2
* Copyright (C) 2001-2002 Michael Niedermayer (michaelni@gmx.at)
4
* This file is part of FFmpeg.
6
* FFmpeg is free software; you can redistribute it and/or modify
7
* it under the terms of the GNU General Public License as published by
8
* the Free Software Foundation; either version 2 of the License, or
9
* (at your option) any later version.
11
* FFmpeg is distributed in the hope that it will be useful,
12
* but WITHOUT ANY WARRANTY; without even the implied warranty of
13
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14
* GNU General Public License for more details.
16
* You should have received a copy of the GNU General Public License
17
* along with FFmpeg; if not, write to the Free Software
18
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22
* @file postprocess_template.c
23
* mmx/mmx2/3dnow postprocess code.
26
#include "libavutil/x86_cpu.h"
28
#define ALIGN_MASK "$-8"
35
#define REAL_PAVGB(a,b) "pavgb " #a ", " #b " \n\t"
36
#elif defined (HAVE_3DNOW)
37
#define REAL_PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"
39
#define PAVGB(a,b) REAL_PAVGB(a,b)
42
#define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"
43
#elif defined (HAVE_MMX)
44
#define PMINUB(b,a,t) \
45
"movq " #a ", " #t " \n\t"\
46
"psubusb " #b ", " #t " \n\t"\
47
"psubb " #t ", " #a " \n\t"
51
#define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"
52
#elif defined (HAVE_MMX)
54
"psubusb " #a ", " #b " \n\t"\
55
"paddb " #a ", " #b " \n\t"
58
//FIXME? |255-0| = 1 (should not be a problem ...)
61
* Check if the middle 8x8 Block in the given 8x16 block is flat
63
static inline int RENAME(vertClassify)(uint8_t src[], int stride, PPContext *c){
65
src+= stride*4; // src points to begin of the 8x8 Block
69
: : "m" (c->mmxDcOffset[c->nonBQP]), "m" (c->mmxDcThreshold[c->nonBQP])
73
"lea (%2, %3), %%"REG_a" \n\t"
74
// 0 1 2 3 4 5 6 7 8 9
75
// %1 eax eax+%2 eax+2%2 %1+4%2 ecx ecx+%2 ecx+2%2 %1+8%2 ecx+4%2
77
"movq (%2), %%mm0 \n\t"
78
"movq (%%"REG_a"), %%mm1 \n\t"
79
"movq %%mm0, %%mm3 \n\t"
80
"movq %%mm0, %%mm4 \n\t"
82
PMINUB(%%mm1, %%mm3, %%mm5)
83
"psubb %%mm1, %%mm0 \n\t" // mm0 = differnece
84
"paddb %%mm7, %%mm0 \n\t"
85
"pcmpgtb %%mm6, %%mm0 \n\t"
87
"movq (%%"REG_a",%3), %%mm2 \n\t"
89
PMINUB(%%mm2, %%mm3, %%mm5)
90
"psubb %%mm2, %%mm1 \n\t"
91
"paddb %%mm7, %%mm1 \n\t"
92
"pcmpgtb %%mm6, %%mm1 \n\t"
93
"paddb %%mm1, %%mm0 \n\t"
95
"movq (%%"REG_a", %3, 2), %%mm1 \n\t"
97
PMINUB(%%mm1, %%mm3, %%mm5)
98
"psubb %%mm1, %%mm2 \n\t"
99
"paddb %%mm7, %%mm2 \n\t"
100
"pcmpgtb %%mm6, %%mm2 \n\t"
101
"paddb %%mm2, %%mm0 \n\t"
103
"lea (%%"REG_a", %3, 4), %%"REG_a" \n\t"
105
"movq (%2, %3, 4), %%mm2 \n\t"
107
PMINUB(%%mm2, %%mm3, %%mm5)
108
"psubb %%mm2, %%mm1 \n\t"
109
"paddb %%mm7, %%mm1 \n\t"
110
"pcmpgtb %%mm6, %%mm1 \n\t"
111
"paddb %%mm1, %%mm0 \n\t"
113
"movq (%%"REG_a"), %%mm1 \n\t"
115
PMINUB(%%mm1, %%mm3, %%mm5)
116
"psubb %%mm1, %%mm2 \n\t"
117
"paddb %%mm7, %%mm2 \n\t"
118
"pcmpgtb %%mm6, %%mm2 \n\t"
119
"paddb %%mm2, %%mm0 \n\t"
121
"movq (%%"REG_a", %3), %%mm2 \n\t"
123
PMINUB(%%mm2, %%mm3, %%mm5)
124
"psubb %%mm2, %%mm1 \n\t"
125
"paddb %%mm7, %%mm1 \n\t"
126
"pcmpgtb %%mm6, %%mm1 \n\t"
127
"paddb %%mm1, %%mm0 \n\t"
129
"movq (%%"REG_a", %3, 2), %%mm1 \n\t"
131
PMINUB(%%mm1, %%mm3, %%mm5)
132
"psubb %%mm1, %%mm2 \n\t"
133
"paddb %%mm7, %%mm2 \n\t"
134
"pcmpgtb %%mm6, %%mm2 \n\t"
135
"paddb %%mm2, %%mm0 \n\t"
136
"psubusb %%mm3, %%mm4 \n\t"
140
"pxor %%mm7, %%mm7 \n\t"
141
"psadbw %%mm7, %%mm0 \n\t"
143
"movq %%mm0, %%mm1 \n\t"
144
"psrlw $8, %%mm0 \n\t"
145
"paddb %%mm1, %%mm0 \n\t"
146
"movq %%mm0, %%mm1 \n\t"
147
"psrlq $16, %%mm0 \n\t"
148
"paddb %%mm1, %%mm0 \n\t"
149
"movq %%mm0, %%mm1 \n\t"
150
"psrlq $32, %%mm0 \n\t"
151
"paddb %%mm1, %%mm0 \n\t"
153
"movq %4, %%mm7 \n\t" // QP,..., QP
154
"paddusb %%mm7, %%mm7 \n\t" // 2QP ... 2QP
155
"psubusb %%mm7, %%mm4 \n\t" // Diff <= 2QP -> 0
156
"packssdw %%mm4, %%mm4 \n\t"
157
"movd %%mm0, %0 \n\t"
158
"movd %%mm4, %1 \n\t"
160
: "=r" (numEq), "=r" (dcOk)
161
: "r" (src), "r" ((long)stride), "m" (c->pQPb)
165
numEq= (-numEq) &0xFF;
166
if(numEq > c->ppMode.flatnessThreshold){
176
* Do a vertical low pass filter on the 8x16 block (only write to the 8x8 block in the middle)
177
* using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16
180
static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)
182
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
184
asm volatile( //"movv %0 %1 %2\n\t"
185
"movq %2, %%mm0 \n\t" // QP,..., QP
186
"pxor %%mm4, %%mm4 \n\t"
188
"movq (%0), %%mm6 \n\t"
189
"movq (%0, %1), %%mm5 \n\t"
190
"movq %%mm5, %%mm1 \n\t"
191
"movq %%mm6, %%mm2 \n\t"
192
"psubusb %%mm6, %%mm5 \n\t"
193
"psubusb %%mm1, %%mm2 \n\t"
194
"por %%mm5, %%mm2 \n\t" // ABS Diff of lines
195
"psubusb %%mm0, %%mm2 \n\t" // diff <= QP -> 0
196
"pcmpeqb %%mm4, %%mm2 \n\t" // diff <= QP -> FF
198
"pand %%mm2, %%mm6 \n\t"
199
"pandn %%mm1, %%mm2 \n\t"
200
"por %%mm2, %%mm6 \n\t"// First Line to Filter
202
"movq (%0, %1, 8), %%mm5 \n\t"
203
"lea (%0, %1, 4), %%"REG_a" \n\t"
204
"lea (%0, %1, 8), %%"REG_c" \n\t"
205
"sub %1, %%"REG_c" \n\t"
206
"add %1, %0 \n\t" // %0 points to line 1 not 0
207
"movq (%0, %1, 8), %%mm7 \n\t"
208
"movq %%mm5, %%mm1 \n\t"
209
"movq %%mm7, %%mm2 \n\t"
210
"psubusb %%mm7, %%mm5 \n\t"
211
"psubusb %%mm1, %%mm2 \n\t"
212
"por %%mm5, %%mm2 \n\t" // ABS Diff of lines
213
"psubusb %%mm0, %%mm2 \n\t" // diff <= QP -> 0
214
"pcmpeqb %%mm4, %%mm2 \n\t" // diff <= QP -> FF
216
"pand %%mm2, %%mm7 \n\t"
217
"pandn %%mm1, %%mm2 \n\t"
218
"por %%mm2, %%mm7 \n\t" // First Line to Filter
222
// %0 %0+%1 %0+2%1 eax %0+4%1 eax+2%1 ecx eax+4%1
227
"movq (%0, %1), %%mm0 \n\t" // 1
228
"movq %%mm0, %%mm1 \n\t" // 1
229
PAVGB(%%mm6, %%mm0) //1 1 /2
230
PAVGB(%%mm6, %%mm0) //3 1 /4
232
"movq (%0, %1, 4), %%mm2 \n\t" // 1
233
"movq %%mm2, %%mm5 \n\t" // 1
234
PAVGB((%%REGa), %%mm2) // 11 /2
235
PAVGB((%0, %1, 2), %%mm2) // 211 /4
236
"movq %%mm2, %%mm3 \n\t" // 211 /4
237
"movq (%0), %%mm4 \n\t" // 1
238
PAVGB(%%mm4, %%mm3) // 4 211 /8
239
PAVGB(%%mm0, %%mm3) //642211 /16
240
"movq %%mm3, (%0) \n\t" // X
241
// mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9
242
"movq %%mm1, %%mm0 \n\t" // 1
243
PAVGB(%%mm6, %%mm0) //1 1 /2
244
"movq %%mm4, %%mm3 \n\t" // 1
245
PAVGB((%0,%1,2), %%mm3) // 1 1 /2
246
PAVGB((%%REGa,%1,2), %%mm5) // 11 /2
247
PAVGB((%%REGa), %%mm5) // 211 /4
248
PAVGB(%%mm5, %%mm3) // 2 2211 /8
249
PAVGB(%%mm0, %%mm3) //4242211 /16
250
"movq %%mm3, (%0,%1) \n\t" // X
251
// mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9
252
PAVGB(%%mm4, %%mm6) //11 /2
253
"movq (%%"REG_c"), %%mm0 \n\t" // 1
254
PAVGB((%%REGa, %1, 2), %%mm0) // 11/2
255
"movq %%mm0, %%mm3 \n\t" // 11/2
256
PAVGB(%%mm1, %%mm0) // 2 11/4
257
PAVGB(%%mm6, %%mm0) //222 11/8
258
PAVGB(%%mm2, %%mm0) //22242211/16
259
"movq (%0, %1, 2), %%mm2 \n\t" // 1
260
"movq %%mm0, (%0, %1, 2) \n\t" // X
261
// mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9
262
"movq (%%"REG_a", %1, 4), %%mm0 \n\t" // 1
263
PAVGB((%%REGc), %%mm0) // 11 /2
264
PAVGB(%%mm0, %%mm6) //11 11 /4
265
PAVGB(%%mm1, %%mm4) // 11 /2
266
PAVGB(%%mm2, %%mm1) // 11 /2
267
PAVGB(%%mm1, %%mm6) //1122 11 /8
268
PAVGB(%%mm5, %%mm6) //112242211 /16
269
"movq (%%"REG_a"), %%mm5 \n\t" // 1
270
"movq %%mm6, (%%"REG_a") \n\t" // X
271
// mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9
272
"movq (%%"REG_a", %1, 4), %%mm6 \n\t" // 1
273
PAVGB(%%mm7, %%mm6) // 11 /2
274
PAVGB(%%mm4, %%mm6) // 11 11 /4
275
PAVGB(%%mm3, %%mm6) // 11 2211 /8
276
PAVGB(%%mm5, %%mm2) // 11 /2
277
"movq (%0, %1, 4), %%mm4 \n\t" // 1
278
PAVGB(%%mm4, %%mm2) // 112 /4
279
PAVGB(%%mm2, %%mm6) // 112242211 /16
280
"movq %%mm6, (%0, %1, 4) \n\t" // X
281
// mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9
282
PAVGB(%%mm7, %%mm1) // 11 2 /4
283
PAVGB(%%mm4, %%mm5) // 11 /2
284
PAVGB(%%mm5, %%mm0) // 11 11 /4
285
"movq (%%"REG_a", %1, 2), %%mm6 \n\t" // 1
286
PAVGB(%%mm6, %%mm1) // 11 4 2 /8
287
PAVGB(%%mm0, %%mm1) // 11224222 /16
288
"movq %%mm1, (%%"REG_a", %1, 2) \n\t" // X
289
// mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9
290
PAVGB((%%REGc), %%mm2) // 112 4 /8
291
"movq (%%"REG_a", %1, 4), %%mm0 \n\t" // 1
292
PAVGB(%%mm0, %%mm6) // 1 1 /2
293
PAVGB(%%mm7, %%mm6) // 1 12 /4
294
PAVGB(%%mm2, %%mm6) // 1122424 /4
295
"movq %%mm6, (%%"REG_c") \n\t" // X
296
// mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9
297
PAVGB(%%mm7, %%mm5) // 11 2 /4
298
PAVGB(%%mm7, %%mm5) // 11 6 /8
300
PAVGB(%%mm3, %%mm0) // 112 /4
301
PAVGB(%%mm0, %%mm5) // 112246 /16
302
"movq %%mm5, (%%"REG_a", %1, 4) \n\t" // X
306
: "r" (src), "r" ((long)stride), "m" (c->pQPb)
309
#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
310
const int l1= stride;
311
const int l2= stride + l1;
312
const int l3= stride + l2;
313
const int l4= stride + l3;
314
const int l5= stride + l4;
315
const int l6= stride + l5;
316
const int l7= stride + l6;
317
const int l8= stride + l7;
318
const int l9= stride + l8;
321
for(x=0; x<BLOCK_SIZE; x++){
322
const int first= FFABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];
323
const int last= FFABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];
326
sums[0] = 4*first + src[l1] + src[l2] + src[l3] + 4;
327
sums[1] = sums[0] - first + src[l4];
328
sums[2] = sums[1] - first + src[l5];
329
sums[3] = sums[2] - first + src[l6];
330
sums[4] = sums[3] - first + src[l7];
331
sums[5] = sums[4] - src[l1] + src[l8];
332
sums[6] = sums[5] - src[l2] + last;
333
sums[7] = sums[6] - src[l3] + last;
334
sums[8] = sums[7] - src[l4] + last;
335
sums[9] = sums[8] - src[l5] + last;
337
src[l1]= (sums[0] + sums[2] + 2*src[l1])>>4;
338
src[l2]= (sums[1] + sums[3] + 2*src[l2])>>4;
339
src[l3]= (sums[2] + sums[4] + 2*src[l3])>>4;
340
src[l4]= (sums[3] + sums[5] + 2*src[l4])>>4;
341
src[l5]= (sums[4] + sums[6] + 2*src[l5])>>4;
342
src[l6]= (sums[5] + sums[7] + 2*src[l6])>>4;
343
src[l7]= (sums[6] + sums[8] + 2*src[l7])>>4;
344
src[l8]= (sums[7] + sums[9] + 2*src[l8])>>4;
348
#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
350
#endif //HAVE_ALTIVEC
354
* Experimental implementation of the filter (Algorithm 1) described in a paper from Ramkishor & Karandikar
355
* values are correctly clipped (MMX2)
356
* values are wraparound (C)
357
* Conclusion: It is fast, but introduces ugly horizontal patterns
358
* if there is a continuous gradient.
365
static inline void RENAME(vertRK1Filter)(uint8_t *src, int stride, int QP)
367
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
371
"pxor %%mm7, %%mm7 \n\t" // 0
372
"movq "MANGLE(b80)", %%mm6 \n\t" // MIN_SIGNED_BYTE
373
"leal (%0, %1), %%"REG_a" \n\t"
374
"leal (%%"REG_a", %1, 4), %%"REG_c" \n\t"
375
// 0 1 2 3 4 5 6 7 8 9
376
// %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1 %0+8%1 ecx+4%1
377
"movq "MANGLE(pQPb)", %%mm0 \n\t" // QP,..., QP
378
"movq %%mm0, %%mm1 \n\t" // QP,..., QP
379
"paddusb "MANGLE(b02)", %%mm0 \n\t"
380
"psrlw $2, %%mm0 \n\t"
381
"pand "MANGLE(b3F)", %%mm0 \n\t" // QP/4,..., QP/4
382
"paddusb %%mm1, %%mm0 \n\t" // QP*1.25 ...
