Newer
Older
Copyright (C) 2001-2003 Michael Niedermayer (michaelni@gmx.at)
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
/**
* @file postprocess.c
* postprocessing.
*/
Michael Niedermayer
committed
C MMX MMX2 3DNow
isVertDC Ec Ec
isVertMinMaxOk Ec Ec
Michael Niedermayer
committed
doVertLowPass E e e
isHorizMinMaxOk a E
doHorizLowPass E e e
Michael Niedermayer
committed
Vertical RKAlgo1 E a a
Michael Niedermayer
committed
Horizontal RKAlgo1 a a
Vertical X1# a E E
Horizontal X1# a E E
LinIpolDeinterlace e E E*
CubicIpolDeinterlace a e e*
LinBlendDeinterlace e E E*
D Richard Felker III
committed
MedianDeinterlace# E Ec Ec
Michael Niedermayer
committed
TempDeNoiser# E e e
* i dont have a 3dnow CPU -> its untested, but noone said it doesnt work so it seems to work
# more or less selfinvented filters so the exactness isnt too meaningfull
E = Exact implementation
e = allmost exact implementation (slightly different rounding,...)
a = alternative / approximate impl
c = checked against the other implementations (-vo md5)
*/
/*
TODO:
reduce the time wasted on the mem transfer
unroll stuff if instructions depend too much on the prior one
move YScale thing to the end instead of fixing QP
write a faster and higher quality deblocking filter :)
make the mainloop more flexible (variable number of blocks at once
(the if/else stuff per block is slowing things down)
compare the quality & speed of all filters
split this huge file
try to unroll inner for(x=0 ... loop to avoid these damn if(x ... checks
#include <inttypes.h>
#include <stdio.h>
#include <stdlib.h>
#ifdef HAVE_MALLOC_H
#include <malloc.h>
#endif
#ifdef USE_FASTMEMCPY
#include "libvo/fastmemcpy.h"
#include "mangle.h" //FIXME should be supressed
#ifndef HAVE_MEMALIGN
#define memalign(a,b) malloc(b)
#endif
#define MIN(a,b) ((a) > (b) ? (b) : (a))
#define MAX(a,b) ((a) < (b) ? (b) : (a))
#define ABS(a) ((a) > 0 ? (a) : (-(a)))
#define SIGN(a) ((a) > 0 ? 1 : -1)
#define GET_MODE_BUFFER_SIZE 500
#define OPTIONS_ARRAY_SIZE 10
#define BLOCK_SIZE 8
#define TEMP_STRIDE 8
//#define NUM_BLOCKS_AT_ONCE 16 //not used yet
static uint64_t __attribute__((aligned(8))) w05= 0x0005000500050005LL;
static uint64_t __attribute__((aligned(8))) w20= 0x0020002000200020LL;
static uint64_t __attribute__((aligned(8))) b00= 0x0000000000000000LL;
static uint64_t __attribute__((aligned(8))) b01= 0x0101010101010101LL;
static uint64_t __attribute__((aligned(8))) b02= 0x0202020202020202LL;
static uint64_t __attribute__((aligned(8))) b08= 0x0808080808080808LL;
static uint64_t __attribute__((aligned(8))) b80= 0x8080808080808080LL;
#endif
static uint8_t clip_table[3*256];
static uint8_t * const clip_tab= clip_table + 256;
static struct PPFilter filters[]=
{
{"hb", "hdeblock", 1, 1, 3, H_DEBLOCK},
{"vb", "vdeblock", 1, 2, 4, V_DEBLOCK},
/* {"hr", "rkhdeblock", 1, 1, 3, H_RK1_FILTER},
{"vr", "rkvdeblock", 1, 2, 4, V_RK1_FILTER},*/
{"h1", "x1hdeblock", 1, 1, 3, H_X1_FILTER},
