1 /*
2 * Copyright (C) 2001-2003 Michael Niedermayer (michaelni@gmx.at)
3 *
4 * AltiVec optimizations (C) 2004 Romain Dolbeau <romain@dolbeau.org>
5 *
6 * This file is part of FFmpeg.
7 *
8 * FFmpeg is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License as published by
10 * the Free Software Foundation; either version 2 of the License, or
11 * (at your option) any later version.
12 *
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23 /**
24 * @file
25 * postprocessing.
26 */
27
28 /*
29 C MMX MMX2 3DNow AltiVec
30 isVertDC Ec Ec Ec
31 isVertMinMaxOk Ec Ec Ec
32 doVertLowPass E e e Ec
33 doVertDefFilter Ec Ec e e Ec
34 isHorizDC Ec Ec Ec
35 isHorizMinMaxOk a E Ec
36 doHorizLowPass E e e Ec
37 doHorizDefFilter Ec Ec e e Ec
38 do_a_deblock Ec E Ec E
39 deRing E e e* Ecp
40 Vertical RKAlgo1 E a a
41 Horizontal RKAlgo1 a a
42 Vertical X1# a E E
43 Horizontal X1# a E E
44 LinIpolDeinterlace e E E*
45 CubicIpolDeinterlace a e e*
46 LinBlendDeinterlace e E E*
47 MedianDeinterlace# E Ec Ec
48 TempDeNoiser# E e e Ec
49
50 * I do not have a 3DNow! CPU -> it is untested, but no one said it does not work so it seems to work
51 # more or less selfinvented filters so the exactness is not too meaningful
52 E = Exact implementation
53 e = almost exact implementation (slightly different rounding,...)
54 a = alternative / approximate impl
55 c = checked against the other implementations (-vo md5)
56 p = partially optimized, still some work to do
57 */
58
59 /*
60 TODO:
61 reduce the time wasted on the mem transfer
62 unroll stuff if instructions depend too much on the prior one
63 move YScale thing to the end instead of fixing QP
64 write a faster and higher quality deblocking filter :)
65 make the mainloop more flexible (variable number of blocks at once
66 (the if/else stuff per block is slowing things down)
67 compare the quality & speed of all filters
68 split this huge file
69 optimize c versions
70 try to unroll inner for(x=0 ... loop to avoid these damn if(x ... checks
71 ...
72 */
73
74 //Changelog: use git log
75
76 #include "config.h"
81 #include <inttypes.h>
82 #include <stdio.h>
83 #include <stdlib.h>
84 #include <string.h>
85 //#undef HAVE_MMXEXT_INLINE
86 //#define HAVE_AMD3DNOW_INLINE
87 //#undef HAVE_MMX_INLINE
88 //#undef ARCH_X86
89 //#define DEBUG_BRIGHTNESS
94
95 #define GET_MODE_BUFFER_SIZE 500
96 #define OPTIONS_ARRAY_SIZE 10
99 //#define NUM_BLOCKS_AT_ONCE 16 //not used yet
100
101 #if ARCH_X86 && HAVE_INLINE_ASM
110 #endif
111
113
114
116 {
119 /* {"hr", "rkhdeblock", 1, 1, 3, H_RK1_FILTER},
120 {"vr", "rkvdeblock", 1, 2, 4, V_RK1_FILTER},*/
125 {
"dr",
"dering", 1, 5, 6,
DERING},
126 {
"al",
"autolevels", 0, 1, 2,
LEVEL_FIX},
135 {
"be",
"bitexact", 1, 0, 0,
BITEXACT},
138 };
139
141 {
142 "default", "hb:a,vb:a,dr:a",
143 "de", "hb:a,vb:a,dr:a",
144 "fast", "h1:a,v1:a,dr:a",
145 "fa", "h1:a,v1:a,dr:a",
146 "ac", "ha:a:128:7,va:a,dr:a",
148 };
149
150 /* The horizontal functions exist only in C because the MMX
151 * code is faster with vertical filters and transposing. */
152
153 /**
154 * Check if the given 8x8 Block is mostly "flat"
155 */
157 {
158 int numEq= 0;
