2 * Copyright (C) 2001-2002 Michael Niedermayer (michaelni@gmx.at)
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 * @file postprocess_template.c
23 * mmx/mmx2/3dnow postprocess code.
29 # define ALIGN_MASK "$0xFFFFFFFFFFFFFFF8"
31 # define ALIGN_MASK "$0xFFFFFFF8"
40 #define REAL_PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
41 #elif defined (HAVE_3DNOW)
42 #define REAL_PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
44 #define PAVGB(a,b) REAL_PAVGB(a,b)
47 #define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
48 #elif defined (HAVE_MMX)
49 #define PMINUB(b,a,t) \
50 "movq " #a ", " #t " \n\t"\
51 "psubusb " #b ", " #t " \n\t"\
52 "psubb " #t ", " #a " \n\t"
56 #define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
57 #elif defined (HAVE_MMX)
59 "psubusb " #a ", " #b " \n\t"\
60 "paddb " #a ", " #b " \n\t"
63 //FIXME? |255-0| = 1 (should not be a problem ...)
66 * Check if the middle 8x8 Block in the given 8x16 block is flat
68 static inline int RENAME(vertClassify)(uint8_t src[], int stride, PPContext *c){
70 src+= stride*4; // src points to begin of the 8x8 Block
74 : : "m" (c->mmxDcOffset[c->nonBQP]), "m" (c->mmxDcThreshold[c->nonBQP])
78 "lea (%2, %3), %%"REG_a" \n\t"
79 // 0 1 2 3 4 5 6 7 8 9
80 // %1 eax eax+%2 eax+2%2 %1+4%2 ecx ecx+%2 ecx+2%2 %1+8%2 ecx+4%2
82 "movq (%2), %%mm0 \n\t"
83 "movq (%%"REG_a"), %%mm1 \n\t"
84 "movq %%mm0, %%mm3 \n\t"
85 "movq %%mm0, %%mm4 \n\t"
87 PMINUB(%%mm1, %%mm3, %%mm5)
88 "psubb %%mm1, %%mm0 \n\t" // mm0 = differnece
89 "paddb %%mm7, %%mm0 \n\t"
90 "pcmpgtb %%mm6, %%mm0 \n\t"
92 "movq (%%"REG_a",%3), %%mm2 \n\t"
94 PMINUB(%%mm2, %%mm3, %%mm5)
95 "psubb %%mm2, %%mm1 \n\t"
96 "paddb %%mm7, %%mm1 \n\t"
97 "pcmpgtb %%mm6, %%mm1 \n\t"
98 "paddb %%mm1, %%mm0 \n\t"
100 "movq (%%"REG_a", %3, 2), %%mm1 \n\t"
102 PMINUB(%%mm1, %%mm3, %%mm5)
103 "psubb %%mm1, %%mm2 \n\t"
104 "paddb %%mm7, %%mm2 \n\t"
105 "pcmpgtb %%mm6, %%mm2 \n\t"
106 "paddb %%mm2, %%mm0 \n\t"
108 "lea (%%"REG_a", %3, 4), %%"REG_a" \n\t"
110 "movq (%2, %3, 4), %%mm2 \n\t"
112 PMINUB(%%mm2, %%mm3, %%mm5)
113 "psubb %%mm2, %%mm1 \n\t"
114 "paddb %%mm7, %%mm1 \n\t"
115 "pcmpgtb %%mm6, %%mm1 \n\t"
116 "paddb %%mm1, %%mm0 \n\t"
118 "movq (%%"REG_a"), %%mm1 \n\t"
120 PMINUB(%%mm1, %%mm3, %%mm5)
121 "psubb %%mm1, %%mm2 \n\t"
122 "paddb %%mm7, %%mm2 \n\t"
123 "pcmpgtb %%mm6, %%mm2 \n\t"
124 "paddb %%mm2, %%mm0 \n\t"
126 "movq (%%"REG_a", %3), %%mm2 \n\t"
128 PMINUB(%%mm2, %%mm3, %%mm5)
129 "psubb %%mm2, %%mm1 \n\t"
130 "paddb %%mm7, %%mm1 \n\t"
131 "pcmpgtb %%mm6, %%mm1 \n\t"
132 "paddb %%mm1, %%mm0 \n\t"
134 "movq (%%"REG_a", %3, 2), %%mm1 \n\t"
136 PMINUB(%%mm1, %%mm3, %%mm5)
137 "psubb %%mm1, %%mm2 \n\t"
138 "paddb %%mm7, %%mm2 \n\t"
139 "pcmpgtb %%mm6, %%mm2 \n\t"
140 "paddb %%mm2, %%mm0 \n\t"
141 "psubusb %%mm3, %%mm4 \n\t"
145 "pxor %%mm7, %%mm7 \n\t"
146 "psadbw %%mm7, %%mm0 \n\t"
148 "movq %%mm0, %%mm1 \n\t"
149 "psrlw $8, %%mm0 \n\t"
150 "paddb %%mm1, %%mm0 \n\t"
151 "movq %%mm0, %%mm1 \n\t"
152 "psrlq $16, %%mm0 \n\t"
153 "paddb %%mm1, %%mm0 \n\t"
154 "movq %%mm0, %%mm1 \n\t"
155 "psrlq $32, %%mm0 \n\t"
156 "paddb %%mm1, %%mm0 \n\t"
158 "movq %4, %%mm7 \n\t" // QP,..., QP
159 "paddusb %%mm7, %%mm7 \n\t" // 2QP ... 2QP
160 "psubusb %%mm7, %%mm4 \n\t" // Diff <= 2QP -> 0
161 "packssdw %%mm4, %%mm4 \n\t"
162 "movd %%mm0, %0 \n\t"
163 "movd %%mm4, %1 \n\t"
165 : "=r" (numEq), "=r" (dcOk)
166 : "r" (src), "r" ((long)stride), "m" (c->pQPb)
170 numEq= (-numEq) &0xFF;
171 if(numEq > c->ppMode.flatnessThreshold){
181 * Do a vertical low pass filter on the 8x16 block (only write to the 8x8 block in the middle)
182 * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16
185 static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
187 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
189 asm volatile( //"movv %0 %1 %2\n\t"
190 "movq %2, %%mm0 \n\t" // QP,..., QP
191 "pxor %%mm4, %%mm4 \n\t"
193 "movq (%0), %%mm6 \n\t"
194 "movq (%0, %1), %%mm5 \n\t"
195 "movq %%mm5, %%mm1 \n\t"
196 "movq %%mm6, %%mm2 \n\t"
197 "psubusb %%mm6, %%mm5 \n\t"
198 "psubusb %%mm1, %%mm2 \n\t"
199 "por %%mm5, %%mm2 \n\t" // ABS Diff of lines
200 "psubusb %%mm0, %%mm2 \n\t" // diff <= QP -> 0
201 "pcmpeqb %%mm4, %%mm2 \n\t" // diff <= QP -> FF
203 "pand %%mm2, %%mm6 \n\t"
204 "pandn %%mm1, %%mm2 \n\t"
205 "por %%mm2, %%mm6 \n\t"// First Line to Filter
207 "movq (%0, %1, 8), %%mm5 \n\t"
208 "lea (%0, %1, 4), %%"REG_a" \n\t"
209 "lea (%0, %1, 8), %%"REG_c" \n\t"
210 "sub %1, %%"REG_c" \n\t"
211 "add %1, %0 \n\t" // %0 points to line 1 not 0
212 "movq (%0, %1, 8), %%mm7 \n\t"
213 "movq %%mm5, %%mm1 \n\t"
214 "movq %%mm7, %%mm2 \n\t"
215 "psubusb %%mm7, %%mm5 \n\t"
216 "psubusb %%mm1, %%mm2 \n\t"
217 "por %%mm5, %%mm2 \n\t" // ABS Diff of lines
218 "psubusb %%mm0, %%mm2 \n\t" // diff <= QP -> 0
219 "pcmpeqb %%mm4, %%mm2 \n\t" // diff <= QP -> FF
221 "pand %%mm2, %%mm7 \n\t"
222 "pandn %%mm1, %%mm2 \n\t"
223 "por %%mm2, %%mm7 \n\t" // First Line to Filter
227 // %0 %0+%1 %0+2%1 eax %0+4%1 eax+2%1 ecx eax+4%1
232 "movq (%0, %1), %%mm0 \n\t" // 1
233 "movq %%mm0, %%mm1 \n\t" // 1
234 PAVGB(%%mm6, %%mm0) //1 1 /2
235 PAVGB(%%mm6, %%mm0) //3 1 /4
237 "movq (%0, %1, 4), %%mm2 \n\t" // 1
238 "movq %%mm2, %%mm5 \n\t" // 1
239 PAVGB((%%REGa), %%mm2) // 11 /2
240 PAVGB((%0, %1, 2), %%mm2) // 211 /4
241 "movq %%mm2, %%mm3 \n\t" // 211 /4
242 "movq (%0), %%mm4 \n\t" // 1
243 PAVGB(%%mm4, %%mm3) // 4 211 /8
244 PAVGB(%%mm0, %%mm3) //642211 /16
245 "movq %%mm3, (%0) \n\t" // X
246 // mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
247 "movq %%mm1, %%mm0 \n\t" // 1
248 PAVGB(%%mm6, %%mm0) //1 1 /2
249 "movq %%mm4, %%mm3 \n\t" // 1
250 PAVGB((%0,%1,2), %%mm3) // 1 1 /2
251 PAVGB((%%REGa,%1,2), %%mm5) // 11 /2
252 PAVGB((%%REGa), %%mm5) // 211 /4
253 PAVGB(%%mm5, %%mm3) // 2 2211 /8
254 PAVGB(%%mm0, %%mm3) //4242211 /16
255 "movq %%mm3, (%0,%1) \n\t" // X
256 // mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
257 PAVGB(%%mm4, %%mm6) //11 /2
258 "movq (%%"REG_c"), %%mm0 \n\t" // 1
259 PAVGB((%%REGa, %1, 2), %%mm0) // 11/2
260 "movq %%mm0, %%mm3 \n\t" // 11/2
261 PAVGB(%%mm1, %%mm0) // 2 11/4
262 PAVGB(%%mm6, %%mm0) //222 11/8
263 PAVGB(%%mm2, %%mm0) //22242211/16
264 "movq (%0, %1, 2), %%mm2 \n\t" // 1
265 "movq %%mm0, (%0, %1, 2) \n\t" // X
266 // mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
267 "movq (%%"REG_a", %1, 4), %%mm0 \n\t" // 1
268 PAVGB((%%REGc), %%mm0) // 11 /2
269 PAVGB(%%mm0, %%mm6) //11 11 /4
270 PAVGB(%%mm1, %%mm4) // 11 /2
271 PAVGB(%%mm2, %%mm1) // 11 /2
272 PAVGB(%%mm1, %%mm6) //1122 11 /8
273 PAVGB(%%mm5, %%mm6) //112242211 /16
274 "movq (%%"REG_a"), %%mm5 \n\t" // 1
275 "movq %%mm6, (%%"REG_a") \n\t" // X
276 // mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
277 "movq (%%"REG_a", %1, 4), %%mm6 \n\t" // 1
278 PAVGB(%%mm7, %%mm6) // 11 /2
279 PAVGB(%%mm4, %%mm6) // 11 11 /4
280 PAVGB(%%mm3, %%mm6) // 11 2211 /8
281 PAVGB(%%mm5, %%mm2) // 11 /2
282 "movq (%0, %1, 4), %%mm4 \n\t" // 1
283 PAVGB(%%mm4, %%mm2) // 112 /4
284 PAVGB(%%mm2, %%mm6) // 112242211 /16
285 "movq %%mm6, (%0, %1, 4) \n\t" // X
286 // mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
287 PAVGB(%%mm7, %%mm1) // 11 2 /4
288 PAVGB(%%mm4, %%mm5) // 11 /2
289 PAVGB(%%mm5, %%mm0) // 11 11 /4
290 "movq (%%"REG_a", %1, 2), %%mm6 \n\t" // 1
291 PAVGB(%%mm6, %%mm1) // 11 4 2 /8
292 PAVGB(%%mm0, %%mm1) // 11224222 /16
293 "movq %%mm1, (%%"REG_a", %1, 2) \n\t" // X
294 // mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
295 PAVGB((%%REGc), %%mm2) // 112 4 /8
296 "movq (%%"REG_a", %1, 4), %%mm0 \n\t" // 1
297 PAVGB(%%mm0, %%mm6) // 1 1 /2
298 PAVGB(%%mm7, %%mm6) // 1 12 /4
299 PAVGB(%%mm2, %%mm6) // 1122424 /4
300 "movq %%mm6, (%%"REG_c") \n\t" // X
301 // mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
302 PAVGB(%%mm7, %%mm5) // 11 2 /4
303 PAVGB(%%mm7, %%mm5) // 11 6 /8
305 PAVGB(%%mm3, %%mm0) // 112 /4
306 PAVGB(%%mm0, %%mm5) // 112246 /16
307 "movq %%mm5, (%%"REG_a", %1, 4) \n\t" // X
311 : "r" (src), "r" ((long)stride), "m" (c->pQPb)
314 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
315 const int l1= stride;
316 const int l2= stride + l1;
317 const int l3= stride + l2;
318 const int l4= stride + l3;
319 const int l5= stride + l4;
320 const int l6= stride + l5;
321 const int l7= stride + l6;
322 const int l8= stride + l7;
323 const int l9= stride + l8;
326 for(x=0; x<BLOCK_SIZE; x++)
328 const int first= FFABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];
329 const int last= FFABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];
332 sums[0] = 4*first + src[l1] + src[l2] + src[l3] + 4;
333 sums[1] = sums[0] - first + src[l4];
334 sums[2] = sums[1] - first + src[l5];
335 sums[3] = sums[2] - first + src[l6];
336 sums[4] = sums[3] - first + src[l7];
337 sums[5] = sums[4] - src[l1] + src[l8];
338 sums[6] = sums[5] - src[l2] + last;
339 sums[7] = sums[6] - src[l3] + last;
340 sums[8] = sums[7] - src[l4] + last;
341 sums[9] = sums[8] - src[l5] + last;
343 src[l1]= (sums[0] + sums[2] + 2*src[l1])>>4;
344 src[l2]= (sums[1] + sums[3] + 2*src[l2])>>4;
345 src[l3]= (sums[2] + sums[4] + 2*src[l3])>>4;
346 src[l4]= (sums[3] + sums[5] + 2*src[l4])>>4;
347 src[l5]= (sums[4] + sums[6] + 2*src[l5])>>4;
348 src[l6]= (sums[5] + sums[7] + 2*src[l6])>>4;
349 src[l7]= (sums[6] + sums[8] + 2*src[l7])>>4;
350 src[l8]= (sums[7] + sums[9] + 2*src[l8])>>4;
354 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
356 #endif //HAVE_ALTIVEC
360 * Experimental implementation of the filter (Algorithm 1) described in a paper from Ramkishor & Karandikar
361 * values are correctly clipped (MMX2)
362 * values are wraparound (C)
363 * Conclusion: It is fast, but introduces ugly horizontal patterns
364 * if there is a continuous gradient.
371 static inline void RENAME(vertRK1Filter)(uint8_t *src, int stride, int QP)
373 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
377 "pxor %%mm7, %%mm7 \n\t" // 0
378 "movq "MANGLE(b80)", %%mm6 \n\t" // MIN_SIGNED_BYTE
379 "leal (%0, %1), %%"REG_a" \n\t"
380 "leal (%%"REG_a", %1, 4), %%"REG_c" \n\t"
381 // 0 1 2 3 4 5 6 7 8 9
382 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1 %0+8%1 ecx+4%1
383 "movq "MANGLE(pQPb)", %%mm0 \n\t" // QP,..., QP
384 "movq %%mm0, %%mm1 \n\t" // QP,..., QP
385 "paddusb "MANGLE(b02)", %%mm0 \n\t"
386 "psrlw $2, %%mm0 \n\t"
387 "pand "MANGLE(b3F)", %%mm0 \n\t" // QP/4,..., QP/4
388 "paddusb %%mm1, %%mm0 \n\t" // QP*1.25 ...