{"v1", "x1vdeblock", 1, 2, 4, V_X1_FILTER},
{"dr", "dering", 1, 5, 6, DERING},
{"al", "autolevels", 0, 1, 2, LEVEL_FIX},
{"lb", "linblenddeint", 1, 1, 4, LINEAR_BLEND_DEINT_FILTER},
{"li", "linipoldeint", 1, 1, 4, LINEAR_IPOL_DEINT_FILTER},
{"ci", "cubicipoldeint", 1, 1, 4, CUBIC_IPOL_DEINT_FILTER},
{"md", "mediandeint", 1, 1, 4, MEDIAN_DEINT_FILTER},
{"fd", "ffmpegdeint", 1, 1, 4, FFMPEG_DEINT_FILTER},
{"l5", "lowpass5", 1, 1, 4, LOWPASS5_DEINT_FILTER},
{"tn", "tmpnoise", 1, 7, 8, TEMP_NOISE_FILTER},
{NULL, NULL,0,0,0,0} //End Marker
};
static char *replaceTable[]=
{
"default", "hdeblock:a,vdeblock:a,dering:a,autolevels,tmpnoise:a:150:200:400",
"de", "hdeblock:a,vdeblock:a,dering:a,autolevels,tmpnoise:a:150:200:400",
"fast", "x1hdeblock:a,x1vdeblock:a,dering:a,autolevels,tmpnoise:a:150:200:400",
"fa", "x1hdeblock:a,x1vdeblock:a,dering:a,autolevels,tmpnoise:a:150:200:400",
Michael Niedermayer
committed
static inline void unusedVariableWarningFixer()
{
if(w05 + w20 + b00 + b01 + b02 + b08 + b80 == 0) b00=0;
Michael Niedermayer
committed
}
#endif
Michael Niedermayer
committed
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
static inline void prefetchnta(void *p)
{
asm volatile( "prefetchnta (%0)\n\t"
: : "r" (p)
);
}
static inline void prefetcht0(void *p)
{
asm volatile( "prefetcht0 (%0)\n\t"
: : "r" (p)
);
}
static inline void prefetcht1(void *p)
{
asm volatile( "prefetcht1 (%0)\n\t"
: : "r" (p)
);
}
static inline void prefetcht2(void *p)
{
asm volatile( "prefetcht2 (%0)\n\t"
: : "r" (p)
);
}
// The horizontal Functions exist only in C cuz the MMX code is faster with vertical filters and transposing
/**
* Check if the given 8x8 Block is mostly "flat"
*/
static inline int isHorizDC(uint8_t src[], int stride, PPContext *c)
const int dcOffset= ((c->QP*c->ppMode.baseDcDiff)>>8) + 1;
const int dcThreshold= dcOffset*2 + 1;
if(((unsigned)(src[0] - src[1] + dcOffset)) < dcThreshold) numEq++;
if(((unsigned)(src[1] - src[2] + dcOffset)) < dcThreshold) numEq++;
if(((unsigned)(src[2] - src[3] + dcOffset)) < dcThreshold) numEq++;
if(((unsigned)(src[3] - src[4] + dcOffset)) < dcThreshold) numEq++;
if(((unsigned)(src[4] - src[5] + dcOffset)) < dcThreshold) numEq++;
if(((unsigned)(src[5] - src[6] + dcOffset)) < dcThreshold) numEq++;
if(((unsigned)(src[6] - src[7] + dcOffset)) < dcThreshold) numEq++;
src+= stride;
}
return numEq > c->ppMode.flatnessThreshold;
}
/**
* Check if the middle 8x8 Block in the given 8x16 block is flat
*/
static inline int isVertDC_C(uint8_t src[], int stride, PPContext *c){
int numEq= 0;
int y;
const int dcOffset= ((c->QP*c->ppMode.baseDcDiff)>>8) + 1;
const int dcThreshold= dcOffset*2 + 1;
src+= stride*4; // src points to begin of the 8x8 Block
for(y=0; y<BLOCK_SIZE-1; y++)
{
if(((unsigned)(src[0] - src[0+stride] + dcOffset)) < dcThreshold) numEq++;
if(((unsigned)(src[1] - src[1+stride] + dcOffset)) < dcThreshold) numEq++;
