projects / libav.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
FOURCC removed, using ff_get_fourcc instead (should be big-endian safe), workarounded...
[libav.git] / postproc / postprocess_template.c
CommitLineData
3057fa66 1/*
bdd677ac 2 Copyright (C) 2001-2002 Michael Niedermayer (michaelni@gmx.at)
3057fa66
A		 3
4 This program is free software; you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation; either version 2 of the License, or
7 (at your option) any later version.
8
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
13
14 You should have received a copy of the GNU General Public License
15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17*/
18
cc9b0679
MN		 19#undef PAVGB
20#undef PMINUB
21#undef PMAXUB
e939e1c3
A		 22
23#ifdef HAVE_MMX2
24#define PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
25#elif defined (HAVE_3DNOW)
26#define PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
27#endif
3057fa66 28
2e212618
MN		 29#ifdef HAVE_MMX2
30#define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
31#elif defined (HAVE_MMX)
32#define PMINUB(b,a,t) \
33 "movq " #a ", " #t " \n\t"\
34 "psubusb " #b ", " #t " \n\t"\
35 "psubb " #t ", " #a " \n\t"
36#endif
37
38#ifdef HAVE_MMX2
39#define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
40#elif defined (HAVE_MMX)
41#define PMAXUB(a,b) \
42 "psubusb " #a ", " #b " \n\t"\
43 "paddb " #a ", " #b " \n\t"
44#endif
45
46
3057fa66
A		 47//FIXME? |255-0| = 1 (shouldnt be a problem ...)
48/**
acced553 49 * Check if the middle 8x8 Block in the given 8x16 block is flat
3057fa66 50 */
cc9b0679 51static inline int RENAME(isVertDC)(uint8_t src[], int stride){
3057fa66 52 int numEq= 0;
e5c30e06 53#ifndef HAVE_MMX
d5a1a995 54 int y;
e5c30e06 55#endif
acced553 56 src+= stride*4; // src points to begin of the 8x8 Block
3057fa66 57#ifdef HAVE_MMX
37da00fc
MN		 58asm volatile(
59 "leal (%1, %2), %%eax \n\t"
60 "leal (%%eax, %2, 4), %%ebx \n\t"
61// 0 1 2 3 4 5 6 7 8 9
62// %1 eax eax+%2 eax+2%2 %1+4%2 ebx ebx+%2 ebx+2%2 %1+8%2 ebx+4%2
9b464428
FB		 63 "movq "MANGLE(mmxDCOffset)", %%mm7 \n\t" // mm7 = 0x7F
64 "movq "MANGLE(mmxDCThreshold)", %%mm6 \n\t" // mm6 = 0x7D
3057fa66 65 "movq (%1), %%mm0 \n\t"
37da00fc 66 "movq (%%eax), %%mm1 \n\t"
3057fa66
A		 67 "psubb %%mm1, %%mm0 \n\t" // mm0 = differnece
68 "paddb %%mm7, %%mm0 \n\t"
69 "pcmpgtb %%mm6, %%mm0 \n\t"
70
37da00fc 71 "movq (%%eax,%2), %%mm2 \n\t"
3057fa66
A		 72 "psubb %%mm2, %%mm1 \n\t"
73 "paddb %%mm7, %%mm1 \n\t"
74 "pcmpgtb %%mm6, %%mm1 \n\t"
75 "paddb %%mm1, %%mm0 \n\t"
76
37da00fc 77 "movq (%%eax, %2, 2), %%mm1 \n\t"
3057fa66
A		 78 "psubb %%mm1, %%mm2 \n\t"
79 "paddb %%mm7, %%mm2 \n\t"
80 "pcmpgtb %%mm6, %%mm2 \n\t"
81 "paddb %%mm2, %%mm0 \n\t"
82
37da00fc 83 "movq (%1, %2, 4), %%mm2 \n\t"
3057fa66
A		 84 "psubb %%mm2, %%mm1 \n\t"
85 "paddb %%mm7, %%mm1 \n\t"
86 "pcmpgtb %%mm6, %%mm1 \n\t"
87 "paddb %%mm1, %%mm0 \n\t"
88
37da00fc 89 "movq (%%ebx), %%mm1 \n\t"
3057fa66
A		 90 "psubb %%mm1, %%mm2 \n\t"
91 "paddb %%mm7, %%mm2 \n\t"
92 "pcmpgtb %%mm6, %%mm2 \n\t"
93 "paddb %%mm2, %%mm0 \n\t"
94
37da00fc 95 "movq (%%ebx, %2), %%mm2 \n\t"
3057fa66
A		 96 "psubb %%mm2, %%mm1 \n\t"
97 "paddb %%mm7, %%mm1 \n\t"
98 "pcmpgtb %%mm6, %%mm1 \n\t"
99 "paddb %%mm1, %%mm0 \n\t"
100
37da00fc 101 "movq (%%ebx, %2, 2), %%mm1 \n\t"
3057fa66
A		 102 "psubb %%mm1, %%mm2 \n\t"
103 "paddb %%mm7, %%mm2 \n\t"
104 "pcmpgtb %%mm6, %%mm2 \n\t"
105 "paddb %%mm2, %%mm0 \n\t"
106
107 " \n\t"
cd38e322
MN		 108#ifdef HAVE_MMX2
109 "pxor %%mm7, %%mm7 \n\t"
110 "psadbw %%mm7, %%mm0 \n\t"
111#else
3057fa66
A		 112 "movq %%mm0, %%mm1 \n\t"
113 "psrlw $8, %%mm0 \n\t"
114 "paddb %%mm1, %%mm0 \n\t"
115 "movq %%mm0, %%mm1 \n\t"
116 "psrlq $16, %%mm0 \n\t"
117 "paddb %%mm1, %%mm0 \n\t"
118 "movq %%mm0, %%mm1 \n\t"
119 "psrlq $32, %%mm0 \n\t"
120 "paddb %%mm1, %%mm0 \n\t"
cd38e322 121#endif
3057fa66
A		 122 "movd %%mm0, %0 \n\t"
123 : "=r" (numEq)
124 : "r" (src), "r" (stride)
47ceabc9 125 : "%eax", "%ebx"
3057fa66 126 );
cd38e322 127 numEq= (-numEq) &0xFF;
3057fa66
A		 128
129#else
d5a1a995 130 for(y=0; y<BLOCK_SIZE-1; y++)
3057fa66 131 {
43d52f76
MN		 132 if(((src[0] - src[0+stride] + dcOffset)&0xFFFF) < dcThreshold) numEq++;
133 if(((src[1] - src[1+stride] + dcOffset)&0xFFFF) < dcThreshold) numEq++;
134 if(((src[2] - src[2+stride] + dcOffset)&0xFFFF) < dcThreshold) numEq++;
135 if(((src[3] - src[3+stride] + dcOffset)&0xFFFF) < dcThreshold) numEq++;
136 if(((src[4] - src[4+stride] + dcOffset)&0xFFFF) < dcThreshold) numEq++;
137 if(((src[5] - src[5+stride] + dcOffset)&0xFFFF) < dcThreshold) numEq++;
138 if(((src[6] - src[6+stride] + dcOffset)&0xFFFF) < dcThreshold) numEq++;
139 if(((src[7] - src[7+stride] + dcOffset)&0xFFFF) < dcThreshold) numEq++;
3057fa66
A		 140 src+= stride;
141 }
142#endif
143/* if(abs(numEq - asmEq) > 0)
144 {
145 printf("\nasm:%d c:%d\n", asmEq, numEq);
146 for(int y=0; y<8; y++)
147 {
148 for(int x=0; x<8; x++)
149 {
150 printf("%d ", temp[x + y*stride]);
151 }
152 printf("\n");
153 }
154 }
155*/
d5a1a995
MN		 156// for(int i=0; i<numEq/8; i++) src[i]=255;
157 return (numEq > vFlatnessThreshold) ? 1 : 0;
3057fa66
A		 158}
159
cc9b0679 160static inline int RENAME(isVertMinMaxOk)(uint8_t src[], int stride, int QP)
3057fa66
A		 161{
162#ifdef HAVE_MMX
163 int isOk;
acced553 164 src+= stride*3;
3057fa66
A		 165 asm volatile(
166// "int $3 \n\t"
167 "movq (%1, %2), %%mm0 \n\t"
168 "movq (%1, %2, 8), %%mm1 \n\t"
169 "movq %%mm0, %%mm2 \n\t"
170 "psubusb %%mm1, %%mm0 \n\t"
171 "psubusb %%mm2, %%mm1 \n\t"
172 "por %%mm1, %%mm0 \n\t" // ABS Diff
173
9b464428 174 "movq "MANGLE(pQPb)", %%mm7 \n\t" // QP,..., QP
3057fa66
A		 175 "paddusb %%mm7, %%mm7 \n\t" // 2QP ... 2QP
176 "psubusb %%mm7, %%mm0 \n\t" // Diff <= 2QP -> 0
9b464428 177 "pcmpeqd "MANGLE(b00)", %%mm0 \n\t"
3057fa66 178 "psrlq $16, %%mm0 \n\t"
9b464428 179 "pcmpeqd "MANGLE(bFF)", %%mm0 \n\t"
3057fa66
A		 180// "movd %%mm0, (%1, %2, 4)\n\t"
181 "movd %%mm0, %0 \n\t"
182 : "=r" (isOk)
183 : "r" (src), "r" (stride)
184 );
ac0b0b2f 185 return isOk;
3057fa66
A		 186#else
187
d5a1a995
MN		 188 int isOk2= 1;
189 int x;
acced553 190 src+= stride*3;
d5a1a995 191 for(x=0; x<BLOCK_SIZE; x++)
3057fa66 192 {
d5a1a995 193 if(abs((int)src[x + stride] - (int)src[x + (stride<<3)]) > 2*QP) isOk2=0;
3057fa66
A		 194 }
195/* if(isOk && !isOk2 || !isOk && isOk2)
196 {
197 printf("\nasm:%d c:%d QP:%d\n", isOk, isOk2, QP);
198 for(int y=0; y<9; y++)
199 {
200 for(int x=0; x<8; x++)
201 {
202 printf("%d ", src[x + y*stride]);
203 }
204 printf("\n");
205 }
206 } */
207
208 return isOk2;
209#endif
210
211}
212
213/**
acced553 214 * Do a vertical low pass filter on the 8x16 block (only write to the 8x8 block in the middle)
a6be8111 215 * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16
3057fa66 216 */
cc9b0679 217static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, int QP)
3057fa66 218{
13e00528 219#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
acced553 220 src+= stride*3;
3057fa66 221 asm volatile( //"movv %0 %1 %2\n\t"
9b464428 222 "movq "MANGLE(pQPb)", %%mm0 \n\t" // QP,..., QP
3057fa66
A		 223
224 "movq (%0), %%mm6 \n\t"
225 "movq (%0, %1), %%mm5 \n\t"
226 "movq %%mm5, %%mm1 \n\t"
227 "movq %%mm6, %%mm2 \n\t"
228 "psubusb %%mm6, %%mm5 \n\t"
229 "psubusb %%mm1, %%mm2 \n\t"
230 "por %%mm5, %%mm2 \n\t" // ABS Diff of lines
231 "psubusb %%mm0, %%mm2 \n\t" // diff <= QP -> 0
9b464428 232 "pcmpeqb "MANGLE(b00)", %%mm2 \n\t" // diff <= QP -> FF
3057fa66
A		 233
234 "pand %%mm2, %%mm6 \n\t"
235 "pandn %%mm1, %%mm2 \n\t"
236 "por %%mm2, %%mm6 \n\t"// First Line to Filter
237
238 "movq (%0, %1, 8), %%mm5 \n\t"
239 "leal (%0, %1, 4), %%eax \n\t"
240 "leal (%0, %1, 8), %%ebx \n\t"
241 "subl %1, %%ebx \n\t"
242 "addl %1, %0 \n\t" // %0 points to line 1 not 0
243 "movq (%0, %1, 8), %%mm7 \n\t"
244 "movq %%mm5, %%mm1 \n\t"
245 "movq %%mm7, %%mm2 \n\t"
246 "psubusb %%mm7, %%mm5 \n\t"
247 "psubusb %%mm1, %%mm2 \n\t"
248 "por %%mm5, %%mm2 \n\t" // ABS Diff of lines
249 "psubusb %%mm0, %%mm2 \n\t" // diff <= QP -> 0
9b464428 250 "pcmpeqb "MANGLE(b00)", %%mm2 \n\t" // diff <= QP -> FF
3057fa66
A		 251
252 "pand %%mm2, %%mm7 \n\t"
253 "pandn %%mm1, %%mm2 \n\t"
254 "por %%mm2, %%mm7 \n\t" // First Line to Filter
255
256
257 // 1 2 3 4 5 6 7 8
258 // %0 %0+%1 %0+2%1 eax %0+4%1 eax+2%1 ebx eax+4%1
259 // 6 4 2 2 1 1
260 // 6 4 4 2
261 // 6 8 2
acced553 262
3057fa66
A		 263 "movq (%0, %1), %%mm0 \n\t" // 1
264 "movq %%mm0, %%mm1 \n\t" // 1
13e00528
A		 265 PAVGB(%%mm6, %%mm0) //1 1 /2
266 PAVGB(%%mm6, %%mm0) //3 1 /4
3057fa66
A		 267
268 "movq (%0, %1, 4), %%mm2 \n\t" // 1
269 "movq %%mm2, %%mm5 \n\t" // 1
13e00528
A		 270 PAVGB((%%eax), %%mm2) // 11 /2
271 PAVGB((%0, %1, 2), %%mm2) // 211 /4
3057fa66
A		 272 "movq %%mm2, %%mm3 \n\t" // 211 /4
273 "movq (%0), %%mm4 \n\t" // 1
13e00528
A		 274 PAVGB(%%mm4, %%mm3) // 4 211 /8
275 PAVGB(%%mm0, %%mm3) //642211 /16
3057fa66
A		 276 "movq %%mm3, (%0) \n\t" // X
277 // mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
278 "movq %%mm1, %%mm0 \n\t" // 1
13e00528 279 PAVGB(%%mm6, %%mm0) //1 1 /2
3057fa66 280 "movq %%mm4, %%mm3 \n\t" // 1
13e00528
A		 281 PAVGB((%0,%1,2), %%mm3) // 1 1 /2
282 PAVGB((%%eax,%1,2), %%mm5) // 11 /2
283 PAVGB((%%eax), %%mm5) // 211 /4
284 PAVGB(%%mm5, %%mm3) // 2 2211 /8
285 PAVGB(%%mm0, %%mm3) //4242211 /16
3057fa66
A		 286 "movq %%mm3, (%0,%1) \n\t" // X
287 // mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
13e00528 288 PAVGB(%%mm4, %%mm6) //11 /2
3057fa66 289 "movq (%%ebx), %%mm0 \n\t" // 1
13e00528 290 PAVGB((%%eax, %1, 2), %%mm0) // 11/2
3057fa66 291 "movq %%mm0, %%mm3 \n\t" // 11/2
13e00528
A		 292 PAVGB(%%mm1, %%mm0) // 2 11/4
293 PAVGB(%%mm6, %%mm0) //222 11/8
294 PAVGB(%%mm2, %%mm0) //22242211/16
3057fa66
A		 295 "movq (%0, %1, 2), %%mm2 \n\t" // 1
296 "movq %%mm0, (%0, %1, 2) \n\t" // X
297 // mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
298 "movq (%%eax, %1, 4), %%mm0 \n\t" // 1
13e00528
A		 299 PAVGB((%%ebx), %%mm0) // 11 /2
300 PAVGB(%%mm0, %%mm6) //11 11 /4
301 PAVGB(%%mm1, %%mm4) // 11 /2
302 PAVGB(%%mm2, %%mm1) // 11 /2
303 PAVGB(%%mm1, %%mm6) //1122 11 /8
304 PAVGB(%%mm5, %%mm6) //112242211 /16
3057fa66
A		 305 "movq (%%eax), %%mm5 \n\t" // 1
306 "movq %%mm6, (%%eax) \n\t" // X
307 // mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
308 "movq (%%eax, %1, 4), %%mm6 \n\t" // 1
13e00528
A		 309 PAVGB(%%mm7, %%mm6) // 11 /2
310 PAVGB(%%mm4, %%mm6) // 11 11 /4
311 PAVGB(%%mm3, %%mm6) // 11 2211 /8
312 PAVGB(%%mm5, %%mm2) // 11 /2
3057fa66 313 "movq (%0, %1, 4), %%mm4 \n\t" // 1
13e00528
A		 314 PAVGB(%%mm4, %%mm2) // 112 /4
315 PAVGB(%%mm2, %%mm6) // 112242211 /16
3057fa66
A		 316 "movq %%mm6, (%0, %1, 4) \n\t" // X
317 // mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
13e00528
A		 318 PAVGB(%%mm7, %%mm1) // 11 2 /4
319 PAVGB(%%mm4, %%mm5) // 11 /2
320 PAVGB(%%mm5, %%mm0) // 11 11 /4
3057fa66 321 "movq (%%eax, %1, 2), %%mm6 \n\t" // 1
13e00528
A		 322 PAVGB(%%mm6, %%mm1) // 11 4 2 /8
323 PAVGB(%%mm0, %%mm1) // 11224222 /16
3057fa66
A		 324 "movq %%mm1, (%%eax, %1, 2) \n\t" // X
325 // mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
13e00528 326 PAVGB((%%ebx), %%mm2) // 112 4 /8
3057fa66 327 "movq (%%eax, %1, 4), %%mm0 \n\t" // 1
13e00528
A		 328 PAVGB(%%mm0, %%mm6) // 1 1 /2
329 PAVGB(%%mm7, %%mm6) // 1 12 /4
330 PAVGB(%%mm2, %%mm6) // 1122424 /4
3057fa66
A		 331 "movq %%mm6, (%%ebx) \n\t" // X
332 // mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
13e00528
A		 333 PAVGB(%%mm7, %%mm5) // 11 2 /4
334 PAVGB(%%mm7, %%mm5) // 11 6 /8
3057fa66 335
13e00528
A		 336 PAVGB(%%mm3, %%mm0) // 112 /4
337 PAVGB(%%mm0, %%mm5) // 112246 /16
3057fa66 338 "movq %%mm5, (%%eax, %1, 4) \n\t" // X
8405b3fd 339 "subl %1, %0 \n\t"
3057fa66
A		 340
341 :
342 : "r" (src), "r" (stride)
343 : "%eax", "%ebx"
344 );
3057fa66
A		 345#else
346 const int l1= stride;
347 const int l2= stride + l1;
348 const int l3= stride + l2;
349 const int l4= stride + l3;
350 const int l5= stride + l4;
351 const int l6= stride + l5;
352 const int l7= stride + l6;
353 const int l8= stride + l7;
354 const int l9= stride + l8;
d5a1a995 355 int x;
acced553 356 src+= stride*3;
d5a1a995 357 for(x=0; x<BLOCK_SIZE; x++)
3057fa66
A		 358 {
359 const int first= ABS(src[0] - src[l1]) < QP ? src[0] : src[l1];
360 const int last= ABS(src[l8] - src[l9]) < QP ? src[l9] : src[l8];
361
362 int sums[9];
363 sums[0] = first + src[l1];
364 sums[1] = src[l1] + src[l2];
365 sums[2] = src[l2] + src[l3];
366 sums[3] = src[l3] + src[l4];
367 sums[4] = src[l4] + src[l5];
368 sums[5] = src[l5] + src[l6];
369 sums[6] = src[l6] + src[l7];
370 sums[7] = src[l7] + src[l8];
371 sums[8] = src[l8] + last;
372
373 src[l1]= ((sums[0]<<2) + ((first + sums[2])<<1) + sums[4] + 8)>>4;
e5c30e06
MN		 374 src[l2]= ((src[l2]<<2) + ((first + sums[0] + sums[3])<<1) + sums[5] + 8)>>4;
375 src[l3]= ((src[l3]<<2) + ((first + sums[1] + sums[4])<<1) + sums[6] + 8)>>4;
376 src[l4]= ((src[l4]<<2) + ((sums[2] + sums[5])<<1) + sums[0] + sums[7] + 8)>>4;
377 src[l5]= ((src[l5]<<2) + ((sums[3] + sums[6])<<1) + sums[1] + sums[8] + 8)>>4;
378 src[l6]= ((src[l6]<<2) + ((last + sums[7] + sums[4])<<1) + sums[2] + 8)>>4;
379 src[l7]= (((last + src[l7])<<2) + ((src[l8] + sums[5])<<1) + sums[3] + 8)>>4;
380 src[l8]= ((sums[8]<<2) + ((last + sums[6])<<1) + sums[4] + 8)>>4;
3057fa66
A		 381
382 src++;
383 }
384
385#endif
386}
387
13e00528
A		 388/**
389 * Experimental implementation of the filter (Algorithm 1) described in a paper from Ramkishor & Karandikar
390 * values are correctly clipped (MMX2)
391 * values are wraparound (C)
392 * conclusion: its fast, but introduces ugly horizontal patterns if there is a continious gradient
393 0 8 16 24
394 x = 8
395 x/2 = 4
396 x/8 = 1
397 1 12 12 23
398 */
cc9b0679 399static inline void RENAME(vertRK1Filter)(uint8_t *src, int stride, int QP)
13e00528 400{
d5a1a995 401#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
acced553 402 src+= stride*3;
13e00528
A		 403// FIXME rounding
404 asm volatile(
405 "pxor %%mm7, %%mm7 \n\t" // 0
9b464428 406 "movq "MANGLE(b80)", %%mm6 \n\t" // MIN_SIGNED_BYTE
13e00528
A		 407 "leal (%0, %1), %%eax \n\t"
408 "leal (%%eax, %1, 4), %%ebx \n\t"
409// 0 1 2 3 4 5 6 7 8 9
410// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
9b464428 411 "movq "MANGLE(pQPb)", %%mm0 \n\t" // QP,..., QP
13e00528 412 "movq %%mm0, %%mm1 \n\t" // QP,..., QP
9b464428 413 "paddusb "MANGLE(b02)", %%mm0 \n\t"
13e00528 414 "psrlw $2, %%mm0 \n\t"
9b464428 415 "pand "MANGLE(b3F)", %%mm0 \n\t" // QP/4,..., QP/4
13e00528
A		 416 "paddusb %%mm1, %%mm0 \n\t" // QP*1.25 ...