159 int y;
160 const int dcOffset= ((
c->nonBQP*
c->ppMode.baseDcDiff)>>8) + 1;
161 const int dcThreshold= dcOffset*2 + 1;
162
164 numEq += ((unsigned)(
src[0] -
src[1] + dcOffset)) < dcThreshold;
165 numEq += ((unsigned)(
src[1] -
src[2] + dcOffset)) < dcThreshold;
166 numEq += ((unsigned)(
src[2] -
src[3] + dcOffset)) < dcThreshold;
167 numEq += ((unsigned)(
src[3] -
src[4] + dcOffset)) < dcThreshold;
168 numEq += ((unsigned)(
src[4] -
src[5] + dcOffset)) < dcThreshold;
169 numEq += ((unsigned)(
src[5] -
src[6] + dcOffset)) < dcThreshold;
170 numEq += ((unsigned)(
src[6] -
src[7] + dcOffset)) < dcThreshold;
172 }
173 return numEq >
c->ppMode.flatnessThreshold;
174 }
175
176 /**
177 * Check if the middle 8x8 Block in the given 8x16 block is flat
178 */
180 {
181 int numEq= 0;
182 int y;
183 const int dcOffset= ((
c->nonBQP*
c->ppMode.baseDcDiff)>>8) + 1;
184 const int dcThreshold= dcOffset*2 + 1;
185
186 src+=
stride*4;
// src points to begin of the 8x8 Block
188 numEq += ((unsigned)(
src[0] -
src[0+
stride] + dcOffset)) < dcThreshold;
189 numEq += ((unsigned)(
src[1] -
src[1+
stride] + dcOffset)) < dcThreshold;
190 numEq += ((unsigned)(
src[2] -
src[2+
stride] + dcOffset)) < dcThreshold;
191 numEq += ((unsigned)(
src[3] -
src[3+
stride] + dcOffset)) < dcThreshold;
192 numEq += ((unsigned)(
src[4] -
src[4+
stride] + dcOffset)) < dcThreshold;
193 numEq += ((unsigned)(
src[5] -
src[5+
stride] + dcOffset)) < dcThreshold;
194 numEq += ((unsigned)(
src[6] -
src[6+
stride] + dcOffset)) < dcThreshold;
195 numEq += ((unsigned)(
src[7] -
src[7+
stride] + dcOffset)) < dcThreshold;
197 }
198 return numEq >
c->ppMode.flatnessThreshold;
199 }
200
202 {
205 if((
unsigned)(
src[0] -
src[5] + 2*
QP) > 4*
QP)
return 0;
207 if((
unsigned)(
src[2] -
src[7] + 2*
QP) > 4*
QP)
return 0;
209 if((
unsigned)(
src[4] -
src[1] + 2*
QP) > 4*
QP)
return 0;
211 if((
unsigned)(
src[6] -
src[3] + 2*
QP) > 4*
QP)
return 0;
213 }
214 return 1;
215 }
216
218 {
219 int x;
226 }
227 return 1;
228 }
229
231 {
234 }else{
235 return 2;
236 }
237 }
238
240 {
243 }else{
244 return 2;
245 }
246 }
247
249 {
250 int y;
252 const int middleEnergy= 5*(dst[4] - dst[3]) + 2*(dst[2] - dst[5]);
253
254 if(
FFABS(middleEnergy) < 8*
c->QP){
255 const int q=(dst[3] - dst[4])/2;
256 const int leftEnergy= 5*(dst[2] - dst[1]) + 2*(dst[0] - dst[3]);
257 const int rightEnergy= 5*(dst[6] - dst[5]) + 2*(dst[4] - dst[7]);
258
261
264
265 if(q>0)
266 {
269 }
270 else
271 {
274 }
275
278 }
280 }
281 }
282
283 /**
284 * Do a horizontal low pass filter on the 10x8 block (dst points to middle 8x8 Block)
285 * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16 (C version)
286 */
288 {
289 int y;
291 const int first=
FFABS(dst[-1] - dst[0]) <
c->QP ? dst[-1] : dst[0];
292 const int last=
FFABS(dst[8] - dst[7]) <
c->QP ? dst[8] : dst[7];
293
294 int sums[10];
295 sums[0] = 4*
first + dst[0] + dst[1] + dst[2] + 4;
296 sums[1] = sums[0] -
first + dst[3];
297 sums[2] = sums[1] -
first + dst[4];
298 sums[3] = sums[2] -
first + dst[5];
299 sums[4] = sums[3] -
first + dst[6];
300 sums[5] = sums[4] - dst[0] + dst[7];
301 sums[6] = sums[5] - dst[1] + last;
