falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
17202 Commits
32 Branches
959MB
Tree: 87609b51da
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '87609b51da'
${ noResults }
FFmpeg/libpostproc/postprocess.c

1091 lines
37KB
Raw Blame History 

  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 postprocess.c

  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 the Subversion log

  75. 

  76. #include "config.h"

  77. #include "libavutil/avutil.h"

  78. #include <inttypes.h>

  79. #include <stdio.h>

  80. #include <stdlib.h>

  81. #include <string.h>

  82. //#undef HAVE_MMX2

  83. //#define HAVE_3DNOW

  84. //#undef HAVE_MMX

  85. //#undef ARCH_X86

  86. //#define DEBUG_BRIGHTNESS

  87. #include "postprocess.h"

  88. #include "postprocess_internal.h"

  89. 

  90. unsigned postproc_version(void)

  91. {

  92.     return LIBPOSTPROC_VERSION_INT;

  93. }

  94. 

  95. #ifdef HAVE_ALTIVEC_H

  96. #include <altivec.h>

  97. #endif

  98. 

  99. #define GET_MODE_BUFFER_SIZE 500

  100. #define OPTIONS_ARRAY_SIZE 10

  101. #define BLOCK_SIZE 8

  102. #define TEMP_STRIDE 8

  103. //#define NUM_BLOCKS_AT_ONCE 16 //not used yet

  104. 

  105. #if defined(ARCH_X86)

  106. DECLARE_ASM_CONST(8, uint64_t, w05)= 0x0005000500050005LL;

  107. DECLARE_ASM_CONST(8, uint64_t, w04)= 0x0004000400040004LL;

  108. DECLARE_ASM_CONST(8, uint64_t, w20)= 0x0020002000200020LL;

  109. DECLARE_ASM_CONST(8, uint64_t, b00)= 0x0000000000000000LL;

  110. DECLARE_ASM_CONST(8, uint64_t, b01)= 0x0101010101010101LL;

  111. DECLARE_ASM_CONST(8, uint64_t, b02)= 0x0202020202020202LL;

  112. DECLARE_ASM_CONST(8, uint64_t, b08)= 0x0808080808080808LL;

  113. DECLARE_ASM_CONST(8, uint64_t, b80)= 0x8080808080808080LL;

  114. #endif

  115. 

  116. DECLARE_ASM_CONST(8, int, deringThreshold)= 20;

  117. 

  118. 

  119. static struct PPFilter filters[]=

  120. {

  121.     {"hb", "hdeblock",              1, 1, 3, H_DEBLOCK},

  122.     {"vb", "vdeblock",              1, 2, 4, V_DEBLOCK},

  123. /*  {"hr", "rkhdeblock",            1, 1, 3, H_RK1_FILTER},

  124.     {"vr", "rkvdeblock",            1, 2, 4, V_RK1_FILTER},*/

  125.     {"h1", "x1hdeblock",            1, 1, 3, H_X1_FILTER},

  126.     {"v1", "x1vdeblock",            1, 2, 4, V_X1_FILTER},

  127.     {"ha", "ahdeblock",             1, 1, 3, H_A_DEBLOCK},

  128.     {"va", "avdeblock",             1, 2, 4, V_A_DEBLOCK},

  129.     {"dr", "dering",                1, 5, 6, DERING},

  130.     {"al", "autolevels",            0, 1, 2, LEVEL_FIX},

  131.     {"lb", "linblenddeint",         1, 1, 4, LINEAR_BLEND_DEINT_FILTER},

  132.     {"li", "linipoldeint",          1, 1, 4, LINEAR_IPOL_DEINT_FILTER},

  133.     {"ci", "cubicipoldeint",        1, 1, 4, CUBIC_IPOL_DEINT_FILTER},

  134.     {"md", "mediandeint",           1, 1, 4, MEDIAN_DEINT_FILTER},

  135.     {"fd", "ffmpegdeint",           1, 1, 4, FFMPEG_DEINT_FILTER},

  136.     {"l5", "lowpass5",              1, 1, 4, LOWPASS5_DEINT_FILTER},

  137.     {"tn", "tmpnoise",              1, 7, 8, TEMP_NOISE_FILTER},

  138.     {"fq", "forcequant",            1, 0, 0, FORCE_QUANT},

  139.     {NULL, NULL,0,0,0,0} //End Marker

  140. };

  141. 

  142. static const char *replaceTable[]=

  143. {

  144.     "default",      "hb:a,vb:a,dr:a",

  145.     "de",           "hb:a,vb:a,dr:a",

  146.     "fast",         "h1:a,v1:a,dr:a",

  147.     "fa",           "h1:a,v1:a,dr:a",

  148.     "ac",           "ha:a:128:7,va:a,dr:a",

  149.     NULL //End Marker

  150. };

  151. 

  152. 

  153. #if defined(ARCH_X86)

  154. static inline void prefetchnta(void *p)

  155. {

  156.     __asm__ volatile(   "prefetchnta (%0)\n\t"

  157.         : : "r" (p)

  158.     );

  159. }

  160. 

  161. static inline void prefetcht0(void *p)

  162. {

  163.     __asm__ volatile(   "prefetcht0 (%0)\n\t"

  164.         : : "r" (p)

  165.     );

  166. }

  167. 

  168. static inline void prefetcht1(void *p)

  169. {

  170.     __asm__ volatile(   "prefetcht1 (%0)\n\t"

  171.         : : "r" (p)

  172.     );

  173. }

  174. 

  175. static inline void prefetcht2(void *p)

  176. {

  177.     __asm__ volatile(   "prefetcht2 (%0)\n\t"

  178.         : : "r" (p)

  179.     );

  180. }

  181. #endif

  182. 

  183. /* The horizontal functions exist only in C because the MMX

  184.  * code is faster with vertical filters and transposing. */

  185. 

  186. /**

  187.  * Check if the given 8x8 Block is mostly "flat"

  188.  */

  189. static inline int isHorizDC_C(uint8_t src[], int stride, PPContext *c)

  190. {

  191.     int numEq= 0;

  192.     int y;

  193.     const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;

  194.     const int dcThreshold= dcOffset*2 + 1;

  195. 