417 "movq (%0, %1, 4), %%mm2 \n\t" // line 4
418 "movq (%%ebx), %%mm3 \n\t" // line 5
419 "movq %%mm2, %%mm4 \n\t" // line 4
420 "pcmpeqb %%mm5, %%mm5 \n\t" // -1
421 "pxor %%mm2, %%mm5 \n\t" // -line 4 - 1
d5a1a995 422 PAVGB(%%mm3, %%mm5)
13e00528
A		 423 "paddb %%mm6, %%mm5 \n\t" // (l5-l4)/2
424 "psubusb %%mm3, %%mm4 \n\t"
425 "psubusb %%mm2, %%mm3 \n\t"
426 "por %%mm3, %%mm4 \n\t" // |l4 - l5|
427 "psubusb %%mm0, %%mm4 \n\t"
428 "pcmpeqb %%mm7, %%mm4 \n\t"
429 "pand %%mm4, %%mm5 \n\t" // d/2
430
431// "paddb %%mm6, %%mm2 \n\t" // line 4 + 0x80
432 "paddb %%mm5, %%mm2 \n\t"
433// "psubb %%mm6, %%mm2 \n\t"
434 "movq %%mm2, (%0,%1, 4) \n\t"
435
436 "movq (%%ebx), %%mm2 \n\t"
437// "paddb %%mm6, %%mm2 \n\t" // line 5 + 0x80
438 "psubb %%mm5, %%mm2 \n\t"
439// "psubb %%mm6, %%mm2 \n\t"
440 "movq %%mm2, (%%ebx) \n\t"
441
442 "paddb %%mm6, %%mm5 \n\t"
443 "psrlw $2, %%mm5 \n\t"
9b464428
FB		 444 "pand "MANGLE(b3F)", %%mm5 \n\t"
445 "psubb "MANGLE(b20)", %%mm5 \n\t" // (l5-l4)/8
13e00528
A		 446
447 "movq (%%eax, %1, 2), %%mm2 \n\t"
448 "paddb %%mm6, %%mm2 \n\t" // line 3 + 0x80
449 "paddsb %%mm5, %%mm2 \n\t"
450 "psubb %%mm6, %%mm2 \n\t"
451 "movq %%mm2, (%%eax, %1, 2) \n\t"
452
453 "movq (%%ebx, %1), %%mm2 \n\t"
454 "paddb %%mm6, %%mm2 \n\t" // line 6 + 0x80
455 "psubsb %%mm5, %%mm2 \n\t"
456 "psubb %%mm6, %%mm2 \n\t"
457 "movq %%mm2, (%%ebx, %1) \n\t"
458
459 :
460 : "r" (src), "r" (stride)
461 : "%eax", "%ebx"
462 );
463#else
464 const int l1= stride;
465 const int l2= stride + l1;
466 const int l3= stride + l2;
467 const int l4= stride + l3;
468 const int l5= stride + l4;
469 const int l6= stride + l5;
e5c30e06
MN		 470// const int l7= stride + l6;
471// const int l8= stride + l7;
472// const int l9= stride + l8;
d5a1a995 473 int x;
3407a972 474 const int QP15= QP + (QP>>2);
acced553 475 src+= stride*3;
d5a1a995 476 for(x=0; x<BLOCK_SIZE; x++)
13e00528 477 {
3407a972
MN		 478 const int v = (src[x+l5] - src[x+l4]);
479 if(ABS(v) < QP15)
13e00528 480 {
3407a972
MN		 481 src[x+l3] +=v>>3;
482 src[x+l4] +=v>>1;
483 src[x+l5] -=v>>1;
484 src[x+l6] -=v>>3;
13e00528 485
13e00528 486 }
13e00528
A		 487 }
488
489#endif
490}
491
492/**
493 * Experimental Filter 1
9f45d04d
MN		 494 * will not damage linear gradients
495 * Flat blocks should look like they where passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
d5a1a995
MN		 496 * can only smooth blocks at the expected locations (it cant smooth them if they did move)
497 * MMX2 version does correct clipping C version doesnt
13e00528 498 */
cc9b0679 499static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, int QP)
13e00528 500{
d5a1a995 501#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
acced553
MN		 502 src+= stride*3;
503
13e00528 504 asm volatile(
d5a1a995 505 "pxor %%mm7, %%mm7 \n\t" // 0
d5a1a995
MN		 506 "leal (%0, %1), %%eax \n\t"
507 "leal (%%eax, %1, 4), %%ebx \n\t"
508// 0 1 2 3 4 5 6 7 8 9
509// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
510 "movq (%%eax, %1, 2), %%mm0 \n\t" // line 3
511 "movq (%0, %1, 4), %%mm1 \n\t" // line 4
512 "movq %%mm1, %%mm2 \n\t" // line 4
513 "psubusb %%mm0, %%mm1 \n\t"
514 "psubusb %%mm2, %%mm0 \n\t"
515 "por %%mm1, %%mm0 \n\t" // |l2 - l3|
516 "movq (%%ebx), %%mm3 \n\t" // line 5
dc16b332 517 "movq (%%ebx, %1), %%mm4 \n\t" // line 6
d5a1a995
MN		 518 "movq %%mm3, %%mm5 \n\t" // line 5
519 "psubusb %%mm4, %%mm3 \n\t"
520 "psubusb %%mm5, %%mm4 \n\t"
521 "por %%mm4, %%mm3 \n\t" // |l5 - l6|
522 PAVGB(%%mm3, %%mm0) // (|l2 - l3| + |l5 - l6|)/2
523 "movq %%mm2, %%mm1 \n\t" // line 4
524 "psubusb %%mm5, %%mm2 \n\t"
525 "movq %%mm2, %%mm4 \n\t"
526 "pcmpeqb %%mm7, %%mm2 \n\t" // (l4 - l5) <= 0 ? -1 : 0
527 "psubusb %%mm1, %%mm5 \n\t"
528 "por %%mm5, %%mm4 \n\t" // |l4 - l5|
529 "psubusb %%mm0, %%mm4 \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
530 "movq %%mm4, %%mm3 \n\t" // d
dc16b332
MN		 531 "movq "MANGLE(pQPb)", %%mm0 \n\t"
532 "paddusb %%mm0, %%mm0 \n\t"
533 "psubusb %%mm0, %%mm4 \n\t"
d5a1a995 534 "pcmpeqb %%mm7, %%mm4 \n\t" // d <= QP ? -1 : 0
9b464428 535 "psubusb "MANGLE(b01)", %%mm3 \n\t"
d5a1a995
MN		 536 "pand %%mm4, %%mm3 \n\t" // d <= QP ? d : 0
537
538 PAVGB(%%mm7, %%mm3) // d/2
9f45d04d
MN		 539 "movq %%mm3, %%mm1 \n\t" // d/2
540 PAVGB(%%mm7, %%mm3) // d/4
541 PAVGB(%%mm1, %%mm3) // 3*d/8
d5a1a995
MN		 542
543 "movq (%0, %1, 4), %%mm0 \n\t" // line 4
544 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
545 "psubusb %%mm3, %%mm0 \n\t"
546 "pxor %%mm2, %%mm0 \n\t"
547 "movq %%mm0, (%0, %1, 4) \n\t" // line 4
548
549 "movq (%%ebx), %%mm0 \n\t" // line 5
550 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
551 "paddusb %%mm3, %%mm0 \n\t"
552 "pxor %%mm2, %%mm0 \n\t"
553 "movq %%mm0, (%%ebx) \n\t" // line 5
554
9f45d04d 555 PAVGB(%%mm7, %%mm1) // d/4
d5a1a995
MN		 556
557 "movq (%%eax, %1, 2), %%mm0 \n\t" // line 3
558 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
9f45d04d 559 "psubusb %%mm1, %%mm0 \n\t"
d5a1a995
MN		 560 "pxor %%mm2, %%mm0 \n\t"
561 "movq %%mm0, (%%eax, %1, 2) \n\t" // line 3
562
563 "movq (%%ebx, %1), %%mm0 \n\t" // line 6
564 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
9f45d04d 565 "paddusb %%mm1, %%mm0 \n\t"
d5a1a995
MN		 566 "pxor %%mm2, %%mm0 \n\t"
567 "movq %%mm0, (%%ebx, %1) \n\t" // line 6
568
9f45d04d 569 PAVGB(%%mm7, %%mm1) // d/8
d5a1a995
MN		 570
571 "movq (%%eax, %1), %%mm0 \n\t" // line 2
572 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
9f45d04d 573 "psubusb %%mm1, %%mm0 \n\t"
d5a1a995
MN		 574 "pxor %%mm2, %%mm0 \n\t"
575 "movq %%mm0, (%%eax, %1) \n\t" // line 2
576
577 "movq (%%ebx, %1, 2), %%mm0 \n\t" // line 7
578 "pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
9f45d04d 579 "paddusb %%mm1, %%mm0 \n\t"
d5a1a995
MN		 580 "pxor %%mm2, %%mm0 \n\t"
581 "movq %%mm0, (%%ebx, %1, 2) \n\t" // line 7
13e00528
A		 582
583 :
584 : "r" (src), "r" (stride)
585 : "%eax", "%ebx"
586 );
587#else
d5a1a995
MN		 588
589 const int l1= stride;
590 const int l2= stride + l1;
591 const int l3= stride + l2;
592 const int l4= stride + l3;
593 const int l5= stride + l4;
594 const int l6= stride + l5;
595 const int l7= stride + l6;
e5c30e06
MN		 596// const int l8= stride + l7;
597// const int l9= stride + l8;
d5a1a995 598 int x;
acced553
MN		 599
600 src+= stride*3;
d5a1a995
MN		 601 for(x=0; x<BLOCK_SIZE; x++)
602 {
603 int a= src[l3] - src[l4];
604 int b= src[l4] - src[l5];
9f45d04d 605 int c= src[l5] - src[l6];
d5a1a995 606
3407a972
MN		 607 int d= ABS(b) - ((ABS(a) + ABS(c))>>1);
608 d= MAX(d, 0);
d5a1a995 609
dc16b332 610 if(d < QP*2)
d5a1a995
MN		 611 {
612 int v = d * SIGN(-b);
613
3407a972
MN		 614 src[l2] +=v>>3;
615 src[l3] +=v>>2;
616 src[l4] +=(3*v)>>3;
617 src[l5] -=(3*v)>>3;
618 src[l6] -=v>>2;
619 src[l7] -=v>>3;
d5a1a995
MN		 620
621 }
622 src++;
623 }
624 /*
13e00528
A		 625 const int l1= stride;
626 const int l2= stride + l1;
627 const int l3= stride + l2;
628 const int l4= stride + l3;
629 const int l5= stride + l4;
630 const int l6= stride + l5;
631 const int l7= stride + l6;
632 const int l8= stride + l7;
633 const int l9= stride + l8;
634 for(int x=0; x<BLOCK_SIZE; x++)
635 {
636 int v2= src[l2];
637 int v3= src[l3];
638 int v4= src[l4];
639 int v5= src[l5];
640 int v6= src[l6];
641 int v7= src[l7];
642
643 if(ABS(v4-v5)<QP && ABS(v4-v5) - (ABS(v3-v4) + ABS(v5-v6))>0 )
644 {
645 src[l3] = (6*v2 + 4*v3 + 3*v4 + 2*v5 + v6 )/16;
646 src[l4] = (3*v2 + 3*v3 + 4*v4 + 3*v5 + 2*v6 + v7 )/16;
647 src[l5] = (1*v2 + 2*v3 + 3*v4 + 4*v5 + 3*v6 + 3*v7)/16;
648 src[l6] = ( 1*v3 + 2*v4 + 3*v5 + 4*v6 + 6*v7)/16;
649 }
650 src++;
651 }
d5a1a995 652*/
13e00528
A		 653#endif
654}
655
cc9b0679 656static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, int QP)
3057fa66 657{
7f16f6e6
MN		 658#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
659/*
660 uint8_t tmp[16];
661 const int l1= stride;
662 const int l2= stride + l1;
663 const int l3= stride + l2;
664 const int l4= (int)tmp - (int)src - stride*3;
665 const int l5= (int)tmp - (int)src - stride*3 + 8;
666 const int l6= stride*3 + l3;
667 const int l7= stride + l6;
668 const int l8= stride + l7;
669
670 memcpy(tmp, src+stride*7, 8);
671 memcpy(tmp+8, src+stride*8, 8);
672*/
673 src+= stride*4;
674 asm volatile(
675
676#if 0 //sligtly more accurate and slightly slower
677 "pxor %%mm7, %%mm7 \n\t" // 0
678 "leal (%0, %1), %%eax \n\t"
679 "leal (%%eax, %1, 4), %%ebx \n\t"
680// 0 1 2 3 4 5 6 7
681// %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 ebx+%1 ebx+2%1
682// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1
683
684
685 "movq (%0, %1, 2), %%mm0 \n\t" // l2
686 "movq (%0), %%mm1 \n\t" // l0
687 "movq %%mm0, %%mm2 \n\t" // l2
688 PAVGB(%%mm7, %%mm0) // ~l2/2
689 PAVGB(%%mm1, %%mm0) // ~(l2 + 2l0)/4
690 PAVGB(%%mm2, %%mm0) // ~(5l2 + 2l0)/8
691
692 "movq (%%eax), %%mm1 \n\t" // l1
693 "movq (%%eax, %1, 2), %%mm3 \n\t" // l3
694 "movq %%mm1, %%mm4 \n\t" // l1
695 PAVGB(%%mm7, %%mm1) // ~l1/2
696 PAVGB(%%mm3, %%mm1) // ~(l1 + 2l3)/4
697 PAVGB(%%mm4, %%mm1) // ~(5l1 + 2l3)/8
698
699 "movq %%mm0, %%mm4 \n\t" // ~(5l2 + 2l0)/8
700 "psubusb %%mm1, %%mm0 \n\t"
701 "psubusb %%mm4, %%mm1 \n\t"
702 "por %%mm0, %%mm1 \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8
703// mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0
704
705 "movq (%0, %1, 4), %%mm0 \n\t" // l4
706 "movq %%mm0, %%mm4 \n\t" // l4
707 PAVGB(%%mm7, %%mm0) // ~l4/2
708 PAVGB(%%mm2, %%mm0) // ~(l4 + 2l2)/4
709 PAVGB(%%mm4, %%mm0) // ~(5l4 + 2l2)/8
710
711 "movq (%%ebx), %%mm2 \n\t" // l5
712 "movq %%mm3, %%mm5 \n\t" // l3
713 PAVGB(%%mm7, %%mm3) // ~l3/2
714 PAVGB(%%mm2, %%mm3) // ~(l3 + 2l5)/4
715 PAVGB(%%mm5, %%mm3) // ~(5l3 + 2l5)/8
716
717 "movq %%mm0, %%mm6 \n\t" // ~(5l4 + 2l2)/8
718 "psubusb %%mm3, %%mm0 \n\t"
719 "psubusb %%mm6, %%mm3 \n\t"
720 "por %%mm0, %%mm3 \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8
721 "pcmpeqb %%mm7, %%mm0 \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)
722// mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0
723
724 "movq (%%ebx, %1), %%mm6 \n\t" // l6
725 "movq %%mm6, %%mm5 \n\t" // l6
726 PAVGB(%%mm7, %%mm6) // ~l6/2
727 PAVGB(%%mm4, %%mm6) // ~(l6 + 2l4)/4
728 PAVGB(%%mm5, %%mm6) // ~(5l6 + 2l4)/8
729
730 "movq (%%ebx, %1, 2), %%mm5 \n\t" // l7
731 "movq %%mm2, %%mm4 \n\t" // l5
732 PAVGB(%%mm7, %%mm2) // ~l5/2
733 PAVGB(%%mm5, %%mm2) // ~(l5 + 2l7)/4
734 PAVGB(%%mm4, %%mm2) // ~(5l5 + 2l7)/8
735
736 "movq %%mm6, %%mm4 \n\t" // ~(5l6 + 2l4)/8
737 "psubusb %%mm2, %%mm6 \n\t"
738 "psubusb %%mm4, %%mm2 \n\t"
739 "por %%mm6, %%mm2 \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8
740// mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0
741
742
743 PMINUB(%%mm2, %%mm1, %%mm4) // MIN(|lenergy|,|renergy|)/8
9b464428
FB		 744 "movq "MANGLE(pQPb)", %%mm4 \n\t" // QP //FIXME QP+1 ?
