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.
32742 Commits
32 Branches
2.1GB
Tree: ecb0edd9d7
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'ecb0edd9d7'
${ noResults }
FFmpeg/libpostproc/postprocess_template.c

3628 lines
157KB
Raw Blame History 

  1. /*

  2.  * Copyright (C) 2001-2002 Michael Niedermayer (michaelni@gmx.at)

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

  6.  * FFmpeg is free software; you can redistribute it and/or modify

  7.  * it under the terms of the GNU General Public License as published by

  8.  * the Free Software Foundation; either version 2 of the License, or

  9.  * (at your option) any later version.

  10.  *

  11.  * FFmpeg is distributed in the hope that it will be useful,

  12.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  13.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

  14.  * GNU General Public License for more details.

  15.  *

  16.  * You should have received a copy of the GNU General Public License

  17.  * along with FFmpeg; if not, write to the Free Software

  18.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  19.  */

  20. 

  21. /**

  22.  * @file

  23.  * mmx/mmx2/3dnow postprocess code.

  24.  */

  25. 

  26. #include "libavutil/x86_cpu.h"

  27. 

  28. #undef REAL_PAVGB

  29. #undef PAVGB

  30. #undef PMINUB

  31. #undef PMAXUB

  32. 

  33. #if   HAVE_MMX2

  34. #define REAL_PAVGB(a,b) "pavgb " #a ", " #b " \n\t"

  35. #elif HAVE_AMD3DNOW

  36. #define REAL_PAVGB(a,b) "pavgusb " #a ", " #b " \n\t"

  37. #endif

  38. #define PAVGB(a,b)  REAL_PAVGB(a,b)

  39. 

  40. #if   HAVE_MMX2

  41. #define PMINUB(a,b,t) "pminub " #a ", " #b " \n\t"

  42. #elif HAVE_MMX

  43. #define PMINUB(b,a,t) \

  44.     "movq " #a ", " #t " \n\t"\

  45.     "psubusb " #b ", " #t " \n\t"\

  46.     "psubb " #t ", " #a " \n\t"

  47. #endif

  48. 

  49. #if   HAVE_MMX2

  50. #define PMAXUB(a,b) "pmaxub " #a ", " #b " \n\t"

  51. #elif HAVE_MMX

  52. #define PMAXUB(a,b) \

  53.     "psubusb " #a ", " #b " \n\t"\

  54.     "paddb " #a ", " #b " \n\t"

  55. #endif

  56. 

  57. //FIXME? |255-0| = 1 (should not be a problem ...)

  58. #if HAVE_MMX

  59. /**

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

  61.  */

  62. static inline int RENAME(vertClassify)(uint8_t src[], int stride, PPContext *c){

  63.     int numEq= 0, dcOk;

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

  65.     __asm__ volatile(

  66.         "movq %0, %%mm7                         \n\t"

  67.         "movq %1, %%mm6                         \n\t"

  68.         : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])

  69.         );

  70. 

  71.     __asm__ volatile(

  72.         "lea (%2, %3), %%"REG_a"                \n\t"

  73. //      0       1       2       3       4       5       6       7       8       9

  74. //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2

  75. 

  76.         "movq (%2), %%mm0                       \n\t"

  77.         "movq (%%"REG_a"), %%mm1                \n\t"

  78.         "movq %%mm0, %%mm3                      \n\t"

  79.         "movq %%mm0, %%mm4                      \n\t"

  80.         PMAXUB(%%mm1, %%mm4)

  81.         PMINUB(%%mm1, %%mm3, %%mm5)

  82.         "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece

  83.         "paddb %%mm7, %%mm0                     \n\t"

  84.         "pcmpgtb %%mm6, %%mm0                   \n\t"

  85. 

  86.         "movq (%%"REG_a",%3), %%mm2             \n\t"

  87.         PMAXUB(%%mm2, %%mm4)

  88.         PMINUB(%%mm2, %%mm3, %%mm5)

  89.         "psubb %%mm2, %%mm1                     \n\t"

  90.         "paddb %%mm7, %%mm1                     \n\t"

  91.         "pcmpgtb %%mm6, %%mm1                   \n\t"

  92.         "paddb %%mm1, %%mm0                     \n\t"

  93. 

  94.         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"

  95.         PMAXUB(%%mm1, %%mm4)

  96.         PMINUB(%%mm1, %%mm3, %%mm5)

  97.         "psubb %%mm1, %%mm2                     \n\t"

  98.         "paddb %%mm7, %%mm2                     \n\t"

  99.         "pcmpgtb %%mm6, %%mm2                   \n\t"

  100.         "paddb %%mm2, %%mm0                     \n\t"

  101. 

  102.         "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"

  103. 

  104.         "movq (%2, %3, 4), %%mm2                \n\t"

  105.         PMAXUB(%%mm2, %%mm4)

  106.         PMINUB(%%mm2, %%mm3, %%mm5)

  107.         "psubb %%mm2, %%mm1                     \n\t"

  108.         "paddb %%mm7, %%mm1                     \n\t"

  109.         "pcmpgtb %%mm6, %%mm1                   \n\t"

  110.         "paddb %%mm1, %%mm0                     \n\t"

  111. 

  112.         "movq (%%"REG_a"), %%mm1                \n\t"

  113.         PMAXUB(%%mm1, %%mm4)

  114.         PMINUB(%%mm1, %%mm3, %%mm5)

  115.         "psubb %%mm1, %%mm2                     \n\t"

  116.         "paddb %%mm7, %%mm2                     \n\t"

  117.         "pcmpgtb %%mm6, %%mm2                   \n\t"

  118.         "paddb %%mm2, %%mm0                     \n\t"

  119. 

  120.         "movq (%%"REG_a", %3), %%mm2            \n\t"

  121.         PMAXUB(%%mm2, %%mm4)

  122.         PMINUB(%%mm2, %%mm3, %%mm5)

  123.         "psubb %%mm2, %%mm1                     \n\t"

  124.         "paddb %%mm7, %%mm1                     \n\t"

  125.         "pcmpgtb %%mm6, %%mm1                   \n\t"

  126.         "paddb %%mm1, %%mm0                     \n\t"

  127. 

  128.         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"

  129.         PMAXUB(%%mm1, %%mm4)

  130.         PMINUB(%%mm1, %%mm3, %%mm5)

  131.         "psubb %%mm1, %%mm2                     \n\t"

  132.         "paddb %%mm7, %%mm2                     \n\t"

  133.         "pcmpgtb %%mm6, %%mm2                   \n\t"

  134.         "paddb %%mm2, %%mm0                     \n\t"

  135.         "psubusb %%mm3, %%mm4                   \n\t"

  136. 

  137.         "                                       \n\t"

  138. #if HAVE_MMX2

  139.         "pxor %%mm7, %%mm7                      \n\t"

  140.         "psadbw %%mm7, %%mm0                    \n\t"

  141. #else

  142.         "movq %%mm0, %%mm1                      \n\t"

  143.         "psrlw $8, %%mm0                        \n\t"

  144.         "paddb %%mm1, %%mm0                     \n\t"

  145.         "movq %%mm0, %%mm1                      \n\t"

  146.         "psrlq $16, %%mm0                       \n\t"

  147.         "paddb %%mm1, %%mm0                     \n\t"

  148.         "movq %%mm0, %%mm1                      \n\t"

  149.         "psrlq $32, %%mm0                       \n\t"

  150.         "paddb %%mm1, %%mm0                     \n\t"

  151. #endif

  152.         "movq %4, %%mm7                         \n\t" // QP,..., QP

  153.         "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP

  154.         "psubusb %%mm7, %%mm4                   \n\t" // Diff <= 2QP -> 0

  155.         "packssdw %%mm4, %%mm4                  \n\t"

  156.         "movd %%mm0, %0                         \n\t"

  157.         "movd %%mm4, %1                         \n\t"

  158. 

  159.         : "=r" (numEq), "=r" (dcOk)

  160.         : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)

  161.         : "%"REG_a

  162.         );

  163. 

  164.     numEq= (-numEq) &0xFF;

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

  166.         if(dcOk) return 0;

  167.         else     return 1;

  168.     }else{

  169.         return 2;

  170.     }

  171. }

  172. #endif //HAVE_MMX

  173. 

  174. /**

  175.  * Do a vertical low pass filter on the 8x16 block (only write to the 8x8 block in the middle)

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

  177.  */

  178. #if !HAVE_ALTIVEC

  179. static inline void RENAME(doVertLowPass)(uint8_t *src, int stride, PPContext *c)

  180. {

  181. #if HAVE_MMX2 || HAVE_AMD3DNOW

  182.     src+= stride*3;

  183.     __asm__ volatile(        //"movv %0 %1 %2\n\t"

  184.         "movq %2, %%mm0                         \n\t"  // QP,..., QP

  185.         "pxor %%mm4, %%mm4                      \n\t"

  186. 

  187.         "movq (%0), %%mm6                       \n\t"

  188.         "movq (%0, %1), %%mm5                   \n\t"

  189.         "movq %%mm5, %%mm1                      \n\t"

  190.         "movq %%mm6, %%mm2                      \n\t"

  191.         "psubusb %%mm6, %%mm5                   \n\t"

  192.         "psubusb %%mm1, %%mm2                   \n\t"

  193.         "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines

  194.         "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0

  195.         "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF

  196. 

  197.         "pand %%mm2, %%mm6                      \n\t"

  198.         "pandn %%mm1, %%mm2                     \n\t"

  199.         "por %%mm2, %%mm6                       \n\t"// First Line to Filter

  200. 

  201.         "movq (%0, %1, 8), %%mm5                \n\t"

  202.         "lea (%0, %1, 4), %%"REG_a"             \n\t"

  203.         "lea (%0, %1, 8), %%"REG_c"             \n\t"

  204.         "sub %1, %%"REG_c"                      \n\t"

  205.         "add %1, %0                             \n\t" // %0 points to line 1 not 0

  206.         "movq (%0, %1, 8), %%mm7                \n\t"

  207.         "movq %%mm5, %%mm1                      \n\t"

  208.         "movq %%mm7, %%mm2                      \n\t"

  209.         "psubusb %%mm7, %%mm5                   \n\t"

  210.         "psubusb %%mm1, %%mm2                   \n\t"

  211.         "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines

  212.         "psubusb %%mm0, %%mm2                   \n\t" // diff <= QP -> 0

  213.         "pcmpeqb %%mm4, %%mm2                   \n\t" // diff <= QP -> FF

  214. 

  215.         "pand %%mm2, %%mm7                      \n\t"

  216.         "pandn %%mm1, %%mm2                     \n\t"

  217.         "por %%mm2, %%mm7                       \n\t" // First Line to Filter

  218. 

  219. 

  220.         //      1       2       3       4       5       6       7       8

  221.         //      %0      %0+%1   %0+2%1  eax     %0+4%1  eax+2%1 ecx     eax+4%1

  222.         // 6 4 2 2 1 1

  223.         // 6 4 4 2

  224.         // 6 8 2

  225. 

  226.         "movq (%0, %1), %%mm0                   \n\t" //  1

  227.         "movq %%mm0, %%mm1                      \n\t" //  1

  228.         PAVGB(%%mm6, %%mm0)                           //1 1        /2

  229.         PAVGB(%%mm6, %%mm0)                           //3 1        /4

  230. 

  231.         "movq (%0, %1, 4), %%mm2                \n\t" //     1

  232.         "movq %%mm2, %%mm5                      \n\t" //     1

  233.         PAVGB((%%REGa), %%mm2)                        //    11        /2

  234.         PAVGB((%0, %1, 2), %%mm2)                     //   211        /4

  235.         "movq %%mm2, %%mm3                      \n\t" //   211        /4

  236.         "movq (%0), %%mm4                       \n\t" // 1

  237.         PAVGB(%%mm4, %%mm3)                           // 4 211        /8

  238.         PAVGB(%%mm0, %%mm3)                           //642211        /16

  239.         "movq %%mm3, (%0)                       \n\t" // X

  240.         // mm1=2 mm2=3(211) mm4=1 mm5=5 mm6=0 mm7=9

  241.         "movq %%mm1, %%mm0                      \n\t" //  1

  242.         PAVGB(%%mm6, %%mm0)                           //1 1        /2

  243.         "movq %%mm4, %%mm3                      \n\t" // 1

  244.         PAVGB((%0,%1,2), %%mm3)                       // 1 1        /2

  245.         PAVGB((%%REGa,%1,2), %%mm5)                   //     11        /2

  246.         PAVGB((%%REGa), %%mm5)                        //    211 /4

  247.         PAVGB(%%mm5, %%mm3)                           // 2 2211 /8

  248.         PAVGB(%%mm0, %%mm3)                           //4242211 /16

  249.         "movq %%mm3, (%0,%1)                    \n\t" //  X

  250.         // mm1=2 mm2=3(211) mm4=1 mm5=4(211) mm6=0 mm7=9

  251.         PAVGB(%%mm4, %%mm6)                                   //11        /2

  252.         "movq (%%"REG_c"), %%mm0                \n\t" //       1

  253.         PAVGB((%%REGa, %1, 2), %%mm0)                 //      11/2

  254.         "movq %%mm0, %%mm3                      \n\t" //      11/2

  255.         PAVGB(%%mm1, %%mm0)                           //  2   11/4

  256.         PAVGB(%%mm6, %%mm0)                           //222   11/8

  257.         PAVGB(%%mm2, %%mm0)                           //22242211/16

  258.         "movq (%0, %1, 2), %%mm2                \n\t" //   1

  259.         "movq %%mm0, (%0, %1, 2)                \n\t" //   X

  260.         // mm1=2 mm2=3 mm3=6(11) mm4=1 mm5=4(211) mm6=0(11) mm7=9

  261.         "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1

  262.         PAVGB((%%REGc), %%mm0)                        //       11        /2

  263.         PAVGB(%%mm0, %%mm6)                           //11     11        /4

  264.         PAVGB(%%mm1, %%mm4)                           // 11                /2

  265.         PAVGB(%%mm2, %%mm1)                           //  11                /2

  266.         PAVGB(%%mm1, %%mm6)                           //1122   11        /8

  267.         PAVGB(%%mm5, %%mm6)                           //112242211        /16

  268.         "movq (%%"REG_a"), %%mm5                \n\t" //    1

  269.         "movq %%mm6, (%%"REG_a")                \n\t" //    X

  270.         // mm0=7(11) mm1=2(11) mm2=3 mm3=6(11) mm4=1(11) mm5=4 mm7=9

  271.         "movq (%%"REG_a", %1, 4), %%mm6         \n\t" //        1

  272.         PAVGB(%%mm7, %%mm6)                           //        11        /2

  273.         PAVGB(%%mm4, %%mm6)                           // 11     11        /4

  274.         PAVGB(%%mm3, %%mm6)                           // 11   2211        /8

  275.         PAVGB(%%mm5, %%mm2)                           //   11                /2

  276.         "movq (%0, %1, 4), %%mm4                \n\t" //     1

  277.         PAVGB(%%mm4, %%mm2)                           //   112                /4

  278.         PAVGB(%%mm2, %%mm6)                           // 112242211        /16

  279.         "movq %%mm6, (%0, %1, 4)                \n\t" //     X

  280.         // mm0=7(11) mm1=2(11) mm2=3(112) mm3=6(11) mm4=5 mm5=4 mm7=9

  281.         PAVGB(%%mm7, %%mm1)                           //  11     2        /4

  282.         PAVGB(%%mm4, %%mm5)                           //    11                /2

  283.         PAVGB(%%mm5, %%mm0)                           //    11 11        /4

  284.         "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //      1

  285.         PAVGB(%%mm6, %%mm1)                           //  11  4  2        /8

  286.         PAVGB(%%mm0, %%mm1)                           //  11224222        /16

  287.         "movq %%mm1, (%%"REG_a", %1, 2)         \n\t" //      X

  288.         // mm2=3(112) mm3=6(11) mm4=5 mm5=4(11) mm6=6 mm7=9

  289.         PAVGB((%%REGc), %%mm2)                        //   112 4        /8

  290.         "movq (%%"REG_a", %1, 4), %%mm0         \n\t" //        1

  291.         PAVGB(%%mm0, %%mm6)                           //      1 1        /2

  292.         PAVGB(%%mm7, %%mm6)                           //      1 12        /4

  293.         PAVGB(%%mm2, %%mm6)                           //   1122424        /4

  294.         "movq %%mm6, (%%"REG_c")                \n\t" //       X

  295.         // mm0=8 mm3=6(11) mm4=5 mm5=4(11) mm7=9

  296.         PAVGB(%%mm7, %%mm5)                           //    11   2        /4

  297.         PAVGB(%%mm7, %%mm5)                           //    11   6        /8

  298. 

  299.         PAVGB(%%mm3, %%mm0)                           //      112        /4

  300.         PAVGB(%%mm0, %%mm5)                           //    112246        /16

  301.         "movq %%mm5, (%%"REG_a", %1, 4)         \n\t" //        X

  302.         "sub %1, %0                             \n\t"

  303. 

  304.         :

  305.         : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)

  306.         : "%"REG_a, "%"REG_c

  307.     );

  308. #else //HAVE_MMX2 || HAVE_AMD3DNOW

  309.     const int l1= stride;

  310.     const int l2= stride + l1;

  311.     const int l3= stride + l2;

  312.     const int l4= stride + l3;

  313.     const int l5= stride + l4;

  314.     const int l6= stride + l5;

  315.     const int l7= stride + l6;

  316.     const int l8= stride + l7;

  317.     const int l9= stride + l8;

  318.     int x;

  319.     src+= stride*3;

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

  321.         const int first= FFABS(src[0] - src[l1]) < c->QP ? src[0] : src[l1];

  322.         const int last= FFABS(src[l8] - src[l9]) < c->QP ? src[l9] : src[l8];

  323. 

  324.         int sums[10];

  325.         sums[0] = 4*first + src[l1] + src[l2] + src[l3] + 4;

  326.         sums[1] = sums[0] - first  + src[l4];

  327.         sums[2] = sums[1] - first  + src[l5];

  328.         sums[3] = sums[2] - first  + src[l6];

  329.         sums[4] = sums[3] - first  + src[l7];

  330.         sums[5] = sums[4] - src[l1] + src[l8];

  331.         sums[6] = sums[5] - src[l2] + last;

  332.         sums[7] = sums[6] - src[l3] + last;

  333.         sums[8] = sums[7] - src[l4] + last;

  334.         sums[9] = sums[8] - src[l5] + last;

  335. 

  336.         src[l1]= (sums[0] + sums[2] + 2*src[l1])>>4;

  337.         src[l2]= (sums[1] + sums[3] + 2*src[l2])>>4;

  338.         src[l3]= (sums[2] + sums[4] + 2*src[l3])>>4;

  339.         src[l4]= (sums[3] + sums[5] + 2*src[l4])>>4;

  340.         src[l5]= (sums[4] + sums[6] + 2*src[l5])>>4;

  341.         src[l6]= (sums[5] + sums[7] + 2*src[l6])>>4;

  342.         src[l7]= (sums[6] + sums[8] + 2*src[l7])>>4;

  343.         src[l8]= (sums[7] + sums[9] + 2*src[l8])>>4;

  344. 

  345.         src++;

  346.     }

  347. #endif //HAVE_MMX2 || HAVE_AMD3DNOW

  348. }

  349. #endif //HAVE_ALTIVEC

  350. 

  351. /**

  352.  * Experimental Filter 1

  353.  * will not damage linear gradients

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

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

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

  357.  */

  358. static inline void RENAME(vertX1Filter)(uint8_t *src, int stride, PPContext *co)

  359. {

  360. #if HAVE_MMX2 || HAVE_AMD3DNOW

  361.     src+= stride*3;

  362. 

  363.     __asm__ volatile(

  364.         "pxor %%mm7, %%mm7                      \n\t" // 0

  365.         "lea (%0, %1), %%"REG_a"                \n\t"

  366.         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"

  367. //      0       1       2       3       4       5       6       7       8       9

  368. //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1

  369.         "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3

  370.         "movq (%0, %1, 4), %%mm1                \n\t" // line 4

  371.         "movq %%mm1, %%mm2                      \n\t" // line 4

  372.         "psubusb %%mm0, %%mm1                   \n\t"

  373.         "psubusb %%mm2, %%mm0                   \n\t"

  374.         "por %%mm1, %%mm0                       \n\t" // |l2 - l3|

  375.         "movq (%%"REG_c"), %%mm3                \n\t" // line 5

  376.         "movq (%%"REG_c", %1), %%mm4            \n\t" // line 6

  377.         "movq %%mm3, %%mm5                      \n\t" // line 5

  378.         "psubusb %%mm4, %%mm3                   \n\t"

  379.         "psubusb %%mm5, %%mm4                   \n\t"

  380.         "por %%mm4, %%mm3                       \n\t" // |l5 - l6|

  381.         PAVGB(%%mm3, %%mm0)                           // (|l2 - l3| + |l5 - l6|)/2

  382.         "movq %%mm2, %%mm1                      \n\t" // line 4

  383.         "psubusb %%mm5, %%mm2                   \n\t"

  384.         "movq %%mm2, %%mm4                      \n\t"

  385.         "pcmpeqb %%mm7, %%mm2                   \n\t" // (l4 - l5) <= 0 ? -1 : 0

  386.         "psubusb %%mm1, %%mm5                   \n\t"

  387.         "por %%mm5, %%mm4                       \n\t" // |l4 - l5|

  388.         "psubusb %%mm0, %%mm4                   \n\t" //d = MAX(0, |l4-l5| - (|l2-l3| + |l5-l6|)/2)

  389.         "movq %%mm4, %%mm3                      \n\t" // d

  390.         "movq %2, %%mm0                         \n\t"

  391.         "paddusb %%mm0, %%mm0                   \n\t"

  392.         "psubusb %%mm0, %%mm4                   \n\t"

  393.         "pcmpeqb %%mm7, %%mm4                   \n\t" // d <= QP ? -1 : 0

  394.         "psubusb "MANGLE(b01)", %%mm3           \n\t"

  395.         "pand %%mm4, %%mm3                      \n\t" // d <= QP ? d : 0

  396. 

  397.         PAVGB(%%mm7, %%mm3)                           // d/2

  398.         "movq %%mm3, %%mm1                      \n\t" // d/2

  399.         PAVGB(%%mm7, %%mm3)                           // d/4

  400.         PAVGB(%%mm1, %%mm3)                           // 3*d/8

  401. 

  402.         "movq (%0, %1, 4), %%mm0                \n\t" // line 4

  403.         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4

  404.         "psubusb %%mm3, %%mm0                   \n\t"

  405.         "pxor %%mm2, %%mm0                      \n\t"

  406.         "movq %%mm0, (%0, %1, 4)                \n\t" // line 4

  407. 

  408.         "movq (%%"REG_c"), %%mm0                \n\t" // line 5

  409.         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5

  410.         "paddusb %%mm3, %%mm0                   \n\t"

  411.         "pxor %%mm2, %%mm0                      \n\t"

  412.         "movq %%mm0, (%%"REG_c")                \n\t" // line 5

  413. 

  414.         PAVGB(%%mm7, %%mm1)                           // d/4

  415. 

  416.         "movq (%%"REG_a", %1, 2), %%mm0         \n\t" // line 3

  417.         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l4-1 : l4

  418.         "psubusb %%mm1, %%mm0                   \n\t"

  419.         "pxor %%mm2, %%mm0                      \n\t"

  420.         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t" // line 3

  421. 

  422.         "movq (%%"REG_c", %1), %%mm0            \n\t" // line 6

  423.         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l5-1 : l5

  424.         "paddusb %%mm1, %%mm0                   \n\t"

  425.         "pxor %%mm2, %%mm0                      \n\t"

  426.         "movq %%mm0, (%%"REG_c", %1)            \n\t" // line 6

  427. 

  428.         PAVGB(%%mm7, %%mm1)                           // d/8

  429. 

