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.
46862 Commits
32 Branches
1.2GB
Tree: e034b07e40
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e034b07e40'
${ noResults }
FFmpeg/libpostproc/postprocess_template.c

3678 lines
159KB
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/asm.h"

  27. 

  28. /* A single TEMPLATE_PP_* should be defined (to 1) when this template is

  29.  * included. The following macros will define its dependencies to 1 as well

  30.  * (like MMX2 depending on MMX), and will define to 0 all the others. Every

  31.  * TEMPLATE_PP_* need to be undef at the end. */

  32. 

  33. #ifdef TEMPLATE_PP_C

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

  35. #else

  36. #   define TEMPLATE_PP_C 0

  37. #endif

  38. 

  39. #ifdef TEMPLATE_PP_ALTIVEC

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

  41. #else

  42. #   define TEMPLATE_PP_ALTIVEC 0

  43. #endif

  44. 

  45. #ifdef TEMPLATE_PP_MMX

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

  47. #else

  48. #   define TEMPLATE_PP_MMX 0

  49. #endif

  50. 

  51. #ifdef TEMPLATE_PP_MMXEXT

  52. #   undef  TEMPLATE_PP_MMX

  53. #   define TEMPLATE_PP_MMX 1

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

  55. #else

  56. #   define TEMPLATE_PP_MMXEXT 0

  57. #endif

  58. 

  59. #ifdef TEMPLATE_PP_3DNOW

  60. #   undef  TEMPLATE_PP_MMX

  61. #   define TEMPLATE_PP_MMX 1

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

  63. #else

  64. #   define TEMPLATE_PP_3DNOW 0

  65. #endif

  66. 

  67. #undef REAL_PAVGB

  68. #undef PAVGB

  69. #undef PMINUB

  70. #undef PMAXUB

  71. 

  72. #if   TEMPLATE_PP_MMXEXT

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

  74. #elif TEMPLATE_PP_3DNOW

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

  76. #endif

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

  78. 

  79. #if   TEMPLATE_PP_MMXEXT

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

  81. #elif TEMPLATE_PP_MMX

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

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

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

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

  86. #endif

  87. 

  88. #if   TEMPLATE_PP_MMXEXT

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

  90. #elif TEMPLATE_PP_MMX

  91. #define PMAXUB(a,b) \

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

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

  94. #endif

  95. 

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

  97. #if TEMPLATE_PP_MMX

  98. /**

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

  100.  */

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

  102.     int numEq= 0, dcOk;

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

  104.     __asm__ volatile(

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

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

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

  108.         );

  109. 

  110.     __asm__ volatile(

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

  112. //      0       1       2       3       4       5       6       7       8       9

  113. //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2

  114. 

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

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

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

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

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

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

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

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

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

  124. 

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

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

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

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

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

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

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

  132. 

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

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

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

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

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

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

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

  140. 

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

  142. 

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

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

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

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

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

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

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

  150. 

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

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

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

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

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

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

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

  158. 

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

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

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

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

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

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

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

  166. 

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

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

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

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

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

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

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

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

  175. 

  176.         "                                       \n\t"

  177. #if TEMPLATE_PP_MMXEXT

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

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

  180. #else

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

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

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

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

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

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

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

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

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

  190. #endif

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

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

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

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

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

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

  197. 

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

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

  200.         : "%"REG_a

  201.         );

  202. 

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

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

  205.         if(dcOk) return 0;

  206.         else     return 1;

  207.     }else{

  208.         return 2;

  209.     }

  210. }

  211. #endif //TEMPLATE_PP_MMX

  212. 

  213. /**

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

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

  216.  */

  217. #if !TEMPLATE_PP_ALTIVEC

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

  219. {

  220. #if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  221.     src+= stride*3;

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

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

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

  225. 

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

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

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

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

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

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

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

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

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

  235. 

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

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

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

  239. 

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

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

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

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

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

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

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

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

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

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

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

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

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

  253. 

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

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

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

  257. 

  258. 

  259.         //      1       2       3       4       5       6       7       8

  260.         //      %0      %0+%1   %0+2%1  eax     %0+4%1  eax+2%1 ecx     eax+4%1

  261.         // 6 4 2 2 1 1

  262.         // 6 4 4 2

  263.         // 6 8 2

  264. 

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

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

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

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

  269. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  337. 

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

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

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

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

  342. 

  343.         :

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

  345.         : "%"REG_a, "%"REG_c

  346.     );

  347. #else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  348.     const int l1= stride;

  349.     const int l2= stride + l1;

  350.     const int l3= stride + l2;

  351.     const int l4= stride + l3;

  352.     const int l5= stride + l4;

  353.     const int l6= stride + l5;

  354.     const int l7= stride + l6;

  355.     const int l8= stride + l7;

  356.     const int l9= stride + l8;

  357.     int x;

  358.     src+= stride*3;

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

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

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

  362. 

  363.         int sums[10];

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

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

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

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

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

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

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

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

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

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

  374. 

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

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

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

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

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

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

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

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

  383. 

  384.         src++;

  385.     }

  386. #endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  387. }

  388. #endif //TEMPLATE_PP_ALTIVEC

  389. 

  390. /**

  391.  * Experimental Filter 1

  392.  * will not damage linear gradients

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

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

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

  396.  */

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

  398. {

  399. #if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  400.     src+= stride*3;

  401. 

  402.     __asm__ volatile(

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

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

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

  406. //      0       1       2       3       4       5       6       7       8       9

  407. //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  435. 

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

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

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

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

  440. 

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

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

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

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

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

  446. 

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

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

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

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

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

  452. 

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

  454. 

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

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

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

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

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

  460. 

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

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

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

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

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

  466. 

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

  468. 

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

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

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

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

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

  474. 

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

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

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

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

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

  480. 

  481.         :

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

  483.         : "%"REG_a, "%"REG_c

  484.     );

  485. #else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  486. 

  487.     const int l1= stride;

  488.     const int l2= stride + l1;

  489.     const int l3= stride + l2;

  490.     const int l4= stride + l3;

  491.     const int l5= stride + l4;

  492.     const int l6= stride + l5;

  493.     const int l7= stride + l6;

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

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

  496.     int x;

  497. 

  498.     src+= stride*3;

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

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

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

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

  503. 

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

  505.         d= FFMAX(d, 0);

  506. 

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

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

  509. 

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

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

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

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

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

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

  516.         }

  517.         src++;

  518.     }

  519. #endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  520. }

  521. 

  522. #if !TEMPLATE_PP_ALTIVEC

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

  524. {

  525. #if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  526. /*

  527.     uint8_t tmp[16];

  528.     const int l1= stride;

  529.     const int l2= stride + l1;

  530.     const int l3= stride + l2;

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

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

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

  534.     const int l7= stride + l6;

  535.     const int l8= stride + l7;

  536. 

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

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

  539. */

  540.     src+= stride*4;

  541.     __asm__ volatile(

  542. 

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

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

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

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

  547. //      0       1       2       3       4       5       6       7

  548. //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1

  549. //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1

  550. 

  551. 

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

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

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

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

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

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

  558. 

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

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

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

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

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

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

  565. 

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

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

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

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

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

  571. 

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

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

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

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

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

  577. 

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

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

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

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

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

  583. 

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

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

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

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

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

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

  590. 

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

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

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

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

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

  596. 

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

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

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

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

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

  602. 

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

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

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

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

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

  608. 

  609. 

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

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

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

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

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

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

  616. 

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

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

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

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

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

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

  623. 

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

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

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

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

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

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

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

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

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

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

  634. 

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

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

  637. 

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

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

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

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

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

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

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

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

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

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

  648. #endif //0

  649. 

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

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

  652. //      0       1       2       3       4       5       6       7

  653. //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 ecx+%1  ecx+2%1

  654. //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1

  655. 

  656. 

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

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

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

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

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

  662. 

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

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

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

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

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

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

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

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

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

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

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

  674. 

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

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

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

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

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

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

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

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

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

  684. 

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

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

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

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

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

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

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

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

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

  694. 

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

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

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

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

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

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

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

  702. 

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

  704. 

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

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

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

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

  709. 

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

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

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

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

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

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

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

  717. 

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

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

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

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

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

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

  724. 

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

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

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

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

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

  730. 

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

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

  733. 

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

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

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

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

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

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

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

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

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

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

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

  745. 

  746.         :

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

  748.         : "%"REG_a, "%"REG_c

  749.     );

  750. 

  751. /*

  752.     {

  753.     int x;

  754.     src-= stride;

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

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

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

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

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

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

  761. 

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

  763.             d= FFMAX(d, 0);

  764. 

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

  766.             d*= FFSIGN(-middleEnergy);

  767. 

  768.             if(q>0){

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

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

  771.             }else{

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

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

  774.             }

  775. 

  776.             src[l4]-= d;

  777.             src[l5]+= d;

  778.         }

  779.         src++;

  780.     }

  781.     src-=8;

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

  783.         int y;

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

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

  786.             int ad= FFABS(d);

  787.             static int max=0;

  788.             static int sum=0;

  789.             static int num=0;

  790.             static int bias=0;

  791. 

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

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

  794.             if(ad>3){

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

  796.             }

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

  798.             num++;

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

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

  801.             }

  802.         }

  803.     }

  804. }

  805. */

  806. #elif TEMPLATE_PP_MMX

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

  808.     src+= stride*4;

  809.     __asm__ volatile(

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

  811. //      0       1       2       3       4       5       6       7

  812. //      %0      %0+%1   %0+2%1  eax+2%1 %0+4%1  eax+4%1 edx+%1  edx+2%1

  813. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1

  814. 

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

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

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

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

  819. 

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

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

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

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

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

  825. 

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

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

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

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

  830. 

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

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

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

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

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

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

  837. 

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

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

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

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

  842. 

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

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

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

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

  847. 

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

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

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

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

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

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

  854. 

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

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

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

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

  859. 

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

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

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

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

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

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

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

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

  868. 

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

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

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

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

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

  874. //50 opcodes so far

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

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

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

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

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

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

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

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

  883. 

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

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

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

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

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

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

  890. 

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

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

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

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

  895. 

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

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

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

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

  900. 

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

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

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

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

  905. 

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

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

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

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

  910. 

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

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

  913. 

  914. #if TEMPLATE_PP_MMXEXT

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

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

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

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

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

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

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

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

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

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

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

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

  927. #else

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  944. #endif

  945. 

  946. #if TEMPLATE_PP_MMXEXT

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

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

  949. #else

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

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

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

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

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

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

  956. #endif

  957. 

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

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

  960. 

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

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

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

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

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

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

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

  968. // 100 opcodes

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

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

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

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

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

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

  975. 

  976. 

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

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

  979. 

  980. 

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

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

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

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

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

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

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

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

  989. 

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

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

  992. 

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

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

  995. 

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

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

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

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

  1000.         "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|

  1001.         "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|

  1002.         "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2

  1003.         "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2

  1004. 

  1005.         "pxor %%mm6, %%mm2                      \n\t"

  1006.         "pxor %%mm7, %%mm3                      \n\t"

  1007.         "pand %%mm2, %%mm4                      \n\t"

  1008.         "pand %%mm3, %%mm5                      \n\t"

  1009. 

  1010. #if TEMPLATE_PP_MMXEXT

  1011.         "pminsw %%mm0, %%mm4                    \n\t"

  1012.         "pminsw %%mm1, %%mm5                    \n\t"

  1013. #else

  1014.         "movq %%mm4, %%mm2                      \n\t"

  1015.         "psubusw %%mm0, %%mm2                   \n\t"

  1016.         "psubw %%mm2, %%mm4                     \n\t"

  1017.         "movq %%mm5, %%mm2                      \n\t"

  1018.         "psubusw %%mm1, %%mm2                   \n\t"

  1019.         "psubw %%mm2, %%mm5                     \n\t"

  1020. #endif

  1021.         "pxor %%mm6, %%mm4                      \n\t"

  1022.         "pxor %%mm7, %%mm5                      \n\t"

  1023.         "psubw %%mm6, %%mm4                     \n\t"

  1024.         "psubw %%mm7, %%mm5                     \n\t"

  1025.         "packsswb %%mm5, %%mm4                  \n\t"

  1026.         "movq (%0), %%mm0                       \n\t"

  1027.         "paddb   %%mm4, %%mm0                   \n\t"

  1028.         "movq %%mm0, (%0)                       \n\t"

  1029.         "movq (%0, %1), %%mm0                   \n\t"

  1030.         "psubb %%mm4, %%mm0                     \n\t"

  1031.         "movq %%mm0, (%0, %1)                   \n\t"

  1032. 

  1033.         : "+r" (src)

  1034.         : "r" ((x86_reg)stride), "m" (c->pQPb), "r"(tmp)

  1035.         : "%"REG_a

  1036.     );

  1037. #else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1038.     const int l1= stride;

  1039.     const int l2= stride + l1;

  1040.     const int l3= stride + l2;

  1041.     const int l4= stride + l3;

  1042.     const int l5= stride + l4;

  1043.     const int l6= stride + l5;

  1044.     const int l7= stride + l6;

  1045.     const int l8= stride + l7;

  1046. //    const int l9= stride + l8;

  1047.     int x;

  1048.     src+= stride*3;

  1049.     for(x=0; x<BLOCK_SIZE; x++){

  1050.         const int middleEnergy= 5*(src[l5] - src[l4]) + 2*(src[l3] - src[l6]);

  1051.         if(FFABS(middleEnergy) < 8*c->QP){

  1052.             const int q=(src[l4] - src[l5])/2;

  1053.             const int leftEnergy=  5*(src[l3] - src[l2]) + 2*(src[l1] - src[l4]);

  1054.             const int rightEnergy= 5*(src[l7] - src[l6]) + 2*(src[l5] - src[l8]);

  1055. 

  1056.             int d= FFABS(middleEnergy) - FFMIN( FFABS(leftEnergy), FFABS(rightEnergy) );

  1057.             d= FFMAX(d, 0);

  1058. 

  1059.             d= (5*d + 32) >> 6;

  1060.             d*= FFSIGN(-middleEnergy);

  1061. 

  1062.             if(q>0){

  1063.                 d= d<0 ? 0 : d;

  1064.                 d= d>q ? q : d;

  1065.             }else{

  1066.                 d= d>0 ? 0 : d;

  1067.                 d= d<q ? q : d;

  1068.             }

  1069. 

  1070.             src[l4]-= d;

  1071.             src[l5]+= d;

  1072.         }

  1073.         src++;

  1074.     }

  1075. #endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1076. }

  1077. #endif //TEMPLATE_PP_ALTIVEC

  1078. 

  1079. #if !TEMPLATE_PP_ALTIVEC

  1080. static inline void RENAME(dering)(uint8_t src[], int stride, PPContext *c)

  1081. {

  1082. #if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1083.     DECLARE_ALIGNED(8, uint64_t, tmp)[3];

  1084.     __asm__ volatile(

  1085.         "pxor %%mm6, %%mm6                      \n\t"

  1086.         "pcmpeqb %%mm7, %%mm7                   \n\t"

  1087.         "movq %2, %%mm0                         \n\t"

  1088.         "punpcklbw %%mm6, %%mm0                 \n\t"

  1089.         "psrlw $1, %%mm0                        \n\t"

  1090.         "psubw %%mm7, %%mm0                     \n\t"

  1091.         "packuswb %%mm0, %%mm0                  \n\t"

  1092.         "movq %%mm0, %3                         \n\t"

  1093. 

  1094.         "lea (%0, %1), %%"REG_a"                \n\t"

  1095.         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"

  1096. 

  1097. //        0        1        2        3        4        5        6        7        8        9

  1098. //        %0        eax        eax+%1        eax+2%1        %0+4%1        edx        edx+%1        edx+2%1        %0+8%1        edx+4%1

  1099. 

  1100. #undef REAL_FIND_MIN_MAX

  1101. #undef FIND_MIN_MAX

  1102. #if TEMPLATE_PP_MMXEXT

  1103. #define REAL_FIND_MIN_MAX(addr)\

  1104.         "movq " #addr ", %%mm0                  \n\t"\

  1105.         "pminub %%mm0, %%mm7                    \n\t"\

  1106.         "pmaxub %%mm0, %%mm6                    \n\t"

  1107. #else

  1108. #define REAL_FIND_MIN_MAX(addr)\

  1109.         "movq " #addr ", %%mm0                  \n\t"\

  1110.         "movq %%mm7, %%mm1                      \n\t"\

  1111.         "psubusb %%mm0, %%mm6                   \n\t"\

  1112.         "paddb %%mm0, %%mm6                     \n\t"\

  1113.         "psubusb %%mm0, %%mm1                   \n\t"\

  1114.         "psubb %%mm1, %%mm7                     \n\t"

  1115. #endif

  1116. #define FIND_MIN_MAX(addr)  REAL_FIND_MIN_MAX(addr)

  1117. 

  1118. FIND_MIN_MAX((%%REGa))

  1119. FIND_MIN_MAX((%%REGa, %1))

  1120. FIND_MIN_MAX((%%REGa, %1, 2))

  1121. FIND_MIN_MAX((%0, %1, 4))

  1122. FIND_MIN_MAX((%%REGd))

  1123. FIND_MIN_MAX((%%REGd, %1))

  1124. FIND_MIN_MAX((%%REGd, %1, 2))

  1125. FIND_MIN_MAX((%0, %1, 8))

  1126. 

  1127.         "movq %%mm7, %%mm4                      \n\t"

  1128.         "psrlq $8, %%mm7                        \n\t"

  1129. #if TEMPLATE_PP_MMXEXT

  1130.         "pminub %%mm4, %%mm7                    \n\t" // min of pixels

  1131.         "pshufw $0xF9, %%mm7, %%mm4             \n\t"

  1132.         "pminub %%mm4, %%mm7                    \n\t" // min of pixels

  1133.         "pshufw $0xFE, %%mm7, %%mm4             \n\t"

  1134.         "pminub %%mm4, %%mm7                    \n\t"

  1135. #else

  1136.         "movq %%mm7, %%mm1                      \n\t"

  1137.         "psubusb %%mm4, %%mm1                   \n\t"

  1138.         "psubb %%mm1, %%mm7                     \n\t"

  1139.         "movq %%mm7, %%mm4                      \n\t"

  1140.         "psrlq $16, %%mm7                       \n\t"

  1141.         "movq %%mm7, %%mm1                      \n\t"

  1142.         "psubusb %%mm4, %%mm1                   \n\t"

  1143.         "psubb %%mm1, %%mm7                     \n\t"

  1144.         "movq %%mm7, %%mm4                      \n\t"

  1145.         "psrlq $32, %%mm7                       \n\t"

  1146.         "movq %%mm7, %%mm1                      \n\t"

  1147.         "psubusb %%mm4, %%mm1                   \n\t"

  1148.         "psubb %%mm1, %%mm7                     \n\t"

  1149. #endif

  1150. 

  1151. 

  1152.         "movq %%mm6, %%mm4                      \n\t"

  1153.         "psrlq $8, %%mm6                        \n\t"

  1154. #if TEMPLATE_PP_MMXEXT

  1155.         "pmaxub %%mm4, %%mm6                    \n\t" // max of pixels

  1156.         "pshufw $0xF9, %%mm6, %%mm4             \n\t"

  1157.         "pmaxub %%mm4, %%mm6                    \n\t"

  1158.         "pshufw $0xFE, %%mm6, %%mm4             \n\t"

  1159.         "pmaxub %%mm4, %%mm6                    \n\t"

  1160. #else

  1161.         "psubusb %%mm4, %%mm6                   \n\t"

  1162.         "paddb %%mm4, %%mm6                     \n\t"

  1163.         "movq %%mm6, %%mm4                      \n\t"

  1164.         "psrlq $16, %%mm6                       \n\t"

  1165.         "psubusb %%mm4, %%mm6                   \n\t"

  1166.         "paddb %%mm4, %%mm6                     \n\t"

  1167.         "movq %%mm6, %%mm4                      \n\t"

  1168.         "psrlq $32, %%mm6                       \n\t"

  1169.         "psubusb %%mm4, %%mm6                   \n\t"

  1170.         "paddb %%mm4, %%mm6                     \n\t"

  1171. #endif

  1172.         "movq %%mm6, %%mm0                      \n\t" // max

  1173.         "psubb %%mm7, %%mm6                     \n\t" // max - min

  1174.         "push %4                              \n\t"

  1175.         "movd %%mm6, %k4                        \n\t"

  1176.         "cmpb "MANGLE(deringThreshold)", %b4    \n\t"

  1177.         "pop %4                               \n\t"

  1178.         " jb 1f                                 \n\t"

  1179.         PAVGB(%%mm0, %%mm7)                           // a=(max + min)/2

  1180.         "punpcklbw %%mm7, %%mm7                 \n\t"

  1181.         "punpcklbw %%mm7, %%mm7                 \n\t"

  1182.         "punpcklbw %%mm7, %%mm7                 \n\t"

  1183.         "movq %%mm7, (%4)                       \n\t"

  1184. 

