq %%mm1, 88(%3)                     \n\t"
2853 
+
2854 
+            PREV //1
2855 
+            "paddw %%mm6, %%mm0                     \n\t"
2856 
+            "paddw %%mm7, %%mm1                     \n\t"
2857 
+            "movq %%mm0, 96(%3)                     \n\t"
2858 
+            "movq %%mm1, 104(%3)                    \n\t"
2859 
+
2860 
+            PREV //2
2861 
+            "paddw %%mm6, %%mm0                     \n\t"
2862 
+            "paddw %%mm7, %%mm1                     \n\t"
2863 
+            "movq %%mm0, 112(%3)                    \n\t"
2864 
+            "movq %%mm1, 120(%3)                    \n\t"
2865 
+
2866 
+            PREV //3
2867 
+            "paddw %%mm6, %%mm0                     \n\t"
2868 
+            "paddw %%mm7, %%mm1                     \n\t"
2869 
+            "movq %%mm0, 128(%3)                    \n\t"
2870 
+            "movq %%mm1, 136(%3)                    \n\t"
2871 
+
2872 
+            PREV //4
2873 
+            "paddw %%mm6, %%mm0                     \n\t"
2874 
+            "paddw %%mm7, %%mm1                     \n\t"
2875 
+            "movq %%mm0, 144(%3)                    \n\t"
2876 
+            "movq %%mm1, 152(%3)                    \n\t"
2877 
+
2878 
+            "mov %4, %0                             \n\t" //FIXME
2879 
+
2880 
+            : "+&r"(src)
2881 
+            : "r" ((long)step), "m" (c->pQPb), "r"(sums), "g"(src)
2882 
+        );
2883 
+
2884 
+        src+= step; // src points to begin of the 8x8 Block
2885 
+
2886 
+        asm volatile(
2887 
+            "movq %4, %%mm6                         \n\t"
2888 
+            "pcmpeqb %%mm5, %%mm5                   \n\t"
2889 
+            "pxor %%mm6, %%mm5                      \n\t"
2890 
+            "pxor %%mm7, %%mm7                      \n\t"
2891 
+
2892 
+            "1:                                     \n\t"
2893 
+            "movq (%1), %%mm0                       \n\t"
2894 
+            "movq 8(%1), %%mm1                      \n\t"
2895 
+            "paddw 32(%1), %%mm0                    \n\t"
2896 
+            "paddw 40(%1), %%mm1                    \n\t"
2897 
+            "movq (%0, %3), %%mm2                   \n\t"
2898 
+            "movq %%mm2, %%mm3                      \n\t"
2899 
+            "movq %%mm2, %%mm4                      \n\t"
2900 
+            "punpcklbw %%mm7, %%mm2                 \n\t"
2901 
+            "punpckhbw %%mm7, %%mm3                 \n\t"
2902 
+            "paddw %%mm2, %%mm0                     \n\t"
2903 
+            "paddw %%mm3, %%mm1                     \n\t"
2904 
+            "paddw %%mm2, %%mm0                     \n\t"
2905 
+            "paddw %%mm3, %%mm1                     \n\t"
2906 
+            "psrlw $4, %%mm0                        \n\t"
2907 
+            "psrlw $4, %%mm1                        \n\t"
2908 
+            "packuswb %%mm1, %%mm0                  \n\t"
2909 
+            "pand %%mm6, %%mm0                      \n\t"
2910 
+            "pand %%mm5, %%mm4                      \n\t"
2911 
+            "por %%mm4, %%mm0                       \n\t"
2912 
+            "movq %%mm0, (%0, %3)                   \n\t"
2913 
+            "add $16, %1                            \n\t"
2914 
+            "add %2, %0                             \n\t"
2915 
+            " js 1b                                 \n\t"
2916 
+
2917 
+            : "+r"(offset), "+r"(temp_sums)
2918 
+            : "r" ((long)step), "r"(src - offset), "m"(both_masks)
2919 
+        );
2920 
+    }else
2921 
+        src+= step; // src points to begin of the 8x8 Block
2922 
+
2923 
+    if(eq_mask != -1LL){
2924 
+        uint8_t *temp_src= src;
2925 
+        asm volatile(
2926 
+            "pxor %%mm7, %%mm7                      \n\t"
2927 
+            "lea -40(%%"REG_SP"), %%"REG_c"         \n\t" // make space for 4 8-byte vars
2928 
+            "and "ALIGN_MASK", %%"REG_c"            \n\t" // align
2929 2929 
 //      0       1       2       3       4       5       6       7       8       9
2930 2930 
 //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %1+8%1  ecx+4%1
2931 2931
2932 
-                "movq (%0), %%mm0                       \n\t"
2933 
-                "movq %%mm0, %%mm1                      \n\t"
2934 
-                "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
2935 
-                "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
2936 
-
2937 
-                "movq (%0, %1), %%mm2                   \n\t"
2938 
-                "lea (%0, %1, 2), %%"REG_a"             \n\t"
2939 
-                "movq %%mm2, %%mm3                      \n\t"
2940 
-                "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
2941 
-                "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
2942 
-
2943 
-                "movq (%%"REG_a"), %%mm4                \n\t"
2944 
-                "movq %%mm4, %%mm5                      \n\t"
2945 
-                "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
2946 
-                "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
2947 
-
2948 
-                "paddw %%mm0, %%mm0                     \n\t" // 2L0
2949 
-                "paddw %%mm1, %%mm1                     \n\t" // 2H0
2950 
-                "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
2951 
-                "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
2952 
-                "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
2953 
-                "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
2954 
-
2955 
-                "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
2956 
-                "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
2957 
-                "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
2958 
-                "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
2959 
-
2960 
-                "movq (%%"REG_a", %1), %%mm2            \n\t"
2961 
-                "movq %%mm2, %%mm3                      \n\t"
2962 
-                "punpcklbw %%mm7, %%mm2                 \n\t" // L3
2963 
-                "punpckhbw %%mm7, %%mm3                 \n\t" // H3
2964 
-
2965 
-                "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
2966 
-                "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
2967 
-                "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2968 
-                "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2969 
-                "movq %%mm0, (%%"REG_c")                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2970 
-                "movq %%mm1, 8(%%"REG_c")               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2971 
-
2972 
-                "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
2973 
-                "movq %%mm0, %%mm1                      \n\t"
2974 
-                "punpcklbw %%mm7, %%mm0                 \n\t" // L4
2975 
-                "punpckhbw %%mm7, %%mm1                 \n\t" // H4
2976 
-
2977 
-                "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
2978 
-                "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
2979 
-                "movq %%mm2, 16(%%"REG_c")              \n\t" // L3 - L4
2980 
-                "movq %%mm3, 24(%%"REG_c")              \n\t" // H3 - H4
2981 
-                "paddw %%mm4, %%mm4                     \n\t" // 2L2
2982 
-                "paddw %%mm5, %%mm5                     \n\t" // 2H2
2983 
-                "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
2984 
-                "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
2985 
-
2986 
-                "lea (%%"REG_a", %1), %0                \n\t"
2987 
-                "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
2988 
-                "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
2989 
-                "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
2990 
-                "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
2932 
+            "movq (%0), %%mm0                       \n\t"
2933 
+            "movq %%mm0, %%mm1                      \n\t"
2934 
+            "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0
2935 
+            "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0
2936 
+
2937 
+            "movq (%0, %1), %%mm2                   \n\t"
2938 
+            "lea (%0, %1, 2), %%"REG_a"             \n\t"
2939 
+            "movq %%mm2, %%mm3                      \n\t"
2940 
+            "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1
2941 
+            "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1
2942 
+
2943 
+            "movq (%%"REG_a"), %%mm4                \n\t"
2944 
+            "movq %%mm4, %%mm5                      \n\t"
2945 
+            "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2
2946 
+            "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2
2947 
+
2948 
+            "paddw %%mm0, %%mm0                     \n\t" // 2L0
2949 
+            "paddw %%mm1, %%mm1                     \n\t" // 2H0
2950 
+            "psubw %%mm4, %%mm2                     \n\t" // L1 - L2
2951 
+            "psubw %%mm5, %%mm3                     \n\t" // H1 - H2
2952 
+            "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2
2953 
+            "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2
2954 
+
2955 
+            "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2
2956 
+            "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2
2957 
+            "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2
2958 
+            "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2
2959 
+
2960 
+            "movq (%%"REG_a", %1), %%mm2            \n\t"
2961 
+            "movq %%mm2, %%mm3                      \n\t"
2962 
+            "punpcklbw %%mm7, %%mm2                 \n\t" // L3
2963 
+            "punpckhbw %%mm7, %%mm3                 \n\t" // H3
2964 
+
2965 
+            "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3
2966 
+            "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3
2967 
+            "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2968 
+            "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2969 
+            "movq %%mm0, (%%"REG_c")                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2970 
+            "movq %%mm1, 8(%%"REG_c")               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2971 
+
2972 
+            "movq (%%"REG_a", %1, 2), %%mm0         \n\t"
2973 
+            "movq %%mm0, %%mm1                      \n\t"
2974 
+            "punpcklbw %%mm7, %%mm0                 \n\t" // L4
2975 
+            "punpckhbw %%mm7, %%mm1                 \n\t" // H4
2976 
+
2977 
+            "psubw %%mm0, %%mm2                     \n\t" // L3 - L4
2978 
+            "psubw %%mm1, %%mm3                     \n\t" // H3 - H4
2979 
+            "movq %%mm2, 16(%%"REG_c")              \n\t" // L3 - L4
2980 
+            "movq %%mm3, 24(%%"REG_c")              \n\t" // H3 - H4
2981 
+            "paddw %%mm4, %%mm4                     \n\t" // 2L2
2982 
+            "paddw %%mm5, %%mm5                     \n\t" // 2H2
2983 
+            "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4
2984 
+            "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4
2985 
+
2986 
+            "lea (%%"REG_a", %1), %0                \n\t"
2987 
+            "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4
2988 
+            "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4
2989 
+            "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4
2990 
+            "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4
2991 2991 
 //50 opcodes so far
2992 
-                "movq (%0, %1, 2), %%mm2                \n\t"
2993 
-                "movq %%mm2, %%mm3                      \n\t"
2994 
-                "punpcklbw %%mm7, %%mm2                 \n\t" // L5
2995 
-                "punpckhbw %%mm7, %%mm3                 \n\t" // H5
2996 
-                "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
2997 
-                "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
2998 
-                "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
2999 
-                "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
3000 
-
3001 
-                "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
3002 
-                "punpcklbw %%mm7, %%mm6                 \n\t" // L6
3003 
-                "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
3004 
-                "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
3005 
-                "punpckhbw %%mm7, %%mm6                 \n\t" // H6
3006 
-                "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
3007 
-
3008 
-                "paddw %%mm0, %%mm0                     \n\t" // 2L4
