2 Copyright (C) 2001-2002 Michael Niedermayer (michaelni@gmx.at)
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
20 * @file postprocess_template.c
21 * mmx/mmx2/3dnow postprocess code.
30 #define PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
31 #elif defined (HAVE_3DNOW)
32 #define PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
36 #define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
37 #elif defined (HAVE_MMX)
38 #define PMINUB(b,a,t) \
39 "movq " #a ", " #t " \n\t"\
40 "psubusb " #b ", " #t " \n\t"\
41 "psubb " #t ", " #a " \n\t"
45 #define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
46 #elif defined (HAVE_MMX)
48 "psubusb " #a ", " #b " \n\t"\
49 "paddb " #a ", " #b " \n\t"
52 //FIXME? |255-0| = 1 (shouldnt be a problem ...)
55 * Check if the middle 8x8 Block in the given 8x16 block is flat
57 static inline int RENAME(vertClassify)(uint8_t src[], int stride, PPContext *c){
59 src+= stride*4; // src points to begin of the 8x8 Block
63 : : "m" (c->mmxDcOffset[c->nonBQP]), "m" (c->mmxDcThreshold[c->nonBQP])
67 "leal (%2, %3), %%eax \n\t"
68 // 0 1 2 3 4 5 6 7 8 9
69 // %1 eax eax+%2 eax+2%2 %1+4%2 ecx ecx+%2 ecx+2%2 %1+8%2 ecx+4%2
71 "movq (%2), %%mm0 \n\t"
72 "movq (%%eax), %%mm1 \n\t"
73 "movq %%mm0, %%mm3 \n\t"
74 "movq %%mm0, %%mm4 \n\t"
76 PMINUB(%%mm1, %%mm3, %%mm5)
77 "psubb %%mm1, %%mm0 \n\t" // mm0 = differnece
78 "paddb %%mm7, %%mm0 \n\t"
79 "pcmpgtb %%mm6, %%mm0 \n\t"
81 "movq (%%eax,%3), %%mm2 \n\t"
83 PMINUB(%%mm2, %%mm3, %%mm5)
84 "psubb %%mm2, %%mm1 \n\t"
85 "paddb %%mm7, %%mm1 \n\t"
86 "pcmpgtb %%mm6, %%mm1 \n\t"
87 "paddb %%mm1, %%mm0 \n\t"
89 "movq (%%eax, %3, 2), %%mm1 \n\t"
91 PMINUB(%%mm1, %%mm3, %%mm5)
92 "psubb %%mm1, %%mm2 \n\t"
93 "paddb %%mm7, %%mm2 \n\t"
94 "pcmpgtb %%mm6, %%mm2 \n\t"
95 "paddb %%mm2, %%mm0 \n\t"
97 "leal (%%eax, %3, 4), %%eax \n\t"
99 "movq (%2, %3, 4), %%mm2 \n\t"
101 PMINUB(%%mm2, %%mm3, %%mm5)
102 "psubb %%mm2, %%mm1 \n\t"
103 "paddb %%mm7, %%mm1 \n\t"
104 "pcmpgtb %%mm6, %%mm1 \n\t"
105 "paddb %%mm1, %%mm0 \n\t"
107 "movq (%%eax), %%mm1 \n\t"
109 PMINUB(%%mm1, %%mm3, %%mm5)
110 "psubb %%mm1, %%mm2 \n\t"
111 "paddb %%mm7, %%mm2 \n\t"
112 "pcmpgtb %%mm6, %%mm2 \n\t"
113 "paddb %%mm2, %%mm0 \n\t"
115 "movq (%%eax, %3), %%mm2 \n\t"
117 PMINUB(%%mm2, %%mm3, %%mm5)
118 "psubb %%mm2, %%mm1 \n\t"
119 "paddb %%mm7, %%mm1 \n\t"
120 "pcmpgtb %%mm6, %%mm1 \n\t"
121 "paddb %%mm1, %%mm0 \n\t"
123 "movq (%%eax, %3, 2), %%mm1 \n\t"
125 PMINUB(%%mm1, %%mm3, %%mm5)
126 "psubb %%mm1, %%mm2 \n\t"
127 "paddb %%mm7, %%mm2 \n\t"
128 "pcmpgtb %%mm6, %%mm2 \n\t"
129 "paddb %%mm2, %%mm0 \n\t"
130 "psubusb %%mm3, %%mm4 \n\t"
134 "pxor %%mm7, %%mm7 \n\t"
135 "psadbw %%mm7, %%mm0 \n\t"
137 "movq %%mm0, %%mm1 \n\t"
138 "psrlw $8, %%mm0 \n\t"
139 "paddb %%mm1, %%mm0 \n\t"
140 "movq %%mm0, %%mm1 \n\t"
141 "psrlq $16, %%mm0 \n\t"
142 "paddb %%mm1, %%mm0 \n\t"
143 "movq %%mm0, %%mm1 \n\t"
144 "psrlq $32, %%mm0 \n\t"
145 "paddb %%mm1, %%mm0 \n\t"
147 "movq %4, %%mm7 \n\t" // QP,..., QP
148 "paddusb %%mm7, %%mm7 \n\t" // 2QP ... 2QP
149 "psubusb %%mm7, %%mm4 \n\t" // Diff <= 2QP -> 0
150 "packssdw %%mm4, %%mm4 \n\t"
151 "movd %%mm0, %0 \n\t"
152 "movd %%mm4, %1 \n\t"
154 : "=r" (numEq), "=r" (dcOk)
155 : "r" (src), "r" (stride), "m" (c->pQPb)
159 numEq= (-numEq) &0xFF;
160 if(numEq > c->ppMode.flatnessThreshold){
170 * Do a vertical low pass filter on the 8x16 block (only write to the 8x8 block in the middle)
171 * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16
173 static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
175 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
177 asm volatile( //"movv %0 %1 %2\n\t"
178 "movq %2, %%mm0 \n\t" // QP,..., QP
179 "pxor %%mm4, %%mm4 \n\t"
181 "movq (%0), %%mm6 \n\t"
182 "movq (%0, %1), %%mm5 \n\t"
183 "movq %%mm5, %%mm1 \n\t"
184 "movq %%mm6, %%mm2 \n\t"
185 "psubusb %%mm6, %%mm5 \n\t"
186 "psubusb %%mm1, %%mm2 \n\t"
187 "por %%mm5, %%mm2 \n\t" // ABS Diff of lines
188 "psubusb %%mm0, %%mm2 \n\t" // diff <= QP -> 0
189 "pcmpeqb %%mm4, %%mm2 \n\t" // diff <= QP -> FF
191 "pand %%mm2, %%mm6 \n\t"
192 "pandn %%mm1, %%mm2 \n\t"
193 "por %%mm2, %%mm6 \n\t"// First Line to Filter
195 "movq (%0, %1, 8), %%mm5 \n\t"
196 "leal (%0, %1, 4), %%eax \n\t"
197 "leal (%0, %1, 8), %%ecx \n\t"
198 "subl %1, %%ecx \n\t"
199 "addl %1, %0 \n\t" // %0 points to line 1 not 0
200 "movq (%0, %1, 8), %%mm7 \n\t"
201 "movq %%mm5, %%mm1 \n\t"
202 "movq %%mm7, %%mm2 \n\t"
203 "psubusb %%mm7, %%mm5 \n\t"
204 "psubusb %%mm1, %%mm2 \n\t"
205 "por %%mm5, %%mm2 \n\t" // ABS Diff of lines
206 "psubusb %%mm0, %%mm2 \n\t" // diff <= QP -> 0
207 "pcmpeqb %%mm4, %%mm2 \n\t" // diff <= QP -> FF
209 "pand %%mm2, %%mm7 \n\t"
210 "pandn %%mm1, %%mm2 \n\t"
211 "por %%mm2, %%mm7 \n\t" // First Line to Filter
215 // %0 %0+%1 %0+2%1 eax %0+4%1 eax+2%1 ecx eax+4%1
220 "movq (%0, %1), %%mm0 \n\t" // 1
221 "movq %%mm0, %%mm1 \n\t" // 1
222 PAVGB(%%mm6, %%mm0) //1 1 /2
223 PAVGB(%%mm6, %%mm0) //3 1 /4
225 "movq (%0, %1, 4), %%mm2 \n\t" // 1
226 "movq %%mm2, %%mm5 \n\t" // 1
227 PAVGB((%%eax), %%mm2) // 11 /2
228 PAVGB((%0, %1, 2), %%mm2) // 211 /4
229 "movq %%mm2, %%mm3 \n\t" // 211 /4
230 "movq (%0), %%mm4 \n\t" // 1
231 PAVGB(%%mm4, %%mm3) // 4 211 /8
232 PAVGB(%%mm0, %%mm3) //642211 /16
233 "movq %%mm3, (%0) \n\t" // X
234 // mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
235 "movq %%mm1, %%mm0 \n\t" // 1
236 PAVGB(%%mm6, %%mm0) //1 1 /2
237 "movq %%mm4, %%mm3 \n\t" // 1
238 PAVGB((%0,%1,2), %%mm3) // 1 1 /2
239 PAVGB((%%eax,%1,2), %%mm5) // 11 /2
240 PAVGB((%%eax), %%mm5) // 211 /4
241 PAVGB(%%mm5, %%mm3) // 2 2211 /8
242 PAVGB(%%mm0, %%mm3) //4242211 /16
243 "movq %%mm3, (%0,%1) \n\t" // X
244 // mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
245 PAVGB(%%mm4, %%mm6) //11 /2
246 "movq (%%ecx), %%mm0 \n\t" // 1
247 PAVGB((%%eax, %1, 2), %%mm0) // 11/2
248 "movq %%mm0, %%mm3 \n\t" // 11/2
249 PAVGB(%%mm1, %%mm0) // 2 11/4
250 PAVGB(%%mm6, %%mm0) //222 11/8
251 PAVGB(%%mm2, %%mm0) //22242211/16
252 "movq (%0, %1, 2), %%mm2 \n\t" // 1
253 "movq %%mm0, (%0, %1, 2) \n\t" // X
254 // mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
255 "movq (%%eax, %1, 4), %%mm0 \n\t" // 1
256 PAVGB((%%ecx), %%mm0) // 11 /2
257 PAVGB(%%mm0, %%mm6) //11 11 /4
258 PAVGB(%%mm1, %%mm4) // 11 /2
259 PAVGB(%%mm2, %%mm1) // 11 /2
260 PAVGB(%%mm1, %%mm6) //1122 11 /8
261 PAVGB(%%mm5, %%mm6) //112242211 /16
262 "movq (%%eax), %%mm5 \n\t" // 1
263 "movq %%mm6, (%%eax) \n\t" // X
264 // mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
265 "movq (%%eax, %1, 4), %%mm6 \n\t" // 1
266 PAVGB(%%mm7, %%mm6) // 11 /2
267 PAVGB(%%mm4, %%mm6) // 11 11 /4
268 PAVGB(%%mm3, %%mm6) // 11 2211 /8
269 PAVGB(%%mm5, %%mm2) // 11 /2
270 "movq (%0, %1, 4), %%mm4 \n\t" // 1
271 PAVGB(%%mm4, %%mm2) // 112 /4
272 PAVGB(%%mm2, %%mm6) // 112242211 /16
273 "movq %%mm6, (%0, %1, 4) \n\t" // X
274 // mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
275 PAVGB(%%mm7, %%mm1) // 11 2 /4
276 PAVGB(%%mm4, %%mm5) // 11 /2
277 PAVGB(%%mm5, %%mm0) // 11 11 /4
278 "movq (%%eax, %1, 2), %%mm6 \n\t" // 1
279 PAVGB(%%mm6, %%mm1) // 11 4 2 /8
280 PAVGB(%%mm0, %%mm1) // 11224222 /16
281 "movq %%mm1, (%%eax, %1, 2) \n\t" // X
282 // mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
283 PAVGB((%%ecx), %%mm2) // 112 4 /8
284 "movq (%%eax, %1, 4), %%mm0 \n\t" // 1
285 PAVGB(%%mm0, %%mm6) // 1 1 /2
286 PAVGB(%%mm7, %%mm6) // 1 12 /4
287 PAVGB(%%mm2, %%mm6) // 1122424 /4
288 "movq %%mm6, (%%ecx) \n\t" // X
289 // mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
290 PAVGB(%%mm7, %%mm5) // 11 2 /4
291 PAVGB(%%mm7, %%mm5) // 11 6 /8
293 PAVGB(%%mm3, %%mm0) // 112 /4
294 PAVGB(%%mm0, %%mm5) // 112246 /16
295 "movq %%mm5, (%%eax, %1, 4) \n\t" // X
299 : "r" (src), "r" (stride), "m" (c->pQPb)
303 const int l1= stride;
304 const int l2= stride + l1;
305 const int l3= stride + l2;
306 const int l4= stride + l3;
307 const int l5= stride + l4;
308 const int l6= stride + l5;
309 const int l7= stride + l6;
310 const int l8= stride + l7;
311 const int l9= stride + l8;
314 for(x=0; x<BLOCK_SIZE; x++)
316 const int first= ABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];
317 const int last= ABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];
320 sums[0] = first + src[l1];
321 sums[1] = src[l1] + src[l2];
322 sums[2] = src[l2] + src[l3];
323 sums[3] = src[l3] + src[l4];
324 sums[4] = src[l4] + src[l5];
325 sums[5] = src[l5] + src[l6];
326 sums[6] = src[l6] + src[l7];
327 sums[7] = src[l7] + src[l8];
328 sums[8] = src[l8] + last;
330 src[l1]= ((sums[0]<<2) + ((first + sums[2])<<1) + sums[4] + 8)>>4;
331 src[l2]= ((src[l2]<<2) + ((first + sums[0] + sums[3])<<1) + sums[5] + 8)>>4;
332 src[l3]= ((src[l3]<<2) + ((first + sums[1] + sums[4])<<1) + sums[6] + 8)>>4;
333 src[l4]= ((src[l4]<<2) + ((sums[2] + sums[5])<<1) + sums[0] + sums[7] + 8)>>4;
334 src[l5]= ((src[l5]<<2) + ((sums[3] + sums[6])<<1) + sums[1] + sums[8] + 8)>>4;
335 src[l6]= ((src[l6]<<2) + ((last + sums[7] + sums[4])<<1) + sums[2] + 8)>>4;
336 src[l7]= (((last + src[l7])<<2) + ((src[l8] + sums[5])<<1) + sums[3] + 8)>>4;
337 src[l8]= ((sums[8]<<2) + ((last + sums[6])<<1) + sums[4] + 8)>>4;
346 * Experimental implementation of the filter (Algorithm 1) described in a paper from Ramkishor & Karandikar
347 * values are correctly clipped (MMX2)
348 * values are wraparound (C)
349 * conclusion: its fast, but introduces ugly horizontal patterns if there is a continious gradient
356 static inline void RENAME(vertRK1Filter)(uint8_t *src, int stride, int QP)
358 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
362 "pxor %%mm7, %%mm7 \n\t" // 0
363 "movq "MANGLE(b80)", %%mm6 \n\t" // MIN_SIGNED_BYTE
364 "leal (%0, %1), %%eax \n\t"
365 "leal (%%eax, %1, 4), %%ecx \n\t"
366 // 0 1 2 3 4 5 6 7 8 9
367 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1 %0+8%1 ecx+4%1
368 "movq "MANGLE(pQPb)", %%mm0 \n\t" // QP,..., QP
369 "movq %%mm0, %%mm1 \n\t" // QP,..., QP
370 "paddusb "MANGLE(b02)", %%mm0 \n\t"
371 "psrlw $2, %%mm0 \n\t"
372 "pand "MANGLE(b3F)", %%mm0 \n\t" // QP/4,..., QP/4
373 "paddusb %%mm1, %%mm0 \n\t" // QP*1.25 ...
