postprocess.c 44 KB
Newer Older
1
/*
2 3 4 5
 * Copyright (C) 2001-2003 Michael Niedermayer (michaelni@gmx.at)
 *
 * AltiVec optimizations (C) 2004 Romain Dolbeau <romain@dolbeau.org>
 *
6
 * This file is part of FFmpeg.
7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
 *
 * FFmpeg 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.
 *
 * FFmpeg 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 FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */
22

Michael Niedermayer's avatar
Michael Niedermayer committed
23 24 25 26
/**
 * @file postprocess.c
 * postprocessing.
 */
27

28
/*
29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48
                        C       MMX     MMX2    3DNow   AltiVec
isVertDC                Ec      Ec                      Ec
isVertMinMaxOk          Ec      Ec                      Ec
doVertLowPass           E               e       e       Ec
doVertDefFilter         Ec      Ec      e       e       Ec
isHorizDC               Ec      Ec                      Ec
isHorizMinMaxOk         a       E                       Ec
doHorizLowPass          E               e       e       Ec
doHorizDefFilter        Ec      Ec      e       e       Ec
do_a_deblock            Ec      E       Ec      E
deRing                  E               e       e*      Ecp
Vertical RKAlgo1        E               a       a
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*
MedianDeinterlace#      E       Ec      Ec
TempDeNoiser#           E               e       e       Ec
49

Diego Biurrun's avatar
Diego Biurrun committed
50 51
* i do not have a 3DNow! CPU -> it is untested, but no one said it does not work so it seems to work
# more or less selfinvented filters so the exactness is not too meaningful
52
E = Exact implementation
53
e = allmost exact implementation (slightly different rounding,...)
54 55
a = alternative / approximate impl
c = checked against the other implementations (-vo md5)
56
p = partially optimized, still some work to do
57 58 59 60 61 62 63
*/

/*
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
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 68
compare the quality & speed of all filters
split this huge file
Michael Niedermayer's avatar
Michael Niedermayer committed
69
optimize c versions
70
try to unroll inner for(x=0 ... loop to avoid these damn if(x ... checks
71
...
72 73
*/

Diego Biurrun's avatar
Diego Biurrun committed
74
//Changelog: use the Subversion log
75

76
#include "config.h"
Luca Barbato's avatar
Luca Barbato committed
77
#include "avutil.h"
78 79
#include <inttypes.h>
#include <stdio.h>
80
#include <stdlib.h>
81
#include <string.h>
82 83 84
#ifdef HAVE_MALLOC_H
#include <malloc.h>
#endif
85
//#undef HAVE_MMX2
86
//#define HAVE_3DNOW
87
//#undef HAVE_MMX
88
//#undef ARCH_X86
89
//#define DEBUG_BRIGHTNESS
90
#include "postprocess.h"
Michael Niedermayer's avatar
Michael Niedermayer committed
91
#include "postprocess_internal.h"
92

93 94 95 96
#ifdef HAVE_ALTIVEC_H
#include <altivec.h>
#endif

97 98
#define GET_MODE_BUFFER_SIZE 500
#define OPTIONS_ARRAY_SIZE 10
99 100 101
#define BLOCK_SIZE 8
#define TEMP_STRIDE 8
//#define NUM_BLOCKS_AT_ONCE 16 //not used yet
102

103
#if defined(ARCH_X86)
104 105 106 107 108 109 110 111
static DECLARE_ALIGNED(8, uint64_t attribute_used, w05)= 0x0005000500050005LL;
static DECLARE_ALIGNED(8, uint64_t attribute_used, w04)= 0x0004000400040004LL;
static DECLARE_ALIGNED(8, uint64_t attribute_used, w20)= 0x0020002000200020LL;
static DECLARE_ALIGNED(8, uint64_t attribute_used, b00)= 0x0000000000000000LL;
static DECLARE_ALIGNED(8, uint64_t attribute_used, b01)= 0x0101010101010101LL;
static DECLARE_ALIGNED(8, uint64_t attribute_used, b02)= 0x0202020202020202LL;
static DECLARE_ALIGNED(8, uint64_t attribute_used, b08)= 0x0808080808080808LL;
static DECLARE_ALIGNED(8, uint64_t attribute_used, b80)= 0x8080808080808080LL;
112
#endif
113

114 115 116
static uint8_t clip_table[3*256];
static uint8_t * const clip_tab= clip_table + 256;