3009 
-                "paddw %%mm1, %%mm1                     \n\t" // 2H4
3010 
-                "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
3011 
-                "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
3012 
-
3013 
-                "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
3014 
-                "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
3015 
-                "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
3016 
-                "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
3017 
-
3018 
-                "movq (%0, %1, 4), %%mm2                \n\t"
3019 
-                "movq %%mm2, %%mm3                      \n\t"
3020 
-                "punpcklbw %%mm7, %%mm2                 \n\t" // L7
3021 
-                "punpckhbw %%mm7, %%mm3                 \n\t" // H7
3022 
-
3023 
-                "paddw %%mm2, %%mm2                     \n\t" // 2L7
3024 
-                "paddw %%mm3, %%mm3                     \n\t" // 2H7
3025 
-                "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
3026 
-                "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
3027 
-
3028 
-                "movq (%%"REG_c"), %%mm2                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
3029 
-                "movq 8(%%"REG_c"), %%mm3               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2992 
+            "movq (%0, %1, 2), %%mm2                \n\t"
2993 
+            "movq %%mm2, %%mm3                      \n\t"
2994 
+            "punpcklbw %%mm7, %%mm2                 \n\t" // L5
2995 
+            "punpckhbw %%mm7, %%mm3                 \n\t" // H5
2996 
+            "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5
2997 
+            "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5
2998 
+            "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5
2999 
+            "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5
3000 
+
3001 
+            "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
3002 
+            "punpcklbw %%mm7, %%mm6                 \n\t" // L6
3003 
+            "psubw %%mm6, %%mm2                     \n\t" // L5 - L6
3004 
+            "movq (%%"REG_a", %1, 4), %%mm6         \n\t"
3005 
+            "punpckhbw %%mm7, %%mm6                 \n\t" // H6
3006 
+            "psubw %%mm6, %%mm3                     \n\t" // H5 - H6
3007 
+
3008 
+            "paddw %%mm0, %%mm0                     \n\t" // 2L4
3009 
+            "paddw %%mm1, %%mm1                     \n\t" // 2H4
3010 
+            "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6
3011 
+            "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6
3012 
+
3013 
+            "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6
3014 
+            "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6
3015 
+            "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6
3016 
+            "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6
3017 
+
3018 
+            "movq (%0, %1, 4), %%mm2                \n\t"
3019 
+            "movq %%mm2, %%mm3                      \n\t"
3020 
+            "punpcklbw %%mm7, %%mm2                 \n\t" // L7
3021 
+            "punpckhbw %%mm7, %%mm3                 \n\t" // H7
3022 
+
3023 
+            "paddw %%mm2, %%mm2                     \n\t" // 2L7
3024 
+            "paddw %%mm3, %%mm3                     \n\t" // 2H7
3025 
+            "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7
3026 
+            "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7
3027 
+
3028 
+            "movq (%%"REG_c"), %%mm2                \n\t" // 2L0 - 5L1 + 5L2 - 2L3
3029 
+            "movq 8(%%"REG_c"), %%mm3               \n\t" // 2H0 - 5H1 + 5H2 - 2H3
3030 3030
3031 3031 
 #ifdef HAVE_MMX2
3032 
-                "movq %%mm7, %%mm6                      \n\t" // 0
3033 
-                "psubw %%mm0, %%mm6                     \n\t"
3034 
-                "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
3035 
-                "movq %%mm7, %%mm6                      \n\t" // 0
3036 
-                "psubw %%mm1, %%mm6                     \n\t"
3037 
-                "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
3038 
-                "movq %%mm7, %%mm6                      \n\t" // 0
3039 
-                "psubw %%mm2, %%mm6                     \n\t"
3040 
-                "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
3041 
-                "movq %%mm7, %%mm6                      \n\t" // 0
3042 
-                "psubw %%mm3, %%mm6                     \n\t"
3043 
-                "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
3032 
+            "movq %%mm7, %%mm6                      \n\t" // 0
3033 
+            "psubw %%mm0, %%mm6                     \n\t"
3034 
+            "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
3035 
+            "movq %%mm7, %%mm6                      \n\t" // 0
3036 
+            "psubw %%mm1, %%mm6                     \n\t"
3037 
+            "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
3038 
+            "movq %%mm7, %%mm6                      \n\t" // 0
3039 
+            "psubw %%mm2, %%mm6                     \n\t"
3040 
+            "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
3041 
+            "movq %%mm7, %%mm6                      \n\t" // 0
3042 
+            "psubw %%mm3, %%mm6                     \n\t"
3043 
+            "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
3044 3044 
 #else
3045 
-                "movq %%mm7, %%mm6                      \n\t" // 0
3046 
-                "pcmpgtw %%mm0, %%mm6                   \n\t"
3047 
-                "pxor %%mm6, %%mm0                      \n\t"
3048 
-                "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
3049 
-                "movq %%mm7, %%mm6                      \n\t" // 0
3050 
-                "pcmpgtw %%mm1, %%mm6                   \n\t"
3051 
-                "pxor %%mm6, %%mm1                      \n\t"
3052 
-                "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
3053 
-                "movq %%mm7, %%mm6                      \n\t" // 0
3054 
-                "pcmpgtw %%mm2, %%mm6                   \n\t"
3055 
-                "pxor %%mm6, %%mm2                      \n\t"
3056 
-                "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
3057 
-                "movq %%mm7, %%mm6                      \n\t" // 0
3058 
-                "pcmpgtw %%mm3, %%mm6                   \n\t"
3059 
-                "pxor %%mm6, %%mm3                      \n\t"
3060 
-                "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
3045 
+            "movq %%mm7, %%mm6                      \n\t" // 0
3046 
+            "pcmpgtw %%mm0, %%mm6                   \n\t"
3047 
+            "pxor %%mm6, %%mm0                      \n\t"
3048 
+            "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
3049 
+            "movq %%mm7, %%mm6                      \n\t" // 0
3050 
+            "pcmpgtw %%mm1, %%mm6                   \n\t"
3051 
+            "pxor %%mm6, %%mm1                      \n\t"
3052 
+            "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
3053 
+            "movq %%mm7, %%mm6                      \n\t" // 0
3054 
+            "pcmpgtw %%mm2, %%mm6                   \n\t"
3055 
+            "pxor %%mm6, %%mm2                      \n\t"
3056 
+            "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
3057 
+            "movq %%mm7, %%mm6                      \n\t" // 0
3058 
+            "pcmpgtw %%mm3, %%mm6                   \n\t"
3059 
+            "pxor %%mm6, %%mm3                      \n\t"
3060 
+            "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
3061 3061 
 #endif
3062 3062
3063 3063 
 #ifdef HAVE_MMX2
3064 
-                "pminsw %%mm2, %%mm0                    \n\t"
3065 
-                "pminsw %%mm3, %%mm1                    \n\t"
3064 
+            "pminsw %%mm2, %%mm0                    \n\t"
3065 
+            "pminsw %%mm3, %%mm1                    \n\t"
3066 3066 
 #else
3067 
-                "movq %%mm0, %%mm6                      \n\t"
3068 
-                "psubusw %%mm2, %%mm6                   \n\t"
3069 
-                "psubw %%mm6, %%mm0                     \n\t"
3070 
-                "movq %%mm1, %%mm6                      \n\t"
3071 
-                "psubusw %%mm3, %%mm6                   \n\t"
3072 
-                "psubw %%mm6, %%mm1                     \n\t"
3067 
+            "movq %%mm0, %%mm6                      \n\t"
3068 
+            "psubusw %%mm2, %%mm6                   \n\t"
3069 
+            "psubw %%mm6, %%mm0                     \n\t"
3070 
+            "movq %%mm1, %%mm6                      \n\t"
3071 
+            "psubusw %%mm3, %%mm6                   \n\t"
3072 
+            "psubw %%mm6, %%mm1                     \n\t"
3073 3073 
 #endif
3074 3074
3075 
-                "movd %2, %%mm2                         \n\t" // QP
3076 
-                "punpcklbw %%mm7, %%mm2                 \n\t"
3075 
+            "movd %2, %%mm2                         \n\t" // QP
3076 
+            "punpcklbw %%mm7, %%mm2                 \n\t"
3077 3077
3078 
-                "movq %%mm7, %%mm6                      \n\t" // 0
3079 
-                "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
3080 
-                "pxor %%mm6, %%mm4                      \n\t"
3081 
-                "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
3082 
-                "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
3083 
-                "pxor %%mm7, %%mm5                      \n\t"
3084 
-                "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
3078 
+            "movq %%mm7, %%mm6                      \n\t" // 0
3079 
+            "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
3080 
+            "pxor %%mm6, %%mm4                      \n\t"
3081 
+            "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
3082 
+            "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
3083 
+            "pxor %%mm7, %%mm5                      \n\t"
3084 
+            "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
3085 3085 
 // 100 opcodes
3086 
-                "psllw $3, %%mm2                        \n\t" // 8QP
3087 
-                "movq %%mm2, %%mm3                      \n\t" // 8QP
3088 
-                "pcmpgtw %%mm4, %%mm2                   \n\t"
3089 
-                "pcmpgtw %%mm5, %%mm3                   \n\t"
3090 
-                "pand %%mm2, %%mm4                      \n\t"
3091 
-                "pand %%mm3, %%mm5                      \n\t"
3092 
-
3093 
-
3094 
-                "psubusw %%mm0, %%mm4                   \n\t" // hd
3095 
-                "psubusw %%mm1, %%mm5                   \n\t" // ld
3096 
-
3097 
-
3098 
-                "movq "MANGLE(w05)", %%mm2              \n\t" // 5
3099 
-                "pmullw %%mm2, %%mm4                    \n\t"
3100 
-                "pmullw %%mm2, %%mm5                    \n\t"
3101 
-                "movq "MANGLE(w20)", %%mm2              \n\t" // 32
3102 
-                "paddw %%mm2, %%mm4                     \n\t"
3103 
-                "paddw %%mm2, %%mm5                     \n\t"
3104 
-                "psrlw $6, %%mm4                        \n\t"
3105 
-                "psrlw $6, %%mm5                        \n\t"
3106 
-
3107 
-                "movq 16(%%"REG_c"), %%mm0              \n\t" // L3 - L4
3108 
-                "movq 24(%%"REG_c"), %%mm1              \n\t" // H3 - H4
3109 
-
3110 
-                "pxor %%mm2, %%mm2                      \n\t"
3111 
-                "pxor %%mm3, %%mm3                      \n\t"
3112 
-
3113 
-                "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
3114 
-                "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
3115 
-                "pxor %%mm2, %%mm0                      \n\t"
3116 
-                "pxor %%mm3, %%mm1                      \n\t"
3117 
-                "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
3118 
-                "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
3119 
-                "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
3120 
-                "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
3121 
-
3122 
-                "pxor %%mm6, %%mm2                      \n\t"
3123 
-                "pxor %%mm7, %%mm3                      \n\t"
3124 
-                "pand %%mm2, %%mm4                      \n\t"
3125 
-                "pand %%mm3, %%mm5                      \n\t"
3086 
+            "psllw $3, %%mm2                        \n\t" // 8QP
3087 
+            "movq %%mm2, %%mm3                      \n\t" // 8QP
3088 
+            "pcmpgtw %%mm4, %%mm2                   \n\t"
3089 
