yading@11: /*
yading@11:  * Copyright (C) 2001-2003 Michael Niedermayer (michaelni@gmx.at)
yading@11:  *
yading@11:  * AltiVec optimizations (C) 2004 Romain Dolbeau <romain@dolbeau.org>
yading@11:  *
yading@11:  * This file is part of FFmpeg.
yading@11:  *
yading@11:  * FFmpeg is free software; you can redistribute it and/or modify
yading@11:  * it under the terms of the GNU General Public License as published by
yading@11:  * the Free Software Foundation; either version 2 of the License, or
yading@11:  * (at your option) any later version.
yading@11:  *
yading@11:  * FFmpeg is distributed in the hope that it will be useful,
yading@11:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@11:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
yading@11:  * GNU General Public License for more details.
yading@11:  *
yading@11:  * You should have received a copy of the GNU General Public License
yading@11:  * along with FFmpeg; if not, write to the Free Software
yading@11:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@11:  */
yading@11: 
yading@11: /**
yading@11:  * @file
yading@11:  * postprocessing.
yading@11:  */
yading@11: 
yading@11: /*
yading@11:                         C       MMX     MMX2    3DNow   AltiVec
yading@11: isVertDC                Ec      Ec                      Ec
yading@11: isVertMinMaxOk          Ec      Ec                      Ec
yading@11: doVertLowPass           E               e       e       Ec
yading@11: doVertDefFilter         Ec      Ec      e       e       Ec
yading@11: isHorizDC               Ec      Ec                      Ec
yading@11: isHorizMinMaxOk         a       E                       Ec
yading@11: doHorizLowPass          E               e       e       Ec
yading@11: doHorizDefFilter        Ec      Ec      e       e       Ec
yading@11: do_a_deblock            Ec      E       Ec      E
yading@11: deRing                  E               e       e*      Ecp
yading@11: Vertical RKAlgo1        E               a       a
yading@11: Horizontal RKAlgo1                      a       a
yading@11: Vertical X1#            a               E       E
yading@11: Horizontal X1#          a               E       E
yading@11: LinIpolDeinterlace      e               E       E*
yading@11: CubicIpolDeinterlace    a               e       e*
yading@11: LinBlendDeinterlace     e               E       E*
yading@11: MedianDeinterlace#      E       Ec      Ec
yading@11: TempDeNoiser#           E               e       e       Ec
yading@11: 
yading@11: * I do not have a 3DNow! CPU -> it is untested, but no one said it does not work so it seems to work
yading@11: # more or less selfinvented filters so the exactness is not too meaningful
yading@11: E = Exact implementation
yading@11: e = almost exact implementation (slightly different rounding,...)
yading@11: a = alternative / approximate impl
yading@11: c = checked against the other implementations (-vo md5)
yading@11: p = partially optimized, still some work to do
yading@11: */
yading@11: 
yading@11: /*
yading@11: TODO:
yading@11: reduce the time wasted on the mem transfer
yading@11: unroll stuff if instructions depend too much on the prior one
yading@11: move YScale thing to the end instead of fixing QP
yading@11: write a faster and higher quality deblocking filter :)
yading@11: make the mainloop more flexible (variable number of blocks at once
yading@11:         (the if/else stuff per block is slowing things down)
yading@11: compare the quality & speed of all filters
yading@11: split this huge file
yading@11: optimize c versions
yading@11: try to unroll inner for(x=0 ... loop to avoid these damn if(x ... checks
yading@11: ...
yading@11: */
yading@11: 
yading@11: //Changelog: use git log
yading@11: 
yading@11: #include "config.h"
yading@11: #include "libavutil/avutil.h"
yading@11: #include "libavutil/avassert.h"
yading@11: #include <inttypes.h>
yading@11: #include <stdio.h>
yading@11: #include <stdlib.h>
yading@11: #include <string.h>
yading@11: //#undef HAVE_MMXEXT_INLINE
yading@11: //#define HAVE_AMD3DNOW_INLINE
yading@11: //#undef HAVE_MMX_INLINE
yading@11: //#undef ARCH_X86
yading@11: //#define DEBUG_BRIGHTNESS
yading@11: #include "postprocess.h"
yading@11: #include "postprocess_internal.h"
yading@11: #include "libavutil/avstring.h"
yading@11: 
yading@11: unsigned postproc_version(void)
yading@11: {
yading@11:     av_assert0(LIBPOSTPROC_VERSION_MICRO >= 100);
yading@11:     return LIBPOSTPROC_VERSION_INT;
yading@11: }
yading@11: 
yading@11: const char *postproc_configuration(void)
yading@11: {
yading@11:     return FFMPEG_CONFIGURATION;
yading@11: }
yading@11: 
yading@11: const char *postproc_license(void)
yading@11: {
yading@11: #define LICENSE_PREFIX "libpostproc license: "
yading@11:     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
yading@11: }
yading@11: 
yading@11: #if HAVE_ALTIVEC_H
yading@11: #include <altivec.h>
yading@11: #endif
yading@11: 
yading@11: #define GET_MODE_BUFFER_SIZE 500
yading@11: #define OPTIONS_ARRAY_SIZE 10
yading@11: #define BLOCK_SIZE 8
yading@11: #define TEMP_STRIDE 8
yading@11: //#define NUM_BLOCKS_AT_ONCE 16 //not used yet
yading@11: 
yading@11: #if ARCH_X86 && HAVE_INLINE_ASM
yading@11: DECLARE_ASM_CONST(8, uint64_t, w05)= 0x0005000500050005LL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, w04)= 0x0004000400040004LL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, w20)= 0x0020002000200020LL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, b00)= 0x0000000000000000LL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, b01)= 0x0101010101010101LL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, b02)= 0x0202020202020202LL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, b08)= 0x0808080808080808LL;
yading@11: DECLARE_ASM_CONST(8, uint64_t, b80)= 0x8080808080808080LL;
yading@11: #endif
yading@11: 
yading@11: DECLARE_ASM_CONST(8, int, deringThreshold)= 20;
yading@11: 
yading@11: 
yading@11: static const struct PPFilter filters[]=
yading@11: {
yading@11:     {"hb", "hdeblock",              1, 1, 3, H_DEBLOCK},