374 "movq (%0, %1, 4), %%mm2 \n\t" // line 4
375 "movq (%%ecx), %%mm3 \n\t" // line 5
376 "movq %%mm2, %%mm4 \n\t" // line 4
377 "pcmpeqb %%mm5, %%mm5 \n\t" // -1
378 "pxor %%mm2, %%mm5 \n\t" // -line 4 - 1
380 "paddb %%mm6, %%mm5 \n\t" // (l5-l4)/2
381 "psubusb %%mm3, %%mm4 \n\t"
382 "psubusb %%mm2, %%mm3 \n\t"
383 "por %%mm3, %%mm4 \n\t" // |l4 - l5|
384 "psubusb %%mm0, %%mm4 \n\t"
385 "pcmpeqb %%mm7, %%mm4 \n\t"
386 "pand %%mm4, %%mm5 \n\t" // d/2
388 // "paddb %%mm6, %%mm2 \n\t" // line 4 + 0x80
389 "paddb %%mm5, %%mm2 \n\t"
390 // "psubb %%mm6, %%mm2 \n\t"
391 "movq %%mm2, (%0,%1, 4) \n\t"
393 "movq (%%ecx), %%mm2 \n\t"
394 // "paddb %%mm6, %%mm2 \n\t" // line 5 + 0x80
395 "psubb %%mm5, %%mm2 \n\t"
396 // "psubb %%mm6, %%mm2 \n\t"
397 "movq %%mm2, (%%ecx) \n\t"
399 "paddb %%mm6, %%mm5 \n\t"
400 "psrlw $2, %%mm5 \n\t"
401 "pand "MANGLE(b3F)", %%mm5 \n\t"
402 "psubb "MANGLE(b20)", %%mm5 \n\t" // (l5-l4)/8
404 "movq (%%eax, %1, 2), %%mm2 \n\t"
405 "paddb %%mm6, %%mm2 \n\t" // line 3 + 0x80
406 "paddsb %%mm5, %%mm2 \n\t"
407 "psubb %%mm6, %%mm2 \n\t"
408 "movq %%mm2, (%%eax, %1, 2) \n\t"
410 "movq (%%ecx, %1), %%mm2 \n\t"
411 "paddb %%mm6, %%mm2 \n\t" // line 6 + 0x80
412 "psubsb %%mm5, %%mm2 \n\t"
413 "psubb %%mm6, %%mm2 \n\t"
414 "movq %%mm2, (%%ecx, %1) \n\t"
417 : "r" (src), "r" (stride)
421 const int l1= stride;
422 const int l2= stride + l1;
423 const int l3= stride + l2;
424 const int l4= stride + l3;
425 const int l5= stride + l4;
426 const int l6= stride + l5;
427 // const int l7= stride + l6;
428 // const int l8= stride + l7;
429 // const int l9= stride + l8;
431 const int QP15= QP + (QP>>2);
433 for(x=0; x<BLOCK_SIZE; x++)
435 const int v = (src[x+l5] - src[x+l4]);
451 * Experimental Filter 1
452 * will not damage linear gradients
453 * Flat blocks should look like they where passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
454 * can only smooth blocks at the expected locations (it cant smooth them if they did move)
455 * MMX2 version does correct clipping C version doesnt
457 static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
459 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
463 "pxor %%mm7, %%mm7 \n\t" // 0
464 "leal (%0, %1), %%eax \n\t"
465 "leal (%%eax, %1, 4), %%ecx \n\t"
466 // 0 1 2 3 4 5 6 7 8 9
467 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1 %0+8%1 ecx+4%1
468 "movq (%%eax, %1, 2), %%mm0 \n\t" // line 3
469 "movq (%0, %1, 4), %%mm1 \n\t" // line 4
470 "movq %%mm1, %%mm2 \n\t" // line 4
471 "psubusb %%mm0, %%mm1 \n\t"
472 "psubusb %%mm2, %%mm0 \n\t"
473 "por %%mm1, %%mm0 \n\t" // |l2 - l3|
474 "movq (%%ecx), %%mm3 \n\t" // line 5
475 "movq (%%ecx, %1), %%mm4 \n\t" // line 6
476 "movq %%mm3, %%mm5 \n\t" // line 5
477 "psubusb %%mm4, %%mm3 \n\t"
478 "psubusb %%mm5, %%mm4 \n\t"
479 "por %%mm4, %%mm3 \n\t" // |l5 - l6|
480 PAVGB(%%mm3, %%mm0) // (|l2 - l3| + |l5 - l6|)/2
481 "movq %%mm2, %%mm1 \n\t" // line 4
482 "psubusb %%mm5, %%mm2 \n\t"
483 "movq %%mm2, %%mm4 \n\t"
484 "pcmpeqb %%mm7, %%mm2 \n\t" // (l4 - l5) <= 0 ? -1 : 0
485 "psubusb %%mm1, %%mm5 \n\t"
486 "por %%mm5, %%mm4 \n\t" // |l4 - l5|
487 "psubusb %%mm0, %%mm4 \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
488 "movq %%mm4, %%mm3 \n\t" // d
489 "movq %2, %%mm0 \n\t"
490 "paddusb %%mm0, %%mm0 \n\t"
491 "psubusb %%mm0, %%mm4 \n\t"
492 "pcmpeqb %%mm7, %%mm4 \n\t" // d <= QP ? -1 : 0
493 "psubusb "MANGLE(b01)", %%mm3 \n\t"
494 "pand %%mm4, %%mm3 \n\t" // d <= QP ? d : 0
496 PAVGB(%%mm7, %%mm3) // d/2
497 "movq %%mm3, %%mm1 \n\t" // d/2
498 PAVGB(%%mm7, %%mm3) // d/4
499 PAVGB(%%mm1, %%mm3) // 3*d/8
501 "movq (%0, %1, 4), %%mm0 \n\t" // line 4
502 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
503 "psubusb %%mm3, %%mm0 \n\t"
504 "pxor %%mm2, %%mm0 \n\t"
505 "movq %%mm0, (%0, %1, 4) \n\t" // line 4
507 "movq (%%ecx), %%mm0 \n\t" // line 5
508 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
509 "paddusb %%mm3, %%mm0 \n\t"
510 "pxor %%mm2, %%mm0 \n\t"
511 "movq %%mm0, (%%ecx) \n\t" // line 5
513 PAVGB(%%mm7, %%mm1) // d/4
515 "movq (%%eax, %1, 2), %%mm0 \n\t" // line 3
516 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
517 "psubusb %%mm1, %%mm0 \n\t"
518 "pxor %%mm2, %%mm0 \n\t"
519 "movq %%mm0, (%%eax, %1, 2) \n\t" // line 3
521 "movq (%%ecx, %1), %%mm0 \n\t" // line 6
522 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
523 "paddusb %%mm1, %%mm0 \n\t"
524 "pxor %%mm2, %%mm0 \n\t"
525 "movq %%mm0, (%%ecx, %1) \n\t" // line 6
527 PAVGB(%%mm7, %%mm1) // d/8
529 "movq (%%eax, %1), %%mm0 \n\t" // line 2
530 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
531 "psubusb %%mm1, %%mm0 \n\t"
532 "pxor %%mm2, %%mm0 \n\t"
533 "movq %%mm0, (%%eax, %1) \n\t" // line 2
535 "movq (%%ecx, %1, 2), %%mm0 \n\t" // line 7
536 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
537 "paddusb %%mm1, %%mm0 \n\t"
538 "pxor %%mm2, %%mm0 \n\t"
539 "movq %%mm0, (%%ecx, %1, 2) \n\t" // line 7
542 : "r" (src), "r" (stride), "m" (co->pQPb)
547 const int l1= stride;
548 const int l2= stride + l1;
549 const int l3= stride + l2;
550 const int l4= stride + l3;
551 const int l5= stride + l4;
552 const int l6= stride + l5;
553 const int l7= stride + l6;
554 // const int l8= stride + l7;
555 // const int l9= stride + l8;
559 for(x=0; x<BLOCK_SIZE; x++)
561 int a= src[l3] - src[l4];
562 int b= src[l4] - src[l5];
563 int c= src[l5] - src[l6];
565 int d= ABS(b) - ((ABS(a) + ABS(c))>>1);
570 int v = d * SIGN(-b);
585 static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext *c)
587 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
590 const int l1= stride;
591 const int l2= stride + l1;
592 const int l3= stride + l2;
593 const int l4= (int)tmp - (int)src - stride*3;
594 const int l5= (int)tmp - (int)src - stride*3 + 8;
595 const int l6= stride*3 + l3;
596 const int l7= stride + l6;
597 const int l8= stride + l7;
599 memcpy(tmp, src+stride*7, 8);
600 memcpy(tmp+8, src+stride*8, 8);
605 #if 0 //sligtly more accurate and slightly slower
606 "pxor %%mm7, %%mm7 \n\t" // 0
607 "leal (%0, %1), %%eax \n\t"
608 "leal (%%eax, %1, 4), %%ecx \n\t"
610 // %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 ecx+%1 ecx+2%1
611 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1
614 "movq (%0, %1, 2), %%mm0 \n\t" // l2
615 "movq (%0), %%mm1 \n\t" // l0
616 "movq %%mm0, %%mm2 \n\t" // l2
617 PAVGB(%%mm7, %%mm0) // ~l2/2
618 PAVGB(%%mm1, %%mm0) // ~(l2 + 2l0)/4
619 PAVGB(%%mm2, %%mm0) // ~(5l2 + 2l0)/8
621 "movq (%%eax), %%mm1 \n\t" // l1
622 "movq (%%eax, %1, 2), %%mm3 \n\t" // l3
623 "movq %%mm1, %%mm4 \n\t" // l1
624 PAVGB(%%mm7, %%mm1) // ~l1/2
625 PAVGB(%%mm3, %%mm1) // ~(l1 + 2l3)/4
626 PAVGB(%%mm4, %%mm1) // ~(5l1 + 2l3)/8
628 "movq %%mm0, %%mm4 \n\t" // ~(5l2 + 2l0)/8
629 "psubusb %%mm1, %%mm0 \n\t"
630 "psubusb %%mm4, %%mm1 \n\t"
631 "por %%mm0, %%mm1 \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8
632 // mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0
634 "movq (%0, %1, 4), %%mm0 \n\t" // l4
635 "movq %%mm0, %%mm4 \n\t" // l4
636 PAVGB(%%mm7, %%mm0) // ~l4/2
637 PAVGB(%%mm2, %%mm0) // ~(l4 + 2l2)/4
638 PAVGB(%%mm4, %%mm0) // ~(5l4 + 2l2)/8
640 "movq (%%ecx), %%mm2 \n\t" // l5
641 "movq %%mm3, %%mm5 \n\t" // l3
642 PAVGB(%%mm7, %%mm3) // ~l3/2
643 PAVGB(%%mm2, %%mm3) // ~(l3 + 2l5)/4
644 PAVGB(%%mm5, %%mm3) // ~(5l3 + 2l5)/8
646 "movq %%mm0, %%mm6 \n\t" // ~(5l4 + 2l2)/8
647 "psubusb %%mm3, %%mm0 \n\t"
648 "psubusb %%mm6, %%mm3 \n\t"
649 "por %%mm0, %%mm3 \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8
650 "pcmpeqb %%mm7, %%mm0 \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)
651 // mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0
653 "movq (%%ecx, %1), %%mm6 \n\t" // l6
654 "movq %%mm6, %%mm5 \n\t" // l6
655 PAVGB(%%mm7, %%mm6) // ~l6/2
656 PAVGB(%%mm4, %%mm6) // ~(l6 + 2l4)/4
657 PAVGB(%%mm5, %%mm6) // ~(5l6 + 2l4)/8
659 "movq (%%ecx, %1, 2), %%mm5 \n\t" // l7
660 "movq %%mm2, %%mm4 \n\t" // l5
661 PAVGB(%%mm7, %%mm2) // ~l5/2
662 PAVGB(%%mm5, %%mm2) // ~(l5 + 2l7)/4
663 PAVGB(%%mm4, %%mm2) // ~(5l5 + 2l7)/8
665 "movq %%mm6, %%mm4 \n\t" // ~(5l6 + 2l4)/8
666 "psubusb %%mm2, %%mm6 \n\t"
667 "psubusb %%mm4, %%mm2 \n\t"
668 "por %%mm6, %%mm2 \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8
669 // mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0
672 PMINUB(%%mm2, %%mm1, %%mm4) // MIN(|lenergy|,|renergy|)/8
673 "movq %2, %%mm4 \n\t" // QP //FIXME QP+1 ?