  1185.         "movq (%0), %%mm0                       \n\t" // L10

  1186.         "movq %%mm0, %%mm1                      \n\t" // L10

  1187.         "movq %%mm0, %%mm2                      \n\t" // L10

  1188.         "psllq $8, %%mm1                        \n\t"

  1189.         "psrlq $8, %%mm2                        \n\t"

  1190.         "movd -4(%0), %%mm3                     \n\t"

  1191.         "movd 8(%0), %%mm4                      \n\t"

  1192.         "psrlq $24, %%mm3                       \n\t"

  1193.         "psllq $56, %%mm4                       \n\t"

  1194.         "por %%mm3, %%mm1                       \n\t" // L00

  1195.         "por %%mm4, %%mm2                       \n\t" // L20

  1196.         "movq %%mm1, %%mm3                      \n\t" // L00

  1197.         PAVGB(%%mm2, %%mm1)                           // (L20 + L00)/2

  1198.         PAVGB(%%mm0, %%mm1)                           // (L20 + L00 + 2L10)/4

  1199.         "psubusb %%mm7, %%mm0                   \n\t"

  1200.         "psubusb %%mm7, %%mm2                   \n\t"

  1201.         "psubusb %%mm7, %%mm3                   \n\t"

  1202.         "pcmpeqb "MANGLE(b00)", %%mm0           \n\t" // L10 > a ? 0 : -1

  1203.         "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L20 > a ? 0 : -1

  1204.         "pcmpeqb "MANGLE(b00)", %%mm3           \n\t" // L00 > a ? 0 : -1

  1205.         "paddb %%mm2, %%mm0                     \n\t"

  1206.         "paddb %%mm3, %%mm0                     \n\t"

  1207. 

  1208.         "movq (%%"REG_a"), %%mm2                \n\t" // L11

  1209.         "movq %%mm2, %%mm3                      \n\t" // L11

  1210.         "movq %%mm2, %%mm4                      \n\t" // L11

  1211.         "psllq $8, %%mm3                        \n\t"

  1212.         "psrlq $8, %%mm4                        \n\t"

  1213.         "movd -4(%%"REG_a"), %%mm5              \n\t"

  1214.         "movd 8(%%"REG_a"), %%mm6               \n\t"

  1215.         "psrlq $24, %%mm5                       \n\t"

  1216.         "psllq $56, %%mm6                       \n\t"

  1217.         "por %%mm5, %%mm3                       \n\t" // L01

  1218.         "por %%mm6, %%mm4                       \n\t" // L21

  1219.         "movq %%mm3, %%mm5                      \n\t" // L01

  1220.         PAVGB(%%mm4, %%mm3)                           // (L21 + L01)/2

  1221.         PAVGB(%%mm2, %%mm3)                           // (L21 + L01 + 2L11)/4

  1222.         "psubusb %%mm7, %%mm2                   \n\t"

  1223.         "psubusb %%mm7, %%mm4                   \n\t"

  1224.         "psubusb %%mm7, %%mm5                   \n\t"

  1225.         "pcmpeqb "MANGLE(b00)", %%mm2           \n\t" // L11 > a ? 0 : -1

  1226.         "pcmpeqb "MANGLE(b00)", %%mm4           \n\t" // L21 > a ? 0 : -1

  1227.         "pcmpeqb "MANGLE(b00)", %%mm5           \n\t" // L01 > a ? 0 : -1

  1228.         "paddb %%mm4, %%mm2                     \n\t"

  1229.         "paddb %%mm5, %%mm2                     \n\t"

  1230. // 0, 2, 3, 1

  1231. #define REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \

  1232.         "movq " #src ", " #sx "                 \n\t" /* src[0] */\

  1233.         "movq " #sx ", " #lx "                  \n\t" /* src[0] */\

  1234.         "movq " #sx ", " #t0 "                  \n\t" /* src[0] */\

  1235.         "psllq $8, " #lx "                      \n\t"\

  1236.         "psrlq $8, " #t0 "                      \n\t"\

  1237.         "movd -4" #src ", " #t1 "               \n\t"\

  1238.         "psrlq $24, " #t1 "                     \n\t"\

  1239.         "por " #t1 ", " #lx "                   \n\t" /* src[-1] */\

  1240.         "movd 8" #src ", " #t1 "                \n\t"\

  1241.         "psllq $56, " #t1 "                     \n\t"\

  1242.         "por " #t1 ", " #t0 "                   \n\t" /* src[+1] */\

  1243.         "movq " #lx ", " #t1 "                  \n\t" /* src[-1] */\

  1244.         PAVGB(t0, lx)                                 /* (src[-1] + src[+1])/2 */\

  1245.         PAVGB(sx, lx)                                 /* (src[-1] + 2src[0] + src[+1])/4 */\

  1246.         PAVGB(lx, pplx)                                     \

  1247.         "movq " #lx ", 8(%4)                    \n\t"\

  1248.         "movq (%4), " #lx "                     \n\t"\

  1249.         "psubusb " #lx ", " #t1 "               \n\t"\

  1250.         "psubusb " #lx ", " #t0 "               \n\t"\

  1251.         "psubusb " #lx ", " #sx "               \n\t"\

  1252.         "movq "MANGLE(b00)", " #lx "            \n\t"\

  1253.         "pcmpeqb " #lx ", " #t1 "               \n\t" /* src[-1] > a ? 0 : -1*/\

  1254.         "pcmpeqb " #lx ", " #t0 "               \n\t" /* src[+1] > a ? 0 : -1*/\

  1255.         "pcmpeqb " #lx ", " #sx "               \n\t" /* src[0]  > a ? 0 : -1*/\

  1256.         "paddb " #t1 ", " #t0 "                 \n\t"\

  1257.         "paddb " #t0 ", " #sx "                 \n\t"\

  1258. \

  1259.         PAVGB(plx, pplx)                              /* filtered */\

  1260.         "movq " #dst ", " #t0 "                 \n\t" /* dst */\

  1261.         "movq " #t0 ", " #t1 "                  \n\t" /* dst */\

  1262.         "psubusb %3, " #t0 "                    \n\t"\

  1263.         "paddusb %3, " #t1 "                    \n\t"\

  1264.         PMAXUB(t0, pplx)\

  1265.         PMINUB(t1, pplx, t0)\

  1266.         "paddb " #sx ", " #ppsx "               \n\t"\

  1267.         "paddb " #psx ", " #ppsx "              \n\t"\

  1268.         "#paddb "MANGLE(b02)", " #ppsx "        \n\t"\

  1269.         "pand "MANGLE(b08)", " #ppsx "          \n\t"\

  1270.         "pcmpeqb " #lx ", " #ppsx "             \n\t"\

  1271.         "pand " #ppsx ", " #pplx "              \n\t"\

  1272.         "pandn " #dst ", " #ppsx "              \n\t"\

  1273.         "por " #pplx ", " #ppsx "               \n\t"\

  1274.         "movq " #ppsx ", " #dst "               \n\t"\

  1275.         "movq 8(%4), " #lx "                    \n\t"

  1276. 

  1277. #define DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1) \

  1278.    REAL_DERING_CORE(dst,src,ppsx,psx,sx,pplx,plx,lx,t0,t1)

  1279. /*

  1280. 0000000

  1281. 1111111

  1282. 

  1283. 1111110

  1284. 1111101

  1285. 1111100

  1286. 1111011

  1287. 1111010

  1288. 1111001

  1289. 

  1290. 1111000

  1291. 1110111

  1292. 

  1293. */

  1294. //DERING_CORE(dst          ,src            ,ppsx ,psx  ,sx   ,pplx ,plx  ,lx   ,t0   ,t1)

  1295. DERING_CORE((%%REGa)       ,(%%REGa, %1)   ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)

  1296. DERING_CORE((%%REGa, %1)   ,(%%REGa, %1, 2),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)

  1297. DERING_CORE((%%REGa, %1, 2),(%0, %1, 4)    ,%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)

  1298. DERING_CORE((%0, %1, 4)    ,(%%REGd)       ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)

  1299. DERING_CORE((%%REGd)       ,(%%REGd, %1)   ,%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)

  1300. DERING_CORE((%%REGd, %1)   ,(%%REGd, %1, 2),%%mm4,%%mm0,%%mm2,%%mm5,%%mm1,%%mm3,%%mm6,%%mm7)

  1301. DERING_CORE((%%REGd, %1, 2),(%0, %1, 8)    ,%%mm0,%%mm2,%%mm4,%%mm1,%%mm3,%%mm5,%%mm6,%%mm7)

  1302. DERING_CORE((%0, %1, 8)    ,(%%REGd, %1, 4),%%mm2,%%mm4,%%mm0,%%mm3,%%mm5,%%mm1,%%mm6,%%mm7)

  1303. 

  1304.         "1:                        \n\t"

  1305.         : : "r" (src), "r" ((x86_reg)stride), "m" (c->pQPb), "m"(c->pQPb2), "q"(tmp)

  1306.         : "%"REG_a, "%"REG_d

  1307.     );

  1308. #else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1309.     int y;

  1310.     int min=255;

  1311.     int max=0;

  1312.     int avg;

  1313.     uint8_t *p;

  1314.     int s[10];

  1315.     const int QP2= c->QP/2 + 1;

  1316. 

  1317.     for(y=1; y<9; y++){

  1318.         int x;

  1319.         p= src + stride*y;

  1320.         for(x=1; x<9; x++){

  1321.             p++;

  1322.             if(*p > max) max= *p;

  1323.             if(*p < min) min= *p;

  1324.         }

  1325.     }

  1326.     avg= (min + max + 1)>>1;

  1327. 

  1328.     if(max - min <deringThreshold) return;

  1329. 

  1330.     for(y=0; y<10; y++){

  1331.         int t = 0;

  1332. 

  1333.         if(src[stride*y + 0] > avg) t+= 1;

  1334.         if(src[stride*y + 1] > avg) t+= 2;

  1335.         if(src[stride*y + 2] > avg) t+= 4;

  1336.         if(src[stride*y + 3] > avg) t+= 8;

  1337.         if(src[stride*y + 4] > avg) t+= 16;

  1338.         if(src[stride*y + 5] > avg) t+= 32;

  1339.         if(src[stride*y + 6] > avg) t+= 64;

  1340.         if(src[stride*y + 7] > avg) t+= 128;

  1341.         if(src[stride*y + 8] > avg) t+= 256;

  1342.         if(src[stride*y + 9] > avg) t+= 512;

  1343. 

  1344.         t |= (~t)<<16;

  1345.         t &= (t<<1) & (t>>1);

  1346.         s[y] = t;

  1347.     }

  1348. 

  1349.     for(y=1; y<9; y++){

  1350.         int t = s[y-1] & s[y] & s[y+1];

  1351.         t|= t>>16;

  1352.         s[y-1]= t;

  1353.     }

  1354. 

  1355.     for(y=1; y<9; y++){

  1356.         int x;

  1357.         int t = s[y-1];

  1358. 

  1359.         p= src + stride*y;

  1360.         for(x=1; x<9; x++){

  1361.             p++;

  1362.             if(t & (1<<x)){

  1363.                 int f= (*(p-stride-1)) + 2*(*(p-stride)) + (*(p-stride+1))

  1364.                       +2*(*(p     -1)) + 4*(*p         ) + 2*(*(p     +1))

  1365.                       +(*(p+stride-1)) + 2*(*(p+stride)) + (*(p+stride+1));

  1366.                 f= (f + 8)>>4;

  1367. 

  1368. #ifdef DEBUG_DERING_THRESHOLD

  1369.                     __asm__ volatile("emms\n\t":);

  1370.                     {

  1371.                     static long long numPixels=0;

  1372.                     if(x!=1 && x!=8 && y!=1 && y!=8) numPixels++;

  1373. //                    if((max-min)<20 || (max-min)*QP<200)

  1374. //                    if((max-min)*QP < 500)

  1375. //                    if(max-min<QP/2)

  1376.                     if(max-min < 20){

  1377.                         static int numSkipped=0;

  1378.                         static int errorSum=0;

  1379.                         static int worstQP=0;

  1380.                         static int worstRange=0;

  1381.                         static int worstDiff=0;

  1382.                         int diff= (f - *p);

  1383.                         int absDiff= FFABS(diff);

  1384.                         int error= diff*diff;

  1385. 

  1386.                         if(x==1 || x==8 || y==1 || y==8) continue;

  1387. 

  1388.                         numSkipped++;

  1389.                         if(absDiff > worstDiff){

  1390.                             worstDiff= absDiff;

  1391.                             worstQP= QP;

  1392.                             worstRange= max-min;

  1393.                         }

  1394.                         errorSum+= error;

  1395. 

  1396.                         if(1024LL*1024LL*1024LL % numSkipped == 0){

  1397.                             av_log(c, AV_LOG_INFO, "sum:%1.3f, skip:%d, wQP:%d, "

  1398.                                    "wRange:%d, wDiff:%d, relSkip:%1.3f\n",

  1399.                                    (float)errorSum/numSkipped, numSkipped, worstQP, worstRange,

  1400.                                    worstDiff, (float)numSkipped/numPixels);

  1401.                         }

  1402.                     }

  1403.                     }

  1404. #endif

  1405.                     if     (*p + QP2 < f) *p= *p + QP2;

  1406.                     else if(*p - QP2 > f) *p= *p - QP2;

  1407.                     else *p=f;

  1408.             }

  1409.         }

  1410.     }

  1411. #ifdef DEBUG_DERING_THRESHOLD

  1412.     if(max-min < 20){

  1413.         for(y=1; y<9; y++){

  1414.             int x;

  1415.             int t = 0;

  1416.             p= src + stride*y;

  1417.             for(x=1; x<9; x++){

  1418.                 p++;

  1419.                 *p = FFMIN(*p + 20, 255);

  1420.             }

  1421.         }

  1422. //        src[0] = src[7]=src[stride*7]=src[stride*7 + 7]=255;

  1423.     }

  1424. #endif

  1425. #endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1426. }

  1427. #endif //TEMPLATE_PP_ALTIVEC

  1428. 

  1429. /**

  1430.  * Deinterlace the given block by linearly interpolating every second line.

  1431.  * will be called for every 8x8 block and can read & write from line 4-15

  1432.  * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.

  1433.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1434.  */

  1435. static inline void RENAME(deInterlaceInterpolateLinear)(uint8_t src[], int stride)

  1436. {

  1437. #if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1438.     src+= 4*stride;

  1439.     __asm__ volatile(

  1440.         "lea (%0, %1), %%"REG_a"                \n\t"

  1441.         "lea (%%"REG_a", %1, 4), %%"REG_c"      \n\t"

  1442. //      0       1       2       3       4       5       6       7       8       9

  1443. //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %0+8%1  ecx+4%1

  1444. 

  1445.         "movq (%0), %%mm0                       \n\t"

  1446.         "movq (%%"REG_a", %1), %%mm1            \n\t"

  1447.         PAVGB(%%mm1, %%mm0)

  1448.         "movq %%mm0, (%%"REG_a")                \n\t"

  1449.         "movq (%0, %1, 4), %%mm0                \n\t"

  1450.         PAVGB(%%mm0, %%mm1)

  1451.         "movq %%mm1, (%%"REG_a", %1, 2)         \n\t"

  1452.         "movq (%%"REG_c", %1), %%mm1            \n\t"

  1453.         PAVGB(%%mm1, %%mm0)

  1454.         "movq %%mm0, (%%"REG_c")                \n\t"

  1455.         "movq (%0, %1, 8), %%mm0                \n\t"

  1456.         PAVGB(%%mm0, %%mm1)

  1457.         "movq %%mm1, (%%"REG_c", %1, 2)         \n\t"

  1458. 

  1459.         : : "r" (src), "r" ((x86_reg)stride)

  1460.         : "%"REG_a, "%"REG_c

  1461.     );

  1462. #else

  1463.     int a, b, x;

  1464.     src+= 4*stride;

  1465. 

  1466.     for(x=0; x<2; x++){

  1467.         a= *(uint32_t*)&src[stride*0];

  1468.         b= *(uint32_t*)&src[stride*2];

  1469.         *(uint32_t*)&src[stride*1]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1470.         a= *(uint32_t*)&src[stride*4];

  1471.         *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1472.         b= *(uint32_t*)&src[stride*6];

  1473.         *(uint32_t*)&src[stride*5]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1474.         a= *(uint32_t*)&src[stride*8];

  1475.         *(uint32_t*)&src[stride*7]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1476.         src += 4;

  1477.     }

  1478. #endif

  1479. }

  1480. 

  1481. /**

  1482.  * Deinterlace the given block by cubic interpolating every second line.

  1483.  * will be called for every 8x8 block and can read & write from line 4-15

  1484.  * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.

  1485.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1486.  * this filter will read lines 3-15 and write 7-13

  1487.  */

  1488. static inline void RENAME(deInterlaceInterpolateCubic)(uint8_t src[], int stride)

  1489. {

  1490. #if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1491.     src+= stride*3;

  1492.     __asm__ volatile(

  1493.         "lea (%0, %1), %%"REG_a"                \n\t"

  1494.         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"

  1495.         "lea (%%"REG_d", %1, 4), %%"REG_c"      \n\t"

  1496.         "add %1, %%"REG_c"                      \n\t"

  1497.         "pxor %%mm7, %%mm7                      \n\t"

  1498. //      0       1       2       3       4       5       6       7       8       9       10

  1499. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx

  1500. 

  1501. #define REAL_DEINT_CUBIC(a,b,c,d,e)\

  1502.         "movq " #a ", %%mm0                     \n\t"\

  1503.         "movq " #b ", %%mm1                     \n\t"\

  1504.         "movq " #d ", %%mm2                     \n\t"\

  1505.         "movq " #e ", %%mm3                     \n\t"\

  1506.         PAVGB(%%mm2, %%mm1)                             /* (b+d) /2 */\

  1507.         PAVGB(%%mm3, %%mm0)                             /* (a+e) /2 */\

  1508.         "movq %%mm0, %%mm2                      \n\t"\

  1509.         "punpcklbw %%mm7, %%mm0                 \n\t"\

  1510.         "punpckhbw %%mm7, %%mm2                 \n\t"\

  1511.         "movq %%mm1, %%mm3                      \n\t"\

  1512.         "punpcklbw %%mm7, %%mm1                 \n\t"\

  1513.         "punpckhbw %%mm7, %%mm3                 \n\t"\

  1514.         "psubw %%mm1, %%mm0                     \n\t"   /* L(a+e - (b+d))/2 */\

  1515.         "psubw %%mm3, %%mm2                     \n\t"   /* H(a+e - (b+d))/2 */\

  1516.         "psraw $3, %%mm0                        \n\t"   /* L(a+e - (b+d))/16 */\

  1517.         "psraw $3, %%mm2                        \n\t"   /* H(a+e - (b+d))/16 */\

  1518.         "psubw %%mm0, %%mm1                     \n\t"   /* L(9b + 9d - a - e)/16 */\

  1519.         "psubw %%mm2, %%mm3                     \n\t"   /* H(9b + 9d - a - e)/16 */\

  1520.         "packuswb %%mm3, %%mm1                  \n\t"\

  1521.         "movq %%mm1, " #c "                     \n\t"

  1522. #define DEINT_CUBIC(a,b,c,d,e)  REAL_DEINT_CUBIC(a,b,c,d,e)

  1523. 

  1524. DEINT_CUBIC((%0)        , (%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd, %1))

  1525. DEINT_CUBIC((%%REGa, %1), (%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%0, %1, 8))

  1526. DEINT_CUBIC((%0, %1, 4) , (%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGc))

  1527. DEINT_CUBIC((%%REGd, %1), (%0, %1, 8) , (%%REGd, %1, 4), (%%REGc)    , (%%REGc, %1, 2))

  1528. 

  1529.         : : "r" (src), "r" ((x86_reg)stride)

  1530.         : "%"REG_a, "%"REG_d, "%"REG_c

  1531.     );

  1532. #else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1533.     int x;

  1534.     src+= stride*3;

  1535.     for(x=0; x<8; x++){

  1536.         src[stride*3] = CLIP((-src[0]        + 9*src[stride*2] + 9*src[stride*4] - src[stride*6])>>4);

  1537.         src[stride*5] = CLIP((-src[stride*2] + 9*src[stride*4] + 9*src[stride*6] - src[stride*8])>>4);

  1538.         src[stride*7] = CLIP((-src[stride*4] + 9*src[stride*6] + 9*src[stride*8] - src[stride*10])>>4);

  1539.         src[stride*9] = CLIP((-src[stride*6] + 9*src[stride*8] + 9*src[stride*10] - src[stride*12])>>4);

  1540.         src++;

  1541.     }

  1542. #endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1543. }

  1544. 