yading@11:     {"vb", "vdeblock",              1, 2, 4, V_DEBLOCK},
yading@11: /*  {"hr", "rkhdeblock",            1, 1, 3, H_RK1_FILTER},
yading@11:     {"vr", "rkvdeblock",            1, 2, 4, V_RK1_FILTER},*/
yading@11:     {"h1", "x1hdeblock",            1, 1, 3, H_X1_FILTER},
yading@11:     {"v1", "x1vdeblock",            1, 2, 4, V_X1_FILTER},
yading@11:     {"ha", "ahdeblock",             1, 1, 3, H_A_DEBLOCK},
yading@11:     {"va", "avdeblock",             1, 2, 4, V_A_DEBLOCK},
yading@11:     {"dr", "dering",                1, 5, 6, DERING},
yading@11:     {"al", "autolevels",            0, 1, 2, LEVEL_FIX},
yading@11:     {"lb", "linblenddeint",         1, 1, 4, LINEAR_BLEND_DEINT_FILTER},
yading@11:     {"li", "linipoldeint",          1, 1, 4, LINEAR_IPOL_DEINT_FILTER},
yading@11:     {"ci", "cubicipoldeint",        1, 1, 4, CUBIC_IPOL_DEINT_FILTER},
yading@11:     {"md", "mediandeint",           1, 1, 4, MEDIAN_DEINT_FILTER},
yading@11:     {"fd", "ffmpegdeint",           1, 1, 4, FFMPEG_DEINT_FILTER},
yading@11:     {"l5", "lowpass5",              1, 1, 4, LOWPASS5_DEINT_FILTER},
yading@11:     {"tn", "tmpnoise",              1, 7, 8, TEMP_NOISE_FILTER},
yading@11:     {"fq", "forcequant",            1, 0, 0, FORCE_QUANT},
yading@11:     {"be", "bitexact",              1, 0, 0, BITEXACT},
yading@11:     {NULL, NULL,0,0,0,0} //End Marker
yading@11: };
yading@11: 
yading@11: static const char *replaceTable[]=
yading@11: {
yading@11:     "default",      "hb:a,vb:a,dr:a",
yading@11:     "de",           "hb:a,vb:a,dr:a",
yading@11:     "fast",         "h1:a,v1:a,dr:a",
yading@11:     "fa",           "h1:a,v1:a,dr:a",
yading@11:     "ac",           "ha:a:128:7,va:a,dr:a",
yading@11:     NULL //End Marker
yading@11: };
yading@11: 
yading@11: 
yading@11: #if ARCH_X86 && HAVE_INLINE_ASM
yading@11: static inline void prefetchnta(void *p)
yading@11: {
yading@11:     __asm__ volatile(   "prefetchnta (%0)\n\t"
yading@11:         : : "r" (p)
yading@11:     );
yading@11: }
yading@11: 
yading@11: static inline void prefetcht0(void *p)
yading@11: {
yading@11:     __asm__ volatile(   "prefetcht0 (%0)\n\t"
yading@11:         : : "r" (p)
yading@11:     );
yading@11: }
yading@11: 
yading@11: static inline void prefetcht1(void *p)
yading@11: {
yading@11:     __asm__ volatile(   "prefetcht1 (%0)\n\t"
yading@11:         : : "r" (p)
yading@11:     );
yading@11: }
yading@11: 
yading@11: static inline void prefetcht2(void *p)
yading@11: {
yading@11:     __asm__ volatile(   "prefetcht2 (%0)\n\t"
yading@11:         : : "r" (p)
yading@11:     );
yading@11: }
yading@11: #endif
yading@11: 
yading@11: /* The horizontal functions exist only in C because the MMX
yading@11:  * code is faster with vertical filters and transposing. */
yading@11: 
yading@11: /**
yading@11:  * Check if the given 8x8 Block is mostly "flat"
yading@11:  */
yading@11: static inline int isHorizDC_C(const uint8_t src[], int stride, const PPContext *c)
yading@11: {
yading@11:     int numEq= 0;
yading@11:     int y;
yading@11:     const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
yading@11:     const int dcThreshold= dcOffset*2 + 1;
yading@11: 
yading@11:     for(y=0; y<BLOCK_SIZE; y++){
yading@11:         if(((unsigned)(src[0] - src[1] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[1] - src[2] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[2] - src[3] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[3] - src[4] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[4] - src[5] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[5] - src[6] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[6] - src[7] + dcOffset)) < dcThreshold) numEq++;
yading@11:         src+= stride;
yading@11:     }
yading@11:     return numEq > c->ppMode.flatnessThreshold;
yading@11: }
yading@11: 
yading@11: /**
yading@11:  * Check if the middle 8x8 Block in the given 8x16 block is flat
yading@11:  */
yading@11: static inline int isVertDC_C(const uint8_t src[], int stride, const PPContext *c)
yading@11: {
yading@11:     int numEq= 0;
yading@11:     int y;
yading@11:     const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
yading@11:     const int dcThreshold= dcOffset*2 + 1;
yading@11: 
yading@11:     src+= stride*4; // src points to begin of the 8x8 Block
yading@11:     for(y=0; y<BLOCK_SIZE-1; y++){
yading@11:         if(((unsigned)(src[0] - src[0+stride] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[1] - src[1+stride] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[2] - src[2+stride] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[3] - src[3+stride] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[4] - src[4+stride] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[5] - src[5+stride] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[6] - src[6+stride] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[7] - src[7+stride] + dcOffset)) < dcThreshold) numEq++;
yading@11:         src+= stride;
yading@11:     }
yading@11:     return numEq > c->ppMode.flatnessThreshold;
yading@11: }
yading@11: 
yading@11: static inline int isHorizMinMaxOk_C(const uint8_t src[], int stride, int QP)
yading@11: {
yading@11:     int i;
yading@11:     for(i=0; i<2; i++){
yading@11:         if((unsigned)(src[0] - src[5] + 2*QP) > 4*QP) return 0;
yading@11:         src += stride;
yading@11:         if((unsigned)(src[2] - src[7] + 2*QP) > 4*QP) return 0;
yading@11:         src += stride;
yading@11:         if((unsigned)(src[4] - src[1] + 2*QP) > 4*QP) return 0;
yading@11:         src += stride;
yading@11:         if((unsigned)(src[6] - src[3] + 2*QP) > 4*QP) return 0;
yading@11:         src += stride;
yading@11:     }
yading@11:     return 1;
yading@11: }
yading@11: 
yading@11: static inline int isVertMinMaxOk_C(const uint8_t src[], int stride, int QP)
yading@11: {
yading@11:     int x;
yading@11:     src+= stride*4;
yading@11:     for(x=0; x<BLOCK_SIZE; x+=4){
yading@11:         if((unsigned)(src[  x + 0*stride] - src[  x + 5*stride] + 2*QP) > 4*QP) return 0;