389 "movq (%0, %1, 4), %%mm2 \n\t" // line 4
390 "movq (%%"REG_c"), %%mm3 \n\t" // line 5
391 "movq %%mm2, %%mm4 \n\t" // line 4
392 "pcmpeqb %%mm5, %%mm5 \n\t" // -1
393 "pxor %%mm2, %%mm5 \n\t" // -line 4 - 1
395 "paddb %%mm6, %%mm5 \n\t" // (l5-l4)/2
396 "psubusb %%mm3, %%mm4 \n\t"
397 "psubusb %%mm2, %%mm3 \n\t"
398 "por %%mm3, %%mm4 \n\t" // |l4 - l5|
399 "psubusb %%mm0, %%mm4 \n\t"
400 "pcmpeqb %%mm7, %%mm4 \n\t"
401 "pand %%mm4, %%mm5 \n\t" // d/2
403 // "paddb %%mm6, %%mm2 \n\t" // line 4 + 0x80
404 "paddb %%mm5, %%mm2 \n\t"
405 // "psubb %%mm6, %%mm2 \n\t"
406 "movq %%mm2, (%0,%1, 4) \n\t"
408 "movq (%%"REG_c"), %%mm2 \n\t"
409 // "paddb %%mm6, %%mm2 \n\t" // line 5 + 0x80
410 "psubb %%mm5, %%mm2 \n\t"
411 // "psubb %%mm6, %%mm2 \n\t"
412 "movq %%mm2, (%%"REG_c") \n\t"
414 "paddb %%mm6, %%mm5 \n\t"
415 "psrlw $2, %%mm5 \n\t"
416 "pand "MANGLE(b3F)", %%mm5 \n\t"
417 "psubb "MANGLE(b20)", %%mm5 \n\t" // (l5-l4)/8
419 "movq (%%"REG_a", %1, 2), %%mm2 \n\t"
420 "paddb %%mm6, %%mm2 \n\t" // line 3 + 0x80
421 "paddsb %%mm5, %%mm2 \n\t"
422 "psubb %%mm6, %%mm2 \n\t"
423 "movq %%mm2, (%%"REG_a", %1, 2) \n\t"
425 "movq (%%"REG_c", %1), %%mm2 \n\t"
426 "paddb %%mm6, %%mm2 \n\t" // line 6 + 0x80
427 "psubsb %%mm5, %%mm2 \n\t"
428 "psubb %%mm6, %%mm2 \n\t"
429 "movq %%mm2, (%%"REG_c", %1) \n\t"
432 : "r" (src), "r" ((long)stride)
435 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
436 const int l1= stride;
437 const int l2= stride + l1;
438 const int l3= stride + l2;
439 const int l4= stride + l3;
440 const int l5= stride + l4;
441 const int l6= stride + l5;
442 // const int l7= stride + l6;
443 // const int l8= stride + l7;
444 // const int l9= stride + l8;
446 const int QP15= QP + (QP>>2);
448 for(x=0; x<BLOCK_SIZE; x++)
450 const int v = (src[x+l5] - src[x+l4]);
461 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
466 * Experimental Filter 1
467 * will not damage linear gradients
468 * Flat blocks should look like they were passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
469 * can only smooth blocks at the expected locations (it cannot smooth them if they did move)
470 * MMX2 version does correct clipping C version does not
472 static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
474 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
478 "pxor %%mm7, %%mm7 \n\t" // 0
479 "lea (%0, %1), %%"REG_a" \n\t"
480 "lea (%%"REG_a", %1, 4), %%"REG_c" \n\t"
481 // 0 1 2 3 4 5 6 7 8 9
482 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1 %0+8%1 ecx+4%1
483 "movq (%%"REG_a", %1, 2), %%mm0 \n\t" // line 3
484 "movq (%0, %1, 4), %%mm1 \n\t" // line 4
485 "movq %%mm1, %%mm2 \n\t" // line 4
486 "psubusb %%mm0, %%mm1 \n\t"
487 "psubusb %%mm2, %%mm0 \n\t"
488 "por %%mm1, %%mm0 \n\t" // |l2 - l3|
489 "movq (%%"REG_c"), %%mm3 \n\t" // line 5
490 "movq (%%"REG_c", %1), %%mm4 \n\t" // line 6
491 "movq %%mm3, %%mm5 \n\t" // line 5
492 "psubusb %%mm4, %%mm3 \n\t"
493 "psubusb %%mm5, %%mm4 \n\t"
494 "por %%mm4, %%mm3 \n\t" // |l5 - l6|
495 PAVGB(%%mm3, %%mm0) // (|l2 - l3| + |l5 - l6|)/2
496 "movq %%mm2, %%mm1 \n\t" // line 4
497 "psubusb %%mm5, %%mm2 \n\t"
498 "movq %%mm2, %%mm4 \n\t"
499 "pcmpeqb %%mm7, %%mm2 \n\t" // (l4 - l5) <= 0 ? -1 : 0
500 "psubusb %%mm1, %%mm5 \n\t"
501 "por %%mm5, %%mm4 \n\t" // |l4 - l5|
502 "psubusb %%mm0, %%mm4 \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
503 "movq %%mm4, %%mm3 \n\t" // d
504 "movq %2, %%mm0 \n\t"
505 "paddusb %%mm0, %%mm0 \n\t"
506 "psubusb %%mm0, %%mm4 \n\t"
507 "pcmpeqb %%mm7, %%mm4 \n\t" // d <= QP ? -1 : 0
508 "psubusb "MANGLE(b01)", %%mm3 \n\t"
509 "pand %%mm4, %%mm3 \n\t" // d <= QP ? d : 0
511 PAVGB(%%mm7, %%mm3) // d/2
512 "movq %%mm3, %%mm1 \n\t" // d/2
513 PAVGB(%%mm7, %%mm3) // d/4
514 PAVGB(%%mm1, %%mm3) // 3*d/8
516 "movq (%0, %1, 4), %%mm0 \n\t" // line 4
517 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
518 "psubusb %%mm3, %%mm0 \n\t"
519 "pxor %%mm2, %%mm0 \n\t"
520 "movq %%mm0, (%0, %1, 4) \n\t" // line 4
522 "movq (%%"REG_c"), %%mm0 \n\t" // line 5
523 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
524 "paddusb %%mm3, %%mm0 \n\t"
525 "pxor %%mm2, %%mm0 \n\t"
526 "movq %%mm0, (%%"REG_c") \n\t" // line 5
528 PAVGB(%%mm7, %%mm1) // d/4
530 "movq (%%"REG_a", %1, 2), %%mm0 \n\t" // line 3
531 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
532 "psubusb %%mm1, %%mm0 \n\t"
533 "pxor %%mm2, %%mm0 \n\t"
534 "movq %%mm0, (%%"REG_a", %1, 2) \n\t" // line 3
536 "movq (%%"REG_c", %1), %%mm0 \n\t" // line 6
537 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
538 "paddusb %%mm1, %%mm0 \n\t"
539 "pxor %%mm2, %%mm0 \n\t"
540 "movq %%mm0, (%%"REG_c", %1) \n\t" // line 6
542 PAVGB(%%mm7, %%mm1) // d/8
544 "movq (%%"REG_a", %1), %%mm0 \n\t" // line 2
545 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
546 "psubusb %%mm1, %%mm0 \n\t"
547 "pxor %%mm2, %%mm0 \n\t"
548 "movq %%mm0, (%%"REG_a", %1) \n\t" // line 2
550 "movq (%%"REG_c", %1, 2), %%mm0 \n\t" // line 7
551 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
552 "paddusb %%mm1, %%mm0 \n\t"
553 "pxor %%mm2, %%mm0 \n\t"
554 "movq %%mm0, (%%"REG_c", %1, 2) \n\t" // line 7
557 : "r" (src), "r" ((long)stride), "m" (co->pQPb)
560 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
562 const int l1= stride;
563 const int l2= stride + l1;
564 const int l3= stride + l2;
565 const int l4= stride + l3;
566 const int l5= stride + l4;
567 const int l6= stride + l5;
568 const int l7= stride + l6;
569 // const int l8= stride + l7;
570 // const int l9= stride + l8;
574 for(x=0; x<BLOCK_SIZE; x++)
576 int a= src[l3] - src[l4];
577 int b= src[l4] - src[l5];
578 int c= src[l5] - src[l6];
580 int d= FFABS(b) - ((FFABS(a) + FFABS(c))>>1);
585 int v = d * FFSIGN(-b);
597 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
601 static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext *c)
603 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
606 const int l1= stride;
607 const int l2= stride + l1;
608 const int l3= stride + l2;
609 const int l4= (int)tmp - (int)src - stride*3;
610 const int l5= (int)tmp - (int)src - stride*3 + 8;
611 const int l6= stride*3 + l3;
612 const int l7= stride + l6;
613 const int l8= stride + l7;
615 memcpy(tmp, src+stride*7, 8);
616 memcpy(tmp+8, src+stride*8, 8);
621 #if 0 //sligtly more accurate and slightly slower
622 "pxor %%mm7, %%mm7 \n\t" // 0
623 "lea (%0, %1), %%"REG_a" \n\t"
624 "lea (%%"REG_a", %1, 4), %%"REG_c" \n\t"
626 // %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 ecx+%1 ecx+2%1
627 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1
630 "movq (%0, %1, 2), %%mm0 \n\t" // l2
631 "movq (%0), %%mm1 \n\t" // l0
632 "movq %%mm0, %%mm2 \n\t" // l2
633 PAVGB(%%mm7, %%mm0) // ~l2/2
634 PAVGB(%%mm1, %%mm0) // ~(l2 + 2l0)/4
635 PAVGB(%%mm2, %%mm0) // ~(5l2 + 2l0)/8
637 "movq (%%"REG_a"), %%mm1 \n\t" // l1
638 "movq (%%"REG_a", %1, 2), %%mm3 \n\t" // l3
639 "movq %%mm1, %%mm4 \n\t" // l1
640 PAVGB(%%mm7, %%mm1) // ~l1/2
641 PAVGB(%%mm3, %%mm1) // ~(l1 + 2l3)/4
642 PAVGB(%%mm4, %%mm1) // ~(5l1 + 2l3)/8
644 "movq %%mm0, %%mm4 \n\t" // ~(5l2 + 2l0)/8
645 "psubusb %%mm1, %%mm0 \n\t"
646 "psubusb %%mm4, %%mm1 \n\t"
647 "por %%mm0, %%mm1 \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8
648 // mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0
650 "movq (%0, %1, 4), %%mm0 \n\t" // l4
651 "movq %%mm0, %%mm4 \n\t" // l4
652 PAVGB(%%mm7, %%mm0) // ~l4/2
653 PAVGB(%%mm2, %%mm0) // ~(l4 + 2l2)/4
654 PAVGB(%%mm4, %%mm0) // ~(5l4 + 2l2)/8
656 "movq (%%"REG_c"), %%mm2 \n\t" // l5
657 "movq %%mm3, %%mm5 \n\t" // l3
658 PAVGB(%%mm7, %%mm3) // ~l3/2
659 PAVGB(%%mm2, %%mm3) // ~(l3 + 2l5)/4
660 PAVGB(%%mm5, %%mm3) // ~(5l3 + 2l5)/8
662 "movq %%mm0, %%mm6 \n\t" // ~(5l4 + 2l2)/8
663 "psubusb %%mm3, %%mm0 \n\t"
664 "psubusb %%mm6, %%mm3 \n\t"
665 "por %%mm0, %%mm3 \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8
666 "pcmpeqb %%mm7, %%mm0 \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)
667 // mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0
669 "movq (%%"REG_c", %1), %%mm6 \n\t" // l6
670 "movq %%mm6, %%mm5 \n\t" // l6
671 PAVGB(%%mm7, %%mm6) // ~l6/2
672 PAVGB(%%mm4, %%mm6) // ~(l6 + 2l4)/4
673 PAVGB(%%mm5, %%mm6) // ~(5l6 + 2l4)/8
675 "movq (%%"REG_c", %1, 2), %%mm5 \n\t" // l7
676 "movq %%mm2, %%mm4 \n\t" // l5
677 PAVGB(%%mm7, %%mm2) // ~l5/2
678 PAVGB(%%mm5, %%mm2) // ~(l5 + 2l7)/4
679 PAVGB(%%mm4, %%mm2) // ~(5l5 + 2l7)/8
681 "movq %%mm6, %%mm4 \n\t" // ~(5l6 + 2l4)/8
682 "psubusb %%mm2, %%mm6 \n\t"
683 "psubusb %%mm4, %%mm2 \n\t"
684 "por %%mm6, %%mm2 \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8
685 // mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0
688 PMINUB(%%mm2, %%mm1, %%mm4) // MIN(|lenergy|,|renergy|)/8
689 "movq %2, %%mm4 \n\t" // QP //FIXME QP+1 ?
690 "paddusb "MANGLE(b01)", %%mm4 \n\t"
691 "pcmpgtb %%mm3, %%mm4 \n\t" // |menergy|/8 < QP
692 "psubusb %%mm1, %%mm3 \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8
693 "pand %%mm4, %%mm3 \n\t"
695 "movq %%mm3, %%mm1 \n\t"
696 // "psubusb "MANGLE(b01)", %%mm3 \n\t"
699 "paddusb %%mm1, %%mm3 \n\t"
700 // "paddusb "MANGLE(b01)", %%mm3 \n\t"
702 "movq (%%"REG_a", %1, 2), %%mm6 \n\t" //l3
703 "movq (%0, %1, 4), %%mm5 \n\t" //l4
704 "movq (%0, %1, 4), %%mm4 \n\t" //l4
705 "psubusb %%mm6, %%mm5 \n\t"
706 "psubusb %%mm4, %%mm6 \n\t"
707 "por %%mm6, %%mm5 \n\t" // |l3-l4|
708 "pcmpeqb %%mm7, %%mm6 \n\t" // SIGN(l3-l4)
709 "pxor %%mm6, %%mm0 \n\t"
710 "pand %%mm0, %%mm3 \n\t"
711 PMINUB(%%mm5, %%mm3, %%mm0)
713 "psubusb "MANGLE(b01)", %%mm3 \n\t"
716 "movq (%%"REG_a", %1, 2), %%mm0 \n\t"
717 "movq (%0, %1, 4), %%mm2 \n\t"
718 "pxor %%mm6, %%mm0 \n\t"
719 "pxor %%mm6, %%mm2 \n\t"
720 "psubb %%mm3, %%mm0 \n\t"
721 "paddb %%mm3, %%mm2 \n\t"
722 "pxor %%mm6, %%mm0 \n\t"
723 "pxor %%mm6, %%mm2 \n\t"
724 "movq %%mm0, (%%"REG_a", %1, 2) \n\t"
725 "movq %%mm2, (%0, %1, 4) \n\t"
728 "lea (%0, %1), %%"REG_a" \n\t"
729 "pcmpeqb %%mm6, %%mm6 \n\t" // -1
731 // %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 ecx+%1 ecx+2%1
732 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1
735 "movq (%%"REG_a", %1, 2), %%mm1 \n\t" // l3
736 "movq (%0, %1, 4), %%mm0 \n\t" // l4
737 "pxor %%mm6, %%mm1 \n\t" // -l3-1
738 PAVGB(%%mm1, %%mm0) // -q+128 = (l4-l3+256)/2
739 // mm1=-l3-1, mm0=128-q
741 "movq (%%"REG_a", %1, 4), %%mm2 \n\t" // l5
742 "movq (%%"REG_a", %1), %%mm3 \n\t" // l2
743 "pxor %%mm6, %%mm2 \n\t" // -l5-1
744 "movq %%mm2, %%mm5 \n\t" // -l5-1
745 "movq "MANGLE(b80)", %%mm4 \n\t" // 128
746 "lea (%%"REG_a", %1, 4), %%"REG_c" \n\t"
747 PAVGB(%%mm3, %%mm2) // (l2-l5+256)/2
748 PAVGB(%%mm0, %%mm4) // ~(l4-l3)/4 + 128
749 PAVGB(%%mm2, %%mm4) // ~(l2-l5)/4 +(l4-l3)/8 + 128
750 PAVGB(%%mm0, %%mm4) // ~(l2-l5)/8 +5(l4-l3)/16 + 128
751 // mm1=-l3-1, mm0=128-q, mm3=l2, mm4=menergy/16 + 128, mm5= -l5-1
753 "movq (%%"REG_a"), %%mm2 \n\t" // l1
754 "pxor %%mm6, %%mm2 \n\t" // -l1-1
755 PAVGB(%%mm3, %%mm2) // (l2-l1+256)/2
756 PAVGB((%0), %%mm1) // (l0-l3+256)/2
757 "movq "MANGLE(b80)", %%mm3 \n\t" // 128
758 PAVGB(%%mm2, %%mm3) // ~(l2-l1)/4 + 128
759 PAVGB(%%mm1, %%mm3) // ~(l0-l3)/4 +(l2-l1)/8 + 128
760 PAVGB(%%mm2, %%mm3) // ~(l0-l3)/8 +5(l2-l1)/16 + 128
761 // mm0=128-q, mm3=lenergy/16 + 128, mm4= menergy/16 + 128, mm5= -l5-1
763 PAVGB((%%REGc, %1), %%mm5) // (l6-l5+256)/2
764 "movq (%%"REG_c", %1, 2), %%mm1 \n\t" // l7
765 "pxor %%mm6, %%mm1 \n\t" // -l7-1
766 PAVGB((%0, %1, 4), %%mm1) // (l4-l7+256)/2
767 "movq "MANGLE(b80)", %%mm2 \n\t" // 128
768 PAVGB(%%mm5, %%mm2) // ~(l6-l5)/4 + 128
769 PAVGB(%%mm1, %%mm2) // ~(l4-l7)/4 +(l6-l5)/8 + 128
770 PAVGB(%%mm5, %%mm2) // ~(l4-l7)/8 +5(l6-l5)/16 + 128
771 // mm0=128-q, mm2=renergy/16 + 128, mm3=lenergy/16 + 128, mm4= menergy/16 + 128
773 "movq "MANGLE(b00)", %%mm1 \n\t" // 0
774 "movq "MANGLE(b00)", %%mm5 \n\t" // 0
775 "psubb %%mm2, %%mm1 \n\t" // 128 - renergy/16
776 "psubb %%mm3, %%mm5 \n\t" // 128 - lenergy/16
777 PMAXUB(%%mm1, %%mm2) // 128 + |renergy/16|
778 PMAXUB(%%mm5, %%mm3) // 128 + |lenergy/16|
779 PMINUB(%%mm2, %%mm3, %%mm1) // 128 + MIN(|lenergy|,|renergy|)/16
781 // mm0=128-q, mm3=128 + MIN(|lenergy|,|renergy|)/16, mm4= menergy/16 + 128
783 "movq "MANGLE(b00)", %%mm7 \n\t" // 0
784 "movq %2, %%mm2 \n\t" // QP
785 PAVGB(%%mm6, %%mm2) // 128 + QP/2
786 "psubb %%mm6, %%mm2 \n\t"
788 "movq %%mm4, %%mm1 \n\t"
789 "pcmpgtb %%mm7, %%mm1 \n\t" // SIGN(menergy)
790 "pxor %%mm1, %%mm4 \n\t"
791 "psubb %%mm1, %%mm4 \n\t" // 128 + |menergy|/16
792 "pcmpgtb %%mm4, %%mm2 \n\t" // |menergy|/16 < QP/2
793 "psubusb %%mm3, %%mm4 \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16
794 // mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16
796 "movq %%mm4, %%mm3 \n\t" // d
797 "psubusb "MANGLE(b01)", %%mm4 \n\t"
798 PAVGB(%%mm7, %%mm4) // d/32
799 PAVGB(%%mm7, %%mm4) // (d + 32)/64
800 "paddb %%mm3, %%mm4 \n\t" // 5d/64
801 "pand %%mm2, %%mm4 \n\t"
803 "movq "MANGLE(b80)", %%mm5 \n\t" // 128
804 "psubb %%mm0, %%mm5 \n\t" // q
805 "paddsb %%mm6, %%mm5 \n\t" // fix bad rounding
806 "pcmpgtb %%mm5, %%mm7 \n\t" // SIGN(q)
807 "pxor %%mm7, %%mm5 \n\t"
809 PMINUB(%%mm5, %%mm4, %%mm3) // MIN(|q|, 5d/64)
810 "pxor %%mm1, %%mm7 \n\t" // SIGN(d*q)
812 "pand %%mm7, %%mm4 \n\t"
813 "movq (%%"REG_a", %1, 2), %%mm0 \n\t"
814 "movq (%0, %1, 4), %%mm2 \n\t"
815 "pxor %%mm1, %%mm0 \n\t"
816 "pxor %%mm1, %%mm2 \n\t"
817 "paddb %%mm4, %%mm0 \n\t"
818 "psubb %%mm4, %%mm2 \n\t"
819 "pxor %%mm1, %%mm0 \n\t"
820 "pxor %%mm1, %%mm2 \n\t"
821 "movq %%mm0, (%%"REG_a", %1, 2) \n\t"
822 "movq %%mm2, (%0, %1, 4) \n\t"
825 : "r" (src), "r" ((long)stride), "m" (c->pQPb)
833 for(x=0; x<BLOCK_SIZE; x++)
835 const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
836 if(FFABS(middleEnergy)< 8*QP)
838 const int q=(src[l4] - src[l5])/2;
839 const int leftEnergy= 5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
840 const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
842 int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
846 d*= FFSIGN(-middleEnergy);
870 int d= src[x+y*stride] - tmp[x+(y-4)*8];
881 src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
887 av_log(c, AV_LOG_INFO, " %d %d %d %d\n", num, sum, max, bias);
893 #elif defined (HAVE_MMX)
896 "pxor %%mm7, %%mm7 \n\t"
897 "lea -40(%%"REG_SP"), %%"REG_c" \n\t" // make space for 4 8-byte vars
898 "and "ALIGN_MASK", %%"REG_c" \n\t" // align
900 // %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 edx+%1 edx+2%1
901 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1
903 "movq (%0), %%mm0 \n\t"
904 "movq %%mm0, %%mm1 \n\t"
905 "punpcklbw %%mm7, %%mm0 \n\t" // low part of line 0
906 "punpckhbw %%mm7, %%mm1 \n\t" // high part of line 0
908 "movq (%0, %1), %%mm2 \n\t"
909 "lea (%0, %1, 2), %%"REG_a" \n\t"
910 "movq %%mm2, %%mm3 \n\t"
911 "punpcklbw %%mm7, %%mm2 \n\t" // low part of line 1
912 "punpckhbw %%mm7, %%mm3 \n\t" // high part of line 1
914 "movq (%%"REG_a"), %%mm4 \n\t"
915 "movq %%mm4, %%mm5 \n\t"
916 "punpcklbw %%mm7, %%mm4 \n\t" // low part of line 2
917 "punpckhbw %%mm7, %%mm5 \n\t" // high part of line 2
919 "paddw %%mm0, %%mm0 \n\t" // 2L0
920 "paddw %%mm1, %%mm1 \n\t" // 2H0
921 "psubw %%mm4, %%mm2 \n\t" // L1 - L2
922 "psubw %%mm5, %%mm3 \n\t" // H1 - H2
923 "psubw %%mm2, %%mm0 \n\t" // 2L0 - L1 + L2
924 "psubw %%mm3, %%mm1 \n\t" // 2H0 - H1 + H2
926 "psllw $2, %%mm2 \n\t" // 4L1 - 4L2
927 "psllw $2, %%mm3 \n\t" // 4H1 - 4H2
928 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2
929 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2
931 "movq (%%"REG_a", %1), %%mm2 \n\t"
932 "movq %%mm2, %%mm3 \n\t"
933 "punpcklbw %%mm7, %%mm2 \n\t" // L3
934 "punpckhbw %%mm7, %%mm3 \n\t" // H3
936 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - L3
937 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - H3
938 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
939 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
940 "movq %%mm0, (%%"REG_c") \n\t" // 2L0 - 5L1 + 5L2 - 2L3
941 "movq %%mm1, 8(%%"REG_c") \n\t" // 2H0 - 5H1 + 5H2 - 2H3
943 "movq (%%"REG_a", %1, 2), %%mm0 \n\t"
944 "movq %%mm0, %%mm1 \n\t"
945 "punpcklbw %%mm7, %%mm0 \n\t" // L4
946 "punpckhbw %%mm7, %%mm1 \n\t" // H4