383
"movq (%0, %1, 4), %%mm2 \n\t" // line 4
384
"movq (%%"REG_c"), %%mm3 \n\t" // line 5
385
"movq %%mm2, %%mm4 \n\t" // line 4
386
"pcmpeqb %%mm5, %%mm5 \n\t" // -1
387
"pxor %%mm2, %%mm5 \n\t" // -line 4 - 1
389
"paddb %%mm6, %%mm5 \n\t" // (l5-l4)/2
390
"psubusb %%mm3, %%mm4 \n\t"
391
"psubusb %%mm2, %%mm3 \n\t"
392
"por %%mm3, %%mm4 \n\t" // |l4 - l5|
393
"psubusb %%mm0, %%mm4 \n\t"
394
"pcmpeqb %%mm7, %%mm4 \n\t"
395
"pand %%mm4, %%mm5 \n\t" // d/2
397
// "paddb %%mm6, %%mm2 \n\t" // line 4 + 0x80
398
"paddb %%mm5, %%mm2 \n\t"
399
// "psubb %%mm6, %%mm2 \n\t"
400
"movq %%mm2, (%0,%1, 4) \n\t"
402
"movq (%%"REG_c"), %%mm2 \n\t"
403
// "paddb %%mm6, %%mm2 \n\t" // line 5 + 0x80
404
"psubb %%mm5, %%mm2 \n\t"
405
// "psubb %%mm6, %%mm2 \n\t"
406
"movq %%mm2, (%%"REG_c") \n\t"
408
"paddb %%mm6, %%mm5 \n\t"
409
"psrlw $2, %%mm5 \n\t"
410
"pand "MANGLE(b3F)", %%mm5 \n\t"
411
"psubb "MANGLE(b20)", %%mm5 \n\t" // (l5-l4)/8
413
"movq (%%"REG_a", %1, 2), %%mm2 \n\t"
414
"paddb %%mm6, %%mm2 \n\t" // line 3 + 0x80
415
"paddsb %%mm5, %%mm2 \n\t"
416
"psubb %%mm6, %%mm2 \n\t"
417
"movq %%mm2, (%%"REG_a", %1, 2) \n\t"
419
"movq (%%"REG_c", %1), %%mm2 \n\t"
420
"paddb %%mm6, %%mm2 \n\t" // line 6 + 0x80
421
"psubsb %%mm5, %%mm2 \n\t"
422
"psubb %%mm6, %%mm2 \n\t"
423
"movq %%mm2, (%%"REG_c", %1) \n\t"
426
: "r" (src), "r" ((long)stride)
429
#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
430
const int l1= stride;
431
const int l2= stride + l1;
432
const int l3= stride + l2;
433
const int l4= stride + l3;
434
const int l5= stride + l4;
435
const int l6= stride + l5;
436
// const int l7= stride + l6;
437
// const int l8= stride + l7;
438
// const int l9= stride + l8;
440
const int QP15= QP + (QP>>2);
442
for(x=0; x<BLOCK_SIZE; x++){
443
const int v = (src[x+l5] - src[x+l4]);
452
#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
457
* Experimental Filter 1
458
* will not damage linear gradients
459
* Flat blocks should look like they were passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
460
* can only smooth blocks at the expected locations (it cannot smooth them if they did move)
461
* MMX2 version does correct clipping C version does not
463
static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)
465
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
469
"pxor %%mm7, %%mm7 \n\t" // 0
470
"lea (%0, %1), %%"REG_a" \n\t"
471
"lea (%%"REG_a", %1, 4), %%"REG_c" \n\t"
472
// 0 1 2 3 4 5 6 7 8 9
473
// %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1 %0+8%1 ecx+4%1
474
"movq (%%"REG_a", %1, 2), %%mm0 \n\t" // line 3
475
"movq (%0, %1, 4), %%mm1 \n\t" // line 4
476
"movq %%mm1, %%mm2 \n\t" // line 4
477
"psubusb %%mm0, %%mm1 \n\t"
478
"psubusb %%mm2, %%mm0 \n\t"
479
"por %%mm1, %%mm0 \n\t" // |l2 - l3|
480
"movq (%%"REG_c"), %%mm3 \n\t" // line 5
481
"movq (%%"REG_c", %1), %%mm4 \n\t" // line 6
482
"movq %%mm3, %%mm5 \n\t" // line 5
483
"psubusb %%mm4, %%mm3 \n\t"
484
"psubusb %%mm5, %%mm4 \n\t"
485
"por %%mm4, %%mm3 \n\t" // |l5 - l6|
486
PAVGB(%%mm3, %%mm0) // (|l2 - l3| + |l5 - l6|)/2
487
"movq %%mm2, %%mm1 \n\t" // line 4
488
"psubusb %%mm5, %%mm2 \n\t"
489
"movq %%mm2, %%mm4 \n\t"
490
"pcmpeqb %%mm7, %%mm2 \n\t" // (l4 - l5) <= 0 ? -1 : 0
491
"psubusb %%mm1, %%mm5 \n\t"
492
"por %%mm5, %%mm4 \n\t" // |l4 - l5|
493
"psubusb %%mm0, %%mm4 \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)
494
"movq %%mm4, %%mm3 \n\t" // d
495
"movq %2, %%mm0 \n\t"
496
"paddusb %%mm0, %%mm0 \n\t"
497
"psubusb %%mm0, %%mm4 \n\t"
498
"pcmpeqb %%mm7, %%mm4 \n\t" // d <= QP ? -1 : 0
499
"psubusb "MANGLE(b01)", %%mm3 \n\t"
500
"pand %%mm4, %%mm3 \n\t" // d <= QP ? d : 0
502
PAVGB(%%mm7, %%mm3) // d/2
503
"movq %%mm3, %%mm1 \n\t" // d/2
504
PAVGB(%%mm7, %%mm3) // d/4
505
PAVGB(%%mm1, %%mm3) // 3*d/8
507
"movq (%0, %1, 4), %%mm0 \n\t" // line 4
508
"pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
509
"psubusb %%mm3, %%mm0 \n\t"
510
"pxor %%mm2, %%mm0 \n\t"
511
"movq %%mm0, (%0, %1, 4) \n\t" // line 4
513
"movq (%%"REG_c"), %%mm0 \n\t" // line 5
514
"pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
515
"paddusb %%mm3, %%mm0 \n\t"
516
"pxor %%mm2, %%mm0 \n\t"
517
"movq %%mm0, (%%"REG_c") \n\t" // line 5
519
PAVGB(%%mm7, %%mm1) // d/4
521
"movq (%%"REG_a", %1, 2), %%mm0 \n\t" // line 3
522
"pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4
523
"psubusb %%mm1, %%mm0 \n\t"
524
"pxor %%mm2, %%mm0 \n\t"
525
"movq %%mm0, (%%"REG_a", %1, 2) \n\t" // line 3
527
"movq (%%"REG_c", %1), %%mm0 \n\t" // line 6
528
"pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5
529
"paddusb %%mm1, %%mm0 \n\t"
530
"pxor %%mm2, %%mm0 \n\t"
531
"movq %%mm0, (%%"REG_c", %1) \n\t" // line 6
533
PAVGB(%%mm7, %%mm1) // d/8
535
"movq (%%"REG_a", %1), %%mm0 \n\t" // line 2
536
"pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2
537
"psubusb %%mm1, %%mm0 \n\t"
538
"pxor %%mm2, %%mm0 \n\t"
539
"movq %%mm0, (%%"REG_a", %1) \n\t" // line 2
541
"movq (%%"REG_c", %1, 2), %%mm0 \n\t" // line 7
542
"pxor %%mm2, %%mm0 \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7
543
"paddusb %%mm1, %%mm0 \n\t"
544
"pxor %%mm2, %%mm0 \n\t"
545
"movq %%mm0, (%%"REG_c", %1, 2) \n\t" // line 7
548
: "r" (src), "r" ((long)stride), "m" (co->pQPb)
551
#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
553
const int l1= stride;
554
const int l2= stride + l1;
555
const int l3= stride + l2;
556
const int l4= stride + l3;
557
const int l5= stride + l4;
558
const int l6= stride + l5;
559
const int l7= stride + l6;
560
// const int l8= stride + l7;
561
// const int l9= stride + l8;
565
for(x=0; x<BLOCK_SIZE; x++){
566
int a= src[l3] - src[l4];
567
int b= src[l4] - src[l5];
568
int c= src[l5] - src[l6];
570
int d= FFABS(b) - ((FFABS(a) + FFABS(c))>>1);
574
int v = d * FFSIGN(-b);
585
#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
589
static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext *c)
591
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
594
const int l1= stride;
595
const int l2= stride + l1;
596
const int l3= stride + l2;
597
const int l4= (int)tmp - (int)src - stride*3;
598
const int l5= (int)tmp - (int)src - stride*3 + 8;
599
const int l6= stride*3 + l3;
600
const int l7= stride + l6;
601
const int l8= stride + l7;
603
memcpy(tmp, src+stride*7, 8);
604
memcpy(tmp+8, src+stride*8, 8);
609
#if 0 //slightly more accurate and slightly slower
610
"pxor %%mm7, %%mm7 \n\t" // 0
611
"lea (%0, %1), %%"REG_a" \n\t"
612
"lea (%%"REG_a", %1, 4), %%"REG_c" \n\t"
614
// %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 ecx+%1 ecx+2%1
615
// %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1
618
"movq (%0, %1, 2), %%mm0 \n\t" // l2
619
"movq (%0), %%mm1 \n\t" // l0
620
"movq %%mm0, %%mm2 \n\t" // l2
621
PAVGB(%%mm7, %%mm0) // ~l2/2
622
PAVGB(%%mm1, %%mm0) // ~(l2 + 2l0)/4
623
PAVGB(%%mm2, %%mm0) // ~(5l2 + 2l0)/8
625
"movq (%%"REG_a"), %%mm1 \n\t" // l1
626
"movq (%%"REG_a", %1, 2), %%mm3 \n\t" // l3
627
"movq %%mm1, %%mm4 \n\t" // l1
628
PAVGB(%%mm7, %%mm1) // ~l1/2
629
PAVGB(%%mm3, %%mm1) // ~(l1 + 2l3)/4
630
PAVGB(%%mm4, %%mm1) // ~(5l1 + 2l3)/8
632
"movq %%mm0, %%mm4 \n\t" // ~(5l2 + 2l0)/8
633
"psubusb %%mm1, %%mm0 \n\t"
634
"psubusb %%mm4, %%mm1 \n\t"
635
"por %%mm0, %%mm1 \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8
636
// mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0
638
"movq (%0, %1, 4), %%mm0 \n\t" // l4
639
"movq %%mm0, %%mm4 \n\t" // l4
640
PAVGB(%%mm7, %%mm0) // ~l4/2
641
PAVGB(%%mm2, %%mm0) // ~(l4 + 2l2)/4
642
PAVGB(%%mm4, %%mm0) // ~(5l4 + 2l2)/8
644
"movq (%%"REG_c"), %%mm2 \n\t" // l5
645
"movq %%mm3, %%mm5 \n\t" // l3
646
PAVGB(%%mm7, %%mm3) // ~l3/2
647
PAVGB(%%mm2, %%mm3) // ~(l3 + 2l5)/4
648
PAVGB(%%mm5, %%mm3) // ~(5l3 + 2l5)/8
650
"movq %%mm0, %%mm6 \n\t" // ~(5l4 + 2l2)/8
651
"psubusb %%mm3, %%mm0 \n\t"
652
"psubusb %%mm6, %%mm3 \n\t"
653
"por %%mm0, %%mm3 \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8
654
"pcmpeqb %%mm7, %%mm0 \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)
655
// mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0
657
"movq (%%"REG_c", %1), %%mm6 \n\t" // l6
658
"movq %%mm6, %%mm5 \n\t" // l6
659
PAVGB(%%mm7, %%mm6) // ~l6/2
660
PAVGB(%%mm4, %%mm6) // ~(l6 + 2l4)/4
661
PAVGB(%%mm5, %%mm6) // ~(5l6 + 2l4)/8
663
"movq (%%"REG_c", %1, 2), %%mm5 \n\t" // l7
664
"movq %%mm2, %%mm4 \n\t" // l5
665
PAVGB(%%mm7, %%mm2) // ~l5/2
666
PAVGB(%%mm5, %%mm2) // ~(l5 + 2l7)/4
667
PAVGB(%%mm4, %%mm2) // ~(5l5 + 2l7)/8
669
"movq %%mm6, %%mm4 \n\t" // ~(5l6 + 2l4)/8
670
"psubusb %%mm2, %%mm6 \n\t"
671
"psubusb %%mm4, %%mm2 \n\t"
672
"por %%mm6, %%mm2 \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8
673
// mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0
676
PMINUB(%%mm2, %%mm1, %%mm4) // MIN(|lenergy|,|renergy|)/8
677
"movq %2, %%mm4 \n\t" // QP //FIXME QP+1 ?
678
"paddusb "MANGLE(b01)", %%mm4 \n\t"
679
"pcmpgtb %%mm3, %%mm4 \n\t" // |menergy|/8 < QP
680
"psubusb %%mm1, %%mm3 \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8
681
"pand %%mm4, %%mm3 \n\t"
683
"movq %%mm3, %%mm1 \n\t"
684
// "psubusb "MANGLE(b01)", %%mm3 \n\t"
687
"paddusb %%mm1, %%mm3 \n\t"
688
// "paddusb "MANGLE(b01)", %%mm3 \n\t"
690
"movq (%%"REG_a", %1, 2), %%mm6 \n\t" //l3
691
"movq (%0, %1, 4), %%mm5 \n\t" //l4
692
"movq (%0, %1, 4), %%mm4 \n\t" //l4
693
"psubusb %%mm6, %%mm5 \n\t"
694
"psubusb %%mm4, %%mm6 \n\t"
695
"por %%mm6, %%mm5 \n\t" // |l3-l4|
696
"pcmpeqb %%mm7, %%mm6 \n\t" // SIGN(l3-l4)
697
"pxor %%mm6, %%mm0 \n\t"
698
"pand %%mm0, %%mm3 \n\t"
699
PMINUB(%%mm5, %%mm3, %%mm0)
701
"psubusb "MANGLE(b01)", %%mm3 \n\t"
704
"movq (%%"REG_a", %1, 2), %%mm0 \n\t"
705
"movq (%0, %1, 4), %%mm2 \n\t"
706
"pxor %%mm6, %%mm0 \n\t"
707
"pxor %%mm6, %%mm2 \n\t"
708
"psubb %%mm3, %%mm0 \n\t"
709
"paddb %%mm3, %%mm2 \n\t"
710
"pxor %%mm6, %%mm0 \n\t"
711
"pxor %%mm6, %%mm2 \n\t"
712
"movq %%mm0, (%%"REG_a", %1, 2) \n\t"
713
"movq %%mm2, (%0, %1, 4) \n\t"
716
"lea (%0, %1), %%"REG_a" \n\t"
717
"pcmpeqb %%mm6, %%mm6 \n\t" // -1
719
// %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 ecx+%1 ecx+2%1
720
// %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1
723
"movq (%%"REG_a", %1, 2), %%mm1 \n\t" // l3
724
"movq (%0, %1, 4), %%mm0 \n\t" // l4
725
"pxor %%mm6, %%mm1 \n\t" // -l3-1
726
PAVGB(%%mm1, %%mm0) // -q+128 = (l4-l3+256)/2
727
// mm1=-l3-1, mm0=128-q
729
"movq (%%"REG_a", %1, 4), %%mm2 \n\t" // l5
730
"movq (%%"REG_a", %1), %%mm3 \n\t" // l2
731
"pxor %%mm6, %%mm2 \n\t" // -l5-1
732
"movq %%mm2, %%mm5 \n\t" // -l5-1
733
"movq "MANGLE(b80)", %%mm4 \n\t" // 128
734
"lea (%%"REG_a", %1, 4), %%"REG_c" \n\t"
735
PAVGB(%%mm3, %%mm2) // (l2-l5+256)/2
736
PAVGB(%%mm0, %%mm4) // ~(l4-l3)/4 + 128
737
PAVGB(%%mm2, %%mm4) // ~(l2-l5)/4 +(l4-l3)/8 + 128
738
PAVGB(%%mm0, %%mm4) // ~(l2-l5)/8 +5(l4-l3)/16 + 128
739
// mm1=-l3-1, mm0=128-q, mm3=l2, mm4=menergy/16 + 128, mm5= -l5-1
741
"movq (%%"REG_a"), %%mm2 \n\t" // l1
742
"pxor %%mm6, %%mm2 \n\t" // -l1-1
743
PAVGB(%%mm3, %%mm2) // (l2-l1+256)/2
744
PAVGB((%0), %%mm1) // (l0-l3+256)/2
745
"movq "MANGLE(b80)", %%mm3 \n\t" // 128
746
PAVGB(%%mm2, %%mm3) // ~(l2-l1)/4 + 128
747
PAVGB(%%mm1, %%mm3) // ~(l0-l3)/4 +(l2-l1)/8 + 128
748
PAVGB(%%mm2, %%mm3) // ~(l0-l3)/8 +5(l2-l1)/16 + 128
749
// mm0=128-q, mm3=lenergy/16 + 128, mm4= menergy/16 + 128, mm5= -l5-1
751
PAVGB((%%REGc, %1), %%mm5) // (l6-l5+256)/2
752
"movq (%%"REG_c", %1, 2), %%mm1 \n\t" // l7
753
"pxor %%mm6, %%mm1 \n\t" // -l7-1
754
PAVGB((%0, %1, 4), %%mm1) // (l4-l7+256)/2
755
"movq "MANGLE(b80)", %%mm2 \n\t" // 128
756
PAVGB(%%mm5, %%mm2) // ~(l6-l5)/4 + 128
757
PAVGB(%%mm1, %%mm2) // ~(l4-l7)/4 +(l6-l5)/8 + 128
758
PAVGB(%%mm5, %%mm2) // ~(l4-l7)/8 +5(l6-l5)/16 + 128
759
// mm0=128-q, mm2=renergy/16 + 128, mm3=lenergy/16 + 128, mm4= menergy/16 + 128
761
"movq "MANGLE(b00)", %%mm1 \n\t" // 0
762
"movq "MANGLE(b00)", %%mm5 \n\t" // 0
763
"psubb %%mm2, %%mm1 \n\t" // 128 - renergy/16
764
"psubb %%mm3, %%mm5 \n\t" // 128 - lenergy/16
765
PMAXUB(%%mm1, %%mm2) // 128 + |renergy/16|
766
PMAXUB(%%mm5, %%mm3) // 128 + |lenergy/16|
767
PMINUB(%%mm2, %%mm3, %%mm1) // 128 + MIN(|lenergy|,|renergy|)/16
769
// mm0=128-q, mm3=128 + MIN(|lenergy|,|renergy|)/16, mm4= menergy/16 + 128
771
"movq "MANGLE(b00)", %%mm7 \n\t" // 0
772
"movq %2, %%mm2 \n\t" // QP
773
PAVGB(%%mm6, %%mm2) // 128 + QP/2
774
"psubb %%mm6, %%mm2 \n\t"
776
"movq %%mm4, %%mm1 \n\t"
777
"pcmpgtb %%mm7, %%mm1 \n\t" // SIGN(menergy)
778
"pxor %%mm1, %%mm4 \n\t"
779
"psubb %%mm1, %%mm4 \n\t" // 128 + |menergy|/16
780
"pcmpgtb %%mm4, %%mm2 \n\t" // |menergy|/16 < QP/2
781
"psubusb %%mm3, %%mm4 \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16
782
// mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16
784
"movq %%mm4, %%mm3 \n\t" // d
785
"psubusb "MANGLE(b01)", %%mm4 \n\t"
786
PAVGB(%%mm7, %%mm4) // d/32
787
PAVGB(%%mm7, %%mm4) // (d + 32)/64
788
"paddb %%mm3, %%mm4 \n\t" // 5d/64
789
"pand %%mm2, %%mm4 \n\t"
791
"movq "MANGLE(b80)", %%mm5 \n\t" // 128
792
"psubb %%mm0, %%mm5 \n\t" // q
793
"paddsb %%mm6, %%mm5 \n\t" // fix bad rounding
794
"pcmpgtb %%mm5, %%mm7 \n\t" // SIGN(q)
795
"pxor %%mm7, %%mm5 \n\t"
797
PMINUB(%%mm5, %%mm4, %%mm3) // MIN(|q|, 5d/64)
798
"pxor %%mm1, %%mm7 \n\t" // SIGN(d*q)
800
"pand %%mm7, %%mm4 \n\t"
801
"movq (%%"REG_a", %1, 2), %%mm0 \n\t"
802
"movq (%0, %1, 4), %%mm2 \n\t"
803
"pxor %%mm1, %%mm0 \n\t"
804
"pxor %%mm1, %%mm2 \n\t"
805
"paddb %%mm4, %%mm0 \n\t"
806
"psubb %%mm4, %%mm2 \n\t"
807
"pxor %%mm1, %%mm0 \n\t"
808
"pxor %%mm1, %%mm2 \n\t"
809
"movq %%mm0, (%%"REG_a", %1, 2) \n\t"
810
"movq %%mm2, (%0, %1, 4) \n\t"
813
: "r" (src), "r" ((long)stride), "m" (c->pQPb)
821
for(x=0; x<BLOCK_SIZE; x++){
822
const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
823
if(FFABS(middleEnergy)< 8*QP){
824
const int q=(src[l4] - src[l5])/2;
825
const int leftEnergy= 5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
826
const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
828
int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
832
d*= FFSIGN(-middleEnergy);
851
int d= src[x+y*stride] - tmp[x+(y-4)*8];
861
src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;
865
if(num%1000000 == 0){
866
av_log(c, AV_LOG_INFO, " %d %d %d %d\n", num, sum, max, bias);
872
#elif defined (HAVE_MMX)
875
"pxor %%mm7, %%mm7 \n\t"
876
"lea -40(%%"REG_SP"), %%"REG_c" \n\t" // make space for 4 8-byte vars
877
"and "ALIGN_MASK", %%"REG_c" \n\t" // align
879
// %0 %0+%1 %0+2%1 eax+2%1 %0+4%1 eax+4%1 edx+%1 edx+2%1
880
// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1
882
"movq (%0), %%mm0 \n\t"
883
"movq %%mm0, %%mm1 \n\t"
884
"punpcklbw %%mm7, %%mm0 \n\t" // low part of line 0
885
"punpckhbw %%mm7, %%mm1 \n\t" // high part of line 0
887
"movq (%0, %1), %%mm2 \n\t"
888
"lea (%0, %1, 2), %%"REG_a" \n\t"
889
"movq %%mm2, %%mm3 \n\t"
890
"punpcklbw %%mm7, %%mm2 \n\t" // low part of line 1
891
"punpckhbw %%mm7, %%mm3 \n\t" // high part of line 1
893
"movq (%%"REG_a"), %%mm4 \n\t"
894
"movq %%mm4, %%mm5 \n\t"
895
"punpcklbw %%mm7, %%mm4 \n\t" // low part of line 2
896
"punpckhbw %%mm7, %%mm5 \n\t" // high part of line 2
898
"paddw %%mm0, %%mm0 \n\t" // 2L0
899
"paddw %%mm1, %%mm1 \n\t" // 2H0
900
"psubw %%mm4, %%mm2 \n\t" // L1 - L2
901
"psubw %%mm5, %%mm3 \n\t" // H1 - H2
902
"psubw %%mm2, %%mm0 \n\t" // 2L0 - L1 + L2
903
"psubw %%mm3, %%mm1 \n\t" // 2H0 - H1 + H2
905
"psllw $2, %%mm2 \n\t" // 4L1 - 4L2
906
"psllw $2, %%mm3 \n\t" // 4H1 - 4H2
907
"psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2
908
"psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2
910
"movq (%%"REG_a", %1), %%mm2 \n\t"
911
"movq %%mm2, %%mm3 \n\t"
912
"punpcklbw %%mm7, %%mm2 \n\t" // L3
913
"punpckhbw %%mm7, %%mm3 \n\t" // H3
915
"psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - L3
916
"psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - H3
917
"psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
918
"psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
919
"movq %%mm0, (%%"REG_c") \n\t" // 2L0 - 5L1 + 5L2 - 2L3
920
"movq %%mm1, 8(%%"REG_c") \n\t" // 2H0 - 5H1 + 5H2 - 2H3
922
"movq (%%"REG_a", %1, 2), %%mm0 \n\t"
923
"movq %%mm0, %%mm1 \n\t"
924
"punpcklbw %%mm7, %%mm0 \n\t" // L4
925
"punpckhbw %%mm7, %%mm1 \n\t" // H4
927
"psubw %%mm0, %%mm2 \n\t" // L3 - L4
