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.
67776 Commits
32 Branches
702MB
Tree: 2ce4f01d59
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '2ce4f01d59'
${ noResults }
FFmpeg/libpostproc/postprocess.c

1054 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

  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. 

  93. #include "libavutil/ffversion.h"

  94. const char postproc_ffversion[] = "FFmpeg version " FFMPEG_VERSION;

  95. 

  96. unsigned postproc_version(void)

  97. {

  98.     av_assert0(LIBPOSTPROC_VERSION_MICRO >= 100);

  99.     return LIBPOSTPROC_VERSION_INT;

  100. }

  101. 

  102. const char *postproc_configuration(void)

  103. {

  104.     return FFMPEG_CONFIGURATION;

  105. }

  106. 

  107. const char *postproc_license(void)

  108. {

  109. #define LICENSE_PREFIX "libpostproc license: "

  110.     return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1;

  111. }

  112. 

  113. #if HAVE_ALTIVEC_H

  114. #include <altivec.h>

  115. #endif

  116. 

  117. #define GET_MODE_BUFFER_SIZE 500

  118. #define OPTIONS_ARRAY_SIZE 10

  119. #define BLOCK_SIZE 8

  120. #define TEMP_STRIDE 8

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

  122. 

  123. #if ARCH_X86 && HAVE_INLINE_ASM

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

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

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

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

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

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

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

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

  132. #endif

  133. 

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

  135. 

  136. 

  137. static const struct PPFilter filters[]=

  138. {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  157.     {"be", "bitexact",              1, 0, 0, BITEXACT},

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

  159. };

  160. 

  161. static const char * const replaceTable[]=

  162. {

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

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

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

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

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

  168.     NULL //End Marker

  169. };

  170. 

  171. 

  172. #if ARCH_X86 && HAVE_INLINE_ASM

  173. static inline void prefetchnta(void *p)

  174. {

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

  176.         : : "r" (p)

  177.     );

  178. }

  179. 

  180. static inline void prefetcht0(void *p)

  181. {

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

  183.         : : "r" (p)

  184.     );

  185. }

  186. 

  187. static inline void prefetcht1(void *p)

  188. {

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

  190.         : : "r" (p)

  191.     );

  192. }

  193. 

  194. static inline void prefetcht2(void *p)

  195. {

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

  197.         : : "r" (p)

  198.     );

  199. }

  200. #endif

  201. 

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

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

  204. 

  205. /**

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

  207.  */

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

  209. {

  210.     int numEq= 0;

  211.     int y;

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

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

  214. 

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

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

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

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

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

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

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

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

  223.         src+= stride;

  224.     }

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

  226. }

  227. 

  228. /**

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

  230.  */

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

  232. {

  233.     int numEq= 0;

  234.     int y;

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

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

  237. 

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

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

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

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

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

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

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

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

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

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

  248.         src+= stride;

  249.     }

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

  251. }

  252. 

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

  254. {

  255.     int i;

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

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

  258.         src += stride;

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

  260.         src += stride;

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

  262.         src += stride;

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

  264.         src += stride;

  265.     }

  266.     return 1;

  267. }

  268. 

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

  270. {

  271.     int x;

  272.     src+= stride*4;

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

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

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

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

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

  278.     }

  279.     return 1;

  280. }

  281. 

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

  283. {

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

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

  286.             return 1;

  287.         else

  288.             return 0;

  289.     }else{

  290.         return 2;

  291.     }

  292. }

  293. 

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

  295. {

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

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

  298.             return 1;

  299.         else

  300.             return 0;

  301.     }else{

  302.         return 2;

  303.     }

  304. }

  305. 

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

  307. {

  308.     int y;

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

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

  311. 

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

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

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

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

  316. 

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

  318.             d= FFMAX(d, 0);

  319. 

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

  321.             d*= FFSIGN(-middleEnergy);

  322. 