  196.     for(y=0; y<BLOCK_SIZE; y++){

  197.         if(((unsigned)(src[0] - src[1] + dcOffset)) < dcThreshold) numEq++;

  198.         if(((unsigned)(src[1] - src[2] + dcOffset)) < dcThreshold) numEq++;

  199.         if(((unsigned)(src[2] - src[3] + dcOffset)) < dcThreshold) numEq++;

  200.         if(((unsigned)(src[3] - src[4] + dcOffset)) < dcThreshold) numEq++;

  201.         if(((unsigned)(src[4] - src[5] + dcOffset)) < dcThreshold) numEq++;

  202.         if(((unsigned)(src[5] - src[6] + dcOffset)) < dcThreshold) numEq++;

  203.         if(((unsigned)(src[6] - src[7] + dcOffset)) < dcThreshold) numEq++;

  204.         src+= stride;

  205.     }

  206.     return numEq > c->ppMode.flatnessThreshold;

  207. }

  208. 

  209. /**

  210.  * Check if the middle 8x8 Block in the given 8x16 block is flat

  211.  */

  212. static inline int isVertDC_C(uint8_t src[], int stride, PPContext *c)

  213. {

  214.     int numEq= 0;

  215.     int y;

  216.     const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;

  217.     const int dcThreshold= dcOffset*2 + 1;

  218. 

  219.     src+= stride*4; // src points to begin of the 8x8 Block

  220.     for(y=0; y<BLOCK_SIZE-1; y++){

  221.         if(((unsigned)(src[0] - src[0+stride] + dcOffset)) < dcThreshold) numEq++;

  222.         if(((unsigned)(src[1] - src[1+stride] + dcOffset)) < dcThreshold) numEq++;

  223.         if(((unsigned)(src[2] - src[2+stride] + dcOffset)) < dcThreshold) numEq++;

  224.         if(((unsigned)(src[3] - src[3+stride] + dcOffset)) < dcThreshold) numEq++;

  225.         if(((unsigned)(src[4] - src[4+stride] + dcOffset)) < dcThreshold) numEq++;

  226.         if(((unsigned)(src[5] - src[5+stride] + dcOffset)) < dcThreshold) numEq++;

  227.         if(((unsigned)(src[6] - src[6+stride] + dcOffset)) < dcThreshold) numEq++;

  228.         if(((unsigned)(src[7] - src[7+stride] + dcOffset)) < dcThreshold) numEq++;

  229.         src+= stride;

  230.     }

  231.     return numEq > c->ppMode.flatnessThreshold;

  232. }

  233. 

  234. static inline int isHorizMinMaxOk_C(uint8_t src[], int stride, int QP)

  235. {

  236.     int i;

  237. #if 1

  238.     for(i=0; i<2; i++){

  239.         if((unsigned)(src[0] - src[5] + 2*QP) > 4*QP) return 0;

  240.         src += stride;

  241.         if((unsigned)(src[2] - src[7] + 2*QP) > 4*QP) return 0;

  242.         src += stride;

  243.         if((unsigned)(src[4] - src[1] + 2*QP) > 4*QP) return 0;

  244.         src += stride;

  245.         if((unsigned)(src[6] - src[3] + 2*QP) > 4*QP) return 0;

  246.         src += stride;

  247.     }

  248. #else

  249.     for(i=0; i<8; i++){

  250.         if((unsigned)(src[0] - src[7] + 2*QP) > 4*QP) return 0;

  251.         src += stride;

  252.     }

  253. #endif

  254.     return 1;

  255. }

  256. 

  257. static inline int isVertMinMaxOk_C(uint8_t src[], int stride, int QP)

  258. {

  259. #if 1

  260. #if 1

  261.     int x;

  262.     src+= stride*4;

  263.     for(x=0; x<BLOCK_SIZE; x+=4){

  264.         if((unsigned)(src[  x + 0*stride] - src[  x + 5*stride] + 2*QP) > 4*QP) return 0;

  265.         if((unsigned)(src[1+x + 2*stride] - src[1+x + 7*stride] + 2*QP) > 4*QP) return 0;

  266.         if((unsigned)(src[2+x + 4*stride] - src[2+x + 1*stride] + 2*QP) > 4*QP) return 0;

  267.         if((unsigned)(src[3+x + 6*stride] - src[3+x + 3*stride] + 2*QP) > 4*QP) return 0;

  268.     }

  269. #else

  270.     int x;

  271.     src+= stride*3;

  272.     for(x=0; x<BLOCK_SIZE; x++){

  273.         if((unsigned)(src[x + stride] - src[x + (stride<<3)] + 2*QP) > 4*QP) return 0;

  274.     }

  275. #endif

  276.     return 1;

  277. #else

  278.     int x;

  279.     src+= stride*4;

  280.     for(x=0; x<BLOCK_SIZE; x++){

  281.         int min=255;

  282.         int max=0;

  283.         int y;

  284.         for(y=0; y<8; y++){

  285.             int v= src[x + y*stride];

  286.             if(v>max) max=v;

  287.             if(v<min) min=v;

  288.         }

  289.         if(max-min > 2*QP) return 0;

  290.     }

  291.     return 1;

  292. #endif

  293. }

  294. 

  295. static inline int horizClassify_C(uint8_t src[], int stride, PPContext *c)

  296. {

  297.     if( isHorizDC_C(src, stride, c) ){

  298.         if( isHorizMinMaxOk_C(src, stride, c->QP) )

  299.             return 1;

  300.         else

  301.             return 0;

  302.     }else{

  303.         return 2;

  304.     }

  305. }

  306. 

  307. static inline int vertClassify_C(uint8_t src[], int stride, PPContext *c)

  308. {

  309.     if( isVertDC_C(src, stride, c) ){

  310.         if( isVertMinMaxOk_C(src, stride, c->QP) )

  311.             return 1;

  312.         else

  313.             return 0;

  314.     }else{

  315.         return 2;

  316.     }

  317. }

  318. 