if(((unsigned)(src[2] - src[2+stride] + dcOffset)) < dcThreshold) numEq++;
if(((unsigned)(src[3] - src[3+stride] + dcOffset)) < dcThreshold) numEq++;
if(((unsigned)(src[4] - src[4+stride] + dcOffset)) < dcThreshold) numEq++;
if(((unsigned)(src[5] - src[5+stride] + dcOffset)) < dcThreshold) numEq++;
if(((unsigned)(src[6] - src[6+stride] + dcOffset)) < dcThreshold) numEq++;
if(((unsigned)(src[7] - src[7+stride] + dcOffset)) < dcThreshold) numEq++;
return numEq > c->ppMode.flatnessThreshold;
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
}
static inline int isHorizMinMaxOk(uint8_t src[], int stride, int QP)
{
if(abs(src[0] - src[7]) > 2*QP) return 0;
return 1;
}
static inline void doHorizDefFilter(uint8_t dst[], int stride, int QP)
{
int y;
for(y=0; y<BLOCK_SIZE; y++)
{
const int middleEnergy= 5*(dst[4] - dst[5]) + 2*(dst[2] - dst[5]);
if(ABS(middleEnergy) < 8*QP)
{
const int q=(dst[3] - dst[4])/2;
const int leftEnergy= 5*(dst[2] - dst[1]) + 2*(dst[0] - dst[3]);
const int rightEnergy= 5*(dst[6] - dst[5]) + 2*(dst[4] - dst[7]);
int d= ABS(middleEnergy) - MIN( ABS(leftEnergy), ABS(rightEnergy) );
d= MAX(d, 0);
d= (5*d + 32) >> 6;
d*= SIGN(-middleEnergy);
if(q>0)
{
d= d<0 ? 0 : d;
d= d>q ? q : d;
}
else
{
d= d>0 ? 0 : d;
d= d<q ? q : d;
}
dst[3]-= d;
dst[4]+= d;
}
dst+= stride;
}
}
/**
* Do a horizontal low pass filter on the 10x8 block (dst points to middle 8x8 Block)
* using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16 (C version)
*/
static inline void doHorizLowPass(uint8_t dst[], int stride, int QP)
{
int y;
for(y=0; y<BLOCK_SIZE; y++)
{
const int first= ABS(dst[-1] - dst[0]) < QP ? dst[-1] : dst[0];
const int last= ABS(dst[8] - dst[7]) < QP ? dst[8] : dst[7];
int sums[9];
sums[0] = first + dst[0];
sums[1] = dst[0] + dst[1];
sums[2] = dst[1] + dst[2];
sums[3] = dst[2] + dst[3];
sums[4] = dst[3] + dst[4];
sums[5] = dst[4] + dst[5];
sums[6] = dst[5] + dst[6];
sums[7] = dst[6] + dst[7];
sums[8] = dst[7] + last;
dst[0]= ((sums[0]<<2) + ((first + sums[2])<<1) + sums[4] + 8)>>4;
dst[1]= ((dst[1]<<2) + ((first + sums[0] + sums[3])<<1) + sums[5] + 8)>>4;
dst[2]= ((dst[2]<<2) + ((first + sums[1] + sums[4])<<1) + sums[6] + 8)>>4;
dst[3]= ((dst[3]<<2) + ((sums[2] + sums[5])<<1) + sums[0] + sums[7] + 8)>>4;
dst[4]= ((dst[4]<<2) + ((sums[3] + sums[6])<<1) + sums[1] + sums[8] + 8)>>4;
dst[5]= ((dst[5]<<2) + ((last + sums[7] + sums[4])<<1) + sums[2] + 8)>>4;
dst[6]= (((last + dst[6])<<2) + ((dst[7] + sums[5])<<1) + sums[3] + 8)>>4;
dst[7]= ((sums[8]<<2) + ((last + sums[6])<<1) + sums[4] + 8)>>4;
dst+= stride;
}
}
/**
* Experimental Filter 1 (Horizontal)
* will not damage linear gradients
* Flat blocks should look like they where passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
* can only smooth blocks at the expected locations (it cant smooth them if they did move)
* MMX2 version does correct clipping C version doesnt
* not identical with the vertical one
*/
static inline void horizX1Filter(uint8_t *src, int stride, int QP)