745 "paddusb "MANGLE(b01)", %%mm4 \n\t"
7f16f6e6
MN		 746 "pcmpgtb %%mm3, %%mm4 \n\t" // |menergy|/8 < QP
747 "psubusb %%mm1, %%mm3 \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8
748 "pand %%mm4, %%mm3 \n\t"
749
750 "movq %%mm3, %%mm1 \n\t"
9b464428 751// "psubusb "MANGLE(b01)", %%mm3 \n\t"
7f16f6e6
MN		 752 PAVGB(%%mm7, %%mm3)
753 PAVGB(%%mm7, %%mm3)
754 "paddusb %%mm1, %%mm3 \n\t"
9b464428 755// "paddusb "MANGLE(b01)", %%mm3 \n\t"
7f16f6e6
MN		 756
757 "movq (%%eax, %1, 2), %%mm6 \n\t" //l3
758 "movq (%0, %1, 4), %%mm5 \n\t" //l4
759 "movq (%0, %1, 4), %%mm4 \n\t" //l4
760 "psubusb %%mm6, %%mm5 \n\t"
761 "psubusb %%mm4, %%mm6 \n\t"
762 "por %%mm6, %%mm5 \n\t" // |l3-l4|
763 "pcmpeqb %%mm7, %%mm6 \n\t" // SIGN(l3-l4)
764 "pxor %%mm6, %%mm0 \n\t"
765 "pand %%mm0, %%mm3 \n\t"
766 PMINUB(%%mm5, %%mm3, %%mm0)
767
9b464428 768 "psubusb "MANGLE(b01)", %%mm3 \n\t"
7f16f6e6
MN		 769 PAVGB(%%mm7, %%mm3)
770
771 "movq (%%eax, %1, 2), %%mm0 \n\t"
772 "movq (%0, %1, 4), %%mm2 \n\t"
773 "pxor %%mm6, %%mm0 \n\t"
774 "pxor %%mm6, %%mm2 \n\t"
775 "psubb %%mm3, %%mm0 \n\t"
776 "paddb %%mm3, %%mm2 \n\t"
777 "pxor %%mm6, %%mm0 \n\t"
778 "pxor %%mm6, %%mm2 \n\t"
779 "movq %%mm0, (%%eax, %1, 2) \n\t"
780 "movq %%mm2, (%0, %1, 4) \n\t"
781#endif
782
783 "leal (%0, %1), %%eax \n\t"
784 "pcmpeqb %%mm6, %%mm6 \n\t" // -1
785// 0 1 2 3 4 5 6 7
786// %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 ebx+%1 ebx+2%1
787// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1
788
789
790 "movq (%%eax, %1, 2), %%mm1 \n\t" // l3
791 "movq (%0, %1, 4), %%mm0 \n\t" // l4
792 "pxor %%mm6, %%mm1 \n\t" // -l3-1
793 PAVGB(%%mm1, %%mm0) // -q+128 = (l4-l3+256)/2
794// mm1=-l3-1, mm0=128-q
795
796 "movq (%%eax, %1, 4), %%mm2 \n\t" // l5
797 "movq (%%eax, %1), %%mm3 \n\t" // l2
798 "pxor %%mm6, %%mm2 \n\t" // -l5-1
799 "movq %%mm2, %%mm5 \n\t" // -l5-1
9b464428 800 "movq "MANGLE(b80)", %%mm4 \n\t" // 128
7f16f6e6
MN		 801 "leal (%%eax, %1, 4), %%ebx \n\t"
802 PAVGB(%%mm3, %%mm2) // (l2-l5+256)/2
803 PAVGB(%%mm0, %%mm4) // ~(l4-l3)/4 + 128
804 PAVGB(%%mm2, %%mm4) // ~(l2-l5)/4 +(l4-l3)/8 + 128
805 PAVGB(%%mm0, %%mm4) // ~(l2-l5)/8 +5(l4-l3)/16 + 128
806// mm1=-l3-1, mm0=128-q, mm3=l2, mm4=menergy/16 + 128, mm5= -l5-1
807
808 "movq (%%eax), %%mm2 \n\t" // l1
809 "pxor %%mm6, %%mm2 \n\t" // -l1-1
810 PAVGB(%%mm3, %%mm2) // (l2-l1+256)/2
811 PAVGB((%0), %%mm1) // (l0-l3+256)/2
9b464428 812 "movq "MANGLE(b80)", %%mm3 \n\t" // 128
7f16f6e6
MN		 813 PAVGB(%%mm2, %%mm3) // ~(l2-l1)/4 + 128
814 PAVGB(%%mm1, %%mm3) // ~(l0-l3)/4 +(l2-l1)/8 + 128
815 PAVGB(%%mm2, %%mm3) // ~(l0-l3)/8 +5(l2-l1)/16 + 128
816// mm0=128-q, mm3=lenergy/16 + 128, mm4= menergy/16 + 128, mm5= -l5-1
817
818 PAVGB((%%ebx, %1), %%mm5) // (l6-l5+256)/2
819 "movq (%%ebx, %1, 2), %%mm1 \n\t" // l7
820 "pxor %%mm6, %%mm1 \n\t" // -l7-1
821 PAVGB((%0, %1, 4), %%mm1) // (l4-l7+256)/2
9b464428 822 "movq "MANGLE(b80)", %%mm2 \n\t" // 128
7f16f6e6
MN		 823 PAVGB(%%mm5, %%mm2) // ~(l6-l5)/4 + 128
824 PAVGB(%%mm1, %%mm2) // ~(l4-l7)/4 +(l6-l5)/8 + 128
825 PAVGB(%%mm5, %%mm2) // ~(l4-l7)/8 +5(l6-l5)/16 + 128
826// mm0=128-q, mm2=renergy/16 + 128, mm3=lenergy/16 + 128, mm4= menergy/16 + 128
827
9b464428
FB		 828 "movq "MANGLE(b00)", %%mm1 \n\t" // 0
829 "movq "MANGLE(b00)", %%mm5 \n\t" // 0
7f16f6e6
MN		 830 "psubb %%mm2, %%mm1 \n\t" // 128 - renergy/16
831 "psubb %%mm3, %%mm5 \n\t" // 128 - lenergy/16
832 PMAXUB(%%mm1, %%mm2) // 128 + |renergy/16|
833 PMAXUB(%%mm5, %%mm3) // 128 + |lenergy/16|
834 PMINUB(%%mm2, %%mm3, %%mm1) // 128 + MIN(|lenergy|,|renergy|)/16
835
836// mm0=128-q, mm3=128 + MIN(|lenergy|,|renergy|)/16, mm4= menergy/16 + 128
837
9b464428
FB		 838 "movq "MANGLE(b00)", %%mm7 \n\t" // 0
839 "movq "MANGLE(pQPb)", %%mm2 \n\t" // QP
7f16f6e6
MN		 840 PAVGB(%%mm6, %%mm2) // 128 + QP/2
841 "psubb %%mm6, %%mm2 \n\t"
842
843 "movq %%mm4, %%mm1 \n\t"
844 "pcmpgtb %%mm7, %%mm1 \n\t" // SIGN(menergy)
845 "pxor %%mm1, %%mm4 \n\t"
846 "psubb %%mm1, %%mm4 \n\t" // 128 + |menergy|/16
847 "pcmpgtb %%mm4, %%mm2 \n\t" // |menergy|/16 < QP/2
848 "psubusb %%mm3, %%mm4 \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16
849// mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16
850
851 "movq %%mm4, %%mm3 \n\t" // d
9b464428 852 "psubusb "MANGLE(b01)", %%mm4 \n\t"
7f16f6e6
MN		 853 PAVGB(%%mm7, %%mm4) // d/32
854 PAVGB(%%mm7, %%mm4) // (d + 32)/64
855 "paddb %%mm3, %%mm4 \n\t" // 5d/64
856 "pand %%mm2, %%mm4 \n\t"
857
9b464428 858 "movq "MANGLE(b80)", %%mm5 \n\t" // 128
7f16f6e6
MN		 859 "psubb %%mm0, %%mm5 \n\t" // q
860 "paddsb %%mm6, %%mm5 \n\t" // fix bad rounding
861 "pcmpgtb %%mm5, %%mm7 \n\t" // SIGN(q)
862 "pxor %%mm7, %%mm5 \n\t"
863
864 PMINUB(%%mm5, %%mm4, %%mm3) // MIN(|q|, 5d/64)
865 "pxor %%mm1, %%mm7 \n\t" // SIGN(d*q)
866
867 "pand %%mm7, %%mm4 \n\t"
868 "movq (%%eax, %1, 2), %%mm0 \n\t"
869 "movq (%0, %1, 4), %%mm2 \n\t"
870 "pxor %%mm1, %%mm0 \n\t"
871 "pxor %%mm1, %%mm2 \n\t"
872 "paddb %%mm4, %%mm0 \n\t"
873 "psubb %%mm4, %%mm2 \n\t"
874 "pxor %%mm1, %%mm0 \n\t"
875 "pxor %%mm1, %%mm2 \n\t"
876 "movq %%mm0, (%%eax, %1, 2) \n\t"
877 "movq %%mm2, (%0, %1, 4) \n\t"
878
879 :
880 : "r" (src), "r" (stride)
881 : "%eax", "%ebx"
882 );
883
884/*
885 {
886 int x;
887 src-= stride;
888 for(x=0; x<BLOCK_SIZE; x++)
889 {
890 const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
891 if(ABS(middleEnergy)< 8*QP)
892 {
893 const int q=(src[l4] - src[l5])/2;
894 const int leftEnergy= 5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
895 const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
896
897 int d= ABS(middleEnergy) - MIN( ABS(leftEnergy), ABS(rightEnergy) );
898 d= MAX(d, 0);
899
900 d= (5*d + 32) >> 6;
901 d*= SIGN(-middleEnergy);
902
903 if(q>0)
904 {
905 d= d<0 ? 0 : d;
906 d= d>q ? q : d;
907 }
908 else
909 {
910 d= d>0 ? 0 : d;
911 d= d<q ? q : d;
912 }
913
914 src[l4]-= d;
915 src[l5]+= d;
916 }
917 src++;
918 }
919src-=8;
920 for(x=0; x<8; x++)
921 {
922 int y;
923 for(y=4; y<6; y++)
924 {
925 int d= src[x+y*stride] - tmp[x+(y-4)*8];
926 int ad= ABS(d);
927 static int max=0;
928 static int sum=0;
929 static int num=0;
930 static int bias=0;
931
932 if(max<ad) max=ad;
933 sum+= ad>3 ? 1 : 0;
934 if(ad>3)
935 {
936 src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
937 }
938 if(y==4) bias+=d;
939 num++;
940 if(num%1000000 == 0)
941 {
942 printf(" %d %d %d %d\n", num, sum, max, bias);
943 }
944 }
945 }
946}
947*/
948#elif defined (HAVE_MMX)
acced553 949 src+= stride*4;
7f16f6e6 950
3057fa66
A		 951 asm volatile(
952 "pxor %%mm7, %%mm7 \n\t"
953 "leal (%0, %1), %%eax \n\t"
954 "leal (%%eax, %1, 4), %%ebx \n\t"
955// 0 1 2 3 4 5 6 7
956// %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 ebx+%1 ebx+2%1
957// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1
958
959 "movq (%0), %%mm0 \n\t"
960 "movq %%mm0, %%mm1 \n\t"
961 "punpcklbw %%mm7, %%mm0 \n\t" // low part of line 0
962 "punpckhbw %%mm7, %%mm1 \n\t" // high part of line 0
963
964 "movq (%%eax), %%mm2 \n\t"
965 "movq %%mm2, %%mm3 \n\t"
966 "punpcklbw %%mm7, %%mm2 \n\t" // low part of line 1
967 "punpckhbw %%mm7, %%mm3 \n\t" // high part of line 1
968
969 "movq (%%eax, %1), %%mm4 \n\t"
970 "movq %%mm4, %%mm5 \n\t"
971 "punpcklbw %%mm7, %%mm4 \n\t" // low part of line 2
972 "punpckhbw %%mm7, %%mm5 \n\t" // high part of line 2
973
974 "paddw %%mm0, %%mm0 \n\t" // 2L0
975 "paddw %%mm1, %%mm1 \n\t" // 2H0
976 "psubw %%mm4, %%mm2 \n\t" // L1 - L2
977 "psubw %%mm5, %%mm3 \n\t" // H1 - H2
978 "psubw %%mm2, %%mm0 \n\t" // 2L0 - L1 + L2
979 "psubw %%mm3, %%mm1 \n\t" // 2H0 - H1 + H2
980
981 "psllw $2, %%mm2 \n\t" // 4L1 - 4L2
982 "psllw $2, %%mm3 \n\t" // 4H1 - 4H2
983 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2
984 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2
985
986 "movq (%%eax, %1, 2), %%mm2 \n\t"
987 "movq %%mm2, %%mm3 \n\t"
988 "punpcklbw %%mm7, %%mm2 \n\t" // L3
989 "punpckhbw %%mm7, %%mm3 \n\t" // H3
990
991 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - L3
992 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - H3
993 "psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
994 "psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
9b464428
FB		 995 "movq %%mm0, "MANGLE(temp0)" \n\t" // 2L0 - 5L1 + 5L2 - 2L3
996 "movq %%mm1, "MANGLE(temp1)" \n\t" // 2H0 - 5H1 + 5H2 - 2H3
3057fa66
A		 997
998 "movq (%0, %1, 4), %%mm0 \n\t"
999 "movq %%mm0, %%mm1 \n\t"
1000 "punpcklbw %%mm7, %%mm0 \n\t" // L4
1001 "punpckhbw %%mm7, %%mm1 \n\t" // H4
1002
1003 "psubw %%mm0, %%mm2 \n\t" // L3 - L4
1004 "psubw %%mm1, %%mm3 \n\t" // H3 - H4
9b464428
FB		 1005 "movq %%mm2, "MANGLE(temp2)" \n\t" // L3 - L4
1006 "movq %%mm3, "MANGLE(temp3)" \n\t" // H3 - H4
3057fa66
A		 1007 "paddw %%mm4, %%mm4 \n\t" // 2L2
1008 "paddw %%mm5, %%mm5 \n\t" // 2H2
1009 "psubw %%mm2, %%mm4 \n\t" // 2L2 - L3 + L4
1010 "psubw %%mm3, %%mm5 \n\t" // 2H2 - H3 + H4
1011
1012 "psllw $2, %%mm2 \n\t" // 4L3 - 4L4
1013 "psllw $2, %%mm3 \n\t" // 4H3 - 4H4
1014 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4
1015 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4
1016//50 opcodes so far
1017 "movq (%%ebx), %%mm2 \n\t"
1018 "movq %%mm2, %%mm3 \n\t"
1019 "punpcklbw %%mm7, %%mm2 \n\t" // L5
1020 "punpckhbw %%mm7, %%mm3 \n\t" // H5
1021 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - L5
1022 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - H5
1023 "psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - 2L5
1024 "psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - 2H5
1025
1026 "movq (%%ebx, %1), %%mm6 \n\t"
1027 "punpcklbw %%mm7, %%mm6 \n\t" // L6
1028 "psubw %%mm6, %%mm2 \n\t" // L5 - L6
1029 "movq (%%ebx, %1), %%mm6 \n\t"
1030 "punpckhbw %%mm7, %%mm6 \n\t" // H6
1031 "psubw %%mm6, %%mm3 \n\t" // H5 - H6
1032
1033 "paddw %%mm0, %%mm0 \n\t" // 2L4
1034 "paddw %%mm1, %%mm1 \n\t" // 2H4
1035 "psubw %%mm2, %%mm0 \n\t" // 2L4 - L5 + L6
1036 "psubw %%mm3, %%mm1 \n\t" // 2H4 - H5 + H6
1037
1038 "psllw $2, %%mm2 \n\t" // 4L5 - 4L6
1039 "psllw $2, %%mm3 \n\t" // 4H5 - 4H6
1040 "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6
1041 "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6
1042
1043 "movq (%%ebx, %1, 2), %%mm2 \n\t"
1044 "movq %%mm2, %%mm3 \n\t"
1045 "punpcklbw %%mm7, %%mm2 \n\t" // L7
1046 "punpckhbw %%mm7, %%mm3 \n\t" // H7
1047
1048 "paddw %%mm2, %%mm2 \n\t" // 2L7
1049 "paddw %%mm3, %%mm3 \n\t" // 2H7
1050 "psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6 - 2L7
1051 "psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6 - 2H7
1052
9b464428
FB		 1053 "movq "MANGLE(temp0)", %%mm2 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
1054 "movq "MANGLE(temp1)", %%mm3 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
8405b3fd
MN		 1055
1056#ifdef HAVE_MMX2
1057 "movq %%mm7, %%mm6 \n\t" // 0
1058 "psubw %%mm0, %%mm6 \n\t"
1059 "pmaxsw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
1060 "movq %%mm7, %%mm6 \n\t" // 0
1061 "psubw %%mm1, %%mm6 \n\t"
1062 "pmaxsw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
1063 "movq %%mm7, %%mm6 \n\t" // 0
1064 "psubw %%mm2, %%mm6 \n\t"