117
static const int attribute_used deringThreshold= 20;
118

119

120 121
static struct PPFilter filters[]=
{
122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140
        {"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},
        {"ha", "ahdeblock",             1, 1, 3, H_A_DEBLOCK},
        {"va", "avdeblock",             1, 2, 4, V_A_DEBLOCK},
        {"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},
        {"fq", "forcequant",            1, 0, 0, FORCE_QUANT},
        {NULL, NULL,0,0,0,0} //End Marker
141 142
};

143
static const char *replaceTable[]=
144
{
145 146 147 148 149 150
        "default",      "hdeblock:a,vdeblock:a,dering:a",
        "de",           "hdeblock:a,vdeblock:a,dering:a",
        "fast",         "x1hdeblock:a,x1vdeblock:a,dering:a",
        "fa",           "x1hdeblock:a,x1vdeblock:a,dering:a",
        "ac",           "ha:a:128:7,va:a,dering:a",
        NULL //End Marker
151 152
};

153

154
#if defined(ARCH_X86)
155 156
static inline void prefetchnta(void *p)
{
157 158 159
        asm volatile(   "prefetchnta (%0)\n\t"
                : : "r" (p)
        );
160 161 162 163
}

static inline void prefetcht0(void *p)
{
164 165 166
        asm volatile(   "prefetcht0 (%0)\n\t"
                : : "r" (p)
        );
167 168 169 170
}

static inline void prefetcht1(void *p)
{
171 172 173
        asm volatile(   "prefetcht1 (%0)\n\t"
                : : "r" (p)
        );
174 175 176 177
}

static inline void prefetcht2(void *p)
{
178 179 180
        asm volatile(   "prefetcht2 (%0)\n\t"
                : : "r" (p)
        );
181
}
182
#endif
183

184
// The horizontal Functions exist only in C cuz the MMX code is faster with vertical filters and transposing
185

186 187 188
/**
 * Check if the given 8x8 Block is mostly "flat"
 */
189
static inline int isHorizDC_C(uint8_t src[], int stride, PPContext *c)
190
{
191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207
        int numEq= 0;
        int y;
        const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
        const int dcThreshold= dcOffset*2 + 1;

        for(y=0; y<BLOCK_SIZE; y++)
        {
                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;
208 209 210 211 212 213
}

/**
 * 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){
214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232
        int numEq= 0;
        int y;
        const int dcOffset= ((c->nonBQP*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++;
                src+= stride;
        }
        return numEq > c->ppMode.flatnessThreshold;
233 234
}

235
static inline int isHorizMinMaxOk_C(uint8_t src[], int stride, int QP)
236
{
237
        int i;
238
#if 1
239 240 241 242 243 244 245 246 247 248
        for(i=0; i<2; i++){
                if((unsigned)(src[0] - src[5] + 2*QP) > 4*QP) return 0;
                src += stride;
                if((unsigned)(src[2] - src[7] + 2*QP) > 4*QP) return 0;
                src += stride;
                if((unsigned)(src[4] - src[1] + 2*QP) > 4*QP) return 0;
                src += stride;
                if((unsigned)(src[6] - src[3] + 2*QP) > 4*QP) return 0;
                src += stride;
        }
249
#else
250 251 252 253
        for(i=0; i<8; i++){
                if((unsigned)(src[0] - src[7] + 2*QP) > 4*QP) return 0;
                src += stride;
        }
254
#endif
255
        return 1;
256
}
257

258 259 260 261
static inline int isVertMinMaxOk_C(uint8_t src[], int stride, int QP)
{
#if 1
#if 1
262 263 264 265 266 267 268 269 270
        int x;
        src+= stride*4;
        for(x=0; x<BLOCK_SIZE; x+=4)
        {
                if((unsigned)(src[  x + 0*stride] - src[  x + 5*stride] + 2*QP) > 4*QP) return 0;
                if((unsigned)(src[1+x + 2*stride] - src[1+x + 7*stride] + 2*QP) > 4*QP) return 0;
                if((unsigned)(src[2+x + 4*stride] - src[2+x + 1*stride] + 2*QP) > 4*QP) return 0;
                if((unsigned)(src[3+x + 6*stride] - src[3+x + 3*stride] + 2*QP) > 4*QP) return 0;
        }
271
#else
272 273 274 275 276 277
        int x;
        src+= stride*3;
        for(x=0; x<BLOCK_SIZE; x++)
        {
                if((unsigned)(src[x + stride] - src[x + (stride<<3)] + 2*QP) > 4*QP) return 0;
        }
278
#endif
279
        return 1;
280
#else
281 282 283 284 285 286 287 288 289 290 291 292 293 294 295
        int x;
        src+= stride*4;
        for(x=0; x<BLOCK_SIZE; x++)
        {
                int min=255;
                int max=0;
                int y;
                for(y=0; y<8; y++){
                        int v= src[x + y*stride];
                        if(v>max) max=v;
                        if(v<min) min=v;
                }
                if(max-min > 2*QP) return 0;
        }
        return 1;
296 297 298
#endif
}