  430.         "movq (%%"REG_a", %1), %%mm0            \n\t" // line 2

  431.         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l2-1 : l2

  432.         "psubusb %%mm1, %%mm0                   \n\t"

  433.         "pxor %%mm2, %%mm0                      \n\t"

  434.         "movq %%mm0, (%%"REG_a", %1)            \n\t" // line 2

  435. 

  436.         "movq (%%"REG_c", %1, 2), %%mm0         \n\t" // line 7

  437.         "pxor %%mm2, %%mm0                      \n\t" //(l4 - l5) <= 0 ? -l7-1 : l7

  438.         "paddusb %%mm1, %%mm0                   \n\t"

  439.         "pxor %%mm2, %%mm0                      \n\t"

  440.         "movq %%mm0, (%%"REG_c", %1, 2)         \n\t" // line 7

  441. 

  442.         :

  443.         : "r" (src), "r" ((x86_reg)stride), "m" (co->pQPb)

  444.         : "%"REG_a, "%"REG_c

  445.     );

  446. #else //HAVE_MMX2 || HAVE_AMD3DNOW

  447. 

  448.     const int l1= stride;

  449.     const int l2= stride + l1;

  450.     const int l3= stride + l2;

  451.     const int l4= stride + l3;

  452.     const int l5= stride + l4;

  453.     const int l6= stride + l5;

  454.     const int l7= stride + l6;

  455. //    const int l8= stride + l7;

  456. //    const int l9= stride + l8;

  457.     int x;

  458. 

  459.     src+= stride*3;

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

  461.         int a= src[l3] - src[l4];

  462.         int b= src[l4] - src[l5];

  463.         int c= src[l5] - src[l6];

  464. 

  465.         int d= FFABS(b) - ((FFABS(a) + FFABS(c))>>1);

  466.         d= FFMAX(d, 0);

  467. 

  468.         if(d < co->QP*2){

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

  470. 

  471.             src[l2] +=v>>3;

  472.             src[l3] +=v>>2;

  473.             src[l4] +=(3*v)>>3;

  474.             src[l5] -=(3*v)>>3;

  475.             src[l6] -=v>>2;

  476.             src[l7] -=v>>3;

  477.         }

  478.         src++;

  479.     }

  480. #endif //HAVE_MMX2 || HAVE_AMD3DNOW

  481. }

  482. 

  483. #if !HAVE_ALTIVEC

  484. static inline void RENAME(doVertDefFilter)(uint8_t src[], int stride, PPContext *c)

  485. {

  486. #if HAVE_MMX2 || HAVE_AMD3DNOW

  487. /*

  488.     uint8_t tmp[16];

  489.     const int l1= stride;

  490.     const int l2= stride + l1;

  491.     const int l3= stride + l2;

  492.     const int l4= (int)tmp - (int)src - stride*3;

  493.     const int l5= (int)tmp - (int)src - stride*3 + 8;

  494.     const int l6= stride*3 + l3;

  495.     const int l7= stride + l6;

  496.     const int l8= stride + l7;

  497. 

  498.     memcpy(tmp, src+stride*7, 8);

  499.     memcpy(tmp+8, src+stride*8, 8);

  500. */

  501.     src+= stride*4;

  502.     __asm__ volatile(

  503. 

  504. #if 0 //slightly more accurate and slightly slower

  505.         "pxor %%mm7, %%mm7                      \n\t" // 0

  506.         "lea (%0, %1), %%"REG_a"                \n\t"

  507.         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"

  508. //      0       1       2       3       4       5       6       7

  509. //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1

  510. //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1

  511. 

  512. 

  513.         "movq (%0, %1, 2), %%mm0                \n\t" // l2

  514.         "movq (%0), %%mm1                       \n\t" // l0

  515.         "movq %%mm0, %%mm2                      \n\t" // l2

  516.         PAVGB(%%mm7, %%mm0)                           // ~l2/2

  517.         PAVGB(%%mm1, %%mm0)                           // ~(l2 + 2l0)/4

  518.         PAVGB(%%mm2, %%mm0)                           // ~(5l2 + 2l0)/8

  519. 

  520.         "movq (%%"REG_a"), %%mm1                \n\t" // l1

  521.         "movq (%%"REG_a", %1, 2), %%mm3         \n\t" // l3

  522.         "movq %%mm1, %%mm4                      \n\t" // l1

  523.         PAVGB(%%mm7, %%mm1)                           // ~l1/2

  524.         PAVGB(%%mm3, %%mm1)                           // ~(l1 + 2l3)/4

  525.         PAVGB(%%mm4, %%mm1)                           // ~(5l1 + 2l3)/8

  526. 

  527.         "movq %%mm0, %%mm4                      \n\t" // ~(5l2 + 2l0)/8

  528.         "psubusb %%mm1, %%mm0                   \n\t"

  529.         "psubusb %%mm4, %%mm1                   \n\t"

  530.         "por %%mm0, %%mm1                       \n\t" // ~|2l0 - 5l1 + 5l2 - 2l3|/8

  531. // mm1= |lenergy|, mm2= l2, mm3= l3, mm7=0

  532. 

  533.         "movq (%0, %1, 4), %%mm0                \n\t" // l4

  534.         "movq %%mm0, %%mm4                      \n\t" // l4

  535.         PAVGB(%%mm7, %%mm0)                           // ~l4/2

  536.         PAVGB(%%mm2, %%mm0)                           // ~(l4 + 2l2)/4

  537.         PAVGB(%%mm4, %%mm0)                           // ~(5l4 + 2l2)/8

  538. 

  539.         "movq (%%"REG_c"), %%mm2                \n\t" // l5

  540.         "movq %%mm3, %%mm5                      \n\t" // l3

  541.         PAVGB(%%mm7, %%mm3)                           // ~l3/2

  542.         PAVGB(%%mm2, %%mm3)                           // ~(l3 + 2l5)/4

  543.         PAVGB(%%mm5, %%mm3)                           // ~(5l3 + 2l5)/8

  544. 

  545.         "movq %%mm0, %%mm6                      \n\t" // ~(5l4 + 2l2)/8

  546.         "psubusb %%mm3, %%mm0                   \n\t"

  547.         "psubusb %%mm6, %%mm3                   \n\t"

  548.         "por %%mm0, %%mm3                       \n\t" // ~|2l2 - 5l3 + 5l4 - 2l5|/8

  549.         "pcmpeqb %%mm7, %%mm0                   \n\t" // SIGN(2l2 - 5l3 + 5l4 - 2l5)

  550. // mm0= SIGN(menergy), mm1= |lenergy|, mm2= l5, mm3= |menergy|, mm4=l4, mm5= l3, mm7=0

  551. 

  552.         "movq (%%"REG_c", %1), %%mm6            \n\t" // l6

  553.         "movq %%mm6, %%mm5                      \n\t" // l6

  554.         PAVGB(%%mm7, %%mm6)                           // ~l6/2

  555.         PAVGB(%%mm4, %%mm6)                           // ~(l6 + 2l4)/4

  556.         PAVGB(%%mm5, %%mm6)                           // ~(5l6 + 2l4)/8

  557. 

  558.         "movq (%%"REG_c", %1, 2), %%mm5         \n\t" // l7

  559.         "movq %%mm2, %%mm4                      \n\t" // l5

  560.         PAVGB(%%mm7, %%mm2)                           // ~l5/2

  561.         PAVGB(%%mm5, %%mm2)                           // ~(l5 + 2l7)/4

  562.         PAVGB(%%mm4, %%mm2)                           // ~(5l5 + 2l7)/8

  563. 

  564.         "movq %%mm6, %%mm4                      \n\t" // ~(5l6 + 2l4)/8

  565.         "psubusb %%mm2, %%mm6                   \n\t"

  566.         "psubusb %%mm4, %%mm2                   \n\t"

  567.         "por %%mm6, %%mm2                       \n\t" // ~|2l4 - 5l5 + 5l6 - 2l7|/8

  568. // mm0= SIGN(menergy), mm1= |lenergy|/8, mm2= |renergy|/8, mm3= |menergy|/8, mm7=0

  569. 

  570. 

  571.         PMINUB(%%mm2, %%mm1, %%mm4)                   // MIN(|lenergy|,|renergy|)/8

  572.         "movq %2, %%mm4                         \n\t" // QP //FIXME QP+1 ?

  573.         "paddusb "MANGLE(b01)", %%mm4           \n\t"

  574.         "pcmpgtb %%mm3, %%mm4                   \n\t" // |menergy|/8 < QP

  575.         "psubusb %%mm1, %%mm3                   \n\t" // d=|menergy|/8-MIN(|lenergy|,|renergy|)/8

  576.         "pand %%mm4, %%mm3                      \n\t"

  577. 

  578.         "movq %%mm3, %%mm1                      \n\t"

  579. //        "psubusb "MANGLE(b01)", %%mm3           \n\t"

  580.         PAVGB(%%mm7, %%mm3)

  581.         PAVGB(%%mm7, %%mm3)

  582.         "paddusb %%mm1, %%mm3                   \n\t"

  583. //        "paddusb "MANGLE(b01)", %%mm3           \n\t"

  584. 

  585.         "movq (%%"REG_a", %1, 2), %%mm6         \n\t" //l3

  586.         "movq (%0, %1, 4), %%mm5                \n\t" //l4

  587.         "movq (%0, %1, 4), %%mm4                \n\t" //l4

  588.         "psubusb %%mm6, %%mm5                   \n\t"

  589.         "psubusb %%mm4, %%mm6                   \n\t"

  590.         "por %%mm6, %%mm5                       \n\t" // |l3-l4|

  591.         "pcmpeqb %%mm7, %%mm6                   \n\t" // SIGN(l3-l4)

  592.         "pxor %%mm6, %%mm0                      \n\t"

  593.         "pand %%mm0, %%mm3                      \n\t"

  594.         PMINUB(%%mm5, %%mm3, %%mm0)

  595. 

  596.         "psubusb "MANGLE(b01)", %%mm3           \n\t"

  597.         PAVGB(%%mm7, %%mm3)

  598. 

  599.         "movq (%%"REG_a", %1, 2), %%mm0         \n\t"

  600.         "movq (%0, %1, 4), %%mm2                \n\t"

  601.         "pxor %%mm6, %%mm0                      \n\t"

  602.         "pxor %%mm6, %%mm2                      \n\t"

  603.         "psubb %%mm3, %%mm0                     \n\t"

  604.         "paddb %%mm3, %%mm2                     \n\t"

  605.         "pxor %%mm6, %%mm0                      \n\t"

  606.         "pxor %%mm6, %%mm2                      \n\t"

  607.         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"

  608.         "movq %%mm2, (%0, %1, 4)                \n\t"

  609. #endif //0

  610. 

  611.         "lea (%0, %1), %%"REG_a"                \n\t"

  612.         "pcmpeqb %%mm6, %%mm6                   \n\t" // -1

  613. //      0       1       2       3       4       5       6       7

  614. //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1

  615. //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1

  616. 

  617. 

  618.         "movq (%%"REG_a", %1, 2), %%mm1         \n\t" // l3

  619.         "movq (%0, %1, 4), %%mm0                \n\t" // l4

  620.         "pxor %%mm6, %%mm1                      \n\t" // -l3-1

  621.         PAVGB(%%mm1, %%mm0)                           // -q+128 = (l4-l3+256)/2

  622. // mm1=-l3-1, mm0=128-q

  623. 

  624.         "movq (%%"REG_a", %1, 4), %%mm2         \n\t" // l5

  625.         "movq (%%"REG_a", %1), %%mm3            \n\t" // l2

  626.         "pxor %%mm6, %%mm2                      \n\t" // -l5-1

  627.         "movq %%mm2, %%mm5                      \n\t" // -l5-1

  628.         "movq "MANGLE(b80)", %%mm4              \n\t" // 128

  629.         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"

  630.         PAVGB(%%mm3, %%mm2)                           // (l2-l5+256)/2

  631.         PAVGB(%%mm0, %%mm4)                           // ~(l4-l3)/4 + 128

  632.         PAVGB(%%mm2, %%mm4)                           // ~(l2-l5)/4 +(l4-l3)/8 + 128

  633.         PAVGB(%%mm0, %%mm4)                           // ~(l2-l5)/8 +5(l4-l3)/16 + 128

  634. // mm1=-l3-1, mm0=128-q, mm3=l2, mm4=menergy/16 + 128, mm5= -l5-1

  635. 

  636.         "movq (%%"REG_a"), %%mm2                \n\t" // l1

  637.         "pxor %%mm6, %%mm2                      \n\t" // -l1-1

  638.         PAVGB(%%mm3, %%mm2)                           // (l2-l1+256)/2

  639.         PAVGB((%0), %%mm1)                            // (l0-l3+256)/2

  640.         "movq "MANGLE(b80)", %%mm3              \n\t" // 128

  641.         PAVGB(%%mm2, %%mm3)                           // ~(l2-l1)/4 + 128

  642.         PAVGB(%%mm1, %%mm3)                           // ~(l0-l3)/4 +(l2-l1)/8 + 128

  643.         PAVGB(%%mm2, %%mm3)                           // ~(l0-l3)/8 +5(l2-l1)/16 + 128

  644. // mm0=128-q, mm3=lenergy/16 + 128, mm4= menergy/16 + 128, mm5= -l5-1

  645. 

  646.         PAVGB((%%REGc, %1), %%mm5)                    // (l6-l5+256)/2

  647.         "movq (%%"REG_c", %1, 2), %%mm1         \n\t" // l7

  648.         "pxor %%mm6, %%mm1                      \n\t" // -l7-1

  649.         PAVGB((%0, %1, 4), %%mm1)                     // (l4-l7+256)/2

  650.         "movq "MANGLE(b80)", %%mm2              \n\t" // 128

  651.         PAVGB(%%mm5, %%mm2)                           // ~(l6-l5)/4 + 128

  652.         PAVGB(%%mm1, %%mm2)                           // ~(l4-l7)/4 +(l6-l5)/8 + 128

  653.         PAVGB(%%mm5, %%mm2)                           // ~(l4-l7)/8 +5(l6-l5)/16 + 128

  654. // mm0=128-q, mm2=renergy/16 + 128, mm3=lenergy/16 + 128, mm4= menergy/16 + 128

  655. 

  656.         "movq "MANGLE(b00)", %%mm1              \n\t" // 0

  657.         "movq "MANGLE(b00)", %%mm5              \n\t" // 0

  658.         "psubb %%mm2, %%mm1                     \n\t" // 128 - renergy/16

  659.         "psubb %%mm3, %%mm5                     \n\t" // 128 - lenergy/16

  660.         PMAXUB(%%mm1, %%mm2)                          // 128 + |renergy/16|

  661.         PMAXUB(%%mm5, %%mm3)                          // 128 + |lenergy/16|

  662.         PMINUB(%%mm2, %%mm3, %%mm1)                   // 128 + MIN(|lenergy|,|renergy|)/16

  663. 

  664. // mm0=128-q, mm3=128 + MIN(|lenergy|,|renergy|)/16, mm4= menergy/16 + 128

  665. 

  666.         "movq "MANGLE(b00)", %%mm7              \n\t" // 0

  667.         "movq %2, %%mm2                         \n\t" // QP

  668.         PAVGB(%%mm6, %%mm2)                           // 128 + QP/2

  669.         "psubb %%mm6, %%mm2                     \n\t"

  670. 

  671.         "movq %%mm4, %%mm1                      \n\t"

  672.         "pcmpgtb %%mm7, %%mm1                   \n\t" // SIGN(menergy)

  673.         "pxor %%mm1, %%mm4                      \n\t"

  674.         "psubb %%mm1, %%mm4                     \n\t" // 128 + |menergy|/16

  675.         "pcmpgtb %%mm4, %%mm2                   \n\t" // |menergy|/16 < QP/2

  676.         "psubusb %%mm3, %%mm4                   \n\t" //d=|menergy|/16 - MIN(|lenergy|,|renergy|)/16

  677. // mm0=128-q, mm1= SIGN(menergy), mm2= |menergy|/16 < QP/2, mm4= d/16

  678. 

  679.         "movq %%mm4, %%mm3                      \n\t" // d

  680.         "psubusb "MANGLE(b01)", %%mm4           \n\t"

  681.         PAVGB(%%mm7, %%mm4)                           // d/32

  682.         PAVGB(%%mm7, %%mm4)                           // (d + 32)/64

  683.         "paddb %%mm3, %%mm4                     \n\t" // 5d/64

  684.         "pand %%mm2, %%mm4                      \n\t"

  685. 

  686.         "movq "MANGLE(b80)", %%mm5              \n\t" // 128

  687.         "psubb %%mm0, %%mm5                     \n\t" // q

  688.         "paddsb %%mm6, %%mm5                    \n\t" // fix bad rounding

  689.         "pcmpgtb %%mm5, %%mm7                   \n\t" // SIGN(q)

  690.         "pxor %%mm7, %%mm5                      \n\t"

  691. 

  692.         PMINUB(%%mm5, %%mm4, %%mm3)                   // MIN(|q|, 5d/64)

  693.         "pxor %%mm1, %%mm7                      \n\t" // SIGN(d*q)

  694. 

  695.         "pand %%mm7, %%mm4                      \n\t"

  696.         "movq (%%"REG_a", %1, 2), %%mm0         \n\t"

  697.         "movq (%0, %1, 4), %%mm2                \n\t"

  698.         "pxor %%mm1, %%mm0                      \n\t"

  699.         "pxor %%mm1, %%mm2                      \n\t"

  700.         "paddb %%mm4, %%mm0                     \n\t"

  701.         "psubb %%mm4, %%mm2                     \n\t"

  702.         "pxor %%mm1, %%mm0                      \n\t"

  703.         "pxor %%mm1, %%mm2                      \n\t"

  704.         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"

  705.         "movq %%mm2, (%0, %1, 4)                \n\t"

  706. 

  707.         :

  708.         : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb)

  709.         : "%"REG_a, "%"REG_c

  710.     );

  711. 

  712. /*

  713.     {

  714.     int x;

  715.     src-= stride;

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

  717.         const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);

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

  719.             const int q=(src[l4] - src[l5])/2;

  720.             const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);

  721.             const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);

  722. 

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

  724.             d= FFMAX(d, 0);

  725. 

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

  727.             d*= FFSIGN(-middleEnergy);

  728. 

  729.             if(q>0){

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

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

  732.             }else{

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

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

  735.             }

  736. 

  737.             src[l4]-= d;

  738.             src[l5]+= d;

  739.         }

  740.         src++;

  741.     }

  742.     src-=8;

  743.     for(x=0; x<8; x++){

  744.         int y;

  745.         for(y=4; y<6; y++){

  746.             int d= src[x+y*stride] - tmp[x+(y-4)*8];

  747.             int ad= FFABS(d);

  748.             static int max=0;

  749.             static int sum=0;

  750.             static int num=0;

  751.             static int bias=0;

  752. 

  753.             if(max<ad) max=ad;

  754.             sum+= ad>3 ? 1 : 0;

  755.             if(ad>3){

  756.                 src[0] = src[7] = src[stride*7] = src[(stride+1)*7]=255;

  757.             }

  758.             if(y==4) bias+=d;

  759.             num++;

  760.             if(num%1000000 == 0){

  761.                 av_log(c, AV_LOG_INFO, " %d %d %d %d\n", num, sum, max, bias);

  762.             }

  763.         }

  764.     }

  765. }

  766. */

  767. #elif HAVE_MMX

  768.     DECLARE_ALIGNED(8, uint64_t, tmp)[4]; // make space for 4 8-byte vars

  769.     src+= stride*4;

  770.     __asm__ volatile(

  771.         "pxor %%mm7, %%mm7                      \n\t"

  772. //      0       1       2       3       4       5       6       7

  773. //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 edx+%1  edx+2%1

  774. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1

  775. 

  776.         "movq (%0), %%mm0                       \n\t"

  777.         "movq %%mm0, %%mm1                      \n\t"

  778.         "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0

  779.         "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0

  780. 

  781.         "movq (%0, %1), %%mm2                   \n\t"

  782.         "lea (%0, %1, 2), %%"REG_a"             \n\t"

  783.         "movq %%mm2, %%mm3                      \n\t"

  784.         "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1

  785.         "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1

  786. 

  787.         "movq (%%"REG_a"), %%mm4                \n\t"

  788.         "movq %%mm4, %%mm5                      \n\t"

  789.         "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2

  790.         "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2

  791. 

  792.         "paddw %%mm0, %%mm0                     \n\t" // 2L0

  793.         "paddw %%mm1, %%mm1                     \n\t" // 2H0

  794.         "psubw %%mm4, %%mm2                     \n\t" // L1 - L2

  795.         "psubw %%mm5, %%mm3                     \n\t" // H1 - H2

  796.         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2

  797.         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2

  798. 

  799.         "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2

  800.         "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2

  801.         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2

  802.         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2

  803. 

  804.         "movq (%%"REG_a", %1), %%mm2            \n\t"

  805.         "movq %%mm2, %%mm3                      \n\t"

  806.         "punpcklbw %%mm7, %%mm2                 \n\t" // L3

  807.         "punpckhbw %%mm7, %%mm3                 \n\t" // H3

  808. 

  809.         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3

  810.         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3

  811.         "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3

  812.         "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3

  813.         "movq %%mm0, (%3)                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3

  814.         "movq %%mm1, 8(%3)                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3

  815. 

  816.         "movq (%%"REG_a", %1, 2), %%mm0         \n\t"

  817.         "movq %%mm0, %%mm1                      \n\t"

  818.         "punpcklbw %%mm7, %%mm0                 \n\t" // L4

  819.         "punpckhbw %%mm7, %%mm1                 \n\t" // H4

  820. 

  821.         "psubw %%mm0, %%mm2                     \n\t" // L3 - L4

  822.         "psubw %%mm1, %%mm3                     \n\t" // H3 - H4

  823.         "movq %%mm2, 16(%3)                     \n\t" // L3 - L4

  824.         "movq %%mm3, 24(%3)                     \n\t" // H3 - H4

  825.         "paddw %%mm4, %%mm4                     \n\t" // 2L2

  826.         "paddw %%mm5, %%mm5                     \n\t" // 2H2

  827.         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4

  828.         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4

  829. 

  830.         "lea (%%"REG_a", %1), %0                \n\t"

  831.         "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4

  832.         "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4

  833.         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4

  834.         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4

  835. //50 opcodes so far

  836.         "movq (%0, %1, 2), %%mm2                \n\t"

  837.         "movq %%mm2, %%mm3                      \n\t"

  838.         "punpcklbw %%mm7, %%mm2                 \n\t" // L5

  839.         "punpckhbw %%mm7, %%mm3                 \n\t" // H5

  840.         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5

  841.         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5

  842.         "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5

  843.         "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5

  844. 

  845.         "movq (%%"REG_a", %1, 4), %%mm6         \n\t"

  846.         "punpcklbw %%mm7, %%mm6                 \n\t" // L6

  847.         "psubw %%mm6, %%mm2                     \n\t" // L5 - L6

  848.         "movq (%%"REG_a", %1, 4), %%mm6         \n\t"

  849.         "punpckhbw %%mm7, %%mm6                 \n\t" // H6

  850.         "psubw %%mm6, %%mm3                     \n\t" // H5 - H6

  851. 

  852.         "paddw %%mm0, %%mm0                     \n\t" // 2L4

  853.         "paddw %%mm1, %%mm1                     \n\t" // 2H4

  854.         "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6

  855.         "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6

  856. 

  857.         "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6

  858.         "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6

  859.         "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6

  860.         "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6

  861. 

  862.         "movq (%0, %1, 4), %%mm2                \n\t"

  863.         "movq %%mm2, %%mm3                      \n\t"

  864.         "punpcklbw %%mm7, %%mm2                 \n\t" // L7

  865.         "punpckhbw %%mm7, %%mm3                 \n\t" // H7

  866. 

  867.         "paddw %%mm2, %%mm2                     \n\t" // 2L7

  868.         "paddw %%mm3, %%mm3                     \n\t" // 2H7

  869.         "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7

  870.         "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7

  871. 