1065 "pmaxsw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
1066 "movq %%mm7, %%mm6 \n\t" // 0
1067 "psubw %%mm3, %%mm6 \n\t"
1068 "pmaxsw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
1069#else
3057fa66
A		 1070 "movq %%mm7, %%mm6 \n\t" // 0
1071 "pcmpgtw %%mm0, %%mm6 \n\t"
1072 "pxor %%mm6, %%mm0 \n\t"
1073 "psubw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
1074 "movq %%mm7, %%mm6 \n\t" // 0
1075 "pcmpgtw %%mm1, %%mm6 \n\t"
1076 "pxor %%mm6, %%mm1 \n\t"
1077 "psubw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
3057fa66
A		 1078 "movq %%mm7, %%mm6 \n\t" // 0
1079 "pcmpgtw %%mm2, %%mm6 \n\t"
1080 "pxor %%mm6, %%mm2 \n\t"
1081 "psubw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
1082 "movq %%mm7, %%mm6 \n\t" // 0
1083 "pcmpgtw %%mm3, %%mm6 \n\t"
1084 "pxor %%mm6, %%mm3 \n\t"
1085 "psubw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
8405b3fd 1086#endif
3057fa66
A		 1087
1088#ifdef HAVE_MMX2
1089 "pminsw %%mm2, %%mm0 \n\t"
1090 "pminsw %%mm3, %%mm1 \n\t"
1091#else
1092 "movq %%mm0, %%mm6 \n\t"
1093 "psubusw %%mm2, %%mm6 \n\t"
1094 "psubw %%mm6, %%mm0 \n\t"
1095 "movq %%mm1, %%mm6 \n\t"
1096 "psubusw %%mm3, %%mm6 \n\t"
1097 "psubw %%mm6, %%mm1 \n\t"
1098#endif
1099
1100 "movq %%mm7, %%mm6 \n\t" // 0
1101 "pcmpgtw %%mm4, %%mm6 \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
1102 "pxor %%mm6, %%mm4 \n\t"
1103 "psubw %%mm6, %%mm4 \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
1104 "pcmpgtw %%mm5, %%mm7 \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
1105 "pxor %%mm7, %%mm5 \n\t"
1106 "psubw %%mm7, %%mm5 \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
1107// 100 opcodes
1108 "movd %2, %%mm2 \n\t" // QP
3057fa66
A		 1109 "punpcklwd %%mm2, %%mm2 \n\t"
1110 "punpcklwd %%mm2, %%mm2 \n\t"
1111 "psllw $3, %%mm2 \n\t" // 8QP
1112 "movq %%mm2, %%mm3 \n\t" // 8QP
1113 "pcmpgtw %%mm4, %%mm2 \n\t"
1114 "pcmpgtw %%mm5, %%mm3 \n\t"
1115 "pand %%mm2, %%mm4 \n\t"
1116 "pand %%mm3, %%mm5 \n\t"
1117
1118
1119 "psubusw %%mm0, %%mm4 \n\t" // hd
1120 "psubusw %%mm1, %%mm5 \n\t" // ld
1121
1122
bf1595c4 1123 "movq "MANGLE(w05)", %%mm2 \n\t" // 5
3057fa66
A		 1124 "pmullw %%mm2, %%mm4 \n\t"
1125 "pmullw %%mm2, %%mm5 \n\t"
bf1595c4 1126 "movq "MANGLE(w20)", %%mm2 \n\t" // 32
3057fa66
A		 1127 "paddw %%mm2, %%mm4 \n\t"
1128 "paddw %%mm2, %%mm5 \n\t"
1129 "psrlw $6, %%mm4 \n\t"
1130 "psrlw $6, %%mm5 \n\t"
1131
1132/*
1133 "movq w06, %%mm2 \n\t" // 6
1134 "paddw %%mm2, %%mm4 \n\t"
1135 "paddw %%mm2, %%mm5 \n\t"
1136 "movq w1400, %%mm2 \n\t" // 1400h = 5120 = 5/64*2^16
1137//FIXME if *5/64 is supposed to be /13 then we should use 5041 instead of 5120
1138 "pmulhw %%mm2, %%mm4 \n\t" // hd/13
1139 "pmulhw %%mm2, %%mm5 \n\t" // ld/13
1140*/
1141
9b464428
FB		 1142 "movq "MANGLE(temp2)", %%mm0 \n\t" // L3 - L4
1143 "movq "MANGLE(temp3)", %%mm1 \n\t" // H3 - H4
3057fa66
A		 1144
1145 "pxor %%mm2, %%mm2 \n\t"
1146 "pxor %%mm3, %%mm3 \n\t"
1147
3057fa66
A		 1148 "pcmpgtw %%mm0, %%mm2 \n\t" // sign (L3-L4)
1149 "pcmpgtw %%mm1, %%mm3 \n\t" // sign (H3-H4)
1150 "pxor %%mm2, %%mm0 \n\t"
1151 "pxor %%mm3, %%mm1 \n\t"
1152 "psubw %%mm2, %%mm0 \n\t" // |L3-L4|
1153 "psubw %%mm3, %%mm1 \n\t" // |H3-H4|
e5c30e06
MN		 1154 "psrlw $1, %%mm0 \n\t" // |L3 - L4|/2
1155 "psrlw $1, %%mm1 \n\t" // |H3 - H4|/2
3057fa66
A		 1156
1157 "pxor %%mm6, %%mm2 \n\t"
1158 "pxor %%mm7, %%mm3 \n\t"
1159 "pand %%mm2, %%mm4 \n\t"
1160 "pand %%mm3, %%mm5 \n\t"
1161
1162#ifdef HAVE_MMX2
1163 "pminsw %%mm0, %%mm4 \n\t"
1164 "pminsw %%mm1, %%mm5 \n\t"
1165#else
1166 "movq %%mm4, %%mm2 \n\t"
1167 "psubusw %%mm0, %%mm2 \n\t"
1168 "psubw %%mm2, %%mm4 \n\t"
1169 "movq %%mm5, %%mm2 \n\t"
1170 "psubusw %%mm1, %%mm2 \n\t"
1171 "psubw %%mm2, %%mm5 \n\t"
1172#endif
1173 "pxor %%mm6, %%mm4 \n\t"
1174 "pxor %%mm7, %%mm5 \n\t"
1175 "psubw %%mm6, %%mm4 \n\t"
1176 "psubw %%mm7, %%mm5 \n\t"
1177 "packsswb %%mm5, %%mm4 \n\t"
1178 "movq (%%eax, %1, 2), %%mm0 \n\t"
1179 "paddb %%mm4, %%mm0 \n\t"
1180 "movq %%mm0, (%%eax, %1, 2) \n\t"
1181 "movq (%0, %1, 4), %%mm0 \n\t"
1182 "psubb %%mm4, %%mm0 \n\t"
3057fa66
A		 1183 "movq %%mm0, (%0, %1, 4) \n\t"
1184
1185 :
1186 : "r" (src), "r" (stride), "r" (QP)
1187 : "%eax", "%ebx"
1188 );
1189#else
1190 const int l1= stride;
1191 const int l2= stride + l1;
1192 const int l3= stride + l2;
1193 const int l4= stride + l3;
1194 const int l5= stride + l4;
1195 const int l6= stride + l5;
1196 const int l7= stride + l6;
1197 const int l8= stride + l7;
1198// const int l9= stride + l8;
d5a1a995 1199 int x;
acced553 1200 src+= stride*3;
d5a1a995 1201 for(x=0; x<BLOCK_SIZE; x++)
3057fa66
A		 1202 {
1203 const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
1204 if(ABS(middleEnergy) < 8*QP)
1205 {
1206 const int q=(src[l4] - src[l5])/2;
1207 const int leftEnergy= 5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
1208 const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
1209
1210 int d= ABS(middleEnergy) - MIN( ABS(leftEnergy), ABS(rightEnergy) );
1211 d= MAX(d, 0);
1212
1213 d= (5*d + 32) >> 6;
1214 d*= SIGN(-middleEnergy);
1215
1216 if(q>0)
1217 {
1218 d= d<0 ? 0 : d;
1219 d= d>q ? q : d;
1220 }
1221 else
1222 {
1223 d= d>0 ? 0 : d;
1224 d= d<q ? q : d;
1225 }
1226
1227 src[l4]-= d;
1228 src[l5]+= d;
1229 }
1230 src++;
1231 }
1232#endif
1233}
1234
cc9b0679 1235static inline void RENAME(dering)(uint8_t src[], int stride, int QP)
3057fa66 1236{
e0f8ffae 1237#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
3057fa66 1238 asm volatile(
9b464428 1239 "movq "MANGLE(pQPb)", %%mm0 \n\t"
70c5ae87 1240 "paddusb %%mm0, %%mm0 \n\t"
9b464428 1241 "movq %%mm0, "MANGLE(pQPb2)" \n\t"
70c5ae87 1242
3057fa66
A		 1243 "leal (%0, %1), %%eax \n\t"
1244 "leal (%%eax, %1, 4), %%ebx \n\t"
1245// 0 1 2 3 4 5 6 7 8 9
1246// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
1247
cd38e322
MN		 1248 "pcmpeqb %%mm7, %%mm7 \n\t"
1249 "pxor %%mm6, %%mm6 \n\t"
cc9b0679 1250#undef FIND_MIN_MAX
e0f8ffae 1251#ifdef HAVE_MMX2
3057fa66 1252#define FIND_MIN_MAX(addr)\
70c5ae87 1253 "movq " #addr ", %%mm0 \n\t"\
cd38e322
MN		 1254 "pminub %%mm0, %%mm7 \n\t"\
1255 "pmaxub %%mm0, %%mm6 \n\t"
e0f8ffae
MN		 1256#else
1257#define FIND_MIN_MAX(addr)\
1258 "movq " #addr ", %%mm0 \n\t"\
cd38e322
MN		 1259 "movq %%mm7, %%mm1 \n\t"\
1260 "psubusb %%mm0, %%mm6 \n\t"\
1261 "paddb %%mm0, %%mm6 \n\t"\
e0f8ffae 1262 "psubusb %%mm0, %%mm1 \n\t"\
cd38e322 1263 "psubb %%mm1, %%mm7 \n\t"
e0f8ffae 1264#endif
3057fa66 1265
70c5ae87
MN		 1266FIND_MIN_MAX((%%eax))
1267FIND_MIN_MAX((%%eax, %1))
1268FIND_MIN_MAX((%%eax, %1, 2))
1269FIND_MIN_MAX((%0, %1, 4))
1270FIND_MIN_MAX((%%ebx))
1271FIND_MIN_MAX((%%ebx, %1))
1272FIND_MIN_MAX((%%ebx, %1, 2))
1273FIND_MIN_MAX((%0, %1, 8))
3057fa66 1274
3057fa66 1275 "movq %%mm7, %%mm4 \n\t"
e5c30e06 1276 "psrlq $8, %%mm7 \n\t"
e5c30e06 1277#ifdef HAVE_MMX2
cd38e322 1278 "pminub %%mm4, %%mm7 \n\t" // min of pixels
e5c30e06 1279 "pshufw $0xF9, %%mm7, %%mm4 \n\t"
cd38e322 1280 "pminub %%mm4, %%mm7 \n\t" // min of pixels
e5c30e06 1281 "pshufw $0xFE, %%mm7, %%mm4 \n\t"
cd38e322 1282 "pminub %%mm4, %%mm7 \n\t"
e5c30e06 1283#else
cd38e322
MN		 1284 "movq %%mm7, %%mm1 \n\t"
1285 "psubusb %%mm4, %%mm1 \n\t"
1286 "psubb %%mm1, %%mm7 \n\t"
3057fa66
A		 1287 "movq %%mm7, %%mm4 \n\t"
1288 "psrlq $16, %%mm7 \n\t"
cd38e322
MN		 1289 "movq %%mm7, %%mm1 \n\t"
1290 "psubusb %%mm4, %%mm1 \n\t"
1291 "psubb %%mm1, %%mm7 \n\t"
3057fa66 1292 "movq %%mm7, %%mm4 \n\t"
e5c30e06 1293 "psrlq $32, %%mm7 \n\t"
cd38e322
MN		 1294 "movq %%mm7, %%mm1 \n\t"
1295 "psubusb %%mm4, %%mm1 \n\t"
1296 "psubb %%mm1, %%mm7 \n\t"
e5c30e06 1297#endif
cd38e322
MN		 1298
1299
1300 "movq %%mm6, %%mm4 \n\t"
1301 "psrlq $8, %%mm6 \n\t"
1302#ifdef HAVE_MMX2
1303 "pmaxub %%mm4, %%mm6 \n\t" // max of pixels
1304 "pshufw $0xF9, %%mm6, %%mm4 \n\t"
1305 "pmaxub %%mm4, %%mm6 \n\t"
1306 "pshufw $0xFE, %%mm6, %%mm4 \n\t"
1307 "pmaxub %%mm4, %%mm6 \n\t"
1308#else
1309 "psubusb %%mm4, %%mm6 \n\t"
1310 "paddb %%mm4, %%mm6 \n\t"
1311 "movq %%mm6, %%mm4 \n\t"
1312 "psrlq $16, %%mm6 \n\t"
1313 "psubusb %%mm4, %%mm6 \n\t"
1314 "paddb %%mm4, %%mm6 \n\t"
1315 "movq %%mm6, %%mm4 \n\t"
1316 "psrlq $32, %%mm6 \n\t"
1317 "psubusb %%mm4, %%mm6 \n\t"
1318 "paddb %%mm4, %%mm6 \n\t"
1319#endif
1320 "movq %%mm6, %%mm0 \n\t" // max
1321 "psubb %%mm7, %%mm6 \n\t" // max - min
1322 "movd %%mm6, %%ecx \n\t"
9b464428 1323 "cmpb "MANGLE(deringThreshold)", %%cl \n\t"
cd38e322
MN		 1324 " jb 1f \n\t"
1325 PAVGB(%%mm0, %%mm7) // a=(max + min)/2
e5c30e06
MN		 1326 "punpcklbw %%mm7, %%mm7 \n\t"
1327 "punpcklbw %%mm7, %%mm7 \n\t"
1328 "punpcklbw %%mm7, %%mm7 \n\t"
9b464428 1329 "movq %%mm7, "MANGLE(temp0)" \n\t"
70c5ae87
MN		 1330
1331 "movq (%0), %%mm0 \n\t" // L10
1332 "movq %%mm0, %%mm1 \n\t" // L10
1333 "movq %%mm0, %%mm2 \n\t" // L10
1334 "psllq $8, %%mm1 \n\t"
1335 "psrlq $8, %%mm2 \n\t"
1336 "movd -4(%0), %%mm3 \n\t"
1337 "movd 8(%0), %%mm4 \n\t"
1338 "psrlq $24, %%mm3 \n\t"
1339 "psllq $56, %%mm4 \n\t"
1340 "por %%mm3, %%mm1 \n\t" // L00
1341 "por %%mm4, %%mm2 \n\t" // L20
1342 "movq %%mm1, %%mm3 \n\t" // L00
1343 PAVGB(%%mm2, %%mm1) // (L20 + L00)/2
1344 PAVGB(%%mm0, %%mm1) // (L20 + L00 + 2L10)/4
1345 "psubusb %%mm7, %%mm0 \n\t"
1346 "psubusb %%mm7, %%mm2 \n\t"
1347 "psubusb %%mm7, %%mm3 \n\t"
9b464428
FB		 1348 "pcmpeqb "MANGLE(b00)", %%mm0 \n\t" // L10 > a ? 0 : -1
1349 "pcmpeqb "MANGLE(b00)", %%mm2 \n\t" // L20 > a ? 0 : -1
1350 "pcmpeqb "MANGLE(b00)", %%mm3 \n\t" // L00 > a ? 0 : -1
70c5ae87
MN		 1351 "paddb %%mm2, %%mm0 \n\t"
1352 "paddb %%mm3, %%mm0 \n\t"
1353
1354 "movq (%%eax), %%mm2 \n\t" // L11
1355 "movq %%mm2, %%mm3 \n\t" // L11
1356 "movq %%mm2, %%mm4 \n\t" // L11
1357 "psllq $8, %%mm3 \n\t"
1358 "psrlq $8, %%mm4 \n\t"
1359 "movd -4(%%eax), %%mm5 \n\t"
1360 "movd 8(%%eax), %%mm6 \n\t"
1361 "psrlq $24, %%mm5 \n\t"
1362 "psllq $56, %%mm6 \n\t"
1363 "por %%mm5, %%mm3 \n\t" // L01
1364 "por %%mm6, %%mm4 \n\t" // L21
1365 "movq %%mm3, %%mm5 \n\t" // L01
1366 PAVGB(%%mm4, %%mm3) // (L21 + L01)/2
1367 PAVGB(%%mm2, %%mm3) // (L21 + L01 + 2L11)/4
1368 "psubusb %%mm7, %%mm2 \n\t"
1369 "psubusb %%mm7, %%mm4 \n\t"
1370 "psubusb %%mm7, %%mm5 \n\t"
9b464428
FB		 1371 "pcmpeqb "MANGLE(b00)", %%mm2 \n\t" // L11 > a ? 0 : -1
1372 "pcmpeqb "MANGLE(b00)", %%mm4 \n\t" // L21 > a ? 0 : -1
1373 "pcmpeqb "MANGLE(b00)", %%mm5 \n\t" // L01 > a ? 0 : -1
70c5ae87
MN		 1374 "paddb %%mm4, %%mm2 \n\t"
1375 "paddb %%mm5, %%mm2 \n\t"
1376// 0, 2, 3, 1
1377#define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
1378 "movq " #src ", " #sx " \n\t" /* src[0] */\
1379 "movq " #sx ", " #lx " \n\t" /* src[0] */\
1380 "movq " #sx ", " #t0 " \n\t" /* src[0] */\
1381 "psllq $8, " #lx " \n\t"\
1382 "psrlq $8, " #t0 " \n\t"\
1383 "movd -4" #src ", " #t1 " \n\t"\
1384 "psrlq $24, " #t1 " \n\t"\
1385 "por " #t1 ", " #lx " \n\t" /* src[-1] */\
1386 "movd 8" #src ", " #t1 " \n\t"\
1387 "psllq $56, " #t1 " \n\t"\
1388 "por " #t1 ", " #t0 " \n\t" /* src[+1] */\
1389 "movq " #lx ", " #t1 " \n\t" /* src[-1] */\
1390 PAVGB(t0, lx) /* (src[-1] + src[+1])/2 */\
1391 PAVGB(sx, lx) /* (src[-1] + 2src[0] + src[+1])/4 */\
9927c7ee 1392 PAVGB(lx, pplx) \