yading@11:         if((unsigned)(src[1+x + 2*stride] - src[1+x + 7*stride] + 2*QP) > 4*QP) return 0;
yading@11:         if((unsigned)(src[2+x + 4*stride] - src[2+x + 1*stride] + 2*QP) > 4*QP) return 0;
yading@11:         if((unsigned)(src[3+x + 6*stride] - src[3+x + 3*stride] + 2*QP) > 4*QP) return 0;
yading@11:     }
yading@11:     return 1;
yading@11: }
yading@11: 
yading@11: static inline int horizClassify_C(const uint8_t src[], int stride, const PPContext *c)
yading@11: {
yading@11:     if( isHorizDC_C(src, stride, c) ){
yading@11:         if( isHorizMinMaxOk_C(src, stride, c->QP) )
yading@11:             return 1;
yading@11:         else
yading@11:             return 0;
yading@11:     }else{
yading@11:         return 2;
yading@11:     }
yading@11: }
yading@11: 
yading@11: static inline int vertClassify_C(const uint8_t src[], int stride, const PPContext *c)
yading@11: {
yading@11:     if( isVertDC_C(src, stride, c) ){
yading@11:         if( isVertMinMaxOk_C(src, stride, c->QP) )
yading@11:             return 1;
yading@11:         else
yading@11:             return 0;
yading@11:     }else{
yading@11:         return 2;
yading@11:     }
yading@11: }
yading@11: 
yading@11: static inline void doHorizDefFilter_C(uint8_t dst[], int stride, const PPContext *c)
yading@11: {
yading@11:     int y;
yading@11:     for(y=0; y<BLOCK_SIZE; y++){
yading@11:         const int middleEnergy= 5*(dst[4] - dst[3]) + 2*(dst[2] - dst[5]);
yading@11: 
yading@11:         if(FFABS(middleEnergy) < 8*c->QP){
yading@11:             const int q=(dst[3] - dst[4])/2;
yading@11:             const int leftEnergy=  5*(dst[2] - dst[1]) + 2*(dst[0] - dst[3]);
yading@11:             const int rightEnergy= 5*(dst[6] - dst[5]) + 2*(dst[4] - dst[7]);
yading@11: 
yading@11:             int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
yading@11:             d= FFMAX(d, 0);
yading@11: 
yading@11:             d= (5*d + 32) >> 6;
yading@11:             d*= FFSIGN(-middleEnergy);
yading@11: 
yading@11:             if(q>0)
yading@11:             {
yading@11:                 d= d<0 ? 0 : d;
yading@11:                 d= d>q ? q : d;
yading@11:             }
yading@11:             else
yading@11:             {
yading@11:                 d= d>0 ? 0 : d;
yading@11:                 d= d<q ? q : d;
yading@11:             }
yading@11: 
yading@11:             dst[3]-= d;
yading@11:             dst[4]+= d;
yading@11:         }
yading@11:         dst+= stride;
yading@11:     }
yading@11: }
yading@11: 
yading@11: /**
yading@11:  * Do a horizontal low pass filter on the 10x8 block (dst points to middle 8x8 Block)
yading@11:  * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16 (C version)
yading@11:  */
yading@11: static inline void doHorizLowPass_C(uint8_t dst[], int stride, const PPContext *c)
yading@11: {
yading@11:     int y;
yading@11:     for(y=0; y<BLOCK_SIZE; y++){
yading@11:         const int first= FFABS(dst[-1] - dst[0]) < c->QP ? dst[-1] : dst[0];
yading@11:         const int last= FFABS(dst[8] - dst[7]) < c->QP ? dst[8] : dst[7];
yading@11: 
yading@11:         int sums[10];
yading@11:         sums[0] = 4*first + dst[0] + dst[1] + dst[2] + 4;
yading@11:         sums[1] = sums[0] - first  + dst[3];
yading@11:         sums[2] = sums[1] - first  + dst[4];
yading@11:         sums[3] = sums[2] - first  + dst[5];
yading@11:         sums[4] = sums[3] - first  + dst[6];
yading@11:         sums[5] = sums[4] - dst[0] + dst[7];
yading@11:         sums[6] = sums[5] - dst[1] + last;
yading@11:         sums[7] = sums[6] - dst[2] + last;
yading@11:         sums[8] = sums[7] - dst[3] + last;
yading@11:         sums[9] = sums[8] - dst[4] + last;
yading@11: 
yading@11:         dst[0]= (sums[0] + sums[2] + 2*dst[0])>>4;
yading@11:         dst[1]= (sums[1] + sums[3] + 2*dst[1])>>4;
yading@11:         dst[2]= (sums[2] + sums[4] + 2*dst[2])>>4;
yading@11:         dst[3]= (sums[3] + sums[5] + 2*dst[3])>>4;
yading@11:         dst[4]= (sums[4] + sums[6] + 2*dst[4])>>4;
yading@11:         dst[5]= (sums[5] + sums[7] + 2*dst[5])>>4;
yading@11:         dst[6]= (sums[6] + sums[8] + 2*dst[6])>>4;
yading@11:         dst[7]= (sums[7] + sums[9] + 2*dst[7])>>4;
yading@11: 
yading@11:         dst+= stride;
yading@11:     }
yading@11: }
yading@11: 
yading@11: /**
yading@11:  * Experimental Filter 1 (Horizontal)
yading@11:  * will not damage linear gradients
yading@11:  * Flat blocks should look like they were passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
yading@11:  * can only smooth blocks at the expected locations (it cannot smooth them if they did move)
yading@11:  * MMX2 version does correct clipping C version does not
yading@11:  * not identical with the vertical one
yading@11:  */
yading@11: static inline void horizX1Filter(uint8_t *src, int stride, int QP)
yading@11: {
yading@11:     int y;
yading@11:     static uint64_t lut[256];
yading@11:     if(!lut[255])
yading@11:     {
yading@11:         int i;
yading@11:         for(i=0; i<256; i++)
yading@11:         {
yading@11:             int v= i < 128 ? 2*i : 2*(i-256);
yading@11: /*
yading@11: //Simulate 112242211 9-Tap filter
yading@11:             uint64_t a= (v/16)  & 0xFF;
yading@11:             uint64_t b= (v/8)   & 0xFF;
yading@11:             uint64_t c= (v/4)   & 0xFF;
yading@11:             uint64_t d= (3*v/8) & 0xFF;
yading@11: */
yading@11: //Simulate piecewise linear interpolation
yading@11:             uint64_t a= (v/16)   & 0xFF;
yading@11:             uint64_t b= (v*3/16) & 0xFF;
yading@11:             uint64_t c= (v*5/16) & 0xFF;
yading@11:             uint64_t d= (7*v/16) & 0xFF;
yading@11:             uint64_t A= (0x100 - a)&0xFF;
yading@11:             uint64_t B= (0x100 - b)&0xFF;
yading@11:             uint64_t C= (0x100 - c)&0xFF;
yading@11:             uint64_t D= (0x100 - c)&0xFF;
yading@11: 
yading@11:             lut[i]   = (a<<56) | (b<<48) | (c<<40) | (d<<32) |
yading@11:                        (D<<24) | (C<<16) | (B<<8)  | (A);
yading@11:             //lut[i] = (v<<32) | (v<<24);
yading@11:         }
yading@11:     }
yading@11: 
yading@11:     for(y=0; y<BLOCK_SIZE; y++){
