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.
35151 Commits
32 Branches
283MB
Tree: 473b297f26
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '473b297f26'
${ noResults }
FFmpeg/libswscale/x86/rgb2rgb_template.c

2514 lines
102KB
Raw Blame History 

  1. /*

  2.  * software RGB to RGB converter

  3.  * pluralize by software PAL8 to RGB converter

  4.  *              software YUV to YUV converter

  5.  *              software YUV to RGB converter

  6.  * Written by Nick Kurshev.

  7.  * palette & YUV & runtime CPU stuff by Michael (michaelni@gmx.at)

  8.  * lot of big-endian byte order fixes by Alex Beregszaszi

  9.  *

  10.  * This file is part of Libav.

  11.  *

  12.  * Libav is free software; you can redistribute it and/or

  13.  * modify it under the terms of the GNU Lesser General Public

  14.  * License as published by the Free Software Foundation; either

  15.  * version 2.1 of the License, or (at your option) any later version.

  16.  *

  17.  * Libav is distributed in the hope that it will be useful,

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

  19.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  20.  * Lesser General Public License for more details.

  21.  *

  22.  * You should have received a copy of the GNU Lesser General Public

  23.  * License along with Libav; if not, write to the Free Software

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

  25.  */

  26. 

  27. #include <stddef.h>

  28. 

  29. #undef PREFETCH

  30. #undef MOVNTQ

  31. #undef EMMS

  32. #undef SFENCE

  33. #undef PAVGB

  34. 

  35. #if COMPILE_TEMPLATE_AMD3DNOW

  36. #define PREFETCH  "prefetch"

  37. #define PAVGB     "pavgusb"

  38. #elif COMPILE_TEMPLATE_MMXEXT

  39. #define PREFETCH "prefetchnta"

  40. #define PAVGB     "pavgb"

  41. #else

  42. #define PREFETCH  " # nop"

  43. #endif

  44. 

  45. #if COMPILE_TEMPLATE_AMD3DNOW

  46. /* On K6 femms is faster than emms. On K7 femms is directly mapped to emms. */

  47. #define EMMS     "femms"

  48. #else

  49. #define EMMS     "emms"

  50. #endif

  51. 

  52. #if COMPILE_TEMPLATE_MMXEXT

  53. #define MOVNTQ "movntq"

  54. #define SFENCE "sfence"

  55. #else

  56. #define MOVNTQ "movq"

  57. #define SFENCE " # nop"

  58. #endif

  59. 

  60. #if !COMPILE_TEMPLATE_SSE2

  61. 

  62. #if !COMPILE_TEMPLATE_AMD3DNOW

  63. 

  64. static inline void RENAME(rgb24tobgr32)(const uint8_t *src, uint8_t *dst, int src_size)

  65. {

  66.     uint8_t *dest = dst;

  67.     const uint8_t *s = src;

  68.     const uint8_t *end;

  69.     const uint8_t *mm_end;

  70.     end = s + src_size;

  71.     __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");

  72.     mm_end = end - 23;

  73.     __asm__ volatile("movq        %0, %%mm7"::"m"(mask32a):"memory");

  74.     while (s < mm_end) {

  75.         __asm__ volatile(

  76.             PREFETCH"  32(%1)           \n\t"

  77.             "movd        (%1), %%mm0    \n\t"

  78.             "punpckldq  3(%1), %%mm0    \n\t"

  79.             "movd       6(%1), %%mm1    \n\t"

  80.             "punpckldq  9(%1), %%mm1    \n\t"

  81.             "movd      12(%1), %%mm2    \n\t"

  82.             "punpckldq 15(%1), %%mm2    \n\t"

  83.             "movd      18(%1), %%mm3    \n\t"

  84.             "punpckldq 21(%1), %%mm3    \n\t"

  85.             "por        %%mm7, %%mm0    \n\t"

  86.             "por        %%mm7, %%mm1    \n\t"

  87.             "por        %%mm7, %%mm2    \n\t"

  88.             "por        %%mm7, %%mm3    \n\t"

  89.             MOVNTQ"     %%mm0,   (%0)   \n\t"

  90.             MOVNTQ"     %%mm1,  8(%0)   \n\t"

  91.             MOVNTQ"     %%mm2, 16(%0)   \n\t"

  92.             MOVNTQ"     %%mm3, 24(%0)"

  93.             :: "r"(dest), "r"(s)

  94.             :"memory");

  95.         dest += 32;

  96.         s += 24;

  97.     }

  98.     __asm__ volatile(SFENCE:::"memory");

  99.     __asm__ volatile(EMMS:::"memory");

  100.     while (s < end) {

  101.         *dest++ = *s++;

  102.         *dest++ = *s++;

  103.         *dest++ = *s++;

  104.         *dest++ = 255;

  105.     }

  106. }

  107. 

  108. #define STORE_BGR24_MMX \

  109.             "psrlq         $8, %%mm2    \n\t" \

  110.             "psrlq         $8, %%mm3    \n\t" \

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

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

  113.             "pand "MANGLE(mask24l)", %%mm0\n\t" \

  114.             "pand "MANGLE(mask24l)", %%mm1\n\t" \

  115.             "pand "MANGLE(mask24l)", %%mm4\n\t" \

  116.             "pand "MANGLE(mask24l)", %%mm5\n\t" \

  117.             "pand "MANGLE(mask24h)", %%mm2\n\t" \

  118.             "pand "MANGLE(mask24h)", %%mm3\n\t" \

  119.             "pand "MANGLE(mask24h)", %%mm6\n\t" \

  120.             "pand "MANGLE(mask24h)", %%mm7\n\t" \

  121.             "por        %%mm2, %%mm0    \n\t" \

  122.             "por        %%mm3, %%mm1    \n\t" \

  123.             "por        %%mm6, %%mm4    \n\t" \

  124.             "por        %%mm7, %%mm5    \n\t" \

  125.  \

  126.             "movq       %%mm1, %%mm2    \n\t" \

  127.             "movq       %%mm4, %%mm3    \n\t" \

  128.             "psllq        $48, %%mm2    \n\t" \

  129.             "psllq        $32, %%mm3    \n\t" \

  130.             "pand "MANGLE(mask24hh)", %%mm2\n\t" \

  131.             "pand "MANGLE(mask24hhh)", %%mm3\n\t" \

  132.             "por        %%mm2, %%mm0    \n\t" \

  133.             "psrlq        $16, %%mm1    \n\t" \

  134.             "psrlq        $32, %%mm4    \n\t" \

  135.             "psllq        $16, %%mm5    \n\t" \

  136.             "por        %%mm3, %%mm1    \n\t" \

  137.             "pand  "MANGLE(mask24hhhh)", %%mm5\n\t" \

  138.             "por        %%mm5, %%mm4    \n\t" \

  139.  \

  140.             MOVNTQ"     %%mm0,   (%0)    \n\t" \

  141.             MOVNTQ"     %%mm1,  8(%0)    \n\t" \

  142.             MOVNTQ"     %%mm4, 16(%0)"

  143. 

  144. 

  145. static inline void RENAME(rgb32tobgr24)(const uint8_t *src, uint8_t *dst, int src_size)

  146. {

  147.     uint8_t *dest = dst;

  148.     const uint8_t *s = src;

  149.     const uint8_t *end;

  150.     const uint8_t *mm_end;

  151.     end = s + src_size;

  152.     __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");

  153.     mm_end = end - 31;

  154.     while (s < mm_end) {

  155.         __asm__ volatile(

  156.             PREFETCH"  32(%1)           \n\t"

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

  158.             "movq       8(%1), %%mm1    \n\t"

  159.             "movq      16(%1), %%mm4    \n\t"

  160.             "movq      24(%1), %%mm5    \n\t"

  161.             "movq       %%mm0, %%mm2    \n\t"

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

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

  164.             "movq       %%mm5, %%mm7    \n\t"

  165.             STORE_BGR24_MMX

  166.             :: "r"(dest), "r"(s)

  167.             :"memory");

  168.         dest += 24;

  169.         s += 32;

  170.     }

  171.     __asm__ volatile(SFENCE:::"memory");

  172.     __asm__ volatile(EMMS:::"memory");

  173.     while (s < end) {

  174.         *dest++ = *s++;

  175.         *dest++ = *s++;

  176.         *dest++ = *s++;

  177.         s++;

  178.     }

  179. }

  180. 

  181. /*

  182.  original by Strepto/Astral

  183.  ported to gcc & bugfixed: A'rpi

  184.  MMX2, 3DNOW optimization by Nick Kurshev

  185.  32-bit C version, and and&add trick by Michael Niedermayer

  186. */

  187. static inline void RENAME(rgb15to16)(const uint8_t *src, uint8_t *dst, int src_size)

  188. {

  189.     register const uint8_t* s=src;

  190.     register uint8_t* d=dst;

  191.     register const uint8_t *end;

  192.     const uint8_t *mm_end;

  193.     end = s + src_size;

  194.     __asm__ volatile(PREFETCH"    %0"::"m"(*s));

  195.     __asm__ volatile("movq        %0, %%mm4"::"m"(mask15s));

  196.     mm_end = end - 15;

  197.     while (s<mm_end) {

  198.         __asm__ volatile(

  199.             PREFETCH" 32(%1)        \n\t"

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

  201.             "movq     8(%1), %%mm2  \n\t"

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

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

  204.             "pand     %%mm4, %%mm0  \n\t"

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

  206.             "paddw    %%mm1, %%mm0  \n\t"

  207.             "paddw    %%mm3, %%mm2  \n\t"

  208.             MOVNTQ"   %%mm0,  (%0)  \n\t"

  209.             MOVNTQ"   %%mm2, 8(%0)"

  210.             :: "r"(d), "r"(s)

  211.         );

  212.         d+=16;

  213.         s+=16;

  214.     }

  215.     __asm__ volatile(SFENCE:::"memory");

  216.     __asm__ volatile(EMMS:::"memory");

  217.     mm_end = end - 3;

  218.     while (s < mm_end) {

  219.         register unsigned x= *((const uint32_t *)s);

  220.         *((uint32_t *)d) = (x&0x7FFF7FFF) + (x&0x7FE07FE0);

  221.         d+=4;

  222.         s+=4;

  223.     }

  224.     if (s < end) {

  225.         register unsigned short x= *((const uint16_t *)s);

  226.         *((uint16_t *)d) = (x&0x7FFF) + (x&0x7FE0);

  227.     }

  228. }

  229. 

  230. static inline void RENAME(rgb16to15)(const uint8_t *src, uint8_t *dst, int src_size)

  231. {

  232.     register const uint8_t* s=src;

  233.     register uint8_t* d=dst;

  234.     register const uint8_t *end;

  235.     const uint8_t *mm_end;

  236.     end = s + src_size;

  237.     __asm__ volatile(PREFETCH"    %0"::"m"(*s));

  238.     __asm__ volatile("movq        %0, %%mm7"::"m"(mask15rg));

  239.     __asm__ volatile("movq        %0, %%mm6"::"m"(mask15b));

  240.     mm_end = end - 15;

  241.     while (s<mm_end) {

  242.         __asm__ volatile(

  243.             PREFETCH" 32(%1)        \n\t"

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

  245.             "movq     8(%1), %%mm2  \n\t"

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

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

  248.             "psrlq       $1, %%mm0  \n\t"

  249.             "psrlq       $1, %%mm2  \n\t"

  250.             "pand     %%mm7, %%mm0  \n\t"

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

  252.             "pand     %%mm6, %%mm1  \n\t"

  253.             "pand     %%mm6, %%mm3  \n\t"

  254.             "por      %%mm1, %%mm0  \n\t"

  255.             "por      %%mm3, %%mm2  \n\t"

  256.             MOVNTQ"   %%mm0,  (%0)  \n\t"

  257.             MOVNTQ"   %%mm2, 8(%0)"

  258.             :: "r"(d), "r"(s)

  259.         );

  260.         d+=16;

  261.         s+=16;

  262.     }

  263.     __asm__ volatile(SFENCE:::"memory");

  264.     __asm__ volatile(EMMS:::"memory");

  265.     mm_end = end - 3;

  266.     while (s < mm_end) {

  267.         register uint32_t x= *((const uint32_t*)s);

  268.         *((uint32_t *)d) = ((x>>1)&0x7FE07FE0) | (x&0x001F001F);

  269.         s+=4;

  270.         d+=4;

  271.     }

  272.     if (s < end) {

  273.         register uint16_t x= *((const uint16_t*)s);

  274.         *((uint16_t *)d) = ((x>>1)&0x7FE0) | (x&0x001F);

  275.     }

  276. }

  277. 

  278. static inline void RENAME(rgb32to16)(const uint8_t *src, uint8_t *dst, int src_size)

  279. {

  280.     const uint8_t *s = src;

  281.     const uint8_t *end;

  282.     const uint8_t *mm_end;

  283.     uint16_t *d = (uint16_t *)dst;

  284.     end = s + src_size;

  285.     mm_end = end - 15;

  286.     __asm__ volatile(

  287.         "movq           %3, %%mm5   \n\t"

  288.         "movq           %4, %%mm6   \n\t"

  289.         "movq           %5, %%mm7   \n\t"

  290.         "jmp 2f                     \n\t"

  291.         ".p2align        4          \n\t"

  292.         "1:                         \n\t"

  293.         PREFETCH"   32(%1)          \n\t"

  294.         "movd         (%1), %%mm0   \n\t"

  295.         "movd        4(%1), %%mm3   \n\t"

  296.         "punpckldq   8(%1), %%mm0   \n\t"

  297.         "punpckldq  12(%1), %%mm3   \n\t"

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

  299.         "movq        %%mm3, %%mm4   \n\t"

  300.         "pand        %%mm6, %%mm0   \n\t"

  301.         "pand        %%mm6, %%mm3   \n\t"

  302.         "pmaddwd     %%mm7, %%mm0   \n\t"

  303.         "pmaddwd     %%mm7, %%mm3   \n\t"

  304.         "pand        %%mm5, %%mm1   \n\t"

  305.         "pand        %%mm5, %%mm4   \n\t"

  306.         "por         %%mm1, %%mm0   \n\t"

  307.         "por         %%mm4, %%mm3   \n\t"

  308.         "psrld          $5, %%mm0   \n\t"

  309.         "pslld         $11, %%mm3   \n\t"

  310.         "por         %%mm3, %%mm0   \n\t"

  311.         MOVNTQ"      %%mm0, (%0)    \n\t"

  312.         "add           $16,  %1     \n\t"

  313.         "add            $8,  %0     \n\t"

  314.         "2:                         \n\t"

  315.         "cmp            %2,  %1     \n\t"

  316.         " jb            1b          \n\t"

  317.         : "+r" (d), "+r"(s)

  318.         : "r" (mm_end), "m" (mask3216g), "m" (mask3216br), "m" (mul3216)

  319.     );

  320.     __asm__ volatile(SFENCE:::"memory");

  321.     __asm__ volatile(EMMS:::"memory");

  322.     while (s < end) {

  323.         register int rgb = *(const uint32_t*)s; s += 4;

  324.         *d++ = ((rgb&0xFF)>>3) + ((rgb&0xFC00)>>5) + ((rgb&0xF80000)>>8);

  325.     }

  326. }

  327. 

  328. static inline void RENAME(rgb32tobgr16)(const uint8_t *src, uint8_t *dst, int src_size)

  329. {

  330.     const uint8_t *s = src;

  331.     const uint8_t *end;

  332.     const uint8_t *mm_end;

  333.     uint16_t *d = (uint16_t *)dst;

  334.     end = s + src_size;

  335.     __asm__ volatile(PREFETCH"    %0"::"m"(*src):"memory");

  336.     __asm__ volatile(

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

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

  339.         ::"m"(red_16mask),"m"(green_16mask));

  340.     mm_end = end - 15;

  341.     while (s < mm_end) {

  342.         __asm__ volatile(

  343.             PREFETCH"  32(%1)           \n\t"

  344.             "movd        (%1), %%mm0    \n\t"

  345.             "movd       4(%1), %%mm3    \n\t"

  346.             "punpckldq  8(%1), %%mm0    \n\t"

  347.             "punpckldq 12(%1), %%mm3    \n\t"

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

  349.             "movq       %%mm0, %%mm2    \n\t"

  350.             "movq       %%mm3, %%mm4    \n\t"

  351.             "movq       %%mm3, %%mm5    \n\t"

  352.             "psllq         $8, %%mm0    \n\t"

  353.             "psllq         $8, %%mm3    \n\t"

  354.             "pand       %%mm7, %%mm0    \n\t"

  355.             "pand       %%mm7, %%mm3    \n\t"

  356.             "psrlq         $5, %%mm1    \n\t"

  357.             "psrlq         $5, %%mm4    \n\t"

  358.             "pand       %%mm6, %%mm1    \n\t"

  359.             "pand       %%mm6, %%mm4    \n\t"

  360.             "psrlq        $19, %%mm2    \n\t"

  361.             "psrlq        $19, %%mm5    \n\t"

  362.             "pand          %2, %%mm2    \n\t"

  363.             "pand          %2, %%mm5    \n\t"

  364.             "por        %%mm1, %%mm0    \n\t"

  365.             "por        %%mm4, %%mm3    \n\t"

  366.             "por        %%mm2, %%mm0    \n\t"

  367.             "por        %%mm5, %%mm3    \n\t"

  368.             "psllq        $16, %%mm3    \n\t"

  369.             "por        %%mm3, %%mm0    \n\t"

  370.             MOVNTQ"     %%mm0, (%0)     \n\t"

  371.             :: "r"(d),"r"(s),"m"(blue_16mask):"memory");

  372.         d += 4;

  373.         s += 16;

  374.     }

  375.     __asm__ volatile(SFENCE:::"memory");

  376.     __asm__ volatile(EMMS:::"memory");

  377.     while (s < end) {

  378.         register int rgb = *(const uint32_t*)s; s += 4;

  379.         *d++ = ((rgb&0xF8)<<8) + ((rgb&0xFC00)>>5) + ((rgb&0xF80000)>>19);

  380.     }

  381. }

  382. 