  323.             if(q>0)

  324.             {

  325.                 d = FFMAX(d, 0);

  326.                 d = FFMIN(d, q);

  327.             }

  328.             else

  329.             {

  330.                 d = FFMIN(d, 0);

  331.                 d = FFMAX(d, q);

  332.             }

  333. 

  334.             dst[3]-= d;

  335.             dst[4]+= d;

  336.         }

  337.         dst+= stride;

  338.     }

  339. }

  340. 

  341. /**

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

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

  344.  */

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

  346. {

  347.     int y;

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

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

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

  351. 

  352.         int sums[10];

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

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

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

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

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

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

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

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

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

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

  363. 

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

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

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

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

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

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

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

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

  372. 

  373.         dst+= stride;

  374.     }

  375. }

  376. 

  377. /**

  378.  * Experimental Filter 1 (Horizontal)

  379.  * will not damage linear gradients

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

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

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

  383.  * not identical with the vertical one

  384.  */

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

  386. {

  387.     int y;

  388.     static uint64_t lut[256];

  389.     if(!lut[255])

  390.     {

  391.         int i;

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

  393.         {

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

  395. /*

  396. //Simulate 112242211 9-Tap filter

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

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

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

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

  401. */

  402. //Simulate piecewise linear interpolation

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

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

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

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

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

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

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

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

  411. 

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

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

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

  415.         }

  416.     }

  417. 

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

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

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

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

  422. 

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

  424. 

  425.         if(d < QP){

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

  427. 

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

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

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

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

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

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

  434.         }

  435.         src+=stride;

  436.     }

  437. }

  438. 

  439. /**

  440.  * accurate deblock filter

  441.  */

  442. static av_always_inline void do_a_deblock_C(uint8_t *src, int step,

  443.                                             int stride, const PPContext *c)

  444. {

  445.     int y;

  446.     const int QP= c->QP;

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

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

  449. //START_TIMER

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

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

  452.         int numEq= 0;

  453. 

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

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

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

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

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

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

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

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

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

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

  464.             int min, max, x;

  465. 

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

  467.                 max= src[0];

  468.                 min= src[step];

  469.             }else{

  470.                 max= src[step];

  471.                 min= src[0];

  472.             }

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

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

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

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

  477.                 }else{

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

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

  480.                 }

  481.             }

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

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

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

  485. 

  486.                 int sums[10];

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

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

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

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

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

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

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

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

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

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

  497. 

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

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

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

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

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

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

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

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

  506.             }

  507.         }else{

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

  509. 

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

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

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

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

  514. 

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

  516.                 d= FFMAX(d, 0);

  517. 

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

  519.                 d*= FFSIGN(-middleEnergy);

  520. 

  521.                 if(q>0){

  522.                     d = FFMAX(d, 0);

  523.                     d = FFMIN(d, q);

  524.                 }else{

  525.                     d = FFMIN(d, 0);

  526.                     d = FFMAX(d, q);

  527.                 }

  528. 

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

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

  531.             }

  532.         }

  533. 

  534.         src += stride;

  535.     }

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

  537.     STOP_TIMER("step16")

  538. }else{

  539.     STOP_TIMER("stepX")

  540. }*/

  541. }

  542. 

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

  544. //Plain C versions

  545. //we always compile C for testing which needs bitexactness

  546. #define TEMPLATE_PP_C 1

  547. #include "postprocess_template.c"

  548. 

  549. #if HAVE_ALTIVEC

  550. #   define TEMPLATE_PP_ALTIVEC 1

  551. #   include "postprocess_altivec_template.c"

  552. #   include "postprocess_template.c"

  553. #endif

  554. 

