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
23 * mmx/mmx2/3dnow postprocess code.
26 #include "libavutil/x86_cpu.h"
28 #define ALIGN_MASK "$-8"
36 #define REAL_PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
38 #define REAL_PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
40 #define PAVGB(a,b) REAL_PAVGB(a,b)
43 #define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
45 #define PMINUB(b,a,t) \
46 "movq " #a ", " #t " \n\t"\
47 "psubusb " #b ", " #t " \n\t"\
48 "psubb " #t ", " #a " \n\t"
52 #define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
55 "psubusb " #a ", " #b " \n\t"\
56 "paddb " #a ", " #b " \n\t"
59 //FIXME? |255-0| = 1 (should not be a problem ...)
62 * Check if the middle 8x8 Block in the given 8x16 block is flat
64 static inline int RENAME(vertClassify)(uint8_t src[], int stride, PPContext *c){
66 src+= stride*4; // src points to begin of the 8x8 Block
70 : : "m" (c->mmxDcOffset[c->nonBQP]), "m" (c->mmxDcThreshold[c->nonBQP])
74 "lea (%2, %3), %%"REG_a" \n\t"
75 // 0 1 2 3 4 5 6 7 8 9
76 // %1 eax eax+%2 eax+2%2 %1+4%2 ecx ecx+%2 ecx+2%2 %1+8%2 ecx+4%2
78 "movq (%2), %%mm0 \n\t"
79 "movq (%%"REG_a"), %%mm1 \n\t"
80 "movq %%mm0, %%mm3 \n\t"
81 "movq %%mm0, %%mm4 \n\t"
83 PMINUB(%%mm1, %%mm3, %%mm5)
84 "psubb %%mm1, %%mm0 \n\t" // mm0 = differnece
85 "paddb %%mm7, %%mm0 \n\t"
86 "pcmpgtb %%mm6, %%mm0 \n\t"
88 "movq (%%"REG_a",%3), %%mm2 \n\t"
90 PMINUB(%%mm2, %%mm3, %%mm5)
91 "psubb %%mm2, %%mm1 \n\t"
92 "paddb %%mm7, %%mm1 \n\t"
93 "pcmpgtb %%mm6, %%mm1 \n\t"
94 "paddb %%mm1, %%mm0 \n\t"
96 "movq (%%"REG_a", %3, 2), %%mm1 \n\t"
98 PMINUB(%%mm1, %%mm3, %%mm5)
99 "psubb %%mm1, %%mm2 \n\t"
100 "paddb %%mm7, %%mm2 \n\t"
101 "pcmpgtb %%mm6, %%mm2 \n\t"
102 "paddb %%mm2, %%mm0 \n\t"
104 "lea (%%"REG_a", %3, 4), %%"REG_a" \n\t"
106 "movq (%2, %3, 4), %%mm2 \n\t"
108 PMINUB(%%mm2, %%mm3, %%mm5)
109 "psubb %%mm2, %%mm1 \n\t"
110 "paddb %%mm7, %%mm1 \n\t"
111 "pcmpgtb %%mm6, %%mm1 \n\t"
112 "paddb %%mm1, %%mm0 \n\t"
114 "movq (%%"REG_a"), %%mm1 \n\t"
116 PMINUB(%%mm1, %%mm3, %%mm5)
117 "psubb %%mm1, %%mm2 \n\t"
118 "paddb %%mm7, %%mm2 \n\t"
119 "pcmpgtb %%mm6, %%mm2 \n\t"
120 "paddb %%mm2, %%mm0 \n\t"
122 "movq (%%"REG_a", %3), %%mm2 \n\t"
124 PMINUB(%%mm2, %%mm3, %%mm5)
125 "psubb %%mm2, %%mm1 \n\t"
126 "paddb %%mm7, %%mm1 \n\t"
127 "pcmpgtb %%mm6, %%mm1 \n\t"
128 "paddb %%mm1, %%mm0 \n\t"
130 "movq (%%"REG_a", %3, 2), %%mm1 \n\t"
132 PMINUB(%%mm1, %%mm3, %%mm5)
133 "psubb %%mm1, %%mm2 \n\t"
134 "paddb %%mm7, %%mm2 \n\t"
135 "pcmpgtb %%mm6, %%mm2 \n\t"
136 "paddb %%mm2, %%mm0 \n\t"
137 "psubusb %%mm3, %%mm4 \n\t"
141 "pxor %%mm7, %%mm7 \n\t"
142 "psadbw %%mm7, %%mm0 \n\t"
144 "movq %%mm0, %%mm1 \n\t"
145 "psrlw $8, %%mm0 \n\t"
146 "paddb %%mm1, %%mm0 \n\t"
147 "movq %%mm0, %%mm1 \n\t"
148 "psrlq $16, %%mm0 \n\t"
149 "paddb %%mm1, %%mm0 \n\t"
150 "movq %%mm0, %%mm1 \n\t"
151 "psrlq $32, %%mm0 \n\t"
152 "paddb %%mm1, %%mm0 \n\t"
154 "movq %4, %%mm7 \n\t" // QP,..., QP
155 "paddusb %%mm7, %%mm7 \n\t" // 2QP ... 2QP
156 "psubusb %%mm7, %%mm4 \n\t" // Diff <= 2QP -> 0
157 "packssdw %%mm4, %%mm4 \n\t"
158 "movd %%mm0, %0 \n\t"
159 "movd %%mm4, %1 \n\t"
161 : "=r" (numEq), "=r" (dcOk)
162 : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
166 numEq= (-numEq) &0xFF;
167 if(numEq > c->ppMode.flatnessThreshold){
177 * Do a vertical low pass filter on the 8x16 block (only write to the 8x8 block in the middle)
178 * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16
181 static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
183 #if HAVE_MMX2 || HAVE_AMD3DNOW
185 __asm__ volatile( //"movv %0 %1 %2\n\t"
186 "movq %2, %%mm0 \n\t" // QP,..., QP
187 "pxor %%mm4, %%mm4 \n\t"
189 "movq (%0), %%mm6 \n\t"
190 "movq (%0, %1), %%mm5 \n\t"
191 "movq %%mm5, %%mm1 \n\t"
192 "movq %%mm6, %%mm2 \n\t"
193 "psubusb %%mm6, %%mm5 \n\t"
194 "psubusb %%mm1, %%mm2 \n\t"
195 "por %%mm5, %%mm2 \n\t" // ABS Diff of lines
196 "psubusb %%mm0, %%mm2 \n\t" // diff <= QP -> 0
197 "pcmpeqb %%mm4, %%mm2 \n\t" // diff <= QP -> FF
199 "pand %%mm2, %%mm6 \n\t"
200 "pandn %%mm1, %%mm2 \n\t"
201 "por %%mm2, %%mm6 \n\t"// First Line to Filter
203 "movq (%0, %1, 8), %%mm5 \n\t"
204 "lea (%0, %1, 4), %%"REG_a" \n\t"
205 "lea (%0, %1, 8), %%"REG_c" \n\t"
206 "sub %1, %%"REG_c" \n\t"
207 "add %1, %0 \n\t" // %0 points to line 1 not 0
208 "movq (%0, %1, 8), %%mm7 \n\t"
209 "movq %%mm5, %%mm1 \n\t"
210 "movq %%mm7, %%mm2 \n\t"
211 "psubusb %%mm7, %%mm5 \n\t"
212 "psubusb %%mm1, %%mm2 \n\t"
213 "por %%mm5, %%mm2 \n\t" // ABS Diff of lines
214 "psubusb %%mm0, %%mm2 \n\t" // diff <= QP -> 0
215 "pcmpeqb %%mm4, %%mm2 \n\t" // diff <= QP -> FF
217 "pand %%mm2, %%mm7 \n\t"
218 "pandn %%mm1, %%mm2 \n\t"
219 "por %%mm2, %%mm7 \n\t" // First Line to Filter
223 // %0 %0+%1 %0+2%1 eax %0+4%1 eax+2%1 ecx eax+4%1
228 "movq (%0, %1), %%mm0 \n\t" // 1
229 "movq %%mm0, %%mm1 \n\t" // 1
230 PAVGB(%%mm6, %%mm0) //1 1 /2
231 PAVGB(%%mm6, %%mm0) //3 1 /4
233 "movq (%0, %1, 4), %%mm2 \n\t" // 1
234 "movq %%mm2, %%mm5 \n\t" // 1
235 PAVGB((%%REGa), %%mm2) // 11 /2
236 PAVGB((%0, %1, 2), %%mm2) // 211 /4
237 "movq %%mm2, %%mm3 \n\t" // 211 /4
238 "movq (%0), %%mm4 \n\t" // 1
239 PAVGB(%%mm4, %%mm3) // 4 211 /8
240 PAVGB(%%mm0, %%mm3) //642211 /16
241 "movq %%mm3, (%0) \n\t" // X
242 // mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
243 "movq %%mm1, %%mm0 \n\t" // 1
244 PAVGB(%%mm6, %%mm0) //1 1 /2
245 "movq %%mm4, %%mm3 \n\t" // 1
246 PAVGB((%0,%1,2), %%mm3) // 1 1 /2
247 PAVGB((%%REGa,%1,2), %%mm5) // 11 /2
248 PAVGB((%%REGa), %%mm5) // 211 /4
249 PAVGB(%%mm5, %%mm3) // 2 2211 /8
250 PAVGB(%%mm0, %%mm3) //4242211 /16
251 "movq %%mm3, (%0,%1) \n\t" // X
252 // mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
253 PAVGB(%%mm4, %%mm6) //11 /2
254 "movq (%%"REG_c"), %%mm0 \n\t" // 1
255 PAVGB((%%REGa, %1, 2), %%mm0) // 11/2
256 "movq %%mm0, %%mm3 \n\t" // 11/2
257 PAVGB(%%mm1, %%mm0) // 2 11/4
258 PAVGB(%%mm6, %%mm0) //222 11/8
259 PAVGB(%%mm2, %%mm0) //22242211/16
260 "movq (%0, %1, 2), %%mm2 \n\t" // 1
261 "movq %%mm0, (%0, %1, 2) \n\t" // X
262 // mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
263 "movq (%%"REG_a", %1, 4), %%mm0 \n\t" // 1
264 PAVGB((%%REGc), %%mm0) // 11 /2
265 PAVGB(%%mm0, %%mm6) //11 11 /4
266 PAVGB(%%mm1, %%mm4) // 11 /2
267 PAVGB(%%mm2, %%mm1) // 11 /2
268 PAVGB(%%mm1, %%mm6) //1122 11 /8
269 PAVGB(%%mm5, %%mm6) //112242211 /16
270 "movq (%%"REG_a"), %%mm5 \n\t" // 1
271 "movq %%mm6, (%%"REG_a") \n\t" // X
272 // mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
273 "movq (%%"REG_a", %1, 4), %%mm6 \n\t" // 1
274 PAVGB(%%mm7, %%mm6) // 11 /2
275 PAVGB(%%mm4, %%mm6) // 11 11 /4
276 PAVGB(%%mm3, %%mm6) // 11 2211 /8
277 PAVGB(%%mm5, %%mm2) // 11 /2
278 "movq (%0, %1, 4), %%mm4 \n\t" // 1
279 PAVGB(%%mm4, %%mm2) // 112 /4
280 PAVGB(%%mm2, %%mm6) // 112242211 /16
281 "movq %%mm6, (%0, %1, 4) \n\t" // X
282 // mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
283 PAVGB(%%mm7, %%mm1) // 11 2 /4
284 PAVGB(%%mm4, %%mm5) // 11 /2
285 PAVGB(%%mm5, %%mm0) // 11 11 /4
286 "movq (%%"REG_a", %1, 2), %%mm6 \n\t" // 1
287 PAVGB(%%mm6, %%mm1) // 11 4 2 /8
288 PAVGB(%%mm0, %%mm1) // 11224222 /16
289 "movq %%mm1, (%%"REG_a", %1, 2) \n\t" // X
290 // mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
291 PAVGB((%%REGc), %%mm2) // 112 4 /8
292 "movq (%%"REG_a", %1, 4), %%mm0 \n\t" // 1
293 PAVGB(%%mm0, %%mm6) // 1 1 /2
294 PAVGB(%%mm7, %%mm6) // 1 12 /4
295 PAVGB(%%mm2, %%mm6) // 1122424 /4
296 "movq %%mm6, (%%"REG_c") \n\t" // X
297 // mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
298 PAVGB(%%mm7, %%mm5) // 11 2 /4
299 PAVGB(%%mm7, %%mm5) // 11 6 /8
301 PAVGB(%%mm3, %%mm0) // 112 /4
302 PAVGB(%%mm0, %%mm5) // 112246 /16
303 "movq %%mm5, (%%"REG_a", %1, 4) \n\t" // X
307 : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
310 #else //HAVE_MMX2 || HAVE_AMD3DNOW
311 const int l1= stride;
312 const int l2= stride + l1;
313 const int l3= stride + l2;
314 const int l4= stride + l3;
315 const int l5= stride + l4;
316 const int l6= stride + l5;
317 const int l7= stride + l6;
318 const int l8= stride + l7;
319 const int l9= stride + l8;
322 for(x=0; x<BLOCK_SIZE; x++){
323 const int first= FFABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];
324 const int last= FFABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];
327 sums[0] = 4*first + src[l1] + src[l2] + src[l3] + 4;
328 sums[1] = sums[0] - first + src[l4];
329 sums[2] = sums[1] - first + src[l5];
330 sums[3] = sums[2] - first + src[l6];
331 sums[4] = sums[3] - first + src[l7];
332 sums[5] = sums[4] - src[l1] + src[l8];
333 sums[6] = sums[5] - src[l2] + last;
334 sums[7] = sums[6] - src[l3] + last;
335 sums[8] = sums[7] - src[l4] + last;
336 sums[9] = sums[8] - src[l5] + last;
338 src[l1]= (sums[0] + sums[2] + 2*src[l1])>>4;
339 src[l2]= (sums[1] + sums[3] + 2*src[l2])>>4;
340 src[l3]= (sums[2] + sums[4] + 2*src[l3])>>4;
341 src[l4]= (sums[3] + sums[5] + 2*src[l4])>>4;
342 src[l5]= (sums[4] + sums[6] + 2*src[l5])>>4;
343 src[l6]= (sums[5] + sums[7] + 2*src[l6])>>4;
344 src[l7]= (sums[6] + sums[8] + 2*src[l7])>>4;
345 src[l8]= (sums[7] + sums[9] + 2*src[l8])>>4;
349 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
351 #endif //HAVE_ALTIVEC
354 * Experimental Filter 1
355 * will not damage linear gradients
356 * Flat blocks should look like they were passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
357 * can only smooth blocks at the expected locations (it cannot smooth them if they did move)
358 * MMX2 version does correct clipping C version does not
360 static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
362 #if HAVE_MMX2 || HAVE_AMD3DNOW
366 "pxor %%mm7, %%mm7 \n\t" // 0
367 "lea (%0, %1), %%"REG_a" \n\t"
368 "lea (%%"REG_a", %1, 4), %%"REG_c" \n\t"
369 // 0 1 2 3 4 5 6 7 8 9
370 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1 %0+8%1 ecx+4%1
371 "movq (%%"REG_a", %1, 2), %%mm0 \n\t" // line 3
372 "movq (%0, %1, 4), %%mm1 \n\t" // line 4
373 "movq %%mm1, %%mm2 \n\t" // line 4
374 "psubusb %%mm0, %%mm1 \n\t"
375 "psubusb %%mm2, %%mm0 \n\t"
376 "por %%mm1, %%mm0 \n\t" // |l2 - l3|
377 "movq (%%"REG_c"), %%mm3 \n\t" // line 5
378 "movq (%%"REG_c", %1), %%mm4 \n\t" // line 6
379 "movq %%mm3, %%mm5 \n\t" // line 5
380 "psubusb %%mm4, %%mm3 \n\t"
381 "psubusb %%mm5, %%mm4 \n\t"
382 "por %%mm4, %%mm3 \n\t" // |l5 - l6|
383 PAVGB(%%mm3, %%mm0) // (|l2 - l3| + |l5 - l6|)/2
384 "movq %%mm2, %%mm1 \n\t" // line 4
385 "psubusb %%mm5, %%mm2 \n\t"
386 "movq %%mm2, %%mm4 \n\t"
387 "pcmpeqb %%mm7, %%mm2 \n\t" // (l4 - l5) <= 0 ? -1 : 0
388 "psubusb %%mm1, %%mm5 \n\t"
389 "por %%mm5, %%mm4 \n\t" // |l4 - l5|
390 "psubusb %%mm0, %%mm4 \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
391 "movq %%mm4, %%mm3 \n\t" // d
392 "movq %2, %%mm0 \n\t"
393 "paddusb %%mm0, %%mm0 \n\t"
394 "psubusb %%mm0, %%mm4 \n\t"
395 "pcmpeqb %%mm7, %%mm4 \n\t" // d <= QP ? -1 : 0
396 "psubusb "MANGLE(b01)", %%mm3 \n\t"
397 "pand %%mm4, %%mm3 \n\t" // d <= QP ? d : 0
399 PAVGB(%%mm7, %%mm3) // d/2
400 "movq %%mm3, %%mm1 \n\t" // d/2
401 PAVGB(%%mm7, %%mm3) // d/4
402 PAVGB(%%mm1, %%mm3) // 3*d/8
404 "movq (%0, %1, 4), %%mm0 \n\t" // line 4
405 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
406 "psubusb %%mm3, %%mm0 \n\t"
407 "pxor %%mm2, %%mm0 \n\t"
408 "movq %%mm0, (%0, %1, 4) \n\t" // line 4
410 "movq (%%"REG_c"), %%mm0 \n\t" // line 5
411 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
412 "paddusb %%mm3, %%mm0 \n\t"
413 "pxor %%mm2, %%mm0 \n\t"
414 "movq %%mm0, (%%"REG_c") \n\t" // line 5
416 PAVGB(%%mm7, %%mm1) // d/4
418 "movq (%%"REG_a", %1, 2), %%mm0 \n\t" // line 3
419 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
420 "psubusb %%mm1, %%mm0 \n\t"
421 "pxor %%mm2, %%mm0 \n\t"
422 "movq %%mm0, (%%"REG_a", %1, 2) \n\t" // line 3
424 "movq (%%"REG_c", %1), %%mm0 \n\t" // line 6
425 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
426 "paddusb %%mm1, %%mm0 \n\t"
427 "pxor %%mm2, %%mm0 \n\t"
428 "movq %%mm0, (%%"REG_c", %1) \n\t" // line 6
430 PAVGB(%%mm7, %%mm1) // d/8
432 "movq (%%"REG_a", %1), %%mm0 \n\t" // line 2
433 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
434 "psubusb %%mm1, %%mm0 \n\t"
435 "pxor %%mm2, %%mm0 \n\t"
436 "movq %%mm0, (%%"REG_a", %1) \n\t" // line 2
438 "movq (%%"REG_c", %1, 2), %%mm0 \n\t" // line 7
439 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
440 "paddusb %%mm1, %%mm0 \n\t"
441 "pxor %%mm2, %%mm0 \n\t"
442 "movq %%mm0, (%%"REG_c", %1, 2) \n\t" // line 7
445 : "r" (src), "r" ((x86_reg)stride), "m" (co->pQPb)
448 #else //HAVE_MMX2 || HAVE_AMD3DNOW
450 const int l1= stride;
451 const int l2= stride + l1;
452 const int l3= stride + l2;
453 const int l4= stride + l3;
454 const int l5= stride + l4;
455 const int l6= stride + l5;
456 const int l7= stride + l6;
457 // const int l8= stride + l7;
458 // const int l9= stride + l8;
462 for(x=0; x<BLOCK_SIZE; x++){
463 int a= src[l3] - src[l4];
464 int b= src[l4] - src[l5];
465 int c= src[l5] - src[l6];
467 int d= FFABS(b) - ((FFABS(a) + FFABS(c))>>1);
471 int v = d * FFSIGN(-b);
482 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
486 static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext *c)
488 #if HAVE_MMX2 || HAVE_AMD3DNOW
491 const int l1= stride;
492 const int l2= stride + l1;
493 const int l3= stride + l2;
494 const int l4= (int)tmp - (int)src - stride*3;
495 const int l5= (int)tmp - (int)src - stride*3 + 8;
496 const int l6= stride*3 + l3;
497 const int l7= stride + l6;
498 const int l8= stride + l7;
500 memcpy(tmp, src+stride*7, 8);
501 memcpy(tmp+8, src+stride*8, 8);
506 #if 0 //slightly more accurate and slightly slower
507 "pxor %%mm7, %%mm7 \n\t" // 0
508 "lea (%0, %1), %%"REG_a" \n\t"
509 "lea (%%"REG_a", %1, 4), %%"REG_c" \n\t"
511 // %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 ecx+%1 ecx+2%1
512 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1
515 "movq (%0, %1, 2), %%mm0 \n\t" // l2
516 "movq (%0), %%mm1 \n\t" // l0
517 "movq %%mm0, %%mm2 \n\t" // l2
518 PAVGB(%%mm7, %%mm0) // ~l2/2
519 PAVGB(%%mm1, %%mm0) // ~(l2 + 2l0)/4
520 PAVGB(%%mm2, %%mm0) // ~(5l2 + 2l0)/8
522 "movq (%%"REG_a"), %%mm1 \n\t" // l1
523 "movq (%%"REG_a", %1, 2), %%mm3 \n\t" // l3
524 "movq %%mm1, %%mm4 \n\t" // l1
525 PAVGB(%%mm7, %%mm1) // ~l1/2
526 PAVGB(%%mm3, %%mm1) // ~(l1 + 2l3)/4
527 PAVGB(%%mm4, %%mm1) // ~(5l1 + 2l3)/8
529 "movq %%mm0, %%mm4 \n\t" // ~(5l2 + 2l0)/8
530 "psubusb %%mm1, %%mm0 \n\t"
531 "psubusb %%mm4, %%mm1 \n\t"
532 "por %%mm0, %%mm1 \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8
533 // mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0
535 "movq (%0, %1, 4), %%mm0 \n\t" // l4
536 "movq %%mm0, %%mm4 \n\t" // l4
537 PAVGB(%%mm7, %%mm0) // ~l4/2
538 PAVGB(%%mm2, %%mm0) // ~(l4 + 2l2)/4
539 PAVGB(%%mm4, %%mm0) // ~(5l4 + 2l2)/8
541 "movq (%%"REG_c"), %%mm2 \n\t" // l5
542 "movq %%mm3, %%mm5 \n\t" // l3
543 PAVGB(%%mm7, %%mm3) // ~l3/2
544 PAVGB(%%mm2, %%mm3) // ~(l3 + 2l5)/4
545 PAVGB(%%mm5, %%mm3) // ~(5l3 + 2l5)/8
547 "movq %%mm0, %%mm6 \n\t" // ~(5l4 + 2l2)/8
548 "psubusb %%mm3, %%mm0 \n\t"
549 "psubusb %%mm6, %%mm3 \n\t"
550 "por %%mm0, %%mm3 \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8
551 "pcmpeqb %%mm7, %%mm0 \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)
552 // mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0
554 "movq (%%"REG_c", %1), %%mm6 \n\t" // l6
555 "movq %%mm6, %%mm5 \n\t" // l6
556 PAVGB(%%mm7, %%mm6) // ~l6/2
557 PAVGB(%%mm4, %%mm6) // ~(l6 + 2l4)/4
558 PAVGB(%%mm5, %%mm6) // ~(5l6 + 2l4)/8
560 "movq (%%"REG_c", %1, 2), %%mm5 \n\t" // l7
561 "movq %%mm2, %%mm4 \n\t" // l5
562 PAVGB(%%mm7, %%mm2) // ~l5/2
563 PAVGB(%%mm5, %%mm2) // ~(l5 + 2l7)/4
564 PAVGB(%%mm4, %%mm2) // ~(5l5 + 2l7)/8
566 "movq %%mm6, %%mm4 \n\t" // ~(5l6 + 2l4)/8
567 "psubusb %%mm2, %%mm6 \n\t"
568 "psubusb %%mm4, %%mm2 \n\t"
569 "por %%mm6, %%mm2 \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8
570 // mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0
573 PMINUB(%%mm2, %%mm1, %%mm4) // MIN(|lenergy|,|renergy|)/8
574 "movq %2, %%mm4 \n\t" // QP //FIXME QP+1 ?
