LCOV - code coverage report
Current view: top level - libpostproc - postprocess.c (source / functions) Hit Total Coverage
Test: coverage.info Lines: 391 467 83.7 %
Date: 2017-10-22 09:09:27 Functions: 20 22 90.9 %

          Line data    Source code
       1             : /*
       2             :  * Copyright (C) 2001-2003 Michael Niedermayer (michaelni@gmx.at)
       3             :  *
       4             :  * AltiVec optimizations (C) 2004 Romain Dolbeau <romain@dolbeau.org>
       5             :  *
       6             :  * This file is part of FFmpeg.
       7             :  *
       8             :  * FFmpeg is free software; you can redistribute it and/or modify
       9             :  * it under the terms of the GNU General Public License as published by
      10             :  * the Free Software Foundation; either version 2 of the License, or
      11             :  * (at your option) any later version.
      12             :  *
      13             :  * FFmpeg is distributed in the hope that it will be useful,
      14             :  * but WITHOUT ANY WARRANTY; without even the implied warranty of
      15             :  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
      16             :  * GNU General Public License for more details.
      17             :  *
      18             :  * You should have received a copy of the GNU General Public License
      19             :  * along with FFmpeg; if not, write to the Free Software
      20             :  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
      21             :  */
      22             : 
      23             : /**
      24             :  * @file
      25             :  * postprocessing.
      26             :  */
      27             : 
      28             : /*
      29             :                         C       MMX     MMX2    3DNow   AltiVec
      30             : isVertDC                Ec      Ec                      Ec
      31             : isVertMinMaxOk          Ec      Ec                      Ec
      32             : doVertLowPass           E               e       e       Ec
      33             : doVertDefFilter         Ec      Ec      e       e       Ec
      34             : isHorizDC               Ec      Ec                      Ec
      35             : isHorizMinMaxOk         a       E                       Ec
      36             : doHorizLowPass          E               e       e       Ec
      37             : doHorizDefFilter        Ec      Ec      e       e       Ec
      38             : do_a_deblock            Ec      E       Ec      E
      39             : deRing                  E               e       e*      Ecp
      40             : Vertical RKAlgo1        E               a       a
      41             : Horizontal RKAlgo1                      a       a
      42             : Vertical X1#            a               E       E
      43             : Horizontal X1#          a               E       E
      44             : LinIpolDeinterlace      e               E       E*
      45             : CubicIpolDeinterlace    a               e       e*
      46             : LinBlendDeinterlace     e               E       E*
      47             : MedianDeinterlace#      E       Ec      Ec
      48             : TempDeNoiser#           E               e       e       Ec
      49             : 
      50             : * I do not have a 3DNow! CPU -> it is untested, but no one said it does not work so it seems to work
      51             : # more or less selfinvented filters so the exactness is not too meaningful
      52             : E = Exact implementation
      53             : e = almost exact implementation (slightly different rounding,...)
      54             : a = alternative / approximate impl
      55             : c = checked against the other implementations (-vo md5)
      56             : p = partially optimized, still some work to do
      57             : */
      58             : 
      59             : /*
      60             : TODO:
      61             : reduce the time wasted on the mem transfer
      62             : unroll stuff if instructions depend too much on the prior one
      63             : move YScale thing to the end instead of fixing QP
      64             : write a faster and higher quality deblocking filter :)
      65             : make the mainloop more flexible (variable number of blocks at once
      66             :         (the if/else stuff per block is slowing things down)
      67             : compare the quality & speed of all filters
      68             : split this huge file
      69             : optimize c versions
      70             : try to unroll inner for(x=0 ... loop to avoid these damn if(x ... checks
      71             : ...
      72             : */
      73             : 
      74             : //Changelog: use git log
      75             : 
      76             : #include "config.h"
      77             : #include "libavutil/avutil.h"
      78             : #include "libavutil/avassert.h"
      79             : #include "libavutil/intreadwrite.h"
      80             : #include <inttypes.h>
      81             : #include <stdio.h>
      82             : #include <stdlib.h>
      83             : #include <string.h>
      84             : //#undef HAVE_MMXEXT_INLINE
      85             : //#define HAVE_AMD3DNOW_INLINE
      86             : //#undef HAVE_MMX_INLINE
      87             : //#undef ARCH_X86
      88             : //#define DEBUG_BRIGHTNESS
      89             : #include "postprocess.h"
      90             : #include "postprocess_internal.h"
      91             : #include "libavutil/avstring.h"
      92             : #include "libavutil/ppc/util_altivec.h"
      93             : 
      94             : #include "libavutil/ffversion.h"
      95             : const char postproc_ffversion[] = "FFmpeg version " FFMPEG_VERSION;
      96             : 
      97        5431 : unsigned postproc_version(void)
      98             : {
      99             :     av_assert0(LIBPOSTPROC_VERSION_MICRO >= 100);
     100        5431 :     return LIBPOSTPROC_VERSION_INT;
     101             : }
     102             : 
     103        5431 : const char *postproc_configuration(void)
     104             : {
     105        5431 :     return FFMPEG_CONFIGURATION;
     106             : }
     107             : 
     108           0 : const char *postproc_license(void)
     109             : {
     110             : #define LICENSE_PREFIX "libpostproc license: "
     111           0 :     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;
     112             : }
     113             : 
     114             : #define GET_MODE_BUFFER_SIZE 500
     115             : #define OPTIONS_ARRAY_SIZE 10
     116             : #define BLOCK_SIZE 8
     117             : #define TEMP_STRIDE 8
     118             : //#define NUM_BLOCKS_AT_ONCE 16 //not used yet
     119             : 
     120             : #if ARCH_X86 && HAVE_INLINE_ASM
     121             : DECLARE_ASM_CONST(8, uint64_t, w05)= 0x0005000500050005LL;
     122             : DECLARE_ASM_CONST(8, uint64_t, w04)= 0x0004000400040004LL;
     123             : DECLARE_ASM_CONST(8, uint64_t, w20)= 0x0020002000200020LL;
     124             : DECLARE_ASM_CONST(8, uint64_t, b00)= 0x0000000000000000LL;
     125             : DECLARE_ASM_CONST(8, uint64_t, b01)= 0x0101010101010101LL;
     126             : DECLARE_ASM_CONST(8, uint64_t, b02)= 0x0202020202020202LL;
     127             : DECLARE_ASM_CONST(8, uint64_t, b08)= 0x0808080808080808LL;
     128             : DECLARE_ASM_CONST(8, uint64_t, b80)= 0x8080808080808080LL;
     129             : #endif
     130             : 