  383. static inline void RENAME(rgb32to15)(const uint8_t *src, uint8_t *dst, int src_size)

  384. {

  385.     const uint8_t *s = src;

  386.     const uint8_t *end;

  387.     const uint8_t *mm_end;

  388.     uint16_t *d = (uint16_t *)dst;

  389.     end = s + src_size;

  390.     mm_end = end - 15;

  391.     __asm__ volatile(

  392.         "movq           %3, %%mm5   \n\t"

  393.         "movq           %4, %%mm6   \n\t"

  394.         "movq           %5, %%mm7   \n\t"

  395.         "jmp            2f          \n\t"

  396.         ".p2align        4          \n\t"

  397.         "1:                         \n\t"

  398.         PREFETCH"   32(%1)          \n\t"

  399.         "movd         (%1), %%mm0   \n\t"

  400.         "movd        4(%1), %%mm3   \n\t"

  401.         "punpckldq   8(%1), %%mm0   \n\t"

  402.         "punpckldq  12(%1), %%mm3   \n\t"

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

  404.         "movq        %%mm3, %%mm4   \n\t"

  405.         "pand        %%mm6, %%mm0   \n\t"

  406.         "pand        %%mm6, %%mm3   \n\t"

  407.         "pmaddwd     %%mm7, %%mm0   \n\t"

  408.         "pmaddwd     %%mm7, %%mm3   \n\t"

  409.         "pand        %%mm5, %%mm1   \n\t"

  410.         "pand        %%mm5, %%mm4   \n\t"

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

  412.         "por         %%mm4, %%mm3   \n\t"

  413.         "psrld          $6, %%mm0   \n\t"

  414.         "pslld         $10, %%mm3   \n\t"

  415.         "por         %%mm3, %%mm0   \n\t"

  416.         MOVNTQ"      %%mm0, (%0)    \n\t"

  417.         "add           $16,  %1     \n\t"

  418.         "add            $8,  %0     \n\t"

  419.         "2:                         \n\t"

  420.         "cmp            %2,  %1     \n\t"

  421.         " jb            1b          \n\t"

  422.         : "+r" (d), "+r"(s)

  423.         : "r" (mm_end), "m" (mask3215g), "m" (mask3216br), "m" (mul3215)

  424.     );

  425.     __asm__ volatile(SFENCE:::"memory");

  426.     __asm__ volatile(EMMS:::"memory");

  427.     while (s < end) {

  428.         register int rgb = *(const uint32_t*)s; s += 4;

  429.         *d++ = ((rgb&0xFF)>>3) + ((rgb&0xF800)>>6) + ((rgb&0xF80000)>>9);

  430.     }

  431. }

  432. 

  433. static inline void RENAME(rgb32tobgr15)(const uint8_t *src, uint8_t *dst, int src_size)

  434. {

  435.     const uint8_t *s = src;

  436.     const uint8_t *end;

  437.     const uint8_t *mm_end;

  438.     uint16_t *d = (uint16_t *)dst;

  439.     end = s + src_size;

  440.     __asm__ volatile(PREFETCH"    %0"::"m"(*src):"memory");

  441.     __asm__ volatile(

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

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

  444.         ::"m"(red_15mask),"m"(green_15mask));

  445.     mm_end = end - 15;

  446.     while (s < mm_end) {

  447.         __asm__ volatile(

  448.             PREFETCH"  32(%1)           \n\t"

  449.             "movd        (%1), %%mm0    \n\t"

  450.             "movd       4(%1), %%mm3    \n\t"

  451.             "punpckldq  8(%1), %%mm0    \n\t"

  452.             "punpckldq 12(%1), %%mm3    \n\t"

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

  454.             "movq       %%mm0, %%mm2    \n\t"

  455.             "movq       %%mm3, %%mm4    \n\t"

  456.             "movq       %%mm3, %%mm5    \n\t"

  457.             "psllq         $7, %%mm0    \n\t"

  458.             "psllq         $7, %%mm3    \n\t"

  459.             "pand       %%mm7, %%mm0    \n\t"

  460.             "pand       %%mm7, %%mm3    \n\t"

  461.             "psrlq         $6, %%mm1    \n\t"

  462.             "psrlq         $6, %%mm4    \n\t"

  463.             "pand       %%mm6, %%mm1    \n\t"

  464.             "pand       %%mm6, %%mm4    \n\t"

  465.             "psrlq        $19, %%mm2    \n\t"

  466.             "psrlq        $19, %%mm5    \n\t"

  467.             "pand          %2, %%mm2    \n\t"

  468.             "pand          %2, %%mm5    \n\t"

  469.             "por        %%mm1, %%mm0    \n\t"

  470.             "por        %%mm4, %%mm3    \n\t"

  471.             "por        %%mm2, %%mm0    \n\t"

  472.             "por        %%mm5, %%mm3    \n\t"

  473.             "psllq        $16, %%mm3    \n\t"

  474.             "por        %%mm3, %%mm0    \n\t"

  475.             MOVNTQ"     %%mm0, (%0)     \n\t"

  476.             ::"r"(d),"r"(s),"m"(blue_15mask):"memory");

  477.         d += 4;

  478.         s += 16;

  479.     }

  480.     __asm__ volatile(SFENCE:::"memory");

  481.     __asm__ volatile(EMMS:::"memory");

  482.     while (s < end) {

  483.         register int rgb = *(const uint32_t*)s; s += 4;

  484.         *d++ = ((rgb&0xF8)<<7) + ((rgb&0xF800)>>6) + ((rgb&0xF80000)>>19);

  485.     }

  486. }

  487. 

  488. static inline void RENAME(rgb24tobgr16)(const uint8_t *src, uint8_t *dst, int src_size)

  489. {

  490.     const uint8_t *s = src;

  491.     const uint8_t *end;

  492.     const uint8_t *mm_end;

  493.     uint16_t *d = (uint16_t *)dst;

  494.     end = s + src_size;

  495.     __asm__ volatile(PREFETCH"    %0"::"m"(*src):"memory");

  496.     __asm__ volatile(

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

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

  499.         ::"m"(red_16mask),"m"(green_16mask));

  500.     mm_end = end - 11;

  501.     while (s < mm_end) {

  502.         __asm__ volatile(

  503.             PREFETCH"  32(%1)           \n\t"

  504.             "movd        (%1), %%mm0    \n\t"

  505.             "movd       3(%1), %%mm3    \n\t"

  506.             "punpckldq  6(%1), %%mm0    \n\t"

  507.             "punpckldq  9(%1), %%mm3    \n\t"

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

  509.             "movq       %%mm0, %%mm2    \n\t"

  510.             "movq       %%mm3, %%mm4    \n\t"

  511.             "movq       %%mm3, %%mm5    \n\t"

  512.             "psrlq         $3, %%mm0    \n\t"

  513.             "psrlq         $3, %%mm3    \n\t"

  514.             "pand          %2, %%mm0    \n\t"

  515.             "pand          %2, %%mm3    \n\t"

  516.             "psrlq         $5, %%mm1    \n\t"

  517.             "psrlq         $5, %%mm4    \n\t"

  518.             "pand       %%mm6, %%mm1    \n\t"

  519.             "pand       %%mm6, %%mm4    \n\t"

  520.             "psrlq         $8, %%mm2    \n\t"

  521.             "psrlq         $8, %%mm5    \n\t"

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

  523.             "pand       %%mm7, %%mm5    \n\t"

  524.             "por        %%mm1, %%mm0    \n\t"

  525.             "por        %%mm4, %%mm3    \n\t"

  526.             "por        %%mm2, %%mm0    \n\t"

  527.             "por        %%mm5, %%mm3    \n\t"

  528.             "psllq        $16, %%mm3    \n\t"

  529.             "por        %%mm3, %%mm0    \n\t"

  530.             MOVNTQ"     %%mm0, (%0)     \n\t"

  531.             ::"r"(d),"r"(s),"m"(blue_16mask):"memory");

  532.         d += 4;

  533.         s += 12;

  534.     }

  535.     __asm__ volatile(SFENCE:::"memory");

  536.     __asm__ volatile(EMMS:::"memory");

  537.     while (s < end) {

  538.         const int b = *s++;

  539.         const int g = *s++;

  540.         const int r = *s++;

  541.         *d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);

  542.     }

  543. }

  544. 

  545. static inline void RENAME(rgb24to16)(const uint8_t *src, uint8_t *dst, int src_size)

  546. {

  547.     const uint8_t *s = src;

  548.     const uint8_t *end;

  549.     const uint8_t *mm_end;

  550.     uint16_t *d = (uint16_t *)dst;

  551.     end = s + src_size;

  552.     __asm__ volatile(PREFETCH"    %0"::"m"(*src):"memory");

  553.     __asm__ volatile(

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

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

  556.         ::"m"(red_16mask),"m"(green_16mask));

  557.     mm_end = end - 15;

  558.     while (s < mm_end) {

  559.         __asm__ volatile(

  560.             PREFETCH"  32(%1)           \n\t"

  561.             "movd        (%1), %%mm0    \n\t"

  562.             "movd       3(%1), %%mm3    \n\t"

  563.             "punpckldq  6(%1), %%mm0    \n\t"

  564.             "punpckldq  9(%1), %%mm3    \n\t"

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

  566.             "movq       %%mm0, %%mm2    \n\t"

  567.             "movq       %%mm3, %%mm4    \n\t"

  568.             "movq       %%mm3, %%mm5    \n\t"

  569.             "psllq         $8, %%mm0    \n\t"

  570.             "psllq         $8, %%mm3    \n\t"

  571.             "pand       %%mm7, %%mm0    \n\t"

  572.             "pand       %%mm7, %%mm3    \n\t"

  573.             "psrlq         $5, %%mm1    \n\t"

  574.             "psrlq         $5, %%mm4    \n\t"

  575.             "pand       %%mm6, %%mm1    \n\t"

  576.             "pand       %%mm6, %%mm4    \n\t"

  577.             "psrlq        $19, %%mm2    \n\t"

  578.             "psrlq        $19, %%mm5    \n\t"

  579.             "pand          %2, %%mm2    \n\t"

  580.             "pand          %2, %%mm5    \n\t"

  581.             "por        %%mm1, %%mm0    \n\t"

  582.             "por        %%mm4, %%mm3    \n\t"

  583.             "por        %%mm2, %%mm0    \n\t"

  584.             "por        %%mm5, %%mm3    \n\t"

  585.             "psllq        $16, %%mm3    \n\t"

  586.             "por        %%mm3, %%mm0    \n\t"

  587.             MOVNTQ"     %%mm0, (%0)     \n\t"

  588.             ::"r"(d),"r"(s),"m"(blue_16mask):"memory");

  589.         d += 4;

  590.         s += 12;

  591.     }

  592.     __asm__ volatile(SFENCE:::"memory");

  593.     __asm__ volatile(EMMS:::"memory");

  594.     while (s < end) {

  595.         const int r = *s++;

  596.         const int g = *s++;

  597.         const int b = *s++;

  598.         *d++ = (b>>3) | ((g&0xFC)<<3) | ((r&0xF8)<<8);

  599.     }

  600. }

  601. 

  602. static inline void RENAME(rgb24tobgr15)(const uint8_t *src, uint8_t *dst, int src_size)

  603. {

  604.     const uint8_t *s = src;

  605.     const uint8_t *end;

  606.     const uint8_t *mm_end;

  607.     uint16_t *d = (uint16_t *)dst;

  608.     end = s + src_size;

  609.     __asm__ volatile(PREFETCH"    %0"::"m"(*src):"memory");

  610.     __asm__ volatile(

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

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

  613.         ::"m"(red_15mask),"m"(green_15mask));

  614.     mm_end = end - 11;

  615.     while (s < mm_end) {

  616.         __asm__ volatile(

  617.             PREFETCH"  32(%1)           \n\t"

  618.             "movd        (%1), %%mm0    \n\t"

  619.             "movd       3(%1), %%mm3    \n\t"

  620.             "punpckldq  6(%1), %%mm0    \n\t"

  621.             "punpckldq  9(%1), %%mm3    \n\t"

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

  623.             "movq       %%mm0, %%mm2    \n\t"

  624.             "movq       %%mm3, %%mm4    \n\t"

  625.             "movq       %%mm3, %%mm5    \n\t"

  626.             "psrlq         $3, %%mm0    \n\t"

  627.             "psrlq         $3, %%mm3    \n\t"

  628.             "pand          %2, %%mm0    \n\t"

  629.             "pand          %2, %%mm3    \n\t"

  630.             "psrlq         $6, %%mm1    \n\t"

  631.             "psrlq         $6, %%mm4    \n\t"

  632.             "pand       %%mm6, %%mm1    \n\t"

  633.             "pand       %%mm6, %%mm4    \n\t"

  634.             "psrlq         $9, %%mm2    \n\t"

  635.             "psrlq         $9, %%mm5    \n\t"

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

  637.             "pand       %%mm7, %%mm5    \n\t"

  638.             "por        %%mm1, %%mm0    \n\t"

  639.             "por        %%mm4, %%mm3    \n\t"

  640.             "por        %%mm2, %%mm0    \n\t"

  641.             "por        %%mm5, %%mm3    \n\t"

  642.             "psllq        $16, %%mm3    \n\t"

  643.             "por        %%mm3, %%mm0    \n\t"

  644.             MOVNTQ"     %%mm0, (%0)     \n\t"

  645.             ::"r"(d),"r"(s),"m"(blue_15mask):"memory");

  646.         d += 4;

  647.         s += 12;

  648.     }

  649.     __asm__ volatile(SFENCE:::"memory");

  650.     __asm__ volatile(EMMS:::"memory");

  651.     while (s < end) {

  652.         const int b = *s++;

  653.         const int g = *s++;

  654.         const int r = *s++;

  655.         *d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);

  656.     }

  657. }

  658. 

  659. static inline void RENAME(rgb24to15)(const uint8_t *src, uint8_t *dst, int src_size)

  660. {

  661.     const uint8_t *s = src;

  662.     const uint8_t *end;

  663.     const uint8_t *mm_end;

  664.     uint16_t *d = (uint16_t *)dst;

  665.     end = s + src_size;

  666.     __asm__ volatile(PREFETCH"    %0"::"m"(*src):"memory");

  667.     __asm__ volatile(

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

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

  670.         ::"m"(red_15mask),"m"(green_15mask));

  671.     mm_end = end - 15;

  672.     while (s < mm_end) {

  673.         __asm__ volatile(

  674.             PREFETCH" 32(%1)            \n\t"

  675.             "movd       (%1), %%mm0     \n\t"

  676.             "movd      3(%1), %%mm3     \n\t"

  677.             "punpckldq 6(%1), %%mm0     \n\t"

  678.             "punpckldq 9(%1), %%mm3     \n\t"

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

  680.             "movq      %%mm0, %%mm2     \n\t"

  681.             "movq      %%mm3, %%mm4     \n\t"

  682.             "movq      %%mm3, %%mm5     \n\t"

  683.             "psllq        $7, %%mm0     \n\t"

  684.             "psllq        $7, %%mm3     \n\t"

  685.             "pand      %%mm7, %%mm0     \n\t"

  686.             "pand      %%mm7, %%mm3     \n\t"

  687.             "psrlq        $6, %%mm1     \n\t"

  688.             "psrlq        $6, %%mm4     \n\t"

  689.             "pand      %%mm6, %%mm1     \n\t"

  690.             "pand      %%mm6, %%mm4     \n\t"

  691.             "psrlq       $19, %%mm2     \n\t"

  692.             "psrlq       $19, %%mm5     \n\t"

  693.             "pand         %2, %%mm2     \n\t"

  694.             "pand         %2, %%mm5     \n\t"

  695.             "por       %%mm1, %%mm0     \n\t"

  696.             "por       %%mm4, %%mm3     \n\t"

  697.             "por       %%mm2, %%mm0     \n\t"

  698.             "por       %%mm5, %%mm3     \n\t"

  699.             "psllq       $16, %%mm3     \n\t"

  700.             "por       %%mm3, %%mm0     \n\t"

  701.             MOVNTQ"    %%mm0, (%0)      \n\t"

  702.             ::"r"(d),"r"(s),"m"(blue_15mask):"memory");

  703.         d += 4;

  704.         s += 12;

  705.     }

  706.     __asm__ volatile(SFENCE:::"memory");

  707.     __asm__ volatile(EMMS:::"memory");

  708.     while (s < end) {

  709.         const int r = *s++;

  710.         const int g = *s++;

  711.         const int b = *s++;

  712.         *d++ = (b>>3) | ((g&0xF8)<<2) | ((r&0xF8)<<7);

  713.     }

  714. }

  715. 

  716. /*

  717.   I use less accurate approximation here by simply left-shifting the input

  718.   value and filling the low order bits with zeroes. This method improves PNG

  719.   compression but this scheme cannot reproduce white exactly, since it does

  720.   not generate an all-ones maximum value; the net effect is to darken the

  721.   image slightly.

  722. 

  723.   The better method should be "left bit replication":

  724. 

  725.    4 3 2 1 0

  726.    ---------

  727.    1 1 0 1 1

  728. 