928
"psubw %%mm1, %%mm3 \n\t" // H3 - H4
929
"movq %%mm2, 16(%%"REG_c") \n\t" // L3 - L4
930
"movq %%mm3, 24(%%"REG_c") \n\t" // H3 - H4
931
"paddw %%mm4, %%mm4 \n\t" // 2L2
932
"paddw %%mm5, %%mm5 \n\t" // 2H2
933
"psubw %%mm2, %%mm4 \n\t" // 2L2 - L3 + L4
934
"psubw %%mm3, %%mm5 \n\t" // 2H2 - H3 + H4
936
"lea (%%"REG_a", %1), %0 \n\t"
937
"psllw $2, %%mm2 \n\t" // 4L3 - 4L4
938
"psllw $2, %%mm3 \n\t" // 4H3 - 4H4
939
"psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4
940
"psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4
942
"movq (%0, %1, 2), %%mm2 \n\t"
943
"movq %%mm2, %%mm3 \n\t"
944
"punpcklbw %%mm7, %%mm2 \n\t" // L5
945
"punpckhbw %%mm7, %%mm3 \n\t" // H5
946
"psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - L5
947
"psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - H5
948
"psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - 2L5
949
"psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - 2H5
951
"movq (%%"REG_a", %1, 4), %%mm6 \n\t"
952
"punpcklbw %%mm7, %%mm6 \n\t" // L6
953
"psubw %%mm6, %%mm2 \n\t" // L5 - L6
954
"movq (%%"REG_a", %1, 4), %%mm6 \n\t"
955
"punpckhbw %%mm7, %%mm6 \n\t" // H6
956
"psubw %%mm6, %%mm3 \n\t" // H5 - H6
958
"paddw %%mm0, %%mm0 \n\t" // 2L4
959
"paddw %%mm1, %%mm1 \n\t" // 2H4
960
"psubw %%mm2, %%mm0 \n\t" // 2L4 - L5 + L6
961
"psubw %%mm3, %%mm1 \n\t" // 2H4 - H5 + H6
963
"psllw $2, %%mm2 \n\t" // 4L5 - 4L6
964
"psllw $2, %%mm3 \n\t" // 4H5 - 4H6
965
"psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6
966
"psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6
968
"movq (%0, %1, 4), %%mm2 \n\t"
969
"movq %%mm2, %%mm3 \n\t"
970
"punpcklbw %%mm7, %%mm2 \n\t" // L7
971
"punpckhbw %%mm7, %%mm3 \n\t" // H7
973
"paddw %%mm2, %%mm2 \n\t" // 2L7
974
"paddw %%mm3, %%mm3 \n\t" // 2H7
975
"psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6 - 2L7
976
"psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6 - 2H7
978
"movq (%%"REG_c"), %%mm2 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
979
"movq 8(%%"REG_c"), %%mm3 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
982
"movq %%mm7, %%mm6 \n\t" // 0
983
"psubw %%mm0, %%mm6 \n\t"
984
"pmaxsw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
985
"movq %%mm7, %%mm6 \n\t" // 0
986
"psubw %%mm1, %%mm6 \n\t"
987
"pmaxsw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
988
"movq %%mm7, %%mm6 \n\t" // 0
989
"psubw %%mm2, %%mm6 \n\t"
990
"pmaxsw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
991
"movq %%mm7, %%mm6 \n\t" // 0
992
"psubw %%mm3, %%mm6 \n\t"
993
"pmaxsw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
995
"movq %%mm7, %%mm6 \n\t" // 0
996
"pcmpgtw %%mm0, %%mm6 \n\t"
997
"pxor %%mm6, %%mm0 \n\t"
998
"psubw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
999
"movq %%mm7, %%mm6 \n\t" // 0
1000
"pcmpgtw %%mm1, %%mm6 \n\t"
1001
"pxor %%mm6, %%mm1 \n\t"
1002
"psubw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
1003
"movq %%mm7, %%mm6 \n\t" // 0
1004
"pcmpgtw %%mm2, %%mm6 \n\t"
1005
"pxor %%mm6, %%mm2 \n\t"
1006
"psubw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
1007
"movq %%mm7, %%mm6 \n\t" // 0
1008
"pcmpgtw %%mm3, %%mm6 \n\t"
1009
"pxor %%mm6, %%mm3 \n\t"
1010
"psubw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
1014
"pminsw %%mm2, %%mm0 \n\t"
1015
"pminsw %%mm3, %%mm1 \n\t"
1017
"movq %%mm0, %%mm6 \n\t"
1018
"psubusw %%mm2, %%mm6 \n\t"
1019
"psubw %%mm6, %%mm0 \n\t"
1020
"movq %%mm1, %%mm6 \n\t"
1021
"psubusw %%mm3, %%mm6 \n\t"
1022
"psubw %%mm6, %%mm1 \n\t"
1025
"movd %2, %%mm2 \n\t" // QP
1026
"punpcklbw %%mm7, %%mm2 \n\t"
1028
"movq %%mm7, %%mm6 \n\t" // 0
1029
"pcmpgtw %%mm4, %%mm6 \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
1030
"pxor %%mm6, %%mm4 \n\t"
1031
"psubw %%mm6, %%mm4 \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
1032
"pcmpgtw %%mm5, %%mm7 \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
1033
"pxor %%mm7, %%mm5 \n\t"
1034
"psubw %%mm7, %%mm5 \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
1036
"psllw $3, %%mm2 \n\t" // 8QP
1037
"movq %%mm2, %%mm3 \n\t" // 8QP
1038
"pcmpgtw %%mm4, %%mm2 \n\t"
1039
"pcmpgtw %%mm5, %%mm3 \n\t"
1040
"pand %%mm2, %%mm4 \n\t"
1041
"pand %%mm3, %%mm5 \n\t"
1044
"psubusw %%mm0, %%mm4 \n\t" // hd
1045
"psubusw %%mm1, %%mm5 \n\t" // ld
1048
"movq "MANGLE(w05)", %%mm2 \n\t" // 5
1049
"pmullw %%mm2, %%mm4 \n\t"
1050
"pmullw %%mm2, %%mm5 \n\t"
1051
"movq "MANGLE(w20)", %%mm2 \n\t" // 32
1052
"paddw %%mm2, %%mm4 \n\t"
1053
"paddw %%mm2, %%mm5 \n\t"
1054
"psrlw $6, %%mm4 \n\t"
1055
"psrlw $6, %%mm5 \n\t"
1057
"movq 16(%%"REG_c"), %%mm0 \n\t" // L3 - L4
1058
"movq 24(%%"REG_c"), %%mm1 \n\t" // H3 - H4
1060
"pxor %%mm2, %%mm2 \n\t"
1061
"pxor %%mm3, %%mm3 \n\t"
1063
"pcmpgtw %%mm0, %%mm2 \n\t" // sign (L3-L4)
1064
"pcmpgtw %%mm1, %%mm3 \n\t" // sign (H3-H4)
1065
"pxor %%mm2, %%mm0 \n\t"
1066
"pxor %%mm3, %%mm1 \n\t"
1067
"psubw %%mm2, %%mm0 \n\t" // |L3-L4|
1068
"psubw %%mm3, %%mm1 \n\t" // |H3-H4|
1069
"psrlw $1, %%mm0 \n\t" // |L3 - L4|/2
1070
"psrlw $1, %%mm1 \n\t" // |H3 - H4|/2
1072
"pxor %%mm6, %%mm2 \n\t"
1073
"pxor %%mm7, %%mm3 \n\t"
1074
"pand %%mm2, %%mm4 \n\t"
1075
"pand %%mm3, %%mm5 \n\t"
1078
"pminsw %%mm0, %%mm4 \n\t"
1079
"pminsw %%mm1, %%mm5 \n\t"
1081
"movq %%mm4, %%mm2 \n\t"
1082
"psubusw %%mm0, %%mm2 \n\t"
1083
"psubw %%mm2, %%mm4 \n\t"
1084
"movq %%mm5, %%mm2 \n\t"
1085
"psubusw %%mm1, %%mm2 \n\t"
1086
"psubw %%mm2, %%mm5 \n\t"
1088
"pxor %%mm6, %%mm4 \n\t"
1089
"pxor %%mm7, %%mm5 \n\t"
1090
"psubw %%mm6, %%mm4 \n\t"
1091
"psubw %%mm7, %%mm5 \n\t"
1092
"packsswb %%mm5, %%mm4 \n\t"
1093
"movq (%0), %%mm0 \n\t"
1094
"paddb %%mm4, %%mm0 \n\t"
1095
"movq %%mm0, (%0) \n\t"
1096
"movq (%0, %1), %%mm0 \n\t"
1097
"psubb %%mm4, %%mm0 \n\t"
1098
"movq %%mm0, (%0, %1) \n\t"
1101
: "r" ((long)stride), "m" (c->pQPb)
1102
: "%"REG_a, "%"REG_c
1104
#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1105
const int l1= stride;
1106
const int l2= stride + l1;
1107
const int l3= stride + l2;
1108
const int l4= stride + l3;
1109
const int l5= stride + l4;
1110
const int l6= stride + l5;
1111
const int l7= stride + l6;
1112
const int l8= stride + l7;
1113
// const int l9= stride + l8;
1116
for(x=0; x<BLOCK_SIZE; x++){
1117
const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);
1118
if(FFABS(middleEnergy) < 8*c->QP){
1119
const int q=(src[l4] - src[l5])/2;
1120
const int leftEnergy= 5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);
1121
const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);
1123
int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
1127
d*= FFSIGN(-middleEnergy);
1142
#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1144
#endif //HAVE_ALTIVEC
1146
#ifndef HAVE_ALTIVEC
1147
static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)
1149
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1151
"pxor %%mm6, %%mm6 \n\t"
1152
"pcmpeqb %%mm7, %%mm7 \n\t"
1153
"movq %2, %%mm0 \n\t"
1154
"punpcklbw %%mm6, %%mm0 \n\t"
1155
"psrlw $1, %%mm0 \n\t"
1156
"psubw %%mm7, %%mm0 \n\t"
1157
"packuswb %%mm0, %%mm0 \n\t"
1158
"movq %%mm0, %3 \n\t"
1160
"lea (%0, %1), %%"REG_a" \n\t"
1161
"lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1163
// 0 1 2 3 4 5 6 7 8 9
1164
// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1168
#define REAL_FIND_MIN_MAX(addr)\
1169
"movq " #addr ", %%mm0 \n\t"\
1170
"pminub %%mm0, %%mm7 \n\t"\
1171
"pmaxub %%mm0, %%mm6 \n\t"
1173
#define REAL_FIND_MIN_MAX(addr)\
1174
"movq " #addr ", %%mm0 \n\t"\
1175
"movq %%mm7, %%mm1 \n\t"\
1176
"psubusb %%mm0, %%mm6 \n\t"\
1177
"paddb %%mm0, %%mm6 \n\t"\
1178
"psubusb %%mm0, %%mm1 \n\t"\
1179
"psubb %%mm1, %%mm7 \n\t"
1181
#define FIND_MIN_MAX(addr) REAL_FIND_MIN_MAX(addr)
1183
FIND_MIN_MAX((%%REGa))
1184
FIND_MIN_MAX((%%REGa, %1))
1185
FIND_MIN_MAX((%%REGa, %1, 2))
1186
FIND_MIN_MAX((%0, %1, 4))
1187
FIND_MIN_MAX((%%REGd))
1188
FIND_MIN_MAX((%%REGd, %1))
1189
FIND_MIN_MAX((%%REGd, %1, 2))
1190
FIND_MIN_MAX((%0, %1, 8))
1192
"movq %%mm7, %%mm4 \n\t"
1193
"psrlq $8, %%mm7 \n\t"
1195
"pminub %%mm4, %%mm7 \n\t" // min of pixels
1196
"pshufw $0xF9, %%mm7, %%mm4 \n\t"
1197
"pminub %%mm4, %%mm7 \n\t" // min of pixels
1198
"pshufw $0xFE, %%mm7, %%mm4 \n\t"
1199
"pminub %%mm4, %%mm7 \n\t"
1201
"movq %%mm7, %%mm1 \n\t"
1202
"psubusb %%mm4, %%mm1 \n\t"
1203
"psubb %%mm1, %%mm7 \n\t"
1204
"movq %%mm7, %%mm4 \n\t"
1205
"psrlq $16, %%mm7 \n\t"
1206
"movq %%mm7, %%mm1 \n\t"
1207
"psubusb %%mm4, %%mm1 \n\t"
1208
"psubb %%mm1, %%mm7 \n\t"
1209
"movq %%mm7, %%mm4 \n\t"
1210
"psrlq $32, %%mm7 \n\t"
1211
"movq %%mm7, %%mm1 \n\t"
1212
"psubusb %%mm4, %%mm1 \n\t"
1213
"psubb %%mm1, %%mm7 \n\t"
1217
"movq %%mm6, %%mm4 \n\t"
1218
"psrlq $8, %%mm6 \n\t"
1220
"pmaxub %%mm4, %%mm6 \n\t" // max of pixels
1221
"pshufw $0xF9, %%mm6, %%mm4 \n\t"
1222
"pmaxub %%mm4, %%mm6 \n\t"
1223
"pshufw $0xFE, %%mm6, %%mm4 \n\t"
1224
"pmaxub %%mm4, %%mm6 \n\t"
1226
"psubusb %%mm4, %%mm6 \n\t"
1227
"paddb %%mm4, %%mm6 \n\t"
1228
"movq %%mm6, %%mm4 \n\t"
1229
"psrlq $16, %%mm6 \n\t"
1230
"psubusb %%mm4, %%mm6 \n\t"
1231
"paddb %%mm4, %%mm6 \n\t"
1232
"movq %%mm6, %%mm4 \n\t"
1233
"psrlq $32, %%mm6 \n\t"
1234
"psubusb %%mm4, %%mm6 \n\t"
1235
"paddb %%mm4, %%mm6 \n\t"
1237
"movq %%mm6, %%mm0 \n\t" // max
1238
"psubb %%mm7, %%mm6 \n\t" // max - min
1239
"movd %%mm6, %%ecx \n\t"
1240
"cmpb "MANGLE(deringThreshold)", %%cl \n\t"
1242
"lea -24(%%"REG_SP"), %%"REG_c" \n\t"
1243
"and "ALIGN_MASK", %%"REG_c" \n\t"
1244
PAVGB(%%mm0, %%mm7) // a=(max + min)/2
1245
"punpcklbw %%mm7, %%mm7 \n\t"
1246
"punpcklbw %%mm7, %%mm7 \n\t"
1247
"punpcklbw %%mm7, %%mm7 \n\t"
1248
"movq %%mm7, (%%"REG_c") \n\t"
1250
"movq (%0), %%mm0 \n\t" // L10
1251
"movq %%mm0, %%mm1 \n\t" // L10
1252
"movq %%mm0, %%mm2 \n\t" // L10
1253
"psllq $8, %%mm1 \n\t"
1254
"psrlq $8, %%mm2 \n\t"
1255
"movd -4(%0), %%mm3 \n\t"
1256
"movd 8(%0), %%mm4 \n\t"
1257
"psrlq $24, %%mm3 \n\t"
1258
"psllq $56, %%mm4 \n\t"
1259
"por %%mm3, %%mm1 \n\t" // L00
1260
"por %%mm4, %%mm2 \n\t" // L20
1261
"movq %%mm1, %%mm3 \n\t" // L00
1262
PAVGB(%%mm2, %%mm1) // (L20 + L00)/2
1263
PAVGB(%%mm0, %%mm1) // (L20 + L00 + 2L10)/4
1264
"psubusb %%mm7, %%mm0 \n\t"
1265
"psubusb %%mm7, %%mm2 \n\t"
1266
"psubusb %%mm7, %%mm3 \n\t"
1267
"pcmpeqb "MANGLE(b00)", %%mm0 \n\t" // L10 > a ? 0 : -1
1268
"pcmpeqb "MANGLE(b00)", %%mm2 \n\t" // L20 > a ? 0 : -1
1269
"pcmpeqb "MANGLE(b00)", %%mm3 \n\t" // L00 > a ? 0 : -1
1270
"paddb %%mm2, %%mm0 \n\t"
1271
"paddb %%mm3, %%mm0 \n\t"
1273
"movq (%%"REG_a"), %%mm2 \n\t" // L11
1274
"movq %%mm2, %%mm3 \n\t" // L11
1275
"movq %%mm2, %%mm4 \n\t" // L11
1276
"psllq $8, %%mm3 \n\t"
1277
"psrlq $8, %%mm4 \n\t"
1278
"movd -4(%%"REG_a"), %%mm5 \n\t"
1279
"movd 8(%%"REG_a"), %%mm6 \n\t"
1280
"psrlq $24, %%mm5 \n\t"
1281
"psllq $56, %%mm6 \n\t"
1282
"por %%mm5, %%mm3 \n\t" // L01
1283
"por %%mm6, %%mm4 \n\t" // L21
1284
"movq %%mm3, %%mm5 \n\t" // L01
1285
PAVGB(%%mm4, %%mm3) // (L21 + L01)/2
1286
PAVGB(%%mm2, %%mm3) // (L21 + L01 + 2L11)/4
1287
"psubusb %%mm7, %%mm2 \n\t"
1288
"psubusb %%mm7, %%mm4 \n\t"
1289
"psubusb %%mm7, %%mm5 \n\t"
1290
"pcmpeqb "MANGLE(b00)", %%mm2 \n\t" // L11 > a ? 0 : -1
1291
"pcmpeqb "MANGLE(b00)", %%mm4 \n\t" // L21 > a ? 0 : -1
1292
"pcmpeqb "MANGLE(b00)", %%mm5 \n\t" // L01 > a ? 0 : -1
1293
"paddb %%mm4, %%mm2 \n\t"
1294
"paddb %%mm5, %%mm2 \n\t"
1296
#define REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
1297
"movq " #src ", " #sx " \n\t" /* src[0] */\
1298
"movq " #sx ", " #lx " \n\t" /* src[0] */\
1299
"movq " #sx ", " #t0 " \n\t" /* src[0] */\
1300
"psllq $8, " #lx " \n\t"\
1301
"psrlq $8, " #t0 " \n\t"\
1302
"movd -4" #src ", " #t1 " \n\t"\
1303
"psrlq $24, " #t1 " \n\t"\
1304
"por " #t1 ", " #lx " \n\t" /* src[-1] */\
1305
"movd 8" #src ", " #t1 " \n\t"\
1306
"psllq $56, " #t1 " \n\t"\
1307
"por " #t1 ", " #t0 " \n\t" /* src[+1] */\
1308
"movq " #lx ", " #t1 " \n\t" /* src[-1] */\
1309
PAVGB(t0, lx) /* (src[-1] + src[+1])/2 */\
1310
PAVGB(sx, lx) /* (src[-1] + 2src[0] + src[+1])/4 */\
1312
"movq " #lx ", 8(%%"REG_c") \n\t"\
1313
"movq (%%"REG_c"), " #lx " \n\t"\
1314
"psubusb " #lx ", " #t1 " \n\t"\
1315
"psubusb " #lx ", " #t0 " \n\t"\
1316
"psubusb " #lx ", " #sx " \n\t"\
1317
"movq "MANGLE(b00)", " #lx " \n\t"\
1318
"pcmpeqb " #lx ", " #t1 " \n\t" /* src[-1] > a ? 0 : -1*/\
1319
"pcmpeqb " #lx ", " #t0 " \n\t" /* src[+1] > a ? 0 : -1*/\
1320
"pcmpeqb " #lx ", " #sx " \n\t" /* src[0] > a ? 0 : -1*/\
1321
"paddb " #t1 ", " #t0 " \n\t"\
1322
"paddb " #t0 ", " #sx " \n\t"\
1324
PAVGB(plx, pplx) /* filtered */\
1325
"movq " #dst ", " #t0 " \n\t" /* dst */\
1326
"movq " #t0 ", " #t1 " \n\t" /* dst */\
1327
"psubusb %3, " #t0 " \n\t"\
1328
"paddusb %3, " #t1 " \n\t"\
1330
PMINUB(t1, pplx, t0)\
1331
"paddb " #sx ", " #ppsx " \n\t"\
1332
"paddb " #psx ", " #ppsx " \n\t"\
1333
"#paddb "MANGLE(b02)", " #ppsx " \n\t"\
1334
"pand "MANGLE(b08)", " #ppsx " \n\t"\
1335
"pcmpeqb " #lx ", " #ppsx " \n\t"\
1336
"pand " #ppsx ", " #pplx " \n\t"\
1337
"pandn " #dst ", " #ppsx " \n\t"\
1338
"por " #pplx ", " #ppsx " \n\t"\
1339
"movq " #ppsx ", " #dst " \n\t"\
1340
"movq 8(%%"REG_c"), " #lx " \n\t"
1342
#define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \
1343
REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1)
1359
//DERING_CORE(dst ,src ,ppsx ,psx ,sx ,pplx ,plx ,lx ,t0 ,t1)
1360
DERING_CORE((%%REGa) ,(%%REGa, %1) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1361
DERING_CORE((%%REGa, %1) ,(%%REGa, %1, 2),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1362
DERING_CORE((%%REGa, %1, 2),(%0, %1, 4) ,%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
1363
DERING_CORE((%0, %1, 4) ,(%%REGd) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1364
DERING_CORE((%%REGd) ,(%%REGd, %1) ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1365
DERING_CORE((%%REGd, %1) ,(%%REGd, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)
1366
DERING_CORE((%%REGd, %1, 2),(%0, %1, 8) ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)
1367
DERING_CORE((%0, %1, 8) ,(%%REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)
1370
: : "r" (src), "r" ((long)stride), "m" (c->pQPb), "m"(c->pQPb2)
1371
: "%"REG_a, "%"REG_d, "%"REG_c
1373
#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1380
const int QP2= c->QP/2 + 1;
1387
if(*p > max) max= *p;
1388
if(*p < min) min= *p;
1391
avg= (min + max + 1)>>1;
1393
if(max - min <deringThreshold) return;
1395
for(y=0; y<10; y++){
1398
if(src[stride*y + 0] > avg) t+= 1;
1399
if(src[stride*y + 1] > avg) t+= 2;
1400
if(src[stride*y + 2] > avg) t+= 4;
1401
if(src[stride*y + 3] > avg) t+= 8;
1402
if(src[stride*y + 4] > avg) t+= 16;
1403
if(src[stride*y + 5] > avg) t+= 32;
1404
if(src[stride*y + 6] > avg) t+= 64;
1405
if(src[stride*y + 7] > avg) t+= 128;
1406
if(src[stride*y + 8] > avg) t+= 256;
1407
if(src[stride*y + 9] > avg) t+= 512;
1410
t &= (t<<1) & (t>>1);
1415
int t = s[y-1] & s[y] & s[y+1];
1428
int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))
1429
+2*(*(p -1)) + 4*(*p ) + 2*(*(p +1))
1430
+(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));
1433
#ifdef DEBUG_DERING_THRESHOLD
1434
asm volatile("emms\n\t":);
1436
static long long numPixels=0;
1437
if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;
1438
// if((max-min)<20 || (max-min)*QP<200)
1439
// if((max-min)*QP < 500)
1442
static int numSkipped=0;
1443
static int errorSum=0;
1444
static int worstQP=0;
1445
static int worstRange=0;
1446
static int worstDiff=0;
1448
int absDiff= FFABS(diff);
1449
int error= diff*diff;
1451
if(x==1 || x==8 || y==1 || y==8) continue;
1454
if(absDiff > worstDiff){
1457
worstRange= max-min;
1461
if(1024LL*1024LL*1024LL % numSkipped == 0){
1462
av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "
1463
"wRange:%d, wDiff:%d, relSkip:%1.3f\n",
1464
(float)errorSum/numSkipped, numSkipped, worstQP, worstRange,
1465
worstDiff, (float)numSkipped/numPixels);
1470
if (*p + QP2 < f) *p= *p + QP2;
1471
else if(*p - QP2 > f) *p= *p - QP2;
1476
#ifdef DEBUG_DERING_THRESHOLD
1484
*p = FFMIN(*p + 20, 255);
1487
// src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;
1490
#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1492
#endif //HAVE_ALTIVEC
1495
* Deinterlaces the given block by linearly interpolating every second line.