+            "pcmpgtw %%mm5, %%mm3                   \n\t"
3090 
+            "pand %%mm2, %%mm4                      \n\t"
3091 
+            "pand %%mm3, %%mm5                      \n\t"
3092 
+
3093 
+
3094 
+            "psubusw %%mm0, %%mm4                   \n\t" // hd
3095 
+            "psubusw %%mm1, %%mm5                   \n\t" // ld
3096 
+
3097 
+
3098 
+            "movq "MANGLE(w05)", %%mm2              \n\t" // 5
3099 
+            "pmullw %%mm2, %%mm4                    \n\t"
3100 
+            "pmullw %%mm2, %%mm5                    \n\t"
3101 
+            "movq "MANGLE(w20)", %%mm2              \n\t" // 32
3102 
+            "paddw %%mm2, %%mm4                     \n\t"
3103 
+            "paddw %%mm2, %%mm5                     \n\t"
3104 
+            "psrlw $6, %%mm4                        \n\t"
3105 
+            "psrlw $6, %%mm5                        \n\t"
3106 
+
3107 
+            "movq 16(%%"REG_c"), %%mm0              \n\t" // L3 - L4
3108 
+            "movq 24(%%"REG_c"), %%mm1              \n\t" // H3 - H4
3109 
+
3110 
+            "pxor %%mm2, %%mm2                      \n\t"
3111 
+            "pxor %%mm3, %%mm3                      \n\t"
3112 
+
3113 
+            "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)
3114 
+            "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)
3115 
+            "pxor %%mm2, %%mm0                      \n\t"
3116 
+            "pxor %%mm3, %%mm1                      \n\t"
3117 
+            "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|
3118 
+            "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|
3119 
+            "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2
3120 
+            "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2
3121 
+
3122 
+            "pxor %%mm6, %%mm2                      \n\t"
3123 
+            "pxor %%mm7, %%mm3                      \n\t"
3124 
+            "pand %%mm2, %%mm4                      \n\t"
3125 
+            "pand %%mm3, %%mm5                      \n\t"
3126 3126
3127 3127 
 #ifdef HAVE_MMX2
3128 
-                "pminsw %%mm0, %%mm4                    \n\t"
3129 
-                "pminsw %%mm1, %%mm5                    \n\t"
3128 
+            "pminsw %%mm0, %%mm4                    \n\t"
3129 
+            "pminsw %%mm1, %%mm5                    \n\t"
3130 3130 
 #else
3131 
-                "movq %%mm4, %%mm2                      \n\t"
3132 
-                "psubusw %%mm0, %%mm2                   \n\t"
3133 
-                "psubw %%mm2, %%mm4                     \n\t"
3134 
-                "movq %%mm5, %%mm2                      \n\t"
3135 
-                "psubusw %%mm1, %%mm2                   \n\t"
3136 
-                "psubw %%mm2, %%mm5                     \n\t"
3131 
+            "movq %%mm4, %%mm2                      \n\t"
3132 
+            "psubusw %%mm0, %%mm2                   \n\t"
3133 
+            "psubw %%mm2, %%mm4                     \n\t"
3134 
+            "movq %%mm5, %%mm2                      \n\t"
3135 
+            "psubusw %%mm1, %%mm2                   \n\t"
3136 
+            "psubw %%mm2, %%mm5                     \n\t"
3137 3137 
 #endif
3138 
-                "pxor %%mm6, %%mm4                      \n\t"
3139 
-                "pxor %%mm7, %%mm5                      \n\t"
3140 
-                "psubw %%mm6, %%mm4                     \n\t"
3141 
-                "psubw %%mm7, %%mm5                     \n\t"
3142 
-                "packsswb %%mm5, %%mm4                  \n\t"
3143 
-                "movq %3, %%mm1                         \n\t"
3144 
-                "pandn %%mm4, %%mm1                     \n\t"
3145 
-                "movq (%0), %%mm0                       \n\t"
3146 
-                "paddb   %%mm1, %%mm0                   \n\t"
3147 
-                "movq %%mm0, (%0)                       \n\t"
3148 
-                "movq (%0, %1), %%mm0                   \n\t"
3149 
-                "psubb %%mm1, %%mm0                     \n\t"
3150 
-                "movq %%mm0, (%0, %1)                   \n\t"
3151 
-
3152 
-                : "+r" (temp_src)
3153 
-                : "r" ((long)step), "m" (c->pQPb), "m"(eq_mask)
3154 
-                : "%"REG_a, "%"REG_c
3155 
-                );
3156 
-        }
3138 
+            "pxor %%mm6, %%mm4                      \n\t"
3139 
+            "pxor %%mm7, %%mm5                      \n\t"
3140 
+            "psubw %%mm6, %%mm4                     \n\t"
3141 
+            "psubw %%mm7, %%mm5                     \n\t"
3142 
+            "packsswb %%mm5, %%mm4                  \n\t"
3143 
+            "movq %3, %%mm1                         \n\t"
3144 
+            "pandn %%mm4, %%mm1                     \n\t"
3145 
+            "movq (%0), %%mm0                       \n\t"
3146 
+            "paddb   %%mm1, %%mm0                   \n\t"
3147 
+            "movq %%mm0, (%0)                       \n\t"
3148 
+            "movq (%0, %1), %%mm0                   \n\t"
3149 
+            "psubb %%mm1, %%mm0                     \n\t"
3150 
+            "movq %%mm0, (%0, %1)                   \n\t"
3151 
+
3152 
+            : "+r" (temp_src)
3153 
+            : "r" ((long)step), "m" (c->pQPb), "m"(eq_mask)
3154 
+            : "%"REG_a, "%"REG_c
3155 
+        );
3156 
+    }
3157 3157 
 /*if(step==16){
3158 3158 
     STOP_TIMER("step16")
3159 3159 
 }else{
... ... 
@@ -3163,7 +3105,7 @@ asm volatile(
3163 3163 
 #endif //HAVE_MMX
3164 3164
3165 3165 
 static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
3166 
-        const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);
3166 
+                                const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);
3167 3167
3168 3168 
 /**
3169 3169 
  * Copies a block from src to dst and fixes the blacklevel.
... ... 
@@ -3172,69 +3114,68 @@ static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[
3172 3172 
 #undef SCALED_CPY
3173 3173
3174 3174 
 static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t src[], int srcStride,
3175 
-        int levelFix, int64_t *packedOffsetAndScale)
3175 
+                                     int levelFix, int64_t *packedOffsetAndScale)
3176 3176 
 {
3177 3177 
 #ifndef HAVE_MMX
3178 
-        int i;
3178 
+    int i;
3179 3179 
 #endif
3180 
-        if(levelFix)
3181 
-        {
3180 
+    if(levelFix){
3182 3181 
 #ifdef HAVE_MMX
3183 
-                asm volatile(
3184 
-                        "movq (%%"REG_a"), %%mm2        \n\t" // packedYOffset
3185 
-                        "movq 8(%%"REG_a"), %%mm3       \n\t" // packedYScale
3186 
-                        "lea (%2,%4), %%"REG_a"         \n\t"
3187 
-                        "lea (%3,%5), %%"REG_d"         \n\t"
3188 
-                        "pxor %%mm4, %%mm4              \n\t"
3182 
+    asm volatile(
3183 
+        "movq (%%"REG_a"), %%mm2        \n\t" // packedYOffset
3184 
+        "movq 8(%%"REG_a"), %%mm3       \n\t" // packedYScale
3185 
+        "lea (%2,%4), %%"REG_a"         \n\t"
3186 
+        "lea (%3,%5), %%"REG_d"         \n\t"
3187 
+        "pxor %%mm4, %%mm4              \n\t"
3189 3188 
 #ifdef HAVE_MMX2
3190 3189 
 #define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                                \
3191 
-                        "movq " #src1 ", %%mm0          \n\t"\
3192 
-                        "movq " #src1 ", %%mm5          \n\t"\
3193 
-                        "movq " #src2 ", %%mm1          \n\t"\
3194 
-                        "movq " #src2 ", %%mm6          \n\t"\
3195 
-                        "punpcklbw %%mm0, %%mm0         \n\t"\
3196 
-                        "punpckhbw %%mm5, %%mm5         \n\t"\
3197 
-                        "punpcklbw %%mm1, %%mm1         \n\t"\
3198 
-                        "punpckhbw %%mm6, %%mm6         \n\t"\
3199 
-                        "pmulhuw %%mm3, %%mm0           \n\t"\
3200 
-                        "pmulhuw %%mm3, %%mm5           \n\t"\
3201 
-                        "pmulhuw %%mm3, %%mm1           \n\t"\
3202 
-                        "pmulhuw %%mm3, %%mm6           \n\t"\
3203 
-                        "psubw %%mm2, %%mm0             \n\t"\
3204 
-                        "psubw %%mm2, %%mm5             \n\t"\
3205 
-                        "psubw %%mm2, %%mm1             \n\t"\
3206 
-                        "psubw %%mm2, %%mm6             \n\t"\
3207 
-                        "packuswb %%mm5, %%mm0          \n\t"\
3208 
-                        "packuswb %%mm6, %%mm1          \n\t"\
3209 
-                        "movq %%mm0, " #dst1 "          \n\t"\
3210 
-                        "movq %%mm1, " #dst2 "          \n\t"\
3190 
+        "movq " #src1 ", %%mm0          \n\t"\
3191 
+        "movq " #src1 ", %%mm5          \n\t"\
3192 
+        "movq " #src2 ", %%mm1          \n\t"\
3193 
+        "movq " #src2 ", %%mm6          \n\t"\
3194 
+        "punpcklbw %%mm0, %%mm0         \n\t"\
3195 
+        "punpckhbw %%mm5, %%mm5         \n\t"\
3196 
+        "punpcklbw %%mm1, %%mm1         \n\t"\
3197 
+        "punpckhbw %%mm6, %%mm6         \n\t"\
3198 
+        "pmulhuw %%mm3, %%mm0           \n\t"\
3199 
+        "pmulhuw %%mm3, %%mm5           \n\t"\
3200 
+        "pmulhuw %%mm3, %%mm1           \n\t"\
3201 
+        "pmulhuw %%mm3, %%mm6           \n\t"\
3202 
+        "psubw %%mm2, %%mm0             \n\t"\
3203 
+        "psubw %%mm2, %%mm5             \n\t"\
3204 
+        "psubw %%mm2, %%mm1             \n\t"\
3205 
+        "psubw %%mm2, %%mm6             \n\t"\
3206 
+        "packuswb %%mm5, %%mm0          \n\t"\
3207 
+        "packuswb %%mm6, %%mm1          \n\t"\
3208 
+        "movq %%mm0, " #dst1 "          \n\t"\
3209 
+        "movq %%mm1, " #dst2 "          \n\t"\
3211 3210
3212 3211 
 #else //HAVE_MMX2
3213 3212 
 #define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                        \
3214 
-                        "movq " #src1 ", %%mm0          \n\t"\
3215 
-                        "movq " #src1 ", %%mm5          \n\t"\
3216 
-                        "punpcklbw %%mm4, %%mm0         \n\t"\
3217 
-                        "punpckhbw %%mm4, %%mm5         \n\t"\
3218 
-                        "psubw %%mm2, %%mm0             \n\t"\
3219 
-                        "psubw %%mm2, %%mm5             \n\t"\
3220 
-                        "movq " #src2 ", %%mm1          \n\t"\
3221 
-                        "psllw $6, %%mm0                \n\t"\
3222 
-                        "psllw $6, %%mm5                \n\t"\
3223 
-                        "pmulhw %%mm3, %%mm0            \n\t"\
3224 
-                        "movq " #src2 ", %%mm6          \n\t"\
3225 
-                        "pmulhw %%mm3, %%mm5            \n\t"\
3226 
-                        "punpcklbw %%mm4, %%mm1         \n\t"\
3227 
-                        "punpckhbw %%mm4, %%mm6         \n\t"\
3228 
-                        "psubw %%mm2, %%mm1             \n\t"\
3229 
-                        "psubw %%mm2, %%mm6             \n\t"\
3230 
-                        "psllw $6, %%mm1                \n\t"\
3231 
-                        "psllw $6, %%mm6                \n\t"\
3232 
-                        "pmulhw %%mm3, %%mm1            \n\t"\
3233 
-                        "pmulhw %%mm3, %%mm6            \n\t"\
3234 
-                        "packuswb %%mm5, %%mm0          \n\t"\
3235 
-                        "packuswb %%mm6, %%mm1          \n\t"\
3236 
-                        "movq %%mm0, " #dst1 "          \n\t"\
3237 
-                        "movq %%mm1, " #dst2 "          \n\t"\
3213 
+        "movq " #src1 ", %%mm0          \n\t"\
3214 
+        "movq " #src1 ", %%mm5          \n\t"\
3215 
+        "punpcklbw %%mm4, %%mm0         \n\t"\
3216 
+        "punpckhbw %%mm4, %%mm5         \n\t"\
3217 
+        "psubw %%mm2, %%mm0             \n\t"\
3218 
+        "psubw %%mm2, %%mm5             \n\t"\
3219 
+        "movq " #src2 ", %%mm1          \n\t"\
3220 
+        "psllw $6, %%mm0                \n\t"\
3221 
+        "psllw $6, %%mm5                \n\t"\
3222 
+        "pmulhw %%mm3, %%mm0            \n\t"\
3223 
+        "movq " #src2 ", %%mm6          \n\t"\
3224 
+        "pmulhw %%mm3, %%mm5            \n\t"\
3225 
+        "punpcklbw %%mm4, %%mm1         \n\t"\
3226 
+        "punpckhbw %%mm4, %%mm6         \n\t"\
3227 
+        "psubw %%mm2, %%mm1             \n\t"\
3228 
+        "psubw %%mm2, %%mm6             \n\t"\
3229 
+        "psllw $6, %%mm1                \n\t"\
3230 
+        "psllw $6, %%mm6                \n\t"\
3231 
+        "pmulhw %%mm3, %%mm1            \n\t"\
3232 
+        "pmulhw %%mm3, %%mm6            \n\t"\
3233 
+        "packuswb %%mm5, %%mm0          \n\t"\
3234 
+        "packuswb %%mm6, %%mm1          \n\t"\
3235 
+        "movq %%mm0, " #dst1 "          \n\t"\
3236 
+        "movq %%mm1, " #dst2 "          \n\t"\
3238 3237
3239 3238 
 #endif //HAVE_MMX2
3240 3239 
 #define SCALED_CPY(src1, src2, dst1, dst2)\
... ... 