yading@11:         int a= src[1] - src[2];
yading@11:         int b= src[3] - src[4];
yading@11:         int c= src[5] - src[6];
yading@11: 
yading@11:         int d= FFMAX(FFABS(b) - (FFABS(a) + FFABS(c))/2, 0);
yading@11: 
yading@11:         if(d < QP){
yading@11:             int v = d * FFSIGN(-b);
yading@11: 
yading@11:             src[1] +=v/8;
yading@11:             src[2] +=v/4;
yading@11:             src[3] +=3*v/8;
yading@11:             src[4] -=3*v/8;
yading@11:             src[5] -=v/4;
yading@11:             src[6] -=v/8;
yading@11:         }
yading@11:         src+=stride;
yading@11:     }
yading@11: }
yading@11: 
yading@11: /**
yading@11:  * accurate deblock filter
yading@11:  */
yading@11: static av_always_inline void do_a_deblock_C(uint8_t *src, int step,
yading@11:                                             int stride, const PPContext *c)
yading@11: {
yading@11:     int y;
yading@11:     const int QP= c->QP;
yading@11:     const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
yading@11:     const int dcThreshold= dcOffset*2 + 1;
yading@11: //START_TIMER
yading@11:     src+= step*4; // src points to begin of the 8x8 Block
yading@11:     for(y=0; y<8; y++){
yading@11:         int numEq= 0;
yading@11: 
yading@11:         if(((unsigned)(src[-1*step] - src[0*step] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[ 0*step] - src[1*step] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[ 1*step] - src[2*step] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[ 2*step] - src[3*step] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[ 3*step] - src[4*step] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[ 4*step] - src[5*step] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[ 5*step] - src[6*step] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[ 6*step] - src[7*step] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(((unsigned)(src[ 7*step] - src[8*step] + dcOffset)) < dcThreshold) numEq++;
yading@11:         if(numEq > c->ppMode.flatnessThreshold){
yading@11:             int min, max, x;
yading@11: 
yading@11:             if(src[0] > src[step]){
yading@11:                 max= src[0];
yading@11:                 min= src[step];
yading@11:             }else{
yading@11:                 max= src[step];
yading@11:                 min= src[0];
yading@11:             }
yading@11:             for(x=2; x<8; x+=2){
yading@11:                 if(src[x*step] > src[(x+1)*step]){
yading@11:                         if(src[x    *step] > max) max= src[ x   *step];
yading@11:                         if(src[(x+1)*step] < min) min= src[(x+1)*step];
yading@11:                 }else{
yading@11:                         if(src[(x+1)*step] > max) max= src[(x+1)*step];
yading@11:                         if(src[ x   *step] < min) min= src[ x   *step];
yading@11:                 }
yading@11:             }
yading@11:             if(max-min < 2*QP){
yading@11:                 const int first= FFABS(src[-1*step] - src[0]) < QP ? src[-1*step] : src[0];
yading@11:                 const int last= FFABS(src[8*step] - src[7*step]) < QP ? src[8*step] : src[7*step];
yading@11: 
yading@11:                 int sums[10];
yading@11:                 sums[0] = 4*first + src[0*step] + src[1*step] + src[2*step] + 4;
yading@11:                 sums[1] = sums[0] - first       + src[3*step];
yading@11:                 sums[2] = sums[1] - first       + src[4*step];
yading@11:                 sums[3] = sums[2] - first       + src[5*step];
yading@11:                 sums[4] = sums[3] - first       + src[6*step];
yading@11:                 sums[5] = sums[4] - src[0*step] + src[7*step];
yading@11:                 sums[6] = sums[5] - src[1*step] + last;
yading@11:                 sums[7] = sums[6] - src[2*step] + last;
yading@11:                 sums[8] = sums[7] - src[3*step] + last;
yading@11:                 sums[9] = sums[8] - src[4*step] + last;
yading@11: 
yading@11:                 src[0*step]= (sums[0] + sums[2] + 2*src[0*step])>>4;
yading@11:                 src[1*step]= (sums[1] + sums[3] + 2*src[1*step])>>4;
yading@11:                 src[2*step]= (sums[2] + sums[4] + 2*src[2*step])>>4;
yading@11:                 src[3*step]= (sums[3] + sums[5] + 2*src[3*step])>>4;
yading@11:                 src[4*step]= (sums[4] + sums[6] + 2*src[4*step])>>4;
yading@11:                 src[5*step]= (sums[5] + sums[7] + 2*src[5*step])>>4;
yading@11:                 src[6*step]= (sums[6] + sums[8] + 2*src[6*step])>>4;
yading@11:                 src[7*step]= (sums[7] + sums[9] + 2*src[7*step])>>4;
yading@11:             }
yading@11:         }else{
yading@11:             const int middleEnergy= 5*(src[4*step] - src[3*step]) + 2*(src[2*step] - src[5*step]);
yading@11: 
yading@11:             if(FFABS(middleEnergy) < 8*QP){
yading@11:                 const int q=(src[3*step] - src[4*step])/2;
yading@11:                 const int leftEnergy=  5*(src[2*step] - src[1*step]) + 2*(src[0*step] - src[3*step]);
yading@11:                 const int rightEnergy= 5*(src[6*step] - src[5*step]) + 2*(src[4*step] - src[7*step]);
yading@11: 
yading@11:                 int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
yading@11:                 d= FFMAX(d, 0);
yading@11: 
yading@11:                 d= (5*d + 32) >> 6;
yading@11:                 d*= FFSIGN(-middleEnergy);
yading@11: 
yading@11:                 if(q>0){
yading@11:                     d= d<0 ? 0 : d;
yading@11:                     d= d>q ? q : d;
yading@11:                 }else{
yading@11:                     d= d>0 ? 0 : d;
yading@11:                     d= d<q ? q : d;
yading@11:                 }
yading@11: 
yading@11:                 src[3*step]-= d;
yading@11:                 src[4*step]+= d;
yading@11:             }
yading@11:         }
yading@11: 
yading@11:         src += stride;
yading@11:     }
yading@11: /*if(step==16){
yading@11:     STOP_TIMER("step16")
yading@11: }else{
yading@11:     STOP_TIMER("stepX")
yading@11: }*/
yading@11: }
yading@11: 
yading@11: //Note: we have C, MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one