  729.    7 6 5 4 3  2 1 0

  730.    ----------------

  731.    1 1 0 1 1  1 1 0

  732.    |=======|  |===|

  733.        |      leftmost bits repeated to fill open bits

  734.        |

  735.    original bits

  736. */

  737. static inline void RENAME(rgb15tobgr24)(const uint8_t *src, uint8_t *dst, int src_size)

  738. {

  739.     const uint16_t *end;

  740.     const uint16_t *mm_end;

  741.     uint8_t *d = dst;

  742.     const uint16_t *s = (const uint16_t*)src;

  743.     end = s + src_size/2;

  744.     __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");

  745.     mm_end = end - 7;

  746.     while (s < mm_end) {

  747.         __asm__ volatile(

  748.             PREFETCH"  32(%1)           \n\t"

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

  750.             "movq        (%1), %%mm1    \n\t"

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

  752.             "pand          %2, %%mm0    \n\t"

  753.             "pand          %3, %%mm1    \n\t"

  754.             "pand          %4, %%mm2    \n\t"

  755.             "psllq         $3, %%mm0    \n\t"

  756.             "psrlq         $2, %%mm1    \n\t"

  757.             "psrlq         $7, %%mm2    \n\t"

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

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

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

  761.             "punpcklwd     %5, %%mm0    \n\t"

  762.             "punpcklwd     %5, %%mm1    \n\t"

  763.             "punpcklwd     %5, %%mm2    \n\t"

  764.             "punpckhwd     %5, %%mm3    \n\t"

  765.             "punpckhwd     %5, %%mm4    \n\t"

  766.             "punpckhwd     %5, %%mm5    \n\t"

  767.             "psllq         $8, %%mm1    \n\t"

  768.             "psllq        $16, %%mm2    \n\t"

  769.             "por        %%mm1, %%mm0    \n\t"

  770.             "por        %%mm2, %%mm0    \n\t"

  771.             "psllq         $8, %%mm4    \n\t"

  772.             "psllq        $16, %%mm5    \n\t"

  773.             "por        %%mm4, %%mm3    \n\t"

  774.             "por        %%mm5, %%mm3    \n\t"

  775. 

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

  777.             "movq       %%mm3, %%mm7    \n\t"

  778. 

  779.             "movq       8(%1), %%mm0    \n\t"

  780.             "movq       8(%1), %%mm1    \n\t"

  781.             "movq       8(%1), %%mm2    \n\t"

  782.             "pand          %2, %%mm0    \n\t"

  783.             "pand          %3, %%mm1    \n\t"

  784.             "pand          %4, %%mm2    \n\t"

  785.             "psllq         $3, %%mm0    \n\t"

  786.             "psrlq         $2, %%mm1    \n\t"

  787.             "psrlq         $7, %%mm2    \n\t"

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

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

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

  791.             "punpcklwd     %5, %%mm0    \n\t"

  792.             "punpcklwd     %5, %%mm1    \n\t"

  793.             "punpcklwd     %5, %%mm2    \n\t"

  794.             "punpckhwd     %5, %%mm3    \n\t"

  795.             "punpckhwd     %5, %%mm4    \n\t"

  796.             "punpckhwd     %5, %%mm5    \n\t"

  797.             "psllq         $8, %%mm1    \n\t"

  798.             "psllq        $16, %%mm2    \n\t"

  799.             "por        %%mm1, %%mm0    \n\t"

  800.             "por        %%mm2, %%mm0    \n\t"

  801.             "psllq         $8, %%mm4    \n\t"

  802.             "psllq        $16, %%mm5    \n\t"

  803.             "por        %%mm4, %%mm3    \n\t"

  804.             "por        %%mm5, %%mm3    \n\t"

  805. 

  806.             :"=m"(*d)

  807.             :"r"(s),"m"(mask15b),"m"(mask15g),"m"(mask15r), "m"(mmx_null)

  808.             :"memory");

  809.         /* borrowed 32 to 24 */

  810.         __asm__ volatile(

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

  812.             "movq       %%mm3, %%mm5    \n\t"

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

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

  815. 

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

  817.             "movq       %%mm5, %%mm7    \n\t"

  818.             "movq       %%mm0, %%mm2    \n\t"

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

  820. 

  821.             STORE_BGR24_MMX

  822. 

  823.             :: "r"(d), "m"(*s)

  824.             :"memory");

  825.         d += 24;

  826.         s += 8;

  827.     }

  828.     __asm__ volatile(SFENCE:::"memory");

  829.     __asm__ volatile(EMMS:::"memory");

  830.     while (s < end) {

  831.         register uint16_t bgr;

  832.         bgr = *s++;

  833.         *d++ = (bgr&0x1F)<<3;

  834.         *d++ = (bgr&0x3E0)>>2;

  835.         *d++ = (bgr&0x7C00)>>7;

  836.     }

  837. }

  838. 

  839. static inline void RENAME(rgb16tobgr24)(const uint8_t *src, uint8_t *dst, int src_size)

  840. {

  841.     const uint16_t *end;

  842.     const uint16_t *mm_end;

  843.     uint8_t *d = (uint8_t *)dst;

  844.     const uint16_t *s = (const uint16_t *)src;

  845.     end = s + src_size/2;

  846.     __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");

  847.     mm_end = end - 7;

  848.     while (s < mm_end) {

  849.         __asm__ volatile(

  850.             PREFETCH"  32(%1)           \n\t"

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

  852.             "movq        (%1), %%mm1    \n\t"

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

  854.             "pand          %2, %%mm0    \n\t"

  855.             "pand          %3, %%mm1    \n\t"

  856.             "pand          %4, %%mm2    \n\t"

  857.             "psllq         $3, %%mm0    \n\t"

  858.             "psrlq         $3, %%mm1    \n\t"

  859.             "psrlq         $8, %%mm2    \n\t"

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

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

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

  863.             "punpcklwd     %5, %%mm0    \n\t"

  864.             "punpcklwd     %5, %%mm1    \n\t"

  865.             "punpcklwd     %5, %%mm2    \n\t"

  866.             "punpckhwd     %5, %%mm3    \n\t"

  867.             "punpckhwd     %5, %%mm4    \n\t"

  868.             "punpckhwd     %5, %%mm5    \n\t"

  869.             "psllq         $8, %%mm1    \n\t"

  870.             "psllq        $16, %%mm2    \n\t"

  871.             "por        %%mm1, %%mm0    \n\t"

  872.             "por        %%mm2, %%mm0    \n\t"

  873.             "psllq         $8, %%mm4    \n\t"

  874.             "psllq        $16, %%mm5    \n\t"

  875.             "por        %%mm4, %%mm3    \n\t"

  876.             "por        %%mm5, %%mm3    \n\t"

  877. 

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

  879.             "movq       %%mm3, %%mm7    \n\t"

  880. 

  881.             "movq       8(%1), %%mm0    \n\t"

  882.             "movq       8(%1), %%mm1    \n\t"

  883.             "movq       8(%1), %%mm2    \n\t"

  884.             "pand          %2, %%mm0    \n\t"

  885.             "pand          %3, %%mm1    \n\t"

  886.             "pand          %4, %%mm2    \n\t"

  887.             "psllq         $3, %%mm0    \n\t"

  888.             "psrlq         $3, %%mm1    \n\t"

  889.             "psrlq         $8, %%mm2    \n\t"

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

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

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

  893.             "punpcklwd     %5, %%mm0    \n\t"

  894.             "punpcklwd     %5, %%mm1    \n\t"

  895.             "punpcklwd     %5, %%mm2    \n\t"

  896.             "punpckhwd     %5, %%mm3    \n\t"

  897.             "punpckhwd     %5, %%mm4    \n\t"

  898.             "punpckhwd     %5, %%mm5    \n\t"

  899.             "psllq         $8, %%mm1    \n\t"

  900.             "psllq        $16, %%mm2    \n\t"

  901.             "por        %%mm1, %%mm0    \n\t"

  902.             "por        %%mm2, %%mm0    \n\t"

  903.             "psllq         $8, %%mm4    \n\t"

  904.             "psllq        $16, %%mm5    \n\t"

  905.             "por        %%mm4, %%mm3    \n\t"

  906.             "por        %%mm5, %%mm3    \n\t"

  907.             :"=m"(*d)

  908.             :"r"(s),"m"(mask16b),"m"(mask16g),"m"(mask16r),"m"(mmx_null)

  909.             :"memory");

  910.         /* borrowed 32 to 24 */

  911.         __asm__ volatile(

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

  913.             "movq       %%mm3, %%mm5    \n\t"

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

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

  916. 

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

  918.             "movq       %%mm5, %%mm7    \n\t"

  919.             "movq       %%mm0, %%mm2    \n\t"

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

  921. 

  922.             STORE_BGR24_MMX

  923. 

  924.             :: "r"(d), "m"(*s)

  925.             :"memory");

  926.         d += 24;

  927.         s += 8;

  928.     }

  929.     __asm__ volatile(SFENCE:::"memory");

  930.     __asm__ volatile(EMMS:::"memory");

  931.     while (s < end) {

  932.         register uint16_t bgr;

  933.         bgr = *s++;

  934.         *d++ = (bgr&0x1F)<<3;

  935.         *d++ = (bgr&0x7E0)>>3;

  936.         *d++ = (bgr&0xF800)>>8;

  937.     }

  938. }

  939. 

  940. /*

  941.  * mm0 = 00 B3 00 B2 00 B1 00 B0

  942.  * mm1 = 00 G3 00 G2 00 G1 00 G0

  943.  * mm2 = 00 R3 00 R2 00 R1 00 R0

  944.  * mm6 = FF FF FF FF FF FF FF FF

  945.  * mm7 = 00 00 00 00 00 00 00 00

  946.  */

  947. #define PACK_RGB32 \

  948.     "packuswb   %%mm7, %%mm0    \n\t" /* 00 00 00 00 B3 B2 B1 B0 */ \

  949.     "packuswb   %%mm7, %%mm1    \n\t" /* 00 00 00 00 G3 G2 G1 G0 */ \

  950.     "packuswb   %%mm7, %%mm2    \n\t" /* 00 00 00 00 R3 R2 R1 R0 */ \

  951.     "punpcklbw  %%mm1, %%mm0    \n\t" /* G3 B3 G2 B2 G1 B1 G0 B0 */ \

  952.     "punpcklbw  %%mm6, %%mm2    \n\t" /* FF R3 FF R2 FF R1 FF R0 */ \

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

  954.     "punpcklwd  %%mm2, %%mm0    \n\t" /* FF R1 G1 B1 FF R0 G0 B0 */ \

  955.     "punpckhwd  %%mm2, %%mm3    \n\t" /* FF R3 G3 B3 FF R2 G2 B2 */ \

  956.     MOVNTQ"     %%mm0,  (%0)    \n\t"                               \

  957.     MOVNTQ"     %%mm3, 8(%0)    \n\t"                               \

  958. 

  959. static inline void RENAME(rgb15to32)(const uint8_t *src, uint8_t *dst, int src_size)

  960. {

  961.     const uint16_t *end;

  962.     const uint16_t *mm_end;

  963.     uint8_t *d = dst;

  964.     const uint16_t *s = (const uint16_t *)src;

  965.     end = s + src_size/2;

  966.     __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");

  967.     __asm__ volatile("pxor    %%mm7,%%mm7    \n\t":::"memory");

  968.     __asm__ volatile("pcmpeqd %%mm6,%%mm6    \n\t":::"memory");

  969.     mm_end = end - 3;

  970.     while (s < mm_end) {

  971.         __asm__ volatile(

  972.             PREFETCH"  32(%1)           \n\t"

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

  974.             "movq        (%1), %%mm1    \n\t"

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

  976.             "pand          %2, %%mm0    \n\t"

  977.             "pand          %3, %%mm1    \n\t"

  978.             "pand          %4, %%mm2    \n\t"

  979.             "psllq         $3, %%mm0    \n\t"

  980.             "psrlq         $2, %%mm1    \n\t"

  981.             "psrlq         $7, %%mm2    \n\t"

  982.             PACK_RGB32

  983.             ::"r"(d),"r"(s),"m"(mask15b),"m"(mask15g),"m"(mask15r)

  984.             :"memory");

  985.         d += 16;

  986.         s += 4;

  987.     }

  988.     __asm__ volatile(SFENCE:::"memory");

  989.     __asm__ volatile(EMMS:::"memory");

  990.     while (s < end) {

  991.         register uint16_t bgr;

  992.         bgr = *s++;

  993.         *d++ = (bgr&0x1F)<<3;

  994.         *d++ = (bgr&0x3E0)>>2;

  995.         *d++ = (bgr&0x7C00)>>7;

  996.         *d++ = 255;

  997.     }

  998. }

  999. 

  1000. static inline void RENAME(rgb16to32)(const uint8_t *src, uint8_t *dst, int src_size)

  1001. {

  1002.     const uint16_t *end;

  1003.     const uint16_t *mm_end;

  1004.     uint8_t *d = dst;

  1005.     const uint16_t *s = (const uint16_t*)src;

  1006.     end = s + src_size/2;

  1007.     __asm__ volatile(PREFETCH"    %0"::"m"(*s):"memory");

  1008.     __asm__ volatile("pxor    %%mm7,%%mm7    \n\t":::"memory");

  1009.     __asm__ volatile("pcmpeqd %%mm6,%%mm6    \n\t":::"memory");

  1010.     mm_end = end - 3;

  1011.     while (s < mm_end) {

  1012.         __asm__ volatile(

  1013.             PREFETCH"  32(%1)           \n\t"

  1014.             "movq        (%1), %%mm0    \n\t"

  1015.             "movq        (%1), %%mm1    \n\t"

  1016.             "movq        (%1), %%mm2    \n\t"

  1017.             "pand          %2, %%mm0    \n\t"

  1018.             "pand          %3, %%mm1    \n\t"

  1019.             "pand          %4, %%mm2    \n\t"

  1020.             "psllq         $3, %%mm0    \n\t"

  1021.             "psrlq         $3, %%mm1    \n\t"

  1022.             "psrlq         $8, %%mm2    \n\t"

  1023.             PACK_RGB32

  1024.             ::"r"(d),"r"(s),"m"(mask16b),"m"(mask16g),"m"(mask16r)

  1025.             :"memory");

  1026.         d += 16;

  1027.         s += 4;

  1028.     }

  1029.     __asm__ volatile(SFENCE:::"memory");

  1030.     __asm__ volatile(EMMS:::"memory");

  1031.     while (s < end) {

  1032.         register uint16_t bgr;

  1033.         bgr = *s++;

  1034.         *d++ = (bgr&0x1F)<<3;

  1035.         *d++ = (bgr&0x7E0)>>3;

  1036.         *d++ = (bgr&0xF800)>>8;

  1037.         *d++ = 255;

  1038.     }

  1039. }

  1040. 

  1041. static inline void RENAME(shuffle_bytes_2103)(const uint8_t *src, uint8_t *dst, int src_size)

  1042. {

  1043.     x86_reg idx = 15 - src_size;

  1044.     const uint8_t *s = src-idx;

  1045.     uint8_t *d = dst-idx;

  1046.     __asm__ volatile(

  1047.         "test          %0, %0           \n\t"

  1048.         "jns           2f               \n\t"

  1049.         PREFETCH"       (%1, %0)        \n\t"

  1050.         "movq          %3, %%mm7        \n\t"

  1051.         "pxor          %4, %%mm7        \n\t"

  1052.         "movq       %%mm7, %%mm6        \n\t"

  1053.         "pxor          %5, %%mm7        \n\t"

  1054.         ".p2align       4               \n\t"

  1055.         "1:                             \n\t"

  1056.         PREFETCH"     32(%1, %0)        \n\t"

  1057.         "movq           (%1, %0), %%mm0 \n\t"

  1058.         "movq          8(%1, %0), %%mm1 \n\t"

  1059. # if COMPILE_TEMPLATE_MMXEXT

  1060.         "pshufw      $177, %%mm0, %%mm3 \n\t"

  1061.         "pshufw      $177, %%mm1, %%mm5 \n\t"

  1062.         "pand       %%mm7, %%mm0        \n\t"

  1063.         "pand       %%mm6, %%mm3        \n\t"

  1064.         "pand       %%mm7, %%mm1        \n\t"

  1065.         "pand       %%mm6, %%mm5        \n\t"

  1066.         "por        %%mm3, %%mm0        \n\t"

  1067.         "por        %%mm5, %%mm1        \n\t"

  1068. # else

  1069.         "movq       %%mm0, %%mm2        \n\t"

  1070.         "movq       %%mm1, %%mm4        \n\t"

  1071.         "pand       %%mm7, %%mm0        \n\t"

  1072.         "pand       %%mm6, %%mm2        \n\t"

  1073.         "pand       %%mm7, %%mm1        \n\t"

  1074.         "pand       %%mm6, %%mm4        \n\t"

  1075.         "movq       %%mm2, %%mm3        \n\t"

  1076.         "movq       %%mm4, %%mm5        \n\t"

  1077.         "pslld        $16, %%mm2        \n\t"

  1078.         "psrld        $16, %%mm3        \n\t"

  1079.         "pslld        $16, %%mm4        \n\t"

  1080.         "psrld        $16, %%mm5        \n\t"

  1081.         "por        %%mm2, %%mm0        \n\t"

  1082.         "por        %%mm4, %%mm1        \n\t"

  1083.         "por        %%mm3, %%mm0        \n\t"

  1084.         "por        %%mm5, %%mm1        \n\t"

  1085. # endif

  1086.         MOVNTQ"     %%mm0,  (%2, %0)    \n\t"

  1087.         MOVNTQ"     %%mm1, 8(%2, %0)    \n\t"

  1088.         "add          $16, %0           \n\t"

  1089.         "js            1b               \n\t"

  1090.         SFENCE"                         \n\t"

  1091.         EMMS"                           \n\t"

  1092.         "2:                             \n\t"

  1093.         : "+&r"(idx)

  1094.         : "r" (s), "r" (d), "m" (mask32b), "m" (mask32r), "m" (mmx_one)

  1095.         : "memory");

  1096.     for (; idx<15; idx+=4) {

  1097.         register int v = *(const uint32_t *)&s[idx], g = v & 0xff00ff00;

  1098.         v &= 0xff00ff;

  1099.         *(uint32_t *)&d[idx] = (v>>16) + g + (v<<16);

  1100.     }

  1101. }

  1102. 