  872.         "movq (%3), %%mm2                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3

  873.         "movq 8(%3), %%mm3                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3

  874. 

  875. #if HAVE_MMX2

  876.         "movq %%mm7, %%mm6                      \n\t" // 0

  877.         "psubw %%mm0, %%mm6                     \n\t"

  878.         "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|

  879.         "movq %%mm7, %%mm6                      \n\t" // 0

  880.         "psubw %%mm1, %%mm6                     \n\t"

  881.         "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|

  882.         "movq %%mm7, %%mm6                      \n\t" // 0

  883.         "psubw %%mm2, %%mm6                     \n\t"

  884.         "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|

  885.         "movq %%mm7, %%mm6                      \n\t" // 0

  886.         "psubw %%mm3, %%mm6                     \n\t"

  887.         "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|

  888. #else

  889.         "movq %%mm7, %%mm6                      \n\t" // 0

  890.         "pcmpgtw %%mm0, %%mm6                   \n\t"

  891.         "pxor %%mm6, %%mm0                      \n\t"

  892.         "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|

  893.         "movq %%mm7, %%mm6                      \n\t" // 0

  894.         "pcmpgtw %%mm1, %%mm6                   \n\t"

  895.         "pxor %%mm6, %%mm1                      \n\t"

  896.         "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|

  897.         "movq %%mm7, %%mm6                      \n\t" // 0

  898.         "pcmpgtw %%mm2, %%mm6                   \n\t"

  899.         "pxor %%mm6, %%mm2                      \n\t"

  900.         "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|

  901.         "movq %%mm7, %%mm6                      \n\t" // 0

  902.         "pcmpgtw %%mm3, %%mm6                   \n\t"

  903.         "pxor %%mm6, %%mm3                      \n\t"

  904.         "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|

  905. #endif

  906. 

  907. #if HAVE_MMX2

  908.         "pminsw %%mm2, %%mm0                    \n\t"

  909.         "pminsw %%mm3, %%mm1                    \n\t"

  910. #else

  911.         "movq %%mm0, %%mm6                      \n\t"

  912.         "psubusw %%mm2, %%mm6                   \n\t"

  913.         "psubw %%mm6, %%mm0                     \n\t"

  914.         "movq %%mm1, %%mm6                      \n\t"

  915.         "psubusw %%mm3, %%mm6                   \n\t"

  916.         "psubw %%mm6, %%mm1                     \n\t"

  917. #endif

  918. 

  919.         "movd %2, %%mm2                         \n\t" // QP

  920.         "punpcklbw %%mm7, %%mm2                 \n\t"

  921. 

  922.         "movq %%mm7, %%mm6                      \n\t" // 0

  923.         "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)

  924.         "pxor %%mm6, %%mm4                      \n\t"

  925.         "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|

  926.         "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)

  927.         "pxor %%mm7, %%mm5                      \n\t"

  928.         "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|

  929. // 100 opcodes

  930.         "psllw $3, %%mm2                        \n\t" // 8QP

  931.         "movq %%mm2, %%mm3                      \n\t" // 8QP

  932.         "pcmpgtw %%mm4, %%mm2                   \n\t"

  933.         "pcmpgtw %%mm5, %%mm3                   \n\t"

  934.         "pand %%mm2, %%mm4                      \n\t"

  935.         "pand %%mm3, %%mm5                      \n\t"

  936. 

  937. 

  938.         "psubusw %%mm0, %%mm4                   \n\t" // hd

  939.         "psubusw %%mm1, %%mm5                   \n\t" // ld

  940. 

  941. 

  942.         "movq "MANGLE(w05)", %%mm2              \n\t" // 5

  943.         "pmullw %%mm2, %%mm4                    \n\t"

  944.         "pmullw %%mm2, %%mm5                    \n\t"

  945.         "movq "MANGLE(w20)", %%mm2              \n\t" // 32

  946.         "paddw %%mm2, %%mm4                     \n\t"

  947.         "paddw %%mm2, %%mm5                     \n\t"

  948.         "psrlw $6, %%mm4                        \n\t"

  949.         "psrlw $6, %%mm5                        \n\t"

  950. 

  951.         "movq 16(%3), %%mm0                     \n\t" // L3 - L4

  952.         "movq 24(%3), %%mm1                     \n\t" // H3 - H4

  953. 

  954.         "pxor %%mm2, %%mm2                      \n\t"

  955.         "pxor %%mm3, %%mm3                      \n\t"

  956. 

  957.         "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)

  958.         "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)

  959.         "pxor %%mm2, %%mm0                      \n\t"

  960.         "pxor %%mm3, %%mm1                      \n\t"

  961.         "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|

  962.         "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|

  963.         "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2

  964.         "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2

  965. 

  966.         "pxor %%mm6, %%mm2                      \n\t"

  967.         "pxor %%mm7, %%mm3                      \n\t"

  968.         "pand %%mm2, %%mm4                      \n\t"

  969.         "pand %%mm3, %%mm5                      \n\t"

  970. 

  971. #if HAVE_MMX2

  972.         "pminsw %%mm0, %%mm4                    \n\t"

  973.         "pminsw %%mm1, %%mm5                    \n\t"

  974. #else

  975.         "movq %%mm4, %%mm2                      \n\t"

  976.         "psubusw %%mm0, %%mm2                   \n\t"

  977.         "psubw %%mm2, %%mm4                     \n\t"

  978.         "movq %%mm5, %%mm2                      \n\t"

  979.         "psubusw %%mm1, %%mm2                   \n\t"

  980.         "psubw %%mm2, %%mm5                     \n\t"

  981. #endif

  982.         "pxor %%mm6, %%mm4                      \n\t"

  983.         "pxor %%mm7, %%mm5                      \n\t"

  984.         "psubw %%mm6, %%mm4                     \n\t"

  985.         "psubw %%mm7, %%mm5                     \n\t"

  986.         "packsswb %%mm5, %%mm4                  \n\t"

  987.         "movq (%0), %%mm0                       \n\t"

  988.         "paddb   %%mm4, %%mm0                   \n\t"

  989.         "movq %%mm0, (%0)                       \n\t"

  990.         "movq (%0, %1), %%mm0                   \n\t"

  991.         "psubb %%mm4, %%mm0                     \n\t"

  992.         "movq %%mm0, (%0, %1)                   \n\t"

  993. 

  994.         : "+r" (src)

  995.         : "r" ((x86_reg)stride), "m" (c->pQPb), "r"(tmp)

  996.         : "%"REG_a

  997.     );

  998. #else //HAVE_MMX2 || HAVE_AMD3DNOW

  999.     const int l1= stride;

  1000.     const int l2= stride + l1;

  1001.     const int l3= stride + l2;

  1002.     const int l4= stride + l3;

  1003.     const int l5= stride + l4;

  1004.     const int l6= stride + l5;

  1005.     const int l7= stride + l6;

  1006.     const int l8= stride + l7;

  1007. //    const int l9= stride + l8;

  1008.     int x;

  1009.     src+= stride*3;

  1010.     for(x=0; x<BLOCK_SIZE; x++){

  1011.         const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);

  1012.         if(FFABS(middleEnergy) < 8*c->QP){

  1013.             const int q=(src[l4] - src[l5])/2;

  1014.             const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);

  1015.             const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);

  1016. 

  1017.             int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );

  1018.             d= FFMAX(d, 0);

  1019. 

  1020.             d= (5*d + 32) >> 6;

  1021.             d*= FFSIGN(-middleEnergy);

  1022. 

  1023.             if(q>0){

  1024.                 d= d<0 ? 0 : d;

  1025.                 d= d>q ? q : d;

  1026.             }else{

  1027.                 d= d>0 ? 0 : d;

  1028.                 d= d<q ? q : d;

  1029.             }

  1030. 

  1031.             src[l4]-= d;

  1032.             src[l5]+= d;

  1033.         }

  1034.         src++;

  1035.     }

  1036. #endif //HAVE_MMX2 || HAVE_AMD3DNOW

  1037. }

  1038. #endif //HAVE_ALTIVEC

  1039. 

  1040. #if !HAVE_ALTIVEC

  1041. static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)

  1042. {

  1043. #if HAVE_MMX2 || HAVE_AMD3DNOW

  1044.     DECLARE_ALIGNED(8, uint64_t, tmp)[3];

  1045.     __asm__ volatile(

  1046.         "pxor %%mm6, %%mm6                      \n\t"

  1047.         "pcmpeqb %%mm7, %%mm7                   \n\t"

  1048.         "movq %2, %%mm0                         \n\t"

  1049.         "punpcklbw %%mm6, %%mm0                 \n\t"

  1050.         "psrlw $1, %%mm0                        \n\t"

  1051.         "psubw %%mm7, %%mm0                     \n\t"

  1052.         "packuswb %%mm0, %%mm0                  \n\t"

  1053.         "movq %%mm0, %3                         \n\t"

  1054. 

  1055.         "lea (%0, %1), %%"REG_a"                \n\t"

  1056.         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"

  1057. 

  1058. //        0        1        2        3        4        5        6        7        8        9

  1059. //        %0        eax        eax+%1        eax+2%1        %0+4%1        edx        edx+%1        edx+2%1        %0+8%1        edx+4%1

  1060. 

  1061. #undef REAL_FIND_MIN_MAX

  1062. #undef FIND_MIN_MAX

  1063. #if HAVE_MMX2

  1064. #define REAL_FIND_MIN_MAX(addr)\

  1065.         "movq " #addr ", %%mm0                  \n\t"\

  1066.         "pminub %%mm0, %%mm7                    \n\t"\

  1067.         "pmaxub %%mm0, %%mm6                    \n\t"

  1068. #else

  1069. #define REAL_FIND_MIN_MAX(addr)\

  1070.         "movq " #addr ", %%mm0                  \n\t"\

  1071.         "movq %%mm7, %%mm1                      \n\t"\

  1072.         "psubusb %%mm0, %%mm6                   \n\t"\

  1073.         "paddb %%mm0, %%mm6                     \n\t"\

  1074.         "psubusb %%mm0, %%mm1                   \n\t"\

  1075.         "psubb %%mm1, %%mm7                     \n\t"

  1076. #endif

  1077. #define FIND_MIN_MAX(addr)  REAL_FIND_MIN_MAX(addr)

  1078. 

  1079. FIND_MIN_MAX((%%REGa))

  1080. FIND_MIN_MAX((%%REGa, %1))

  1081. FIND_MIN_MAX((%%REGa, %1, 2))

  1082. FIND_MIN_MAX((%0, %1, 4))

  1083. FIND_MIN_MAX((%%REGd))

  1084. FIND_MIN_MAX((%%REGd, %1))

  1085. FIND_MIN_MAX((%%REGd, %1, 2))

  1086. FIND_MIN_MAX((%0, %1, 8))

  1087. 

  1088.         "movq %%mm7, %%mm4                      \n\t"

  1089.         "psrlq $8, %%mm7                        \n\t"

  1090. #if HAVE_MMX2

  1091.         "pminub %%mm4, %%mm7                    \n\t" // min of pixels

  1092.         "pshufw $0xF9, %%mm7, %%mm4             \n\t"

  1093.         "pminub %%mm4, %%mm7                    \n\t" // min of pixels

  1094.         "pshufw $0xFE, %%mm7, %%mm4             \n\t"

  1095.         "pminub %%mm4, %%mm7                    \n\t"

  1096. #else

  1097.         "movq %%mm7, %%mm1                      \n\t"

  1098.         "psubusb %%mm4, %%mm1                   \n\t"

  1099.         "psubb %%mm1, %%mm7                     \n\t"

  1100.         "movq %%mm7, %%mm4                      \n\t"

  1101.         "psrlq $16, %%mm7                       \n\t"

  1102.         "movq %%mm7, %%mm1                      \n\t"

  1103.         "psubusb %%mm4, %%mm1                   \n\t"

  1104.         "psubb %%mm1, %%mm7                     \n\t"

  1105.         "movq %%mm7, %%mm4                      \n\t"

  1106.         "psrlq $32, %%mm7                       \n\t"

  1107.         "movq %%mm7, %%mm1                      \n\t"

  1108.         "psubusb %%mm4, %%mm1                   \n\t"

  1109.         "psubb %%mm1, %%mm7                     \n\t"

  1110. #endif

  1111. 

  1112. 

  1113.         "movq %%mm6, %%mm4                      \n\t"

  1114.         "psrlq $8, %%mm6                        \n\t"

  1115. #if HAVE_MMX2

  1116.         "pmaxub %%mm4, %%mm6                    \n\t" // max of pixels

  1117.         "pshufw $0xF9, %%mm6, %%mm4             \n\t"

  1118.         "pmaxub %%mm4, %%mm6                    \n\t"

  1119.         "pshufw $0xFE, %%mm6, %%mm4             \n\t"

  1120.         "pmaxub %%mm4, %%mm6                    \n\t"

  1121. #else

  1122.         "psubusb %%mm4, %%mm6                   \n\t"

  1123.         "paddb %%mm4, %%mm6                     \n\t"

  1124.         "movq %%mm6, %%mm4                      \n\t"

  1125.         "psrlq $16, %%mm6                       \n\t"

  1126.         "psubusb %%mm4, %%mm6                   \n\t"

  1127.         "paddb %%mm4, %%mm6                     \n\t"

  1128.         "movq %%mm6, %%mm4                      \n\t"

  1129.         "psrlq $32, %%mm6                       \n\t"

  1130.         "psubusb %%mm4, %%mm6                   \n\t"

  1131.         "paddb %%mm4, %%mm6                     \n\t"

  1132. #endif

  1133.         "movq %%mm6, %%mm0                      \n\t" // max

  1134.         "psubb %%mm7, %%mm6                     \n\t" // max - min

  1135.         "movd %%mm6, %%ecx                      \n\t"

  1136.         "cmpb "MANGLE(deringThreshold)", %%cl   \n\t"

  1137.         " jb 1f                                 \n\t"

  1138.         PAVGB(%%mm0, %%mm7)                           // a=(max + min)/2

  1139.         "punpcklbw %%mm7, %%mm7                 \n\t"

  1140.         "punpcklbw %%mm7, %%mm7                 \n\t"

  1141.         "punpcklbw %%mm7, %%mm7                 \n\t"

  1142.         "movq %%mm7, (%4)                       \n\t"

  1143. 

  1144.         "movq (%0), %%mm0                       \n\t" // L10

  1145.         "movq %%mm0, %%mm1                      \n\t" // L10

  1146.         "movq %%mm0, %%mm2                      \n\t" // L10

  1147.         "psllq $8, %%mm1                        \n\t"

  1148.         "psrlq $8, %%mm2                        \n\t"

  1149.         "movd -4(%0), %%mm3                     \n\t"

  1150.         "movd 8(%0), %%mm4                      \n\t"

  1151.         "psrlq $24, %%mm3                       \n\t"

  1152.         "psllq $56, %%mm4                       \n\t"

  1153.         "por %%mm3, %%mm1                       \n\t" // L00

  1154.         "por %%mm4, %%mm2                       \n\t" // L20

  1155.         "movq %%mm1, %%mm3                      \n\t" // L00

  1156.         PAVGB(%%mm2, %%mm1)                           // (L20 + L00)/2

  1157.         PAVGB(%%mm0, %%mm1)                           // (L20 + L00 + 2L10)/4

  1158.         "psubusb %%mm7, %%mm0                   \n\t"

  1159.         "psubusb %%mm7, %%mm2                   \n\t"

  1160.         "psubusb %%mm7, %%mm3                   \n\t"

  1161.         "pcmpeqb "MANGLE(b00)", %%mm0           \n\t" // L10 > a ? 0 : -1

  1162.         "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L20 > a ? 0 : -1

  1163.         "pcmpeqb "MANGLE(b00)", %%mm3           \n\t" // L00 > a ? 0 : -1

  1164.         "paddb %%mm2, %%mm0                     \n\t"

  1165.         "paddb %%mm3, %%mm0                     \n\t"

  1166. 

  1167.         "movq (%%"REG_a"), %%mm2                \n\t" // L11

  1168.         "movq %%mm2, %%mm3                      \n\t" // L11

  1169.         "movq %%mm2, %%mm4                      \n\t" // L11

  1170.         "psllq $8, %%mm3                        \n\t"

  1171.         "psrlq $8, %%mm4                        \n\t"

  1172.         "movd -4(%%"REG_a"), %%mm5              \n\t"

  1173.         "movd 8(%%"REG_a"), %%mm6               \n\t"

  1174.         "psrlq $24, %%mm5                       \n\t"

  1175.         "psllq $56, %%mm6                       \n\t"

  1176.         "por %%mm5, %%mm3                       \n\t" // L01

  1177.         "por %%mm6, %%mm4                       \n\t" // L21

  1178.         "movq %%mm3, %%mm5                      \n\t" // L01

  1179.         PAVGB(%%mm4, %%mm3)                           // (L21 + L01)/2

  1180.         PAVGB(%%mm2, %%mm3)                           // (L21 + L01 + 2L11)/4

  1181.         "psubusb %%mm7, %%mm2                   \n\t"

  1182.         "psubusb %%mm7, %%mm4                   \n\t"

  1183.         "psubusb %%mm7, %%mm5                   \n\t"

  1184.         "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L11 > a ? 0 : -1

  1185.         "pcmpeqb "MANGLE(b00)", %%mm4           \n\t" // L21 > a ? 0 : -1

  1186.         "pcmpeqb "MANGLE(b00)", %%mm5           \n\t" // L01 > a ? 0 : -1

  1187.         "paddb %%mm4, %%mm2                     \n\t"

  1188.         "paddb %%mm5, %%mm2                     \n\t"

  1189. // 0, 2, 3, 1

  1190. #define REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \

  1191.         "movq " #src ", " #sx "                 \n\t" /* src[0] */\

  1192.         "movq " #sx ", " #lx "                  \n\t" /* src[0] */\

  1193.         "movq " #sx ", " #t0 "                  \n\t" /* src[0] */\

  1194.         "psllq $8, " #lx "                      \n\t"\

  1195.         "psrlq $8, " #t0 "                      \n\t"\

  1196.         "movd -4" #src ", " #t1 "               \n\t"\

  1197.         "psrlq $24, " #t1 "                     \n\t"\

  1198.         "por " #t1 ", " #lx "                   \n\t" /* src[-1] */\

  1199.         "movd 8" #src ", " #t1 "                \n\t"\

  1200.         "psllq $56, " #t1 "                     \n\t"\

  1201.         "por " #t1 ", " #t0 "                   \n\t" /* src[+1] */\

  1202.         "movq " #lx ", " #t1 "                  \n\t" /* src[-1] */\

  1203.         PAVGB(t0, lx)                                 /* (src[-1] + src[+1])/2 */\

  1204.         PAVGB(sx, lx)                                 /* (src[-1] + 2src[0] + src[+1])/4 */\

  1205.         PAVGB(lx, pplx)                                     \

  1206.         "movq " #lx ", 8(%4)                    \n\t"\

  1207.         "movq (%4), " #lx "                     \n\t"\

  1208.         "psubusb " #lx ", " #t1 "               \n\t"\

  1209.         "psubusb " #lx ", " #t0 "               \n\t"\

  1210.         "psubusb " #lx ", " #sx "               \n\t"\

  1211.         "movq "MANGLE(b00)", " #lx "            \n\t"\

  1212.         "pcmpeqb " #lx ", " #t1 "               \n\t" /* src[-1] > a ? 0 : -1*/\

  1213.         "pcmpeqb " #lx ", " #t0 "               \n\t" /* src[+1] > a ? 0 : -1*/\

  1214.         "pcmpeqb " #lx ", " #sx "               \n\t" /* src[0]  > a ? 0 : -1*/\

  1215.         "paddb " #t1 ", " #t0 "                 \n\t"\

  1216.         "paddb " #t0 ", " #sx "                 \n\t"\

  1217. \

  1218.         PAVGB(plx, pplx)                              /* filtered */\

  1219.         "movq " #dst ", " #t0 "                 \n\t" /* dst */\

  1220.         "movq " #t0 ", " #t1 "                  \n\t" /* dst */\

  1221.         "psubusb %3, " #t0 "                    \n\t"\

  1222.         "paddusb %3, " #t1 "                    \n\t"\

  1223.         PMAXUB(t0, pplx)\

  1224.         PMINUB(t1, pplx, t0)\

  1225.         "paddb " #sx ", " #ppsx "               \n\t"\

  1226.         "paddb " #psx ", " #ppsx "              \n\t"\

  1227.         "#paddb "MANGLE(b02)", " #ppsx "        \n\t"\

  1228.         "pand "MANGLE(b08)", " #ppsx "          \n\t"\

  1229.         "pcmpeqb " #lx ", " #ppsx "             \n\t"\

  1230.         "pand " #ppsx ", " #pplx "              \n\t"\

  1231.         "pandn " #dst ", " #ppsx "              \n\t"\

  1232.         "por " #pplx ", " #ppsx "               \n\t"\

  1233.         "movq " #ppsx ", " #dst "               \n\t"\

  1234.         "movq 8(%4), " #lx "                    \n\t"

  1235. 

  1236. #define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \

  1237.    REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1)

  1238. /*

  1239. 0000000

  1240. 1111111

  1241. 

  1242. 1111110

  1243. 1111101

  1244. 1111100

  1245. 1111011

  1246. 1111010

  1247. 1111001

  1248. 

  1249. 1111000

  1250. 1110111

  1251. 

  1252. */

  1253. //DERING_CORE(dst          ,src            ,ppsx ,psx  ,sx   ,pplx ,plx  ,lx   ,t0   ,t1)

  1254. DERING_CORE((%%REGa)       ,(%%REGa, %1)   ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)

  1255. DERING_CORE((%%REGa, %1)   ,(%%REGa, %1, 2),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)

  1256. DERING_CORE((%%REGa, %1, 2),(%0, %1, 4)    ,%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)

  1257. DERING_CORE((%0, %1, 4)    ,(%%REGd)       ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)

  1258. DERING_CORE((%%REGd)       ,(%%REGd, %1)   ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)

  1259. DERING_CORE((%%REGd, %1)   ,(%%REGd, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)

  1260. DERING_CORE((%%REGd, %1, 2),(%0, %1, 8)    ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)

  1261. DERING_CORE((%0, %1, 8)    ,(%%REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)

  1262. 

  1263.         "1:                        \n\t"

  1264.         : : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb), "m"(c->pQPb2), "r"(tmp)

  1265.         : "%"REG_a, "%"REG_d, "%"REG_c

  1266.     );

  1267. #else //HAVE_MMX2 || HAVE_AMD3DNOW

  1268.     int y;

  1269.     int min=255;

  1270.     int max=0;

  1271.     int avg;

  1272.     uint8_t *p;

  1273.     int s[10];

  1274.     const int QP2= c->QP/2 + 1;

  1275. 

  1276.     for(y=1; y<9; y++){

  1277.         int x;

  1278.         p= src + stride*y;

  1279.         for(x=1; x<9; x++){

  1280.             p++;

  1281.             if(*p > max) max= *p;

  1282.             if(*p < min) min= *p;

  1283.         }

  1284.     }

  1285.     avg= (min + max + 1)>>1;

  1286. 

  1287.     if(max - min <deringThreshold) return;

  1288. 

  1289.     for(y=0; y<10; y++){

  1290.         int t = 0;

  1291. 

  1292.         if(src[stride*y + 0] > avg) t+= 1;

  1293.         if(src[stride*y + 1] > avg) t+= 2;

  1294.         if(src[stride*y + 2] > avg) t+= 4;

  1295.         if(src[stride*y + 3] > avg) t+= 8;

  1296.         if(src[stride*y + 4] > avg) t+= 16;

  1297.         if(src[stride*y + 5] > avg) t+= 32;

  1298.         if(src[stride*y + 6] > avg) t+= 64;

  1299.         if(src[stride*y + 7] > avg) t+= 128;

  1300.         if(src[stride*y + 8] > avg) t+= 256;

  1301.         if(src[stride*y + 9] > avg) t+= 512;

  1302. 