yading@11: //Plain C versions
yading@11: //we always compile C for testing which needs bitexactness
yading@11: #define TEMPLATE_PP_C 1
yading@11: #include "postprocess_template.c"
yading@11: 
yading@11: #if HAVE_ALTIVEC
yading@11: #   define TEMPLATE_PP_ALTIVEC 1
yading@11: #   include "postprocess_altivec_template.c"
yading@11: #   include "postprocess_template.c"
yading@11: #endif
yading@11: 
yading@11: #if ARCH_X86 && HAVE_INLINE_ASM
yading@11: #    if CONFIG_RUNTIME_CPUDETECT
yading@11: #        define TEMPLATE_PP_MMX 1
yading@11: #        include "postprocess_template.c"
yading@11: #        define TEMPLATE_PP_MMXEXT 1
yading@11: #        include "postprocess_template.c"
yading@11: #        define TEMPLATE_PP_3DNOW 1
yading@11: #        include "postprocess_template.c"
yading@11: #        define TEMPLATE_PP_SSE2 1
yading@11: #        include "postprocess_template.c"
yading@11: #    else
yading@11: #        if HAVE_SSE2_INLINE
yading@11: #            define TEMPLATE_PP_SSE2 1
yading@11: #            include "postprocess_template.c"
yading@11: #        elif HAVE_MMXEXT_INLINE
yading@11: #            define TEMPLATE_PP_MMXEXT 1
yading@11: #            include "postprocess_template.c"
yading@11: #        elif HAVE_AMD3DNOW_INLINE
yading@11: #            define TEMPLATE_PP_3DNOW 1
yading@11: #            include "postprocess_template.c"
yading@11: #        elif HAVE_MMX_INLINE
yading@11: #            define TEMPLATE_PP_MMX 1
yading@11: #            include "postprocess_template.c"
yading@11: #        endif
yading@11: #    endif
yading@11: #endif
yading@11: 
yading@11: typedef void (*pp_fn)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
yading@11:                       const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2);
yading@11: 
yading@11: static inline void postProcess(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
yading@11:         const QP_STORE_T QPs[], int QPStride, int isColor, pp_mode *vm, pp_context *vc)
yading@11: {
yading@11:     pp_fn pp = postProcess_C;
yading@11:     PPContext *c= (PPContext *)vc;
yading@11:     PPMode *ppMode= (PPMode *)vm;
yading@11:     c->ppMode= *ppMode; //FIXME
yading@11: 
yading@11:     if (!(ppMode->lumMode & BITEXACT)) {
yading@11: #if CONFIG_RUNTIME_CPUDETECT
yading@11: #if ARCH_X86 && HAVE_INLINE_ASM
yading@11:         // ordered per speed fastest first
yading@11:         if      (c->cpuCaps & AV_CPU_FLAG_SSE2)     pp = postProcess_SSE2;
yading@11:         else if (c->cpuCaps & AV_CPU_FLAG_MMXEXT)   pp = postProcess_MMX2;
yading@11:         else if (c->cpuCaps & AV_CPU_FLAG_3DNOW)    pp = postProcess_3DNow;
yading@11:         else if (c->cpuCaps & AV_CPU_FLAG_MMX)      pp = postProcess_MMX;
yading@11: #elif HAVE_ALTIVEC
yading@11:         if      (c->cpuCaps & AV_CPU_FLAG_ALTIVEC)  pp = postProcess_altivec;
yading@11: #endif
yading@11: #else /* CONFIG_RUNTIME_CPUDETECT */
yading@11: #if     HAVE_SSE2_INLINE
yading@11:         pp = postProcess_SSE2;
yading@11: #elif   HAVE_MMXEXT_INLINE
yading@11:         pp = postProcess_MMX2;
yading@11: #elif HAVE_AMD3DNOW_INLINE
yading@11:         pp = postProcess_3DNow;
yading@11: #elif HAVE_MMX_INLINE
yading@11:         pp = postProcess_MMX;
yading@11: #elif HAVE_ALTIVEC
yading@11:         pp = postProcess_altivec;
yading@11: #endif
yading@11: #endif /* !CONFIG_RUNTIME_CPUDETECT */
yading@11:     }
yading@11: 
yading@11:     pp(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
yading@11: }
yading@11: 
yading@11: /* -pp Command line Help
yading@11: */
yading@11: const char pp_help[] =
yading@11: "Available postprocessing filters:\n"
yading@11: "Filters                        Options\n"
yading@11: "short  long name       short   long option     Description\n"
yading@11: "*      *               a       autoq           CPU power dependent enabler\n"
yading@11: "                       c       chrom           chrominance filtering enabled\n"
yading@11: "                       y       nochrom         chrominance filtering disabled\n"
yading@11: "                       n       noluma          luma filtering disabled\n"
yading@11: "hb     hdeblock        (2 threshold)           horizontal deblocking filter\n"
yading@11: "       1. difference factor: default=32, higher -> more deblocking\n"
yading@11: "       2. flatness threshold: default=39, lower -> more deblocking\n"
yading@11: "                       the h & v deblocking filters share these\n"
yading@11: "                       so you can't set different thresholds for h / v\n"
yading@11: "vb     vdeblock        (2 threshold)           vertical deblocking filter\n"
yading@11: "ha     hadeblock       (2 threshold)           horizontal deblocking filter\n"
yading@11: "va     vadeblock       (2 threshold)           vertical deblocking filter\n"
yading@11: "h1     x1hdeblock                              experimental h deblock filter 1\n"
yading@11: "v1     x1vdeblock                              experimental v deblock filter 1\n"
yading@11: "dr     dering                                  deringing filter\n"
yading@11: "al     autolevels                              automatic brightness / contrast\n"
yading@11: "                       f        fullyrange     stretch luminance to (0..255)\n"
yading@11: "lb     linblenddeint                           linear blend deinterlacer\n"
yading@11: "li     linipoldeint                            linear interpolating deinterlace\n"
yading@11: "ci     cubicipoldeint                          cubic interpolating deinterlacer\n"
yading@11: "md     mediandeint                             median deinterlacer\n"
yading@11: "fd     ffmpegdeint                             ffmpeg deinterlacer\n"
yading@11: "l5     lowpass5                                FIR lowpass deinterlacer\n"
yading@11: "de     default                                 hb:a,vb:a,dr:a\n"
yading@11: "fa     fast                                    h1:a,v1:a,dr:a\n"
yading@11: "ac                                             ha:a:128:7,va:a,dr:a\n"