  1103. static inline void RENAME(rgb24tobgr24)(const uint8_t *src, uint8_t *dst, int src_size)

  1104. {

  1105.     unsigned i;

  1106.     x86_reg mmx_size= 23 - src_size;

  1107.     __asm__ volatile (

  1108.         "test             %%"REG_a", %%"REG_a"          \n\t"

  1109.         "jns                     2f                     \n\t"

  1110.         "movq     "MANGLE(mask24r)", %%mm5              \n\t"

  1111.         "movq     "MANGLE(mask24g)", %%mm6              \n\t"

  1112.         "movq     "MANGLE(mask24b)", %%mm7              \n\t"

  1113.         ".p2align                 4                     \n\t"

  1114.         "1:                                             \n\t"

  1115.         PREFETCH" 32(%1, %%"REG_a")                     \n\t"

  1116.         "movq       (%1, %%"REG_a"), %%mm0              \n\t" // BGR BGR BG

  1117.         "movq       (%1, %%"REG_a"), %%mm1              \n\t" // BGR BGR BG

  1118.         "movq      2(%1, %%"REG_a"), %%mm2              \n\t" // R BGR BGR B

  1119.         "psllq                  $16, %%mm0              \n\t" // 00 BGR BGR

  1120.         "pand                 %%mm5, %%mm0              \n\t"

  1121.         "pand                 %%mm6, %%mm1              \n\t"

  1122.         "pand                 %%mm7, %%mm2              \n\t"

  1123.         "por                  %%mm0, %%mm1              \n\t"

  1124.         "por                  %%mm2, %%mm1              \n\t"

  1125.         "movq      6(%1, %%"REG_a"), %%mm0              \n\t" // BGR BGR BG

  1126.         MOVNTQ"               %%mm1,   (%2, %%"REG_a")  \n\t" // RGB RGB RG

  1127.         "movq      8(%1, %%"REG_a"), %%mm1              \n\t" // R BGR BGR B

  1128.         "movq     10(%1, %%"REG_a"), %%mm2              \n\t" // GR BGR BGR

  1129.         "pand                 %%mm7, %%mm0              \n\t"

  1130.         "pand                 %%mm5, %%mm1              \n\t"

  1131.         "pand                 %%mm6, %%mm2              \n\t"

  1132.         "por                  %%mm0, %%mm1              \n\t"

  1133.         "por                  %%mm2, %%mm1              \n\t"

  1134.         "movq     14(%1, %%"REG_a"), %%mm0              \n\t" // R BGR BGR B

  1135.         MOVNTQ"               %%mm1,  8(%2, %%"REG_a")  \n\t" // B RGB RGB R

  1136.         "movq     16(%1, %%"REG_a"), %%mm1              \n\t" // GR BGR BGR

  1137.         "movq     18(%1, %%"REG_a"), %%mm2              \n\t" // BGR BGR BG

  1138.         "pand                 %%mm6, %%mm0              \n\t"

  1139.         "pand                 %%mm7, %%mm1              \n\t"

  1140.         "pand                 %%mm5, %%mm2              \n\t"

  1141.         "por                  %%mm0, %%mm1              \n\t"

  1142.         "por                  %%mm2, %%mm1              \n\t"

  1143.         MOVNTQ"               %%mm1, 16(%2, %%"REG_a")  \n\t"

  1144.         "add                    $24, %%"REG_a"          \n\t"

  1145.         " js                     1b                     \n\t"

  1146.         "2:                                             \n\t"

  1147.         : "+a" (mmx_size)

  1148.         : "r" (src-mmx_size), "r"(dst-mmx_size)

  1149.     );

  1150. 

  1151.     __asm__ volatile(SFENCE:::"memory");

  1152.     __asm__ volatile(EMMS:::"memory");

  1153. 

  1154.     if (mmx_size==23) return; //finished, was multiple of 8

  1155. 

  1156.     src+= src_size;

  1157.     dst+= src_size;

  1158.     src_size= 23-mmx_size;

  1159.     src-= src_size;

  1160.     dst-= src_size;

  1161.     for (i=0; i<src_size; i+=3) {

  1162.         register uint8_t x;

  1163.         x          = src[i + 2];

  1164.         dst[i + 1] = src[i + 1];

  1165.         dst[i + 2] = src[i + 0];

  1166.         dst[i + 0] = x;

  1167.     }

  1168. }

  1169. 

  1170. static inline void RENAME(yuvPlanartoyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,

  1171.                                            int width, int height,

  1172.                                            int lumStride, int chromStride, int dstStride, int vertLumPerChroma)

  1173. {

  1174.     int y;

  1175.     const x86_reg chromWidth= width>>1;

  1176.     for (y=0; y<height; y++) {

  1177.         //FIXME handle 2 lines at once (fewer prefetches, reuse some chroma, but very likely memory-limited anyway)

  1178.         __asm__ volatile(

  1179.             "xor                 %%"REG_a", %%"REG_a"   \n\t"

  1180.             ".p2align                    4              \n\t"

  1181.             "1:                                         \n\t"

  1182.             PREFETCH"    32(%1, %%"REG_a", 2)           \n\t"

  1183.             PREFETCH"    32(%2, %%"REG_a")              \n\t"

  1184.             PREFETCH"    32(%3, %%"REG_a")              \n\t"

  1185.             "movq          (%2, %%"REG_a"), %%mm0       \n\t" // U(0)

  1186.             "movq                    %%mm0, %%mm2       \n\t" // U(0)

  1187.             "movq          (%3, %%"REG_a"), %%mm1       \n\t" // V(0)

  1188.             "punpcklbw               %%mm1, %%mm0       \n\t" // UVUV UVUV(0)

  1189.             "punpckhbw               %%mm1, %%mm2       \n\t" // UVUV UVUV(8)

  1190. 

  1191.             "movq        (%1, %%"REG_a",2), %%mm3       \n\t" // Y(0)

  1192.             "movq       8(%1, %%"REG_a",2), %%mm5       \n\t" // Y(8)

  1193.             "movq                    %%mm3, %%mm4       \n\t" // Y(0)

  1194.             "movq                    %%mm5, %%mm6       \n\t" // Y(8)

  1195.             "punpcklbw               %%mm0, %%mm3       \n\t" // YUYV YUYV(0)

  1196.             "punpckhbw               %%mm0, %%mm4       \n\t" // YUYV YUYV(4)

  1197.             "punpcklbw               %%mm2, %%mm5       \n\t" // YUYV YUYV(8)

  1198.             "punpckhbw               %%mm2, %%mm6       \n\t" // YUYV YUYV(12)

  1199. 

  1200.             MOVNTQ"                  %%mm3,   (%0, %%"REG_a", 4)    \n\t"

  1201.             MOVNTQ"                  %%mm4,  8(%0, %%"REG_a", 4)    \n\t"

  1202.             MOVNTQ"                  %%mm5, 16(%0, %%"REG_a", 4)    \n\t"

  1203.             MOVNTQ"                  %%mm6, 24(%0, %%"REG_a", 4)    \n\t"

  1204. 

  1205.             "add                        $8, %%"REG_a"   \n\t"

  1206.             "cmp                        %4, %%"REG_a"   \n\t"

  1207.             " jb                        1b              \n\t"

  1208.             ::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "g" (chromWidth)

  1209.             : "%"REG_a

  1210.         );

  1211.         if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) {

  1212.             usrc += chromStride;

  1213.             vsrc += chromStride;

  1214.         }

  1215.         ysrc += lumStride;

  1216.         dst  += dstStride;

  1217.     }

  1218.     __asm__(EMMS"       \n\t"

  1219.             SFENCE"     \n\t"

  1220.             :::"memory");

  1221. }

  1222. 

  1223. /**

  1224.  * Height should be a multiple of 2 and width should be a multiple of 16.

  1225.  * (If this is a problem for anyone then tell me, and I will fix it.)

  1226.  */

  1227. static inline void RENAME(yv12toyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,

  1228.                                       int width, int height,

  1229.                                       int lumStride, int chromStride, int dstStride)

  1230. {

  1231.     //FIXME interpolate chroma

  1232.     RENAME(yuvPlanartoyuy2)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 2);

  1233. }

  1234. 

  1235. static inline void RENAME(yuvPlanartouyvy)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,

  1236.                                            int width, int height,

  1237.                                            int lumStride, int chromStride, int dstStride, int vertLumPerChroma)

  1238. {

  1239.     int y;

  1240.     const x86_reg chromWidth= width>>1;

  1241.     for (y=0; y<height; y++) {

  1242.         //FIXME handle 2 lines at once (fewer prefetches, reuse some chroma, but very likely memory-limited anyway)

  1243.         __asm__ volatile(

  1244.             "xor                %%"REG_a", %%"REG_a"    \n\t"

  1245.             ".p2align                   4               \n\t"

  1246.             "1:                                         \n\t"

  1247.             PREFETCH"   32(%1, %%"REG_a", 2)            \n\t"

  1248.             PREFETCH"   32(%2, %%"REG_a")               \n\t"

  1249.             PREFETCH"   32(%3, %%"REG_a")               \n\t"

  1250.             "movq         (%2, %%"REG_a"), %%mm0        \n\t" // U(0)

  1251.             "movq                   %%mm0, %%mm2        \n\t" // U(0)

  1252.             "movq         (%3, %%"REG_a"), %%mm1        \n\t" // V(0)

  1253.             "punpcklbw              %%mm1, %%mm0        \n\t" // UVUV UVUV(0)

  1254.             "punpckhbw              %%mm1, %%mm2        \n\t" // UVUV UVUV(8)

  1255. 

  1256.             "movq       (%1, %%"REG_a",2), %%mm3        \n\t" // Y(0)

  1257.             "movq      8(%1, %%"REG_a",2), %%mm5        \n\t" // Y(8)

  1258.             "movq                   %%mm0, %%mm4        \n\t" // Y(0)

  1259.             "movq                   %%mm2, %%mm6        \n\t" // Y(8)

  1260.             "punpcklbw              %%mm3, %%mm0        \n\t" // YUYV YUYV(0)

  1261.             "punpckhbw              %%mm3, %%mm4        \n\t" // YUYV YUYV(4)

  1262.             "punpcklbw              %%mm5, %%mm2        \n\t" // YUYV YUYV(8)

  1263.             "punpckhbw              %%mm5, %%mm6        \n\t" // YUYV YUYV(12)

  1264. 

  1265.             MOVNTQ"                 %%mm0,   (%0, %%"REG_a", 4)     \n\t"

  1266.             MOVNTQ"                 %%mm4,  8(%0, %%"REG_a", 4)     \n\t"

  1267.             MOVNTQ"                 %%mm2, 16(%0, %%"REG_a", 4)     \n\t"

  1268.             MOVNTQ"                 %%mm6, 24(%0, %%"REG_a", 4)     \n\t"

  1269. 

  1270.             "add                       $8, %%"REG_a"    \n\t"

  1271.             "cmp                       %4, %%"REG_a"    \n\t"

  1272.             " jb                       1b               \n\t"

  1273.             ::"r"(dst), "r"(ysrc), "r"(usrc), "r"(vsrc), "g" (chromWidth)

  1274.             : "%"REG_a

  1275.         );

  1276.         if ((y&(vertLumPerChroma-1)) == vertLumPerChroma-1) {

  1277.             usrc += chromStride;

  1278.             vsrc += chromStride;

  1279.         }

  1280.         ysrc += lumStride;

  1281.         dst += dstStride;

  1282.     }

  1283.     __asm__(EMMS"       \n\t"

  1284.             SFENCE"     \n\t"

  1285.             :::"memory");

  1286. }

  1287. 

  1288. /**

  1289.  * Height should be a multiple of 2 and width should be a multiple of 16

  1290.  * (If this is a problem for anyone then tell me, and I will fix it.)

  1291.  */

  1292. static inline void RENAME(yv12touyvy)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,

  1293.                                       int width, int height,

  1294.                                       int lumStride, int chromStride, int dstStride)

  1295. {

  1296.     //FIXME interpolate chroma

  1297.     RENAME(yuvPlanartouyvy)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 2);

  1298. }

  1299. 

  1300. /**

  1301.  * Width should be a multiple of 16.

  1302.  */

  1303. static inline void RENAME(yuv422ptouyvy)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,

  1304.                                          int width, int height,

  1305.                                          int lumStride, int chromStride, int dstStride)

  1306. {

  1307.     RENAME(yuvPlanartouyvy)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 1);

  1308. }

  1309. 

  1310. /**

  1311.  * Width should be a multiple of 16.

  1312.  */

  1313. static inline void RENAME(yuv422ptoyuy2)(const uint8_t *ysrc, const uint8_t *usrc, const uint8_t *vsrc, uint8_t *dst,

  1314.                                          int width, int height,

  1315.                                          int lumStride, int chromStride, int dstStride)

  1316. {

  1317.     RENAME(yuvPlanartoyuy2)(ysrc, usrc, vsrc, dst, width, height, lumStride, chromStride, dstStride, 1);

  1318. }

  1319. 

  1320. /**

  1321.  * Height should be a multiple of 2 and width should be a multiple of 16.

  1322.  * (If this is a problem for anyone then tell me, and I will fix it.)

  1323.  */

  1324. static inline void RENAME(yuy2toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,

  1325.                                       int width, int height,

  1326.                                       int lumStride, int chromStride, int srcStride)

  1327. {

  1328.     int y;

  1329.     const x86_reg chromWidth= width>>1;

  1330.     for (y=0; y<height; y+=2) {

  1331.         __asm__ volatile(

  1332.             "xor                 %%"REG_a", %%"REG_a"   \n\t"

  1333.             "pcmpeqw                 %%mm7, %%mm7       \n\t"

  1334.             "psrlw                      $8, %%mm7       \n\t" // FF,00,FF,00...

  1335.             ".p2align                    4              \n\t"

  1336.             "1:                \n\t"

  1337.             PREFETCH" 64(%0, %%"REG_a", 4)              \n\t"

  1338.             "movq       (%0, %%"REG_a", 4), %%mm0       \n\t" // YUYV YUYV(0)

  1339.             "movq      8(%0, %%"REG_a", 4), %%mm1       \n\t" // YUYV YUYV(4)

  1340.             "movq                    %%mm0, %%mm2       \n\t" // YUYV YUYV(0)

  1341.             "movq                    %%mm1, %%mm3       \n\t" // YUYV YUYV(4)

  1342.             "psrlw                      $8, %%mm0       \n\t" // U0V0 U0V0(0)

  1343.             "psrlw                      $8, %%mm1       \n\t" // U0V0 U0V0(4)

  1344.             "pand                    %%mm7, %%mm2       \n\t" // Y0Y0 Y0Y0(0)

  1345.             "pand                    %%mm7, %%mm3       \n\t" // Y0Y0 Y0Y0(4)

  1346.             "packuswb                %%mm1, %%mm0       \n\t" // UVUV UVUV(0)

  1347.             "packuswb                %%mm3, %%mm2       \n\t" // YYYY YYYY(0)

  1348. 

  1349.             MOVNTQ"                  %%mm2, (%1, %%"REG_a", 2)  \n\t"

  1350. 

  1351.             "movq     16(%0, %%"REG_a", 4), %%mm1       \n\t" // YUYV YUYV(8)

  1352.             "movq     24(%0, %%"REG_a", 4), %%mm2       \n\t" // YUYV YUYV(12)

  1353.             "movq                    %%mm1, %%mm3       \n\t" // YUYV YUYV(8)

  1354.             "movq                    %%mm2, %%mm4       \n\t" // YUYV YUYV(12)

  1355.             "psrlw                      $8, %%mm1       \n\t" // U0V0 U0V0(8)

  1356.             "psrlw                      $8, %%mm2       \n\t" // U0V0 U0V0(12)

  1357.             "pand                    %%mm7, %%mm3       \n\t" // Y0Y0 Y0Y0(8)

  1358.             "pand                    %%mm7, %%mm4       \n\t" // Y0Y0 Y0Y0(12)

  1359.             "packuswb                %%mm2, %%mm1       \n\t" // UVUV UVUV(8)

  1360.             "packuswb                %%mm4, %%mm3       \n\t" // YYYY YYYY(8)

  1361. 

  1362.             MOVNTQ"                  %%mm3, 8(%1, %%"REG_a", 2) \n\t"

  1363. 

  1364.             "movq                    %%mm0, %%mm2       \n\t" // UVUV UVUV(0)

  1365.             "movq                    %%mm1, %%mm3       \n\t" // UVUV UVUV(8)

  1366.             "psrlw                      $8, %%mm0       \n\t" // V0V0 V0V0(0)

  1367.             "psrlw                      $8, %%mm1       \n\t" // V0V0 V0V0(8)

  1368.             "pand                    %%mm7, %%mm2       \n\t" // U0U0 U0U0(0)

  1369.             "pand                    %%mm7, %%mm3       \n\t" // U0U0 U0U0(8)

  1370.             "packuswb                %%mm1, %%mm0       \n\t" // VVVV VVVV(0)

  1371.             "packuswb                %%mm3, %%mm2       \n\t" // UUUU UUUU(0)

  1372. 