@@ -3243,37 +3184,35 @@ static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t
3243 3243 
 SCALED_CPY((%2)       , (%2, %4)      , (%3)       , (%3, %5))
3244 3244 
 SCALED_CPY((%2, %4, 2), (%%REGa, %4, 2), (%3, %5, 2), (%%REGd, %5, 2))
3245 3245 
 SCALED_CPY((%2, %4, 4), (%%REGa, %4, 4), (%3, %5, 4), (%%REGd, %5, 4))
3246 
-                        "lea (%%"REG_a",%4,4), %%"REG_a"        \n\t"
3247 
-                        "lea (%%"REG_d",%5,4), %%"REG_d"        \n\t"
3246 
+        "lea (%%"REG_a",%4,4), %%"REG_a"        \n\t"
3247 
+        "lea (%%"REG_d",%5,4), %%"REG_d"        \n\t"
3248 3248 
 SCALED_CPY((%%REGa, %4), (%%REGa, %4, 2), (%%REGd, %5), (%%REGd, %5, 2))
3249 3249
3250 3250
3251 
-                        : "=&a" (packedOffsetAndScale)
3252 
-                        : "0" (packedOffsetAndScale),
3253 
-                        "r"(src),
3254 
-                        "r"(dst),
3255 
-                        "r" ((long)srcStride),
3256 
-                        "r" ((long)dstStride)
3257 
-                        : "%"REG_d
3258 
-                                        );
3251 
+        : "=&a" (packedOffsetAndScale)
3252 
+        : "0" (packedOffsetAndScale),
3253 
+        "r"(src),
3254 
+        "r"(dst),
3255 
+        "r" ((long)srcStride),
3256 
+        "r" ((long)dstStride)
3257 
+        : "%"REG_d
3258 
+    );
3259 3259 
 #else //HAVE_MMX
3260 
-        for(i=0; i<8; i++)
3261 
-                memcpy( &(dst[dstStride*i]),
3262 
-                        &(src[srcStride*i]), BLOCK_SIZE);
3260 
+    for(i=0; i<8; i++)
3261 
+        memcpy( &(dst[dstStride*i]),
3262 
+                &(src[srcStride*i]), BLOCK_SIZE);
3263 3263 
 #endif //HAVE_MMX
3264 
-        }
3265 
-        else
3266 
-        {
3264 
+    }else{
3267 3265 
 #ifdef HAVE_MMX
3268 
-        asm volatile(
3269 
-                "lea (%0,%2), %%"REG_a"                 \n\t"
3270 
-                "lea (%1,%3), %%"REG_d"                 \n\t"
3266 
+    asm volatile(
3267 
+        "lea (%0,%2), %%"REG_a"                 \n\t"
3268 
+        "lea (%1,%3), %%"REG_d"                 \n\t"
3271 3269
3272 3270 
 #define REAL_SIMPLE_CPY(src1, src2, dst1, dst2)                              \
3273 
-                "movq " #src1 ", %%mm0          \n\t"\
3274 
-                "movq " #src2 ", %%mm1          \n\t"\
3275 
-                "movq %%mm0, " #dst1 "          \n\t"\
3276 
-                "movq %%mm1, " #dst2 "          \n\t"\
3271 
+        "movq " #src1 ", %%mm0          \n\t"\
3272 
+        "movq " #src2 ", %%mm1          \n\t"\
3273 
+        "movq %%mm0, " #dst1 "          \n\t"\
3274 
+        "movq %%mm1, " #dst2 "          \n\t"\
3277 3275
3278 3276 
 #define SIMPLE_CPY(src1, src2, dst1, dst2)\
3279 3277 
    REAL_SIMPLE_CPY(src1, src2, dst1, dst2)
... ... 
@@ -3281,22 +3220,22 @@ SCALED_CPY((%%REGa, %4), (%%REGa, %4, 2), (%%REGd, %5), (%%REGd, %5, 2))
3281 3281 
 SIMPLE_CPY((%0)       , (%0, %2)       , (%1)       , (%1, %3))
3282 3282 
 SIMPLE_CPY((%0, %2, 2), (%%REGa, %2, 2), (%1, %3, 2), (%%REGd, %3, 2))
3283 3283 
 SIMPLE_CPY((%0, %2, 4), (%%REGa, %2, 4), (%1, %3, 4), (%%REGd, %3, 4))
3284 
-                "lea (%%"REG_a",%2,4), %%"REG_a"        \n\t"
3285 
-                "lea (%%"REG_d",%3,4), %%"REG_d"        \n\t"
3284 
+        "lea (%%"REG_a",%2,4), %%"REG_a"        \n\t"
3285 
+        "lea (%%"REG_d",%3,4), %%"REG_d"        \n\t"
3286 3286 
 SIMPLE_CPY((%%REGa, %2), (%%REGa, %2, 2), (%%REGd, %3), (%%REGd, %3, 2))
3287 3287
3288 
-                : : "r" (src),
3289 
-                "r" (dst),
3290 
-                "r" ((long)srcStride),
3291 
-                "r" ((long)dstStride)
3292 
-                : "%"REG_a, "%"REG_d
3293 
-        );
3288 
+        : : "r" (src),
3289 
+        "r" (dst),
3290 
+        "r" ((long)srcStride),
3291 
+        "r" ((long)dstStride)
3292 
+        : "%"REG_a, "%"REG_d
3293 
+    );
3294 3294 
 #else //HAVE_MMX
3295 
-        for(i=0; i<8; i++)
3296 
-                memcpy( &(dst[dstStride*i]),
3297 
-                        &(src[srcStride*i]), BLOCK_SIZE);
3295 
+    for(i=0; i<8; i++)
3296 
+        memcpy( &(dst[dstStride*i]),
3297 
+                &(src[srcStride*i]), BLOCK_SIZE);
3298 3298 
 #endif //HAVE_MMX
3299 
-        }
3299 
+    }
3300 3300 
 }
3301 3301
3302 3302 
 /**
... ... 
@@ -3305,23 +3244,22 @@ SIMPLE_CPY((%%REGa, %2), (%%REGa, %2, 2), (%%REGd, %3), (%%REGd, %3, 2))
3305 3305 
 static inline void RENAME(duplicate)(uint8_t src[], int stride)
3306 3306 
 {
3307 3307 
 #ifdef HAVE_MMX
3308 
-        asm volatile(
3309 
-                "movq (%0), %%mm0               \n\t"
3310 
-                "add %1, %0                     \n\t"
3311 
-                "movq %%mm0, (%0)               \n\t"
3312 
-                "movq %%mm0, (%0, %1)           \n\t"
3313 
-                "movq %%mm0, (%0, %1, 2)        \n\t"
3314 
-                : "+r" (src)
3315 
-                : "r" ((long)-stride)
3316 
-        );
3308 
+    asm volatile(
3309 
+        "movq (%0), %%mm0               \n\t"
3310 
+        "add %1, %0                     \n\t"
3311 
+        "movq %%mm0, (%0)               \n\t"
3312 
+        "movq %%mm0, (%0, %1)           \n\t"
3313 
+        "movq %%mm0, (%0, %1, 2)        \n\t"
3314 
+        : "+r" (src)
3315 
+        : "r" ((long)-stride)
3316 
+    );
3317 3317 
 #else
3318 
-        int i;
3319 
-        uint8_t *p=src;
3320 
-        for(i=0; i<3; i++)
3321 
-        {
3322 
-                p-= stride;
3323 
-                memcpy(p, src, 8);
3324 
-        }
3318 
+    int i;
3319 
+    uint8_t *p=src;
3320 
+    for(i=0; i<3; i++){
3321 
+        p-= stride;
3322 
+        memcpy(p, src, 8);
3323 
+    }
3325 3324 
 #endif
3326 3325 
 }
3327 3326
... ... 