  319. static inline void doHorizDefFilter_C(uint8_t dst[], int stride, PPContext *c)

  320. {

  321.     int y;

  322.     for(y=0; y<BLOCK_SIZE; y++){

  323.         const int middleEnergy= 5*(dst[4] - dst[3]) + 2*(dst[2] - dst[5]);

  324. 

  325.         if(FFABS(middleEnergy) < 8*c->QP){

  326.             const int q=(dst[3] - dst[4])/2;

  327.             const int leftEnergy=  5*(dst[2] - dst[1]) + 2*(dst[0] - dst[3]);

  328.             const int rightEnergy= 5*(dst[6] - dst[5]) + 2*(dst[4] - dst[7]);

  329. 

  330.             int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );

  331.             d= FFMAX(d, 0);

  332. 

  333.             d= (5*d + 32) >> 6;

  334.             d*= FFSIGN(-middleEnergy);

  335. 

  336.             if(q>0)

  337.             {

  338.                 d= d<0 ? 0 : d;

  339.                 d= d>q ? q : d;

  340.             }

  341.             else

  342.             {

  343.                 d= d>0 ? 0 : d;

  344.                 d= d<q ? q : d;

  345.             }

  346. 

  347.             dst[3]-= d;

  348.             dst[4]+= d;

  349.         }

  350.         dst+= stride;

  351.     }

  352. }

  353. 

  354. /**

  355.  * Do a horizontal low pass filter on the 10x8 block (dst points to middle 8x8 Block)

  356.  * using the 9-Tap Filter (1,1,2,2,4,2,2,1,1)/16 (C version)

  357.  */

  358. static inline void doHorizLowPass_C(uint8_t dst[], int stride, PPContext *c)

  359. {

  360.     int y;

  361.     for(y=0; y<BLOCK_SIZE; y++){

  362.         const int first= FFABS(dst[-1] - dst[0]) < c->QP ? dst[-1] : dst[0];

  363.         const int last= FFABS(dst[8] - dst[7]) < c->QP ? dst[8] : dst[7];

  364. 

  365.         int sums[10];

  366.         sums[0] = 4*first + dst[0] + dst[1] + dst[2] + 4;

  367.         sums[1] = sums[0] - first  + dst[3];

  368.         sums[2] = sums[1] - first  + dst[4];

  369.         sums[3] = sums[2] - first  + dst[5];

  370.         sums[4] = sums[3] - first  + dst[6];

  371.         sums[5] = sums[4] - dst[0] + dst[7];

  372.         sums[6] = sums[5] - dst[1] + last;

  373.         sums[7] = sums[6] - dst[2] + last;

  374.         sums[8] = sums[7] - dst[3] + last;

  375.         sums[9] = sums[8] - dst[4] + last;

  376. 

  377.         dst[0]= (sums[0] + sums[2] + 2*dst[0])>>4;

  378.         dst[1]= (sums[1] + sums[3] + 2*dst[1])>>4;

  379.         dst[2]= (sums[2] + sums[4] + 2*dst[2])>>4;

  380.         dst[3]= (sums[3] + sums[5] + 2*dst[3])>>4;

  381.         dst[4]= (sums[4] + sums[6] + 2*dst[4])>>4;

  382.         dst[5]= (sums[5] + sums[7] + 2*dst[5])>>4;

  383.         dst[6]= (sums[6] + sums[8] + 2*dst[6])>>4;

  384.         dst[7]= (sums[7] + sums[9] + 2*dst[7])>>4;

  385. 

  386.         dst+= stride;

  387.     }

  388. }

  389. 

  390. /**

  391.  * Experimental Filter 1 (Horizontal)

  392.  * will not damage linear gradients

  393.  * Flat blocks should look like they were passed through the (1,1,2,2,4,2,2,1,1) 9-Tap filter

  394.  * can only smooth blocks at the expected locations (it cannot smooth them if they did move)

  395.  * MMX2 version does correct clipping C version does not

  396.  * not identical with the vertical one

  397.  */

  398. static inline void horizX1Filter(uint8_t *src, int stride, int QP)

  399. {

  400.     int y;

  401.     static uint64_t *lut= NULL;

  402.     if(lut==NULL)

  403.     {

  404.         int i;

  405.         lut = av_malloc(256*8);

  406.         for(i=0; i<256; i++)

  407.         {

  408.             int v= i < 128 ? 2*i : 2*(i-256);

  409. /*

  410. //Simulate 112242211 9-Tap filter

  411.             uint64_t a= (v/16)  & 0xFF;

  412.             uint64_t b= (v/8)   & 0xFF;

  413.             uint64_t c= (v/4)   & 0xFF;

  414.             uint64_t d= (3*v/8) & 0xFF;

  415. */

  416. //Simulate piecewise linear interpolation

  417.             uint64_t a= (v/16)   & 0xFF;

  418.             uint64_t b= (v*3/16) & 0xFF;

  419.             uint64_t c= (v*5/16) & 0xFF;

  420.             uint64_t d= (7*v/16) & 0xFF;

  421.             uint64_t A= (0x100 - a)&0xFF;

  422.             uint64_t B= (0x100 - b)&0xFF;

  423.             uint64_t C= (0x100 - c)&0xFF;

  424.             uint64_t D= (0x100 - c)&0xFF;

  425. 

  426.             lut[i]   = (a<<56) | (b<<48) | (c<<40) | (d<<32) |

  427.                        (D<<24) | (C<<16) | (B<<8)  | (A);

  428.             //lut[i] = (v<<32) | (v<<24);

  429.         }

  430.     }

  431. 

  432.     for(y=0; y<BLOCK_SIZE; y++){

  433.         int a= src[1] - src[2];

  434.         int b= src[3] - src[4];

  435.         int c= src[5] - src[6];

  436. 

  437.         int d= FFMAX(FFABS(b) - (FFABS(a) + FFABS(c))/2, 0);

  438. 

  439.         if(d < QP){

  440.             int v = d * FFSIGN(-b);

  441. 