  1373.             MOVNTQ"                  %%mm0, (%3, %%"REG_a")     \n\t"

  1374.             MOVNTQ"                  %%mm2, (%2, %%"REG_a")     \n\t"

  1375. 

  1376.             "add                        $8, %%"REG_a"   \n\t"

  1377.             "cmp                        %4, %%"REG_a"   \n\t"

  1378.             " jb                        1b              \n\t"

  1379.             ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)

  1380.             : "memory", "%"REG_a

  1381.         );

  1382. 

  1383.         ydst += lumStride;

  1384.         src  += srcStride;

  1385. 

  1386.         __asm__ volatile(

  1387.             "xor                 %%"REG_a", %%"REG_a"   \n\t"

  1388.             ".p2align                    4              \n\t"

  1389.             "1:                                         \n\t"

  1390.             PREFETCH" 64(%0, %%"REG_a", 4)              \n\t"

  1391.             "movq       (%0, %%"REG_a", 4), %%mm0       \n\t" // YUYV YUYV(0)

  1392.             "movq      8(%0, %%"REG_a", 4), %%mm1       \n\t" // YUYV YUYV(4)

  1393.             "movq     16(%0, %%"REG_a", 4), %%mm2       \n\t" // YUYV YUYV(8)

  1394.             "movq     24(%0, %%"REG_a", 4), %%mm3       \n\t" // YUYV YUYV(12)

  1395.             "pand                    %%mm7, %%mm0       \n\t" // Y0Y0 Y0Y0(0)

  1396.             "pand                    %%mm7, %%mm1       \n\t" // Y0Y0 Y0Y0(4)

  1397.             "pand                    %%mm7, %%mm2       \n\t" // Y0Y0 Y0Y0(8)

  1398.             "pand                    %%mm7, %%mm3       \n\t" // Y0Y0 Y0Y0(12)

  1399.             "packuswb                %%mm1, %%mm0       \n\t" // YYYY YYYY(0)

  1400.             "packuswb                %%mm3, %%mm2       \n\t" // YYYY YYYY(8)

  1401. 

  1402.             MOVNTQ"                  %%mm0,  (%1, %%"REG_a", 2) \n\t"

  1403.             MOVNTQ"                  %%mm2, 8(%1, %%"REG_a", 2) \n\t"

  1404. 

  1405.             "add                        $8, %%"REG_a"   \n\t"

  1406.             "cmp                        %4, %%"REG_a"   \n\t"

  1407.             " jb                        1b              \n\t"

  1408. 

  1409.             ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)

  1410.             : "memory", "%"REG_a

  1411.         );

  1412.         udst += chromStride;

  1413.         vdst += chromStride;

  1414.         ydst += lumStride;

  1415.         src  += srcStride;

  1416.     }

  1417.     __asm__ volatile(EMMS"       \n\t"

  1418.                      SFENCE"     \n\t"

  1419.                      :::"memory");

  1420. }

  1421. #endif /* !COMPILE_TEMPLATE_AMD3DNOW */

  1422. 

  1423. #if COMPILE_TEMPLATE_MMXEXT || COMPILE_TEMPLATE_AMD3DNOW

  1424. static inline void RENAME(planar2x)(const uint8_t *src, uint8_t *dst, int srcWidth, int srcHeight, int srcStride, int dstStride)

  1425. {

  1426.     int x,y;

  1427. 

  1428.     dst[0]= src[0];

  1429. 

  1430.     // first line

  1431.     for (x=0; x<srcWidth-1; x++) {

  1432.         dst[2*x+1]= (3*src[x] +   src[x+1])>>2;

  1433.         dst[2*x+2]= (  src[x] + 3*src[x+1])>>2;

  1434.     }

  1435.     dst[2*srcWidth-1]= src[srcWidth-1];

  1436. 

  1437.     dst+= dstStride;

  1438. 

  1439.     for (y=1; y<srcHeight; y++) {

  1440.         const x86_reg mmxSize= srcWidth&~15;

  1441.         __asm__ volatile(

  1442.             "mov           %4, %%"REG_a"            \n\t"

  1443.             "movq        "MANGLE(mmx_ff)", %%mm0    \n\t"

  1444.             "movq         (%0, %%"REG_a"), %%mm4    \n\t"

  1445.             "movq                   %%mm4, %%mm2    \n\t"

  1446.             "psllq                     $8, %%mm4    \n\t"

  1447.             "pand                   %%mm0, %%mm2    \n\t"

  1448.             "por                    %%mm2, %%mm4    \n\t"

  1449.             "movq         (%1, %%"REG_a"), %%mm5    \n\t"

  1450.             "movq                   %%mm5, %%mm3    \n\t"

  1451.             "psllq                     $8, %%mm5    \n\t"

  1452.             "pand                   %%mm0, %%mm3    \n\t"

  1453.             "por                    %%mm3, %%mm5    \n\t"

  1454.             "1:                                     \n\t"

  1455.             "movq         (%0, %%"REG_a"), %%mm0    \n\t"

  1456.             "movq         (%1, %%"REG_a"), %%mm1    \n\t"

  1457.             "movq        1(%0, %%"REG_a"), %%mm2    \n\t"

  1458.             "movq        1(%1, %%"REG_a"), %%mm3    \n\t"

  1459.             PAVGB"                  %%mm0, %%mm5    \n\t"

  1460.             PAVGB"                  %%mm0, %%mm3    \n\t"

  1461.             PAVGB"                  %%mm0, %%mm5    \n\t"

  1462.             PAVGB"                  %%mm0, %%mm3    \n\t"

  1463.             PAVGB"                  %%mm1, %%mm4    \n\t"

  1464.             PAVGB"                  %%mm1, %%mm2    \n\t"

  1465.             PAVGB"                  %%mm1, %%mm4    \n\t"

  1466.             PAVGB"                  %%mm1, %%mm2    \n\t"

  1467.             "movq                   %%mm5, %%mm7    \n\t"

  1468.             "movq                   %%mm4, %%mm6    \n\t"

  1469.             "punpcklbw              %%mm3, %%mm5    \n\t"

  1470.             "punpckhbw              %%mm3, %%mm7    \n\t"

  1471.             "punpcklbw              %%mm2, %%mm4    \n\t"

  1472.             "punpckhbw              %%mm2, %%mm6    \n\t"

  1473.             MOVNTQ"                 %%mm5,  (%2, %%"REG_a", 2)  \n\t"

  1474.             MOVNTQ"                 %%mm7, 8(%2, %%"REG_a", 2)  \n\t"

  1475.             MOVNTQ"                 %%mm4,  (%3, %%"REG_a", 2)  \n\t"

  1476.             MOVNTQ"                 %%mm6, 8(%3, %%"REG_a", 2)  \n\t"

  1477.             "add                       $8, %%"REG_a"            \n\t"

  1478.             "movq       -1(%0, %%"REG_a"), %%mm4    \n\t"

  1479.             "movq       -1(%1, %%"REG_a"), %%mm5    \n\t"

  1480.             " js                       1b                       \n\t"

  1481.             :: "r" (src + mmxSize  ), "r" (src + srcStride + mmxSize  ),

  1482.                "r" (dst + mmxSize*2), "r" (dst + dstStride + mmxSize*2),

  1483.                "g" (-mmxSize)

  1484.             : "%"REG_a

  1485.         );

  1486. 

  1487.         for (x=mmxSize-1; x<srcWidth-1; x++) {

  1488.             dst[2*x          +1]= (3*src[x+0] +   src[x+srcStride+1])>>2;

  1489.             dst[2*x+dstStride+2]= (  src[x+0] + 3*src[x+srcStride+1])>>2;

  1490.             dst[2*x+dstStride+1]= (  src[x+1] + 3*src[x+srcStride  ])>>2;

  1491.             dst[2*x          +2]= (3*src[x+1] +   src[x+srcStride  ])>>2;

  1492.         }

  1493.         dst[srcWidth*2 -1            ]= (3*src[srcWidth-1] +   src[srcWidth-1 + srcStride])>>2;

  1494.         dst[srcWidth*2 -1 + dstStride]= (  src[srcWidth-1] + 3*src[srcWidth-1 + srcStride])>>2;

  1495. 

  1496.         dst+=dstStride*2;

  1497.         src+=srcStride;

  1498.     }

  1499. 

  1500.     // last line

  1501.     dst[0]= src[0];

  1502. 

  1503.     for (x=0; x<srcWidth-1; x++) {

  1504.         dst[2*x+1]= (3*src[x] +   src[x+1])>>2;

  1505.         dst[2*x+2]= (  src[x] + 3*src[x+1])>>2;

  1506.     }

  1507.     dst[2*srcWidth-1]= src[srcWidth-1];

  1508. 

  1509.     __asm__ volatile(EMMS"       \n\t"

  1510.                      SFENCE"     \n\t"

  1511.                      :::"memory");

  1512. }

  1513. #endif /* COMPILE_TEMPLATE_MMXEXT || COMPILE_TEMPLATE_AMD3DNOW */

  1514. 

  1515. #if !COMPILE_TEMPLATE_AMD3DNOW

  1516. /**

  1517.  * Height should be a multiple of 2 and width should be a multiple of 16.

  1518.  * (If this is a problem for anyone then tell me, and I will fix it.)

  1519.  * Chrominance data is only taken from every second line, others are ignored.

  1520.  * FIXME: Write HQ version.

  1521.  */

  1522. static inline void RENAME(uyvytoyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,

  1523.                                       int width, int height,

  1524.                                       int lumStride, int chromStride, int srcStride)

  1525. {

  1526.     int y;

  1527.     const x86_reg chromWidth= width>>1;

  1528.     for (y=0; y<height; y+=2) {

  1529.         __asm__ volatile(

  1530.             "xor                 %%"REG_a", %%"REG_a"   \n\t"

  1531.             "pcmpeqw             %%mm7, %%mm7   \n\t"

  1532.             "psrlw                  $8, %%mm7   \n\t" // FF,00,FF,00...

  1533.             ".p2align                4          \n\t"

  1534.             "1:                                 \n\t"

  1535.             PREFETCH" 64(%0, %%"REG_a", 4)          \n\t"

  1536.             "movq       (%0, %%"REG_a", 4), %%mm0   \n\t" // UYVY UYVY(0)

  1537.             "movq      8(%0, %%"REG_a", 4), %%mm1   \n\t" // UYVY UYVY(4)

  1538.             "movq                %%mm0, %%mm2   \n\t" // UYVY UYVY(0)

  1539.             "movq                %%mm1, %%mm3   \n\t" // UYVY UYVY(4)

  1540.             "pand                %%mm7, %%mm0   \n\t" // U0V0 U0V0(0)

  1541.             "pand                %%mm7, %%mm1   \n\t" // U0V0 U0V0(4)

  1542.             "psrlw                  $8, %%mm2   \n\t" // Y0Y0 Y0Y0(0)

  1543.             "psrlw                  $8, %%mm3   \n\t" // Y0Y0 Y0Y0(4)

  1544.             "packuswb            %%mm1, %%mm0   \n\t" // UVUV UVUV(0)

  1545.             "packuswb            %%mm3, %%mm2   \n\t" // YYYY YYYY(0)

  1546. 

  1547.             MOVNTQ"              %%mm2,  (%1, %%"REG_a", 2) \n\t"

  1548. 

  1549.             "movq     16(%0, %%"REG_a", 4), %%mm1   \n\t" // UYVY UYVY(8)

  1550.             "movq     24(%0, %%"REG_a", 4), %%mm2   \n\t" // UYVY UYVY(12)

  1551.             "movq                %%mm1, %%mm3   \n\t" // UYVY UYVY(8)

  1552.             "movq                %%mm2, %%mm4   \n\t" // UYVY UYVY(12)

  1553.             "pand                %%mm7, %%mm1   \n\t" // U0V0 U0V0(8)

  1554.             "pand                %%mm7, %%mm2   \n\t" // U0V0 U0V0(12)

  1555.             "psrlw                  $8, %%mm3   \n\t" // Y0Y0 Y0Y0(8)

  1556.             "psrlw                  $8, %%mm4   \n\t" // Y0Y0 Y0Y0(12)

  1557.             "packuswb            %%mm2, %%mm1   \n\t" // UVUV UVUV(8)

  1558.             "packuswb            %%mm4, %%mm3   \n\t" // YYYY YYYY(8)

  1559. 

  1560.             MOVNTQ"              %%mm3, 8(%1, %%"REG_a", 2) \n\t"

  1561. 

  1562.             "movq                %%mm0, %%mm2   \n\t" // UVUV UVUV(0)

  1563.             "movq                %%mm1, %%mm3   \n\t" // UVUV UVUV(8)

  1564.             "psrlw                  $8, %%mm0   \n\t" // V0V0 V0V0(0)

  1565.             "psrlw                  $8, %%mm1   \n\t" // V0V0 V0V0(8)

  1566.             "pand                %%mm7, %%mm2   \n\t" // U0U0 U0U0(0)

  1567.             "pand                %%mm7, %%mm3   \n\t" // U0U0 U0U0(8)

  1568.             "packuswb            %%mm1, %%mm0   \n\t" // VVVV VVVV(0)

  1569.             "packuswb            %%mm3, %%mm2   \n\t" // UUUU UUUU(0)

  1570. 

  1571.             MOVNTQ"              %%mm0, (%3, %%"REG_a") \n\t"

  1572.             MOVNTQ"              %%mm2, (%2, %%"REG_a") \n\t"

  1573. 

  1574.             "add                    $8, %%"REG_a"   \n\t"

  1575.             "cmp                    %4, %%"REG_a"   \n\t"

  1576.             " jb                    1b          \n\t"

  1577.             ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)

  1578.             : "memory", "%"REG_a

  1579.         );

  1580. 

  1581.         ydst += lumStride;

  1582.         src  += srcStride;

  1583. 

  1584.         __asm__ volatile(

  1585.             "xor                 %%"REG_a", %%"REG_a"   \n\t"

  1586.             ".p2align                    4              \n\t"

  1587.             "1:                                 \n\t"

  1588.             PREFETCH" 64(%0, %%"REG_a", 4)          \n\t"

  1589.             "movq       (%0, %%"REG_a", 4), %%mm0   \n\t" // YUYV YUYV(0)

  1590.             "movq      8(%0, %%"REG_a", 4), %%mm1   \n\t" // YUYV YUYV(4)

  1591.             "movq     16(%0, %%"REG_a", 4), %%mm2   \n\t" // YUYV YUYV(8)

  1592.             "movq     24(%0, %%"REG_a", 4), %%mm3   \n\t" // YUYV YUYV(12)

  1593.             "psrlw                  $8, %%mm0   \n\t" // Y0Y0 Y0Y0(0)

  1594.             "psrlw                  $8, %%mm1   \n\t" // Y0Y0 Y0Y0(4)

  1595.             "psrlw                  $8, %%mm2   \n\t" // Y0Y0 Y0Y0(8)

  1596.             "psrlw                  $8, %%mm3   \n\t" // Y0Y0 Y0Y0(12)

  1597.             "packuswb            %%mm1, %%mm0   \n\t" // YYYY YYYY(0)

  1598.             "packuswb            %%mm3, %%mm2   \n\t" // YYYY YYYY(8)

  1599. 

  1600.             MOVNTQ"              %%mm0,  (%1, %%"REG_a", 2) \n\t"

  1601.             MOVNTQ"              %%mm2, 8(%1, %%"REG_a", 2) \n\t"

  1602. 

  1603.             "add                    $8, %%"REG_a"   \n\t"

  1604.             "cmp                    %4, %%"REG_a"   \n\t"

  1605.             " jb                    1b          \n\t"

  1606. 

  1607.             ::"r"(src), "r"(ydst), "r"(udst), "r"(vdst), "g" (chromWidth)

  1608.             : "memory", "%"REG_a

  1609.         );

  1610.         udst += chromStride;

  1611.         vdst += chromStride;

  1612.         ydst += lumStride;

  1613.         src  += srcStride;

  1614.     }

  1615.     __asm__ volatile(EMMS"       \n\t"

  1616.                      SFENCE"     \n\t"

  1617.                      :::"memory");

  1618. }

  1619. #endif /* !COMPILE_TEMPLATE_AMD3DNOW */

  1620. 

  1621. /**

  1622.  * Height should be a multiple of 2 and width should be a multiple of 2.

  1623.  * (If this is a problem for anyone then tell me, and I will fix it.)

  1624.  * Chrominance data is only taken from every second line,

  1625.  * others are ignored in the C version.

  1626.  * FIXME: Write HQ version.