     131             : DECLARE_ASM_CONST(8, int, deringThreshold)= 20;
     132             : 
     133             : 
     134             : static const struct PPFilter filters[]=
     135             : {
     136             :     {"hb", "hdeblock",              1, 1, 3, H_DEBLOCK},
     137             :     {"vb", "vdeblock",              1, 2, 4, V_DEBLOCK},
     138             : /*  {"hr", "rkhdeblock",            1, 1, 3, H_RK1_FILTER},
     139             :     {"vr", "rkvdeblock",            1, 2, 4, V_RK1_FILTER},*/
     140             :     {"h1", "x1hdeblock",            1, 1, 3, H_X1_FILTER},
     141             :     {"v1", "x1vdeblock",            1, 2, 4, V_X1_FILTER},
     142             :     {"ha", "ahdeblock",             1, 1, 3, H_A_DEBLOCK},
     143             :     {"va", "avdeblock",             1, 2, 4, V_A_DEBLOCK},
     144             :     {"dr", "dering",                1, 5, 6, DERING},
     145             :     {"al", "autolevels",            0, 1, 2, LEVEL_FIX},
     146             :     {"lb", "linblenddeint",         1, 1, 4, LINEAR_BLEND_DEINT_FILTER},
     147             :     {"li", "linipoldeint",          1, 1, 4, LINEAR_IPOL_DEINT_FILTER},
     148             :     {"ci", "cubicipoldeint",        1, 1, 4, CUBIC_IPOL_DEINT_FILTER},
     149             :     {"md", "mediandeint",           1, 1, 4, MEDIAN_DEINT_FILTER},
     150             :     {"fd", "ffmpegdeint",           1, 1, 4, FFMPEG_DEINT_FILTER},
     151             :     {"l5", "lowpass5",              1, 1, 4, LOWPASS5_DEINT_FILTER},
     152             :     {"tn", "tmpnoise",              1, 7, 8, TEMP_NOISE_FILTER},
     153             :     {"fq", "forcequant",            1, 0, 0, FORCE_QUANT},
     154             :     {"be", "bitexact",              1, 0, 0, BITEXACT},
     155             :     {"vi", "visualize",             1, 0, 0, VISUALIZE},
     156             :     {NULL, NULL,0,0,0,0} //End Marker
     157             : };
     158             : 
     159             : static const char * const replaceTable[]=
     160             : {
     161             :     "default",      "hb:a,vb:a,dr:a",
     162             :     "de",           "hb:a,vb:a,dr:a",
     163             :     "fast",         "h1:a,v1:a,dr:a",
     164             :     "fa",           "h1:a,v1:a,dr:a",
     165             :     "ac",           "ha:a:128:7,va:a,dr:a",
     166             :     NULL //End Marker
     167             : };
     168             : 
     169             : /* The horizontal functions exist only in C because the MMX
     170             :  * code is faster with vertical filters and transposing. */
     171             : 
     172             : /**
     173             :  * Check if the given 8x8 Block is mostly "flat"
     174             :  */
     175       36864 : static inline int isHorizDC_C(const uint8_t src[], int stride, const PPContext *c)
     176             : {
     177       36864 :     int numEq= 0;
     178             :     int y;
     179       36864 :     const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
     180       36864 :     const int dcThreshold= dcOffset*2 + 1;
     181             : 
     182      331776 :     for(y=0; y<BLOCK_SIZE; y++){
     183      294912 :         numEq += ((unsigned)(src[0] - src[1] + dcOffset)) < dcThreshold;
     184      294912 :         numEq += ((unsigned)(src[1] - src[2] + dcOffset)) < dcThreshold;
     185      294912 :         numEq += ((unsigned)(src[2] - src[3] + dcOffset)) < dcThreshold;
     186      294912 :         numEq += ((unsigned)(src[3] - src[4] + dcOffset)) < dcThreshold;
     187      294912 :         numEq += ((unsigned)(src[4] - src[5] + dcOffset)) < dcThreshold;
     188      294912 :         numEq += ((unsigned)(src[5] - src[6] + dcOffset)) < dcThreshold;
     189      294912 :         numEq += ((unsigned)(src[6] - src[7] + dcOffset)) < dcThreshold;
     190      294912 :         src+= stride;
     191             :     }
     192       36864 :     return numEq > c->ppMode.flatnessThreshold;
     193             : }
     194             : 
     195             : /**
     196             :  * Check if the middle 8x8 Block in the given 8x16 block is flat
     197             :  */
     198       36608 : static inline int isVertDC_C(const uint8_t src[], int stride, const PPContext *c)
     199             : {
     200       36608 :     int numEq= 0;
     201             :     int y;
     202       36608 :     const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
     203       36608 :     const int dcThreshold= dcOffset*2 + 1;
     204             : 
     205       36608 :     src+= stride*4; // src points to begin of the 8x8 Block
     206      292864 :     for(y=0; y<BLOCK_SIZE-1; y++){
     207      256256 :         numEq += ((unsigned)(src[0] - src[0+stride] + dcOffset)) < dcThreshold;
     208      256256 :         numEq += ((unsigned)(src[1] - src[1+stride] + dcOffset)) < dcThreshold;
     209      256256 :         numEq += ((unsigned)(src[2] - src[2+stride] + dcOffset)) < dcThreshold;
     210      256256 :         numEq += ((unsigned)(src[3] - src[3+stride] + dcOffset)) < dcThreshold;
     211      256256 :         numEq += ((unsigned)(src[4] - src[4+stride] + dcOffset)) < dcThreshold;
     212      256256 :         numEq += ((unsigned)(src[5] - src[5+stride] + dcOffset)) < dcThreshold;
     213      256256 :         numEq += ((unsigned)(src[6] - src[6+stride] + dcOffset)) < dcThreshold;
     214      256256 :         numEq += ((unsigned)(src[7] - src[7+stride] + dcOffset)) < dcThreshold;
     215      256256 :         src+= stride;
     216             :     }
     217       36608 :     return numEq > c->ppMode.flatnessThreshold;
     218             : }
     219             : 
     220         701 : static inline int isHorizMinMaxOk_C(const uint8_t src[], int stride, int QP)
     221             : {
     222             :     int i;
     223        1149 :     for(i=0; i<2; i++){
     224        1020 :         if((unsigned)(src[0] - src[5] + 2*QP) > 4*QP) return 0;
     225         652 :         src += stride;
     226         652 :         if((unsigned)(src[2] - src[7] + 2*QP) > 4*QP) return 0;
     227         510 :         src += stride;
     228         510 :         if((unsigned)(src[4] - src[1] + 2*QP) > 4*QP) return 0;
     229         508 :         src += stride;
     230         508 :         if((unsigned)(src[6] - src[3] + 2*QP) > 4*QP) return 0;
     231         448 :         src += stride;
     232             :     }
     233         129 :     return 1;
     234             : }
     235             : 
     236         352 : static inline int isVertMinMaxOk_C(const uint8_t src[], int stride, int QP)
     237             : {
     238             :     int x;
     239         352 :     src+= stride*4;
     240         611 :     for(x=0; x<BLOCK_SIZE; x+=4){
     241         526 :         if((unsigned)(src[  x + 0*stride] - src[  x + 5*stride] + 2*QP) > 4*QP) return 0;
     242         349 :         if((unsigned)(src[1+x + 2*stride] - src[1+x + 7*stride] + 2*QP) > 4*QP) return 0;
     243         291 :         if((unsigned)(src[2+x + 4*stride] - src[2+x + 1*stride] + 2*QP) > 4*QP) return 0;
     244         291 :         if((unsigned)(src[3+x + 6*stride] - src[3+x + 3*stride] + 2*QP) > 4*QP) return 0;
     245             :     }
     246          85 :     return 1;
     247             : }
     248             : 
     249       36864 : static inline int horizClassify_C(const uint8_t src[], int stride, const PPContext *c)
     250             : {
     251       36864 :     if( isHorizDC_C(src, stride, c) ){
     252         701 :         return isHorizMinMaxOk_C(src, stride, c->QP);