1496
* will be called for every 8x8 block and can read & write from line 4-15
1497
* lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1498
* lines 4-12 will be read into the deblocking filter and should be deinterlaced
1500
static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)
1502
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1505
"lea (%0, %1), %%"REG_a" \n\t"
1506
"lea (%%"REG_a", %1, 4), %%"REG_c" \n\t"
1507
// 0 1 2 3 4 5 6 7 8 9
1508
// %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1 %0+8%1 ecx+4%1
1510
"movq (%0), %%mm0 \n\t"
1511
"movq (%%"REG_a", %1), %%mm1 \n\t"
1513
"movq %%mm0, (%%"REG_a") \n\t"
1514
"movq (%0, %1, 4), %%mm0 \n\t"
1516
"movq %%mm1, (%%"REG_a", %1, 2) \n\t"
1517
"movq (%%"REG_c", %1), %%mm1 \n\t"
1519
"movq %%mm0, (%%"REG_c") \n\t"
1520
"movq (%0, %1, 8), %%mm0 \n\t"
1522
"movq %%mm1, (%%"REG_c", %1, 2) \n\t"
1524
: : "r" (src), "r" ((long)stride)
1525
: "%"REG_a, "%"REG_c
1532
a= *(uint32_t*)&src[stride*0];
1533
b= *(uint32_t*)&src[stride*2];
1534
*(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1535
a= *(uint32_t*)&src[stride*4];
1536
*(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1537
b= *(uint32_t*)&src[stride*6];
1538
*(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1539
a= *(uint32_t*)&src[stride*8];
1540
*(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1547
* Deinterlaces the given block by cubic interpolating every second line.
1548
* will be called for every 8x8 block and can read & write from line 4-15
1549
* lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1550
* lines 4-12 will be read into the deblocking filter and should be deinterlaced
1551
* this filter will read lines 3-15 and write 7-13
1553
static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)
1555
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1558
"lea (%0, %1), %%"REG_a" \n\t"
1559
"lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1560
"lea (%%"REG_d", %1, 4), %%"REG_c" \n\t"
1561
"add %1, %%"REG_c" \n\t"
1562
"pxor %%mm7, %%mm7 \n\t"
1563
// 0 1 2 3 4 5 6 7 8 9 10
1564
// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1 ecx
1566
#define REAL_DEINT_CUBIC(a,b,c,d,e)\
1567
"movq " #a ", %%mm0 \n\t"\
1568
"movq " #b ", %%mm1 \n\t"\
1569
"movq " #d ", %%mm2 \n\t"\
1570
"movq " #e ", %%mm3 \n\t"\
1571
PAVGB(%%mm2, %%mm1) /* (b+d) /2 */\
1572
PAVGB(%%mm3, %%mm0) /* a(a+e) /2 */\
1573
"movq %%mm0, %%mm2 \n\t"\
1574
"punpcklbw %%mm7, %%mm0 \n\t"\
1575
"punpckhbw %%mm7, %%mm2 \n\t"\
1576
"movq %%mm1, %%mm3 \n\t"\
1577
"punpcklbw %%mm7, %%mm1 \n\t"\
1578
"punpckhbw %%mm7, %%mm3 \n\t"\
1579
"psubw %%mm1, %%mm0 \n\t" /* L(a+e - (b+d))/2 */\
1580
"psubw %%mm3, %%mm2 \n\t" /* H(a+e - (b+d))/2 */\
1581
"psraw $3, %%mm0 \n\t" /* L(a+e - (b+d))/16 */\
1582
"psraw $3, %%mm2 \n\t" /* H(a+e - (b+d))/16 */\
1583
"psubw %%mm0, %%mm1 \n\t" /* L(9b + 9d - a - e)/16 */\
1584
"psubw %%mm2, %%mm3 \n\t" /* H(9b + 9d - a - e)/16 */\
1585
"packuswb %%mm3, %%mm1 \n\t"\
1586
"movq %%mm1, " #c " \n\t"
1587
#define DEINT_CUBIC(a,b,c,d,e) REAL_DEINT_CUBIC(a,b,c,d,e)
1589
DEINT_CUBIC((%0) , (%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd, %1))
1590
DEINT_CUBIC((%%REGa, %1), (%0, %1, 4) , (%%REGd) , (%%REGd, %1), (%0, %1, 8))
1591
DEINT_CUBIC((%0, %1, 4) , (%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGc))
1592
DEINT_CUBIC((%%REGd, %1), (%0, %1, 8) , (%%REGd, %1, 4), (%%REGc) , (%%REGc, %1, 2))
1594
: : "r" (src), "r" ((long)stride)
1595
: "%"REG_a, "%"REG_d, "%"REG_c
1597
#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1601
src[stride*3] = CLIP((-src[0] + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);
1602
src[stride*5] = CLIP((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);
1603
src[stride*7] = CLIP((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);
1604
src[stride*9] = CLIP((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);
1607
#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1611
* Deinterlaces the given block by filtering every second line with a (-1 4 2 4 -1) filter.
1612
* will be called for every 8x8 block and can read & write from line 4-15
1613
* lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1614
* lines 4-12 will be read into the deblocking filter and should be deinterlaced
1615
* this filter will read lines 4-13 and write 5-11
1617
static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)
1619
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1622
"lea (%0, %1), %%"REG_a" \n\t"
1623
"lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1624
"pxor %%mm7, %%mm7 \n\t"
1625
"movq (%2), %%mm0 \n\t"
1626
// 0 1 2 3 4 5 6 7 8 9 10
1627
// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1 ecx
1629
#define REAL_DEINT_FF(a,b,c,d)\
1630
"movq " #a ", %%mm1 \n\t"\
1631
"movq " #b ", %%mm2 \n\t"\
1632
"movq " #c ", %%mm3 \n\t"\
1633
"movq " #d ", %%mm4 \n\t"\
1634
PAVGB(%%mm3, %%mm1) \
1635
PAVGB(%%mm4, %%mm0) \
1636
"movq %%mm0, %%mm3 \n\t"\
1637
"punpcklbw %%mm7, %%mm0 \n\t"\
1638
"punpckhbw %%mm7, %%mm3 \n\t"\
1639
"movq %%mm1, %%mm4 \n\t"\
1640
"punpcklbw %%mm7, %%mm1 \n\t"\
1641
"punpckhbw %%mm7, %%mm4 \n\t"\
1642
"psllw $2, %%mm1 \n\t"\
1643
"psllw $2, %%mm4 \n\t"\
1644
"psubw %%mm0, %%mm1 \n\t"\
1645
"psubw %%mm3, %%mm4 \n\t"\
1646
"movq %%mm2, %%mm5 \n\t"\
1647
"movq %%mm2, %%mm0 \n\t"\
1648
"punpcklbw %%mm7, %%mm2 \n\t"\
1649
"punpckhbw %%mm7, %%mm5 \n\t"\
1650
"paddw %%mm2, %%mm1 \n\t"\
1651
"paddw %%mm5, %%mm4 \n\t"\
1652
"psraw $2, %%mm1 \n\t"\
1653
"psraw $2, %%mm4 \n\t"\
1654
"packuswb %%mm4, %%mm1 \n\t"\
1655
"movq %%mm1, " #b " \n\t"\
1657
#define DEINT_FF(a,b,c,d) REAL_DEINT_FF(a,b,c,d)
1659
DEINT_FF((%0) , (%%REGa) , (%%REGa, %1), (%%REGa, %1, 2))
1660
DEINT_FF((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd) )
1661
DEINT_FF((%0, %1, 4) , (%%REGd) , (%%REGd, %1), (%%REGd, %1, 2))
1662
DEINT_FF((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))
1664
"movq %%mm0, (%2) \n\t"
1665
: : "r" (src), "r" ((long)stride), "r"(tmp)
1666
: "%"REG_a, "%"REG_d
1668
#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1673
int t2= src[stride*1];
1675
src[stride*1]= CLIP((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);
1677
src[stride*3]= CLIP((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);
1679
src[stride*5]= CLIP((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);
1681
src[stride*7]= CLIP((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);
1686
#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1690
* Deinterlaces the given block by filtering every line with a (-1 2 6 2 -1) filter.
1691
* will be called for every 8x8 block and can read & write from line 4-15
1692
* lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1693
* lines 4-12 will be read into the deblocking filter and should be deinterlaced
1694
* this filter will read lines 4-13 and write 4-11
1696
static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)
1698
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1701
"lea (%0, %1), %%"REG_a" \n\t"
1702
"lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1703
"pxor %%mm7, %%mm7 \n\t"
1704
"movq (%2), %%mm0 \n\t"
1705
"movq (%3), %%mm1 \n\t"
1706
// 0 1 2 3 4 5 6 7 8 9 10
1707
// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1 ecx
1709
#define REAL_DEINT_L5(t1,t2,a,b,c)\
1710
"movq " #a ", %%mm2 \n\t"\
1711
"movq " #b ", %%mm3 \n\t"\
1712
"movq " #c ", %%mm4 \n\t"\
1715
"movq %%mm2, %%mm5 \n\t"\
1716
"movq %%mm2, " #t1 " \n\t"\
1717
"punpcklbw %%mm7, %%mm2 \n\t"\
1718
"punpckhbw %%mm7, %%mm5 \n\t"\
1719
"movq %%mm2, %%mm6 \n\t"\
1720
"paddw %%mm2, %%mm2 \n\t"\
1721
"paddw %%mm6, %%mm2 \n\t"\
1722
"movq %%mm5, %%mm6 \n\t"\
1723
"paddw %%mm5, %%mm5 \n\t"\
1724
"paddw %%mm6, %%mm5 \n\t"\
1725
"movq %%mm3, %%mm6 \n\t"\
1726
"punpcklbw %%mm7, %%mm3 \n\t"\
1727
"punpckhbw %%mm7, %%mm6 \n\t"\
1728
"paddw %%mm3, %%mm3 \n\t"\
1729
"paddw %%mm6, %%mm6 \n\t"\
1730
"paddw %%mm3, %%mm2 \n\t"\
1731
"paddw %%mm6, %%mm5 \n\t"\
1732
"movq %%mm4, %%mm6 \n\t"\
1733
"punpcklbw %%mm7, %%mm4 \n\t"\
1734
"punpckhbw %%mm7, %%mm6 \n\t"\
1735
"psubw %%mm4, %%mm2 \n\t"\
1736
"psubw %%mm6, %%mm5 \n\t"\
1737
"psraw $2, %%mm2 \n\t"\
1738
"psraw $2, %%mm5 \n\t"\
1739
"packuswb %%mm5, %%mm2 \n\t"\
1740
"movq %%mm2, " #a " \n\t"\
1742
#define DEINT_L5(t1,t2,a,b,c) REAL_DEINT_L5(t1,t2,a,b,c)
1744
DEINT_L5(%%mm0, %%mm1, (%0) , (%%REGa) , (%%REGa, %1) )
1745
DEINT_L5(%%mm1, %%mm0, (%%REGa) , (%%REGa, %1) , (%%REGa, %1, 2))
1746
DEINT_L5(%%mm0, %%mm1, (%%REGa, %1) , (%%REGa, %1, 2), (%0, %1, 4) )
1747
DEINT_L5(%%mm1, %%mm0, (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd) )
1748
DEINT_L5(%%mm0, %%mm1, (%0, %1, 4) , (%%REGd) , (%%REGd, %1) )
1749
DEINT_L5(%%mm1, %%mm0, (%%REGd) , (%%REGd, %1) , (%%REGd, %1, 2))
1750
DEINT_L5(%%mm0, %%mm1, (%%REGd, %1) , (%%REGd, %1, 2), (%0, %1, 8) )
1751
DEINT_L5(%%mm1, %%mm0, (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))
1753
"movq %%mm0, (%2) \n\t"
1754
"movq %%mm1, (%3) \n\t"
1755
: : "r" (src), "r" ((long)stride), "r"(tmp), "r"(tmp2)
1756
: "%"REG_a, "%"REG_d
1758
#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1766
src[stride*0]= CLIP((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);
1768
src[stride*1]= CLIP((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);
1770
src[stride*2]= CLIP((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);
1772
src[stride*3]= CLIP((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);
1774
src[stride*4]= CLIP((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);
1776
src[stride*5]= CLIP((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);
1778
src[stride*6]= CLIP((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);
1780
src[stride*7]= CLIP((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);
1787
#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1791
* Deinterlaces the given block by filtering all lines with a (1 2 1) filter.
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
1795
* this filter will read lines 4-13 and write 4-11
1797
static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uint8_t *tmp)
1799
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
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 (%2), %%mm0 \n\t" // L0
1808
"movq (%%"REG_a"), %%mm1 \n\t" // L2
1809
PAVGB(%%mm1, %%mm0) // L0+L2
1810
"movq (%0), %%mm2 \n\t" // L1
1812
"movq %%mm0, (%0) \n\t"
1813
"movq (%%"REG_a", %1), %%mm0 \n\t" // L3
1814
PAVGB(%%mm0, %%mm2) // L1+L3
1815
PAVGB(%%mm1, %%mm2) // 2L2 + L1 + L3
1816
"movq %%mm2, (%%"REG_a") \n\t"
1817
"movq (%%"REG_a", %1, 2), %%mm2 \n\t" // L4
1818
PAVGB(%%mm2, %%mm1) // L2+L4
1819
PAVGB(%%mm0, %%mm1) // 2L3 + L2 + L4
1820
"movq %%mm1, (%%"REG_a", %1) \n\t"
1821
"movq (%0, %1, 4), %%mm1 \n\t" // L5
1822
PAVGB(%%mm1, %%mm0) // L3+L5
1823
PAVGB(%%mm2, %%mm0) // 2L4 + L3 + L5
1824
"movq %%mm0, (%%"REG_a", %1, 2) \n\t"
1825
"movq (%%"REG_d"), %%mm0 \n\t" // L6
1826
PAVGB(%%mm0, %%mm2) // L4+L6
1827
PAVGB(%%mm1, %%mm2) // 2L5 + L4 + L6
1828
"movq %%mm2, (%0, %1, 4) \n\t"
1829
"movq (%%"REG_d", %1), %%mm2 \n\t" // L7
1830
PAVGB(%%mm2, %%mm1) // L5+L7
1831
PAVGB(%%mm0, %%mm1) // 2L6 + L5 + L7
1832
"movq %%mm1, (%%"REG_d") \n\t"
1833
"movq (%%"REG_d", %1, 2), %%mm1 \n\t" // L8
1834
PAVGB(%%mm1, %%mm0) // L6+L8
1835
PAVGB(%%mm2, %%mm0) // 2L7 + L6 + L8
1836
"movq %%mm0, (%%"REG_d", %1) \n\t"
1837
"movq (%0, %1, 8), %%mm0 \n\t" // L9
1838
PAVGB(%%mm0, %%mm2) // L7+L9
1839
PAVGB(%%mm1, %%mm2) // 2L8 + L7 + L9
1840
"movq %%mm2, (%%"REG_d", %1, 2) \n\t"
1841
"movq %%mm1, (%2) \n\t"
1843
: : "r" (src), "r" ((long)stride), "r" (tmp)
1844
: "%"REG_a, "%"REG_d
1846
#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1851
a= *(uint32_t*)&tmp[stride*0];
1852
b= *(uint32_t*)&src[stride*0];
1853
c= *(uint32_t*)&src[stride*1];
1854
a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1855
*(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1857
a= *(uint32_t*)&src[stride*2];
1858
b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1859
*(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1861
b= *(uint32_t*)&src[stride*3];
1862
c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1863
*(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1865
c= *(uint32_t*)&src[stride*4];
1866
a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1867
*(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1869
a= *(uint32_t*)&src[stride*5];
1870
b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1871
*(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1873
b= *(uint32_t*)&src[stride*6];
1874
c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);
1875
*(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);
1877
c= *(uint32_t*)&src[stride*7];
1878
a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);
1879
*(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);
1881
a= *(uint32_t*)&src[stride*8];
1882
b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);
1883
*(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);
1885
*(uint32_t*)&tmp[stride*0]= c;
1889
#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
1893
* Deinterlaces the given block by applying a median filter to every second line.