@@ -3329,498 +3267,467 @@ static inline void RENAME(duplicate)(uint8_t src[], int stride)
3329 3329 
  * Filters array of bytes (Y or U or V values)
3330 3330 
  */
3331 3331 
 static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
3332 
-        const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)
3332 
+                                const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)
3333 3333 
 {
3334 
-        DECLARE_ALIGNED(8, PPContext, c)= *c2; //copy to stack for faster access
3335 
-        int x,y;
3334 
+    DECLARE_ALIGNED(8, PPContext, c)= *c2; //copy to stack for faster access
3335 
+    int x,y;
3336 3336 
 #ifdef COMPILE_TIME_MODE
3337 
-        const int mode= COMPILE_TIME_MODE;
3337 
+    const int mode= COMPILE_TIME_MODE;
3338 3338 
 #else
3339 
-        const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;
3339 
+    const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;
3340 3340 
 #endif
3341 
-        int black=0, white=255; // blackest black and whitest white in the picture
3342 
-        int QPCorrecture= 256*256;
3341 
+    int black=0, white=255; // blackest black and whitest white in the picture
3342 
+    int QPCorrecture= 256*256;
3343 3343
3344 
-        int copyAhead;
3344 
+    int copyAhead;
3345 3345 
 #ifdef HAVE_MMX
3346 
-        int i;
3346 
+    int i;
3347 3347 
 #endif
3348 3348
3349 
-        const int qpHShift= isColor ? 4-c.hChromaSubSample : 4;
3350 
-        const int qpVShift= isColor ? 4-c.vChromaSubSample : 4;
3349 
+    const int qpHShift= isColor ? 4-c.hChromaSubSample : 4;
3350 
+    const int qpVShift= isColor ? 4-c.vChromaSubSample : 4;
3351 3351
3352 
-        //FIXME remove
3353 
-        uint64_t * const yHistogram= c.yHistogram;
3354 
-        uint8_t * const tempSrc= srcStride > 0 ? c.tempSrc : c.tempSrc - 23*srcStride;
3355 
-        uint8_t * const tempDst= dstStride > 0 ? c.tempDst : c.tempDst - 23*dstStride;
3356 
-        //const int mbWidth= isColor ? (width+7)>>3 : (width+15)>>4;
3352 
+    //FIXME remove
3353 
+    uint64_t * const yHistogram= c.yHistogram;
3354 
+    uint8_t * const tempSrc= srcStride > 0 ? c.tempSrc : c.tempSrc - 23*srcStride;
3355 
+    uint8_t * const tempDst= dstStride > 0 ? c.tempDst : c.tempDst - 23*dstStride;
3356 
+    //const int mbWidth= isColor ? (width+7)>>3 : (width+15)>>4;
3357 3357
3358 3358 
 #ifdef HAVE_MMX
3359 
-        for(i=0; i<57; i++){
3360 
-                int offset= ((i*c.ppMode.baseDcDiff)>>8) + 1;
3361 
-                int threshold= offset*2 + 1;
3362 
-                c.mmxDcOffset[i]= 0x7F - offset;
3363 
-                c.mmxDcThreshold[i]= 0x7F - threshold;
3364 
-                c.mmxDcOffset[i]*= 0x0101010101010101LL;
3365 
-                c.mmxDcThreshold[i]*= 0x0101010101010101LL;
3366 
-        }
3359 
+    for(i=0; i<57; i++){
3360 
+        int offset= ((i*c.ppMode.baseDcDiff)>>8) + 1;
3361 
+        int threshold= offset*2 + 1;
3362 
+        c.mmxDcOffset[i]= 0x7F - offset;
3363 
+        c.mmxDcThreshold[i]= 0x7F - threshold;
3364 
+        c.mmxDcOffset[i]*= 0x0101010101010101LL;
3365 
+        c.mmxDcThreshold[i]*= 0x0101010101010101LL;
3366 
+    }
3367 3367 
 #endif
3368 3368
3369 
-        if(mode & CUBIC_IPOL_DEINT_FILTER) copyAhead=16;
3370 
-        else if(   (mode & LINEAR_BLEND_DEINT_FILTER)
3371 
-                || (mode & FFMPEG_DEINT_FILTER)
3372 
-                || (mode & LOWPASS5_DEINT_FILTER)) copyAhead=14;
3373 
-        else if(   (mode & V_DEBLOCK)
3374 
-                || (mode & LINEAR_IPOL_DEINT_FILTER)
3375 
-                || (mode & MEDIAN_DEINT_FILTER)
3376 
-                || (mode & V_A_DEBLOCK)) copyAhead=13;
3377 
-        else if(mode & V_X1_FILTER) copyAhead=11;
3378 
-//        else if(mode & V_RK1_FILTER) copyAhead=10;
3379 
-        else if(mode & DERING) copyAhead=9;
3380 
-        else copyAhead=8;
3381 
-
3382 
-        copyAhead-= 8;
3383 
-
3384 
-        if(!isColor)
3385 
-        {
3386 
-                uint64_t sum= 0;
3387 
-                int i;
3388 
-                uint64_t maxClipped;
3389 
-                uint64_t clipped;
3390 
-                double scale;
3369 
+    if(mode & CUBIC_IPOL_DEINT_FILTER) copyAhead=16;
3370 
+    else if(   (mode & LINEAR_BLEND_DEINT_FILTER)
3371 
+            || (mode & FFMPEG_DEINT_FILTER)
3372 
+            || (mode & LOWPASS5_DEINT_FILTER)) copyAhead=14;
3373 
+    else if(   (mode & V_DEBLOCK)
3374 
+            || (mode & LINEAR_IPOL_DEINT_FILTER)
3375 
+            || (mode & MEDIAN_DEINT_FILTER)
3376 
+            || (mode & V_A_DEBLOCK)) copyAhead=13;
3377 
+    else if(mode & V_X1_FILTER) copyAhead=11;
3378 
+//    else if(mode & V_RK1_FILTER) copyAhead=10;
3379 
+    else if(mode & DERING) copyAhead=9;
3380 
+    else copyAhead=8;
3381 
+
3382 
+    copyAhead-= 8;
3383 
+
3384 
+    if(!isColor){
3385 
+        uint64_t sum= 0;
3386 
+        int i;
3387 
+        uint64_t maxClipped;
3388 
+        uint64_t clipped;
3389 
+        double scale;
3391 3390
3392 
-                c.frameNum++;
3393 
-                // first frame is fscked so we ignore it
3394 
-                if(c.frameNum == 1) yHistogram[0]= width*height/64*15/256;
3391 
+        c.frameNum++;
3392 
+        // first frame is fscked so we ignore it
3393 
+        if(c.frameNum == 1) yHistogram[0]= width*height/64*15/256;
3395 3394
3396 
-                for(i=0; i<256; i++)
3397 
-                {
3398 
-                        sum+= yHistogram[i];
3399 
-                }
3395 
+        for(i=0; i<256; i++){
3396 
+            sum+= yHistogram[i];
3397 
+        }
3400 3398
3401 
-                /* We always get a completely black picture first. */
3402 
-                maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);
3399 
+        /* We always get a completely black picture first. */
3400 
+        maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);
3403 3401
3404 
-                clipped= sum;
3405 
-                for(black=255; black>0; black--)
3406 
-                {
3407 
-                        if(clipped < maxClipped) break;
3408 
-                        clipped-= yHistogram[black];
3409 
-                }
3402 
+        clipped= sum;
3403 
+        for(black=255; black>0; black--){
3404 
+            if(clipped < maxClipped) break;
3405 
+            clipped-= yHistogram[black];
3406 
+        }
3410 3407
3411 
-                clipped= sum;
3412 
-                for(white=0; white<256; white++)
3413 
-                {
3414 
-                        if(clipped < maxClipped) break;
3415 
-                        clipped-= yHistogram[white];
3416 
-                }
3408 
+        clipped= sum;
3409 
+        for(white=0; white<256; white++){
3410 
+            if(clipped < maxClipped) break;
3411 
+            clipped-= yHistogram[white];
3412 
+        }
3417 3413
3418 
-                scale= (double)(c.ppMode.maxAllowedY - c.ppMode.minAllowedY) / (double)(white-black);
3414 
+        scale= (double)(c.ppMode.maxAllowedY - c.ppMode.minAllowedY) / (double)(white-black);
3419 3415
3420 3416 
 #ifdef HAVE_MMX2
3421 
-                c.packedYScale= (uint16_t)(scale*256.0 + 0.5);
3422 
-                c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF;
3417 
+        c.packedYScale= (uint16_t)(scale*256.0 + 0.5);
3418 
+        c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF;
3423 3419 
 #else
3424 
-                c.packedYScale= (uint16_t)(scale*1024.0 + 0.5);
3425 
-                c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF;
3420 
+        c.packedYScale= (uint16_t)(scale*1024.0 + 0.5);
3421 
+        c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF;
3426 3422 
 #endif
3427 3423
3428 
-                c.packedYOffset|= c.packedYOffset<<32;
3429 
-                c.packedYOffset|= c.packedYOffset<<16;
3424 
+        c.packedYOffset|= c.packedYOffset<<32;
3425 
+        c.packedYOffset|= c.packedYOffset<<16;
3430 3426
3431 
-                c.packedYScale|= c.packedYScale<<32;
3432 
-                c.packedYScale|= c.packedYScale<<16;
3427 
+        c.packedYScale|= c.packedYScale<<32;
3428 
+        c.packedYScale|= c.packedYScale<<16;
3433 3429
3434 
-                if(mode & LEVEL_FIX)        QPCorrecture= (int)(scale*256*256 + 0.5);
3435 
-                else                        QPCorrecture= 256*256;
3436 
-        }
3437 
-        else
3438 
-        {
3439 
-                c.packedYScale= 0x0100010001000100LL;
3440 
-                c.packedYOffset= 0;
3441 
-                QPCorrecture= 256*256;
3442 
-        }
3430 
+        if(mode & LEVEL_FIX)        QPCorrecture= (int)(scale*256*256 + 0.5);
3431 
+        else                        QPCorrecture= 256*256;
3432 
+    }else{
3433 
+        c.packedYScale= 0x0100010001000100LL;
3434 
+        c.packedYOffset= 0;
3435 
+        QPCorrecture= 256*256;
3436 
+    }
3443 3437
3444 
-        /* copy & deinterlace first row of blocks */
3445 
-        y=-BLOCK_SIZE;
3446 
-        {
3447 
-                const uint8_t *srcBlock= &(src[y*srcStride]);
3448 
-                uint8_t *dstBlock= tempDst + dstStride;
3438 
+    /* copy & deinterlace first row of blocks */
3439 
+    y=-BLOCK_SIZE;
3440 
+    {
3441 
+        const uint8_t *srcBlock= &(src[y*srcStride]);
3442 
+        uint8_t *dstBlock= tempDst + dstStride;
3449 3443
3450 
-                // From this point on it is guaranteed that we can read and write 16 lines downward
3451 
-                // finish 1 block before the next otherwise we might have a problem
3452 
-                // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
3453 
-                for(x=0; x<width; x+=BLOCK_SIZE)
3454 
-                {
3444 
+        // From this point on it is guaranteed that we can read and write 16 lines downward
3445 
+        // finish 1 block before the next otherwise we might have a problem
3446 
+        // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
3447 
+        for(x=0; x<width; x+=BLOCK_SIZE){
3455 3448
3456 3449 
 #ifdef HAVE_MMX2
3457 3450 
 /*
3458 
-                        prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
3459 
-                        prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
3460 
-                        prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
3461 
-                        prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
3451 
+            prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
3452 
+            prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
3453 
+            prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
3454 
+            prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
3462 3455 
 */
3463 3456
3464 
-                        asm(
3465 
-                                "mov %4, %%"REG_a"              \n\t"
3466 
-                                "shr $2, %%"REG_a"              \n\t"
3467 
-                                "and $6, %%"REG_a"              \n\t"
3468 
-                                "add %5, %%"REG_a"              \n\t"
3469 
-                                "mov %%"REG_a", %%"REG_d"       \n\t"
3470 
-                                "imul %1, %%"REG_a"             \n\t"
3471 
-                                "imul %3, %%"REG_d"             \n\t"
3472 
-                                "prefetchnta 32(%%"REG_a", %0)  \n\t"
3473 
-                                "prefetcht0 32(%%"REG_d", %2)   \n\t"
3474 
-                                "add %1, %%"REG_a"              \n\t"
3475 
-                                "add %3, %%"REG_d"              \n\t"