302 sums[7] = sums[6] - dst[2] + last;
303 sums[8] = sums[7] - dst[3] + last;
304 sums[9] = sums[8] - dst[4] + last;
305
306 dst[0]= (sums[0] + sums[2] + 2*dst[0])>>4;
307 dst[1]= (sums[1] + sums[3] + 2*dst[1])>>4;
308 dst[2]= (sums[2] + sums[4] + 2*dst[2])>>4;
309 dst[3]= (sums[3] + sums[5] + 2*dst[3])>>4;
310 dst[4]= (sums[4] + sums[6] + 2*dst[4])>>4;
311 dst[5]= (sums[5] + sums[7] + 2*dst[5])>>4;
312 dst[6]= (sums[6] + sums[8] + 2*dst[6])>>4;
313 dst[7]= (sums[7] + sums[9] + 2*dst[7])>>4;
314
316 }
317 }
318
319 /**
320 * Experimental Filter 1 (Horizontal)
321 * will not damage linear gradients
322 * Flat blocks should look like they were passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
323 * can only smooth blocks at the expected locations (it cannot smooth them if they did move)
324 * MMX2 version does correct clipping C version does not
325 * not identical with the vertical one
326 */
328 {
329 int y;
330 static uint64_t lut[256];
331 if(!lut[255])
332 {
335 {
336 int v=
i < 128 ? 2*
i : 2*(
i-256);
337 /*
338 //Simulate 112242211 9-Tap filter
339 uint64_t a= (v/16) & 0xFF;
340 uint64_t b= (v/8) & 0xFF;
341 uint64_t c= (v/4) & 0xFF;
342 uint64_t d= (3*v/8) & 0xFF;
343 */
344 //Simulate piecewise linear interpolation
345 uint64_t
a= (v/16) & 0xFF;
346 uint64_t
b= (v*3/16) & 0xFF;
347 uint64_t
c= (v*5/16) & 0xFF;
348 uint64_t
d= (7*v/16) & 0xFF;
349 uint64_t
A= (0x100 -
a)&0xFF;
350 uint64_t
B= (0x100 -
b)&0xFF;
351 uint64_t
C= (0x100 -
c)&0xFF;
352 uint64_t
D= (0x100 -
c)&0xFF;
353
354 lut[
i] = (
a<<56) | (
b<<48) | (
c<<40) | (
d<<32) |
355 (
D<<24) | (
C<<16) | (
B<<8) | (
A);
356 //lut[i] = (v<<32) | (v<<24);
357 }
358 }
359
364
366
369
376 }
378 }
379 }
380
381 /**
382 * accurate deblock filter
383 */
386 {
387 int y;
389 const int dcOffset= ((
c->nonBQP*
c->ppMode.baseDcDiff)>>8) + 1;
390 const int dcThreshold= dcOffset*2 + 1;
391
392 src+=
step*4;
// src points to begin of the 8x8 Block
393 for(y=0; y<8; y++){
394 int numEq= 0;
395
396 numEq += ((unsigned)(
src[-1*
step] -
src[0*
step] + dcOffset)) < dcThreshold;
397 numEq += ((unsigned)(
src[ 0*
step] -
src[1*
step] + dcOffset)) < dcThreshold;
398 numEq += ((unsigned)(
src[ 1*
step] -
src[2*
step] + dcOffset)) < dcThreshold;
399 numEq += ((unsigned)(
src[ 2*
step] -
src[3*
step] + dcOffset)) < dcThreshold;
400 numEq += ((unsigned)(
src[ 3*
step] -
src[4*
step] + dcOffset)) < dcThreshold;
401 numEq += ((unsigned)(
src[ 4*
step] -
src[5*
step] + dcOffset)) < dcThreshold;
402 numEq += ((unsigned)(
src[ 5*
step] -
src[6*
step] + dcOffset)) < dcThreshold;
403 numEq += ((unsigned)(
src[ 6*
step] -
src[7*
step] + dcOffset)) < dcThreshold;
404 numEq += ((unsigned)(
src[ 7*
step] -
src[8*
step] + dcOffset)) < dcThreshold;
405 if(numEq >
c->ppMode.flatnessThreshold){
407
411 }else{
414 }
415 for(x=2; x<8; x+=2){
419 }else{
422 }
423 }
427
428 int sums[10];
435 sums[6] = sums[5] -
src[1*
step] + last;
436 sums[7] = sums[6] -
src[2*
step] + last;
437 sums[8] = sums[7] -
src[3*
step] + last;
438 sums[9] = sums[8] -
src[4*
step] + last;
439
449 }
458 }
459 }else{
461
462 if(
FFABS(middleEnergy) < 8*
QP){
466
469
472
473 if(q>0){