674 "paddusb "MANGLE(b01)", %%mm4 \n\t"
675 "pcmpgtb %%mm3, %%mm4 \n\t" // |menergy|/8 < QP
676 "psubusb %%mm1, %%mm3 \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8
677 "pand %%mm4, %%mm3 \n\t"
679 "movq %%mm3, %%mm1 \n\t"
680 // "psubusb "MANGLE(b01)", %%mm3 \n\t"
683 "paddusb %%mm1, %%mm3 \n\t"
684 // "paddusb "MANGLE(b01)", %%mm3 \n\t"
686 "movq (%%eax, %1, 2), %%mm6 \n\t" //l3
687 "movq (%0, %1, 4), %%mm5 \n\t" //l4
688 "movq (%0, %1, 4), %%mm4 \n\t" //l4
689 "psubusb %%mm6, %%mm5 \n\t"
690 "psubusb %%mm4, %%mm6 \n\t"
691 "por %%mm6, %%mm5 \n\t" // |l3-l4|
692 "pcmpeqb %%mm7, %%mm6 \n\t" // SIGN(l3-l4)
693 "pxor %%mm6, %%mm0 \n\t"
694 "pand %%mm0, %%mm3 \n\t"
695 PMINUB(%%mm5, %%mm3, %%mm0)
697 "psubusb "MANGLE(b01)", %%mm3 \n\t"
700 "movq (%%eax, %1, 2), %%mm0 \n\t"
701 "movq (%0, %1, 4), %%mm2 \n\t"
702 "pxor %%mm6, %%mm0 \n\t"
703 "pxor %%mm6, %%mm2 \n\t"
704 "psubb %%mm3, %%mm0 \n\t"
705 "paddb %%mm3, %%mm2 \n\t"
706 "pxor %%mm6, %%mm0 \n\t"
707 "pxor %%mm6, %%mm2 \n\t"
708 "movq %%mm0, (%%eax, %1, 2) \n\t"
709 "movq %%mm2, (%0, %1, 4) \n\t"
712 "leal (%0, %1), %%eax \n\t"
713 "pcmpeqb %%mm6, %%mm6 \n\t" // -1
715 // %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 ecx+%1 ecx+2%1
716 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1
719 "movq (%%eax, %1, 2), %%mm1 \n\t" // l3
720 "movq (%0, %1, 4), %%mm0 \n\t" // l4
721 "pxor %%mm6, %%mm1 \n\t" // -l3-1
722 PAVGB(%%mm1, %%mm0) // -q+128 = (l4-l3+256)/2
723 // mm1=-l3-1, mm0=128-q
725 "movq (%%eax, %1, 4), %%mm2 \n\t" // l5
726 "movq (%%eax, %1), %%mm3 \n\t" // l2
727 "pxor %%mm6, %%mm2 \n\t" // -l5-1
728 "movq %%mm2, %%mm5 \n\t" // -l5-1
729 "movq "MANGLE(b80)", %%mm4 \n\t" // 128
730 "leal (%%eax, %1, 4), %%ecx \n\t"
731 PAVGB(%%mm3, %%mm2) // (l2-l5+256)/2
732 PAVGB(%%mm0, %%mm4) // ~(l4-l3)/4 + 128
733 PAVGB(%%mm2, %%mm4) // ~(l2-l5)/4 +(l4-l3)/8 + 128
734 PAVGB(%%mm0, %%mm4) // ~(l2-l5)/8 +5(l4-l3)/16 + 128
735 // mm1=-l3-1, mm0=128-q, mm3=l2, mm4=menergy/16 + 128, mm5= -l5-1
737 "movq (%%eax), %%mm2 \n\t" // l1
738 "pxor %%mm6, %%mm2 \n\t" // -l1-1
739 PAVGB(%%mm3, %%mm2) // (l2-l1+256)/2
740 PAVGB((%0), %%mm1) // (l0-l3+256)/2
741 "movq "MANGLE(b80)", %%mm3 \n\t" // 128
742 PAVGB(%%mm2, %%mm3) // ~(l2-l1)/4 + 128
743 PAVGB(%%mm1, %%mm3) // ~(l0-l3)/4 +(l2-l1)/8 + 128
744 PAVGB(%%mm2, %%mm3) // ~(l0-l3)/8 +5(l2-l1)/16 + 128
745 // mm0=128-q, mm3=lenergy/16 + 128, mm4= menergy/16 + 128, mm5= -l5-1
747 PAVGB((%%ecx, %1), %%mm5) // (l6-l5+256)/2
748 "movq (%%ecx, %1, 2), %%mm1 \n\t" // l7
749 "pxor %%mm6, %%mm1 \n\t" // -l7-1
750 PAVGB((%0, %1, 4), %%mm1) // (l4-l7+256)/2
751 "movq "MANGLE(b80)", %%mm2 \n\t" // 128
752 PAVGB(%%mm5, %%mm2) // ~(l6-l5)/4 + 128
753 PAVGB(%%mm1, %%mm2) // ~(l4-l7)/4 +(l6-l5)/8 + 128
754 PAVGB(%%mm5, %%mm2) // ~(l4-l7)/8 +5(l6-l5)/16 + 128
755 // mm0=128-q, mm2=renergy/16 + 128, mm3=lenergy/16 + 128, mm4= menergy/16 + 128
757 "movq "MANGLE(b00)", %%mm1 \n\t" // 0
758 "movq "MANGLE(b00)", %%mm5 \n\t" // 0
759 "psubb %%mm2, %%mm1 \n\t" // 128 - renergy/16
760 "psubb %%mm3, %%mm5 \n\t" // 128 - lenergy/16
761 PMAXUB(%%mm1, %%mm2) // 128 + |renergy/16|
762 PMAXUB(%%mm5, %%mm3) // 128 + |lenergy/16|
763 PMINUB(%%mm2, %%mm3, %%mm1) // 128 + MIN(|lenergy|,|renergy|)/16
765 // mm0=128-q, mm3=128 + MIN(|lenergy|,|renergy|)/16, mm4= menergy/16 + 128
767 "movq "MANGLE(b00)", %%mm7 \n\t" // 0
768 "movq %2, %%mm2 \n\t" // QP
769 PAVGB(%%mm6, %%mm2) // 128 + QP/2
770 "psubb %%mm6, %%mm2 \n\t"
772 "movq %%mm4, %%mm1 \n\t"
773 "pcmpgtb %%mm7, %%mm1 \n\t" // SIGN(menergy)
774 "pxor %%mm1, %%mm4 \n\t"
775 "psubb %%mm1, %%mm4 \n\t" // 128 + |menergy|/16
776 "pcmpgtb %%mm4, %%mm2 \n\t" // |menergy|/16 < QP/2
777 "psubusb %%mm3, %%mm4 \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16
778 // mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16
780 "movq %%mm4, %%mm3 \n\t" // d
781 "psubusb "MANGLE(b01)", %%mm4 \n\t"
782 PAVGB(%%mm7, %%mm4) // d/32
783 PAVGB(%%mm7, %%mm4) // (d + 32)/64
784 "paddb %%mm3, %%mm4 \n\t" // 5d/64
785 "pand %%mm2, %%mm4 \n\t"
787 "movq "MANGLE(b80)", %%mm5 \n\t" // 128
788 "psubb %%mm0, %%mm5 \n\t" // q
789 "paddsb %%mm6, %%mm5 \n\t" // fix bad rounding
790 "pcmpgtb %%mm5, %%mm7 \n\t" // SIGN(q)
791 "pxor %%mm7, %%mm5 \n\t"
793 PMINUB(%%mm5, %%mm4, %%mm3) // MIN(|q|, 5d/64)
794 "pxor %%mm1, %%mm7 \n\t" // SIGN(d*q)
796 "pand %%mm7, %%mm4 \n\t"
797 "movq (%%eax, %1, 2), %%mm0 \n\t"
798 "movq (%0, %1, 4), %%mm2 \n\t"
799 "pxor %%mm1, %%mm0 \n\t"
800 "pxor %%mm1, %%mm2 \n\t"
801 "paddb %%mm4, %%mm0 \n\t"
802 "psubb %%mm4, %%mm2 \n\t"
803 "pxor %%mm1, %%mm0 \n\t"
804 "pxor %%mm1, %%mm2 \n\t"
805 "movq %%mm0, (%%eax, %1, 2) \n\t"
806 "movq %%mm2, (%0, %1, 4) \n\t"
809 : "r" (src), "r" (stride), "m" (c->pQPb)
817 for(x=0; x<BLOCK_SIZE; x++)
819 const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
820 if(ABS(middleEnergy)< 8*QP)
822 const int q=(src[l4] - src[l5])/2;
823 const int leftEnergy= 5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
824 const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
826 int d= ABS(middleEnergy) - MIN( ABS(leftEnergy), ABS(rightEnergy) );
830 d*= SIGN(-middleEnergy);
854 int d= src[x+y*stride] - tmp[x+(y-4)*8];
865 src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
871 printf(" %d %d %d %d\n", num, sum, max, bias);
877 #elif defined (HAVE_MMX)
880 "pxor %%mm7, %%mm7 \n\t"
881 "leal -40(%%esp), %%ecx \n\t" // make space for 4 8-byte vars
882 "andl $0xFFFFFFF8, %%ecx \n\t" // align
884 // %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 edx+%1 edx+2%1
885 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1
887 "movq (%0), %%mm0 \n\t"
888 "movq %%mm0, %%mm1 \n\t"
889 "punpcklbw %%mm7, %%mm0 \n\t" // low part of line 0
890 "punpckhbw %%mm7, %%mm1 \n\t" // high part of line 0
892 "movq (%0, %1), %%mm2 \n\t"
893 "leal (%0, %1, 2), %%eax \n\t"
894 "movq %%mm2, %%mm3 \n\t"
895 "punpcklbw %%mm7, %%mm2 \n\t" // low part of line 1
896 "punpckhbw %%mm7, %%mm3 \n\t" // high part of line 1
898 "movq (%%eax), %%mm4 \n\t"
899 "movq %%mm4, %%mm5 \n\t"
900 "punpcklbw %%mm7, %%mm4 \n\t" // low part of line 2
901 "punpckhbw %%mm7, %%mm5 \n\t" // high part of line 2
903 "paddw %%mm0, %%mm0 \n\t" // 2L0
904 "paddw %%mm1, %%mm1 \n\t" // 2H0
905 "psubw %%mm4, %%mm2 \n\t" // L1 - L2
906 "psubw %%mm5, %%mm3 \n\t" // H1 - H2
907 "psubw %%mm2, %%mm0 \n\t" // 2L0 - L1 + L2
908 "psubw %%mm3, %%mm1 \n\t" // 2H0 - H1 + H2
910 "psllw $2, %%mm2 \n\t" // 4L1 - 4L2
911 "psllw $2, %%mm3 \n\t" // 4H1 - 4H2
912 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2
913 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2
915 "movq (%%eax, %1), %%mm2 \n\t"
916 "movq %%mm2, %%mm3 \n\t"
917 "punpcklbw %%mm7, %%mm2 \n\t" // L3
918 "punpckhbw %%mm7, %%mm3 \n\t" // H3
920 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - L3
921 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - H3
922 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
923 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
924 "movq %%mm0, (%%ecx) \n\t" // 2L0 - 5L1 + 5L2 - 2L3
925 "movq %%mm1, 8(%%ecx) \n\t" // 2H0 - 5H1 + 5H2 - 2H3
927 "movq (%%eax, %1, 2), %%mm0 \n\t"
928 "movq %%mm0, %%mm1 \n\t"
929 "punpcklbw %%mm7, %%mm0 \n\t" // L4
930 "punpckhbw %%mm7, %%mm1 \n\t" // H4
932 "psubw %%mm0, %%mm2 \n\t" // L3 - L4
933 "psubw %%mm1, %%mm3 \n\t" // H3 - H4
934 "movq %%mm2, 16(%%ecx) \n\t" // L3 - L4
935 "movq %%mm3, 24(%%ecx) \n\t" // H3 - H4
936 "paddw %%mm4, %%mm4 \n\t" // 2L2
937 "paddw %%mm5, %%mm5 \n\t" // 2H2
938 "psubw %%mm2, %%mm4 \n\t" // 2L2 - L3 + L4
939 "psubw %%mm3, %%mm5 \n\t" // 2H2 - H3 + H4
941 "leal (%%eax, %1), %0 \n\t"
942 "psllw $2, %%mm2 \n\t" // 4L3 - 4L4
943 "psllw $2, %%mm3 \n\t" // 4H3 - 4H4
944 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4
945 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4
947 "movq (%0, %1, 2), %%mm2 \n\t"
948 "movq %%mm2, %%mm3 \n\t"
949 "punpcklbw %%mm7, %%mm2 \n\t" // L5
950 "punpckhbw %%mm7, %%mm3 \n\t" // H5
951 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - L5
952 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - H5
953 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - 2L5
954 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - 2H5
956 "movq (%%eax, %1, 4), %%mm6 \n\t"
957 "punpcklbw %%mm7, %%mm6 \n\t" // L6
958 "psubw %%mm6, %%mm2 \n\t" // L5 - L6
959 "movq (%%eax, %1, 4), %%mm6 \n\t"
960 "punpckhbw %%mm7, %%mm6 \n\t" // H6
961 "psubw %%mm6, %%mm3 \n\t" // H5 - H6
963 "paddw %%mm0, %%mm0 \n\t" // 2L4
964 "paddw %%mm1, %%mm1 \n\t" // 2H4
965 "psubw %%mm2, %%mm0 \n\t" // 2L4 - L5 + L6
966 "psubw %%mm3, %%mm1 \n\t" // 2H4 - H5 + H6
968 "psllw $2, %%mm2 \n\t" // 4L5 - 4L6
969 "psllw $2, %%mm3 \n\t" // 4H5 - 4H6
970 "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6
971 "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6
973 "movq (%0, %1, 4), %%mm2 \n\t"
974 "movq %%mm2, %%mm3 \n\t"
975 "punpcklbw %%mm7, %%mm2 \n\t" // L7
976 "punpckhbw %%mm7, %%mm3 \n\t" // H7
978 "paddw %%mm2, %%mm2 \n\t" // 2L7
979 "paddw %%mm3, %%mm3 \n\t" // 2H7
980 "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6 - 2L7
981 "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6 - 2H7
983 "movq (%%ecx), %%mm2 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
984 "movq 8(%%ecx), %%mm3 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
987 "movq %%mm7, %%mm6 \n\t" // 0
988 "psubw %%mm0, %%mm6 \n\t"
989 "pmaxsw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
990 "movq %%mm7, %%mm6 \n\t" // 0
991 "psubw %%mm1, %%mm6 \n\t"
992 "pmaxsw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
993 "movq %%mm7, %%mm6 \n\t" // 0
994 "psubw %%mm2, %%mm6 \n\t"
995 "pmaxsw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
996 "movq %%mm7, %%mm6 \n\t" // 0
997 "psubw %%mm3, %%mm6 \n\t"
998 "pmaxsw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
1000 "movq %%mm7, %%mm6 \n\t" // 0
1001 "pcmpgtw %%mm0, %%mm6 \n\t"
1002 "pxor %%mm6, %%mm0 \n\t"
1003 "psubw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
1004 "movq %%mm7, %%mm6 \n\t" // 0
1005 "pcmpgtw %%mm1, %%mm6 \n\t"
1006 "pxor %%mm6, %%mm1 \n\t"
1007 "psubw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
1008 "movq %%mm7, %%mm6 \n\t" // 0
1009 "pcmpgtw %%mm2, %%mm6 \n\t"
1010 "pxor %%mm6, %%mm2 \n\t"
1011 "psubw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
1012 "movq %%mm7, %%mm6 \n\t" // 0
1013 "pcmpgtw %%mm3, %%mm6 \n\t"
1014 "pxor %%mm6, %%mm3 \n\t"
1015 "psubw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
1019 "pminsw %%mm2, %%mm0 \n\t"
1020 "pminsw %%mm3, %%mm1 \n\t"
1022 "movq %%mm0, %%mm6 \n\t"
1023 "psubusw %%mm2, %%mm6 \n\t"
1024 "psubw %%mm6, %%mm0 \n\t"
1025 "movq %%mm1, %%mm6 \n\t"
1026 "psubusw %%mm3, %%mm6 \n\t"
1027 "psubw %%mm6, %%mm1 \n\t"
1030 "movq %%mm7, %%mm6 \n\t" // 0
1031 "pcmpgtw %%mm4, %%mm6 \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
1032 "pxor %%mm6, %%mm4 \n\t"
1033 "psubw %%mm6, %%mm4 \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
1034 "pcmpgtw %%mm5, %%mm7 \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
1035 "pxor %%mm7, %%mm5 \n\t"
1036 "psubw %%mm7, %%mm5 \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
1038 "movd %2, %%mm2 \n\t" // QP
1039 "psllw $3, %%mm2 \n\t" // 8QP
1040 "movq %%mm2, %%mm3 \n\t" // 8QP
1041 "pcmpgtw %%mm4, %%mm2 \n\t"
1042 "pcmpgtw %%mm5, %%mm3 \n\t"
1043 "pand %%mm2, %%mm4 \n\t"
1044 "pand %%mm3, %%mm5 \n\t"
1047 "psubusw %%mm0, %%mm4 \n\t" // hd
1048 "psubusw %%mm1, %%mm5 \n\t" // ld
1051 "movq "MANGLE(w05)", %%mm2 \n\t" // 5
1052 "pmullw %%mm2, %%mm4 \n\t"
1053 "pmullw %%mm2, %%mm5 \n\t"
1054 "movq "MANGLE(w20)", %%mm2 \n\t" // 32
1055 "paddw %%mm2, %%mm4 \n\t"
1056 "paddw %%mm2, %%mm5 \n\t"
1057 "psrlw $6, %%mm4 \n\t"
1058 "psrlw $6, %%mm5 \n\t"
1060 "movq 16(%%ecx), %%mm0 \n\t" // L3 - L4
1061 "movq 24(%%ecx), %%mm1 \n\t" // H3 - H4
1063 "pxor %%mm2, %%mm2 \n\t"
1064 "pxor %%mm3, %%mm3 \n\t"
1066 "pcmpgtw %%mm0, %%mm2 \n\t" // sign (L3-L4)
1067 "pcmpgtw %%mm1, %%mm3 \n\t" // sign (H3-H4)
1068 "pxor %%mm2, %%mm0 \n\t"