{
static uint64_t *lut= NULL;
if(lut==NULL)
int i;
lut= (uint64_t*)memalign(8, 256*8);
for(i=0; i<256; i++)
//Simulate 112242211 9-Tap filter
uint64_t a= (v/16) & 0xFF;
uint64_t b= (v/8) & 0xFF;
uint64_t c= (v/4) & 0xFF;
uint64_t d= (3*v/8) & 0xFF;
//Simulate piecewise linear interpolation
uint64_t a= (v/16) & 0xFF;
uint64_t b= (v*3/16) & 0xFF;
uint64_t c= (v*5/16) & 0xFF;
uint64_t d= (7*v/16) & 0xFF;
uint64_t A= (0x100 - a)&0xFF;
uint64_t B= (0x100 - b)&0xFF;
uint64_t C= (0x100 - c)&0xFF;
uint64_t D= (0x100 - c)&0xFF;
lut[i] = (a<<56) | (b<<48) | (c<<40) | (d<<32) |
(D<<24) | (C<<16) | (B<<8) | (A);
//lut[i] = (v<<32) | (v<<24);
int a= src[1] - src[2];
int b= src[3] - src[4];
int c= src[5] - src[6];
int d= MAX(ABS(b) - (ABS(a) + ABS(c))/2, 0);
if(d < QP)
int v = d * SIGN(-b);
src[1] +=v/8;
src[2] +=v/4;
src[3] +=3*v/8;
src[4] -=3*v/8;
src[5] -=v/4;
src[6] -=v/8;
//Note: we have C, MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one
#if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
#define COMPILE_C
#endif
#if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
#define COMPILE_MMX
#endif
#if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
#define COMPILE_MMX2
#endif
#if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
#define COMPILE_3DNOW
#endif
#undef HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_3DNOW
#undef ARCH_X86
#ifdef COMPILE_C
#undef HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_3DNOW
#undef ARCH_X86
#define RENAME(a) a ## _C
#include "postprocess_template.c"
#undef RENAME
#define HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_3DNOW
#define ARCH_X86
#define RENAME(a) a ## _MMX
#include "postprocess_template.c"
#undef RENAME
#define HAVE_MMX
#define HAVE_MMX2
#undef HAVE_3DNOW
#define ARCH_X86
#define RENAME(a) a ## _MMX2
#include "postprocess_template.c"
#ifdef COMPILE_3DNOW
#undef RENAME
#define HAVE_MMX
#undef HAVE_MMX2
#define HAVE_3DNOW
#define ARCH_X86
#define RENAME(a) a ## _3DNow
#include "postprocess_template.c"
// minor note: the HAVE_xyz is messed up after that line so dont use it
static inline void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
QP_STORE_T QPs[], int QPStride, int isColor, pp_mode_t *vm, pp_context_t *vc)
// useing ifs here as they are faster than function pointers allthough the
// difference wouldnt be messureable here but its much better because
// someone might exchange the cpu whithout restarting mplayer ;)
#ifdef RUNTIME_CPUDETECT
if(c->cpuCaps & PP_CPU_CAPS_MMX2)
postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
else if(c->cpuCaps & PP_CPU_CAPS_3DNOW)
postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
else if(c->cpuCaps & PP_CPU_CAPS_MMX)
postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
Michael Niedermayer
committed
#endif
#else //RUNTIME_CPUDETECT
#ifdef HAVE_MMX2
postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