     253             :     }else{
     254       36163 :         return 2;
     255             :     }
     256             : }
     257             : 
     258       36608 : static inline int vertClassify_C(const uint8_t src[], int stride, const PPContext *c)
     259             : {
     260       36608 :     if( isVertDC_C(src, stride, c) ){
     261         352 :         return isVertMinMaxOk_C(src, stride, c->QP);
     262             :     }else{
     263       36256 :         return 2;
     264             :     }
     265             : }
     266             : 
     267       36163 : static inline void doHorizDefFilter_C(uint8_t dst[], int stride, const PPContext *c)
     268             : {
     269             :     int y;
     270      325467 :     for(y=0; y<BLOCK_SIZE; y++){
     271      289304 :         const int middleEnergy= 5*(dst[4] - dst[3]) + 2*(dst[2] - dst[5]);
     272             : 
     273      289304 :         if(FFABS(middleEnergy) < 8*c->QP){
     274      188700 :             const int q=(dst[3] - dst[4])/2;
     275      188700 :             const int leftEnergy=  5*(dst[2] - dst[1]) + 2*(dst[0] - dst[3]);
     276      188700 :             const int rightEnergy= 5*(dst[6] - dst[5]) + 2*(dst[4] - dst[7]);
     277             : 
     278      188700 :             int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
     279      188700 :             d= FFMAX(d, 0);
     280             : 
     281      188700 :             d= (5*d + 32) >> 6;
     282      188700 :             d*= FFSIGN(-middleEnergy);
     283             : 
     284      188700 :             if(q>0)
     285             :             {
     286       43347 :                 d = FFMAX(d, 0);
     287       43347 :                 d = FFMIN(d, q);
     288             :             }
     289             :             else
     290             :             {
     291      145353 :                 d = FFMIN(d, 0);
     292      145353 :                 d = FFMAX(d, q);
     293             :             }
     294             : 
     295      188700 :             dst[3]-= d;
     296      188700 :             dst[4]+= d;
     297             :         }
     298      289304 :         dst+= stride;
     299             :     }
     300       36163 : }
     301             : 
     302             : /**
     303             :  * Do a horizontal low pass filter on the 10x8 block (dst points to middle 8x8 Block)
     304             :  * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16 (C version)
     305             :  */
     306         129 : static inline void doHorizLowPass_C(uint8_t dst[], int stride, const PPContext *c)
     307             : {
     308             :     int y;
     309        1161 :     for(y=0; y<BLOCK_SIZE; y++){
     310        1032 :         const int first= FFABS(dst[-1] - dst[0]) < c->QP ? dst[-1] : dst[0];
     311        1032 :         const int last= FFABS(dst[8] - dst[7]) < c->QP ? dst[8] : dst[7];
     312             : 
     313             :         int sums[10];
     314        1032 :         sums[0] = 4*first + dst[0] + dst[1] + dst[2] + 4;
     315        1032 :         sums[1] = sums[0] - first  + dst[3];
     316        1032 :         sums[2] = sums[1] - first  + dst[4];
     317        1032 :         sums[3] = sums[2] - first  + dst[5];
     318        1032 :         sums[4] = sums[3] - first  + dst[6];
     319        1032 :         sums[5] = sums[4] - dst[0] + dst[7];
     320        1032 :         sums[6] = sums[5] - dst[1] + last;
     321        1032 :         sums[7] = sums[6] - dst[2] + last;
     322        1032 :         sums[8] = sums[7] - dst[3] + last;
     323        1032 :         sums[9] = sums[8] - dst[4] + last;
     324             : 
     325        1032 :         dst[0]= (sums[0] + sums[2] + 2*dst[0])>>4;
     326        1032 :         dst[1]= (sums[1] + sums[3] + 2*dst[1])>>4;
     327        1032 :         dst[2]= (sums[2] + sums[4] + 2*dst[2])>>4;
     328        1032 :         dst[3]= (sums[3] + sums[5] + 2*dst[3])>>4;
     329        1032 :         dst[4]= (sums[4] + sums[6] + 2*dst[4])>>4;
     330        1032 :         dst[5]= (sums[5] + sums[7] + 2*dst[5])>>4;
     331        1032 :         dst[6]= (sums[6] + sums[8] + 2*dst[6])>>4;
     332        1032 :         dst[7]= (sums[7] + sums[9] + 2*dst[7])>>4;
     333             : 
     334        1032 :         dst+= stride;
     335             :     }
     336         129 : }
     337             : 
     338             : /**
     339             :  * Experimental Filter 1 (Horizontal)
     340             :  * will not damage linear gradients
     341             :  * Flat blocks should look like they were passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter
     342             :  * can only smooth blocks at the expected locations (it cannot smooth them if they did move)
     343             :  * MMX2 version does correct clipping C version does not
     344             :  * not identical with the vertical one
     345             :  */
     346       11520 : static inline void horizX1Filter(uint8_t *src, int stride, int QP)
     347             : {
     348             :     int y;
     349             :     static uint64_t lut[256];
     350       11520 :     if(!lut[255])
     351             :     {
     352             :         int i;
     353     2960640 :         for(i=0; i<256; i++)
     354             :         {
     355     2949120 :             int v= i < 128 ? 2*i : 2*(i-256);
     356             : /*
     357             : //Simulate 112242211 9-Tap filter
     358             :             uint64_t a= (v/16)  & 0xFF;
     359             :             uint64_t b= (v/8)   & 0xFF;
     360             :             uint64_t c= (v/4)   & 0xFF;
     361             :             uint64_t d= (3*v/8) & 0xFF;
     362             : */
     363             : //Simulate piecewise linear interpolation
     364     2949120 :             uint64_t a= (v/16)   & 0xFF;
     365     2949120 :             uint64_t b= (v*3/16) & 0xFF;
     366     2949120 :             uint64_t c= (v*5/16) & 0xFF;
     367     2949120 :             uint64_t d= (7*v/16) & 0xFF;
     368     2949120 :             uint64_t A= (0x100 - a)&0xFF;
     369     2949120 :             uint64_t B= (0x100 - b)&0xFF;
     370     2949120 :             uint64_t C= (0x100 - c)&0xFF;
     371     2949120 :             uint64_t D= (0x100 - c)&0xFF;
     372             : 
     373     8847360 :             lut[i]   = (a<<56) | (b<<48) | (c<<40) | (d<<32) |
     374     5898240 :                        (D<<24) | (C<<16) | (B<<8)  | (A);
     375             :             //lut[i] = (v<<32) | (v<<24);
     376             :         }
     377             :     }
     378             : 
     379      103680 :     for(y=0; y<BLOCK_SIZE; y++){
     380       92160 :         int a= src[1] - src[2];
     381       92160 :         int b= src[3] - src[4];
     382       92160 :         int c= src[5] - src[6];
     383             : 
     384       92160 :         int d= FFMAX(FFABS(b) - (FFABS(a) + FFABS(c))/2, 0);
     385             : 
     386       92160 :         if(d < QP){
     387       84620 :             int v = d * FFSIGN(-b);
     388             : 
     389       84620 :             src[1] +=v/8;
     390       84620 :             src[2] +=v/4;
     391       84620 :             src[3] +=3*v/8;
     392       84620 :             src[4] -=3*v/8;
     393       84620 :             src[5] -=v/4;
     394       84620 :             src[6] -=v/8;
     395             :         }
     396       92160 :         src+=stride;
     397             :     }
     398       11520 : }
