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.
23232 Commits
32 Branches
593MB
Tree: 88b51ea948
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '88b51ea948'
${ noResults }
FFmpeg/libavcodec/x86/snowdsp_mmx.c

873 lines
39KB
Raw Blame History 

  1. /*

  2.  * MMX and SSE2 optimized snow DSP utils

  3.  * Copyright (c) 2005-2006 Robert Edele <yartrebo@earthlink.net>

  4.  *

  5.  * This file is part of FFmpeg.

  6.  *

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

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

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

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

  11.  *

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

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

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

  15.  * Lesser General Public License for more details.

  16.  *

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

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

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

  20.  */

  21. 

  22. #include "libavutil/x86_cpu.h"

  23. #include "libavcodec/avcodec.h"

  24. #include "libavcodec/snow.h"

  25. #include "dsputil_mmx.h"

  26. 

  27. void ff_snow_horizontal_compose97i_sse2(IDWTELEM *b, int width){

  28.     const int w2= (width+1)>>1;

  29.     DECLARE_ALIGNED(16, IDWTELEM, temp)[width>>1];

  30.     const int w_l= (width>>1);

  31.     const int w_r= w2 - 1;

  32.     int i;

  33. 

  34.     { // Lift 0

  35.         IDWTELEM * const ref = b + w2 - 1;

  36.         IDWTELEM b_0 = b[0]; //By allowing the first entry in b[0] to be calculated twice

  37.         // (the first time erroneously), we allow the SSE2 code to run an extra pass.

  38.         // The savings in code and time are well worth having to store this value and

  39.         // calculate b[0] correctly afterwards.

  40. 

  41.         i = 0;

  42.         __asm__ volatile(

  43.             "pcmpeqd   %%xmm7, %%xmm7         \n\t"

  44.             "pcmpeqd   %%xmm3, %%xmm3         \n\t"

  45.             "psllw         $1, %%xmm3         \n\t"

  46.             "paddw     %%xmm7, %%xmm3         \n\t"

  47.             "psllw        $13, %%xmm3         \n\t"

  48.         ::);

  49.         for(; i<w_l-15; i+=16){

  50.             __asm__ volatile(

  51.                 "movdqu   (%1), %%xmm1        \n\t"

  52.                 "movdqu 16(%1), %%xmm5        \n\t"

  53.                 "movdqu  2(%1), %%xmm2        \n\t"

  54.                 "movdqu 18(%1), %%xmm6        \n\t"

  55.                 "paddw  %%xmm1, %%xmm2        \n\t"

  56.                 "paddw  %%xmm5, %%xmm6        \n\t"

  57.                 "paddw  %%xmm7, %%xmm2        \n\t"

  58.                 "paddw  %%xmm7, %%xmm6        \n\t"

  59.                 "pmulhw %%xmm3, %%xmm2        \n\t"

  60.                 "pmulhw %%xmm3, %%xmm6        \n\t"

  61.                 "paddw    (%0), %%xmm2        \n\t"

  62.                 "paddw  16(%0), %%xmm6        \n\t"

  63.                 "movdqa %%xmm2, (%0)          \n\t"

  64.                 "movdqa %%xmm6, 16(%0)        \n\t"

  65.                 :: "r"(&b[i]), "r"(&ref[i])

  66.                 : "memory"

  67.             );

  68.         }

  69.         snow_horizontal_compose_lift_lead_out(i, b, b, ref, width, w_l, 0, W_DM, W_DO, W_DS);

  70.         b[0] = b_0 - ((W_DM * 2 * ref[1]+W_DO)>>W_DS);

  71.     }

  72. 

  73.     { // Lift 1

  74.         IDWTELEM * const dst = b+w2;

  75. 

  76.         i = 0;

  77.         for(; (((x86_reg)&dst[i]) & 0x1F) && i<w_r; i++){

  78.             dst[i] = dst[i] - (b[i] + b[i + 1]);

  79.         }

  80.         for(; i<w_r-15; i+=16){

  81.             __asm__ volatile(

  82.                 "movdqu   (%1), %%xmm1        \n\t"

  83.                 "movdqu 16(%1), %%xmm5        \n\t"

  84.                 "movdqu  2(%1), %%xmm2        \n\t"

  85.                 "movdqu 18(%1), %%xmm6        \n\t"

  86.                 "paddw  %%xmm1, %%xmm2        \n\t"

  87.                 "paddw  %%xmm5, %%xmm6        \n\t"

  88.                 "movdqa   (%0), %%xmm0        \n\t"

  89.                 "movdqa 16(%0), %%xmm4        \n\t"

  90.                 "psubw  %%xmm2, %%xmm0        \n\t"

  91.                 "psubw  %%xmm6, %%xmm4        \n\t"

  92.                 "movdqa %%xmm0, (%0)          \n\t"

  93.                 "movdqa %%xmm4, 16(%0)        \n\t"

  94.                 :: "r"(&dst[i]), "r"(&b[i])

  95.                 : "memory"

  96.             );

  97.         }

  98.         snow_horizontal_compose_lift_lead_out(i, dst, dst, b, width, w_r, 1, W_CM, W_CO, W_CS);

  99.     }

  100. 