1894
* will be called for every 8x8 block and can read & write from line 4-15,
1895
* lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.
1896
* lines 4-12 will be read into the deblocking filter and should be deinterlaced
1898
static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)
1904
"lea (%0, %1), %%"REG_a" \n\t"
1905
"lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1906
// 0 1 2 3 4 5 6 7 8 9
1907
// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1909
"movq (%0), %%mm0 \n\t" //
1910
"movq (%%"REG_a", %1), %%mm2 \n\t" //
1911
"movq (%%"REG_a"), %%mm1 \n\t" //
1912
"movq %%mm0, %%mm3 \n\t"
1913
"pmaxub %%mm1, %%mm0 \n\t" //
1914
"pminub %%mm3, %%mm1 \n\t" //
1915
"pmaxub %%mm2, %%mm1 \n\t" //
1916
"pminub %%mm1, %%mm0 \n\t"
1917
"movq %%mm0, (%%"REG_a") \n\t"
1919
"movq (%0, %1, 4), %%mm0 \n\t" //
1920
"movq (%%"REG_a", %1, 2), %%mm1 \n\t" //
1921
"movq %%mm2, %%mm3 \n\t"
1922
"pmaxub %%mm1, %%mm2 \n\t" //
1923
"pminub %%mm3, %%mm1 \n\t" //
1924
"pmaxub %%mm0, %%mm1 \n\t" //
1925
"pminub %%mm1, %%mm2 \n\t"
1926
"movq %%mm2, (%%"REG_a", %1, 2) \n\t"
1928
"movq (%%"REG_d"), %%mm2 \n\t" //
1929
"movq (%%"REG_d", %1), %%mm1 \n\t" //
1930
"movq %%mm2, %%mm3 \n\t"
1931
"pmaxub %%mm0, %%mm2 \n\t" //
1932
"pminub %%mm3, %%mm0 \n\t" //
1933
"pmaxub %%mm1, %%mm0 \n\t" //
1934
"pminub %%mm0, %%mm2 \n\t"
1935
"movq %%mm2, (%%"REG_d") \n\t"
1937
"movq (%%"REG_d", %1, 2), %%mm2 \n\t" //
1938
"movq (%0, %1, 8), %%mm0 \n\t" //
1939
"movq %%mm2, %%mm3 \n\t"
1940
"pmaxub %%mm0, %%mm2 \n\t" //
1941
"pminub %%mm3, %%mm0 \n\t" //
1942
"pmaxub %%mm1, %%mm0 \n\t" //
1943
"pminub %%mm0, %%mm2 \n\t"
1944
"movq %%mm2, (%%"REG_d", %1, 2) \n\t"
1947
: : "r" (src), "r" ((long)stride)
1948
: "%"REG_a, "%"REG_d
1951
#else // MMX without MMX2
1953
"lea (%0, %1), %%"REG_a" \n\t"
1954
"lea (%%"REG_a", %1, 4), %%"REG_d" \n\t"
1955
// 0 1 2 3 4 5 6 7 8 9
1956
// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
1957
"pxor %%mm7, %%mm7 \n\t"
1959
#define REAL_MEDIAN(a,b,c)\
1960
"movq " #a ", %%mm0 \n\t"\
1961
"movq " #b ", %%mm2 \n\t"\
1962
"movq " #c ", %%mm1 \n\t"\
1963
"movq %%mm0, %%mm3 \n\t"\
1964
"movq %%mm1, %%mm4 \n\t"\
1965
"movq %%mm2, %%mm5 \n\t"\
1966
"psubusb %%mm1, %%mm3 \n\t"\
1967
"psubusb %%mm2, %%mm4 \n\t"\
1968
"psubusb %%mm0, %%mm5 \n\t"\
1969
"pcmpeqb %%mm7, %%mm3 \n\t"\
1970
"pcmpeqb %%mm7, %%mm4 \n\t"\
1971
"pcmpeqb %%mm7, %%mm5 \n\t"\
1972
"movq %%mm3, %%mm6 \n\t"\
1973
"pxor %%mm4, %%mm3 \n\t"\
1974
"pxor %%mm5, %%mm4 \n\t"\
1975
"pxor %%mm6, %%mm5 \n\t"\
1976
"por %%mm3, %%mm1 \n\t"\
1977
"por %%mm4, %%mm2 \n\t"\
1978
"por %%mm5, %%mm0 \n\t"\
1979
"pand %%mm2, %%mm0 \n\t"\
1980
"pand %%mm1, %%mm0 \n\t"\
1981
"movq %%mm0, " #b " \n\t"
1982
#define MEDIAN(a,b,c) REAL_MEDIAN(a,b,c)
1984
MEDIAN((%0) , (%%REGa) , (%%REGa, %1))
1985
MEDIAN((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4))
1986
MEDIAN((%0, %1, 4) , (%%REGd) , (%%REGd, %1))
1987
MEDIAN((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8))
1989
: : "r" (src), "r" ((long)stride)
1990
: "%"REG_a, "%"REG_d
1996
// FIXME - there should be a way to do a few columns in parallel like w/mmx
1998
uint8_t *colsrc = src;
1999
for (y=0; y<4; y++){
2000
int a, b, c, d, e, f;
2002
b = colsrc[stride ];
2003
c = colsrc[stride*2];
2007
colsrc[stride ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));
2017
* transposes and shift the given 8x8 Block into dst1 and dst2
2019
static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)
2022
"lea (%0, %1), %%"REG_a" \n\t"
2023
// 0 1 2 3 4 5 6 7 8 9
2024
// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
2025
"movq (%0), %%mm0 \n\t" // 12345678
2026
"movq (%%"REG_a"), %%mm1 \n\t" // abcdefgh
2027
"movq %%mm0, %%mm2 \n\t" // 12345678
2028
"punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
2029
"punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
2031
"movq (%%"REG_a", %1), %%mm1 \n\t"
2032
"movq (%%"REG_a", %1, 2), %%mm3 \n\t"
2033
"movq %%mm1, %%mm4 \n\t"
2034
"punpcklbw %%mm3, %%mm1 \n\t"
2035
"punpckhbw %%mm3, %%mm4 \n\t"
2037
"movq %%mm0, %%mm3 \n\t"
2038
"punpcklwd %%mm1, %%mm0 \n\t"
2039
"punpckhwd %%mm1, %%mm3 \n\t"
2040
"movq %%mm2, %%mm1 \n\t"
2041
"punpcklwd %%mm4, %%mm2 \n\t"
2042
"punpckhwd %%mm4, %%mm1 \n\t"
2044
"movd %%mm0, 128(%2) \n\t"
2045
"psrlq $32, %%mm0 \n\t"
2046
"movd %%mm0, 144(%2) \n\t"
2047
"movd %%mm3, 160(%2) \n\t"
2048
"psrlq $32, %%mm3 \n\t"
2049
"movd %%mm3, 176(%2) \n\t"
2050
"movd %%mm3, 48(%3) \n\t"
2051
"movd %%mm2, 192(%2) \n\t"
2052
"movd %%mm2, 64(%3) \n\t"
2053
"psrlq $32, %%mm2 \n\t"
2054
"movd %%mm2, 80(%3) \n\t"
2055
"movd %%mm1, 96(%3) \n\t"
2056
"psrlq $32, %%mm1 \n\t"
2057
"movd %%mm1, 112(%3) \n\t"
2059
"lea (%%"REG_a", %1, 4), %%"REG_a" \n\t"
2061
"movq (%0, %1, 4), %%mm0 \n\t" // 12345678
2062
"movq (%%"REG_a"), %%mm1 \n\t" // abcdefgh
2063
"movq %%mm0, %%mm2 \n\t" // 12345678
2064
"punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
2065
"punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
2067
"movq (%%"REG_a", %1), %%mm1 \n\t"
2068
"movq (%%"REG_a", %1, 2), %%mm3 \n\t"
2069
"movq %%mm1, %%mm4 \n\t"
2070
"punpcklbw %%mm3, %%mm1 \n\t"
2071
"punpckhbw %%mm3, %%mm4 \n\t"
2073
"movq %%mm0, %%mm3 \n\t"
2074
"punpcklwd %%mm1, %%mm0 \n\t"
2075
"punpckhwd %%mm1, %%mm3 \n\t"
2076
"movq %%mm2, %%mm1 \n\t"
2077
"punpcklwd %%mm4, %%mm2 \n\t"
2078
"punpckhwd %%mm4, %%mm1 \n\t"
2080
"movd %%mm0, 132(%2) \n\t"
2081
"psrlq $32, %%mm0 \n\t"
2082
"movd %%mm0, 148(%2) \n\t"
2083
"movd %%mm3, 164(%2) \n\t"
2084
"psrlq $32, %%mm3 \n\t"
2085
"movd %%mm3, 180(%2) \n\t"
2086
"movd %%mm3, 52(%3) \n\t"
2087
"movd %%mm2, 196(%2) \n\t"
2088
"movd %%mm2, 68(%3) \n\t"
2089
"psrlq $32, %%mm2 \n\t"
2090
"movd %%mm2, 84(%3) \n\t"
2091
"movd %%mm1, 100(%3) \n\t"
2092
"psrlq $32, %%mm1 \n\t"
2093
"movd %%mm1, 116(%3) \n\t"
2096
:: "r" (src), "r" ((long)srcStride), "r" (dst1), "r" (dst2)
2102
* transposes the given 8x8 block
2104
static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)
2107
"lea (%0, %1), %%"REG_a" \n\t"
2108
"lea (%%"REG_a",%1,4), %%"REG_d" \n\t"
2109
// 0 1 2 3 4 5 6 7 8 9
2110
// %0 eax eax+%1 eax+2%1 %0+4%1 edx edx+%1 edx+2%1 %0+8%1 edx+4%1
2111
"movq (%2), %%mm0 \n\t" // 12345678
2112
"movq 16(%2), %%mm1 \n\t" // abcdefgh
2113
"movq %%mm0, %%mm2 \n\t" // 12345678
2114
"punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
2115
"punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
2117
"movq 32(%2), %%mm1 \n\t"
2118
"movq 48(%2), %%mm3 \n\t"
2119
"movq %%mm1, %%mm4 \n\t"
2120
"punpcklbw %%mm3, %%mm1 \n\t"
2121
"punpckhbw %%mm3, %%mm4 \n\t"
2123
"movq %%mm0, %%mm3 \n\t"
2124
"punpcklwd %%mm1, %%mm0 \n\t"
2125
"punpckhwd %%mm1, %%mm3 \n\t"
2126
"movq %%mm2, %%mm1 \n\t"
2127
"punpcklwd %%mm4, %%mm2 \n\t"
2128
"punpckhwd %%mm4, %%mm1 \n\t"
2130
"movd %%mm0, (%0) \n\t"
2131
"psrlq $32, %%mm0 \n\t"
2132
"movd %%mm0, (%%"REG_a") \n\t"
2133
"movd %%mm3, (%%"REG_a", %1) \n\t"
2134
"psrlq $32, %%mm3 \n\t"
2135
"movd %%mm3, (%%"REG_a", %1, 2) \n\t"
2136
"movd %%mm2, (%0, %1, 4) \n\t"
2137
"psrlq $32, %%mm2 \n\t"
2138
"movd %%mm2, (%%"REG_d") \n\t"
2139
"movd %%mm1, (%%"REG_d", %1) \n\t"
2140
"psrlq $32, %%mm1 \n\t"
2141
"movd %%mm1, (%%"REG_d", %1, 2) \n\t"
2144
"movq 64(%2), %%mm0 \n\t" // 12345678
2145
"movq 80(%2), %%mm1 \n\t" // abcdefgh
2146
"movq %%mm0, %%mm2 \n\t" // 12345678
2147
"punpcklbw %%mm1, %%mm0 \n\t" // 1a2b3c4d
2148
"punpckhbw %%mm1, %%mm2 \n\t" // 5e6f7g8h
2150
"movq 96(%2), %%mm1 \n\t"
2151
"movq 112(%2), %%mm3 \n\t"
2152
"movq %%mm1, %%mm4 \n\t"
2153
"punpcklbw %%mm3, %%mm1 \n\t"
2154
"punpckhbw %%mm3, %%mm4 \n\t"
2156
"movq %%mm0, %%mm3 \n\t"
2157
"punpcklwd %%mm1, %%mm0 \n\t"
2158
"punpckhwd %%mm1, %%mm3 \n\t"
2159
"movq %%mm2, %%mm1 \n\t"
2160
"punpcklwd %%mm4, %%mm2 \n\t"
2161
"punpckhwd %%mm4, %%mm1 \n\t"
2163
"movd %%mm0, 4(%0) \n\t"
2164
"psrlq $32, %%mm0 \n\t"
2165
"movd %%mm0, 4(%%"REG_a") \n\t"
2166
"movd %%mm3, 4(%%"REG_a", %1) \n\t"
2167
"psrlq $32, %%mm3 \n\t"
2168
"movd %%mm3, 4(%%"REG_a", %1, 2) \n\t"
2169
"movd %%mm2, 4(%0, %1, 4) \n\t"
2170
"psrlq $32, %%mm2 \n\t"
2171
"movd %%mm2, 4(%%"REG_d") \n\t"
2172
"movd %%mm1, 4(%%"REG_d", %1) \n\t"
2173
"psrlq $32, %%mm1 \n\t"
2174
"movd %%mm1, 4(%%"REG_d", %1, 2) \n\t"
2176
:: "r" (dst), "r" ((long)dstStride), "r" (src)
2177
: "%"REG_a, "%"REG_d
2181
//static long test=0;
2183
#ifndef HAVE_ALTIVEC
2184
static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,
2185
uint8_t *tempBlurred, uint32_t *tempBlurredPast, int *maxNoise)
2187
// to save a register (FIXME do this outside of the loops)
2188
tempBlurredPast[127]= maxNoise[0];
2189
tempBlurredPast[128]= maxNoise[1];
2190
tempBlurredPast[129]= maxNoise[2];
2192
#define FAST_L2_DIFF
2193
//#define L1_DIFF //u should change the thresholds too if u try that one
2194
#if defined (HAVE_MMX2) || defined (HAVE_3DNOW)
2196
"lea (%2, %2, 2), %%"REG_a" \n\t" // 3*stride
2197
"lea (%2, %2, 4), %%"REG_d" \n\t" // 5*stride
2198
"lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
2199
// 0 1 2 3 4 5 6 7 8 9
2200
// %x %x+%2 %x+2%2 %x+eax %x+4%2 %x+edx %x+2eax %x+ecx %x+8%2
2202
#ifdef L1_DIFF //needs mmx2
2203
"movq (%0), %%mm0 \n\t" // L0
2204
"psadbw (%1), %%mm0 \n\t" // |L0-R0|
2205
"movq (%0, %2), %%mm1 \n\t" // L1
2206
"psadbw (%1, %2), %%mm1 \n\t" // |L1-R1|
2207
"movq (%0, %2, 2), %%mm2 \n\t" // L2
2208
"psadbw (%1, %2, 2), %%mm2 \n\t" // |L2-R2|
2209
"movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
2210
"psadbw (%1, %%"REG_a"), %%mm3 \n\t" // |L3-R3|
2212
"movq (%0, %2, 4), %%mm4 \n\t" // L4
2213
"paddw %%mm1, %%mm0 \n\t"
2214
"psadbw (%1, %2, 4), %%mm4 \n\t" // |L4-R4|
2215
"movq (%0, %%"REG_d"), %%mm5 \n\t" // L5
2216
"paddw %%mm2, %%mm0 \n\t"
2217
"psadbw (%1, %%"REG_d"), %%mm5 \n\t" // |L5-R5|
2218
"movq (%0, %%"REG_a", 2), %%mm6 \n\t" // L6
2219
"paddw %%mm3, %%mm0 \n\t"
2220
"psadbw (%1, %%"REG_a", 2), %%mm6 \n\t" // |L6-R6|
2221
"movq (%0, %%"REG_c"), %%mm7 \n\t" // L7
2222
"paddw %%mm4, %%mm0 \n\t"
2223
"psadbw (%1, %%"REG_c"), %%mm7 \n\t" // |L7-R7|
2224
"paddw %%mm5, %%mm6 \n\t"
2225
"paddw %%mm7, %%mm6 \n\t"
2226
"paddw %%mm6, %%mm0 \n\t"
2228
#if defined (FAST_L2_DIFF)
2229
"pcmpeqb %%mm7, %%mm7 \n\t"
2230
"movq "MANGLE(b80)", %%mm6 \n\t"
2231
"pxor %%mm0, %%mm0 \n\t"
2232
#define REAL_L2_DIFF_CORE(a, b)\
2233
"movq " #a ", %%mm5 \n\t"\
2234
"movq " #b ", %%mm2 \n\t"\
2235
"pxor %%mm7, %%mm2 \n\t"\
2236
PAVGB(%%mm2, %%mm5)\
2237
"paddb %%mm6, %%mm5 \n\t"\
2238
"movq %%mm5, %%mm2 \n\t"\
2239
"psllw $8, %%mm5 \n\t"\
2240
"pmaddwd %%mm5, %%mm5 \n\t"\
2241
"pmaddwd %%mm2, %%mm2 \n\t"\
2242
"paddd %%mm2, %%mm5 \n\t"\
2243
"psrld $14, %%mm5 \n\t"\
2244
"paddd %%mm5, %%mm0 \n\t"
2246
#else //defined (FAST_L2_DIFF)
2247
"pxor %%mm7, %%mm7 \n\t"
2248
"pxor %%mm0, %%mm0 \n\t"
2249
#define REAL_L2_DIFF_CORE(a, b)\
2250
"movq " #a ", %%mm5 \n\t"\
2251
"movq " #b ", %%mm2 \n\t"\
2252
"movq %%mm5, %%mm1 \n\t"\
2253
"movq %%mm2, %%mm3 \n\t"\
2254
"punpcklbw %%mm7, %%mm5 \n\t"\
2255
"punpckhbw %%mm7, %%mm1 \n\t"\
2256
"punpcklbw %%mm7, %%mm2 \n\t"\
2257
"punpckhbw %%mm7, %%mm3 \n\t"\
2258
"psubw %%mm2, %%mm5 \n\t"\
2259
"psubw %%mm3, %%mm1 \n\t"\
2260
"pmaddwd %%mm5, %%mm5 \n\t"\
2261
"pmaddwd %%mm1, %%mm1 \n\t"\
2262
"paddd %%mm1, %%mm5 \n\t"\
2263
"paddd %%mm5, %%mm0 \n\t"
2265
#endif //defined (FAST_L2_DIFF)
2267
#define L2_DIFF_CORE(a, b) REAL_L2_DIFF_CORE(a, b)
2269
L2_DIFF_CORE((%0) , (%1))
2270
L2_DIFF_CORE((%0, %2) , (%1, %2))
2271
L2_DIFF_CORE((%0, %2, 2) , (%1, %2, 2))
2272
L2_DIFF_CORE((%0, %%REGa) , (%1, %%REGa))