  1545. /**

  1546.  * Deinterlace the given block by filtering every second line with a (-1 4 2 4 -1) filter.

  1547.  * will be called for every 8x8 block and can read & write from line 4-15

  1548.  * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.

  1549.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1550.  * this filter will read lines 4-13 and write 5-11

  1551.  */

  1552. static inline void RENAME(deInterlaceFF)(uint8_t src[], int stride, uint8_t *tmp)

  1553. {

  1554. #if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1555.     src+= stride*4;

  1556.     __asm__ volatile(

  1557.         "lea (%0, %1), %%"REG_a"                \n\t"

  1558.         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"

  1559.         "pxor %%mm7, %%mm7                      \n\t"

  1560.         "movq (%2), %%mm0                       \n\t"

  1561. //      0       1       2       3       4       5       6       7       8       9       10

  1562. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx

  1563. 

  1564. #define REAL_DEINT_FF(a,b,c,d)\

  1565.         "movq " #a ", %%mm1                     \n\t"\

  1566.         "movq " #b ", %%mm2                     \n\t"\

  1567.         "movq " #c ", %%mm3                     \n\t"\

  1568.         "movq " #d ", %%mm4                     \n\t"\

  1569.         PAVGB(%%mm3, %%mm1)                          \

  1570.         PAVGB(%%mm4, %%mm0)                          \

  1571.         "movq %%mm0, %%mm3                      \n\t"\

  1572.         "punpcklbw %%mm7, %%mm0                 \n\t"\

  1573.         "punpckhbw %%mm7, %%mm3                 \n\t"\

  1574.         "movq %%mm1, %%mm4                      \n\t"\

  1575.         "punpcklbw %%mm7, %%mm1                 \n\t"\

  1576.         "punpckhbw %%mm7, %%mm4                 \n\t"\

  1577.         "psllw $2, %%mm1                        \n\t"\

  1578.         "psllw $2, %%mm4                        \n\t"\

  1579.         "psubw %%mm0, %%mm1                     \n\t"\

  1580.         "psubw %%mm3, %%mm4                     \n\t"\

  1581.         "movq %%mm2, %%mm5                      \n\t"\

  1582.         "movq %%mm2, %%mm0                      \n\t"\

  1583.         "punpcklbw %%mm7, %%mm2                 \n\t"\

  1584.         "punpckhbw %%mm7, %%mm5                 \n\t"\

  1585.         "paddw %%mm2, %%mm1                     \n\t"\

  1586.         "paddw %%mm5, %%mm4                     \n\t"\

  1587.         "psraw $2, %%mm1                        \n\t"\

  1588.         "psraw $2, %%mm4                        \n\t"\

  1589.         "packuswb %%mm4, %%mm1                  \n\t"\

  1590.         "movq %%mm1, " #b "                     \n\t"\

  1591. 

  1592. #define DEINT_FF(a,b,c,d)  REAL_DEINT_FF(a,b,c,d)

  1593. 

  1594. DEINT_FF((%0)        , (%%REGa)       , (%%REGa, %1), (%%REGa, %1, 2))

  1595. DEINT_FF((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4) , (%%REGd)       )

  1596. DEINT_FF((%0, %1, 4) , (%%REGd)       , (%%REGd, %1), (%%REGd, %1, 2))

  1597. DEINT_FF((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8) , (%%REGd, %1, 4))

  1598. 

  1599.         "movq %%mm0, (%2)                       \n\t"

  1600.         : : "r" (src), "r" ((x86_reg)stride), "r"(tmp)

  1601.         : "%"REG_a, "%"REG_d

  1602.     );

  1603. #else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1604.     int x;

  1605.     src+= stride*4;

  1606.     for(x=0; x<8; x++){

  1607.         int t1= tmp[x];

  1608.         int t2= src[stride*1];

  1609. 

  1610.         src[stride*1]= CLIP((-t1 + 4*src[stride*0] + 2*t2 + 4*src[stride*2] - src[stride*3] + 4)>>3);

  1611.         t1= src[stride*4];

  1612.         src[stride*3]= CLIP((-t2 + 4*src[stride*2] + 2*t1 + 4*src[stride*4] - src[stride*5] + 4)>>3);

  1613.         t2= src[stride*6];

  1614.         src[stride*5]= CLIP((-t1 + 4*src[stride*4] + 2*t2 + 4*src[stride*6] - src[stride*7] + 4)>>3);

  1615.         t1= src[stride*8];

  1616.         src[stride*7]= CLIP((-t2 + 4*src[stride*6] + 2*t1 + 4*src[stride*8] - src[stride*9] + 4)>>3);

  1617.         tmp[x]= t1;

  1618. 

  1619.         src++;

  1620.     }

  1621. #endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1622. }

  1623. 

  1624. /**

  1625.  * Deinterlace the given block by filtering every line with a (-1 2 6 2 -1) filter.

  1626.  * will be called for every 8x8 block and can read & write from line 4-15

  1627.  * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.

  1628.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1629.  * this filter will read lines 4-13 and write 4-11

  1630.  */

  1631. static inline void RENAME(deInterlaceL5)(uint8_t src[], int stride, uint8_t *tmp, uint8_t *tmp2)

  1632. {

  1633. #if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1634.     src+= stride*4;

  1635.     __asm__ volatile(

  1636.         "lea (%0, %1), %%"REG_a"                \n\t"

  1637.         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"

  1638.         "pxor %%mm7, %%mm7                      \n\t"

  1639.         "movq (%2), %%mm0                       \n\t"

  1640.         "movq (%3), %%mm1                       \n\t"

  1641. //      0       1       2       3       4       5       6       7       8       9       10

  1642. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1 ecx

  1643. 

  1644. #define REAL_DEINT_L5(t1,t2,a,b,c)\

  1645.         "movq " #a ", %%mm2                     \n\t"\

  1646.         "movq " #b ", %%mm3                     \n\t"\

  1647.         "movq " #c ", %%mm4                     \n\t"\

  1648.         PAVGB(t2, %%mm3)                             \

  1649.         PAVGB(t1, %%mm4)                             \

  1650.         "movq %%mm2, %%mm5                      \n\t"\

  1651.         "movq %%mm2, " #t1 "                    \n\t"\

  1652.         "punpcklbw %%mm7, %%mm2                 \n\t"\

  1653.         "punpckhbw %%mm7, %%mm5                 \n\t"\

  1654.         "movq %%mm2, %%mm6                      \n\t"\

  1655.         "paddw %%mm2, %%mm2                     \n\t"\

  1656.         "paddw %%mm6, %%mm2                     \n\t"\

  1657.         "movq %%mm5, %%mm6                      \n\t"\

  1658.         "paddw %%mm5, %%mm5                     \n\t"\

  1659.         "paddw %%mm6, %%mm5                     \n\t"\

  1660.         "movq %%mm3, %%mm6                      \n\t"\

  1661.         "punpcklbw %%mm7, %%mm3                 \n\t"\

  1662.         "punpckhbw %%mm7, %%mm6                 \n\t"\

  1663.         "paddw %%mm3, %%mm3                     \n\t"\

  1664.         "paddw %%mm6, %%mm6                     \n\t"\

  1665.         "paddw %%mm3, %%mm2                     \n\t"\

  1666.         "paddw %%mm6, %%mm5                     \n\t"\

  1667.         "movq %%mm4, %%mm6                      \n\t"\

  1668.         "punpcklbw %%mm7, %%mm4                 \n\t"\

  1669.         "punpckhbw %%mm7, %%mm6                 \n\t"\

  1670.         "psubw %%mm4, %%mm2                     \n\t"\

  1671.         "psubw %%mm6, %%mm5                     \n\t"\

  1672.         "psraw $2, %%mm2                        \n\t"\

  1673.         "psraw $2, %%mm5                        \n\t"\

  1674.         "packuswb %%mm5, %%mm2                  \n\t"\

  1675.         "movq %%mm2, " #a "                     \n\t"\

  1676. 

  1677. #define DEINT_L5(t1,t2,a,b,c)  REAL_DEINT_L5(t1,t2,a,b,c)

  1678. 

  1679. DEINT_L5(%%mm0, %%mm1, (%0)           , (%%REGa)       , (%%REGa, %1)   )

  1680. DEINT_L5(%%mm1, %%mm0, (%%REGa)       , (%%REGa, %1)   , (%%REGa, %1, 2))

  1681. DEINT_L5(%%mm0, %%mm1, (%%REGa, %1)   , (%%REGa, %1, 2), (%0, %1, 4)   )

  1682. DEINT_L5(%%mm1, %%mm0, (%%REGa, %1, 2), (%0, %1, 4)    , (%%REGd)       )

  1683. DEINT_L5(%%mm0, %%mm1, (%0, %1, 4)    , (%%REGd)       , (%%REGd, %1)   )

  1684. DEINT_L5(%%mm1, %%mm0, (%%REGd)       , (%%REGd, %1)   , (%%REGd, %1, 2))

  1685. DEINT_L5(%%mm0, %%mm1, (%%REGd, %1)   , (%%REGd, %1, 2), (%0, %1, 8)   )

  1686. DEINT_L5(%%mm1, %%mm0, (%%REGd, %1, 2), (%0, %1, 8)    , (%%REGd, %1, 4))

  1687. 

  1688.         "movq %%mm0, (%2)                       \n\t"

  1689.         "movq %%mm1, (%3)                       \n\t"

  1690.         : : "r" (src), "r" ((x86_reg)stride), "r"(tmp), "r"(tmp2)

  1691.         : "%"REG_a, "%"REG_d

  1692.     );

  1693. #else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1694.     int x;

  1695.     src+= stride*4;

  1696.     for(x=0; x<8; x++){

  1697.         int t1= tmp[x];

  1698.         int t2= tmp2[x];

  1699.         int t3= src[0];

  1700. 

  1701.         src[stride*0]= CLIP((-(t1 + src[stride*2]) + 2*(t2 + src[stride*1]) + 6*t3 + 4)>>3);

  1702.         t1= src[stride*1];

  1703.         src[stride*1]= CLIP((-(t2 + src[stride*3]) + 2*(t3 + src[stride*2]) + 6*t1 + 4)>>3);

  1704.         t2= src[stride*2];

  1705.         src[stride*2]= CLIP((-(t3 + src[stride*4]) + 2*(t1 + src[stride*3]) + 6*t2 + 4)>>3);

  1706.         t3= src[stride*3];

  1707.         src[stride*3]= CLIP((-(t1 + src[stride*5]) + 2*(t2 + src[stride*4]) + 6*t3 + 4)>>3);

  1708.         t1= src[stride*4];

  1709.         src[stride*4]= CLIP((-(t2 + src[stride*6]) + 2*(t3 + src[stride*5]) + 6*t1 + 4)>>3);

  1710.         t2= src[stride*5];

  1711.         src[stride*5]= CLIP((-(t3 + src[stride*7]) + 2*(t1 + src[stride*6]) + 6*t2 + 4)>>3);

  1712.         t3= src[stride*6];

  1713.         src[stride*6]= CLIP((-(t1 + src[stride*8]) + 2*(t2 + src[stride*7]) + 6*t3 + 4)>>3);

  1714.         t1= src[stride*7];

  1715.         src[stride*7]= CLIP((-(t2 + src[stride*9]) + 2*(t3 + src[stride*8]) + 6*t1 + 4)>>3);

  1716. 

  1717.         tmp[x]= t3;

  1718.         tmp2[x]= t1;

  1719. 

  1720.         src++;

  1721.     }

  1722. #endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1723. }

  1724. 

  1725. /**

  1726.  * Deinterlace the given block by filtering all lines with a (1 2 1) filter.

  1727.  * will be called for every 8x8 block and can read & write from line 4-15

  1728.  * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.

  1729.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1730.  * this filter will read lines 4-13 and write 4-11

  1731.  */

  1732. static inline void RENAME(deInterlaceBlendLinear)(uint8_t src[], int stride, uint8_t *tmp)

  1733. {

  1734. #if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1735.     src+= 4*stride;

  1736.     __asm__ volatile(

  1737.         "lea (%0, %1), %%"REG_a"                \n\t"

  1738.         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"

  1739. //      0       1       2       3       4       5       6       7       8       9

  1740. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1

  1741. 

  1742.         "movq (%2), %%mm0                       \n\t" // L0

  1743.         "movq (%%"REG_a"), %%mm1                \n\t" // L2

  1744.         PAVGB(%%mm1, %%mm0)                           // L0+L2

  1745.         "movq (%0), %%mm2                       \n\t" // L1

  1746.         PAVGB(%%mm2, %%mm0)

  1747.         "movq %%mm0, (%0)                       \n\t"

  1748.         "movq (%%"REG_a", %1), %%mm0            \n\t" // L3

  1749.         PAVGB(%%mm0, %%mm2)                           // L1+L3

  1750.         PAVGB(%%mm1, %%mm2)                           // 2L2 + L1 + L3

  1751.         "movq %%mm2, (%%"REG_a")                \n\t"

  1752.         "movq (%%"REG_a", %1, 2), %%mm2         \n\t" // L4

  1753.         PAVGB(%%mm2, %%mm1)                           // L2+L4

  1754.         PAVGB(%%mm0, %%mm1)                           // 2L3 + L2 + L4

  1755.         "movq %%mm1, (%%"REG_a", %1)            \n\t"

  1756.         "movq (%0, %1, 4), %%mm1                \n\t" // L5

  1757.         PAVGB(%%mm1, %%mm0)                           // L3+L5

  1758.         PAVGB(%%mm2, %%mm0)                           // 2L4 + L3 + L5

  1759.         "movq %%mm0, (%%"REG_a", %1, 2)         \n\t"

  1760.         "movq (%%"REG_d"), %%mm0                \n\t" // L6

  1761.         PAVGB(%%mm0, %%mm2)                           // L4+L6

  1762.         PAVGB(%%mm1, %%mm2)                           // 2L5 + L4 + L6

  1763.         "movq %%mm2, (%0, %1, 4)                \n\t"

  1764.         "movq (%%"REG_d", %1), %%mm2            \n\t" // L7

  1765.         PAVGB(%%mm2, %%mm1)                           // L5+L7

  1766.         PAVGB(%%mm0, %%mm1)                           // 2L6 + L5 + L7

  1767.         "movq %%mm1, (%%"REG_d")                \n\t"

  1768.         "movq (%%"REG_d", %1, 2), %%mm1         \n\t" // L8

  1769.         PAVGB(%%mm1, %%mm0)                           // L6+L8

  1770.         PAVGB(%%mm2, %%mm0)                           // 2L7 + L6 + L8

  1771.         "movq %%mm0, (%%"REG_d", %1)            \n\t"

  1772.         "movq (%0, %1, 8), %%mm0                \n\t" // L9

  1773.         PAVGB(%%mm0, %%mm2)                           // L7+L9

  1774.         PAVGB(%%mm1, %%mm2)                           // 2L8 + L7 + L9

  1775.         "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"

  1776.         "movq %%mm1, (%2)                       \n\t"

  1777. 

  1778.         : : "r" (src), "r" ((x86_reg)stride), "r" (tmp)

  1779.         : "%"REG_a, "%"REG_d

  1780.     );

  1781. #else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1782.     int a, b, c, x;

  1783.     src+= 4*stride;

  1784. 

  1785.     for(x=0; x<2; x++){

  1786.         a= *(uint32_t*)&tmp[stride*0];

  1787.         b= *(uint32_t*)&src[stride*0];

  1788.         c= *(uint32_t*)&src[stride*1];

  1789.         a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);

  1790.         *(uint32_t*)&src[stride*0]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1791. 

  1792.         a= *(uint32_t*)&src[stride*2];

  1793.         b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);

  1794.         *(uint32_t*)&src[stride*1]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);

  1795. 

  1796.         b= *(uint32_t*)&src[stride*3];

  1797.         c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);

  1798.         *(uint32_t*)&src[stride*2]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);

  1799. 

  1800.         c= *(uint32_t*)&src[stride*4];

  1801.         a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);

  1802.         *(uint32_t*)&src[stride*3]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1803. 

  1804.         a= *(uint32_t*)&src[stride*5];

  1805.         b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);

  1806.         *(uint32_t*)&src[stride*4]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);

  1807. 

  1808.         b= *(uint32_t*)&src[stride*6];

  1809.         c= (b&c) + (((b^c)&0xFEFEFEFEUL)>>1);

  1810.         *(uint32_t*)&src[stride*5]= (c|a) - (((c^a)&0xFEFEFEFEUL)>>1);

  1811. 

  1812.         c= *(uint32_t*)&src[stride*7];

  1813.         a= (a&c) + (((a^c)&0xFEFEFEFEUL)>>1);

  1814.         *(uint32_t*)&src[stride*6]= (a|b) - (((a^b)&0xFEFEFEFEUL)>>1);

  1815. 

  1816.         a= *(uint32_t*)&src[stride*8];

  1817.         b= (a&b) + (((a^b)&0xFEFEFEFEUL)>>1);

  1818.         *(uint32_t*)&src[stride*7]= (c|b) - (((c^b)&0xFEFEFEFEUL)>>1);

  1819. 

  1820.         *(uint32_t*)&tmp[stride*0]= c;

  1821.         src += 4;

  1822.         tmp += 4;

  1823.     }

  1824. #endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  1825. }

  1826. 

  1827. /**

  1828.  * Deinterlace the given block by applying a median filter to every second line.

  1829.  * will be called for every 8x8 block and can read & write from line 4-15,

  1830.  * lines 0-3 have been passed through the deblock / dering filters already, but can be read, too.

  1831.  * lines 4-12 will be read into the deblocking filter and should be deinterlaced

  1832.  */

  1833. static inline void RENAME(deInterlaceMedian)(uint8_t src[], int stride)

  1834. {

  1835. #if TEMPLATE_PP_MMX

  1836.     src+= 4*stride;

  1837. #if TEMPLATE_PP_MMXEXT

  1838.     __asm__ volatile(

  1839.         "lea (%0, %1), %%"REG_a"                \n\t"

  1840.         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"

  1841. //      0       1       2       3       4       5       6       7       8       9

  1842. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1

  1843. 

  1844.         "movq (%0), %%mm0                       \n\t" //

  1845.         "movq (%%"REG_a", %1), %%mm2            \n\t" //

  1846.         "movq (%%"REG_a"), %%mm1                \n\t" //

  1847.         "movq %%mm0, %%mm3                      \n\t"

  1848.         "pmaxub %%mm1, %%mm0                    \n\t" //

  1849.         "pminub %%mm3, %%mm1                    \n\t" //

  1850.         "pmaxub %%mm2, %%mm1                    \n\t" //

  1851.         "pminub %%mm1, %%mm0                    \n\t"

  1852.         "movq %%mm0, (%%"REG_a")                \n\t"

  1853. 

  1854.         "movq (%0, %1, 4), %%mm0                \n\t" //

  1855.         "movq (%%"REG_a", %1, 2), %%mm1         \n\t" //

  1856.         "movq %%mm2, %%mm3                      \n\t"

  1857.         "pmaxub %%mm1, %%mm2                    \n\t" //

  1858.         "pminub %%mm3, %%mm1                    \n\t" //

  1859.         "pmaxub %%mm0, %%mm1                    \n\t" //

  1860.         "pminub %%mm1, %%mm2                    \n\t"

  1861.         "movq %%mm2, (%%"REG_a", %1, 2)         \n\t"

  1862. 

  1863.         "movq (%%"REG_d"), %%mm2                \n\t" //

  1864.         "movq (%%"REG_d", %1), %%mm1            \n\t" //

  1865.         "movq %%mm2, %%mm3                      \n\t"

  1866.         "pmaxub %%mm0, %%mm2                    \n\t" //

  1867.         "pminub %%mm3, %%mm0                    \n\t" //

  1868.         "pmaxub %%mm1, %%mm0                    \n\t" //

  1869.         "pminub %%mm0, %%mm2                    \n\t"

  1870.         "movq %%mm2, (%%"REG_d")                \n\t"

  1871. 

  1872.         "movq (%%"REG_d", %1, 2), %%mm2         \n\t" //

  1873.         "movq (%0, %1, 8), %%mm0                \n\t" //

  1874.         "movq %%mm2, %%mm3                      \n\t"

  1875.         "pmaxub %%mm0, %%mm2                    \n\t" //

  1876.         "pminub %%mm3, %%mm0                    \n\t" //

  1877.         "pmaxub %%mm1, %%mm0                    \n\t" //

  1878.         "pminub %%mm0, %%mm2                    \n\t"

  1879.         "movq %%mm2, (%%"REG_d", %1, 2)         \n\t"

  1880. 