  101.     { // Lift 2

  102.         IDWTELEM * const ref = b+w2 - 1;

  103.         IDWTELEM b_0 = b[0];

  104. 

  105.         i = 0;

  106.         __asm__ volatile(

  107.             "psllw         $15, %%xmm7        \n\t"

  108.             "pcmpeqw    %%xmm6, %%xmm6        \n\t"

  109.             "psrlw         $13, %%xmm6        \n\t"

  110.             "paddw      %%xmm7, %%xmm6        \n\t"

  111.         ::);

  112.         for(; i<w_l-15; i+=16){

  113.             __asm__ volatile(

  114.                 "movdqu   (%1), %%xmm0        \n\t"

  115.                 "movdqu 16(%1), %%xmm4        \n\t"

  116.                 "movdqu  2(%1), %%xmm1        \n\t"

  117.                 "movdqu 18(%1), %%xmm5        \n\t" //FIXME try aligned reads and shifts

  118.                 "paddw  %%xmm6, %%xmm0        \n\t"

  119.                 "paddw  %%xmm6, %%xmm4        \n\t"

  120.                 "paddw  %%xmm7, %%xmm1        \n\t"

  121.                 "paddw  %%xmm7, %%xmm5        \n\t"

  122.                 "pavgw  %%xmm1, %%xmm0        \n\t"

  123.                 "pavgw  %%xmm5, %%xmm4        \n\t"

  124.                 "psubw  %%xmm7, %%xmm0        \n\t"

  125.                 "psubw  %%xmm7, %%xmm4        \n\t"

  126.                 "psraw      $1, %%xmm0        \n\t"

  127.                 "psraw      $1, %%xmm4        \n\t"

  128.                 "movdqa   (%0), %%xmm1        \n\t"

  129.                 "movdqa 16(%0), %%xmm5        \n\t"

  130.                 "paddw  %%xmm1, %%xmm0        \n\t"

  131.                 "paddw  %%xmm5, %%xmm4        \n\t"

  132.                 "psraw      $2, %%xmm0        \n\t"

  133.                 "psraw      $2, %%xmm4        \n\t"

  134.                 "paddw  %%xmm1, %%xmm0        \n\t"

  135.                 "paddw  %%xmm5, %%xmm4        \n\t"

  136.                 "movdqa %%xmm0, (%0)          \n\t"

  137.                 "movdqa %%xmm4, 16(%0)        \n\t"

  138.                 :: "r"(&b[i]), "r"(&ref[i])

  139.                 : "memory"

  140.             );

  141.         }

  142.         snow_horizontal_compose_liftS_lead_out(i, b, b, ref, width, w_l);

  143.         b[0] = b_0 + ((2 * ref[1] + W_BO-1 + 4 * b_0) >> W_BS);

  144.     }

  145. 

  146.     { // Lift 3

  147.         IDWTELEM * const src = b+w2;

  148. 

  149.         i = 0;

  150.         for(; (((x86_reg)&temp[i]) & 0x1F) && i<w_r; i++){

  151.             temp[i] = src[i] - ((-W_AM*(b[i] + b[i+1]))>>W_AS);

  152.         }

  153.         for(; i<w_r-7; i+=8){

  154.             __asm__ volatile(

  155.                 "movdqu  2(%1), %%xmm2        \n\t"

  156.                 "movdqu 18(%1), %%xmm6        \n\t"

  157.                 "paddw    (%1), %%xmm2        \n\t"

  158.                 "paddw  16(%1), %%xmm6        \n\t"

  159.                 "movdqu   (%0), %%xmm0        \n\t"

  160.                 "movdqu 16(%0), %%xmm4        \n\t"

  161.                 "paddw  %%xmm2, %%xmm0        \n\t"

  162.                 "paddw  %%xmm6, %%xmm4        \n\t"

  163.                 "psraw      $1, %%xmm2        \n\t"

  164.                 "psraw      $1, %%xmm6        \n\t"

  165.                 "paddw  %%xmm0, %%xmm2        \n\t"

  166.                 "paddw  %%xmm4, %%xmm6        \n\t"

  167.                 "movdqa %%xmm2, (%2)          \n\t"

  168.                 "movdqa %%xmm6, 16(%2)        \n\t"

  169.                 :: "r"(&src[i]), "r"(&b[i]), "r"(&temp[i])

  170.                  : "memory"

  171.                );

  172.         }

  173.         snow_horizontal_compose_lift_lead_out(i, temp, src, b, width, w_r, 1, -W_AM, W_AO+1, W_AS);

  174.     }

  175. 