948 "psubw %%mm0, %%mm2 \n\t" // L3 - L4
949 "psubw %%mm1, %%mm3 \n\t" // H3 - H4
950 "movq %%mm2, 16(%%"REG_c") \n\t" // L3 - L4
951 "movq %%mm3, 24(%%"REG_c") \n\t" // H3 - H4
952 "paddw %%mm4, %%mm4 \n\t" // 2L2
953 "paddw %%mm5, %%mm5 \n\t" // 2H2
954 "psubw %%mm2, %%mm4 \n\t" // 2L2 - L3 + L4
955 "psubw %%mm3, %%mm5 \n\t" // 2H2 - H3 + H4
957 "lea (%%"REG_a", %1), %0 \n\t"
958 "psllw $2, %%mm2 \n\t" // 4L3 - 4L4
959 "psllw $2, %%mm3 \n\t" // 4H3 - 4H4
960 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4
961 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4
963 "movq (%0, %1, 2), %%mm2 \n\t"
964 "movq %%mm2, %%mm3 \n\t"
965 "punpcklbw %%mm7, %%mm2 \n\t" // L5
966 "punpckhbw %%mm7, %%mm3 \n\t" // H5
967 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - L5
968 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - H5
969 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - 2L5
970 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - 2H5
972 "movq (%%"REG_a", %1, 4), %%mm6 \n\t"
973 "punpcklbw %%mm7, %%mm6 \n\t" // L6
974 "psubw %%mm6, %%mm2 \n\t" // L5 - L6
975 "movq (%%"REG_a", %1, 4), %%mm6 \n\t"
976 "punpckhbw %%mm7, %%mm6 \n\t" // H6
977 "psubw %%mm6, %%mm3 \n\t" // H5 - H6
979 "paddw %%mm0, %%mm0 \n\t" // 2L4
980 "paddw %%mm1, %%mm1 \n\t" // 2H4
981 "psubw %%mm2, %%mm0 \n\t" // 2L4 - L5 + L6
982 "psubw %%mm3, %%mm1 \n\t" // 2H4 - H5 + H6
984 "psllw $2, %%mm2 \n\t" // 4L5 - 4L6
985 "psllw $2, %%mm3 \n\t" // 4H5 - 4H6
986 "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6
987 "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6
989 "movq (%0, %1, 4), %%mm2 \n\t"
990 "movq %%mm2, %%mm3 \n\t"
991 "punpcklbw %%mm7, %%mm2 \n\t" // L7
992 "punpckhbw %%mm7, %%mm3 \n\t" // H7
994 "paddw %%mm2, %%mm2 \n\t" // 2L7
995 "paddw %%mm3, %%mm3 \n\t" // 2H7
996 "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6 - 2L7
997 "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6 - 2H7
999 "movq (%%"REG_c"), %%mm2 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
1000 "movq 8(%%"REG_c"), %%mm3 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
1003 "movq %%mm7, %%mm6 \n\t" // 0
1004 "psubw %%mm0, %%mm6 \n\t"
1005 "pmaxsw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
1006 "movq %%mm7, %%mm6 \n\t" // 0
1007 "psubw %%mm1, %%mm6 \n\t"
1008 "pmaxsw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
1009 "movq %%mm7, %%mm6 \n\t" // 0
1010 "psubw %%mm2, %%mm6 \n\t"
1011 "pmaxsw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
1012 "movq %%mm7, %%mm6 \n\t" // 0
1013 "psubw %%mm3, %%mm6 \n\t"
1014 "pmaxsw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
1016 "movq %%mm7, %%mm6 \n\t" // 0
1017 "pcmpgtw %%mm0, %%mm6 \n\t"
1018 "pxor %%mm6, %%mm0 \n\t"
1019 "psubw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
1020 "movq %%mm7, %%mm6 \n\t" // 0
1021 "pcmpgtw %%mm1, %%mm6 \n\t"
1022 "pxor %%mm6, %%mm1 \n\t"
1023 "psubw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
1024 "movq %%mm7, %%mm6 \n\t" // 0
1025 "pcmpgtw %%mm2, %%mm6 \n\t"
1026 "pxor %%mm6, %%mm2 \n\t"
1027 "psubw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
1028 "movq %%mm7, %%mm6 \n\t" // 0
1029 "pcmpgtw %%mm3, %%mm6 \n\t"
1030 "pxor %%mm6, %%mm3 \n\t"
1031 "psubw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
1035 "pminsw %%mm2, %%mm0 \n\t"
1036 "pminsw %%mm3, %%mm1 \n\t"
1038 "movq %%mm0, %%mm6 \n\t"
1039 "psubusw %%mm2, %%mm6 \n\t"
1040 "psubw %%mm6, %%mm0 \n\t"
1041 "movq %%mm1, %%mm6 \n\t"
1042 "psubusw %%mm3, %%mm6 \n\t"
1043 "psubw %%mm6, %%mm1 \n\t"
1046 "movd %2, %%mm2 \n\t" // QP
1047 "punpcklbw %%mm7, %%mm2 \n\t"
1049 "movq %%mm7, %%mm6 \n\t" // 0
1050 "pcmpgtw %%mm4, %%mm6 \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
1051 "pxor %%mm6, %%mm4 \n\t"
1052 "psubw %%mm6, %%mm4 \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
1053 "pcmpgtw %%mm5, %%mm7 \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
1054 "pxor %%mm7, %%mm5 \n\t"
1055 "psubw %%mm7, %%mm5 \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
1057 "psllw $3, %%mm2 \n\t" // 8QP
1058 "movq %%mm2, %%mm3 \n\t" // 8QP
1059 "pcmpgtw %%mm4, %%mm2 \n\t"
1060 "pcmpgtw %%mm5, %%mm3 \n\t"
1061 "pand %%mm2, %%mm4 \n\t"
1062 "pand %%mm3, %%mm5 \n\t"
1065 "psubusw %%mm0, %%mm4 \n\t" // hd
1066 "psubusw %%mm1, %%mm5 \n\t" // ld
1069 "movq "MANGLE(w05)", %%mm2 \n\t" // 5
1070 "pmullw %%mm2, %%mm4 \n\t"
1071 "pmullw %%mm2, %%mm5 \n\t"
1072 "movq "MANGLE(w20)", %%mm2 \n\t" // 32
1073 "paddw %%mm2, %%mm4 \n\t"
1074 "paddw %%mm2, %%mm5 \n\t"
1075 "psrlw $6, %%mm4 \n\t"
1076 "psrlw $6, %%mm5 \n\t"
1078 "movq 16(%%"REG_c"), %%mm0 \n\t" // L3 - L4
1079 "movq 24(%%"REG_c"), %%mm1 \n\t" // H3 - H4
1081 "pxor %%mm2, %%mm2 \n\t"
1082 "pxor %%mm3, %%mm3 \n\t"
1084 "pcmpgtw %%mm0, %%mm2 \n\t" // sign (L3-L4)
1085 "pcmpgtw %%mm1, %%mm3 \n\t" // sign (H3-H4)
1086 "pxor %%mm2, %%mm0 \n\t"
1087 "pxor %%mm3, %%mm1 \n\t"
1088 "psubw %%mm2, %%mm0 \n\t" // |L3-L4|
1089 "psubw %%mm3, %%mm1 \n\t" // |H3-H4|
1090 "psrlw $1, %%mm0 \n\t" // |L3 - L4|/2
1091 "psrlw $1, %%mm1 \n\t" // |H3 - H4|/2
1093 "pxor %%mm6, %%mm2 \n\t"
1094 "pxor %%mm7, %%mm3 \n\t"
1095 "pand %%mm2, %%mm4 \n\t"
1096 "pand %%mm3, %%mm5 \n\t"
1099 "pminsw %%mm0, %%mm4 \n\t"
1100 "pminsw %%mm1, %%mm5 \n\t"
1102 "movq %%mm4, %%mm2 \n\t"
1103 "psubusw %%mm0, %%mm2 \n\t"
1104 "psubw %%mm2, %%mm4 \n\t"
1105 "movq %%mm5, %%mm2 \n\t"
1106 "psubusw %%mm1, %%mm2 \n\t"
1107 "psubw %%mm2, %%mm5 \n\t"
1109 "pxor %%mm6, %%mm4 \n\t"
1110 "pxor %%mm7, %%mm5 \n\t"
1111 "psubw %%mm6, %%mm4 \n\t"
1112 "psubw %%mm7, %%mm5 \n\t"
1113 "packsswb %%mm5, %%mm4 \n\t"
1114 "movq (%0), %%mm0 \n\t"
1115 "paddb %%mm4, %%mm0 \n\t"
1116 "movq %%mm0, (%0) \n\t"
1117 "movq (%0, %1), %%mm0 \n\t"
1118 "psubb %%mm4, %%mm0 \n\t"
1119 "movq %%mm0, (%0, %1) \n\t"
1122 : "r" ((long)stride), "m" (c->pQPb)
1123 : "%"REG_a, "%"REG_c
1125 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1126 const int l1= stride;
1127 const int l2= stride + l1;
1128 const int l3= stride + l2;
1129 const int l4= stride + l3;
1130 const int l5= stride + l4;
1131 const int l6= stride + l5;
1132 const int l7= stride + l6;
1133 const int l8= stride + l7;
1134 // const int l9= stride + l8;
1137 for(x=0; x<BLOCK_SIZE; x++)
1139 const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
1140 if(FFABS(middleEnergy) < 8*c->QP)
1142 const int q=(src[l4] - src[l5])/2;
1143 const int leftEnergy= 5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
1144 const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
1146 int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
1150 d*= FFSIGN(-middleEnergy);
1168 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1170 #endif //HAVE_ALTIVEC
1172 #ifndef HAVE_ALTIVEC
1173 static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
1175 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1177 "pxor %%mm6, %%mm6 \n\t"
1178 "pcmpeqb %%mm7, %%mm7 \n\t"
1179 "movq %2, %%mm0 \n\t"
1180 "punpcklbw %%mm6, %%mm0 \n\t"
1181 "psrlw $1, %%mm0 \n\t"
1182 "psubw %%mm7, %%mm0 \n\t"
1183 "packuswb %%mm0, %%mm0 \n\t"
1184 "movq %%mm0, %3 \n\t"
1186 "lea (%0, %1), %%"REG_a" \n\t"
1187 "lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1189 // 0 1 2 3 4 5 6 7 8 9
1190 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1194 #define REAL_FIND_MIN_MAX(addr)\
1195 "movq " #addr ", %%mm0 \n\t"\
1196 "pminub %%mm0, %%mm7 \n\t"\
1197 "pmaxub %%mm0, %%mm6 \n\t"
1199 #define REAL_FIND_MIN_MAX(addr)\
1200 "movq " #addr ", %%mm0 \n\t"\
1201 "movq %%mm7, %%mm1 \n\t"\
1202 "psubusb %%mm0, %%mm6 \n\t"\
1203 "paddb %%mm0, %%mm6 \n\t"\
1204 "psubusb %%mm0, %%mm1 \n\t"\
1205 "psubb %%mm1, %%mm7 \n\t"
1207 #define FIND_MIN_MAX(addr) REAL_FIND_MIN_MAX(addr)
1209 FIND_MIN_MAX((%%REGa))
1210 FIND_MIN_MAX((%%REGa, %1))
1211 FIND_MIN_MAX((%%REGa, %1, 2))
1212 FIND_MIN_MAX((%0, %1, 4))
1213 FIND_MIN_MAX((%%REGd))
1214 FIND_MIN_MAX((%%REGd, %1))
1215 FIND_MIN_MAX((%%REGd, %1, 2))
1216 FIND_MIN_MAX((%0, %1, 8))
1218 "movq %%mm7, %%mm4 \n\t"
1219 "psrlq $8, %%mm7 \n\t"
1221 "pminub %%mm4, %%mm7 \n\t" // min of pixels
1222 "pshufw $0xF9, %%mm7, %%mm4 \n\t"
1223 "pminub %%mm4, %%mm7 \n\t" // min of pixels
1224 "pshufw $0xFE, %%mm7, %%mm4 \n\t"
1225 "pminub %%mm4, %%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 $16, %%mm7 \n\t"
1232 "movq %%mm7, %%mm1 \n\t"
1233 "psubusb %%mm4, %%mm1 \n\t"
1234 "psubb %%mm1, %%mm7 \n\t"
1235 "movq %%mm7, %%mm4 \n\t"
1236 "psrlq $32, %%mm7 \n\t"
1237 "movq %%mm7, %%mm1 \n\t"
1238 "psubusb %%mm4, %%mm1 \n\t"
1239 "psubb %%mm1, %%mm7 \n\t"
1243 "movq %%mm6, %%mm4 \n\t"
1244 "psrlq $8, %%mm6 \n\t"
1246 "pmaxub %%mm4, %%mm6 \n\t" // max of pixels
1247 "pshufw $0xF9, %%mm6, %%mm4 \n\t"
1248 "pmaxub %%mm4, %%mm6 \n\t"
1249 "pshufw $0xFE, %%mm6, %%mm4 \n\t"
1250 "pmaxub %%mm4, %%mm6 \n\t"
1252 "psubusb %%mm4, %%mm6 \n\t"
1253 "paddb %%mm4, %%mm6 \n\t"
1254 "movq %%mm6, %%mm4 \n\t"
1255 "psrlq $16, %%mm6 \n\t"
1256 "psubusb %%mm4, %%mm6 \n\t"
1257 "paddb %%mm4, %%mm6 \n\t"
1258 "movq %%mm6, %%mm4 \n\t"
1259 "psrlq $32, %%mm6 \n\t"
1260 "psubusb %%mm4, %%mm6 \n\t"
1261 "paddb %%mm4, %%mm6 \n\t"
1263 "movq %%mm6, %%mm0 \n\t" // max
1264 "psubb %%mm7, %%mm6 \n\t" // max - min
1265 "movd %%mm6, %%ecx \n\t"
1266 "cmpb "MANGLE(deringThreshold)", %%cl \n\t"
1268 "lea -24(%%"REG_SP"), %%"REG_c" \n\t"
1269 "and "ALIGN_MASK", %%"REG_c" \n\t"
1270 PAVGB(%%mm0, %%mm7) // a=(max + min)/2
1271 "punpcklbw %%mm7, %%mm7 \n\t"
1272 "punpcklbw %%mm7, %%mm7 \n\t"
1273 "punpcklbw %%mm7, %%mm7 \n\t"
1274 "movq %%mm7, (%%"REG_c") \n\t"
1276 "movq (%0), %%mm0 \n\t" // L10
1277 "movq %%mm0, %%mm1 \n\t" // L10
1278 "movq %%mm0, %%mm2 \n\t" // L10
1279 "psllq $8, %%mm1 \n\t"
1280 "psrlq $8, %%mm2 \n\t"
1281 "movd -4(%0), %%mm3 \n\t"
1282 "movd 8(%0), %%mm4 \n\t"
1283 "psrlq $24, %%mm3 \n\t"
1284 "psllq $56, %%mm4 \n\t"
1285 "por %%mm3, %%mm1 \n\t" // L00
1286 "por %%mm4, %%mm2 \n\t" // L20
1287 "movq %%mm1, %%mm3 \n\t" // L00
1288 PAVGB(%%mm2, %%mm1) // (L20 + L00)/2
1289 PAVGB(%%mm0, %%mm1) // (L20 + L00 + 2L10)/4
1290 "psubusb %%mm7, %%mm0 \n\t"
1291 "psubusb %%mm7, %%mm2 \n\t"
1292 "psubusb %%mm7, %%mm3 \n\t"
1293 "pcmpeqb "MANGLE(b00)", %%mm0 \n\t" // L10 > a ? 0 : -1
1294 "pcmpeqb "MANGLE(b00)", %%mm2 \n\t" // L20 > a ? 0 : -1
1295 "pcmpeqb "MANGLE(b00)", %%mm3 \n\t" // L00 > a ? 0 : -1
1296 "paddb %%mm2, %%mm0 \n\t"
1297 "paddb %%mm3, %%mm0 \n\t"
1299 "movq (%%"REG_a"), %%mm2 \n\t" // L11
1300 "movq %%mm2, %%mm3 \n\t" // L11
1301 "movq %%mm2, %%mm4 \n\t" // L11
1302 "psllq $8, %%mm3 \n\t"
1303 "psrlq $8, %%mm4 \n\t"
1304 "movd -4(%%"REG_a"), %%mm5 \n\t"
1305 "movd 8(%%"REG_a"), %%mm6 \n\t"
1306 "psrlq $24, %%mm5 \n\t"
1307 "psllq $56, %%mm6 \n\t"
1308 "por %%mm5, %%mm3 \n\t" // L01
1309 "por %%mm6, %%mm4 \n\t" // L21
1310 "movq %%mm3, %%mm5 \n\t" // L01
1311 PAVGB(%%mm4, %%mm3) // (L21 + L01)/2
1312 PAVGB(%%mm2, %%mm3) // (L21 + L01 + 2L11)/4
1313 "psubusb %%mm7, %%mm2 \n\t"
1314 "psubusb %%mm7, %%mm4 \n\t"
1315 "psubusb %%mm7, %%mm5 \n\t"
1316 "pcmpeqb "MANGLE(b00)", %%mm2 \n\t" // L11 > a ? 0 : -1
1317 "pcmpeqb "MANGLE(b00)", %%mm4 \n\t" // L21 > a ? 0 : -1
1318 "pcmpeqb "MANGLE(b00)", %%mm5 \n\t" // L01 > a ? 0 : -1
1319 "paddb %%mm4, %%mm2 \n\t"
1320 "paddb %%mm5, %%mm2 \n\t"
1322 #define REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
1323 "movq " #src ", " #sx " \n\t" /* src[0] */\
1324 "movq " #sx ", " #lx " \n\t" /* src[0] */\
1325 "movq " #sx ", " #t0 " \n\t" /* src[0] */\
1326 "psllq $8, " #lx " \n\t"\
1327 "psrlq $8, " #t0 " \n\t"\
1328 "movd -4" #src ", " #t1 " \n\t"\
1329 "psrlq $24, " #t1 " \n\t"\
1330 "por " #t1 ", " #lx " \n\t" /* src[-1] */\
1331 "movd 8" #src ", " #t1 " \n\t"\
1332 "psllq $56, " #t1 " \n\t"\
1333 "por " #t1 ", " #t0 " \n\t" /* src[+1] */\
1334 "movq " #lx ", " #t1 " \n\t" /* src[-1] */\
1335 PAVGB(t0, lx) /* (src[-1] + src[+1])/2 */\
1336 PAVGB(sx, lx) /* (src[-1] + 2src[0] + src[+1])/4 */\
1338 "movq " #lx ", 8(%%"REG_c") \n\t"\
1339 "movq (%%"REG_c"), " #lx " \n\t"\
1340 "psubusb " #lx ", " #t1 " \n\t"\
1341 "psubusb " #lx ", " #t0 " \n\t"\
1342 "psubusb " #lx ", " #sx " \n\t"\
1343 "movq "MANGLE(b00)", " #lx " \n\t"\
1344 "pcmpeqb " #lx ", " #t1 " \n\t" /* src[-1] > a ? 0 : -1*/\
1345 "pcmpeqb " #lx ", " #t0 " \n\t" /* src[+1] > a ? 0 : -1*/\
1346 "pcmpeqb " #lx ", " #sx " \n\t" /* src[0] > a ? 0 : -1*/\
1347 "paddb " #t1 ", " #t0 " \n\t"\
1348 "paddb " #t0 ", " #sx " \n\t"\
1350 PAVGB(plx, pplx) /* filtered */\
1351 "movq " #dst ", " #t0 " \n\t" /* dst */\
1352 "movq " #t0 ", " #t1 " \n\t" /* dst */\
1353 "psubusb %3, " #t0 " \n\t"\
1354 "paddusb %3, " #t1 " \n\t"\
1356 PMINUB(t1, pplx, t0)\
1357 "paddb " #sx ", " #ppsx " \n\t"\
1358 "paddb " #psx ", " #ppsx " \n\t"\
1359 "#paddb "MANGLE(b02)", " #ppsx " \n\t"\
1360 "pand "MANGLE(b08)", " #ppsx " \n\t"\
1361 "pcmpeqb " #lx ", " #ppsx " \n\t"\
1362 "pand " #ppsx ", " #pplx " \n\t"\
1363 "pandn " #dst ", " #ppsx " \n\t"\
1364 "por " #pplx ", " #ppsx " \n\t"\
1365 "movq " #ppsx ", " #dst " \n\t"\
1366 "movq 8(%%"REG_c"), " #lx " \n\t"
1368 #define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
1369 REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1)
1385 //DERING_CORE(dst ,src ,ppsx ,psx ,sx ,pplx ,plx ,lx ,t0 ,t1)
1386 DERING_CORE((%%REGa) ,(%%REGa, %1) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1387 DERING_CORE((%%REGa, %1) ,(%%REGa, %1, 2),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1388 DERING_CORE((%%REGa, %1, 2),(%0, %1, 4) ,%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
1389 DERING_CORE((%0, %1, 4) ,(%%REGd) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1390 DERING_CORE((%%REGd) ,(%%REGd, %1) ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1391 DERING_CORE((%%REGd, %1) ,(%%REGd, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
1392 DERING_CORE((%%REGd, %1, 2),(%0, %1, 8) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1393 DERING_CORE((%0, %1, 8) ,(%%REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1396 : : "r" (src), "r" ((long)stride), "m" (c->pQPb), "m"(c->pQPb2)
1397 : "%"REG_a, "%"REG_d, "%"REG_c
1399 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1406 const int QP2= c->QP/2 + 1;
1415 if(*p > max) max= *p;
1416 if(*p < min) min= *p;
1419 avg= (min + max + 1)>>1;
1421 if(max - min <deringThreshold) return;
1427 if(src[stride*y + 0] > avg) t+= 1;
1428 if(src[stride*y + 1] > avg) t+= 2;
1429 if(src[stride*y + 2] > avg) t+= 4;
1430 if(src[stride*y + 3] > avg) t+= 8;
1431 if(src[stride*y + 4] > avg) t+= 16;
1432 if(src[stride*y + 5] > avg) t+= 32;
1433 if(src[stride*y + 6] > avg) t+= 64;
1434 if(src[stride*y + 7] > avg) t+= 128;
1435 if(src[stride*y + 8] > avg) t+= 256;
1436 if(src[stride*y + 9] > avg) t+= 512;
1439 t &= (t<<1) & (t>>1);
1445 int t = s[y-1] & s[y] & s[y+1];
1461 int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))
1462 +2*(*(p -1)) + 4*(*p ) + 2*(*(p +1))
1463 +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));
1466 #ifdef DEBUG_DERING_THRESHOLD
1467 asm volatile("emms\n\t":);
1469 static long long numPixels=0;
1470 if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;
1471 // if((max-min)<20 || (max-min)*QP<200)
1472 // if((max-min)*QP < 500)
1476 static int numSkiped=0;
1477 static int errorSum=0;
1478 static int worstQP=0;
1479 static int worstRange=0;
1480 static int worstDiff=0;
1482 int absDiff= FFABS(diff);
1483 int error= diff*diff;
1485 if(x==1 || x==8 || y==1 || y==8) continue;
1488 if(absDiff > worstDiff)
1492 worstRange= max-min;
1496 if(1024LL*1024LL*1024LL % numSkiped == 0)
1498 av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "
1499 "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
1500 (float)errorSum/numSkiped, numSkiped, worstQP, worstRange,
1501 worstDiff, (float)numSkiped/numPixels);
1506 if (*p + QP2 < f) *p= *p + QP2;
1507 else if(*p - QP2 > f) *p= *p - QP2;
1512 #ifdef DEBUG_DERING_THRESHOLD
1523 *p = FFMIN(*p + 20, 255);
1526 // src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
1529 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1531 #endif //HAVE_ALTIVEC
1534 * Deinterlaces the given block by linearly interpolating every second line.