  1303.         t |= (~t)<<16;

  1304.         t &= (t<<1) & (t>>1);

  1305.         s[y] = t;

  1306.     }

  1307. 

  1308.     for(y=1; y<9; y++){

  1309.         int t = s[y-1] & s[y] & s[y+1];

  1310.         t|= t>>16;

  1311.         s[y-1]= t;

  1312.     }

  1313. 

  1314.     for(y=1; y<9; y++){

  1315.         int x;

  1316.         int t = s[y-1];

  1317. 

  1318.         p= src + stride*y;

  1319.         for(x=1; x<9; x++){

  1320.             p++;

  1321.             if(t & (1<<x)){

  1322.                 int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))

  1323.                       +2*(*(p     -1)) + 4*(*p         ) + 2*(*(p     +1))

  1324.                       +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));

  1325.                 f= (f + 8)>>4;

  1326. 

  1327. #ifdef DEBUG_DERING_THRESHOLD

  1328.                     __asm__ volatile("emms\n\t":);

  1329.                     {

  1330.                     static long long numPixels=0;

  1331.                     if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;

  1332. //                    if((max-min)<20 || (max-min)*QP<200)

  1333. //                    if((max-min)*QP < 500)

  1334. //                    if(max-min<QP/2)

  1335.                     if(max-min < 20){

  1336.                         static int numSkipped=0;

  1337.                         static int errorSum=0;

  1338.                         static int worstQP=0;

  1339.                         static int worstRange=0;

  1340.                         static int worstDiff=0;

  1341.                         int diff= (f - *p);

  1342.                         int absDiff= FFABS(diff);

  1343.                         int error= diff*diff;

  1344. 

  1345.                         if(x==1 || x==8 || y==1 || y==8) continue;

  1346. 

  1347.                         numSkipped++;

  1348.                         if(absDiff > worstDiff){

  1349.                             worstDiff= absDiff;

  1350.                             worstQP= QP;

  1351.                             worstRange= max-min;

  1352.                         }

  1353.                         errorSum+= error;

  1354. 

  1355.                         if(1024LL*1024LL*1024LL % numSkipped == 0){

  1356.                             av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "

  1357.                                    "wRange:%d, wDiff:%d, relSkip:%1.3f\n",

  1358.                                    (float)errorSum/numSkipped, numSkipped, worstQP, worstRange,

  1359.                                    worstDiff, (float)numSkipped/numPixels);

  1360.                         }

  1361.                     }

  1362.                     }

  1363. #endif

  1364.                     if     (*p + QP2 < f) *p= *p + QP2;

  1365.                     else if(*p - QP2 > f) *p= *p - QP2;

  1366.                     else *p=f;

  1367.             }

  1368.         }

  1369.     }

  1370. #ifdef DEBUG_DERING_THRESHOLD

  1371.     if(max-min < 20){

  1372.         for(y=1; y<9; y++){

  1373.             int x;

  1374.             int t = 0;

  1375.             p= src + stride*y;

  1376.             for(x=1; x<9; x++){

  1377.                 p++;

  1378.                 *p = FFMIN(*p + 20, 255);

  1379.             }

  1380.         }

  1381. //        src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;

  1382.     }

  1383. #endif

  1384. #endif //HAVE_MMX2 || HAVE_AMD3DNOW

  1385. }

  1386. #endif //HAVE_ALTIVEC

  1387. 

  1388. /**

  1389.  * Deinterlace the given block by linearly interpolating every second line.

  1390.  * will be called for every 8x8 block and can read & write from line 4-15

  1391.  * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.

  1392.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1393.  */

  1394. static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)

  1395. {

  1396. #if HAVE_MMX2 || HAVE_AMD3DNOW

  1397.     src+= 4*stride;

  1398.     __asm__ volatile(

  1399.         "lea (%0, %1), %%"REG_a"                \n\t"

  1400.         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"

  1401. //      0       1       2       3       4       5       6       7       8       9

  1402. //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1

  1403. 

  1404.         "movq (%0), %%mm0                       \n\t"

  1405.         "movq (%%"REG_a", %1), %%mm1            \n\t"

  1406.         PAVGB(%%mm1, %%mm0)

  1407.         "movq %%mm0, (%%"REG_a")                \n\t"

  1408.         "movq (%0, %1, 4), %%mm0                \n\t"

  1409.         PAVGB(%%mm0, %%mm1)

  1410.         "movq %%mm1, (%%"REG_a", %1, 2)         \n\t"

  1411.         "movq (%%"REG_c", %1), %%mm1            \n\t"

  1412.         PAVGB(%%mm1, %%mm0)

  1413.         "movq %%mm0, (%%"REG_c")                \n\t"

  1414.         "movq (%0, %1, 8), %%mm0                \n\t"

  1415.         PAVGB(%%mm0, %%mm1)

  1416.         "movq %%mm1, (%%"REG_c", %1, 2)         \n\t"

  1417. 

  1418.         : : "r" (src), "r" ((x86_reg)stride)

  1419.         : "%"REG_a, "%"REG_c

  1420.     );

  1421. #else

  1422.     int a, b, x;

  1423.     src+= 4*stride;

  1424. 

  1425.     for(x=0; x<2; x++){

  1426.         a= *(uint32_t*)&src[stride*0];

  1427.         b= *(uint32_t*)&src[stride*2];

  1428.         *(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1429.         a= *(uint32_t*)&src[stride*4];

  1430.         *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1431.         b= *(uint32_t*)&src[stride*6];

  1432.         *(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1433.         a= *(uint32_t*)&src[stride*8];

  1434.         *(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1435.         src += 4;

  1436.     }

  1437. #endif

  1438. }

  1439. 

  1440. /**

  1441.  * Deinterlace the given block by cubic interpolating every second line.

  1442.  * will be called for every 8x8 block and can read & write from line 4-15

  1443.  * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.

  1444.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1445.  * this filter will read lines 3-15 and write 7-13

  1446.  */

  1447. static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)

  1448. {

  1449. #if HAVE_MMX2 || HAVE_AMD3DNOW

  1450.     src+= stride*3;

  1451.     __asm__ volatile(

  1452.         "lea (%0, %1), %%"REG_a"                \n\t"

  1453.         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"

  1454.         "lea (%%"REG_d", %1, 4), %%"REG_c"      \n\t"

  1455.         "add %1, %%"REG_c"                      \n\t"

  1456.         "pxor %%mm7, %%mm7                      \n\t"

  1457. //      0       1       2       3       4       5       6       7       8       9       10

  1458. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx

  1459. 

  1460. #define REAL_DEINT_CUBIC(a,b,c,d,e)\

  1461.         "movq " #a ", %%mm0                     \n\t"\

  1462.         "movq " #b ", %%mm1                     \n\t"\

  1463.         "movq " #d ", %%mm2                     \n\t"\

  1464.         "movq " #e ", %%mm3                     \n\t"\

  1465.         PAVGB(%%mm2, %%mm1)                             /* (b+d) /2 */\

  1466.         PAVGB(%%mm3, %%mm0)                             /* a(a+e) /2 */\

  1467.         "movq %%mm0, %%mm2                      \n\t"\

  1468.         "punpcklbw %%mm7, %%mm0                 \n\t"\

  1469.         "punpckhbw %%mm7, %%mm2                 \n\t"\

  1470.         "movq %%mm1, %%mm3                      \n\t"\

  1471.         "punpcklbw %%mm7, %%mm1                 \n\t"\

  1472.         "punpckhbw %%mm7, %%mm3                 \n\t"\

  1473.         "psubw %%mm1, %%mm0                     \n\t"   /* L(a+e - (b+d))/2 */\

  1474.         "psubw %%mm3, %%mm2                     \n\t"   /* H(a+e - (b+d))/2 */\

  1475.         "psraw $3, %%mm0                        \n\t"   /* L(a+e - (b+d))/16 */\

  1476.         "psraw $3, %%mm2                        \n\t"   /* H(a+e - (b+d))/16 */\

  1477.         "psubw %%mm0, %%mm1                     \n\t"   /* L(9b + 9d - a - e)/16 */\

  1478.         "psubw %%mm2, %%mm3                     \n\t"   /* H(9b + 9d - a - e)/16 */\

  1479.         "packuswb %%mm3, %%mm1                  \n\t"\

  1480.         "movq %%mm1, " #c "                     \n\t"

  1481. #define DEINT_CUBIC(a,b,c,d,e)  REAL_DEINT_CUBIC(a,b,c,d,e)

  1482. 

  1483. DEINT_CUBIC((%0)        , (%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd, %1))

  1484. DEINT_CUBIC((%%REGa, %1), (%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%0, %1, 8))

  1485. DEINT_CUBIC((%0, %1, 4) , (%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGc))

  1486. DEINT_CUBIC((%%REGd, %1), (%0, %1, 8) , (%%REGd, %1, 4), (%%REGc)    , (%%REGc, %1, 2))

  1487. 

  1488.         : : "r" (src), "r" ((x86_reg)stride)

  1489.         : "%"REG_a, "%"REG_d, "%"REG_c

  1490.     );

  1491. #else //HAVE_MMX2 || HAVE_AMD3DNOW

  1492.     int x;

  1493.     src+= stride*3;

  1494.     for(x=0; x<8; x++){

  1495.         src[stride*3] = CLIP((-src[0]        + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);

  1496.         src[stride*5] = CLIP((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);

  1497.         src[stride*7] = CLIP((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);

  1498.         src[stride*9] = CLIP((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);

  1499.         src++;

  1500.     }

  1501. #endif //HAVE_MMX2 || HAVE_AMD3DNOW

  1502. }

  1503. 

  1504. /**

  1505.  * Deinterlace the given block by filtering every second line with a (-1 4 2 4 -1) filter.

  1506.  * will be called for every 8x8 block and can read & write from line 4-15

  1507.  * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.

  1508.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1509.  * this filter will read lines 4-13 and write 5-11

  1510.  */

  1511. static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)

  1512. {

  1513. #if HAVE_MMX2 || HAVE_AMD3DNOW

  1514.     src+= stride*4;

  1515.     __asm__ volatile(

  1516.         "lea (%0, %1), %%"REG_a"                \n\t"

  1517.         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"

  1518.         "pxor %%mm7, %%mm7                      \n\t"

  1519.         "movq (%2), %%mm0                       \n\t"

  1520. //      0       1       2       3       4       5       6       7       8       9       10

  1521. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx

  1522. 

  1523. #define REAL_DEINT_FF(a,b,c,d)\

  1524.         "movq " #a ", %%mm1                     \n\t"\

  1525.         "movq " #b ", %%mm2                     \n\t"\

  1526.         "movq " #c ", %%mm3                     \n\t"\

  1527.         "movq " #d ", %%mm4                     \n\t"\

  1528.         PAVGB(%%mm3, %%mm1)                          \

  1529.         PAVGB(%%mm4, %%mm0)                          \

  1530.         "movq %%mm0, %%mm3                      \n\t"\

  1531.         "punpcklbw %%mm7, %%mm0                 \n\t"\

  1532.         "punpckhbw %%mm7, %%mm3                 \n\t"\

  1533.         "movq %%mm1, %%mm4                      \n\t"\

  1534.         "punpcklbw %%mm7, %%mm1                 \n\t"\

  1535.         "punpckhbw %%mm7, %%mm4                 \n\t"\

  1536.         "psllw $2, %%mm1                        \n\t"\

  1537.         "psllw $2, %%mm4                        \n\t"\

  1538.         "psubw %%mm0, %%mm1                     \n\t"\

  1539.         "psubw %%mm3, %%mm4                     \n\t"\

  1540.         "movq %%mm2, %%mm5                      \n\t"\

  1541.         "movq %%mm2, %%mm0                      \n\t"\

  1542.         "punpcklbw %%mm7, %%mm2                 \n\t"\

  1543.         "punpckhbw %%mm7, %%mm5                 \n\t"\

  1544.         "paddw %%mm2, %%mm1                     \n\t"\

  1545.         "paddw %%mm5, %%mm4                     \n\t"\

  1546.         "psraw $2, %%mm1                        \n\t"\

  1547.         "psraw $2, %%mm4                        \n\t"\

  1548.         "packuswb %%mm4, %%mm1                  \n\t"\

  1549.         "movq %%mm1, " #b "                     \n\t"\

  1550. 

  1551. #define DEINT_FF(a,b,c,d)  REAL_DEINT_FF(a,b,c,d)

  1552. 

  1553. DEINT_FF((%0)        , (%%REGa)       , (%%REGa, %1), (%%REGa, %1, 2))

  1554. DEINT_FF((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd)       )

  1555. DEINT_FF((%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%%REGd, %1, 2))

  1556. DEINT_FF((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))

  1557. 

  1558.         "movq %%mm0, (%2)                       \n\t"

  1559.         : : "r" (src), "r" ((x86_reg)stride), "r"(tmp)

  1560.         : "%"REG_a, "%"REG_d

  1561.     );

  1562. #else //HAVE_MMX2 || HAVE_AMD3DNOW

  1563.     int x;

  1564.     src+= stride*4;

  1565.     for(x=0; x<8; x++){

  1566.         int t1= tmp[x];

  1567.         int t2= src[stride*1];

  1568. 

  1569.         src[stride*1]= CLIP((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);

  1570.         t1= src[stride*4];

  1571.         src[stride*3]= CLIP((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);

  1572.         t2= src[stride*6];

  1573.         src[stride*5]= CLIP((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);

  1574.         t1= src[stride*8];

  1575.         src[stride*7]= CLIP((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);

  1576.         tmp[x]= t1;

  1577. 

  1578.         src++;

  1579.     }

  1580. #endif //HAVE_MMX2 || HAVE_AMD3DNOW

  1581. }

  1582. 

  1583. /**

  1584.  * Deinterlace the given block by filtering every line with a (-1 2 6 2 -1) filter.

  1585.  * will be called for every 8x8 block and can read & write from line 4-15

  1586.  * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.

  1587.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1588.  * this filter will read lines 4-13 and write 4-11

  1589.  */

  1590. static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)

  1591. {

  1592. #if HAVE_MMX2 || HAVE_AMD3DNOW

  1593.     src+= stride*4;

  1594.     __asm__ volatile(

  1595.         "lea (%0, %1), %%"REG_a"                \n\t"

  1596.         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"

  1597.         "pxor %%mm7, %%mm7                      \n\t"

  1598.         "movq (%2), %%mm0                       \n\t"

  1599.         "movq (%3), %%mm1                       \n\t"

  1600. //      0       1       2       3       4       5       6       7       8       9       10

  1601. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx

  1602. 

  1603. #define REAL_DEINT_L5(t1,t2,a,b,c)\

  1604.         "movq " #a ", %%mm2                     \n\t"\

  1605.         "movq " #b ", %%mm3                     \n\t"\

  1606.         "movq " #c ", %%mm4                     \n\t"\

  1607.         PAVGB(t2, %%mm3)                             \

  1608.         PAVGB(t1, %%mm4)                             \

  1609.         "movq %%mm2, %%mm5                      \n\t"\

  1610.         "movq %%mm2, " #t1 "                    \n\t"\

  1611.         "punpcklbw %%mm7, %%mm2                 \n\t"\

  1612.         "punpckhbw %%mm7, %%mm5                 \n\t"\

  1613.         "movq %%mm2, %%mm6                      \n\t"\

  1614.         "paddw %%mm2, %%mm2                     \n\t"\

  1615.         "paddw %%mm6, %%mm2                     \n\t"\

  1616.         "movq %%mm5, %%mm6                      \n\t"\

  1617.         "paddw %%mm5, %%mm5                     \n\t"\

  1618.         "paddw %%mm6, %%mm5                     \n\t"\

  1619.         "movq %%mm3, %%mm6                      \n\t"\

  1620.         "punpcklbw %%mm7, %%mm3                 \n\t"\

  1621.         "punpckhbw %%mm7, %%mm6                 \n\t"\

  1622.         "paddw %%mm3, %%mm3                     \n\t"\

  1623.         "paddw %%mm6, %%mm6                     \n\t"\

  1624.         "paddw %%mm3, %%mm2                     \n\t"\

  1625.         "paddw %%mm6, %%mm5                     \n\t"\

  1626.         "movq %%mm4, %%mm6                      \n\t"\

  1627.         "punpcklbw %%mm7, %%mm4                 \n\t"\

  1628.         "punpckhbw %%mm7, %%mm6                 \n\t"\

  1629.         "psubw %%mm4, %%mm2                     \n\t"\

  1630.         "psubw %%mm6, %%mm5                     \n\t"\

  1631.         "psraw $2, %%mm2                        \n\t"\

  1632.         "psraw $2, %%mm5                        \n\t"\

  1633.         "packuswb %%mm5, %%mm2                  \n\t"\

  1634.         "movq %%mm2, " #a "                     \n\t"\

  1635. 

  1636. #define DEINT_L5(t1,t2,a,b,c)  REAL_DEINT_L5(t1,t2,a,b,c)

  1637. 

  1638. DEINT_L5(%%mm0, %%mm1, (%0)           , (%%REGa)       , (%%REGa, %1)   )

  1639. DEINT_L5(%%mm1, %%mm0, (%%REGa)       , (%%REGa, %1)   , (%%REGa, %1, 2))

  1640. DEINT_L5(%%mm0, %%mm1, (%%REGa, %1)   , (%%REGa, %1, 2), (%0, %1, 4)   )

  1641. DEINT_L5(%%mm1, %%mm0, (%%REGa, %1, 2), (%0, %1, 4)    , (%%REGd)       )

  1642. DEINT_L5(%%mm0, %%mm1, (%0, %1, 4)    , (%%REGd)       , (%%REGd, %1)   )

  1643. DEINT_L5(%%mm1, %%mm0, (%%REGd)       , (%%REGd, %1)   , (%%REGd, %1, 2))

  1644. DEINT_L5(%%mm0, %%mm1, (%%REGd, %1)   , (%%REGd, %1, 2), (%0, %1, 8)   )

  1645. DEINT_L5(%%mm1, %%mm0, (%%REGd, %1, 2), (%0, %1, 8)    , (%%REGd, %1, 4))

  1646. 

  1647.         "movq %%mm0, (%2)                       \n\t"

  1648.         "movq %%mm1, (%3)                       \n\t"

  1649.         : : "r" (src), "r" ((x86_reg)stride), "r"(tmp), "r"(tmp2)

  1650.         : "%"REG_a, "%"REG_d

  1651.     );

  1652. #else //HAVE_MMX2 || HAVE_AMD3DNOW

  1653.     int x;

  1654.     src+= stride*4;

  1655.     for(x=0; x<8; x++){

  1656.         int t1= tmp[x];

  1657.         int t2= tmp2[x];

  1658.         int t3= src[0];

  1659. 

  1660.         src[stride*0]= CLIP((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);

  1661.         t1= src[stride*1];

  1662.         src[stride*1]= CLIP((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);

  1663.         t2= src[stride*2];

  1664.         src[stride*2]= CLIP((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);

  1665.         t3= src[stride*3];

  1666.         src[stride*3]= CLIP((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);

  1667.         t1= src[stride*4];

  1668.         src[stride*4]= CLIP((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);

  1669.         t2= src[stride*5];

  1670.         src[stride*5]= CLIP((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);

  1671.         t3= src[stride*6];

  1672.         src[stride*6]= CLIP((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);

  1673.         t1= src[stride*7];

  1674.         src[stride*7]= CLIP((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);

  1675. 

  1676.         tmp[x]= t3;

  1677.         tmp2[x]= t1;

  1678. 

  1679.         src++;

  1680.     }

  1681. #endif //HAVE_MMX2 || HAVE_AMD3DNOW

  1682. }

  1683. 

  1684. /**

  1685.  * Deinterlace the given block by filtering all lines with a (1 2 1) filter.

  1686.  * will be called for every 8x8 block and can read & write from line 4-15

  1687.  * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.

  1688.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1689.  * this filter will read lines 4-13 and write 4-11

  1690.  */

  1691. static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uint8_t *tmp)

  1692. {

  1693. #if HAVE_MMX2 || HAVE_AMD3DNOW

  1694.     src+= 4*stride;

  1695.     __asm__ volatile(

  1696.         "lea (%0, %1), %%"REG_a"                \n\t"

  1697.         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"

  1698. //      0       1       2       3       4       5       6       7       8       9

  1699. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1

  1700. 

  1701.         "movq (%2), %%mm0                       \n\t" // L0

  1702.         "movq (%%"REG_a"), %%mm1                \n\t" // L2

  1703.         PAVGB(%%mm1, %%mm0)                           // L0+L2

  1704.         "movq (%0), %%mm2                       \n\t" // L1

  1705.         PAVGB(%%mm2, %%mm0)

  1706.         "movq %%mm0, (%0)                       \n\t"

  1707.         "movq (%%"REG_a", %1), %%mm0            \n\t" // L3

  1708.         PAVGB(%%mm0, %%mm2)                           // L1+L3

  1709.         PAVGB(%%mm1, %%mm2)                           // 2L2 + L1 + L3

  1710.         "movq %%mm2, (%%"REG_a")                \n\t"

  1711.         "movq (%%"REG_a", %1, 2), %%mm2         \n\t" // L4

  1712.         PAVGB(%%mm2, %%mm1)                           // L2+L4

  1713.         PAVGB(%%mm0, %%mm1)                           // 2L3 + L2 + L4

  1714.         "movq %%mm1, (%%"REG_a", %1)            \n\t"

  1715.         "movq (%0, %1, 4), %%mm1                \n\t" // L5

  1716.         PAVGB(%%mm1, %%mm0)                           // L3+L5

  1717.         PAVGB(%%mm2, %%mm0)                           // 2L4 + L3 + L5

  1718.         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"

  1719.         "movq (%%"REG_d"), %%mm0                \n\t" // L6

  1720.         PAVGB(%%mm0, %%mm2)                           // L4+L6

  1721.         PAVGB(%%mm1, %%mm2)                           // 2L5 + L4 + L6

  1722.         "movq %%mm2, (%0, %1, 4)                \n\t"

  1723.         "movq (%%"REG_d", %1), %%mm2            \n\t" // L7

  1724.         PAVGB(%%mm2, %%mm1)                           // L5+L7

  1725.         PAVGB(%%mm0, %%mm1)                           // 2L6 + L5 + L7

  1726.         "movq %%mm1, (%%"REG_d")                \n\t"

  1727.         "movq (%%"REG_d", %1, 2), %%mm1         \n\t" // L8

  1728.         PAVGB(%%mm1, %%mm0)                           // L6+L8

  1729.         PAVGB(%%mm2, %%mm0)                           // 2L7 + L6 + L8

  1730.         "movq %%mm0, (%%"REG_d", %1)            \n\t"

  1731.         "movq (%0, %1, 8), %%mm0                \n\t" // L9

  1732.         PAVGB(%%mm0, %%mm2)                           // L7+L9

  1733.         PAVGB(%%mm1, %%mm2)                           // 2L8 + L7 + L9

  1734.         "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"

  1735.         "movq %%mm1, (%2)                       \n\t"

  1736. 

  1737.         : : "r" (src), "r" ((x86_reg)stride), "r" (tmp)

  1738.         : "%"REG_a, "%"REG_d

  1739.     );

  1740. #else //HAVE_MMX2 || HAVE_AMD3DNOW

  1741.     int a, b, c, x;

  1742.     src+= 4*stride;

  1743. 

  1744.     for(x=0; x<2; x++){

  1745.         a= *(uint32_t*)&tmp[stride*0];

  1746.         b= *(uint32_t*)&src[stride*0];

  1747.         c= *(uint32_t*)&src[stride*1];

  1748.         a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);

  1749.         *(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1750. 

  1751.         a= *(uint32_t*)&src[stride*2];

  1752.         b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);

  1753.         *(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);

  1754. 

  1755.         b= *(uint32_t*)&src[stride*3];

  1756.         c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);

  1757.         *(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);

  1758. 

  1759.         c= *(uint32_t*)&src[stride*4];

  1760.         a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);

  1761.         *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1762. 