  1881. 

  1882.         : : "r" (src), "r" ((x86_reg)stride)

  1883.         : "%"REG_a, "%"REG_d

  1884.     );

  1885. 

  1886. #else // MMX without MMX2

  1887.     __asm__ volatile(

  1888.         "lea (%0, %1), %%"REG_a"                \n\t"

  1889.         "lea (%%"REG_a", %1, 4), %%"REG_d"      \n\t"

  1890. //      0       1       2       3       4       5       6       7       8       9

  1891. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1

  1892.         "pxor %%mm7, %%mm7                      \n\t"

  1893. 

  1894. #define REAL_MEDIAN(a,b,c)\

  1895.         "movq " #a ", %%mm0                     \n\t"\

  1896.         "movq " #b ", %%mm2                     \n\t"\

  1897.         "movq " #c ", %%mm1                     \n\t"\

  1898.         "movq %%mm0, %%mm3                      \n\t"\

  1899.         "movq %%mm1, %%mm4                      \n\t"\

  1900.         "movq %%mm2, %%mm5                      \n\t"\

  1901.         "psubusb %%mm1, %%mm3                   \n\t"\

  1902.         "psubusb %%mm2, %%mm4                   \n\t"\

  1903.         "psubusb %%mm0, %%mm5                   \n\t"\

  1904.         "pcmpeqb %%mm7, %%mm3                   \n\t"\

  1905.         "pcmpeqb %%mm7, %%mm4                   \n\t"\

  1906.         "pcmpeqb %%mm7, %%mm5                   \n\t"\

  1907.         "movq %%mm3, %%mm6                      \n\t"\

  1908.         "pxor %%mm4, %%mm3                      \n\t"\

  1909.         "pxor %%mm5, %%mm4                      \n\t"\

  1910.         "pxor %%mm6, %%mm5                      \n\t"\

  1911.         "por %%mm3, %%mm1                       \n\t"\

  1912.         "por %%mm4, %%mm2                       \n\t"\

  1913.         "por %%mm5, %%mm0                       \n\t"\

  1914.         "pand %%mm2, %%mm0                      \n\t"\

  1915.         "pand %%mm1, %%mm0                      \n\t"\

  1916.         "movq %%mm0, " #b "                     \n\t"

  1917. #define MEDIAN(a,b,c)  REAL_MEDIAN(a,b,c)

  1918. 

  1919. MEDIAN((%0)        , (%%REGa)       , (%%REGa, %1))

  1920. MEDIAN((%%REGa, %1), (%%REGa, %1, 2), (%0, %1, 4))

  1921. MEDIAN((%0, %1, 4) , (%%REGd)       , (%%REGd, %1))

  1922. MEDIAN((%%REGd, %1), (%%REGd, %1, 2), (%0, %1, 8))

  1923. 

  1924.         : : "r" (src), "r" ((x86_reg)stride)

  1925.         : "%"REG_a, "%"REG_d

  1926.     );

  1927. #endif //TEMPLATE_PP_MMXEXT

  1928. #else //TEMPLATE_PP_MMX

  1929.     int x, y;

  1930.     src+= 4*stride;

  1931.     // FIXME - there should be a way to do a few columns in parallel like w/mmx

  1932.     for(x=0; x<8; x++){

  1933.         uint8_t *colsrc = src;

  1934.         for (y=0; y<4; y++){

  1935.             int a, b, c, d, e, f;

  1936.             a = colsrc[0       ];

  1937.             b = colsrc[stride  ];

  1938.             c = colsrc[stride*2];

  1939.             d = (a-b)>>31;

  1940.             e = (b-c)>>31;

  1941.             f = (c-a)>>31;

  1942.             colsrc[stride  ] = (a|(d^f)) & (b|(d^e)) & (c|(e^f));

  1943.             colsrc += stride*2;

  1944.         }

  1945.         src++;

  1946.     }

  1947. #endif //TEMPLATE_PP_MMX

  1948. }

  1949. 

  1950. #if TEMPLATE_PP_MMX

  1951. /**

  1952.  * Transpose and shift the given 8x8 Block into dst1 and dst2.

  1953.  */

  1954. static inline void RENAME(transpose1)(uint8_t *dst1, uint8_t *dst2, uint8_t *src, int srcStride)

  1955. {

  1956.     __asm__(

  1957.         "lea (%0, %1), %%"REG_a"                \n\t"

  1958. //      0       1       2       3       4       5       6       7       8       9

  1959. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1

  1960.         "movq (%0), %%mm0                       \n\t" // 12345678

  1961.         "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh

  1962.         "movq %%mm0, %%mm2                      \n\t" // 12345678

  1963.         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d

  1964.         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h

  1965. 

  1966.         "movq (%%"REG_a", %1), %%mm1            \n\t"

  1967.         "movq (%%"REG_a", %1, 2), %%mm3         \n\t"

  1968.         "movq %%mm1, %%mm4                      \n\t"

  1969.         "punpcklbw %%mm3, %%mm1                 \n\t"

  1970.         "punpckhbw %%mm3, %%mm4                 \n\t"

  1971. 

  1972.         "movq %%mm0, %%mm3                      \n\t"

  1973.         "punpcklwd %%mm1, %%mm0                 \n\t"

  1974.         "punpckhwd %%mm1, %%mm3                 \n\t"

  1975.         "movq %%mm2, %%mm1                      \n\t"

  1976.         "punpcklwd %%mm4, %%mm2                 \n\t"

  1977.         "punpckhwd %%mm4, %%mm1                 \n\t"

  1978. 

  1979.         "movd %%mm0, 128(%2)                    \n\t"

  1980.         "psrlq $32, %%mm0                       \n\t"

  1981.         "movd %%mm0, 144(%2)                    \n\t"

  1982.         "movd %%mm3, 160(%2)                    \n\t"

  1983.         "psrlq $32, %%mm3                       \n\t"

  1984.         "movd %%mm3, 176(%2)                    \n\t"

  1985.         "movd %%mm3, 48(%3)                     \n\t"

  1986.         "movd %%mm2, 192(%2)                    \n\t"

  1987.         "movd %%mm2, 64(%3)                     \n\t"

  1988.         "psrlq $32, %%mm2                       \n\t"

  1989.         "movd %%mm2, 80(%3)                     \n\t"

  1990.         "movd %%mm1, 96(%3)                     \n\t"

  1991.         "psrlq $32, %%mm1                       \n\t"

  1992.         "movd %%mm1, 112(%3)                    \n\t"

  1993. 

  1994.         "lea (%%"REG_a", %1, 4), %%"REG_a"      \n\t"

  1995. 

  1996.         "movq (%0, %1, 4), %%mm0                \n\t" // 12345678

  1997.         "movq (%%"REG_a"), %%mm1                \n\t" // abcdefgh

  1998.         "movq %%mm0, %%mm2                      \n\t" // 12345678

  1999.         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d

  2000.         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h

  2001. 

  2002.         "movq (%%"REG_a", %1), %%mm1            \n\t"

  2003.         "movq (%%"REG_a", %1, 2), %%mm3         \n\t"

  2004.         "movq %%mm1, %%mm4                      \n\t"

  2005.         "punpcklbw %%mm3, %%mm1                 \n\t"

  2006.         "punpckhbw %%mm3, %%mm4                 \n\t"

  2007. 

  2008.         "movq %%mm0, %%mm3                      \n\t"

  2009.         "punpcklwd %%mm1, %%mm0                 \n\t"

  2010.         "punpckhwd %%mm1, %%mm3                 \n\t"

  2011.         "movq %%mm2, %%mm1                      \n\t"

  2012.         "punpcklwd %%mm4, %%mm2                 \n\t"

  2013.         "punpckhwd %%mm4, %%mm1                 \n\t"

  2014. 

  2015.         "movd %%mm0, 132(%2)                    \n\t"

  2016.         "psrlq $32, %%mm0                       \n\t"

  2017.         "movd %%mm0, 148(%2)                    \n\t"

  2018.         "movd %%mm3, 164(%2)                    \n\t"

  2019.         "psrlq $32, %%mm3                       \n\t"

  2020.         "movd %%mm3, 180(%2)                    \n\t"

  2021.         "movd %%mm3, 52(%3)                     \n\t"

  2022.         "movd %%mm2, 196(%2)                    \n\t"

  2023.         "movd %%mm2, 68(%3)                     \n\t"

  2024.         "psrlq $32, %%mm2                       \n\t"

  2025.         "movd %%mm2, 84(%3)                     \n\t"

  2026.         "movd %%mm1, 100(%3)                    \n\t"

  2027.         "psrlq $32, %%mm1                       \n\t"

  2028.         "movd %%mm1, 116(%3)                    \n\t"

  2029. 

  2030. 

  2031.         :: "r" (src), "r" ((x86_reg)srcStride), "r" (dst1), "r" (dst2)

  2032.         : "%"REG_a

  2033.     );

  2034. }

  2035. 

  2036. /**

  2037.  * Transpose the given 8x8 block.

  2038.  */

  2039. static inline void RENAME(transpose2)(uint8_t *dst, int dstStride, uint8_t *src)

  2040. {

  2041.     __asm__(

  2042.         "lea (%0, %1), %%"REG_a"                \n\t"

  2043.         "lea (%%"REG_a",%1,4), %%"REG_d"        \n\t"

  2044. //      0       1       2       3       4       5       6       7       8       9

  2045. //      %0      eax     eax+%1  eax+2%1 %0+4%1  edx     edx+%1  edx+2%1 %0+8%1  edx+4%1

  2046.         "movq (%2), %%mm0                       \n\t" // 12345678

  2047.         "movq 16(%2), %%mm1                     \n\t" // abcdefgh

  2048.         "movq %%mm0, %%mm2                      \n\t" // 12345678

  2049.         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d

  2050.         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h

  2051. 

  2052.         "movq 32(%2), %%mm1                     \n\t"

  2053.         "movq 48(%2), %%mm3                     \n\t"

  2054.         "movq %%mm1, %%mm4                      \n\t"

  2055.         "punpcklbw %%mm3, %%mm1                 \n\t"

  2056.         "punpckhbw %%mm3, %%mm4                 \n\t"

  2057. 

  2058.         "movq %%mm0, %%mm3                      \n\t"

  2059.         "punpcklwd %%mm1, %%mm0                 \n\t"

  2060.         "punpckhwd %%mm1, %%mm3                 \n\t"

  2061.         "movq %%mm2, %%mm1                      \n\t"

  2062.         "punpcklwd %%mm4, %%mm2                 \n\t"

  2063.         "punpckhwd %%mm4, %%mm1                 \n\t"

  2064. 

  2065.         "movd %%mm0, (%0)                       \n\t"

  2066.         "psrlq $32, %%mm0                       \n\t"

  2067.         "movd %%mm0, (%%"REG_a")                \n\t"

  2068.         "movd %%mm3, (%%"REG_a", %1)            \n\t"

  2069.         "psrlq $32, %%mm3                       \n\t"

  2070.         "movd %%mm3, (%%"REG_a", %1, 2)         \n\t"

  2071.         "movd %%mm2, (%0, %1, 4)                \n\t"

  2072.         "psrlq $32, %%mm2                       \n\t"

  2073.         "movd %%mm2, (%%"REG_d")                \n\t"

  2074.         "movd %%mm1, (%%"REG_d", %1)            \n\t"

  2075.         "psrlq $32, %%mm1                       \n\t"

  2076.         "movd %%mm1, (%%"REG_d", %1, 2)         \n\t"

  2077. 

  2078. 

  2079.         "movq 64(%2), %%mm0                     \n\t" // 12345678

  2080.         "movq 80(%2), %%mm1                     \n\t" // abcdefgh

  2081.         "movq %%mm0, %%mm2                      \n\t" // 12345678

  2082.         "punpcklbw %%mm1, %%mm0                 \n\t" // 1a2b3c4d

  2083.         "punpckhbw %%mm1, %%mm2                 \n\t" // 5e6f7g8h

  2084. 

  2085.         "movq 96(%2), %%mm1                     \n\t"

  2086.         "movq 112(%2), %%mm3                    \n\t"

  2087.         "movq %%mm1, %%mm4                      \n\t"

  2088.         "punpcklbw %%mm3, %%mm1                 \n\t"

  2089.         "punpckhbw %%mm3, %%mm4                 \n\t"

  2090. 

  2091.         "movq %%mm0, %%mm3                      \n\t"

  2092.         "punpcklwd %%mm1, %%mm0                 \n\t"

  2093.         "punpckhwd %%mm1, %%mm3                 \n\t"

  2094.         "movq %%mm2, %%mm1                      \n\t"

  2095.         "punpcklwd %%mm4, %%mm2                 \n\t"

  2096.         "punpckhwd %%mm4, %%mm1                 \n\t"

  2097. 

  2098.         "movd %%mm0, 4(%0)                      \n\t"

  2099.         "psrlq $32, %%mm0                       \n\t"

  2100.         "movd %%mm0, 4(%%"REG_a")               \n\t"

  2101.         "movd %%mm3, 4(%%"REG_a", %1)           \n\t"

  2102.         "psrlq $32, %%mm3                       \n\t"

  2103.         "movd %%mm3, 4(%%"REG_a", %1, 2)        \n\t"

  2104.         "movd %%mm2, 4(%0, %1, 4)               \n\t"

  2105.         "psrlq $32, %%mm2                       \n\t"

  2106.         "movd %%mm2, 4(%%"REG_d")               \n\t"

  2107.         "movd %%mm1, 4(%%"REG_d", %1)           \n\t"

  2108.         "psrlq $32, %%mm1                       \n\t"

  2109.         "movd %%mm1, 4(%%"REG_d", %1, 2)        \n\t"

  2110. 

  2111.         :: "r" (dst), "r" ((x86_reg)dstStride), "r" (src)

  2112.         : "%"REG_a, "%"REG_d

  2113.     );

  2114. }

  2115. #endif //TEMPLATE_PP_MMX

  2116. //static long test=0;

  2117. 

  2118. #if !TEMPLATE_PP_ALTIVEC

  2119. static inline void RENAME(tempNoiseReducer)(uint8_t *src, int stride,

  2120.                                     uint8_t *tempBlurred, uint32_t *tempBlurredPast, int *maxNoise)

  2121. {

  2122.     // to save a register (FIXME do this outside of the loops)

  2123.     tempBlurredPast[127]= maxNoise[0];

  2124.     tempBlurredPast[128]= maxNoise[1];

  2125.     tempBlurredPast[129]= maxNoise[2];

  2126. 

  2127. #define FAST_L2_DIFF

  2128. //#define L1_DIFF //u should change the thresholds too if u try that one

  2129. #if TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  2130.     __asm__ volatile(

  2131.         "lea (%2, %2, 2), %%"REG_a"             \n\t" // 3*stride

  2132.         "lea (%2, %2, 4), %%"REG_d"             \n\t" // 5*stride

  2133.         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride

  2134. //      0       1       2       3       4       5       6       7       8       9

  2135. //      %x      %x+%2   %x+2%2  %x+eax  %x+4%2  %x+edx  %x+2eax %x+ecx  %x+8%2

  2136. //FIXME reorder?

  2137. #ifdef L1_DIFF //needs mmx2

  2138.         "movq (%0), %%mm0                       \n\t" // L0

  2139.         "psadbw (%1), %%mm0                     \n\t" // |L0-R0|

  2140.         "movq (%0, %2), %%mm1                   \n\t" // L1

  2141.         "psadbw (%1, %2), %%mm1                 \n\t" // |L1-R1|

  2142.         "movq (%0, %2, 2), %%mm2                \n\t" // L2

  2143.         "psadbw (%1, %2, 2), %%mm2              \n\t" // |L2-R2|

  2144.         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3

  2145.         "psadbw (%1, %%"REG_a"), %%mm3          \n\t" // |L3-R3|

  2146. 

  2147.         "movq (%0, %2, 4), %%mm4                \n\t" // L4

  2148.         "paddw %%mm1, %%mm0                     \n\t"

  2149.         "psadbw (%1, %2, 4), %%mm4              \n\t" // |L4-R4|

  2150.         "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5

  2151.         "paddw %%mm2, %%mm0                     \n\t"

  2152.         "psadbw (%1, %%"REG_d"), %%mm5          \n\t" // |L5-R5|

  2153.         "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6

  2154.         "paddw %%mm3, %%mm0                     \n\t"

  2155.         "psadbw (%1, %%"REG_a", 2), %%mm6       \n\t" // |L6-R6|

  2156.         "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7

  2157.         "paddw %%mm4, %%mm0                     \n\t"

  2158.         "psadbw (%1, %%"REG_c"), %%mm7          \n\t" // |L7-R7|

  2159.         "paddw %%mm5, %%mm6                     \n\t"

  2160.         "paddw %%mm7, %%mm6                     \n\t"

  2161.         "paddw %%mm6, %%mm0                     \n\t"

  2162. #else //L1_DIFF

  2163. #if defined (FAST_L2_DIFF)

  2164.         "pcmpeqb %%mm7, %%mm7                   \n\t"

  2165.         "movq "MANGLE(b80)", %%mm6              \n\t"

  2166.         "pxor %%mm0, %%mm0                      \n\t"

  2167. #define REAL_L2_DIFF_CORE(a, b)\

  2168.         "movq " #a ", %%mm5                     \n\t"\

  2169.         "movq " #b ", %%mm2                     \n\t"\

  2170.         "pxor %%mm7, %%mm2                      \n\t"\

  2171.         PAVGB(%%mm2, %%mm5)\

  2172.         "paddb %%mm6, %%mm5                     \n\t"\

  2173.         "movq %%mm5, %%mm2                      \n\t"\

  2174.         "psllw $8, %%mm5                        \n\t"\

  2175.         "pmaddwd %%mm5, %%mm5                   \n\t"\

  2176.         "pmaddwd %%mm2, %%mm2                   \n\t"\

  2177.         "paddd %%mm2, %%mm5                     \n\t"\

  2178.         "psrld $14, %%mm5                       \n\t"\

  2179.         "paddd %%mm5, %%mm0                     \n\t"

  2180. 

  2181. #else //defined (FAST_L2_DIFF)

  2182.         "pxor %%mm7, %%mm7                      \n\t"

  2183.         "pxor %%mm0, %%mm0                      \n\t"

  2184. #define REAL_L2_DIFF_CORE(a, b)\

  2185.         "movq " #a ", %%mm5                     \n\t"\

  2186.         "movq " #b ", %%mm2                     \n\t"\

  2187.         "movq %%mm5, %%mm1                      \n\t"\

  2188.         "movq %%mm2, %%mm3                      \n\t"\

  2189.         "punpcklbw %%mm7, %%mm5                 \n\t"\

  2190.         "punpckhbw %%mm7, %%mm1                 \n\t"\

  2191.         "punpcklbw %%mm7, %%mm2                 \n\t"\

  2192.         "punpckhbw %%mm7, %%mm3                 \n\t"\

  2193.         "psubw %%mm2, %%mm5                     \n\t"\

  2194.         "psubw %%mm3, %%mm1                     \n\t"\

  2195.         "pmaddwd %%mm5, %%mm5                   \n\t"\

  2196.         "pmaddwd %%mm1, %%mm1                   \n\t"\

  2197.         "paddd %%mm1, %%mm5                     \n\t"\

  2198.         "paddd %%mm5, %%mm0                     \n\t"

  2199. 

  2200. #endif //defined (FAST_L2_DIFF)

  2201. 

  2202. #define L2_DIFF_CORE(a, b)  REAL_L2_DIFF_CORE(a, b)

  2203. 

  2204. L2_DIFF_CORE((%0)          , (%1))

  2205. L2_DIFF_CORE((%0, %2)      , (%1, %2))

  2206. L2_DIFF_CORE((%0, %2, 2)   , (%1, %2, 2))

  2207. L2_DIFF_CORE((%0, %%REGa)  , (%1, %%REGa))

  2208. L2_DIFF_CORE((%0, %2, 4)   , (%1, %2, 4))

  2209. L2_DIFF_CORE((%0, %%REGd)  , (%1, %%REGd))

  2210. L2_DIFF_CORE((%0, %%REGa,2), (%1, %%REGa,2))

  2211. L2_DIFF_CORE((%0, %%REGc)  , (%1, %%REGc))

  2212. 

  2213. #endif //L1_DIFF

  2214. 