#elif defined (HAVE_3DNOW)
postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
#elif defined (HAVE_MMX)
postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
#endif
#endif //!RUNTIME_CPUDETECT
//static void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
// QP_STORE_T QPs[], int QPStride, int isColor, struct PPMode *ppMode);
"<filterName>[:<option>[:<option>...]][[,|/][-]<filterName>[:<option>...]]...\n"
"vdeblock:autoq/hdeblock:autoq/linblenddeint default,-vdeblock\n"
"Filters Options\n"
"short long name short long option Description\n"
"* * a autoq cpu power dependant enabler\n"
" c chrom chrominance filtring enabled\n"
" y nochrom chrominance filtring disabled\n"
"hb hdeblock (2 Threshold) horizontal deblocking filter\n"
" 1. difference factor: default=64, higher -> more deblocking\n"
" 2. flatness threshold: default=40, lower -> more deblocking\n"
" the h & v deblocking filters share these\n"
" so u cant set different thresholds for h / v\n"
"vb vdeblock (2 Threshold) vertical deblocking filter\n"
"h1 x1hdeblock Experimental h deblock filter 1\n"
"v1 x1vdeblock Experimental v deblock filter 1\n"
"dr dering Deringing filter\n"
"al autolevels automatic brightness / contrast\n"
" f fullyrange stretch luminance to (0..255)\n"
"lb linblenddeint linear blend deinterlacer\n"
"li linipoldeint linear interpolating deinterlace\n"
"ci cubicipoldeint cubic interpolating deinterlacer\n"
"md mediandeint median deinterlacer\n"
"de default hb:a,vb:a,dr:a,al\n"
"fa fast h1:a,v1:a,dr:a,al\n"
"tn tmpnoise (3 Thresholds) Temporal Noise Reducer\n"
" 1. <= 2. <= 3. larger -> stronger filtering\n"
"fq forceQuant <quantizer> Force quantizer\n"
;
pp_mode_t *pp_get_mode_by_name_and_quality(char *name, int quality)
{
char temp[GET_MODE_BUFFER_SIZE];
char *p= temp;
ppMode= memalign(8, sizeof(PPMode));
ppMode->lumMode= 0;
ppMode->chromMode= 0;
ppMode->maxTmpNoise[0]= 700;
ppMode->maxTmpNoise[1]= 1500;
ppMode->maxTmpNoise[2]= 3000;
ppMode->maxAllowedY= 234;
ppMode->minAllowedY= 16;
ppMode->baseDcDiff= 256/4;
ppMode->flatnessThreshold= 56-16;
ppMode->maxClippedThreshold= 0.01;
ppMode->error=0;
strncpy(temp, name, GET_MODE_BUFFER_SIZE);
if(verbose>1) printf("pp: %s\n", name);
int chrom=-1;
char *option;
char *options[OPTIONS_ARRAY_SIZE];
int i;
int filterNameOk=0;
int numOfUnknownOptions=0;
int enable=1; //does the user want us to enabled or disabled the filter
filterToken= strtok(p, filterDelimiters);
if(filterToken == NULL) break;
p+= strlen(filterToken) + 1; // p points to next filterToken
filterName= strtok(filterToken, optionDelimiters);
if(verbose>1) printf("pp: %s::%s\n", filterToken, filterName);
if(*filterName == '-')
{
enable=0;
filterName++;
}
for(;;){ //for all options
option= strtok(NULL, optionDelimiters);
if(option == NULL) break;
if(verbose>1) printf("pp: option: %s\n", option);
if(!strcmp("autoq", option) || !strcmp("a", option)) q= quality;
else if(!strcmp("nochrom", option) || !strcmp("y", option)) chrom=0;