575 "paddusb "MANGLE(b01)", %%mm4 \n\t"
576 "pcmpgtb %%mm3, %%mm4 \n\t" // |menergy|/8 < QP
577 "psubusb %%mm1, %%mm3 \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8
578 "pand %%mm4, %%mm3 \n\t"
580 "movq %%mm3, %%mm1 \n\t"
581 // "psubusb "MANGLE(b01)", %%mm3 \n\t"
584 "paddusb %%mm1, %%mm3 \n\t"
585 // "paddusb "MANGLE(b01)", %%mm3 \n\t"
587 "movq (%%"REG_a", %1, 2), %%mm6 \n\t" //l3
588 "movq (%0, %1, 4), %%mm5 \n\t" //l4
589 "movq (%0, %1, 4), %%mm4 \n\t" //l4
590 "psubusb %%mm6, %%mm5 \n\t"
591 "psubusb %%mm4, %%mm6 \n\t"
592 "por %%mm6, %%mm5 \n\t" // |l3-l4|
593 "pcmpeqb %%mm7, %%mm6 \n\t" // SIGN(l3-l4)
594 "pxor %%mm6, %%mm0 \n\t"
595 "pand %%mm0, %%mm3 \n\t"
596 PMINUB(%%mm5, %%mm3, %%mm0)
598 "psubusb "MANGLE(b01)", %%mm3 \n\t"
601 "movq (%%"REG_a", %1, 2), %%mm0 \n\t"
602 "movq (%0, %1, 4), %%mm2 \n\t"
603 "pxor %%mm6, %%mm0 \n\t"
604 "pxor %%mm6, %%mm2 \n\t"
605 "psubb %%mm3, %%mm0 \n\t"
606 "paddb %%mm3, %%mm2 \n\t"
607 "pxor %%mm6, %%mm0 \n\t"
608 "pxor %%mm6, %%mm2 \n\t"
609 "movq %%mm0, (%%"REG_a", %1, 2) \n\t"
610 "movq %%mm2, (%0, %1, 4) \n\t"
613 "lea (%0, %1), %%"REG_a" \n\t"
614 "pcmpeqb %%mm6, %%mm6 \n\t" // -1
616 // %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 ecx+%1 ecx+2%1
617 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1
620 "movq (%%"REG_a", %1, 2), %%mm1 \n\t" // l3
621 "movq (%0, %1, 4), %%mm0 \n\t" // l4
622 "pxor %%mm6, %%mm1 \n\t" // -l3-1
623 PAVGB(%%mm1, %%mm0) // -q+128 = (l4-l3+256)/2
624 // mm1=-l3-1, mm0=128-q
626 "movq (%%"REG_a", %1, 4), %%mm2 \n\t" // l5
627 "movq (%%"REG_a", %1), %%mm3 \n\t" // l2
628 "pxor %%mm6, %%mm2 \n\t" // -l5-1
629 "movq %%mm2, %%mm5 \n\t" // -l5-1
630 "movq "MANGLE(b80)", %%mm4 \n\t" // 128
631 "lea (%%"REG_a", %1, 4), %%"REG_c" \n\t"
632 PAVGB(%%mm3, %%mm2) // (l2-l5+256)/2
633 PAVGB(%%mm0, %%mm4) // ~(l4-l3)/4 + 128
634 PAVGB(%%mm2, %%mm4) // ~(l2-l5)/4 +(l4-l3)/8 + 128
635 PAVGB(%%mm0, %%mm4) // ~(l2-l5)/8 +5(l4-l3)/16 + 128
636 // mm1=-l3-1, mm0=128-q, mm3=l2, mm4=menergy/16 + 128, mm5= -l5-1
638 "movq (%%"REG_a"), %%mm2 \n\t" // l1
639 "pxor %%mm6, %%mm2 \n\t" // -l1-1
640 PAVGB(%%mm3, %%mm2) // (l2-l1+256)/2
641 PAVGB((%0), %%mm1) // (l0-l3+256)/2
642 "movq "MANGLE(b80)", %%mm3 \n\t" // 128
643 PAVGB(%%mm2, %%mm3) // ~(l2-l1)/4 + 128
644 PAVGB(%%mm1, %%mm3) // ~(l0-l3)/4 +(l2-l1)/8 + 128
645 PAVGB(%%mm2, %%mm3) // ~(l0-l3)/8 +5(l2-l1)/16 + 128
646 // mm0=128-q, mm3=lenergy/16 + 128, mm4= menergy/16 + 128, mm5= -l5-1
648 PAVGB((%%REGc, %1), %%mm5) // (l6-l5+256)/2
649 "movq (%%"REG_c", %1, 2), %%mm1 \n\t" // l7
650 "pxor %%mm6, %%mm1 \n\t" // -l7-1
651 PAVGB((%0, %1, 4), %%mm1) // (l4-l7+256)/2
652 "movq "MANGLE(b80)", %%mm2 \n\t" // 128
653 PAVGB(%%mm5, %%mm2) // ~(l6-l5)/4 + 128
654 PAVGB(%%mm1, %%mm2) // ~(l4-l7)/4 +(l6-l5)/8 + 128
655 PAVGB(%%mm5, %%mm2) // ~(l4-l7)/8 +5(l6-l5)/16 + 128
656 // mm0=128-q, mm2=renergy/16 + 128, mm3=lenergy/16 + 128, mm4= menergy/16 + 128
658 "movq "MANGLE(b00)", %%mm1 \n\t" // 0
659 "movq "MANGLE(b00)", %%mm5 \n\t" // 0
660 "psubb %%mm2, %%mm1 \n\t" // 128 - renergy/16
661 "psubb %%mm3, %%mm5 \n\t" // 128 - lenergy/16
662 PMAXUB(%%mm1, %%mm2) // 128 + |renergy/16|
663 PMAXUB(%%mm5, %%mm3) // 128 + |lenergy/16|
664 PMINUB(%%mm2, %%mm3, %%mm1) // 128 + MIN(|lenergy|,|renergy|)/16
666 // mm0=128-q, mm3=128 + MIN(|lenergy|,|renergy|)/16, mm4= menergy/16 + 128
668 "movq "MANGLE(b00)", %%mm7 \n\t" // 0
669 "movq %2, %%mm2 \n\t" // QP
670 PAVGB(%%mm6, %%mm2) // 128 + QP/2
671 "psubb %%mm6, %%mm2 \n\t"
673 "movq %%mm4, %%mm1 \n\t"
674 "pcmpgtb %%mm7, %%mm1 \n\t" // SIGN(menergy)
675 "pxor %%mm1, %%mm4 \n\t"
676 "psubb %%mm1, %%mm4 \n\t" // 128 + |menergy|/16
677 "pcmpgtb %%mm4, %%mm2 \n\t" // |menergy|/16 < QP/2
678 "psubusb %%mm3, %%mm4 \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16
679 // mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16
681 "movq %%mm4, %%mm3 \n\t" // d
682 "psubusb "MANGLE(b01)", %%mm4 \n\t"
683 PAVGB(%%mm7, %%mm4) // d/32
684 PAVGB(%%mm7, %%mm4) // (d + 32)/64
685 "paddb %%mm3, %%mm4 \n\t" // 5d/64
686 "pand %%mm2, %%mm4 \n\t"
688 "movq "MANGLE(b80)", %%mm5 \n\t" // 128
689 "psubb %%mm0, %%mm5 \n\t" // q
690 "paddsb %%mm6, %%mm5 \n\t" // fix bad rounding
691 "pcmpgtb %%mm5, %%mm7 \n\t" // SIGN(q)
692 "pxor %%mm7, %%mm5 \n\t"
694 PMINUB(%%mm5, %%mm4, %%mm3) // MIN(|q|, 5d/64)
695 "pxor %%mm1, %%mm7 \n\t" // SIGN(d*q)
697 "pand %%mm7, %%mm4 \n\t"
698 "movq (%%"REG_a", %1, 2), %%mm0 \n\t"
699 "movq (%0, %1, 4), %%mm2 \n\t"
700 "pxor %%mm1, %%mm0 \n\t"
701 "pxor %%mm1, %%mm2 \n\t"
702 "paddb %%mm4, %%mm0 \n\t"
703 "psubb %%mm4, %%mm2 \n\t"
704 "pxor %%mm1, %%mm0 \n\t"
705 "pxor %%mm1, %%mm2 \n\t"
706 "movq %%mm0, (%%"REG_a", %1, 2) \n\t"
707 "movq %%mm2, (%0, %1, 4) \n\t"
710 : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)
718 for(x=0; x<BLOCK_SIZE; x++){
719 const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
720 if(FFABS(middleEnergy)< 8*QP){
721 const int q=(src[l4] - src[l5])/2;
722 const int leftEnergy= 5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
723 const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
725 int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
729 d*= FFSIGN(-middleEnergy);
748 int d= src[x+y*stride] - tmp[x+(y-4)*8];
758 src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
762 if(num%1000000 == 0){
763 av_log(c, AV_LOG_INFO, " %d %d %d %d\n", num, sum, max, bias);
772 "pxor %%mm7, %%mm7 \n\t"
773 "lea -40(%%"REG_SP"), %%"REG_c" \n\t" // make space for 4 8-byte vars
774 "and "ALIGN_MASK", %%"REG_c" \n\t" // align
776 // %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 edx+%1 edx+2%1
777 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1
779 "movq (%0), %%mm0 \n\t"
780 "movq %%mm0, %%mm1 \n\t"
781 "punpcklbw %%mm7, %%mm0 \n\t" // low part of line 0
782 "punpckhbw %%mm7, %%mm1 \n\t" // high part of line 0
784 "movq (%0, %1), %%mm2 \n\t"
785 "lea (%0, %1, 2), %%"REG_a" \n\t"
786 "movq %%mm2, %%mm3 \n\t"
787 "punpcklbw %%mm7, %%mm2 \n\t" // low part of line 1
788 "punpckhbw %%mm7, %%mm3 \n\t" // high part of line 1
790 "movq (%%"REG_a"), %%mm4 \n\t"
791 "movq %%mm4, %%mm5 \n\t"
792 "punpcklbw %%mm7, %%mm4 \n\t" // low part of line 2
793 "punpckhbw %%mm7, %%mm5 \n\t" // high part of line 2
795 "paddw %%mm0, %%mm0 \n\t" // 2L0
796 "paddw %%mm1, %%mm1 \n\t" // 2H0
797 "psubw %%mm4, %%mm2 \n\t" // L1 - L2
798 "psubw %%mm5, %%mm3 \n\t" // H1 - H2
799 "psubw %%mm2, %%mm0 \n\t" // 2L0 - L1 + L2
800 "psubw %%mm3, %%mm1 \n\t" // 2H0 - H1 + H2
802 "psllw $2, %%mm2 \n\t" // 4L1 - 4L2
803 "psllw $2, %%mm3 \n\t" // 4H1 - 4H2
804 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2
805 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2
807 "movq (%%"REG_a", %1), %%mm2 \n\t"
808 "movq %%mm2, %%mm3 \n\t"
809 "punpcklbw %%mm7, %%mm2 \n\t" // L3
810 "punpckhbw %%mm7, %%mm3 \n\t" // H3
812 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - L3
813 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - H3
814 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
815 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
816 "movq %%mm0, (%%"REG_c") \n\t" // 2L0 - 5L1 + 5L2 - 2L3
817 "movq %%mm1, 8(%%"REG_c") \n\t" // 2H0 - 5H1 + 5H2 - 2H3
819 "movq (%%"REG_a", %1, 2), %%mm0 \n\t"
820 "movq %%mm0, %%mm1 \n\t"
821 "punpcklbw %%mm7, %%mm0 \n\t" // L4
822 "punpckhbw %%mm7, %%mm1 \n\t" // H4
824 "psubw %%mm0, %%mm2 \n\t" // L3 - L4
825 "psubw %%mm1, %%mm3 \n\t" // H3 - H4
826 "movq %%mm2, 16(%%"REG_c") \n\t" // L3 - L4
827 "movq %%mm3, 24(%%"REG_c") \n\t" // H3 - H4
828 "paddw %%mm4, %%mm4 \n\t" // 2L2
829 "paddw %%mm5, %%mm5 \n\t" // 2H2
830 "psubw %%mm2, %%mm4 \n\t" // 2L2 - L3 + L4
831 "psubw %%mm3, %%mm5 \n\t" // 2H2 - H3 + H4
833 "lea (%%"REG_a", %1), %0 \n\t"
834 "psllw $2, %%mm2 \n\t" // 4L3 - 4L4
835 "psllw $2, %%mm3 \n\t" // 4H3 - 4H4
836 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4
837 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4
839 "movq (%0, %1, 2), %%mm2 \n\t"
840 "movq %%mm2, %%mm3 \n\t"
841 "punpcklbw %%mm7, %%mm2 \n\t" // L5
842 "punpckhbw %%mm7, %%mm3 \n\t" // H5
843 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - L5
844 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - H5
845 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - 2L5
846 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - 2H5
848 "movq (%%"REG_a", %1, 4), %%mm6 \n\t"
849 "punpcklbw %%mm7, %%mm6 \n\t" // L6
850 "psubw %%mm6, %%mm2 \n\t" // L5 - L6
851 "movq (%%"REG_a", %1, 4), %%mm6 \n\t"
852 "punpckhbw %%mm7, %%mm6 \n\t" // H6
853 "psubw %%mm6, %%mm3 \n\t" // H5 - H6
855 "paddw %%mm0, %%mm0 \n\t" // 2L4
856 "paddw %%mm1, %%mm1 \n\t" // 2H4
857 "psubw %%mm2, %%mm0 \n\t" // 2L4 - L5 + L6
858 "psubw %%mm3, %%mm1 \n\t" // 2H4 - H5 + H6
860 "psllw $2, %%mm2 \n\t" // 4L5 - 4L6
861 "psllw $2, %%mm3 \n\t" // 4H5 - 4H6
862 "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6
863 "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6
865 "movq (%0, %1, 4), %%mm2 \n\t"
866 "movq %%mm2, %%mm3 \n\t"
867 "punpcklbw %%mm7, %%mm2 \n\t" // L7
868 "punpckhbw %%mm7, %%mm3 \n\t" // H7
870 "paddw %%mm2, %%mm2 \n\t" // 2L7
871 "paddw %%mm3, %%mm3 \n\t" // 2H7
872 "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6 - 2L7
873 "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6 - 2H7
875 "movq (%%"REG_c"), %%mm2 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
876 "movq 8(%%"REG_c"), %%mm3 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
879 "movq %%mm7, %%mm6 \n\t" // 0
880 "psubw %%mm0, %%mm6 \n\t"
881 "pmaxsw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
882 "movq %%mm7, %%mm6 \n\t" // 0
883 "psubw %%mm1, %%mm6 \n\t"
884 "pmaxsw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
885 "movq %%mm7, %%mm6 \n\t" // 0
886 "psubw %%mm2, %%mm6 \n\t"
887 "pmaxsw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
888 "movq %%mm7, %%mm6 \n\t" // 0
889 "psubw %%mm3, %%mm6 \n\t"
890 "pmaxsw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
892 "movq %%mm7, %%mm6 \n\t" // 0
893 "pcmpgtw %%mm0, %%mm6 \n\t"
894 "pxor %%mm6, %%mm0 \n\t"
895 "psubw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
896 "movq %%mm7, %%mm6 \n\t" // 0
897 "pcmpgtw %%mm1, %%mm6 \n\t"
898 "pxor %%mm6, %%mm1 \n\t"
899 "psubw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
900 "movq %%mm7, %%mm6 \n\t" // 0
901 "pcmpgtw %%mm2, %%mm6 \n\t"
902 "pxor %%mm6, %%mm2 \n\t"
903 "psubw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
904 "movq %%mm7, %%mm6 \n\t" // 0
905 "pcmpgtw %%mm3, %%mm6 \n\t"
906 "pxor %%mm6, %%mm3 \n\t"
907 "psubw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
911 "pminsw %%mm2, %%mm0 \n\t"
912 "pminsw %%mm3, %%mm1 \n\t"
914 "movq %%mm0, %%mm6 \n\t"
915 "psubusw %%mm2, %%mm6 \n\t"
916 "psubw %%mm6, %%mm0 \n\t"
917 "movq %%mm1, %%mm6 \n\t"
918 "psubusw %%mm3, %%mm6 \n\t"
919 "psubw %%mm6, %%mm1 \n\t"
922 "movd %2, %%mm2 \n\t" // QP
923 "punpcklbw %%mm7, %%mm2 \n\t"
925 "movq %%mm7, %%mm6 \n\t" // 0
926 "pcmpgtw %%mm4, %%mm6 \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
927 "pxor %%mm6, %%mm4 \n\t"
928 "psubw %%mm6, %%mm4 \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
929 "pcmpgtw %%mm5, %%mm7 \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
930 "pxor %%mm7, %%mm5 \n\t"
931 "psubw %%mm7, %%mm5 \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
933 "psllw $3, %%mm2 \n\t" // 8QP
934 "movq %%mm2, %%mm3 \n\t" // 8QP
935 "pcmpgtw %%mm4, %%mm2 \n\t"
936 "pcmpgtw %%mm5, %%mm3 \n\t"
937 "pand %%mm2, %%mm4 \n\t"
938 "pand %%mm3, %%mm5 \n\t"
941 "psubusw %%mm0, %%mm4 \n\t" // hd
942 "psubusw %%mm1, %%mm5 \n\t" // ld
945 "movq "MANGLE(w05)", %%mm2 \n\t" // 5
946 "pmullw %%mm2, %%mm4 \n\t"
947 "pmullw %%mm2, %%mm5 \n\t"