  442.             src[1] +=v/8;

  443.             src[2] +=v/4;

  444.             src[3] +=3*v/8;

  445.             src[4] -=3*v/8;

  446.             src[5] -=v/4;

  447.             src[6] -=v/8;

  448.         }

  449.         src+=stride;

  450.     }

  451. }

  452. 

  453. /**

  454.  * accurate deblock filter

  455.  */

  456. static av_always_inline void do_a_deblock_C(uint8_t *src, int step, int stride, PPContext *c){

  457.     int y;

  458.     const int QP= c->QP;

  459.     const int dcOffset= ((c->nonBQP*c->ppMode.baseDcDiff)>>8) + 1;

  460.     const int dcThreshold= dcOffset*2 + 1;

  461. //START_TIMER

  462.     src+= step*4; // src points to begin of the 8x8 Block

  463.     for(y=0; y<8; y++){

  464.         int numEq= 0;

  465. 

  466.         if(((unsigned)(src[-1*step] - src[0*step] + dcOffset)) < dcThreshold) numEq++;

  467.         if(((unsigned)(src[ 0*step] - src[1*step] + dcOffset)) < dcThreshold) numEq++;

  468.         if(((unsigned)(src[ 1*step] - src[2*step] + dcOffset)) < dcThreshold) numEq++;

  469.         if(((unsigned)(src[ 2*step] - src[3*step] + dcOffset)) < dcThreshold) numEq++;

  470.         if(((unsigned)(src[ 3*step] - src[4*step] + dcOffset)) < dcThreshold) numEq++;

  471.         if(((unsigned)(src[ 4*step] - src[5*step] + dcOffset)) < dcThreshold) numEq++;

  472.         if(((unsigned)(src[ 5*step] - src[6*step] + dcOffset)) < dcThreshold) numEq++;

  473.         if(((unsigned)(src[ 6*step] - src[7*step] + dcOffset)) < dcThreshold) numEq++;

  474.         if(((unsigned)(src[ 7*step] - src[8*step] + dcOffset)) < dcThreshold) numEq++;

  475.         if(numEq > c->ppMode.flatnessThreshold){

  476.             int min, max, x;

  477. 

  478.             if(src[0] > src[step]){

  479.                 max= src[0];

  480.                 min= src[step];

  481.             }else{

  482.                 max= src[step];

  483.                 min= src[0];

  484.             }

  485.             for(x=2; x<8; x+=2){

  486.                 if(src[x*step] > src[(x+1)*step]){

  487.                         if(src[x    *step] > max) max= src[ x   *step];

  488.                         if(src[(x+1)*step] < min) min= src[(x+1)*step];

  489.                 }else{

  490.                         if(src[(x+1)*step] > max) max= src[(x+1)*step];

  491.                         if(src[ x   *step] < min) min= src[ x   *step];

  492.                 }

  493.             }

  494.             if(max-min < 2*QP){

  495.                 const int first= FFABS(src[-1*step] - src[0]) < QP ? src[-1*step] : src[0];

  496.                 const int last= FFABS(src[8*step] - src[7*step]) < QP ? src[8*step] : src[7*step];

  497. 

  498.                 int sums[10];

  499.                 sums[0] = 4*first + src[0*step] + src[1*step] + src[2*step] + 4;

  500.                 sums[1] = sums[0] - first       + src[3*step];

  501.                 sums[2] = sums[1] - first       + src[4*step];

  502.                 sums[3] = sums[2] - first       + src[5*step];

  503.                 sums[4] = sums[3] - first       + src[6*step];

  504.                 sums[5] = sums[4] - src[0*step] + src[7*step];

  505.                 sums[6] = sums[5] - src[1*step] + last;

  506.                 sums[7] = sums[6] - src[2*step] + last;

  507.                 sums[8] = sums[7] - src[3*step] + last;

  508.                 sums[9] = sums[8] - src[4*step] + last;

  509. 

  510.                 src[0*step]= (sums[0] + sums[2] + 2*src[0*step])>>4;

  511.                 src[1*step]= (sums[1] + sums[3] + 2*src[1*step])>>4;

  512.                 src[2*step]= (sums[2] + sums[4] + 2*src[2*step])>>4;

  513.                 src[3*step]= (sums[3] + sums[5] + 2*src[3*step])>>4;

  514.                 src[4*step]= (sums[4] + sums[6] + 2*src[4*step])>>4;

  515.                 src[5*step]= (sums[5] + sums[7] + 2*src[5*step])>>4;

  516.                 src[6*step]= (sums[6] + sums[8] + 2*src[6*step])>>4;

  517.                 src[7*step]= (sums[7] + sums[9] + 2*src[7*step])>>4;

  518.             }

  519.         }else{

  520.             const int middleEnergy= 5*(src[4*step] - src[3*step]) + 2*(src[2*step] - src[5*step]);

  521. 

  522.             if(FFABS(middleEnergy) < 8*QP){

  523.                 const int q=(src[3*step] - src[4*step])/2;

  524.                 const int leftEnergy=  5*(src[2*step] - src[1*step]) + 2*(src[0*step] - src[3*step]);

  525.                 const int rightEnergy= 5*(src[6*step] - src[5*step]) + 2*(src[4*step] - src[7*step]);

  526. 

  527.                 int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );

  528.                 d= FFMAX(d, 0);

  529. 

  530.                 d= (5*d + 32) >> 6;

  531.                 d*= FFSIGN(-middleEnergy);

  532. 

  533.                 if(q>0){

  534.                     d= d<0 ? 0 : d;

  535.                     d= d>q ? q : d;

  536.                 }else{

  537.                     d= d>0 ? 0 : d;

  538.                     d= d<q ? q : d;

  539.                 }

  540. 

  541.                 src[3*step]-= d;

  542.                 src[4*step]+= d;

  543.             }

  544.         }

  545. 

  546.         src += stride;

  547.     }

  548. /*if(step==16){

  549.     STOP_TIMER("step16")

  550. }else{

  551.     STOP_TIMER("stepX")

  552. }*/

  553. }

  554. 

  555. //Note: we have C, MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one

  556. //Plain C versions

  557. #if !(defined (HAVE_MMX) || defined (HAVE_ALTIVEC)) || defined (RUNTIME_CPUDETECT)

  558. #define COMPILE_C

  559. #endif

  560. 