  2215.         "movq %%mm0, %%mm4                      \n\t"

  2216.         "psrlq $32, %%mm0                       \n\t"

  2217.         "paddd %%mm0, %%mm4                     \n\t"

  2218.         "movd %%mm4, %%ecx                      \n\t"

  2219.         "shll $2, %%ecx                         \n\t"

  2220.         "mov %3, %%"REG_d"                      \n\t"

  2221.         "addl -4(%%"REG_d"), %%ecx              \n\t"

  2222.         "addl 4(%%"REG_d"), %%ecx               \n\t"

  2223.         "addl -1024(%%"REG_d"), %%ecx           \n\t"

  2224.         "addl $4, %%ecx                         \n\t"

  2225.         "addl 1024(%%"REG_d"), %%ecx            \n\t"

  2226.         "shrl $3, %%ecx                         \n\t"

  2227.         "movl %%ecx, (%%"REG_d")                \n\t"

  2228. 

  2229. //        "mov %3, %%"REG_c"                      \n\t"

  2230. //        "mov %%"REG_c", test                    \n\t"

  2231. //        "jmp 4f                                 \n\t"

  2232.         "cmpl 512(%%"REG_d"), %%ecx             \n\t"

  2233.         " jb 2f                                 \n\t"

  2234.         "cmpl 516(%%"REG_d"), %%ecx             \n\t"

  2235.         " jb 1f                                 \n\t"

  2236. 

  2237.         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride

  2238.         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride

  2239.         "movq (%0), %%mm0                       \n\t" // L0

  2240.         "movq (%0, %2), %%mm1                   \n\t" // L1

  2241.         "movq (%0, %2, 2), %%mm2                \n\t" // L2

  2242.         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3

  2243.         "movq (%0, %2, 4), %%mm4                \n\t" // L4

  2244.         "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5

  2245.         "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6

  2246.         "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7

  2247.         "movq %%mm0, (%1)                       \n\t" // L0

  2248.         "movq %%mm1, (%1, %2)                   \n\t" // L1

  2249.         "movq %%mm2, (%1, %2, 2)                \n\t" // L2

  2250.         "movq %%mm3, (%1, %%"REG_a")            \n\t" // L3

  2251.         "movq %%mm4, (%1, %2, 4)                \n\t" // L4

  2252.         "movq %%mm5, (%1, %%"REG_d")            \n\t" // L5

  2253.         "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // L6

  2254.         "movq %%mm7, (%1, %%"REG_c")            \n\t" // L7

  2255.         "jmp 4f                                 \n\t"

  2256. 

  2257.         "1:                                     \n\t"

  2258.         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride

  2259.         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride

  2260.         "movq (%0), %%mm0                       \n\t" // L0

  2261.         PAVGB((%1), %%mm0)                            // L0

  2262.         "movq (%0, %2), %%mm1                   \n\t" // L1

  2263.         PAVGB((%1, %2), %%mm1)                        // L1

  2264.         "movq (%0, %2, 2), %%mm2                \n\t" // L2

  2265.         PAVGB((%1, %2, 2), %%mm2)                     // L2

  2266.         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3

  2267.         PAVGB((%1, %%REGa), %%mm3)                    // L3

  2268.         "movq (%0, %2, 4), %%mm4                \n\t" // L4

  2269.         PAVGB((%1, %2, 4), %%mm4)                     // L4

  2270.         "movq (%0, %%"REG_d"), %%mm5            \n\t" // L5

  2271.         PAVGB((%1, %%REGd), %%mm5)                    // L5

  2272.         "movq (%0, %%"REG_a", 2), %%mm6         \n\t" // L6

  2273.         PAVGB((%1, %%REGa, 2), %%mm6)                 // L6

  2274.         "movq (%0, %%"REG_c"), %%mm7            \n\t" // L7

  2275.         PAVGB((%1, %%REGc), %%mm7)                    // L7

  2276.         "movq %%mm0, (%1)                       \n\t" // R0

  2277.         "movq %%mm1, (%1, %2)                   \n\t" // R1

  2278.         "movq %%mm2, (%1, %2, 2)                \n\t" // R2

  2279.         "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3

  2280.         "movq %%mm4, (%1, %2, 4)                \n\t" // R4

  2281.         "movq %%mm5, (%1, %%"REG_d")            \n\t" // R5

  2282.         "movq %%mm6, (%1, %%"REG_a", 2)         \n\t" // R6

  2283.         "movq %%mm7, (%1, %%"REG_c")            \n\t" // R7

  2284.         "movq %%mm0, (%0)                       \n\t" // L0

  2285.         "movq %%mm1, (%0, %2)                   \n\t" // L1

  2286.         "movq %%mm2, (%0, %2, 2)                \n\t" // L2

  2287.         "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3

  2288.         "movq %%mm4, (%0, %2, 4)                \n\t" // L4

  2289.         "movq %%mm5, (%0, %%"REG_d")            \n\t" // L5

  2290.         "movq %%mm6, (%0, %%"REG_a", 2)         \n\t" // L6

  2291.         "movq %%mm7, (%0, %%"REG_c")            \n\t" // L7

  2292.         "jmp 4f                                 \n\t"

  2293. 

  2294.         "2:                                     \n\t"

  2295.         "cmpl 508(%%"REG_d"), %%ecx             \n\t"

  2296.         " jb 3f                                 \n\t"

  2297. 

  2298.         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride

  2299.         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride

  2300.         "movq (%0), %%mm0                       \n\t" // L0

  2301.         "movq (%0, %2), %%mm1                   \n\t" // L1

  2302.         "movq (%0, %2, 2), %%mm2                \n\t" // L2

  2303.         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3

  2304.         "movq (%1), %%mm4                       \n\t" // R0

  2305.         "movq (%1, %2), %%mm5                   \n\t" // R1

  2306.         "movq (%1, %2, 2), %%mm6                \n\t" // R2

  2307.         "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3

  2308.         PAVGB(%%mm4, %%mm0)

  2309.         PAVGB(%%mm5, %%mm1)

  2310.         PAVGB(%%mm6, %%mm2)

  2311.         PAVGB(%%mm7, %%mm3)

  2312.         PAVGB(%%mm4, %%mm0)

  2313.         PAVGB(%%mm5, %%mm1)

  2314.         PAVGB(%%mm6, %%mm2)

  2315.         PAVGB(%%mm7, %%mm3)

  2316.         "movq %%mm0, (%1)                       \n\t" // R0

  2317.         "movq %%mm1, (%1, %2)                   \n\t" // R1

  2318.         "movq %%mm2, (%1, %2, 2)                \n\t" // R2

  2319.         "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3

  2320.         "movq %%mm0, (%0)                       \n\t" // L0

  2321.         "movq %%mm1, (%0, %2)                   \n\t" // L1

  2322.         "movq %%mm2, (%0, %2, 2)                \n\t" // L2

  2323.         "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3

  2324. 

  2325.         "movq (%0, %2, 4), %%mm0                \n\t" // L4

  2326.         "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5

  2327.         "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6

  2328.         "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7

  2329.         "movq (%1, %2, 4), %%mm4                \n\t" // R4

  2330.         "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5

  2331.         "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6

  2332.         "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7

  2333.         PAVGB(%%mm4, %%mm0)

  2334.         PAVGB(%%mm5, %%mm1)

  2335.         PAVGB(%%mm6, %%mm2)

  2336.         PAVGB(%%mm7, %%mm3)

  2337.         PAVGB(%%mm4, %%mm0)

  2338.         PAVGB(%%mm5, %%mm1)

  2339.         PAVGB(%%mm6, %%mm2)

  2340.         PAVGB(%%mm7, %%mm3)

  2341.         "movq %%mm0, (%1, %2, 4)                \n\t" // R4

  2342.         "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5

  2343.         "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6

  2344.         "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7

  2345.         "movq %%mm0, (%0, %2, 4)                \n\t" // L4

  2346.         "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5

  2347.         "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6

  2348.         "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7

  2349.         "jmp 4f                                 \n\t"

  2350. 

  2351.         "3:                                     \n\t"

  2352.         "lea (%%"REG_a", %2, 2), %%"REG_d"      \n\t" // 5*stride

  2353.         "lea (%%"REG_d", %2, 2), %%"REG_c"      \n\t" // 7*stride

  2354.         "movq (%0), %%mm0                       \n\t" // L0

  2355.         "movq (%0, %2), %%mm1                   \n\t" // L1

  2356.         "movq (%0, %2, 2), %%mm2                \n\t" // L2

  2357.         "movq (%0, %%"REG_a"), %%mm3            \n\t" // L3

  2358.         "movq (%1), %%mm4                       \n\t" // R0

  2359.         "movq (%1, %2), %%mm5                   \n\t" // R1

  2360.         "movq (%1, %2, 2), %%mm6                \n\t" // R2

  2361.         "movq (%1, %%"REG_a"), %%mm7            \n\t" // R3

  2362.         PAVGB(%%mm4, %%mm0)

  2363.         PAVGB(%%mm5, %%mm1)

  2364.         PAVGB(%%mm6, %%mm2)

  2365.         PAVGB(%%mm7, %%mm3)

  2366.         PAVGB(%%mm4, %%mm0)

  2367.         PAVGB(%%mm5, %%mm1)

  2368.         PAVGB(%%mm6, %%mm2)

  2369.         PAVGB(%%mm7, %%mm3)

  2370.         PAVGB(%%mm4, %%mm0)

  2371.         PAVGB(%%mm5, %%mm1)

  2372.         PAVGB(%%mm6, %%mm2)

  2373.         PAVGB(%%mm7, %%mm3)

  2374.         "movq %%mm0, (%1)                       \n\t" // R0

  2375.         "movq %%mm1, (%1, %2)                   \n\t" // R1

  2376.         "movq %%mm2, (%1, %2, 2)                \n\t" // R2

  2377.         "movq %%mm3, (%1, %%"REG_a")            \n\t" // R3

  2378.         "movq %%mm0, (%0)                       \n\t" // L0

  2379.         "movq %%mm1, (%0, %2)                   \n\t" // L1

  2380.         "movq %%mm2, (%0, %2, 2)                \n\t" // L2

  2381.         "movq %%mm3, (%0, %%"REG_a")            \n\t" // L3

  2382. 

  2383.         "movq (%0, %2, 4), %%mm0                \n\t" // L4

  2384.         "movq (%0, %%"REG_d"), %%mm1            \n\t" // L5

  2385.         "movq (%0, %%"REG_a", 2), %%mm2         \n\t" // L6

  2386.         "movq (%0, %%"REG_c"), %%mm3            \n\t" // L7

  2387.         "movq (%1, %2, 4), %%mm4                \n\t" // R4

  2388.         "movq (%1, %%"REG_d"), %%mm5            \n\t" // R5

  2389.         "movq (%1, %%"REG_a", 2), %%mm6         \n\t" // R6

  2390.         "movq (%1, %%"REG_c"), %%mm7            \n\t" // R7

  2391.         PAVGB(%%mm4, %%mm0)

  2392.         PAVGB(%%mm5, %%mm1)

  2393.         PAVGB(%%mm6, %%mm2)

  2394.         PAVGB(%%mm7, %%mm3)

  2395.         PAVGB(%%mm4, %%mm0)

  2396.         PAVGB(%%mm5, %%mm1)

  2397.         PAVGB(%%mm6, %%mm2)

  2398.         PAVGB(%%mm7, %%mm3)

  2399.         PAVGB(%%mm4, %%mm0)

  2400.         PAVGB(%%mm5, %%mm1)

  2401.         PAVGB(%%mm6, %%mm2)

  2402.         PAVGB(%%mm7, %%mm3)

  2403.         "movq %%mm0, (%1, %2, 4)                \n\t" // R4

  2404.         "movq %%mm1, (%1, %%"REG_d")            \n\t" // R5

  2405.         "movq %%mm2, (%1, %%"REG_a", 2)         \n\t" // R6

  2406.         "movq %%mm3, (%1, %%"REG_c")            \n\t" // R7

  2407.         "movq %%mm0, (%0, %2, 4)                \n\t" // L4

  2408.         "movq %%mm1, (%0, %%"REG_d")            \n\t" // L5

  2409.         "movq %%mm2, (%0, %%"REG_a", 2)         \n\t" // L6

  2410.         "movq %%mm3, (%0, %%"REG_c")            \n\t" // L7

  2411. 

  2412.         "4:                                     \n\t"

  2413. 

  2414.         :: "r" (src), "r" (tempBlurred), "r"((x86_reg)stride), "m" (tempBlurredPast)

  2415.         : "%"REG_a, "%"REG_d, "%"REG_c, "memory"

  2416.     );

  2417. #else //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  2418. {

  2419.     int y;

  2420.     int d=0;

  2421. //    int sysd=0;

  2422.     int i;

  2423. 

  2424.     for(y=0; y<8; y++){

  2425.         int x;

  2426.         for(x=0; x<8; x++){

  2427.             int ref= tempBlurred[ x + y*stride ];

  2428.             int cur= src[ x + y*stride ];

  2429.             int d1=ref - cur;

  2430. //            if(x==0 || x==7) d1+= d1>>1;

  2431. //            if(y==0 || y==7) d1+= d1>>1;

  2432. //            d+= FFABS(d1);

  2433.             d+= d1*d1;

  2434. //            sysd+= d1;

  2435.         }

  2436.     }

  2437.     i=d;

  2438.     d=  (

  2439.         4*d

  2440.         +(*(tempBlurredPast-256))

  2441.         +(*(tempBlurredPast-1))+ (*(tempBlurredPast+1))

  2442.         +(*(tempBlurredPast+256))

  2443.         +4)>>3;

  2444.     *tempBlurredPast=i;

  2445. //    ((*tempBlurredPast)*3 + d + 2)>>2;

  2446. 

  2447. /*

  2448. Switch between

  2449.  1  0  0  0  0  0  0  (0)

  2450. 64 32 16  8  4  2  1  (1)

  2451. 64 48 36 27 20 15 11 (33) (approx)

  2452. 64 56 49 43 37 33 29 (200) (approx)

  2453. */

  2454.     if(d > maxNoise[1]){

  2455.         if(d < maxNoise[2]){

  2456.             for(y=0; y<8; y++){

  2457.                 int x;

  2458.                 for(x=0; x<8; x++){

  2459.                     int ref= tempBlurred[ x + y*stride ];

  2460.                     int cur= src[ x + y*stride ];

  2461.                     tempBlurred[ x + y*stride ]=

  2462.                     src[ x + y*stride ]=

  2463.                         (ref + cur + 1)>>1;

  2464.                 }

  2465.             }

  2466.         }else{

  2467.             for(y=0; y<8; y++){

  2468.                 int x;

  2469.                 for(x=0; x<8; x++){

  2470.                     tempBlurred[ x + y*stride ]= src[ x + y*stride ];

  2471.                 }

  2472.             }

  2473.         }

  2474.     }else{

  2475.         if(d < maxNoise[0]){

  2476.             for(y=0; y<8; y++){

  2477.                 int x;

  2478.                 for(x=0; x<8; x++){

  2479.                     int ref= tempBlurred[ x + y*stride ];

  2480.                     int cur= src[ x + y*stride ];

  2481.                     tempBlurred[ x + y*stride ]=

  2482.                     src[ x + y*stride ]=

  2483.                         (ref*7 + cur + 4)>>3;

  2484.                 }

  2485.             }

  2486.         }else{

  2487.             for(y=0; y<8; y++){

  2488.                 int x;

  2489.                 for(x=0; x<8; x++){

  2490.                     int ref= tempBlurred[ x + y*stride ];

  2491.                     int cur= src[ x + y*stride ];

  2492.                     tempBlurred[ x + y*stride ]=

  2493.                     src[ x + y*stride ]=

  2494.                         (ref*3 + cur + 2)>>2;

  2495.                 }

  2496.             }

  2497.         }

  2498.     }

  2499. }

  2500. #endif //TEMPLATE_PP_MMXEXT || TEMPLATE_PP_3DNOW

  2501. }

  2502. #endif //TEMPLATE_PP_ALTIVEC

  2503. 

  2504. #if TEMPLATE_PP_MMX

  2505. /**

  2506.  * accurate deblock filter

  2507.  */

  2508. static av_always_inline void RENAME(do_a_deblock)(uint8_t *src, int step, int stride, PPContext *c){

  2509.     int64_t dc_mask, eq_mask, both_masks;

  2510.     int64_t sums[10*8*2];

  2511.     src+= step*3; // src points to begin of the 8x8 Block

  2512.     //{ START_TIMER

  2513.     __asm__ volatile(

  2514.         "movq %0, %%mm7                         \n\t"

  2515.         "movq %1, %%mm6                         \n\t"

  2516.         : : "m" (c->mmxDcOffset[c->nonBQP]),  "m" (c->mmxDcThreshold[c->nonBQP])

  2517.         );

  2518. 

  2519.     __asm__ volatile(

  2520.         "lea (%2, %3), %%"REG_a"                \n\t"

  2521. //      0       1       2       3       4       5       6       7       8       9

  2522. //      %1      eax     eax+%2  eax+2%2 %1+4%2  ecx     ecx+%2  ecx+2%2 %1+8%2  ecx+4%2

  2523. 

  2524.         "movq (%2), %%mm0                       \n\t"

  2525.         "movq (%%"REG_a"), %%mm1                \n\t"

  2526.         "movq %%mm1, %%mm3                      \n\t"

  2527.         "movq %%mm1, %%mm4                      \n\t"

  2528.         "psubb %%mm1, %%mm0                     \n\t" // mm0 = differnece

  2529.         "paddb %%mm7, %%mm0                     \n\t"

  2530.         "pcmpgtb %%mm6, %%mm0                   \n\t"

  2531. 

  2532.         "movq (%%"REG_a",%3), %%mm2             \n\t"

  2533.         PMAXUB(%%mm2, %%mm4)

  2534.         PMINUB(%%mm2, %%mm3, %%mm5)

  2535.         "psubb %%mm2, %%mm1                     \n\t"

  2536.         "paddb %%mm7, %%mm1                     \n\t"

  2537.         "pcmpgtb %%mm6, %%mm1                   \n\t"

  2538.         "paddb %%mm1, %%mm0                     \n\t"

  2539. 

  2540.         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"

  2541.         PMAXUB(%%mm1, %%mm4)

  2542.         PMINUB(%%mm1, %%mm3, %%mm5)

  2543.         "psubb %%mm1, %%mm2                     \n\t"

  2544.         "paddb %%mm7, %%mm2                     \n\t"

  2545.         "pcmpgtb %%mm6, %%mm2                   \n\t"

  2546.         "paddb %%mm2, %%mm0                     \n\t"

  2547. 

  2548.         "lea (%%"REG_a", %3, 4), %%"REG_a"      \n\t"

  2549. 

  2550.         "movq (%2, %3, 4), %%mm2                \n\t"

  2551.         PMAXUB(%%mm2, %%mm4)

  2552.         PMINUB(%%mm2, %%mm3, %%mm5)

  2553.         "psubb %%mm2, %%mm1                     \n\t"

  2554.         "paddb %%mm7, %%mm1                     \n\t"

  2555.         "pcmpgtb %%mm6, %%mm1                   \n\t"

  2556.         "paddb %%mm1, %%mm0                     \n\t"

  2557. 

  2558.         "movq (%%"REG_a"), %%mm1                \n\t"

  2559.         PMAXUB(%%mm1, %%mm4)

  2560.         PMINUB(%%mm1, %%mm3, %%mm5)

  2561.         "psubb %%mm1, %%mm2                     \n\t"

  2562.         "paddb %%mm7, %%mm2                     \n\t"

  2563.         "pcmpgtb %%mm6, %%mm2                   \n\t"

  2564.         "paddb %%mm2, %%mm0                     \n\t"

  2565. 

  2566.         "movq (%%"REG_a", %3), %%mm2            \n\t"

  2567.         PMAXUB(%%mm2, %%mm4)

  2568.         PMINUB(%%mm2, %%mm3, %%mm5)

  2569.         "psubb %%mm2, %%mm1                     \n\t"

  2570.         "paddb %%mm7, %%mm1                     \n\t"

  2571.         "pcmpgtb %%mm6, %%mm1                   \n\t"

  2572.         "paddb %%mm1, %%mm0                     \n\t"

  2573. 

  2574.         "movq (%%"REG_a", %3, 2), %%mm1         \n\t"

  2575.         PMAXUB(%%mm1, %%mm4)

  2576.         PMINUB(%%mm1, %%mm3, %%mm5)

  2577.         "psubb %%mm1, %%mm2                     \n\t"

  2578.         "paddb %%mm7, %%mm2                     \n\t"

  2579.         "pcmpgtb %%mm6, %%mm2                   \n\t"

  2580.         "paddb %%mm2, %%mm0                     \n\t"

  2581. 

  2582.         "movq (%2, %3, 8), %%mm2                \n\t"

  2583.         PMAXUB(%%mm2, %%mm4)

  2584.         PMINUB(%%mm2, %%mm3, %%mm5)

  2585.         "psubb %%mm2, %%mm1                     \n\t"

  2586.         "paddb %%mm7, %%mm1                     \n\t"

  2587.         "pcmpgtb %%mm6, %%mm1                   \n\t"

  2588.         "paddb %%mm1, %%mm0                     \n\t"

  2589. 