948 "movq "MANGLE(w20)", %%mm2 \n\t" // 32
949 "paddw %%mm2, %%mm4 \n\t"
950 "paddw %%mm2, %%mm5 \n\t"
951 "psrlw $6, %%mm4 \n\t"
952 "psrlw $6, %%mm5 \n\t"
954 "movq 16(%%"REG_c"), %%mm0 \n\t" // L3 - L4
955 "movq 24(%%"REG_c"), %%mm1 \n\t" // H3 - H4
957 "pxor %%mm2, %%mm2 \n\t"
958 "pxor %%mm3, %%mm3 \n\t"
960 "pcmpgtw %%mm0, %%mm2 \n\t" // sign (L3-L4)
961 "pcmpgtw %%mm1, %%mm3 \n\t" // sign (H3-H4)
962 "pxor %%mm2, %%mm0 \n\t"
963 "pxor %%mm3, %%mm1 \n\t"
964 "psubw %%mm2, %%mm0 \n\t" // |L3-L4|
965 "psubw %%mm3, %%mm1 \n\t" // |H3-H4|
966 "psrlw $1, %%mm0 \n\t" // |L3 - L4|/2
967 "psrlw $1, %%mm1 \n\t" // |H3 - H4|/2
969 "pxor %%mm6, %%mm2 \n\t"
970 "pxor %%mm7, %%mm3 \n\t"
971 "pand %%mm2, %%mm4 \n\t"
972 "pand %%mm3, %%mm5 \n\t"
975 "pminsw %%mm0, %%mm4 \n\t"
976 "pminsw %%mm1, %%mm5 \n\t"
978 "movq %%mm4, %%mm2 \n\t"
979 "psubusw %%mm0, %%mm2 \n\t"
980 "psubw %%mm2, %%mm4 \n\t"
981 "movq %%mm5, %%mm2 \n\t"
982 "psubusw %%mm1, %%mm2 \n\t"
983 "psubw %%mm2, %%mm5 \n\t"
985 "pxor %%mm6, %%mm4 \n\t"
986 "pxor %%mm7, %%mm5 \n\t"
987 "psubw %%mm6, %%mm4 \n\t"
988 "psubw %%mm7, %%mm5 \n\t"
989 "packsswb %%mm5, %%mm4 \n\t"
990 "movq (%0), %%mm0 \n\t"
991 "paddb %%mm4, %%mm0 \n\t"
992 "movq %%mm0, (%0) \n\t"
993 "movq (%0, %1), %%mm0 \n\t"
994 "psubb %%mm4, %%mm0 \n\t"
995 "movq %%mm0, (%0, %1) \n\t"
998 : "r" ((x86_reg)stride), "m" (c->pQPb)
1001 #else //HAVE_MMX2 || HAVE_AMD3DNOW
1002 const int l1= stride;
1003 const int l2= stride + l1;
1004 const int l3= stride + l2;
1005 const int l4= stride + l3;
1006 const int l5= stride + l4;
1007 const int l6= stride + l5;
1008 const int l7= stride + l6;
1009 const int l8= stride + l7;
1010 // const int l9= stride + l8;
1013 for(x=0; x<BLOCK_SIZE; x++){
1014 const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
1015 if(FFABS(middleEnergy) < 8*c->QP){
1016 const int q=(src[l4] - src[l5])/2;
1017 const int leftEnergy= 5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
1018 const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
1020 int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
1024 d*= FFSIGN(-middleEnergy);
1039 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
1041 #endif //HAVE_ALTIVEC
1044 static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
1046 #if HAVE_MMX2 || HAVE_AMD3DNOW
1048 "pxor %%mm6, %%mm6 \n\t"
1049 "pcmpeqb %%mm7, %%mm7 \n\t"
1050 "movq %2, %%mm0 \n\t"
1051 "punpcklbw %%mm6, %%mm0 \n\t"
1052 "psrlw $1, %%mm0 \n\t"
1053 "psubw %%mm7, %%mm0 \n\t"
1054 "packuswb %%mm0, %%mm0 \n\t"
1055 "movq %%mm0, %3 \n\t"
1057 "lea (%0, %1), %%"REG_a" \n\t"
1058 "lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1060 // 0 1 2 3 4 5 6 7 8 9
1061 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1063 #undef REAL_FIND_MIN_MAX
1066 #define REAL_FIND_MIN_MAX(addr)\
1067 "movq " #addr ", %%mm0 \n\t"\
1068 "pminub %%mm0, %%mm7 \n\t"\
1069 "pmaxub %%mm0, %%mm6 \n\t"
1071 #define REAL_FIND_MIN_MAX(addr)\
1072 "movq " #addr ", %%mm0 \n\t"\
1073 "movq %%mm7, %%mm1 \n\t"\
1074 "psubusb %%mm0, %%mm6 \n\t"\
1075 "paddb %%mm0, %%mm6 \n\t"\
1076 "psubusb %%mm0, %%mm1 \n\t"\
1077 "psubb %%mm1, %%mm7 \n\t"
1079 #define FIND_MIN_MAX(addr) REAL_FIND_MIN_MAX(addr)
1081 FIND_MIN_MAX((%%REGa))
1082 FIND_MIN_MAX((%%REGa, %1))
1083 FIND_MIN_MAX((%%REGa, %1, 2))
1084 FIND_MIN_MAX((%0, %1, 4))
1085 FIND_MIN_MAX((%%REGd))
1086 FIND_MIN_MAX((%%REGd, %1))
1087 FIND_MIN_MAX((%%REGd, %1, 2))
1088 FIND_MIN_MAX((%0, %1, 8))
1090 "movq %%mm7, %%mm4 \n\t"
1091 "psrlq $8, %%mm7 \n\t"
1093 "pminub %%mm4, %%mm7 \n\t" // min of pixels
1094 "pshufw $0xF9, %%mm7, %%mm4 \n\t"
1095 "pminub %%mm4, %%mm7 \n\t" // min of pixels
1096 "pshufw $0xFE, %%mm7, %%mm4 \n\t"
1097 "pminub %%mm4, %%mm7 \n\t"
1099 "movq %%mm7, %%mm1 \n\t"
1100 "psubusb %%mm4, %%mm1 \n\t"
1101 "psubb %%mm1, %%mm7 \n\t"
1102 "movq %%mm7, %%mm4 \n\t"
1103 "psrlq $16, %%mm7 \n\t"
1104 "movq %%mm7, %%mm1 \n\t"
1105 "psubusb %%mm4, %%mm1 \n\t"
1106 "psubb %%mm1, %%mm7 \n\t"
1107 "movq %%mm7, %%mm4 \n\t"
1108 "psrlq $32, %%mm7 \n\t"
1109 "movq %%mm7, %%mm1 \n\t"
1110 "psubusb %%mm4, %%mm1 \n\t"
1111 "psubb %%mm1, %%mm7 \n\t"
1115 "movq %%mm6, %%mm4 \n\t"
1116 "psrlq $8, %%mm6 \n\t"
1118 "pmaxub %%mm4, %%mm6 \n\t" // max of pixels
1119 "pshufw $0xF9, %%mm6, %%mm4 \n\t"
1120 "pmaxub %%mm4, %%mm6 \n\t"
1121 "pshufw $0xFE, %%mm6, %%mm4 \n\t"
1122 "pmaxub %%mm4, %%mm6 \n\t"
1124 "psubusb %%mm4, %%mm6 \n\t"
1125 "paddb %%mm4, %%mm6 \n\t"
1126 "movq %%mm6, %%mm4 \n\t"
1127 "psrlq $16, %%mm6 \n\t"
1128 "psubusb %%mm4, %%mm6 \n\t"
1129 "paddb %%mm4, %%mm6 \n\t"
1130 "movq %%mm6, %%mm4 \n\t"
1131 "psrlq $32, %%mm6 \n\t"
1132 "psubusb %%mm4, %%mm6 \n\t"
1133 "paddb %%mm4, %%mm6 \n\t"
1135 "movq %%mm6, %%mm0 \n\t" // max
1136 "psubb %%mm7, %%mm6 \n\t" // max - min
1137 "movd %%mm6, %%ecx \n\t"
1138 "cmpb "MANGLE(deringThreshold)", %%cl \n\t"
1140 "lea -24(%%"REG_SP"), %%"REG_c" \n\t"
1141 "and "ALIGN_MASK", %%"REG_c" \n\t"
1142 PAVGB(%%mm0, %%mm7) // a=(max + min)/2
1143 "punpcklbw %%mm7, %%mm7 \n\t"
1144 "punpcklbw %%mm7, %%mm7 \n\t"
1145 "punpcklbw %%mm7, %%mm7 \n\t"
1146 "movq %%mm7, (%%"REG_c") \n\t"
1148 "movq (%0), %%mm0 \n\t" // L10
1149 "movq %%mm0, %%mm1 \n\t" // L10
1150 "movq %%mm0, %%mm2 \n\t" // L10
1151 "psllq $8, %%mm1 \n\t"
1152 "psrlq $8, %%mm2 \n\t"
1153 "movd -4(%0), %%mm3 \n\t"
1154 "movd 8(%0), %%mm4 \n\t"
1155 "psrlq $24, %%mm3 \n\t"
1156 "psllq $56, %%mm4 \n\t"
1157 "por %%mm3, %%mm1 \n\t" // L00
1158 "por %%mm4, %%mm2 \n\t" // L20
1159 "movq %%mm1, %%mm3 \n\t" // L00
1160 PAVGB(%%mm2, %%mm1) // (L20 + L00)/2
1161 PAVGB(%%mm0, %%mm1) // (L20 + L00 + 2L10)/4
1162 "psubusb %%mm7, %%mm0 \n\t"
1163 "psubusb %%mm7, %%mm2 \n\t"
1164 "psubusb %%mm7, %%mm3 \n\t"
1165 "pcmpeqb "MANGLE(b00)", %%mm0 \n\t" // L10 > a ? 0 : -1
1166 "pcmpeqb "MANGLE(b00)", %%mm2 \n\t" // L20 > a ? 0 : -1
1167 "pcmpeqb "MANGLE(b00)", %%mm3 \n\t" // L00 > a ? 0 : -1
1168 "paddb %%mm2, %%mm0 \n\t"
1169 "paddb %%mm3, %%mm0 \n\t"
1171 "movq (%%"REG_a"), %%mm2 \n\t" // L11
1172 "movq %%mm2, %%mm3 \n\t" // L11
1173 "movq %%mm2, %%mm4 \n\t" // L11
1174 "psllq $8, %%mm3 \n\t"
1175 "psrlq $8, %%mm4 \n\t"
1176 "movd -4(%%"REG_a"), %%mm5 \n\t"
1177 "movd 8(%%"REG_a"), %%mm6 \n\t"
1178 "psrlq $24, %%mm5 \n\t"
1179 "psllq $56, %%mm6 \n\t"
1180 "por %%mm5, %%mm3 \n\t" // L01
1181 "por %%mm6, %%mm4 \n\t" // L21
1182 "movq %%mm3, %%mm5 \n\t" // L01
1183 PAVGB(%%mm4, %%mm3) // (L21 + L01)/2
1184 PAVGB(%%mm2, %%mm3) // (L21 + L01 + 2L11)/4
1185 "psubusb %%mm7, %%mm2 \n\t"
1186 "psubusb %%mm7, %%mm4 \n\t"
1187 "psubusb %%mm7, %%mm5 \n\t"
1188 "pcmpeqb "MANGLE(b00)", %%mm2 \n\t" // L11 > a ? 0 : -1
1189 "pcmpeqb "MANGLE(b00)", %%mm4 \n\t" // L21 > a ? 0 : -1
1190 "pcmpeqb "MANGLE(b00)", %%mm5 \n\t" // L01 > a ? 0 : -1
1191 "paddb %%mm4, %%mm2 \n\t"
1192 "paddb %%mm5, %%mm2 \n\t"
1194 #define REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
1195 "movq " #src ", " #sx " \n\t" /* src[0] */\
1196 "movq " #sx ", " #lx " \n\t" /* src[0] */\
1197 "movq " #sx ", " #t0 " \n\t" /* src[0] */\
1198 "psllq $8, " #lx " \n\t"\
1199 "psrlq $8, " #t0 " \n\t"\
1200 "movd -4" #src ", " #t1 " \n\t"\
1201 "psrlq $24, " #t1 " \n\t"\
1202 "por " #t1 ", " #lx " \n\t" /* src[-1] */\
1203 "movd 8" #src ", " #t1 " \n\t"\
1204 "psllq $56, " #t1 " \n\t"\
1205 "por " #t1 ", " #t0 " \n\t" /* src[+1] */\
1206 "movq " #lx ", " #t1 " \n\t" /* src[-1] */\
1207 PAVGB(t0, lx) /* (src[-1] + src[+1])/2 */\
1208 PAVGB(sx, lx) /* (src[-1] + 2src[0] + src[+1])/4 */\
1210 "movq " #lx ", 8(%%"REG_c") \n\t"\
1211 "movq (%%"REG_c"), " #lx " \n\t"\
1212 "psubusb " #lx ", " #t1 " \n\t"\
1213 "psubusb " #lx ", " #t0 " \n\t"\
1214 "psubusb " #lx ", " #sx " \n\t"\
1215 "movq "MANGLE(b00)", " #lx " \n\t"\
1216 "pcmpeqb " #lx ", " #t1 " \n\t" /* src[-1] > a ? 0 : -1*/\
1217 "pcmpeqb " #lx ", " #t0 " \n\t" /* src[+1] > a ? 0 : -1*/\
1218 "pcmpeqb " #lx ", " #sx " \n\t" /* src[0] > a ? 0 : -1*/\
1219 "paddb " #t1 ", " #t0 " \n\t"\
1220 "paddb " #t0 ", " #sx " \n\t"\
1222 PAVGB(plx, pplx) /* filtered */\
1223 "movq " #dst ", " #t0 " \n\t" /* dst */\
1224 "movq " #t0 ", " #t1 " \n\t" /* dst */\
1225 "psubusb %3, " #t0 " \n\t"\
1226 "paddusb %3, " #t1 " \n\t"\
1228 PMINUB(t1, pplx, t0)\
1229 "paddb " #sx ", " #ppsx " \n\t"\
1230 "paddb " #psx ", " #ppsx " \n\t"\
1231 "#paddb "MANGLE(b02)", " #ppsx " \n\t"\
1232 "pand "MANGLE(b08)", " #ppsx " \n\t"\
1233 "pcmpeqb " #lx ", " #ppsx " \n\t"\
1234 "pand " #ppsx ", " #pplx " \n\t"\
1235 "pandn " #dst ", " #ppsx " \n\t"\
1236 "por " #pplx ", " #ppsx " \n\t"\
1237 "movq " #ppsx ", " #dst " \n\t"\
1238 "movq 8(%%"REG_c"), " #lx " \n\t"
1240 #define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
1241 REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1)
1257 //DERING_CORE(dst ,src ,ppsx ,psx ,sx ,pplx ,plx ,lx ,t0 ,t1)
1258 DERING_CORE((%%REGa) ,(%%REGa, %1) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1259 DERING_CORE((%%REGa, %1) ,(%%REGa, %1, 2),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1260 DERING_CORE((%%REGa, %1, 2),(%0, %1, 4) ,%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
1261 DERING_CORE((%0, %1, 4) ,(%%REGd) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1262 DERING_CORE((%%REGd) ,(%%REGd, %1) ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1263 DERING_CORE((%%REGd, %1) ,(%%REGd, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
1264 DERING_CORE((%%REGd, %1, 2),(%0, %1, 8) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1265 DERING_CORE((%0, %1, 8) ,(%%REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1268 : : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb), "m"(c->pQPb2)
1269 : "%"REG_a, "%"REG_d, "%"REG_c
1271 #else //HAVE_MMX2 || HAVE_AMD3DNOW
1278 const int QP2= c->QP/2 + 1;
1285 if(*p > max) max= *p;
1286 if(*p < min) min= *p;
1289 avg= (min + max + 1)>>1;
1291 if(max - min <deringThreshold) return;
1293 for(y=0; y<10; y++){
1296 if(src[stride*y + 0] > avg) t+= 1;
1297 if(src[stride*y + 1] > avg) t+= 2;
1298 if(src[stride*y + 2] > avg) t+= 4;
1299 if(src[stride*y + 3] > avg) t+= 8;
1300 if(src[stride*y + 4] > avg) t+= 16;
1301 if(src[stride*y + 5] > avg) t+= 32;
1302 if(src[stride*y + 6] > avg) t+= 64;
1303 if(src[stride*y + 7] > avg) t+= 128;
1304 if(src[stride*y + 8] > avg) t+= 256;
1305 if(src[stride*y + 9] > avg) t+= 512;
1308 t &= (t<<1) & (t>>1);
1313 int t = s[y-1] & s[y] & s[y+1];
1326 int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))
1327 +2*(*(p -1)) + 4*(*p ) + 2*(*(p +1))
1328 +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));
1331 #ifdef DEBUG_DERING_THRESHOLD
1332 __asm__ volatile("emms\n\t":);
1334 static long long numPixels=0;
1335 if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;
1336 // if((max-min)<20 || (max-min)*QP<200)
1337 // if((max-min)*QP < 500)
1340 static int numSkipped=0;
1341 static int errorSum=0;
1342 static int worstQP=0;
1343 static int worstRange=0;
1344 static int worstDiff=0;
1346 int absDiff= FFABS(diff);
1347 int error= diff*diff;
1349 if(x==1 || x==8 || y==1 || y==8) continue;
1352 if(absDiff > worstDiff){
1355 worstRange= max-min;
1359 if(1024LL*1024LL*1024LL % numSkipped == 0){
1360 av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "
1361 "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
1362 (float)errorSum/numSkipped, numSkipped, worstQP, worstRange,
1363 worstDiff, (float)numSkipped/numPixels);
1368 if (*p + QP2 < f) *p= *p + QP2;
1369 else if(*p - QP2 > f) *p= *p - QP2;
1374 #ifdef DEBUG_DERING_THRESHOLD
1382 *p = FFMIN(*p + 20, 255);
1385 // src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
1388 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
1390 #endif //HAVE_ALTIVEC
1393 * Deinterlace the given block by linearly interpolating every second line.