299
static inline int horizClassify_C(uint8_t src[], int stride, PPContext *c){
300 301 302 303 304 305 306 307
        if( isHorizDC_C(src, stride, c) ){
                if( isHorizMinMaxOk_C(src, stride, c->QP) )
                        return 1;
                else
                        return 0;
        }else{
                return 2;
        }
308 309
}

310
static inline int vertClassify_C(uint8_t src[], int stride, PPContext *c){
311 312 313 314 315 316 317 318
        if( isVertDC_C(src, stride, c) ){
                if( isVertMinMaxOk_C(src, stride, c->QP) )
                        return 1;
                else
                        return 0;
        }else{
                return 2;
        }
319 320
}

321
static inline void doHorizDefFilter_C(uint8_t dst[], int stride, PPContext *c)
322
{
323 324 325 326 327
        int y;
        for(y=0; y<BLOCK_SIZE; y++)
        {
                const int middleEnergy= 5*(dst[4] - dst[3]) + 2*(dst[2] - dst[5]);

328
                if(FFABS(middleEnergy) < 8*c->QP)
329 330 331 332 333
                {
                        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]);

334
                        int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
335
                        d= FFMAX(d, 0);
336 337

                        d= (5*d + 32) >> 6;
338
                        d*= FFSIGN(-middleEnergy);
339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355

                        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;
        }
356 357 358 359 360 361
}

/**
 * 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)
 */
362
static inline void doHorizLowPass_C(uint8_t dst[], int stride, PPContext *c)
363
{
364 365 366
        int y;
        for(y=0; y<BLOCK_SIZE; y++)
        {
367 368
                const int first= FFABS(dst[-1] - dst[0]) < c->QP ? dst[-1] : dst[0];
                const int last= FFABS(dst[8] - dst[7]) < c->QP ? dst[8] : dst[7];
369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392

                int sums[10];
                sums[0] = 4*first + dst[0] + dst[1] + dst[2] + 4;
                sums[1] = sums[0] - first  + dst[3];
                sums[2] = sums[1] - first  + dst[4];
                sums[3] = sums[2] - first  + dst[5];
                sums[4] = sums[3] - first  + dst[6];
                sums[5] = sums[4] - dst[0] + dst[7];
                sums[6] = sums[5] - dst[1] + last;
                sums[7] = sums[6] - dst[2] + last;
                sums[8] = sums[7] - dst[3] + last;
                sums[9] = sums[8] - dst[4] + last;

                dst[0]= (sums[0] + sums[2] + 2*dst[0])>>4;
                dst[1]= (sums[1] + sums[3] + 2*dst[1])>>4;
                dst[2]= (sums[2] + sums[4] + 2*dst[2])>>4;
                dst[3]= (sums[3] + sums[5] + 2*dst[3])>>4;
                dst[4]= (sums[4] + sums[6] + 2*dst[4])>>4;
                dst[5]= (sums[5] + sums[7] + 2*dst[5])>>4;
                dst[6]= (sums[6] + sums[8] + 2*dst[6])>>4;
                dst[7]= (sums[7] + sums[9] + 2*dst[7])>>4;

                dst+= stride;
        }
393 394
}

395
/**
396 397 398
 * 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
Diego Biurrun's avatar
Diego Biurrun committed
399 400
 * can only smooth blocks at the expected locations (it cannot smooth them if they did move)
 * MMX2 version does correct clipping C version does not
401
 * not identical with the vertical one
402
 */
403 404
static inline void horizX1Filter(uint8_t *src, int stride, int QP)
{
405 406 407 408 409
        int y;
        static uint64_t *lut= NULL;
        if(lut==NULL)
        {
                int i;
Luca Barbato's avatar
Luca Barbato committed
410
                lut = av_malloc(256*8);
411 412 413
                for(i=0; i<256; i++)
                {
                        int v= i < 128 ? 2*i : 2*(i-256);
414
/*
415
//Simulate 112242211 9-Tap filter
416 417 418 419
                        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;
420
*/
421
//Simulate piecewise linear interpolation
422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442
                        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);
                }
        }

        for(y=0; y<BLOCK_SIZE; y++)
        {
                int a= src[1] - src[2];
                int b= src[3] - src[4];
                int c= src[5] - src[6];

443
                int d= FFMAX(FFABS(b) - (FFABS(a) + FFABS(c))/2, 0);
444 445 446

                if(d < QP)
                {
447
                        int v = d * FFSIGN(-b);
448 449 450 451 452 453 454 455 456 457 458