1069 "pxor %%mm3, %%mm1 \n\t"
1070 "psubw %%mm2, %%mm0 \n\t" // |L3-L4|
1071 "psubw %%mm3, %%mm1 \n\t" // |H3-H4|
1072 "psrlw $1, %%mm0 \n\t" // |L3 - L4|/2
1073 "psrlw $1, %%mm1 \n\t" // |H3 - H4|/2
1075 "pxor %%mm6, %%mm2 \n\t"
1076 "pxor %%mm7, %%mm3 \n\t"
1077 "pand %%mm2, %%mm4 \n\t"
1078 "pand %%mm3, %%mm5 \n\t"
1081 "pminsw %%mm0, %%mm4 \n\t"
1082 "pminsw %%mm1, %%mm5 \n\t"
1084 "movq %%mm4, %%mm2 \n\t"
1085 "psubusw %%mm0, %%mm2 \n\t"
1086 "psubw %%mm2, %%mm4 \n\t"
1087 "movq %%mm5, %%mm2 \n\t"
1088 "psubusw %%mm1, %%mm2 \n\t"
1089 "psubw %%mm2, %%mm5 \n\t"
1091 "pxor %%mm6, %%mm4 \n\t"
1092 "pxor %%mm7, %%mm5 \n\t"
1093 "psubw %%mm6, %%mm4 \n\t"
1094 "psubw %%mm7, %%mm5 \n\t"
1095 "packsswb %%mm5, %%mm4 \n\t"
1096 "movq (%0), %%mm0 \n\t"
1097 "paddb %%mm4, %%mm0 \n\t"
1098 "movq %%mm0, (%0) \n\t"
1099 "movq (%0, %1), %%mm0 \n\t"
1100 "psubb %%mm4, %%mm0 \n\t"
1101 "movq %%mm0, (%0, %1) \n\t"
1104 : "r" (stride), "m" (c->pQPb)
1108 const int l1= stride;
1109 const int l2= stride + l1;
1110 const int l3= stride + l2;
1111 const int l4= stride + l3;
1112 const int l5= stride + l4;
1113 const int l6= stride + l5;
1114 const int l7= stride + l6;
1115 const int l8= stride + l7;
1116 // const int l9= stride + l8;
1119 for(x=0; x<BLOCK_SIZE; x++)
1121 const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
1122 if(ABS(middleEnergy) < 8*c->QP)
1124 const int q=(src[l4] - src[l5])/2;
1125 const int leftEnergy= 5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
1126 const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
1128 int d= ABS(middleEnergy) - MIN( ABS(leftEnergy), ABS(rightEnergy) );
1132 d*= SIGN(-middleEnergy);
1153 static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
1155 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1157 "pxor %%mm6, %%mm6 \n\t"
1158 "pcmpeqb %%mm7, %%mm7 \n\t"
1159 "movq %2, %%mm0 \n\t"
1160 "punpcklbw %%mm6, %%mm0 \n\t"
1161 "psrlw $1, %%mm0 \n\t"
1162 "psubw %%mm7, %%mm0 \n\t"
1163 "packuswb %%mm0, %%mm0 \n\t"
1164 "movq %%mm0, %3 \n\t"
1166 "leal (%0, %1), %%eax \n\t"
1167 "leal (%%eax, %1, 4), %%edx \n\t"
1169 // 0 1 2 3 4 5 6 7 8 9
1170 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1174 #define FIND_MIN_MAX(addr)\
1175 "movq " #addr ", %%mm0 \n\t"\
1176 "pminub %%mm0, %%mm7 \n\t"\
1177 "pmaxub %%mm0, %%mm6 \n\t"
1179 #define FIND_MIN_MAX(addr)\
1180 "movq " #addr ", %%mm0 \n\t"\
1181 "movq %%mm7, %%mm1 \n\t"\
1182 "psubusb %%mm0, %%mm6 \n\t"\
1183 "paddb %%mm0, %%mm6 \n\t"\
1184 "psubusb %%mm0, %%mm1 \n\t"\
1185 "psubb %%mm1, %%mm7 \n\t"
1188 FIND_MIN_MAX((%%eax))
1189 FIND_MIN_MAX((%%eax, %1))
1190 FIND_MIN_MAX((%%eax, %1, 2))
1191 FIND_MIN_MAX((%0, %1, 4))
1192 FIND_MIN_MAX((%%edx))
1193 FIND_MIN_MAX((%%edx, %1))
1194 FIND_MIN_MAX((%%edx, %1, 2))
1195 FIND_MIN_MAX((%0, %1, 8))
1197 "movq %%mm7, %%mm4 \n\t"
1198 "psrlq $8, %%mm7 \n\t"
1200 "pminub %%mm4, %%mm7 \n\t" // min of pixels
1201 "pshufw $0xF9, %%mm7, %%mm4 \n\t"
1202 "pminub %%mm4, %%mm7 \n\t" // min of pixels
1203 "pshufw $0xFE, %%mm7, %%mm4 \n\t"
1204 "pminub %%mm4, %%mm7 \n\t"
1206 "movq %%mm7, %%mm1 \n\t"
1207 "psubusb %%mm4, %%mm1 \n\t"
1208 "psubb %%mm1, %%mm7 \n\t"
1209 "movq %%mm7, %%mm4 \n\t"
1210 "psrlq $16, %%mm7 \n\t"
1211 "movq %%mm7, %%mm1 \n\t"
1212 "psubusb %%mm4, %%mm1 \n\t"
1213 "psubb %%mm1, %%mm7 \n\t"
1214 "movq %%mm7, %%mm4 \n\t"
1215 "psrlq $32, %%mm7 \n\t"
1216 "movq %%mm7, %%mm1 \n\t"
1217 "psubusb %%mm4, %%mm1 \n\t"
1218 "psubb %%mm1, %%mm7 \n\t"
1222 "movq %%mm6, %%mm4 \n\t"
1223 "psrlq $8, %%mm6 \n\t"
1225 "pmaxub %%mm4, %%mm6 \n\t" // max of pixels
1226 "pshufw $0xF9, %%mm6, %%mm4 \n\t"
1227 "pmaxub %%mm4, %%mm6 \n\t"
1228 "pshufw $0xFE, %%mm6, %%mm4 \n\t"
1229 "pmaxub %%mm4, %%mm6 \n\t"
1231 "psubusb %%mm4, %%mm6 \n\t"
1232 "paddb %%mm4, %%mm6 \n\t"
1233 "movq %%mm6, %%mm4 \n\t"
1234 "psrlq $16, %%mm6 \n\t"
1235 "psubusb %%mm4, %%mm6 \n\t"
1236 "paddb %%mm4, %%mm6 \n\t"
1237 "movq %%mm6, %%mm4 \n\t"
1238 "psrlq $32, %%mm6 \n\t"
1239 "psubusb %%mm4, %%mm6 \n\t"
1240 "paddb %%mm4, %%mm6 \n\t"
1242 "movq %%mm6, %%mm0 \n\t" // max
1243 "psubb %%mm7, %%mm6 \n\t" // max - min
1244 "movd %%mm6, %%ecx \n\t"
1245 "cmpb "MANGLE(deringThreshold)", %%cl \n\t"
1247 "leal -24(%%esp), %%ecx \n\t"
1248 "andl $0xFFFFFFF8, %%ecx \n\t"
1249 PAVGB(%%mm0, %%mm7) // a=(max + min)/2
1250 "punpcklbw %%mm7, %%mm7 \n\t"
1251 "punpcklbw %%mm7, %%mm7 \n\t"
1252 "punpcklbw %%mm7, %%mm7 \n\t"
1253 "movq %%mm7, (%%ecx) \n\t"
1255 "movq (%0), %%mm0 \n\t" // L10
1256 "movq %%mm0, %%mm1 \n\t" // L10
1257 "movq %%mm0, %%mm2 \n\t" // L10
1258 "psllq $8, %%mm1 \n\t"
1259 "psrlq $8, %%mm2 \n\t"
1260 "movd -4(%0), %%mm3 \n\t"
1261 "movd 8(%0), %%mm4 \n\t"
1262 "psrlq $24, %%mm3 \n\t"
1263 "psllq $56, %%mm4 \n\t"
1264 "por %%mm3, %%mm1 \n\t" // L00
1265 "por %%mm4, %%mm2 \n\t" // L20
1266 "movq %%mm1, %%mm3 \n\t" // L00
1267 PAVGB(%%mm2, %%mm1) // (L20 + L00)/2
1268 PAVGB(%%mm0, %%mm1) // (L20 + L00 + 2L10)/4
1269 "psubusb %%mm7, %%mm0 \n\t"
1270 "psubusb %%mm7, %%mm2 \n\t"
1271 "psubusb %%mm7, %%mm3 \n\t"
1272 "pcmpeqb "MANGLE(b00)", %%mm0 \n\t" // L10 > a ? 0 : -1
1273 "pcmpeqb "MANGLE(b00)", %%mm2 \n\t" // L20 > a ? 0 : -1
1274 "pcmpeqb "MANGLE(b00)", %%mm3 \n\t" // L00 > a ? 0 : -1
1275 "paddb %%mm2, %%mm0 \n\t"
1276 "paddb %%mm3, %%mm0 \n\t"
1278 "movq (%%eax), %%mm2 \n\t" // L11
1279 "movq %%mm2, %%mm3 \n\t" // L11
1280 "movq %%mm2, %%mm4 \n\t" // L11
1281 "psllq $8, %%mm3 \n\t"
1282 "psrlq $8, %%mm4 \n\t"
1283 "movd -4(%%eax), %%mm5 \n\t"
1284 "movd 8(%%eax), %%mm6 \n\t"
1285 "psrlq $24, %%mm5 \n\t"
1286 "psllq $56, %%mm6 \n\t"
1287 "por %%mm5, %%mm3 \n\t" // L01
1288 "por %%mm6, %%mm4 \n\t" // L21
1289 "movq %%mm3, %%mm5 \n\t" // L01
1290 PAVGB(%%mm4, %%mm3) // (L21 + L01)/2
1291 PAVGB(%%mm2, %%mm3) // (L21 + L01 + 2L11)/4
1292 "psubusb %%mm7, %%mm2 \n\t"
1293 "psubusb %%mm7, %%mm4 \n\t"
1294 "psubusb %%mm7, %%mm5 \n\t"
1295 "pcmpeqb "MANGLE(b00)", %%mm2 \n\t" // L11 > a ? 0 : -1
1296 "pcmpeqb "MANGLE(b00)", %%mm4 \n\t" // L21 > a ? 0 : -1
1297 "pcmpeqb "MANGLE(b00)", %%mm5 \n\t" // L01 > a ? 0 : -1
1298 "paddb %%mm4, %%mm2 \n\t"
1299 "paddb %%mm5, %%mm2 \n\t"
1301 #define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
1302 "movq " #src ", " #sx " \n\t" /* src[0] */\
1303 "movq " #sx ", " #lx " \n\t" /* src[0] */\
1304 "movq " #sx ", " #t0 " \n\t" /* src[0] */\
1305 "psllq $8, " #lx " \n\t"\
1306 "psrlq $8, " #t0 " \n\t"\
1307 "movd -4" #src ", " #t1 " \n\t"\
1308 "psrlq $24, " #t1 " \n\t"\
1309 "por " #t1 ", " #lx " \n\t" /* src[-1] */\
1310 "movd 8" #src ", " #t1 " \n\t"\
1311 "psllq $56, " #t1 " \n\t"\
1312 "por " #t1 ", " #t0 " \n\t" /* src[+1] */\
1313 "movq " #lx ", " #t1 " \n\t" /* src[-1] */\
1314 PAVGB(t0, lx) /* (src[-1] + src[+1])/2 */\
1315 PAVGB(sx, lx) /* (src[-1] + 2src[0] + src[+1])/4 */\
1317 "movq " #lx ", 8(%%ecx) \n\t"\
1318 "movq (%%ecx), " #lx " \n\t"\
1319 "psubusb " #lx ", " #t1 " \n\t"\
1320 "psubusb " #lx ", " #t0 " \n\t"\
1321 "psubusb " #lx ", " #sx " \n\t"\
1322 "movq "MANGLE(b00)", " #lx " \n\t"\
1323 "pcmpeqb " #lx ", " #t1 " \n\t" /* src[-1] > a ? 0 : -1*/\
1324 "pcmpeqb " #lx ", " #t0 " \n\t" /* src[+1] > a ? 0 : -1*/\
1325 "pcmpeqb " #lx ", " #sx " \n\t" /* src[0] > a ? 0 : -1*/\
1326 "paddb " #t1 ", " #t0 " \n\t"\
1327 "paddb " #t0 ", " #sx " \n\t"\
1329 PAVGB(plx, pplx) /* filtered */\
1330 "movq " #dst ", " #t0 " \n\t" /* dst */\
1331 "movq " #t0 ", " #t1 " \n\t" /* dst */\
1332 "psubusb %3, " #t0 " \n\t"\
1333 "paddusb %3, " #t1 " \n\t"\
1335 PMINUB(t1, pplx, t0)\
1336 "paddb " #sx ", " #ppsx " \n\t"\
1337 "paddb " #psx ", " #ppsx " \n\t"\
1338 "#paddb "MANGLE(b02)", " #ppsx " \n\t"\
1339 "pand "MANGLE(b08)", " #ppsx " \n\t"\
1340 "pcmpeqb " #lx ", " #ppsx " \n\t"\
1341 "pand " #ppsx ", " #pplx " \n\t"\
1342 "pandn " #dst ", " #ppsx " \n\t"\
1343 "por " #pplx ", " #ppsx " \n\t"\
1344 "movq " #ppsx ", " #dst " \n\t"\
1345 "movq 8(%%ecx), " #lx " \n\t"
1362 //DERING_CORE(dst,src ,ppsx ,psx ,sx ,pplx ,plx ,lx ,t0 ,t1)
1363 DERING_CORE((%%eax),(%%eax, %1) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1364 DERING_CORE((%%eax, %1),(%%eax, %1, 2) ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1365 DERING_CORE((%%eax, %1, 2),(%0, %1, 4) ,%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
1366 DERING_CORE((%0, %1, 4),(%%edx) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1367 DERING_CORE((%%edx),(%%edx, %1) ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1368 DERING_CORE((%%edx, %1), (%%edx, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
1369 DERING_CORE((%%edx, %1, 2),(%0, %1, 8) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1370 DERING_CORE((%0, %1, 8),(%%edx, %1, 4) ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1373 : : "r" (src), "r" (stride), "m" (c->pQPb), "m"(c->pQPb2)
1374 : "%eax", "%edx", "%ecx"
1383 const int QP2= c->QP/2 + 1;
1392 if(*p > max) max= *p;
1393 if(*p < min) min= *p;
1396 avg= (min + max + 1)>>1;
1398 if(max - min <deringThreshold) return;
1404 if(src[stride*y + 0] > avg) t+= 1;
1405 if(src[stride*y + 1] > avg) t+= 2;
1406 if(src[stride*y + 2] > avg) t+= 4;
1407 if(src[stride*y + 3] > avg) t+= 8;
1408 if(src[stride*y + 4] > avg) t+= 16;
1409 if(src[stride*y + 5] > avg) t+= 32;
1410 if(src[stride*y + 6] > avg) t+= 64;
1411 if(src[stride*y + 7] > avg) t+= 128;
1412 if(src[stride*y + 8] > avg) t+= 256;
1413 if(src[stride*y + 9] > avg) t+= 512;
1416 t &= (t<<1) & (t>>1);
1422 int t = s[y-1] & s[y] & s[y+1];
1438 int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))
1439 +2*(*(p -1)) + 4*(*p ) + 2*(*(p +1))
1440 +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));
1443 #ifdef DEBUG_DERING_THRESHOLD
1444 asm volatile("emms\n\t":);
1446 static long long numPixels=0;
1447 if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;
1448 // if((max-min)<20 || (max-min)*QP<200)
1449 // if((max-min)*QP < 500)
1453 static int numSkiped=0;
1454 static int errorSum=0;
1455 static int worstQP=0;
1456 static int worstRange=0;
1457 static int worstDiff=0;
1459 int absDiff= ABS(diff);
1460 int error= diff*diff;
1462 if(x==1 || x==8 || y==1 || y==8) continue;
1465 if(absDiff > worstDiff)
1469 worstRange= max-min;
1473 if(1024LL*1024LL*1024LL % numSkiped == 0)
1475 printf( "sum:%1.3f, skip:%d, wQP:%d, "
1476 "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
1477 (float)errorSum/numSkiped, numSkiped, worstQP, worstRange,
1478 worstDiff, (float)numSkiped/numPixels);
1483 if (*p + QP2 < f) *p= *p + QP2;
1484 else if(*p - QP2 > f) *p= *p - QP2;
1489 #ifdef DEBUG_DERING_THRESHOLD
1500 *p = MIN(*p + 20, 255);
1503 // src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
1510 * Deinterlaces the given block by linearly interpolating every second line.
1511 * will be called for every 8x8 block and can read & write from line 4-15
1512 * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1513 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1515 static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)
1517 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1520 "leal (%0, %1), %%eax \n\t"
1521 "leal (%%eax, %1, 4), %%ecx \n\t"
1522 // 0 1 2 3 4 5 6 7 8 9
1523 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1 %0+8%1 ecx+4%1
1525 "movq (%0), %%mm0 \n\t"
1526 "movq (%%eax, %1), %%mm1 \n\t"
1528 "movq %%mm0, (%%eax) \n\t"
1529 "movq (%0, %1, 4), %%mm0 \n\t"
1531 "movq %%mm1, (%%eax, %1, 2) \n\t"
1532 "movq (%%ecx, %1), %%mm1 \n\t"
1534 "movq %%mm0, (%%ecx) \n\t"
1535 "movq (%0, %1, 8), %%mm0 \n\t"
1537 "movq %%mm1, (%%ecx, %1, 2) \n\t"
1539 : : "r" (src), "r" (stride)
1547 a= *(uint32_t*)&src[stride*0];
1548 b= *(uint32_t*)&src[stride*2];
1549 *(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1550 a= *(uint32_t*)&src[stride*4];
1551 *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1552 b= *(uint32_t*)&src[stride*6];
1553 *(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1554 a= *(uint32_t*)&src[stride*8];
1555 *(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1562 * Deinterlaces the given block by cubic interpolating every second line.