  561. #ifdef HAVE_ALTIVEC

  562. #define COMPILE_ALTIVEC

  563. #endif //HAVE_ALTIVEC

  564. 

  565. #if defined(ARCH_X86)

  566. 

  567. #if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)

  568. #define COMPILE_MMX

  569. #endif

  570. 

  571. #if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)

  572. #define COMPILE_MMX2

  573. #endif

  574. 

  575. #if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)

  576. #define COMPILE_3DNOW

  577. #endif

  578. #endif /* defined(ARCH_X86) */

  579. 

  580. #undef HAVE_MMX

  581. #undef HAVE_MMX2

  582. #undef HAVE_3DNOW

  583. #undef HAVE_ALTIVEC

  584. 

  585. #ifdef COMPILE_C

  586. #undef HAVE_MMX

  587. #undef HAVE_MMX2

  588. #undef HAVE_3DNOW

  589. #define RENAME(a) a ## _C

  590. #include "postprocess_template.c"

  591. #endif

  592. 

  593. #ifdef COMPILE_ALTIVEC

  594. #undef RENAME

  595. #define HAVE_ALTIVEC

  596. #define RENAME(a) a ## _altivec

  597. #include "postprocess_altivec_template.c"

  598. #include "postprocess_template.c"

  599. #endif

  600. 

  601. //MMX versions

  602. #ifdef COMPILE_MMX

  603. #undef RENAME

  604. #define HAVE_MMX

  605. #undef HAVE_MMX2

  606. #undef HAVE_3DNOW

  607. #define RENAME(a) a ## _MMX

  608. #include "postprocess_template.c"

  609. #endif

  610. 

  611. //MMX2 versions

  612. #ifdef COMPILE_MMX2

  613. #undef RENAME

  614. #define HAVE_MMX

  615. #define HAVE_MMX2

  616. #undef HAVE_3DNOW

  617. #define RENAME(a) a ## _MMX2

  618. #include "postprocess_template.c"

  619. #endif

  620. 

  621. //3DNOW versions

  622. #ifdef COMPILE_3DNOW

  623. #undef RENAME

  624. #define HAVE_MMX

  625. #undef HAVE_MMX2

  626. #define HAVE_3DNOW

  627. #define RENAME(a) a ## _3DNow

  628. #include "postprocess_template.c"

  629. #endif

  630. 

  631. // minor note: the HAVE_xyz is messed up after that line so do not use it.

  632. 

  633. static inline void postProcess(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,

  634.         const QP_STORE_T QPs[], int QPStride, int isColor, pp_mode *vm, pp_context *vc)

  635. {

  636.     PPContext *c= (PPContext *)vc;

  637.     PPMode *ppMode= (PPMode *)vm;

  638.     c->ppMode= *ppMode; //FIXME

  639. 

  640.     // Using ifs here as they are faster than function pointers although the

  641.     // difference would not be measurable here but it is much better because

  642.     // someone might exchange the CPU whithout restarting MPlayer ;)

  643. #ifdef RUNTIME_CPUDETECT

  644. #if defined(ARCH_X86)

  645.     // ordered per speed fastest first

  646.     if(c->cpuCaps & PP_CPU_CAPS_MMX2)

  647.         postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  648.     else if(c->cpuCaps & PP_CPU_CAPS_3DNOW)

  649.         postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  650.     else if(c->cpuCaps & PP_CPU_CAPS_MMX)

  651.         postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  652.     else

  653.         postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  654. #else

  655. #ifdef HAVE_ALTIVEC

  656.     if(c->cpuCaps & PP_CPU_CAPS_ALTIVEC)

  657.             postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  658.     else

  659. #endif

  660.             postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  661. #endif

  662. #else //RUNTIME_CPUDETECT

  663. #ifdef HAVE_MMX2

  664.             postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  665. #elif defined (HAVE_3DNOW)

  666.             postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  667. #elif defined (HAVE_MMX)

  668.             postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  669. #elif defined (HAVE_ALTIVEC)

  670.             postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  671. #else

  672.             postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  673. #endif

  674. #endif //!RUNTIME_CPUDETECT

  675. }

  676. 

  677. //static void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,

  678. //        QP_STORE_T QPs[], int QPStride, int isColor, struct PPMode *ppMode);

  679. 

  680. /* -pp Command line Help

  681. */

  682. #if LIBPOSTPROC_VERSION_INT < (52<<16)

  683. const char *const pp_help=

  684. #else

  685. const char pp_help[] =

  686. #endif

  687. "Available postprocessing filters:\n"

  688. "Filters                        Options\n"

  689. "short  long name       short   long option     Description\n"

  690. "*      *               a       autoq           CPU power dependent enabler\n"

  691. "                       c       chrom           chrominance filtering enabled\n"

  692. "                       y       nochrom         chrominance filtering disabled\n"

  693. "                       n       noluma          luma filtering disabled\n"

  694. "hb     hdeblock        (2 threshold)           horizontal deblocking filter\n"

  695. "       1. difference factor: default=32, higher -> more deblocking\n"

  696. "       2. flatness threshold: default=39, lower -> more deblocking\n"

  697. "                       the h & v deblocking filters share these\n"

  698. "                       so you can't set different thresholds for h / v\n"

  699. "vb     vdeblock        (2 threshold)           vertical deblocking filter\n"

  700. "ha     hadeblock       (2 threshold)           horizontal deblocking filter\n"

  701. "va     vadeblock       (2 threshold)           vertical deblocking filter\n"

  702. "h1     x1hdeblock                              experimental h deblock filter 1\n"

  703. "v1     x1vdeblock                              experimental v deblock filter 1\n"

  704. "dr     dering                                  deringing filter\n"

  705. "al     autolevels                              automatic brightness / contrast\n"

  706. "                       f        fullyrange     stretch luminance to (0..255)\n"

  707. "lb     linblenddeint                           linear blend deinterlacer\n"

  708. "li     linipoldeint                            linear interpolating deinterlace\n"