     399             : 
     400             : /**
     401             :  * accurate deblock filter
     402             :  */
     403       22960 : static av_always_inline void do_a_deblock_C(uint8_t *src, int step,
     404             :                                             int stride, const PPContext *c, int mode)
     405             : {
     406             :     int y;
     407       22960 :     const int QP= c->QP;
     408       22960 :     const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;
     409       22960 :     const int dcThreshold= dcOffset*2 + 1;
     410             : //START_TIMER
     411       22960 :     src+= step*4; // src points to begin of the 8x8 Block
     412      206640 :     for(y=0; y<8; y++){
     413      183680 :         int numEq= 0;
     414             : 
     415      183680 :         numEq += ((unsigned)(src[-1*step] - src[0*step] + dcOffset)) < dcThreshold;
     416      183680 :         numEq += ((unsigned)(src[ 0*step] - src[1*step] + dcOffset)) < dcThreshold;
     417      183680 :         numEq += ((unsigned)(src[ 1*step] - src[2*step] + dcOffset)) < dcThreshold;
     418      183680 :         numEq += ((unsigned)(src[ 2*step] - src[3*step] + dcOffset)) < dcThreshold;
     419      183680 :         numEq += ((unsigned)(src[ 3*step] - src[4*step] + dcOffset)) < dcThreshold;
     420      183680 :         numEq += ((unsigned)(src[ 4*step] - src[5*step] + dcOffset)) < dcThreshold;
     421      183680 :         numEq += ((unsigned)(src[ 5*step] - src[6*step] + dcOffset)) < dcThreshold;
     422      183680 :         numEq += ((unsigned)(src[ 6*step] - src[7*step] + dcOffset)) < dcThreshold;
     423      183680 :         numEq += ((unsigned)(src[ 7*step] - src[8*step] + dcOffset)) < dcThreshold;
     424      183680 :         if(numEq > c->ppMode.flatnessThreshold){
     425             :             int min, max, x;
     426             : 
     427       82036 :             if(src[0] > src[step]){
     428       17783 :                 max= src[0];
     429       17783 :                 min= src[step];
     430             :             }else{
     431       64253 :                 max= src[step];
     432       64253 :                 min= src[0];
     433             :             }
     434      328144 :             for(x=2; x<8; x+=2){
     435      246108 :                 if(src[x*step] > src[(x+1)*step]){
     436       53874 :                         if(src[x    *step] > max) max= src[ x   *step];
     437       53874 :                         if(src[(x+1)*step] < min) min= src[(x+1)*step];
     438             :                 }else{
     439      192234 :                         if(src[(x+1)*step] > max) max= src[(x+1)*step];
     440      192234 :                         if(src[ x   *step] < min) min= src[ x   *step];
     441             :                 }
     442             :             }
     443       82036 :             if(max-min < 2*QP){
     444       42585 :                 const int first= FFABS(src[-1*step] - src[0]) < QP ? src[-1*step] : src[0];
     445       42585 :                 const int last= FFABS(src[8*step] - src[7*step]) < QP ? src[8*step] : src[7*step];
     446             : 
     447             :                 int sums[10];
     448       42585 :                 sums[0] = 4*first + src[0*step] + src[1*step] + src[2*step] + 4;
     449       42585 :                 sums[1] = sums[0] - first       + src[3*step];
     450       42585 :                 sums[2] = sums[1] - first       + src[4*step];
     451       42585 :                 sums[3] = sums[2] - first       + src[5*step];
     452       42585 :                 sums[4] = sums[3] - first       + src[6*step];
     453       42585 :                 sums[5] = sums[4] - src[0*step] + src[7*step];
     454       42585 :                 sums[6] = sums[5] - src[1*step] + last;
     455       42585 :                 sums[7] = sums[6] - src[2*step] + last;
     456       42585 :                 sums[8] = sums[7] - src[3*step] + last;
     457       42585 :                 sums[9] = sums[8] - src[4*step] + last;
     458             : 
     459       42585 :                 if (mode & VISUALIZE) {
     460           0 :                     src[0*step] =
     461           0 :                     src[1*step] =
     462           0 :                     src[2*step] =
     463           0 :                     src[3*step] =
     464           0 :                     src[4*step] =
     465           0 :                     src[5*step] =
     466           0 :                     src[6*step] =
     467           0 :                     src[7*step] = 128;
     468             :                 }
     469       42585 :                 src[0*step]= (sums[0] + sums[2] + 2*src[0*step])>>4;
     470       42585 :                 src[1*step]= (sums[1] + sums[3] + 2*src[1*step])>>4;
     471       42585 :                 src[2*step]= (sums[2] + sums[4] + 2*src[2*step])>>4;
     472       42585 :                 src[3*step]= (sums[3] + sums[5] + 2*src[3*step])>>4;
     473       42585 :                 src[4*step]= (sums[4] + sums[6] + 2*src[4*step])>>4;
     474       42585 :                 src[5*step]= (sums[5] + sums[7] + 2*src[5*step])>>4;
     475       42585 :                 src[6*step]= (sums[6] + sums[8] + 2*src[6*step])>>4;
     476       42585 :                 src[7*step]= (sums[7] + sums[9] + 2*src[7*step])>>4;
     477             :             }
     478             :         }else{
     479      101644 :             const int middleEnergy= 5*(src[4*step] - src[3*step]) + 2*(src[2*step] - src[5*step]);
     480             : 
     481      101644 :             if(FFABS(middleEnergy) < 8*QP){
     482       86530 :                 const int q=(src[3*step] - src[4*step])/2;
     483       86530 :                 const int leftEnergy=  5*(src[2*step] - src[1*step]) + 2*(src[0*step] - src[3*step]);
     484       86530 :                 const int rightEnergy= 5*(src[6*step] - src[5*step]) + 2*(src[4*step] - src[7*step]);
     485             : 
     486       86530 :                 int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );
     487       86530 :                 d= FFMAX(d, 0);
     488             : 
     489       86530 :                 d= (5*d + 32) >> 6;
     490       86530 :                 d*= FFSIGN(-middleEnergy);
     491             : 
     492       86530 :                 if(q>0){
     493       27119 :                     d = FFMAX(d, 0);
     494       27119 :                     d = FFMIN(d, q);
     495             :                 }else{
     496       59411 :                     d = FFMIN(d, 0);
     497       59411 :                     d = FFMAX(d, q);
     498             :                 }
     499             : 
     500       86530 :                 if ((mode & VISUALIZE) && d) {
     501           0 :                     d= (d < 0) ? 32 : -32;
     502           0 :                     src[3*step]= av_clip_uint8(src[3*step] - d);
     503           0 :                     src[4*step]= av_clip_uint8(src[4*step] + d);
     504           0 :                     d = 0;
     505             :                 }
     506             : 
     507       86530 :                 src[3*step]-= d;
     508       86530 :                 src[4*step]+= d;
     509             :             }
     510             :         }
     511             : 
     512      183680 :         src += stride;
     513             :     }
     514             : /*if(step==16){
     515             :     STOP_TIMER("step16")
     516             : }else{
     517             :     STOP_TIMER("stepX")
     518             : }*/
     519       22960 : }