yading@11: "tn     tmpnoise        (3 threshold)           temporal noise reducer\n"
yading@11: "                     1. <= 2. <= 3.            larger -> stronger filtering\n"
yading@11: "fq     forceQuant      <quantizer>             force quantizer\n"
yading@11: "Usage:\n"
yading@11: "<filterName>[:<option>[:<option>...]][[,|/][-]<filterName>[:<option>...]]...\n"
yading@11: "long form example:\n"
yading@11: "vdeblock:autoq/hdeblock:autoq/linblenddeint    default,-vdeblock\n"
yading@11: "short form example:\n"
yading@11: "vb:a/hb:a/lb                                   de,-vb\n"
yading@11: "more examples:\n"
yading@11: "tn:64:128:256\n"
yading@11: "\n"
yading@11: ;
yading@11: 
yading@11: pp_mode *pp_get_mode_by_name_and_quality(const char *name, int quality)
yading@11: {
yading@11:     char temp[GET_MODE_BUFFER_SIZE];
yading@11:     char *p= temp;
yading@11:     static const char filterDelimiters[] = ",/";
yading@11:     static const char optionDelimiters[] = ":|";
yading@11:     struct PPMode *ppMode;
yading@11:     char *filterToken;
yading@11: 
yading@11:     if (!name)  {
yading@11:         av_log(NULL, AV_LOG_ERROR, "pp: Missing argument\n");
yading@11:         return NULL;
yading@11:     }
yading@11: 
yading@11:     if (!strcmp(name, "help")) {
yading@11:         const char *p;
yading@11:         for (p = pp_help; strchr(p, '\n'); p = strchr(p, '\n') + 1) {
yading@11:             av_strlcpy(temp, p, FFMIN(sizeof(temp), strchr(p, '\n') - p + 2));
yading@11:             av_log(NULL, AV_LOG_INFO, "%s", temp);
yading@11:         }
yading@11:         return NULL;
yading@11:     }
yading@11: 
yading@11:     ppMode= av_malloc(sizeof(PPMode));
yading@11: 
yading@11:     ppMode->lumMode= 0;
yading@11:     ppMode->chromMode= 0;
yading@11:     ppMode->maxTmpNoise[0]= 700;
yading@11:     ppMode->maxTmpNoise[1]= 1500;
yading@11:     ppMode->maxTmpNoise[2]= 3000;
yading@11:     ppMode->maxAllowedY= 234;
yading@11:     ppMode->minAllowedY= 16;
yading@11:     ppMode->baseDcDiff= 256/8;
yading@11:     ppMode->flatnessThreshold= 56-16-1;
yading@11:     ppMode->maxClippedThreshold= 0.01;
yading@11:     ppMode->error=0;
yading@11: 
yading@11:     memset(temp, 0, GET_MODE_BUFFER_SIZE);
yading@11:     av_strlcpy(temp, name, GET_MODE_BUFFER_SIZE - 1);
yading@11: 
yading@11:     av_log(NULL, AV_LOG_DEBUG, "pp: %s\n", name);
yading@11: 
yading@11:     for(;;){
yading@11:         char *filterName;
yading@11:         int q= 1000000; //PP_QUALITY_MAX;
yading@11:         int chrom=-1;
yading@11:         int luma=-1;
yading@11:         char *option;
yading@11:         char *options[OPTIONS_ARRAY_SIZE];
yading@11:         int i;
yading@11:         int filterNameOk=0;
yading@11:         int numOfUnknownOptions=0;
yading@11:         int enable=1; //does the user want us to enabled or disabled the filter
yading@11: 
yading@11:         filterToken= strtok(p, filterDelimiters);
yading@11:         if(filterToken == NULL) break;
yading@11:         p+= strlen(filterToken) + 1; // p points to next filterToken
yading@11:         filterName= strtok(filterToken, optionDelimiters);
yading@11:         av_log(NULL, AV_LOG_DEBUG, "pp: %s::%s\n", filterToken, filterName);
yading@11: 
yading@11:         if(*filterName == '-'){
yading@11:             enable=0;
yading@11:             filterName++;
yading@11:         }
yading@11: 
yading@11:         for(;;){ //for all options
yading@11:             option= strtok(NULL, optionDelimiters);
yading@11:             if(option == NULL) break;
yading@11: 
yading@11:             av_log(NULL, AV_LOG_DEBUG, "pp: option: %s\n", option);
yading@11:             if(!strcmp("autoq", option) || !strcmp("a", option)) q= quality;
yading@11:             else if(!strcmp("nochrom", option) || !strcmp("y", option)) chrom=0;
yading@11:             else if(!strcmp("chrom", option) || !strcmp("c", option)) chrom=1;
yading@11:             else if(!strcmp("noluma", option) || !strcmp("n", option)) luma=0;
yading@11:             else{
yading@11:                 options[numOfUnknownOptions] = option;
yading@11:                 numOfUnknownOptions++;
yading@11:             }
yading@11:             if(numOfUnknownOptions >= OPTIONS_ARRAY_SIZE-1) break;
yading@11:         }
yading@11:         options[numOfUnknownOptions] = NULL;
yading@11: 
yading@11:         /* replace stuff from the replace Table */
yading@11:         for(i=0; replaceTable[2*i]!=NULL; i++){
yading@11:             if(!strcmp(replaceTable[2*i], filterName)){
yading@11:                 int newlen= strlen(replaceTable[2*i + 1]);
yading@11:                 int plen;
yading@11:                 int spaceLeft;
yading@11: 
yading@11:                 p--, *p=',';
yading@11: 
yading@11:                 plen= strlen(p);
yading@11:                 spaceLeft= p - temp + plen;
yading@11:                 if(spaceLeft + newlen  >= GET_MODE_BUFFER_SIZE - 1){
yading@11:                     ppMode->error++;
yading@11:                     break;
yading@11:                 }
yading@11:                 memmove(p + newlen, p, plen+1);
yading@11:                 memcpy(p, replaceTable[2*i + 1], newlen);
yading@11:                 filterNameOk=1;
yading@11:             }
yading@11:         }
yading@11: 
yading@11:         for(i=0; filters[i].shortName!=NULL; i++){
yading@11:             if(   !strcmp(filters[i].longName, filterName)
yading@11:                || !strcmp(filters[i].shortName, filterName)){
yading@11:                 ppMode->lumMode &= ~filters[i].mask;
yading@11:                 ppMode->chromMode &= ~filters[i].mask;
yading@11: 
yading@11:                 filterNameOk=1;
yading@11:                 if(!enable) break; // user wants to disable it
yading@11: 
yading@11:                 if(q >= filters[i].minLumQuality && luma)
yading@11:                     ppMode->lumMode|= filters[i].mask;
yading@11:                 if(chrom==1 || (chrom==-1 && filters[i].chromDefault))
yading@11:                     if(q >= filters[i].minChromQuality)
yading@11:                             ppMode->chromMode|= filters[i].mask;
yading@11: 