  2590.         "movq (%%"REG_a", %3, 4), %%mm1         \n\t"

  2591.         "psubb %%mm1, %%mm2                     \n\t"

  2592.         "paddb %%mm7, %%mm2                     \n\t"

  2593.         "pcmpgtb %%mm6, %%mm2                   \n\t"

  2594.         "paddb %%mm2, %%mm0                     \n\t"

  2595.         "psubusb %%mm3, %%mm4                   \n\t"

  2596. 

  2597.         "pxor %%mm6, %%mm6                      \n\t"

  2598.         "movq %4, %%mm7                         \n\t" // QP,..., QP

  2599.         "paddusb %%mm7, %%mm7                   \n\t" // 2QP ... 2QP

  2600.         "psubusb %%mm4, %%mm7                   \n\t" // Diff >=2QP -> 0

  2601.         "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0

  2602.         "pcmpeqb %%mm6, %%mm7                   \n\t" // Diff < 2QP -> 0

  2603.         "movq %%mm7, %1                         \n\t"

  2604. 

  2605.         "movq %5, %%mm7                         \n\t"

  2606.         "punpcklbw %%mm7, %%mm7                 \n\t"

  2607.         "punpcklbw %%mm7, %%mm7                 \n\t"

  2608.         "punpcklbw %%mm7, %%mm7                 \n\t"

  2609.         "psubb %%mm0, %%mm6                     \n\t"

  2610.         "pcmpgtb %%mm7, %%mm6                   \n\t"

  2611.         "movq %%mm6, %0                         \n\t"

  2612. 

  2613.         : "=m" (eq_mask), "=m" (dc_mask)

  2614.         : "r" (src), "r" ((x86_reg)step), "m" (c->pQPb), "m"(c->ppMode.flatnessThreshold)

  2615.         : "%"REG_a

  2616.     );

  2617. 

  2618.     both_masks = dc_mask & eq_mask;

  2619. 

  2620.     if(both_masks){

  2621.         x86_reg offset= -8*step;

  2622.         int64_t *temp_sums= sums;

  2623. 

  2624.         __asm__ volatile(

  2625.             "movq %2, %%mm0                         \n\t"  // QP,..., QP

  2626.             "pxor %%mm4, %%mm4                      \n\t"

  2627. 

  2628.             "movq (%0), %%mm6                       \n\t"

  2629.             "movq (%0, %1), %%mm5                   \n\t"

  2630.             "movq %%mm5, %%mm1                      \n\t"

  2631.             "movq %%mm6, %%mm2                      \n\t"

  2632.             "psubusb %%mm6, %%mm5                   \n\t"

  2633.             "psubusb %%mm1, %%mm2                   \n\t"

  2634.             "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines

  2635.             "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0

  2636.             "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF

  2637. 

  2638.             "pxor %%mm6, %%mm1                      \n\t"

  2639.             "pand %%mm0, %%mm1                      \n\t"

  2640.             "pxor %%mm1, %%mm6                      \n\t"

  2641.             // 0:QP  6:First

  2642. 

  2643.             "movq (%0, %1, 8), %%mm5                \n\t"

  2644.             "add %1, %0                             \n\t" // %0 points to line 1 not 0

  2645.             "movq (%0, %1, 8), %%mm7                \n\t"

  2646.             "movq %%mm5, %%mm1                      \n\t"

  2647.             "movq %%mm7, %%mm2                      \n\t"

  2648.             "psubusb %%mm7, %%mm5                   \n\t"

  2649.             "psubusb %%mm1, %%mm2                   \n\t"

  2650.             "por %%mm5, %%mm2                       \n\t" // ABS Diff of lines

  2651.             "movq %2, %%mm0                         \n\t"  // QP,..., QP

  2652.             "psubusb %%mm2, %%mm0                   \n\t" // diff >= QP -> 0

  2653.             "pcmpeqb %%mm4, %%mm0                   \n\t" // diff >= QP -> FF

  2654. 

  2655.             "pxor %%mm7, %%mm1                      \n\t"

  2656.             "pand %%mm0, %%mm1                      \n\t"

  2657.             "pxor %%mm1, %%mm7                      \n\t"

  2658. 

  2659.             "movq %%mm6, %%mm5                      \n\t"

  2660.             "punpckhbw %%mm4, %%mm6                 \n\t"

  2661.             "punpcklbw %%mm4, %%mm5                 \n\t"

  2662.             // 4:0 5/6:First 7:Last

  2663. 

  2664.             "movq %%mm5, %%mm0                      \n\t"

  2665.             "movq %%mm6, %%mm1                      \n\t"

  2666.             "psllw $2, %%mm0                        \n\t"

  2667.             "psllw $2, %%mm1                        \n\t"

  2668.             "paddw "MANGLE(w04)", %%mm0             \n\t"

  2669.             "paddw "MANGLE(w04)", %%mm1             \n\t"

  2670. 

  2671. #define NEXT\

  2672.             "movq (%0), %%mm2                       \n\t"\

  2673.             "movq (%0), %%mm3                       \n\t"\

  2674.             "add %1, %0                             \n\t"\

  2675.             "punpcklbw %%mm4, %%mm2                 \n\t"\

  2676.             "punpckhbw %%mm4, %%mm3                 \n\t"\

  2677.             "paddw %%mm2, %%mm0                     \n\t"\

  2678.             "paddw %%mm3, %%mm1                     \n\t"

  2679. 

  2680. #define PREV\

  2681.             "movq (%0), %%mm2                       \n\t"\

  2682.             "movq (%0), %%mm3                       \n\t"\

  2683.             "add %1, %0                             \n\t"\

  2684.             "punpcklbw %%mm4, %%mm2                 \n\t"\

  2685.             "punpckhbw %%mm4, %%mm3                 \n\t"\

  2686.             "psubw %%mm2, %%mm0                     \n\t"\

  2687.             "psubw %%mm3, %%mm1                     \n\t"

  2688. 

  2689. 

  2690.             NEXT //0

  2691.             NEXT //1

  2692.             NEXT //2

  2693.             "movq %%mm0, (%3)                       \n\t"

  2694.             "movq %%mm1, 8(%3)                      \n\t"

  2695. 

  2696.             NEXT //3

  2697.             "psubw %%mm5, %%mm0                     \n\t"

  2698.             "psubw %%mm6, %%mm1                     \n\t"

  2699.             "movq %%mm0, 16(%3)                     \n\t"

  2700.             "movq %%mm1, 24(%3)                     \n\t"

  2701. 

  2702.             NEXT //4

  2703.             "psubw %%mm5, %%mm0                     \n\t"

  2704.             "psubw %%mm6, %%mm1                     \n\t"

  2705.             "movq %%mm0, 32(%3)                     \n\t"

  2706.             "movq %%mm1, 40(%3)                     \n\t"

  2707. 

  2708.             NEXT //5

  2709.             "psubw %%mm5, %%mm0                     \n\t"

  2710.             "psubw %%mm6, %%mm1                     \n\t"

  2711.             "movq %%mm0, 48(%3)                     \n\t"

  2712.             "movq %%mm1, 56(%3)                     \n\t"

  2713. 

  2714.             NEXT //6

  2715.             "psubw %%mm5, %%mm0                     \n\t"

  2716.             "psubw %%mm6, %%mm1                     \n\t"

  2717.             "movq %%mm0, 64(%3)                     \n\t"

  2718.             "movq %%mm1, 72(%3)                     \n\t"

  2719. 

  2720.             "movq %%mm7, %%mm6                      \n\t"

  2721.             "punpckhbw %%mm4, %%mm7                 \n\t"

  2722.             "punpcklbw %%mm4, %%mm6                 \n\t"

  2723. 

  2724.             NEXT //7

  2725.             "mov %4, %0                             \n\t"

  2726.             "add %1, %0                             \n\t"

  2727.             PREV //0

  2728.             "movq %%mm0, 80(%3)                     \n\t"

  2729.             "movq %%mm1, 88(%3)                     \n\t"

  2730. 

  2731.             PREV //1

  2732.             "paddw %%mm6, %%mm0                     \n\t"

  2733.             "paddw %%mm7, %%mm1                     \n\t"

  2734.             "movq %%mm0, 96(%3)                     \n\t"

  2735.             "movq %%mm1, 104(%3)                    \n\t"

  2736. 

  2737.             PREV //2

  2738.             "paddw %%mm6, %%mm0                     \n\t"

  2739.             "paddw %%mm7, %%mm1                     \n\t"

  2740.             "movq %%mm0, 112(%3)                    \n\t"

  2741.             "movq %%mm1, 120(%3)                    \n\t"

  2742. 

  2743.             PREV //3

  2744.             "paddw %%mm6, %%mm0                     \n\t"

  2745.             "paddw %%mm7, %%mm1                     \n\t"

  2746.             "movq %%mm0, 128(%3)                    \n\t"

  2747.             "movq %%mm1, 136(%3)                    \n\t"

  2748. 

  2749.             PREV //4

  2750.             "paddw %%mm6, %%mm0                     \n\t"

  2751.             "paddw %%mm7, %%mm1                     \n\t"

  2752.             "movq %%mm0, 144(%3)                    \n\t"

  2753.             "movq %%mm1, 152(%3)                    \n\t"

  2754. 

  2755.             "mov %4, %0                             \n\t" //FIXME

  2756. 

  2757.             : "+&r"(src)

  2758.             : "r" ((x86_reg)step), "m" (c->pQPb), "r"(sums), "g"(src)

  2759.         );

  2760. 

  2761.         src+= step; // src points to begin of the 8x8 Block

  2762. 

  2763.         __asm__ volatile(

  2764.             "movq %4, %%mm6                         \n\t"

  2765.             "pcmpeqb %%mm5, %%mm5                   \n\t"

  2766.             "pxor %%mm6, %%mm5                      \n\t"

  2767.             "pxor %%mm7, %%mm7                      \n\t"

  2768. 

  2769.             "1:                                     \n\t"

  2770.             "movq (%1), %%mm0                       \n\t"

  2771.             "movq 8(%1), %%mm1                      \n\t"

  2772.             "paddw 32(%1), %%mm0                    \n\t"

  2773.             "paddw 40(%1), %%mm1                    \n\t"

  2774.             "movq (%0, %3), %%mm2                   \n\t"

  2775.             "movq %%mm2, %%mm3                      \n\t"

  2776.             "movq %%mm2, %%mm4                      \n\t"

  2777.             "punpcklbw %%mm7, %%mm2                 \n\t"

  2778.             "punpckhbw %%mm7, %%mm3                 \n\t"

  2779.             "paddw %%mm2, %%mm0                     \n\t"

  2780.             "paddw %%mm3, %%mm1                     \n\t"

  2781.             "paddw %%mm2, %%mm0                     \n\t"

  2782.             "paddw %%mm3, %%mm1                     \n\t"

  2783.             "psrlw $4, %%mm0                        \n\t"

  2784.             "psrlw $4, %%mm1                        \n\t"

  2785.             "packuswb %%mm1, %%mm0                  \n\t"

  2786.             "pand %%mm6, %%mm0                      \n\t"

  2787.             "pand %%mm5, %%mm4                      \n\t"

  2788.             "por %%mm4, %%mm0                       \n\t"

  2789.             "movq %%mm0, (%0, %3)                   \n\t"

  2790.             "add $16, %1                            \n\t"

  2791.             "add %2, %0                             \n\t"

  2792.             " js 1b                                 \n\t"

  2793. 

  2794.             : "+r"(offset), "+r"(temp_sums)

  2795.             : "r" ((x86_reg)step), "r"(src - offset), "m"(both_masks)

  2796.         );

  2797.     }else

  2798.         src+= step; // src points to begin of the 8x8 Block

  2799. 

  2800.     if(eq_mask != -1LL){

  2801.         uint8_t *temp_src= src;

  2802.         DECLARE_ALIGNED(8, uint64_t, tmp)[4]; // make space for 4 8-byte vars

  2803.         __asm__ volatile(

  2804.             "pxor %%mm7, %%mm7                      \n\t"

  2805. //      0       1       2       3       4       5       6       7       8       9

  2806. //      %0      eax     eax+%1  eax+2%1 %0+4%1  ecx     ecx+%1  ecx+2%1 %1+8%1  ecx+4%1

  2807. 

  2808.             "movq (%0), %%mm0                       \n\t"

  2809.             "movq %%mm0, %%mm1                      \n\t"

  2810.             "punpcklbw %%mm7, %%mm0                 \n\t" // low part of line 0

  2811.             "punpckhbw %%mm7, %%mm1                 \n\t" // high part of line 0

  2812. 

  2813.             "movq (%0, %1), %%mm2                   \n\t"

  2814.             "lea (%0, %1, 2), %%"REG_a"             \n\t"

  2815.             "movq %%mm2, %%mm3                      \n\t"

  2816.             "punpcklbw %%mm7, %%mm2                 \n\t" // low part of line 1

  2817.             "punpckhbw %%mm7, %%mm3                 \n\t" // high part of line 1

  2818. 

  2819.             "movq (%%"REG_a"), %%mm4                \n\t"

  2820.             "movq %%mm4, %%mm5                      \n\t"

  2821.             "punpcklbw %%mm7, %%mm4                 \n\t" // low part of line 2

  2822.             "punpckhbw %%mm7, %%mm5                 \n\t" // high part of line 2

  2823. 

  2824.             "paddw %%mm0, %%mm0                     \n\t" // 2L0

  2825.             "paddw %%mm1, %%mm1                     \n\t" // 2H0

  2826.             "psubw %%mm4, %%mm2                     \n\t" // L1 - L2

  2827.             "psubw %%mm5, %%mm3                     \n\t" // H1 - H2

  2828.             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - L1 + L2

  2829.             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - H1 + H2

  2830. 

  2831.             "psllw $2, %%mm2                        \n\t" // 4L1 - 4L2

  2832.             "psllw $2, %%mm3                        \n\t" // 4H1 - 4H2

  2833.             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2

  2834.             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2

  2835. 

  2836.             "movq (%%"REG_a", %1), %%mm2            \n\t"

  2837.             "movq %%mm2, %%mm3                      \n\t"

  2838.             "punpcklbw %%mm7, %%mm2                 \n\t" // L3

  2839.             "punpckhbw %%mm7, %%mm3                 \n\t" // H3

  2840. 

  2841.             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - L3

  2842.             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - H3

  2843.             "psubw %%mm2, %%mm0                     \n\t" // 2L0 - 5L1 + 5L2 - 2L3

  2844.             "psubw %%mm3, %%mm1                     \n\t" // 2H0 - 5H1 + 5H2 - 2H3

  2845.             "movq %%mm0, (%4)                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3

  2846.             "movq %%mm1, 8(%4)                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3

  2847. 

  2848.             "movq (%%"REG_a", %1, 2), %%mm0         \n\t"

  2849.             "movq %%mm0, %%mm1                      \n\t"

  2850.             "punpcklbw %%mm7, %%mm0                 \n\t" // L4

  2851.             "punpckhbw %%mm7, %%mm1                 \n\t" // H4

  2852. 

  2853.             "psubw %%mm0, %%mm2                     \n\t" // L3 - L4

  2854.             "psubw %%mm1, %%mm3                     \n\t" // H3 - H4

  2855.             "movq %%mm2, 16(%4)                     \n\t" // L3 - L4

  2856.             "movq %%mm3, 24(%4)                     \n\t" // H3 - H4

  2857.             "paddw %%mm4, %%mm4                     \n\t" // 2L2

  2858.             "paddw %%mm5, %%mm5                     \n\t" // 2H2

  2859.             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - L3 + L4

  2860.             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - H3 + H4

  2861. 

  2862.             "lea (%%"REG_a", %1), %0                \n\t"

  2863.             "psllw $2, %%mm2                        \n\t" // 4L3 - 4L4

  2864.             "psllw $2, %%mm3                        \n\t" // 4H3 - 4H4

  2865.             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4

  2866.             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4

  2867. //50 opcodes so far

  2868.             "movq (%0, %1, 2), %%mm2                \n\t"

  2869.             "movq %%mm2, %%mm3                      \n\t"

  2870.             "punpcklbw %%mm7, %%mm2                 \n\t" // L5

  2871.             "punpckhbw %%mm7, %%mm3                 \n\t" // H5

  2872.             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - L5

  2873.             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - H5

  2874.             "psubw %%mm2, %%mm4                     \n\t" // 2L2 - 5L3 + 5L4 - 2L5

  2875.             "psubw %%mm3, %%mm5                     \n\t" // 2H2 - 5H3 + 5H4 - 2H5

  2876. 

  2877.             "movq (%%"REG_a", %1, 4), %%mm6         \n\t"

  2878.             "punpcklbw %%mm7, %%mm6                 \n\t" // L6

  2879.             "psubw %%mm6, %%mm2                     \n\t" // L5 - L6

  2880.             "movq (%%"REG_a", %1, 4), %%mm6         \n\t"

  2881.             "punpckhbw %%mm7, %%mm6                 \n\t" // H6

  2882.             "psubw %%mm6, %%mm3                     \n\t" // H5 - H6

  2883. 

  2884.             "paddw %%mm0, %%mm0                     \n\t" // 2L4

  2885.             "paddw %%mm1, %%mm1                     \n\t" // 2H4

  2886.             "psubw %%mm2, %%mm0                     \n\t" // 2L4 - L5 + L6

  2887.             "psubw %%mm3, %%mm1                     \n\t" // 2H4 - H5 + H6

  2888. 

  2889.             "psllw $2, %%mm2                        \n\t" // 4L5 - 4L6

  2890.             "psllw $2, %%mm3                        \n\t" // 4H5 - 4H6

  2891.             "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6

  2892.             "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6

  2893. 

  2894.             "movq (%0, %1, 4), %%mm2                \n\t"

  2895.             "movq %%mm2, %%mm3                      \n\t"

  2896.             "punpcklbw %%mm7, %%mm2                 \n\t" // L7

  2897.             "punpckhbw %%mm7, %%mm3                 \n\t" // H7

  2898. 

  2899.             "paddw %%mm2, %%mm2                     \n\t" // 2L7

  2900.             "paddw %%mm3, %%mm3                     \n\t" // 2H7

  2901.             "psubw %%mm2, %%mm0                     \n\t" // 2L4 - 5L5 + 5L6 - 2L7

  2902.             "psubw %%mm3, %%mm1                     \n\t" // 2H4 - 5H5 + 5H6 - 2H7

  2903. 

  2904.             "movq (%4), %%mm2                       \n\t" // 2L0 - 5L1 + 5L2 - 2L3

  2905.             "movq 8(%4), %%mm3                      \n\t" // 2H0 - 5H1 + 5H2 - 2H3

  2906. 

  2907. #if TEMPLATE_PP_MMXEXT

  2908.             "movq %%mm7, %%mm6                      \n\t" // 0

  2909.             "psubw %%mm0, %%mm6                     \n\t"

  2910.             "pmaxsw %%mm6, %%mm0                    \n\t" // |2L4 - 5L5 + 5L6 - 2L7|

  2911.             "movq %%mm7, %%mm6                      \n\t" // 0

  2912.             "psubw %%mm1, %%mm6                     \n\t"

  2913.             "pmaxsw %%mm6, %%mm1                    \n\t" // |2H4 - 5H5 + 5H6 - 2H7|

  2914.             "movq %%mm7, %%mm6                      \n\t" // 0

  2915.             "psubw %%mm2, %%mm6                     \n\t"

  2916.             "pmaxsw %%mm6, %%mm2                    \n\t" // |2L0 - 5L1 + 5L2 - 2L3|

  2917.             "movq %%mm7, %%mm6                      \n\t" // 0

  2918.             "psubw %%mm3, %%mm6                     \n\t"

  2919.             "pmaxsw %%mm6, %%mm3                    \n\t" // |2H0 - 5H1 + 5H2 - 2H3|

  2920. #else

  2921.             "movq %%mm7, %%mm6                      \n\t" // 0

  2922.             "pcmpgtw %%mm0, %%mm6                   \n\t"

  2923.             "pxor %%mm6, %%mm0                      \n\t"

  2924.             "psubw %%mm6, %%mm0                     \n\t" // |2L4 - 5L5 + 5L6 - 2L7|

  2925.             "movq %%mm7, %%mm6                      \n\t" // 0

  2926.             "pcmpgtw %%mm1, %%mm6                   \n\t"

  2927.             "pxor %%mm6, %%mm1                      \n\t"

  2928.             "psubw %%mm6, %%mm1                     \n\t" // |2H4 - 5H5 + 5H6 - 2H7|

  2929.             "movq %%mm7, %%mm6                      \n\t" // 0

  2930.             "pcmpgtw %%mm2, %%mm6                   \n\t"

  2931.             "pxor %%mm6, %%mm2                      \n\t"

  2932.             "psubw %%mm6, %%mm2                     \n\t" // |2L0 - 5L1 + 5L2 - 2L3|

  2933.             "movq %%mm7, %%mm6                      \n\t" // 0

  2934.             "pcmpgtw %%mm3, %%mm6                   \n\t"

  2935.             "pxor %%mm6, %%mm3                      \n\t"

  2936.             "psubw %%mm6, %%mm3                     \n\t" // |2H0 - 5H1 + 5H2 - 2H3|

  2937. #endif

  2938. 