  1627.  */

  1628. static inline void RENAME(rgb24toyv12)(const uint8_t *src, uint8_t *ydst, uint8_t *udst, uint8_t *vdst,

  1629.                                        int width, int height,

  1630.                                        int lumStride, int chromStride, int srcStride)

  1631. {

  1632.     int y;

  1633.     const x86_reg chromWidth= width>>1;

  1634.     for (y=0; y<height-2; y+=2) {

  1635.         int i;

  1636.         for (i=0; i<2; i++) {

  1637.             __asm__ volatile(

  1638.                 "mov                        %2, %%"REG_a"   \n\t"

  1639.                 "movq  "MANGLE(ff_bgr2YCoeff)", %%mm6       \n\t"

  1640.                 "movq       "MANGLE(ff_w1111)", %%mm5       \n\t"

  1641.                 "pxor                    %%mm7, %%mm7       \n\t"

  1642.                 "lea (%%"REG_a", %%"REG_a", 2), %%"REG_d"   \n\t"

  1643.                 ".p2align                    4              \n\t"

  1644.                 "1:                                         \n\t"

  1645.                 PREFETCH"    64(%0, %%"REG_d")              \n\t"

  1646.                 "movd          (%0, %%"REG_d"), %%mm0       \n\t"

  1647.                 "movd         3(%0, %%"REG_d"), %%mm1       \n\t"

  1648.                 "punpcklbw               %%mm7, %%mm0       \n\t"

  1649.                 "punpcklbw               %%mm7, %%mm1       \n\t"

  1650.                 "movd         6(%0, %%"REG_d"), %%mm2       \n\t"

  1651.                 "movd         9(%0, %%"REG_d"), %%mm3       \n\t"

  1652.                 "punpcklbw               %%mm7, %%mm2       \n\t"

  1653.                 "punpcklbw               %%mm7, %%mm3       \n\t"

  1654.                 "pmaddwd                 %%mm6, %%mm0       \n\t"

  1655.                 "pmaddwd                 %%mm6, %%mm1       \n\t"

  1656.                 "pmaddwd                 %%mm6, %%mm2       \n\t"

  1657.                 "pmaddwd                 %%mm6, %%mm3       \n\t"

  1658. #ifndef FAST_BGR2YV12

  1659.                 "psrad                      $8, %%mm0       \n\t"

  1660.                 "psrad                      $8, %%mm1       \n\t"

  1661.                 "psrad                      $8, %%mm2       \n\t"

  1662.                 "psrad                      $8, %%mm3       \n\t"

  1663. #endif

  1664.                 "packssdw                %%mm1, %%mm0       \n\t"

  1665.                 "packssdw                %%mm3, %%mm2       \n\t"

  1666.                 "pmaddwd                 %%mm5, %%mm0       \n\t"

  1667.                 "pmaddwd                 %%mm5, %%mm2       \n\t"

  1668.                 "packssdw                %%mm2, %%mm0       \n\t"

  1669.                 "psraw                      $7, %%mm0       \n\t"

  1670. 

  1671.                 "movd        12(%0, %%"REG_d"), %%mm4       \n\t"

  1672.                 "movd        15(%0, %%"REG_d"), %%mm1       \n\t"

  1673.                 "punpcklbw               %%mm7, %%mm4       \n\t"

  1674.                 "punpcklbw               %%mm7, %%mm1       \n\t"

  1675.                 "movd        18(%0, %%"REG_d"), %%mm2       \n\t"

  1676.                 "movd        21(%0, %%"REG_d"), %%mm3       \n\t"

  1677.                 "punpcklbw               %%mm7, %%mm2       \n\t"

  1678.                 "punpcklbw               %%mm7, %%mm3       \n\t"

  1679.                 "pmaddwd                 %%mm6, %%mm4       \n\t"

  1680.                 "pmaddwd                 %%mm6, %%mm1       \n\t"

  1681.                 "pmaddwd                 %%mm6, %%mm2       \n\t"

  1682.                 "pmaddwd                 %%mm6, %%mm3       \n\t"

  1683. #ifndef FAST_BGR2YV12

  1684.                 "psrad                      $8, %%mm4       \n\t"

  1685.                 "psrad                      $8, %%mm1       \n\t"

  1686.                 "psrad                      $8, %%mm2       \n\t"

  1687.                 "psrad                      $8, %%mm3       \n\t"

  1688. #endif

  1689.                 "packssdw                %%mm1, %%mm4       \n\t"

  1690.                 "packssdw                %%mm3, %%mm2       \n\t"

  1691.                 "pmaddwd                 %%mm5, %%mm4       \n\t"

  1692.                 "pmaddwd                 %%mm5, %%mm2       \n\t"

  1693.                 "add                       $24, %%"REG_d"   \n\t"

  1694.                 "packssdw                %%mm2, %%mm4       \n\t"

  1695.                 "psraw                      $7, %%mm4       \n\t"

  1696. 

  1697.                 "packuswb                %%mm4, %%mm0       \n\t"

  1698.                 "paddusb "MANGLE(ff_bgr2YOffset)", %%mm0    \n\t"

  1699. 

  1700.                 MOVNTQ"                  %%mm0, (%1, %%"REG_a") \n\t"

  1701.                 "add                        $8,      %%"REG_a"  \n\t"

  1702.                 " js                        1b                  \n\t"

  1703.                 : : "r" (src+width*3), "r" (ydst+width), "g" ((x86_reg)-width)

  1704.                 : "%"REG_a, "%"REG_d

  1705.             );

  1706.             ydst += lumStride;

  1707.             src  += srcStride;

  1708.         }

  1709.         src -= srcStride*2;

  1710.         __asm__ volatile(

  1711.             "mov                        %4, %%"REG_a"   \n\t"

  1712.             "movq       "MANGLE(ff_w1111)", %%mm5       \n\t"

  1713.             "movq  "MANGLE(ff_bgr2UCoeff)", %%mm6       \n\t"

  1714.             "pxor                    %%mm7, %%mm7       \n\t"

  1715.             "lea (%%"REG_a", %%"REG_a", 2), %%"REG_d"   \n\t"

  1716.             "add                 %%"REG_d", %%"REG_d"   \n\t"

  1717.             ".p2align                    4              \n\t"

  1718.             "1:                                         \n\t"

  1719.             PREFETCH"    64(%0, %%"REG_d")              \n\t"

  1720.             PREFETCH"    64(%1, %%"REG_d")              \n\t"

  1721. #if COMPILE_TEMPLATE_MMXEXT || COMPILE_TEMPLATE_AMD3DNOW

  1722.             "movq          (%0, %%"REG_d"), %%mm0       \n\t"

  1723.             "movq          (%1, %%"REG_d"), %%mm1       \n\t"

  1724.             "movq         6(%0, %%"REG_d"), %%mm2       \n\t"

  1725.             "movq         6(%1, %%"REG_d"), %%mm3       \n\t"

  1726.             PAVGB"                   %%mm1, %%mm0       \n\t"

  1727.             PAVGB"                   %%mm3, %%mm2       \n\t"

  1728.             "movq                    %%mm0, %%mm1       \n\t"

  1729.             "movq                    %%mm2, %%mm3       \n\t"

  1730.             "psrlq                     $24, %%mm0       \n\t"

  1731.             "psrlq                     $24, %%mm2       \n\t"

  1732.             PAVGB"                   %%mm1, %%mm0       \n\t"

  1733.             PAVGB"                   %%mm3, %%mm2       \n\t"

  1734.             "punpcklbw               %%mm7, %%mm0       \n\t"

  1735.             "punpcklbw               %%mm7, %%mm2       \n\t"

  1736. #else

  1737.             "movd          (%0, %%"REG_d"), %%mm0       \n\t"

  1738.             "movd          (%1, %%"REG_d"), %%mm1       \n\t"

  1739.             "movd         3(%0, %%"REG_d"), %%mm2       \n\t"

  1740.             "movd         3(%1, %%"REG_d"), %%mm3       \n\t"

  1741.             "punpcklbw               %%mm7, %%mm0       \n\t"

  1742.             "punpcklbw               %%mm7, %%mm1       \n\t"

  1743.             "punpcklbw               %%mm7, %%mm2       \n\t"

  1744.             "punpcklbw               %%mm7, %%mm3       \n\t"

  1745.             "paddw                   %%mm1, %%mm0       \n\t"

  1746.             "paddw                   %%mm3, %%mm2       \n\t"

  1747.             "paddw                   %%mm2, %%mm0       \n\t"

  1748.             "movd         6(%0, %%"REG_d"), %%mm4       \n\t"

  1749.             "movd         6(%1, %%"REG_d"), %%mm1       \n\t"

  1750.             "movd         9(%0, %%"REG_d"), %%mm2       \n\t"

  1751.             "movd         9(%1, %%"REG_d"), %%mm3       \n\t"

  1752.             "punpcklbw               %%mm7, %%mm4       \n\t"

  1753.             "punpcklbw               %%mm7, %%mm1       \n\t"

  1754.             "punpcklbw               %%mm7, %%mm2       \n\t"

  1755.             "punpcklbw               %%mm7, %%mm3       \n\t"

  1756.             "paddw                   %%mm1, %%mm4       \n\t"

  1757.             "paddw                   %%mm3, %%mm2       \n\t"

  1758.             "paddw                   %%mm4, %%mm2       \n\t"

  1759.             "psrlw                      $2, %%mm0       \n\t"

  1760.             "psrlw                      $2, %%mm2       \n\t"

  1761. #endif

  1762.             "movq  "MANGLE(ff_bgr2VCoeff)", %%mm1       \n\t"

  1763.             "movq  "MANGLE(ff_bgr2VCoeff)", %%mm3       \n\t"

  1764. 

  1765.             "pmaddwd                 %%mm0, %%mm1       \n\t"

  1766.             "pmaddwd                 %%mm2, %%mm3       \n\t"

  1767.             "pmaddwd                 %%mm6, %%mm0       \n\t"

  1768.             "pmaddwd                 %%mm6, %%mm2       \n\t"

  1769. #ifndef FAST_BGR2YV12

  1770.             "psrad                      $8, %%mm0       \n\t"

  1771.             "psrad                      $8, %%mm1       \n\t"

  1772.             "psrad                      $8, %%mm2       \n\t"

  1773.             "psrad                      $8, %%mm3       \n\t"

  1774. #endif

  1775.             "packssdw                %%mm2, %%mm0       \n\t"

  1776.             "packssdw                %%mm3, %%mm1       \n\t"

  1777.             "pmaddwd                 %%mm5, %%mm0       \n\t"

  1778.             "pmaddwd                 %%mm5, %%mm1       \n\t"

  1779.             "packssdw                %%mm1, %%mm0       \n\t" // V1 V0 U1 U0

  1780.             "psraw                      $7, %%mm0       \n\t"

  1781. 

  1782. #if COMPILE_TEMPLATE_MMXEXT || COMPILE_TEMPLATE_AMD3DNOW

  1783.             "movq        12(%0, %%"REG_d"), %%mm4       \n\t"

  1784.             "movq        12(%1, %%"REG_d"), %%mm1       \n\t"

  1785.             "movq        18(%0, %%"REG_d"), %%mm2       \n\t"

  1786.             "movq        18(%1, %%"REG_d"), %%mm3       \n\t"

  1787.             PAVGB"                   %%mm1, %%mm4       \n\t"

  1788.             PAVGB"                   %%mm3, %%mm2       \n\t"

  1789.             "movq                    %%mm4, %%mm1       \n\t"

  1790.             "movq                    %%mm2, %%mm3       \n\t"

  1791.             "psrlq                     $24, %%mm4       \n\t"

  1792.             "psrlq                     $24, %%mm2       \n\t"

  1793.             PAVGB"                   %%mm1, %%mm4       \n\t"

  1794.             PAVGB"                   %%mm3, %%mm2       \n\t"

  1795.             "punpcklbw               %%mm7, %%mm4       \n\t"

  1796.             "punpcklbw               %%mm7, %%mm2       \n\t"

  1797. #else

  1798.             "movd        12(%0, %%"REG_d"), %%mm4       \n\t"

  1799.             "movd        12(%1, %%"REG_d"), %%mm1       \n\t"

  1800.             "movd        15(%0, %%"REG_d"), %%mm2       \n\t"

  1801.             "movd        15(%1, %%"REG_d"), %%mm3       \n\t"

  1802.             "punpcklbw               %%mm7, %%mm4       \n\t"

  1803.             "punpcklbw               %%mm7, %%mm1       \n\t"

  1804.             "punpcklbw               %%mm7, %%mm2       \n\t"

  1805.             "punpcklbw               %%mm7, %%mm3       \n\t"

  1806.             "paddw                   %%mm1, %%mm4       \n\t"

  1807.             "paddw                   %%mm3, %%mm2       \n\t"

  1808.             "paddw                   %%mm2, %%mm4       \n\t"

  1809.             "movd        18(%0, %%"REG_d"), %%mm5       \n\t"

  1810.             "movd        18(%1, %%"REG_d"), %%mm1       \n\t"

  1811.             "movd        21(%0, %%"REG_d"), %%mm2       \n\t"

  1812.             "movd        21(%1, %%"REG_d"), %%mm3       \n\t"

  1813.             "punpcklbw               %%mm7, %%mm5       \n\t"

  1814.             "punpcklbw               %%mm7, %%mm1       \n\t"

  1815.             "punpcklbw               %%mm7, %%mm2       \n\t"

  1816.             "punpcklbw               %%mm7, %%mm3       \n\t"

  1817.             "paddw                   %%mm1, %%mm5       \n\t"

  1818.             "paddw                   %%mm3, %%mm2       \n\t"

  1819.             "paddw                   %%mm5, %%mm2       \n\t"

  1820.             "movq       "MANGLE(ff_w1111)", %%mm5       \n\t"

  1821.             "psrlw                      $2, %%mm4       \n\t"

  1822.             "psrlw                      $2, %%mm2       \n\t"

  1823. #endif

  1824.             "movq  "MANGLE(ff_bgr2VCoeff)", %%mm1       \n\t"

  1825.             "movq  "MANGLE(ff_bgr2VCoeff)", %%mm3       \n\t"

  1826. 

  1827.             "pmaddwd                 %%mm4, %%mm1       \n\t"

  1828.             "pmaddwd                 %%mm2, %%mm3       \n\t"

  1829.             "pmaddwd                 %%mm6, %%mm4       \n\t"

  1830.             "pmaddwd                 %%mm6, %%mm2       \n\t"

  1831. #ifndef FAST_BGR2YV12

  1832.             "psrad                      $8, %%mm4       \n\t"

  1833.             "psrad                      $8, %%mm1       \n\t"

  1834.             "psrad                      $8, %%mm2       \n\t"

  1835.             "psrad                      $8, %%mm3       \n\t"

  1836. #endif

  1837.             "packssdw                %%mm2, %%mm4       \n\t"

  1838.             "packssdw                %%mm3, %%mm1       \n\t"

  1839.             "pmaddwd                 %%mm5, %%mm4       \n\t"

  1840.             "pmaddwd                 %%mm5, %%mm1       \n\t"

  1841.             "add                       $24, %%"REG_d"   \n\t"

  1842.             "packssdw                %%mm1, %%mm4       \n\t" // V3 V2 U3 U2

  1843.             "psraw                      $7, %%mm4       \n\t"

  1844. 

  1845.             "movq                    %%mm0, %%mm1           \n\t"

  1846.             "punpckldq               %%mm4, %%mm0           \n\t"

  1847.             "punpckhdq               %%mm4, %%mm1           \n\t"

  1848.             "packsswb                %%mm1, %%mm0           \n\t"

  1849.             "paddb "MANGLE(ff_bgr2UVOffset)", %%mm0         \n\t"

  1850.             "movd                    %%mm0, (%2, %%"REG_a") \n\t"

  1851.             "punpckhdq               %%mm0, %%mm0           \n\t"

  1852.             "movd                    %%mm0, (%3, %%"REG_a") \n\t"

  1853.             "add                        $4, %%"REG_a"       \n\t"

  1854.             " js                        1b                  \n\t"

  1855.             : : "r" (src+chromWidth*6), "r" (src+srcStride+chromWidth*6), "r" (udst+chromWidth), "r" (vdst+chromWidth), "g" (-chromWidth)

  1856.             : "%"REG_a, "%"REG_d

  1857.         );

  1858. 

  1859.         udst += chromStride;

  1860.         vdst += chromStride;

  1861.         src  += srcStride*2;

  1862.     }

  1863. 

  1864.     __asm__ volatile(EMMS"       \n\t"

  1865.                      SFENCE"     \n\t"

  1866.                      :::"memory");

  1867. 

  1868.      rgb24toyv12_c(src, ydst, udst, vdst, width, height-y, lumStride, chromStride, srcStride);

  1869. }

  1870. #endif /* !COMPILE_TEMPLATE_SSE2 */

  1871. 

  1872. #if !COMPILE_TEMPLATE_AMD3DNOW

  1873. static void RENAME(interleaveBytes)(const uint8_t *src1, const uint8_t *src2, uint8_t *dest,

  1874.                                     int width, int height, int src1Stride,

  1875.                                     int src2Stride, int dstStride)

  1876. {

  1877.     int h;

  1878. 

  1879.     for (h=0; h < height; h++) {

  1880.         int w;

  1881. 