     520             : 
     521             : //Note: we have C, MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one
     522             : //Plain C versions
     523             : //we always compile C for testing which needs bitexactness
     524             : #define TEMPLATE_PP_C 1
     525             : #include "postprocess_template.c"
     526             : 
     527             : #if HAVE_ALTIVEC
     528             : #   define TEMPLATE_PP_ALTIVEC 1
     529             : #   include "postprocess_altivec_template.c"
     530             : #   include "postprocess_template.c"
     531             : #endif
     532             : 
     533             : #if ARCH_X86 && HAVE_INLINE_ASM
     534             : #    if CONFIG_RUNTIME_CPUDETECT
     535             : #        define TEMPLATE_PP_MMX 1
     536             : #        include "postprocess_template.c"
     537             : #        define TEMPLATE_PP_MMXEXT 1
     538             : #        include "postprocess_template.c"
     539             : #        define TEMPLATE_PP_3DNOW 1
     540             : #        include "postprocess_template.c"
     541             : #        define TEMPLATE_PP_SSE2 1
     542             : #        include "postprocess_template.c"
     543             : #    else
     544             : #        if HAVE_SSE2_INLINE
     545             : #            define TEMPLATE_PP_SSE2 1
     546             : #            include "postprocess_template.c"
     547             : #        elif HAVE_MMXEXT_INLINE
     548             : #            define TEMPLATE_PP_MMXEXT 1
     549             : #            include "postprocess_template.c"
     550             : #        elif HAVE_AMD3DNOW_INLINE
     551             : #            define TEMPLATE_PP_3DNOW 1
     552             : #            include "postprocess_template.c"
     553             : #        elif HAVE_MMX_INLINE
     554             : #            define TEMPLATE_PP_MMX 1
     555             : #            include "postprocess_template.c"
     556             : #        endif
     557             : #    endif
     558             : #endif
     559             : 
     560             : typedef void (*pp_fn)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
     561             :                       const int8_t QPs[], int QPStride, int isColor, PPContext *c2);
     562             : 
     563         123 : static inline void postProcess(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
     564             :         const int8_t QPs[], int QPStride, int isColor, pp_mode *vm, pp_context *vc)
     565             : {
     566         123 :     pp_fn pp = postProcess_C;
     567         123 :     PPContext *c= (PPContext *)vc;
     568         123 :     PPMode *ppMode= (PPMode *)vm;
     569         123 :     c->ppMode= *ppMode; //FIXME
     570             : 
     571         123 :     if (!(ppMode->lumMode & BITEXACT)) {
     572             : #if CONFIG_RUNTIME_CPUDETECT
     573             : #if ARCH_X86 && HAVE_INLINE_ASM
     574             :         // ordered per speed fastest first
     575          15 :         if      (c->cpuCaps & AV_CPU_FLAG_SSE2)     pp = postProcess_SSE2;
     576          15 :         else if (c->cpuCaps & AV_CPU_FLAG_MMXEXT)   pp = postProcess_MMX2;
     577          15 :         else if (c->cpuCaps & AV_CPU_FLAG_3DNOW)    pp = postProcess_3DNow;
     578          15 :         else if (c->cpuCaps & AV_CPU_FLAG_MMX)      pp = postProcess_MMX;
     579             : #elif HAVE_ALTIVEC
     580             :         if      (c->cpuCaps & AV_CPU_FLAG_ALTIVEC)  pp = postProcess_altivec;
     581             : #endif
     582             : #else /* CONFIG_RUNTIME_CPUDETECT */
     583             : #if     HAVE_SSE2_INLINE
     584             :         pp = postProcess_SSE2;
     585             : #elif   HAVE_MMXEXT_INLINE
     586             :         pp = postProcess_MMX2;
     587             : #elif HAVE_AMD3DNOW_INLINE
     588             :         pp = postProcess_3DNow;
     589             : #elif HAVE_MMX_INLINE
     590             :         pp = postProcess_MMX;
     591             : #elif HAVE_ALTIVEC
     592             :         pp = postProcess_altivec;
     593             : #endif
     594             : #endif /* !CONFIG_RUNTIME_CPUDETECT */
     595             :     }
     596             : 
     597         123 :     pp(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
     598         123 : }
     599             : 
     600             : /* -pp Command line Help
     601             : */
     602             : const char pp_help[] =
     603             : "Available postprocessing filters:\n"
     604             : "Filters                        Options\n"
     605             : "short  long name       short   long option     Description\n"
     606             : "*      *               a       autoq           CPU power dependent enabler\n"
     607             : "                       c       chrom           chrominance filtering enabled\n"
     608             : "                       y       nochrom         chrominance filtering disabled\n"
     609             : "                       n       noluma          luma filtering disabled\n"
     610             : "hb     hdeblock        (2 threshold)           horizontal deblocking filter\n"
     611             : "       1. difference factor: default=32, higher -> more deblocking\n"
     612             : "       2. flatness threshold: default=39, lower -> more deblocking\n"
     613             : "                       the h & v deblocking filters share these\n"
     614             : "                       so you can't set different thresholds for h / v\n"
     615             : "vb     vdeblock        (2 threshold)           vertical deblocking filter\n"
     616             : "ha     hadeblock       (2 threshold)           horizontal deblocking filter\n"
     617             : "va     vadeblock       (2 threshold)           vertical deblocking filter\n"
     618             : "h1     x1hdeblock                              experimental h deblock filter 1\n"
     619             : "v1     x1vdeblock                              experimental v deblock filter 1\n"
     620             : "dr     dering                                  deringing filter\n"
     621             : "al     autolevels                              automatic brightness / contrast\n"
     622             : "                       f        fullyrange     stretch luminance to (0..255)\n"
     623             : "lb     linblenddeint                           linear blend deinterlacer\n"
     624             : "li     linipoldeint                            linear interpolating deinterlace\n"
     625             : "ci     cubicipoldeint                          cubic interpolating deinterlacer\n"
     626             : "md     mediandeint                             median deinterlacer\n"
     627             : "fd     ffmpegdeint                             ffmpeg deinterlacer\n"
     628             : "l5     lowpass5                                FIR lowpass deinterlacer\n"
     629             : "de     default                                 hb:a,vb:a,dr:a\n"
     630             : "fa     fast                                    h1:a,v1:a,dr:a\n"
     631             : "ac                                             ha:a:128:7,va:a,dr:a\n"
     632             : "tn     tmpnoise        (3 threshold)           temporal noise reducer\n"
     633             : "                     1. <= 2. <= 3.            larger -> stronger filtering\n"
     634             : "fq     forceQuant      <quantizer>             force quantizer\n"
     635             : "Usage:\n"