1563 * will be called for every 8x8 block and can read & write from line 4-15
1564 * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1565 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1566 * this filter will read lines 3-15 and write 7-13
1568 static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)
1570 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1573 "leal (%0, %1), %%eax \n\t"
1574 "leal (%%eax, %1, 4), %%edx \n\t"
1575 "leal (%%edx, %1, 4), %%ecx \n\t"
1576 "addl %1, %%ecx \n\t"
1577 "pxor %%mm7, %%mm7 \n\t"
1578 // 0 1 2 3 4 5 6 7 8 9 10
1579 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1 ecx
1581 #define DEINT_CUBIC(a,b,c,d,e)\
1582 "movq " #a ", %%mm0 \n\t"\
1583 "movq " #b ", %%mm1 \n\t"\
1584 "movq " #d ", %%mm2 \n\t"\
1585 "movq " #e ", %%mm3 \n\t"\
1586 PAVGB(%%mm2, %%mm1) /* (b+d) /2 */\
1587 PAVGB(%%mm3, %%mm0) /* a(a+e) /2 */\
1588 "movq %%mm0, %%mm2 \n\t"\
1589 "punpcklbw %%mm7, %%mm0 \n\t"\
1590 "punpckhbw %%mm7, %%mm2 \n\t"\
1591 "movq %%mm1, %%mm3 \n\t"\
1592 "punpcklbw %%mm7, %%mm1 \n\t"\
1593 "punpckhbw %%mm7, %%mm3 \n\t"\
1594 "psubw %%mm1, %%mm0 \n\t" /* L(a+e - (b+d))/2 */\
1595 "psubw %%mm3, %%mm2 \n\t" /* H(a+e - (b+d))/2 */\
1596 "psraw $3, %%mm0 \n\t" /* L(a+e - (b+d))/16 */\
1597 "psraw $3, %%mm2 \n\t" /* H(a+e - (b+d))/16 */\
1598 "psubw %%mm0, %%mm1 \n\t" /* L(9b + 9d - a - e)/16 */\
1599 "psubw %%mm2, %%mm3 \n\t" /* H(9b + 9d - a - e)/16 */\
1600 "packuswb %%mm3, %%mm1 \n\t"\
1601 "movq %%mm1, " #c " \n\t"
1603 DEINT_CUBIC((%0), (%%eax, %1), (%%eax, %1, 2), (%0, %1, 4), (%%edx, %1))
1604 DEINT_CUBIC((%%eax, %1), (%0, %1, 4), (%%edx), (%%edx, %1), (%0, %1, 8))
1605 DEINT_CUBIC((%0, %1, 4), (%%edx, %1), (%%edx, %1, 2), (%0, %1, 8), (%%ecx))
1606 DEINT_CUBIC((%%edx, %1), (%0, %1, 8), (%%edx, %1, 4), (%%ecx), (%%ecx, %1, 2))
1608 : : "r" (src), "r" (stride)
1609 : "%eax", "%edx", "ecx"
1616 src[stride*3] = CLIP((-src[0] + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);
1617 src[stride*5] = CLIP((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);
1618 src[stride*7] = CLIP((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);
1619 src[stride*9] = CLIP((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);
1626 * Deinterlaces the given block by filtering every second line with a (-1 4 2 4 -1) filter.
1627 * will be called for every 8x8 block and can read & write from line 4-15
1628 * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1629 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1630 * this filter will read lines 4-13 and write 5-11
1632 static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)
1634 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1637 "leal (%0, %1), %%eax \n\t"
1638 "leal (%%eax, %1, 4), %%edx \n\t"
1639 "pxor %%mm7, %%mm7 \n\t"
1640 "movq (%2), %%mm0 \n\t"
1641 // 0 1 2 3 4 5 6 7 8 9 10
1642 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1 ecx
1644 #define DEINT_FF(a,b,c,d)\
1645 "movq " #a ", %%mm1 \n\t"\
1646 "movq " #b ", %%mm2 \n\t"\
1647 "movq " #c ", %%mm3 \n\t"\
1648 "movq " #d ", %%mm4 \n\t"\
1649 PAVGB(%%mm3, %%mm1) \
1650 PAVGB(%%mm4, %%mm0) \
1651 "movq %%mm0, %%mm3 \n\t"\
1652 "punpcklbw %%mm7, %%mm0 \n\t"\
1653 "punpckhbw %%mm7, %%mm3 \n\t"\
1654 "movq %%mm1, %%mm4 \n\t"\
1655 "punpcklbw %%mm7, %%mm1 \n\t"\
1656 "punpckhbw %%mm7, %%mm4 \n\t"\
1657 "psllw $2, %%mm1 \n\t"\
1658 "psllw $2, %%mm4 \n\t"\
1659 "psubw %%mm0, %%mm1 \n\t"\
1660 "psubw %%mm3, %%mm4 \n\t"\
1661 "movq %%mm2, %%mm5 \n\t"\
1662 "movq %%mm2, %%mm0 \n\t"\
1663 "punpcklbw %%mm7, %%mm2 \n\t"\
1664 "punpckhbw %%mm7, %%mm5 \n\t"\
1665 "paddw %%mm2, %%mm1 \n\t"\
1666 "paddw %%mm5, %%mm4 \n\t"\
1667 "psraw $2, %%mm1 \n\t"\
1668 "psraw $2, %%mm4 \n\t"\
1669 "packuswb %%mm4, %%mm1 \n\t"\
1670 "movq %%mm1, " #b " \n\t"\
1672 DEINT_FF((%0) , (%%eax) , (%%eax, %1), (%%eax, %1, 2))
1673 DEINT_FF((%%eax, %1), (%%eax, %1, 2), (%0, %1, 4), (%%edx) )
1674 DEINT_FF((%0, %1, 4), (%%edx) , (%%edx, %1), (%%edx, %1, 2))
1675 DEINT_FF((%%edx, %1), (%%edx, %1, 2), (%0, %1, 8), (%%edx, %1, 4))
1677 "movq %%mm0, (%2) \n\t"
1678 : : "r" (src), "r" (stride), "r"(tmp)
1687 int t2= src[stride*1];
1689 src[stride*1]= CLIP((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);
1691 src[stride*3]= CLIP((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);
1693 src[stride*5]= CLIP((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);
1695 src[stride*7]= CLIP((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);
1704 * Deinterlaces the given block by filtering every line with a (-1 2 6 2 -1) filter.
1705 * will be called for every 8x8 block and can read & write from line 4-15
1706 * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1707 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1708 * this filter will read lines 4-13 and write 4-11
1710 static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)
1712 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1715 "leal (%0, %1), %%eax \n\t"
1716 "leal (%%eax, %1, 4), %%edx \n\t"
1717 "pxor %%mm7, %%mm7 \n\t"
1718 "movq (%2), %%mm0 \n\t"
1719 "movq (%3), %%mm1 \n\t"
1720 // 0 1 2 3 4 5 6 7 8 9 10
1721 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1 ecx
1723 #define DEINT_L5(t1,t2,a,b,c)\
1724 "movq " #a ", %%mm2 \n\t"\
1725 "movq " #b ", %%mm3 \n\t"\
1726 "movq " #c ", %%mm4 \n\t"\
1729 "movq %%mm2, %%mm5 \n\t"\
1730 "movq %%mm2, " #t1 " \n\t"\
1731 "punpcklbw %%mm7, %%mm2 \n\t"\
1732 "punpckhbw %%mm7, %%mm5 \n\t"\
1733 "movq %%mm2, %%mm6 \n\t"\
1734 "paddw %%mm2, %%mm2 \n\t"\
1735 "paddw %%mm6, %%mm2 \n\t"\
1736 "movq %%mm5, %%mm6 \n\t"\
1737 "paddw %%mm5, %%mm5 \n\t"\
1738 "paddw %%mm6, %%mm5 \n\t"\
1739 "movq %%mm3, %%mm6 \n\t"\
1740 "punpcklbw %%mm7, %%mm3 \n\t"\
1741 "punpckhbw %%mm7, %%mm6 \n\t"\
1742 "paddw %%mm3, %%mm3 \n\t"\
1743 "paddw %%mm6, %%mm6 \n\t"\
1744 "paddw %%mm3, %%mm2 \n\t"\
1745 "paddw %%mm6, %%mm5 \n\t"\
1746 "movq %%mm4, %%mm6 \n\t"\
1747 "punpcklbw %%mm7, %%mm4 \n\t"\
1748 "punpckhbw %%mm7, %%mm6 \n\t"\
1749 "psubw %%mm4, %%mm2 \n\t"\
1750 "psubw %%mm6, %%mm5 \n\t"\
1751 "psraw $2, %%mm2 \n\t"\
1752 "psraw $2, %%mm5 \n\t"\
1753 "packuswb %%mm5, %%mm2 \n\t"\
1754 "movq %%mm2, " #a " \n\t"\
1756 DEINT_L5(%%mm0, %%mm1, (%0) , (%%eax) , (%%eax, %1) )
1757 DEINT_L5(%%mm1, %%mm0, (%%eax) , (%%eax, %1) , (%%eax, %1, 2))
1758 DEINT_L5(%%mm0, %%mm1, (%%eax, %1) , (%%eax, %1, 2), (%0, %1, 4) )
1759 DEINT_L5(%%mm1, %%mm0, (%%eax, %1, 2), (%0, %1, 4) , (%%edx) )
1760 DEINT_L5(%%mm0, %%mm1, (%0, %1, 4) , (%%edx) , (%%edx, %1) )
1761 DEINT_L5(%%mm1, %%mm0, (%%edx) , (%%edx, %1) , (%%edx, %1, 2))
1762 DEINT_L5(%%mm0, %%mm1, (%%edx, %1) , (%%edx, %1, 2), (%0, %1, 8) )
1763 DEINT_L5(%%mm1, %%mm0, (%%edx, %1, 2), (%0, %1, 8) , (%%edx, %1, 4))
1765 "movq %%mm0, (%2) \n\t"
1766 "movq %%mm1, (%3) \n\t"
1767 : : "r" (src), "r" (stride), "r"(tmp), "r"(tmp2)
1779 src[stride*0]= CLIP((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);
1781 src[stride*1]= CLIP((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);
1783 src[stride*2]= CLIP((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);
1785 src[stride*3]= CLIP((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);
1787 src[stride*4]= CLIP((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);
1789 src[stride*5]= CLIP((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);
1791 src[stride*6]= CLIP((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);
1793 src[stride*7]= CLIP((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);
1804 * Deinterlaces the given block by filtering all lines with a (1 2 1) filter.
1805 * will be called for every 8x8 block and can read & write from line 4-15
1806 * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1807 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1808 * will shift the image up by 1 line (FIXME if this is a problem)
1809 * this filter will read lines 4-13 and write 4-11
1811 static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride)
1813 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1816 "leal (%0, %1), %%eax \n\t"
1817 "leal (%%eax, %1, 4), %%edx \n\t"
1818 // 0 1 2 3 4 5 6 7 8 9
1819 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1821 "movq (%0), %%mm0 \n\t" // L0
1822 "movq (%%eax, %1), %%mm1 \n\t" // L2
1823 PAVGB(%%mm1, %%mm0) // L0+L2
1824 "movq (%%eax), %%mm2 \n\t" // L1
1826 "movq %%mm0, (%0) \n\t"
1827 "movq (%%eax, %1, 2), %%mm0 \n\t" // L3
1828 PAVGB(%%mm0, %%mm2) // L1+L3
1829 PAVGB(%%mm1, %%mm2) // 2L2 + L1 + L3
1830 "movq %%mm2, (%%eax) \n\t"
1831 "movq (%0, %1, 4), %%mm2 \n\t" // L4
1832 PAVGB(%%mm2, %%mm1) // L2+L4
1833 PAVGB(%%mm0, %%mm1) // 2L3 + L2 + L4
1834 "movq %%mm1, (%%eax, %1) \n\t"
1835 "movq (%%edx), %%mm1 \n\t" // L5
1836 PAVGB(%%mm1, %%mm0) // L3+L5
1837 PAVGB(%%mm2, %%mm0) // 2L4 + L3 + L5
1838 "movq %%mm0, (%%eax, %1, 2) \n\t"
1839 "movq (%%edx, %1), %%mm0 \n\t" // L6
1840 PAVGB(%%mm0, %%mm2) // L4+L6
1841 PAVGB(%%mm1, %%mm2) // 2L5 + L4 + L6
1842 "movq %%mm2, (%0, %1, 4) \n\t"
1843 "movq (%%edx, %1, 2), %%mm2 \n\t" // L7
1844 PAVGB(%%mm2, %%mm1) // L5+L7
1845 PAVGB(%%mm0, %%mm1) // 2L6 + L5 + L7
1846 "movq %%mm1, (%%edx) \n\t"
1847 "movq (%0, %1, 8), %%mm1 \n\t" // L8
1848 PAVGB(%%mm1, %%mm0) // L6+L8
1849 PAVGB(%%mm2, %%mm0) // 2L7 + L6 + L8
1850 "movq %%mm0, (%%edx, %1) \n\t"
1851 "movq (%%edx, %1, 4), %%mm0 \n\t" // L9
1852 PAVGB(%%mm0, %%mm2) // L7+L9
1853 PAVGB(%%mm1, %%mm2) // 2L8 + L7 + L9
1854 "movq %%mm2, (%%edx, %1, 2) \n\t"
1857 : : "r" (src), "r" (stride)
1865 a= *(uint32_t*)&src[stride*0];
1866 b= *(uint32_t*)&src[stride*1];
1867 c= *(uint32_t*)&src[stride*2];
1868 a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1869 *(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1871 a= *(uint32_t*)&src[stride*3];
1872 b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1873 *(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1875 b= *(uint32_t*)&src[stride*4];
1876 c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1877 *(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1879 c= *(uint32_t*)&src[stride*5];
1880 a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1881 *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1883 a= *(uint32_t*)&src[stride*6];
1884 b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1885 *(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1887 b= *(uint32_t*)&src[stride*7];
1888 c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1889 *(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1891 c= *(uint32_t*)&src[stride*8];
1892 a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1893 *(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1895 a= *(uint32_t*)&src[stride*9];
1896 b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1897 *(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1905 * Deinterlaces the given block by applying a median filter to every second line.