9b464428
FB		 1393 "movq " #lx ", "MANGLE(temp1)" \n\t"\
1394 "movq "MANGLE(temp0)", " #lx " \n\t"\
8405b3fd
MN		 1395 "psubusb " #lx ", " #t1 " \n\t"\
1396 "psubusb " #lx ", " #t0 " \n\t"\
1397 "psubusb " #lx ", " #sx " \n\t"\
9b464428 1398 "movq "MANGLE(b00)", " #lx " \n\t"\
8405b3fd
MN		 1399 "pcmpeqb " #lx ", " #t1 " \n\t" /* src[-1] > a ? 0 : -1*/\
1400 "pcmpeqb " #lx ", " #t0 " \n\t" /* src[+1] > a ? 0 : -1*/\
1401 "pcmpeqb " #lx ", " #sx " \n\t" /* src[0] > a ? 0 : -1*/\
70c5ae87
MN		 1402 "paddb " #t1 ", " #t0 " \n\t"\
1403 "paddb " #t0 ", " #sx " \n\t"\
1404\
70c5ae87
MN		 1405 PAVGB(plx, pplx) /* filtered */\
1406 "movq " #dst ", " #t0 " \n\t" /* dst */\
2e212618 1407 "movq " #t0 ", " #t1 " \n\t" /* dst */\
9b464428
FB		 1408 "psubusb "MANGLE(pQPb2)", " #t0 " \n\t"\
1409 "paddusb "MANGLE(pQPb2)", " #t1 " \n\t"\
2e212618
MN		 1410 PMAXUB(t0, pplx)\
1411 PMINUB(t1, pplx, t0)\
70c5ae87
MN		 1412 "paddb " #sx ", " #ppsx " \n\t"\
1413 "paddb " #psx ", " #ppsx " \n\t"\
9b464428
FB		 1414 "#paddb "MANGLE(b02)", " #ppsx " \n\t"\
1415 "pand "MANGLE(b08)", " #ppsx " \n\t"\
8405b3fd 1416 "pcmpeqb " #lx ", " #ppsx " \n\t"\
2e212618 1417 "pand " #ppsx ", " #pplx " \n\t"\
70c5ae87 1418 "pandn " #dst ", " #ppsx " \n\t"\
8405b3fd 1419 "por " #pplx ", " #ppsx " \n\t"\
9927c7ee 1420 "movq " #ppsx ", " #dst " \n\t"\
9b464428 1421 "movq "MANGLE(temp1)", " #lx " \n\t"
2e212618 1422
70c5ae87
MN		 1423/*
14240000000
14251111111
e5c30e06 1426
70c5ae87
MN		 14271111110
14281111101
14291111100
14301111011
14311111010
14321111001
e5c30e06 1433
70c5ae87
MN		 14341111000
14351110111
e5c30e06 1436
70c5ae87
MN		 1437*/
1438//DERING_CORE(dst,src ,ppsx ,psx ,sx ,pplx ,plx ,lx ,t0 ,t1)
1439DERING_CORE((%%eax),(%%eax, %1) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1440DERING_CORE((%%eax, %1),(%%eax, %1, 2) ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1441DERING_CORE((%%eax, %1, 2),(%0, %1, 4) ,%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
1442DERING_CORE((%0, %1, 4),(%%ebx) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1443DERING_CORE((%%ebx),(%%ebx, %1) ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1444DERING_CORE((%%ebx, %1), (%%ebx, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
1445DERING_CORE((%%ebx, %1, 2),(%0, %1, 8) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1446DERING_CORE((%0, %1, 8),(%%ebx, %1, 4) ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
3057fa66 1447
cd38e322 1448 "1: \n\t"
3057fa66 1449 : : "r" (src), "r" (stride), "r" (QP)
cd38e322 1450 : "%eax", "%ebx", "%ecx"
3057fa66
A		 1451 );
1452#else
2e212618
MN		 1453 int y;
1454 int min=255;
1455 int max=0;
1456 int avg;
1457 uint8_t *p;
1458 int s[10];
1459
1460 for(y=1; y<9; y++)
1461 {
1462 int x;
1463 p= src + stride*y;
1464 for(x=1; x<9; x++)
1465 {
1466 p++;
1467 if(*p > max) max= *p;
1468 if(*p < min) min= *p;
1469 }
1470 }
1471 avg= (min + max + 1)/2;
1472
cd38e322
MN		 1473 if(max - min <deringThreshold) return;
1474
2e212618
MN		 1475 for(y=0; y<10; y++)
1476 {
1477 int x;
1478 int t = 0;
1479 p= src + stride*y;
1480 for(x=0; x<10; x++)
1481 {
1482 if(*p > avg) t |= (1<<x);
1483 p++;
1484 }
1485 t |= (~t)<<16;
1486 t &= (t<<1) & (t>>1);
1487 s[y] = t;
1488 }
1489
1490 for(y=1; y<9; y++)
1491 {
1492 int x;
1493 int t = s[y-1] & s[y] & s[y+1];
1494 t|= t>>16;
1495
1496 p= src + stride*y;
1497 for(x=1; x<9; x++)
1498 {
1499 p++;
1500 if(t & (1<<x))
1501 {
1502 int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))
1503 +2*(*(p -1)) + 4*(*p ) + 2*(*(p +1))
1504 +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));
1505 f= (f + 8)>>4;
1506
cd38e322
MN		 1507#ifdef DEBUG_DERING_THRESHOLD
1508 asm volatile("emms\n\t":);
1509 {
1510 static long long numPixels=0;
1511 if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;
1512// if((max-min)<20 || (max-min)*QP<200)
1513// if((max-min)*QP < 500)
1514// if(max-min<QP/2)
1515 if(max-min < 20)
1516 {
1517 static int numSkiped=0;
1518 static int errorSum=0;
1519 static int worstQP=0;
1520 static int worstRange=0;
1521 static int worstDiff=0;
1522 int diff= (f - *p);
1523 int absDiff= ABS(diff);
1524 int error= diff*diff;
1525
1526 if(x==1 || x==8 || y==1 || y==8) continue;
1527
1528 numSkiped++;
1529 if(absDiff > worstDiff)
1530 {
1531 worstDiff= absDiff;
1532 worstQP= QP;
1533 worstRange= max-min;
1534 }
1535 errorSum+= error;
1536
1537 if(1024LL*1024LL*1024LL % numSkiped == 0)
1538 {
1539 printf( "sum:%1.3f, skip:%d, wQP:%d, "
1540 "wRange:%d, wDiff:%d, relSkip:%1.3f\n",
1541 (float)errorSum/numSkiped, numSkiped, worstQP, worstRange,
1542 worstDiff, (float)numSkiped/numPixels);
1543 }
1544 }
1545 }
1546#endif
2e212618
MN		 1547 if (*p + 2*QP < f) *p= *p + 2*QP;
1548 else if(*p - 2*QP > f) *p= *p - 2*QP;
1549 else *p=f;
1550 }
1551 }
1552 }
cd38e322
MN		 1553#ifdef DEBUG_DERING_THRESHOLD
1554 if(max-min < 20)
1555 {
1556 for(y=1; y<9; y++)
1557 {
1558 int x;
1559 int t = 0;
1560 p= src + stride*y;
1561 for(x=1; x<9; x++)
1562 {
1563 p++;
1564 *p = MIN(*p + 20, 255);
1565 }
1566 }
1567// src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
1568 }
1569#endif
3057fa66
A		 1570#endif
1571}
1572
3b58b885
MN		 1573/**
1574 * Deinterlaces the given block
7fb36f6c
MN		 1575 * will be called for every 8x8 block and can read & write from line 4-15
1576 * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1577 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
3b58b885 1578 */
cc9b0679 1579static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)
3b58b885
MN		 1580{
1581#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
7fb36f6c 1582 src+= 4*stride;
3b58b885
MN		 1583 asm volatile(
1584 "leal (%0, %1), %%eax \n\t"
1585 "leal (%%eax, %1, 4), %%ebx \n\t"
1586// 0 1 2 3 4 5 6 7 8 9
1587// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
1588
1589 "movq (%0), %%mm0 \n\t"
1590 "movq (%%eax, %1), %%mm1 \n\t"
acced553 1591 PAVGB(%%mm1, %%mm0)
3b58b885
MN		 1592 "movq %%mm0, (%%eax) \n\t"
1593 "movq (%0, %1, 4), %%mm0 \n\t"
acced553 1594 PAVGB(%%mm0, %%mm1)
3b58b885
MN		 1595 "movq %%mm1, (%%eax, %1, 2) \n\t"
1596 "movq (%%ebx, %1), %%mm1 \n\t"
acced553 1597 PAVGB(%%mm1, %%mm0)
3b58b885
MN		 1598 "movq %%mm0, (%%ebx) \n\t"
1599 "movq (%0, %1, 8), %%mm0 \n\t"
acced553 1600 PAVGB(%%mm0, %%mm1)
3b58b885
MN		 1601 "movq %%mm1, (%%ebx, %1, 2) \n\t"
1602
1603 : : "r" (src), "r" (stride)
1604 : "%eax", "%ebx"
1605 );
1606#else
1607 int x;
7fb36f6c 1608 src+= 4*stride;
3b58b885
MN		 1609 for(x=0; x<8; x++)
1610 {
1611 src[stride] = (src[0] + src[stride*2])>>1;
1612 src[stride*3] = (src[stride*2] + src[stride*4])>>1;
1613 src[stride*5] = (src[stride*4] + src[stride*6])>>1;
1614 src[stride*7] = (src[stride*6] + src[stride*8])>>1;
1615 src++;
1616 }
1617#endif
1618}
1619
1620/**
1621 * Deinterlaces the given block
7fb36f6c
MN		 1622 * will be called for every 8x8 block and can read & write from line 4-15
1623 * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1624 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
1625 * this filter will read lines 3-15 and write 7-13
acced553 1626 * no cliping in C version
3b58b885 1627 */
cc9b0679 1628static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)
3b58b885
MN		 1629{
1630#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
7fb36f6c 1631 src+= stride*3;
3b58b885
MN		 1632 asm volatile(
1633 "leal (%0, %1), %%eax \n\t"
1634 "leal (%%eax, %1, 4), %%ebx \n\t"
acced553
MN		 1635 "leal (%%ebx, %1, 4), %%ecx \n\t"
1636 "addl %1, %%ecx \n\t"
1637 "pxor %%mm7, %%mm7 \n\t"
1638// 0 1 2 3 4 5 6 7 8 9 10
1639// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1 ecx
3b58b885 1640
acced553
MN		 1641#define DEINT_CUBIC(a,b,c,d,e)\
1642 "movq " #a ", %%mm0 \n\t"\
1643 "movq " #b ", %%mm1 \n\t"\
1644 "movq " #d ", %%mm2 \n\t"\
1645 "movq " #e ", %%mm3 \n\t"\
1646 PAVGB(%%mm2, %%mm1) /* (b+d) /2 */\
1647 PAVGB(%%mm3, %%mm0) /* a(a+e) /2 */\
1648 "movq %%mm0, %%mm2 \n\t"\
1649 "punpcklbw %%mm7, %%mm0 \n\t"\
1650 "punpckhbw %%mm7, %%mm2 \n\t"\
1651 "movq %%mm1, %%mm3 \n\t"\
1652 "punpcklbw %%mm7, %%mm1 \n\t"\
1653 "punpckhbw %%mm7, %%mm3 \n\t"\
1654 "psubw %%mm1, %%mm0 \n\t" /* L(a+e - (b+d))/2 */\
1655 "psubw %%mm3, %%mm2 \n\t" /* H(a+e - (b+d))/2 */\
1656 "psraw $3, %%mm0 \n\t" /* L(a+e - (b+d))/16 */\
1657 "psraw $3, %%mm2 \n\t" /* H(a+e - (b+d))/16 */\
1658 "psubw %%mm0, %%mm1 \n\t" /* L(9b + 9d - a - e)/16 */\
1659 "psubw %%mm2, %%mm3 \n\t" /* H(9b + 9d - a - e)/16 */\
1660 "packuswb %%mm3, %%mm1 \n\t"\
1661 "movq %%mm1, " #c " \n\t"
1662
1663DEINT_CUBIC((%0), (%%eax, %1), (%%eax, %1, 2), (%0, %1, 4), (%%ebx, %1))
1664DEINT_CUBIC((%%eax, %1), (%0, %1, 4), (%%ebx), (%%ebx, %1), (%0, %1, 8))
1665DEINT_CUBIC((%0, %1, 4), (%%ebx, %1), (%%ebx, %1, 2), (%0, %1, 8), (%%ecx))
1666DEINT_CUBIC((%%ebx, %1), (%0, %1, 8), (%%ebx, %1, 4), (%%ecx), (%%ecx, %1, 2))
3b58b885
MN		 1667
1668 : : "r" (src), "r" (stride)
acced553 1669 : "%eax", "%ebx", "ecx"
3b58b885
MN		 1670 );
1671#else
1672 int x;
7fb36f6c 1673 src+= stride*3;
3b58b885
MN		 1674 for(x=0; x<8; x++)
1675 {
acced553
MN		 1676 src[stride*3] = (-src[0] + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4;
1677 src[stride*5] = (-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4;
1678 src[stride*7] = (-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4;
1679 src[stride*9] = (-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4;
3b58b885
MN		 1680 src++;
1681 }
1682#endif
1683}
1684
1685/**
1686 * Deinterlaces the given block
7fb36f6c
MN		 1687 * will be called for every 8x8 block and can read & write from line 4-15
1688 * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1689 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
3b58b885 1690 * will shift the image up by 1 line (FIXME if this is a problem)
7fb36f6c 1691 * this filter will read lines 4-13 and write 4-11
3b58b885 1692 */
cc9b0679 1693static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride)
3b58b885
MN		 1694{
1695#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
7fb36f6c 1696 src+= 4*stride;
3b58b885
MN		 1697 asm volatile(
1698 "leal (%0, %1), %%eax \n\t"
1699 "leal (%%eax, %1, 4), %%ebx \n\t"
1700// 0 1 2 3 4 5 6 7 8 9
1701// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
1702
1703 "movq (%0), %%mm0 \n\t" // L0
1704 "movq (%%eax, %1), %%mm1 \n\t" // L2
1705 PAVGB(%%mm1, %%mm0) // L0+L2
1706 "movq (%%eax), %%mm2 \n\t" // L1
1707 PAVGB(%%mm2, %%mm0)
1708 "movq %%mm0, (%0) \n\t"
1709 "movq (%%eax, %1, 2), %%mm0 \n\t" // L3
1710 PAVGB(%%mm0, %%mm2) // L1+L3
1711 PAVGB(%%mm1, %%mm2) // 2L2 + L1 + L3
1712 "movq %%mm2, (%%eax) \n\t"
1713 "movq (%0, %1, 4), %%mm2 \n\t" // L4
1714 PAVGB(%%mm2, %%mm1) // L2+L4
1715 PAVGB(%%mm0, %%mm1) // 2L3 + L2 + L4
1716 "movq %%mm1, (%%eax, %1) \n\t"
1717 "movq (%%ebx), %%mm1 \n\t" // L5
1718 PAVGB(%%mm1, %%mm0) // L3+L5
1719 PAVGB(%%mm2, %%mm0) // 2L4 + L3 + L5
1720 "movq %%mm0, (%%eax, %1, 2) \n\t"
1721 "movq (%%ebx, %1), %%mm0 \n\t" // L6
1722 PAVGB(%%mm0, %%mm2) // L4+L6
1723 PAVGB(%%mm1, %%mm2) // 2L5 + L4 + L6
1724 "movq %%mm2, (%0, %1, 4) \n\t"