1535 * will be called for every 8x8 block and can read & write from line 4-15
1536 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1537 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1539 static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)
1541 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1544 "lea (%0, %1), %%"REG_a" \n\t"
1545 "lea (%%"REG_a", %1, 4), %%"REG_c" \n\t"
1546 // 0 1 2 3 4 5 6 7 8 9
1547 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1 %0+8%1 ecx+4%1
1549 "movq (%0), %%mm0 \n\t"
1550 "movq (%%"REG_a", %1), %%mm1 \n\t"
1552 "movq %%mm0, (%%"REG_a") \n\t"
1553 "movq (%0, %1, 4), %%mm0 \n\t"
1555 "movq %%mm1, (%%"REG_a", %1, 2) \n\t"
1556 "movq (%%"REG_c", %1), %%mm1 \n\t"
1558 "movq %%mm0, (%%"REG_c") \n\t"
1559 "movq (%0, %1, 8), %%mm0 \n\t"
1561 "movq %%mm1, (%%"REG_c", %1, 2) \n\t"
1563 : : "r" (src), "r" ((long)stride)
1564 : "%"REG_a, "%"REG_c
1571 a= *(uint32_t*)&src[stride*0];
1572 b= *(uint32_t*)&src[stride*2];
1573 *(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1574 a= *(uint32_t*)&src[stride*4];
1575 *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1576 b= *(uint32_t*)&src[stride*6];
1577 *(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1578 a= *(uint32_t*)&src[stride*8];
1579 *(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1586 * Deinterlaces the given block by cubic interpolating every second line.
1587 * will be called for every 8x8 block and can read & write from line 4-15
1588 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1589 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1590 * this filter will read lines 3-15 and write 7-13
1592 static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)
1594 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1597 "lea (%0, %1), %%"REG_a" \n\t"
1598 "lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1599 "lea (%%"REG_d", %1, 4), %%"REG_c" \n\t"
1600 "add %1, %%"REG_c" \n\t"
1601 "pxor %%mm7, %%mm7 \n\t"
1602 // 0 1 2 3 4 5 6 7 8 9 10
1603 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1 ecx
1605 #define REAL_DEINT_CUBIC(a,b,c,d,e)\
1606 "movq " #a ", %%mm0 \n\t"\
1607 "movq " #b ", %%mm1 \n\t"\
1608 "movq " #d ", %%mm2 \n\t"\
1609 "movq " #e ", %%mm3 \n\t"\
1610 PAVGB(%%mm2, %%mm1) /* (b+d) /2 */\
1611 PAVGB(%%mm3, %%mm0) /* a(a+e) /2 */\
1612 "movq %%mm0, %%mm2 \n\t"\
1613 "punpcklbw %%mm7, %%mm0 \n\t"\
1614 "punpckhbw %%mm7, %%mm2 \n\t"\
1615 "movq %%mm1, %%mm3 \n\t"\
1616 "punpcklbw %%mm7, %%mm1 \n\t"\
1617 "punpckhbw %%mm7, %%mm3 \n\t"\
1618 "psubw %%mm1, %%mm0 \n\t" /* L(a+e - (b+d))/2 */\
1619 "psubw %%mm3, %%mm2 \n\t" /* H(a+e - (b+d))/2 */\
1620 "psraw $3, %%mm0 \n\t" /* L(a+e - (b+d))/16 */\
1621 "psraw $3, %%mm2 \n\t" /* H(a+e - (b+d))/16 */\
1622 "psubw %%mm0, %%mm1 \n\t" /* L(9b + 9d - a - e)/16 */\
1623 "psubw %%mm2, %%mm3 \n\t" /* H(9b + 9d - a - e)/16 */\
1624 "packuswb %%mm3, %%mm1 \n\t"\
1625 "movq %%mm1, " #c " \n\t"
1626 #define DEINT_CUBIC(a,b,c,d,e) REAL_DEINT_CUBIC(a,b,c,d,e)
1628 DEINT_CUBIC((%0) , (%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd, %1))
1629 DEINT_CUBIC((%%REGa, %1), (%0, %1, 4) , (%%REGd) , (%%REGd, %1), (%0, %1, 8))
1630 DEINT_CUBIC((%0, %1, 4) , (%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGc))
1631 DEINT_CUBIC((%%REGd, %1), (%0, %1, 8) , (%%REGd, %1, 4), (%%REGc) , (%%REGc, %1, 2))
1633 : : "r" (src), "r" ((long)stride)
1634 : "%"REG_a, "%"REG_d, "%"REG_c
1636 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1641 src[stride*3] = CLIP((-src[0] + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);
1642 src[stride*5] = CLIP((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);
1643 src[stride*7] = CLIP((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);
1644 src[stride*9] = CLIP((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);
1647 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1651 * Deinterlaces the given block by filtering every second line with a (-1 4 2 4 -1) filter.
1652 * will be called for every 8x8 block and can read & write from line 4-15
1653 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1654 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1655 * this filter will read lines 4-13 and write 5-11
1657 static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)
1659 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1662 "lea (%0, %1), %%"REG_a" \n\t"
1663 "lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1664 "pxor %%mm7, %%mm7 \n\t"
1665 "movq (%2), %%mm0 \n\t"
1666 // 0 1 2 3 4 5 6 7 8 9 10
1667 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1 ecx
1669 #define REAL_DEINT_FF(a,b,c,d)\
1670 "movq " #a ", %%mm1 \n\t"\
1671 "movq " #b ", %%mm2 \n\t"\
1672 "movq " #c ", %%mm3 \n\t"\
1673 "movq " #d ", %%mm4 \n\t"\
1674 PAVGB(%%mm3, %%mm1) \
1675 PAVGB(%%mm4, %%mm0) \
1676 "movq %%mm0, %%mm3 \n\t"\
1677 "punpcklbw %%mm7, %%mm0 \n\t"\
1678 "punpckhbw %%mm7, %%mm3 \n\t"\
1679 "movq %%mm1, %%mm4 \n\t"\
1680 "punpcklbw %%mm7, %%mm1 \n\t"\
1681 "punpckhbw %%mm7, %%mm4 \n\t"\
1682 "psllw $2, %%mm1 \n\t"\
1683 "psllw $2, %%mm4 \n\t"\
1684 "psubw %%mm0, %%mm1 \n\t"\
1685 "psubw %%mm3, %%mm4 \n\t"\
1686 "movq %%mm2, %%mm5 \n\t"\
1687 "movq %%mm2, %%mm0 \n\t"\
1688 "punpcklbw %%mm7, %%mm2 \n\t"\
1689 "punpckhbw %%mm7, %%mm5 \n\t"\
1690 "paddw %%mm2, %%mm1 \n\t"\
1691 "paddw %%mm5, %%mm4 \n\t"\
1692 "psraw $2, %%mm1 \n\t"\
1693 "psraw $2, %%mm4 \n\t"\
1694 "packuswb %%mm4, %%mm1 \n\t"\
1695 "movq %%mm1, " #b " \n\t"\
1697 #define DEINT_FF(a,b,c,d) REAL_DEINT_FF(a,b,c,d)
1699 DEINT_FF((%0) , (%%REGa) , (%%REGa, %1), (%%REGa, %1, 2))
1700 DEINT_FF((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd) )
1701 DEINT_FF((%0, %1, 4) , (%%REGd) , (%%REGd, %1), (%%REGd, %1, 2))
1702 DEINT_FF((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))
1704 "movq %%mm0, (%2) \n\t"
1705 : : "r" (src), "r" ((long)stride), "r"(tmp)
1706 : "%"REG_a, "%"REG_d
1708 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1714 int t2= src[stride*1];
1716 src[stride*1]= CLIP((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);
1718 src[stride*3]= CLIP((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);
1720 src[stride*5]= CLIP((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);
1722 src[stride*7]= CLIP((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);
1727 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1731 * Deinterlaces the given block by filtering every line with a (-1 2 6 2 -1) filter.
1732 * will be called for every 8x8 block and can read & write from line 4-15
1733 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1734 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1735 * this filter will read lines 4-13 and write 4-11
1737 static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)
1739 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1742 "lea (%0, %1), %%"REG_a" \n\t"
1743 "lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1744 "pxor %%mm7, %%mm7 \n\t"
1745 "movq (%2), %%mm0 \n\t"
1746 "movq (%3), %%mm1 \n\t"
1747 // 0 1 2 3 4 5 6 7 8 9 10
1748 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1 ecx
1750 #define REAL_DEINT_L5(t1,t2,a,b,c)\
1751 "movq " #a ", %%mm2 \n\t"\
1752 "movq " #b ", %%mm3 \n\t"\
1753 "movq " #c ", %%mm4 \n\t"\
1756 "movq %%mm2, %%mm5 \n\t"\
1757 "movq %%mm2, " #t1 " \n\t"\
1758 "punpcklbw %%mm7, %%mm2 \n\t"\
1759 "punpckhbw %%mm7, %%mm5 \n\t"\
1760 "movq %%mm2, %%mm6 \n\t"\
1761 "paddw %%mm2, %%mm2 \n\t"\
1762 "paddw %%mm6, %%mm2 \n\t"\
1763 "movq %%mm5, %%mm6 \n\t"\
1764 "paddw %%mm5, %%mm5 \n\t"\
1765 "paddw %%mm6, %%mm5 \n\t"\
1766 "movq %%mm3, %%mm6 \n\t"\
1767 "punpcklbw %%mm7, %%mm3 \n\t"\
1768 "punpckhbw %%mm7, %%mm6 \n\t"\
1769 "paddw %%mm3, %%mm3 \n\t"\
1770 "paddw %%mm6, %%mm6 \n\t"\
1771 "paddw %%mm3, %%mm2 \n\t"\
1772 "paddw %%mm6, %%mm5 \n\t"\
1773 "movq %%mm4, %%mm6 \n\t"\
1774 "punpcklbw %%mm7, %%mm4 \n\t"\
1775 "punpckhbw %%mm7, %%mm6 \n\t"\
1776 "psubw %%mm4, %%mm2 \n\t"\
1777 "psubw %%mm6, %%mm5 \n\t"\
1778 "psraw $2, %%mm2 \n\t"\
1779 "psraw $2, %%mm5 \n\t"\
1780 "packuswb %%mm5, %%mm2 \n\t"\
1781 "movq %%mm2, " #a " \n\t"\
1783 #define DEINT_L5(t1,t2,a,b,c) REAL_DEINT_L5(t1,t2,a,b,c)
1785 DEINT_L5(%%mm0, %%mm1, (%0) , (%%REGa) , (%%REGa, %1) )
1786 DEINT_L5(%%mm1, %%mm0, (%%REGa) , (%%REGa, %1) , (%%REGa, %1, 2))
1787 DEINT_L5(%%mm0, %%mm1, (%%REGa, %1) , (%%REGa, %1, 2), (%0, %1, 4) )
1788 DEINT_L5(%%mm1, %%mm0, (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd) )
1789 DEINT_L5(%%mm0, %%mm1, (%0, %1, 4) , (%%REGd) , (%%REGd, %1) )
1790 DEINT_L5(%%mm1, %%mm0, (%%REGd) , (%%REGd, %1) , (%%REGd, %1, 2))
1791 DEINT_L5(%%mm0, %%mm1, (%%REGd, %1) , (%%REGd, %1, 2), (%0, %1, 8) )
1792 DEINT_L5(%%mm1, %%mm0, (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))
1794 "movq %%mm0, (%2) \n\t"
1795 "movq %%mm1, (%3) \n\t"
1796 : : "r" (src), "r" ((long)stride), "r"(tmp), "r"(tmp2)
1797 : "%"REG_a, "%"REG_d
1799 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1808 src[stride*0]= CLIP((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);
1810 src[stride*1]= CLIP((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);
1812 src[stride*2]= CLIP((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);
1814 src[stride*3]= CLIP((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);
1816 src[stride*4]= CLIP((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);
1818 src[stride*5]= CLIP((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);
1820 src[stride*6]= CLIP((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);
1822 src[stride*7]= CLIP((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);
1829 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1833 * Deinterlaces the given block by filtering all lines with a (1 2 1) filter.
1834 * will be called for every 8x8 block and can read & write from line 4-15
1835 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1836 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1837 * this filter will read lines 4-13 and write 4-11
1839 static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uint8_t *tmp)
1841 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1844 "lea (%0, %1), %%"REG_a" \n\t"
1845 "lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1846 // 0 1 2 3 4 5 6 7 8 9
1847 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1849 "movq (%2), %%mm0 \n\t" // L0
1850 "movq (%%"REG_a"), %%mm1 \n\t" // L2
1851 PAVGB(%%mm1, %%mm0) // L0+L2
1852 "movq (%0), %%mm2 \n\t" // L1
1854 "movq %%mm0, (%0) \n\t"
1855 "movq (%%"REG_a", %1), %%mm0 \n\t" // L3
1856 PAVGB(%%mm0, %%mm2) // L1+L3
1857 PAVGB(%%mm1, %%mm2) // 2L2 + L1 + L3
1858 "movq %%mm2, (%%"REG_a") \n\t"
1859 "movq (%%"REG_a", %1, 2), %%mm2 \n\t" // L4
1860 PAVGB(%%mm2, %%mm1) // L2+L4
1861 PAVGB(%%mm0, %%mm1) // 2L3 + L2 + L4
1862 "movq %%mm1, (%%"REG_a", %1) \n\t"
1863 "movq (%0, %1, 4), %%mm1 \n\t" // L5
1864 PAVGB(%%mm1, %%mm0) // L3+L5
1865 PAVGB(%%mm2, %%mm0) // 2L4 + L3 + L5
1866 "movq %%mm0, (%%"REG_a", %1, 2) \n\t"
1867 "movq (%%"REG_d"), %%mm0 \n\t" // L6
1868 PAVGB(%%mm0, %%mm2) // L4+L6
1869 PAVGB(%%mm1, %%mm2) // 2L5 + L4 + L6
1870 "movq %%mm2, (%0, %1, 4) \n\t"
1871 "movq (%%"REG_d", %1), %%mm2 \n\t" // L7
1872 PAVGB(%%mm2, %%mm1) // L5+L7
1873 PAVGB(%%mm0, %%mm1) // 2L6 + L5 + L7
1874 "movq %%mm1, (%%"REG_d") \n\t"
1875 "movq (%%"REG_d", %1, 2), %%mm1 \n\t" // L8
1876 PAVGB(%%mm1, %%mm0) // L6+L8
1877 PAVGB(%%mm2, %%mm0) // 2L7 + L6 + L8
1878 "movq %%mm0, (%%"REG_d", %1) \n\t"
1879 "movq (%0, %1, 8), %%mm0 \n\t" // L9
1880 PAVGB(%%mm0, %%mm2) // L7+L9
1881 PAVGB(%%mm1, %%mm2) // 2L8 + L7 + L9
1882 "movq %%mm2, (%%"REG_d", %1, 2) \n\t"
1883 "movq %%mm1, (%2) \n\t"
1885 : : "r" (src), "r" ((long)stride), "r" (tmp)
1886 : "%"REG_a, "%"REG_d
1888 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1893 a= *(uint32_t*)&tmp[stride*0];
1894 b= *(uint32_t*)&src[stride*0];
1895 c= *(uint32_t*)&src[stride*1];
1896 a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1897 *(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1899 a= *(uint32_t*)&src[stride*2];
1900 b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1901 *(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1903 b= *(uint32_t*)&src[stride*3];
1904 c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1905 *(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1907 c= *(uint32_t*)&src[stride*4];
1908 a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1909 *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1911 a= *(uint32_t*)&src[stride*5];
1912 b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1913 *(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1915 b= *(uint32_t*)&src[stride*6];
1916 c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1917 *(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1919 c= *(uint32_t*)&src[stride*7];
1920 a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1921 *(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1923 a= *(uint32_t*)&src[stride*8];
1924 b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1925 *(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1927 *(uint32_t*)&tmp[stride*0]= c;
1931 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1935 * Deinterlaces the given block by applying a median filter to every second line.