  1763.         a= *(uint32_t*)&src[stride*5];

  1764.         b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);

  1765.         *(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);

  1766. 

  1767.         b= *(uint32_t*)&src[stride*6];

  1768.         c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);

  1769.         *(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);

  1770. 

  1771.         c= *(uint32_t*)&src[stride*7];

  1772.         a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);

  1773.         *(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1774. 

  1775.         a= *(uint32_t*)&src[stride*8];

  1776.         b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);

  1777.         *(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);

  1778. 

  1779.         *(uint32_t*)&tmp[stride*0]= c;

  1780.         src += 4;

  1781.         tmp += 4;

  1782.     }

  1783. #endif //HAVE_MMX2 || HAVE_AMD3DNOW

  1784. }

  1785. 

  1786. /**

  1787.  * Deinterlace the given block by applying a median filter to every second line.

  1788.  * will be called for every 8x8 block and can read & write from line 4-15,

  1789.  * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.

  1790.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1791.  */

  1792. static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)

  1793. {

  1794. #if HAVE_MMX

  1795.     src+= 4*stride;

  1796. #if HAVE_MMX2

  1797.     __asm__ volatile(

  1798.         "lea (%0, %1), %%"REG_a"                \n\t"

  1799.         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"

  1800. //      0       1       2       3       4       5       6       7       8       9

  1801. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1

  1802. 

  1803.         "movq (%0), %%mm0                       \n\t" //

  1804.         "movq (%%"REG_a", %1), %%mm2            \n\t" //

  1805.         "movq (%%"REG_a"), %%mm1                \n\t" //

  1806.         "movq %%mm0, %%mm3                      \n\t"

  1807.         "pmaxub %%mm1, %%mm0                    \n\t" //

  1808.         "pminub %%mm3, %%mm1                    \n\t" //

  1809.         "pmaxub %%mm2, %%mm1                    \n\t" //

  1810.         "pminub %%mm1, %%mm0                    \n\t"

  1811.         "movq %%mm0, (%%"REG_a")                \n\t"

  1812. 

  1813.         "movq (%0, %1, 4), %%mm0                \n\t" //

  1814.         "movq (%%"REG_a", %1, 2), %%mm1         \n\t" //

  1815.         "movq %%mm2, %%mm3                      \n\t"

  1816.         "pmaxub %%mm1, %%mm2                    \n\t" //

  1817.         "pminub %%mm3, %%mm1                    \n\t" //

  1818.         "pmaxub %%mm0, %%mm1                    \n\t" //

  1819.         "pminub %%mm1, %%mm2                    \n\t"

  1820.         "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"

  1821. 

  1822.         "movq (%%"REG_d"), %%mm2                \n\t" //

  1823.         "movq (%%"REG_d", %1), %%mm1            \n\t" //

  1824.         "movq %%mm2, %%mm3                      \n\t"

  1825.         "pmaxub %%mm0, %%mm2                    \n\t" //

  1826.         "pminub %%mm3, %%mm0                    \n\t" //

  1827.         "pmaxub %%mm1, %%mm0                    \n\t" //

  1828.         "pminub %%mm0, %%mm2                    \n\t"

  1829.         "movq %%mm2, (%%"REG_d")                \n\t"

  1830. 

  1831.         "movq (%%"REG_d", %1, 2), %%mm2         \n\t" //

  1832.         "movq (%0, %1, 8), %%mm0                \n\t" //

  1833.         "movq %%mm2, %%mm3                      \n\t"

  1834.         "pmaxub %%mm0, %%mm2                    \n\t" //

  1835.         "pminub %%mm3, %%mm0                    \n\t" //

  1836.         "pmaxub %%mm1, %%mm0                    \n\t" //

  1837.         "pminub %%mm0, %%mm2                    \n\t"

  1838.         "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"

  1839. 

  1840. 

  1841.         : : "r" (src), "r" ((x86_reg)stride)

  1842.         : "%"REG_a, "%"REG_d

  1843.     );

  1844. 

  1845. #else // MMX without MMX2

  1846.     __asm__ volatile(

  1847.         "lea (%0, %1), %%"REG_a"                \n\t"

  1848.         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"

  1849. //      0       1       2       3       4       5       6       7       8       9

  1850. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1

  1851.         "pxor %%mm7, %%mm7                      \n\t"

  1852. 

  1853. #define REAL_MEDIAN(a,b,c)\

  1854.         "movq " #a ", %%mm0                     \n\t"\

  1855.         "movq " #b ", %%mm2                     \n\t"\

  1856.         "movq " #c ", %%mm1                     \n\t"\

  1857.         "movq %%mm0, %%mm3                      \n\t"\

  1858.         "movq %%mm1, %%mm4                      \n\t"\

  1859.         "movq %%mm2, %%mm5                      \n\t"\

  1860.         "psubusb %%mm1, %%mm3                   \n\t"\

  1861.         "psubusb %%mm2, %%mm4                   \n\t"\

  1862.         "psubusb %%mm0, %%mm5                   \n\t"\

  1863.         "pcmpeqb %%mm7, %%mm3                   \n\t"\

  1864.         "pcmpeqb %%mm7, %%mm4                   \n\t"\

  1865.         "pcmpeqb %%mm7, %%mm5                   \n\t"\

  1866.         "movq %%mm3, %%mm6                      \n\t"\

  1867.         "pxor %%mm4, %%mm3                      \n\t"\

  1868.         "pxor %%mm5, %%mm4                      \n\t"\

  1869.         "pxor %%mm6, %%mm5                      \n\t"\

  1870.         "por %%mm3, %%mm1                       \n\t"\

  1871.         "por %%mm4, %%mm2                       \n\t"\

  1872.         "por %%mm5, %%mm0                       \n\t"\

  1873.         "pand %%mm2, %%mm0                      \n\t"\

  1874.         "pand %%mm1, %%mm0                      \n\t"\

  1875.         "movq %%mm0, " #b "                     \n\t"

  1876. #define MEDIAN(a,b,c)  REAL_MEDIAN(a,b,c)

  1877. 

  1878. MEDIAN((%0)        , (%%REGa)       , (%%REGa, %1))

  1879. MEDIAN((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4))

  1880. MEDIAN((%0, %1, 4) , (%%REGd)       , (%%REGd, %1))

  1881. MEDIAN((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8))

  1882. 

  1883.         : : "r" (src), "r" ((x86_reg)stride)

  1884.         : "%"REG_a, "%"REG_d

  1885.     );

  1886. #endif //HAVE_MMX2

  1887. #else //HAVE_MMX

  1888.     int x, y;

  1889.     src+= 4*stride;

  1890.     // FIXME - there should be a way to do a few columns in parallel like w/mmx

  1891.     for(x=0; x<8; x++){

  1892.         uint8_t *colsrc = src;

  1893.         for (y=0; y<4; y++){

  1894.             int a, b, c, d, e, f;

  1895.             a = colsrc[0       ];

  1896.             b = colsrc[stride  ];

  1897.             c = colsrc[stride*2];

  1898.             d = (a-b)>>31;

  1899.             e = (b-c)>>31;

  1900.             f = (c-a)>>31;

  1901.             colsrc[stride  ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));

  1902.             colsrc += stride*2;

  1903.         }

  1904.         src++;

  1905.     }

  1906. #endif //HAVE_MMX

  1907. }

  1908. 

  1909. #if HAVE_MMX

  1910. /**

  1911.  * transposes and shift the given 8x8 Block into dst1 and dst2

  1912.  */

  1913. static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)

  1914. {

  1915.     __asm__(

  1916.         "lea (%0, %1), %%"REG_a"                \n\t"

  1917. //      0       1       2       3       4       5       6       7       8       9

  1918. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1

  1919.         "movq (%0), %%mm0                       \n\t" // 12345678

  1920.         "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh

  1921.         "movq %%mm0, %%mm2                      \n\t" // 12345678

  1922.         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d

  1923.         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h

  1924. 

  1925.         "movq (%%"REG_a", %1), %%mm1            \n\t"

  1926.         "movq (%%"REG_a", %1, 2), %%mm3         \n\t"

  1927.         "movq %%mm1, %%mm4                      \n\t"

  1928.         "punpcklbw %%mm3, %%mm1                 \n\t"

  1929.         "punpckhbw %%mm3, %%mm4                 \n\t"

  1930. 

  1931.         "movq %%mm0, %%mm3                      \n\t"

  1932.         "punpcklwd %%mm1, %%mm0                 \n\t"

  1933.         "punpckhwd %%mm1, %%mm3                 \n\t"

  1934.         "movq %%mm2, %%mm1                      \n\t"

  1935.         "punpcklwd %%mm4, %%mm2                 \n\t"

  1936.         "punpckhwd %%mm4, %%mm1                 \n\t"

  1937. 

  1938.         "movd %%mm0, 128(%2)                    \n\t"

  1939.         "psrlq $32, %%mm0                       \n\t"

  1940.         "movd %%mm0, 144(%2)                    \n\t"

  1941.         "movd %%mm3, 160(%2)                    \n\t"

  1942.         "psrlq $32, %%mm3                       \n\t"

  1943.         "movd %%mm3, 176(%2)                    \n\t"

  1944.         "movd %%mm3, 48(%3)                     \n\t"

  1945.         "movd %%mm2, 192(%2)                    \n\t"

  1946.         "movd %%mm2, 64(%3)                     \n\t"

  1947.         "psrlq $32, %%mm2                       \n\t"

  1948.         "movd %%mm2, 80(%3)                     \n\t"

  1949.         "movd %%mm1, 96(%3)                     \n\t"

  1950.         "psrlq $32, %%mm1                       \n\t"

  1951.         "movd %%mm1, 112(%3)                    \n\t"

  1952. 

  1953.         "lea (%%"REG_a", %1, 4), %%"REG_a"      \n\t"

  1954. 

  1955.         "movq (%0, %1, 4), %%mm0                \n\t" // 12345678

  1956.         "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh

  1957.         "movq %%mm0, %%mm2                      \n\t" // 12345678

  1958.         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d

  1959.         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h

  1960. 

  1961.         "movq (%%"REG_a", %1), %%mm1            \n\t"

  1962.         "movq (%%"REG_a", %1, 2), %%mm3         \n\t"

  1963.         "movq %%mm1, %%mm4                      \n\t"

  1964.         "punpcklbw %%mm3, %%mm1                 \n\t"

  1965.         "punpckhbw %%mm3, %%mm4                 \n\t"

  1966. 

  1967.         "movq %%mm0, %%mm3                      \n\t"

  1968.         "punpcklwd %%mm1, %%mm0                 \n\t"

  1969.         "punpckhwd %%mm1, %%mm3                 \n\t"

  1970.         "movq %%mm2, %%mm1                      \n\t"

  1971.         "punpcklwd %%mm4, %%mm2                 \n\t"

  1972.         "punpckhwd %%mm4, %%mm1                 \n\t"

  1973. 

  1974.         "movd %%mm0, 132(%2)                    \n\t"

  1975.         "psrlq $32, %%mm0                       \n\t"

  1976.         "movd %%mm0, 148(%2)                    \n\t"

  1977.         "movd %%mm3, 164(%2)                    \n\t"

  1978.         "psrlq $32, %%mm3                       \n\t"

  1979.         "movd %%mm3, 180(%2)                    \n\t"

  1980.         "movd %%mm3, 52(%3)                     \n\t"

  1981.         "movd %%mm2, 196(%2)                    \n\t"

  1982.         "movd %%mm2, 68(%3)                     \n\t"

  1983.         "psrlq $32, %%mm2                       \n\t"

  1984.         "movd %%mm2, 84(%3)                     \n\t"

  1985.         "movd %%mm1, 100(%3)                    \n\t"

  1986.         "psrlq $32, %%mm1                       \n\t"

  1987.         "movd %%mm1, 116(%3)                    \n\t"

  1988. 

  1989. 

  1990.         :: "r" (src), "r" ((x86_reg)srcStride), "r" (dst1), "r" (dst2)

  1991.         : "%"REG_a

  1992.     );

  1993. }

  1994. 

  1995. /**

  1996.  * transposes the given 8x8 block

  1997.  */

  1998. static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)

  1999. {

  2000.     __asm__(

  2001.         "lea (%0, %1), %%"REG_a"                \n\t"

  2002.         "lea (%%"REG_a",%1,4), %%"REG_d"        \n\t"

  2003. //      0       1       2       3       4       5       6       7       8       9

  2004. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1

  2005.         "movq (%2), %%mm0                       \n\t" // 12345678

  2006.         "movq 16(%2), %%mm1                     \n\t" // abcdefgh

  2007.         "movq %%mm0, %%mm2                      \n\t" // 12345678

  2008.         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d

  2009.         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h

  2010. 

  2011.         "movq 32(%2), %%mm1                     \n\t"

  2012.         "movq 48(%2), %%mm3                     \n\t"

  2013.         "movq %%mm1, %%mm4                      \n\t"

  2014.         "punpcklbw %%mm3, %%mm1                 \n\t"

  2015.         "punpckhbw %%mm3, %%mm4                 \n\t"

  2016. 

  2017.         "movq %%mm0, %%mm3                      \n\t"

  2018.         "punpcklwd %%mm1, %%mm0                 \n\t"

  2019.         "punpckhwd %%mm1, %%mm3                 \n\t"

  2020.         "movq %%mm2, %%mm1                      \n\t"

  2021.         "punpcklwd %%mm4, %%mm2                 \n\t"

  2022.         "punpckhwd %%mm4, %%mm1                 \n\t"

  2023. 

  2024.         "movd %%mm0, (%0)                       \n\t"

  2025.         "psrlq $32, %%mm0                       \n\t"

  2026.         "movd %%mm0, (%%"REG_a")                \n\t"

  2027.         "movd %%mm3, (%%"REG_a", %1)            \n\t"

  2028.         "psrlq $32, %%mm3                       \n\t"

  2029.         "movd %%mm3, (%%"REG_a", %1, 2)         \n\t"

  2030.         "movd %%mm2, (%0, %1, 4)                \n\t"

  2031.         "psrlq $32, %%mm2                       \n\t"

  2032.         "movd %%mm2, (%%"REG_d")                \n\t"

  2033.         "movd %%mm1, (%%"REG_d", %1)            \n\t"

  2034.         "psrlq $32, %%mm1                       \n\t"

  2035.         "movd %%mm1, (%%"REG_d", %1, 2)         \n\t"

  2036. 

  2037. 

  2038.         "movq 64(%2), %%mm0                     \n\t" // 12345678

  2039.         "movq 80(%2), %%mm1                     \n\t" // abcdefgh

  2040.         "movq %%mm0, %%mm2                      \n\t" // 12345678

  2041.         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d

  2042.         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h

  2043. 

  2044.         "movq 96(%2), %%mm1                     \n\t"

  2045.         "movq 112(%2), %%mm3                    \n\t"

  2046.         "movq %%mm1, %%mm4                      \n\t"

  2047.         "punpcklbw %%mm3, %%mm1                 \n\t"

  2048.         "punpckhbw %%mm3, %%mm4                 \n\t"

  2049. 

  2050.         "movq %%mm0, %%mm3                      \n\t"

  2051.         "punpcklwd %%mm1, %%mm0                 \n\t"

  2052.         "punpckhwd %%mm1, %%mm3                 \n\t"

  2053.         "movq %%mm2, %%mm1                      \n\t"

  2054.         "punpcklwd %%mm4, %%mm2                 \n\t"

  2055.         "punpckhwd %%mm4, %%mm1                 \n\t"

  2056. 

  2057.         "movd %%mm0, 4(%0)                      \n\t"

  2058.         "psrlq $32, %%mm0                       \n\t"

  2059.         "movd %%mm0, 4(%%"REG_a")               \n\t"

  2060.         "movd %%mm3, 4(%%"REG_a", %1)           \n\t"

  2061.         "psrlq $32, %%mm3                       \n\t"

  2062.         "movd %%mm3, 4(%%"REG_a", %1, 2)        \n\t"

  2063.         "movd %%mm2, 4(%0, %1, 4)               \n\t"

  2064.         "psrlq $32, %%mm2                       \n\t"

  2065.         "movd %%mm2, 4(%%"REG_d")               \n\t"

  2066.         "movd %%mm1, 4(%%"REG_d", %1)           \n\t"

  2067.         "psrlq $32, %%mm1                       \n\t"

  2068.         "movd %%mm1, 4(%%"REG_d", %1, 2)        \n\t"

  2069. 

  2070.         :: "r" (dst), "r" ((x86_reg)dstStride), "r" (src)

  2071.         : "%"REG_a, "%"REG_d

  2072.     );

  2073. }

  2074. #endif //HAVE_MMX

  2075. //static long test=0;

  2076. 

  2077. #if !HAVE_ALTIVEC

  2078. static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,

  2079.                                     uint8_t *tempBlurred, uint32_t *tempBlurredPast, int *maxNoise)

  2080. {

  2081.     // to save a register (FIXME do this outside of the loops)

  2082.     tempBlurredPast[127]= maxNoise[0];

  2083.     tempBlurredPast[128]= maxNoise[1];

  2084.     tempBlurredPast[129]= maxNoise[2];

  2085. 

  2086. #define FAST_L2_DIFF

  2087. //#define L1_DIFF //u should change the thresholds too if u try that one

  2088. #if HAVE_MMX2 || HAVE_AMD3DNOW

  2089.     __asm__ volatile(

  2090.         "lea (%2, %2, 2), %%"REG_a"             \n\t" // 3*stride

  2091.         "lea (%2, %2, 4), %%"REG_d"             \n\t" // 5*stride

  2092.         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride

  2093. //      0       1       2       3       4       5       6       7       8       9

  2094. //      %x      %x+%2   %x+2%2  %x+eax  %x+4%2  %x+edx  %x+2eax %x+ecx  %x+8%2

  2095. //FIXME reorder?

  2096. #ifdef L1_DIFF //needs mmx2

  2097.         "movq (%0), %%mm0                       \n\t" // L0

  2098.         "psadbw (%1), %%mm0                     \n\t" // |L0-R0|

  2099.         "movq (%0, %2), %%mm1                   \n\t" // L1

  2100.         "psadbw (%1, %2), %%mm1                 \n\t" // |L1-R1|

  2101.         "movq (%0, %2, 2), %%mm2                \n\t" // L2

  2102.         "psadbw (%1, %2, 2), %%mm2              \n\t" // |L2-R2|

  2103.         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3

  2104.         "psadbw (%1, %%"REG_a"), %%mm3          \n\t" // |L3-R3|

  2105. 

  2106.         "movq (%0, %2, 4), %%mm4                \n\t" // L4

  2107.         "paddw %%mm1, %%mm0                     \n\t"

  2108.         "psadbw (%1, %2, 4), %%mm4              \n\t" // |L4-R4|

  2109.         "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5

  2110.         "paddw %%mm2, %%mm0                     \n\t"

  2111.         "psadbw (%1, %%"REG_d"), %%mm5          \n\t" // |L5-R5|

  2112.         "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6

  2113.         "paddw %%mm3, %%mm0                     \n\t"

  2114.         "psadbw (%1, %%"REG_a", 2), %%mm6       \n\t" // |L6-R6|

  2115.         "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7

  2116.         "paddw %%mm4, %%mm0                     \n\t"

  2117.         "psadbw (%1, %%"REG_c"), %%mm7          \n\t" // |L7-R7|

  2118.         "paddw %%mm5, %%mm6                     \n\t"

  2119.         "paddw %%mm7, %%mm6                     \n\t"

  2120.         "paddw %%mm6, %%mm0                     \n\t"

  2121. #else //L1_DIFF

  2122. #if defined (FAST_L2_DIFF)

  2123.         "pcmpeqb %%mm7, %%mm7                   \n\t"

  2124.         "movq "MANGLE(b80)", %%mm6              \n\t"

  2125.         "pxor %%mm0, %%mm0                      \n\t"

  2126. #define REAL_L2_DIFF_CORE(a, b)\

  2127.         "movq " #a ", %%mm5                     \n\t"\

  2128.         "movq " #b ", %%mm2                     \n\t"\

  2129.         "pxor %%mm7, %%mm2                      \n\t"\

  2130.         PAVGB(%%mm2, %%mm5)\

  2131.         "paddb %%mm6, %%mm5                     \n\t"\

  2132.         "movq %%mm5, %%mm2                      \n\t"\

  2133.         "psllw $8, %%mm5                        \n\t"\

  2134.         "pmaddwd %%mm5, %%mm5                   \n\t"\

  2135.         "pmaddwd %%mm2, %%mm2                   \n\t"\

  2136.         "paddd %%mm2, %%mm5                     \n\t"\

  2137.         "psrld $14, %%mm5                       \n\t"\

  2138.         "paddd %%mm5, %%mm0                     \n\t"

  2139. 

  2140. #else //defined (FAST_L2_DIFF)

  2141.         "pxor %%mm7, %%mm7                      \n\t"

  2142.         "pxor %%mm0, %%mm0                      \n\t"

  2143. #define REAL_L2_DIFF_CORE(a, b)\

  2144.         "movq " #a ", %%mm5                     \n\t"\

  2145.         "movq " #b ", %%mm2                     \n\t"\

  2146.         "movq %%mm5, %%mm1                      \n\t"\

  2147.         "movq %%mm2, %%mm3                      \n\t"\

  2148.         "punpcklbw %%mm7, %%mm5                 \n\t"\

  2149.         "punpckhbw %%mm7, %%mm1                 \n\t"\

  2150.         "punpcklbw %%mm7, %%mm2                 \n\t"\

  2151.         "punpckhbw %%mm7, %%mm3                 \n\t"\

  2152.         "psubw %%mm2, %%mm5                     \n\t"\

  2153.         "psubw %%mm3, %%mm1                     \n\t"\

  2154.         "pmaddwd %%mm5, %%mm5                   \n\t"\

  2155.         "pmaddwd %%mm1, %%mm1                   \n\t"\

  2156.         "paddd %%mm1, %%mm5                     \n\t"\

  2157.         "paddd %%mm5, %%mm0                     \n\t"

  2158. 

  2159. #endif //defined (FAST_L2_DIFF)

  2160. 

  2161. #define L2_DIFF_CORE(a, b)  REAL_L2_DIFF_CORE(a, b)

  2162. 

  2163. L2_DIFF_CORE((%0)          , (%1))

  2164. L2_DIFF_CORE((%0, %2)      , (%1, %2))

  2165. L2_DIFF_CORE((%0, %2, 2)   , (%1, %2, 2))

  2166. L2_DIFF_CORE((%0, %%REGa)  , (%1, %%REGa))

  2167. L2_DIFF_CORE((%0, %2, 4)   , (%1, %2, 4))

  2168. L2_DIFF_CORE((%0, %%REGd)  , (%1, %%REGd))

  2169. L2_DIFF_CORE((%0, %%REGa,2), (%1, %%REGa,2))

  2170. L2_DIFF_CORE((%0, %%REGc)  , (%1, %%REGc))

  2171. 

  2172. #endif //L1_DIFF

  2173. 

  2174.         "movq %%mm0, %%mm4                      \n\t"

  2175.         "psrlq $32, %%mm0                       \n\t"

  2176.         "paddd %%mm0, %%mm4                     \n\t"

  2177.         "movd %%mm4, %%ecx                      \n\t"

  2178.         "shll $2, %%ecx                         \n\t"

  2179.         "mov %3, %%"REG_d"                      \n\t"

  2180.         "addl -4(%%"REG_d"), %%ecx              \n\t"

  2181.         "addl 4(%%"REG_d"), %%ecx               \n\t"

  2182.         "addl -1024(%%"REG_d"), %%ecx           \n\t"

  2183.         "addl $4, %%ecx                         \n\t"

  2184.         "addl 1024(%%"REG_d"), %%ecx            \n\t"

  2185.         "shrl $3, %%ecx                         \n\t"

  2186.         "movl %%ecx, (%%"REG_d")                \n\t"

  2187. 