  176.     {

  177.         snow_interleave_line_header(&i, width, b, temp);

  178. 

  179.         for (; (i & 0x3E) != 0x3E; i-=2){

  180.             b[i+1] = temp[i>>1];

  181.             b[i] = b[i>>1];

  182.         }

  183.         for (i-=62; i>=0; i-=64){

  184.             __asm__ volatile(

  185.                 "movdqa      (%1), %%xmm0       \n\t"

  186.                 "movdqa    16(%1), %%xmm2       \n\t"

  187.                 "movdqa    32(%1), %%xmm4       \n\t"

  188.                 "movdqa    48(%1), %%xmm6       \n\t"

  189.                 "movdqa      (%1), %%xmm1       \n\t"

  190.                 "movdqa    16(%1), %%xmm3       \n\t"

  191.                 "movdqa    32(%1), %%xmm5       \n\t"

  192.                 "movdqa    48(%1), %%xmm7       \n\t"

  193.                 "punpcklwd   (%2), %%xmm0       \n\t"

  194.                 "punpcklwd 16(%2), %%xmm2       \n\t"

  195.                 "punpcklwd 32(%2), %%xmm4       \n\t"

  196.                 "punpcklwd 48(%2), %%xmm6       \n\t"

  197.                 "movdqa    %%xmm0, (%0)         \n\t"

  198.                 "movdqa    %%xmm2, 32(%0)       \n\t"

  199.                 "movdqa    %%xmm4, 64(%0)       \n\t"

  200.                 "movdqa    %%xmm6, 96(%0)       \n\t"

  201.                 "punpckhwd   (%2), %%xmm1       \n\t"

  202.                 "punpckhwd 16(%2), %%xmm3       \n\t"

  203.                 "punpckhwd 32(%2), %%xmm5       \n\t"

  204.                 "punpckhwd 48(%2), %%xmm7       \n\t"

  205.                 "movdqa    %%xmm1, 16(%0)       \n\t"

  206.                 "movdqa    %%xmm3, 48(%0)       \n\t"

  207.                 "movdqa    %%xmm5, 80(%0)       \n\t"

  208.                 "movdqa    %%xmm7, 112(%0)      \n\t"

  209.                 :: "r"(&(b)[i]), "r"(&(b)[i>>1]), "r"(&(temp)[i>>1])

  210.                  : "memory"

  211.                );

  212.         }

  213.     }

  214. }

  215. 

  216. void ff_snow_horizontal_compose97i_mmx(IDWTELEM *b, int width){

  217.     const int w2= (width+1)>>1;

  218.     IDWTELEM temp[width >> 1];

  219.     const int w_l= (width>>1);

  220.     const int w_r= w2 - 1;

  221.     int i;

  222. 

  223.     { // Lift 0

  224.         IDWTELEM * const ref = b + w2 - 1;

  225. 

  226.         i = 1;

  227.         b[0] = b[0] - ((W_DM * 2 * ref[1]+W_DO)>>W_DS);

  228.         __asm__ volatile(

  229.             "pcmpeqw    %%mm7, %%mm7         \n\t"

  230.             "pcmpeqw    %%mm3, %%mm3         \n\t"

  231.             "psllw         $1, %%mm3         \n\t"

  232.             "paddw      %%mm7, %%mm3         \n\t"

  233.             "psllw        $13, %%mm3         \n\t"

  234.            ::);

  235.         for(; i<w_l-7; i+=8){

  236.             __asm__ volatile(

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

  238.                 "movq    8(%1), %%mm6        \n\t"

  239.                 "paddw   2(%1), %%mm2        \n\t"

  240.                 "paddw  10(%1), %%mm6        \n\t"

  241.                 "paddw   %%mm7, %%mm2        \n\t"

  242.                 "paddw   %%mm7, %%mm6        \n\t"

  243.                 "pmulhw  %%mm3, %%mm2        \n\t"

  244.                 "pmulhw  %%mm3, %%mm6        \n\t"

  245.                 "paddw    (%0), %%mm2        \n\t"

  246.                 "paddw   8(%0), %%mm6        \n\t"

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

  248.                 "movq    %%mm6, 8(%0)        \n\t"

  249.                 :: "r"(&b[i]), "r"(&ref[i])

  250.                  : "memory"

  251.                );

  252.         }

  253.         snow_horizontal_compose_lift_lead_out(i, b, b, ref, width, w_l, 0, W_DM, W_DO, W_DS);

  254.     }

  255. 