1936 * will be called for every 8x8 block and can read & write from line 4-15,
1937 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1938 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1940 static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)
1946 "lea (%0, %1), %%"REG_a" \n\t"
1947 "lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1948 // 0 1 2 3 4 5 6 7 8 9
1949 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1951 "movq (%0), %%mm0 \n\t" //
1952 "movq (%%"REG_a", %1), %%mm2 \n\t" //
1953 "movq (%%"REG_a"), %%mm1 \n\t" //
1954 "movq %%mm0, %%mm3 \n\t"
1955 "pmaxub %%mm1, %%mm0 \n\t" //
1956 "pminub %%mm3, %%mm1 \n\t" //
1957 "pmaxub %%mm2, %%mm1 \n\t" //
1958 "pminub %%mm1, %%mm0 \n\t"
1959 "movq %%mm0, (%%"REG_a") \n\t"
1961 "movq (%0, %1, 4), %%mm0 \n\t" //
1962 "movq (%%"REG_a", %1, 2), %%mm1 \n\t" //
1963 "movq %%mm2, %%mm3 \n\t"
1964 "pmaxub %%mm1, %%mm2 \n\t" //
1965 "pminub %%mm3, %%mm1 \n\t" //
1966 "pmaxub %%mm0, %%mm1 \n\t" //
1967 "pminub %%mm1, %%mm2 \n\t"
1968 "movq %%mm2, (%%"REG_a", %1, 2) \n\t"
1970 "movq (%%"REG_d"), %%mm2 \n\t" //
1971 "movq (%%"REG_d", %1), %%mm1 \n\t" //
1972 "movq %%mm2, %%mm3 \n\t"
1973 "pmaxub %%mm0, %%mm2 \n\t" //
1974 "pminub %%mm3, %%mm0 \n\t" //
1975 "pmaxub %%mm1, %%mm0 \n\t" //
1976 "pminub %%mm0, %%mm2 \n\t"
1977 "movq %%mm2, (%%"REG_d") \n\t"
1979 "movq (%%"REG_d", %1, 2), %%mm2 \n\t" //
1980 "movq (%0, %1, 8), %%mm0 \n\t" //
1981 "movq %%mm2, %%mm3 \n\t"
1982 "pmaxub %%mm0, %%mm2 \n\t" //
1983 "pminub %%mm3, %%mm0 \n\t" //
1984 "pmaxub %%mm1, %%mm0 \n\t" //
1985 "pminub %%mm0, %%mm2 \n\t"
1986 "movq %%mm2, (%%"REG_d", %1, 2) \n\t"
1989 : : "r" (src), "r" ((long)stride)
1990 : "%"REG_a, "%"REG_d
1993 #else // MMX without MMX2
1995 "lea (%0, %1), %%"REG_a" \n\t"
1996 "lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1997 // 0 1 2 3 4 5 6 7 8 9
1998 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1999 "pxor %%mm7, %%mm7 \n\t"
2001 #define REAL_MEDIAN(a,b,c)\
2002 "movq " #a ", %%mm0 \n\t"\
2003 "movq " #b ", %%mm2 \n\t"\
2004 "movq " #c ", %%mm1 \n\t"\
2005 "movq %%mm0, %%mm3 \n\t"\
2006 "movq %%mm1, %%mm4 \n\t"\
2007 "movq %%mm2, %%mm5 \n\t"\
2008 "psubusb %%mm1, %%mm3 \n\t"\
2009 "psubusb %%mm2, %%mm4 \n\t"\
2010 "psubusb %%mm0, %%mm5 \n\t"\
2011 "pcmpeqb %%mm7, %%mm3 \n\t"\
2012 "pcmpeqb %%mm7, %%mm4 \n\t"\
2013 "pcmpeqb %%mm7, %%mm5 \n\t"\
2014 "movq %%mm3, %%mm6 \n\t"\
2015 "pxor %%mm4, %%mm3 \n\t"\
2016 "pxor %%mm5, %%mm4 \n\t"\
2017 "pxor %%mm6, %%mm5 \n\t"\
2018 "por %%mm3, %%mm1 \n\t"\
2019 "por %%mm4, %%mm2 \n\t"\
2020 "por %%mm5, %%mm0 \n\t"\
2021 "pand %%mm2, %%mm0 \n\t"\
2022 "pand %%mm1, %%mm0 \n\t"\
2023 "movq %%mm0, " #b " \n\t"
2024 #define MEDIAN(a,b,c) REAL_MEDIAN(a,b,c)
2026 MEDIAN((%0) , (%%REGa) , (%%REGa, %1))
2027 MEDIAN((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4))
2028 MEDIAN((%0, %1, 4) , (%%REGd) , (%%REGd, %1))
2029 MEDIAN((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8))
2031 : : "r" (src), "r" ((long)stride)
2032 : "%"REG_a, "%"REG_d
2038 // FIXME - there should be a way to do a few columns in parallel like w/mmx
2041 uint8_t *colsrc = src;
2044 int a, b, c, d, e, f;
2046 b = colsrc[stride ];
2047 c = colsrc[stride*2];
2051 colsrc[stride ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));
2061 * transposes and shift the given 8x8 Block into dst1 and dst2
2063 static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)
2066 "lea (%0, %1), %%"REG_a" \n\t"
2067 // 0 1 2 3 4 5 6 7 8 9
2068 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
2069 "movq (%0), %%mm0 \n\t" // 12345678
2070 "movq (%%"REG_a"), %%mm1 \n\t" // abcdefgh
2071 "movq %%mm0, %%mm2 \n\t" // 12345678
2072 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
2073 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
2075 "movq (%%"REG_a", %1), %%mm1 \n\t"
2076 "movq (%%"REG_a", %1, 2), %%mm3 \n\t"
2077 "movq %%mm1, %%mm4 \n\t"
2078 "punpcklbw %%mm3, %%mm1 \n\t"
2079 "punpckhbw %%mm3, %%mm4 \n\t"
2081 "movq %%mm0, %%mm3 \n\t"
2082 "punpcklwd %%mm1, %%mm0 \n\t"
2083 "punpckhwd %%mm1, %%mm3 \n\t"
2084 "movq %%mm2, %%mm1 \n\t"
2085 "punpcklwd %%mm4, %%mm2 \n\t"
2086 "punpckhwd %%mm4, %%mm1 \n\t"
2088 "movd %%mm0, 128(%2) \n\t"
2089 "psrlq $32, %%mm0 \n\t"
2090 "movd %%mm0, 144(%2) \n\t"
2091 "movd %%mm3, 160(%2) \n\t"
2092 "psrlq $32, %%mm3 \n\t"
2093 "movd %%mm3, 176(%2) \n\t"
2094 "movd %%mm3, 48(%3) \n\t"
2095 "movd %%mm2, 192(%2) \n\t"
2096 "movd %%mm2, 64(%3) \n\t"
2097 "psrlq $32, %%mm2 \n\t"
2098 "movd %%mm2, 80(%3) \n\t"
2099 "movd %%mm1, 96(%3) \n\t"
2100 "psrlq $32, %%mm1 \n\t"
2101 "movd %%mm1, 112(%3) \n\t"
2103 "lea (%%"REG_a", %1, 4), %%"REG_a" \n\t"
2105 "movq (%0, %1, 4), %%mm0 \n\t" // 12345678
2106 "movq (%%"REG_a"), %%mm1 \n\t" // abcdefgh
2107 "movq %%mm0, %%mm2 \n\t" // 12345678
2108 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
2109 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
2111 "movq (%%"REG_a", %1), %%mm1 \n\t"
2112 "movq (%%"REG_a", %1, 2), %%mm3 \n\t"
2113 "movq %%mm1, %%mm4 \n\t"
2114 "punpcklbw %%mm3, %%mm1 \n\t"
2115 "punpckhbw %%mm3, %%mm4 \n\t"
2117 "movq %%mm0, %%mm3 \n\t"
2118 "punpcklwd %%mm1, %%mm0 \n\t"
2119 "punpckhwd %%mm1, %%mm3 \n\t"
2120 "movq %%mm2, %%mm1 \n\t"
2121 "punpcklwd %%mm4, %%mm2 \n\t"
2122 "punpckhwd %%mm4, %%mm1 \n\t"
2124 "movd %%mm0, 132(%2) \n\t"
2125 "psrlq $32, %%mm0 \n\t"
2126 "movd %%mm0, 148(%2) \n\t"
2127 "movd %%mm3, 164(%2) \n\t"
2128 "psrlq $32, %%mm3 \n\t"
2129 "movd %%mm3, 180(%2) \n\t"
2130 "movd %%mm3, 52(%3) \n\t"
2131 "movd %%mm2, 196(%2) \n\t"
2132 "movd %%mm2, 68(%3) \n\t"
2133 "psrlq $32, %%mm2 \n\t"
2134 "movd %%mm2, 84(%3) \n\t"
2135 "movd %%mm1, 100(%3) \n\t"
2136 "psrlq $32, %%mm1 \n\t"
2137 "movd %%mm1, 116(%3) \n\t"
2140 :: "r" (src), "r" ((long)srcStride), "r" (dst1), "r" (dst2)
2146 * transposes the given 8x8 block
2148 static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)
2151 "lea (%0, %1), %%"REG_a" \n\t"
2152 "lea (%%"REG_a",%1,4), %%"REG_d" \n\t"
2153 // 0 1 2 3 4 5 6 7 8 9
2154 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
2155 "movq (%2), %%mm0 \n\t" // 12345678
2156 "movq 16(%2), %%mm1 \n\t" // abcdefgh
2157 "movq %%mm0, %%mm2 \n\t" // 12345678
2158 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
2159 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
2161 "movq 32(%2), %%mm1 \n\t"
2162 "movq 48(%2), %%mm3 \n\t"
2163 "movq %%mm1, %%mm4 \n\t"
2164 "punpcklbw %%mm3, %%mm1 \n\t"
2165 "punpckhbw %%mm3, %%mm4 \n\t"
2167 "movq %%mm0, %%mm3 \n\t"
2168 "punpcklwd %%mm1, %%mm0 \n\t"
2169 "punpckhwd %%mm1, %%mm3 \n\t"
2170 "movq %%mm2, %%mm1 \n\t"
2171 "punpcklwd %%mm4, %%mm2 \n\t"
2172 "punpckhwd %%mm4, %%mm1 \n\t"
2174 "movd %%mm0, (%0) \n\t"
2175 "psrlq $32, %%mm0 \n\t"
2176 "movd %%mm0, (%%"REG_a") \n\t"
2177 "movd %%mm3, (%%"REG_a", %1) \n\t"
2178 "psrlq $32, %%mm3 \n\t"
2179 "movd %%mm3, (%%"REG_a", %1, 2) \n\t"
2180 "movd %%mm2, (%0, %1, 4) \n\t"
2181 "psrlq $32, %%mm2 \n\t"
2182 "movd %%mm2, (%%"REG_d") \n\t"
2183 "movd %%mm1, (%%"REG_d", %1) \n\t"
2184 "psrlq $32, %%mm1 \n\t"
2185 "movd %%mm1, (%%"REG_d", %1, 2) \n\t"
2188 "movq 64(%2), %%mm0 \n\t" // 12345678
2189 "movq 80(%2), %%mm1 \n\t" // abcdefgh
2190 "movq %%mm0, %%mm2 \n\t" // 12345678
2191 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
2192 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
2194 "movq 96(%2), %%mm1 \n\t"
2195 "movq 112(%2), %%mm3 \n\t"
2196 "movq %%mm1, %%mm4 \n\t"
2197 "punpcklbw %%mm3, %%mm1 \n\t"
2198 "punpckhbw %%mm3, %%mm4 \n\t"
2200 "movq %%mm0, %%mm3 \n\t"
2201 "punpcklwd %%mm1, %%mm0 \n\t"
2202 "punpckhwd %%mm1, %%mm3 \n\t"
2203 "movq %%mm2, %%mm1 \n\t"
2204 "punpcklwd %%mm4, %%mm2 \n\t"
2205 "punpckhwd %%mm4, %%mm1 \n\t"
2207 "movd %%mm0, 4(%0) \n\t"
2208 "psrlq $32, %%mm0 \n\t"
2209 "movd %%mm0, 4(%%"REG_a") \n\t"
2210 "movd %%mm3, 4(%%"REG_a", %1) \n\t"
2211 "psrlq $32, %%mm3 \n\t"
2212 "movd %%mm3, 4(%%"REG_a", %1, 2) \n\t"
2213 "movd %%mm2, 4(%0, %1, 4) \n\t"
2214 "psrlq $32, %%mm2 \n\t"
2215 "movd %%mm2, 4(%%"REG_d") \n\t"
2216 "movd %%mm1, 4(%%"REG_d", %1) \n\t"
2217 "psrlq $32, %%mm1 \n\t"
2218 "movd %%mm1, 4(%%"REG_d", %1, 2) \n\t"
2220 :: "r" (dst), "r" ((long)dstStride), "r" (src)
2221 : "%"REG_a, "%"REG_d
2225 //static long test=0;
2227 #ifndef HAVE_ALTIVEC
2228 static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
2229 uint8_t *tempBlured, uint32_t *tempBluredPast, int *maxNoise)
2231 // to save a register (FIXME do this outside of the loops)
2232 tempBluredPast[127]= maxNoise[0];
2233 tempBluredPast[128]= maxNoise[1];
2234 tempBluredPast[129]= maxNoise[2];
2236 #define FAST_L2_DIFF
2237 //#define L1_DIFF //u should change the thresholds too if u try that one
2238 #if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
2240 "lea (%2, %2, 2), %%"REG_a" \n\t" // 3*stride
2241 "lea (%2, %2, 4), %%"REG_d" \n\t" // 5*stride
2242 "lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
2243 // 0 1 2 3 4 5 6 7 8 9
2244 // %x %x+%2 %x+2%2 %x+eax %x+4%2 %x+edx %x+2eax %x+ecx %x+8%2
2246 #ifdef L1_DIFF //needs mmx2
2247 "movq (%0), %%mm0 \n\t" // L0
2248 "psadbw (%1), %%mm0 \n\t" // |L0-R0|
2249 "movq (%0, %2), %%mm1 \n\t" // L1
2250 "psadbw (%1, %2), %%mm1 \n\t" // |L1-R1|
2251 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2252 "psadbw (%1, %2, 2), %%mm2 \n\t" // |L2-R2|
2253 "movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
2254 "psadbw (%1, %%"REG_a"), %%mm3 \n\t" // |L3-R3|
2256 "movq (%0, %2, 4), %%mm4 \n\t" // L4
2257 "paddw %%mm1, %%mm0 \n\t"
2258 "psadbw (%1, %2, 4), %%mm4 \n\t" // |L4-R4|
2259 "movq (%0, %%"REG_d"), %%mm5 \n\t" // L5
2260 "paddw %%mm2, %%mm0 \n\t"
2261 "psadbw (%1, %%"REG_d"), %%mm5 \n\t" // |L5-R5|
2262 "movq (%0, %%"REG_a", 2), %%mm6 \n\t" // L6
2263 "paddw %%mm3, %%mm0 \n\t"
2264 "psadbw (%1, %%"REG_a", 2), %%mm6 \n\t" // |L6-R6|
2265 "movq (%0, %%"REG_c"), %%mm7 \n\t" // L7
2266 "paddw %%mm4, %%mm0 \n\t"
2267 "psadbw (%1, %%"REG_c"), %%mm7 \n\t" // |L7-R7|
2268 "paddw %%mm5, %%mm6 \n\t"
2269 "paddw %%mm7, %%mm6 \n\t"
2270 "paddw %%mm6, %%mm0 \n\t"
2272 #if defined (FAST_L2_DIFF)
2273 "pcmpeqb %%mm7, %%mm7 \n\t"
2274 "movq "MANGLE(b80)", %%mm6 \n\t"
2275 "pxor %%mm0, %%mm0 \n\t"
2276 #define REAL_L2_DIFF_CORE(a, b)\
2277 "movq " #a ", %%mm5 \n\t"\
2278 "movq " #b ", %%mm2 \n\t"\
2279 "pxor %%mm7, %%mm2 \n\t"\
2280 PAVGB(%%mm2, %%mm5)\
2281 "paddb %%mm6, %%mm5 \n\t"\
2282 "movq %%mm5, %%mm2 \n\t"\
2283 "psllw $8, %%mm5 \n\t"\
2284 "pmaddwd %%mm5, %%mm5 \n\t"\
2285 "pmaddwd %%mm2, %%mm2 \n\t"\
2286 "paddd %%mm2, %%mm5 \n\t"\
2287 "psrld $14, %%mm5 \n\t"\
2288 "paddd %%mm5, %%mm0 \n\t"
2290 #else //defined (FAST_L2_DIFF)
2291 "pxor %%mm7, %%mm7 \n\t"
2292 "pxor %%mm0, %%mm0 \n\t"
2293 #define REAL_L2_DIFF_CORE(a, b)\
2294 "movq " #a ", %%mm5 \n\t"\
2295 "movq " #b ", %%mm2 \n\t"\
2296 "movq %%mm5, %%mm1 \n\t"\
2297 "movq %%mm2, %%mm3 \n\t"\
2298 "punpcklbw %%mm7, %%mm5 \n\t"\
2299 "punpckhbw %%mm7, %%mm1 \n\t"\
2300 "punpcklbw %%mm7, %%mm2 \n\t"\
2301 "punpckhbw %%mm7, %%mm3 \n\t"\
2302 "psubw %%mm2, %%mm5 \n\t"\
2303 "psubw %%mm3, %%mm1 \n\t"\
2304 "pmaddwd %%mm5, %%mm5 \n\t"\
2305 "pmaddwd %%mm1, %%mm1 \n\t"\
2306 "paddd %%mm1, %%mm5 \n\t"\
2307 "paddd %%mm5, %%mm0 \n\t"
2309 #endif //defined (FAST_L2_DIFF)
2311 #define L2_DIFF_CORE(a, b) REAL_L2_DIFF_CORE(a, b)
2313 L2_DIFF_CORE((%0) , (%1))
2314 L2_DIFF_CORE((%0, %2) , (%1, %2))
2315 L2_DIFF_CORE((%0, %2, 2) , (%1, %2, 2))
2316 L2_DIFF_CORE((%0, %%REGa) , (%1, %%REGa))
2317 L2_DIFF_CORE((%0, %2, 4) , (%1, %2, 4))