  1882. #if COMPILE_TEMPLATE_SSE2

  1883.         __asm__(

  1884.             "xor              %%"REG_a", %%"REG_a"  \n\t"

  1885.             "1:                                     \n\t"

  1886.             PREFETCH" 64(%1, %%"REG_a")             \n\t"

  1887.             PREFETCH" 64(%2, %%"REG_a")             \n\t"

  1888.             "movdqa     (%1, %%"REG_a"), %%xmm0     \n\t"

  1889.             "movdqa     (%1, %%"REG_a"), %%xmm1     \n\t"

  1890.             "movdqa     (%2, %%"REG_a"), %%xmm2     \n\t"

  1891.             "punpcklbw           %%xmm2, %%xmm0     \n\t"

  1892.             "punpckhbw           %%xmm2, %%xmm1     \n\t"

  1893.             "movntdq             %%xmm0,   (%0, %%"REG_a", 2)   \n\t"

  1894.             "movntdq             %%xmm1, 16(%0, %%"REG_a", 2)   \n\t"

  1895.             "add                    $16, %%"REG_a"  \n\t"

  1896.             "cmp                     %3, %%"REG_a"  \n\t"

  1897.             " jb                     1b             \n\t"

  1898.             ::"r"(dest), "r"(src1), "r"(src2), "r" ((x86_reg)width-15)

  1899.             : "memory", "%"REG_a""

  1900.         );

  1901. #else

  1902.         __asm__(

  1903.             "xor %%"REG_a", %%"REG_a"               \n\t"

  1904.             "1:                                     \n\t"

  1905.             PREFETCH" 64(%1, %%"REG_a")             \n\t"

  1906.             PREFETCH" 64(%2, %%"REG_a")             \n\t"

  1907.             "movq       (%1, %%"REG_a"), %%mm0      \n\t"

  1908.             "movq      8(%1, %%"REG_a"), %%mm2      \n\t"

  1909.             "movq                 %%mm0, %%mm1      \n\t"

  1910.             "movq                 %%mm2, %%mm3      \n\t"

  1911.             "movq       (%2, %%"REG_a"), %%mm4      \n\t"

  1912.             "movq      8(%2, %%"REG_a"), %%mm5      \n\t"

  1913.             "punpcklbw            %%mm4, %%mm0      \n\t"

  1914.             "punpckhbw            %%mm4, %%mm1      \n\t"

  1915.             "punpcklbw            %%mm5, %%mm2      \n\t"

  1916.             "punpckhbw            %%mm5, %%mm3      \n\t"

  1917.             MOVNTQ"               %%mm0,   (%0, %%"REG_a", 2)   \n\t"

  1918.             MOVNTQ"               %%mm1,  8(%0, %%"REG_a", 2)   \n\t"

  1919.             MOVNTQ"               %%mm2, 16(%0, %%"REG_a", 2)   \n\t"

  1920.             MOVNTQ"               %%mm3, 24(%0, %%"REG_a", 2)   \n\t"

  1921.             "add                    $16, %%"REG_a"  \n\t"

  1922.             "cmp                     %3, %%"REG_a"  \n\t"

  1923.             " jb                     1b             \n\t"

  1924.             ::"r"(dest), "r"(src1), "r"(src2), "r" ((x86_reg)width-15)

  1925.             : "memory", "%"REG_a

  1926.         );

  1927. #endif

  1928.         for (w= (width&(~15)); w < width; w++) {

  1929.             dest[2*w+0] = src1[w];

  1930.             dest[2*w+1] = src2[w];

  1931.         }

  1932.         dest += dstStride;

  1933.         src1 += src1Stride;

  1934.         src2 += src2Stride;

  1935.     }

  1936.     __asm__(

  1937.             EMMS"       \n\t"

  1938.             SFENCE"     \n\t"

  1939.             ::: "memory"

  1940.             );

  1941. }

  1942. #endif /* !COMPILE_TEMPLATE_AMD3DNOW */

  1943. 

  1944. #if !COMPILE_TEMPLATE_SSE2

  1945. #if !COMPILE_TEMPLATE_AMD3DNOW

  1946. static inline void RENAME(vu9_to_vu12)(const uint8_t *src1, const uint8_t *src2,

  1947.                                        uint8_t *dst1, uint8_t *dst2,

  1948.                                        int width, int height,

  1949.                                        int srcStride1, int srcStride2,

  1950.                                        int dstStride1, int dstStride2)

  1951. {

  1952.     x86_reg x, y;

  1953.     int w,h;

  1954.     w=width/2; h=height/2;

  1955.     __asm__ volatile(

  1956.         PREFETCH" %0    \n\t"

  1957.         PREFETCH" %1    \n\t"

  1958.         ::"m"(*(src1+srcStride1)),"m"(*(src2+srcStride2)):"memory");

  1959.     for (y=0;y<h;y++) {

  1960.         const uint8_t* s1=src1+srcStride1*(y>>1);

  1961.         uint8_t* d=dst1+dstStride1*y;

  1962.         x=0;

  1963.         for (;x<w-31;x+=32) {

  1964.             __asm__ volatile(

  1965.                 PREFETCH"   32(%1,%2)        \n\t"

  1966.                 "movq         (%1,%2), %%mm0 \n\t"

  1967.                 "movq        8(%1,%2), %%mm2 \n\t"

  1968.                 "movq       16(%1,%2), %%mm4 \n\t"

  1969.                 "movq       24(%1,%2), %%mm6 \n\t"

  1970.                 "movq      %%mm0, %%mm1 \n\t"

  1971.                 "movq      %%mm2, %%mm3 \n\t"

  1972.                 "movq      %%mm4, %%mm5 \n\t"

  1973.                 "movq      %%mm6, %%mm7 \n\t"

  1974.                 "punpcklbw %%mm0, %%mm0 \n\t"

  1975.                 "punpckhbw %%mm1, %%mm1 \n\t"

  1976.                 "punpcklbw %%mm2, %%mm2 \n\t"

  1977.                 "punpckhbw %%mm3, %%mm3 \n\t"

  1978.                 "punpcklbw %%mm4, %%mm4 \n\t"

  1979.                 "punpckhbw %%mm5, %%mm5 \n\t"

  1980.                 "punpcklbw %%mm6, %%mm6 \n\t"

  1981.                 "punpckhbw %%mm7, %%mm7 \n\t"

  1982.                 MOVNTQ"    %%mm0,   (%0,%2,2)  \n\t"

  1983.                 MOVNTQ"    %%mm1,  8(%0,%2,2)  \n\t"

  1984.                 MOVNTQ"    %%mm2, 16(%0,%2,2)  \n\t"

  1985.                 MOVNTQ"    %%mm3, 24(%0,%2,2)  \n\t"

  1986.                 MOVNTQ"    %%mm4, 32(%0,%2,2)  \n\t"

  1987.                 MOVNTQ"    %%mm5, 40(%0,%2,2)  \n\t"

  1988.                 MOVNTQ"    %%mm6, 48(%0,%2,2)  \n\t"

  1989.                 MOVNTQ"    %%mm7, 56(%0,%2,2)"

  1990.                 :: "r"(d), "r"(s1), "r"(x)

  1991.                 :"memory");

  1992.         }

  1993.         for (;x<w;x++) d[2*x]=d[2*x+1]=s1[x];

  1994.     }

  1995.     for (y=0;y<h;y++) {

  1996.         const uint8_t* s2=src2+srcStride2*(y>>1);

  1997.         uint8_t* d=dst2+dstStride2*y;

  1998.         x=0;

  1999.         for (;x<w-31;x+=32) {

  2000.             __asm__ volatile(

  2001.                 PREFETCH"   32(%1,%2)        \n\t"

  2002.                 "movq         (%1,%2), %%mm0 \n\t"

  2003.                 "movq        8(%1,%2), %%mm2 \n\t"

  2004.                 "movq       16(%1,%2), %%mm4 \n\t"

  2005.                 "movq       24(%1,%2), %%mm6 \n\t"

  2006.                 "movq      %%mm0, %%mm1 \n\t"

  2007.                 "movq      %%mm2, %%mm3 \n\t"

  2008.                 "movq      %%mm4, %%mm5 \n\t"

  2009.                 "movq      %%mm6, %%mm7 \n\t"

  2010.                 "punpcklbw %%mm0, %%mm0 \n\t"

  2011.                 "punpckhbw %%mm1, %%mm1 \n\t"

  2012.                 "punpcklbw %%mm2, %%mm2 \n\t"

  2013.                 "punpckhbw %%mm3, %%mm3 \n\t"

  2014.                 "punpcklbw %%mm4, %%mm4 \n\t"

  2015.                 "punpckhbw %%mm5, %%mm5 \n\t"

  2016.                 "punpcklbw %%mm6, %%mm6 \n\t"

  2017.                 "punpckhbw %%mm7, %%mm7 \n\t"

  2018.                 MOVNTQ"    %%mm0,   (%0,%2,2)  \n\t"

  2019.                 MOVNTQ"    %%mm1,  8(%0,%2,2)  \n\t"

  2020.                 MOVNTQ"    %%mm2, 16(%0,%2,2)  \n\t"

  2021.                 MOVNTQ"    %%mm3, 24(%0,%2,2)  \n\t"

  2022.                 MOVNTQ"    %%mm4, 32(%0,%2,2)  \n\t"

  2023.                 MOVNTQ"    %%mm5, 40(%0,%2,2)  \n\t"

  2024.                 MOVNTQ"    %%mm6, 48(%0,%2,2)  \n\t"

  2025.                 MOVNTQ"    %%mm7, 56(%0,%2,2)"

  2026.                 :: "r"(d), "r"(s2), "r"(x)

  2027.                 :"memory");

  2028.         }

  2029.         for (;x<w;x++) d[2*x]=d[2*x+1]=s2[x];

  2030.     }

  2031.     __asm__(

  2032.             EMMS"       \n\t"

  2033.             SFENCE"     \n\t"

  2034.             ::: "memory"

  2035.         );

  2036. }

  2037. 

  2038. static inline void RENAME(yvu9_to_yuy2)(const uint8_t *src1, const uint8_t *src2, const uint8_t *src3,

  2039.                                         uint8_t *dst,

  2040.                                         int width, int height,

  2041.                                         int srcStride1, int srcStride2,

  2042.                                         int srcStride3, int dstStride)

  2043. {

  2044.     x86_reg x;

  2045.     int y,w,h;

  2046.     w=width/2; h=height;

  2047.     for (y=0;y<h;y++) {

  2048.         const uint8_t* yp=src1+srcStride1*y;

  2049.         const uint8_t* up=src2+srcStride2*(y>>2);

  2050.         const uint8_t* vp=src3+srcStride3*(y>>2);

  2051.         uint8_t* d=dst+dstStride*y;

  2052.         x=0;

  2053.         for (;x<w-7;x+=8) {

  2054.             __asm__ volatile(

  2055.                 PREFETCH"   32(%1, %0)          \n\t"

  2056.                 PREFETCH"   32(%2, %0)          \n\t"

  2057.                 PREFETCH"   32(%3, %0)          \n\t"

  2058.                 "movq      (%1, %0, 4), %%mm0   \n\t" /* Y0Y1Y2Y3Y4Y5Y6Y7 */

  2059.                 "movq         (%2, %0), %%mm1   \n\t" /* U0U1U2U3U4U5U6U7 */

  2060.                 "movq         (%3, %0), %%mm2   \n\t" /* V0V1V2V3V4V5V6V7 */

  2061.                 "movq            %%mm0, %%mm3   \n\t" /* Y0Y1Y2Y3Y4Y5Y6Y7 */

  2062.                 "movq            %%mm1, %%mm4   \n\t" /* U0U1U2U3U4U5U6U7 */

  2063.                 "movq            %%mm2, %%mm5   \n\t" /* V0V1V2V3V4V5V6V7 */

  2064.                 "punpcklbw       %%mm1, %%mm1   \n\t" /* U0U0 U1U1 U2U2 U3U3 */

  2065.                 "punpcklbw       %%mm2, %%mm2   \n\t" /* V0V0 V1V1 V2V2 V3V3 */

  2066.                 "punpckhbw       %%mm4, %%mm4   \n\t" /* U4U4 U5U5 U6U6 U7U7 */

  2067.                 "punpckhbw       %%mm5, %%mm5   \n\t" /* V4V4 V5V5 V6V6 V7V7 */

  2068. 

  2069.                 "movq            %%mm1, %%mm6   \n\t"

  2070.                 "punpcklbw       %%mm2, %%mm1   \n\t" /* U0V0 U0V0 U1V1 U1V1*/

  2071.                 "punpcklbw       %%mm1, %%mm0   \n\t" /* Y0U0 Y1V0 Y2U0 Y3V0*/

  2072.                 "punpckhbw       %%mm1, %%mm3   \n\t" /* Y4U1 Y5V1 Y6U1 Y7V1*/

  2073.                 MOVNTQ"          %%mm0,  (%4, %0, 8)    \n\t"

  2074.                 MOVNTQ"          %%mm3, 8(%4, %0, 8)    \n\t"

  2075. 

  2076.                 "punpckhbw       %%mm2, %%mm6   \n\t" /* U2V2 U2V2 U3V3 U3V3*/

  2077.                 "movq     8(%1, %0, 4), %%mm0   \n\t"

  2078.                 "movq            %%mm0, %%mm3   \n\t"

  2079.                 "punpcklbw       %%mm6, %%mm0   \n\t" /* Y U2 Y V2 Y U2 Y V2*/

  2080.                 "punpckhbw       %%mm6, %%mm3   \n\t" /* Y U3 Y V3 Y U3 Y V3*/

  2081.                 MOVNTQ"          %%mm0, 16(%4, %0, 8)   \n\t"

  2082.                 MOVNTQ"          %%mm3, 24(%4, %0, 8)   \n\t"

  2083. 

  2084.                 "movq            %%mm4, %%mm6   \n\t"

  2085.                 "movq    16(%1, %0, 4), %%mm0   \n\t"

  2086.                 "movq            %%mm0, %%mm3   \n\t"

  2087.                 "punpcklbw       %%mm5, %%mm4   \n\t"

  2088.                 "punpcklbw       %%mm4, %%mm0   \n\t" /* Y U4 Y V4 Y U4 Y V4*/

  2089.                 "punpckhbw       %%mm4, %%mm3   \n\t" /* Y U5 Y V5 Y U5 Y V5*/

  2090.                 MOVNTQ"          %%mm0, 32(%4, %0, 8)   \n\t"

  2091.                 MOVNTQ"          %%mm3, 40(%4, %0, 8)   \n\t"

  2092. 

  2093.                 "punpckhbw       %%mm5, %%mm6   \n\t"

  2094.                 "movq    24(%1, %0, 4), %%mm0   \n\t"

  2095.                 "movq            %%mm0, %%mm3   \n\t"

  2096.                 "punpcklbw       %%mm6, %%mm0   \n\t" /* Y U6 Y V6 Y U6 Y V6*/

  2097.                 "punpckhbw       %%mm6, %%mm3   \n\t" /* Y U7 Y V7 Y U7 Y V7*/

  2098.                 MOVNTQ"          %%mm0, 48(%4, %0, 8)   \n\t"

  2099.                 MOVNTQ"          %%mm3, 56(%4, %0, 8)   \n\t"

  2100. 

  2101.                 : "+r" (x)

  2102.                 : "r"(yp), "r" (up), "r"(vp), "r"(d)

  2103.                 :"memory");

  2104.         }

  2105.         for (; x<w; x++) {

  2106.             const int x2 = x<<2;

  2107.             d[8*x+0] = yp[x2];

  2108.             d[8*x+1] = up[x];

  2109.             d[8*x+2] = yp[x2+1];

  2110.             d[8*x+3] = vp[x];

  2111.             d[8*x+4] = yp[x2+2];

  2112.             d[8*x+5] = up[x];

  2113.             d[8*x+6] = yp[x2+3];

  2114.             d[8*x+7] = vp[x];

  2115.         }

  2116.     }

  2117.     __asm__(

  2118.             EMMS"       \n\t"

  2119.             SFENCE"     \n\t"

  2120.             ::: "memory"

  2121.         );

  2122. }

  2123. #endif /* !COMPILE_TEMPLATE_AMD3DNOW */

  2124. 

  2125. static void RENAME(extract_even)(const uint8_t *src, uint8_t *dst, x86_reg count)

  2126. {

  2127.     dst +=   count;

  2128.     src += 2*count;

  2129.     count= - count;

  2130. 

  2131.     if(count <= -16) {

  2132.         count += 15;

  2133.         __asm__ volatile(

  2134.             "pcmpeqw       %%mm7, %%mm7        \n\t"

  2135.             "psrlw            $8, %%mm7        \n\t"

  2136.             "1:                                \n\t"

  2137.             "movq -30(%1, %0, 2), %%mm0        \n\t"

  2138.             "movq -22(%1, %0, 2), %%mm1        \n\t"

  2139.             "movq -14(%1, %0, 2), %%mm2        \n\t"

  2140.             "movq  -6(%1, %0, 2), %%mm3        \n\t"

  2141.             "pand          %%mm7, %%mm0        \n\t"

  2142.             "pand          %%mm7, %%mm1        \n\t"

  2143.             "pand          %%mm7, %%mm2        \n\t"

  2144.             "pand          %%mm7, %%mm3        \n\t"

  2145.             "packuswb      %%mm1, %%mm0        \n\t"

  2146.             "packuswb      %%mm3, %%mm2        \n\t"

  2147.             MOVNTQ"        %%mm0,-15(%2, %0)   \n\t"

  2148.             MOVNTQ"        %%mm2,- 7(%2, %0)   \n\t"

  2149.             "add             $16, %0           \n\t"

  2150.             " js 1b                            \n\t"

  2151.             : "+r"(count)

  2152.             : "r"(src), "r"(dst)

  2153.         );

  2154.         count -= 15;

  2155.     }

  2156.     while(count<0) {

  2157.         dst[count]= src[2*count];

  2158.         count++;

  2159.     }

  2160. }

  2161. 