3476 
-                                "prefetchnta 32(%%"REG_a", %0)  \n\t"
3477 
-                                "prefetcht0 32(%%"REG_d", %2)   \n\t"
3478 
-                        :: "r" (srcBlock), "r" ((long)srcStride), "r" (dstBlock), "r" ((long)dstStride),
3479 
-                        "g" ((long)x), "g" ((long)copyAhead)
3480 
-                        : "%"REG_a, "%"REG_d
3481 
-                        );
3457 
+            asm(
3458 
+                "mov %4, %%"REG_a"              \n\t"
3459 
+                "shr $2, %%"REG_a"              \n\t"
3460 
+                "and $6, %%"REG_a"              \n\t"
3461 
+                "add %5, %%"REG_a"              \n\t"
3462 
+                "mov %%"REG_a", %%"REG_d"       \n\t"
3463 
+                "imul %1, %%"REG_a"             \n\t"
3464 
+                "imul %3, %%"REG_d"             \n\t"
3465 
+                "prefetchnta 32(%%"REG_a", %0)  \n\t"
3466 
+                "prefetcht0 32(%%"REG_d", %2)   \n\t"
3467 
+                "add %1, %%"REG_a"              \n\t"
3468 
+                "add %3, %%"REG_d"              \n\t"
3469 
+                "prefetchnta 32(%%"REG_a", %0)  \n\t"
3470 
+                "prefetcht0 32(%%"REG_d", %2)   \n\t"
3471 
+                :: "r" (srcBlock), "r" ((long)srcStride), "r" (dstBlock), "r" ((long)dstStride),
3472 
+                "g" ((long)x), "g" ((long)copyAhead)
3473 
+                : "%"REG_a, "%"REG_d
3474 
+            );
3482 3475
3483 3476 
 #elif defined(HAVE_3DNOW)
3484 3477 
 //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
3485 
-/*                        prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
3486 
-                        prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
3487 
-                        prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
3488 
-                        prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
3478 
+/*          prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
3479 
+            prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
3480 
+            prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
3481 
+            prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
3489 3482 
 */
3490 3483 
 #endif
3491 3484
3492 
-                        RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,
3493 
-                                srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
3494 
-
3495 
-                        RENAME(duplicate)(dstBlock + dstStride*8, dstStride);
3496 
-
3497 
-                        if(mode & LINEAR_IPOL_DEINT_FILTER)
3498 
-                                RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
3499 
-                        else if(mode & LINEAR_BLEND_DEINT_FILTER)
3500 
-                                RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
3501 
-                        else if(mode & MEDIAN_DEINT_FILTER)
3502 
-                                RENAME(deInterlaceMedian)(dstBlock, dstStride);
3503 
-                        else if(mode & CUBIC_IPOL_DEINT_FILTER)
3504 
-                                RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
3505 
-                        else if(mode & FFMPEG_DEINT_FILTER)
3506 
-                                RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
3507 
-                        else if(mode & LOWPASS5_DEINT_FILTER)
3508 
-                                RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
3509 
-/*                        else if(mode & CUBIC_BLEND_DEINT_FILTER)
3510 
-                                RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3485 
+            RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,
3486 
+                              srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
3487 
+
3488 
+            RENAME(duplicate)(dstBlock + dstStride*8, dstStride);
3489 
+
3490 
+            if(mode & LINEAR_IPOL_DEINT_FILTER)
3491 
+                RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
3492 
+            else if(mode & LINEAR_BLEND_DEINT_FILTER)
3493 
+                RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
3494 
+            else if(mode & MEDIAN_DEINT_FILTER)
3495 
+                RENAME(deInterlaceMedian)(dstBlock, dstStride);
3496 
+            else if(mode & CUBIC_IPOL_DEINT_FILTER)
3497 
+                RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
3498 
+            else if(mode & FFMPEG_DEINT_FILTER)
3499 
+                RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
3500 
+            else if(mode & LOWPASS5_DEINT_FILTER)
3501 
+                RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
3502 
+/*          else if(mode & CUBIC_BLEND_DEINT_FILTER)
3503 
+                RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3511 3504 
 */
3512 
-                        dstBlock+=8;
3513 
-                        srcBlock+=8;
3514 
-                }
3515 
-                if(width==FFABS(dstStride))
3516 
-                        linecpy(dst, tempDst + 9*dstStride, copyAhead, dstStride);
3517 
-                else
3518 
-                {
3519 
-                        int i;
3520 
-                        for(i=0; i<copyAhead; i++)
3521 
-                        {
3522 
-                                memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);
3523 
-                        }
3524 
-                }
3505 
+            dstBlock+=8;
3506 
+            srcBlock+=8;
3507 
+        }
3508 
+        if(width==FFABS(dstStride))
3509 
+            linecpy(dst, tempDst + 9*dstStride, copyAhead, dstStride);
3510 
+        else{
3511 
+            int i;
3512 
+            for(i=0; i<copyAhead; i++){
3513 
+                memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);
3514 
+            }
3525 3515 
         }
3516 
+    }
3526 3517
3527 
-        for(y=0; y<height; y+=BLOCK_SIZE)
3528 
-        {
3529 
-                //1% speedup if these are here instead of the inner loop
3530 
-                const uint8_t *srcBlock= &(src[y*srcStride]);
3531 
-                uint8_t *dstBlock= &(dst[y*dstStride]);
3518 
+    for(y=0; y<height; y+=BLOCK_SIZE){
3519 
+        //1% speedup if these are here instead of the inner loop
3520 
+        const uint8_t *srcBlock= &(src[y*srcStride]);
3521 
+        uint8_t *dstBlock= &(dst[y*dstStride]);
3532 3522 
 #ifdef HAVE_MMX
3533 
-                uint8_t *tempBlock1= c.tempBlocks;
3534 
-                uint8_t *tempBlock2= c.tempBlocks + 8;
3523 
+        uint8_t *tempBlock1= c.tempBlocks;
3524 
+        uint8_t *tempBlock2= c.tempBlocks + 8;
3535 3525 
 #endif
3536 
-                const int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];
3537 
-                int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*FFABS(QPStride)];
3538 
-                int QP=0;
3539 
-                /* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards
3540 
-                   if not than use a temporary buffer */
3541 
-                if(y+15 >= height)
3542 
-                {
3543 
-                        int i;
3544 
-                        /* copy from line (copyAhead) to (copyAhead+7) of src, these will be copied with
3545 
-                           blockcopy to dst later */
3546 
-                        linecpy(tempSrc + srcStride*copyAhead, srcBlock + srcStride*copyAhead,
3547 
-                                FFMAX(height-y-copyAhead, 0), srcStride);
3548 
-
3549 
-                        /* duplicate last line of src to fill the void upto line (copyAhead+7) */
3550 
-                        for(i=FFMAX(height-y, 8); i<copyAhead+8; i++)
3551 
-                                memcpy(tempSrc + srcStride*i, src + srcStride*(height-1), FFABS(srcStride));
3552 
-
3553 
-                        /* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/
3554 
-                        linecpy(tempDst, dstBlock - dstStride, FFMIN(height-y+1, copyAhead+1), dstStride);
3555 
-
3556 
-                        /* duplicate last line of dst to fill the void upto line (copyAhead) */
3557 
-                        for(i=height-y+1; i<=copyAhead; i++)
3558 
-                                memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), FFABS(dstStride));
3559 
-
3560 
-                        dstBlock= tempDst + dstStride;
3561 
-                        srcBlock= tempSrc;
3562 
-                }
3526 
+        const int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];
3527 
+        int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*FFABS(QPStride)];
3528 
+        int QP=0;
3529 
+        /* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards
3530 
+           if not than use a temporary buffer */
3531 
+        if(y+15 >= height){
3532 
+            int i;
3533 
+            /* copy from line (copyAhead) to (copyAhead+7) of src, these will be copied with
3534 
+               blockcopy to dst later */
3535 
+            linecpy(tempSrc + srcStride*copyAhead, srcBlock + srcStride*copyAhead,
3536 
+                    FFMAX(height-y-copyAhead, 0), srcStride);
3537 
+
3538 
+            /* duplicate last line of src to fill the void upto line (copyAhead+7) */
3539 
+            for(i=FFMAX(height-y, 8); i<copyAhead+8; i++)
3540 
+                    memcpy(tempSrc + srcStride*i, src + srcStride*(height-1), FFABS(srcStride));
3541 
+
3542 
+            /* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/
3543 
+            linecpy(tempDst, dstBlock - dstStride, FFMIN(height-y+1, copyAhead+1), dstStride);
3544 
+
3545 
+            /* duplicate last line of dst to fill the void upto line (copyAhead) */
3546 
+            for(i=height-y+1; i<=copyAhead; i++)
3547 
+                    memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), FFABS(dstStride));
3548 
+
3549 
+            dstBlock= tempDst + dstStride;
3550 
+            srcBlock= tempSrc;
3551 
+        }
3563 3552
3564 
-                // From this point on it is guaranteed that we can read and write 16 lines downward
3565 
-                // finish 1 block before the next otherwise we might have a problem
3566 
-                // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
3567 
-                for(x=0; x<width; x+=BLOCK_SIZE)
3568 
-                {
3569 
-                        const int stride= dstStride;
3553 
+        // From this point on it is guaranteed that we can read and write 16 lines downward
3554 
+        // finish 1 block before the next otherwise we might have a problem
3555 
+        // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
3556 
+        for(x=0; x<width; x+=BLOCK_SIZE){
3557 
+            const int stride= dstStride;
3570 3558 
 #ifdef HAVE_MMX
3571 
-                        uint8_t *tmpXchg;
3559 
+            uint8_t *tmpXchg;
3572 3560 
 #endif
3573 
-                        if(isColor)
3574 
-                        {
3575 
-                                QP= QPptr[x>>qpHShift];
3576 
-                                c.nonBQP= nonBQPptr[x>>qpHShift];
3577 
-                        }
3578 
-                        else
3579 
-                        {
3580 
-                                QP= QPptr[x>>4];
3581 
-                                QP= (QP* QPCorrecture + 256*128)>>16;
3582 
-                                c.nonBQP= nonBQPptr[x>>4];
3583 
-                                c.nonBQP= (c.nonBQP* QPCorrecture + 256*128)>>16;
3584 
-                                yHistogram[ srcBlock[srcStride*12 + 4] ]++;