1394 * will be called for every 8x8 block and can read & write from line 4-15
1395 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1396 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1398 static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)
1400 #if HAVE_MMX2 || HAVE_AMD3DNOW
1403 "lea (%0, %1), %%"REG_a" \n\t"
1404 "lea (%%"REG_a", %1, 4), %%"REG_c" \n\t"
1405 // 0 1 2 3 4 5 6 7 8 9
1406 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1 %0+8%1 ecx+4%1
1408 "movq (%0), %%mm0 \n\t"
1409 "movq (%%"REG_a", %1), %%mm1 \n\t"
1411 "movq %%mm0, (%%"REG_a") \n\t"
1412 "movq (%0, %1, 4), %%mm0 \n\t"
1414 "movq %%mm1, (%%"REG_a", %1, 2) \n\t"
1415 "movq (%%"REG_c", %1), %%mm1 \n\t"
1417 "movq %%mm0, (%%"REG_c") \n\t"
1418 "movq (%0, %1, 8), %%mm0 \n\t"
1420 "movq %%mm1, (%%"REG_c", %1, 2) \n\t"
1422 : : "r" (src), "r" ((x86_reg)stride)
1423 : "%"REG_a, "%"REG_c
1430 a= *(uint32_t*)&src[stride*0];
1431 b= *(uint32_t*)&src[stride*2];
1432 *(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1433 a= *(uint32_t*)&src[stride*4];
1434 *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1435 b= *(uint32_t*)&src[stride*6];
1436 *(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1437 a= *(uint32_t*)&src[stride*8];
1438 *(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1445 * Deinterlace the given block by cubic interpolating every second line.
1446 * will be called for every 8x8 block and can read & write from line 4-15
1447 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1448 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1449 * this filter will read lines 3-15 and write 7-13
1451 static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)
1453 #if HAVE_MMX2 || HAVE_AMD3DNOW
1456 "lea (%0, %1), %%"REG_a" \n\t"
1457 "lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1458 "lea (%%"REG_d", %1, 4), %%"REG_c" \n\t"
1459 "add %1, %%"REG_c" \n\t"
1460 "pxor %%mm7, %%mm7 \n\t"
1461 // 0 1 2 3 4 5 6 7 8 9 10
1462 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1 ecx
1464 #define REAL_DEINT_CUBIC(a,b,c,d,e)\
1465 "movq " #a ", %%mm0 \n\t"\
1466 "movq " #b ", %%mm1 \n\t"\
1467 "movq " #d ", %%mm2 \n\t"\
1468 "movq " #e ", %%mm3 \n\t"\
1469 PAVGB(%%mm2, %%mm1) /* (b+d) /2 */\
1470 PAVGB(%%mm3, %%mm0) /* a(a+e) /2 */\
1471 "movq %%mm0, %%mm2 \n\t"\
1472 "punpcklbw %%mm7, %%mm0 \n\t"\
1473 "punpckhbw %%mm7, %%mm2 \n\t"\
1474 "movq %%mm1, %%mm3 \n\t"\
1475 "punpcklbw %%mm7, %%mm1 \n\t"\
1476 "punpckhbw %%mm7, %%mm3 \n\t"\
1477 "psubw %%mm1, %%mm0 \n\t" /* L(a+e - (b+d))/2 */\
1478 "psubw %%mm3, %%mm2 \n\t" /* H(a+e - (b+d))/2 */\
1479 "psraw $3, %%mm0 \n\t" /* L(a+e - (b+d))/16 */\
1480 "psraw $3, %%mm2 \n\t" /* H(a+e - (b+d))/16 */\
1481 "psubw %%mm0, %%mm1 \n\t" /* L(9b + 9d - a - e)/16 */\
1482 "psubw %%mm2, %%mm3 \n\t" /* H(9b + 9d - a - e)/16 */\
1483 "packuswb %%mm3, %%mm1 \n\t"\
1484 "movq %%mm1, " #c " \n\t"
1485 #define DEINT_CUBIC(a,b,c,d,e) REAL_DEINT_CUBIC(a,b,c,d,e)
1487 DEINT_CUBIC((%0) , (%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd, %1))
1488 DEINT_CUBIC((%%REGa, %1), (%0, %1, 4) , (%%REGd) , (%%REGd, %1), (%0, %1, 8))
1489 DEINT_CUBIC((%0, %1, 4) , (%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGc))
1490 DEINT_CUBIC((%%REGd, %1), (%0, %1, 8) , (%%REGd, %1, 4), (%%REGc) , (%%REGc, %1, 2))
1492 : : "r" (src), "r" ((x86_reg)stride)
1493 : "%"REG_a, "%"REG_d, "%"REG_c
1495 #else //HAVE_MMX2 || HAVE_AMD3DNOW
1499 src[stride*3] = CLIP((-src[0] + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);
1500 src[stride*5] = CLIP((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);
1501 src[stride*7] = CLIP((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);
1502 src[stride*9] = CLIP((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);
1505 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
1509 * Deinterlace the given block by filtering every second line with a (-1 4 2 4 -1) filter.
1510 * will be called for every 8x8 block and can read & write from line 4-15
1511 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1512 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1513 * this filter will read lines 4-13 and write 5-11
1515 static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)
1517 #if HAVE_MMX2 || HAVE_AMD3DNOW
1520 "lea (%0, %1), %%"REG_a" \n\t"
1521 "lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1522 "pxor %%mm7, %%mm7 \n\t"
1523 "movq (%2), %%mm0 \n\t"
1524 // 0 1 2 3 4 5 6 7 8 9 10
1525 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1 ecx
1527 #define REAL_DEINT_FF(a,b,c,d)\
1528 "movq " #a ", %%mm1 \n\t"\
1529 "movq " #b ", %%mm2 \n\t"\
1530 "movq " #c ", %%mm3 \n\t"\
1531 "movq " #d ", %%mm4 \n\t"\
1532 PAVGB(%%mm3, %%mm1) \
1533 PAVGB(%%mm4, %%mm0) \
1534 "movq %%mm0, %%mm3 \n\t"\
1535 "punpcklbw %%mm7, %%mm0 \n\t"\
1536 "punpckhbw %%mm7, %%mm3 \n\t"\
1537 "movq %%mm1, %%mm4 \n\t"\
1538 "punpcklbw %%mm7, %%mm1 \n\t"\
1539 "punpckhbw %%mm7, %%mm4 \n\t"\
1540 "psllw $2, %%mm1 \n\t"\
1541 "psllw $2, %%mm4 \n\t"\
1542 "psubw %%mm0, %%mm1 \n\t"\
1543 "psubw %%mm3, %%mm4 \n\t"\
1544 "movq %%mm2, %%mm5 \n\t"\
1545 "movq %%mm2, %%mm0 \n\t"\
1546 "punpcklbw %%mm7, %%mm2 \n\t"\
1547 "punpckhbw %%mm7, %%mm5 \n\t"\
1548 "paddw %%mm2, %%mm1 \n\t"\
1549 "paddw %%mm5, %%mm4 \n\t"\
1550 "psraw $2, %%mm1 \n\t"\
1551 "psraw $2, %%mm4 \n\t"\
1552 "packuswb %%mm4, %%mm1 \n\t"\
1553 "movq %%mm1, " #b " \n\t"\
1555 #define DEINT_FF(a,b,c,d) REAL_DEINT_FF(a,b,c,d)
1557 DEINT_FF((%0) , (%%REGa) , (%%REGa, %1), (%%REGa, %1, 2))
1558 DEINT_FF((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd) )
1559 DEINT_FF((%0, %1, 4) , (%%REGd) , (%%REGd, %1), (%%REGd, %1, 2))
1560 DEINT_FF((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))
1562 "movq %%mm0, (%2) \n\t"
1563 : : "r" (src), "r" ((x86_reg)stride), "r"(tmp)
1564 : "%"REG_a, "%"REG_d
1566 #else //HAVE_MMX2 || HAVE_AMD3DNOW
1571 int t2= src[stride*1];
1573 src[stride*1]= CLIP((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);
1575 src[stride*3]= CLIP((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);
1577 src[stride*5]= CLIP((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);
1579 src[stride*7]= CLIP((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);
1584 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
1588 * Deinterlace the given block by filtering every line with a (-1 2 6 2 -1) filter.
1589 * will be called for every 8x8 block and can read & write from line 4-15
1590 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1591 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1592 * this filter will read lines 4-13 and write 4-11
1594 static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)
1596 #if HAVE_MMX2 || HAVE_AMD3DNOW
1599 "lea (%0, %1), %%"REG_a" \n\t"
1600 "lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1601 "pxor %%mm7, %%mm7 \n\t"
1602 "movq (%2), %%mm0 \n\t"
1603 "movq (%3), %%mm1 \n\t"
1604 // 0 1 2 3 4 5 6 7 8 9 10
1605 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1 ecx
1607 #define REAL_DEINT_L5(t1,t2,a,b,c)\
1608 "movq " #a ", %%mm2 \n\t"\
1609 "movq " #b ", %%mm3 \n\t"\
1610 "movq " #c ", %%mm4 \n\t"\
1613 "movq %%mm2, %%mm5 \n\t"\
1614 "movq %%mm2, " #t1 " \n\t"\
1615 "punpcklbw %%mm7, %%mm2 \n\t"\
1616 "punpckhbw %%mm7, %%mm5 \n\t"\
1617 "movq %%mm2, %%mm6 \n\t"\
1618 "paddw %%mm2, %%mm2 \n\t"\
1619 "paddw %%mm6, %%mm2 \n\t"\
1620 "movq %%mm5, %%mm6 \n\t"\
1621 "paddw %%mm5, %%mm5 \n\t"\
1622 "paddw %%mm6, %%mm5 \n\t"\
1623 "movq %%mm3, %%mm6 \n\t"\
1624 "punpcklbw %%mm7, %%mm3 \n\t"\
1625 "punpckhbw %%mm7, %%mm6 \n\t"\
1626 "paddw %%mm3, %%mm3 \n\t"\
1627 "paddw %%mm6, %%mm6 \n\t"\
1628 "paddw %%mm3, %%mm2 \n\t"\
1629 "paddw %%mm6, %%mm5 \n\t"\
1630 "movq %%mm4, %%mm6 \n\t"\
1631 "punpcklbw %%mm7, %%mm4 \n\t"\
1632 "punpckhbw %%mm7, %%mm6 \n\t"\
1633 "psubw %%mm4, %%mm2 \n\t"\
1634 "psubw %%mm6, %%mm5 \n\t"\
1635 "psraw $2, %%mm2 \n\t"\
1636 "psraw $2, %%mm5 \n\t"\
1637 "packuswb %%mm5, %%mm2 \n\t"\
1638 "movq %%mm2, " #a " \n\t"\
1640 #define DEINT_L5(t1,t2,a,b,c) REAL_DEINT_L5(t1,t2,a,b,c)
1642 DEINT_L5(%%mm0, %%mm1, (%0) , (%%REGa) , (%%REGa, %1) )
1643 DEINT_L5(%%mm1, %%mm0, (%%REGa) , (%%REGa, %1) , (%%REGa, %1, 2))
1644 DEINT_L5(%%mm0, %%mm1, (%%REGa, %1) , (%%REGa, %1, 2), (%0, %1, 4) )
1645 DEINT_L5(%%mm1, %%mm0, (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd) )
1646 DEINT_L5(%%mm0, %%mm1, (%0, %1, 4) , (%%REGd) , (%%REGd, %1) )
1647 DEINT_L5(%%mm1, %%mm0, (%%REGd) , (%%REGd, %1) , (%%REGd, %1, 2))
1648 DEINT_L5(%%mm0, %%mm1, (%%REGd, %1) , (%%REGd, %1, 2), (%0, %1, 8) )
1649 DEINT_L5(%%mm1, %%mm0, (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))
1651 "movq %%mm0, (%2) \n\t"
1652 "movq %%mm1, (%3) \n\t"
1653 : : "r" (src), "r" ((x86_reg)stride), "r"(tmp), "r"(tmp2)
1654 : "%"REG_a, "%"REG_d
1656 #else //HAVE_MMX2 || HAVE_AMD3DNOW
1664 src[stride*0]= CLIP((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);
1666 src[stride*1]= CLIP((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);
1668 src[stride*2]= CLIP((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);
1670 src[stride*3]= CLIP((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);
1672 src[stride*4]= CLIP((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);
1674 src[stride*5]= CLIP((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);
1676 src[stride*6]= CLIP((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);
1678 src[stride*7]= CLIP((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);
1685 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
1689 * Deinterlace the given block by filtering all lines with a (1 2 1) filter.
1690 * will be called for every 8x8 block and can read & write from line 4-15
1691 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1692 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1693 * this filter will read lines 4-13 and write 4-11
1695 static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uint8_t *tmp)
1697 #if HAVE_MMX2 || HAVE_AMD3DNOW
1700 "lea (%0, %1), %%"REG_a" \n\t"
1701 "lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1702 // 0 1 2 3 4 5 6 7 8 9
1703 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1705 "movq (%2), %%mm0 \n\t" // L0
1706 "movq (%%"REG_a"), %%mm1 \n\t" // L2
1707 PAVGB(%%mm1, %%mm0) // L0+L2
1708 "movq (%0), %%mm2 \n\t" // L1
1710 "movq %%mm0, (%0) \n\t"
1711 "movq (%%"REG_a", %1), %%mm0 \n\t" // L3
1712 PAVGB(%%mm0, %%mm2) // L1+L3
1713 PAVGB(%%mm1, %%mm2) // 2L2 + L1 + L3
1714 "movq %%mm2, (%%"REG_a") \n\t"
1715 "movq (%%"REG_a", %1, 2), %%mm2 \n\t" // L4
1716 PAVGB(%%mm2, %%mm1) // L2+L4
1717 PAVGB(%%mm0, %%mm1) // 2L3 + L2 + L4
1718 "movq %%mm1, (%%"REG_a", %1) \n\t"
1719 "movq (%0, %1, 4), %%mm1 \n\t" // L5
1720 PAVGB(%%mm1, %%mm0) // L3+L5
1721 PAVGB(%%mm2, %%mm0) // 2L4 + L3 + L5
1722 "movq %%mm0, (%%"REG_a", %1, 2) \n\t"
1723 "movq (%%"REG_d"), %%mm0 \n\t" // L6
1724 PAVGB(%%mm0, %%mm2) // L4+L6
1725 PAVGB(%%mm1, %%mm2) // 2L5 + L4 + L6
1726 "movq %%mm2, (%0, %1, 4) \n\t"
1727 "movq (%%"REG_d", %1), %%mm2 \n\t" // L7
1728 PAVGB(%%mm2, %%mm1) // L5+L7
1729 PAVGB(%%mm0, %%mm1) // 2L6 + L5 + L7
1730 "movq %%mm1, (%%"REG_d") \n\t"
1731 "movq (%%"REG_d", %1, 2), %%mm1 \n\t" // L8
1732 PAVGB(%%mm1, %%mm0) // L6+L8
1733 PAVGB(%%mm2, %%mm0) // 2L7 + L6 + L8
1734 "movq %%mm0, (%%"REG_d", %1) \n\t"
1735 "movq (%0, %1, 8), %%mm0 \n\t" // L9
1736 PAVGB(%%mm0, %%mm2) // L7+L9
1737 PAVGB(%%mm1, %%mm2) // 2L8 + L7 + L9
1738 "movq %%mm2, (%%"REG_d", %1, 2) \n\t"
1739 "movq %%mm1, (%2) \n\t"
1741 : : "r" (src), "r" ((x86_reg)stride), "r" (tmp)
1742 : "%"REG_a, "%"REG_d
1744 #else //HAVE_MMX2 || HAVE_AMD3DNOW
1749 a= *(uint32_t*)&tmp[stride*0];
1750 b= *(uint32_t*)&src[stride*0];
1751 c= *(uint32_t*)&src[stride*1];
1752 a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1753 *(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1755 a= *(uint32_t*)&src[stride*2];
1756 b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1757 *(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1759 b= *(uint32_t*)&src[stride*3];
1760 c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1761 *(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1763 c= *(uint32_t*)&src[stride*4];
1764 a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1765 *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1767 a= *(uint32_t*)&src[stride*5];
1768 b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1769 *(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1771 b= *(uint32_t*)&src[stride*6];
1772 c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1773 *(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1775 c= *(uint32_t*)&src[stride*7];
1776 a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1777 *(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1779 a= *(uint32_t*)&src[stride*8];
1780 b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1781 *(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1783 *(uint32_t*)&tmp[stride*0]= c;
1787 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
1791 * Deinterlace the given block by applying a median filter to every second line.