                        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;

                }
                src+=stride;
        }
459 460
}

461 462 463
/**
 * accurate deblock filter
 */
464
static av_always_inline void do_a_deblock_C(uint8_t *src, int step, int stride, PPContext *c){
465 466 467 468
        int y;
        const int QP= c->QP;
        const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
        const int dcThreshold= dcOffset*2 + 1;
469
//START_TIMER
470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502
        src+= step*4; // src points to begin of the 8x8 Block
        for(y=0; y<8; y++){
                int numEq= 0;

                if(((unsigned)(src[-1*step] - src[0*step] + dcOffset)) < dcThreshold) numEq++;
                if(((unsigned)(src[ 0*step] - src[1*step] + dcOffset)) < dcThreshold) numEq++;
                if(((unsigned)(src[ 1*step] - src[2*step] + dcOffset)) < dcThreshold) numEq++;
                if(((unsigned)(src[ 2*step] - src[3*step] + dcOffset)) < dcThreshold) numEq++;
                if(((unsigned)(src[ 3*step] - src[4*step] + dcOffset)) < dcThreshold) numEq++;
                if(((unsigned)(src[ 4*step] - src[5*step] + dcOffset)) < dcThreshold) numEq++;
                if(((unsigned)(src[ 5*step] - src[6*step] + dcOffset)) < dcThreshold) numEq++;
                if(((unsigned)(src[ 6*step] - src[7*step] + dcOffset)) < dcThreshold) numEq++;
                if(((unsigned)(src[ 7*step] - src[8*step] + dcOffset)) < dcThreshold) numEq++;
                if(numEq > c->ppMode.flatnessThreshold){
                        int min, max, x;

                        if(src[0] > src[step]){
                            max= src[0];
                            min= src[step];
                        }else{
                            max= src[step];
                            min= src[0];
                        }
                        for(x=2; x<8; x+=2){
                                if(src[x*step] > src[(x+1)*step]){
                                        if(src[x    *step] > max) max= src[ x   *step];
                                        if(src[(x+1)*step] < min) min= src[(x+1)*step];
                                }else{
                                        if(src[(x+1)*step] > max) max= src[(x+1)*step];
                                        if(src[ x   *step] < min) min= src[ x   *step];
                                }
                        }
                        if(max-min < 2*QP){
503 504
                                const int first= FFABS(src[-1*step] - src[0]) < QP ? src[-1*step] : src[0];
                                const int last= FFABS(src[8*step] - src[7*step]) < QP ? src[8*step] : src[7*step];
505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529

                                int sums[10];
                                sums[0] = 4*first + src[0*step] + src[1*step] + src[2*step] + 4;
                                sums[1] = sums[0] - first       + src[3*step];
                                sums[2] = sums[1] - first       + src[4*step];
                                sums[3] = sums[2] - first       + src[5*step];
                                sums[4] = sums[3] - first       + src[6*step];
                                sums[5] = sums[4] - src[0*step] + src[7*step];
                                sums[6] = sums[5] - src[1*step] + last;
                                sums[7] = sums[6] - src[2*step] + last;
                                sums[8] = sums[7] - src[3*step] + last;
                                sums[9] = sums[8] - src[4*step] + last;

                                src[0*step]= (sums[0] + sums[2] + 2*src[0*step])>>4;
                                src[1*step]= (sums[1] + sums[3] + 2*src[1*step])>>4;
                                src[2*step]= (sums[2] + sums[4] + 2*src[2*step])>>4;
                                src[3*step]= (sums[3] + sums[5] + 2*src[3*step])>>4;
                                src[4*step]= (sums[4] + sums[6] + 2*src[4*step])>>4;
                                src[5*step]= (sums[5] + sums[7] + 2*src[5*step])>>4;
                                src[6*step]= (sums[6] + sums[8] + 2*src[6*step])>>4;
                                src[7*step]= (sums[7] + sums[9] + 2*src[7*step])>>4;
                        }
                }else{
                        const int middleEnergy= 5*(src[4*step] - src[3*step]) + 2*(src[2*step] - src[5*step]);

530
                        if(FFABS(middleEnergy) < 8*QP)
531 532 533 534 535
                        {
                                const int q=(src[3*step] - src[4*step])/2;
                                const int leftEnergy=  5*(src[2*step] - src[1*step]) + 2*(src[0*step] - src[3*step]);
                                const int rightEnergy= 5*(src[6*step] - src[5*step]) + 2*(src[4*step] - src[7*step]);

536
                                int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
537
                                d= FFMAX(d, 0);
538 539