  709. "ci     cubicipoldeint                          cubic interpolating deinterlacer\n"

  710. "md     mediandeint                             median deinterlacer\n"

  711. "fd     ffmpegdeint                             ffmpeg deinterlacer\n"

  712. "l5     lowpass5                                FIR lowpass deinterlacer\n"

  713. "de     default                                 hb:a,vb:a,dr:a\n"

  714. "fa     fast                                    h1:a,v1:a,dr:a\n"

  715. "ac                                             ha:a:128:7,va:a,dr:a\n"

  716. "tn     tmpnoise        (3 threshold)           temporal noise reducer\n"

  717. "                     1. <= 2. <= 3.            larger -> stronger filtering\n"

  718. "fq     forceQuant      <quantizer>             force quantizer\n"

  719. "Usage:\n"

  720. "<filterName>[:<option>[:<option>...]][[,|/][-]<filterName>[:<option>...]]...\n"

  721. "long form example:\n"

  722. "vdeblock:autoq/hdeblock:autoq/linblenddeint    default,-vdeblock\n"

  723. "short form example:\n"

  724. "vb:a/hb:a/lb                                   de,-vb\n"

  725. "more examples:\n"

  726. "tn:64:128:256\n"

  727. "\n"

  728. ;

  729. 

  730. pp_mode *pp_get_mode_by_name_and_quality(const char *name, int quality)

  731. {

  732.     char temp[GET_MODE_BUFFER_SIZE];

  733.     char *p= temp;

  734.     static const char filterDelimiters[] = ",/";

  735.     static const char optionDelimiters[] = ":";

  736.     struct PPMode *ppMode;

  737.     char *filterToken;

  738. 

  739.     ppMode= av_malloc(sizeof(PPMode));

  740. 

  741.     ppMode->lumMode= 0;

  742.     ppMode->chromMode= 0;

  743.     ppMode->maxTmpNoise[0]= 700;

  744.     ppMode->maxTmpNoise[1]= 1500;

  745.     ppMode->maxTmpNoise[2]= 3000;

  746.     ppMode->maxAllowedY= 234;

  747.     ppMode->minAllowedY= 16;

  748.     ppMode->baseDcDiff= 256/8;

  749.     ppMode->flatnessThreshold= 56-16-1;

  750.     ppMode->maxClippedThreshold= 0.01;

  751.     ppMode->error=0;

  752. 

  753.     strncpy(temp, name, GET_MODE_BUFFER_SIZE);

  754. 

  755.     av_log(NULL, AV_LOG_DEBUG, "pp: %s\n", name);

  756. 

  757.     for(;;){

  758.         char *filterName;

  759.         int q= 1000000; //PP_QUALITY_MAX;

  760.         int chrom=-1;

  761.         int luma=-1;

  762.         char *option;

  763.         char *options[OPTIONS_ARRAY_SIZE];

  764.         int i;

  765.         int filterNameOk=0;

  766.         int numOfUnknownOptions=0;

  767.         int enable=1; //does the user want us to enabled or disabled the filter

  768. 

  769.         filterToken= strtok(p, filterDelimiters);

  770.         if(filterToken == NULL) break;

  771.         p+= strlen(filterToken) + 1; // p points to next filterToken

  772.         filterName= strtok(filterToken, optionDelimiters);

  773.         av_log(NULL, AV_LOG_DEBUG, "pp: %s::%s\n", filterToken, filterName);

  774. 

  775.         if(*filterName == '-'){

  776.             enable=0;

  777.             filterName++;

  778.         }

  779. 

  780.         for(;;){ //for all options

  781.             option= strtok(NULL, optionDelimiters);

  782.             if(option == NULL) break;

  783. 

  784.             av_log(NULL, AV_LOG_DEBUG, "pp: option: %s\n", option);

  785.             if(!strcmp("autoq", option) || !strcmp("a", option)) q= quality;

  786.             else if(!strcmp("nochrom", option) || !strcmp("y", option)) chrom=0;

  787.             else if(!strcmp("chrom", option) || !strcmp("c", option)) chrom=1;

  788.             else if(!strcmp("noluma", option) || !strcmp("n", option)) luma=0;

  789.             else{

  790.                 options[numOfUnknownOptions] = option;

  791.                 numOfUnknownOptions++;

  792.             }

  793.             if(numOfUnknownOptions >= OPTIONS_ARRAY_SIZE-1) break;

  794.         }

  795.         options[numOfUnknownOptions] = NULL;

  796. 

  797.         /* replace stuff from the replace Table */

  798.         for(i=0; replaceTable[2*i]!=NULL; i++){

  799.             if(!strcmp(replaceTable[2*i], filterName)){

  800.                 int newlen= strlen(replaceTable[2*i + 1]);

  801.                 int plen;

  802.                 int spaceLeft;

  803. 

  804.                 if(p==NULL) p= temp, *p=0;      //last filter

  805.                 else p--, *p=',';               //not last filter

  806. 

  807.                 plen= strlen(p);

  808.                 spaceLeft= p - temp + plen;

  809.                 if(spaceLeft + newlen  >= GET_MODE_BUFFER_SIZE){

  810.                     ppMode->error++;

  811.                     break;

  812.                 }

  813.                 memmove(p + newlen, p, plen+1);

  814.                 memcpy(p, replaceTable[2*i + 1], newlen);

  815.                 filterNameOk=1;

  816.             }

  817.         }

  818. 

  819.         for(i=0; filters[i].shortName!=NULL; i++){

  820.             if(   !strcmp(filters[i].longName, filterName)

  821.                || !strcmp(filters[i].shortName, filterName)){

  822.                 ppMode->lumMode &= ~filters[i].mask;

  823.                 ppMode->chromMode &= ~filters[i].mask;

  824. 

  825.                 filterNameOk=1;

  826.                 if(!enable) break; // user wants to disable it

  827. 

  828.                 if(q >= filters[i].minLumQuality && luma)

  829.                     ppMode->lumMode|= filters[i].mask;

  830.                 if(chrom==1 || (chrom==-1 && filters[i].chromDefault))

  831.                     if(q >= filters[i].minChromQuality)

  832.                             ppMode->chromMode|= filters[i].mask;

  833. 