476 }else{
479 }
480
482 d= (
d < 0) ? 32 : -32;
486 }
487
490 }
491 }
492
494 }
495 }
496
497 //Note: we have C, MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one
498 //Plain C versions
499 //we always compile C for testing which needs bitexactness
500 #define TEMPLATE_PP_C 1
502
503 #if HAVE_ALTIVEC
504 # define TEMPLATE_PP_ALTIVEC 1
507 #endif
508
509 #if ARCH_X86 && HAVE_INLINE_ASM
510 # if CONFIG_RUNTIME_CPUDETECT
511 # define TEMPLATE_PP_MMX 1
513 # define TEMPLATE_PP_MMXEXT 1
515 # define TEMPLATE_PP_3DNOW 1
517 # define TEMPLATE_PP_SSE2 1
519 # else
520 # if HAVE_SSE2_INLINE
521 # define TEMPLATE_PP_SSE2 1
523 # elif HAVE_MMXEXT_INLINE
524 # define TEMPLATE_PP_MMXEXT 1
526 # elif HAVE_AMD3DNOW_INLINE
527 # define TEMPLATE_PP_3DNOW 1
529 # elif HAVE_MMX_INLINE
530 # define TEMPLATE_PP_MMX 1
532 # endif
533 # endif
534 #endif
535
536 typedef void (*
pp_fn)(
const uint8_t
src[],
int srcStride, uint8_t dst[],
int dstStride,
int width,
int height,
537 const int8_t QPs[],
int QPStride,
int isColor,
PPContext *
c2);
538
541 {
542 pp_fn pp = postProcess_C;
545 c->ppMode= *ppMode;
//FIXME
546
548 #if CONFIG_RUNTIME_CPUDETECT
549 #if ARCH_X86 && HAVE_INLINE_ASM
550 // ordered per speed fastest first
555 #elif HAVE_ALTIVEC
557 #endif
558 #else /* CONFIG_RUNTIME_CPUDETECT */
559 #if HAVE_SSE2_INLINE
560 pp = postProcess_SSE2;
561 #elif HAVE_MMXEXT_INLINE
562 pp = postProcess_MMX2;
563 #elif HAVE_AMD3DNOW_INLINE
564 pp = postProcess_3DNow;
565 #elif HAVE_MMX_INLINE
566 pp = postProcess_MMX;
567 #elif HAVE_ALTIVEC
568 pp = postProcess_altivec;
569 #endif
570 #endif /* !CONFIG_RUNTIME_CPUDETECT */
571 }
572
573 pp(
src, srcStride, dst, dstStride,
width,
height, QPs, QPStride, isColor,
c);
574 }
575
576 /* -pp Command line Help
577 */
579 "Available postprocessing filters:\n"
580 "Filters Options\n"
581 "short long name short long option Description\n"
582 "* * a autoq CPU power dependent enabler\n"
583 " c chrom chrominance filtering enabled\n"
584 " y nochrom chrominance filtering disabled\n"
585 " n noluma luma filtering disabled\n"
586 "hb hdeblock (2 threshold) horizontal deblocking filter\n"
587 " 1. difference factor: default=32, higher -> more deblocking\n"
588 " 2. flatness threshold: default=39, lower -> more deblocking\n"
589 " the h & v deblocking filters share these\n"
590 " so you can't set different thresholds for h / v\n"
591 "vb vdeblock (2 threshold) vertical deblocking filter\n"
592 "ha hadeblock (2 threshold) horizontal deblocking filter\n"
593 "va vadeblock (2 threshold) vertical deblocking filter\n"
594 "h1 x1hdeblock experimental h deblock filter 1\n"
595 "v1 x1vdeblock experimental v deblock filter 1\n"
596 "dr dering deringing filter\n"
597 "al autolevels automatic brightness / contrast\n"
598 " f fullyrange stretch luminance to (0..255)\n"
599 "lb linblenddeint linear blend deinterlacer\n"
600 "li linipoldeint linear interpolating deinterlace\n"
601 "ci cubicipoldeint cubic interpolating deinterlacer\n"
602 "md mediandeint median deinterlacer\n"
603 "fd ffmpegdeint ffmpeg deinterlacer\n"
604 "l5 lowpass5 FIR lowpass deinterlacer\n"
605 "de default hb:a,vb:a,dr:a\n"