     636             : "<filterName>[:<option>[:<option>...]][[,|/][-]<filterName>[:<option>...]]...\n"
     637             : "long form example:\n"
     638             : "vdeblock:autoq/hdeblock:autoq/linblenddeint    default,-vdeblock\n"
     639             : "short form example:\n"
     640             : "vb:a/hb:a/lb                                   de,-vb\n"
     641             : "more examples:\n"
     642             : "tn:64:128:256\n"
     643             : "\n"
     644             : ;
     645             : 
     646          56 : pp_mode *pp_get_mode_by_name_and_quality(const char *name, int quality)
     647             : {
     648             :     char temp[GET_MODE_BUFFER_SIZE];
     649          56 :     char *p= temp;
     650             :     static const char filterDelimiters[] = ",/";
     651             :     static const char optionDelimiters[] = ":|";
     652             :     struct PPMode *ppMode;
     653             :     char *filterToken;
     654             : 
     655          56 :     if (!name)  {
     656           0 :         av_log(NULL, AV_LOG_ERROR, "pp: Missing argument\n");
     657           0 :         return NULL;
     658             :     }
     659             : 
     660          56 :     if (!strcmp(name, "help")) {
     661             :         const char *p;
     662           0 :         for (p = pp_help; strchr(p, '\n'); p = strchr(p, '\n') + 1) {
     663           0 :             av_strlcpy(temp, p, FFMIN(sizeof(temp), strchr(p, '\n') - p + 2));
     664           0 :             av_log(NULL, AV_LOG_INFO, "%s", temp);
     665             :         }
     666           0 :         return NULL;
     667             :     }
     668             : 
     669          56 :     ppMode= av_malloc(sizeof(PPMode));
     670          56 :     if (!ppMode)
     671           0 :         return NULL;
     672             : 
     673          56 :     ppMode->lumMode= 0;
     674          56 :     ppMode->chromMode= 0;
     675          56 :     ppMode->maxTmpNoise[0]= 700;
     676          56 :     ppMode->maxTmpNoise[1]= 1500;
     677          56 :     ppMode->maxTmpNoise[2]= 3000;
     678          56 :     ppMode->maxAllowedY= 234;
     679          56 :     ppMode->minAllowedY= 16;
     680          56 :     ppMode->baseDcDiff= 256/8;
     681          56 :     ppMode->flatnessThreshold= 56-16-1;
     682          56 :     ppMode->maxClippedThreshold= (AVRational){1,100};
     683          56 :     ppMode->error=0;
     684             : 
     685          56 :     memset(temp, 0, GET_MODE_BUFFER_SIZE);
     686          56 :     av_strlcpy(temp, name, GET_MODE_BUFFER_SIZE - 1);
     687             : 
     688          56 :     av_log(NULL, AV_LOG_DEBUG, "pp: %s\n", name);
     689             : 
     690         224 :     for(;;){
     691             :         const char *filterName;
     692         280 :         int q= 1000000; //PP_QUALITY_MAX;
     693         280 :         int chrom=-1;
     694         280 :         int luma=-1;
     695             :         const char *option;
     696             :         const char *options[OPTIONS_ARRAY_SIZE];
     697             :         int i;
     698         280 :         int filterNameOk=0;
     699         280 :         int numOfUnknownOptions=0;
     700         280 :         int enable=1; //does the user want us to enabled or disabled the filter
     701             :         char *tokstate;
     702             : 
     703         280 :         filterToken= av_strtok(p, filterDelimiters, &tokstate);
     704         336 :         if(!filterToken) break;
     705         224 :         p+= strlen(filterToken) + 1; // p points to next filterToken
     706         224 :         filterName= av_strtok(filterToken, optionDelimiters, &tokstate);
     707         224 :         if (!filterName) {
     708           0 :             ppMode->error++;
     709           0 :             break;
     710             :         }
     711         224 :         av_log(NULL, AV_LOG_DEBUG, "pp: %s::%s\n", filterToken, filterName);
     712             : 
     713         224 :         if(*filterName == '-'){
     714           0 :             enable=0;
     715           0 :             filterName++;
     716             :         }
     717             : 
     718             :         for(;;){ //for all options
     719         266 :             option= av_strtok(NULL, optionDelimiters, &tokstate);
     720         245 :             if(!option) break;
     721             : 
     722          21 :             av_log(NULL, AV_LOG_DEBUG, "pp: option: %s\n", option);
     723          21 :             if(!strcmp("autoq", option) || !strcmp("a", option)) q= quality;
     724          21 :             else if(!strcmp("nochrom", option) || !strcmp("y", option)) chrom=0;
     725          21 :             else if(!strcmp("chrom", option) || !strcmp("c", option)) chrom=1;
     726          21 :             else if(!strcmp("noluma", option) || !strcmp("n", option)) luma=0;
     727             :             else{
     728          21 :                 options[numOfUnknownOptions] = option;
     729          21 :                 numOfUnknownOptions++;
     730             :             }
     731          21 :             if(numOfUnknownOptions >= OPTIONS_ARRAY_SIZE-1) break;
     732             :         }
     733         224 :         options[numOfUnknownOptions] = NULL;
     734             : 
     735             :         /* replace stuff from the replace Table */
     736        1344 :         for(i=0; replaceTable[2*i]; i++){
     737        1120 :             if(!strcmp(replaceTable[2*i], filterName)){
     738           0 :                 size_t newlen = strlen(replaceTable[2*i + 1]);
     739             :                 int plen;
     740             :                 int spaceLeft;
     741             : 
     742           0 :                 p--, *p=',';
     743             : 
     744           0 :                 plen= strlen(p);
     745           0 :                 spaceLeft= p - temp + plen;
     746           0 :                 if(spaceLeft + newlen  >= GET_MODE_BUFFER_SIZE - 1){
     747           0 :                     ppMode->error++;
     748           0 :                     break;
     749             :                 }
     750           0 :                 memmove(p + newlen, p, plen+1);
     751           0 :                 memcpy(p, replaceTable[2*i + 1], newlen);
     752           0 :                 filterNameOk=1;
     753             :             }
     754             :         }
     755             : 
     756        4256 :         for(i=0; filters[i].shortName; i++){
     757        4032 :             if(   !strcmp(filters[i].longName, filterName)
     758        4032 :                || !strcmp(filters[i].shortName, filterName)){
     759         224 :                 ppMode->lumMode &= ~filters[i].mask;
     760         224 :                 ppMode->chromMode &= ~filters[i].mask;
     761             : 
     762         224 :                 filterNameOk=1;
     763         224 :                 if(!enable) break; // user wants to disable it
     764             : 
     765         224 :                 if(q >= filters[i].minLumQuality && luma)
     766         224 :                     ppMode->lumMode|= filters[i].mask;
     767         224 :                 if(chrom==1 || (chrom==-1 && filters[i].chromDefault))
     768         203 :                     if(q >= filters[i].minChromQuality)
     769         203 :                             ppMode->chromMode|= filters[i].mask;
     770             : 
     771         224 :                 if(filters[i].mask == LEVEL_FIX){