  2188. //        "mov %3, %%"REG_c"                      \n\t"

  2189. //        "mov %%"REG_c", test                    \n\t"

  2190. //        "jmp 4f                                 \n\t"

  2191.         "cmpl 512(%%"REG_d"), %%ecx             \n\t"

  2192.         " jb 2f                                 \n\t"

  2193.         "cmpl 516(%%"REG_d"), %%ecx             \n\t"

  2194.         " jb 1f                                 \n\t"

  2195. 

  2196.         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride

  2197.         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride

  2198.         "movq (%0), %%mm0                       \n\t" // L0

  2199.         "movq (%0, %2), %%mm1                   \n\t" // L1

  2200.         "movq (%0, %2, 2), %%mm2                \n\t" // L2

  2201.         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3

  2202.         "movq (%0, %2, 4), %%mm4                \n\t" // L4

  2203.         "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5

  2204.         "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6

  2205.         "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7

  2206.         "movq %%mm0, (%1)                       \n\t" // L0

  2207.         "movq %%mm1, (%1, %2)                   \n\t" // L1

  2208.         "movq %%mm2, (%1, %2, 2)                \n\t" // L2

  2209.         "movq %%mm3, (%1, %%"REG_a")            \n\t" // L3

  2210.         "movq %%mm4, (%1, %2, 4)                \n\t" // L4

  2211.         "movq %%mm5, (%1, %%"REG_d")            \n\t" // L5

  2212.         "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // L6

  2213.         "movq %%mm7, (%1, %%"REG_c")            \n\t" // L7

  2214.         "jmp 4f                                 \n\t"

  2215. 

  2216.         "1:                                     \n\t"

  2217.         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride

  2218.         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride

  2219.         "movq (%0), %%mm0                       \n\t" // L0

  2220.         PAVGB((%1), %%mm0)                            // L0

  2221.         "movq (%0, %2), %%mm1                   \n\t" // L1

  2222.         PAVGB((%1, %2), %%mm1)                        // L1

  2223.         "movq (%0, %2, 2), %%mm2                \n\t" // L2

  2224.         PAVGB((%1, %2, 2), %%mm2)                     // L2

  2225.         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3

  2226.         PAVGB((%1, %%REGa), %%mm3)                    // L3

  2227.         "movq (%0, %2, 4), %%mm4                \n\t" // L4

  2228.         PAVGB((%1, %2, 4), %%mm4)                     // L4

  2229.         "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5

  2230.         PAVGB((%1, %%REGd), %%mm5)                    // L5

  2231.         "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6

  2232.         PAVGB((%1, %%REGa, 2), %%mm6)                 // L6

  2233.         "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7

  2234.         PAVGB((%1, %%REGc), %%mm7)                    // L7

  2235.         "movq %%mm0, (%1)                       \n\t" // R0

  2236.         "movq %%mm1, (%1, %2)                   \n\t" // R1

  2237.         "movq %%mm2, (%1, %2, 2)                \n\t" // R2

  2238.         "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3

  2239.         "movq %%mm4, (%1, %2, 4)                \n\t" // R4

  2240.         "movq %%mm5, (%1, %%"REG_d")            \n\t" // R5

  2241.         "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // R6

  2242.         "movq %%mm7, (%1, %%"REG_c")            \n\t" // R7

  2243.         "movq %%mm0, (%0)                       \n\t" // L0

  2244.         "movq %%mm1, (%0, %2)                   \n\t" // L1

  2245.         "movq %%mm2, (%0, %2, 2)                \n\t" // L2

  2246.         "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3

  2247.         "movq %%mm4, (%0, %2, 4)                \n\t" // L4

  2248.         "movq %%mm5, (%0, %%"REG_d")            \n\t" // L5

  2249.         "movq %%mm6, (%0, %%"REG_a", 2)         \n\t" // L6

  2250.         "movq %%mm7, (%0, %%"REG_c")            \n\t" // L7

  2251.         "jmp 4f                                 \n\t"

  2252. 

  2253.         "2:                                     \n\t"

  2254.         "cmpl 508(%%"REG_d"), %%ecx             \n\t"

  2255.         " jb 3f                                 \n\t"

  2256. 

  2257.         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride

  2258.         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride

  2259.         "movq (%0), %%mm0                       \n\t" // L0

  2260.         "movq (%0, %2), %%mm1                   \n\t" // L1

  2261.         "movq (%0, %2, 2), %%mm2                \n\t" // L2

  2262.         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3

  2263.         "movq (%1), %%mm4                       \n\t" // R0

  2264.         "movq (%1, %2), %%mm5                   \n\t" // R1

  2265.         "movq (%1, %2, 2), %%mm6                \n\t" // R2

  2266.         "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3

  2267.         PAVGB(%%mm4, %%mm0)

  2268.         PAVGB(%%mm5, %%mm1)

  2269.         PAVGB(%%mm6, %%mm2)

  2270.         PAVGB(%%mm7, %%mm3)

  2271.         PAVGB(%%mm4, %%mm0)

  2272.         PAVGB(%%mm5, %%mm1)

  2273.         PAVGB(%%mm6, %%mm2)

  2274.         PAVGB(%%mm7, %%mm3)

  2275.         "movq %%mm0, (%1)                       \n\t" // R0

  2276.         "movq %%mm1, (%1, %2)                   \n\t" // R1

  2277.         "movq %%mm2, (%1, %2, 2)                \n\t" // R2

  2278.         "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3

  2279.         "movq %%mm0, (%0)                       \n\t" // L0

  2280.         "movq %%mm1, (%0, %2)                   \n\t" // L1

  2281.         "movq %%mm2, (%0, %2, 2)                \n\t" // L2

  2282.         "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3

  2283. 

  2284.         "movq (%0, %2, 4), %%mm0                \n\t" // L4

  2285.         "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5

  2286.         "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6

  2287.         "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7

  2288.         "movq (%1, %2, 4), %%mm4                \n\t" // R4

  2289.         "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5

  2290.         "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6

  2291.         "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7

  2292.         PAVGB(%%mm4, %%mm0)

  2293.         PAVGB(%%mm5, %%mm1)

  2294.         PAVGB(%%mm6, %%mm2)

  2295.         PAVGB(%%mm7, %%mm3)

  2296.         PAVGB(%%mm4, %%mm0)

  2297.         PAVGB(%%mm5, %%mm1)

  2298.         PAVGB(%%mm6, %%mm2)

  2299.         PAVGB(%%mm7, %%mm3)

  2300.         "movq %%mm0, (%1, %2, 4)                \n\t" // R4

  2301.         "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5

  2302.         "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6

  2303.         "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7

  2304.         "movq %%mm0, (%0, %2, 4)                \n\t" // L4

  2305.         "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5

  2306.         "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6

  2307.         "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7

  2308.         "jmp 4f                                 \n\t"

  2309. 

  2310.         "3:                                     \n\t"

  2311.         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride

  2312.         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride

  2313.         "movq (%0), %%mm0                       \n\t" // L0

  2314.         "movq (%0, %2), %%mm1                   \n\t" // L1

  2315.         "movq (%0, %2, 2), %%mm2                \n\t" // L2

  2316.         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3

  2317.         "movq (%1), %%mm4                       \n\t" // R0

  2318.         "movq (%1, %2), %%mm5                   \n\t" // R1

  2319.         "movq (%1, %2, 2), %%mm6                \n\t" // R2

  2320.         "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3

  2321.         PAVGB(%%mm4, %%mm0)

  2322.         PAVGB(%%mm5, %%mm1)

  2323.         PAVGB(%%mm6, %%mm2)

  2324.         PAVGB(%%mm7, %%mm3)

  2325.         PAVGB(%%mm4, %%mm0)

  2326.         PAVGB(%%mm5, %%mm1)

  2327.         PAVGB(%%mm6, %%mm2)

  2328.         PAVGB(%%mm7, %%mm3)

  2329.         PAVGB(%%mm4, %%mm0)

  2330.         PAVGB(%%mm5, %%mm1)

  2331.         PAVGB(%%mm6, %%mm2)

  2332.         PAVGB(%%mm7, %%mm3)

  2333.         "movq %%mm0, (%1)                       \n\t" // R0

  2334.         "movq %%mm1, (%1, %2)                   \n\t" // R1

  2335.         "movq %%mm2, (%1, %2, 2)                \n\t" // R2

  2336.         "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3

  2337.         "movq %%mm0, (%0)                       \n\t" // L0

  2338.         "movq %%mm1, (%0, %2)                   \n\t" // L1

  2339.         "movq %%mm2, (%0, %2, 2)                \n\t" // L2

  2340.         "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3

  2341. 

  2342.         "movq (%0, %2, 4), %%mm0                \n\t" // L4

  2343.         "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5

  2344.         "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6

  2345.         "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7

  2346.         "movq (%1, %2, 4), %%mm4                \n\t" // R4

  2347.         "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5

  2348.         "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6

  2349.         "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7

  2350.         PAVGB(%%mm4, %%mm0)

  2351.         PAVGB(%%mm5, %%mm1)

  2352.         PAVGB(%%mm6, %%mm2)

  2353.         PAVGB(%%mm7, %%mm3)

  2354.         PAVGB(%%mm4, %%mm0)

  2355.         PAVGB(%%mm5, %%mm1)

  2356.         PAVGB(%%mm6, %%mm2)

  2357.         PAVGB(%%mm7, %%mm3)

  2358.         PAVGB(%%mm4, %%mm0)

  2359.         PAVGB(%%mm5, %%mm1)

  2360.         PAVGB(%%mm6, %%mm2)

  2361.         PAVGB(%%mm7, %%mm3)

  2362.         "movq %%mm0, (%1, %2, 4)                \n\t" // R4

  2363.         "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5

  2364.         "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6

  2365.         "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7

  2366.         "movq %%mm0, (%0, %2, 4)                \n\t" // L4

  2367.         "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5

  2368.         "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6

  2369.         "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7

  2370. 

  2371.         "4:                                     \n\t"

  2372. 

  2373.         :: "r" (src), "r" (tempBlurred), "r"((x86_reg)stride), "m" (tempBlurredPast)

  2374.         : "%"REG_a, "%"REG_d, "%"REG_c, "memory"

  2375.     );

  2376. #else //HAVE_MMX2 || HAVE_AMD3DNOW

  2377. {

  2378.     int y;

  2379.     int d=0;

  2380. //    int sysd=0;

  2381.     int i;

  2382. 

  2383.     for(y=0; y<8; y++){

  2384.         int x;

  2385.         for(x=0; x<8; x++){

  2386.             int ref= tempBlurred[ x + y*stride ];

  2387.             int cur= src[ x + y*stride ];

  2388.             int d1=ref - cur;

  2389. //            if(x==0 || x==7) d1+= d1>>1;

  2390. //            if(y==0 || y==7) d1+= d1>>1;

  2391. //            d+= FFABS(d1);

  2392.             d+= d1*d1;

  2393. //            sysd+= d1;

  2394.         }

  2395.     }

  2396.     i=d;

  2397.     d=  (

  2398.         4*d

  2399.         +(*(tempBlurredPast-256))

  2400.         +(*(tempBlurredPast-1))+ (*(tempBlurredPast+1))

  2401.         +(*(tempBlurredPast+256))

  2402.         +4)>>3;

  2403.     *tempBlurredPast=i;

  2404. //    ((*tempBlurredPast)*3 + d + 2)>>2;

  2405. 

  2406. /*

  2407. Switch between

  2408.  1  0  0  0  0  0  0  (0)

  2409. 64 32 16  8  4  2  1  (1)

  2410. 64 48 36 27 20 15 11 (33) (approx)

  2411. 64 56 49 43 37 33 29 (200) (approx)

  2412. */

  2413.     if(d > maxNoise[1]){

  2414.         if(d < maxNoise[2]){

  2415.             for(y=0; y<8; y++){

  2416.                 int x;

  2417.                 for(x=0; x<8; x++){

  2418.                     int ref= tempBlurred[ x + y*stride ];

  2419.                     int cur= src[ x + y*stride ];

  2420.                     tempBlurred[ x + y*stride ]=

  2421.                     src[ x + y*stride ]=

  2422.                         (ref + cur + 1)>>1;

  2423.                 }

  2424.             }

  2425.         }else{

  2426.             for(y=0; y<8; y++){

  2427.                 int x;

  2428.                 for(x=0; x<8; x++){

  2429.                     tempBlurred[ x + y*stride ]= src[ x + y*stride ];

  2430.                 }

  2431.             }

  2432.         }

  2433.     }else{

  2434.         if(d < maxNoise[0]){

  2435.             for(y=0; y<8; y++){

  2436.                 int x;

  2437.                 for(x=0; x<8; x++){

  2438.                     int ref= tempBlurred[ x + y*stride ];

  2439.                     int cur= src[ x + y*stride ];

  2440.                     tempBlurred[ x + y*stride ]=

  2441.                     src[ x + y*stride ]=

  2442.                         (ref*7 + cur + 4)>>3;

  2443.                 }

  2444.             }

  2445.         }else{

  2446.             for(y=0; y<8; y++){

  2447.                 int x;

  2448.                 for(x=0; x<8; x++){

  2449.                     int ref= tempBlurred[ x + y*stride ];

  2450.                     int cur= src[ x + y*stride ];

  2451.                     tempBlurred[ x + y*stride ]=

  2452.                     src[ x + y*stride ]=

  2453.                         (ref*3 + cur + 2)>>2;

  2454.                 }

  2455.             }

  2456.         }

  2457.     }

  2458. }

  2459. #endif //HAVE_MMX2 || HAVE_AMD3DNOW

  2460. }

  2461. #endif //HAVE_ALTIVEC

  2462. 

  2463. #if HAVE_MMX

  2464. /**

  2465.  * accurate deblock filter

  2466.  */

  2467. static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, PPContext *c){

  2468.     int64_t dc_mask, eq_mask, both_masks;

  2469.     int64_t sums[10*8*2];

  2470.     src+= step*3; // src points to begin of the 8x8 Block

  2471. //START_TIMER

  2472.     __asm__ volatile(

  2473.         "movq %0, %%mm7                         \n\t"

  2474.         "movq %1, %%mm6                         \n\t"

  2475.         : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])

  2476.         );

  2477. 

  2478.     __asm__ volatile(

  2479.         "lea (%2, %3), %%"REG_a"                \n\t"

  2480. //      0       1       2       3       4       5       6       7       8       9

  2481. //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2

  2482. 

  2483.         "movq (%2), %%mm0                       \n\t"

  2484.         "movq (%%"REG_a"), %%mm1                \n\t"

  2485.         "movq %%mm1, %%mm3                      \n\t"

  2486.         "movq %%mm1, %%mm4                      \n\t"

  2487.         "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece

  2488.         "paddb %%mm7, %%mm0                     \n\t"

  2489.         "pcmpgtb %%mm6, %%mm0                   \n\t"

  2490. 

  2491.         "movq (%%"REG_a",%3), %%mm2             \n\t"

  2492.         PMAXUB(%%mm2, %%mm4)

  2493.         PMINUB(%%mm2, %%mm3, %%mm5)

  2494.         "psubb %%mm2, %%mm1                     \n\t"

  2495.         "paddb %%mm7, %%mm1                     \n\t"

  2496.         "pcmpgtb %%mm6, %%mm1                   \n\t"

  2497.         "paddb %%mm1, %%mm0                     \n\t"

  2498. 

  2499.         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"

  2500.         PMAXUB(%%mm1, %%mm4)

  2501.         PMINUB(%%mm1, %%mm3, %%mm5)

  2502.         "psubb %%mm1, %%mm2                     \n\t"

  2503.         "paddb %%mm7, %%mm2                     \n\t"

  2504.         "pcmpgtb %%mm6, %%mm2                   \n\t"

  2505.         "paddb %%mm2, %%mm0                     \n\t"

  2506. 

  2507.         "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"

  2508. 

  2509.         "movq (%2, %3, 4), %%mm2                \n\t"

  2510.         PMAXUB(%%mm2, %%mm4)

  2511.         PMINUB(%%mm2, %%mm3, %%mm5)

  2512.         "psubb %%mm2, %%mm1                     \n\t"

  2513.         "paddb %%mm7, %%mm1                     \n\t"

  2514.         "pcmpgtb %%mm6, %%mm1                   \n\t"

  2515.         "paddb %%mm1, %%mm0                     \n\t"

  2516. 

  2517.         "movq (%%"REG_a"), %%mm1                \n\t"

  2518.         PMAXUB(%%mm1, %%mm4)

  2519.         PMINUB(%%mm1, %%mm3, %%mm5)

  2520.         "psubb %%mm1, %%mm2                     \n\t"

  2521.         "paddb %%mm7, %%mm2                     \n\t"

  2522.         "pcmpgtb %%mm6, %%mm2                   \n\t"

  2523.         "paddb %%mm2, %%mm0                     \n\t"

  2524. 

  2525.         "movq (%%"REG_a", %3), %%mm2            \n\t"

  2526.         PMAXUB(%%mm2, %%mm4)

  2527.         PMINUB(%%mm2, %%mm3, %%mm5)

  2528.         "psubb %%mm2, %%mm1                     \n\t"

  2529.         "paddb %%mm7, %%mm1                     \n\t"

  2530.         "pcmpgtb %%mm6, %%mm1                   \n\t"

  2531.         "paddb %%mm1, %%mm0                     \n\t"

  2532. 

  2533.         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"

  2534.         PMAXUB(%%mm1, %%mm4)

  2535.         PMINUB(%%mm1, %%mm3, %%mm5)

  2536.         "psubb %%mm1, %%mm2                     \n\t"

  2537.         "paddb %%mm7, %%mm2                     \n\t"

  2538.         "pcmpgtb %%mm6, %%mm2                   \n\t"

  2539.         "paddb %%mm2, %%mm0                     \n\t"

  2540. 

  2541.         "movq (%2, %3, 8), %%mm2                \n\t"

  2542.         PMAXUB(%%mm2, %%mm4)

  2543.         PMINUB(%%mm2, %%mm3, %%mm5)

  2544.         "psubb %%mm2, %%mm1                     \n\t"

  2545.         "paddb %%mm7, %%mm1                     \n\t"

  2546.         "pcmpgtb %%mm6, %%mm1                   \n\t"

  2547.         "paddb %%mm1, %%mm0                     \n\t"

  2548. 

  2549.         "movq (%%"REG_a", %3, 4), %%mm1         \n\t"

  2550.         "psubb %%mm1, %%mm2                     \n\t"

  2551.         "paddb %%mm7, %%mm2                     \n\t"

  2552.         "pcmpgtb %%mm6, %%mm2                   \n\t"

  2553.         "paddb %%mm2, %%mm0                     \n\t"

  2554.         "psubusb %%mm3, %%mm4                   \n\t"

  2555. 

  2556.         "pxor %%mm6, %%mm6                      \n\t"

  2557.         "movq %4, %%mm7                         \n\t" // QP,..., QP

  2558.         "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP

  2559.         "psubusb %%mm4, %%mm7                   \n\t" // Diff >=2QP -> 0

  2560.         "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0

  2561.         "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0

  2562.         "movq %%mm7, %1                         \n\t"

  2563. 

  2564.         "movq %5, %%mm7                         \n\t"

  2565.         "punpcklbw %%mm7, %%mm7                 \n\t"

  2566.         "punpcklbw %%mm7, %%mm7                 \n\t"

  2567.         "punpcklbw %%mm7, %%mm7                 \n\t"

  2568.         "psubb %%mm0, %%mm6                     \n\t"

  2569.         "pcmpgtb %%mm7, %%mm6                   \n\t"

  2570.         "movq %%mm6, %0                         \n\t"

  2571. 

  2572.         : "=m" (eq_mask), "=m" (dc_mask)

  2573.         : "r" (src), "r" ((x86_reg)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)

  2574.         : "%"REG_a

  2575.     );

  2576. 

  2577.     both_masks = dc_mask & eq_mask;

  2578. 

  2579.     if(both_masks){

  2580.         x86_reg offset= -8*step;

  2581.         int64_t *temp_sums= sums;

  2582. 

  2583.         __asm__ volatile(

  2584.             "movq %2, %%mm0                         \n\t"  // QP,..., QP

  2585.             "pxor %%mm4, %%mm4                      \n\t"

  2586. 

  2587.             "movq (%0), %%mm6                       \n\t"

  2588.             "movq (%0, %1), %%mm5                   \n\t"

  2589.             "movq %%mm5, %%mm1                      \n\t"

  2590.             "movq %%mm6, %%mm2                      \n\t"

  2591.             "psubusb %%mm6, %%mm5                   \n\t"

  2592.             "psubusb %%mm1, %%mm2                   \n\t"

  2593.             "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines

  2594.             "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0

  2595.             "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF

  2596. 

  2597.             "pxor %%mm6, %%mm1                      \n\t"

  2598.             "pand %%mm0, %%mm1                      \n\t"

  2599.             "pxor %%mm1, %%mm6                      \n\t"

  2600.             // 0:QP  6:First

  2601. 

  2602.             "movq (%0, %1, 8), %%mm5                \n\t"

  2603.             "add %1, %0                             \n\t" // %0 points to line 1 not 0

  2604.             "movq (%0, %1, 8), %%mm7                \n\t"

  2605.             "movq %%mm5, %%mm1                      \n\t"

  2606.             "movq %%mm7, %%mm2                      \n\t"

  2607.             "psubusb %%mm7, %%mm5                   \n\t"

  2608.             "psubusb %%mm1, %%mm2                   \n\t"

  2609.             "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines

  2610.             "movq %2, %%mm0                         \n\t"  // QP,..., QP

  2611.             "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0

  2612.             "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF

  2613. 

  2614.             "pxor %%mm7, %%mm1                      \n\t"

  2615.             "pand %%mm0, %%mm1                      \n\t"

  2616.             "pxor %%mm1, %%mm7                      \n\t"

  2617. 

  2618.             "movq %%mm6, %%mm5                      \n\t"

  2619.             "punpckhbw %%mm4, %%mm6                 \n\t"

  2620.             "punpcklbw %%mm4, %%mm5                 \n\t"

  2621.             // 4:0 5/6:First 7:Last

  2622. 

  2623.             "movq %%mm5, %%mm0                      \n\t"

  2624.             "movq %%mm6, %%mm1                      \n\t"

  2625.             "psllw $2, %%mm0                        \n\t"

  2626.             "psllw $2, %%mm1                        \n\t"

  2627.             "paddw "MANGLE(w04)", %%mm0             \n\t"

  2628.             "paddw "MANGLE(w04)", %%mm1             \n\t"

  2629. 

  2630. #define NEXT\

  2631.             "movq (%0), %%mm2                       \n\t"\

  2632.             "movq (%0), %%mm3                       \n\t"\

  2633.             "add %1, %0                             \n\t"\

  2634.             "punpcklbw %%mm4, %%mm2                 \n\t"\

  2635.             "punpckhbw %%mm4, %%mm3                 \n\t"\

  2636.             "paddw %%mm2, %%mm0                     \n\t"\

  2637.             "paddw %%mm3, %%mm1                     \n\t"

  2638. 

  2639. #define PREV\

  2640.             "movq (%0), %%mm2                       \n\t"\

  2641.             "movq (%0), %%mm3                       \n\t"\

  2642.             "add %1, %0                             \n\t"\

  2643.             "punpcklbw %%mm4, %%mm2                 \n\t"\

  2644.             "punpckhbw %%mm4, %%mm3                 \n\t"\

  2645.             "psubw %%mm2, %%mm0                     \n\t"\

  2646.             "psubw %%mm3, %%mm1                     \n\t"

  2647. 

  2648. 

  2649.             NEXT //0

  2650.             NEXT //1

  2651.             NEXT //2

  2652.             "movq %%mm0, (%3)                       \n\t"

  2653.             "movq %%mm1, 8(%3)                      \n\t"

  2654. 

  2655.             NEXT //3

  2656.             "psubw %%mm5, %%mm0                     \n\t"

  2657.             "psubw %%mm6, %%mm1                     \n\t"

  2658.             "movq %%mm0, 16(%3)                     \n\t"

  2659.             "movq %%mm1, 24(%3)                     \n\t"

  2660. 