1906 * will be called for every 8x8 block and can read & write from line 4-15,
1907 * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1908 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1910 static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)
1916 "leal (%0, %1), %%eax \n\t"
1917 "leal (%%eax, %1, 4), %%edx \n\t"
1918 // 0 1 2 3 4 5 6 7 8 9
1919 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1921 "movq (%0), %%mm0 \n\t" //
1922 "movq (%%eax, %1), %%mm2 \n\t" //
1923 "movq (%%eax), %%mm1 \n\t" //
1924 "movq %%mm0, %%mm3 \n\t"
1925 "pmaxub %%mm1, %%mm0 \n\t" //
1926 "pminub %%mm3, %%mm1 \n\t" //
1927 "pmaxub %%mm2, %%mm1 \n\t" //
1928 "pminub %%mm1, %%mm0 \n\t"
1929 "movq %%mm0, (%%eax) \n\t"
1931 "movq (%0, %1, 4), %%mm0 \n\t" //
1932 "movq (%%eax, %1, 2), %%mm1 \n\t" //
1933 "movq %%mm2, %%mm3 \n\t"
1934 "pmaxub %%mm1, %%mm2 \n\t" //
1935 "pminub %%mm3, %%mm1 \n\t" //
1936 "pmaxub %%mm0, %%mm1 \n\t" //
1937 "pminub %%mm1, %%mm2 \n\t"
1938 "movq %%mm2, (%%eax, %1, 2) \n\t"
1940 "movq (%%edx), %%mm2 \n\t" //
1941 "movq (%%edx, %1), %%mm1 \n\t" //
1942 "movq %%mm2, %%mm3 \n\t"
1943 "pmaxub %%mm0, %%mm2 \n\t" //
1944 "pminub %%mm3, %%mm0 \n\t" //
1945 "pmaxub %%mm1, %%mm0 \n\t" //
1946 "pminub %%mm0, %%mm2 \n\t"
1947 "movq %%mm2, (%%edx) \n\t"
1949 "movq (%%edx, %1, 2), %%mm2 \n\t" //
1950 "movq (%0, %1, 8), %%mm0 \n\t" //
1951 "movq %%mm2, %%mm3 \n\t"
1952 "pmaxub %%mm0, %%mm2 \n\t" //
1953 "pminub %%mm3, %%mm0 \n\t" //
1954 "pmaxub %%mm1, %%mm0 \n\t" //
1955 "pminub %%mm0, %%mm2 \n\t"
1956 "movq %%mm2, (%%edx, %1, 2) \n\t"
1959 : : "r" (src), "r" (stride)
1963 #else // MMX without MMX2
1965 "leal (%0, %1), %%eax \n\t"
1966 "leal (%%eax, %1, 4), %%edx \n\t"
1967 // 0 1 2 3 4 5 6 7 8 9
1968 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1969 "pxor %%mm7, %%mm7 \n\t"
1971 #define MEDIAN(a,b,c)\
1972 "movq " #a ", %%mm0 \n\t"\
1973 "movq " #b ", %%mm2 \n\t"\
1974 "movq " #c ", %%mm1 \n\t"\
1975 "movq %%mm0, %%mm3 \n\t"\
1976 "movq %%mm1, %%mm4 \n\t"\
1977 "movq %%mm2, %%mm5 \n\t"\
1978 "psubusb %%mm1, %%mm3 \n\t"\
1979 "psubusb %%mm2, %%mm4 \n\t"\
1980 "psubusb %%mm0, %%mm5 \n\t"\
1981 "pcmpeqb %%mm7, %%mm3 \n\t"\
1982 "pcmpeqb %%mm7, %%mm4 \n\t"\
1983 "pcmpeqb %%mm7, %%mm5 \n\t"\
1984 "movq %%mm3, %%mm6 \n\t"\
1985 "pxor %%mm4, %%mm3 \n\t"\
1986 "pxor %%mm5, %%mm4 \n\t"\
1987 "pxor %%mm6, %%mm5 \n\t"\
1988 "por %%mm3, %%mm1 \n\t"\
1989 "por %%mm4, %%mm2 \n\t"\
1990 "por %%mm5, %%mm0 \n\t"\
1991 "pand %%mm2, %%mm0 \n\t"\
1992 "pand %%mm1, %%mm0 \n\t"\
1993 "movq %%mm0, " #b " \n\t"
1995 MEDIAN((%0), (%%eax), (%%eax, %1))
1996 MEDIAN((%%eax, %1), (%%eax, %1, 2), (%0, %1, 4))
1997 MEDIAN((%0, %1, 4), (%%edx), (%%edx, %1))
1998 MEDIAN((%%edx, %1), (%%edx, %1, 2), (%0, %1, 8))
2000 : : "r" (src), "r" (stride)
2007 // FIXME - there should be a way to do a few columns in parallel like w/mmx
2010 uint8_t *colsrc = src;
2013 int a, b, c, d, e, f;
2015 b = colsrc[stride ];
2016 c = colsrc[stride*2];
2020 colsrc[stride ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));
2030 * transposes and shift the given 8x8 Block into dst1 and dst2
2032 static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)
2035 "leal (%0, %1), %%eax \n\t"
2036 // 0 1 2 3 4 5 6 7 8 9
2037 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
2038 "movq (%0), %%mm0 \n\t" // 12345678
2039 "movq (%%eax), %%mm1 \n\t" // abcdefgh
2040 "movq %%mm0, %%mm2 \n\t" // 12345678
2041 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
2042 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
2044 "movq (%%eax, %1), %%mm1 \n\t"
2045 "movq (%%eax, %1, 2), %%mm3 \n\t"
2046 "movq %%mm1, %%mm4 \n\t"
2047 "punpcklbw %%mm3, %%mm1 \n\t"
2048 "punpckhbw %%mm3, %%mm4 \n\t"
2050 "movq %%mm0, %%mm3 \n\t"
2051 "punpcklwd %%mm1, %%mm0 \n\t"
2052 "punpckhwd %%mm1, %%mm3 \n\t"
2053 "movq %%mm2, %%mm1 \n\t"
2054 "punpcklwd %%mm4, %%mm2 \n\t"
2055 "punpckhwd %%mm4, %%mm1 \n\t"
2057 "movd %%mm0, 128(%2) \n\t"
2058 "psrlq $32, %%mm0 \n\t"
2059 "movd %%mm0, 144(%2) \n\t"
2060 "movd %%mm3, 160(%2) \n\t"
2061 "psrlq $32, %%mm3 \n\t"
2062 "movd %%mm3, 176(%2) \n\t"
2063 "movd %%mm3, 48(%3) \n\t"
2064 "movd %%mm2, 192(%2) \n\t"
2065 "movd %%mm2, 64(%3) \n\t"
2066 "psrlq $32, %%mm2 \n\t"
2067 "movd %%mm2, 80(%3) \n\t"
2068 "movd %%mm1, 96(%3) \n\t"
2069 "psrlq $32, %%mm1 \n\t"
2070 "movd %%mm1, 112(%3) \n\t"
2072 "leal (%%eax, %1, 4), %%eax \n\t"
2074 "movq (%0, %1, 4), %%mm0 \n\t" // 12345678
2075 "movq (%%eax), %%mm1 \n\t" // abcdefgh
2076 "movq %%mm0, %%mm2 \n\t" // 12345678
2077 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
2078 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
2080 "movq (%%eax, %1), %%mm1 \n\t"
2081 "movq (%%eax, %1, 2), %%mm3 \n\t"
2082 "movq %%mm1, %%mm4 \n\t"
2083 "punpcklbw %%mm3, %%mm1 \n\t"
2084 "punpckhbw %%mm3, %%mm4 \n\t"
2086 "movq %%mm0, %%mm3 \n\t"
2087 "punpcklwd %%mm1, %%mm0 \n\t"
2088 "punpckhwd %%mm1, %%mm3 \n\t"
2089 "movq %%mm2, %%mm1 \n\t"
2090 "punpcklwd %%mm4, %%mm2 \n\t"
2091 "punpckhwd %%mm4, %%mm1 \n\t"
2093 "movd %%mm0, 132(%2) \n\t"
2094 "psrlq $32, %%mm0 \n\t"
2095 "movd %%mm0, 148(%2) \n\t"
2096 "movd %%mm3, 164(%2) \n\t"
2097 "psrlq $32, %%mm3 \n\t"
2098 "movd %%mm3, 180(%2) \n\t"
2099 "movd %%mm3, 52(%3) \n\t"
2100 "movd %%mm2, 196(%2) \n\t"
2101 "movd %%mm2, 68(%3) \n\t"
2102 "psrlq $32, %%mm2 \n\t"
2103 "movd %%mm2, 84(%3) \n\t"
2104 "movd %%mm1, 100(%3) \n\t"
2105 "psrlq $32, %%mm1 \n\t"
2106 "movd %%mm1, 116(%3) \n\t"
2109 :: "r" (src), "r" (srcStride), "r" (dst1), "r" (dst2)
2115 * transposes the given 8x8 block
2117 static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)
2120 "leal (%0, %1), %%eax \n\t"
2121 "leal (%%eax, %1, 4), %%edx \n\t"
2122 // 0 1 2 3 4 5 6 7 8 9
2123 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
2124 "movq (%2), %%mm0 \n\t" // 12345678
2125 "movq 16(%2), %%mm1 \n\t" // abcdefgh
2126 "movq %%mm0, %%mm2 \n\t" // 12345678
2127 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
2128 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
2130 "movq 32(%2), %%mm1 \n\t"
2131 "movq 48(%2), %%mm3 \n\t"
2132 "movq %%mm1, %%mm4 \n\t"
2133 "punpcklbw %%mm3, %%mm1 \n\t"
2134 "punpckhbw %%mm3, %%mm4 \n\t"
2136 "movq %%mm0, %%mm3 \n\t"
2137 "punpcklwd %%mm1, %%mm0 \n\t"
2138 "punpckhwd %%mm1, %%mm3 \n\t"
2139 "movq %%mm2, %%mm1 \n\t"
2140 "punpcklwd %%mm4, %%mm2 \n\t"
2141 "punpckhwd %%mm4, %%mm1 \n\t"
2143 "movd %%mm0, (%0) \n\t"
2144 "psrlq $32, %%mm0 \n\t"
2145 "movd %%mm0, (%%eax) \n\t"
2146 "movd %%mm3, (%%eax, %1) \n\t"
2147 "psrlq $32, %%mm3 \n\t"
2148 "movd %%mm3, (%%eax, %1, 2) \n\t"
2149 "movd %%mm2, (%0, %1, 4) \n\t"
2150 "psrlq $32, %%mm2 \n\t"
2151 "movd %%mm2, (%%edx) \n\t"
2152 "movd %%mm1, (%%edx, %1) \n\t"
2153 "psrlq $32, %%mm1 \n\t"
2154 "movd %%mm1, (%%edx, %1, 2) \n\t"
2157 "movq 64(%2), %%mm0 \n\t" // 12345678
2158 "movq 80(%2), %%mm1 \n\t" // abcdefgh
2159 "movq %%mm0, %%mm2 \n\t" // 12345678
2160 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
2161 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
2163 "movq 96(%2), %%mm1 \n\t"
2164 "movq 112(%2), %%mm3 \n\t"
2165 "movq %%mm1, %%mm4 \n\t"
2166 "punpcklbw %%mm3, %%mm1 \n\t"
2167 "punpckhbw %%mm3, %%mm4 \n\t"
2169 "movq %%mm0, %%mm3 \n\t"
2170 "punpcklwd %%mm1, %%mm0 \n\t"
2171 "punpckhwd %%mm1, %%mm3 \n\t"
2172 "movq %%mm2, %%mm1 \n\t"
2173 "punpcklwd %%mm4, %%mm2 \n\t"
2174 "punpckhwd %%mm4, %%mm1 \n\t"
2176 "movd %%mm0, 4(%0) \n\t"
2177 "psrlq $32, %%mm0 \n\t"
2178 "movd %%mm0, 4(%%eax) \n\t"
2179 "movd %%mm3, 4(%%eax, %1) \n\t"
2180 "psrlq $32, %%mm3 \n\t"
2181 "movd %%mm3, 4(%%eax, %1, 2) \n\t"
2182 "movd %%mm2, 4(%0, %1, 4) \n\t"
2183 "psrlq $32, %%mm2 \n\t"
2184 "movd %%mm2, 4(%%edx) \n\t"
2185 "movd %%mm1, 4(%%edx, %1) \n\t"
2186 "psrlq $32, %%mm1 \n\t"
2187 "movd %%mm1, 4(%%edx, %1, 2) \n\t"
2189 :: "r" (dst), "r" (dstStride), "r" (src)
2194 //static int test=0;
2196 static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
2197 uint8_t *tempBlured, uint32_t *tempBluredPast, int *maxNoise)
2199 // to save a register (FIXME do this outside of the loops)
2200 tempBluredPast[127]= maxNoise[0];
2201 tempBluredPast[128]= maxNoise[1];
2202 tempBluredPast[129]= maxNoise[2];
2204 #define FAST_L2_DIFF
2205 //#define L1_DIFF //u should change the thresholds too if u try that one
2206 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
2208 "leal (%2, %2, 2), %%eax \n\t" // 3*stride
2209 "leal (%2, %2, 4), %%edx \n\t" // 5*stride
2210 "leal (%%edx, %2, 2), %%ecx \n\t" // 7*stride
2211 // 0 1 2 3 4 5 6 7 8 9
2212 // %x %x+%2 %x+2%2 %x+eax %x+4%2 %x+edx %x+2eax %x+ecx %x+8%2
2214 #ifdef L1_DIFF //needs mmx2
2215 "movq (%0), %%mm0 \n\t" // L0
2216 "psadbw (%1), %%mm0 \n\t" // |L0-R0|
2217 "movq (%0, %2), %%mm1 \n\t" // L1
2218 "psadbw (%1, %2), %%mm1 \n\t" // |L1-R1|
2219 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2220 "psadbw (%1, %2, 2), %%mm2 \n\t" // |L2-R2|
2221 "movq (%0, %%eax), %%mm3 \n\t" // L3
2222 "psadbw (%1, %%eax), %%mm3 \n\t" // |L3-R3|
2224 "movq (%0, %2, 4), %%mm4 \n\t" // L4
2225 "paddw %%mm1, %%mm0 \n\t"
2226 "psadbw (%1, %2, 4), %%mm4 \n\t" // |L4-R4|
2227 "movq (%0, %%edx), %%mm5 \n\t" // L5
2228 "paddw %%mm2, %%mm0 \n\t"
2229 "psadbw (%1, %%edx), %%mm5 \n\t" // |L5-R5|
2230 "movq (%0, %%eax, 2), %%mm6 \n\t" // L6
2231 "paddw %%mm3, %%mm0 \n\t"
2232 "psadbw (%1, %%eax, 2), %%mm6 \n\t" // |L6-R6|
2233 "movq (%0, %%ecx), %%mm7 \n\t" // L7
2234 "paddw %%mm4, %%mm0 \n\t"
2235 "psadbw (%1, %%ecx), %%mm7 \n\t" // |L7-R7|
2236 "paddw %%mm5, %%mm6 \n\t"
2237 "paddw %%mm7, %%mm6 \n\t"
2238 "paddw %%mm6, %%mm0 \n\t"
2239 #elif defined (FAST_L2_DIFF)
2240 "pcmpeqb %%mm7, %%mm7 \n\t"
2241 "movq "MANGLE(b80)", %%mm6 \n\t"
2242 "pxor %%mm0, %%mm0 \n\t"
2243 #define L2_DIFF_CORE(a, b)\
2244 "movq " #a ", %%mm5 \n\t"\
2245 "movq " #b ", %%mm2 \n\t"\
2246 "pxor %%mm7, %%mm2 \n\t"\
2247 PAVGB(%%mm2, %%mm5)\
2248 "paddb %%mm6, %%mm5 \n\t"\
2249 "movq %%mm5, %%mm2 \n\t"\
2250 "psllw $8, %%mm5 \n\t"\
2251 "pmaddwd %%mm5, %%mm5 \n\t"\
2252 "pmaddwd %%mm2, %%mm2 \n\t"\
2253 "paddd %%mm2, %%mm5 \n\t"\
2254 "psrld $14, %%mm5 \n\t"\
2255 "paddd %%mm5, %%mm0 \n\t"
2257 L2_DIFF_CORE((%0), (%1))
2258 L2_DIFF_CORE((%0, %2), (%1, %2))
2259 L2_DIFF_CORE((%0, %2, 2), (%1, %2, 2))
2260 L2_DIFF_CORE((%0, %%eax), (%1, %%eax))
2261 L2_DIFF_CORE((%0, %2, 4), (%1, %2, 4))
2262 L2_DIFF_CORE((%0, %%edx), (%1, %%edx))
2263 L2_DIFF_CORE((%0, %%eax,2), (%1, %%eax,2))
2264 L2_DIFF_CORE((%0, %%ecx), (%1, %%ecx))
2267 "pxor %%mm7, %%mm7 \n\t"
2268 "pxor %%mm0, %%mm0 \n\t"
2269 #define L2_DIFF_CORE(a, b)\
2270 "movq " #a ", %%mm5 \n\t"\
2271 "movq " #b ", %%mm2 \n\t"\
2272 "movq %%mm5, %%mm1 \n\t"\
2273 "movq %%mm2, %%mm3 \n\t"\
2274 "punpcklbw %%mm7, %%mm5 \n\t"\
2275 "punpckhbw %%mm7, %%mm1 \n\t"\
2276 "punpcklbw %%mm7, %%mm2 \n\t"\
2277 "punpckhbw %%mm7, %%mm3 \n\t"\
2278 "psubw %%mm2, %%mm5 \n\t"\
2279 "psubw %%mm3, %%mm1 \n\t"\
2280 "pmaddwd %%mm5, %%mm5 \n\t"\
2281 "pmaddwd %%mm1, %%mm1 \n\t"\
2282 "paddd %%mm1, %%mm5 \n\t"\
2283 "paddd %%mm5, %%mm0 \n\t"
2285 L2_DIFF_CORE((%0), (%1))
2286 L2_DIFF_CORE((%0, %2), (%1, %2))
2287 L2_DIFF_CORE((%0, %2, 2), (%1, %2, 2))
2288 L2_DIFF_CORE((%0, %%eax), (%1, %%eax))
2289 L2_DIFF_CORE((%0, %2, 4), (%1, %2, 4))
2290 L2_DIFF_CORE((%0, %%edx), (%1, %%edx))
2291 L2_DIFF_CORE((%0, %%eax,2), (%1, %%eax,2))
2292 L2_DIFF_CORE((%0, %%ecx), (%1, %%ecx))
2296 "movq %%mm0, %%mm4 \n\t"
2297 "psrlq $32, %%mm0 \n\t"
2298 "paddd %%mm0, %%mm4 \n\t"