2273
L2_DIFF_CORE((%0, %2, 4) , (%1, %2, 4))
2274
L2_DIFF_CORE((%0, %%REGd) , (%1, %%REGd))
2275
L2_DIFF_CORE((%0, %%REGa,2), (%1, %%REGa,2))
2276
L2_DIFF_CORE((%0, %%REGc) , (%1, %%REGc))
2280
"movq %%mm0, %%mm4 \n\t"
2281
"psrlq $32, %%mm0 \n\t"
2282
"paddd %%mm0, %%mm4 \n\t"
2283
"movd %%mm4, %%ecx \n\t"
2284
"shll $2, %%ecx \n\t"
2285
"mov %3, %%"REG_d" \n\t"
2286
"addl -4(%%"REG_d"), %%ecx \n\t"
2287
"addl 4(%%"REG_d"), %%ecx \n\t"
2288
"addl -1024(%%"REG_d"), %%ecx \n\t"
2289
"addl $4, %%ecx \n\t"
2290
"addl 1024(%%"REG_d"), %%ecx \n\t"
2291
"shrl $3, %%ecx \n\t"
2292
"movl %%ecx, (%%"REG_d") \n\t"
2294
// "mov %3, %%"REG_c" \n\t"
2295
// "mov %%"REG_c", test \n\t"
2297
"cmpl 512(%%"REG_d"), %%ecx \n\t"
2299
"cmpl 516(%%"REG_d"), %%ecx \n\t"
2302
"lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
2303
"lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
2304
"movq (%0), %%mm0 \n\t" // L0
2305
"movq (%0, %2), %%mm1 \n\t" // L1
2306
"movq (%0, %2, 2), %%mm2 \n\t" // L2
2307
"movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
2308
"movq (%0, %2, 4), %%mm4 \n\t" // L4
2309
"movq (%0, %%"REG_d"), %%mm5 \n\t" // L5
2310
"movq (%0, %%"REG_a", 2), %%mm6 \n\t" // L6
2311
"movq (%0, %%"REG_c"), %%mm7 \n\t" // L7
2312
"movq %%mm0, (%1) \n\t" // L0
2313
"movq %%mm1, (%1, %2) \n\t" // L1
2314
"movq %%mm2, (%1, %2, 2) \n\t" // L2
2315
"movq %%mm3, (%1, %%"REG_a") \n\t" // L3
2316
"movq %%mm4, (%1, %2, 4) \n\t" // L4
2317
"movq %%mm5, (%1, %%"REG_d") \n\t" // L5
2318
"movq %%mm6, (%1, %%"REG_a", 2) \n\t" // L6
2319
"movq %%mm7, (%1, %%"REG_c") \n\t" // L7
2323
"lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
2324
"lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
2325
"movq (%0), %%mm0 \n\t" // L0
2326
PAVGB((%1), %%mm0) // L0
2327
"movq (%0, %2), %%mm1 \n\t" // L1
2328
PAVGB((%1, %2), %%mm1) // L1
2329
"movq (%0, %2, 2), %%mm2 \n\t" // L2
2330
PAVGB((%1, %2, 2), %%mm2) // L2
2331
"movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
2332
PAVGB((%1, %%REGa), %%mm3) // L3
2333
"movq (%0, %2, 4), %%mm4 \n\t" // L4
2334
PAVGB((%1, %2, 4), %%mm4) // L4
2335
"movq (%0, %%"REG_d"), %%mm5 \n\t" // L5
2336
PAVGB((%1, %%REGd), %%mm5) // L5
2337
"movq (%0, %%"REG_a", 2), %%mm6 \n\t" // L6
2338
PAVGB((%1, %%REGa, 2), %%mm6) // L6
2339
"movq (%0, %%"REG_c"), %%mm7 \n\t" // L7
2340
PAVGB((%1, %%REGc), %%mm7) // L7
2341
"movq %%mm0, (%1) \n\t" // R0
2342
"movq %%mm1, (%1, %2) \n\t" // R1
2343
"movq %%mm2, (%1, %2, 2) \n\t" // R2
2344
"movq %%mm3, (%1, %%"REG_a") \n\t" // R3
2345
"movq %%mm4, (%1, %2, 4) \n\t" // R4
2346
"movq %%mm5, (%1, %%"REG_d") \n\t" // R5
2347
"movq %%mm6, (%1, %%"REG_a", 2) \n\t" // R6
2348
"movq %%mm7, (%1, %%"REG_c") \n\t" // R7
2349
"movq %%mm0, (%0) \n\t" // L0
2350
"movq %%mm1, (%0, %2) \n\t" // L1
2351
"movq %%mm2, (%0, %2, 2) \n\t" // L2
2352
"movq %%mm3, (%0, %%"REG_a") \n\t" // L3
2353
"movq %%mm4, (%0, %2, 4) \n\t" // L4
2354
"movq %%mm5, (%0, %%"REG_d") \n\t" // L5
2355
"movq %%mm6, (%0, %%"REG_a", 2) \n\t" // L6
2356
"movq %%mm7, (%0, %%"REG_c") \n\t" // L7
2360
"cmpl 508(%%"REG_d"), %%ecx \n\t"
2363
"lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
2364
"lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
2365
"movq (%0), %%mm0 \n\t" // L0
2366
"movq (%0, %2), %%mm1 \n\t" // L1
2367
"movq (%0, %2, 2), %%mm2 \n\t" // L2
2368
"movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
2369
"movq (%1), %%mm4 \n\t" // R0
2370
"movq (%1, %2), %%mm5 \n\t" // R1
2371
"movq (%1, %2, 2), %%mm6 \n\t" // R2
2372
"movq (%1, %%"REG_a"), %%mm7 \n\t" // R3
2381
"movq %%mm0, (%1) \n\t" // R0
2382
"movq %%mm1, (%1, %2) \n\t" // R1
2383
"movq %%mm2, (%1, %2, 2) \n\t" // R2
2384
"movq %%mm3, (%1, %%"REG_a") \n\t" // R3
2385
"movq %%mm0, (%0) \n\t" // L0
2386
"movq %%mm1, (%0, %2) \n\t" // L1
2387
"movq %%mm2, (%0, %2, 2) \n\t" // L2
2388
"movq %%mm3, (%0, %%"REG_a") \n\t" // L3
2390
"movq (%0, %2, 4), %%mm0 \n\t" // L4
2391
"movq (%0, %%"REG_d"), %%mm1 \n\t" // L5
2392
"movq (%0, %%"REG_a", 2), %%mm2 \n\t" // L6
2393
"movq (%0, %%"REG_c"), %%mm3 \n\t" // L7
2394
"movq (%1, %2, 4), %%mm4 \n\t" // R4
2395
"movq (%1, %%"REG_d"), %%mm5 \n\t" // R5
2396
"movq (%1, %%"REG_a", 2), %%mm6 \n\t" // R6
2397
"movq (%1, %%"REG_c"), %%mm7 \n\t" // R7
2406
"movq %%mm0, (%1, %2, 4) \n\t" // R4
2407
"movq %%mm1, (%1, %%"REG_d") \n\t" // R5
2408
"movq %%mm2, (%1, %%"REG_a", 2) \n\t" // R6
2409
"movq %%mm3, (%1, %%"REG_c") \n\t" // R7
2410
"movq %%mm0, (%0, %2, 4) \n\t" // L4
2411
"movq %%mm1, (%0, %%"REG_d") \n\t" // L5
2412
"movq %%mm2, (%0, %%"REG_a", 2) \n\t" // L6
2413
"movq %%mm3, (%0, %%"REG_c") \n\t" // L7
2417
"lea (%%"REG_a", %2, 2), %%"REG_d" \n\t" // 5*stride
2418
"lea (%%"REG_d", %2, 2), %%"REG_c" \n\t" // 7*stride
2419
"movq (%0), %%mm0 \n\t" // L0
2420
"movq (%0, %2), %%mm1 \n\t" // L1
2421
"movq (%0, %2, 2), %%mm2 \n\t" // L2
2422
"movq (%0, %%"REG_a"), %%mm3 \n\t" // L3
2423
"movq (%1), %%mm4 \n\t" // R0
2424
"movq (%1, %2), %%mm5 \n\t" // R1
2425
"movq (%1, %2, 2), %%mm6 \n\t" // R2
2426
"movq (%1, %%"REG_a"), %%mm7 \n\t" // R3
2439
"movq %%mm0, (%1) \n\t" // R0
2440
"movq %%mm1, (%1, %2) \n\t" // R1
2441
"movq %%mm2, (%1, %2, 2) \n\t" // R2
2442
"movq %%mm3, (%1, %%"REG_a") \n\t" // R3
2443
"movq %%mm0, (%0) \n\t" // L0
2444
"movq %%mm1, (%0, %2) \n\t" // L1
2445
"movq %%mm2, (%0, %2, 2) \n\t" // L2
2446
"movq %%mm3, (%0, %%"REG_a") \n\t" // L3
2448
"movq (%0, %2, 4), %%mm0 \n\t" // L4
2449
"movq (%0, %%"REG_d"), %%mm1 \n\t" // L5
2450
"movq (%0, %%"REG_a", 2), %%mm2 \n\t" // L6
2451
"movq (%0, %%"REG_c"), %%mm3 \n\t" // L7
2452
"movq (%1, %2, 4), %%mm4 \n\t" // R4
2453
"movq (%1, %%"REG_d"), %%mm5 \n\t" // R5
2454
"movq (%1, %%"REG_a", 2), %%mm6 \n\t" // R6
2455
"movq (%1, %%"REG_c"), %%mm7 \n\t" // R7
2468
"movq %%mm0, (%1, %2, 4) \n\t" // R4
2469
"movq %%mm1, (%1, %%"REG_d") \n\t" // R5
2470
"movq %%mm2, (%1, %%"REG_a", 2) \n\t" // R6
2471
"movq %%mm3, (%1, %%"REG_c") \n\t" // R7
2472
"movq %%mm0, (%0, %2, 4) \n\t" // L4
2473
"movq %%mm1, (%0, %%"REG_d") \n\t" // L5
2474
"movq %%mm2, (%0, %%"REG_a", 2) \n\t" // L6
2475
"movq %%mm3, (%0, %%"REG_c") \n\t" // L7
2479
:: "r" (src), "r" (tempBlurred), "r"((long)stride), "m" (tempBlurredPast)
2480
: "%"REG_a, "%"REG_d, "%"REG_c, "memory"
2482
#else //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
2492
int ref= tempBlurred[ x + y*stride ];
2493
int cur= src[ x + y*stride ];
2495
// if(x==0 || x==7) d1+= d1>>1;
2496
// if(y==0 || y==7) d1+= d1>>1;
2505
+(*(tempBlurredPast-256))
2506
+(*(tempBlurredPast-1))+ (*(tempBlurredPast+1))
2507
+(*(tempBlurredPast+256))
2510
// ((*tempBlurredPast)*3 + d + 2)>>2;
2515
64 32 16 8 4 2 1 (1)
2516
64 48 36 27 20 15 11 (33) (approx)
2517
64 56 49 43 37 33 29 (200) (approx)
2519
if(d > maxNoise[1]){
2520
if(d < maxNoise[2]){
2524
int ref= tempBlurred[ x + y*stride ];
2525
int cur= src[ x + y*stride ];
2526
tempBlurred[ x + y*stride ]=
2527
src[ x + y*stride ]=
2535
tempBlurred[ x + y*stride ]= src[ x + y*stride ];
2540
if(d < maxNoise[0]){
2544
int ref= tempBlurred[ x + y*stride ];
2545
int cur= src[ x + y*stride ];
2546
tempBlurred[ x + y*stride ]=
2547
src[ x + y*stride ]=
2548
(ref*7 + cur + 4)>>3;
2555
int ref= tempBlurred[ x + y*stride ];
2556
int cur= src[ x + y*stride ];
2557
tempBlurred[ x + y*stride ]=
2558
src[ x + y*stride ]=
2559
(ref*3 + cur + 2)>>2;
2565
#endif //defined (HAVE_MMX2) || defined (HAVE_3DNOW)
2567
#endif //HAVE_ALTIVEC
2571
* accurate deblock filter
2573
static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, PPContext *c){
2574
int64_t dc_mask, eq_mask, both_masks;
2575
int64_t sums[10*8*2];
2576
src+= step*3; // src points to begin of the 8x8 Block
2579
"movq %0, %%mm7 \n\t"
2580
"movq %1, %%mm6 \n\t"
2581
: : "m" (c->mmxDcOffset[c->nonBQP]), "m" (c->mmxDcThreshold[c->nonBQP])
2585
"lea (%2, %3), %%"REG_a" \n\t"
2586
// 0 1 2 3 4 5 6 7 8 9
2587
// %1 eax eax+%2 eax+2%2 %1+4%2 ecx ecx+%2 ecx+2%2 %1+8%2 ecx+4%2
2589
"movq (%2), %%mm0 \n\t"
2590
"movq (%%"REG_a"), %%mm1 \n\t"
2591
"movq %%mm1, %%mm3 \n\t"
2592
"movq %%mm1, %%mm4 \n\t"
2593
"psubb %%mm1, %%mm0 \n\t" // mm0 = differnece
2594
"paddb %%mm7, %%mm0 \n\t"
2595
"pcmpgtb %%mm6, %%mm0 \n\t"
2597
"movq (%%"REG_a",%3), %%mm2 \n\t"
2598
PMAXUB(%%mm2, %%mm4)
2599
PMINUB(%%mm2, %%mm3, %%mm5)
2600
"psubb %%mm2, %%mm1 \n\t"
2601
"paddb %%mm7, %%mm1 \n\t"
2602
"pcmpgtb %%mm6, %%mm1 \n\t"
2603
"paddb %%mm1, %%mm0 \n\t"
2605
"movq (%%"REG_a", %3, 2), %%mm1 \n\t"
2606
PMAXUB(%%mm1, %%mm4)
2607
PMINUB(%%mm1, %%mm3, %%mm5)
2608
"psubb %%mm1, %%mm2 \n\t"
2609
"paddb %%mm7, %%mm2 \n\t"
2610
"pcmpgtb %%mm6, %%mm2 \n\t"
2611
"paddb %%mm2, %%mm0 \n\t"
2613
"lea (%%"REG_a", %3, 4), %%"REG_a" \n\t"
2615
"movq (%2, %3, 4), %%mm2 \n\t"
2616
PMAXUB(%%mm2, %%mm4)
2617
PMINUB(%%mm2, %%mm3, %%mm5)
2618
"psubb %%mm2, %%mm1 \n\t"
2619
"paddb %%mm7, %%mm1 \n\t"
2620
"pcmpgtb %%mm6, %%mm1 \n\t"
2621
"paddb %%mm1, %%mm0 \n\t"
2623
"movq (%%"REG_a"), %%mm1 \n\t"
2624
PMAXUB(%%mm1, %%mm4)
2625
PMINUB(%%mm1, %%mm3, %%mm5)
2626
"psubb %%mm1, %%mm2 \n\t"
2627
"paddb %%mm7, %%mm2 \n\t"
2628
"pcmpgtb %%mm6, %%mm2 \n\t"
2629
"paddb %%mm2, %%mm0 \n\t"
2631
"movq (%%"REG_a", %3), %%mm2 \n\t"
2632
PMAXUB(%%mm2, %%mm4)
2633
PMINUB(%%mm2, %%mm3, %%mm5)
2634
"psubb %%mm2, %%mm1 \n\t"
2635
"paddb %%mm7, %%mm1 \n\t"
2636
"pcmpgtb %%mm6, %%mm1 \n\t"
2637
"paddb %%mm1, %%mm0 \n\t"
2639
"movq (%%"REG_a", %3, 2), %%mm1 \n\t"
2640
PMAXUB(%%mm1, %%mm4)
2641
PMINUB(%%mm1, %%mm3, %%mm5)
2642
"psubb %%mm1, %%mm2 \n\t"
2643
"paddb %%mm7, %%mm2 \n\t"
2644
"pcmpgtb %%mm6, %%mm2 \n\t"
2645
"paddb %%mm2, %%mm0 \n\t"
2647
"movq (%2, %3, 8), %%mm2 \n\t"
2648
PMAXUB(%%mm2, %%mm4)
2649
PMINUB(%%mm2, %%mm3, %%mm5)
2650
"psubb %%mm2, %%mm1 \n\t"
2651
"paddb %%mm7, %%mm1 \n\t"
2652
"pcmpgtb %%mm6, %%mm1 \n\t"
2653
"paddb %%mm1, %%mm0 \n\t"
2655
"movq (%%"REG_a", %3, 4), %%mm1 \n\t"
2656
"psubb %%mm1, %%mm2 \n\t"
2657
"paddb %%mm7, %%mm2 \n\t"
2658
"pcmpgtb %%mm6, %%mm2 \n\t"
2659
"paddb %%mm2, %%mm0 \n\t"
2660
"psubusb %%mm3, %%mm4 \n\t"
2662
"pxor %%mm6, %%mm6 \n\t"
2663
"movq %4, %%mm7 \n\t" // QP,..., QP
2664
"paddusb %%mm7, %%mm7 \n\t" // 2QP ... 2QP
2665
"psubusb %%mm4, %%mm7 \n\t" // Diff >=2QP -> 0
2666
"pcmpeqb %%mm6, %%mm7 \n\t" // Diff < 2QP -> 0
2667
"pcmpeqb %%mm6, %%mm7 \n\t" // Diff < 2QP -> 0
2668
"movq %%mm7, %1 \n\t"
2670
"movq %5, %%mm7 \n\t"
2671
"punpcklbw %%mm7, %%mm7 \n\t"
2672
"punpcklbw %%mm7, %%mm7 \n\t"
2673
"punpcklbw %%mm7, %%mm7 \n\t"
2674
"psubb %%mm0, %%mm6 \n\t"
2675
"pcmpgtb %%mm7, %%mm6 \n\t"
2676
"movq %%mm6, %0 \n\t"
2678
: "=m" (eq_mask), "=m" (dc_mask)
2679
: "r" (src), "r" ((long)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)
2683
both_masks = dc_mask & eq_mask;
2686
long offset= -8*step;
2687
int64_t *temp_sums= sums;
2690
"movq %2, %%mm0 \n\t" // QP,..., QP
2691
"pxor %%mm4, %%mm4 \n\t"
2693
"movq (%0), %%mm6 \n\t"
2694
"movq (%0, %1), %%mm5 \n\t"
2695
"movq %%mm5, %%mm1 \n\t"
2696
"movq %%mm6, %%mm2 \n\t"
2697
"psubusb %%mm6, %%mm5 \n\t"
2698
"psubusb %%mm1, %%mm2 \n\t"
2699
"por %%mm5, %%mm2 \n\t" // ABS Diff of lines
2700
"psubusb %%mm2, %%mm0 \n\t" // diff >= QP -> 0
2701
"pcmpeqb %%mm4, %%mm0 \n\t" // diff >= QP -> FF
2703
"pxor %%mm6, %%mm1 \n\t"
2704
"pand %%mm0, %%mm1 \n\t"
2705
"pxor %%mm1, %%mm6 \n\t"
2708
"movq (%0, %1, 8), %%mm5 \n\t"
2709
"add %1, %0 \n\t" // %0 points to line 1 not 0
2710
"movq (%0, %1, 8), %%mm7 \n\t"
2711
"movq %%mm5, %%mm1 \n\t"
2712
"movq %%mm7, %%mm2 \n\t"
2713
"psubusb %%mm7, %%mm5 \n\t"