                                d= (5*d + 32) >> 6;
540
                                d*= FFSIGN(-middleEnergy);
541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559

                                if(q>0)
                                {
                                        d= d<0 ? 0 : d;
                                        d= d>q ? q : d;
                                }
                                else
                                {
                                        d= d>0 ? 0 : d;
                                        d= d<q ? q : d;
                                }

                                src[3*step]-= d;
                                src[4*step]+= d;
                        }
                }

                src += stride;
        }
560 561 562 563 564 565
/*if(step==16){
    STOP_TIMER("step16")
}else{
    STOP_TIMER("stepX")
}*/
}
566

567
//Note: we have C, MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one
568
//Plain C versions
569 570 571 572
#if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
#define COMPILE_C
#endif

573 574 575 576 577 578
#ifdef ARCH_POWERPC
#ifdef HAVE_ALTIVEC
#define COMPILE_ALTIVEC
#endif //HAVE_ALTIVEC
#endif //ARCH_POWERPC

579
#if defined(ARCH_X86)
580 581 582 583 584 585 586 587 588 589 590 591

#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
592
#endif /* defined(ARCH_X86) */
593 594 595 596

#undef HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_3DNOW
597
#undef HAVE_ALTIVEC
598 599

#ifdef COMPILE_C
600 601 602 603 604
#undef HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_3DNOW
#define RENAME(a) a ## _C
#include "postprocess_template.c"
605
#endif
606

607 608 609 610 611 612 613 614 615 616
#ifdef ARCH_POWERPC
#ifdef COMPILE_ALTIVEC
#undef RENAME
#define HAVE_ALTIVEC
#define RENAME(a) a ## _altivec
#include "postprocess_altivec_template.c"
#include "postprocess_template.c"
#endif
#endif //ARCH_POWERPC

617
//MMX versions
618
#ifdef COMPILE_MMX
619 620 621 622 623 624
#undef RENAME
#define HAVE_MMX
#undef HAVE_MMX2
#undef HAVE_3DNOW
#define RENAME(a) a ## _MMX
#include "postprocess_template.c"
625
#endif
626 627

//MMX2 versions
628
#ifdef COMPILE_MMX2
629 630 631 632 633 634
#undef RENAME
#define HAVE_MMX
#define HAVE_MMX2
#undef HAVE_3DNOW
#define RENAME(a) a ## _MMX2
#include "postprocess_template.c"
635
#endif
636 637

//3DNOW versions
638
#ifdef COMPILE_3DNOW
639 640 641 642 643 644
#undef RENAME
#define HAVE_MMX
#undef HAVE_MMX2
#define HAVE_3DNOW
#define RENAME(a) a ## _3DNow
#include "postprocess_template.c"
645
#endif
646

Diego Biurrun's avatar
Diego Biurrun committed
647
// minor note: the HAVE_xyz is messed up after that line so do not use it.
648 649

static inline void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
650
        QP_STORE_T QPs[], int QPStride, int isColor, pp_mode_t *vm, pp_context_t *vc)
651
{
652 653 654
        PPContext *c= (PPContext *)vc;
        PPMode *ppMode= (PPMode *)vm;
        c->ppMode= *ppMode; //FIXME
655

Diego Biurrun's avatar
Diego Biurrun committed
656 657 658
        // Using ifs here as they are faster than function pointers although the
        // difference would not be measureable here but it is much better because
        // someone might exchange the CPU whithout restarting MPlayer ;)
659
#ifdef RUNTIME_CPUDETECT
660
#if defined(ARCH_X86)
661 662 663 664 665 666 667 668 669
        // ordered per speed fasterst first
        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);
        else
                postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
670
#else
671 672
#ifdef ARCH_POWERPC
#ifdef HAVE_ALTIVEC
673
        if(c->cpuCaps & PP_CPU_CAPS_ALTIVEC)
674
                postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
675 676 677
        else
#endif
#endif
678
                postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
679
#endif
680 681
#else //RUNTIME_CPUDETECT
#ifdef HAVE_MMX2
682
                postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
683
#elif defined (HAVE_3DNOW)
684
                postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
685
#elif defined (HAVE_MMX)
686
                postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
687
#elif defined (HAVE_ALTIVEC)
688
                postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
689
#else
690
                postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
691 692
#endif
#endif //!RUNTIME_CPUDETECT
693 694
}

695
//static void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
696
//        QP_STORE_T QPs[], int QPStride, int isColor, struct PPMode *ppMode);
697