  2661.             NEXT //4

  2662.             "psubw %%mm5, %%mm0                     \n\t"

  2663.             "psubw %%mm6, %%mm1                     \n\t"

  2664.             "movq %%mm0, 32(%3)                     \n\t"

  2665.             "movq %%mm1, 40(%3)                     \n\t"

  2666. 

  2667.             NEXT //5

  2668.             "psubw %%mm5, %%mm0                     \n\t"

  2669.             "psubw %%mm6, %%mm1                     \n\t"

  2670.             "movq %%mm0, 48(%3)                     \n\t"

  2671.             "movq %%mm1, 56(%3)                     \n\t"

  2672. 

  2673.             NEXT //6

  2674.             "psubw %%mm5, %%mm0                     \n\t"

  2675.             "psubw %%mm6, %%mm1                     \n\t"

  2676.             "movq %%mm0, 64(%3)                     \n\t"

  2677.             "movq %%mm1, 72(%3)                     \n\t"

  2678. 

  2679.             "movq %%mm7, %%mm6                      \n\t"

  2680.             "punpckhbw %%mm4, %%mm7                 \n\t"

  2681.             "punpcklbw %%mm4, %%mm6                 \n\t"

  2682. 

  2683.             NEXT //7

  2684.             "mov %4, %0                             \n\t"

  2685.             "add %1, %0                             \n\t"

  2686.             PREV //0

  2687.             "movq %%mm0, 80(%3)                     \n\t"

  2688.             "movq %%mm1, 88(%3)                     \n\t"

  2689. 

  2690.             PREV //1

  2691.             "paddw %%mm6, %%mm0                     \n\t"

  2692.             "paddw %%mm7, %%mm1                     \n\t"

  2693.             "movq %%mm0, 96(%3)                     \n\t"

  2694.             "movq %%mm1, 104(%3)                    \n\t"

  2695. 

  2696.             PREV //2

  2697.             "paddw %%mm6, %%mm0                     \n\t"

  2698.             "paddw %%mm7, %%mm1                     \n\t"

  2699.             "movq %%mm0, 112(%3)                    \n\t"

  2700.             "movq %%mm1, 120(%3)                    \n\t"

  2701. 

  2702.             PREV //3

  2703.             "paddw %%mm6, %%mm0                     \n\t"

  2704.             "paddw %%mm7, %%mm1                     \n\t"

  2705.             "movq %%mm0, 128(%3)                    \n\t"

  2706.             "movq %%mm1, 136(%3)                    \n\t"

  2707. 

  2708.             PREV //4

  2709.             "paddw %%mm6, %%mm0                     \n\t"

  2710.             "paddw %%mm7, %%mm1                     \n\t"

  2711.             "movq %%mm0, 144(%3)                    \n\t"

  2712.             "movq %%mm1, 152(%3)                    \n\t"

  2713. 

  2714.             "mov %4, %0                             \n\t" //FIXME

  2715. 

  2716.             : "+&r"(src)

  2717.             : "r" ((x86_reg)step), "m" (c->pQPb), "r"(sums), "g"(src)

  2718.         );

  2719. 

  2720.         src+= step; // src points to begin of the 8x8 Block

  2721. 

  2722.         __asm__ volatile(

  2723.             "movq %4, %%mm6                         \n\t"

  2724.             "pcmpeqb %%mm5, %%mm5                   \n\t"

  2725.             "pxor %%mm6, %%mm5                      \n\t"

  2726.             "pxor %%mm7, %%mm7                      \n\t"

  2727. 

  2728.             "1:                                     \n\t"

  2729.             "movq (%1), %%mm0                       \n\t"

  2730.             "movq 8(%1), %%mm1                      \n\t"

  2731.             "paddw 32(%1), %%mm0                    \n\t"

  2732.             "paddw 40(%1), %%mm1                    \n\t"

  2733.             "movq (%0, %3), %%mm2                   \n\t"

  2734.             "movq %%mm2, %%mm3                      \n\t"

  2735.             "movq %%mm2, %%mm4                      \n\t"

  2736.             "punpcklbw %%mm7, %%mm2                 \n\t"

  2737.             "punpckhbw %%mm7, %%mm3                 \n\t"

  2738.             "paddw %%mm2, %%mm0                     \n\t"

  2739.             "paddw %%mm3, %%mm1                     \n\t"

  2740.             "paddw %%mm2, %%mm0                     \n\t"

  2741.             "paddw %%mm3, %%mm1                     \n\t"

  2742.             "psrlw $4, %%mm0                        \n\t"

  2743.             "psrlw $4, %%mm1                        \n\t"

  2744.             "packuswb %%mm1, %%mm0                  \n\t"

  2745.             "pand %%mm6, %%mm0                      \n\t"

  2746.             "pand %%mm5, %%mm4                      \n\t"

  2747.             "por %%mm4, %%mm0                       \n\t"

  2748.             "movq %%mm0, (%0, %3)                   \n\t"

  2749.             "add $16, %1                            \n\t"

  2750.             "add %2, %0                             \n\t"

  2751.             " js 1b                                 \n\t"

  2752. 

  2753.             : "+r"(offset), "+r"(temp_sums)

  2754.             : "r" ((x86_reg)step), "r"(src - offset), "m"(both_masks)

  2755.         );

  2756.     }else

  2757.         src+= step; // src points to begin of the 8x8 Block

  2758. 

  2759.     if(eq_mask != -1LL){

  2760.         uint8_t *temp_src= src;

  2761.         DECLARE_ALIGNED(8, uint64_t, tmp)[4]; // make space for 4 8-byte vars

  2762.         __asm__ volatile(

  2763.             "pxor %%mm7, %%mm7                      \n\t"

  2764. //      0       1       2       3       4       5       6       7       8       9

  2765. //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %1+8%1  ecx+4%1

  2766. 

  2767.             "movq (%0), %%mm0                       \n\t"

  2768.             "movq %%mm0, %%mm1                      \n\t"

  2769.             "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0

  2770.             "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0

  2771. 

  2772.             "movq (%0, %1), %%mm2                   \n\t"

  2773.             "lea (%0, %1, 2), %%"REG_a"             \n\t"

  2774.             "movq %%mm2, %%mm3                      \n\t"

  2775.             "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1

  2776.             "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1

  2777. 

  2778.             "movq (%%"REG_a"), %%mm4                \n\t"

  2779.             "movq %%mm4, %%mm5                      \n\t"

  2780.             "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2

  2781.             "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2

  2782. 

  2783.             "paddw %%mm0, %%mm0                     \n\t" // 2L0

  2784.             "paddw %%mm1, %%mm1                     \n\t" // 2H0

  2785.             "psubw %%mm4, %%mm2                     \n\t" // L1 - L2

  2786.             "psubw %%mm5, %%mm3                     \n\t" // H1 - H2

  2787.             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2

  2788.             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2

  2789. 

  2790.             "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2

  2791.             "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2

  2792.             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2

  2793.             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2

  2794. 

  2795.             "movq (%%"REG_a", %1), %%mm2            \n\t"

  2796.             "movq %%mm2, %%mm3                      \n\t"

  2797.             "punpcklbw %%mm7, %%mm2                 \n\t" // L3

  2798.             "punpckhbw %%mm7, %%mm3                 \n\t" // H3

  2799. 

  2800.             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3

  2801.             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3

  2802.             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3

  2803.             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3

  2804.             "movq %%mm0, (%4)                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3

  2805.             "movq %%mm1, 8(%4)                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3

  2806. 

  2807.             "movq (%%"REG_a", %1, 2), %%mm0         \n\t"

  2808.             "movq %%mm0, %%mm1                      \n\t"

  2809.             "punpcklbw %%mm7, %%mm0                 \n\t" // L4

  2810.             "punpckhbw %%mm7, %%mm1                 \n\t" // H4

  2811. 

  2812.             "psubw %%mm0, %%mm2                     \n\t" // L3 - L4

  2813.             "psubw %%mm1, %%mm3                     \n\t" // H3 - H4

  2814.             "movq %%mm2, 16(%4)                     \n\t" // L3 - L4

  2815.             "movq %%mm3, 24(%4)                     \n\t" // H3 - H4

  2816.             "paddw %%mm4, %%mm4                     \n\t" // 2L2

  2817.             "paddw %%mm5, %%mm5                     \n\t" // 2H2

  2818.             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4

  2819.             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4

  2820. 

  2821.             "lea (%%"REG_a", %1), %0                \n\t"

  2822.             "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4

  2823.             "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4

  2824.             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4

  2825.             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4

  2826. //50 opcodes so far

  2827.             "movq (%0, %1, 2), %%mm2                \n\t"

  2828.             "movq %%mm2, %%mm3                      \n\t"

  2829.             "punpcklbw %%mm7, %%mm2                 \n\t" // L5

  2830.             "punpckhbw %%mm7, %%mm3                 \n\t" // H5

  2831.             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5

  2832.             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5

  2833.             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5

  2834.             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5

  2835. 

  2836.             "movq (%%"REG_a", %1, 4), %%mm6         \n\t"

  2837.             "punpcklbw %%mm7, %%mm6                 \n\t" // L6

  2838.             "psubw %%mm6, %%mm2                     \n\t" // L5 - L6

  2839.             "movq (%%"REG_a", %1, 4), %%mm6         \n\t"

  2840.             "punpckhbw %%mm7, %%mm6                 \n\t" // H6

  2841.             "psubw %%mm6, %%mm3                     \n\t" // H5 - H6

  2842. 

  2843.             "paddw %%mm0, %%mm0                     \n\t" // 2L4

  2844.             "paddw %%mm1, %%mm1                     \n\t" // 2H4

  2845.             "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6

  2846.             "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6

  2847. 

  2848.             "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6

  2849.             "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6

  2850.             "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6

  2851.             "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6

  2852. 

  2853.             "movq (%0, %1, 4), %%mm2                \n\t"

  2854.             "movq %%mm2, %%mm3                      \n\t"

  2855.             "punpcklbw %%mm7, %%mm2                 \n\t" // L7

  2856.             "punpckhbw %%mm7, %%mm3                 \n\t" // H7

  2857. 

  2858.             "paddw %%mm2, %%mm2                     \n\t" // 2L7

  2859.             "paddw %%mm3, %%mm3                     \n\t" // 2H7

  2860.             "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7

  2861.             "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7

  2862. 

  2863.             "movq (%4), %%mm2                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3

  2864.             "movq 8(%4), %%mm3                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3

  2865. 

  2866. #if HAVE_MMX2

  2867.             "movq %%mm7, %%mm6                      \n\t" // 0

  2868.             "psubw %%mm0, %%mm6                     \n\t"

  2869.             "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|

  2870.             "movq %%mm7, %%mm6                      \n\t" // 0

  2871.             "psubw %%mm1, %%mm6                     \n\t"

  2872.             "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|

  2873.             "movq %%mm7, %%mm6                      \n\t" // 0

  2874.             "psubw %%mm2, %%mm6                     \n\t"

  2875.             "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|

  2876.             "movq %%mm7, %%mm6                      \n\t" // 0

  2877.             "psubw %%mm3, %%mm6                     \n\t"

  2878.             "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|

  2879. #else

  2880.             "movq %%mm7, %%mm6                      \n\t" // 0

  2881.             "pcmpgtw %%mm0, %%mm6                   \n\t"

  2882.             "pxor %%mm6, %%mm0                      \n\t"

  2883.             "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|

  2884.             "movq %%mm7, %%mm6                      \n\t" // 0

  2885.             "pcmpgtw %%mm1, %%mm6                   \n\t"

  2886.             "pxor %%mm6, %%mm1                      \n\t"

  2887.             "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|

  2888.             "movq %%mm7, %%mm6                      \n\t" // 0

  2889.             "pcmpgtw %%mm2, %%mm6                   \n\t"

  2890.             "pxor %%mm6, %%mm2                      \n\t"

  2891.             "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|

  2892.             "movq %%mm7, %%mm6                      \n\t" // 0

  2893.             "pcmpgtw %%mm3, %%mm6                   \n\t"

  2894.             "pxor %%mm6, %%mm3                      \n\t"

  2895.             "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|

  2896. #endif

  2897. 

  2898. #if HAVE_MMX2

  2899.             "pminsw %%mm2, %%mm0                    \n\t"

  2900.             "pminsw %%mm3, %%mm1                    \n\t"

  2901. #else

  2902.             "movq %%mm0, %%mm6                      \n\t"

  2903.             "psubusw %%mm2, %%mm6                   \n\t"

  2904.             "psubw %%mm6, %%mm0                     \n\t"

  2905.             "movq %%mm1, %%mm6                      \n\t"

  2906.             "psubusw %%mm3, %%mm6                   \n\t"

  2907.             "psubw %%mm6, %%mm1                     \n\t"

  2908. #endif

  2909. 

  2910.             "movd %2, %%mm2                         \n\t" // QP

  2911.             "punpcklbw %%mm7, %%mm2                 \n\t"

  2912. 

  2913.             "movq %%mm7, %%mm6                      \n\t" // 0

  2914.             "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)

  2915.             "pxor %%mm6, %%mm4                      \n\t"

  2916.             "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|

  2917.             "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)

  2918.             "pxor %%mm7, %%mm5                      \n\t"

  2919.             "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|

  2920. // 100 opcodes

  2921.             "psllw $3, %%mm2                        \n\t" // 8QP

  2922.             "movq %%mm2, %%mm3                      \n\t" // 8QP

  2923.             "pcmpgtw %%mm4, %%mm2                   \n\t"

  2924.             "pcmpgtw %%mm5, %%mm3                   \n\t"

  2925.             "pand %%mm2, %%mm4                      \n\t"

  2926.             "pand %%mm3, %%mm5                      \n\t"

  2927. 

  2928. 

  2929.             "psubusw %%mm0, %%mm4                   \n\t" // hd

  2930.             "psubusw %%mm1, %%mm5                   \n\t" // ld

  2931. 

  2932. 

  2933.             "movq "MANGLE(w05)", %%mm2              \n\t" // 5

  2934.             "pmullw %%mm2, %%mm4                    \n\t"

  2935.             "pmullw %%mm2, %%mm5                    \n\t"

  2936.             "movq "MANGLE(w20)", %%mm2              \n\t" // 32

  2937.             "paddw %%mm2, %%mm4                     \n\t"

  2938.             "paddw %%mm2, %%mm5                     \n\t"

  2939.             "psrlw $6, %%mm4                        \n\t"

  2940.             "psrlw $6, %%mm5                        \n\t"

  2941. 

  2942.             "movq 16(%4), %%mm0                     \n\t" // L3 - L4

  2943.             "movq 24(%4), %%mm1                     \n\t" // H3 - H4

  2944. 

  2945.             "pxor %%mm2, %%mm2                      \n\t"

  2946.             "pxor %%mm3, %%mm3                      \n\t"

  2947. 

  2948.             "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)

  2949.             "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)

  2950.             "pxor %%mm2, %%mm0                      \n\t"

  2951.             "pxor %%mm3, %%mm1                      \n\t"

  2952.             "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|

  2953.             "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|

  2954.             "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2

  2955.             "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2

  2956. 

  2957.             "pxor %%mm6, %%mm2                      \n\t"

  2958.             "pxor %%mm7, %%mm3                      \n\t"

  2959.             "pand %%mm2, %%mm4                      \n\t"

  2960.             "pand %%mm3, %%mm5                      \n\t"

  2961. 

  2962. #if HAVE_MMX2

  2963.             "pminsw %%mm0, %%mm4                    \n\t"

  2964.             "pminsw %%mm1, %%mm5                    \n\t"

  2965. #else

  2966.             "movq %%mm4, %%mm2                      \n\t"

  2967.             "psubusw %%mm0, %%mm2                   \n\t"

  2968.             "psubw %%mm2, %%mm4                     \n\t"

  2969.             "movq %%mm5, %%mm2                      \n\t"

  2970.             "psubusw %%mm1, %%mm2                   \n\t"

  2971.             "psubw %%mm2, %%mm5                     \n\t"

  2972. #endif

  2973.             "pxor %%mm6, %%mm4                      \n\t"

  2974.             "pxor %%mm7, %%mm5                      \n\t"

  2975.             "psubw %%mm6, %%mm4                     \n\t"

  2976.             "psubw %%mm7, %%mm5                     \n\t"

  2977.             "packsswb %%mm5, %%mm4                  \n\t"

  2978.             "movq %3, %%mm1                         \n\t"

  2979.             "pandn %%mm4, %%mm1                     \n\t"

  2980.             "movq (%0), %%mm0                       \n\t"

  2981.             "paddb   %%mm1, %%mm0                   \n\t"

  2982.             "movq %%mm0, (%0)                       \n\t"

  2983.             "movq (%0, %1), %%mm0                   \n\t"

  2984.             "psubb %%mm1, %%mm0                     \n\t"

  2985.             "movq %%mm0, (%0, %1)                   \n\t"

  2986. 

  2987.             : "+r" (temp_src)

  2988.             : "r" ((x86_reg)step), "m" (c->pQPb), "m"(eq_mask), "r"(tmp)

  2989.             : "%"REG_a

  2990.         );

  2991.     }

  2992. /*if(step==16){

  2993.     STOP_TIMER("step16")

  2994. }else{

  2995.     STOP_TIMER("stepX")

  2996. }*/

  2997. }

  2998. #endif //HAVE_MMX

  2999. 

  3000. static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,

  3001.                                 const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);

  3002. 

  3003. /**

  3004.  * Copy a block from src to dst and fixes the blacklevel.

  3005.  * levelFix == 0 -> do not touch the brighness & contrast

  3006.  */

  3007. #undef REAL_SCALED_CPY

  3008. #undef SCALED_CPY

  3009. 

  3010. static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t src[], int srcStride,

  3011.                                      int levelFix, int64_t *packedOffsetAndScale)

  3012. {

  3013. #if !HAVE_MMX

  3014.     int i;

  3015. #endif

  3016.     if(levelFix){

  3017. #if HAVE_MMX

  3018.     __asm__ volatile(

  3019.         "movq (%%"REG_a"), %%mm2        \n\t" // packedYOffset

  3020.         "movq 8(%%"REG_a"), %%mm3       \n\t" // packedYScale

  3021.         "lea (%2,%4), %%"REG_a"         \n\t"

  3022.         "lea (%3,%5), %%"REG_d"         \n\t"

  3023.         "pxor %%mm4, %%mm4              \n\t"

  3024. #if HAVE_MMX2

  3025. #define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                                \

  3026.         "movq " #src1 ", %%mm0          \n\t"\

  3027.         "movq " #src1 ", %%mm5          \n\t"\

  3028.         "movq " #src2 ", %%mm1          \n\t"\

  3029.         "movq " #src2 ", %%mm6          \n\t"\

  3030.         "punpcklbw %%mm0, %%mm0         \n\t"\

  3031.         "punpckhbw %%mm5, %%mm5         \n\t"\

  3032.         "punpcklbw %%mm1, %%mm1         \n\t"\

  3033.         "punpckhbw %%mm6, %%mm6         \n\t"\

  3034.         "pmulhuw %%mm3, %%mm0           \n\t"\

  3035.         "pmulhuw %%mm3, %%mm5           \n\t"\

  3036.         "pmulhuw %%mm3, %%mm1           \n\t"\

  3037.         "pmulhuw %%mm3, %%mm6           \n\t"\

  3038.         "psubw %%mm2, %%mm0             \n\t"\

  3039.         "psubw %%mm2, %%mm5             \n\t"\

  3040.         "psubw %%mm2, %%mm1             \n\t"\

  3041.         "psubw %%mm2, %%mm6             \n\t"\

  3042.         "packuswb %%mm5, %%mm0          \n\t"\

  3043.         "packuswb %%mm6, %%mm1          \n\t"\

  3044.         "movq %%mm0, " #dst1 "          \n\t"\

  3045.         "movq %%mm1, " #dst2 "          \n\t"\

  3046. 

  3047. #else //HAVE_MMX2

  3048. #define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                        \

  3049.         "movq " #src1 ", %%mm0          \n\t"\

  3050.         "movq " #src1 ", %%mm5          \n\t"\

  3051.         "punpcklbw %%mm4, %%mm0         \n\t"\

  3052.         "punpckhbw %%mm4, %%mm5         \n\t"\

  3053.         "psubw %%mm2, %%mm0             \n\t"\

  3054.         "psubw %%mm2, %%mm5             \n\t"\

  3055.         "movq " #src2 ", %%mm1          \n\t"\

  3056.         "psllw $6, %%mm0                \n\t"\

  3057.         "psllw $6, %%mm5                \n\t"\

  3058.         "pmulhw %%mm3, %%mm0            \n\t"\

  3059.         "movq " #src2 ", %%mm6          \n\t"\

  3060.         "pmulhw %%mm3, %%mm5            \n\t"\

  3061.         "punpcklbw %%mm4, %%mm1         \n\t"\

  3062.         "punpckhbw %%mm4, %%mm6         \n\t"\

  3063.         "psubw %%mm2, %%mm1             \n\t"\

  3064.         "psubw %%mm2, %%mm6             \n\t"\

  3065.         "psllw $6, %%mm1                \n\t"\

  3066.         "psllw $6, %%mm6                \n\t"\

  3067.         "pmulhw %%mm3, %%mm1            \n\t"\

  3068.         "pmulhw %%mm3, %%mm6            \n\t"\

  3069.         "packuswb %%mm5, %%mm0          \n\t"\

  3070.         "packuswb %%mm6, %%mm1          \n\t"\

  3071.         "movq %%mm0, " #dst1 "          \n\t"\

  3072.         "movq %%mm1, " #dst2 "          \n\t"\

  3073. 

  3074. #endif //HAVE_MMX2

  3075. #define SCALED_CPY(src1, src2, dst1, dst2)\

  3076.    REAL_SCALED_CPY(src1, src2, dst1, dst2)

  3077. 

  3078. SCALED_CPY((%2)       , (%2, %4)      , (%3)       , (%3, %5))

  3079. SCALED_CPY((%2, %4, 2), (%%REGa, %4, 2), (%3, %5, 2), (%%REGd, %5, 2))

  3080. SCALED_CPY((%2, %4, 4), (%%REGa, %4, 4), (%3, %5, 4), (%%REGd, %5, 4))

  3081.         "lea (%%"REG_a",%4,4), %%"REG_a"        \n\t"

  3082.         "lea (%%"REG_d",%5,4), %%"REG_d"        \n\t"

  3083. SCALED_CPY((%%REGa, %4), (%%REGa, %4, 2), (%%REGd, %5), (%%REGd, %5, 2))

  3084. 

  3085. 

  3086.         : "=&a" (packedOffsetAndScale)

  3087.         : "0" (packedOffsetAndScale),

  3088.         "r"(src),

  3089.         "r"(dst),

  3090.         "r" ((x86_reg)srcStride),

  3091.         "r" ((x86_reg)dstStride)

  3092.         : "%"REG_d

  3093.     );

  3094. #else //HAVE_MMX

  3095.     for(i=0; i<8; i++)

  3096.         memcpy( &(dst[dstStride*i]),

  3097.                 &(src[srcStride*i]), BLOCK_SIZE);

  3098. #endif //HAVE_MMX

  3099.     }else{

  3100. #if HAVE_MMX

  3101.     __asm__ volatile(

  3102.         "lea (%0,%2), %%"REG_a"                 \n\t"

  3103.         "lea (%1,%3), %%"REG_d"                 \n\t"

  3104. 

  3105. #define REAL_SIMPLE_CPY(src1, src2, dst1, dst2)                              \

  3106.         "movq " #src1 ", %%mm0          \n\t"\

  3107.         "movq " #src2 ", %%mm1          \n\t"\

  3108.         "movq %%mm0, " #dst1 "          \n\t"\

  3109.         "movq %%mm1, " #dst2 "          \n\t"\

  3110. 

  3111. #define SIMPLE_CPY(src1, src2, dst1, dst2)\

  3112.    REAL_SIMPLE_CPY(src1, src2, dst1, dst2)

  3113. 