2714
"psubusb %%mm1, %%mm2 \n\t"
2715
"por %%mm5, %%mm2 \n\t" // ABS Diff of lines
2716
"movq %2, %%mm0 \n\t" // QP,..., QP
2717
"psubusb %%mm2, %%mm0 \n\t" // diff >= QP -> 0
2718
"pcmpeqb %%mm4, %%mm0 \n\t" // diff >= QP -> FF
2720
"pxor %%mm7, %%mm1 \n\t"
2721
"pand %%mm0, %%mm1 \n\t"
2722
"pxor %%mm1, %%mm7 \n\t"
2724
"movq %%mm6, %%mm5 \n\t"
2725
"punpckhbw %%mm4, %%mm6 \n\t"
2726
"punpcklbw %%mm4, %%mm5 \n\t"
2727
// 4:0 5/6:First 7:Last
2729
"movq %%mm5, %%mm0 \n\t"
2730
"movq %%mm6, %%mm1 \n\t"
2731
"psllw $2, %%mm0 \n\t"
2732
"psllw $2, %%mm1 \n\t"
2733
"paddw "MANGLE(w04)", %%mm0 \n\t"
2734
"paddw "MANGLE(w04)", %%mm1 \n\t"
2737
"movq (%0), %%mm2 \n\t"\
2738
"movq (%0), %%mm3 \n\t"\
2740
"punpcklbw %%mm4, %%mm2 \n\t"\
2741
"punpckhbw %%mm4, %%mm3 \n\t"\
2742
"paddw %%mm2, %%mm0 \n\t"\
2743
"paddw %%mm3, %%mm1 \n\t"
2746
"movq (%0), %%mm2 \n\t"\
2747
"movq (%0), %%mm3 \n\t"\
2749
"punpcklbw %%mm4, %%mm2 \n\t"\
2750
"punpckhbw %%mm4, %%mm3 \n\t"\
2751
"psubw %%mm2, %%mm0 \n\t"\
2752
"psubw %%mm3, %%mm1 \n\t"
2758
"movq %%mm0, (%3) \n\t"
2759
"movq %%mm1, 8(%3) \n\t"
2762
"psubw %%mm5, %%mm0 \n\t"
2763
"psubw %%mm6, %%mm1 \n\t"
2764
"movq %%mm0, 16(%3) \n\t"
2765
"movq %%mm1, 24(%3) \n\t"
2768
"psubw %%mm5, %%mm0 \n\t"
2769
"psubw %%mm6, %%mm1 \n\t"
2770
"movq %%mm0, 32(%3) \n\t"
2771
"movq %%mm1, 40(%3) \n\t"
2774
"psubw %%mm5, %%mm0 \n\t"
2775
"psubw %%mm6, %%mm1 \n\t"
2776
"movq %%mm0, 48(%3) \n\t"
2777
"movq %%mm1, 56(%3) \n\t"
2780
"psubw %%mm5, %%mm0 \n\t"
2781
"psubw %%mm6, %%mm1 \n\t"
2782
"movq %%mm0, 64(%3) \n\t"
2783
"movq %%mm1, 72(%3) \n\t"
2785
"movq %%mm7, %%mm6 \n\t"
2786
"punpckhbw %%mm4, %%mm7 \n\t"
2787
"punpcklbw %%mm4, %%mm6 \n\t"
2793
"movq %%mm0, 80(%3) \n\t"
2794
"movq %%mm1, 88(%3) \n\t"
2797
"paddw %%mm6, %%mm0 \n\t"
2798
"paddw %%mm7, %%mm1 \n\t"
2799
"movq %%mm0, 96(%3) \n\t"
2800
"movq %%mm1, 104(%3) \n\t"
2803
"paddw %%mm6, %%mm0 \n\t"
2804
"paddw %%mm7, %%mm1 \n\t"
2805
"movq %%mm0, 112(%3) \n\t"
2806
"movq %%mm1, 120(%3) \n\t"
2809
"paddw %%mm6, %%mm0 \n\t"
2810
"paddw %%mm7, %%mm1 \n\t"
2811
"movq %%mm0, 128(%3) \n\t"
2812
"movq %%mm1, 136(%3) \n\t"
2815
"paddw %%mm6, %%mm0 \n\t"
2816
"paddw %%mm7, %%mm1 \n\t"
2817
"movq %%mm0, 144(%3) \n\t"
2818
"movq %%mm1, 152(%3) \n\t"
2820
"mov %4, %0 \n\t" //FIXME
2823
: "r" ((long)step), "m" (c->pQPb), "r"(sums), "g"(src)
2826
src+= step; // src points to begin of the 8x8 Block
2829
"movq %4, %%mm6 \n\t"
2830
"pcmpeqb %%mm5, %%mm5 \n\t"
2831
"pxor %%mm6, %%mm5 \n\t"
2832
"pxor %%mm7, %%mm7 \n\t"
2835
"movq (%1), %%mm0 \n\t"
2836
"movq 8(%1), %%mm1 \n\t"
2837
"paddw 32(%1), %%mm0 \n\t"
2838
"paddw 40(%1), %%mm1 \n\t"
2839
"movq (%0, %3), %%mm2 \n\t"
2840
"movq %%mm2, %%mm3 \n\t"
2841
"movq %%mm2, %%mm4 \n\t"
2842
"punpcklbw %%mm7, %%mm2 \n\t"
2843
"punpckhbw %%mm7, %%mm3 \n\t"
2844
"paddw %%mm2, %%mm0 \n\t"
2845
"paddw %%mm3, %%mm1 \n\t"
2846
"paddw %%mm2, %%mm0 \n\t"
2847
"paddw %%mm3, %%mm1 \n\t"
2848
"psrlw $4, %%mm0 \n\t"
2849
"psrlw $4, %%mm1 \n\t"
2850
"packuswb %%mm1, %%mm0 \n\t"
2851
"pand %%mm6, %%mm0 \n\t"
2852
"pand %%mm5, %%mm4 \n\t"
2853
"por %%mm4, %%mm0 \n\t"
2854
"movq %%mm0, (%0, %3) \n\t"
2859
: "+r"(offset), "+r"(temp_sums)
2860
: "r" ((long)step), "r"(src - offset), "m"(both_masks)
2863
src+= step; // src points to begin of the 8x8 Block
2865
if(eq_mask != -1LL){
2866
uint8_t *temp_src= src;
2868
"pxor %%mm7, %%mm7 \n\t"
2869
"lea -40(%%"REG_SP"), %%"REG_c" \n\t" // make space for 4 8-byte vars
2870
"and "ALIGN_MASK", %%"REG_c" \n\t" // align
2871
// 0 1 2 3 4 5 6 7 8 9
2872
// %0 eax eax+%1 eax+2%1 %0+4%1 ecx ecx+%1 ecx+2%1 %1+8%1 ecx+4%1
2874
"movq (%0), %%mm0 \n\t"
2875
"movq %%mm0, %%mm1 \n\t"
2876
"punpcklbw %%mm7, %%mm0 \n\t" // low part of line 0
2877
"punpckhbw %%mm7, %%mm1 \n\t" // high part of line 0
2879
"movq (%0, %1), %%mm2 \n\t"
2880
"lea (%0, %1, 2), %%"REG_a" \n\t"
2881
"movq %%mm2, %%mm3 \n\t"
2882
"punpcklbw %%mm7, %%mm2 \n\t" // low part of line 1
2883
"punpckhbw %%mm7, %%mm3 \n\t" // high part of line 1
2885
"movq (%%"REG_a"), %%mm4 \n\t"
2886
"movq %%mm4, %%mm5 \n\t"
2887
"punpcklbw %%mm7, %%mm4 \n\t" // low part of line 2
2888
"punpckhbw %%mm7, %%mm5 \n\t" // high part of line 2
2890
"paddw %%mm0, %%mm0 \n\t" // 2L0
2891
"paddw %%mm1, %%mm1 \n\t" // 2H0
2892
"psubw %%mm4, %%mm2 \n\t" // L1 - L2
2893
"psubw %%mm5, %%mm3 \n\t" // H1 - H2
2894
"psubw %%mm2, %%mm0 \n\t" // 2L0 - L1 + L2
2895
"psubw %%mm3, %%mm1 \n\t" // 2H0 - H1 + H2
2897
"psllw $2, %%mm2 \n\t" // 4L1 - 4L2
2898
"psllw $2, %%mm3 \n\t" // 4H1 - 4H2
2899
"psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2
2900
"psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2
2902
"movq (%%"REG_a", %1), %%mm2 \n\t"
2903
"movq %%mm2, %%mm3 \n\t"
2904
"punpcklbw %%mm7, %%mm2 \n\t" // L3
2905
"punpckhbw %%mm7, %%mm3 \n\t" // H3
2907
"psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - L3
2908
"psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - H3
2909
"psubw %%mm2, %%mm0 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2910
"psubw %%mm3, %%mm1 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2911
"movq %%mm0, (%%"REG_c") \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2912
"movq %%mm1, 8(%%"REG_c") \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2914
"movq (%%"REG_a", %1, 2), %%mm0 \n\t"
2915
"movq %%mm0, %%mm1 \n\t"
2916
"punpcklbw %%mm7, %%mm0 \n\t" // L4
2917
"punpckhbw %%mm7, %%mm1 \n\t" // H4
2919
"psubw %%mm0, %%mm2 \n\t" // L3 - L4
2920
"psubw %%mm1, %%mm3 \n\t" // H3 - H4
2921
"movq %%mm2, 16(%%"REG_c") \n\t" // L3 - L4
2922
"movq %%mm3, 24(%%"REG_c") \n\t" // H3 - H4
2923
"paddw %%mm4, %%mm4 \n\t" // 2L2
2924
"paddw %%mm5, %%mm5 \n\t" // 2H2
2925
"psubw %%mm2, %%mm4 \n\t" // 2L2 - L3 + L4
2926
"psubw %%mm3, %%mm5 \n\t" // 2H2 - H3 + H4
2928
"lea (%%"REG_a", %1), %0 \n\t"
2929
"psllw $2, %%mm2 \n\t" // 4L3 - 4L4
2930
"psllw $2, %%mm3 \n\t" // 4H3 - 4H4
2931
"psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4
2932
"psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4
2934
"movq (%0, %1, 2), %%mm2 \n\t"
2935
"movq %%mm2, %%mm3 \n\t"
2936
"punpcklbw %%mm7, %%mm2 \n\t" // L5
2937
"punpckhbw %%mm7, %%mm3 \n\t" // H5
2938
"psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - L5
2939
"psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - H5
2940
"psubw %%mm2, %%mm4 \n\t" // 2L2 - 5L3 + 5L4 - 2L5
2941
"psubw %%mm3, %%mm5 \n\t" // 2H2 - 5H3 + 5H4 - 2H5
2943
"movq (%%"REG_a", %1, 4), %%mm6 \n\t"
2944
"punpcklbw %%mm7, %%mm6 \n\t" // L6
2945
"psubw %%mm6, %%mm2 \n\t" // L5 - L6
2946
"movq (%%"REG_a", %1, 4), %%mm6 \n\t"
2947
"punpckhbw %%mm7, %%mm6 \n\t" // H6
2948
"psubw %%mm6, %%mm3 \n\t" // H5 - H6
2950
"paddw %%mm0, %%mm0 \n\t" // 2L4
2951
"paddw %%mm1, %%mm1 \n\t" // 2H4
2952
"psubw %%mm2, %%mm0 \n\t" // 2L4 - L5 + L6
2953
"psubw %%mm3, %%mm1 \n\t" // 2H4 - H5 + H6
2955
"psllw $2, %%mm2 \n\t" // 4L5 - 4L6
2956
"psllw $2, %%mm3 \n\t" // 4H5 - 4H6
2957
"psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6
2958
"psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6
2960
"movq (%0, %1, 4), %%mm2 \n\t"
2961
"movq %%mm2, %%mm3 \n\t"
2962
"punpcklbw %%mm7, %%mm2 \n\t" // L7
2963
"punpckhbw %%mm7, %%mm3 \n\t" // H7
2965
"paddw %%mm2, %%mm2 \n\t" // 2L7
2966
"paddw %%mm3, %%mm3 \n\t" // 2H7
2967
"psubw %%mm2, %%mm0 \n\t" // 2L4 - 5L5 + 5L6 - 2L7
2968
"psubw %%mm3, %%mm1 \n\t" // 2H4 - 5H5 + 5H6 - 2H7
2970
"movq (%%"REG_c"), %%mm2 \n\t" // 2L0 - 5L1 + 5L2 - 2L3
2971
"movq 8(%%"REG_c"), %%mm3 \n\t" // 2H0 - 5H1 + 5H2 - 2H3
2974
"movq %%mm7, %%mm6 \n\t" // 0
2975
"psubw %%mm0, %%mm6 \n\t"
2976
"pmaxsw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
2977
"movq %%mm7, %%mm6 \n\t" // 0
2978
"psubw %%mm1, %%mm6 \n\t"
2979
"pmaxsw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
2980
"movq %%mm7, %%mm6 \n\t" // 0
2981
"psubw %%mm2, %%mm6 \n\t"
2982
"pmaxsw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
2983
"movq %%mm7, %%mm6 \n\t" // 0
2984
"psubw %%mm3, %%mm6 \n\t"
2985
"pmaxsw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
2987
"movq %%mm7, %%mm6 \n\t" // 0
2988
"pcmpgtw %%mm0, %%mm6 \n\t"
2989
"pxor %%mm6, %%mm0 \n\t"
2990
"psubw %%mm6, %%mm0 \n\t" // |2L4 - 5L5 + 5L6 - 2L7|
2991
"movq %%mm7, %%mm6 \n\t" // 0
2992
"pcmpgtw %%mm1, %%mm6 \n\t"
2993
"pxor %%mm6, %%mm1 \n\t"
2994
"psubw %%mm6, %%mm1 \n\t" // |2H4 - 5H5 + 5H6 - 2H7|
2995
"movq %%mm7, %%mm6 \n\t" // 0
2996
"pcmpgtw %%mm2, %%mm6 \n\t"
2997
"pxor %%mm6, %%mm2 \n\t"
2998
"psubw %%mm6, %%mm2 \n\t" // |2L0 - 5L1 + 5L2 - 2L3|
2999
"movq %%mm7, %%mm6 \n\t" // 0
3000
"pcmpgtw %%mm3, %%mm6 \n\t"
3001
"pxor %%mm6, %%mm3 \n\t"
3002
"psubw %%mm6, %%mm3 \n\t" // |2H0 - 5H1 + 5H2 - 2H3|
3006
"pminsw %%mm2, %%mm0 \n\t"
3007
"pminsw %%mm3, %%mm1 \n\t"
3009
"movq %%mm0, %%mm6 \n\t"
3010
"psubusw %%mm2, %%mm6 \n\t"
3011
"psubw %%mm6, %%mm0 \n\t"
3012
"movq %%mm1, %%mm6 \n\t"
3013
"psubusw %%mm3, %%mm6 \n\t"
3014
"psubw %%mm6, %%mm1 \n\t"
3017
"movd %2, %%mm2 \n\t" // QP
3018
"punpcklbw %%mm7, %%mm2 \n\t"
3020
"movq %%mm7, %%mm6 \n\t" // 0
3021
"pcmpgtw %%mm4, %%mm6 \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)
3022
"pxor %%mm6, %%mm4 \n\t"
3023
"psubw %%mm6, %%mm4 \n\t" // |2L2 - 5L3 + 5L4 - 2L5|
3024
"pcmpgtw %%mm5, %%mm7 \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)
3025
"pxor %%mm7, %%mm5 \n\t"
3026
"psubw %%mm7, %%mm5 \n\t" // |2H2 - 5H3 + 5H4 - 2H5|
3028
"psllw $3, %%mm2 \n\t" // 8QP
3029
"movq %%mm2, %%mm3 \n\t" // 8QP
3030
"pcmpgtw %%mm4, %%mm2 \n\t"
3031
"pcmpgtw %%mm5, %%mm3 \n\t"
3032
"pand %%mm2, %%mm4 \n\t"
3033
"pand %%mm3, %%mm5 \n\t"
3036
"psubusw %%mm0, %%mm4 \n\t" // hd
3037
"psubusw %%mm1, %%mm5 \n\t" // ld
3040
"movq "MANGLE(w05)", %%mm2 \n\t" // 5
3041
"pmullw %%mm2, %%mm4 \n\t"
3042
"pmullw %%mm2, %%mm5 \n\t"
3043
"movq "MANGLE(w20)", %%mm2 \n\t" // 32
3044
"paddw %%mm2, %%mm4 \n\t"
3045
"paddw %%mm2, %%mm5 \n\t"
3046
"psrlw $6, %%mm4 \n\t"
3047
"psrlw $6, %%mm5 \n\t"
3049
"movq 16(%%"REG_c"), %%mm0 \n\t" // L3 - L4
3050
"movq 24(%%"REG_c"), %%mm1 \n\t" // H3 - H4
3052
"pxor %%mm2, %%mm2 \n\t"
3053
"pxor %%mm3, %%mm3 \n\t"
3055
"pcmpgtw %%mm0, %%mm2 \n\t" // sign (L3-L4)
3056
"pcmpgtw %%mm1, %%mm3 \n\t" // sign (H3-H4)
3057
"pxor %%mm2, %%mm0 \n\t"
3058
"pxor %%mm3, %%mm1 \n\t"
3059
"psubw %%mm2, %%mm0 \n\t" // |L3-L4|
3060
"psubw %%mm3, %%mm1 \n\t" // |H3-H4|
3061
"psrlw $1, %%mm0 \n\t" // |L3 - L4|/2
3062
"psrlw $1, %%mm1 \n\t" // |H3 - H4|/2
3064
"pxor %%mm6, %%mm2 \n\t"
3065
"pxor %%mm7, %%mm3 \n\t"
3066
"pand %%mm2, %%mm4 \n\t"
3067
"pand %%mm3, %%mm5 \n\t"
3070
"pminsw %%mm0, %%mm4 \n\t"
3071
"pminsw %%mm1, %%mm5 \n\t"
3073
"movq %%mm4, %%mm2 \n\t"
3074
"psubusw %%mm0, %%mm2 \n\t"
3075
"psubw %%mm2, %%mm4 \n\t"
3076
"movq %%mm5, %%mm2 \n\t"
3077
"psubusw %%mm1, %%mm2 \n\t"
3078
"psubw %%mm2, %%mm5 \n\t"
3080
"pxor %%mm6, %%mm4 \n\t"
3081
"pxor %%mm7, %%mm5 \n\t"
3082
"psubw %%mm6, %%mm4 \n\t"
3083
"psubw %%mm7, %%mm5 \n\t"
3084
"packsswb %%mm5, %%mm4 \n\t"
3085
"movq %3, %%mm1 \n\t"
3086
"pandn %%mm4, %%mm1 \n\t"
3087
"movq (%0), %%mm0 \n\t"
3088
"paddb %%mm1, %%mm0 \n\t"
3089
"movq %%mm0, (%0) \n\t"
3090
"movq (%0, %1), %%mm0 \n\t"
3091
"psubb %%mm1, %%mm0 \n\t"
3092
"movq %%mm0, (%0, %1) \n\t"
3095
: "r" ((long)step), "m" (c->pQPb), "m"(eq_mask)
3096
: "%"REG_a, "%"REG_c
3100
STOP_TIMER("step16")
3107
static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
3108
const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);
3111
* Copies a block from src to dst and fixes the blacklevel.