yading@11:                 if(filters[i].mask == LEVEL_FIX){
yading@11:                     int o;
yading@11:                     ppMode->minAllowedY= 16;
yading@11:                     ppMode->maxAllowedY= 234;
yading@11:                     for(o=0; options[o]!=NULL; o++){
yading@11:                         if(  !strcmp(options[o],"fullyrange")
yading@11:                            ||!strcmp(options[o],"f")){
yading@11:                             ppMode->minAllowedY= 0;
yading@11:                             ppMode->maxAllowedY= 255;
yading@11:                             numOfUnknownOptions--;
yading@11:                         }
yading@11:                     }
yading@11:                 }
yading@11:                 else if(filters[i].mask == TEMP_NOISE_FILTER)
yading@11:                 {
yading@11:                     int o;
yading@11:                     int numOfNoises=0;
yading@11: 
yading@11:                     for(o=0; options[o]!=NULL; o++){
yading@11:                         char *tail;
yading@11:                         ppMode->maxTmpNoise[numOfNoises]=
yading@11:                             strtol(options[o], &tail, 0);
yading@11:                         if(tail!=options[o]){
yading@11:                             numOfNoises++;
yading@11:                             numOfUnknownOptions--;
yading@11:                             if(numOfNoises >= 3) break;
yading@11:                         }
yading@11:                     }
yading@11:                 }
yading@11:                 else if(filters[i].mask == V_DEBLOCK   || filters[i].mask == H_DEBLOCK
yading@11:                      || filters[i].mask == V_A_DEBLOCK || filters[i].mask == H_A_DEBLOCK){
yading@11:                     int o;
yading@11: 
yading@11:                     for(o=0; options[o]!=NULL && o<2; o++){
yading@11:                         char *tail;
yading@11:                         int val= strtol(options[o], &tail, 0);
yading@11:                         if(tail==options[o]) break;
yading@11: 
yading@11:                         numOfUnknownOptions--;
yading@11:                         if(o==0) ppMode->baseDcDiff= val;
yading@11:                         else ppMode->flatnessThreshold= val;
yading@11:                     }
yading@11:                 }
yading@11:                 else if(filters[i].mask == FORCE_QUANT){
yading@11:                     int o;
yading@11:                     ppMode->forcedQuant= 15;
yading@11: 
yading@11:                     for(o=0; options[o]!=NULL && o<1; o++){
yading@11:                         char *tail;
yading@11:                         int val= strtol(options[o], &tail, 0);
yading@11:                         if(tail==options[o]) break;
yading@11: 
yading@11:                         numOfUnknownOptions--;
yading@11:                         ppMode->forcedQuant= val;
yading@11:                     }
yading@11:                 }
yading@11:             }
yading@11:         }
yading@11:         if(!filterNameOk) ppMode->error++;
yading@11:         ppMode->error += numOfUnknownOptions;
yading@11:     }
yading@11: 
yading@11:     av_log(NULL, AV_LOG_DEBUG, "pp: lumMode=%X, chromMode=%X\n", ppMode->lumMode, ppMode->chromMode);
yading@11:     if(ppMode->error){
yading@11:         av_log(NULL, AV_LOG_ERROR, "%d errors in postprocess string \"%s\"\n", ppMode->error, name);
yading@11:         av_free(ppMode);
yading@11:         return NULL;
yading@11:     }
yading@11:     return ppMode;
yading@11: }
yading@11: 
yading@11: void pp_free_mode(pp_mode *mode){
yading@11:     av_free(mode);
yading@11: }
yading@11: 
yading@11: static void reallocAlign(void **p, int alignment, int size){
yading@11:     av_free(*p);
yading@11:     *p= av_mallocz(size);
yading@11: }
yading@11: 
yading@11: static void reallocBuffers(PPContext *c, int width, int height, int stride, int qpStride){
yading@11:     int mbWidth = (width+15)>>4;
yading@11:     int mbHeight= (height+15)>>4;
yading@11:     int i;
yading@11: 
yading@11:     c->stride= stride;
yading@11:     c->qpStride= qpStride;
yading@11: 
yading@11:     reallocAlign((void **)&c->tempDst, 8, stride*24+32);
yading@11:     reallocAlign((void **)&c->tempSrc, 8, stride*24);
yading@11:     reallocAlign((void **)&c->tempBlocks, 8, 2*16*8);
yading@11:     reallocAlign((void **)&c->yHistogram, 8, 256*sizeof(uint64_t));
yading@11:     for(i=0; i<256; i++)
yading@11:             c->yHistogram[i]= width*height/64*15/256;
yading@11: 
yading@11:     for(i=0; i<3; i++){
yading@11:         //Note: The +17*1024 is just there so I do not have to worry about r/w over the end.
yading@11:         reallocAlign((void **)&c->tempBlurred[i], 8, stride*mbHeight*16 + 17*1024);
yading@11:         reallocAlign((void **)&c->tempBlurredPast[i], 8, 256*((height+7)&(~7))/2 + 17*1024);//FIXME size
yading@11:     }
yading@11: 
yading@11:     reallocAlign((void **)&c->deintTemp, 8, 2*width+32);
yading@11:     reallocAlign((void **)&c->nonBQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T));
yading@11:     reallocAlign((void **)&c->stdQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T));
yading@11:     reallocAlign((void **)&c->forcedQPTable, 8, mbWidth*sizeof(QP_STORE_T));
yading@11: }
yading@11: 
yading@11: static const char * context_to_name(void * ptr) {
yading@11:     return "postproc";
yading@11: }
yading@11: 
yading@11: static const AVClass av_codec_context_class = { "Postproc", context_to_name, NULL };
yading@11: 
yading@11: pp_context *pp_get_context(int width, int height, int cpuCaps){
yading@11:     PPContext *c= av_malloc(sizeof(PPContext));
yading@11:     int stride= FFALIGN(width, 16);  //assumed / will realloc if needed
yading@11:     int qpStride= (width+15)/16 + 2; //assumed / will realloc if needed
yading@11: 
yading@11:     memset(c, 0, sizeof(PPContext));
yading@11:     c->av_class = &av_codec_context_class;
yading@11:     if(cpuCaps&PP_FORMAT){
yading@11:         c->hChromaSubSample= cpuCaps&0x3;
yading@11:         c->vChromaSubSample= (cpuCaps>>4)&0x3;
yading@11:     }else{
yading@11:         c->hChromaSubSample= 1;