2299 "movd %%mm4, %%ecx \n\t"
2300 "shll $2, %%ecx \n\t"
2301 "movl %3, %%edx \n\t"
2302 "addl -4(%%edx), %%ecx \n\t"
2303 "addl 4(%%edx), %%ecx \n\t"
2304 "addl -1024(%%edx), %%ecx \n\t"
2305 "addl $4, %%ecx \n\t"
2306 "addl 1024(%%edx), %%ecx \n\t"
2307 "shrl $3, %%ecx \n\t"
2308 "movl %%ecx, (%%edx) \n\t"
2310 // "movl %3, %%ecx \n\t"
2311 // "movl %%ecx, test \n\t"
2313 "cmpl 512(%%edx), %%ecx \n\t"
2315 "cmpl 516(%%edx), %%ecx \n\t"
2318 "leal (%%eax, %2, 2), %%edx \n\t" // 5*stride
2319 "leal (%%edx, %2, 2), %%ecx \n\t" // 7*stride
2320 "movq (%0), %%mm0 \n\t" // L0
2321 "movq (%0, %2), %%mm1 \n\t" // L1
2322 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2323 "movq (%0, %%eax), %%mm3 \n\t" // L3
2324 "movq (%0, %2, 4), %%mm4 \n\t" // L4
2325 "movq (%0, %%edx), %%mm5 \n\t" // L5
2326 "movq (%0, %%eax, 2), %%mm6 \n\t" // L6
2327 "movq (%0, %%ecx), %%mm7 \n\t" // L7
2328 "movq %%mm0, (%1) \n\t" // L0
2329 "movq %%mm1, (%1, %2) \n\t" // L1
2330 "movq %%mm2, (%1, %2, 2) \n\t" // L2
2331 "movq %%mm3, (%1, %%eax) \n\t" // L3
2332 "movq %%mm4, (%1, %2, 4) \n\t" // L4
2333 "movq %%mm5, (%1, %%edx) \n\t" // L5
2334 "movq %%mm6, (%1, %%eax, 2) \n\t" // L6
2335 "movq %%mm7, (%1, %%ecx) \n\t" // L7
2339 "leal (%%eax, %2, 2), %%edx \n\t" // 5*stride
2340 "leal (%%edx, %2, 2), %%ecx \n\t" // 7*stride
2341 "movq (%0), %%mm0 \n\t" // L0
2342 PAVGB((%1), %%mm0) // L0
2343 "movq (%0, %2), %%mm1 \n\t" // L1
2344 PAVGB((%1, %2), %%mm1) // L1
2345 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2346 PAVGB((%1, %2, 2), %%mm2) // L2
2347 "movq (%0, %%eax), %%mm3 \n\t" // L3
2348 PAVGB((%1, %%eax), %%mm3) // L3
2349 "movq (%0, %2, 4), %%mm4 \n\t" // L4
2350 PAVGB((%1, %2, 4), %%mm4) // L4
2351 "movq (%0, %%edx), %%mm5 \n\t" // L5
2352 PAVGB((%1, %%edx), %%mm5) // L5
2353 "movq (%0, %%eax, 2), %%mm6 \n\t" // L6
2354 PAVGB((%1, %%eax, 2), %%mm6) // L6
2355 "movq (%0, %%ecx), %%mm7 \n\t" // L7
2356 PAVGB((%1, %%ecx), %%mm7) // L7
2357 "movq %%mm0, (%1) \n\t" // R0
2358 "movq %%mm1, (%1, %2) \n\t" // R1
2359 "movq %%mm2, (%1, %2, 2) \n\t" // R2
2360 "movq %%mm3, (%1, %%eax) \n\t" // R3
2361 "movq %%mm4, (%1, %2, 4) \n\t" // R4
2362 "movq %%mm5, (%1, %%edx) \n\t" // R5
2363 "movq %%mm6, (%1, %%eax, 2) \n\t" // R6
2364 "movq %%mm7, (%1, %%ecx) \n\t" // R7
2365 "movq %%mm0, (%0) \n\t" // L0
2366 "movq %%mm1, (%0, %2) \n\t" // L1
2367 "movq %%mm2, (%0, %2, 2) \n\t" // L2
2368 "movq %%mm3, (%0, %%eax) \n\t" // L3
2369 "movq %%mm4, (%0, %2, 4) \n\t" // L4
2370 "movq %%mm5, (%0, %%edx) \n\t" // L5
2371 "movq %%mm6, (%0, %%eax, 2) \n\t" // L6
2372 "movq %%mm7, (%0, %%ecx) \n\t" // L7
2376 "cmpl 508(%%edx), %%ecx \n\t"
2379 "leal (%%eax, %2, 2), %%edx \n\t" // 5*stride
2380 "leal (%%edx, %2, 2), %%ecx \n\t" // 7*stride
2381 "movq (%0), %%mm0 \n\t" // L0
2382 "movq (%0, %2), %%mm1 \n\t" // L1
2383 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2384 "movq (%0, %%eax), %%mm3 \n\t" // L3
2385 "movq (%1), %%mm4 \n\t" // R0
2386 "movq (%1, %2), %%mm5 \n\t" // R1
2387 "movq (%1, %2, 2), %%mm6 \n\t" // R2
2388 "movq (%1, %%eax), %%mm7 \n\t" // R3
2397 "movq %%mm0, (%1) \n\t" // R0
2398 "movq %%mm1, (%1, %2) \n\t" // R1
2399 "movq %%mm2, (%1, %2, 2) \n\t" // R2
2400 "movq %%mm3, (%1, %%eax) \n\t" // R3
2401 "movq %%mm0, (%0) \n\t" // L0
2402 "movq %%mm1, (%0, %2) \n\t" // L1
2403 "movq %%mm2, (%0, %2, 2) \n\t" // L2
2404 "movq %%mm3, (%0, %%eax) \n\t" // L3
2406 "movq (%0, %2, 4), %%mm0 \n\t" // L4
2407 "movq (%0, %%edx), %%mm1 \n\t" // L5
2408 "movq (%0, %%eax, 2), %%mm2 \n\t" // L6
2409 "movq (%0, %%ecx), %%mm3 \n\t" // L7
2410 "movq (%1, %2, 4), %%mm4 \n\t" // R4
2411 "movq (%1, %%edx), %%mm5 \n\t" // R5
2412 "movq (%1, %%eax, 2), %%mm6 \n\t" // R6
2413 "movq (%1, %%ecx), %%mm7 \n\t" // R7
2422 "movq %%mm0, (%1, %2, 4) \n\t" // R4
2423 "movq %%mm1, (%1, %%edx) \n\t" // R5
2424 "movq %%mm2, (%1, %%eax, 2) \n\t" // R6
2425 "movq %%mm3, (%1, %%ecx) \n\t" // R7
2426 "movq %%mm0, (%0, %2, 4) \n\t" // L4
2427 "movq %%mm1, (%0, %%edx) \n\t" // L5
2428 "movq %%mm2, (%0, %%eax, 2) \n\t" // L6
2429 "movq %%mm3, (%0, %%ecx) \n\t" // L7
2433 "leal (%%eax, %2, 2), %%edx \n\t" // 5*stride
2434 "leal (%%edx, %2, 2), %%ecx \n\t" // 7*stride
2435 "movq (%0), %%mm0 \n\t" // L0
2436 "movq (%0, %2), %%mm1 \n\t" // L1
2437 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2438 "movq (%0, %%eax), %%mm3 \n\t" // L3
2439 "movq (%1), %%mm4 \n\t" // R0
2440 "movq (%1, %2), %%mm5 \n\t" // R1
2441 "movq (%1, %2, 2), %%mm6 \n\t" // R2
2442 "movq (%1, %%eax), %%mm7 \n\t" // R3
2455 "movq %%mm0, (%1) \n\t" // R0
2456 "movq %%mm1, (%1, %2) \n\t" // R1
2457 "movq %%mm2, (%1, %2, 2) \n\t" // R2
2458 "movq %%mm3, (%1, %%eax) \n\t" // R3
2459 "movq %%mm0, (%0) \n\t" // L0
2460 "movq %%mm1, (%0, %2) \n\t" // L1
2461 "movq %%mm2, (%0, %2, 2) \n\t" // L2
2462 "movq %%mm3, (%0, %%eax) \n\t" // L3
2464 "movq (%0, %2, 4), %%mm0 \n\t" // L4
2465 "movq (%0, %%edx), %%mm1 \n\t" // L5
2466 "movq (%0, %%eax, 2), %%mm2 \n\t" // L6
2467 "movq (%0, %%ecx), %%mm3 \n\t" // L7
2468 "movq (%1, %2, 4), %%mm4 \n\t" // R4
2469 "movq (%1, %%edx), %%mm5 \n\t" // R5
2470 "movq (%1, %%eax, 2), %%mm6 \n\t" // R6
2471 "movq (%1, %%ecx), %%mm7 \n\t" // R7
2484 "movq %%mm0, (%1, %2, 4) \n\t" // R4
2485 "movq %%mm1, (%1, %%edx) \n\t" // R5
2486 "movq %%mm2, (%1, %%eax, 2) \n\t" // R6
2487 "movq %%mm3, (%1, %%ecx) \n\t" // R7
2488 "movq %%mm0, (%0, %2, 4) \n\t" // L4
2489 "movq %%mm1, (%0, %%edx) \n\t" // L5
2490 "movq %%mm2, (%0, %%eax, 2) \n\t" // L6
2491 "movq %%mm3, (%0, %%ecx) \n\t" // L7
2495 :: "r" (src), "r" (tempBlured), "r"(stride), "m" (tempBluredPast)
2496 : "%eax", "%edx", "%ecx", "memory"
2498 //printf("%d\n", test);
2511 int ref= tempBlured[ x + y*stride ];
2512 int cur= src[ x + y*stride ];
2514 // if(x==0 || x==7) d1+= d1>>1;
2515 // if(y==0 || y==7) d1+= d1>>1;
2524 +(*(tempBluredPast-256))
2525 +(*(tempBluredPast-1))+ (*(tempBluredPast+1))
2526 +(*(tempBluredPast+256))
2529 // ((*tempBluredPast)*3 + d + 2)>>2;
2531 //printf("%d %d %d\n", maxNoise[0], maxNoise[1], maxNoise[2]);
2535 64 32 16 8 4 2 1 (1)
2536 64 48 36 27 20 15 11 (33) (approx)
2537 64 56 49 43 37 33 29 (200) (approx)
2548 int ref= tempBlured[ x + y*stride ];
2549 int cur= src[ x + y*stride ];
2550 tempBlured[ x + y*stride ]=
2551 src[ x + y*stride ]=
2563 tempBlured[ x + y*stride ]= src[ x + y*stride ];
2577 int ref= tempBlured[ x + y*stride ];
2578 int cur= src[ x + y*stride ];
2579 tempBlured[ x + y*stride ]=
2580 src[ x + y*stride ]=
2581 (ref*7 + cur + 4)>>3;
2592 int ref= tempBlured[ x + y*stride ];
2593 int cur= src[ x + y*stride ];
2594 tempBlured[ x + y*stride ]=
2595 src[ x + y*stride ]=
2596 (ref*3 + cur + 2)>>2;
2605 static void RENAME(postProcess)(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
2606 QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);
2609 * Copies a block from src to dst and fixes the blacklevel
2610 * levelFix == 0 -> dont touch the brighness & contrast
2614 static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, uint8_t src[], int srcStride,
2615 int levelFix, int64_t *packedOffsetAndScale)
2624 "movq (%%eax), %%mm2 \n\t" // packedYOffset
2625 "movq 8(%%eax), %%mm3 \n\t" // packedYScale
2626 "leal (%2,%4), %%eax \n\t"
2627 "leal (%3,%5), %%edx \n\t"
2628 "pxor %%mm4, %%mm4 \n\t"
2630 #define SCALED_CPY(src1, src2, dst1, dst2) \
2631 "movq " #src1 ", %%mm0 \n\t"\
2632 "movq " #src1 ", %%mm5 \n\t"\
2633 "movq " #src2 ", %%mm1 \n\t"\
2634 "movq " #src2 ", %%mm6 \n\t"\
2635 "punpcklbw %%mm0, %%mm0 \n\t"\
2636 "punpckhbw %%mm5, %%mm5 \n\t"\
2637 "punpcklbw %%mm1, %%mm1 \n\t"\
2638 "punpckhbw %%mm6, %%mm6 \n\t"\
2639 "pmulhuw %%mm3, %%mm0 \n\t"\
2640 "pmulhuw %%mm3, %%mm5 \n\t"\
2641 "pmulhuw %%mm3, %%mm1 \n\t"\
2642 "pmulhuw %%mm3, %%mm6 \n\t"\
2643 "psubw %%mm2, %%mm0 \n\t"\
2644 "psubw %%mm2, %%mm5 \n\t"\
2645 "psubw %%mm2, %%mm1 \n\t"\
2646 "psubw %%mm2, %%mm6 \n\t"\
2647 "packuswb %%mm5, %%mm0 \n\t"\
2648 "packuswb %%mm6, %%mm1 \n\t"\
2649 "movq %%mm0, " #dst1 " \n\t"\
2650 "movq %%mm1, " #dst2 " \n\t"\
2653 #define SCALED_CPY(src1, src2, dst1, dst2) \
2654 "movq " #src1 ", %%mm0 \n\t"\
2655 "movq " #src1 ", %%mm5 \n\t"\
2656 "punpcklbw %%mm4, %%mm0 \n\t"\
2657 "punpckhbw %%mm4, %%mm5 \n\t"\
2658 "psubw %%mm2, %%mm0 \n\t"\
2659 "psubw %%mm2, %%mm5 \n\t"\
2660 "movq " #src2 ", %%mm1 \n\t"\
2661 "psllw $6, %%mm0 \n\t"\
2662 "psllw $6, %%mm5 \n\t"\
2663 "pmulhw %%mm3, %%mm0 \n\t"\
2664 "movq " #src2 ", %%mm6 \n\t"\
2665 "pmulhw %%mm3, %%mm5 \n\t"\
2666 "punpcklbw %%mm4, %%mm1 \n\t"\
2667 "punpckhbw %%mm4, %%mm6 \n\t"\
2668 "psubw %%mm2, %%mm1 \n\t"\
2669 "psubw %%mm2, %%mm6 \n\t"\
2670 "psllw $6, %%mm1 \n\t"\
2671 "psllw $6, %%mm6 \n\t"\
2672 "pmulhw %%mm3, %%mm1 \n\t"\
2673 "pmulhw %%mm3, %%mm6 \n\t"\
2674 "packuswb %%mm5, %%mm0 \n\t"\
2675 "packuswb %%mm6, %%mm1 \n\t"\
2676 "movq %%mm0, " #dst1 " \n\t"\
2677 "movq %%mm1, " #dst2 " \n\t"\
2681 SCALED_CPY((%2) , (%2, %4) , (%3) , (%3, %5))
2682 SCALED_CPY((%2, %4, 2), (%%eax, %4, 2), (%3, %5, 2), (%%edx, %5, 2))
2683 SCALED_CPY((%2, %4, 4), (%%eax, %4, 4), (%3, %5, 4), (%%edx, %5, 4))
2684 "leal (%%eax,%4,4), %%eax \n\t"
2685 "leal (%%edx,%5,4), %%edx \n\t"
2686 SCALED_CPY((%%eax, %4), (%%eax, %4, 2), (%%edx, %5), (%%edx, %5, 2))
2689 : "=&a" (packedOffsetAndScale)
2690 : "0" (packedOffsetAndScale),
2699 memcpy( &(dst[dstStride*i]),
2700 &(src[srcStride*i]), BLOCK_SIZE);
2707 "leal (%0,%2), %%eax \n\t"
2708 "leal (%1,%3), %%edx \n\t"
2710 #define SIMPLE_CPY(src1, src2, dst1, dst2) \
2711 "movq " #src1 ", %%mm0 \n\t"\
2712 "movq " #src2 ", %%mm1 \n\t"\
2713 "movq %%mm0, " #dst1 " \n\t"\
2714 "movq %%mm1, " #dst2 " \n\t"\
2716 SIMPLE_CPY((%0) , (%0, %2) , (%1) , (%1, %3))
2717 SIMPLE_CPY((%0, %2, 2), (%%eax, %2, 2), (%1, %3, 2), (%%edx, %3, 2))
2718 SIMPLE_CPY((%0, %2, 4), (%%eax, %2, 4), (%1, %3, 4), (%%edx, %3, 4))
2719 "leal (%%eax,%2,4), %%eax \n\t"
2720 "leal (%%edx,%3,4), %%edx \n\t"
2721 SIMPLE_CPY((%%eax, %2), (%%eax, %2, 2), (%%edx, %3), (%%edx, %3, 2))
2731 memcpy( &(dst[dstStride*i]),
2732 &(src[srcStride*i]), BLOCK_SIZE);
2738 * Duplicates the given 8 src pixels ? times upward
2740 static inline void RENAME(duplicate)(uint8_t src[], int stride)
2744 "movq (%0), %%mm0 \n\t"
2746 "movq %%mm0, (%0) \n\t"
2747 "movq %%mm0, (%0, %1) \n\t"
2748 "movq %%mm0, (%0, %1, 2) \n\t"
2764 * Filters array of bytes (Y or U or V values)
2766 static void RENAME(postProcess)(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
2767 QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)
2769 PPContext __attribute__((aligned(8))) c= *c2; //copy to stack for faster access
2771 #ifdef COMPILE_TIME_MODE
2772 const int mode= COMPILE_TIME_MODE;
2774 const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;
2776 int black=0, white=255; // blackest black and whitest white in the picture
2777 int QPCorrecture= 256*256;
2784 const int qpHShift= isColor ? 4-c.hChromaSubSample : 4;
2785 const int qpVShift= isColor ? 4-c.vChromaSubSample : 4;
2788 uint64_t * const yHistogram= c.yHistogram;
2789 uint8_t * const tempSrc= c.tempSrc;
2790 uint8_t * const tempDst= c.tempDst;
2791 const int mbWidth= isColor ? (width+7)>>3 : (width+15)>>4;
2794 for(i=0; i<32; i++){
2795 int offset= ((i*c.ppMode.baseDcDiff)>>8) + 1;
2796 int threshold= offset*2 + 1;
2797 c.mmxDcOffset[i]= 0x7F - offset;
2798 c.mmxDcThreshold[i]= 0x7F - threshold;
2799 c.mmxDcOffset[i]*= 0x0101010101010101LL;
2800 c.mmxDcThreshold[i]*= 0x0101010101010101LL;