698 699
/* -pp Command line Help
*/
700
char *pp_help=
701
"Available postprocessing filters:\n"
702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732
"Filters                        Options\n"
"short  long name       short   long option     Description\n"
"*      *               a       autoq           CPU power dependent enabler\n"
"                       c       chrom           chrominance filtering enabled\n"
"                       y       nochrom         chrominance filtering disabled\n"
"                       n       noluma          luma filtering disabled\n"
"hb     hdeblock        (2 threshold)           horizontal deblocking filter\n"
"       1. difference factor: default=32, higher -> more deblocking\n"
"       2. flatness threshold: default=39, lower -> more deblocking\n"
"                       the h & v deblocking filters share these\n"
"                       so you can't set different thresholds for h / v\n"
"vb     vdeblock        (2 threshold)           vertical deblocking filter\n"
"ha     hadeblock       (2 threshold)           horizontal deblocking filter\n"
"va     vadeblock       (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"
"fd     ffmpegdeint                             ffmpeg deinterlacer\n"
"l5     lowpass5                                FIR lowpass deinterlacer\n"
"de     default                                 hb:a,vb:a,dr:a\n"
"fa     fast                                    h1:a,v1:a,dr:a\n"
"ac                                             ha:a:128:7,va:a,dr:a\n"
"tn     tmpnoise        (3 threshold)           temporal noise reducer\n"
"                     1. <= 2. <= 3.            larger -> stronger filtering\n"
"fq     forceQuant      <quantizer>             force quantizer\n"
733 734 735
"Usage:\n"
"<filterName>[:<option>[:<option>...]][[,|/][-]<filterName>[:<option>...]]...\n"
"long form example:\n"
736
"vdeblock:autoq/hdeblock:autoq/linblenddeint    default,-vdeblock\n"
737
"short form example:\n"
738
"vb:a/hb:a/lb                                   de,-vb\n"
739 740
"more examples:\n"
"tn:64:128:256\n"
741
"\n"
Michael Niedermayer's avatar
Michael Niedermayer committed
742
;
743

Michael Niedermayer's avatar
Michael Niedermayer committed
744
pp_mode_t *pp_get_mode_by_name_and_quality(char *name, int quality)
745
{
746 747
        char temp[GET_MODE_BUFFER_SIZE];
        char *p= temp;
748 749
        const char *filterDelimiters= ",/";
        const char *optionDelimiters= ":";
750 751 752
        struct PPMode *ppMode;
        char *filterToken;

Luca Barbato's avatar
Luca Barbato committed
753
        ppMode= av_malloc(sizeof(PPMode));
754 755 756 757 758 759 760 761 762 763 764 765 766 767 768

        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= 256/8;
        ppMode->flatnessThreshold= 56-16-1;
        ppMode->maxClippedThreshold= 0.01;
        ppMode->error=0;

        strncpy(temp, name, GET_MODE_BUFFER_SIZE);

769
        av_log(NULL, AV_LOG_DEBUG, "pp: %s\n", name);
770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786

        for(;;){
                char *filterName;
                int q= 1000000; //PP_QUALITY_MAX;
                int chrom=-1;
                int luma=-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);
787
                av_log(NULL, AV_LOG_DEBUG, "pp: %s::%s\n", filterToken, filterName);
788 789 790 791 792 793 794 795 796 797 798

                if(*filterName == '-')
                {
                        enable=0;
                        filterName++;
                }

                for(;;){ //for all options
                        option= strtok(NULL, optionDelimiters);
                        if(option == NULL) break;

799
                        av_log(NULL, AV_LOG_DEBUG, "pp: option: %s\n", option);
800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925
                        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 if(!strcmp("noluma", option) || !strcmp("n", option)) luma=0;
                        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)
                                {
                                        ppMode->error++;
                                        break;
                                }
                                memmove(p + newlen, p, plen+1);
                                memcpy(p, replaceTable[2*i + 1], newlen);
                                filterNameOk=1;
                        }
                }

                for(i=0; filters[i].shortName!=NULL; i++)
                {
                        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 && luma)
                                        ppMode->lumMode|= filters[i].mask;
                                if(chrom==1 || (chrom==-1 && filters[i].chromDefault))
                                        if(q >= filters[i].minChromQuality)
                                                ppMode->chromMode|= filters[i].mask;

                                if(filters[i].mask == LEVEL_FIX)
                                {
                                        int o;
                                        ppMode->minAllowedY= 16;
                                        ppMode->maxAllowedY= 234;
                                        for(o=0; options[o]!=NULL; o++)
                                        {
                                                if(  !strcmp(options[o],"fullyrange")
                                                   ||!strcmp(options[o],"f"))
                                                {
                                                        ppMode->minAllowedY= 0;
                                                        ppMode->maxAllowedY= 255;
                                                        numOfUnknownOptions--;
                                                }
                                        }
                                }
                                else if(filters[i].mask == TEMP_NOISE_FILTER)
                                {
                                        int o;
                                        int numOfNoises=0;