else if(!strcmp("chrom", option) || !strcmp("c", option)) chrom=1;
else
{
options[numOfUnknownOptions] = option;
numOfUnknownOptions++;
}
if(numOfUnknownOptions >= OPTIONS_ARRAY_SIZE-1) break;
}
options[numOfUnknownOptions] = NULL;
/* replace stuff from the replace Table */
for(i=0; replaceTable[2*i]!=NULL; i++)
{
if(!strcmp(replaceTable[2*i], filterName))
{
int newlen= strlen(replaceTable[2*i + 1]);
int plen;
int spaceLeft;
if(p==NULL) p= temp, *p=0; //last filter
else p--, *p=','; //not last filter
plen= strlen(p);
spaceLeft= p - temp + plen;
if(spaceLeft + newlen >= GET_MODE_BUFFER_SIZE)
{
break;
}
memmove(p + newlen, p, plen+1);
memcpy(p, replaceTable[2*i + 1], newlen);
filterNameOk=1;
}
}
for(i=0; filters[i].shortName!=NULL; i++)
{
// printf("Compareing %s, %s, %s\n", filters[i].shortName,filters[i].longName, filterName);
if( !strcmp(filters[i].longName, filterName)
|| !strcmp(filters[i].shortName, filterName))
{
ppMode->lumMode &= ~filters[i].mask;
ppMode->chromMode &= ~filters[i].mask;
filterNameOk=1;
if(!enable) break; // user wants to disable it
if(q >= filters[i].minLumQuality)
if(chrom==1 || (chrom==-1 && filters[i].chromDefault))
if(q >= filters[i].minChromQuality)
if(filters[i].mask == LEVEL_FIX)
{
int o;
{
if( !strcmp(options[o],"fullyrange")
||!strcmp(options[o],"f"))
{
}
else if(filters[i].mask == TEMP_NOISE_FILTER)
{
int o;
int numOfNoises=0;
for(o=0; options[o]!=NULL; o++)
{
char *tail;
strtol(options[o], &tail, 0);
if(tail!=options[o])
{
numOfNoises++;
numOfUnknownOptions--;
if(numOfNoises >= 3) break;
}
}
}
else if(filters[i].mask == V_DEBLOCK || filters[i].mask == H_DEBLOCK)
{
int o;
for(o=0; options[o]!=NULL && o<2; o++)
{
char *tail;
int val= strtol(options[o], &tail, 0);
if(tail==options[o]) break;
numOfUnknownOptions--;
if(o==0) ppMode->baseDcDiff= val;
else ppMode->flatnessThreshold= val;
else if(filters[i].mask == FORCE_QUANT)
{
int o;
for(o=0; options[o]!=NULL && o<1; o++)
{
char *tail;
int val= strtol(options[o], &tail, 0);
if(tail==options[o]) break;
numOfUnknownOptions--;
if(!filterNameOk) ppMode->error++;
ppMode->error += numOfUnknownOptions;
if(verbose>1) printf("pp: lumMode=%X, chromMode=%X\n", ppMode->lumMode, ppMode->chromMode);
if(ppMode->error)
{
fprintf(stderr, "%d errors in postprocess string \"%s\"\n", ppMode->error, name);
free(ppMode);
return NULL;
}
void pp_free_mode(pp_mode_t *mode){
if(mode) free(mode);
}
static void reallocAlign(void **p, int alignment, int size){
if(*p) free(*p);
*p= memalign(alignment, size);
memset(*p, 0, size);
}
static void reallocBuffers(PPContext *c, int width, int height, int stride){
int mbWidth = (width+15)>>4;
int mbHeight= (height+15)>>4;
reallocAlign((void **)&c->tempDst, 8, stride*24);
reallocAlign((void **)&c->tempSrc, 8, stride*24);
reallocAlign((void **)&c->tempBlocks, 8, 2*16*8);
reallocAlign((void **)&c->yHistogram, 8, 256*sizeof(uint64_t));
for(i=0; i<256; i++)