1725 "movq (%%ebx, %1, 2), %%mm2 \n\t" // L7
1726 PAVGB(%%mm2, %%mm1) // L5+L7
1727 PAVGB(%%mm0, %%mm1) // 2L6 + L5 + L7
1728 "movq %%mm1, (%%ebx) \n\t"
1729 "movq (%0, %1, 8), %%mm1 \n\t" // L8
1730 PAVGB(%%mm1, %%mm0) // L6+L8
1731 PAVGB(%%mm2, %%mm0) // 2L7 + L6 + L8
1732 "movq %%mm0, (%%ebx, %1) \n\t"
1733 "movq (%%ebx, %1, 4), %%mm0 \n\t" // L9
1734 PAVGB(%%mm0, %%mm2) // L7+L9
1735 PAVGB(%%mm1, %%mm2) // 2L8 + L7 + L9
1736 "movq %%mm2, (%%ebx, %1, 2) \n\t"
1737
1738
1739 : : "r" (src), "r" (stride)
1740 : "%eax", "%ebx"
1741 );
1742#else
1743 int x;
7fb36f6c 1744 src+= 4*stride;
3b58b885
MN		 1745 for(x=0; x<8; x++)
1746 {
1747 src[0 ] = (src[0 ] + 2*src[stride ] + src[stride*2])>>2;
1748 src[stride ] = (src[stride ] + 2*src[stride*2] + src[stride*3])>>2;
1749 src[stride*2] = (src[stride*2] + 2*src[stride*3] + src[stride*4])>>2;
1750 src[stride*3] = (src[stride*3] + 2*src[stride*4] + src[stride*5])>>2;
1751 src[stride*4] = (src[stride*4] + 2*src[stride*5] + src[stride*6])>>2;
1752 src[stride*5] = (src[stride*5] + 2*src[stride*6] + src[stride*7])>>2;
1753 src[stride*6] = (src[stride*6] + 2*src[stride*7] + src[stride*8])>>2;
1754 src[stride*7] = (src[stride*7] + 2*src[stride*8] + src[stride*9])>>2;
1755 src++;
1756 }
1757#endif
1758}
1759
1760/**
1761 * Deinterlaces the given block
7fb36f6c
MN		 1762 * will be called for every 8x8 block and can read & write from line 4-15,
1763 * lines 0-3 have been passed through the deblock / dering filters allready, but can be read too
1764 * lines 4-12 will be read into the deblocking filter and should be deinterlaced
3b58b885 1765 */
cc9b0679 1766static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)
3b58b885 1767{
a6be8111 1768#ifdef HAVE_MMX
7fb36f6c 1769 src+= 4*stride;
a6be8111 1770#ifdef HAVE_MMX2
3b58b885
MN		 1771 asm volatile(
1772 "leal (%0, %1), %%eax \n\t"
1773 "leal (%%eax, %1, 4), %%ebx \n\t"
1774// 0 1 2 3 4 5 6 7 8 9
1775// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
1776
1777 "movq (%0), %%mm0 \n\t" //
1778 "movq (%%eax, %1), %%mm2 \n\t" //
1779 "movq (%%eax), %%mm1 \n\t" //
1780 "movq %%mm0, %%mm3 \n\t"
1781 "pmaxub %%mm1, %%mm0 \n\t" //
1782 "pminub %%mm3, %%mm1 \n\t" //
1783 "pmaxub %%mm2, %%mm1 \n\t" //
1784 "pminub %%mm1, %%mm0 \n\t"
1785 "movq %%mm0, (%%eax) \n\t"
1786
1787 "movq (%0, %1, 4), %%mm0 \n\t" //
1788 "movq (%%eax, %1, 2), %%mm1 \n\t" //
1789 "movq %%mm2, %%mm3 \n\t"
1790 "pmaxub %%mm1, %%mm2 \n\t" //
1791 "pminub %%mm3, %%mm1 \n\t" //
1792 "pmaxub %%mm0, %%mm1 \n\t" //
1793 "pminub %%mm1, %%mm2 \n\t"
1794 "movq %%mm2, (%%eax, %1, 2) \n\t"
1795
1796 "movq (%%ebx), %%mm2 \n\t" //
1797 "movq (%%ebx, %1), %%mm1 \n\t" //
1798 "movq %%mm2, %%mm3 \n\t"
1799 "pmaxub %%mm0, %%mm2 \n\t" //
1800 "pminub %%mm3, %%mm0 \n\t" //
1801 "pmaxub %%mm1, %%mm0 \n\t" //
1802 "pminub %%mm0, %%mm2 \n\t"
1803 "movq %%mm2, (%%ebx) \n\t"
1804
1805 "movq (%%ebx, %1, 2), %%mm2 \n\t" //
1806 "movq (%0, %1, 8), %%mm0 \n\t" //
1807 "movq %%mm2, %%mm3 \n\t"
1808 "pmaxub %%mm0, %%mm2 \n\t" //
1809 "pminub %%mm3, %%mm0 \n\t" //
1810 "pmaxub %%mm1, %%mm0 \n\t" //
1811 "pminub %%mm0, %%mm2 \n\t"
1812 "movq %%mm2, (%%ebx, %1, 2) \n\t"
1813
1814
1815 : : "r" (src), "r" (stride)
1816 : "%eax", "%ebx"
1817 );
a6be8111
MN		 1818
1819#else // MMX without MMX2
1820 asm volatile(
1821 "leal (%0, %1), %%eax \n\t"
1822 "leal (%%eax, %1, 4), %%ebx \n\t"
1823// 0 1 2 3 4 5 6 7 8 9
1824// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
1825 "pxor %%mm7, %%mm7 \n\t"
1826
1827#define MEDIAN(a,b,c)\
1828 "movq " #a ", %%mm0 \n\t"\
1829 "movq " #b ", %%mm2 \n\t"\
1830 "movq " #c ", %%mm1 \n\t"\
1831 "movq %%mm0, %%mm3 \n\t"\
1832 "movq %%mm1, %%mm4 \n\t"\
1833 "movq %%mm2, %%mm5 \n\t"\
1834 "psubusb %%mm1, %%mm3 \n\t"\
1835 "psubusb %%mm2, %%mm4 \n\t"\
1836 "psubusb %%mm0, %%mm5 \n\t"\
1837 "pcmpeqb %%mm7, %%mm3 \n\t"\
1838 "pcmpeqb %%mm7, %%mm4 \n\t"\
1839 "pcmpeqb %%mm7, %%mm5 \n\t"\
1840 "movq %%mm3, %%mm6 \n\t"\
1841 "pxor %%mm4, %%mm3 \n\t"\
1842 "pxor %%mm5, %%mm4 \n\t"\
1843 "pxor %%mm6, %%mm5 \n\t"\
1844 "por %%mm3, %%mm1 \n\t"\
1845 "por %%mm4, %%mm2 \n\t"\
1846 "por %%mm5, %%mm0 \n\t"\
1847 "pand %%mm2, %%mm0 \n\t"\
1848 "pand %%mm1, %%mm0 \n\t"\
1849 "movq %%mm0, " #b " \n\t"
1850
1851MEDIAN((%0), (%%eax), (%%eax, %1))
1852MEDIAN((%%eax, %1), (%%eax, %1, 2), (%0, %1, 4))
1853MEDIAN((%0, %1, 4), (%%ebx), (%%ebx, %1))
1854MEDIAN((%%ebx, %1), (%%ebx, %1, 2), (%0, %1, 8))
1855
1856 : : "r" (src), "r" (stride)
1857 : "%eax", "%ebx"
1858 );
1859#endif // MMX
3b58b885
MN		 1860#else
1861 //FIXME
1862 int x;
7fb36f6c 1863 src+= 4*stride;
3b58b885
MN		 1864 for(x=0; x<8; x++)
1865 {
1866 src[0 ] = (src[0 ] + 2*src[stride ] + src[stride*2])>>2;
1867 src[stride ] = (src[stride ] + 2*src[stride*2] + src[stride*3])>>2;
1868 src[stride*2] = (src[stride*2] + 2*src[stride*3] + src[stride*4])>>2;
1869 src[stride*3] = (src[stride*3] + 2*src[stride*4] + src[stride*5])>>2;
1870 src[stride*4] = (src[stride*4] + 2*src[stride*5] + src[stride*6])>>2;
1871 src[stride*5] = (src[stride*5] + 2*src[stride*6] + src[stride*7])>>2;
1872 src[stride*6] = (src[stride*6] + 2*src[stride*7] + src[stride*8])>>2;
1873 src[stride*7] = (src[stride*7] + 2*src[stride*8] + src[stride*9])>>2;
1874 src++;
1875 }
1876#endif
1877}
1878
e5c30e06 1879#ifdef HAVE_MMX
4e4dcbc5
MN		 1880/**
1881 * transposes and shift the given 8x8 Block into dst1 and dst2
1882 */
cc9b0679 1883static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)
4e4dcbc5
MN		 1884{
1885 asm(
1886 "leal (%0, %1), %%eax \n\t"
1887 "leal (%%eax, %1, 4), %%ebx \n\t"
1888// 0 1 2 3 4 5 6 7 8 9
1889// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
1890 "movq (%0), %%mm0 \n\t" // 12345678
1891 "movq (%%eax), %%mm1 \n\t" // abcdefgh
1892 "movq %%mm0, %%mm2 \n\t" // 12345678
1893 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
1894 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
1895
1896 "movq (%%eax, %1), %%mm1 \n\t"
1897 "movq (%%eax, %1, 2), %%mm3 \n\t"
1898 "movq %%mm1, %%mm4 \n\t"
1899 "punpcklbw %%mm3, %%mm1 \n\t"
1900 "punpckhbw %%mm3, %%mm4 \n\t"
1901
1902 "movq %%mm0, %%mm3 \n\t"
1903 "punpcklwd %%mm1, %%mm0 \n\t"
1904 "punpckhwd %%mm1, %%mm3 \n\t"
1905 "movq %%mm2, %%mm1 \n\t"
1906 "punpcklwd %%mm4, %%mm2 \n\t"
1907 "punpckhwd %%mm4, %%mm1 \n\t"
1908
1909 "movd %%mm0, 128(%2) \n\t"
1910 "psrlq $32, %%mm0 \n\t"
1911 "movd %%mm0, 144(%2) \n\t"
1912 "movd %%mm3, 160(%2) \n\t"
1913 "psrlq $32, %%mm3 \n\t"
1914 "movd %%mm3, 176(%2) \n\t"
1915 "movd %%mm3, 48(%3) \n\t"
1916 "movd %%mm2, 192(%2) \n\t"
1917 "movd %%mm2, 64(%3) \n\t"
1918 "psrlq $32, %%mm2 \n\t"
1919 "movd %%mm2, 80(%3) \n\t"
1920 "movd %%mm1, 96(%3) \n\t"
1921 "psrlq $32, %%mm1 \n\t"
1922 "movd %%mm1, 112(%3) \n\t"
1923
1924 "movq (%0, %1, 4), %%mm0 \n\t" // 12345678
1925 "movq (%%ebx), %%mm1 \n\t" // abcdefgh
1926 "movq %%mm0, %%mm2 \n\t" // 12345678
1927 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
1928 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
1929
1930 "movq (%%ebx, %1), %%mm1 \n\t"
1931 "movq (%%ebx, %1, 2), %%mm3 \n\t"
1932 "movq %%mm1, %%mm4 \n\t"
1933 "punpcklbw %%mm3, %%mm1 \n\t"
1934 "punpckhbw %%mm3, %%mm4 \n\t"
1935
1936 "movq %%mm0, %%mm3 \n\t"
1937 "punpcklwd %%mm1, %%mm0 \n\t"
1938 "punpckhwd %%mm1, %%mm3 \n\t"
1939 "movq %%mm2, %%mm1 \n\t"
1940 "punpcklwd %%mm4, %%mm2 \n\t"
1941 "punpckhwd %%mm4, %%mm1 \n\t"
1942
1943 "movd %%mm0, 132(%2) \n\t"
1944 "psrlq $32, %%mm0 \n\t"
1945 "movd %%mm0, 148(%2) \n\t"
1946 "movd %%mm3, 164(%2) \n\t"
1947 "psrlq $32, %%mm3 \n\t"
1948 "movd %%mm3, 180(%2) \n\t"
1949 "movd %%mm3, 52(%3) \n\t"
1950 "movd %%mm2, 196(%2) \n\t"
1951 "movd %%mm2, 68(%3) \n\t"
1952 "psrlq $32, %%mm2 \n\t"
1953 "movd %%mm2, 84(%3) \n\t"
1954 "movd %%mm1, 100(%3) \n\t"
1955 "psrlq $32, %%mm1 \n\t"
1956 "movd %%mm1, 116(%3) \n\t"
1957
1958
1959 :: "r" (src), "r" (srcStride), "r" (dst1), "r" (dst2)
1960 : "%eax", "%ebx"
1961 );
1962}
1963
1964/**
1965 * transposes the given 8x8 block
1966 */
cc9b0679 1967static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)
4e4dcbc5
MN		 1968{
1969 asm(
1970 "leal (%0, %1), %%eax \n\t"
1971 "leal (%%eax, %1, 4), %%ebx \n\t"
1972// 0 1 2 3 4 5 6 7 8 9
1973// %0 eax eax+%1 eax+2%1 %0+4%1 ebx ebx+%1 ebx+2%1 %0+8%1 ebx+4%1
1974 "movq (%2), %%mm0 \n\t" // 12345678
1975 "movq 16(%2), %%mm1 \n\t" // abcdefgh
1976 "movq %%mm0, %%mm2 \n\t" // 12345678
1977 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
1978 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
1979
1980 "movq 32(%2), %%mm1 \n\t"
1981 "movq 48(%2), %%mm3 \n\t"
1982 "movq %%mm1, %%mm4 \n\t"
1983 "punpcklbw %%mm3, %%mm1 \n\t"
1984 "punpckhbw %%mm3, %%mm4 \n\t"
1985
1986 "movq %%mm0, %%mm3 \n\t"
1987 "punpcklwd %%mm1, %%mm0 \n\t"
1988 "punpckhwd %%mm1, %%mm3 \n\t"
1989 "movq %%mm2, %%mm1 \n\t"
1990 "punpcklwd %%mm4, %%mm2 \n\t"
1991 "punpckhwd %%mm4, %%mm1 \n\t"
1992
1993 "movd %%mm0, (%0) \n\t"
1994 "psrlq $32, %%mm0 \n\t"
1995 "movd %%mm0, (%%eax) \n\t"
1996 "movd %%mm3, (%%eax, %1) \n\t"
1997 "psrlq $32, %%mm3 \n\t"
1998 "movd %%mm3, (%%eax, %1, 2) \n\t"
1999 "movd %%mm2, (%0, %1, 4) \n\t"
2000 "psrlq $32, %%mm2 \n\t"
2001 "movd %%mm2, (%%ebx) \n\t"
2002 "movd %%mm1, (%%ebx, %1) \n\t"
2003 "psrlq $32, %%mm1 \n\t"
2004 "movd %%mm1, (%%ebx, %1, 2) \n\t"
2005
2006
2007 "movq 64(%2), %%mm0 \n\t" // 12345678
2008 "movq 80(%2), %%mm1 \n\t" // abcdefgh
2009 "movq %%mm0, %%mm2 \n\t" // 12345678
2010 "punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
2011 "punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
2012
2013 "movq 96(%2), %%mm1 \n\t"
2014 "movq 112(%2), %%mm3 \n\t"
2015 "movq %%mm1, %%mm4 \n\t"
2016 "punpcklbw %%mm3, %%mm1 \n\t"
2017 "punpckhbw %%mm3, %%mm4 \n\t"
2018
2019 "movq %%mm0, %%mm3 \n\t"
2020 "punpcklwd %%mm1, %%mm0 \n\t"
2021 "punpckhwd %%mm1, %%mm3 \n\t"
2022 "movq %%mm2, %%mm1 \n\t"
2023 "punpcklwd %%mm4, %%mm2 \n\t"
2024 "punpckhwd %%mm4, %%mm1 \n\t"
2025
2026 "movd %%mm0, 4(%0) \n\t"
2027 "psrlq $32, %%mm0 \n\t"
2028 "movd %%mm0, 4(%%eax) \n\t"
2029 "movd %%mm3, 4(%%eax, %1) \n\t"
2030 "psrlq $32, %%mm3 \n\t"
2031 "movd %%mm3, 4(%%eax, %1, 2) \n\t"
2032 "movd %%mm2, 4(%0, %1, 4) \n\t"
2033 "psrlq $32, %%mm2 \n\t"
2034 "movd %%mm2, 4(%%ebx) \n\t"
2035 "movd %%mm1, 4(%%ebx, %1) \n\t"
2036 "psrlq $32, %%mm1 \n\t"
2037 "movd %%mm1, 4(%%ebx, %1, 2) \n\t"
2038
2039 :: "r" (dst), "r" (dstStride), "r" (src)
2040 : "%eax", "%ebx"
2041 );
2042}
e5c30e06 2043#endif
be44a4d7 2044//static int test=0;
4e4dcbc5 2045
cc9b0679 2046static void inline RENAME(tempNoiseReducer)(uint8_t *src, int stride,
a9c77978 2047 uint8_t *tempBlured, uint32_t *tempBluredPast, int *maxNoise)
117e45b0 2048{
be44a4d7
MN		 2049#define FAST_L2_DIFF
2050//#define L1_DIFF //u should change the thresholds too if u try that one
2051#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
2052 asm volatile(
2053 "leal (%2, %2, 2), %%eax \n\t" // 3*stride
2054 "leal (%2, %2, 4), %%ebx \n\t" // 5*stride
2055 "leal (%%ebx, %2, 2), %%ecx \n\t" // 7*stride
2056// 0 1 2 3 4 5 6 7 8 9
2057// %x %x+%2 %x+2%2 %x+eax %x+4%2 %x+ebx %x+2eax %x+ecx %x+8%2
2058//FIXME reorder?