  3114. SIMPLE_CPY((%0)       , (%0, %2)       , (%1)       , (%1, %3))

  3115. SIMPLE_CPY((%0, %2, 2), (%%REGa, %2, 2), (%1, %3, 2), (%%REGd, %3, 2))

  3116. SIMPLE_CPY((%0, %2, 4), (%%REGa, %2, 4), (%1, %3, 4), (%%REGd, %3, 4))

  3117.         "lea (%%"REG_a",%2,4), %%"REG_a"        \n\t"

  3118.         "lea (%%"REG_d",%3,4), %%"REG_d"        \n\t"

  3119. SIMPLE_CPY((%%REGa, %2), (%%REGa, %2, 2), (%%REGd, %3), (%%REGd, %3, 2))

  3120. 

  3121.         : : "r" (src),

  3122.         "r" (dst),

  3123.         "r" ((x86_reg)srcStride),

  3124.         "r" ((x86_reg)dstStride)

  3125.         : "%"REG_a, "%"REG_d

  3126.     );

  3127. #else //HAVE_MMX

  3128.     for(i=0; i<8; i++)

  3129.         memcpy( &(dst[dstStride*i]),

  3130.                 &(src[srcStride*i]), BLOCK_SIZE);

  3131. #endif //HAVE_MMX

  3132.     }

  3133. }

  3134. 

  3135. /**

  3136.  * Duplicate the given 8 src pixels ? times upward

  3137.  */

  3138. static inline void RENAME(duplicate)(uint8_t src[], int stride)

  3139. {

  3140. #if HAVE_MMX

  3141.     __asm__ volatile(

  3142.         "movq (%0), %%mm0               \n\t"

  3143.         "add %1, %0                     \n\t"

  3144.         "movq %%mm0, (%0)               \n\t"

  3145.         "movq %%mm0, (%0, %1)           \n\t"

  3146.         "movq %%mm0, (%0, %1, 2)        \n\t"

  3147.         : "+r" (src)

  3148.         : "r" ((x86_reg)-stride)

  3149.     );

  3150. #else

  3151.     int i;

  3152.     uint8_t *p=src;

  3153.     for(i=0; i<3; i++){

  3154.         p-= stride;

  3155.         memcpy(p, src, 8);

  3156.     }

  3157. #endif

  3158. }

  3159. 

  3160. /**

  3161.  * Filter array of bytes (Y or U or V values)

  3162.  */

  3163. static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,

  3164.                                 const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)

  3165. {

  3166.     DECLARE_ALIGNED(8, PPContext, c)= *c2; //copy to stack for faster access

  3167.     int x,y;

  3168. #ifdef COMPILE_TIME_MODE

  3169.     const int mode= COMPILE_TIME_MODE;

  3170. #else

  3171.     const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;

  3172. #endif

  3173.     int black=0, white=255; // blackest black and whitest white in the picture

  3174.     int QPCorrecture= 256*256;

  3175. 

  3176.     int copyAhead;

  3177. #if HAVE_MMX

  3178.     int i;

  3179. #endif

  3180. 

  3181.     const int qpHShift= isColor ? 4-c.hChromaSubSample : 4;

  3182.     const int qpVShift= isColor ? 4-c.vChromaSubSample : 4;

  3183. 

  3184.     //FIXME remove

  3185.     uint64_t * const yHistogram= c.yHistogram;

  3186.     uint8_t * const tempSrc= srcStride > 0 ? c.tempSrc : c.tempSrc - 23*srcStride;

  3187.     uint8_t * const tempDst= dstStride > 0 ? c.tempDst : c.tempDst - 23*dstStride;

  3188.     //const int mbWidth= isColor ? (width+7)>>3 : (width+15)>>4;

  3189. 

  3190. #if HAVE_MMX

  3191.     for(i=0; i<57; i++){

  3192.         int offset= ((i*c.ppMode.baseDcDiff)>>8) + 1;

  3193.         int threshold= offset*2 + 1;

  3194.         c.mmxDcOffset[i]= 0x7F - offset;

  3195.         c.mmxDcThreshold[i]= 0x7F - threshold;

  3196.         c.mmxDcOffset[i]*= 0x0101010101010101LL;

  3197.         c.mmxDcThreshold[i]*= 0x0101010101010101LL;

  3198.     }

  3199. #endif

  3200. 

  3201.     if(mode & CUBIC_IPOL_DEINT_FILTER) copyAhead=16;

  3202.     else if(   (mode & LINEAR_BLEND_DEINT_FILTER)

  3203.             || (mode & FFMPEG_DEINT_FILTER)

  3204.             || (mode & LOWPASS5_DEINT_FILTER)) copyAhead=14;

  3205.     else if(   (mode & V_DEBLOCK)

  3206.             || (mode & LINEAR_IPOL_DEINT_FILTER)

  3207.             || (mode & MEDIAN_DEINT_FILTER)

  3208.             || (mode & V_A_DEBLOCK)) copyAhead=13;

  3209.     else if(mode & V_X1_FILTER) copyAhead=11;

  3210. //    else if(mode & V_RK1_FILTER) copyAhead=10;

  3211.     else if(mode & DERING) copyAhead=9;

  3212.     else copyAhead=8;

  3213. 

  3214.     copyAhead-= 8;

  3215. 

  3216.     if(!isColor){

  3217.         uint64_t sum= 0;

  3218.         int i;

  3219.         uint64_t maxClipped;

  3220.         uint64_t clipped;

  3221.         double scale;

  3222. 

  3223.         c.frameNum++;

  3224.         // first frame is fscked so we ignore it

  3225.         if(c.frameNum == 1) yHistogram[0]= width*height/64*15/256;

  3226. 

  3227.         for(i=0; i<256; i++){

  3228.             sum+= yHistogram[i];

  3229.         }

  3230. 

  3231.         /* We always get a completely black picture first. */

  3232.         maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);

  3233. 

  3234.         clipped= sum;

  3235.         for(black=255; black>0; black--){

  3236.             if(clipped < maxClipped) break;

  3237.             clipped-= yHistogram[black];

  3238.         }

  3239. 

  3240.         clipped= sum;

  3241.         for(white=0; white<256; white++){

  3242.             if(clipped < maxClipped) break;

  3243.             clipped-= yHistogram[white];

  3244.         }

  3245. 

  3246.         scale= (double)(c.ppMode.maxAllowedY - c.ppMode.minAllowedY) / (double)(white-black);

  3247. 

  3248. #if HAVE_MMX2

  3249.         c.packedYScale= (uint16_t)(scale*256.0 + 0.5);

  3250.         c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF;

  3251. #else

  3252.         c.packedYScale= (uint16_t)(scale*1024.0 + 0.5);

  3253.         c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF;

  3254. #endif

  3255. 

  3256.         c.packedYOffset|= c.packedYOffset<<32;

  3257.         c.packedYOffset|= c.packedYOffset<<16;

  3258. 

  3259.         c.packedYScale|= c.packedYScale<<32;

  3260.         c.packedYScale|= c.packedYScale<<16;

  3261. 

  3262.         if(mode & LEVEL_FIX)        QPCorrecture= (int)(scale*256*256 + 0.5);

  3263.         else                        QPCorrecture= 256*256;

  3264.     }else{

  3265.         c.packedYScale= 0x0100010001000100LL;

  3266.         c.packedYOffset= 0;

  3267.         QPCorrecture= 256*256;

  3268.     }

  3269. 

  3270.     /* copy & deinterlace first row of blocks */

  3271.     y=-BLOCK_SIZE;

  3272.     {

  3273.         const uint8_t *srcBlock= &(src[y*srcStride]);

  3274.         uint8_t *dstBlock= tempDst + dstStride;

  3275. 

  3276.         // From this point on it is guaranteed that we can read and write 16 lines downward

  3277.         // finish 1 block before the next otherwise we might have a problem

  3278.         // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing

  3279.         for(x=0; x<width; x+=BLOCK_SIZE){

  3280. 

  3281. #if HAVE_MMX2

  3282. /*

  3283.             prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);

  3284.             prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);

  3285.             prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);

  3286.             prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);

  3287. */

  3288. 

  3289.             __asm__(

  3290.                 "mov %4, %%"REG_a"              \n\t"

  3291.                 "shr $2, %%"REG_a"              \n\t"

  3292.                 "and $6, %%"REG_a"              \n\t"

  3293.                 "add %5, %%"REG_a"              \n\t"

  3294.                 "mov %%"REG_a", %%"REG_d"       \n\t"

  3295.                 "imul %1, %%"REG_a"             \n\t"

  3296.                 "imul %3, %%"REG_d"             \n\t"

  3297.                 "prefetchnta 32(%%"REG_a", %0)  \n\t"

  3298.                 "prefetcht0 32(%%"REG_d", %2)   \n\t"

  3299.                 "add %1, %%"REG_a"              \n\t"

  3300.                 "add %3, %%"REG_d"              \n\t"

  3301.                 "prefetchnta 32(%%"REG_a", %0)  \n\t"

  3302.                 "prefetcht0 32(%%"REG_d", %2)   \n\t"

  3303.                 :: "r" (srcBlock), "r" ((x86_reg)srcStride), "r" (dstBlock), "r" ((x86_reg)dstStride),

  3304.                 "g" ((x86_reg)x), "g" ((x86_reg)copyAhead)

  3305.                 : "%"REG_a, "%"REG_d

  3306.             );

  3307. 

  3308. #elif HAVE_AMD3DNOW

  3309. //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...

  3310. /*          prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);

  3311.             prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);

  3312.             prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);

  3313.             prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);

  3314. */

  3315. #endif

  3316. 

  3317.             RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,

  3318.                               srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);

  3319. 

  3320.             RENAME(duplicate)(dstBlock + dstStride*8, dstStride);

  3321. 

  3322.             if(mode & LINEAR_IPOL_DEINT_FILTER)

  3323.                 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);

  3324.             else if(mode & LINEAR_BLEND_DEINT_FILTER)

  3325.                 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);

  3326.             else if(mode & MEDIAN_DEINT_FILTER)

  3327.                 RENAME(deInterlaceMedian)(dstBlock, dstStride);

  3328.             else if(mode & CUBIC_IPOL_DEINT_FILTER)

  3329.                 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);

  3330.             else if(mode & FFMPEG_DEINT_FILTER)

  3331.                 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);

  3332.             else if(mode & LOWPASS5_DEINT_FILTER)

  3333.                 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);

  3334. /*          else if(mode & CUBIC_BLEND_DEINT_FILTER)

  3335.                 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);

  3336. */

  3337.             dstBlock+=8;

  3338.             srcBlock+=8;

  3339.         }

  3340.         if(width==FFABS(dstStride))

  3341.             linecpy(dst, tempDst + 9*dstStride, copyAhead, dstStride);

  3342.         else{

  3343.             int i;

  3344.             for(i=0; i<copyAhead; i++){

  3345.                 memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);

  3346.             }

  3347.         }

  3348.     }

  3349. 

  3350.     for(y=0; y<height; y+=BLOCK_SIZE){

  3351.         //1% speedup if these are here instead of the inner loop

  3352.         const uint8_t *srcBlock= &(src[y*srcStride]);

  3353.         uint8_t *dstBlock= &(dst[y*dstStride]);

  3354. #if HAVE_MMX

  3355.         uint8_t *tempBlock1= c.tempBlocks;

  3356.         uint8_t *tempBlock2= c.tempBlocks + 8;

  3357. #endif

  3358.         const int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];

  3359.         int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*FFABS(QPStride)];

  3360.         int QP=0;

  3361.         /* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards

  3362.            if not than use a temporary buffer */

  3363.         if(y+15 >= height){

  3364.             int i;

  3365.             /* copy from line (copyAhead) to (copyAhead+7) of src, these will be copied with

  3366.                blockcopy to dst later */

  3367.             linecpy(tempSrc + srcStride*copyAhead, srcBlock + srcStride*copyAhead,

  3368.                     FFMAX(height-y-copyAhead, 0), srcStride);

  3369. 

  3370.             /* duplicate last line of src to fill the void upto line (copyAhead+7) */

  3371.             for(i=FFMAX(height-y, 8); i<copyAhead+8; i++)

  3372.                     memcpy(tempSrc + srcStride*i, src + srcStride*(height-1), FFABS(srcStride));

  3373. 

  3374.             /* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/

  3375.             linecpy(tempDst, dstBlock - dstStride, FFMIN(height-y+1, copyAhead+1), dstStride);

  3376. 

  3377.             /* duplicate last line of dst to fill the void upto line (copyAhead) */

  3378.             for(i=height-y+1; i<=copyAhead; i++)

  3379.                     memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), FFABS(dstStride));

  3380. 

  3381.             dstBlock= tempDst + dstStride;

  3382.             srcBlock= tempSrc;

  3383.         }

  3384. 

  3385.         // From this point on it is guaranteed that we can read and write 16 lines downward

  3386.         // finish 1 block before the next otherwise we might have a problem

  3387.         // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing

  3388.         for(x=0; x<width; x+=BLOCK_SIZE){

  3389.             const int stride= dstStride;

  3390. #if HAVE_MMX

  3391.             uint8_t *tmpXchg;

  3392. #endif

  3393.             if(isColor){

  3394.                 QP= QPptr[x>>qpHShift];

  3395.                 c.nonBQP= nonBQPptr[x>>qpHShift];

  3396.             }else{

  3397.                 QP= QPptr[x>>4];

  3398.                 QP= (QP* QPCorrecture + 256*128)>>16;

  3399.                 c.nonBQP= nonBQPptr[x>>4];

  3400.                 c.nonBQP= (c.nonBQP* QPCorrecture + 256*128)>>16;

  3401.                 yHistogram[ srcBlock[srcStride*12 + 4] ]++;

  3402.             }

  3403.             c.QP= QP;

  3404. #if HAVE_MMX

  3405.             __asm__ volatile(

  3406.                 "movd %1, %%mm7         \n\t"

  3407.                 "packuswb %%mm7, %%mm7  \n\t" // 0, 0, 0, QP, 0, 0, 0, QP

  3408.                 "packuswb %%mm7, %%mm7  \n\t" // 0,QP, 0, QP, 0,QP, 0, QP

  3409.                 "packuswb %%mm7, %%mm7  \n\t" // QP,..., QP

  3410.                 "movq %%mm7, %0         \n\t"

  3411.                 : "=m" (c.pQPb)

  3412.                 : "r" (QP)

  3413.             );

  3414. #endif

  3415. 

  3416. 

  3417. #if HAVE_MMX2

  3418. /*

  3419.             prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);

  3420.             prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);

  3421.             prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);

  3422.             prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);

  3423. */

  3424. 

  3425.             __asm__(

  3426.                 "mov %4, %%"REG_a"              \n\t"

  3427.                 "shr $2, %%"REG_a"              \n\t"

  3428.                 "and $6, %%"REG_a"              \n\t"

  3429.                 "add %5, %%"REG_a"              \n\t"

  3430.                 "mov %%"REG_a", %%"REG_d"       \n\t"

  3431.                 "imul %1, %%"REG_a"             \n\t"

  3432.                 "imul %3, %%"REG_d"             \n\t"

  3433.                 "prefetchnta 32(%%"REG_a", %0)  \n\t"

  3434.                 "prefetcht0 32(%%"REG_d", %2)   \n\t"

  3435.                 "add %1, %%"REG_a"              \n\t"

  3436.                 "add %3, %%"REG_d"              \n\t"

  3437.                 "prefetchnta 32(%%"REG_a", %0)  \n\t"

  3438.                 "prefetcht0 32(%%"REG_d", %2)   \n\t"

  3439.                 :: "r" (srcBlock), "r" ((x86_reg)srcStride), "r" (dstBlock), "r" ((x86_reg)dstStride),

  3440.                 "g" ((x86_reg)x), "g" ((x86_reg)copyAhead)

  3441.                 : "%"REG_a, "%"REG_d

  3442.             );

  3443. 

  3444. #elif HAVE_AMD3DNOW

  3445. //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...

  3446. /*          prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);

  3447.             prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);

  3448.             prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);

  3449.             prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);

  3450. */

  3451. #endif

  3452. 

  3453.             RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,

  3454.                               srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);

  3455. 

  3456.             if(mode & LINEAR_IPOL_DEINT_FILTER)

  3457.                 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);

  3458.             else if(mode & LINEAR_BLEND_DEINT_FILTER)

  3459.                 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);

  3460.             else if(mode & MEDIAN_DEINT_FILTER)

  3461.                 RENAME(deInterlaceMedian)(dstBlock, dstStride);

  3462.             else if(mode & CUBIC_IPOL_DEINT_FILTER)

  3463.                 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);

  3464.             else if(mode & FFMPEG_DEINT_FILTER)

  3465.                 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);

  3466.             else if(mode & LOWPASS5_DEINT_FILTER)

  3467.                 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);

  3468. /*          else if(mode & CUBIC_BLEND_DEINT_FILTER)

  3469.                 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);

  3470. */

  3471. 

  3472.             /* only deblock if we have 2 blocks */

  3473.             if(y + 8 < height){

  3474.                 if(mode & V_X1_FILTER)

  3475.                     RENAME(vertX1Filter)(dstBlock, stride, &c);

  3476.                 else if(mode & V_DEBLOCK){

  3477.                     const int t= RENAME(vertClassify)(dstBlock, stride, &c);

  3478. 

  3479.                     if(t==1)

  3480.                         RENAME(doVertLowPass)(dstBlock, stride, &c);

  3481.                     else if(t==2)

  3482.                         RENAME(doVertDefFilter)(dstBlock, stride, &c);

  3483.                 }else if(mode & V_A_DEBLOCK){

  3484.                     RENAME(do_a_deblock)(dstBlock, stride, 1, &c);

  3485.                 }

  3486.             }

  3487. 

  3488. #if HAVE_MMX

  3489.             RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);

  3490. #endif

  3491.             /* check if we have a previous block to deblock it with dstBlock */

  3492.             if(x - 8 >= 0){

  3493. #if HAVE_MMX

  3494.                 if(mode & H_X1_FILTER)

  3495.                         RENAME(vertX1Filter)(tempBlock1, 16, &c);

  3496.                 else if(mode & H_DEBLOCK){

  3497. //START_TIMER

  3498.                     const int t= RENAME(vertClassify)(tempBlock1, 16, &c);

  3499. //STOP_TIMER("dc & minmax")

  3500.                     if(t==1)

  3501.                         RENAME(doVertLowPass)(tempBlock1, 16, &c);

  3502.                     else if(t==2)

  3503.                         RENAME(doVertDefFilter)(tempBlock1, 16, &c);

  3504.                 }else if(mode & H_A_DEBLOCK){

  3505.                         RENAME(do_a_deblock)(tempBlock1, 16, 1, &c);

  3506.                 }

  3507. 

  3508.                 RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16);

  3509. 

  3510. #else

  3511.                 if(mode & H_X1_FILTER)

  3512.                     horizX1Filter(dstBlock-4, stride, QP);

  3513.                 else if(mode & H_DEBLOCK){

  3514. #if HAVE_ALTIVEC

  3515.                     DECLARE_ALIGNED(16, unsigned char, tempBlock)[272];

  3516.                     transpose_16x8_char_toPackedAlign_altivec(tempBlock, dstBlock - (4 + 1), stride);

  3517. 

  3518.                     const int t=vertClassify_altivec(tempBlock-48, 16, &c);

  3519.                     if(t==1) {

  3520.                         doVertLowPass_altivec(tempBlock-48, 16, &c);

  3521.                         transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);

  3522.                     }

  3523.                     else if(t==2) {

  3524.                         doVertDefFilter_altivec(tempBlock-48, 16, &c);

  3525.                         transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);

  3526.                     }

  3527. #else

  3528.                     const int t= RENAME(horizClassify)(dstBlock-4, stride, &c);

  3529. 

  3530.                     if(t==1)

  3531.                         RENAME(doHorizLowPass)(dstBlock-4, stride, &c);

  3532.                     else if(t==2)

  3533.                         RENAME(doHorizDefFilter)(dstBlock-4, stride, &c);

  3534. #endif

  3535.                 }else if(mode & H_A_DEBLOCK){

  3536.                     RENAME(do_a_deblock)(dstBlock-8, 1, stride, &c);

  3537.                 }

  3538. #endif //HAVE_MMX

  3539.                 if(mode & DERING){

  3540.                 //FIXME filter first line

  3541.                     if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);

  3542.                 }

  3543. 

  3544.                 if(mode & TEMP_NOISE_FILTER)

  3545.                 {

  3546.                     RENAME(tempNoiseReducer)(dstBlock-8, stride,

  3547.                             c.tempBlurred[isColor] + y*dstStride + x,

  3548.                             c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3),

  3549.                             c.ppMode.maxTmpNoise);

  3550.                 }

  3551.             }

  3552. 

  3553.             dstBlock+=8;

  3554.             srcBlock+=8;

  3555. 

  3556. #if HAVE_MMX

  3557.             tmpXchg= tempBlock1;

  3558.             tempBlock1= tempBlock2;

  3559.             tempBlock2 = tmpXchg;

  3560. #endif

  3561.         }

  3562. 

  3563.         if(mode & DERING){

  3564.             if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, &c);

  3565.         }

  3566. 

  3567.         if((mode & TEMP_NOISE_FILTER)){

  3568.             RENAME(tempNoiseReducer)(dstBlock-8, dstStride,

  3569.                     c.tempBlurred[isColor] + y*dstStride + x,

  3570.                     c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3),

  3571.                     c.ppMode.maxTmpNoise);

  3572.         }

  3573. 

  3574.         /* did we use a tmp buffer for the last lines*/

  3575.         if(y+15 >= height){

  3576.             uint8_t *dstBlock= &(dst[y*dstStride]);

  3577.             if(width==FFABS(dstStride))

  3578.                 linecpy(dstBlock, tempDst + dstStride, height-y, dstStride);

  3579.             else{

  3580.                 int i;

  3581.                 for(i=0; i<height-y; i++){

  3582.                     memcpy(dstBlock + i*dstStride, tempDst + (i+1)*dstStride, width);

  3583.                 }

  3584.             }

  3585.         }

  3586. /*

  3587.         for(x=0; x<width; x+=32){

  3588.             volatile int i;

  3589.             i+= + dstBlock[x + 7*dstStride] + dstBlock[x + 8*dstStride]

  3590.                 + dstBlock[x + 9*dstStride] + dstBlock[x +10*dstStride]

  3591.                 + dstBlock[x +11*dstStride] + dstBlock[x +12*dstStride];

  3592.                 + dstBlock[x +13*dstStride]

  3593.                 + dstBlock[x +14*dstStride] + dstBlock[x +15*dstStride];

  3594.         }*/

  3595.     }

  3596. #if   HAVE_AMD3DNOW

  3597.     __asm__ volatile("femms");

  3598. #elif HAVE_MMX

  3599.     __asm__ volatile("emms");

  3600. #endif

  3601. 

  3602. #ifdef DEBUG_BRIGHTNESS

  3603.     if(!isColor){

  3604.         int max=1;

  3605.         int i;

  3606.         for(i=0; i<256; i++)

  3607.             if(yHistogram[i] > max) max=yHistogram[i];

  3608. 

  3609.         for(i=1; i<256; i++){

  3610.             int x;

  3611.             int start=yHistogram[i-1]/(max/256+1);

  3612.             int end=yHistogram[i]/(max/256+1);

  3613.             int inc= end > start ? 1 : -1;

  3614.             for(x=start; x!=end+inc; x+=inc)

  3615.                 dst[ i*dstStride + x]+=128;

  3616.         }

  3617. 

  3618.         for(i=0; i<100; i+=2){

  3619.             dst[ (white)*dstStride + i]+=128;

  3620.             dst[ (black)*dstStride + i]+=128;

  3621.         }

  3622.     }

  3623. #endif

  3624. 

  3625.     *c2= c; //copy local context back

  3626. 

  3627. }