c->yHistogram[i]= width*height/64*15/256;
for(i=0; i<3; i++)
//Note:the +17*1024 is just there so i dont have to worry about r/w over te end
reallocAlign((void **)&c->tempBlured[i], 8, stride*mbHeight*16 + 17*1024);
reallocAlign((void **)&c->tempBluredPast[i], 8, 256*((height+7)&(~7))/2 + 17*1024);//FIXME size
reallocAlign((void **)&c->deintTemp, 8, 2*width+32);
reallocAlign((void **)&c->nonBQPTable, 8, mbWidth*mbHeight*sizeof(QP_STORE_T));
reallocAlign((void **)&c->forcedQPTable, 8, mbWidth*sizeof(QP_STORE_T));
}
static void global_init(){
int i;
memset(clip_table, 0, 256);
for(i=256; i<512; i++)
clip_table[i]= i;
memset(clip_table+512, 0, 256);
}
pp_context_t *pp_get_context(int width, int height, int cpuCaps){
PPContext *c= memalign(32, sizeof(PPContext));
int stride= (width+15)&(~15); //assumed / will realloc if needed
memset(c, 0, sizeof(PPContext));
c->cpuCaps= cpuCaps;
if(cpuCaps&PP_FORMAT){
c->hChromaSubSample= cpuCaps&0x3;
c->vChromaSubSample= (cpuCaps>>4)&0x3;
}else{
c->hChromaSubSample= 1;
c->vChromaSubSample= 1;
}
reallocBuffers(c, width, height, stride);
PPContext *c = (PPContext*)vc;
int i;
for(i=0; i<3; i++) free(c->tempBlured[i]);
for(i=0; i<3; i++) free(c->tempBluredPast[i]);
free(c->tempBlocks);
free(c->yHistogram);
free(c->tempDst);
free(c->tempSrc);
free(c->deintTemp);
free(c->forcedQPTable);
memset(c, 0, sizeof(PPContext));
void pp_postprocess(uint8_t * src[3], int srcStride[3],
int mbWidth = (width+15)>>4;
int mbHeight= (height+15)>>4;
int minStride= MAX(srcStride[0], dstStride[0]);
if(c->stride < minStride)
reallocBuffers(c, width, height, minStride);
if(QP_store==NULL || (mode->lumMode & FORCE_QUANT))
for(i=0; i<mbWidth; i++) QP_store[i]= mode->forcedQuant;
for(i=0; i<mbWidth; i++) QP_store[i]= 1;
if(0){
int x,y;
for(y=0; y<mbHeight; y++){
for(x=0; x<mbWidth; x++){
printf("%2d ", QP_store[x + y*QPStride]);
}
printf("\n");
}
printf("\n");
}
//printf("pict_type:%d\n", pict_type);
if(pict_type!=3)
{
int x,y;
for(y=0; y<mbHeight; y++){
for(x=0; x<mbWidth; x++){
int qscale= QP_store[x + y*QPStride];
if(qscale&~31)
qscale=31;
c->nonBQPTable[y*mbWidth + x]= qscale;
}
}
}
{
printf("using npp filters 0x%X/0x%X\n", mode->lumMode, mode->chromMode);
}
postProcess(src[0], srcStride[0], dst[0], dstStride[0],
width, height, QP_store, QPStride, 0, mode, c);
width = (width )>>c->hChromaSubSample;
height = (height)>>c->vChromaSubSample;
if(mode->chromMode)
{
postProcess(src[1], srcStride[1], dst[1], dstStride[1],
width, height, QP_store, QPStride, 1, mode, c);
postProcess(src[2], srcStride[2], dst[2], dstStride[2],
width, height, QP_store, QPStride, 2, mode, c);
}
else if(srcStride[1] == dstStride[1] && srcStride[2] == dstStride[2])
{
memcpy(dst[1], src[1], srcStride[1]*height);
memcpy(dst[2], src[2], srcStride[2]*height);
}
else
{
int y;
{
memcpy(&(dst[1][y*dstStride[1]]), &(src[1][y*srcStride[1]]), width);
memcpy(&(dst[2][y*dstStride[2]]), &(src[2][y*srcStride[2]]), width);
}
}