  834.                 if(filters[i].mask == LEVEL_FIX){

  835.                     int o;

  836.                     ppMode->minAllowedY= 16;

  837.                     ppMode->maxAllowedY= 234;

  838.                     for(o=0; options[o]!=NULL; o++){

  839.                         if(  !strcmp(options[o],"fullyrange")

  840.                            ||!strcmp(options[o],"f")){

  841.                             ppMode->minAllowedY= 0;

  842.                             ppMode->maxAllowedY= 255;

  843.                             numOfUnknownOptions--;

  844.                         }

  845.                     }

  846.                 }

  847.                 else if(filters[i].mask == TEMP_NOISE_FILTER)

  848.                 {

  849.                     int o;

  850.                     int numOfNoises=0;

  851. 

  852.                     for(o=0; options[o]!=NULL; o++){

  853.                         char *tail;

  854.                         ppMode->maxTmpNoise[numOfNoises]=

  855.                             strtol(options[o], &tail, 0);

  856.                         if(tail!=options[o]){

  857.                             numOfNoises++;

  858.                             numOfUnknownOptions--;

  859.                             if(numOfNoises >= 3) break;

  860.                         }

  861.                     }

  862.                 }

  863.                 else if(filters[i].mask == V_DEBLOCK   || filters[i].mask == H_DEBLOCK

  864.                      || filters[i].mask == V_A_DEBLOCK || filters[i].mask == H_A_DEBLOCK){

  865.                     int o;

  866. 

  867.                     for(o=0; options[o]!=NULL && o<2; o++){

  868.                         char *tail;

  869.                         int val= strtol(options[o], &tail, 0);

  870.                         if(tail==options[o]) break;

  871. 

  872.                         numOfUnknownOptions--;

  873.                         if(o==0) ppMode->baseDcDiff= val;

  874.                         else ppMode->flatnessThreshold= val;

  875.                     }

  876.                 }

  877.                 else if(filters[i].mask == FORCE_QUANT){

  878.                     int o;

  879.                     ppMode->forcedQuant= 15;

  880. 

  881.                     for(o=0; options[o]!=NULL && o<1; o++){

  882.                         char *tail;

  883.                         int val= strtol(options[o], &tail, 0);

  884.                         if(tail==options[o]) break;

  885. 

  886.                         numOfUnknownOptions--;

  887.                         ppMode->forcedQuant= val;

  888.                     }

  889.                 }

  890.             }

  891.         }

  892.         if(!filterNameOk) ppMode->error++;

  893.         ppMode->error += numOfUnknownOptions;

  894.     }

  895. 

  896.     av_log(NULL, AV_LOG_DEBUG, "pp: lumMode=%X, chromMode=%X\n", ppMode->lumMode, ppMode->chromMode);

  897.     if(ppMode->error){

  898.         av_log(NULL, AV_LOG_ERROR, "%d errors in postprocess string \"%s\"\n", ppMode->error, name);

  899.         av_free(ppMode);

  900.         return NULL;

  901.     }

  902.     return ppMode;

  903. }

  904. 

  905. void pp_free_mode(pp_mode *mode){

  906.     av_free(mode);

  907. }

  908. 

  909. static void reallocAlign(void **p, int alignment, int size){

  910.     av_free(*p);

  911.     *p= av_mallocz(size);

  912. }

  913. 

  914. static void reallocBuffers(PPContext *c, int width, int height, int stride, int qpStride){

  915.     int mbWidth = (width+15)>>4;

  916.     int mbHeight= (height+15)>>4;

  917.     int i;

  918. 

  919.     c->stride= stride;

  920.     c->qpStride= qpStride;

  921. 

  922.     reallocAlign((void **)&c->tempDst, 8, stride*24);

  923.     reallocAlign((void **)&c->tempSrc, 8, stride*24);

  924.     reallocAlign((void **)&c->tempBlocks, 8, 2*16*8);

  925.     reallocAlign((void **)&c->yHistogram, 8, 256*sizeof(uint64_t));

  926.     for(i=0; i<256; i++)

  927.             c->yHistogram[i]= width*height/64*15/256;

  928. 

  929.     for(i=0; i<3; i++){

  930.         //Note: The +17*1024 is just there so i do not have to worry about r/w over the end.

  931.         reallocAlign((void **)&c->tempBlurred[i], 8, stride*mbHeight*16 + 17*1024);

  932.         reallocAlign((void **)&c->tempBlurredPast[i], 8, 256*((height+7)&(~7))/2 + 17*1024);//FIXME size

  933.     }

  934. 

  935.     reallocAlign((void **)&c->deintTemp, 8, 2*width+32);

  936.     reallocAlign((void **)&c->nonBQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T));

  937.     reallocAlign((void **)&c->stdQPTable, 8, qpStride*mbHeight*sizeof(QP_STORE_T));

  938.     reallocAlign((void **)&c->forcedQPTable, 8, mbWidth*sizeof(QP_STORE_T));

  939. }

  940. 

  941. static const char * context_to_name(void * ptr) {

  942.     return "postproc";

  943. }

  944. 

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

  946. 

  947. pp_context *pp_get_context(int width, int height, int cpuCaps){

  948.     PPContext *c= av_malloc(sizeof(PPContext));

  949.     int stride= (width+15)&(~15);    //assumed / will realloc if needed

  950.     int qpStride= (width+15)/16 + 2; //assumed / will realloc if needed

  951. 

  952.     memset(c, 0, sizeof(PPContext));

  953.     c->av_class = &av_codec_context_class;

  954.     c->cpuCaps= cpuCaps;

  955.     if(cpuCaps&PP_FORMAT){

  956.         c->hChromaSubSample= cpuCaps&0x3;

  957.         c->vChromaSubSample= (cpuCaps>>4)&0x3;

  958.     }else{

  959.         c->hChromaSubSample= 1;

  960.         c->vChromaSubSample= 1;

  961.     }

  962. 

  963.     reallocBuffers(c, width, height, stride, qpStride);

  964. 