  256.     { // Lift 1

  257.         IDWTELEM * const dst = b+w2;

  258. 

  259.         i = 0;

  260.         for(; i<w_r-7; i+=8){

  261.             __asm__ volatile(

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

  263.                 "movq    8(%1), %%mm6        \n\t"

  264.                 "paddw   2(%1), %%mm2        \n\t"

  265.                 "paddw  10(%1), %%mm6        \n\t"

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

  267.                 "movq    8(%0), %%mm4        \n\t"

  268.                 "psubw   %%mm2, %%mm0        \n\t"

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

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

  271.                 "movq    %%mm4, 8(%0)        \n\t"

  272.                 :: "r"(&dst[i]), "r"(&b[i])

  273.                  : "memory"

  274.                );

  275.         }

  276.         snow_horizontal_compose_lift_lead_out(i, dst, dst, b, width, w_r, 1, W_CM, W_CO, W_CS);

  277.     }

  278. 

  279.     { // Lift 2

  280.         IDWTELEM * const ref = b+w2 - 1;

  281. 

  282.         i = 1;

  283.         b[0] = b[0] + (((2 * ref[1] + W_BO) + 4 * b[0]) >> W_BS);

  284.         __asm__ volatile(

  285.             "psllw         $15, %%mm7        \n\t"

  286.             "pcmpeqw     %%mm6, %%mm6        \n\t"

  287.             "psrlw         $13, %%mm6        \n\t"

  288.             "paddw       %%mm7, %%mm6        \n\t"

  289.            ::);

  290.         for(; i<w_l-7; i+=8){

  291.             __asm__ volatile(

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

  293.                 "movq    8(%1), %%mm4        \n\t"

  294.                 "movq    2(%1), %%mm1        \n\t"

  295.                 "movq   10(%1), %%mm5        \n\t"

  296.                 "paddw   %%mm6, %%mm0        \n\t"

  297.                 "paddw   %%mm6, %%mm4        \n\t"

  298.                 "paddw   %%mm7, %%mm1        \n\t"

  299.                 "paddw   %%mm7, %%mm5        \n\t"

  300.                 "pavgw   %%mm1, %%mm0        \n\t"

  301.                 "pavgw   %%mm5, %%mm4        \n\t"

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

  303.                 "psubw   %%mm7, %%mm4        \n\t"

  304.                 "psraw      $1, %%mm0        \n\t"

  305.                 "psraw      $1, %%mm4        \n\t"

  306.                 "movq     (%0), %%mm1        \n\t"

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

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

  309.                 "paddw   %%mm5, %%mm4        \n\t"

  310.                 "psraw      $2, %%mm0        \n\t"

  311.                 "psraw      $2, %%mm4        \n\t"

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

  313.                 "paddw   %%mm5, %%mm4        \n\t"

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

  315.                 "movq    %%mm4, 8(%0)        \n\t"

  316.                 :: "r"(&b[i]), "r"(&ref[i])

  317.                  : "memory"

  318.                );

  319.         }

  320.         snow_horizontal_compose_liftS_lead_out(i, b, b, ref, width, w_l);

  321.     }

  322. 

  323.     { // Lift 3

  324.         IDWTELEM * const src = b+w2;

  325.         i = 0;

  326. 

  327.         for(; i<w_r-7; i+=8){

  328.             __asm__ volatile(

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

  330.                 "movq   10(%1), %%mm6        \n\t"

  331.                 "paddw    (%1), %%mm2        \n\t"

  332.                 "paddw   8(%1), %%mm6        \n\t"

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

  334.                 "movq    8(%0), %%mm4        \n\t"

  335.                 "paddw   %%mm2, %%mm0        \n\t"

  336.                 "paddw   %%mm6, %%mm4        \n\t"

  337.                 "psraw      $1, %%mm2        \n\t"

  338.                 "psraw      $1, %%mm6        \n\t"

  339.                 "paddw   %%mm0, %%mm2        \n\t"

  340.                 "paddw   %%mm4, %%mm6        \n\t"

  341.                 "movq    %%mm2, (%2)         \n\t"

  342.                 "movq    %%mm6, 8(%2)        \n\t"

  343.                 :: "r"(&src[i]), "r"(&b[i]), "r"(&temp[i])

  344.                  : "memory"

  345.                );

  346.         }

  347.         snow_horizontal_compose_lift_lead_out(i, temp, src, b, width, w_r, 1, -W_AM, W_AO+1, W_AS);

  348.     }

  349. 