3585 
-                        }
3586 
-                        c.QP= QP;
3561 
+            if(isColor){
3562 
+                QP= QPptr[x>>qpHShift];
3563 
+                c.nonBQP= nonBQPptr[x>>qpHShift];
3564 
+            }else{
3565 
+                QP= QPptr[x>>4];
3566 
+                QP= (QP* QPCorrecture + 256*128)>>16;
3567 
+                c.nonBQP= nonBQPptr[x>>4];
3568 
+                c.nonBQP= (c.nonBQP* QPCorrecture + 256*128)>>16;
3569 
+                yHistogram[ srcBlock[srcStride*12 + 4] ]++;
3570 
+            }
3571 
+            c.QP= QP;
3587 3572 
 #ifdef HAVE_MMX
3588 
-                        asm volatile(
3589 
-                                "movd %1, %%mm7         \n\t"
3590 
-                                "packuswb %%mm7, %%mm7  \n\t" // 0, 0, 0, QP, 0, 0, 0, QP
3591 
-                                "packuswb %%mm7, %%mm7  \n\t" // 0,QP, 0, QP, 0,QP, 0, QP
3592 
-                                "packuswb %%mm7, %%mm7  \n\t" // QP,..., QP
3593 
-                                "movq %%mm7, %0         \n\t"
3594 
-                                : "=m" (c.pQPb)
3595 
-                                : "r" (QP)
3596 
-                        );
3573 
+            asm volatile(
3574 
+                "movd %1, %%mm7         \n\t"
3575 
+                "packuswb %%mm7, %%mm7  \n\t" // 0, 0, 0, QP, 0, 0, 0, QP
3576 
+                "packuswb %%mm7, %%mm7  \n\t" // 0,QP, 0, QP, 0,QP, 0, QP
3577 
+                "packuswb %%mm7, %%mm7  \n\t" // QP,..., QP
3578 
+                "movq %%mm7, %0         \n\t"
3579 
+                : "=m" (c.pQPb)
3580 
+                : "r" (QP)
3581 
+            );
3597 3582 
 #endif
3598 3583
3599 3584
3600 3585 
 #ifdef HAVE_MMX2
3601 3586 
 /*
3602 
-                        prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
3603 
-                        prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
3604 
-                        prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
3605 
-                        prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
3587 
+            prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
3588 
+            prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
3589 
+            prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
3590 
+            prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
3606 3591 
 */
3607 3592
3608 
-                        asm(
3609 
-                                "mov %4, %%"REG_a"              \n\t"
3610 
-                                "shr $2, %%"REG_a"              \n\t"
3611 
-                                "and $6, %%"REG_a"              \n\t"
3612 
-                                "add %5, %%"REG_a"              \n\t"
3613 
-                                "mov %%"REG_a", %%"REG_d"       \n\t"
3614 
-                                "imul %1, %%"REG_a"             \n\t"
3615 
-                                "imul %3, %%"REG_d"             \n\t"
3616 
-                                "prefetchnta 32(%%"REG_a", %0)  \n\t"
3617 
-                                "prefetcht0 32(%%"REG_d", %2)   \n\t"
3618 
-                                "add %1, %%"REG_a"              \n\t"
3619 
-                                "add %3, %%"REG_d"              \n\t"
3620 
-                                "prefetchnta 32(%%"REG_a", %0)  \n\t"
3621 
-                                "prefetcht0 32(%%"REG_d", %2)   \n\t"
3622 
-                        :: "r" (srcBlock), "r" ((long)srcStride), "r" (dstBlock), "r" ((long)dstStride),
3623 
-                         "g" ((long)x), "g" ((long)copyAhead)
3624 
-                        : "%"REG_a, "%"REG_d
3625 
-                        );
3593 
+            asm(
3594 
+                "mov %4, %%"REG_a"              \n\t"
3595 
+                "shr $2, %%"REG_a"              \n\t"
3596 
+                "and $6, %%"REG_a"              \n\t"
3597 
+                "add %5, %%"REG_a"              \n\t"
3598 
+                "mov %%"REG_a", %%"REG_d"       \n\t"
3599 
+                "imul %1, %%"REG_a"             \n\t"
3600 
+                "imul %3, %%"REG_d"             \n\t"
3601 
+                "prefetchnta 32(%%"REG_a", %0)  \n\t"
3602 
+                "prefetcht0 32(%%"REG_d", %2)   \n\t"
3603 
+                "add %1, %%"REG_a"              \n\t"
3604 
+                "add %3, %%"REG_d"              \n\t"
3605 
+                "prefetchnta 32(%%"REG_a", %0)  \n\t"
3606 
+                "prefetcht0 32(%%"REG_d", %2)   \n\t"
3607 
+                :: "r" (srcBlock), "r" ((long)srcStride), "r" (dstBlock), "r" ((long)dstStride),
3608 
+                "g" ((long)x), "g" ((long)copyAhead)
3609 
+                : "%"REG_a, "%"REG_d
3610 
+            );
3626 3611
3627 3612 
 #elif defined(HAVE_3DNOW)
3628 3613 
 //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
3629 
-/*                        prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
3630 
-                        prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
3631 
-                        prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
3632 
-                        prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
3614 
+/*          prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
3615 
+            prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
3616 
+            prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
3617 
+            prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
3633 3618 
 */
3634 3619 
 #endif
3635 3620
3636 
-                        RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,
3637 
-                                srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
3638 
-
3639 
-                        if(mode & LINEAR_IPOL_DEINT_FILTER)
3640 
-                                RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
3641 
-                        else if(mode & LINEAR_BLEND_DEINT_FILTER)
3642 
-                                RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
3643 
-                        else if(mode & MEDIAN_DEINT_FILTER)
3644 
-                                RENAME(deInterlaceMedian)(dstBlock, dstStride);
3645 
-                        else if(mode & CUBIC_IPOL_DEINT_FILTER)
3646 
-                                RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
3647 
-                        else if(mode & FFMPEG_DEINT_FILTER)
3648 
-                                RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
3649 
-                        else if(mode & LOWPASS5_DEINT_FILTER)
3650 
-                                RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
3651 
-/*                        else if(mode & CUBIC_BLEND_DEINT_FILTER)
3652 
-                                RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3621 
+            RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,
3622 
+                              srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
3623 
+
3624 
+            if(mode & LINEAR_IPOL_DEINT_FILTER)
3625 
+                RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
3626 
+            else if(mode & LINEAR_BLEND_DEINT_FILTER)
3627 
+                RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
3628 
+            else if(mode & MEDIAN_DEINT_FILTER)
3629 
+                RENAME(deInterlaceMedian)(dstBlock, dstStride);
3630 
+            else if(mode & CUBIC_IPOL_DEINT_FILTER)
3631 
+                RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
3632 
+            else if(mode & FFMPEG_DEINT_FILTER)
3633 
+                RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
3634 
+            else if(mode & LOWPASS5_DEINT_FILTER)
3635 
+                RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
3636 
+/*          else if(mode & CUBIC_BLEND_DEINT_FILTER)
3637 
+                RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3653 3638 
 */
3654 3639
3655 
-                        /* only deblock if we have 2 blocks */
3656 
-                        if(y + 8 < height)
3657 
-                        {
3658 
-                                if(mode & V_X1_FILTER)
3659 
-                                        RENAME(vertX1Filter)(dstBlock, stride, &c);
3660 
-                                else if(mode & V_DEBLOCK)
3661 
-                                {
3662 
-                                        const int t= RENAME(vertClassify)(dstBlock, stride, &c);
3663 
-
3664 
-                                        if(t==1)
3665 
-                                                RENAME(doVertLowPass)(dstBlock, stride, &c);
3666 
-                                        else if(t==2)
3667 
-                                                RENAME(doVertDefFilter)(dstBlock, stride, &c);
3668 
-                                }else if(mode & V_A_DEBLOCK){
3669 
-                                        RENAME(do_a_deblock)(dstBlock, stride, 1, &c);
3670 
-                                }
3671 
-                        }
3640 
+            /* only deblock if we have 2 blocks */
3641 
+            if(y + 8 < height){
3642 
+                if(mode & V_X1_FILTER)
3643 
+                    RENAME(vertX1Filter)(dstBlock, stride, &c);
3644 
+                else if(mode & V_DEBLOCK){
3645 
+                    const int t= RENAME(vertClassify)(dstBlock, stride, &c);
3646 
+
3647 
+                    if(t==1)
3648 
+                        RENAME(doVertLowPass)(dstBlock, stride, &c);
3649 
+                    else if(t==2)
3650 
+                        RENAME(doVertDefFilter)(dstBlock, stride, &c);
3651 
+                }else if(mode & V_A_DEBLOCK){
3652 
+                    RENAME(do_a_deblock)(dstBlock, stride, 1, &c);
3653 
+                }
3654 
+            }
3672 3655
3673 3656 
 #ifdef HAVE_MMX
3674 
-                        RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);
3657 
+            RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);
3675 3658 
 #endif
3676 
-                        /* check if we have a previous block to deblock it with dstBlock */
3677 
-                        if(x - 8 >= 0)
3678 
-                        {
3659 
+            /* check if we have a previous block to deblock it with dstBlock */
3660 
+            if(x - 8 >= 0){
3679 3661 
 #ifdef HAVE_MMX
3680 
-                                if(mode & H_X1_FILTER)
3681 
-                                        RENAME(vertX1Filter)(tempBlock1, 16, &c);
3682 
-                                else if(mode & H_DEBLOCK)
3683 
-                                {
3662 
+                if(mode & H_X1_FILTER)
3663 
+                        RENAME(vertX1Filter)(tempBlock1, 16, &c);
3664 
+                else if(mode & H_DEBLOCK){
3684 3665 
 //START_TIMER
3685 
-                                        const int t= RENAME(vertClassify)(tempBlock1, 16, &c);
3666 
+                    const int t= RENAME(vertClassify)(tempBlock1, 16, &c);
3686 3667 
 //STOP_TIMER("dc & minmax")
3687 
-                                        if(t==1)
3688 
-                                                RENAME(doVertLowPass)(tempBlock1, 16, &c);
3689 
-                                        else if(t==2)
3690 
-                                                RENAME(doVertDefFilter)(tempBlock1, 16, &c);
3691 
-                                }else if(mode & H_A_DEBLOCK){
3692 
-                                        RENAME(do_a_deblock)(tempBlock1, 16, 1, &c);
3693 