  2939. #if TEMPLATE_PP_MMXEXT

  2940.             "pminsw %%mm2, %%mm0                    \n\t"

  2941.             "pminsw %%mm3, %%mm1                    \n\t"

  2942. #else

  2943.             "movq %%mm0, %%mm6                      \n\t"

  2944.             "psubusw %%mm2, %%mm6                   \n\t"

  2945.             "psubw %%mm6, %%mm0                     \n\t"

  2946.             "movq %%mm1, %%mm6                      \n\t"

  2947.             "psubusw %%mm3, %%mm6                   \n\t"

  2948.             "psubw %%mm6, %%mm1                     \n\t"

  2949. #endif

  2950. 

  2951.             "movd %2, %%mm2                         \n\t" // QP

  2952.             "punpcklbw %%mm7, %%mm2                 \n\t"

  2953. 

  2954.             "movq %%mm7, %%mm6                      \n\t" // 0

  2955.             "pcmpgtw %%mm4, %%mm6                   \n\t" // sign(2L2 - 5L3 + 5L4 - 2L5)

  2956.             "pxor %%mm6, %%mm4                      \n\t"

  2957.             "psubw %%mm6, %%mm4                     \n\t" // |2L2 - 5L3 + 5L4 - 2L5|

  2958.             "pcmpgtw %%mm5, %%mm7                   \n\t" // sign(2H2 - 5H3 + 5H4 - 2H5)

  2959.             "pxor %%mm7, %%mm5                      \n\t"

  2960.             "psubw %%mm7, %%mm5                     \n\t" // |2H2 - 5H3 + 5H4 - 2H5|

  2961. // 100 opcodes

  2962.             "psllw $3, %%mm2                        \n\t" // 8QP

  2963.             "movq %%mm2, %%mm3                      \n\t" // 8QP

  2964.             "pcmpgtw %%mm4, %%mm2                   \n\t"

  2965.             "pcmpgtw %%mm5, %%mm3                   \n\t"

  2966.             "pand %%mm2, %%mm4                      \n\t"

  2967.             "pand %%mm3, %%mm5                      \n\t"

  2968. 

  2969. 

  2970.             "psubusw %%mm0, %%mm4                   \n\t" // hd

  2971.             "psubusw %%mm1, %%mm5                   \n\t" // ld

  2972. 

  2973. 

  2974.             "movq "MANGLE(w05)", %%mm2              \n\t" // 5

  2975.             "pmullw %%mm2, %%mm4                    \n\t"

  2976.             "pmullw %%mm2, %%mm5                    \n\t"

  2977.             "movq "MANGLE(w20)", %%mm2              \n\t" // 32

  2978.             "paddw %%mm2, %%mm4                     \n\t"

  2979.             "paddw %%mm2, %%mm5                     \n\t"

  2980.             "psrlw $6, %%mm4                        \n\t"

  2981.             "psrlw $6, %%mm5                        \n\t"

  2982. 

  2983.             "movq 16(%4), %%mm0                     \n\t" // L3 - L4

  2984.             "movq 24(%4), %%mm1                     \n\t" // H3 - H4

  2985. 

  2986.             "pxor %%mm2, %%mm2                      \n\t"

  2987.             "pxor %%mm3, %%mm3                      \n\t"

  2988. 

  2989.             "pcmpgtw %%mm0, %%mm2                   \n\t" // sign (L3-L4)

  2990.             "pcmpgtw %%mm1, %%mm3                   \n\t" // sign (H3-H4)

  2991.             "pxor %%mm2, %%mm0                      \n\t"

  2992.             "pxor %%mm3, %%mm1                      \n\t"

  2993.             "psubw %%mm2, %%mm0                     \n\t" // |L3-L4|

  2994.             "psubw %%mm3, %%mm1                     \n\t" // |H3-H4|

  2995.             "psrlw $1, %%mm0                        \n\t" // |L3 - L4|/2

  2996.             "psrlw $1, %%mm1                        \n\t" // |H3 - H4|/2

  2997. 

  2998.             "pxor %%mm6, %%mm2                      \n\t"

  2999.             "pxor %%mm7, %%mm3                      \n\t"

  3000.             "pand %%mm2, %%mm4                      \n\t"

  3001.             "pand %%mm3, %%mm5                      \n\t"

  3002. 

  3003. #if TEMPLATE_PP_MMXEXT

  3004.             "pminsw %%mm0, %%mm4                    \n\t"

  3005.             "pminsw %%mm1, %%mm5                    \n\t"

  3006. #else

  3007.             "movq %%mm4, %%mm2                      \n\t"

  3008.             "psubusw %%mm0, %%mm2                   \n\t"

  3009.             "psubw %%mm2, %%mm4                     \n\t"

  3010.             "movq %%mm5, %%mm2                      \n\t"

  3011.             "psubusw %%mm1, %%mm2                   \n\t"

  3012.             "psubw %%mm2, %%mm5                     \n\t"

  3013. #endif

  3014.             "pxor %%mm6, %%mm4                      \n\t"

  3015.             "pxor %%mm7, %%mm5                      \n\t"

  3016.             "psubw %%mm6, %%mm4                     \n\t"

  3017.             "psubw %%mm7, %%mm5                     \n\t"

  3018.             "packsswb %%mm5, %%mm4                  \n\t"

  3019.             "movq %3, %%mm1                         \n\t"

  3020.             "pandn %%mm4, %%mm1                     \n\t"

  3021.             "movq (%0), %%mm0                       \n\t"

  3022.             "paddb   %%mm1, %%mm0                   \n\t"

  3023.             "movq %%mm0, (%0)                       \n\t"

  3024.             "movq (%0, %1), %%mm0                   \n\t"

  3025.             "psubb %%mm1, %%mm0                     \n\t"

  3026.             "movq %%mm0, (%0, %1)                   \n\t"

  3027. 

  3028.             : "+r" (temp_src)

  3029.             : "r" ((x86_reg)step), "m" (c->pQPb), "m"(eq_mask), "r"(tmp)

  3030.             : "%"REG_a

  3031.         );

  3032.     }

  3033. /*if(step==16){

  3034.     STOP_TIMER("step16")

  3035. }else{

  3036.     STOP_TIMER("stepX")

  3037. }

  3038.     } */

  3039. }

  3040. #endif //TEMPLATE_PP_MMX

  3041. 

  3042. static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,

  3043.                                 const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c);

  3044. 

  3045. /**

  3046.  * Copy a block from src to dst and fixes the blacklevel.

  3047.  * levelFix == 0 -> do not touch the brighness & contrast

  3048.  */

  3049. #undef REAL_SCALED_CPY

  3050. #undef SCALED_CPY

  3051. 

  3052. static inline void RENAME(blockCopy)(uint8_t dst[], int dstStride, const uint8_t src[], int srcStride,

  3053.                                      int levelFix, int64_t *packedOffsetAndScale)

  3054. {

  3055. #if !TEMPLATE_PP_MMX

  3056.     int i;

  3057. #endif

  3058.     if(levelFix){

  3059. #if TEMPLATE_PP_MMX

  3060.     __asm__ volatile(

  3061.         "movq (%%"REG_a"), %%mm2        \n\t" // packedYOffset

  3062.         "movq 8(%%"REG_a"), %%mm3       \n\t" // packedYScale

  3063.         "lea (%2,%4), %%"REG_a"         \n\t"

  3064.         "lea (%3,%5), %%"REG_d"         \n\t"

  3065.         "pxor %%mm4, %%mm4              \n\t"

  3066. #if TEMPLATE_PP_MMXEXT

  3067. #define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                                \

  3068.         "movq " #src1 ", %%mm0          \n\t"\

  3069.         "movq " #src1 ", %%mm5          \n\t"\

  3070.         "movq " #src2 ", %%mm1          \n\t"\

  3071.         "movq " #src2 ", %%mm6          \n\t"\

  3072.         "punpcklbw %%mm0, %%mm0         \n\t"\

  3073.         "punpckhbw %%mm5, %%mm5         \n\t"\

  3074.         "punpcklbw %%mm1, %%mm1         \n\t"\

  3075.         "punpckhbw %%mm6, %%mm6         \n\t"\

  3076.         "pmulhuw %%mm3, %%mm0           \n\t"\

  3077.         "pmulhuw %%mm3, %%mm5           \n\t"\

  3078.         "pmulhuw %%mm3, %%mm1           \n\t"\

  3079.         "pmulhuw %%mm3, %%mm6           \n\t"\

  3080.         "psubw %%mm2, %%mm0             \n\t"\

  3081.         "psubw %%mm2, %%mm5             \n\t"\

  3082.         "psubw %%mm2, %%mm1             \n\t"\

  3083.         "psubw %%mm2, %%mm6             \n\t"\

  3084.         "packuswb %%mm5, %%mm0          \n\t"\

  3085.         "packuswb %%mm6, %%mm1          \n\t"\

  3086.         "movq %%mm0, " #dst1 "          \n\t"\

  3087.         "movq %%mm1, " #dst2 "          \n\t"\

  3088. 

  3089. #else //TEMPLATE_PP_MMXEXT

  3090. #define REAL_SCALED_CPY(src1, src2, dst1, dst2)                                        \

  3091.         "movq " #src1 ", %%mm0          \n\t"\

  3092.         "movq " #src1 ", %%mm5          \n\t"\

  3093.         "punpcklbw %%mm4, %%mm0         \n\t"\

  3094.         "punpckhbw %%mm4, %%mm5         \n\t"\

  3095.         "psubw %%mm2, %%mm0             \n\t"\

  3096.         "psubw %%mm2, %%mm5             \n\t"\

  3097.         "movq " #src2 ", %%mm1          \n\t"\

  3098.         "psllw $6, %%mm0                \n\t"\

  3099.         "psllw $6, %%mm5                \n\t"\

  3100.         "pmulhw %%mm3, %%mm0            \n\t"\

  3101.         "movq " #src2 ", %%mm6          \n\t"\

  3102.         "pmulhw %%mm3, %%mm5            \n\t"\

  3103.         "punpcklbw %%mm4, %%mm1         \n\t"\

  3104.         "punpckhbw %%mm4, %%mm6         \n\t"\

  3105.         "psubw %%mm2, %%mm1             \n\t"\

  3106.         "psubw %%mm2, %%mm6             \n\t"\

  3107.         "psllw $6, %%mm1                \n\t"\

  3108.         "psllw $6, %%mm6                \n\t"\

  3109.         "pmulhw %%mm3, %%mm1            \n\t"\

  3110.         "pmulhw %%mm3, %%mm6            \n\t"\

  3111.         "packuswb %%mm5, %%mm0          \n\t"\

  3112.         "packuswb %%mm6, %%mm1          \n\t"\

  3113.         "movq %%mm0, " #dst1 "          \n\t"\

  3114.         "movq %%mm1, " #dst2 "          \n\t"\

  3115. 

  3116. #endif //TEMPLATE_PP_MMXEXT

  3117. #define SCALED_CPY(src1, src2, dst1, dst2)\

  3118.    REAL_SCALED_CPY(src1, src2, dst1, dst2)

  3119. 

  3120. SCALED_CPY((%2)       , (%2, %4)      , (%3)       , (%3, %5))

  3121. SCALED_CPY((%2, %4, 2), (%%REGa, %4, 2), (%3, %5, 2), (%%REGd, %5, 2))

  3122. SCALED_CPY((%2, %4, 4), (%%REGa, %4, 4), (%3, %5, 4), (%%REGd, %5, 4))

  3123.         "lea (%%"REG_a",%4,4), %%"REG_a"        \n\t"

  3124.         "lea (%%"REG_d",%5,4), %%"REG_d"        \n\t"

  3125. SCALED_CPY((%%REGa, %4), (%%REGa, %4, 2), (%%REGd, %5), (%%REGd, %5, 2))

  3126. 

  3127. 

  3128.         : "=&a" (packedOffsetAndScale)

  3129.         : "0" (packedOffsetAndScale),

  3130.         "r"(src),

  3131.         "r"(dst),

  3132.         "r" ((x86_reg)srcStride),

  3133.         "r" ((x86_reg)dstStride)

  3134.         : "%"REG_d

  3135.     );

  3136. #else //TEMPLATE_PP_MMX

  3137.     for(i=0; i<8; i++)

  3138.         memcpy( &(dst[dstStride*i]),

  3139.                 &(src[srcStride*i]), BLOCK_SIZE);

  3140. #endif //TEMPLATE_PP_MMX

  3141.     }else{

  3142. #if TEMPLATE_PP_MMX

  3143.     __asm__ volatile(

  3144.         "lea (%0,%2), %%"REG_a"                 \n\t"

  3145.         "lea (%1,%3), %%"REG_d"                 \n\t"

  3146. 

  3147. #define REAL_SIMPLE_CPY(src1, src2, dst1, dst2)                              \

  3148.         "movq " #src1 ", %%mm0          \n\t"\

  3149.         "movq " #src2 ", %%mm1          \n\t"\

  3150.         "movq %%mm0, " #dst1 "          \n\t"\

  3151.         "movq %%mm1, " #dst2 "          \n\t"\

  3152. 

  3153. #define SIMPLE_CPY(src1, src2, dst1, dst2)\

  3154.    REAL_SIMPLE_CPY(src1, src2, dst1, dst2)

  3155. 

  3156. SIMPLE_CPY((%0)       , (%0, %2)       , (%1)       , (%1, %3))

  3157. SIMPLE_CPY((%0, %2, 2), (%%REGa, %2, 2), (%1, %3, 2), (%%REGd, %3, 2))

  3158. SIMPLE_CPY((%0, %2, 4), (%%REGa, %2, 4), (%1, %3, 4), (%%REGd, %3, 4))

  3159.         "lea (%%"REG_a",%2,4), %%"REG_a"        \n\t"

  3160.         "lea (%%"REG_d",%3,4), %%"REG_d"        \n\t"

  3161. SIMPLE_CPY((%%REGa, %2), (%%REGa, %2, 2), (%%REGd, %3), (%%REGd, %3, 2))

  3162. 

  3163.         : : "r" (src),

  3164.         "r" (dst),

  3165.         "r" ((x86_reg)srcStride),

  3166.         "r" ((x86_reg)dstStride)

  3167.         : "%"REG_a, "%"REG_d

  3168.     );

  3169. #else //TEMPLATE_PP_MMX

  3170.     for(i=0; i<8; i++)

  3171.         memcpy( &(dst[dstStride*i]),

  3172.                 &(src[srcStride*i]), BLOCK_SIZE);

  3173. #endif //TEMPLATE_PP_MMX

  3174.     }

  3175. }

  3176. 

  3177. /**

  3178.  * Duplicate the given 8 src pixels ? times upward

  3179.  */

  3180. static inline void RENAME(duplicate)(uint8_t src[], int stride)

  3181. {

  3182. #if TEMPLATE_PP_MMX

  3183.     __asm__ volatile(

  3184.         "movq (%0), %%mm0               \n\t"

  3185.         "add %1, %0                     \n\t"

  3186.         "movq %%mm0, (%0)               \n\t"

  3187.         "movq %%mm0, (%0, %1)           \n\t"

  3188.         "movq %%mm0, (%0, %1, 2)        \n\t"

  3189.         : "+r" (src)

  3190.         : "r" ((x86_reg)-stride)

  3191.     );

  3192. #else

  3193.     int i;

  3194.     uint8_t *p=src;

  3195.     for(i=0; i<3; i++){

  3196.         p-= stride;

  3197.         memcpy(p, src, 8);

  3198.     }

  3199. #endif

  3200. }

  3201. 

  3202. /**

  3203.  * Filter array of bytes (Y or U or V values)

  3204.  */

  3205. static void RENAME(postProcess)(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,

  3206.                                 const QP_STORE_T QPs[], int QPStride, int isColor, PPContext *c2)

  3207. {

  3208.     DECLARE_ALIGNED(8, PPContext, c)= *c2; //copy to stack for faster access

  3209.     int x,y;

  3210. #ifdef TEMPLATE_PP_TIME_MODE

  3211.     const int mode= TEMPLATE_PP_TIME_MODE;

  3212. #else

  3213.     const int mode= isColor ? c.ppMode.chromMode : c.ppMode.lumMode;

  3214. #endif

  3215.     int black=0, white=255; // blackest black and whitest white in the picture

  3216.     int QPCorrecture= 256*256;

  3217. 

  3218.     int copyAhead;

  3219. #if TEMPLATE_PP_MMX

  3220.     int i;

  3221. #endif

  3222. 

  3223.     const int qpHShift= isColor ? 4-c.hChromaSubSample : 4;

  3224.     const int qpVShift= isColor ? 4-c.vChromaSubSample : 4;

  3225. 

  3226.     //FIXME remove

  3227.     uint64_t * const yHistogram= c.yHistogram;

  3228.     uint8_t * const tempSrc= srcStride > 0 ? c.tempSrc : c.tempSrc - 23*srcStride;

  3229.     uint8_t * const tempDst= dstStride > 0 ? c.tempDst : c.tempDst - 23*dstStride;

  3230.     //const int mbWidth= isColor ? (width+7)>>3 : (width+15)>>4;

  3231. 

  3232. #if TEMPLATE_PP_MMX

  3233.     for(i=0; i<57; i++){

  3234.         int offset= ((i*c.ppMode.baseDcDiff)>>8) + 1;

  3235.         int threshold= offset*2 + 1;

  3236.         c.mmxDcOffset[i]= 0x7F - offset;

  3237.         c.mmxDcThreshold[i]= 0x7F - threshold;

  3238.         c.mmxDcOffset[i]*= 0x0101010101010101LL;

  3239.         c.mmxDcThreshold[i]*= 0x0101010101010101LL;

  3240.     }

  3241. #endif

  3242. 

  3243.     if(mode & CUBIC_IPOL_DEINT_FILTER) copyAhead=16;

  3244.     else if(   (mode & LINEAR_BLEND_DEINT_FILTER)

  3245.             || (mode & FFMPEG_DEINT_FILTER)

  3246.             || (mode & LOWPASS5_DEINT_FILTER)) copyAhead=14;

  3247.     else if(   (mode & V_DEBLOCK)

  3248.             || (mode & LINEAR_IPOL_DEINT_FILTER)

  3249.             || (mode & MEDIAN_DEINT_FILTER)

  3250.             || (mode & V_A_DEBLOCK)) copyAhead=13;

  3251.     else if(mode & V_X1_FILTER) copyAhead=11;

  3252. //    else if(mode & V_RK1_FILTER) copyAhead=10;

  3253.     else if(mode & DERING) copyAhead=9;

  3254.     else copyAhead=8;

  3255. 

  3256.     copyAhead-= 8;

  3257. 

  3258.     if(!isColor){

  3259.         uint64_t sum= 0;

  3260.         int i;

  3261.         uint64_t maxClipped;

  3262.         uint64_t clipped;

  3263.         double scale;

  3264. 

  3265.         c.frameNum++;

  3266.         // first frame is fscked so we ignore it

  3267.         if(c.frameNum == 1) yHistogram[0]= width*(uint64_t)height/64*15/256;

  3268. 

  3269.         for(i=0; i<256; i++){

  3270.             sum+= yHistogram[i];

  3271.         }

  3272. 

  3273.         /* We always get a completely black picture first. */

  3274.         maxClipped= (uint64_t)(sum * c.ppMode.maxClippedThreshold);

  3275. 

  3276.         clipped= sum;

  3277.         for(black=255; black>0; black--){

  3278.             if(clipped < maxClipped) break;

  3279.             clipped-= yHistogram[black];

  3280.         }

  3281. 

  3282.         clipped= sum;

  3283.         for(white=0; white<256; white++){

  3284.             if(clipped < maxClipped) break;

  3285.             clipped-= yHistogram[white];

  3286.         }

  3287. 

  3288.         scale= (double)(c.ppMode.maxAllowedY - c.ppMode.minAllowedY) / (double)(white-black);

  3289. 