                                        for(o=0; options[o]!=NULL; o++)
                                        {
                                                char *tail;
                                                ppMode->maxTmpNoise[numOfNoises]=
                                                        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
                                     || filters[i].mask == V_A_DEBLOCK || filters[i].mask == H_A_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;
                                        ppMode->forcedQuant= 15;

                                        for(o=0; options[o]!=NULL && o<1; o++)
                                        {
                                                char *tail;
                                                int val= strtol(options[o], &tail, 0);
                                                if(tail==options[o]) break;

                                                numOfUnknownOptions--;
                                                ppMode->forcedQuant= val;
                                        }
                                }
                        }
                }
                if(!filterNameOk) ppMode->error++;
                ppMode->error += numOfUnknownOptions;
        }

926
        av_log(NULL, AV_LOG_DEBUG, "pp: lumMode=%X, chromMode=%X\n", ppMode->lumMode, ppMode->chromMode);
927 928
        if(ppMode->error)
        {
929
                av_log(NULL, AV_LOG_ERROR, "%d errors in postprocess string \"%s\"\n", ppMode->error, name);
Luca Barbato's avatar
Luca Barbato committed
930
                av_free(ppMode);
931 932 933
                return NULL;
        }
        return ppMode;
934 935
}

Michael Niedermayer's avatar
Michael Niedermayer committed
936
void pp_free_mode(pp_mode_t *mode){
Luca Barbato's avatar
Luca Barbato committed
937
    av_free(mode);
Michael Niedermayer's avatar
Michael Niedermayer committed
938 939
}

940
static void reallocAlign(void **p, int alignment, int size){
941
        av_free(*p);
Luca Barbato's avatar
Luca Barbato committed
942
        *p= av_mallocz(size);
943 944
}

Michael Niedermayer's avatar
Michael Niedermayer committed
945
static void reallocBuffers(PPContext *c, int width, int height, int stride, int qpStride){
946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961
        int mbWidth = (width+15)>>4;
        int mbHeight= (height+15)>>4;
        int i;

        c->stride= stride;
        c->qpStride= qpStride;

        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++)
        {
Diego Biurrun's avatar
Diego Biurrun committed
962
                //Note: The +17*1024 is just there so i do not have to worry about r/w over the end.
963 964 965 966 967 968 969 970
                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, qpStride*mbHeight*sizeof(QP_STORE_T));
        reallocAlign((void **)&c->stdQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T));
        reallocAlign((void **)&c->forcedQPTable, 8, mbWidth*sizeof(QP_STORE_T));
971 972
}

973
static void global_init(void){
974 975 976 977 978
        int i;
        memset(clip_table, 0, 256);
        for(i=256; i<512; i++)
                clip_table[i]= i;
        memset(clip_table+512, 0, 256);
979 980
}

981 982 983 984 985 986
static const char * context_to_name(void * ptr) {
    return "postproc";
}

static AVClass av_codec_context_class = { "Postproc", context_to_name, NULL };

987
pp_context_t *pp_get_context(int width, int height, int cpuCaps){
Luca Barbato's avatar
Luca Barbato committed
988
        PPContext *c= av_malloc(sizeof(PPContext));
989 990
        int stride= (width+15)&(~15);    //assumed / will realloc if needed
        int qpStride= (width+15)/16 + 2; //assumed / will realloc if needed
991

992
        global_init();
993

994
        memset(c, 0, sizeof(PPContext));
995
        c->av_class = &av_codec_context_class;
996 997 998 999 1000 1001 1002 1003
        c->cpuCaps= cpuCaps;
        if(cpuCaps&PP_FORMAT){
                c->hChromaSubSample= cpuCaps&0x3;
                c->vChromaSubSample= (cpuCaps>>4)&0x3;
        }else{
                c->hChromaSubSample= 1;
                c->vChromaSubSample= 1;
        }
1004

1005
        reallocBuffers(c, width, height, stride, qpStride);
1006

1007
        c->frameNum=-1;
1008

1009
        return c;
1010 1011
}

Michael Niedermayer's avatar
Michael Niedermayer committed
1012
void pp_free_context(void *vc){
1013 1014
        PPContext *c = (PPContext*)vc;
        int i;
1015

Luca Barbato's avatar
Luca Barbato committed
1016 1017
        for(i=0; i<3; i++) av_free(c->tempBlured[i]);
        for(i=0; i<3; i++) av_free(c->tempBluredPast[i]);
1018