2804 if(mode & CUBIC_IPOL_DEINT_FILTER) copyAhead=16;
2805 else if( (mode & INTERPOLATION_LINEAR_BLEND_DEINT_FILTER)
2806 || (mode & FFMPEG_DEINT_FILTER)
2807 || (mode & LOWPASS5_DEINT_FILTER)) copyAhead=14;
2808 else if( (mode & V_DEBLOCK)
2809 || (mode & INTERPOLATION_LINEAR_IPOL_DEINT_FILTER)
2810 || (mode & MEDIAN_DEINT_FILTER)) copyAhead=13;
2811 else if(mode & V_X1_FILTER) copyAhead=11;
2812 // else if(mode & V_RK1_FILTER) copyAhead=10;
2813 else if(mode & DERING) copyAhead=9;
2822 uint64_t maxClipped;
2827 // first frame is fscked so we ignore it
2828 if(c.frameNum == 1) yHistogram[0]= width*height/64*15/256;
2830 for(i=0; i<256; i++)
2832 sum+= yHistogram[i];
2833 // printf("%d ", yHistogram[i]);
2837 /* we allways get a completly black picture first */
2838 maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);
2841 for(black=255; black>0; black--)
2843 if(clipped < maxClipped) break;
2844 clipped-= yHistogram[black];
2848 for(white=0; white<256; white++)
2850 if(clipped < maxClipped) break;
2851 clipped-= yHistogram[white];
2854 scale= (double)(c.ppMode.maxAllowedY - c.ppMode.minAllowedY) / (double)(white-black);
2857 c.packedYScale= (uint16_t)(scale*256.0 + 0.5);
2858 c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF;
2860 c.packedYScale= (uint16_t)(scale*1024.0 + 0.5);
2861 c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF;
2864 c.packedYOffset|= c.packedYOffset<<32;
2865 c.packedYOffset|= c.packedYOffset<<16;
2867 c.packedYScale|= c.packedYScale<<32;
2868 c.packedYScale|= c.packedYScale<<16;
2870 if(mode & LEVEL_FIX) QPCorrecture= (int)(scale*256*256 + 0.5);
2871 else QPCorrecture= 256*256;
2875 c.packedYScale= 0x0100010001000100LL;
2877 QPCorrecture= 256*256;
2880 /* copy & deinterlace first row of blocks */
2883 uint8_t *srcBlock= &(src[y*srcStride]);
2884 uint8_t *dstBlock= tempDst + dstStride;
2886 // From this point on it is guranteed that we can read and write 16 lines downward
2887 // finish 1 block before the next otherwise we´ll might have a problem
2888 // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
2889 for(x=0; x<width; x+=BLOCK_SIZE)
2894 prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
2895 prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
2896 prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
2897 prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
2901 "movl %4, %%eax \n\t"
2902 "shrl $2, %%eax \n\t"
2903 "andl $6, %%eax \n\t"
2904 "addl %5, %%eax \n\t"
2905 "movl %%eax, %%edx \n\t"
2906 "imul %1, %%eax \n\t"
2907 "imul %3, %%edx \n\t"
2908 "prefetchnta 32(%%eax, %0) \n\t"
2909 "prefetcht0 32(%%edx, %2) \n\t"
2910 "addl %1, %%eax \n\t"
2911 "addl %3, %%edx \n\t"
2912 "prefetchnta 32(%%eax, %0) \n\t"
2913 "prefetcht0 32(%%edx, %2) \n\t"
2914 :: "r" (srcBlock), "r" (srcStride), "r" (dstBlock), "r" (dstStride),
2915 "m" (x), "m" (copyAhead)
2919 #elif defined(HAVE_3DNOW)
2920 //FIXME check if this is faster on an 3dnow chip or if its faster without the prefetch or ...
2921 /* prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
2922 prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
2923 prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
2924 prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
2928 RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,
2929 srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
2931 RENAME(duplicate)(dstBlock + dstStride*8, dstStride);
2933 if(mode & INTERPOLATION_LINEAR_IPOL_DEINT_FILTER)
2934 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
2935 else if(mode & INTERPOLATION_LINEAR_BLEND_DEINT_FILTER)
2936 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride);
2937 else if(mode & MEDIAN_DEINT_FILTER)
2938 RENAME(deInterlaceMedian)(dstBlock, dstStride);
2939 else if(mode & CUBIC_IPOL_DEINT_FILTER)
2940 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
2941 else if(mode & FFMPEG_DEINT_FILTER)
2942 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
2943 else if(mode & LOWPASS5_DEINT_FILTER)
2944 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
2945 /* else if(mode & CUBIC_BLEND_DEINT_FILTER)
2946 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
2951 if(width==dstStride)
2952 memcpy(dst, tempDst + 9*dstStride, copyAhead*dstStride);
2956 for(i=0; i<copyAhead; i++)
2958 memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);
2964 for(y=0; y<height; y+=BLOCK_SIZE)
2966 //1% speedup if these are here instead of the inner loop
2967 uint8_t *srcBlock= &(src[y*srcStride]);
2968 uint8_t *dstBlock= &(dst[y*dstStride]);
2970 uint8_t *tempBlock1= c.tempBlocks;
2971 uint8_t *tempBlock2= c.tempBlocks + 8;
2973 int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];
2974 int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*QPStride];
2976 /* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards
2977 if not than use a temporary buffer */
2981 /* copy from line (copyAhead) to (copyAhead+7) of src, these will be copied with
2982 blockcopy to dst later */
2983 memcpy(tempSrc + srcStride*copyAhead, srcBlock + srcStride*copyAhead,
2984 srcStride*MAX(height-y-copyAhead, 0) );
2986 /* duplicate last line of src to fill the void upto line (copyAhead+7) */
2987 for(i=MAX(height-y, 8); i<copyAhead+8; i++)
2988 memcpy(tempSrc + srcStride*i, src + srcStride*(height-1), srcStride);
2990 /* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/
2991 memcpy(tempDst, dstBlock - dstStride, dstStride*MIN(height-y+1, copyAhead+1) );
2993 /* duplicate last line of dst to fill the void upto line (copyAhead) */
2994 for(i=height-y+1; i<=copyAhead; i++)
2995 memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), dstStride);
2997 dstBlock= tempDst + dstStride;
3002 // From this point on it is guranteed that we can read and write 16 lines downward
3003 // finish 1 block before the next otherwise we´ll might have a problem
3004 // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
3005 for(x=0; x<width; x+=BLOCK_SIZE)
3007 const int stride= dstStride;
3013 QP= QPptr[x>>qpHShift];
3014 c.nonBQP= nonBQPptr[x>>qpHShift];
3019 QP= (QP* QPCorrecture + 256*128)>>16;
3020 c.nonBQP= nonBQPptr[x>>4];
3021 c.nonBQP= (c.nonBQP* QPCorrecture + 256*128)>>16;
3022 yHistogram[ srcBlock[srcStride*12 + 4] ]++;
3027 "movd %1, %%mm7 \n\t"
3028 "packuswb %%mm7, %%mm7 \n\t" // 0, 0, 0, QP, 0, 0, 0, QP
3029 "packuswb %%mm7, %%mm7 \n\t" // 0,QP, 0, QP, 0,QP, 0, QP
3030 "packuswb %%mm7, %%mm7 \n\t" // QP,..., QP
3031 "movq %%mm7, %0 \n\t"
3040 prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
3041 prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
3042 prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
3043 prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
3047 "movl %4, %%eax \n\t"
3048 "shrl $2, %%eax \n\t"
3049 "andl $6, %%eax \n\t"
3050 "addl %5, %%eax \n\t"
3051 "movl %%eax, %%edx \n\t"
3052 "imul %1, %%eax \n\t"
3053 "imul %3, %%edx \n\t"
3054 "prefetchnta 32(%%eax, %0) \n\t"
3055 "prefetcht0 32(%%edx, %2) \n\t"
3056 "addl %1, %%eax \n\t"
3057 "addl %3, %%edx \n\t"
3058 "prefetchnta 32(%%eax, %0) \n\t"
3059 "prefetcht0 32(%%edx, %2) \n\t"
3060 :: "r" (srcBlock), "r" (srcStride), "r" (dstBlock), "r" (dstStride),
3061 "m" (x), "m" (copyAhead)
3065 #elif defined(HAVE_3DNOW)
3066 //FIXME check if this is faster on an 3dnow chip or if its faster without the prefetch or ...
3067 /* prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
3068 prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
3069 prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
3070 prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
3074 RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,
3075 srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
3077 if(mode & INTERPOLATION_LINEAR_IPOL_DEINT_FILTER)
3078 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
3079 else if(mode & INTERPOLATION_LINEAR_BLEND_DEINT_FILTER)
3080 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride);
3081 else if(mode & MEDIAN_DEINT_FILTER)
3082 RENAME(deInterlaceMedian)(dstBlock, dstStride);
3083 else if(mode & CUBIC_IPOL_DEINT_FILTER)
3084 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
3085 else if(mode & FFMPEG_DEINT_FILTER)
3086 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
3087 else if(mode & LOWPASS5_DEINT_FILTER)
3088 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
3089 /* else if(mode & CUBIC_BLEND_DEINT_FILTER)
3090 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3093 /* only deblock if we have 2 blocks */
3096 if(mode & V_X1_FILTER)
3097 RENAME(vertX1Filter)(dstBlock, stride, &c);
3098 else if(mode & V_DEBLOCK)
3100 const int t= RENAME(vertClassify)(dstBlock, stride, &c);
3103 RENAME(doVertLowPass)(dstBlock, stride, &c);
3105 RENAME(doVertDefFilter)(dstBlock, stride, &c);
3110 RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);
3112 /* check if we have a previous block to deblock it with dstBlock */
3116 if(mode & H_X1_FILTER)
3117 RENAME(vertX1Filter)(tempBlock1, 16, &c);
3118 else if(mode & H_DEBLOCK)
3121 const int t= RENAME(vertClassify)(tempBlock1, 16, &c);
3122 //STOP_TIMER("dc & minmax")
3124 RENAME(doVertLowPass)(tempBlock1, 16, &c);
3126 RENAME(doVertDefFilter)(tempBlock1, 16, &c);
3129 RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16);
3132 if(mode & H_X1_FILTER)
3133 horizX1Filter(dstBlock-4, stride, QP);
3134 else if(mode & H_DEBLOCK)
3136 if( isHorizDC(dstBlock-4, stride, &c))
3138 if(isHorizMinMaxOk(dstBlock-4, stride, QP))
3139 doHorizLowPass(dstBlock-4, stride, QP);
3142 doHorizDefFilter(dstBlock-4, stride, QP);
3147 //FIXME filter first line
3148 if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);
3151 if(mode & TEMP_NOISE_FILTER)
3153 RENAME(tempNoiseReducer)(dstBlock-8, stride,
3154 c.tempBlured[isColor] + y*dstStride + x,
3155 c.tempBluredPast[isColor] + (y>>3)*256 + (x>>3),
3156 c.ppMode.maxTmpNoise);
3164 tmpXchg= tempBlock1;
3165 tempBlock1= tempBlock2;
3166 tempBlock2 = tmpXchg;
3172 if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, &c);
3175 if((mode & TEMP_NOISE_FILTER))
3177 RENAME(tempNoiseReducer)(dstBlock-8, dstStride,
3178 c.tempBlured[isColor] + y*dstStride + x,
3179 c.tempBluredPast[isColor] + (y>>3)*256 + (x>>3),
3180 c.ppMode.maxTmpNoise);
3183 /* did we use a tmp buffer for the last lines*/
3186 uint8_t *dstBlock= &(dst[y*dstStride]);
3187 if(width==dstStride)
3188 memcpy(dstBlock, tempDst + dstStride, dstStride*(height-y));
3192 for(i=0; i<height-y; i++)
3194 memcpy(dstBlock + i*dstStride, tempDst + (i+1)*dstStride, width);
3199 for(x=0; x<width; x+=32)
3202 i+= + dstBlock[x + 7*dstStride] + dstBlock[x + 8*dstStride]
3203 + dstBlock[x + 9*dstStride] + dstBlock[x +10*dstStride]
3204 + dstBlock[x +11*dstStride] + dstBlock[x +12*dstStride];
3205 // + dstBlock[x +13*dstStride]
3206 // + dstBlock[x +14*dstStride] + dstBlock[x +15*dstStride];
3210 asm volatile("femms");
3211 #elif defined (HAVE_MMX)
3212 asm volatile("emms");
3215 #ifdef DEBUG_BRIGHTNESS
3220 for(i=0; i<256; i++)
3221 if(yHistogram[i] > max) max=yHistogram[i];
3223 for(i=1; i<256; i++)
3226 int start=yHistogram[i-1]/(max/256+1);
3227 int end=yHistogram[i]/(max/256+1);
3228 int inc= end > start ? 1 : -1;
3229 for(x=start; x!=end+inc; x+=inc)
3230 dst[ i*dstStride + x]+=128;
3233 for(i=0; i<100; i+=2)
3235 dst[ (white)*dstStride + i]+=128;
3236 dst[ (black)*dstStride + i]+=128;
3242 *c2= c; //copy local context back