2059#ifdef L1_DIFF //needs mmx2
2060 "movq (%0), %%mm0 \n\t" // L0
2061 "psadbw (%1), %%mm0 \n\t" // |L0-R0|
2062 "movq (%0, %2), %%mm1 \n\t" // L1
2063 "psadbw (%1, %2), %%mm1 \n\t" // |L1-R1|
2064 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2065 "psadbw (%1, %2, 2), %%mm2 \n\t" // |L2-R2|
2066 "movq (%0, %%eax), %%mm3 \n\t" // L3
2067 "psadbw (%1, %%eax), %%mm3 \n\t" // |L3-R3|
2068
2069 "movq (%0, %2, 4), %%mm4 \n\t" // L4
2070 "paddw %%mm1, %%mm0 \n\t"
2071 "psadbw (%1, %2, 4), %%mm4 \n\t" // |L4-R4|
2072 "movq (%0, %%ebx), %%mm5 \n\t" // L5
2073 "paddw %%mm2, %%mm0 \n\t"
2074 "psadbw (%1, %%ebx), %%mm5 \n\t" // |L5-R5|
2075 "movq (%0, %%eax, 2), %%mm6 \n\t" // L6
2076 "paddw %%mm3, %%mm0 \n\t"
2077 "psadbw (%1, %%eax, 2), %%mm6 \n\t" // |L6-R6|
2078 "movq (%0, %%ecx), %%mm7 \n\t" // L7
2079 "paddw %%mm4, %%mm0 \n\t"
2080 "psadbw (%1, %%ecx), %%mm7 \n\t" // |L7-R7|
2081 "paddw %%mm5, %%mm6 \n\t"
2082 "paddw %%mm7, %%mm6 \n\t"
2083 "paddw %%mm6, %%mm0 \n\t"
2084#elif defined (FAST_L2_DIFF)
2085 "pcmpeqb %%mm7, %%mm7 \n\t"
9b464428 2086 "movq "MANGLE(b80)", %%mm6 \n\t"
be44a4d7
MN		 2087 "pxor %%mm0, %%mm0 \n\t"
2088#define L2_DIFF_CORE(a, b)\
2089 "movq " #a ", %%mm5 \n\t"\
2090 "movq " #b ", %%mm2 \n\t"\
2091 "pxor %%mm7, %%mm2 \n\t"\
2092 PAVGB(%%mm2, %%mm5)\
2093 "paddb %%mm6, %%mm5 \n\t"\
2094 "movq %%mm5, %%mm2 \n\t"\
2095 "psllw $8, %%mm5 \n\t"\
2096 "pmaddwd %%mm5, %%mm5 \n\t"\
2097 "pmaddwd %%mm2, %%mm2 \n\t"\
2098 "paddd %%mm2, %%mm5 \n\t"\
2099 "psrld $14, %%mm5 \n\t"\
2100 "paddd %%mm5, %%mm0 \n\t"
2101
2102L2_DIFF_CORE((%0), (%1))
2103L2_DIFF_CORE((%0, %2), (%1, %2))
2104L2_DIFF_CORE((%0, %2, 2), (%1, %2, 2))
2105L2_DIFF_CORE((%0, %%eax), (%1, %%eax))
2106L2_DIFF_CORE((%0, %2, 4), (%1, %2, 4))
2107L2_DIFF_CORE((%0, %%ebx), (%1, %%ebx))
2108L2_DIFF_CORE((%0, %%eax,2), (%1, %%eax,2))
2109L2_DIFF_CORE((%0, %%ecx), (%1, %%ecx))
2110
2111#else
2112 "pxor %%mm7, %%mm7 \n\t"
2113 "pxor %%mm0, %%mm0 \n\t"
2114#define L2_DIFF_CORE(a, b)\
2115 "movq " #a ", %%mm5 \n\t"\
2116 "movq " #b ", %%mm2 \n\t"\
2117 "movq %%mm5, %%mm1 \n\t"\
2118 "movq %%mm2, %%mm3 \n\t"\
2119 "punpcklbw %%mm7, %%mm5 \n\t"\
2120 "punpckhbw %%mm7, %%mm1 \n\t"\
2121 "punpcklbw %%mm7, %%mm2 \n\t"\
2122 "punpckhbw %%mm7, %%mm3 \n\t"\
2123 "psubw %%mm2, %%mm5 \n\t"\
2124 "psubw %%mm3, %%mm1 \n\t"\
2125 "pmaddwd %%mm5, %%mm5 \n\t"\
2126 "pmaddwd %%mm1, %%mm1 \n\t"\
2127 "paddd %%mm1, %%mm5 \n\t"\
2128 "paddd %%mm5, %%mm0 \n\t"
2129
2130L2_DIFF_CORE((%0), (%1))
2131L2_DIFF_CORE((%0, %2), (%1, %2))
2132L2_DIFF_CORE((%0, %2, 2), (%1, %2, 2))
2133L2_DIFF_CORE((%0, %%eax), (%1, %%eax))
2134L2_DIFF_CORE((%0, %2, 4), (%1, %2, 4))
2135L2_DIFF_CORE((%0, %%ebx), (%1, %%ebx))
2136L2_DIFF_CORE((%0, %%eax,2), (%1, %%eax,2))
2137L2_DIFF_CORE((%0, %%ecx), (%1, %%ecx))
2138
2139#endif
2140
2141 "movq %%mm0, %%mm4 \n\t"
2142 "psrlq $32, %%mm0 \n\t"
2143 "paddd %%mm0, %%mm4 \n\t"
2144 "movd %%mm4, %%ecx \n\t"
a9c77978
MN		 2145 "shll $2, %%ecx \n\t"
2146 "movl %3, %%ebx \n\t"
2147 "addl -4(%%ebx), %%ecx \n\t"
2148 "addl 4(%%ebx), %%ecx \n\t"
2149 "addl -1024(%%ebx), %%ecx \n\t"
2150 "addl $4, %%ecx \n\t"
2151 "addl 1024(%%ebx), %%ecx \n\t"
2152 "shrl $3, %%ecx \n\t"
2153 "movl %%ecx, (%%ebx) \n\t"
2154 "leal (%%eax, %2, 2), %%ebx \n\t" // 5*stride
2155
9b464428 2156// "movl %3, %%ecx \n\t"
be44a4d7
MN		 2157// "movl %%ecx, test \n\t"
2158// "jmp 4f \n\t"
9b464428 2159 "cmpl 4+"MANGLE(maxTmpNoise)", %%ecx \n\t"
be44a4d7 2160 " jb 2f \n\t"
9b464428 2161 "cmpl 8+"MANGLE(maxTmpNoise)", %%ecx \n\t"
be44a4d7
MN		 2162 " jb 1f \n\t"
2163
2164 "leal (%%ebx, %2, 2), %%ecx \n\t" // 7*stride
2165 "movq (%0), %%mm0 \n\t" // L0
2166 "movq (%0, %2), %%mm1 \n\t" // L1
2167 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2168 "movq (%0, %%eax), %%mm3 \n\t" // L3
2169 "movq (%0, %2, 4), %%mm4 \n\t" // L4
2170 "movq (%0, %%ebx), %%mm5 \n\t" // L5
2171 "movq (%0, %%eax, 2), %%mm6 \n\t" // L6
2172 "movq (%0, %%ecx), %%mm7 \n\t" // L7
2173 "movq %%mm0, (%1) \n\t" // L0
2174 "movq %%mm1, (%1, %2) \n\t" // L1
2175 "movq %%mm2, (%1, %2, 2) \n\t" // L2
2176 "movq %%mm3, (%1, %%eax) \n\t" // L3
2177 "movq %%mm4, (%1, %2, 4) \n\t" // L4
2178 "movq %%mm5, (%1, %%ebx) \n\t" // L5
2179 "movq %%mm6, (%1, %%eax, 2) \n\t" // L6
2180 "movq %%mm7, (%1, %%ecx) \n\t" // L7
2181 "jmp 4f \n\t"
2182
2183 "1: \n\t"
2184 "leal (%%ebx, %2, 2), %%ecx \n\t" // 7*stride
2185 "movq (%0), %%mm0 \n\t" // L0
413dec62 2186 PAVGB((%1), %%mm0) // L0
be44a4d7 2187 "movq (%0, %2), %%mm1 \n\t" // L1
413dec62 2188 PAVGB((%1, %2), %%mm1) // L1
be44a4d7 2189 "movq (%0, %2, 2), %%mm2 \n\t" // L2
413dec62 2190 PAVGB((%1, %2, 2), %%mm2) // L2
be44a4d7 2191 "movq (%0, %%eax), %%mm3 \n\t" // L3
413dec62 2192 PAVGB((%1, %%eax), %%mm3) // L3
be44a4d7 2193 "movq (%0, %2, 4), %%mm4 \n\t" // L4
413dec62 2194 PAVGB((%1, %2, 4), %%mm4) // L4
be44a4d7 2195 "movq (%0, %%ebx), %%mm5 \n\t" // L5
413dec62 2196 PAVGB((%1, %%ebx), %%mm5) // L5
be44a4d7 2197 "movq (%0, %%eax, 2), %%mm6 \n\t" // L6
413dec62 2198 PAVGB((%1, %%eax, 2), %%mm6) // L6
be44a4d7 2199 "movq (%0, %%ecx), %%mm7 \n\t" // L7
413dec62 2200 PAVGB((%1, %%ecx), %%mm7) // L7
be44a4d7
MN		 2201 "movq %%mm0, (%1) \n\t" // R0
2202 "movq %%mm1, (%1, %2) \n\t" // R1
2203 "movq %%mm2, (%1, %2, 2) \n\t" // R2
2204 "movq %%mm3, (%1, %%eax) \n\t" // R3
2205 "movq %%mm4, (%1, %2, 4) \n\t" // R4
2206 "movq %%mm5, (%1, %%ebx) \n\t" // R5
2207 "movq %%mm6, (%1, %%eax, 2) \n\t" // R6
2208 "movq %%mm7, (%1, %%ecx) \n\t" // R7
2209 "movq %%mm0, (%0) \n\t" // L0
2210 "movq %%mm1, (%0, %2) \n\t" // L1
2211 "movq %%mm2, (%0, %2, 2) \n\t" // L2
2212 "movq %%mm3, (%0, %%eax) \n\t" // L3
2213 "movq %%mm4, (%0, %2, 4) \n\t" // L4
2214 "movq %%mm5, (%0, %%ebx) \n\t" // L5
2215 "movq %%mm6, (%0, %%eax, 2) \n\t" // L6
2216 "movq %%mm7, (%0, %%ecx) \n\t" // L7
2217 "jmp 4f \n\t"
2218
2219 "2: \n\t"
9b464428 2220 "cmpl "MANGLE(maxTmpNoise)", %%ecx \n\t"
be44a4d7
MN		 2221 " jb 3f \n\t"
2222
2223 "leal (%%ebx, %2, 2), %%ecx \n\t" // 7*stride
2224 "movq (%0), %%mm0 \n\t" // L0
2225 "movq (%0, %2), %%mm1 \n\t" // L1
2226 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2227 "movq (%0, %%eax), %%mm3 \n\t" // L3
2228 "movq (%1), %%mm4 \n\t" // R0
2229 "movq (%1, %2), %%mm5 \n\t" // R1
2230 "movq (%1, %2, 2), %%mm6 \n\t" // R2
2231 "movq (%1, %%eax), %%mm7 \n\t" // R3
2232 PAVGB(%%mm4, %%mm0)
2233 PAVGB(%%mm5, %%mm1)
2234 PAVGB(%%mm6, %%mm2)
2235 PAVGB(%%mm7, %%mm3)
2236 PAVGB(%%mm4, %%mm0)
2237 PAVGB(%%mm5, %%mm1)
2238 PAVGB(%%mm6, %%mm2)
2239 PAVGB(%%mm7, %%mm3)
2240 "movq %%mm0, (%1) \n\t" // R0
2241 "movq %%mm1, (%1, %2) \n\t" // R1
2242 "movq %%mm2, (%1, %2, 2) \n\t" // R2
2243 "movq %%mm3, (%1, %%eax) \n\t" // R3
2244 "movq %%mm0, (%0) \n\t" // L0
2245 "movq %%mm1, (%0, %2) \n\t" // L1
2246 "movq %%mm2, (%0, %2, 2) \n\t" // L2
2247 "movq %%mm3, (%0, %%eax) \n\t" // L3
2248
2249 "movq (%0, %2, 4), %%mm0 \n\t" // L4
2250 "movq (%0, %%ebx), %%mm1 \n\t" // L5
2251 "movq (%0, %%eax, 2), %%mm2 \n\t" // L6
2252 "movq (%0, %%ecx), %%mm3 \n\t" // L7
2253 "movq (%1, %2, 4), %%mm4 \n\t" // R4
2254 "movq (%1, %%ebx), %%mm5 \n\t" // R5
2255 "movq (%1, %%eax, 2), %%mm6 \n\t" // R6
2256 "movq (%1, %%ecx), %%mm7 \n\t" // R7
2257 PAVGB(%%mm4, %%mm0)
2258 PAVGB(%%mm5, %%mm1)
2259 PAVGB(%%mm6, %%mm2)
2260 PAVGB(%%mm7, %%mm3)
2261 PAVGB(%%mm4, %%mm0)
2262 PAVGB(%%mm5, %%mm1)
2263 PAVGB(%%mm6, %%mm2)
2264 PAVGB(%%mm7, %%mm3)
2265 "movq %%mm0, (%1, %2, 4) \n\t" // R4
2266 "movq %%mm1, (%1, %%ebx) \n\t" // R5
2267 "movq %%mm2, (%1, %%eax, 2) \n\t" // R6
2268 "movq %%mm3, (%1, %%ecx) \n\t" // R7
2269 "movq %%mm0, (%0, %2, 4) \n\t" // L4
2270 "movq %%mm1, (%0, %%ebx) \n\t" // L5
2271 "movq %%mm2, (%0, %%eax, 2) \n\t" // L6
2272 "movq %%mm3, (%0, %%ecx) \n\t" // L7
2273 "jmp 4f \n\t"
2274
2275 "3: \n\t"
2276 "leal (%%ebx, %2, 2), %%ecx \n\t" // 7*stride
2277 "movq (%0), %%mm0 \n\t" // L0
2278 "movq (%0, %2), %%mm1 \n\t" // L1
2279 "movq (%0, %2, 2), %%mm2 \n\t" // L2
2280 "movq (%0, %%eax), %%mm3 \n\t" // L3
2281 "movq (%1), %%mm4 \n\t" // R0
2282 "movq (%1, %2), %%mm5 \n\t" // R1
2283 "movq (%1, %2, 2), %%mm6 \n\t" // R2
2284 "movq (%1, %%eax), %%mm7 \n\t" // R3
2285 PAVGB(%%mm4, %%mm0)
2286 PAVGB(%%mm5, %%mm1)
2287 PAVGB(%%mm6, %%mm2)
2288 PAVGB(%%mm7, %%mm3)
2289 PAVGB(%%mm4, %%mm0)
2290 PAVGB(%%mm5, %%mm1)
2291 PAVGB(%%mm6, %%mm2)
2292 PAVGB(%%mm7, %%mm3)
2293 PAVGB(%%mm4, %%mm0)
2294 PAVGB(%%mm5, %%mm1)
2295 PAVGB(%%mm6, %%mm2)
2296 PAVGB(%%mm7, %%mm3)
2297 "movq %%mm0, (%1) \n\t" // R0
2298 "movq %%mm1, (%1, %2) \n\t" // R1
2299 "movq %%mm2, (%1, %2, 2) \n\t" // R2
2300 "movq %%mm3, (%1, %%eax) \n\t" // R3
2301 "movq %%mm0, (%0) \n\t" // L0
2302 "movq %%mm1, (%0, %2) \n\t" // L1
2303 "movq %%mm2, (%0, %2, 2) \n\t" // L2
2304 "movq %%mm3, (%0, %%eax) \n\t" // L3
2305