1792 * will be called for every 8x8 block and can read & write from line 4-15,
1793 * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1794 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1796 static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)
1802 "lea (%0, %1), %%"REG_a" \n\t"
1803 "lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1804 // 0 1 2 3 4 5 6 7 8 9
1805 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1807 "movq (%0), %%mm0 \n\t" //
1808 "movq (%%"REG_a", %1), %%mm2 \n\t" //
1809 "movq (%%"REG_a"), %%mm1 \n\t" //
1810 "movq %%mm0, %%mm3 \n\t"
1811 "pmaxub %%mm1, %%mm0 \n\t" //
1812 "pminub %%mm3, %%mm1 \n\t" //
1813 "pmaxub %%mm2, %%mm1 \n\t" //
1814 "pminub %%mm1, %%mm0 \n\t"
1815 "movq %%mm0, (%%"REG_a") \n\t"
1817 "movq (%0, %1, 4), %%mm0 \n\t" //
1818 "movq (%%"REG_a", %1, 2), %%mm1 \n\t" //
1819 "movq %%mm2, %%mm3 \n\t"
1820 "pmaxub %%mm1, %%mm2 \n\t" //
1821 "pminub %%mm3, %%mm1 \n\t" //
1822 "pmaxub %%mm0, %%mm1 \n\t" //
1823 "pminub %%mm1, %%mm2 \n\t"
1824 "movq %%mm2, (%%"REG_a", %1, 2) \n\t"
1826 "movq (%%"REG_d"), %%mm2 \n\t" //
1827 "movq (%%"REG_d", %1), %%mm1 \n\t" //
1828 "movq %%mm2, %%mm3 \n\t"
1829 "pmaxub %%mm0, %%mm2 \n\t" //
1830 "pminub %%mm3, %%mm0 \n\t" //
1831 "pmaxub %%mm1, %%mm0 \n\t" //
1832 "pminub %%mm0, %%mm2 \n\t"
1833 "movq %%mm2, (%%"REG_d") \n\t"
1835 "movq (%%"REG_d", %1, 2), %%mm2 \n\t" //
1836 "movq (%0, %1, 8), %%mm0 \n\t" //
1837 "movq %%mm2, %%mm3 \n\t"
1838 "pmaxub %%mm0, %%mm2 \n\t" //
1839 "pminub %%mm3, %%mm0 \n\t" //
1840 "pmaxub %%mm1, %%mm0 \n\t" //
1841 "pminub %%mm0, %%mm2 \n\t"
1842 "movq %%mm2, (%%"REG_d", %1, 2) \n\t"
1845 : : "r" (src), "r" ((x86_reg)stride)
1846 : "%"REG_a, "%"REG_d
1849 #else // MMX without MMX2
1851 "lea (%0, %1), %%"REG_a" \n\t"
1852 "lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1853 // 0 1 2 3 4 5 6 7 8 9
1854 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1855 "pxor %%mm7, %%mm7 \n\t"
1857 #define REAL_MEDIAN(a,b,c)\
1858 "movq " #a ", %%mm0 \n\t"\
1859 "movq " #b ", %%mm2 \n\t"\
1860 "movq " #c ", %%mm1 \n\t"\
1861 "movq %%mm0, %%mm3 \n\t"\
1862 "movq %%mm1, %%mm4 \n\t"\
1863 "movq %%mm2, %%mm5 \n\t"\
1864 "psubusb %%mm1, %%mm3 \n\t"\
1865 "psubusb %%mm2, %%mm4 \n\t"\
1866 "psubusb %%mm0, %%mm5 \n\t"\
1867 "pcmpeqb %%mm7, %%mm3 \n\t"\
1868 "pcmpeqb %%mm7, %%mm4 \n\t"\
1869 "pcmpeqb %%mm7, %%mm5 \n\t"\
1870 "movq %%mm3, %%mm6 \n\t"\
1871 "pxor %%mm4, %%mm3 \n\t"\
1872 "pxor %%mm5, %%mm4 \n\t"\
1873 "pxor %%mm6, %%mm5 \n\t"\
1874 "por %%mm3, %%mm1 \n\t"\
1875 "por %%mm4, %%mm2 \n\t"\
1876 "por %%mm5, %%mm0 \n\t"\
1877 "pand %%mm2, %%mm0 \n\t"\
1878 "pand %%mm1, %%mm0 \n\t"\
1879 "movq %%mm0, " #b " \n\t"
1880 #define MEDIAN(a,b,c) REAL_MEDIAN(a,b,c)
1882 MEDIAN((%0) , (%%REGa) , (%%REGa, %1))
1883 MEDIAN((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4))
1884 MEDIAN((%0, %1, 4) , (%%REGd) , (%%REGd, %1))
1885 MEDIAN((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8))
1887 : : "r" (src), "r" ((x86_reg)stride)
1888 : "%"REG_a, "%"REG_d
1894 // FIXME - there should be a way to do a few columns in parallel like w/mmx
1896 uint8_t *colsrc = src;
1897 for (y=0; y<4; y++){
1898 int a, b, c, d, e, f;
1900 b = colsrc[stride ];
1901 c = colsrc[stride*2];
1905 colsrc[stride ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));
1915 * transposes and shift the given 8x8 Block into dst1 and dst2
1917 static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)
1920 "lea (%0, %1), %%"REG_a" \n\t"
1921 // 0 1 2 3 4 5 6 7 8 9
1922 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1923 "movq (%0), %%mm0 \n\t" // 12345678
1924 "movq (%%"REG_a"), %%mm1 \n\t" // abcdefgh
1925 "movq %%mm0, %%mm2 \n\t" // 12345678
1926 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
1927 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
1929 "movq (%%"REG_a", %1), %%mm1 \n\t"
1930 "movq (%%"REG_a", %1, 2), %%mm3 \n\t"
1931 "movq %%mm1, %%mm4 \n\t"
1932 "punpcklbw %%mm3, %%mm1 \n\t"
1933 "punpckhbw %%mm3, %%mm4 \n\t"
1935 "movq %%mm0, %%mm3 \n\t"
1936 "punpcklwd %%mm1, %%mm0 \n\t"
1937 "punpckhwd %%mm1, %%mm3 \n\t"
1938 "movq %%mm2, %%mm1 \n\t"
1939 "punpcklwd %%mm4, %%mm2 \n\t"
1940 "punpckhwd %%mm4, %%mm1 \n\t"
1942 "movd %%mm0, 128(%2) \n\t"
1943 "psrlq $32, %%mm0 \n\t"
1944 "movd %%mm0, 144(%2) \n\t"
1945 "movd %%mm3, 160(%2) \n\t"
1946 "psrlq $32, %%mm3 \n\t"
1947 "movd %%mm3, 176(%2) \n\t"
1948 "movd %%mm3, 48(%3) \n\t"
1949 "movd %%mm2, 192(%2) \n\t"
1950 "movd %%mm2, 64(%3) \n\t"
1951 "psrlq $32, %%mm2 \n\t"
1952 "movd %%mm2, 80(%3) \n\t"
1953 "movd %%mm1, 96(%3) \n\t"
1954 "psrlq $32, %%mm1 \n\t"
1955 "movd %%mm1, 112(%3) \n\t"
1957 "lea (%%"REG_a", %1, 4), %%"REG_a" \n\t"
1959 "movq (%0, %1, 4), %%mm0 \n\t" // 12345678
1960 "movq (%%"REG_a"), %%mm1 \n\t" // abcdefgh
1961 "movq %%mm0, %%mm2 \n\t" // 12345678
1962 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
1963 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
1965 "movq (%%"REG_a", %1), %%mm1 \n\t"
1966 "movq (%%"REG_a", %1, 2), %%mm3 \n\t"
1967 "movq %%mm1, %%mm4 \n\t"
1968 "punpcklbw %%mm3, %%mm1 \n\t"
1969 "punpckhbw %%mm3, %%mm4 \n\t"
1971 "movq %%mm0, %%mm3 \n\t"
1972 "punpcklwd %%mm1, %%mm0 \n\t"
1973 "punpckhwd %%mm1, %%mm3 \n\t"
1974 "movq %%mm2, %%mm1 \n\t"
1975 "punpcklwd %%mm4, %%mm2 \n\t"
1976 "punpckhwd %%mm4, %%mm1 \n\t"
1978 "movd %%mm0, 132(%2) \n\t"
1979 "psrlq $32, %%mm0 \n\t"
1980 "movd %%mm0, 148(%2) \n\t"
1981 "movd %%mm3, 164(%2) \n\t"
1982 "psrlq $32, %%mm3 \n\t"
1983 "movd %%mm3, 180(%2) \n\t"
1984 "movd %%mm3, 52(%3) \n\t"
1985 "movd %%mm2, 196(%2) \n\t"
1986 "movd %%mm2, 68(%3) \n\t"
1987 "psrlq $32, %%mm2 \n\t"
1988 "movd %%mm2, 84(%3) \n\t"
1989 "movd %%mm1, 100(%3) \n\t"
1990 "psrlq $32, %%mm1 \n\t"
1991 "movd %%mm1, 116(%3) \n\t"
1994 :: "r" (src), "r" ((x86_reg)srcStride), "r" (dst1), "r" (dst2)
2000 * transposes the given 8x8 block
2002 static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)
2005 "lea (%0, %1), %%"REG_a" \n\t"
2006 "lea (%%"REG_a",%1,4), %%"REG_d" \n\t"
2007 // 0 1 2 3 4 5 6 7 8 9
2008 // %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
2009 "movq (%2), %%mm0 \n\t" // 12345678
2010 "movq 16(%2), %%mm1 \n\t" // abcdefgh
2011 "movq %%mm0, %%mm2 \n\t" // 12345678
2012 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
2013 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
2015 "movq 32(%2), %%mm1 \n\t"
2016 "movq 48(%2), %%mm3 \n\t"
2017 "movq %%mm1, %%mm4 \n\t"
2018 "punpcklbw %%mm3, %%mm1 \n\t"
2019 "punpckhbw %%mm3, %%mm4 \n\t"
2021 "movq %%mm0, %%mm3 \n\t"
2022 "punpcklwd %%mm1, %%mm0 \n\t"
2023 "punpckhwd %%mm1, %%mm3 \n\t"
2024 "movq %%mm2, %%mm1 \n\t"
2025 "punpcklwd %%mm4, %%mm2 \n\t"
2026 "punpckhwd %%mm4, %%mm1 \n\t"
2028 "movd %%mm0, (%0) \n\t"
2029 "psrlq $32, %%mm0 \n\t"
2030 "movd %%mm0, (%%"REG_a") \n\t"
2031 "movd %%mm3, (%%"REG_a", %1) \n\t"
2032 "psrlq $32, %%mm3 \n\t"
2033 "movd %%mm3, (%%"REG_a", %1, 2) \n\t"
2034 "movd %%mm2, (%0, %1, 4) \n\t"
2035 "psrlq $32, %%mm2 \n\t"
2036 "movd %%mm2, (%%"REG_d") \n\t"
2037 "movd %%mm1, (%%"REG_d", %1) \n\t"
2038 "psrlq $32, %%mm1 \n\t"
2039 "movd %%mm1, (%%"REG_d", %1, 2) \n\t"
2042 "movq 64(%2), %%mm0 \n\t" // 12345678
2043 "movq 80(%2), %%mm1 \n\t" // abcdefgh
2044 "movq %%mm0, %%mm2 \n\t" // 12345678
2045 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
2046 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
2048 "movq 96(%2), %%mm1 \n\t"
2049 "movq 112(%2), %%mm3 \n\t"
2050 "movq %%mm1, %%mm4 \n\t"
2051 "punpcklbw %%mm3, %%mm1 \n\t"
2052 "punpckhbw %%mm3, %%mm4 \n\t"
2054 "movq %%mm0, %%mm3 \n\t"
2055 "punpcklwd %%mm1, %%mm0 \n\t"
2056 "punpckhwd %%mm1, %%mm3 \n\t"
2057 "movq %%mm2, %%mm1 \n\t"
2058 "punpcklwd %%mm4, %%mm2 \n\t"
2059 "punpckhwd %%mm4, %%mm1 \n\t"
2061 "movd %%mm0, 4(%0) \n\t"
2062 "psrlq $32, %%mm0 \n\t"
2063 "movd %%mm0, 4(%%"REG_a") \n\t"
2064 "movd %%mm3, 4(%%"REG_a", %1) \n\t"
2065 "psrlq $32, %%mm3 \n\t"
2066 "movd %%mm3, 4(%%"REG_a", %1, 2) \n\t"
2067 "movd %%mm2, 4(%0, %1, 4) \n\t"
2068 "psrlq $32, %%mm2 \n\t"
2069 "movd %%mm2, 4(%%"REG_d") \n\t"
2070 "movd %%mm1, 4(%%"REG_d", %1) \n\t"
2071 "psrlq $32, %%mm1 \n\t"
2072 "movd %%mm1, 4(%%"REG_d", %1, 2) \n\t"
2074 :: "r" (dst), "r" ((x86_reg)dstStride), "r" (src)
2075 : "%"REG_a, "%"REG_d
2079 //static long test=0;
2082 static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
2083 uint8_t *tempBlurred, uint32_t *tempBlurredPast, int *maxNoise)
2085 // to save a register (FIXME do this outside of the loops)
2086 tempBlurredPast[127]= maxNoise[0];
2087 tempBlurredPast[128]= maxNoise[1];
2088 tempBlurredPast[129]= maxNoise[2];
2090 #define FAST_L2_DIFF
2091 //#define L1_DIFF //u should change the thresholds too if u try that one
2092 #if HAVE_MMX2 || HAVE_AMD3DNOW
2094 "lea (%2, %2, 2), %%"REG_a" \n\t" // 3*stride
2095 "lea (%2, %2, 4), %%"REG_d" \n\t" // 5*stride
2096 "lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
2097 // 0 1 2 3 4 5 6 7 8 9
2098 // %x %x+%2 %x+2%2 %x+eax %x+4%2 %x+edx %x+2eax %x+ecx %x+8%2
2100 #ifdef L1_DIFF //needs mmx2
2101 "movq (%0), %%mm0 \n\t" // L0
2102 "psadbw (%1), %%mm0 \n\t" // |L0-R0|
2103 "movq (%0, %2), %%mm1 \n\t" // L1
2104 "psadbw (%1, %2), %%mm1 \n\t" // |L1-R1|
2105 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2106 "psadbw (%1, %2, 2), %%mm2 \n\t" // |L2-R2|
2107 "movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
2108 "psadbw (%1, %%"REG_a"), %%mm3 \n\t" // |L3-R3|
2110 "movq (%0, %2, 4), %%mm4 \n\t" // L4
2111 "paddw %%mm1, %%mm0 \n\t"
2112 "psadbw (%1, %2, 4), %%mm4 \n\t" // |L4-R4|
2113 "movq (%0, %%"REG_d"), %%mm5 \n\t" // L5
2114 "paddw %%mm2, %%mm0 \n\t"
2115 "psadbw (%1, %%"REG_d"), %%mm5 \n\t" // |L5-R5|
2116 "movq (%0, %%"REG_a", 2), %%mm6 \n\t" // L6
2117 "paddw %%mm3, %%mm0 \n\t"
2118 "psadbw (%1, %%"REG_a", 2), %%mm6 \n\t" // |L6-R6|
2119 "movq (%0, %%"REG_c"), %%mm7 \n\t" // L7
2120 "paddw %%mm4, %%mm0 \n\t"
2121 "psadbw (%1, %%"REG_c"), %%mm7 \n\t" // |L7-R7|
2122 "paddw %%mm5, %%mm6 \n\t"
2123 "paddw %%mm7, %%mm6 \n\t"
2124 "paddw %%mm6, %%mm0 \n\t"
2126 #if defined (FAST_L2_DIFF)
2127 "pcmpeqb %%mm7, %%mm7 \n\t"
2128 "movq "MANGLE(b80)", %%mm6 \n\t"
2129 "pxor %%mm0, %%mm0 \n\t"
2130 #define REAL_L2_DIFF_CORE(a, b)\
2131 "movq " #a ", %%mm5 \n\t"\
2132 "movq " #b ", %%mm2 \n\t"\
2133 "pxor %%mm7, %%mm2 \n\t"\
2134 PAVGB(%%mm2, %%mm5)\
2135 "paddb %%mm6, %%mm5 \n\t"\
2136 "movq %%mm5, %%mm2 \n\t"\
2137 "psllw $8, %%mm5 \n\t"\
2138 "pmaddwd %%mm5, %%mm5 \n\t"\
2139 "pmaddwd %%mm2, %%mm2 \n\t"\
2140 "paddd %%mm2, %%mm5 \n\t"\
2141 "psrld $14, %%mm5 \n\t"\
2142 "paddd %%mm5, %%mm0 \n\t"
2144 #else //defined (FAST_L2_DIFF)
2145 "pxor %%mm7, %%mm7 \n\t"
2146 "pxor %%mm0, %%mm0 \n\t"
2147 #define REAL_L2_DIFF_CORE(a, b)\
2148 "movq " #a ", %%mm5 \n\t"\
2149 "movq " #b ", %%mm2 \n\t"\
2150 "movq %%mm5, %%mm1 \n\t"\
2151 "movq %%mm2, %%mm3 \n\t"\
2152 "punpcklbw %%mm7, %%mm5 \n\t"\
2153 "punpckhbw %%mm7, %%mm1 \n\t"\
2154 "punpcklbw %%mm7, %%mm2 \n\t"\
2155 "punpckhbw %%mm7, %%mm3 \n\t"\
2156 "psubw %%mm2, %%mm5 \n\t"\
2157 "psubw %%mm3, %%mm1 \n\t"\
2158 "pmaddwd %%mm5, %%mm5 \n\t"\
2159 "pmaddwd %%mm1, %%mm1 \n\t"\
2160 "paddd %%mm1, %%mm5 \n\t"\
2161 "paddd %%mm5, %%mm0 \n\t"
2163 #endif //defined (FAST_L2_DIFF)
2165 #define L2_DIFF_CORE(a, b) REAL_L2_DIFF_CORE(a, b)
2167 L2_DIFF_CORE((%0) , (%1))
2168 L2_DIFF_CORE((%0, %2) , (%1, %2))
2169 L2_DIFF_CORE((%0, %2, 2) , (%1, %2, 2))
2170 L2_DIFF_CORE((%0, %%REGa) , (%1, %%REGa))
2171 L2_DIFF_CORE((%0, %2, 4) , (%1, %2, 4))