  555. #if ARCH_X86 && HAVE_INLINE_ASM

  556. #    if CONFIG_RUNTIME_CPUDETECT

  557. #        define TEMPLATE_PP_MMX 1

  558. #        include "postprocess_template.c"

  559. #        define TEMPLATE_PP_MMXEXT 1

  560. #        include "postprocess_template.c"

  561. #        define TEMPLATE_PP_3DNOW 1

  562. #        include "postprocess_template.c"

  563. #        define TEMPLATE_PP_SSE2 1

  564. #        include "postprocess_template.c"

  565. #    else

  566. #        if HAVE_SSE2_INLINE

  567. #            define TEMPLATE_PP_SSE2 1

  568. #            include "postprocess_template.c"

  569. #        elif HAVE_MMXEXT_INLINE

  570. #            define TEMPLATE_PP_MMXEXT 1

  571. #            include "postprocess_template.c"

  572. #        elif HAVE_AMD3DNOW_INLINE

  573. #            define TEMPLATE_PP_3DNOW 1

  574. #            include "postprocess_template.c"

  575. #        elif HAVE_MMX_INLINE

  576. #            define TEMPLATE_PP_MMX 1

  577. #            include "postprocess_template.c"

  578. #        endif

  579. #    endif

  580. #endif

  581. 

  582. typedef void (*pp_fn)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,

  583.                       const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2);

  584. 

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

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

  587. {

  588.     pp_fn pp = postProcess_C;

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

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

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

  592. 

  593.     if (!(ppMode->lumMode & BITEXACT)) {

  594. #if CONFIG_RUNTIME_CPUDETECT

  595. #if ARCH_X86 && HAVE_INLINE_ASM

  596.         // ordered per speed fastest first

  597.         if      (c->cpuCaps & AV_CPU_FLAG_SSE2)     pp = postProcess_SSE2;

  598.         else if (c->cpuCaps & AV_CPU_FLAG_MMXEXT)   pp = postProcess_MMX2;

  599.         else if (c->cpuCaps & AV_CPU_FLAG_3DNOW)    pp = postProcess_3DNow;

  600.         else if (c->cpuCaps & AV_CPU_FLAG_MMX)      pp = postProcess_MMX;

  601. #elif HAVE_ALTIVEC

  602.         if      (c->cpuCaps & AV_CPU_FLAG_ALTIVEC)  pp = postProcess_altivec;

  603. #endif

  604. #else /* CONFIG_RUNTIME_CPUDETECT */

  605. #if     HAVE_SSE2_INLINE

  606.         pp = postProcess_SSE2;

  607. #elif   HAVE_MMXEXT_INLINE

  608.         pp = postProcess_MMX2;

  609. #elif HAVE_AMD3DNOW_INLINE

  610.         pp = postProcess_3DNow;

  611. #elif HAVE_MMX_INLINE

  612.         pp = postProcess_MMX;

  613. #elif HAVE_ALTIVEC

  614.         pp = postProcess_altivec;

  615. #endif

  616. #endif /* !CONFIG_RUNTIME_CPUDETECT */

  617.     }

  618. 

  619.     pp(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);

  620. }

  621. 

  622. /* -pp Command line Help

  623. */

  624. const char pp_help[] =

  625. "Available postprocessing filters:\n"

  626. "Filters                        Options\n"

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

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

  629. "                       c       chrom           chrominance filtering enabled\n"

  630. "                       y       nochrom         chrominance filtering disabled\n"

  631. "                       n       noluma          luma filtering disabled\n"

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

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

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

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

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

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

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

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

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

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

  642. "dr     dering                                  deringing filter\n"

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

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

  645. "lb     linblenddeint                           linear blend deinterlacer\n"

  646. "li     linipoldeint                            linear interpolating deinterlace\n"

  647. "ci     cubicipoldeint                          cubic interpolating deinterlacer\n"

  648. "md     mediandeint                             median deinterlacer\n"

  649. "fd     ffmpegdeint                             ffmpeg deinterlacer\n"

  650. "l5     lowpass5                                FIR lowpass deinterlacer\n"

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

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

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

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

  655. "                     1. <= 2. <= 3.            larger -> stronger filtering\n"

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

  657. "Usage:\n"

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

  659. "long form example:\n"

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

  661. "short form example:\n"

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

  663. "more examples:\n"

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

  665. "\n"

  666. ;

  667. 