2318 L2_DIFF_CORE((%0, %%REGd) , (%1, %%REGd))
2319 L2_DIFF_CORE((%0, %%REGa,2), (%1, %%REGa,2))
2320 L2_DIFF_CORE((%0, %%REGc) , (%1, %%REGc))
2324 "movq %%mm0, %%mm4 \n\t"
2325 "psrlq $32, %%mm0 \n\t"
2326 "paddd %%mm0, %%mm4 \n\t"
2327 "movd %%mm4, %%ecx \n\t"
2328 "shll $2, %%ecx \n\t"
2329 "mov %3, %%"REG_d" \n\t"
2330 "addl -4(%%"REG_d"), %%ecx \n\t"
2331 "addl 4(%%"REG_d"), %%ecx \n\t"
2332 "addl -1024(%%"REG_d"), %%ecx \n\t"
2333 "addl $4, %%ecx \n\t"
2334 "addl 1024(%%"REG_d"), %%ecx \n\t"
2335 "shrl $3, %%ecx \n\t"
2336 "movl %%ecx, (%%"REG_d") \n\t"
2338 // "mov %3, %%"REG_c" \n\t"
2339 // "mov %%"REG_c", test \n\t"
2341 "cmpl 512(%%"REG_d"), %%ecx \n\t"
2343 "cmpl 516(%%"REG_d"), %%ecx \n\t"
2346 "lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
2347 "lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
2348 "movq (%0), %%mm0 \n\t" // L0
2349 "movq (%0, %2), %%mm1 \n\t" // L1
2350 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2351 "movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
2352 "movq (%0, %2, 4), %%mm4 \n\t" // L4
2353 "movq (%0, %%"REG_d"), %%mm5 \n\t" // L5
2354 "movq (%0, %%"REG_a", 2), %%mm6 \n\t" // L6
2355 "movq (%0, %%"REG_c"), %%mm7 \n\t" // L7
2356 "movq %%mm0, (%1) \n\t" // L0
2357 "movq %%mm1, (%1, %2) \n\t" // L1
2358 "movq %%mm2, (%1, %2, 2) \n\t" // L2
2359 "movq %%mm3, (%1, %%"REG_a") \n\t" // L3
2360 "movq %%mm4, (%1, %2, 4) \n\t" // L4
2361 "movq %%mm5, (%1, %%"REG_d") \n\t" // L5
2362 "movq %%mm6, (%1, %%"REG_a", 2) \n\t" // L6
2363 "movq %%mm7, (%1, %%"REG_c") \n\t" // L7
2367 "lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
2368 "lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
2369 "movq (%0), %%mm0 \n\t" // L0
2370 PAVGB((%1), %%mm0) // L0
2371 "movq (%0, %2), %%mm1 \n\t" // L1
2372 PAVGB((%1, %2), %%mm1) // L1
2373 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2374 PAVGB((%1, %2, 2), %%mm2) // L2
2375 "movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
2376 PAVGB((%1, %%REGa), %%mm3) // L3
2377 "movq (%0, %2, 4), %%mm4 \n\t" // L4
2378 PAVGB((%1, %2, 4), %%mm4) // L4
2379 "movq (%0, %%"REG_d"), %%mm5 \n\t" // L5
2380 PAVGB((%1, %%REGd), %%mm5) // L5
2381 "movq (%0, %%"REG_a", 2), %%mm6 \n\t" // L6
2382 PAVGB((%1, %%REGa, 2), %%mm6) // L6
2383 "movq (%0, %%"REG_c"), %%mm7 \n\t" // L7
2384 PAVGB((%1, %%REGc), %%mm7) // L7
2385 "movq %%mm0, (%1) \n\t" // R0
2386 "movq %%mm1, (%1, %2) \n\t" // R1
2387 "movq %%mm2, (%1, %2, 2) \n\t" // R2
2388 "movq %%mm3, (%1, %%"REG_a") \n\t" // R3
2389 "movq %%mm4, (%1, %2, 4) \n\t" // R4
2390 "movq %%mm5, (%1, %%"REG_d") \n\t" // R5
2391 "movq %%mm6, (%1, %%"REG_a", 2) \n\t" // R6
2392 "movq %%mm7, (%1, %%"REG_c") \n\t" // R7
2393 "movq %%mm0, (%0) \n\t" // L0
2394 "movq %%mm1, (%0, %2) \n\t" // L1
2395 "movq %%mm2, (%0, %2, 2) \n\t" // L2
2396 "movq %%mm3, (%0, %%"REG_a") \n\t" // L3
2397 "movq %%mm4, (%0, %2, 4) \n\t" // L4
2398 "movq %%mm5, (%0, %%"REG_d") \n\t" // L5
2399 "movq %%mm6, (%0, %%"REG_a", 2) \n\t" // L6
2400 "movq %%mm7, (%0, %%"REG_c") \n\t" // L7
2404 "cmpl 508(%%"REG_d"), %%ecx \n\t"
2407 "lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
2408 "lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
2409 "movq (%0), %%mm0 \n\t" // L0
2410 "movq (%0, %2), %%mm1 \n\t" // L1
2411 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2412 "movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
2413 "movq (%1), %%mm4 \n\t" // R0
2414 "movq (%1, %2), %%mm5 \n\t" // R1
2415 "movq (%1, %2, 2), %%mm6 \n\t" // R2
2416 "movq (%1, %%"REG_a"), %%mm7 \n\t" // R3
2425 "movq %%mm0, (%1) \n\t" // R0
2426 "movq %%mm1, (%1, %2) \n\t" // R1
2427 "movq %%mm2, (%1, %2, 2) \n\t" // R2
2428 "movq %%mm3, (%1, %%"REG_a") \n\t" // R3
2429 "movq %%mm0, (%0) \n\t" // L0
2430 "movq %%mm1, (%0, %2) \n\t" // L1
2431 "movq %%mm2, (%0, %2, 2) \n\t" // L2
2432 "movq %%mm3, (%0, %%"REG_a") \n\t" // L3
2434 "movq (%0, %2, 4), %%mm0 \n\t" // L4
2435 "movq (%0, %%"REG_d"), %%mm1 \n\t" // L5
2436 "movq (%0, %%"REG_a", 2), %%mm2 \n\t" // L6
2437 "movq (%0, %%"REG_c"), %%mm3 \n\t" // L7
2438 "movq (%1, %2, 4), %%mm4 \n\t" // R4
2439 "movq (%1, %%"REG_d"), %%mm5 \n\t" // R5
2440 "movq (%1, %%"REG_a", 2), %%mm6 \n\t" // R6
2441 "movq (%1, %%"REG_c"), %%mm7 \n\t" // R7
2450 "movq %%mm0, (%1, %2, 4) \n\t" // R4
2451 "movq %%mm1, (%1, %%"REG_d") \n\t" // R5
2452 "movq %%mm2, (%1, %%"REG_a", 2) \n\t" // R6
2453 "movq %%mm3, (%1, %%"REG_c") \n\t" // R7
2454 "movq %%mm0, (%0, %2, 4) \n\t" // L4
2455 "movq %%mm1, (%0, %%"REG_d") \n\t" // L5
2456 "movq %%mm2, (%0, %%"REG_a", 2) \n\t" // L6
2457 "movq %%mm3, (%0, %%"REG_c") \n\t" // L7
2461 "lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
2462 "lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
2463 "movq (%0), %%mm0 \n\t" // L0
2464 "movq (%0, %2), %%mm1 \n\t" // L1
2465 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2466 "movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
2467 "movq (%1), %%mm4 \n\t" // R0
2468 "movq (%1, %2), %%mm5 \n\t" // R1
2469 "movq (%1, %2, 2), %%mm6 \n\t" // R2
2470 "movq (%1, %%"REG_a"), %%mm7 \n\t" // R3
2483 "movq %%mm0, (%1) \n\t" // R0
2484 "movq %%mm1, (%1, %2) \n\t" // R1
2485 "movq %%mm2, (%1, %2, 2) \n\t" // R2
2486 "movq %%mm3, (%1, %%"REG_a") \n\t" // R3
2487 "movq %%mm0, (%0) \n\t" // L0
2488 "movq %%mm1, (%0, %2) \n\t" // L1
2489 "movq %%mm2, (%0, %2, 2) \n\t" // L2
2490 "movq %%mm3, (%0, %%"REG_a") \n\t" // L3
2492 "movq (%0, %2, 4), %%mm0 \n\t" // L4
2493 "movq (%0, %%"REG_d"), %%mm1 \n\t" // L5
2494 "movq (%0, %%"REG_a", 2), %%mm2 \n\t" // L6
2495 "movq (%0, %%"REG_c"), %%mm3 \n\t" // L7
2496 "movq (%1, %2, 4), %%mm4 \n\t" // R4
2497 "movq (%1, %%"REG_d"), %%mm5 \n\t" // R5
2498 "movq (%1, %%"REG_a", 2), %%mm6 \n\t" // R6
2499 "movq (%1, %%"REG_c"), %%mm7 \n\t" // R7
2512 "movq %%mm0, (%1, %2, 4) \n\t" // R4
2513 "movq %%mm1, (%1, %%"REG_d") \n\t" // R5
2514 "movq %%mm2, (%1, %%"REG_a", 2) \n\t" // R6
2515 "movq %%mm3, (%1, %%"REG_c") \n\t" // R7
2516 "movq %%mm0, (%0, %2, 4) \n\t" // L4
2517 "movq %%mm1, (%0, %%"REG_d") \n\t" // L5
2518 "movq %%mm2, (%0, %%"REG_a", 2) \n\t" // L6
2519 "movq %%mm3, (%0, %%"REG_c") \n\t" // L7
2523 :: "r" (src), "r" (tempBlured), "r"((long)stride), "m" (tempBluredPast)
2524 : "%"REG_a, "%"REG_d, "%"REG_c, "memory"
2526 #else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
2538 int ref= tempBlured[ x + y*stride ];
2539 int cur= src[ x + y*stride ];
2541 // if(x==0 || x==7) d1+= d1>>1;
2542 // if(y==0 || y==7) d1+= d1>>1;
2551 +(*(tempBluredPast-256))
2552 +(*(tempBluredPast-1))+ (*(tempBluredPast+1))
2553 +(*(tempBluredPast+256))
2556 // ((*tempBluredPast)*3 + d + 2)>>2;
2561 64 32 16 8 4 2 1 (1)
2562 64 48 36 27 20 15 11 (33) (approx)
2563 64 56 49 43 37 33 29 (200) (approx)
2574 int ref= tempBlured[ x + y*stride ];
2575 int cur= src[ x + y*stride ];
2576 tempBlured[ x + y*stride ]=
2577 src[ x + y*stride ]=
2589 tempBlured[ x + y*stride ]= src[ x + y*stride ];
2603 int ref= tempBlured[ x + y*stride ];
2604 int cur= src[ x + y*stride ];
2605 tempBlured[ x + y*stride ]=
2606 src[ x + y*stride ]=
2607 (ref*7 + cur + 4)>>3;
2618 int ref= tempBlured[ x + y*stride ];
2619 int cur= src[ x + y*stride ];
2620 tempBlured[ x + y*stride ]=
2621 src[ x + y*stride ]=
2622 (ref*3 + cur + 2)>>2;
2628 #endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
2630 #endif //HAVE_ALTIVEC
2634 * accurate deblock filter
2636 static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, PPContext *c){
2637 int64_t dc_mask, eq_mask, both_masks;
2638 int64_t sums[10*8*2];
2639 src+= step*3; // src points to begin of the 8x8 Block
2642 "movq %0, %%mm7 \n\t"
2643 "movq %1, %%mm6 \n\t"
2644 : : "m" (c->mmxDcOffset[c->nonBQP]), "m" (c->mmxDcThreshold[c->nonBQP])
2648 "lea (%2, %3), %%"REG_a" \n\t"
2649 // 0 1 2 3 4 5 6 7 8 9
2650 // %1 eax eax+%2 eax+2%2 %1+4%2 ecx ecx+%2 ecx+2%2 %1+8%2 ecx+4%2
2652 "movq (%2), %%mm0 \n\t"
2653 "movq (%%"REG_a"), %%mm1 \n\t"
2654 "movq %%mm1, %%mm3 \n\t"
2655 "movq %%mm1, %%mm4 \n\t"
2656 "psubb %%mm1, %%mm0 \n\t" // mm0 = differnece
2657 "paddb %%mm7, %%mm0 \n\t"
2658 "pcmpgtb %%mm6, %%mm0 \n\t"
2660 "movq (%%"REG_a",%3), %%mm2 \n\t"
2661 PMAXUB(%%mm2, %%mm4)
2662 PMINUB(%%mm2, %%mm3, %%mm5)
2663 "psubb %%mm2, %%mm1 \n\t"
2664 "paddb %%mm7, %%mm1 \n\t"
2665 "pcmpgtb %%mm6, %%mm1 \n\t"
2666 "paddb %%mm1, %%mm0 \n\t"
2668 "movq (%%"REG_a", %3, 2), %%mm1 \n\t"
2669 PMAXUB(%%mm1, %%mm4)
2670 PMINUB(%%mm1, %%mm3, %%mm5)
2671 "psubb %%mm1, %%mm2 \n\t"
2672 "paddb %%mm7, %%mm2 \n\t"
2673 "pcmpgtb %%mm6, %%mm2 \n\t"
2674 "paddb %%mm2, %%mm0 \n\t"
2676 "lea (%%"REG_a", %3, 4), %%"REG_a" \n\t"
2678 "movq (%2, %3, 4), %%mm2 \n\t"
2679 PMAXUB(%%mm2, %%mm4)
2680 PMINUB(%%mm2, %%mm3, %%mm5)
2681 "psubb %%mm2, %%mm1 \n\t"
2682 "paddb %%mm7, %%mm1 \n\t"
2683 "pcmpgtb %%mm6, %%mm1 \n\t"
2684 "paddb %%mm1, %%mm0 \n\t"
2686 "movq (%%"REG_a"), %%mm1 \n\t"
2687 PMAXUB(%%mm1, %%mm4)
2688 PMINUB(%%mm1, %%mm3, %%mm5)
2689 "psubb %%mm1, %%mm2 \n\t"
2690 "paddb %%mm7, %%mm2 \n\t"
2691 "pcmpgtb %%mm6, %%mm2 \n\t"
2692 "paddb %%mm2, %%mm0 \n\t"
2694 "movq (%%"REG_a", %3), %%mm2 \n\t"
2695 PMAXUB(%%mm2, %%mm4)
2696 PMINUB(%%mm2, %%mm3, %%mm5)
2697 "psubb %%mm2, %%mm1 \n\t"
2698 "paddb %%mm7, %%mm1 \n\t"
2699 "pcmpgtb %%mm6, %%mm1 \n\t"
2700 "paddb %%mm1, %%mm0 \n\t"
2702 "movq (%%"REG_a", %3, 2), %%mm1 \n\t"
2703 PMAXUB(%%mm1, %%mm4)
2704 PMINUB(%%mm1, %%mm3, %%mm5)
2705 "psubb %%mm1, %%mm2 \n\t"
2706 "paddb %%mm7, %%mm2 \n\t"
2707 "pcmpgtb %%mm6, %%mm2 \n\t"
2708 "paddb %%mm2, %%mm0 \n\t"
2710 "movq (%2, %3, 8), %%mm2 \n\t"
2711 PMAXUB(%%mm2, %%mm4)
2712 PMINUB(%%mm2, %%mm3, %%mm5)
2713 "psubb %%mm2, %%mm1 \n\t"
2714 "paddb %%mm7, %%mm1 \n\t"
2715 "pcmpgtb %%mm6, %%mm1 \n\t"
2716 "paddb %%mm1, %%mm0 \n\t"
2718 "movq (%%"REG_a", %3, 4), %%mm1 \n\t"
2719 "psubb %%mm1, %%mm2 \n\t"
2720 "paddb %%mm7, %%mm2 \n\t"
2721 "pcmpgtb %%mm6, %%mm2 \n\t"
2722 "paddb %%mm2, %%mm0 \n\t"
2723 "psubusb %%mm3, %%mm4 \n\t"
2725 "pxor %%mm6, %%mm6 \n\t"
2726 "movq %4, %%mm7 \n\t" // QP,..., QP
2727 "paddusb %%mm7, %%mm7 \n\t" // 2QP ... 2QP
2728 "psubusb %%mm4, %%mm7 \n\t" // Diff >=2QP -> 0
2729 "pcmpeqb %%mm6, %%mm7 \n\t" // Diff < 2QP -> 0
2730 "pcmpeqb %%mm6, %%mm7 \n\t" // Diff < 2QP -> 0
2731 "movq %%mm7, %1 \n\t"
2733 "movq %5, %%mm7 \n\t"
2734 "punpcklbw %%mm7, %%mm7 \n\t"
2735 "punpcklbw %%mm7, %%mm7 \n\t"
2736 "punpcklbw %%mm7, %%mm7 \n\t"
2737 "psubb %%mm0, %%mm6 \n\t"
2738 "pcmpgtb %%mm7, %%mm6 \n\t"
2739 "movq %%mm6, %0 \n\t"
2741 : "=m" (eq_mask), "=m" (dc_mask)
2742 : "r" (src), "r" ((long)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)
2746 both_masks = dc_mask & eq_mask;
2749 long offset= -8*step;
2750 int64_t *temp_sums= sums;
2753 "movq %2, %%mm0 \n\t" // QP,..., QP
2754 "pxor %%mm4, %%mm4 \n\t"
2756 "movq (%0), %%mm6 \n\t"
2757 "movq (%0, %1), %%mm5 \n\t"
2758 "movq %%mm5, %%mm1 \n\t"
2759 "movq %%mm6, %%mm2 \n\t"
2760 "psubusb %%mm6, %%mm5 \n\t"
2761 "psubusb %%mm1, %%mm2 \n\t"
2762 "por %%mm5, %%mm2 \n\t" // ABS Diff of lines
2763 "psubusb %%mm2, %%mm0 \n\t" // diff >= QP -> 0
2764 "pcmpeqb %%mm4, %%mm0 \n\t" // diff >= QP -> FF
2766 "pxor %%mm6, %%mm1 \n\t"
2767 "pand %%mm0, %%mm1 \n\t"
2768 "pxor %%mm1, %%mm6 \n\t"
2771 "movq (%0, %1, 8), %%mm5 \n\t"
2772 "add %1, %0 \n\t" // %0 points to line 1 not 0
2773 "movq (%0, %1, 8), %%mm7 \n\t"
2774 "movq %%mm5, %%mm1 \n\t"
2775 "movq %%mm7, %%mm2 \n\t"
2776 "psubusb %%mm7, %%mm5 \n\t"
2777 "psubusb %%mm1, %%mm2 \n\t"
2778 "por %%mm5, %%mm2 \n\t" // ABS Diff of lines