     772             :                     int o;
     773          21 :                     ppMode->minAllowedY= 16;
     774          21 :                     ppMode->maxAllowedY= 234;
     775          21 :                     for(o=0; options[o]; o++){
     776           0 :                         if(  !strcmp(options[o],"fullyrange")
     777           0 :                            ||!strcmp(options[o],"f")){
     778           0 :                             ppMode->minAllowedY= 0;
     779           0 :                             ppMode->maxAllowedY= 255;
     780           0 :                             numOfUnknownOptions--;
     781             :                         }
     782             :                     }
     783             :                 }
     784         203 :                 else if(filters[i].mask == TEMP_NOISE_FILTER)
     785             :                 {
     786             :                     int o;
     787          21 :                     int numOfNoises=0;
     788             : 
     789          21 :                     for(o=0; options[o]; o++){
     790             :                         char *tail;
     791           0 :                         ppMode->maxTmpNoise[numOfNoises]=
     792           0 :                             strtol(options[o], &tail, 0);
     793           0 :                         if(tail!=options[o]){
     794           0 :                             numOfNoises++;
     795           0 :                             numOfUnknownOptions--;
     796           0 :                             if(numOfNoises >= 3) break;
     797             :                         }
     798             :                     }
     799             :                 }
     800         182 :                 else if(filters[i].mask == V_DEBLOCK   || filters[i].mask == H_DEBLOCK
     801         196 :                      || filters[i].mask == V_A_DEBLOCK || filters[i].mask == H_A_DEBLOCK){
     802             :                     int o;
     803             : 
     804          70 :                     for(o=0; options[o] && o<2; o++){
     805             :                         char *tail;
     806          14 :                         int val= strtol(options[o], &tail, 0);
     807          14 :                         if(tail==options[o]) break;
     808             : 
     809          14 :                         numOfUnknownOptions--;
     810          14 :                         if(o==0) ppMode->baseDcDiff= val;
     811           7 :                         else ppMode->flatnessThreshold= val;
     812             :                     }
     813             :                 }
     814         126 :                 else if(filters[i].mask == FORCE_QUANT){
     815             :                     int o;
     816           7 :                     ppMode->forcedQuant= 15;
     817             : 
     818          14 :                     for(o=0; options[o] && o<1; o++){
     819             :                         char *tail;
     820           7 :                         int val= strtol(options[o], &tail, 0);
     821           7 :                         if(tail==options[o]) break;
     822             : 
     823           7 :                         numOfUnknownOptions--;
     824           7 :                         ppMode->forcedQuant= val;
     825             :                     }
     826             :                 }
     827             :             }
     828             :         }
     829         224 :         if(!filterNameOk) ppMode->error++;
     830         224 :         ppMode->error += numOfUnknownOptions;
     831             :     }
     832             : 
     833          56 :     av_log(NULL, AV_LOG_DEBUG, "pp: lumMode=%X, chromMode=%X\n", ppMode->lumMode, ppMode->chromMode);
     834          56 :     if(ppMode->error){
     835           0 :         av_log(NULL, AV_LOG_ERROR, "%d errors in postprocess string \"%s\"\n", ppMode->error, name);
     836           0 :         av_free(ppMode);
     837           0 :         return NULL;
     838             :     }
     839          56 :     return ppMode;
     840             : }
     841             : 
     842          56 : void pp_free_mode(pp_mode *mode){
     843          56 :     av_free(mode);
     844          56 : }
     845             : 
     846         224 : static void reallocAlign(void **p, int size){
     847         224 :     av_free(*p);
     848         224 :     *p= av_mallocz(size);
     849         224 : }
     850             : 
     851          16 : static void reallocBuffers(PPContext *c, int width, int height, int stride, int qpStride){
     852          16 :     int mbWidth = (width+15)>>4;
     853          16 :     int mbHeight= (height+15)>>4;
     854             :     int i;
     855             : 
     856          16 :     c->stride= stride;
     857          16 :     c->qpStride= qpStride;
     858             : 
     859          16 :     reallocAlign((void **)&c->tempDst, stride*24+32);
     860          16 :     reallocAlign((void **)&c->tempSrc, stride*24);
     861          16 :     reallocAlign((void **)&c->tempBlocks, 2*16*8);
     862          16 :     reallocAlign((void **)&c->yHistogram, 256*sizeof(uint64_t));
     863        4112 :     for(i=0; i<256; i++)
     864        4096 :             c->yHistogram[i]= width*height/64*15/256;
     865             : 
     866          64 :     for(i=0; i<3; i++){
     867             :         //Note: The +17*1024 is just there so I do not have to worry about r/w over the end.
     868          48 :         reallocAlign((void **)&c->tempBlurred[i], stride*mbHeight*16 + 17*1024);
     869          48 :         reallocAlign((void **)&c->tempBlurredPast[i], 256*((height+7)&(~7))/2 + 17*1024);//FIXME size
     870             :     }
     871             : 
     872          16 :     reallocAlign((void **)&c->deintTemp, 2*width+32);
     873          16 :     reallocAlign((void **)&c->nonBQPTable, qpStride*mbHeight*sizeof(int8_t));
     874          16 :     reallocAlign((void **)&c->stdQPTable, qpStride*mbHeight*sizeof(int8_t));
     875          16 :     reallocAlign((void **)&c->forcedQPTable, mbWidth*sizeof(int8_t));
     876          16 : }
     877             : 
     878           0 : static const char * context_to_name(void * ptr) {
     879           0 :     return "postproc";
     880             : }
     881             : 
     882             : static const AVClass av_codec_context_class = { "Postproc", context_to_name, NULL };
     883             : 
     884           8 : av_cold pp_context *pp_get_context(int width, int height, int cpuCaps){
     885           8 :     PPContext *c= av_mallocz(sizeof(PPContext));
     886           8 :     int stride= FFALIGN(width, 16);  //assumed / will realloc if needed
     887           8 :     int qpStride= (width+15)/16 + 2; //assumed / will realloc if needed
     888             : 
     889           8 :     if (!c)
     890           0 :         return NULL;
     891             : 
     892           8 :     c->av_class = &av_codec_context_class;
     893           8 :     if(cpuCaps&PP_FORMAT){
     894           8 :         c->hChromaSubSample= cpuCaps&0x3;
     895           8 :         c->vChromaSubSample= (cpuCaps>>4)&0x3;
     896             :     }else{
     897           0 :         c->hChromaSubSample= 1;
     898           0 :         c->vChromaSubSample= 1;
     899             :     }
     900           8 :     if (cpuCaps & PP_CPU_CAPS_AUTO) {
     901           8 :         c->cpuCaps = av_get_cpu_flags();
     902             :     } else {
     903           0 :         c->cpuCaps = 0;