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

  669. {

  670.     char temp[GET_MODE_BUFFER_SIZE];

  671.     char *p= temp;

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

  673.     static const char optionDelimiters[] = ":|";

  674.     struct PPMode *ppMode;

  675.     char *filterToken;

  676. 

  677.     if (!name)  {

  678.         av_log(NULL, AV_LOG_ERROR, "pp: Missing argument\n");

  679.         return NULL;

  680.     }

  681. 

  682.     if (!strcmp(name, "help")) {

  683.         const char *p;

  684.         for (p = pp_help; strchr(p, '\n'); p = strchr(p, '\n') + 1) {

  685.             av_strlcpy(temp, p, FFMIN(sizeof(temp), strchr(p, '\n') - p + 2));

  686.             av_log(NULL, AV_LOG_INFO, "%s", temp);

  687.         }

  688.         return NULL;

  689.     }

  690. 

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

  692. 

  693.     ppMode->lumMode= 0;

  694.     ppMode->chromMode= 0;

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

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

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

  698.     ppMode->maxAllowedY= 234;

  699.     ppMode->minAllowedY= 16;

  700.     ppMode->baseDcDiff= 256/8;

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

  702.     ppMode->maxClippedThreshold= 0.01;

  703.     ppMode->error=0;

  704. 

  705.     memset(temp, 0, GET_MODE_BUFFER_SIZE);

  706.     av_strlcpy(temp, name, GET_MODE_BUFFER_SIZE - 1);

  707. 

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

  709. 

  710.     for(;;){

  711.         char *filterName;

  712.         int q= 1000000; //PP_QUALITY_MAX;

  713.         int chrom=-1;

  714.         int luma=-1;

  715.         char *option;

  716.         char *options[OPTIONS_ARRAY_SIZE];

  717.         int i;

  718.         int filterNameOk=0;

  719.         int numOfUnknownOptions=0;

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

  721. 

  722.         filterToken= strtok(p, filterDelimiters);

  723.         if(!filterToken) break;

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

  725.         filterName= strtok(filterToken, optionDelimiters);

  726.         if (!filterName) {

  727.             ppMode->error++;

  728.             break;

  729.         }

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

  731. 

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

  733.             enable=0;

  734.             filterName++;

  735.         }

  736. 

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

  738.             option= strtok(NULL, optionDelimiters);

  739.             if(!option) break;

  740. 

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

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

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

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

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

  746.             else{

  747.                 options[numOfUnknownOptions] = option;

  748.                 numOfUnknownOptions++;

  749.             }

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

  751.         }

  752.         options[numOfUnknownOptions] = NULL;

  753. 

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

  755.         for(i=0; replaceTable[2*i]; i++){

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

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

  758.                 int plen;

  759.                 int spaceLeft;

  760. 

  761.                 p--, *p=',';

  762. 

  763.                 plen= strlen(p);

  764.                 spaceLeft= p - temp + plen;

  765.                 if(spaceLeft + newlen  >= GET_MODE_BUFFER_SIZE - 1){

  766.                     ppMode->error++;

  767.                     break;

  768.                 }

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

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

  771.                 filterNameOk=1;

  772.             }

  773.         }

  774. 

  775.         for(i=0; filters[i].shortName; i++){

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

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

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

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

  780. 

  781.                 filterNameOk=1;

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

  783. 

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

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

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

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

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

  789. 

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

  791.                     int o;

  792.                     ppMode->minAllowedY= 16;

  793.                     ppMode->maxAllowedY= 234;

  794.                     for(o=0; options[o]; o++){

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

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

  797.                             ppMode->minAllowedY= 0;

  798.                             ppMode->maxAllowedY= 255;

  799.                             numOfUnknownOptions--;

  800.                         }

  801.                     }

  802.                 }

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

  804.                 {

  805.                     int o;

  806.                     int numOfNoises=0;

  807. 