2779 "movq %2, %%mm0 \n\t" // QP,..., QP
2780 "psubusb %%mm2, %%mm0 \n\t" // diff >= QP -> 0
2781 "pcmpeqb %%mm4, %%mm0 \n\t" // diff >= QP -> FF
2783 "pxor %%mm7, %%mm1 \n\t"
2784 "pand %%mm0, %%mm1 \n\t"
2785 "pxor %%mm1, %%mm7 \n\t"
2787 "movq %%mm6, %%mm5 \n\t"
2788 "punpckhbw %%mm4, %%mm6 \n\t"
2789 "punpcklbw %%mm4, %%mm5 \n\t"
2790 // 4:0 5/6:First 7:Last
2792 "movq %%mm5, %%mm0 \n\t"
2793 "movq %%mm6, %%mm1 \n\t"
2794 "psllw $2, %%mm0 \n\t"
2795 "psllw $2, %%mm1 \n\t"
2796 "paddw "MANGLE(w04)", %%mm0 \n\t"
2797 "paddw "MANGLE(w04)", %%mm1 \n\t"
2800 "movq (%0), %%mm2 \n\t"\
2801 "movq (%0), %%mm3 \n\t"\
2803 "punpcklbw %%mm4, %%mm2 \n\t"\
2804 "punpckhbw %%mm4, %%mm3 \n\t"\
2805 "paddw %%mm2, %%mm0 \n\t"\
2806 "paddw %%mm3, %%mm1 \n\t"
2809 "movq (%0), %%mm2 \n\t"\
2810 "movq (%0), %%mm3 \n\t"\
2812 "punpcklbw %%mm4, %%mm2 \n\t"\
2813 "punpckhbw %%mm4, %%mm3 \n\t"\
2814 "psubw %%mm2, %%mm0 \n\t"\
2815 "psubw %%mm3, %%mm1 \n\t"
2821 "movq %%mm0, (%3) \n\t"
2822 "movq %%mm1, 8(%3) \n\t"
2825 "psubw %%mm5, %%mm0 \n\t"
2826 "psubw %%mm6, %%mm1 \n\t"
2827 "movq %%mm0, 16(%3) \n\t"
2828 "movq %%mm1, 24(%3) \n\t"
2831 "psubw %%mm5, %%mm0 \n\t"
2832 "psubw %%mm6, %%mm1 \n\t"
2833 "movq %%mm0, 32(%3) \n\t"
2834 "movq %%mm1, 40(%3) \n\t"
2837 "psubw %%mm5, %%mm0 \n\t"
2838 "psubw %%mm6, %%mm1 \n\t"
2839 "movq %%mm0, 48(%3) \n\t"
2840 "movq %%mm1, 56(%3) \n\t"
2843 "psubw %%mm5, %%mm0 \n\t"
2844 "psubw %%mm6, %%mm1 \n\t"
2845 "movq %%mm0, 64(%3) \n\t"
2846 "movq %%mm1, 72(%3) \n\t"
2848 "movq %%mm7, %%mm6 \n\t"
2849 "punpckhbw %%mm4, %%mm7 \n\t"
2850 "punpcklbw %%mm4, %%mm6 \n\t"
2856 "movq %%mm0, 80(%3) \n\t"
2857 "movq %%mm1, 88(%3) \n\t"
2860 "paddw %%mm6, %%mm0 \n\t"
2861 "paddw %%mm7, %%mm1 \n\t"
2862 "movq %%mm0, 96(%3) \n\t"
2863 "movq %%mm1, 104(%3) \n\t"
2866 "paddw %%mm6, %%mm0 \n\t"
2867 "paddw %%mm7, %%mm1 \n\t"
2868 "movq %%mm0, 112(%3) \n\t"
2869 "movq %%mm1, 120(%3) \n\t"
2872 "paddw %%mm6, %%mm0 \n\t"
2873 "paddw %%mm7, %%mm1 \n\t"
2874 "movq %%mm0, 128(%3) \n\t"
2875 "movq %%mm1, 136(%3) \n\t"
2878 "paddw %%mm6, %%mm0 \n\t"
2879 "paddw %%mm7, %%mm1 \n\t"
2880 "movq %%mm0, 144(%3) \n\t"
2881 "movq %%mm1, 152(%3) \n\t"
2883 "mov %4, %0 \n\t" //FIXME
2886 : "r" ((long)step), "m" (c->pQPb), "r"(sums), "g"(src)
2889 src+= step; // src points to begin of the 8x8 Block
2892 "movq %4, %%mm6 \n\t"
2893 "pcmpeqb %%mm5, %%mm5 \n\t"
2894 "pxor %%mm6, %%mm5 \n\t"
2895 "pxor %%mm7, %%mm7 \n\t"
2898 "movq (%1), %%mm0 \n\t"
2899 "movq 8(%1), %%mm1 \n\t"
2900 "paddw 32(%1), %%mm0 \n\t"
2901 "paddw 40(%1), %%mm1 \n\t"
2902 "movq (%0, %3), %%mm2 \n\t"
2903 "movq %%mm2, %%mm3 \n\t"
2904 "movq %%mm2, %%mm4 \n\t"
2905 "punpcklbw %%mm7, %%mm2 \n\t"
2906 "punpckhbw %%mm7, %%mm3 \n\t"
2907 "paddw %%mm2, %%mm0 \n\t"
2908 "paddw %%mm3, %%mm1 \n\t"
2909 "paddw %%mm2, %%mm0 \n\t"
2910 "paddw %%mm3, %%mm1 \n\t"
2911 "psrlw $4, %%mm0 \n\t"
2912 "psrlw $4, %%mm1 \n\t"
2913 "packuswb %%mm1, %%mm0 \n\t"
2914 "pand %%mm6, %%mm0 \n\t"
2915 "pand %%mm5, %%mm4 \n\t"
2916 "por %%mm4, %%mm0 \n\t"
2917 "movq %%mm0, (%0, %3) \n\t"
2922 : "+r"(offset), "+r"(temp_sums)
2923 : "r" ((long)step), "r"(src - offset), "m"(both_masks)
2926 src+= step; // src points to begin of the 8x8 Block
2928 if(eq_mask != -1LL){
2929 uint8_t *temp_src= src;
2931 "pxor %%mm7, %%mm7 \n\t"
2932 "lea -40(%%"REG_SP"), %%"REG_c" \n\t" // make space for 4 8-byte vars
2933 "and "ALIGN_MASK", %%"REG_c" \n\t" // align
2934 // 0 1 2 3 4 5 6 7 8 9
2935 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1 %1+8%1 ecx+4%1
2937 "movq (%0), %%mm0 \n\t"
2938 "movq %%mm0, %%mm1 \n\t"
2939 "punpcklbw %%mm7, %%mm0 \n\t" // low part of line 0
2940 "punpckhbw %%mm7, %%mm1 \n\t" // high part of line 0
2942 "movq (%0, %1), %%mm2 \n\t"
2943 "lea (%0, %1, 2), %%"REG_a" \n\t"
2944 "movq %%mm2, %%mm3 \n\t"
2945 "punpcklbw %%mm7, %%mm2 \n\t" // low part of line 1
2946 "punpckhbw %%mm7, %%mm3 \n\t" // high part of line 1
2948 "movq (%%"REG_a"), %%mm4 \n\t"
2949 "movq %%mm4, %%mm5 \n\t"
2950 "punpcklbw %%mm7, %%mm4 \n\t" // low part of line 2
2951 "punpckhbw %%mm7, %%mm5 \n\t" // high part of line 2
2953 "paddw %%mm0, %%mm0 \n\t" // 2L0
2954 "paddw %%mm1, %%mm1 \n\t" // 2H0
2955 "psubw %%mm4, %%mm2 \n\t" // L1 - L2
2956 "psubw %%mm5, %%mm3 \n\t" // H1 - H2
2957 "psubw %%mm2, %%mm0 \n\t" // 2L0 - L1 + L2
2958 "psubw %%mm3, %%mm1 \n\t" // 2H0 - H1 + H2
2960 "psllw $2, %%mm2 \n\t" // 4L1 - 4L2
2961 "psllw $2, %%mm3 \n\t" // 4H1 - 4H2
2962 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2
2963 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2
2965 "movq (%%"REG_a", %1), %%mm2 \n\t"
2966 "movq %%mm2, %%mm3 \n\t"
2967 "punpcklbw %%mm7, %%mm2 \n\t" // L3
2968 "punpckhbw %%mm7, %%mm3 \n\t" // H3
2970 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - L3
2971 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - H3
2972 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2973 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2974 "movq %%mm0, (%%"REG_c") \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2975 "movq %%mm1, 8(%%"REG_c") \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2977 "movq (%%"REG_a", %1, 2), %%mm0 \n\t"
2978 "movq %%mm0, %%mm1 \n\t"
2979 "punpcklbw %%mm7, %%mm0 \n\t" // L4
2980 "punpckhbw %%mm7, %%mm1 \n\t" // H4
2982 "psubw %%mm0, %%mm2 \n\t" // L3 - L4
2983 "psubw %%mm1, %%mm3 \n\t" // H3 - H4
2984 "movq %%mm2, 16(%%"REG_c") \n\t" // L3 - L4
2985 "movq %%mm3, 24(%%"REG_c") \n\t" // H3 - H4
2986 "paddw %%mm4, %%mm4 \n\t" // 2L2
2987 "paddw %%mm5, %%mm5 \n\t" // 2H2
2988 "psubw %%mm2, %%mm4 \n\t" // 2L2 - L3 + L4
2989 "psubw %%mm3, %%mm5 \n\t" // 2H2 - H3 + H4
2991 "lea (%%"REG_a", %1), %0 \n\t"
2992 "psllw $2, %%mm2 \n\t" // 4L3 - 4L4
2993 "psllw $2, %%mm3 \n\t" // 4H3 - 4H4
2994 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4
2995 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4
2997 "movq (%0, %1, 2), %%mm2 \n\t"
2998 "movq %%mm2, %%mm3 \n\t"
2999 "punpcklbw %%mm7, %%mm2 \n\t" // L5
3000 "punpckhbw %%mm7, %%mm3 \n\t" // H5
3001 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - L5
3002 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - H5
3003 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - 2L5
3004 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - 2H5
3006 "movq (%%"REG_a", %1, 4), %%mm6 \n\t"
3007 "punpcklbw %%mm7, %%mm6 \n\t" // L6
3008 "psubw %%mm6, %%mm2 \n\t" // L5 - L6
3009 "movq (%%"REG_a", %1, 4), %%mm6 \n\t"
3010 "punpckhbw %%mm7, %%mm6 \n\t" // H6
3011 "psubw %%mm6, %%mm3 \n\t" // H5 - H6
3013 "paddw %%mm0, %%mm0 \n\t" // 2L4
3014 "paddw %%mm1, %%mm1 \n\t" // 2H4
3015 "psubw %%mm2, %%mm0 \n\t" // 2L4 - L5 + L6
3016 "psubw %%mm3, %%mm1 \n\t" // 2H4 - H5 + H6
3018 "psllw $2, %%mm2 \n\t" // 4L5 - 4L6
3019 "psllw $2, %%mm3 \n\t" // 4H5 - 4H6
3020 "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6
3021 "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6
3023 "movq (%0, %1, 4), %%mm2 \n\t"
3024 "movq %%mm2, %%mm3 \n\t"
3025 "punpcklbw %%mm7, %%mm2 \n\t" // L7
3026 "punpckhbw %%mm7, %%mm3 \n\t" // H7
3028 "paddw %%mm2, %%mm2 \n\t" // 2L7
3029 "paddw %%mm3, %%mm3 \n\t" // 2H7
3030 "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6 - 2L7
3031 "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6 - 2H7
3033 "movq (%%"REG_c"), %%mm2 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
3034 "movq 8(%%"REG_c"), %%mm3 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
3037 "movq %%mm7, %%mm6 \n\t" // 0
3038 "psubw %%mm0, %%mm6 \n\t"
3039 "pmaxsw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
3040 "movq %%mm7, %%mm6 \n\t" // 0
3041 "psubw %%mm1, %%mm6 \n\t"
3042 "pmaxsw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
3043 "movq %%mm7, %%mm6 \n\t" // 0
3044 "psubw %%mm2, %%mm6 \n\t"
3045 "pmaxsw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
3046 "movq %%mm7, %%mm6 \n\t" // 0
3047 "psubw %%mm3, %%mm6 \n\t"
3048 "pmaxsw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
3050 "movq %%mm7, %%mm6 \n\t" // 0
3051 "pcmpgtw %%mm0, %%mm6 \n\t"
3052 "pxor %%mm6, %%mm0 \n\t"
3053 "psubw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
3054 "movq %%mm7, %%mm6 \n\t" // 0
3055 "pcmpgtw %%mm1, %%mm6 \n\t"
3056 "pxor %%mm6, %%mm1 \n\t"
3057 "psubw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
3058 "movq %%mm7, %%mm6 \n\t" // 0
3059 "pcmpgtw %%mm2, %%mm6 \n\t"
3060 "pxor %%mm6, %%mm2 \n\t"
3061 "psubw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
3062 "movq %%mm7, %%mm6 \n\t" // 0
3063 "pcmpgtw %%mm3, %%mm6 \n\t"
3064 "pxor %%mm6, %%mm3 \n\t"
3065 "psubw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
3069 "pminsw %%mm2, %%mm0 \n\t"
3070 "pminsw %%mm3, %%mm1 \n\t"
3072 "movq %%mm0, %%mm6 \n\t"
3073 "psubusw %%mm2, %%mm6 \n\t"
3074 "psubw %%mm6, %%mm0 \n\t"
3075 "movq %%mm1, %%mm6 \n\t"
3076 "psubusw %%mm3, %%mm6 \n\t"
3077 "psubw %%mm6, %%mm1 \n\t"
3080 "movd %2, %%mm2 \n\t" // QP
3081 "punpcklbw %%mm7, %%mm2 \n\t"
3083 "movq %%mm7, %%mm6 \n\t" // 0
3084 "pcmpgtw %%mm4, %%mm6 \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
3085 "pxor %%mm6, %%mm4 \n\t"
3086 "psubw %%mm6, %%mm4 \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
3087 "pcmpgtw %%mm5, %%mm7 \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
3088 "pxor %%mm7, %%mm5 \n\t"
3089 "psubw %%mm7, %%mm5 \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
3091 "psllw $3, %%mm2 \n\t" // 8QP
3092 "movq %%mm2, %%mm3 \n\t" // 8QP
3093 "pcmpgtw %%mm4, %%mm2 \n\t"
3094 "pcmpgtw %%mm5, %%mm3 \n\t"
3095 "pand %%mm2, %%mm4 \n\t"
3096 "pand %%mm3, %%mm5 \n\t"
3099 "psubusw %%mm0, %%mm4 \n\t" // hd
3100 "psubusw %%mm1, %%mm5 \n\t" // ld
3103 "movq "MANGLE(w05)", %%mm2 \n\t" // 5
3104 "pmullw %%mm2, %%mm4 \n\t"
3105 "pmullw %%mm2, %%mm5 \n\t"
3106 "movq "MANGLE(w20)", %%mm2 \n\t" // 32
3107 "paddw %%mm2, %%mm4 \n\t"
3108 "paddw %%mm2, %%mm5 \n\t"
3109 "psrlw $6, %%mm4 \n\t"
3110 "psrlw $6, %%mm5 \n\t"
3112 "movq 16(%%"REG_c"), %%mm0 \n\t" // L3 - L4
3113 "movq 24(%%"REG_c"), %%mm1 \n\t" // H3 - H4
3115 "pxor %%mm2, %%mm2 \n\t"
3116 "pxor %%mm3, %%mm3 \n\t"
3118 "pcmpgtw %%mm0, %%mm2 \n\t" // sign (L3-L4)
3119 "pcmpgtw %%mm1, %%mm3 \n\t" // sign (H3-H4)
3120 "pxor %%mm2, %%mm0 \n\t"
3121 "pxor %%mm3, %%mm1 \n\t"
3122 "psubw %%mm2, %%mm0 \n\t" // |L3-L4|
3123 "psubw %%mm3, %%mm1 \n\t" // |H3-H4|
3124 "psrlw $1, %%mm0 \n\t" // |L3 - L4|/2
3125 "psrlw $1, %%mm1 \n\t" // |H3 - H4|/2
3127 "pxor %%mm6, %%mm2 \n\t"
3128 "pxor %%mm7, %%mm3 \n\t"
3129 "pand %%mm2, %%mm4 \n\t"
3130 "pand %%mm3, %%mm5 \n\t"
3133 "pminsw %%mm0, %%mm4 \n\t"
3134 "pminsw %%mm1, %%mm5 \n\t"
3136 "movq %%mm4, %%mm2 \n\t"
3137 "psubusw %%mm0, %%mm2 \n\t"
3138 "psubw %%mm2, %%mm4 \n\t"
3139 "movq %%mm5, %%mm2 \n\t"
3140 "psubusw %%mm1, %%mm2 \n\t"
3141 "psubw %%mm2, %%mm5 \n\t"
3143 "pxor %%mm6, %%mm4 \n\t"
3144 "pxor %%mm7, %%mm5 \n\t"
3145 "psubw %%mm6, %%mm4 \n\t"
3146 "psubw %%mm7, %%mm5 \n\t"
3147 "packsswb %%mm5, %%mm4 \n\t"
3148 "movq %3, %%mm1 \n\t"
3149 "pandn %%mm4, %%mm1 \n\t"
3150 "movq (%0), %%mm0 \n\t"
3151 "paddb %%mm1, %%mm0 \n\t"
3152 "movq %%mm0, (%0) \n\t"
3153 "movq (%0, %1), %%mm0 \n\t"
3154 "psubb %%mm1, %%mm0 \n\t"
3155 "movq %%mm0, (%0, %1) \n\t"
3158 : "r" ((long)step), "m" (c->pQPb), "m"(eq_mask)
3159 : "%"REG_a, "%"REG_c
3163 STOP_TIMER("step16")
3170 static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
3171 const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);
3174 * Copies a block from src to dst and fixes the blacklevel.