yading@11:         c->vChromaSubSample= 1;
yading@11:     }
yading@11:     if (cpuCaps & PP_CPU_CAPS_AUTO) {
yading@11:         c->cpuCaps = av_get_cpu_flags();
yading@11:     } else {
yading@11:         c->cpuCaps = 0;
yading@11:         if (cpuCaps & PP_CPU_CAPS_MMX)      c->cpuCaps |= AV_CPU_FLAG_MMX;
yading@11:         if (cpuCaps & PP_CPU_CAPS_MMX2)     c->cpuCaps |= AV_CPU_FLAG_MMXEXT;
yading@11:         if (cpuCaps & PP_CPU_CAPS_3DNOW)    c->cpuCaps |= AV_CPU_FLAG_3DNOW;
yading@11:         if (cpuCaps & PP_CPU_CAPS_ALTIVEC)  c->cpuCaps |= AV_CPU_FLAG_ALTIVEC;
yading@11:     }
yading@11: 
yading@11:     reallocBuffers(c, width, height, stride, qpStride);
yading@11: 
yading@11:     c->frameNum=-1;
yading@11: 
yading@11:     return c;
yading@11: }
yading@11: 
yading@11: void pp_free_context(void *vc){
yading@11:     PPContext *c = (PPContext*)vc;
yading@11:     int i;
yading@11: 
yading@11:     for(i=0; i<3; i++) av_free(c->tempBlurred[i]);
yading@11:     for(i=0; i<3; i++) av_free(c->tempBlurredPast[i]);
yading@11: 
yading@11:     av_free(c->tempBlocks);
yading@11:     av_free(c->yHistogram);
yading@11:     av_free(c->tempDst);
yading@11:     av_free(c->tempSrc);
yading@11:     av_free(c->deintTemp);
yading@11:     av_free(c->stdQPTable);
yading@11:     av_free(c->nonBQPTable);
yading@11:     av_free(c->forcedQPTable);
yading@11: 
yading@11:     memset(c, 0, sizeof(PPContext));
yading@11: 
yading@11:     av_free(c);
yading@11: }
yading@11: 
yading@11: void  pp_postprocess(const uint8_t * src[3], const int srcStride[3],
yading@11:                      uint8_t * dst[3], const int dstStride[3],
yading@11:                      int width, int height,
yading@11:                      const QP_STORE_T *QP_store,  int QPStride,
yading@11:                      pp_mode *vm,  void *vc, int pict_type)
yading@11: {
yading@11:     int mbWidth = (width+15)>>4;
yading@11:     int mbHeight= (height+15)>>4;
yading@11:     PPMode *mode = (PPMode*)vm;
yading@11:     PPContext *c = (PPContext*)vc;
yading@11:     int minStride= FFMAX(FFABS(srcStride[0]), FFABS(dstStride[0]));
yading@11:     int absQPStride = FFABS(QPStride);
yading@11: 
yading@11:     // c->stride and c->QPStride are always positive
yading@11:     if(c->stride < minStride || c->qpStride < absQPStride)
yading@11:         reallocBuffers(c, width, height,
yading@11:                        FFMAX(minStride, c->stride),
yading@11:                        FFMAX(c->qpStride, absQPStride));
yading@11: 
yading@11:     if(QP_store==NULL || (mode->lumMode & FORCE_QUANT)){
yading@11:         int i;
yading@11:         QP_store= c->forcedQPTable;
yading@11:         absQPStride = QPStride = 0;
yading@11:         if(mode->lumMode & FORCE_QUANT)
yading@11:             for(i=0; i<mbWidth; i++) c->forcedQPTable[i]= mode->forcedQuant;
yading@11:         else
yading@11:             for(i=0; i<mbWidth; i++) c->forcedQPTable[i]= 1;
yading@11:     }
yading@11: 
yading@11:     if(pict_type & PP_PICT_TYPE_QP2){
yading@11:         int i;
yading@11:         const int count= mbHeight * absQPStride;
yading@11:         for(i=0; i<(count>>2); i++){
yading@11:             ((uint32_t*)c->stdQPTable)[i] = (((const uint32_t*)QP_store)[i]>>1) & 0x7F7F7F7F;
yading@11:         }
yading@11:         for(i<<=2; i<count; i++){
yading@11:             c->stdQPTable[i] = QP_store[i]>>1;
yading@11:         }
yading@11:         QP_store= c->stdQPTable;
yading@11:         QPStride= absQPStride;
yading@11:     }
yading@11: 
yading@11:     if(0){
yading@11:         int x,y;
yading@11:         for(y=0; y<mbHeight; y++){
yading@11:             for(x=0; x<mbWidth; x++){
yading@11:                 av_log(c, AV_LOG_INFO, "%2d ", QP_store[x + y*QPStride]);
yading@11:             }
yading@11:             av_log(c, AV_LOG_INFO, "\n");
yading@11:         }
yading@11:         av_log(c, AV_LOG_INFO, "\n");
yading@11:     }
yading@11: 
yading@11:     if((pict_type&7)!=3){
yading@11:         if (QPStride >= 0){
yading@11:             int i;
yading@11:             const int count= mbHeight * QPStride;
yading@11:             for(i=0; i<(count>>2); i++){
yading@11:                 ((uint32_t*)c->nonBQPTable)[i] = ((const uint32_t*)QP_store)[i] & 0x3F3F3F3F;
yading@11:             }
yading@11:             for(i<<=2; i<count; i++){
yading@11:                 c->nonBQPTable[i] = QP_store[i] & 0x3F;
yading@11:             }
yading@11:         } else {
yading@11:             int i,j;
yading@11:             for(i=0; i<mbHeight; i++) {
yading@11:                 for(j=0; j<absQPStride; j++) {
yading@11:                     c->nonBQPTable[i*absQPStride+j] = QP_store[i*QPStride+j] & 0x3F;
yading@11:                 }
yading@11:             }
yading@11:         }
yading@11:     }
yading@11: 
yading@11:     av_log(c, AV_LOG_DEBUG, "using npp filters 0x%X/0x%X\n",
yading@11:            mode->lumMode, mode->chromMode);
yading@11: 
yading@11:     postProcess(src[0], srcStride[0], dst[0], dstStride[0],
yading@11:                 width, height, QP_store, QPStride, 0, mode, c);
yading@11: 
yading@11:     width  = (width )>>c->hChromaSubSample;
yading@11:     height = (height)>>c->vChromaSubSample;
yading@11: 
yading@11:     if(mode->chromMode){
yading@11:         postProcess(src[1], srcStride[1], dst[1], dstStride[1],
yading@11:                     width, height, QP_store, QPStride, 1, mode, c);
yading@11:         postProcess(src[2], srcStride[2], dst[2], dstStride[2],
yading@11:                     width, height, QP_store, QPStride, 2, mode, c);
yading@11:     }
yading@11:     else if(srcStride[1] == dstStride[1] && srcStride[2] == dstStride[2]){
yading@11:         linecpy(dst[1], src[1], height, srcStride[1]);
yading@11:         linecpy(dst[2], src[2], height, srcStride[2]);
yading@11:     }else{
yading@11:         int y;
yading@11:         for(y=0; y<height; y++){
yading@11:             memcpy(&(dst[1][y*dstStride[1]]), &(src[1][y*srcStride[1]]), width);
yading@11:             memcpy(&(dst[2][y*dstStride[2]]), &(src[2][y*srcStride[2]]), width);
yading@11:         }
yading@11:     }
yading@11: }