Luca Barbato's avatar
Luca Barbato committed
1019 1020 1021 1022 1023 1024 1025 1026
        av_free(c->tempBlocks);
        av_free(c->yHistogram);
        av_free(c->tempDst);
        av_free(c->tempSrc);
        av_free(c->deintTemp);
        av_free(c->stdQPTable);
        av_free(c->nonBQPTable);
        av_free(c->forcedQPTable);
1027

1028
        memset(c, 0, sizeof(PPContext));
1029

Luca Barbato's avatar
Luca Barbato committed
1030
        av_free(c);
1031 1032
}

Michael Niedermayer's avatar
Michael Niedermayer committed
1033
void  pp_postprocess(uint8_t * src[3], int srcStride[3],
1034
                 uint8_t * dst[3], int dstStride[3],
1035
                 int width, int height,
1036
                 QP_STORE_T *QP_store,  int QPStride,
1037
                 pp_mode_t *vm,  void *vc, int pict_type)
1038
{
1039 1040 1041 1042
        int mbWidth = (width+15)>>4;
        int mbHeight= (height+15)>>4;
        PPMode *mode = (PPMode*)vm;
        PPContext *c = (PPContext*)vc;
1043 1044
        int minStride= FFMAX(FFABS(srcStride[0]), FFABS(dstStride[0]));
        int absQPStride = FFABS(QPStride);
1045 1046 1047 1048

        // c->stride and c->QPStride are always positive
        if(c->stride < minStride || c->qpStride < absQPStride)
                reallocBuffers(c, width, height,
1049 1050
                                FFMAX(minStride, c->stride),
                                FFMAX(c->qpStride, absQPStride));
1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061

        if(QP_store==NULL || (mode->lumMode & FORCE_QUANT))
        {
                int i;
                QP_store= c->forcedQPTable;
                absQPStride = QPStride = 0;
                if(mode->lumMode & FORCE_QUANT)
                        for(i=0; i<mbWidth; i++) QP_store[i]= mode->forcedQuant;
                else
                        for(i=0; i<mbWidth; i++) QP_store[i]= 1;
        }
Michael Niedermayer's avatar
Michael Niedermayer committed
1062

1063 1064 1065 1066 1067 1068 1069 1070 1071
        if(pict_type & PP_PICT_TYPE_QP2){
                int i;
                const int count= mbHeight * absQPStride;
                for(i=0; i<(count>>2); i++){
                        ((uint32_t*)c->stdQPTable)[i] = (((uint32_t*)QP_store)[i]>>1) & 0x7F7F7F7F;
                }
                for(i<<=2; i<count; i++){
                        c->stdQPTable[i] = QP_store[i]>>1;
                }
Michael Niedermayer's avatar
Michael Niedermayer committed
1072
                QP_store= c->stdQPTable;
1073 1074
                QPStride= absQPStride;
        }
Michael Niedermayer's avatar
Michael Niedermayer committed
1075

1076 1077 1078
if(0){
int x,y;
for(y=0; y<mbHeight; y++){
1079
        for(x=0; x<mbWidth; x++){
1080
                av_log(c, AV_LOG_INFO, "%2d ", QP_store[x + y*QPStride]);
1081
        }
1082
        av_log(c, AV_LOG_INFO, "\n");
1083
}
1084
        av_log(c, AV_LOG_INFO, "\n");
1085
}
Michael Niedermayer's avatar
Michael Niedermayer committed
1086

1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107
        if((pict_type&7)!=3)
        {
                if (QPStride >= 0) {
                        int i;
                        const int count= mbHeight * QPStride;
                        for(i=0; i<(count>>2); i++){
                                ((uint32_t*)c->nonBQPTable)[i] = ((uint32_t*)QP_store)[i] & 0x3F3F3F3F;
                        }
                        for(i<<=2; i<count; i++){
                                c->nonBQPTable[i] = QP_store[i] & 0x3F;
                        }
                } else {
                        int i,j;
                        for(i=0; i<mbHeight; i++) {
                                    for(j=0; j<absQPStride; j++) {
                                        c->nonBQPTable[i*absQPStride+j] = QP_store[i*QPStride+j] & 0x3F;
                                }
                        }
                }
        }

1108 1109
        av_log(c, AV_LOG_DEBUG, "using npp filters 0x%X/0x%X\n",
               mode->lumMode, mode->chromMode);
1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137

        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])
        {
                linecpy(dst[1], src[1], height, srcStride[1]);
                linecpy(dst[2], src[2], height, srcStride[2]);
        }
        else
        {
                int y;
                for(y=0; y<height; 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);
                }
        }
1138 1139
}