3112
* levelFix == 0 -> do not touch the brighness & contrast
3116
static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t src[], int srcStride,
3117
int levelFix, int64_t *packedOffsetAndScale)
3125
"movq (%%"REG_a"), %%mm2 \n\t" // packedYOffset
3126
"movq 8(%%"REG_a"), %%mm3 \n\t" // packedYScale
3127
"lea (%2,%4), %%"REG_a" \n\t"
3128
"lea (%3,%5), %%"REG_d" \n\t"
3129
"pxor %%mm4, %%mm4 \n\t"
3131
#define REAL_SCALED_CPY(src1, src2, dst1, dst2) \
3132
"movq " #src1 ", %%mm0 \n\t"\
3133
"movq " #src1 ", %%mm5 \n\t"\
3134
"movq " #src2 ", %%mm1 \n\t"\
3135
"movq " #src2 ", %%mm6 \n\t"\
3136
"punpcklbw %%mm0, %%mm0 \n\t"\
3137
"punpckhbw %%mm5, %%mm5 \n\t"\
3138
"punpcklbw %%mm1, %%mm1 \n\t"\
3139
"punpckhbw %%mm6, %%mm6 \n\t"\
3140
"pmulhuw %%mm3, %%mm0 \n\t"\
3141
"pmulhuw %%mm3, %%mm5 \n\t"\
3142
"pmulhuw %%mm3, %%mm1 \n\t"\
3143
"pmulhuw %%mm3, %%mm6 \n\t"\
3144
"psubw %%mm2, %%mm0 \n\t"\
3145
"psubw %%mm2, %%mm5 \n\t"\
3146
"psubw %%mm2, %%mm1 \n\t"\
3147
"psubw %%mm2, %%mm6 \n\t"\
3148
"packuswb %%mm5, %%mm0 \n\t"\
3149
"packuswb %%mm6, %%mm1 \n\t"\
3150
"movq %%mm0, " #dst1 " \n\t"\
3151
"movq %%mm1, " #dst2 " \n\t"\
3154
#define REAL_SCALED_CPY(src1, src2, dst1, dst2) \
3155
"movq " #src1 ", %%mm0 \n\t"\
3156
"movq " #src1 ", %%mm5 \n\t"\
3157
"punpcklbw %%mm4, %%mm0 \n\t"\
3158
"punpckhbw %%mm4, %%mm5 \n\t"\
3159
"psubw %%mm2, %%mm0 \n\t"\
3160
"psubw %%mm2, %%mm5 \n\t"\
3161
"movq " #src2 ", %%mm1 \n\t"\
3162
"psllw $6, %%mm0 \n\t"\
3163
"psllw $6, %%mm5 \n\t"\
3164
"pmulhw %%mm3, %%mm0 \n\t"\
3165
"movq " #src2 ", %%mm6 \n\t"\
3166
"pmulhw %%mm3, %%mm5 \n\t"\
3167
"punpcklbw %%mm4, %%mm1 \n\t"\
3168
"punpckhbw %%mm4, %%mm6 \n\t"\
3169
"psubw %%mm2, %%mm1 \n\t"\
3170
"psubw %%mm2, %%mm6 \n\t"\
3171
"psllw $6, %%mm1 \n\t"\
3172
"psllw $6, %%mm6 \n\t"\
3173
"pmulhw %%mm3, %%mm1 \n\t"\
3174
"pmulhw %%mm3, %%mm6 \n\t"\
3175
"packuswb %%mm5, %%mm0 \n\t"\
3176
"packuswb %%mm6, %%mm1 \n\t"\
3177
"movq %%mm0, " #dst1 " \n\t"\
3178
"movq %%mm1, " #dst2 " \n\t"\
3181
#define SCALED_CPY(src1, src2, dst1, dst2)\
3182
REAL_SCALED_CPY(src1, src2, dst1, dst2)
3184
SCALED_CPY((%2) , (%2, %4) , (%3) , (%3, %5))
3185
SCALED_CPY((%2, %4, 2), (%%REGa, %4, 2), (%3, %5, 2), (%%REGd, %5, 2))
3186
SCALED_CPY((%2, %4, 4), (%%REGa, %4, 4), (%3, %5, 4), (%%REGd, %5, 4))
3187
"lea (%%"REG_a",%4,4), %%"REG_a" \n\t"
3188
"lea (%%"REG_d",%5,4), %%"REG_d" \n\t"
3189
SCALED_CPY((%%REGa, %4), (%%REGa, %4, 2), (%%REGd, %5), (%%REGd, %5, 2))
3192
: "=&a" (packedOffsetAndScale)
3193
: "0" (packedOffsetAndScale),
3196
"r" ((long)srcStride),
3197
"r" ((long)dstStride)
3202
memcpy( &(dst[dstStride*i]),
3203
&(src[srcStride*i]), BLOCK_SIZE);
3208
"lea (%0,%2), %%"REG_a" \n\t"
3209
"lea (%1,%3), %%"REG_d" \n\t"
3211
#define REAL_SIMPLE_CPY(src1, src2, dst1, dst2) \
3212
"movq " #src1 ", %%mm0 \n\t"\
3213
"movq " #src2 ", %%mm1 \n\t"\
3214
"movq %%mm0, " #dst1 " \n\t"\
3215
"movq %%mm1, " #dst2 " \n\t"\
3217
#define SIMPLE_CPY(src1, src2, dst1, dst2)\
3218
REAL_SIMPLE_CPY(src1, src2, dst1, dst2)
3220
SIMPLE_CPY((%0) , (%0, %2) , (%1) , (%1, %3))
3221
SIMPLE_CPY((%0, %2, 2), (%%REGa, %2, 2), (%1, %3, 2), (%%REGd, %3, 2))
3222
SIMPLE_CPY((%0, %2, 4), (%%REGa, %2, 4), (%1, %3, 4), (%%REGd, %3, 4))
3223
"lea (%%"REG_a",%2,4), %%"REG_a" \n\t"
3224
"lea (%%"REG_d",%3,4), %%"REG_d" \n\t"
3225
SIMPLE_CPY((%%REGa, %2), (%%REGa, %2, 2), (%%REGd, %3), (%%REGd, %3, 2))
3229
"r" ((long)srcStride),
3230
"r" ((long)dstStride)
3231
: "%"REG_a, "%"REG_d
3235
memcpy( &(dst[dstStride*i]),
3236
&(src[srcStride*i]), BLOCK_SIZE);
3242
* Duplicates the given 8 src pixels ? times upward
3244
static inline void RENAME(duplicate)(uint8_t src[], int stride)
3248
"movq (%0), %%mm0 \n\t"
3250
"movq %%mm0, (%0) \n\t"
3251
"movq %%mm0, (%0, %1) \n\t"
3252
"movq %%mm0, (%0, %1, 2) \n\t"
3254
: "r" ((long)-stride)
3267
* Filters array of bytes (Y or U or V values)
3269
static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
3270
const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)
3272
DECLARE_ALIGNED(8, PPContext, c)= *c2; //copy to stack for faster access
3274
#ifdef COMPILE_TIME_MODE
3275
const int mode= COMPILE_TIME_MODE;
3277
const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;
3279
int black=0, white=255; // blackest black and whitest white in the picture
3280
int QPCorrecture= 256*256;
3287
const int qpHShift= isColor ? 4-c.hChromaSubSample : 4;
3288
const int qpVShift= isColor ? 4-c.vChromaSubSample : 4;
3291
uint64_t * const yHistogram= c.yHistogram;
3292
uint8_t * const tempSrc= srcStride > 0 ? c.tempSrc : c.tempSrc - 23*srcStride;
3293
uint8_t * const tempDst= dstStride > 0 ? c.tempDst : c.tempDst - 23*dstStride;
3294
//const int mbWidth= isColor ? (width+7)>>3 : (width+15)>>4;
3297
for(i=0; i<57; i++){
3298
int offset= ((i*c.ppMode.baseDcDiff)>>8) + 1;
3299
int threshold= offset*2 + 1;
3300
c.mmxDcOffset[i]= 0x7F - offset;
3301
c.mmxDcThreshold[i]= 0x7F - threshold;
3302
c.mmxDcOffset[i]*= 0x0101010101010101LL;
3303
c.mmxDcThreshold[i]*= 0x0101010101010101LL;
3307
if(mode & CUBIC_IPOL_DEINT_FILTER) copyAhead=16;
3308
else if( (mode & LINEAR_BLEND_DEINT_FILTER)
3309
|| (mode & FFMPEG_DEINT_FILTER)
3310
|| (mode & LOWPASS5_DEINT_FILTER)) copyAhead=14;
3311
else if( (mode & V_DEBLOCK)
3312
|| (mode & LINEAR_IPOL_DEINT_FILTER)
3313
|| (mode & MEDIAN_DEINT_FILTER)
3314
|| (mode & V_A_DEBLOCK)) copyAhead=13;
3315
else if(mode & V_X1_FILTER) copyAhead=11;
3316
// else if(mode & V_RK1_FILTER) copyAhead=10;
3317
else if(mode & DERING) copyAhead=9;
3325
uint64_t maxClipped;
3330
// first frame is fscked so we ignore it
3331
if(c.frameNum == 1) yHistogram[0]= width*height/64*15/256;
3333
for(i=0; i<256; i++){
3334
sum+= yHistogram[i];
3337
/* We always get a completely black picture first. */
3338
maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);
3341
for(black=255; black>0; black--){
3342
if(clipped < maxClipped) break;
3343
clipped-= yHistogram[black];
3347
for(white=0; white<256; white++){
3348
if(clipped < maxClipped) break;
3349
clipped-= yHistogram[white];
3352
scale= (double)(c.ppMode.maxAllowedY - c.ppMode.minAllowedY) / (double)(white-black);
3355
c.packedYScale= (uint16_t)(scale*256.0 + 0.5);
3356
c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF;
3358
c.packedYScale= (uint16_t)(scale*1024.0 + 0.5);
3359
c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF;
3362
c.packedYOffset|= c.packedYOffset<<32;
3363
c.packedYOffset|= c.packedYOffset<<16;
3365
c.packedYScale|= c.packedYScale<<32;
3366
c.packedYScale|= c.packedYScale<<16;
3368
if(mode & LEVEL_FIX) QPCorrecture= (int)(scale*256*256 + 0.5);
3369
else QPCorrecture= 256*256;
3371
c.packedYScale= 0x0100010001000100LL;
3373
QPCorrecture= 256*256;
3376
/* copy & deinterlace first row of blocks */
3379
const uint8_t *srcBlock= &(src[y*srcStride]);
3380
uint8_t *dstBlock= tempDst + dstStride;
3382
// From this point on it is guaranteed that we can read and write 16 lines downward
3383
// finish 1 block before the next otherwise we might have a problem
3384
// with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
3385
for(x=0; x<width; x+=BLOCK_SIZE){
3389
prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
3390
prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
3391
prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
3392
prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
3396
"mov %4, %%"REG_a" \n\t"
3397
"shr $2, %%"REG_a" \n\t"
3398
"and $6, %%"REG_a" \n\t"
3399
"add %5, %%"REG_a" \n\t"
3400
"mov %%"REG_a", %%"REG_d" \n\t"
3401
"imul %1, %%"REG_a" \n\t"
3402
"imul %3, %%"REG_d" \n\t"
3403
"prefetchnta 32(%%"REG_a", %0) \n\t"
3404
"prefetcht0 32(%%"REG_d", %2) \n\t"
3405
"add %1, %%"REG_a" \n\t"
3406
"add %3, %%"REG_d" \n\t"
3407
"prefetchnta 32(%%"REG_a", %0) \n\t"
3408
"prefetcht0 32(%%"REG_d", %2) \n\t"
3409
:: "r" (srcBlock), "r" ((long)srcStride), "r" (dstBlock), "r" ((long)dstStride),
3410
"g" ((long)x), "g" ((long)copyAhead)
3411
: "%"REG_a, "%"REG_d
3414
#elif defined(HAVE_3DNOW)
3415
//FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
3416
/* prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
3417
prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
3418
prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
3419
prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
3423
RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,
3424
srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
3426
RENAME(duplicate)(dstBlock + dstStride*8, dstStride);
3428
if(mode & LINEAR_IPOL_DEINT_FILTER)
3429
RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
3430
else if(mode & LINEAR_BLEND_DEINT_FILTER)
3431
RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
3432
else if(mode & MEDIAN_DEINT_FILTER)
3433
RENAME(deInterlaceMedian)(dstBlock, dstStride);
3434
else if(mode & CUBIC_IPOL_DEINT_FILTER)
3435
RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
3436
else if(mode & FFMPEG_DEINT_FILTER)
3437
RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
3438
else if(mode & LOWPASS5_DEINT_FILTER)
3439
RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
3440
/* else if(mode & CUBIC_BLEND_DEINT_FILTER)
3441
RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3446
if(width==FFABS(dstStride))
3447
linecpy(dst, tempDst + 9*dstStride, copyAhead, dstStride);
3450
for(i=0; i<copyAhead; i++){
3451
memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);
3456
for(y=0; y<height; y+=BLOCK_SIZE){
3457
//1% speedup if these are here instead of the inner loop
3458
const uint8_t *srcBlock= &(src[y*srcStride]);
3459
uint8_t *dstBlock= &(dst[y*dstStride]);
3461
uint8_t *tempBlock1= c.tempBlocks;
3462
uint8_t *tempBlock2= c.tempBlocks + 8;
3464
const int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];
3465
int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*FFABS(QPStride)];
3467
/* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards
3468
if not than use a temporary buffer */
3471
/* copy from line (copyAhead) to (copyAhead+7) of src, these will be copied with
3472
blockcopy to dst later */
3473
linecpy(tempSrc + srcStride*copyAhead, srcBlock + srcStride*copyAhead,
3474
FFMAX(height-y-copyAhead, 0), srcStride);
3476
/* duplicate last line of src to fill the void upto line (copyAhead+7) */
3477
for(i=FFMAX(height-y, 8); i<copyAhead+8; i++)
3478
memcpy(tempSrc + srcStride*i, src + srcStride*(height-1), FFABS(srcStride));
3480
/* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/
3481
linecpy(tempDst, dstBlock - dstStride, FFMIN(height-y+1, copyAhead+1), dstStride);
3483
/* duplicate last line of dst to fill the void upto line (copyAhead) */
3484
for(i=height-y+1; i<=copyAhead; i++)
3485
memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), FFABS(dstStride));
3487
dstBlock= tempDst + dstStride;
3491
// From this point on it is guaranteed that we can read and write 16 lines downward
3492
// finish 1 block before the next otherwise we might have a problem
3493
// with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing
3494
for(x=0; x<width; x+=BLOCK_SIZE){
3495
const int stride= dstStride;
3500
QP= QPptr[x>>qpHShift];
3501
c.nonBQP= nonBQPptr[x>>qpHShift];
3504
QP= (QP* QPCorrecture + 256*128)>>16;
3505
c.nonBQP= nonBQPptr[x>>4];
3506
c.nonBQP= (c.nonBQP* QPCorrecture + 256*128)>>16;
3507
yHistogram[ srcBlock[srcStride*12 + 4] ]++;
3512
"movd %1, %%mm7 \n\t"
3513
"packuswb %%mm7, %%mm7 \n\t" // 0, 0, 0, QP, 0, 0, 0, QP
3514
"packuswb %%mm7, %%mm7 \n\t" // 0,QP, 0, QP, 0,QP, 0, QP
3515
"packuswb %%mm7, %%mm7 \n\t" // QP,..., QP
3516
"movq %%mm7, %0 \n\t"
3525
prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);
3526
prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);
3527
prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);
3528
prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);
3532
"mov %4, %%"REG_a" \n\t"
3533
"shr $2, %%"REG_a" \n\t"
3534
"and $6, %%"REG_a" \n\t"
3535
"add %5, %%"REG_a" \n\t"
3536
"mov %%"REG_a", %%"REG_d" \n\t"
3537
"imul %1, %%"REG_a" \n\t"
3538
"imul %3, %%"REG_d" \n\t"
3539
"prefetchnta 32(%%"REG_a", %0) \n\t"
3540
"prefetcht0 32(%%"REG_d", %2) \n\t"
3541
"add %1, %%"REG_a" \n\t"
3542
"add %3, %%"REG_d" \n\t"
3543
"prefetchnta 32(%%"REG_a", %0) \n\t"
3544
"prefetcht0 32(%%"REG_d", %2) \n\t"
3545
:: "r" (srcBlock), "r" ((long)srcStride), "r" (dstBlock), "r" ((long)dstStride),
3546
"g" ((long)x), "g" ((long)copyAhead)
3547
: "%"REG_a, "%"REG_d
3550
#elif defined(HAVE_3DNOW)
3551
//FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...
3552
/* prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);
3553
prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);
3554
prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);
3555
prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);
3559
RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,
3560
srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);
3562
if(mode & LINEAR_IPOL_DEINT_FILTER)
3563
RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);
3564
else if(mode & LINEAR_BLEND_DEINT_FILTER)
3565
RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);
3566
else if(mode & MEDIAN_DEINT_FILTER)
3567
RENAME(deInterlaceMedian)(dstBlock, dstStride);
3568
else if(mode & CUBIC_IPOL_DEINT_FILTER)
3569
RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);
3570
else if(mode & FFMPEG_DEINT_FILTER)
3571
RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);
3572
else if(mode & LOWPASS5_DEINT_FILTER)
3573
RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);
3574
/* else if(mode & CUBIC_BLEND_DEINT_FILTER)
3575
RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);
3578
/* only deblock if we have 2 blocks */
3580
if(mode & V_X1_FILTER)
3581
RENAME(vertX1Filter)(dstBlock, stride, &c);
3582
else if(mode & V_DEBLOCK){
3583
const int t= RENAME(vertClassify)(dstBlock, stride, &c);
3586
RENAME(doVertLowPass)(dstBlock, stride, &c);
3588
RENAME(doVertDefFilter)(dstBlock, stride, &c);
3589
}else if(mode & V_A_DEBLOCK){
3590
RENAME(do_a_deblock)(dstBlock, stride, 1, &c);
3595
RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);
3597
/* check if we have a previous block to deblock it with dstBlock */
3600
if(mode & H_X1_FILTER)
3601
RENAME(vertX1Filter)(tempBlock1, 16, &c);
3602
else if(mode & H_DEBLOCK){
3604
const int t= RENAME(vertClassify)(tempBlock1, 16, &c);
3605
//STOP_TIMER("dc & minmax")
3607
RENAME(doVertLowPass)(tempBlock1, 16, &c);
3609
RENAME(doVertDefFilter)(tempBlock1, 16, &c);
3610
}else if(mode & H_A_DEBLOCK){
3611
RENAME(do_a_deblock)(tempBlock1, 16, 1, &c);
3614
RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16);
3617
if(mode & H_X1_FILTER)
3618
horizX1Filter(dstBlock-4, stride, QP);
3619
else if(mode & H_DEBLOCK){
3621
DECLARE_ALIGNED(16, unsigned char, tempBlock[272]);
3622
transpose_16x8_char_toPackedAlign_altivec(tempBlock, dstBlock - (4 + 1), stride);
3624
const int t=vertClassify_altivec(tempBlock-48, 16, &c);
3626
doVertLowPass_altivec(tempBlock-48, 16, &c);
3627
transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
3630
doVertDefFilter_altivec(tempBlock-48, 16, &c);
3631
transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);
3634
const int t= RENAME(horizClassify)(dstBlock-4, stride, &c);
3637
RENAME(doHorizLowPass)(dstBlock-4, stride, &c);
3639
RENAME(doHorizDefFilter)(dstBlock-4, stride, &c);
3641
}else if(mode & H_A_DEBLOCK){
3642
RENAME(do_a_deblock)(dstBlock-8, 1, stride, &c);
3646
//FIXME filter first line
3647
if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);
3650
if(mode & TEMP_NOISE_FILTER)
3652
RENAME(tempNoiseReducer)(dstBlock-8, stride,
3653
c.tempBlurred[isColor] + y*dstStride + x,
3654
c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3),
3655
c.ppMode.maxTmpNoise);
3663
tmpXchg= tempBlock1;
3664
tempBlock1= tempBlock2;
3665
tempBlock2 = tmpXchg;
3670
if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, &c);
3673
if((mode & TEMP_NOISE_FILTER)){
3674
RENAME(tempNoiseReducer)(dstBlock-8, dstStride,
3675
c.tempBlurred[isColor] + y*dstStride + x,
3676
c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3),
3677
c.ppMode.maxTmpNoise);
3680
/* did we use a tmp buffer for the last lines*/
3682
uint8_t *dstBlock= &(dst[y*dstStride]);
3683
if(width==FFABS(dstStride))
3684
linecpy(dstBlock, tempDst + dstStride, height-y, dstStride);
3687
for(i=0; i<height-y; i++){
3688
memcpy(dstBlock + i*dstStride, tempDst + (i+1)*dstStride, width);
3693
for(x=0; x<width; x+=32){
3695
i+= + dstBlock[x + 7*dstStride] + dstBlock[x + 8*dstStride]
3696
+ dstBlock[x + 9*dstStride] + dstBlock[x +10*dstStride]
3697
+ dstBlock[x +11*dstStride] + dstBlock[x +12*dstStride];
3698
+ dstBlock[x +13*dstStride]
3699
+ dstBlock[x +14*dstStride] + dstBlock[x +15*dstStride];
3703
asm volatile("femms");
3704
#elif defined (HAVE_MMX)
3705
asm volatile("emms");
3708
#ifdef DEBUG_BRIGHTNESS
3712
for(i=0; i<256; i++)
3713
if(yHistogram[i] > max) max=yHistogram[i];
3715
for(i=1; i<256; i++){
3717
int start=yHistogram[i-1]/(max/256+1);
3718
int end=yHistogram[i]/(max/256+1);
3719
int inc= end > start ? 1 : -1;
3720
for(x=start; x!=end+inc; x+=inc)
3721
dst[ i*dstStride + x]+=128;
3724
for(i=0; i<100; i+=2){
3725
dst[ (white)*dstStride + i]+=128;
3726
dst[ (black)*dstStride + i]+=128;
3731
*c2= c; //copy local context back
added
removed
quicktime/ffmpeg/libpostproc/postprocess_template.c