3175 * levelFix == 0 -> do not touch the brighness & contrast
3179 static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t src[], int srcStride,
3180 int levelFix, int64_t *packedOffsetAndScale)
3189 "movq (%%"REG_a"), %%mm2 \n\t" // packedYOffset
3190 "movq 8(%%"REG_a"), %%mm3 \n\t" // packedYScale
3191 "lea (%2,%4), %%"REG_a" \n\t"
3192 "lea (%3,%5), %%"REG_d" \n\t"
3193 "pxor %%mm4, %%mm4 \n\t"
3195 #define REAL_SCALED_CPY(src1, src2, dst1, dst2) \
3196 "movq " #src1 ", %%mm0 \n\t"\
3197 "movq " #src1 ", %%mm5 \n\t"\
3198 "movq " #src2 ", %%mm1 \n\t"\
3199 "movq " #src2 ", %%mm6 \n\t"\
3200 "punpcklbw %%mm0, %%mm0 \n\t"\
3201 "punpckhbw %%mm5, %%mm5 \n\t"\
3202 "punpcklbw %%mm1, %%mm1 \n\t"\
3203 "punpckhbw %%mm6, %%mm6 \n\t"\
3204 "pmulhuw %%mm3, %%mm0 \n\t"\
3205 "pmulhuw %%mm3, %%mm5 \n\t"\
3206 "pmulhuw %%mm3, %%mm1 \n\t"\
3207 "pmulhuw %%mm3, %%mm6 \n\t"\
3208 "psubw %%mm2, %%mm0 \n\t"\
3209 "psubw %%mm2, %%mm5 \n\t"\
3210 "psubw %%mm2, %%mm1 \n\t"\
3211 "psubw %%mm2, %%mm6 \n\t"\
3212 "packuswb %%mm5, %%mm0 \n\t"\
3213 "packuswb %%mm6, %%mm1 \n\t"\
3214 "movq %%mm0, " #dst1 " \n\t"\
3215 "movq %%mm1, " #dst2 " \n\t"\
3218 #define REAL_SCALED_CPY(src1, src2, dst1, dst2) \
3219 "movq " #src1 ", %%mm0 \n\t"\
3220 "movq " #src1 ", %%mm5 \n\t"\
3221 "punpcklbw %%mm4, %%mm0 \n\t"\
3222 "punpckhbw %%mm4, %%mm5 \n\t"\
3223 "psubw %%mm2, %%mm0 \n\t"\
3224 "psubw %%mm2, %%mm5 \n\t"\
3225 "movq " #src2 ", %%mm1 \n\t"\
3226 "psllw $6, %%mm0 \n\t"\
3227 "psllw $6, %%mm5 \n\t"\
3228 "pmulhw %%mm3, %%mm0 \n\t"\
3229 "movq " #src2 ", %%mm6 \n\t"\
3230 "pmulhw %%mm3, %%mm5 \n\t"\
3231 "punpcklbw %%mm4, %%mm1 \n\t"\
3232 "punpckhbw %%mm4, %%mm6 \n\t"\
3233 "psubw %%mm2, %%mm1 \n\t"\
3234 "psubw %%mm2, %%mm6 \n\t"\
3235 "psllw $6, %%mm1 \n\t"\
3236 "psllw $6, %%mm6 \n\t"\
3237 "pmulhw %%mm3, %%mm1 \n\t"\
3238 "pmulhw %%mm3, %%mm6 \n\t"\
3239 "packuswb %%mm5, %%mm0 \n\t"\
3240 "packuswb %%mm6, %%mm1 \n\t"\
3241 "movq %%mm0, " #dst1 " \n\t"\
3242 "movq %%mm1, " #dst2 " \n\t"\
3245 #define SCALED_CPY(src1, src2, dst1, dst2)\
3246 REAL_SCALED_CPY(src1, src2, dst1, dst2)
3248 SCALED_CPY((%2) , (%2, %4) , (%3) , (%3, %5))
3249 SCALED_CPY((%2, %4, 2), (%%REGa, %4, 2), (%3, %5, 2), (%%REGd, %5, 2))
3250 SCALED_CPY((%2, %4, 4), (%%REGa, %4, 4), (%3, %5, 4), (%%REGd, %5, 4))
3251 "lea (%%"REG_a",%4,4), %%"REG_a" \n\t"
3252 "lea (%%"REG_d",%5,4), %%"REG_d" \n\t"
3253 SCALED_CPY((%%REGa, %4), (%%REGa, %4, 2), (%%REGd, %5), (%%REGd, %5, 2))
3256 : "=&a" (packedOffsetAndScale)
3257 : "0" (packedOffsetAndScale),
3260 "r" ((long)srcStride),
3261 "r" ((long)dstStride)
3266 memcpy( &(dst[dstStride*i]),
3267 &(src[srcStride*i]), BLOCK_SIZE);
3274 "lea (%0,%2), %%"REG_a" \n\t"
3275 "lea (%1,%3), %%"REG_d" \n\t"
3277 #define REAL_SIMPLE_CPY(src1, src2, dst1, dst2) \
3278 "movq " #src1 ", %%mm0 \n\t"\
3279 "movq " #src2 ", %%mm1 \n\t"\
3280 "movq %%mm0, " #dst1 " \n\t"\
3281 "movq %%mm1, " #dst2 " \n\t"\
3283 #define SIMPLE_CPY(src1, src2, dst1, dst2)\
3284 REAL_SIMPLE_CPY(src1, src2, dst1, dst2)
3286 SIMPLE_CPY((%0) , (%0, %2) , (%1) , (%1, %3))
3287 SIMPLE_CPY((%0, %2, 2), (%%REGa, %2, 2), (%1, %3, 2), (%%REGd, %3, 2))
3288 SIMPLE_CPY((%0, %2, 4), (%%REGa, %2, 4), (%1, %3, 4), (%%REGd, %3, 4))
3289 "lea (%%"REG_a",%2,4), %%"REG_a" \n\t"
3290 "lea (%%"REG_d",%3,4), %%"REG_d" \n\t"
3291 SIMPLE_CPY((%%REGa, %2), (%%REGa, %2, 2), (%%REGd, %3), (%%REGd, %3, 2))
3295 "r" ((long)srcStride),
3296 "r" ((long)dstStride)
3297 : "%"REG_a, "%"REG_d
3301 memcpy( &(dst[dstStride*i]),
3302 &(src[srcStride*i]), BLOCK_SIZE);
3308 * Duplicates the given 8 src pixels ? times upward
3310 static inline void RENAME(duplicate)(uint8_t src[], int stride)
3314 "movq (%0), %%mm0 \n\t"
3316 "movq %%mm0, (%0) \n\t"
3317 "movq %%mm0, (%0, %1) \n\t"
3318 "movq %%mm0, (%0, %1, 2) \n\t"
3320 : "r" ((long)-stride)
3334 * Filters array of bytes (Y or U or V values)
3336 static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
3337 const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)
3339 DECLARE_ALIGNED(8, PPContext, c)= *c2; //copy to stack for faster access
3341 #ifdef COMPILE_TIME_MODE
3342 const int mode= COMPILE_TIME_MODE;
3344 const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;
3346 int black=0, white=255; // blackest black and whitest white in the picture
3347 int QPCorrecture= 256*256;
3354 const int qpHShift= isColor ? 4-c.hChromaSubSample : 4;
3355 const int qpVShift= isColor ? 4-c.vChromaSubSample : 4;
3358 uint64_t * const yHistogram= c.yHistogram;
3359 uint8_t * const tempSrc= srcStride > 0 ? c.tempSrc : c.tempSrc - 23*srcStride;
3360 uint8_t * const tempDst= dstStride > 0 ? c.tempDst : c.tempDst - 23*dstStride;
3361 //const int mbWidth= isColor ? (width+7)>>3 : (width+15)>>4;
3364 for(i=0; i<57; i++){
3365 int offset= ((i*c.ppMode.baseDcDiff)>>8) + 1;
3366 int threshold= offset*2 + 1;
3367 c.mmxDcOffset[i]= 0x7F - offset;
3368 c.mmxDcThreshold[i]= 0x7F - threshold;
3369 c.mmxDcOffset[i]*= 0x0101010101010101LL;
3370 c.mmxDcThreshold[i]*= 0x0101010101010101LL;
3374 if(mode & CUBIC_IPOL_DEINT_FILTER) copyAhead=16;
3375 else if( (mode & LINEAR_BLEND_DEINT_FILTER)
3376 || (mode & FFMPEG_DEINT_FILTER)
3377 || (mode & LOWPASS5_DEINT_FILTER)) copyAhead=14;
3378 else if( (mode & V_DEBLOCK)
3379 || (mode & LINEAR_IPOL_DEINT_FILTER)
3380 || (mode & MEDIAN_DEINT_FILTER)
3381 || (mode & V_A_DEBLOCK)) copyAhead=13;
3382 else if(mode & V_X1_FILTER) copyAhead=11;
3383 // else if(mode & V_RK1_FILTER) copyAhead=10;
3384 else if(mode & DERING) copyAhead=9;
3393 uint64_t maxClipped;
3398 // first frame is fscked so we ignore it
3399 if(c.frameNum == 1) yHistogram[0]= width*height/64*15/256;
3401 for(i=0; i<256; i++)
3403 sum+= yHistogram[i];
3406 /* We always get a completely black picture first. */
3407 maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);
3410 for(black=255; black>0; black--)
3412 if(clipped < maxClipped) break;
3413 clipped-= yHistogram[black];
3417 for(white=0; white<256; white++)
3419 if(clipped < maxClipped) break;
3420 clipped-= yHistogram[white];
3423 scale= (double)(c.ppMode.maxAllowedY - c.ppMode.minAllowedY) / (double)(white-black);
3426 c.packedYScale= (uint16_t)(scale*256.0 + 0.5);
3427 c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF;
3429 c.packedYScale= (uint16_t)(scale*1024.0 + 0.5);
3430 c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF;
3433 c.packedYOffset|= c.packedYOffset<<32;
3434 c.packedYOffset|= c.packedYOffset<<16;
3436 c.packedYScale|= c.packedYScale<<32;
3437 c.packedYScale|= c.packedYScale<<16;
3439 if(mode & LEVEL_FIX) QPCorrecture= (int)(scale*256*256 + 0.5);
3440 else QPCorrecture= 256*256;
3444 c.packedYScale= 0x0100010001000100LL;
3446 QPCorrecture= 256*256;
3449 /* copy & deinterlace first row of blocks */
3452 const uint8_t *srcBlock= &(src[y*srcStride]);
3453 uint8_t *dstBlock= tempDst + dstStride;
3455 // From this point on it is guaranteed that we can read and write 16 lines downward
3456 // finish 1 block before the next otherwise we might have a problem
3457 // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
3458 for(x=0; x<width; x+=BLOCK_SIZE)
3463 prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
3464 prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
3465 prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
3466 prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
3470 "mov %4, %%"REG_a" \n\t"
3471 "shr $2, %%"REG_a" \n\t"
3472 "and $6, %%"REG_a" \n\t"
3473 "add %5, %%"REG_a" \n\t"
3474 "mov %%"REG_a", %%"REG_d" \n\t"
3475 "imul %1, %%"REG_a" \n\t"
3476 "imul %3, %%"REG_d" \n\t"
3477 "prefetchnta 32(%%"REG_a", %0) \n\t"
3478 "prefetcht0 32(%%"REG_d", %2) \n\t"
3479 "add %1, %%"REG_a" \n\t"
3480 "add %3, %%"REG_d" \n\t"
3481 "prefetchnta 32(%%"REG_a", %0) \n\t"
3482 "prefetcht0 32(%%"REG_d", %2) \n\t"
3483 :: "r" (srcBlock), "r" ((long)srcStride), "r" (dstBlock), "r" ((long)dstStride),
3484 "g" ((long)x), "g" ((long)copyAhead)
3485 : "%"REG_a, "%"REG_d
3488 #elif defined(HAVE_3DNOW)
3489 //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
3490 /* prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
3491 prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
3492 prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
3493 prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
3497 RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,
3498 srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
3500 RENAME(duplicate)(dstBlock + dstStride*8, dstStride);
3502 if(mode & LINEAR_IPOL_DEINT_FILTER)
3503 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
3504 else if(mode & LINEAR_BLEND_DEINT_FILTER)
3505 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
3506 else if(mode & MEDIAN_DEINT_FILTER)
3507 RENAME(deInterlaceMedian)(dstBlock, dstStride);
3508 else if(mode & CUBIC_IPOL_DEINT_FILTER)
3509 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
3510 else if(mode & FFMPEG_DEINT_FILTER)
3511 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
3512 else if(mode & LOWPASS5_DEINT_FILTER)
3513 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
3514 /* else if(mode & CUBIC_BLEND_DEINT_FILTER)
3515 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3520 if(width==FFABS(dstStride))
3521 linecpy(dst, tempDst + 9*dstStride, copyAhead, dstStride);
3525 for(i=0; i<copyAhead; i++)
3527 memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);
3532 for(y=0; y<height; y+=BLOCK_SIZE)
3534 //1% speedup if these are here instead of the inner loop
3535 const uint8_t *srcBlock= &(src[y*srcStride]);
3536 uint8_t *dstBlock= &(dst[y*dstStride]);
3538 uint8_t *tempBlock1= c.tempBlocks;
3539 uint8_t *tempBlock2= c.tempBlocks + 8;
3541 const int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];
3542 int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*FFABS(QPStride)];
3544 /* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards
3545 if not than use a temporary buffer */
3549 /* copy from line (copyAhead) to (copyAhead+7) of src, these will be copied with
3550 blockcopy to dst later */
3551 linecpy(tempSrc + srcStride*copyAhead, srcBlock + srcStride*copyAhead,
3552 FFMAX(height-y-copyAhead, 0), srcStride);
3554 /* duplicate last line of src to fill the void upto line (copyAhead+7) */
3555 for(i=FFMAX(height-y, 8); i<copyAhead+8; i++)
3556 memcpy(tempSrc + srcStride*i, src + srcStride*(height-1), FFABS(srcStride));
3558 /* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/
3559 linecpy(tempDst, dstBlock - dstStride, FFMIN(height-y+1, copyAhead+1), dstStride);
3561 /* duplicate last line of dst to fill the void upto line (copyAhead) */
3562 for(i=height-y+1; i<=copyAhead; i++)
3563 memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), FFABS(dstStride));
3565 dstBlock= tempDst + dstStride;
3569 // From this point on it is guaranteed that we can read and write 16 lines downward
3570 // finish 1 block before the next otherwise we might have a problem
3571 // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
3572 for(x=0; x<width; x+=BLOCK_SIZE)
3574 const int stride= dstStride;
3580 QP= QPptr[x>>qpHShift];
3581 c.nonBQP= nonBQPptr[x>>qpHShift];
3586 QP= (QP* QPCorrecture + 256*128)>>16;
3587 c.nonBQP= nonBQPptr[x>>4];
3588 c.nonBQP= (c.nonBQP* QPCorrecture + 256*128)>>16;
3589 yHistogram[ srcBlock[srcStride*12 + 4] ]++;
3594 "movd %1, %%mm7 \n\t"
3595 "packuswb %%mm7, %%mm7 \n\t" // 0, 0, 0, QP, 0, 0, 0, QP
3596 "packuswb %%mm7, %%mm7 \n\t" // 0,QP, 0, QP, 0,QP, 0, QP
3597 "packuswb %%mm7, %%mm7 \n\t" // QP,..., QP
3598 "movq %%mm7, %0 \n\t"
3607 prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
3608 prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
3609 prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
3610 prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
3614 "mov %4, %%"REG_a" \n\t"
3615 "shr $2, %%"REG_a" \n\t"
3616 "and $6, %%"REG_a" \n\t"
3617 "add %5, %%"REG_a" \n\t"
3618 "mov %%"REG_a", %%"REG_d" \n\t"
3619 "imul %1, %%"REG_a" \n\t"
3620 "imul %3, %%"REG_d" \n\t"
3621 "prefetchnta 32(%%"REG_a", %0) \n\t"
3622 "prefetcht0 32(%%"REG_d", %2) \n\t"
3623 "add %1, %%"REG_a" \n\t"
3624 "add %3, %%"REG_d" \n\t"
3625 "prefetchnta 32(%%"REG_a", %0) \n\t"
3626 "prefetcht0 32(%%"REG_d", %2) \n\t"
3627 :: "r" (srcBlock), "r" ((long)srcStride), "r" (dstBlock), "r" ((long)dstStride),
3628 "g" ((long)x), "g" ((long)copyAhead)
3629 : "%"REG_a, "%"REG_d
3632 #elif defined(HAVE_3DNOW)
3633 //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
3634 /* prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
3635 prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
3636 prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
3637 prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
3641 RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,
3642 srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
3644 if(mode & LINEAR_IPOL_DEINT_FILTER)
3645 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
3646 else if(mode & LINEAR_BLEND_DEINT_FILTER)
3647 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
3648 else if(mode & MEDIAN_DEINT_FILTER)
3649 RENAME(deInterlaceMedian)(dstBlock, dstStride);
3650 else if(mode & CUBIC_IPOL_DEINT_FILTER)
3651 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
3652 else if(mode & FFMPEG_DEINT_FILTER)
3653 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
3654 else if(mode & LOWPASS5_DEINT_FILTER)
3655 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
3656 /* else if(mode & CUBIC_BLEND_DEINT_FILTER)
3657 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3660 /* only deblock if we have 2 blocks */
3663 if(mode & V_X1_FILTER)
3664 RENAME(vertX1Filter)(dstBlock, stride, &c);
3665 else if(mode & V_DEBLOCK)
3667 const int t= RENAME(vertClassify)(dstBlock, stride, &c);
3670 RENAME(doVertLowPass)(dstBlock, stride, &c);
3672 RENAME(doVertDefFilter)(dstBlock, stride, &c);
3673 }else if(mode & V_A_DEBLOCK){
3674 RENAME(do_a_deblock)(dstBlock, stride, 1, &c);
3679 RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);
3681 /* check if we have a previous block to deblock it with dstBlock */
3685 if(mode & H_X1_FILTER)
3686 RENAME(vertX1Filter)(tempBlock1, 16, &c);
3687 else if(mode & H_DEBLOCK)
3690 const int t= RENAME(vertClassify)(tempBlock1, 16, &c);
3691 //STOP_TIMER("dc & minmax")
3693 RENAME(doVertLowPass)(tempBlock1, 16, &c);
3695 RENAME(doVertDefFilter)(tempBlock1, 16, &c);
3696 }else if(mode & H_A_DEBLOCK){
3697 RENAME(do_a_deblock)(tempBlock1, 16, 1, &c);
3700 RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16);
3703 if(mode & H_X1_FILTER)
3704 horizX1Filter(dstBlock-4, stride, QP);
3705 else if(mode & H_DEBLOCK)
3708 DECLARE_ALIGNED(16, unsigned char, tempBlock[272]);
3709 transpose_16x8_char_toPackedAlign_altivec(tempBlock, dstBlock - (4 + 1), stride);
3711 const int t=vertClassify_altivec(tempBlock-48, 16, &c);
3713 doVertLowPass_altivec(tempBlock-48, 16, &c);
3714 transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
3717 doVertDefFilter_altivec(tempBlock-48, 16, &c);
3718 transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
3721 const int t= RENAME(horizClassify)(dstBlock-4, stride, &c);
3724 RENAME(doHorizLowPass)(dstBlock-4, stride, &c);
3726 RENAME(doHorizDefFilter)(dstBlock-4, stride, &c);
3728 }else if(mode & H_A_DEBLOCK){
3729 RENAME(do_a_deblock)(dstBlock-8, 1, stride, &c);
3734 //FIXME filter first line
3735 if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);
3738 if(mode & TEMP_NOISE_FILTER)
3740 RENAME(tempNoiseReducer)(dstBlock-8, stride,
3741 c.tempBlured[isColor] + y*dstStride + x,
3742 c.tempBluredPast[isColor] + (y>>3)*256 + (x>>3),
3743 c.ppMode.maxTmpNoise);
3751 tmpXchg= tempBlock1;
3752 tempBlock1= tempBlock2;
3753 tempBlock2 = tmpXchg;
3759 if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, &c);
3762 if((mode & TEMP_NOISE_FILTER))
3764 RENAME(tempNoiseReducer)(dstBlock-8, dstStride,
3765 c.tempBlured[isColor] + y*dstStride + x,
3766 c.tempBluredPast[isColor] + (y>>3)*256 + (x>>3),
3767 c.ppMode.maxTmpNoise);
3770 /* did we use a tmp buffer for the last lines*/
3773 uint8_t *dstBlock= &(dst[y*dstStride]);
3774 if(width==FFABS(dstStride))
3775 linecpy(dstBlock, tempDst + dstStride, height-y, dstStride);
3779 for(i=0; i<height-y; i++)
3781 memcpy(dstBlock + i*dstStride, tempDst + (i+1)*dstStride, width);
3786 for(x=0; x<width; x+=32)
3789 i+= + dstBlock[x + 7*dstStride] + dstBlock[x + 8*dstStride]
3790 + dstBlock[x + 9*dstStride] + dstBlock[x +10*dstStride]
3791 + dstBlock[x +11*dstStride] + dstBlock[x +12*dstStride];
3792 // + dstBlock[x +13*dstStride]
3793 // + dstBlock[x +14*dstStride] + dstBlock[x +15*dstStride];
3797 asm volatile("femms");
3798 #elif defined (HAVE_MMX)
3799 asm volatile("emms");
3802 #ifdef DEBUG_BRIGHTNESS
3807 for(i=0; i<256; i++)
3808 if(yHistogram[i] > max) max=yHistogram[i];
3810 for(i=1; i<256; i++)
3813 int start=yHistogram[i-1]/(max/256+1);
3814 int end=yHistogram[i]/(max/256+1);
3815 int inc= end > start ? 1 : -1;
3816 for(x=start; x!=end+inc; x+=inc)
3817 dst[ i*dstStride + x]+=128;
3820 for(i=0; i<100; i+=2)
3822 dst[ (white)*dstStride + i]+=128;
3823 dst[ (black)*dstStride + i]+=128;
3829 *c2= c; //copy local context back
projects / frescor / ffmpeg.git / blob