  3290. #if TEMPLATE_PP_MMXEXT

  3291.         c.packedYScale= (uint16_t)(scale*256.0 + 0.5);

  3292.         c.packedYOffset= (((black*c.packedYScale)>>8) - c.ppMode.minAllowedY) & 0xFFFF;

  3293. #else

  3294.         c.packedYScale= (uint16_t)(scale*1024.0 + 0.5);

  3295.         c.packedYOffset= (black - c.ppMode.minAllowedY) & 0xFFFF;

  3296. #endif

  3297. 

  3298.         c.packedYOffset|= c.packedYOffset<<32;

  3299.         c.packedYOffset|= c.packedYOffset<<16;

  3300. 

  3301.         c.packedYScale|= c.packedYScale<<32;

  3302.         c.packedYScale|= c.packedYScale<<16;

  3303. 

  3304.         if(mode & LEVEL_FIX)        QPCorrecture= (int)(scale*256*256 + 0.5);

  3305.         else                        QPCorrecture= 256*256;

  3306.     }else{

  3307.         c.packedYScale= 0x0100010001000100LL;

  3308.         c.packedYOffset= 0;

  3309.         QPCorrecture= 256*256;

  3310.     }

  3311. 

  3312.     /* copy & deinterlace first row of blocks */

  3313.     y=-BLOCK_SIZE;

  3314.     {

  3315.         const uint8_t *srcBlock= &(src[y*srcStride]);

  3316.         uint8_t *dstBlock= tempDst + dstStride;

  3317. 

  3318.         // From this point on it is guaranteed that we can read and write 16 lines downward

  3319.         // finish 1 block before the next otherwise we might have a problem

  3320.         // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing

  3321.         for(x=0; x<width; x+=BLOCK_SIZE){

  3322. 

  3323. #if TEMPLATE_PP_MMXEXT

  3324. /*

  3325.             prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);

  3326.             prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);

  3327.             prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);

  3328.             prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);

  3329. */

  3330. 

  3331.             __asm__(

  3332.                 "mov %4, %%"REG_a"              \n\t"

  3333.                 "shr $2, %%"REG_a"              \n\t"

  3334.                 "and $6, %%"REG_a"              \n\t"

  3335.                 "add %5, %%"REG_a"              \n\t"

  3336.                 "mov %%"REG_a", %%"REG_d"       \n\t"

  3337.                 "imul %1, %%"REG_a"             \n\t"

  3338.                 "imul %3, %%"REG_d"             \n\t"

  3339.                 "prefetchnta 32(%%"REG_a", %0)  \n\t"

  3340.                 "prefetcht0 32(%%"REG_d", %2)   \n\t"

  3341.                 "add %1, %%"REG_a"              \n\t"

  3342.                 "add %3, %%"REG_d"              \n\t"

  3343.                 "prefetchnta 32(%%"REG_a", %0)  \n\t"

  3344.                 "prefetcht0 32(%%"REG_d", %2)   \n\t"

  3345.                 :: "r" (srcBlock), "r" ((x86_reg)srcStride), "r" (dstBlock), "r" ((x86_reg)dstStride),

  3346.                 "g" ((x86_reg)x), "g" ((x86_reg)copyAhead)

  3347.                 : "%"REG_a, "%"REG_d

  3348.             );

  3349. 

  3350. #elif TEMPLATE_PP_3DNOW

  3351. //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...

  3352. /*          prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);

  3353.             prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);

  3354.             prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);

  3355.             prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);

  3356. */

  3357. #endif

  3358. 

  3359.             RENAME(blockCopy)(dstBlock + dstStride*8, dstStride,

  3360.                               srcBlock + srcStride*8, srcStride, mode & LEVEL_FIX, &c.packedYOffset);

  3361. 

  3362.             RENAME(duplicate)(dstBlock + dstStride*8, dstStride);

  3363. 

  3364.             if(mode & LINEAR_IPOL_DEINT_FILTER)

  3365.                 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);

  3366.             else if(mode & LINEAR_BLEND_DEINT_FILTER)

  3367.                 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);

  3368.             else if(mode & MEDIAN_DEINT_FILTER)

  3369.                 RENAME(deInterlaceMedian)(dstBlock, dstStride);

  3370.             else if(mode & CUBIC_IPOL_DEINT_FILTER)

  3371.                 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);

  3372.             else if(mode & FFMPEG_DEINT_FILTER)

  3373.                 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);

  3374.             else if(mode & LOWPASS5_DEINT_FILTER)

  3375.                 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);

  3376. /*          else if(mode & CUBIC_BLEND_DEINT_FILTER)

  3377.                 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);

  3378. */

  3379.             dstBlock+=8;

  3380.             srcBlock+=8;

  3381.         }

  3382.         if(width==FFABS(dstStride))

  3383.             linecpy(dst, tempDst + 9*dstStride, copyAhead, dstStride);

  3384.         else{

  3385.             int i;

  3386.             for(i=0; i<copyAhead; i++){

  3387.                 memcpy(dst + i*dstStride, tempDst + (9+i)*dstStride, width);

  3388.             }

  3389.         }

  3390.     }

  3391. 

  3392.     for(y=0; y<height; y+=BLOCK_SIZE){

  3393.         //1% speedup if these are here instead of the inner loop

  3394.         const uint8_t *srcBlock= &(src[y*srcStride]);

  3395.         uint8_t *dstBlock= &(dst[y*dstStride]);

  3396. #if TEMPLATE_PP_MMX

  3397.         uint8_t *tempBlock1= c.tempBlocks;

  3398.         uint8_t *tempBlock2= c.tempBlocks + 8;

  3399. #endif

  3400.         const int8_t *QPptr= &QPs[(y>>qpVShift)*QPStride];

  3401.         int8_t *nonBQPptr= &c.nonBQPTable[(y>>qpVShift)*FFABS(QPStride)];

  3402.         int QP=0;

  3403.         /* can we mess with a 8x16 block from srcBlock/dstBlock downwards and 1 line upwards

  3404.            if not than use a temporary buffer */

  3405.         if(y+15 >= height){

  3406.             int i;

  3407.             /* copy from line (copyAhead) to (copyAhead+7) of src, these will be copied with

  3408.                blockcopy to dst later */

  3409.             linecpy(tempSrc + srcStride*copyAhead, srcBlock + srcStride*copyAhead,

  3410.                     FFMAX(height-y-copyAhead, 0), srcStride);

  3411. 

  3412.             /* duplicate last line of src to fill the void up to line (copyAhead+7) */

  3413.             for(i=FFMAX(height-y, 8); i<copyAhead+8; i++)

  3414.                     memcpy(tempSrc + srcStride*i, src + srcStride*(height-1), FFABS(srcStride));

  3415. 

  3416.             /* copy up to (copyAhead+1) lines of dst (line -1 to (copyAhead-1))*/

  3417.             linecpy(tempDst, dstBlock - dstStride, FFMIN(height-y+1, copyAhead+1), dstStride);

  3418. 

  3419.             /* duplicate last line of dst to fill the void up to line (copyAhead) */

  3420.             for(i=height-y+1; i<=copyAhead; i++)

  3421.                     memcpy(tempDst + dstStride*i, dst + dstStride*(height-1), FFABS(dstStride));

  3422. 

  3423.             dstBlock= tempDst + dstStride;

  3424.             srcBlock= tempSrc;

  3425.         }

  3426. 

  3427.         // From this point on it is guaranteed that we can read and write 16 lines downward

  3428.         // finish 1 block before the next otherwise we might have a problem

  3429.         // with the L1 Cache of the P4 ... or only a few blocks at a time or soemthing

  3430.         for(x=0; x<width; x+=BLOCK_SIZE){

  3431.             const int stride= dstStride;

  3432. #if TEMPLATE_PP_MMX

  3433.             uint8_t *tmpXchg;

  3434. #endif

  3435.             if(isColor){

  3436.                 QP= QPptr[x>>qpHShift];

  3437.                 c.nonBQP= nonBQPptr[x>>qpHShift];

  3438.             }else{

  3439.                 QP= QPptr[x>>4];

  3440.                 QP= (QP* QPCorrecture + 256*128)>>16;

  3441.                 c.nonBQP= nonBQPptr[x>>4];

  3442.                 c.nonBQP= (c.nonBQP* QPCorrecture + 256*128)>>16;

  3443.                 yHistogram[ srcBlock[srcStride*12 + 4] ]++;

  3444.             }

  3445.             c.QP= QP;

  3446. #if TEMPLATE_PP_MMX

  3447.             __asm__ volatile(

  3448.                 "movd %1, %%mm7         \n\t"

  3449.                 "packuswb %%mm7, %%mm7  \n\t" // 0, 0, 0, QP, 0, 0, 0, QP

  3450.                 "packuswb %%mm7, %%mm7  \n\t" // 0,QP, 0, QP, 0,QP, 0, QP

  3451.                 "packuswb %%mm7, %%mm7  \n\t" // QP,..., QP

  3452.                 "movq %%mm7, %0         \n\t"

  3453.                 : "=m" (c.pQPb)

  3454.                 : "r" (QP)

  3455.             );

  3456. #endif

  3457. 

  3458. 

  3459. #if TEMPLATE_PP_MMXEXT

  3460. /*

  3461.             prefetchnta(srcBlock + (((x>>2)&6) + 5)*srcStride + 32);

  3462.             prefetchnta(srcBlock + (((x>>2)&6) + 6)*srcStride + 32);

  3463.             prefetcht0(dstBlock + (((x>>2)&6) + 5)*dstStride + 32);

  3464.             prefetcht0(dstBlock + (((x>>2)&6) + 6)*dstStride + 32);

  3465. */

  3466. 

  3467.             __asm__(

  3468.                 "mov %4, %%"REG_a"              \n\t"

  3469.                 "shr $2, %%"REG_a"              \n\t"

  3470.                 "and $6, %%"REG_a"              \n\t"

  3471.                 "add %5, %%"REG_a"              \n\t"

  3472.                 "mov %%"REG_a", %%"REG_d"       \n\t"

  3473.                 "imul %1, %%"REG_a"             \n\t"

  3474.                 "imul %3, %%"REG_d"             \n\t"

  3475.                 "prefetchnta 32(%%"REG_a", %0)  \n\t"

  3476.                 "prefetcht0 32(%%"REG_d", %2)   \n\t"

  3477.                 "add %1, %%"REG_a"              \n\t"

  3478.                 "add %3, %%"REG_d"              \n\t"

  3479.                 "prefetchnta 32(%%"REG_a", %0)  \n\t"

  3480.                 "prefetcht0 32(%%"REG_d", %2)   \n\t"

  3481.                 :: "r" (srcBlock), "r" ((x86_reg)srcStride), "r" (dstBlock), "r" ((x86_reg)dstStride),

  3482.                 "g" ((x86_reg)x), "g" ((x86_reg)copyAhead)

  3483.                 : "%"REG_a, "%"REG_d

  3484.             );

  3485. 

  3486. #elif TEMPLATE_PP_3DNOW

  3487. //FIXME check if this is faster on an 3dnow chip or if it is faster without the prefetch or ...

  3488. /*          prefetch(srcBlock + (((x>>3)&3) + 5)*srcStride + 32);

  3489.             prefetch(srcBlock + (((x>>3)&3) + 9)*srcStride + 32);

  3490.             prefetchw(dstBlock + (((x>>3)&3) + 5)*dstStride + 32);

  3491.             prefetchw(dstBlock + (((x>>3)&3) + 9)*dstStride + 32);

  3492. */

  3493. #endif

  3494. 

  3495.             RENAME(blockCopy)(dstBlock + dstStride*copyAhead, dstStride,

  3496.                               srcBlock + srcStride*copyAhead, srcStride, mode & LEVEL_FIX, &c.packedYOffset);

  3497. 

  3498.             if(mode & LINEAR_IPOL_DEINT_FILTER)

  3499.                 RENAME(deInterlaceInterpolateLinear)(dstBlock, dstStride);

  3500.             else if(mode & LINEAR_BLEND_DEINT_FILTER)

  3501.                 RENAME(deInterlaceBlendLinear)(dstBlock, dstStride, c.deintTemp + x);

  3502.             else if(mode & MEDIAN_DEINT_FILTER)

  3503.                 RENAME(deInterlaceMedian)(dstBlock, dstStride);

  3504.             else if(mode & CUBIC_IPOL_DEINT_FILTER)

  3505.                 RENAME(deInterlaceInterpolateCubic)(dstBlock, dstStride);

  3506.             else if(mode & FFMPEG_DEINT_FILTER)

  3507.                 RENAME(deInterlaceFF)(dstBlock, dstStride, c.deintTemp + x);

  3508.             else if(mode & LOWPASS5_DEINT_FILTER)

  3509.                 RENAME(deInterlaceL5)(dstBlock, dstStride, c.deintTemp + x, c.deintTemp + width + x);

  3510. /*          else if(mode & CUBIC_BLEND_DEINT_FILTER)

  3511.                 RENAME(deInterlaceBlendCubic)(dstBlock, dstStride);

  3512. */

  3513. 

  3514.             /* only deblock if we have 2 blocks */

  3515.             if(y + 8 < height){

  3516.                 if(mode & V_X1_FILTER)

  3517.                     RENAME(vertX1Filter)(dstBlock, stride, &c);

  3518.                 else if(mode & V_DEBLOCK){

  3519.                     const int t= RENAME(vertClassify)(dstBlock, stride, &c);

  3520. 

  3521.                     if(t==1)

  3522.                         RENAME(doVertLowPass)(dstBlock, stride, &c);

  3523.                     else if(t==2)

  3524.                         RENAME(doVertDefFilter)(dstBlock, stride, &c);

  3525.                 }else if(mode & V_A_DEBLOCK){

  3526.                     RENAME(do_a_deblock)(dstBlock, stride, 1, &c);

  3527.                 }

  3528.             }

  3529. 

  3530. #if TEMPLATE_PP_MMX

  3531.             RENAME(transpose1)(tempBlock1, tempBlock2, dstBlock, dstStride);

  3532. #endif

  3533.             /* check if we have a previous block to deblock it with dstBlock */

  3534.             if(x - 8 >= 0){

  3535. #if TEMPLATE_PP_MMX

  3536.                 if(mode & H_X1_FILTER)

  3537.                         RENAME(vertX1Filter)(tempBlock1, 16, &c);

  3538.                 else if(mode & H_DEBLOCK){

  3539. //START_TIMER

  3540.                     const int t= RENAME(vertClassify)(tempBlock1, 16, &c);

  3541. //STOP_TIMER("dc & minmax")

  3542.                     if(t==1)

  3543.                         RENAME(doVertLowPass)(tempBlock1, 16, &c);

  3544.                     else if(t==2)

  3545.                         RENAME(doVertDefFilter)(tempBlock1, 16, &c);

  3546.                 }else if(mode & H_A_DEBLOCK){

  3547.                         RENAME(do_a_deblock)(tempBlock1, 16, 1, &c);

  3548.                 }

  3549. 

  3550.                 RENAME(transpose2)(dstBlock-4, dstStride, tempBlock1 + 4*16);

  3551. 

  3552. #else

  3553.                 if(mode & H_X1_FILTER)

  3554.                     horizX1Filter(dstBlock-4, stride, QP);

  3555.                 else if(mode & H_DEBLOCK){

  3556. #if TEMPLATE_PP_ALTIVEC

  3557.                     DECLARE_ALIGNED(16, unsigned char, tempBlock)[272];

  3558.                     int t;

  3559.                     transpose_16x8_char_toPackedAlign_altivec(tempBlock, dstBlock - (4 + 1), stride);

  3560. 

  3561.                     t = vertClassify_altivec(tempBlock-48, 16, &c);

  3562.                     if(t==1) {

  3563.                         doVertLowPass_altivec(tempBlock-48, 16, &c);

  3564.                         transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);

  3565.                     }

  3566.                     else if(t==2) {

  3567.                         doVertDefFilter_altivec(tempBlock-48, 16, &c);

  3568.                         transpose_8x16_char_fromPackedAlign_altivec(dstBlock - (4 + 1), tempBlock, stride);

  3569.                     }

  3570. #else

  3571.                     const int t= RENAME(horizClassify)(dstBlock-4, stride, &c);

  3572. 

  3573.                     if(t==1)

  3574.                         RENAME(doHorizLowPass)(dstBlock-4, stride, &c);

  3575.                     else if(t==2)

  3576.                         RENAME(doHorizDefFilter)(dstBlock-4, stride, &c);

  3577. #endif

  3578.                 }else if(mode & H_A_DEBLOCK){

  3579.                     RENAME(do_a_deblock)(dstBlock-8, 1, stride, &c);

  3580.                 }

  3581. #endif //TEMPLATE_PP_MMX

  3582.                 if(mode & DERING){

  3583.                 //FIXME filter first line

  3584.                     if(y>0) RENAME(dering)(dstBlock - stride - 8, stride, &c);

  3585.                 }

  3586. 

  3587.                 if(mode & TEMP_NOISE_FILTER)

  3588.                 {

  3589.                     RENAME(tempNoiseReducer)(dstBlock-8, stride,

  3590.                             c.tempBlurred[isColor] + y*dstStride + x,

  3591.                             c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3) + 256,

  3592.                             c.ppMode.maxTmpNoise);

  3593.                 }

  3594.             }

  3595. 

  3596.             dstBlock+=8;

  3597.             srcBlock+=8;

  3598. 

  3599. #if TEMPLATE_PP_MMX

  3600.             tmpXchg= tempBlock1;

  3601.             tempBlock1= tempBlock2;

  3602.             tempBlock2 = tmpXchg;

  3603. #endif

  3604.         }

  3605. 

  3606.         if(mode & DERING){

  3607.             if(y > 0) RENAME(dering)(dstBlock - dstStride - 8, dstStride, &c);

  3608.         }

  3609. 

  3610.         if((mode & TEMP_NOISE_FILTER)){

  3611.             RENAME(tempNoiseReducer)(dstBlock-8, dstStride,

  3612.                     c.tempBlurred[isColor] + y*dstStride + x,

  3613.                     c.tempBlurredPast[isColor] + (y>>3)*256 + (x>>3) + 256,

  3614.                     c.ppMode.maxTmpNoise);

  3615.         }

  3616. 

  3617.         /* did we use a tmp buffer for the last lines*/

  3618.         if(y+15 >= height){

  3619.             uint8_t *dstBlock= &(dst[y*dstStride]);

  3620.             if(width==FFABS(dstStride))

  3621.                 linecpy(dstBlock, tempDst + dstStride, height-y, dstStride);

  3622.             else{

  3623.                 int i;

  3624.                 for(i=0; i<height-y; i++){

  3625.                     memcpy(dstBlock + i*dstStride, tempDst + (i+1)*dstStride, width);

  3626.                 }

  3627.             }

  3628.         }

  3629. /*

  3630.         for(x=0; x<width; x+=32){

  3631.             volatile int i;

  3632.             i+=   dstBlock[x + 7*dstStride] + dstBlock[x + 8*dstStride]

  3633.                 + dstBlock[x + 9*dstStride] + dstBlock[x +10*dstStride]

  3634.                 + dstBlock[x +11*dstStride] + dstBlock[x +12*dstStride];

  3635.                 + dstBlock[x +13*dstStride]

  3636.                 + dstBlock[x +14*dstStride] + dstBlock[x +15*dstStride];

  3637.         }*/

  3638.     }

  3639. #if   TEMPLATE_PP_3DNOW

  3640.     __asm__ volatile("femms");

  3641. #elif TEMPLATE_PP_MMX

  3642.     __asm__ volatile("emms");

  3643. #endif

  3644. 

  3645. #ifdef DEBUG_BRIGHTNESS

  3646.     if(!isColor){

  3647.         int max=1;

  3648.         int i;

  3649.         for(i=0; i<256; i++)

  3650.             if(yHistogram[i] > max) max=yHistogram[i];

  3651. 

  3652.         for(i=1; i<256; i++){

  3653.             int x;

  3654.             int start=yHistogram[i-1]/(max/256+1);

  3655.             int end=yHistogram[i]/(max/256+1);

  3656.             int inc= end > start ? 1 : -1;

  3657.             for(x=start; x!=end+inc; x+=inc)

  3658.                 dst[ i*dstStride + x]+=128;

  3659.         }

  3660. 

  3661.         for(i=0; i<100; i+=2){

  3662.             dst[ (white)*dstStride + i]+=128;

  3663.             dst[ (black)*dstStride + i]+=128;

  3664.         }

  3665.     }

  3666. #endif

  3667. 

  3668.     *c2= c; //copy local context back

  3669. 

  3670. }

  3671. 

  3672. #undef RENAME

  3673. #undef TEMPLATE_PP_C

  3674. #undef TEMPLATE_PP_ALTIVEC

  3675. #undef TEMPLATE_PP_MMX

  3676. #undef TEMPLATE_PP_MMXEXT

  3677. #undef TEMPLATE_PP_3DNOW