  808.                     for(o=0; options[o]; o++){

  809.                         char *tail;

  810.                         ppMode->maxTmpNoise[numOfNoises]=

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

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

  813.                             numOfNoises++;

  814.                             numOfUnknownOptions--;

  815.                             if(numOfNoises >= 3) break;

  816.                         }

  817.                     }

  818.                 }

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

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

  821.                     int o;

  822. 

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

  824.                         char *tail;

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

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

  827. 

  828.                         numOfUnknownOptions--;

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

  830.                         else ppMode->flatnessThreshold= val;

  831.                     }

  832.                 }

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

  834.                     int o;

  835.                     ppMode->forcedQuant= 15;

  836. 

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

  838.                         char *tail;

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

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

  841. 

  842.                         numOfUnknownOptions--;

  843.                         ppMode->forcedQuant= val;

  844.                     }

  845.                 }

  846.             }

  847.         }

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

  849.         ppMode->error += numOfUnknownOptions;

  850.     }

  851. 

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

  853.     if(ppMode->error){

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

  855.         av_free(ppMode);

  856.         return NULL;

  857.     }

  858.     return ppMode;

  859. }

  860. 

  861. void pp_free_mode(pp_mode *mode){

  862.     av_free(mode);

  863. }

  864. 

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

  866.     av_free(*p);

  867.     *p= av_mallocz(size);

  868. }

  869. 

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

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

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

  873.     int i;

  874. 

  875.     c->stride= stride;

  876.     c->qpStride= qpStride;

  877. 

  878.     reallocAlign((void **)&c->tempDst, 8, stride*24+32);

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

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

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

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

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

  884. 

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

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

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

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

  889.     }

  890. 

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

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

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

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

  895. }

  896. 

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

  898.     return "postproc";

  899. }

  900. 

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

  902. 

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

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

  905.     int stride= FFALIGN(width, 16);  //assumed / will realloc if needed

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

  907. 

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

  909.     c->av_class = &av_codec_context_class;

  910.     if(cpuCaps&PP_FORMAT){

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

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

  913.     }else{

  914.         c->hChromaSubSample= 1;

  915.         c->vChromaSubSample= 1;

  916.     }

  917.     if (cpuCaps & PP_CPU_CAPS_AUTO) {

  918.         c->cpuCaps = av_get_cpu_flags();

  919.     } else {

  920.         c->cpuCaps = 0;

  921.         if (cpuCaps & PP_CPU_CAPS_MMX)      c->cpuCaps |= AV_CPU_FLAG_MMX;

  922.         if (cpuCaps & PP_CPU_CAPS_MMX2)     c->cpuCaps |= AV_CPU_FLAG_MMXEXT;

  923.         if (cpuCaps & PP_CPU_CAPS_3DNOW)    c->cpuCaps |= AV_CPU_FLAG_3DNOW;

  924.         if (cpuCaps & PP_CPU_CAPS_ALTIVEC)  c->cpuCaps |= AV_CPU_FLAG_ALTIVEC;

  925.     }

  926. 

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

  928. 

  929.     c->frameNum=-1;

  930. 

  931.     return c;

  932. }

  933. 

  934. void pp_free_context(void *vc){

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

  936.     int i;

  937. 

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

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

  940. 

  941.     av_free(c->tempBlocks);

  942.     av_free(c->yHistogram);

  943.     av_free(c->tempDst);

  944.     av_free(c->tempSrc);

  945.     av_free(c->deintTemp);

  946.     av_free(c->stdQPTable);

  947.     av_free(c->nonBQPTable);

  948.     av_free(c->forcedQPTable);

  949. 

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

  951. 

  952.     av_free(c);

  953. }

  954. 

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

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

  957.                      int width, int height,

  958.                      const QP_STORE_T *QP_store,  int QPStride,

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

  960. {

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

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

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

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

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

  966.     int absQPStride = FFABS(QPStride);

  967. 