  2162. #if !COMPILE_TEMPLATE_AMD3DNOW

  2163. static void RENAME(extract_even2)(const uint8_t *src, uint8_t *dst0, uint8_t *dst1, x86_reg count)

  2164. {

  2165.     dst0+=   count;

  2166.     dst1+=   count;

  2167.     src += 4*count;

  2168.     count= - count;

  2169.     if(count <= -8) {

  2170.         count += 7;

  2171.         __asm__ volatile(

  2172.             "pcmpeqw       %%mm7, %%mm7        \n\t"

  2173.             "psrlw            $8, %%mm7        \n\t"

  2174.             "1:                                \n\t"

  2175.             "movq -28(%1, %0, 4), %%mm0        \n\t"

  2176.             "movq -20(%1, %0, 4), %%mm1        \n\t"

  2177.             "movq -12(%1, %0, 4), %%mm2        \n\t"

  2178.             "movq  -4(%1, %0, 4), %%mm3        \n\t"

  2179.             "pand          %%mm7, %%mm0        \n\t"

  2180.             "pand          %%mm7, %%mm1        \n\t"

  2181.             "pand          %%mm7, %%mm2        \n\t"

  2182.             "pand          %%mm7, %%mm3        \n\t"

  2183.             "packuswb      %%mm1, %%mm0        \n\t"

  2184.             "packuswb      %%mm3, %%mm2        \n\t"

  2185.             "movq          %%mm0, %%mm1        \n\t"

  2186.             "movq          %%mm2, %%mm3        \n\t"

  2187.             "psrlw            $8, %%mm0        \n\t"

  2188.             "psrlw            $8, %%mm2        \n\t"

  2189.             "pand          %%mm7, %%mm1        \n\t"

  2190.             "pand          %%mm7, %%mm3        \n\t"

  2191.             "packuswb      %%mm2, %%mm0        \n\t"

  2192.             "packuswb      %%mm3, %%mm1        \n\t"

  2193.             MOVNTQ"        %%mm0,- 7(%3, %0)   \n\t"

  2194.             MOVNTQ"        %%mm1,- 7(%2, %0)   \n\t"

  2195.             "add              $8, %0           \n\t"

  2196.             " js 1b                            \n\t"

  2197.             : "+r"(count)

  2198.             : "r"(src), "r"(dst0), "r"(dst1)

  2199.         );

  2200.         count -= 7;

  2201.     }

  2202.     while(count<0) {

  2203.         dst0[count]= src[4*count+0];

  2204.         dst1[count]= src[4*count+2];

  2205.         count++;

  2206.     }

  2207. }

  2208. #endif /* !COMPILE_TEMPLATE_AMD3DNOW */

  2209. 

  2210. static void RENAME(extract_even2avg)(const uint8_t *src0, const uint8_t *src1, uint8_t *dst0, uint8_t *dst1, x86_reg count)

  2211. {

  2212.     dst0 +=   count;

  2213.     dst1 +=   count;

  2214.     src0 += 4*count;

  2215.     src1 += 4*count;

  2216.     count= - count;

  2217. #ifdef PAVGB

  2218.     if(count <= -8) {

  2219.         count += 7;

  2220.         __asm__ volatile(

  2221.             "pcmpeqw        %%mm7, %%mm7        \n\t"

  2222.             "psrlw             $8, %%mm7        \n\t"

  2223.             "1:                                \n\t"

  2224.             "movq  -28(%1, %0, 4), %%mm0        \n\t"

  2225.             "movq  -20(%1, %0, 4), %%mm1        \n\t"

  2226.             "movq  -12(%1, %0, 4), %%mm2        \n\t"

  2227.             "movq   -4(%1, %0, 4), %%mm3        \n\t"

  2228.             PAVGB" -28(%2, %0, 4), %%mm0        \n\t"

  2229.             PAVGB" -20(%2, %0, 4), %%mm1        \n\t"

  2230.             PAVGB" -12(%2, %0, 4), %%mm2        \n\t"

  2231.             PAVGB" - 4(%2, %0, 4), %%mm3        \n\t"

  2232.             "pand           %%mm7, %%mm0        \n\t"

  2233.             "pand           %%mm7, %%mm1        \n\t"

  2234.             "pand           %%mm7, %%mm2        \n\t"

  2235.             "pand           %%mm7, %%mm3        \n\t"

  2236.             "packuswb       %%mm1, %%mm0        \n\t"

  2237.             "packuswb       %%mm3, %%mm2        \n\t"

  2238.             "movq           %%mm0, %%mm1        \n\t"

  2239.             "movq           %%mm2, %%mm3        \n\t"

  2240.             "psrlw             $8, %%mm0        \n\t"

  2241.             "psrlw             $8, %%mm2        \n\t"

  2242.             "pand           %%mm7, %%mm1        \n\t"

  2243.             "pand           %%mm7, %%mm3        \n\t"

  2244.             "packuswb       %%mm2, %%mm0        \n\t"

  2245.             "packuswb       %%mm3, %%mm1        \n\t"

  2246.             MOVNTQ"         %%mm0,- 7(%4, %0)   \n\t"

  2247.             MOVNTQ"         %%mm1,- 7(%3, %0)   \n\t"

  2248.             "add               $8, %0           \n\t"

  2249.             " js 1b                            \n\t"

  2250.             : "+r"(count)

  2251.             : "r"(src0), "r"(src1), "r"(dst0), "r"(dst1)

  2252.         );

  2253.         count -= 7;

  2254.     }

  2255. #endif

  2256.     while(count<0) {

  2257.         dst0[count]= (src0[4*count+0]+src1[4*count+0])>>1;

  2258.         dst1[count]= (src0[4*count+2]+src1[4*count+2])>>1;

  2259.         count++;

  2260.     }

  2261. }

  2262. 

  2263. #if !COMPILE_TEMPLATE_AMD3DNOW

  2264. static void RENAME(extract_odd2)(const uint8_t *src, uint8_t *dst0, uint8_t *dst1, x86_reg count)

  2265. {

  2266.     dst0+=   count;

  2267.     dst1+=   count;

  2268.     src += 4*count;

  2269.     count= - count;

  2270.     if(count <= -8) {

  2271.         count += 7;

  2272.         __asm__ volatile(

  2273.             "pcmpeqw       %%mm7, %%mm7        \n\t"

  2274.             "psrlw            $8, %%mm7        \n\t"

  2275.             "1:                                \n\t"

  2276.             "movq -28(%1, %0, 4), %%mm0        \n\t"

  2277.             "movq -20(%1, %0, 4), %%mm1        \n\t"

  2278.             "movq -12(%1, %0, 4), %%mm2        \n\t"

  2279.             "movq  -4(%1, %0, 4), %%mm3        \n\t"

  2280.             "psrlw            $8, %%mm0        \n\t"

  2281.             "psrlw            $8, %%mm1        \n\t"

  2282.             "psrlw            $8, %%mm2        \n\t"

  2283.             "psrlw            $8, %%mm3        \n\t"

  2284.             "packuswb      %%mm1, %%mm0        \n\t"

  2285.             "packuswb      %%mm3, %%mm2        \n\t"

  2286.             "movq          %%mm0, %%mm1        \n\t"

  2287.             "movq          %%mm2, %%mm3        \n\t"

  2288.             "psrlw            $8, %%mm0        \n\t"

  2289.             "psrlw            $8, %%mm2        \n\t"

  2290.             "pand          %%mm7, %%mm1        \n\t"

  2291.             "pand          %%mm7, %%mm3        \n\t"

  2292.             "packuswb      %%mm2, %%mm0        \n\t"

  2293.             "packuswb      %%mm3, %%mm1        \n\t"

  2294.             MOVNTQ"        %%mm0,- 7(%3, %0)   \n\t"

  2295.             MOVNTQ"        %%mm1,- 7(%2, %0)   \n\t"

  2296.             "add              $8, %0           \n\t"

  2297.             " js 1b                            \n\t"

  2298.             : "+r"(count)

  2299.             : "r"(src), "r"(dst0), "r"(dst1)

  2300.         );

  2301.         count -= 7;

  2302.     }

  2303.     src++;

  2304.     while(count<0) {

  2305.         dst0[count]= src[4*count+0];

  2306.         dst1[count]= src[4*count+2];

  2307.         count++;

  2308.     }

  2309. }

  2310. #endif /* !COMPILE_TEMPLATE_AMD3DNOW */

  2311. 

  2312. static void RENAME(extract_odd2avg)(const uint8_t *src0, const uint8_t *src1, uint8_t *dst0, uint8_t *dst1, x86_reg count)

  2313. {

  2314.     dst0 +=   count;

  2315.     dst1 +=   count;

  2316.     src0 += 4*count;

  2317.     src1 += 4*count;

  2318.     count= - count;

  2319. #ifdef PAVGB

  2320.     if(count <= -8) {

  2321.         count += 7;

  2322.         __asm__ volatile(

  2323.             "pcmpeqw        %%mm7, %%mm7        \n\t"

  2324.             "psrlw             $8, %%mm7        \n\t"

  2325.             "1:                                \n\t"

  2326.             "movq  -28(%1, %0, 4), %%mm0        \n\t"

  2327.             "movq  -20(%1, %0, 4), %%mm1        \n\t"

  2328.             "movq  -12(%1, %0, 4), %%mm2        \n\t"

  2329.             "movq   -4(%1, %0, 4), %%mm3        \n\t"

  2330.             PAVGB" -28(%2, %0, 4), %%mm0        \n\t"

  2331.             PAVGB" -20(%2, %0, 4), %%mm1        \n\t"

  2332.             PAVGB" -12(%2, %0, 4), %%mm2        \n\t"

  2333.             PAVGB" - 4(%2, %0, 4), %%mm3        \n\t"

  2334.             "psrlw             $8, %%mm0        \n\t"

  2335.             "psrlw             $8, %%mm1        \n\t"

  2336.             "psrlw             $8, %%mm2        \n\t"

  2337.             "psrlw             $8, %%mm3        \n\t"

  2338.             "packuswb       %%mm1, %%mm0        \n\t"

  2339.             "packuswb       %%mm3, %%mm2        \n\t"

  2340.             "movq           %%mm0, %%mm1        \n\t"

  2341.             "movq           %%mm2, %%mm3        \n\t"

  2342.             "psrlw             $8, %%mm0        \n\t"

  2343.             "psrlw             $8, %%mm2        \n\t"

  2344.             "pand           %%mm7, %%mm1        \n\t"

  2345.             "pand           %%mm7, %%mm3        \n\t"

  2346.             "packuswb       %%mm2, %%mm0        \n\t"

  2347.             "packuswb       %%mm3, %%mm1        \n\t"

  2348.             MOVNTQ"         %%mm0,- 7(%4, %0)   \n\t"

  2349.             MOVNTQ"         %%mm1,- 7(%3, %0)   \n\t"

  2350.             "add               $8, %0           \n\t"

  2351.             " js 1b                            \n\t"

  2352.             : "+r"(count)

  2353.             : "r"(src0), "r"(src1), "r"(dst0), "r"(dst1)

  2354.         );

  2355.         count -= 7;

  2356.     }

  2357. #endif

  2358.     src0++;

  2359.     src1++;

  2360.     while(count<0) {

  2361.         dst0[count]= (src0[4*count+0]+src1[4*count+0])>>1;

  2362.         dst1[count]= (src0[4*count+2]+src1[4*count+2])>>1;

  2363.         count++;

  2364.     }

  2365. }

  2366. 

  2367. static void RENAME(yuyvtoyuv420)(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src,

  2368.                                  int width, int height,

  2369.                                  int lumStride, int chromStride, int srcStride)

  2370. {

  2371.     int y;

  2372.     const int chromWidth= -((-width)>>1);

  2373. 

  2374.     for (y=0; y<height; y++) {

  2375.         RENAME(extract_even)(src, ydst, width);

  2376.         if(y&1) {

  2377.             RENAME(extract_odd2avg)(src-srcStride, src, udst, vdst, chromWidth);

  2378.             udst+= chromStride;

  2379.             vdst+= chromStride;

  2380.         }

  2381. 

  2382.         src += srcStride;

  2383.         ydst+= lumStride;

  2384.     }

  2385.     __asm__(

  2386.             EMMS"       \n\t"

  2387.             SFENCE"     \n\t"

  2388.             ::: "memory"

  2389.         );

  2390. }

  2391. 

  2392. #if !COMPILE_TEMPLATE_AMD3DNOW

  2393. static void RENAME(yuyvtoyuv422)(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src,

  2394.                                  int width, int height,

  2395.                                  int lumStride, int chromStride, int srcStride)

  2396. {

  2397.     int y;

  2398.     const int chromWidth= -((-width)>>1);

  2399. 

  2400.     for (y=0; y<height; y++) {

  2401.         RENAME(extract_even)(src, ydst, width);

  2402.         RENAME(extract_odd2)(src, udst, vdst, chromWidth);

  2403. 

  2404.         src += srcStride;

  2405.         ydst+= lumStride;

  2406.         udst+= chromStride;

  2407.         vdst+= chromStride;

  2408.     }

  2409.     __asm__(

  2410.             EMMS"       \n\t"

  2411.             SFENCE"     \n\t"

  2412.             ::: "memory"

  2413.         );

  2414. }

  2415. #endif /* !COMPILE_TEMPLATE_AMD3DNOW */

  2416. 

  2417. static void RENAME(uyvytoyuv420)(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src,

  2418.                                  int width, int height,

  2419.                                  int lumStride, int chromStride, int srcStride)

  2420. {

  2421.     int y;

  2422.     const int chromWidth= -((-width)>>1);

  2423. 

  2424.     for (y=0; y<height; y++) {

  2425.         RENAME(extract_even)(src+1, ydst, width);

  2426.         if(y&1) {

  2427.             RENAME(extract_even2avg)(src-srcStride, src, udst, vdst, chromWidth);

  2428.             udst+= chromStride;

  2429.             vdst+= chromStride;

  2430.         }

  2431. 

  2432.         src += srcStride;

  2433.         ydst+= lumStride;

  2434.     }

  2435.     __asm__(

  2436.             EMMS"       \n\t"

  2437.             SFENCE"     \n\t"

  2438.             ::: "memory"

  2439.         );

  2440. }

  2441. 

  2442. #if !COMPILE_TEMPLATE_AMD3DNOW

  2443. static void RENAME(uyvytoyuv422)(uint8_t *ydst, uint8_t *udst, uint8_t *vdst, const uint8_t *src,

  2444.                                  int width, int height,

  2445.                                  int lumStride, int chromStride, int srcStride)

  2446. {

  2447.     int y;

  2448.     const int chromWidth= -((-width)>>1);

  2449. 

  2450.     for (y=0; y<height; y++) {

  2451.         RENAME(extract_even)(src+1, ydst, width);

  2452.         RENAME(extract_even2)(src, udst, vdst, chromWidth);

  2453. 

  2454.         src += srcStride;

  2455.         ydst+= lumStride;

  2456.         udst+= chromStride;

  2457.         vdst+= chromStride;

  2458.     }

  2459.     __asm__(

  2460.             EMMS"       \n\t"

  2461.             SFENCE"     \n\t"

  2462.             ::: "memory"

  2463.         );

  2464. }

  2465. #endif /* !COMPILE_TEMPLATE_AMD3DNOW */

  2466. #endif /* !COMPILE_TEMPLATE_SSE2 */

  2467. 

  2468. static inline void RENAME(rgb2rgb_init)(void)

  2469. {

  2470. #if !COMPILE_TEMPLATE_SSE2

  2471. #if !COMPILE_TEMPLATE_AMD3DNOW

  2472.     rgb15to16          = RENAME(rgb15to16);

  2473.     rgb15tobgr24       = RENAME(rgb15tobgr24);

  2474.     rgb15to32          = RENAME(rgb15to32);

  2475.     rgb16tobgr24       = RENAME(rgb16tobgr24);

  2476.     rgb16to32          = RENAME(rgb16to32);

  2477.     rgb16to15          = RENAME(rgb16to15);

  2478.     rgb24tobgr16       = RENAME(rgb24tobgr16);

  2479.     rgb24tobgr15       = RENAME(rgb24tobgr15);

  2480.     rgb24tobgr32       = RENAME(rgb24tobgr32);

  2481.     rgb32to16          = RENAME(rgb32to16);

  2482.     rgb32to15          = RENAME(rgb32to15);

  2483.     rgb32tobgr24       = RENAME(rgb32tobgr24);

  2484.     rgb24to15          = RENAME(rgb24to15);

  2485.     rgb24to16          = RENAME(rgb24to16);

  2486.     rgb24tobgr24       = RENAME(rgb24tobgr24);

  2487.     shuffle_bytes_2103 = RENAME(shuffle_bytes_2103);

  2488.     rgb32tobgr16       = RENAME(rgb32tobgr16);

  2489.     rgb32tobgr15       = RENAME(rgb32tobgr15);

  2490.     yv12toyuy2         = RENAME(yv12toyuy2);

  2491.     yv12touyvy         = RENAME(yv12touyvy);

  2492.     yuv422ptoyuy2      = RENAME(yuv422ptoyuy2);

  2493.     yuv422ptouyvy      = RENAME(yuv422ptouyvy);

  2494.     yuy2toyv12         = RENAME(yuy2toyv12);

  2495.     vu9_to_vu12        = RENAME(vu9_to_vu12);

  2496.     yvu9_to_yuy2       = RENAME(yvu9_to_yuy2);

  2497.     uyvytoyuv422       = RENAME(uyvytoyuv422);

  2498.     yuyvtoyuv422       = RENAME(yuyvtoyuv422);

  2499. #endif /* !COMPILE_TEMPLATE_AMD3DNOW */

  2500. 

  2501. #if COMPILE_TEMPLATE_MMXEXT || COMPILE_TEMPLATE_AMD3DNOW

  2502.     planar2x           = RENAME(planar2x);

  2503. #endif /* COMPILE_TEMPLATE_MMXEXT || COMPILE_TEMPLATE_AMD3DNOW */

  2504.     rgb24toyv12        = RENAME(rgb24toyv12);

  2505. 

  2506.     yuyvtoyuv420       = RENAME(yuyvtoyuv420);

  2507.     uyvytoyuv420       = RENAME(uyvytoyuv420);

  2508. #endif /* !COMPILE_TEMPLATE_SSE2 */

  2509. 

  2510. #if !COMPILE_TEMPLATE_AMD3DNOW

  2511.     interleaveBytes    = RENAME(interleaveBytes);

  2512. #endif /* !COMPILE_TEMPLATE_AMD3DNOW */

  2513. }