606 "fa fast h1:a,v1:a,dr:a\n"
607 "ac ha:a:128:7,va:a,dr:a\n"
608 "tn tmpnoise (3 threshold) temporal noise reducer\n"
609 " 1. <= 2. <= 3. larger -> stronger filtering\n"
610 "fq forceQuant <quantizer> force quantizer\n"
611 "Usage:\n"
612 "<filterName>[:<option>[:<option>...]][[,|/][-]<filterName>[:<option>...]]...\n"
613 "long form example:\n"
614 "vdeblock:autoq/hdeblock:autoq/linblenddeint default,-vdeblock\n"
615 "short form example:\n"
616 "vb:a/hb:a/lb de,-vb\n"
617 "more examples:\n"
618 "tn:64:128:256\n"
619 "\n"
620 ;
621
623 {
626 static const char filterDelimiters[] = ",/";
627 static const char optionDelimiters[] = ":|";
629 char *filterToken;
630
634 }
635
636 if (!strcmp(
name,
"help")) {
637 const char *p;
638 for (p =
pp_help; strchr(p,
'\n'); p = strchr(p,
'\n') + 1) {
641 }
643 }
644
646 if (!ppMode)
648
660
663
665
666 for(;;){
667 const char *filterName;
668 int q= 1000000; //PP_QUALITY_MAX;
669 int chrom=-1;
670 int luma=-1;
674 int filterNameOk=0;
675 int numOfUnknownOptions=0;
676 int enable=1; //does the user want us to enabled or disabled the filter
677 char *tokstate;
678
679 filterToken=
av_strtok(p, filterDelimiters, &tokstate);
680 if(!filterToken) break;
681 p+= strlen(filterToken) + 1; // p points to next filterToken
682 filterName=
av_strtok(filterToken, optionDelimiters, &tokstate);
683 if (!filterName) {
685 break;
686 }
688
689 if(*filterName == '-'){
690 enable=0;
691 filterName++;
692 }
693
694 for(;;){ //for all options
697
700 else if(!strcmp(
"nochrom",
option) || !strcmp(
"y",
option)) chrom=0;
701 else if(!strcmp(
"chrom",
option) || !strcmp(
"c",
option)) chrom=1;
702 else if(!strcmp(
"noluma",
option) || !strcmp(
"n",
option)) luma=0;
703 else{
705 numOfUnknownOptions++;
706 }
708 }
710
711 /* replace stuff from the replace Table */
715 int plen;
716 int spaceLeft;
717
718 p--, *p=',';
719
720 plen= strlen(p);
721 spaceLeft= p -
temp + plen;
724 break;
725 }
726 memmove(p + newlen, p, plen+1);
728 filterNameOk=1;
729 }
730 }
731
733 if( !strcmp(
filters[
i].longName, filterName)
734 || !strcmp(
filters[
i].shortName, filterName)){
737
738 filterNameOk=1;
739 if(!enable) break; // user wants to disable it
740
741 if(q >=
filters[
i].minLumQuality && luma)
743 if(chrom==1 || (chrom==-1 &&
filters[
i].chromDefault))
746
748 int o;
752 if( !strcmp(
options[o],
"fullyrange")
756 numOfUnknownOptions--;
757 }
758 }
759 }
761 {
762 int o;
763 int numOfNoises=0;
764
766 char *tail;
770 numOfNoises++;
771 numOfUnknownOptions--;
772 if(numOfNoises >= 3) break;
773 }
774 }
775 }
778 int o;
779
780 for(o=0;
options[o] && o<2; o++){
781 char *tail;
784
785 numOfUnknownOptions--;
788 }
789 }
791 int o;
793
794 for(o=0;
options[o] && o<1; o++){
795 char *tail;
798
799 numOfUnknownOptions--;
801 }
802 }
803 }
804 }
805 if(!filterNameOk) ppMode->
error++;
806 ppMode->
error += numOfUnknownOptions;
807 }
808
814 }
815 return ppMode;
816 }
817
820 }
821
825 }
826
828 int mbWidth = (
width+15)>>4;
829 int mbHeight= (
height+15)>>4;
831
833 c->qpStride= qpStride;
834
841
843 //Note: The +17*1024 is just there so I do not have to worry about r/w over the end.