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

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

  970.         reallocBuffers(c, width, height,

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

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

  973. 

  974.     if(!QP_store || (mode->lumMode & FORCE_QUANT)){

  975.         int i;

  976.         QP_store= c->forcedQPTable;

  977.         absQPStride = QPStride = 0;

  978.         if(mode->lumMode & FORCE_QUANT)

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

  980.         else

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

  982.     }

  983. 

  984.     if(pict_type & PP_PICT_TYPE_QP2){

  985.         int i;

  986.         const int count= FFMAX(mbHeight * absQPStride, mbWidth);

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

  988.             AV_WN32(c->stdQPTable + (i<<2), AV_RN32(QP_store + (i<<2)) >> 1 & 0x7F7F7F7F);

  989.         }

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

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

  992.         }

  993.         QP_store= c->stdQPTable;

  994.         QPStride= absQPStride;

  995.     }

  996. 

  997.     if(0){

  998.         int x,y;

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

  1000.             for(x=0; x<mbWidth; x++){

  1001.                 av_log(c, AV_LOG_INFO, "%2d ", QP_store[x + y*QPStride]);

  1002.             }

  1003.             av_log(c, AV_LOG_INFO, "\n");

  1004.         }

  1005.         av_log(c, AV_LOG_INFO, "\n");

  1006.     }

  1007. 

  1008.     if((pict_type&7)!=3){

  1009.         if (QPStride >= 0){

  1010.             int i;

  1011.             const int count= FFMAX(mbHeight * QPStride, mbWidth);

  1012.             for(i=0; i<(count>>2); i++){

  1013.                 AV_WN32(c->nonBQPTable + (i<<2), AV_RN32(QP_store + (i<<2)) & 0x3F3F3F3F);

  1014.             }

  1015.             for(i<<=2; i<count; i++){

  1016.                 c->nonBQPTable[i] = QP_store[i] & 0x3F;

  1017.             }

  1018.         } else {

  1019.             int i,j;

  1020.             for(i=0; i<mbHeight; i++) {

  1021.                 for(j=0; j<absQPStride; j++) {

  1022.                     c->nonBQPTable[i*absQPStride+j] = QP_store[i*QPStride+j] & 0x3F;

  1023.                 }

  1024.             }

  1025.         }

  1026.     }

  1027. 

  1028.     av_log(c, AV_LOG_DEBUG, "using npp filters 0x%X/0x%X\n",

  1029.            mode->lumMode, mode->chromMode);

  1030. 

  1031.     postProcess(src[0], srcStride[0], dst[0], dstStride[0],

  1032.                 width, height, QP_store, QPStride, 0, mode, c);

  1033. 

  1034.     width  = (width )>>c->hChromaSubSample;

  1035.     height = (height)>>c->vChromaSubSample;

  1036. 

  1037.     if(mode->chromMode){

  1038.         postProcess(src[1], srcStride[1], dst[1], dstStride[1],

  1039.                     width, height, QP_store, QPStride, 1, mode, c);

  1040.         postProcess(src[2], srcStride[2], dst[2], dstStride[2],

  1041.                     width, height, QP_store, QPStride, 2, mode, c);

  1042.     }

  1043.     else if(srcStride[1] == dstStride[1] && srcStride[2] == dstStride[2]){

  1044.         linecpy(dst[1], src[1], height, srcStride[1]);

  1045.         linecpy(dst[2], src[2], height, srcStride[2]);

  1046.     }else{

  1047.         int y;

  1048.         for(y=0; y<height; y++){

  1049.             memcpy(&(dst[1][y*dstStride[1]]), &(src[1][y*srcStride[1]]), width);

  1050.             memcpy(&(dst[2][y*dstStride[2]]), &(src[2][y*srcStride[2]]), width);

  1051.         }

  1052.     }

  1053. }