2172 L2_DIFF_CORE((%0, %%REGd) , (%1, %%REGd))
2173 L2_DIFF_CORE((%0, %%REGa,2), (%1, %%REGa,2))
2174 L2_DIFF_CORE((%0, %%REGc) , (%1, %%REGc))
2178 "movq %%mm0, %%mm4 \n\t"
2179 "psrlq $32, %%mm0 \n\t"
2180 "paddd %%mm0, %%mm4 \n\t"
2181 "movd %%mm4, %%ecx \n\t"
2182 "shll $2, %%ecx \n\t"
2183 "mov %3, %%"REG_d" \n\t"
2184 "addl -4(%%"REG_d"), %%ecx \n\t"
2185 "addl 4(%%"REG_d"), %%ecx \n\t"
2186 "addl -1024(%%"REG_d"), %%ecx \n\t"
2187 "addl $4, %%ecx \n\t"
2188 "addl 1024(%%"REG_d"), %%ecx \n\t"
2189 "shrl $3, %%ecx \n\t"
2190 "movl %%ecx, (%%"REG_d") \n\t"
2192 // "mov %3, %%"REG_c" \n\t"
2193 // "mov %%"REG_c", test \n\t"
2195 "cmpl 512(%%"REG_d"), %%ecx \n\t"
2197 "cmpl 516(%%"REG_d"), %%ecx \n\t"
2200 "lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
2201 "lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
2202 "movq (%0), %%mm0 \n\t" // L0
2203 "movq (%0, %2), %%mm1 \n\t" // L1
2204 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2205 "movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
2206 "movq (%0, %2, 4), %%mm4 \n\t" // L4
2207 "movq (%0, %%"REG_d"), %%mm5 \n\t" // L5
2208 "movq (%0, %%"REG_a", 2), %%mm6 \n\t" // L6
2209 "movq (%0, %%"REG_c"), %%mm7 \n\t" // L7
2210 "movq %%mm0, (%1) \n\t" // L0
2211 "movq %%mm1, (%1, %2) \n\t" // L1
2212 "movq %%mm2, (%1, %2, 2) \n\t" // L2
2213 "movq %%mm3, (%1, %%"REG_a") \n\t" // L3
2214 "movq %%mm4, (%1, %2, 4) \n\t" // L4
2215 "movq %%mm5, (%1, %%"REG_d") \n\t" // L5
2216 "movq %%mm6, (%1, %%"REG_a", 2) \n\t" // L6
2217 "movq %%mm7, (%1, %%"REG_c") \n\t" // L7
2221 "lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
2222 "lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
2223 "movq (%0), %%mm0 \n\t" // L0
2224 PAVGB((%1), %%mm0) // L0
2225 "movq (%0, %2), %%mm1 \n\t" // L1
2226 PAVGB((%1, %2), %%mm1) // L1
2227 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2228 PAVGB((%1, %2, 2), %%mm2) // L2
2229 "movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
2230 PAVGB((%1, %%REGa), %%mm3) // L3
2231 "movq (%0, %2, 4), %%mm4 \n\t" // L4
2232 PAVGB((%1, %2, 4), %%mm4) // L4
2233 "movq (%0, %%"REG_d"), %%mm5 \n\t" // L5
2234 PAVGB((%1, %%REGd), %%mm5) // L5
2235 "movq (%0, %%"REG_a", 2), %%mm6 \n\t" // L6
2236 PAVGB((%1, %%REGa, 2), %%mm6) // L6
2237 "movq (%0, %%"REG_c"), %%mm7 \n\t" // L7
2238 PAVGB((%1, %%REGc), %%mm7) // L7
2239 "movq %%mm0, (%1) \n\t" // R0
2240 "movq %%mm1, (%1, %2) \n\t" // R1
2241 "movq %%mm2, (%1, %2, 2) \n\t" // R2
2242 "movq %%mm3, (%1, %%"REG_a") \n\t" // R3
2243 "movq %%mm4, (%1, %2, 4) \n\t" // R4
2244 "movq %%mm5, (%1, %%"REG_d") \n\t" // R5
2245 "movq %%mm6, (%1, %%"REG_a", 2) \n\t" // R6
2246 "movq %%mm7, (%1, %%"REG_c") \n\t" // R7
2247 "movq %%mm0, (%0) \n\t" // L0
2248 "movq %%mm1, (%0, %2) \n\t" // L1
2249 "movq %%mm2, (%0, %2, 2) \n\t" // L2
2250 "movq %%mm3, (%0, %%"REG_a") \n\t" // L3
2251 "movq %%mm4, (%0, %2, 4) \n\t" // L4
2252 "movq %%mm5, (%0, %%"REG_d") \n\t" // L5
2253 "movq %%mm6, (%0, %%"REG_a", 2) \n\t" // L6
2254 "movq %%mm7, (%0, %%"REG_c") \n\t" // L7
2258 "cmpl 508(%%"REG_d"), %%ecx \n\t"
2261 "lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
2262 "lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
2263 "movq (%0), %%mm0 \n\t" // L0
2264 "movq (%0, %2), %%mm1 \n\t" // L1
2265 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2266 "movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
2267 "movq (%1), %%mm4 \n\t" // R0
2268 "movq (%1, %2), %%mm5 \n\t" // R1
2269 "movq (%1, %2, 2), %%mm6 \n\t" // R2
2270 "movq (%1, %%"REG_a"), %%mm7 \n\t" // R3
2279 "movq %%mm0, (%1) \n\t" // R0
2280 "movq %%mm1, (%1, %2) \n\t" // R1
2281 "movq %%mm2, (%1, %2, 2) \n\t" // R2
2282 "movq %%mm3, (%1, %%"REG_a") \n\t" // R3
2283 "movq %%mm0, (%0) \n\t" // L0
2284 "movq %%mm1, (%0, %2) \n\t" // L1
2285 "movq %%mm2, (%0, %2, 2) \n\t" // L2
2286 "movq %%mm3, (%0, %%"REG_a") \n\t" // L3
2288 "movq (%0, %2, 4), %%mm0 \n\t" // L4
2289 "movq (%0, %%"REG_d"), %%mm1 \n\t" // L5
2290 "movq (%0, %%"REG_a", 2), %%mm2 \n\t" // L6
2291 "movq (%0, %%"REG_c"), %%mm3 \n\t" // L7
2292 "movq (%1, %2, 4), %%mm4 \n\t" // R4
2293 "movq (%1, %%"REG_d"), %%mm5 \n\t" // R5
2294 "movq (%1, %%"REG_a", 2), %%mm6 \n\t" // R6
2295 "movq (%1, %%"REG_c"), %%mm7 \n\t" // R7
2304 "movq %%mm0, (%1, %2, 4) \n\t" // R4
2305 "movq %%mm1, (%1, %%"REG_d") \n\t" // R5
2306 "movq %%mm2, (%1, %%"REG_a", 2) \n\t" // R6
2307 "movq %%mm3, (%1, %%"REG_c") \n\t" // R7
2308 "movq %%mm0, (%0, %2, 4) \n\t" // L4
2309 "movq %%mm1, (%0, %%"REG_d") \n\t" // L5
2310 "movq %%mm2, (%0, %%"REG_a", 2) \n\t" // L6
2311 "movq %%mm3, (%0, %%"REG_c") \n\t" // L7
2315 "lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
2316 "lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
2317 "movq (%0), %%mm0 \n\t" // L0
2318 "movq (%0, %2), %%mm1 \n\t" // L1
2319 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2320 "movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
2321 "movq (%1), %%mm4 \n\t" // R0
2322 "movq (%1, %2), %%mm5 \n\t" // R1
2323 "movq (%1, %2, 2), %%mm6 \n\t" // R2
2324 "movq (%1, %%"REG_a"), %%mm7 \n\t" // R3
2337 "movq %%mm0, (%1) \n\t" // R0
2338 "movq %%mm1, (%1, %2) \n\t" // R1
2339 "movq %%mm2, (%1, %2, 2) \n\t" // R2
2340 "movq %%mm3, (%1, %%"REG_a") \n\t" // R3
2341 "movq %%mm0, (%0) \n\t" // L0
2342 "movq %%mm1, (%0, %2) \n\t" // L1
2343 "movq %%mm2, (%0, %2, 2) \n\t" // L2
2344 "movq %%mm3, (%0, %%"REG_a") \n\t" // L3
2346 "movq (%0, %2, 4), %%mm0 \n\t" // L4
2347 "movq (%0, %%"REG_d"), %%mm1 \n\t" // L5
2348 "movq (%0, %%"REG_a", 2), %%mm2 \n\t" // L6
2349 "movq (%0, %%"REG_c"), %%mm3 \n\t" // L7
2350 "movq (%1, %2, 4), %%mm4 \n\t" // R4
2351 "movq (%1, %%"REG_d"), %%mm5 \n\t" // R5
2352 "movq (%1, %%"REG_a", 2), %%mm6 \n\t" // R6
2353 "movq (%1, %%"REG_c"), %%mm7 \n\t" // R7
2366 "movq %%mm0, (%1, %2, 4) \n\t" // R4
2367 "movq %%mm1, (%1, %%"REG_d") \n\t" // R5
2368 "movq %%mm2, (%1, %%"REG_a", 2) \n\t" // R6
2369 "movq %%mm3, (%1, %%"REG_c") \n\t" // R7
2370 "movq %%mm0, (%0, %2, 4) \n\t" // L4
2371 "movq %%mm1, (%0, %%"REG_d") \n\t" // L5
2372 "movq %%mm2, (%0, %%"REG_a", 2) \n\t" // L6
2373 "movq %%mm3, (%0, %%"REG_c") \n\t" // L7
2377 :: "r" (src), "r" (tempBlurred), "r"((x86_reg)stride), "m" (tempBlurredPast)
2378 : "%"REG_a, "%"REG_d, "%"REG_c, "memory"
2380 #else //HAVE_MMX2 || HAVE_AMD3DNOW
2390 int ref= tempBlurred[ x + y*stride ];
2391 int cur= src[ x + y*stride ];
2393 // if(x==0 || x==7) d1+= d1>>1;
2394 // if(y==0 || y==7) d1+= d1>>1;
2403 +(*(tempBlurredPast-256))
2404 +(*(tempBlurredPast-1))+ (*(tempBlurredPast+1))
2405 +(*(tempBlurredPast+256))
2408 // ((*tempBlurredPast)*3 + d + 2)>>2;
2413 64 32 16 8 4 2 1 (1)
2414 64 48 36 27 20 15 11 (33) (approx)
2415 64 56 49 43 37 33 29 (200) (approx)
2417 if(d > maxNoise[1]){
2418 if(d < maxNoise[2]){
2422 int ref= tempBlurred[ x + y*stride ];
2423 int cur= src[ x + y*stride ];
2424 tempBlurred[ x + y*stride ]=
2425 src[ x + y*stride ]=
2433 tempBlurred[ x + y*stride ]= src[ x + y*stride ];
2438 if(d < maxNoise[0]){
2442 int ref= tempBlurred[ x + y*stride ];
2443 int cur= src[ x + y*stride ];
2444 tempBlurred[ x + y*stride ]=
2445 src[ x + y*stride ]=
2446 (ref*7 + cur + 4)>>3;
2453 int ref= tempBlurred[ x + y*stride ];
2454 int cur= src[ x + y*stride ];
2455 tempBlurred[ x + y*stride ]=
2456 src[ x + y*stride ]=
2457 (ref*3 + cur + 2)>>2;
2463 #endif //HAVE_MMX2 || HAVE_AMD3DNOW
2465 #endif //HAVE_ALTIVEC
2469 * accurate deblock filter
2471 static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, PPContext *c){
2472 int64_t dc_mask, eq_mask, both_masks;
2473 int64_t sums[10*8*2];
2474 src+= step*3; // src points to begin of the 8x8 Block
2477 "movq %0, %%mm7 \n\t"
2478 "movq %1, %%mm6 \n\t"
2479 : : "m" (c->mmxDcOffset[c->nonBQP]), "m" (c->mmxDcThreshold[c->nonBQP])
2483 "lea (%2, %3), %%"REG_a" \n\t"
2484 // 0 1 2 3 4 5 6 7 8 9
2485 // %1 eax eax+%2 eax+2%2 %1+4%2 ecx ecx+%2 ecx+2%2 %1+8%2 ecx+4%2
2487 "movq (%2), %%mm0 \n\t"
2488 "movq (%%"REG_a"), %%mm1 \n\t"
2489 "movq %%mm1, %%mm3 \n\t"
2490 "movq %%mm1, %%mm4 \n\t"
2491 "psubb %%mm1, %%mm0 \n\t" // mm0 = differnece
2492 "paddb %%mm7, %%mm0 \n\t"
2493 "pcmpgtb %%mm6, %%mm0 \n\t"
2495 "movq (%%"REG_a",%3), %%mm2 \n\t"
2496 PMAXUB(%%mm2, %%mm4)
2497 PMINUB(%%mm2, %%mm3, %%mm5)
2498 "psubb %%mm2, %%mm1 \n\t"
2499 "paddb %%mm7, %%mm1 \n\t"
2500 "pcmpgtb %%mm6, %%mm1 \n\t"
2501 "paddb %%mm1, %%mm0 \n\t"
2503 "movq (%%"REG_a", %3, 2), %%mm1 \n\t"
2504 PMAXUB(%%mm1, %%mm4)
2505 PMINUB(%%mm1, %%mm3, %%mm5)
2506 "psubb %%mm1, %%mm2 \n\t"
2507 "paddb %%mm7, %%mm2 \n\t"
2508 "pcmpgtb %%mm6, %%mm2 \n\t"
2509 "paddb %%mm2, %%mm0 \n\t"
2511 "lea (%%"REG_a", %3, 4), %%"REG_a" \n\t"
2513 "movq (%2, %3, 4), %%mm2 \n\t"
2514 PMAXUB(%%mm2, %%mm4)
2515 PMINUB(%%mm2, %%mm3, %%mm5)
2516 "psubb %%mm2, %%mm1 \n\t"
2517 "paddb %%mm7, %%mm1 \n\t"
2518 "pcmpgtb %%mm6, %%mm1 \n\t"
2519 "paddb %%mm1, %%mm0 \n\t"
2521 "movq (%%"REG_a"), %%mm1 \n\t"
2522 PMAXUB(%%mm1, %%mm4)
2523 PMINUB(%%mm1, %%mm3, %%mm5)
2524 "psubb %%mm1, %%mm2 \n\t"
2525 "paddb %%mm7, %%mm2 \n\t"
2526 "pcmpgtb %%mm6, %%mm2 \n\t"
2527 "paddb %%mm2, %%mm0 \n\t"
2529 "movq (%%"REG_a", %3), %%mm2 \n\t"
2530 PMAXUB(%%mm2, %%mm4)
2531 PMINUB(%%mm2, %%mm3, %%mm5)
2532 "psubb %%mm2, %%mm1 \n\t"
2533 "paddb %%mm7, %%mm1 \n\t"
2534 "pcmpgtb %%mm6, %%mm1 \n\t"
2535 "paddb %%mm1, %%mm0 \n\t"
2537 "movq (%%"REG_a", %3, 2), %%mm1 \n\t"
2538 PMAXUB(%%mm1, %%mm4)
2539 PMINUB(%%mm1, %%mm3, %%mm5)
2540 "psubb %%mm1, %%mm2 \n\t"
2541 "paddb %%mm7, %%mm2 \n\t"
2542 "pcmpgtb %%mm6, %%mm2 \n\t"
2543 "paddb %%mm2, %%mm0 \n\t"
2545 "movq (%2, %3, 8), %%mm2 \n\t"
2546 PMAXUB(%%mm2, %%mm4)
2547 PMINUB(%%mm2, %%mm3, %%mm5)
2548 "psubb %%mm2, %%mm1 \n\t"
2549 "paddb %%mm7, %%mm1 \n\t"
2550 "pcmpgtb %%mm6, %%mm1 \n\t"
2551 "paddb %%mm1, %%mm0 \n\t"
2553 "movq (%%"REG_a", %3, 4), %%mm1 \n\t"
2554 "psubb %%mm1, %%mm2 \n\t"
2555 "paddb %%mm7, %%mm2 \n\t"
2556 "pcmpgtb %%mm6, %%mm2 \n\t"
2557 "paddb %%mm2, %%mm0 \n\t"
2558 "psubusb %%mm3, %%mm4 \n\t"
2560 "pxor %%mm6, %%mm6 \n\t"
2561 "movq %4, %%mm7 \n\t" // QP,..., QP
2562 "paddusb %%mm7, %%mm7 \n\t" // 2QP ... 2QP
2563 "psubusb %%mm4, %%mm7 \n\t" // Diff >=2QP -> 0
2564 "pcmpeqb %%mm6, %%mm7 \n\t" // Diff < 2QP -> 0
2565 "pcmpeqb %%mm6, %%mm7 \n\t" // Diff < 2QP -> 0
2566 "movq %%mm7, %1 \n\t"
2568 "movq %5, %%mm7 \n\t"
2569 "punpcklbw %%mm7, %%mm7 \n\t"
2570 "punpcklbw %%mm7, %%mm7 \n\t"
2571 "punpcklbw %%mm7, %%mm7 \n\t"
2572 "psubb %%mm0, %%mm6 \n\t"
2573 "pcmpgtb %%mm7, %%mm6 \n\t"
2574 "movq %%mm6, %0 \n\t"
2576 : "=m" (eq_mask), "=m" (dc_mask)
2577 : "r" (src), "r" ((x86_reg)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)
2581 both_masks = dc_mask & eq_mask;
2584 x86_reg offset= -8*step;
2585 int64_t *temp_sums= sums;
2588 "movq %2, %%mm0 \n\t" // QP,..., QP
2589 "pxor %%mm4, %%mm4 \n\t"
2591 "movq (%0), %%mm6 \n\t"
2592 "movq (%0, %1), %%mm5 \n\t"
2593 "movq %%mm5, %%mm1 \n\t"
2594 "movq %%mm6, %%mm2 \n\t"
2595 "psubusb %%mm6, %%mm5 \n\t"
2596 "psubusb %%mm1, %%mm2 \n\t"
2597 "por %%mm5, %%mm2 \n\t" // ABS Diff of lines
2598 "psubusb %%mm2, %%mm0 \n\t" // diff >= QP -> 0
2599 "pcmpeqb %%mm4, %%mm0 \n\t" // diff >= QP -> FF
2601 "pxor %%mm6, %%mm1 \n\t"
2602 "pand %%mm0, %%mm1 \n\t"
2603 "pxor %%mm1, %%mm6 \n\t"
2606 "movq (%0, %1, 8), %%mm5 \n\t"
2607 "add %1, %0 \n\t" // %0 points to line 1 not 0
2608 "movq (%0, %1, 8), %%mm7 \n\t"
2609 "movq %%mm5, %%mm1 \n\t"
2610 "movq %%mm7, %%mm2 \n\t"
2611 "psubusb %%mm7, %%mm5 \n\t"
2612 "psubusb %%mm1, %%mm2 \n\t"
2613 "por %%mm5, %%mm2 \n\t" // ABS Diff of lines