-                                }
3668 
+                    if(t==1)
3669 
+                        RENAME(doVertLowPass)(tempBlock1, 16, &c);
3670 
+                    else if(t==2)
3671 
+                        RENAME(doVertDefFilter)(tempBlock1, 16, &c);
3672 
+                }else if(mode & H_A_DEBLOCK){
3673 
+                        RENAME(do_a_deblock)(tempBlock1, 16, 1, &c);
3674 
+                }
3694 3675
3695 
-                                RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16);
3676 
+                RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16);
3696 3677
3697 3678 
 #else
3698 
-                                if(mode & H_X1_FILTER)
3699 
-                                        horizX1Filter(dstBlock-4, stride, QP);
3700 
-                                else if(mode & H_DEBLOCK)
3701 
-                                {
3679 
+                if(mode & H_X1_FILTER)
3680 
+                    horizX1Filter(dstBlock-4, stride, QP);
3681 
+                else if(mode & H_DEBLOCK){
3702 3682 
 #ifdef HAVE_ALTIVEC
3703 
-                                        DECLARE_ALIGNED(16, unsigned char, tempBlock[272]);
3704 
-                                        transpose_16x8_char_toPackedAlign_altivec(tempBlock, dstBlock - (4 + 1), stride);
3705 
-
3706 
-                                        const int t=vertClassify_altivec(tempBlock-48, 16, &c);
3707 
-                                        if(t==1) {
3708 
-                                                doVertLowPass_altivec(tempBlock-48, 16, &c);
3709 
-                                                transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
3710 
-                                        }
3711 
-                                        else if(t==2) {
3712 
-                                                doVertDefFilter_altivec(tempBlock-48, 16, &c);
3713 
-                                                transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
3714 
-                                        }
3683 
+                    DECLARE_ALIGNED(16, unsigned char, tempBlock[272]);
3684 
+                    transpose_16x8_char_toPackedAlign_altivec(tempBlock, dstBlock - (4 + 1), stride);
3685 
+
3686 
+                    const int t=vertClassify_altivec(tempBlock-48, 16, &c);
3687 
+                    if(t==1) {
3688 
+                        doVertLowPass_altivec(tempBlock-48, 16, &c);
3689 
+                        transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
3690 
+                    }
3691 
+                    else if(t==2) {
3692 
+                        doVertDefFilter_altivec(tempBlock-48, 16, &c);
3693 
+                        transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
3694 
+                    }
3715 3695 
 #else
3716 
-                                        const int t= RENAME(horizClassify)(dstBlock-4, stride, &c);
3696 
+                    const int t= RENAME(horizClassify)(dstBlock-4, stride, &c);
3717 3697
3718 
-                                        if(t==1)
3719 
-                                                RENAME(doHorizLowPass)(dstBlock-4, stride, &c);
3720 
-                                        else if(t==2)
3721 
-                                                RENAME(doHorizDefFilter)(dstBlock-4, stride, &c);
3698 
+                    if(t==1)
3699 
+                        RENAME(doHorizLowPass)(dstBlock-4, stride, &c);
3700 
+                    else if(t==2)
3701 
+                        RENAME(doHorizDefFilter)(dstBlock-4, stride, &c);
3722 3702 
 #endif
3723 
-                                }else if(mode & H_A_DEBLOCK){
3724 
-                                        RENAME(do_a_deblock)(dstBlock-8, 1, stride, &c);
3725 
-                                }
3703 
+                }else if(mode & H_A_DEBLOCK){
3704 
+                    RENAME(do_a_deblock)(dstBlock-8, 1, stride, &c);
3705 
+                }
3726 3706 
 #endif //HAVE_MMX
3727 
-                                if(mode & DERING)
3728 
-                                {
3729 
-                                //FIXME filter first line
3730 
-                                        if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);
3731 
-                                }
3732 
-
3733 
-                                if(mode & TEMP_NOISE_FILTER)
3734 
-                                {
3735 
-                                        RENAME(tempNoiseReducer)(dstBlock-8, stride,
3736 
-                                                c.tempBlured[isColor] + y*dstStride + x,
3737 
-                                                c.tempBluredPast[isColor] + (y>>3)*256 + (x>>3),
3738 
-                                                c.ppMode.maxTmpNoise);
3739 
-                                }
3740 
-                        }
3707 
+                if(mode & DERING){
3708 
+                //FIXME filter first line
3709 
+                    if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);
3710 
+                }
3741 3711
3742 
-                        dstBlock+=8;
3743 
-                        srcBlock+=8;
3712 
+                if(mode & TEMP_NOISE_FILTER)
3713 
+                {
3714 
+                    RENAME(tempNoiseReducer)(dstBlock-8, stride,
3715 
+                            c.tempBlured[isColor] + y*dstStride + x,
3716 
+                            c.tempBluredPast[isColor] + (y>>3)*256 + (x>>3),
3717 
+                            c.ppMode.maxTmpNoise);
3718 
+                }
3719 
+            }
3720 
+
3721 
+            dstBlock+=8;
3722 
+            srcBlock+=8;
3744 3723
3745 3724 
 #ifdef HAVE_MMX
3746 
-                        tmpXchg= tempBlock1;
3747 
-                        tempBlock1= tempBlock2;
3748 
-                        tempBlock2 = tmpXchg;
3725 
+            tmpXchg= tempBlock1;
3726 
+            tempBlock1= tempBlock2;
3727 
+            tempBlock2 = tmpXchg;
3749 3728 
 #endif
3750 
-                }
3729 
+        }
3751 3730
3752 
-                if(mode & DERING)
3753 
-                {
3754 
-                                if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, &c);
3755 
-                }
3731 
+        if(mode & DERING){
3732 
+            if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, &c);
3733 
+        }
3756 3734
3757 
-                if((mode & TEMP_NOISE_FILTER))
3758 
-                {
3759 
-                        RENAME(tempNoiseReducer)(dstBlock-8, dstStride,
3760 
-                                c.tempBlured[isColor] + y*dstStride + x,
3761 
-                                c.tempBluredPast[isColor] + (y>>3)*256 + (x>>3),
3762 
-                                c.ppMode.maxTmpNoise);
3763 
-                }
3735 
+        if((mode & TEMP_NOISE_FILTER)){
3736 
+            RENAME(tempNoiseReducer)(dstBlock-8, dstStride,
3737 
+                    c.tempBlured[isColor] + y*dstStride + x,
3738 
+                    c.tempBluredPast[isColor] + (y>>3)*256 + (x>>3),
3739 
+                    c.ppMode.maxTmpNoise);
3740 
+        }
3764 3741
3765 
-                /* did we use a tmp buffer for the last lines*/
3766 
-                if(y+15 >= height)
3767 
-                {
3768 
-                        uint8_t *dstBlock= &(dst[y*dstStride]);
3769 
-                        if(width==FFABS(dstStride))
3770 
-                                linecpy(dstBlock, tempDst + dstStride, height-y, dstStride);
3771 
-                        else
3772 
-                        {
3773 
-                                int i;
3774 
-                                for(i=0; i<height-y; i++)
3775 
-                                {
3776 
-                                        memcpy(dstBlock + i*dstStride, tempDst + (i+1)*dstStride, width);
3777 
-                                }
3778 
-                        }
3742 
+        /* did we use a tmp buffer for the last lines*/
3743 
+        if(y+15 >= height){
3744 
+            uint8_t *dstBlock= &(dst[y*dstStride]);
3745 
+            if(width==FFABS(dstStride))
3746 
+                linecpy(dstBlock, tempDst + dstStride, height-y, dstStride);
3747 
+            else{
3748 
+                int i;
3749 
+                for(i=0; i<height-y; i++){
3750 
+                    memcpy(dstBlock + i*dstStride, tempDst + (i+1)*dstStride, width);
3779 3751 
                 }
3780 
-/*
3781 
-                for(x=0; x<width; x+=32)
3782 
-                {
3783 
-                        volatile int i;
3784 
-                        i+=        + dstBlock[x + 7*dstStride] + dstBlock[x + 8*dstStride]
3785 
-                                + dstBlock[x + 9*dstStride] + dstBlock[x +10*dstStride]
3786 
-                                + dstBlock[x +11*dstStride] + dstBlock[x +12*dstStride];
3787 
-//                                + dstBlock[x +13*dstStride]
3788 
-//                                + dstBlock[x +14*dstStride] + dstBlock[x +15*dstStride];
3789 
-                }*/
3752 
+            }
3790 3753 
         }
3754 
+/*
3755 
+        for(x=0; x<width; x+=32){
3756 
+            volatile int i;
3757 
+            i+= + dstBlock[x + 7*dstStride] + dstBlock[x + 8*dstStride]
3758 
+                + dstBlock[x + 9*dstStride] + dstBlock[x +10*dstStride]
3759 
+                + dstBlock[x +11*dstStride] + dstBlock[x +12*dstStride];
3760 
+                + dstBlock[x +13*dstStride]
3761 
+                + dstBlock[x +14*dstStride] + dstBlock[x +15*dstStride];
3762 
+        }*/
3763 
+    }
3791 3764 
 #ifdef HAVE_3DNOW
3792 
-        asm volatile("femms");
3765 
+    asm volatile("femms");
3793 3766 
 #elif defined (HAVE_MMX)
3794 
-        asm volatile("emms");
3767 
+    asm volatile("emms");
3795 3768 
 #endif
3796 3769
3797 3770 
 #ifdef DEBUG_BRIGHTNESS
3798 
-        if(!isColor)
3799 
-        {
3800 
-                int max=1;
3801 
-                int i;
3802 
-                for(i=0; i<256; i++)
3803 
-                        if(yHistogram[i] > max) max=yHistogram[i];
3804 
-
3805 
-                for(i=1; i<256; i++)
3806 
-                {
3807 
-                        int x;
3808 
-                        int start=yHistogram[i-1]/(max/256+1);
3809 
-                        int end=yHistogram[i]/(max/256+1);
3810 
-                        int inc= end > start ? 1 : -1;
3811 
-                        for(x=start; x!=end+inc; x+=inc)
3812 
-                                dst[ i*dstStride + x]+=128;
3813 
-                }
3814 
-
3815 
-                for(i=0; i<100; i+=2)
3816 
-                {
3817 
-                        dst[ (white)*dstStride + i]+=128;
3818 
-                        dst[ (black)*dstStride + i]+=128;
3819 
-                }
3771 
+    if(!isColor){
3772 
+        int max=1;
3773 
+        int i;
3774 
+        for(i=0; i<256; i++)
3775 
+            if(yHistogram[i] > max) max=yHistogram[i];
3776 
+
3777 
+        for(i=1; i<256; i++){
3778 
+            int x;
3779 
+            int start=yHistogram[i-1]/(max/256+1);
3780 
+            int end=yHistogram[i]/(max/256+1);
3781 
+            int inc= end > start ? 1 : -1;
3782 
+            for(x=start; x!=end+inc; x+=inc)
3783 
+                dst[ i*dstStride + x]+=128;
3784 
+        }
3820 3785
3786 
+        for(i=0; i<100; i+=2){
3787 
+            dst[ (white)*dstStride + i]+=128;
3788 
+            dst[ (black)*dstStride + i]+=128;
3821 3789 
         }
3790 
+    }
3822 3791 
 #endif
3823 3792
3824 
-        *c2= c; //copy local context back
3793 
+    *c2= c; //copy local context back
3825 3794
3826 3795 
 }