846 }
847
849 reallocAlign((
void **)&
c->nonBQPTable, qpStride*mbHeight*
sizeof(int8_t));
850 reallocAlign((
void **)&
c->stdQPTable, qpStride*mbHeight*
sizeof(int8_t));
851 reallocAlign((
void **)&
c->forcedQPTable, mbWidth*
sizeof(int8_t));
852 }
853
855 return "postproc";
856 }
857
859
863 int qpStride= (
width+15)/16 + 2;
//assumed / will realloc if needed
864
867
870 c->hChromaSubSample= cpuCaps&0x3;
871 c->vChromaSubSample= (cpuCaps>>4)&0x3;
872 }else{
873 c->hChromaSubSample= 1;
874 c->vChromaSubSample= 1;
875 }
878 } else {
884 }
885
887
889
891 }
892
896
901
910
912
914 }
915
917 uint8_t * dst[3], const int dstStride[3],
919 const int8_t *QP_store, int QPStride,
920 pp_mode *vm,
void *vc,
int pict_type)
921 {
922 int mbWidth = (
width+15)>>4;
923 int mbHeight= (
height+15)>>4;
927 int absQPStride =
FFABS(QPStride);
928
929 // c->stride and c->QPStride are always positive
930 if(
c->stride < minStride ||
c->qpStride < absQPStride)
932 FFMAX(minStride,
c->stride),
933 FFMAX(
c->qpStride, absQPStride));
934
937 QP_store=
c->forcedQPTable;
938 absQPStride = QPStride = 0;
940 for(
i=0;
i<mbWidth;
i++)
c->forcedQPTable[
i]=
mode->forcedQuant;
941 else
942 for(
i=0;
i<mbWidth;
i++)
c->forcedQPTable[
i]= 1;
943 }
944
947 const int count=
FFMAX(mbHeight * absQPStride, mbWidth);
948 for(
i=0;
i<(count>>2);
i++){
950 }
951 for(
i<<=2;
i<count;
i++){
952 c->stdQPTable[
i] = QP_store[
i]>>1;
953 }
954 QP_store=
c->stdQPTable;
955 QPStride= absQPStride;
956 }
957
958 if(0){
959 int x,y;
960 for(y=0; y<mbHeight; y++){
961 for(x=0; x<mbWidth; x++){
963 }
965 }
967 }
968
969 if((pict_type&7)!=3){
970 if (QPStride >= 0){
972 const int count=
FFMAX(mbHeight * QPStride, mbWidth);
973 for(
i=0;
i<(count>>2);
i++){
975 }
976 for(
i<<=2;
i<count;
i++){
977 c->nonBQPTable[
i] = QP_store[
i] & 0x3F;
978 }
979 } else {
981 for(
i=0;
i<mbHeight;
i++) {
982 for(j=0; j<absQPStride; j++) {
983 c->nonBQPTable[
i*absQPStride+j] = QP_store[
i*QPStride+j] & 0x3F;
984 }
985 }
986 }
987 }
988
991
994
995 if (!(
src[1] &&
src[2] && dst[1] && dst[2]))
996 return;
997
1000
1001 if(
mode->chromMode){
1006 }
1007 else if(srcStride[1] == dstStride[1] && srcStride[2] == dstStride[2]){
1010 }else{
1011 int y;
1013 memcpy(&(dst[1][y*dstStride[1]]), &(
src[1][y*srcStride[1]]),
width);
1014 memcpy(&(dst[2][y*dstStride[2]]), &(
src[2][y*srcStride[2]]),
width);
1015 }
1016 }
1017 }