2614 "movq %2, %%mm0 \n\t" // QP,..., QP
2615 "psubusb %%mm2, %%mm0 \n\t" // diff >= QP -> 0
2616 "pcmpeqb %%mm4, %%mm0 \n\t" // diff >= QP -> FF
2618 "pxor %%mm7, %%mm1 \n\t"
2619 "pand %%mm0, %%mm1 \n\t"
2620 "pxor %%mm1, %%mm7 \n\t"
2622 "movq %%mm6, %%mm5 \n\t"
2623 "punpckhbw %%mm4, %%mm6 \n\t"
2624 "punpcklbw %%mm4, %%mm5 \n\t"
2625 // 4:0 5/6:First 7:Last
2627 "movq %%mm5, %%mm0 \n\t"
2628 "movq %%mm6, %%mm1 \n\t"
2629 "psllw $2, %%mm0 \n\t"
2630 "psllw $2, %%mm1 \n\t"
2631 "paddw "MANGLE(w04)", %%mm0 \n\t"
2632 "paddw "MANGLE(w04)", %%mm1 \n\t"
2635 "movq (%0), %%mm2 \n\t"\
2636 "movq (%0), %%mm3 \n\t"\
2638 "punpcklbw %%mm4, %%mm2 \n\t"\
2639 "punpckhbw %%mm4, %%mm3 \n\t"\
2640 "paddw %%mm2, %%mm0 \n\t"\
2641 "paddw %%mm3, %%mm1 \n\t"
2644 "movq (%0), %%mm2 \n\t"\
2645 "movq (%0), %%mm3 \n\t"\
2647 "punpcklbw %%mm4, %%mm2 \n\t"\
2648 "punpckhbw %%mm4, %%mm3 \n\t"\
2649 "psubw %%mm2, %%mm0 \n\t"\
2650 "psubw %%mm3, %%mm1 \n\t"
2656 "movq %%mm0, (%3) \n\t"
2657 "movq %%mm1, 8(%3) \n\t"
2660 "psubw %%mm5, %%mm0 \n\t"
2661 "psubw %%mm6, %%mm1 \n\t"
2662 "movq %%mm0, 16(%3) \n\t"
2663 "movq %%mm1, 24(%3) \n\t"
2666 "psubw %%mm5, %%mm0 \n\t"
2667 "psubw %%mm6, %%mm1 \n\t"
2668 "movq %%mm0, 32(%3) \n\t"
2669 "movq %%mm1, 40(%3) \n\t"
2672 "psubw %%mm5, %%mm0 \n\t"
2673 "psubw %%mm6, %%mm1 \n\t"
2674 "movq %%mm0, 48(%3) \n\t"
2675 "movq %%mm1, 56(%3) \n\t"
2678 "psubw %%mm5, %%mm0 \n\t"
2679 "psubw %%mm6, %%mm1 \n\t"
2680 "movq %%mm0, 64(%3) \n\t"
2681 "movq %%mm1, 72(%3) \n\t"
2683 "movq %%mm7, %%mm6 \n\t"
2684 "punpckhbw %%mm4, %%mm7 \n\t"
2685 "punpcklbw %%mm4, %%mm6 \n\t"
2691 "movq %%mm0, 80(%3) \n\t"
2692 "movq %%mm1, 88(%3) \n\t"
2695 "paddw %%mm6, %%mm0 \n\t"
2696 "paddw %%mm7, %%mm1 \n\t"
2697 "movq %%mm0, 96(%3) \n\t"
2698 "movq %%mm1, 104(%3) \n\t"
2701 "paddw %%mm6, %%mm0 \n\t"
2702 "paddw %%mm7, %%mm1 \n\t"
2703 "movq %%mm0, 112(%3) \n\t"
2704 "movq %%mm1, 120(%3) \n\t"
2707 "paddw %%mm6, %%mm0 \n\t"
2708 "paddw %%mm7, %%mm1 \n\t"
2709 "movq %%mm0, 128(%3) \n\t"
2710 "movq %%mm1, 136(%3) \n\t"
2713 "paddw %%mm6, %%mm0 \n\t"
2714 "paddw %%mm7, %%mm1 \n\t"
2715 "movq %%mm0, 144(%3) \n\t"
2716 "movq %%mm1, 152(%3) \n\t"
2718 "mov %4, %0 \n\t" //FIXME
2721 : "r" ((x86_reg)step), "m" (c->pQPb), "r"(sums), "g"(src)
2724 src+= step; // src points to begin of the 8x8 Block
2727 "movq %4, %%mm6 \n\t"
2728 "pcmpeqb %%mm5, %%mm5 \n\t"
2729 "pxor %%mm6, %%mm5 \n\t"
2730 "pxor %%mm7, %%mm7 \n\t"
2733 "movq (%1), %%mm0 \n\t"
2734 "movq 8(%1), %%mm1 \n\t"
2735 "paddw 32(%1), %%mm0 \n\t"
2736 "paddw 40(%1), %%mm1 \n\t"
2737 "movq (%0, %3), %%mm2 \n\t"
2738 "movq %%mm2, %%mm3 \n\t"
2739 "movq %%mm2, %%mm4 \n\t"
2740 "punpcklbw %%mm7, %%mm2 \n\t"
2741 "punpckhbw %%mm7, %%mm3 \n\t"
2742 "paddw %%mm2, %%mm0 \n\t"
2743 "paddw %%mm3, %%mm1 \n\t"
2744 "paddw %%mm2, %%mm0 \n\t"
2745 "paddw %%mm3, %%mm1 \n\t"
2746 "psrlw $4, %%mm0 \n\t"
2747 "psrlw $4, %%mm1 \n\t"
2748 "packuswb %%mm1, %%mm0 \n\t"
2749 "pand %%mm6, %%mm0 \n\t"
2750 "pand %%mm5, %%mm4 \n\t"
2751 "por %%mm4, %%mm0 \n\t"
2752 "movq %%mm0, (%0, %3) \n\t"
2757 : "+r"(offset), "+r"(temp_sums)
2758 : "r" ((x86_reg)step), "r"(src - offset), "m"(both_masks)
2761 src+= step; // src points to begin of the 8x8 Block
2763 if(eq_mask != -1LL){
2764 uint8_t *temp_src= src;
2766 "pxor %%mm7, %%mm7 \n\t"
2767 "lea -40(%%"REG_SP"), %%"REG_c" \n\t" // make space for 4 8-byte vars
2768 "and "ALIGN_MASK", %%"REG_c" \n\t" // align
2769 // 0 1 2 3 4 5 6 7 8 9
2770 // %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1 %1+8%1 ecx+4%1
2772 "movq (%0), %%mm0 \n\t"
2773 "movq %%mm0, %%mm1 \n\t"
2774 "punpcklbw %%mm7, %%mm0 \n\t" // low part of line 0
2775 "punpckhbw %%mm7, %%mm1 \n\t" // high part of line 0
2777 "movq (%0, %1), %%mm2 \n\t"
2778 "lea (%0, %1, 2), %%"REG_a" \n\t"
2779 "movq %%mm2, %%mm3 \n\t"
2780 "punpcklbw %%mm7, %%mm2 \n\t" // low part of line 1
2781 "punpckhbw %%mm7, %%mm3 \n\t" // high part of line 1
2783 "movq (%%"REG_a"), %%mm4 \n\t"
2784 "movq %%mm4, %%mm5 \n\t"
2785 "punpcklbw %%mm7, %%mm4 \n\t" // low part of line 2
2786 "punpckhbw %%mm7, %%mm5 \n\t" // high part of line 2
2788 "paddw %%mm0, %%mm0 \n\t" // 2L0
2789 "paddw %%mm1, %%mm1 \n\t" // 2H0
2790 "psubw %%mm4, %%mm2 \n\t" // L1 - L2
2791 "psubw %%mm5, %%mm3 \n\t" // H1 - H2
2792 "psubw %%mm2, %%mm0 \n\t" // 2L0 - L1 + L2
2793 "psubw %%mm3, %%mm1 \n\t" // 2H0 - H1 + H2
2795 "psllw $2, %%mm2 \n\t" // 4L1 - 4L2
2796 "psllw $2, %%mm3 \n\t" // 4H1 - 4H2
2797 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2
2798 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2
2800 "movq (%%"REG_a", %1), %%mm2 \n\t"
2801 "movq %%mm2, %%mm3 \n\t"
2802 "punpcklbw %%mm7, %%mm2 \n\t" // L3
2803 "punpckhbw %%mm7, %%mm3 \n\t" // H3
2805 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - L3
2806 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - H3
2807 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2808 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2809 "movq %%mm0, (%%"REG_c") \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2810 "movq %%mm1, 8(%%"REG_c") \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2812 "movq (%%"REG_a", %1, 2), %%mm0 \n\t"
2813 "movq %%mm0, %%mm1 \n\t"
2814 "punpcklbw %%mm7, %%mm0 \n\t" // L4
2815 "punpckhbw %%mm7, %%mm1 \n\t" // H4
2817 "psubw %%mm0, %%mm2 \n\t" // L3 - L4
2818 "psubw %%mm1, %%mm3 \n\t" // H3 - H4
2819 "movq %%mm2, 16(%%"REG_c") \n\t" // L3 - L4
2820 "movq %%mm3, 24(%%"REG_c") \n\t" // H3 - H4
2821 "paddw %%mm4, %%mm4 \n\t" // 2L2
2822 "paddw %%mm5, %%mm5 \n\t" // 2H2
2823 "psubw %%mm2, %%mm4 \n\t" // 2L2 - L3 + L4
2824 "psubw %%mm3, %%mm5 \n\t" // 2H2 - H3 + H4
2826 "lea (%%"REG_a", %1), %0 \n\t"
2827 "psllw $2, %%mm2 \n\t" // 4L3 - 4L4
2828 "psllw $2, %%mm3 \n\t" // 4H3 - 4H4
2829 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4
2830 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4
2832 "movq (%0, %1, 2), %%mm2 \n\t"
2833 "movq %%mm2, %%mm3 \n\t"
2834 "punpcklbw %%mm7, %%mm2 \n\t" // L5
2835 "punpckhbw %%mm7, %%mm3 \n\t" // H5
2836 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - L5
2837 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - H5
2838 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - 2L5
2839 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - 2H5
2841 "movq (%%"REG_a", %1, 4), %%mm6 \n\t"
2842 "punpcklbw %%mm7, %%mm6 \n\t" // L6
2843 "psubw %%mm6, %%mm2 \n\t" // L5 - L6
2844 "movq (%%"REG_a", %1, 4), %%mm6 \n\t"
2845 "punpckhbw %%mm7, %%mm6 \n\t" // H6
2846 "psubw %%mm6, %%mm3 \n\t" // H5 - H6
2848 "paddw %%mm0, %%mm0 \n\t" // 2L4
2849 "paddw %%mm1, %%mm1 \n\t" // 2H4
2850 "psubw %%mm2, %%mm0 \n\t" // 2L4 - L5 + L6
2851 "psubw %%mm3, %%mm1 \n\t" // 2H4 - H5 + H6
2853 "psllw $2, %%mm2 \n\t" // 4L5 - 4L6
2854 "psllw $2, %%mm3 \n\t" // 4H5 - 4H6
2855 "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6
2856 "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6
2858 "movq (%0, %1, 4), %%mm2 \n\t"
2859 "movq %%mm2, %%mm3 \n\t"
2860 "punpcklbw %%mm7, %%mm2 \n\t" // L7
2861 "punpckhbw %%mm7, %%mm3 \n\t" // H7
2863 "paddw %%mm2, %%mm2 \n\t" // 2L7
2864 "paddw %%mm3, %%mm3 \n\t" // 2H7
2865 "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6 - 2L7
2866 "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6 - 2H7
2868 "movq (%%"REG_c"), %%mm2 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2869 "movq 8(%%"REG_c"), %%mm3 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2872 "movq %%mm7, %%mm6 \n\t" // 0
2873 "psubw %%mm0, %%mm6 \n\t"
2874 "pmaxsw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
2875 "movq %%mm7, %%mm6 \n\t" // 0
2876 "psubw %%mm1, %%mm6 \n\t"
2877 "pmaxsw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
2878 "movq %%mm7, %%mm6 \n\t" // 0
2879 "psubw %%mm2, %%mm6 \n\t"
2880 "pmaxsw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
2881 "movq %%mm7, %%mm6 \n\t" // 0
2882 "psubw %%mm3, %%mm6 \n\t"
2883 "pmaxsw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
2885 "movq %%mm7, %%mm6 \n\t" // 0
2886 "pcmpgtw %%mm0, %%mm6 \n\t"
2887 "pxor %%mm6, %%mm0 \n\t"
2888 "psubw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
2889 "movq %%mm7, %%mm6 \n\t" // 0
2890 "pcmpgtw %%mm1, %%mm6 \n\t"
2891 "pxor %%mm6, %%mm1 \n\t"
2892 "psubw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
2893 "movq %%mm7, %%mm6 \n\t" // 0
2894 "pcmpgtw %%mm2, %%mm6 \n\t"
2895 "pxor %%mm6, %%mm2 \n\t"
2896 "psubw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
2897 "movq %%mm7, %%mm6 \n\t" // 0
2898 "pcmpgtw %%mm3, %%mm6 \n\t"
2899 "pxor %%mm6, %%mm3 \n\t"
2900 "psubw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
2904 "pminsw %%mm2, %%mm0 \n\t"
2905 "pminsw %%mm3, %%mm1 \n\t"
2907 "movq %%mm0, %%mm6 \n\t"
2908 "psubusw %%mm2, %%mm6 \n\t"
2909 "psubw %%mm6, %%mm0 \n\t"
2910 "movq %%mm1, %%mm6 \n\t"
2911 "psubusw %%mm3, %%mm6 \n\t"
2912 "psubw %%mm6, %%mm1 \n\t"
2915 "movd %2, %%mm2 \n\t" // QP
2916 "punpcklbw %%mm7, %%mm2 \n\t"
2918 "movq %%mm7, %%mm6 \n\t" // 0
2919 "pcmpgtw %%mm4, %%mm6 \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
2920 "pxor %%mm6, %%mm4 \n\t"
2921 "psubw %%mm6, %%mm4 \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
2922 "pcmpgtw %%mm5, %%mm7 \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
2923 "pxor %%mm7, %%mm5 \n\t"
2924 "psubw %%mm7, %%mm5 \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
2926 "psllw $3, %%mm2 \n\t" // 8QP
2927 "movq %%mm2, %%mm3 \n\t" // 8QP
2928 "pcmpgtw %%mm4, %%mm2 \n\t"
2929 "pcmpgtw %%mm5, %%mm3 \n\t"
2930 "pand %%mm2, %%mm4 \n\t"
2931 "pand %%mm3, %%mm5 \n\t"
2934 "psubusw %%mm0, %%mm4 \n\t" // hd
2935 "psubusw %%mm1, %%mm5 \n\t" // ld
2938 "movq "MANGLE(w05)", %%mm2 \n\t" // 5
2939 "pmullw %%mm2, %%mm4 \n\t"
2940 "pmullw %%mm2, %%mm5 \n\t"
2941 "movq "MANGLE(w20)", %%mm2 \n\t" // 32
2942 "paddw %%mm2, %%mm4 \n\t"
2943 "paddw %%mm2, %%mm5 \n\t"
2944 "psrlw $6, %%mm4 \n\t"
2945 "psrlw $6, %%mm5 \n\t"
2947 "movq 16(%%"REG_c"), %%mm0 \n\t" // L3 - L4
2948 "movq 24(%%"REG_c"), %%mm1 \n\t" // H3 - H4
2950 "pxor %%mm2, %%mm2 \n\t"
2951 "pxor %%mm3, %%mm3 \n\t"
2953 "pcmpgtw %%mm0, %%mm2 \n\t" // sign (L3-L4)
2954 "pcmpgtw %%mm1, %%mm3 \n\t" // sign (H3-H4)
2955 "pxor %%mm2, %%mm0 \n\t"
2956 "pxor %%mm3, %%mm1 \n\t"
2957 "psubw %%mm2, %%mm0 \n\t" // |L3-L4|
2958 "psubw %%mm3, %%mm1 \n\t" // |H3-H4|
2959 "psrlw $1, %%mm0 \n\t" // |L3 - L4|/2
2960 "psrlw $1, %%mm1 \n\t" // |H3 - H4|/2
2962 "pxor %%mm6, %%mm2 \n\t"
2963 "pxor %%mm7, %%mm3 \n\t"
2964 "pand %%mm2, %%mm4 \n\t"
2965 "pand %%mm3, %%mm5 \n\t"
2968 "pminsw %%mm0, %%mm4 \n\t"
2969 "pminsw %%mm1, %%mm5 \n\t"
2971 "movq %%mm4, %%mm2 \n\t"
2972 "psubusw %%mm0, %%mm2 \n\t"
2973 "psubw %%mm2, %%mm4 \n\t"
2974 "movq %%mm5, %%mm2 \n\t"
2975 "psubusw %%mm1, %%mm2 \n\t"
2976 "psubw %%mm2, %%mm5 \n\t"
2978 "pxor %%mm6, %%mm4 \n\t"
2979 "pxor %%mm7, %%mm5 \n\t"
2980 "psubw %%mm6, %%mm4 \n\t"
2981 "psubw %%mm7, %%mm5 \n\t"
2982 "packsswb %%mm5, %%mm4 \n\t"
2983 "movq %3, %%mm1 \n\t"
2984 "pandn %%mm4, %%mm1 \n\t"
2985 "movq (%0), %%mm0 \n\t"
2986 "paddb %%mm1, %%mm0 \n\t"
2987 "movq %%mm0, (%0) \n\t"
2988 "movq (%0, %1), %%mm0 \n\t"
2989 "psubb %%mm1, %%mm0 \n\t"
2990 "movq %%mm0, (%0, %1) \n\t"
2993 : "r" ((x86_reg)step), "m" (c->pQPb), "m"(eq_mask)
2994 : "%"REG_a, "%"REG_c
2998 STOP_TIMER("step16")
3005 static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
3006 const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);
3009 * Copy a block from src to dst and fixes the blacklevel.