  965.     c->frameNum=-1;

  966. 

  967.     return c;

  968. }

  969. 

  970. void pp_free_context(void *vc){

  971.     PPContext *c = (PPContext*)vc;

  972.     int i;

  973. 

  974.     for(i=0; i<3; i++) av_free(c->tempBlurred[i]);

  975.     for(i=0; i<3; i++) av_free(c->tempBlurredPast[i]);

  976. 

  977.     av_free(c->tempBlocks);

  978.     av_free(c->yHistogram);

  979.     av_free(c->tempDst);

  980.     av_free(c->tempSrc);

  981.     av_free(c->deintTemp);

  982.     av_free(c->stdQPTable);

  983.     av_free(c->nonBQPTable);

  984.     av_free(c->forcedQPTable);

  985. 

  986.     memset(c, 0, sizeof(PPContext));

  987. 

  988.     av_free(c);

  989. }

  990. 

  991. void  pp_postprocess(const uint8_t * src[3], const int srcStride[3],

  992.                      uint8_t * dst[3], const int dstStride[3],

  993.                      int width, int height,

  994.                      const QP_STORE_T *QP_store,  int QPStride,

  995.                      pp_mode *vm,  void *vc, int pict_type)

  996. {

  997.     int mbWidth = (width+15)>>4;

  998.     int mbHeight= (height+15)>>4;

  999.     PPMode *mode = (PPMode*)vm;

  1000.     PPContext *c = (PPContext*)vc;

  1001.     int minStride= FFMAX(FFABS(srcStride[0]), FFABS(dstStride[0]));

  1002.     int absQPStride = FFABS(QPStride);

  1003. 

  1004.     // c->stride and c->QPStride are always positive

  1005.     if(c->stride < minStride || c->qpStride < absQPStride)

  1006.         reallocBuffers(c, width, height,

  1007.                        FFMAX(minStride, c->stride),

  1008.                        FFMAX(c->qpStride, absQPStride));

  1009. 

  1010.     if(QP_store==NULL || (mode->lumMode & FORCE_QUANT)){

  1011.         int i;

  1012.         QP_store= c->forcedQPTable;

  1013.         absQPStride = QPStride = 0;

  1014.         if(mode->lumMode & FORCE_QUANT)

  1015.             for(i=0; i<mbWidth; i++) c->forcedQPTable[i]= mode->forcedQuant;

  1016.         else

  1017.             for(i=0; i<mbWidth; i++) c->forcedQPTable[i]= 1;

  1018.     }

  1019. 

  1020.     if(pict_type & PP_PICT_TYPE_QP2){

  1021.         int i;

  1022.         const int count= mbHeight * absQPStride;

  1023.         for(i=0; i<(count>>2); i++){

  1024.             ((uint32_t*)c->stdQPTable)[i] = (((const uint32_t*)QP_store)[i]>>1) & 0x7F7F7F7F;

  1025.         }

  1026.         for(i<<=2; i<count; i++){

  1027.             c->stdQPTable[i] = QP_store[i]>>1;

  1028.         }

  1029.         QP_store= c->stdQPTable;

  1030.         QPStride= absQPStride;

  1031.     }

  1032. 

  1033.     if(0){

  1034.         int x,y;

  1035.         for(y=0; y<mbHeight; y++){

  1036.             for(x=0; x<mbWidth; x++){

  1037.                 av_log(c, AV_LOG_INFO, "%2d ", QP_store[x + y*QPStride]);

  1038.             }

  1039.             av_log(c, AV_LOG_INFO, "\n");

  1040.         }

  1041.         av_log(c, AV_LOG_INFO, "\n");

  1042.     }

  1043. 

  1044.     if((pict_type&7)!=3){

  1045.         if (QPStride >= 0){

  1046.             int i;

  1047.             const int count= mbHeight * QPStride;

  1048.             for(i=0; i<(count>>2); i++){

  1049.                 ((uint32_t*)c->nonBQPTable)[i] = ((const uint32_t*)QP_store)[i] & 0x3F3F3F3F;

  1050.             }

  1051.             for(i<<=2; i<count; i++){

  1052.                 c->nonBQPTable[i] = QP_store[i] & 0x3F;

  1053.             }

  1054.         } else {

  1055.             int i,j;

  1056.             for(i=0; i<mbHeight; i++) {

  1057.                 for(j=0; j<absQPStride; j++) {

  1058.                     c->nonBQPTable[i*absQPStride+j] = QP_store[i*QPStride+j] & 0x3F;

  1059.                 }

  1060.             }

  1061.         }

  1062.     }

  1063. 

  1064.     av_log(c, AV_LOG_DEBUG, "using npp filters 0x%X/0x%X\n",

  1065.            mode->lumMode, mode->chromMode);

  1066. 

  1067.     postProcess(src[0], srcStride[0], dst[0], dstStride[0],

  1068.                 width, height, QP_store, QPStride, 0, mode, c);

  1069. 

  1070.     width  = (width )>>c->hChromaSubSample;

  1071.     height = (height)>>c->vChromaSubSample;

  1072. 

  1073.     if(mode->chromMode){

  1074.         postProcess(src[1], srcStride[1], dst[1], dstStride[1],

  1075.                     width, height, QP_store, QPStride, 1, mode, c);

  1076.         postProcess(src[2], srcStride[2], dst[2], dstStride[2],

  1077.                     width, height, QP_store, QPStride, 2, mode, c);

  1078.     }

  1079.     else if(srcStride[1] == dstStride[1] && srcStride[2] == dstStride[2]){

  1080.         linecpy(dst[1], src[1], height, srcStride[1]);

  1081.         linecpy(dst[2], src[2], height, srcStride[2]);

  1082.     }else{

  1083.         int y;

  1084.         for(y=0; y<height; y++){

  1085.             memcpy(&(dst[1][y*dstStride[1]]), &(src[1][y*srcStride[1]]), width);

  1086.             memcpy(&(dst[2][y*dstStride[2]]), &(src[2][y*srcStride[2]]), width);

  1087.         }

  1088.     }

  1089. }