2306 "movq (%0, %2, 4), %%mm0 \n\t" // L4
2307 "movq (%0, %%ebx), %%mm1 \n\t" // L5
2308 "movq (%0, %%eax, 2), %%mm2 \n\t" // L6
2309 "movq (%0, %%ecx), %%mm3 \n\t" // L7
2310 "movq (%1, %2, 4), %%mm4 \n\t" // R4
2311 "movq (%1, %%ebx), %%mm5 \n\t" // R5
2312 "movq (%1, %%eax, 2), %%mm6 \n\t" // R6
2313 "movq (%1, %%ecx), %%mm7 \n\t" // R7
2314 PAVGB(%%mm4, %%mm0)
2315 PAVGB(%%mm5, %%mm1)
2316 PAVGB(%%mm6, %%mm2)
2317 PAVGB(%%mm7, %%mm3)
2318 PAVGB(%%mm4, %%mm0)
2319 PAVGB(%%mm5, %%mm1)
2320 PAVGB(%%mm6, %%mm2)
2321 PAVGB(%%mm7, %%mm3)
2322 PAVGB(%%mm4, %%mm0)
2323 PAVGB(%%mm5, %%mm1)
2324 PAVGB(%%mm6, %%mm2)
2325 PAVGB(%%mm7, %%mm3)
2326 "movq %%mm0, (%1, %2, 4) \n\t" // R4
2327 "movq %%mm1, (%1, %%ebx) \n\t" // R5
2328 "movq %%mm2, (%1, %%eax, 2) \n\t" // R6
2329 "movq %%mm3, (%1, %%ecx) \n\t" // R7
2330 "movq %%mm0, (%0, %2, 4) \n\t" // L4
2331 "movq %%mm1, (%0, %%ebx) \n\t" // L5
2332 "movq %%mm2, (%0, %%eax, 2) \n\t" // L6
2333 "movq %%mm3, (%0, %%ecx) \n\t" // L7
2334
2335 "4: \n\t"
2336
a9c77978 2337 :: "r" (src), "r" (tempBlured), "r"(stride), "m" (tempBluredPast)
be44a4d7
MN		 2338 : "%eax", "%ebx", "%ecx", "memory"
2339 );
2340//printf("%d\n", test);
2341#else
117e45b0
MN		 2342 int y;
2343 int d=0;
2344 int sysd=0;
a9c77978 2345 int i;
117e45b0
MN		 2346
2347 for(y=0; y<8; y++)
2348 {
2349 int x;
2350 for(x=0; x<8; x++)
2351 {
2352 int ref= tempBlured[ x + y*stride ];
2353 int cur= src[ x + y*stride ];
2354 int d1=ref - cur;
be44a4d7
MN		 2355// if(x==0 || x==7) d1+= d1>>1;
2356// if(y==0 || y==7) d1+= d1>>1;
2357// d+= ABS(d1);
2358 d+= d1*d1;
117e45b0
MN		 2359 sysd+= d1;
2360 }
2361 }
a9c77978
MN		 2362 i=d;
2363 d= (
2364 4*d
2365 +(*(tempBluredPast-256))
2366 +(*(tempBluredPast-1))+ (*(tempBluredPast+1))
2367 +(*(tempBluredPast+256))
2368 +4)>>3;
2369 *tempBluredPast=i;
2370// ((*tempBluredPast)*3 + d + 2)>>2;
2371
117e45b0
MN		 2372//printf("%d %d %d\n", maxNoise[0], maxNoise[1], maxNoise[2]);
2373/*
2374Switch between
2375 1 0 0 0 0 0 0 (0)
237664 32 16 8 4 2 1 (1)
237764 48 36 27 20 15 11 (33) (approx)
237864 56 49 43 37 33 29 (200) (approx)
2379*/
2380 if(d > maxNoise[1])
2381 {
2382 if(d < maxNoise[2])
2383 {
2384 for(y=0; y<8; y++)
2385 {
2386 int x;
2387 for(x=0; x<8; x++)
2388 {
2389 int ref= tempBlured[ x + y*stride ];
2390 int cur= src[ x + y*stride ];
2391 tempBlured[ x + y*stride ]=
2392 src[ x + y*stride ]=
2393 (ref + cur + 1)>>1;
2394 }
2395 }
2396 }
2397 else
2398 {
2399 for(y=0; y<8; y++)
2400 {
2401 int x;
2402 for(x=0; x<8; x++)
2403 {
2404 tempBlured[ x + y*stride ]= src[ x + y*stride ];
2405 }
2406 }
2407 }
2408 }
2409 else
2410 {
2411 if(d < maxNoise[0])
2412 {
2413 for(y=0; y<8; y++)
2414 {
2415 int x;
2416 for(x=0; x<8; x++)
2417 {
2418 int ref= tempBlured[ x + y*stride ];
2419 int cur= src[ x + y*stride ];
2420 tempBlured[ x + y*stride ]=
2421 src[ x + y*stride ]=
2422 (ref*7 + cur + 4)>>3;
2423 }
2424 }
2425 }
2426 else
2427 {
2428 for(y=0; y<8; y++)
2429 {
2430 int x;
2431 for(x=0; x<8; x++)
2432 {
2433 int ref= tempBlured[ x + y*stride ];
2434 int cur= src[ x + y*stride ];
2435 tempBlured[ x + y*stride ]=
2436 src[ x + y*stride ]=
2437 (ref*3 + cur + 2)>>2;
2438 }
2439 }
2440 }
2441 }
be44a4d7 2442#endif
117e45b0
MN		 2443}
2444
cc9b0679 2445static void RENAME(postProcess)(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
117e45b0 2446 QP_STORE_T QPs[], int QPStride, int isColor, struct PPMode *ppMode);
13e00528 2447
3057fa66
A		 2448/**
2449 * Copies a block from src to dst and fixes the blacklevel
d5a1a995 2450 * levelFix == 0 -> dont touch the brighness & contrast
3057fa66 2451 */
6b791538
A		 2452#undef SCALED_CPY
2453
cc9b0679 2454static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, uint8_t src[], int srcStride,
c09dc465 2455 int levelFix)
3057fa66 2456{
e5c30e06 2457#ifndef HAVE_MMX
d5a1a995 2458 int i;
e5c30e06 2459#endif
d5a1a995
MN		 2460 if(levelFix)
2461 {
3057fa66
A		 2462#ifdef HAVE_MMX
2463 asm volatile(
043ba56f
MN		 2464 "leal (%0,%2), %%eax \n\t"
2465 "leal (%1,%3), %%ebx \n\t"
9b464428
FB		 2466 "movq "MANGLE(packedYOffset)", %%mm2\n\t"
2467 "movq "MANGLE(packedYScale)", %%mm3\n\t"
5b65f0df 2468 "pxor %%mm4, %%mm4 \n\t"
6e9b4840
MN		 2469#ifdef HAVE_MMX2
2470#define SCALED_CPY(src1, src2, dst1, dst2) \
2471 "movq " #src1 ", %%mm0 \n\t"\
2472 "movq " #src1 ", %%mm5 \n\t"\
2473 "movq " #src2 ", %%mm1 \n\t"\
2474 "movq " #src2 ", %%mm6 \n\t"\
2475 "punpcklbw %%mm0, %%mm0 \n\t"\
2476 "punpckhbw %%mm5, %%mm5 \n\t"\
2477 "punpcklbw %%mm1, %%mm1 \n\t"\
2478 "punpckhbw %%mm6, %%mm6 \n\t"\
2479 "pmulhuw %%mm3, %%mm0 \n\t"\
2480 "pmulhuw %%mm3, %%mm5 \n\t"\
2481 "pmulhuw %%mm3, %%mm1 \n\t"\
2482 "pmulhuw %%mm3, %%mm6 \n\t"\
2483 "psubw %%mm2, %%mm0 \n\t"\
2484 "psubw %%mm2, %%mm5 \n\t"\
2485 "psubw %%mm2, %%mm1 \n\t"\
2486 "psubw %%mm2, %%mm6 \n\t"\
2487 "packuswb %%mm5, %%mm0 \n\t"\
2488 "packuswb %%mm6, %%mm1 \n\t"\
2489 "movq %%mm0, " #dst1 " \n\t"\
2490 "movq %%mm1, " #dst2 " \n\t"\
3057fa66 2491
6e9b4840 2492#else //HAVE_MMX2
043ba56f
MN		 2493#define SCALED_CPY(src1, src2, dst1, dst2) \
2494 "movq " #src1 ", %%mm0 \n\t"\
2495 "movq " #src1 ", %%mm5 \n\t"\
5b65f0df
MN		 2496 "punpcklbw %%mm4, %%mm0 \n\t"\
2497 "punpckhbw %%mm4, %%mm5 \n\t"\
57d04d3f
MN		 2498 "psubw %%mm2, %%mm0 \n\t"\
2499 "psubw %%mm2, %%mm5 \n\t"\
043ba56f 2500 "movq " #src2 ", %%mm1 \n\t"\
57d04d3f
MN		 2501 "psllw $6, %%mm0 \n\t"\
2502 "psllw $6, %%mm5 \n\t"\
5b65f0df 2503 "pmulhw %%mm3, %%mm0 \n\t"\
043ba56f 2504 "movq " #src2 ", %%mm6 \n\t"\
5b65f0df 2505 "pmulhw %%mm3, %%mm5 \n\t"\
5b65f0df 2506 "punpcklbw %%mm4, %%mm1 \n\t"\
534a602d 2507 "punpckhbw %%mm4, %%mm6 \n\t"\
57d04d3f 2508 "psubw %%mm2, %%mm1 \n\t"\
534a602d 2509 "psubw %%mm2, %%mm6 \n\t"\
57d04d3f 2510 "psllw $6, %%mm1 \n\t"\
534a602d 2511 "psllw $6, %%mm6 \n\t"\
5b65f0df 2512 "pmulhw %%mm3, %%mm1 \n\t"\
534a602d 2513 "pmulhw %%mm3, %%mm6 \n\t"\
534a602d
MN		 2514 "packuswb %%mm5, %%mm0 \n\t"\
2515 "packuswb %%mm6, %%mm1 \n\t"\
043ba56f
MN		 2516 "movq %%mm0, " #dst1 " \n\t"\
2517 "movq %%mm1, " #dst2 " \n\t"\
2518
6e9b4840
MN		 2519#endif //!HAVE_MMX2
2520
043ba56f
MN		 2521SCALED_CPY((%0) , (%0, %2) , (%1) , (%1, %3))
2522SCALED_CPY((%0, %2, 2), (%%eax, %2, 2), (%1, %3, 2), (%%ebx, %3, 2))
2523SCALED_CPY((%0, %2, 4), (%%eax, %2, 4), (%1, %3, 4), (%%ebx, %3, 4))
2524 "leal (%%eax,%2,4), %%eax \n\t"
2525 "leal (%%ebx,%3,4), %%ebx \n\t"
2526SCALED_CPY((%%eax, %2), (%%eax, %2, 2), (%%ebx, %3), (%%ebx, %3, 2))
2527
2528
2529 : : "r"(src),
2530 "r"(dst),
2531 "r" (srcStride),
534a602d 2532 "r" (dstStride)
d5a1a995
MN		 2533 : "%eax", "%ebx"
2534 );
2535#else
c09dc465 2536 for(i=0; i<8; i++)
d5a1a995
MN		 2537 memcpy( &(dst[dstStride*i]),
2538 &(src[srcStride*i]), BLOCK_SIZE);
2539#endif
2540 }
2541 else
2542 {
2543#ifdef HAVE_MMX
2544 asm volatile(
043ba56f
MN		 2545 "leal (%0,%2), %%eax \n\t"
2546 "leal (%1,%3), %%ebx \n\t"
2547
2548#define SIMPLE_CPY(src1, src2, dst1, dst2) \
2549 "movq " #src1 ", %%mm0 \n\t"\
2550 "movq " #src2 ", %%mm1 \n\t"\
2551 "movq %%mm0, " #dst1 " \n\t"\
2552 "movq %%mm1, " #dst2 " \n\t"\
2553
2554SIMPLE_CPY((%0) , (%0, %2) , (%1) , (%1, %3))
2555SIMPLE_CPY((%0, %2, 2), (%%eax, %2, 2), (%1, %3, 2), (%%ebx, %3, 2))
2556SIMPLE_CPY((%0, %2, 4), (%%eax, %2, 4), (%1, %3, 4), (%%ebx, %3, 4))
2557 "leal (%%eax,%2,4), %%eax \n\t"
2558 "leal (%%ebx,%3,4), %%ebx \n\t"
2559SIMPLE_CPY((%%eax, %2), (%%eax, %2, 2), (%%ebx, %3), (%%ebx, %3, 2))
2560
3057fa66
A		 2561 : : "r" (src),
2562 "r" (dst),
2563 "r" (srcStride),
c09dc465 2564 "r" (dstStride)
3057fa66
A		 2565 : "%eax", "%ebx"
2566 );
2567#else
c09dc465 2568 for(i=0; i<8; i++)
3057fa66
A		 2569 memcpy( &(dst[dstStride*i]),
2570 &(src[srcStride*i]), BLOCK_SIZE);
2571#endif
d5a1a995 2572 }
3057fa66
A		 2573}
2574
4b6388d1
MN		 2575/**
2576 * Duplicates the given 8 src pixels ? times upward
2577 */
2578static inline void RENAME(duplicate)(uint8_t src[], int stride)
2579{
2580#ifdef HAVE_MMX
2581 asm volatile(
2582 "movq (%0), %%mm0 \n\t"
2583 "addl %1, %0 \n\t"
2584 "movq %%mm0, (%0) \n\t"
2585 "movq %%mm0, (%0, %1) \n\t"
2586 "movq %%mm0, (%0, %1, 2) \n\t"
2587 : "+r" (src)
2588 : "r" (-stride)
2589 );
2590#else
2591 int i;
2592 uint8_t *p=src;
2593 for(i=0; i<3; i++)
2594 {
2595 p-= stride;
2596 memcpy(p, src, 8);
2597 }
2598#endif
2599}
3057fa66
A		 2600
2601/**
2602 * Filters array of bytes (Y or U or V values)
2603 */
cc9b0679 2604static void RENAME(postProcess)(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
117e45b0 2605 QP_STORE_T QPs[], int QPStride, int isColor, struct PPMode *ppMode)
3057fa66 2606{
d5a1a995 2607 int x,y;
06c4804a
MN		 2608#ifdef COMPILE_TIME_MODE
2609 const int mode= COMPILE_TIME_MODE;
2610#else
117e45b0 2611 const int mode= isColor ? ppMode->chromMode : ppMode->lumMode;
06c4804a 2612#endif
d5a1a995
MN		 2613