3010 * levelFix == 0 -> do not touch the brighness & contrast
3012 #undef REAL_SCALED_CPY
3015 static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t src[], int srcStride,
3016 int levelFix, int64_t *packedOffsetAndScale)
3024 "movq (%%"REG_a"), %%mm2 \n\t" // packedYOffset
3025 "movq 8(%%"REG_a"), %%mm3 \n\t" // packedYScale
3026 "lea (%2,%4), %%"REG_a" \n\t"
3027 "lea (%3,%5), %%"REG_d" \n\t"
3028 "pxor %%mm4, %%mm4 \n\t"
3030 #define REAL_SCALED_CPY(src1, src2, dst1, dst2) \
3031 "movq " #src1 ", %%mm0 \n\t"\
3032 "movq " #src1 ", %%mm5 \n\t"\
3033 "movq " #src2 ", %%mm1 \n\t"\
3034 "movq " #src2 ", %%mm6 \n\t"\
3035 "punpcklbw %%mm0, %%mm0 \n\t"\
3036 "punpckhbw %%mm5, %%mm5 \n\t"\
3037 "punpcklbw %%mm1, %%mm1 \n\t"\
3038 "punpckhbw %%mm6, %%mm6 \n\t"\
3039 "pmulhuw %%mm3, %%mm0 \n\t"\
3040 "pmulhuw %%mm3, %%mm5 \n\t"\
3041 "pmulhuw %%mm3, %%mm1 \n\t"\
3042 "pmulhuw %%mm3, %%mm6 \n\t"\
3043 "psubw %%mm2, %%mm0 \n\t"\
3044 "psubw %%mm2, %%mm5 \n\t"\
3045 "psubw %%mm2, %%mm1 \n\t"\
3046 "psubw %%mm2, %%mm6 \n\t"\
3047 "packuswb %%mm5, %%mm0 \n\t"\
3048 "packuswb %%mm6, %%mm1 \n\t"\
3049 "movq %%mm0, " #dst1 " \n\t"\
3050 "movq %%mm1, " #dst2 " \n\t"\
3053 #define REAL_SCALED_CPY(src1, src2, dst1, dst2) \
3054 "movq " #src1 ", %%mm0 \n\t"\
3055 "movq " #src1 ", %%mm5 \n\t"\
3056 "punpcklbw %%mm4, %%mm0 \n\t"\
3057 "punpckhbw %%mm4, %%mm5 \n\t"\
3058 "psubw %%mm2, %%mm0 \n\t"\
3059 "psubw %%mm2, %%mm5 \n\t"\
3060 "movq " #src2 ", %%mm1 \n\t"\
3061 "psllw $6, %%mm0 \n\t"\
3062 "psllw $6, %%mm5 \n\t"\
3063 "pmulhw %%mm3, %%mm0 \n\t"\
3064 "movq " #src2 ", %%mm6 \n\t"\
3065 "pmulhw %%mm3, %%mm5 \n\t"\
3066 "punpcklbw %%mm4, %%mm1 \n\t"\
3067 "punpckhbw %%mm4, %%mm6 \n\t"\
3068 "psubw %%mm2, %%mm1 \n\t"\
3069 "psubw %%mm2, %%mm6 \n\t"\
3070 "psllw $6, %%mm1 \n\t"\
3071 "psllw $6, %%mm6 \n\t"\
3072 "pmulhw %%mm3, %%mm1 \n\t"\
3073 "pmulhw %%mm3, %%mm6 \n\t"\
3074 "packuswb %%mm5, %%mm0 \n\t"\
3075 "packuswb %%mm6, %%mm1 \n\t"\
3076 "movq %%mm0, " #dst1 " \n\t"\
3077 "movq %%mm1, " #dst2 " \n\t"\
3080 #define SCALED_CPY(src1, src2, dst1, dst2)\
3081 REAL_SCALED_CPY(src1, src2, dst1, dst2)
3083 SCALED_CPY((%2) , (%2, %4) , (%3) , (%3, %5))
3084 SCALED_CPY((%2, %4, 2), (%%REGa, %4, 2), (%3, %5, 2), (%%REGd, %5, 2))
3085 SCALED_CPY((%2, %4, 4), (%%REGa, %4, 4), (%3, %5, 4), (%%REGd, %5, 4))
3086 "lea (%%"REG_a",%4,4), %%"REG_a" \n\t"
3087 "lea (%%"REG_d",%5,4), %%"REG_d" \n\t"
3088 SCALED_CPY((%%REGa, %4), (%%REGa, %4, 2), (%%REGd, %5), (%%REGd, %5, 2))
3091 : "=&a" (packedOffsetAndScale)
3092 : "0" (packedOffsetAndScale),
3095 "r" ((x86_reg)srcStride),
3096 "r" ((x86_reg)dstStride)
3101 memcpy( &(dst[dstStride*i]),
3102 &(src[srcStride*i]), BLOCK_SIZE);
3107 "lea (%0,%2), %%"REG_a" \n\t"
3108 "lea (%1,%3), %%"REG_d" \n\t"
3110 #define REAL_SIMPLE_CPY(src1, src2, dst1, dst2) \
3111 "movq " #src1 ", %%mm0 \n\t"\
3112 "movq " #src2 ", %%mm1 \n\t"\
3113 "movq %%mm0, " #dst1 " \n\t"\
3114 "movq %%mm1, " #dst2 " \n\t"\
3116 #define SIMPLE_CPY(src1, src2, dst1, dst2)\
3117 REAL_SIMPLE_CPY(src1, src2, dst1, dst2)
3119 SIMPLE_CPY((%0) , (%0, %2) , (%1) , (%1, %3))
3120 SIMPLE_CPY((%0, %2, 2), (%%REGa, %2, 2), (%1, %3, 2), (%%REGd, %3, 2))
3121 SIMPLE_CPY((%0, %2, 4), (%%REGa, %2, 4), (%1, %3, 4), (%%REGd, %3, 4))
3122 "lea (%%"REG_a",%2,4), %%"REG_a" \n\t"
3123 "lea (%%"REG_d",%3,4), %%"REG_d" \n\t"
3124 SIMPLE_CPY((%%REGa, %2), (%%REGa, %2, 2), (%%REGd, %3), (%%REGd, %3, 2))
3128 "r" ((x86_reg)srcStride),
3129 "r" ((x86_reg)dstStride)
3130 : "%"REG_a, "%"REG_d
3134 memcpy( &(dst[dstStride*i]),
3135 &(src[srcStride*i]), BLOCK_SIZE);
3141 * Duplicate the given 8 src pixels ? times upward
3143 static inline void RENAME(duplicate)(uint8_t src[], int stride)
3147 "movq (%0), %%mm0 \n\t"
3149 "movq %%mm0, (%0) \n\t"
3150 "movq %%mm0, (%0, %1) \n\t"
3151 "movq %%mm0, (%0, %1, 2) \n\t"
3153 : "r" ((x86_reg)-stride)
3166 * Filter array of bytes (Y or U or V values)
3168 static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
3169 const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)
3171 DECLARE_ALIGNED(8, PPContext, c)= *c2; //copy to stack for faster access
3173 #ifdef COMPILE_TIME_MODE
3174 const int mode= COMPILE_TIME_MODE;
3176 const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;
3178 int black=0, white=255; // blackest black and whitest white in the picture
3179 int QPCorrecture= 256*256;
3186 const int qpHShift= isColor ? 4-c.hChromaSubSample : 4;
3187 const int qpVShift= isColor ? 4-c.vChromaSubSample : 4;
3190 uint64_t * const yHistogram= c.yHistogram;
3191 uint8_t * const tempSrc= srcStride > 0 ? c.tempSrc : c.tempSrc - 23*srcStride;
3192 uint8_t * const tempDst= dstStride > 0 ? c.tempDst : c.tempDst - 23*dstStride;
3193 //const int mbWidth= isColor ? (width+7)>>3 : (width+15)>>4;
3196 for(i=0; i<57; i++){
3197 int offset= ((i*c.ppMode.baseDcDiff)>>8) + 1;
3198 int threshold= offset*2 + 1;
3199 c.mmxDcOffset[i]= 0x7F - offset;
3200 c.mmxDcThreshold[i]= 0x7F - threshold;
3201 c.mmxDcOffset[i]*= 0x0101010101010101LL;
3202 c.mmxDcThreshold[i]*= 0x0101010101010101LL;
3206 if(mode & CUBIC_IPOL_DEINT_FILTER) copyAhead=16;
3207 else if( (mode & LINEAR_BLEND_DEINT_FILTER)
3208 || (mode & FFMPEG_DEINT_FILTER)
3209 || (mode & LOWPASS5_DEINT_FILTER)) copyAhead=14;
3210 else if( (mode & V_DEBLOCK)
3211 || (mode & LINEAR_IPOL_DEINT_FILTER)
3212 || (mode & MEDIAN_DEINT_FILTER)
3213 || (mode & V_A_DEBLOCK)) copyAhead=13;
3214 else if(mode & V_X1_FILTER) copyAhead=11;
3215 // else if(mode & V_RK1_FILTER) copyAhead=10;
3216 else if(mode & DERING) copyAhead=9;
3224 uint64_t maxClipped;
3229 // first frame is fscked so we ignore it
3230 if(c.frameNum == 1) yHistogram[0]= width*height/64*15/256;
3232 for(i=0; i<256; i++){
3233 sum+= yHistogram[i];
3236 /* We always get a completely black picture first. */
3237 maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);
3240 for(black=255; black>0; black--){
3241 if(clipped < maxClipped) break;
3242 clipped-= yHistogram[black];
3246 for(white=0; white<256; white++){
3247 if(clipped < maxClipped) break;
3248 clipped-= yHistogram[white];
3251 scale= (double)(c.ppMode.maxAllowedY - c.ppMode.minAllowedY) / (double)(white-black);
3254 c.packedYScale= (uint16_t)(scale*256.0 + 0.5);
3255 c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF;
3257 c.packedYScale= (uint16_t)(scale*1024.0 + 0.5);
3258 c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF;
3261 c.packedYOffset|= c.packedYOffset<<32;
3262 c.packedYOffset|= c.packedYOffset<<16;
3264 c.packedYScale|= c.packedYScale<<32;
3265 c.packedYScale|= c.packedYScale<<16;
3267 if(mode & LEVEL_FIX) QPCorrecture= (int)(scale*256*256 + 0.5);
3268 else QPCorrecture= 256*256;
3270 c.packedYScale= 0x0100010001000100LL;
3272 QPCorrecture= 256*256;
3275 /* copy & deinterlace first row of blocks */
3278 const uint8_t *srcBlock= &(src[y*srcStride]);
3279 uint8_t *dstBlock= tempDst + dstStride;
3281 // From this point on it is guaranteed that we can read and write 16 lines downward
3282 // finish 1 block before the next otherwise we might have a problem
3283 // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
3284 for(x=0; x<width; x+=BLOCK_SIZE){
3288 prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
3289 prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
3290 prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
3291 prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
3295 "mov %4, %%"REG_a" \n\t"
3296 "shr $2, %%"REG_a" \n\t"
3297 "and $6, %%"REG_a" \n\t"
3298 "add %5, %%"REG_a" \n\t"
3299 "mov %%"REG_a", %%"REG_d" \n\t"
3300 "imul %1, %%"REG_a" \n\t"
3301 "imul %3, %%"REG_d" \n\t"
3302 "prefetchnta 32(%%"REG_a", %0) \n\t"
3303 "prefetcht0 32(%%"REG_d", %2) \n\t"
3304 "add %1, %%"REG_a" \n\t"
3305 "add %3, %%"REG_d" \n\t"
3306 "prefetchnta 32(%%"REG_a", %0) \n\t"
3307 "prefetcht0 32(%%"REG_d", %2) \n\t"
3308 :: "r" (srcBlock), "r" ((x86_reg)srcStride), "r" (dstBlock), "r" ((x86_reg)dstStride),
3309 "g" ((x86_reg)x), "g" ((x86_reg)copyAhead)
3310 : "%"REG_a, "%"REG_d
3314 //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
3315 /* prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
3316 prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
3317 prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
3318 prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
3322 RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,
3323 srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
3325 RENAME(duplicate)(dstBlock + dstStride*8, dstStride);
3327 if(mode & LINEAR_IPOL_DEINT_FILTER)
3328 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
3329 else if(mode & LINEAR_BLEND_DEINT_FILTER)
3330 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
3331 else if(mode & MEDIAN_DEINT_FILTER)
3332 RENAME(deInterlaceMedian)(dstBlock, dstStride);
3333 else if(mode & CUBIC_IPOL_DEINT_FILTER)
3334 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
3335 else if(mode & FFMPEG_DEINT_FILTER)
3336 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
3337 else if(mode & LOWPASS5_DEINT_FILTER)
3338 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
3339 /* else if(mode & CUBIC_BLEND_DEINT_FILTER)
3340 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3345 if(width==FFABS(dstStride))
3346 linecpy(dst, tempDst + 9*dstStride, copyAhead, dstStride);
3349 for(i=0; i<copyAhead; i++){
3350 memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);
3355 for(y=0; y<height; y+=BLOCK_SIZE){
3356 //1% speedup if these are here instead of the inner loop
3357 const uint8_t *srcBlock= &(src[y*srcStride]);
3358 uint8_t *dstBlock= &(dst[y*dstStride]);
3360 uint8_t *tempBlock1= c.tempBlocks;
3361 uint8_t *tempBlock2= c.tempBlocks + 8;
3363 const int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];
3364 int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*FFABS(QPStride)];
3366 /* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards
3367 if not than use a temporary buffer */
3370 /* copy from line (copyAhead) to (copyAhead+7) of src, these will be copied with
3371 blockcopy to dst later */
3372 linecpy(tempSrc + srcStride*copyAhead, srcBlock + srcStride*copyAhead,
3373 FFMAX(height-y-copyAhead, 0), srcStride);
3375 /* duplicate last line of src to fill the void upto line (copyAhead+7) */
3376 for(i=FFMAX(height-y, 8); i<copyAhead+8; i++)
3377 memcpy(tempSrc + srcStride*i, src + srcStride*(height-1), FFABS(srcStride));
3379 /* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/
3380 linecpy(tempDst, dstBlock - dstStride, FFMIN(height-y+1, copyAhead+1), dstStride);
3382 /* duplicate last line of dst to fill the void upto line (copyAhead) */
3383 for(i=height-y+1; i<=copyAhead; i++)
3384 memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), FFABS(dstStride));
3386 dstBlock= tempDst + dstStride;
3390 // From this point on it is guaranteed that we can read and write 16 lines downward
3391 // finish 1 block before the next otherwise we might have a problem
3392 // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
3393 for(x=0; x<width; x+=BLOCK_SIZE){
3394 const int stride= dstStride;
3399 QP= QPptr[x>>qpHShift];
3400 c.nonBQP= nonBQPptr[x>>qpHShift];
3403 QP= (QP* QPCorrecture + 256*128)>>16;
3404 c.nonBQP= nonBQPptr[x>>4];
3405 c.nonBQP= (c.nonBQP* QPCorrecture + 256*128)>>16;
3406 yHistogram[ srcBlock[srcStride*12 + 4] ]++;
3411 "movd %1, %%mm7 \n\t"
3412 "packuswb %%mm7, %%mm7 \n\t" // 0, 0, 0, QP, 0, 0, 0, QP
3413 "packuswb %%mm7, %%mm7 \n\t" // 0,QP, 0, QP, 0,QP, 0, QP
3414 "packuswb %%mm7, %%mm7 \n\t" // QP,..., QP
3415 "movq %%mm7, %0 \n\t"
3424 prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
3425 prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
3426 prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
3427 prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
3431 "mov %4, %%"REG_a" \n\t"
3432 "shr $2, %%"REG_a" \n\t"
3433 "and $6, %%"REG_a" \n\t"
3434 "add %5, %%"REG_a" \n\t"
3435 "mov %%"REG_a", %%"REG_d" \n\t"
3436 "imul %1, %%"REG_a" \n\t"
3437 "imul %3, %%"REG_d" \n\t"
3438 "prefetchnta 32(%%"REG_a", %0) \n\t"
3439 "prefetcht0 32(%%"REG_d", %2) \n\t"
3440 "add %1, %%"REG_a" \n\t"
3441 "add %3, %%"REG_d" \n\t"
3442 "prefetchnta 32(%%"REG_a", %0) \n\t"
3443 "prefetcht0 32(%%"REG_d", %2) \n\t"
3444 :: "r" (srcBlock), "r" ((x86_reg)srcStride), "r" (dstBlock), "r" ((x86_reg)dstStride),
3445 "g" ((x86_reg)x), "g" ((x86_reg)copyAhead)
3446 : "%"REG_a, "%"REG_d
3450 //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
3451 /* prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
3452 prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
3453 prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
3454 prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
3458 RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,
3459 srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
3461 if(mode & LINEAR_IPOL_DEINT_FILTER)
3462 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
3463 else if(mode & LINEAR_BLEND_DEINT_FILTER)
3464 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
3465 else if(mode & MEDIAN_DEINT_FILTER)
3466 RENAME(deInterlaceMedian)(dstBlock, dstStride);
3467 else if(mode & CUBIC_IPOL_DEINT_FILTER)
3468 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
3469 else if(mode & FFMPEG_DEINT_FILTER)
3470 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
3471 else if(mode & LOWPASS5_DEINT_FILTER)
3472 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
3473 /* else if(mode & CUBIC_BLEND_DEINT_FILTER)
3474 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3477 /* only deblock if we have 2 blocks */
3479 if(mode & V_X1_FILTER)
3480 RENAME(vertX1Filter)(dstBlock, stride, &c);
3481 else if(mode & V_DEBLOCK){
3482 const int t= RENAME(vertClassify)(dstBlock, stride, &c);
3485 RENAME(doVertLowPass)(dstBlock, stride, &c);
3487 RENAME(doVertDefFilter)(dstBlock, stride, &c);
3488 }else if(mode & V_A_DEBLOCK){
3489 RENAME(do_a_deblock)(dstBlock, stride, 1, &c);
3494 RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);
3496 /* check if we have a previous block to deblock it with dstBlock */
3499 if(mode & H_X1_FILTER)
3500 RENAME(vertX1Filter)(tempBlock1, 16, &c);
3501 else if(mode & H_DEBLOCK){
3503 const int t= RENAME(vertClassify)(tempBlock1, 16, &c);
3504 //STOP_TIMER("dc & minmax")
3506 RENAME(doVertLowPass)(tempBlock1, 16, &c);
3508 RENAME(doVertDefFilter)(tempBlock1, 16, &c);
3509 }else if(mode & H_A_DEBLOCK){
3510 RENAME(do_a_deblock)(tempBlock1, 16, 1, &c);
3513 RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16);
3516 if(mode & H_X1_FILTER)
3517 horizX1Filter(dstBlock-4, stride, QP);
3518 else if(mode & H_DEBLOCK){
3520 DECLARE_ALIGNED(16, unsigned char, tempBlock)[272];
3521 transpose_16x8_char_toPackedAlign_altivec(tempBlock, dstBlock - (4 + 1), stride);
3523 const int t=vertClassify_altivec(tempBlock-48, 16, &c);
3525 doVertLowPass_altivec(tempBlock-48, 16, &c);
3526 transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
3529 doVertDefFilter_altivec(tempBlock-48, 16, &c);
3530 transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
3533 const int t= RENAME(horizClassify)(dstBlock-4, stride, &c);
3536 RENAME(doHorizLowPass)(dstBlock-4, stride, &c);
3538 RENAME(doHorizDefFilter)(dstBlock-4, stride, &c);
3540 }else if(mode & H_A_DEBLOCK){
3541 RENAME(do_a_deblock)(dstBlock-8, 1, stride, &c);
3545 //FIXME filter first line
3546 if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);
3549 if(mode & TEMP_NOISE_FILTER)
3551 RENAME(tempNoiseReducer)(dstBlock-8, stride,
3552 c.tempBlurred[isColor] + y*dstStride + x,
3553 c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3),
3554 c.ppMode.maxTmpNoise);
3562 tmpXchg= tempBlock1;
3563 tempBlock1= tempBlock2;
3564 tempBlock2 = tmpXchg;
3569 if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, &c);
3572 if((mode & TEMP_NOISE_FILTER)){
3573 RENAME(tempNoiseReducer)(dstBlock-8, dstStride,
3574 c.tempBlurred[isColor] + y*dstStride + x,
3575 c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3),
3576 c.ppMode.maxTmpNoise);
3579 /* did we use a tmp buffer for the last lines*/
3581 uint8_t *dstBlock= &(dst[y*dstStride]);
3582 if(width==FFABS(dstStride))
3583 linecpy(dstBlock, tempDst + dstStride, height-y, dstStride);
3586 for(i=0; i<height-y; i++){
3587 memcpy(dstBlock + i*dstStride, tempDst + (i+1)*dstStride, width);
3592 for(x=0; x<width; x+=32){
3594 i+= + dstBlock[x + 7*dstStride] + dstBlock[x + 8*dstStride]
3595 + dstBlock[x + 9*dstStride] + dstBlock[x +10*dstStride]
3596 + dstBlock[x +11*dstStride] + dstBlock[x +12*dstStride];
3597 + dstBlock[x +13*dstStride]
3598 + dstBlock[x +14*dstStride] + dstBlock[x +15*dstStride];
3602 __asm__ volatile("femms");
3604 __asm__ volatile("emms");
3607 #ifdef DEBUG_BRIGHTNESS
3611 for(i=0; i<256; i++)
3612 if(yHistogram[i] > max) max=yHistogram[i];
3614 for(i=1; i<256; i++){
3616 int start=yHistogram[i-1]/(max/256+1);
3617 int end=yHistogram[i]/(max/256+1);
3618 int inc= end > start ? 1 : -1;
3619 for(x=start; x!=end+inc; x+=inc)
3620 dst[ i*dstStride + x]+=128;
3623 for(i=0; i<100; i+=2){
3624 dst[ (white)*dstStride + i]+=128;
3625 dst[ (black)*dstStride + i]+=128;
3630 *c2= c; //copy local context back