     904           0 :         if (cpuCaps & PP_CPU_CAPS_MMX)      c->cpuCaps |= AV_CPU_FLAG_MMX;
     905           0 :         if (cpuCaps & PP_CPU_CAPS_MMX2)     c->cpuCaps |= AV_CPU_FLAG_MMXEXT;
     906           0 :         if (cpuCaps & PP_CPU_CAPS_3DNOW)    c->cpuCaps |= AV_CPU_FLAG_3DNOW;
     907           0 :         if (cpuCaps & PP_CPU_CAPS_ALTIVEC)  c->cpuCaps |= AV_CPU_FLAG_ALTIVEC;
     908             :     }
     909             : 
     910           8 :     reallocBuffers(c, width, height, stride, qpStride);
     911             : 
     912           8 :     c->frameNum=-1;
     913             : 
     914           8 :     return c;
     915             : }
     916             : 
     917           8 : av_cold void pp_free_context(void *vc){
     918           8 :     PPContext *c = (PPContext*)vc;
     919             :     int i;
     920             : 
     921          32 :     for(i=0; i<FF_ARRAY_ELEMS(c->tempBlurred); i++)
     922          24 :         av_free(c->tempBlurred[i]);
     923          32 :     for(i=0; i<FF_ARRAY_ELEMS(c->tempBlurredPast); i++)
     924          24 :         av_free(c->tempBlurredPast[i]);
     925             : 
     926           8 :     av_free(c->tempBlocks);
     927           8 :     av_free(c->yHistogram);
     928           8 :     av_free(c->tempDst);
     929           8 :     av_free(c->tempSrc);
     930           8 :     av_free(c->deintTemp);
     931           8 :     av_free(c->stdQPTable);
     932           8 :     av_free(c->nonBQPTable);
     933           8 :     av_free(c->forcedQPTable);
     934             : 
     935           8 :     memset(c, 0, sizeof(PPContext));
     936             : 
     937           8 :     av_free(c);
     938           8 : }
     939             : 
     940          41 : void  pp_postprocess(const uint8_t * src[3], const int srcStride[3],
     941             :                      uint8_t * dst[3], const int dstStride[3],
     942             :                      int width, int height,
     943             :                      const int8_t *QP_store,  int QPStride,
     944             :                      pp_mode *vm,  void *vc, int pict_type)
     945             : {
     946          41 :     int mbWidth = (width+15)>>4;
     947          41 :     int mbHeight= (height+15)>>4;
     948          41 :     PPMode *mode = vm;
     949          41 :     PPContext *c = vc;
     950          41 :     int minStride= FFMAX(FFABS(srcStride[0]), FFABS(dstStride[0]));
     951          41 :     int absQPStride = FFABS(QPStride);
     952             : 
     953             :     // c->stride and c->QPStride are always positive
     954          41 :     if(c->stride < minStride || c->qpStride < absQPStride)
     955          16 :         reallocBuffers(c, width, height,
     956           8 :                        FFMAX(minStride, c->stride),
     957           8 :                        FFMAX(c->qpStride, absQPStride));
     958             : 
     959          41 :     if(!QP_store || (mode->lumMode & FORCE_QUANT)){
     960             :         int i;
     961          21 :         QP_store= c->forcedQPTable;
     962          21 :         absQPStride = QPStride = 0;
     963          21 :         if(mode->lumMode & FORCE_QUANT)
     964           5 :             for(i=0; i<mbWidth; i++) c->forcedQPTable[i]= mode->forcedQuant;
     965             :         else
     966          16 :             for(i=0; i<mbWidth; i++) c->forcedQPTable[i]= 1;
     967             :     }
     968             : 
     969          41 :     if(pict_type & PP_PICT_TYPE_QP2){
     970             :         int i;
     971           0 :         const int count= FFMAX(mbHeight * absQPStride, mbWidth);
     972           0 :         for(i=0; i<(count>>2); i++){
     973           0 :             AV_WN32(c->stdQPTable + (i<<2), AV_RN32(QP_store + (i<<2)) >> 1 & 0x7F7F7F7F);
     974             :         }
     975           0 :         for(i<<=2; i<count; i++){
     976           0 :             c->stdQPTable[i] = QP_store[i]>>1;
     977             :         }
     978           0 :         QP_store= c->stdQPTable;
     979           0 :         QPStride= absQPStride;
     980             :     }
     981             : 
     982             :     if(0){
     983             :         int x,y;
     984             :         for(y=0; y<mbHeight; y++){
     985             :             for(x=0; x<mbWidth; x++){
     986             :                 av_log(c, AV_LOG_INFO, "%2d ", QP_store[x + y*QPStride]);
     987             :             }
     988             :             av_log(c, AV_LOG_INFO, "\n");
     989             :         }
     990             :         av_log(c, AV_LOG_INFO, "\n");
     991             :     }
     992             : 
     993          41 :     if((pict_type&7)!=3){
     994          38 :         if (QPStride >= 0){
     995             :             int i;
     996          38 :             const int count= FFMAX(mbHeight * QPStride, mbWidth);
     997        1834 :             for(i=0; i<(count>>2); i++){
     998        1796 :                 AV_WN32(c->nonBQPTable + (i<<2), AV_RN32(QP_store + (i<<2)) & 0x3F3F3F3F);
     999             :             }
    1000          84 :             for(i<<=2; i<count; i++){
    1001          46 :                 c->nonBQPTable[i] = QP_store[i] & 0x3F;
    1002             :             }
    1003             :         } else {
    1004             :             int i,j;
    1005           0 :             for(i=0; i<mbHeight; i++) {
    1006           0 :                 for(j=0; j<absQPStride; j++) {
    1007           0 :                     c->nonBQPTable[i*absQPStride+j] = QP_store[i*QPStride+j] & 0x3F;
    1008             :                 }
    1009             :             }
    1010             :         }
    1011             :     }
    1012             : 
    1013          41 :     av_log(c, AV_LOG_DEBUG, "using npp filters 0x%X/0x%X\n",
    1014             :            mode->lumMode, mode->chromMode);
    1015             : 
    1016          41 :     postProcess(src[0], srcStride[0], dst[0], dstStride[0],
    1017             :                 width, height, QP_store, QPStride, 0, mode, c);
    1018             : 
    1019          41 :     if (!(src[1] && src[2] && dst[1] && dst[2]))
    1020           0 :         return;
    1021             : 
    1022          41 :     width  = (width )>>c->hChromaSubSample;
    1023          41 :     height = (height)>>c->vChromaSubSample;
    1024             : 
    1025          41 :     if(mode->chromMode){
    1026          41 :         postProcess(src[1], srcStride[1], dst[1], dstStride[1],
    1027             :                     width, height, QP_store, QPStride, 1, mode, c);
    1028          41 :         postProcess(src[2], srcStride[2], dst[2], dstStride[2],
    1029             :                     width, height, QP_store, QPStride, 2, mode, c);
    1030             :     }
    1031           0 :     else if(srcStride[1] == dstStride[1] && srcStride[2] == dstStride[2]){
    1032           0 :         linecpy(dst[1], src[1], height, srcStride[1]);
    1033           0 :         linecpy(dst[2], src[2], height, srcStride[2]);
    1034             :     }else{
    1035             :         int y;
    1036           0 :         for(y=0; y<height; y++){
    1037           0 :             memcpy(&(dst[1][y*dstStride[1]]), &(src[1][y*srcStride[1]]), width);
    1038           0 :             memcpy(&(dst[2][y*dstStride[2]]), &(src[2][y*srcStride[2]]), width);
    1039             :         }
    1040             :     }
    1041             : }

Generated by: LCOV version 1.13