  350.     {

  351.         snow_interleave_line_header(&i, width, b, temp);

  352. 

  353.         for (; (i & 0x1E) != 0x1E; i-=2){

  354.             b[i+1] = temp[i>>1];

  355.             b[i] = b[i>>1];

  356.         }

  357.         for (i-=30; i>=0; i-=32){

  358.             __asm__ volatile(

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

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

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

  362.                 "movq      24(%1), %%mm6       \n\t"

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

  364.                 "movq       8(%1), %%mm3       \n\t"

  365.                 "movq      16(%1), %%mm5       \n\t"

  366.                 "movq      24(%1), %%mm7       \n\t"

  367.                 "punpcklwd   (%2), %%mm0       \n\t"

  368.                 "punpcklwd  8(%2), %%mm2       \n\t"

  369.                 "punpcklwd 16(%2), %%mm4       \n\t"

  370.                 "punpcklwd 24(%2), %%mm6       \n\t"

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

  372.                 "movq       %%mm2, 16(%0)      \n\t"

  373.                 "movq       %%mm4, 32(%0)      \n\t"

  374.                 "movq       %%mm6, 48(%0)      \n\t"

  375.                 "punpckhwd   (%2), %%mm1       \n\t"

  376.                 "punpckhwd  8(%2), %%mm3       \n\t"

  377.                 "punpckhwd 16(%2), %%mm5       \n\t"

  378.                 "punpckhwd 24(%2), %%mm7       \n\t"

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

  380.                 "movq       %%mm3, 24(%0)      \n\t"

  381.                 "movq       %%mm5, 40(%0)      \n\t"

  382.                 "movq       %%mm7, 56(%0)      \n\t"

  383.                 :: "r"(&b[i]), "r"(&b[i>>1]), "r"(&temp[i>>1])

  384.                  : "memory"

  385.                );

  386.         }

  387.     }

  388. }

  389. 

  390. #if HAVE_7REGS

  391. #define snow_vertical_compose_sse2_load_add(op,r,t0,t1,t2,t3)\

  392.         ""op" ("r",%%"REG_d"), %%"t0"      \n\t"\

  393.         ""op" 16("r",%%"REG_d"), %%"t1"    \n\t"\

  394.         ""op" 32("r",%%"REG_d"), %%"t2"    \n\t"\

  395.         ""op" 48("r",%%"REG_d"), %%"t3"    \n\t"

  396. 

  397. #define snow_vertical_compose_sse2_load(r,t0,t1,t2,t3)\

  398.         snow_vertical_compose_sse2_load_add("movdqa",r,t0,t1,t2,t3)

  399. 

  400. #define snow_vertical_compose_sse2_add(r,t0,t1,t2,t3)\

  401.         snow_vertical_compose_sse2_load_add("paddw",r,t0,t1,t2,t3)

  402. 

  403. #define snow_vertical_compose_r2r_sub(s0,s1,s2,s3,t0,t1,t2,t3)\

  404.         "psubw %%"s0", %%"t0" \n\t"\

  405.         "psubw %%"s1", %%"t1" \n\t"\

  406.         "psubw %%"s2", %%"t2" \n\t"\

  407.         "psubw %%"s3", %%"t3" \n\t"

  408. 

  409. #define snow_vertical_compose_sse2_store(w,s0,s1,s2,s3)\

  410.         "movdqa %%"s0", ("w",%%"REG_d")      \n\t"\

  411.         "movdqa %%"s1", 16("w",%%"REG_d")    \n\t"\

  412.         "movdqa %%"s2", 32("w",%%"REG_d")    \n\t"\

  413.         "movdqa %%"s3", 48("w",%%"REG_d")    \n\t"

  414. 

  415. #define snow_vertical_compose_sra(n,t0,t1,t2,t3)\

  416.         "psraw $"n", %%"t0" \n\t"\

  417.         "psraw $"n", %%"t1" \n\t"\

  418.         "psraw $"n", %%"t2" \n\t"\

  419.         "psraw $"n", %%"t3" \n\t"

  420. 

  421. #define snow_vertical_compose_r2r_add(s0,s1,s2,s3,t0,t1,t2,t3)\

  422.         "paddw %%"s0", %%"t0" \n\t"\

  423.         "paddw %%"s1", %%"t1" \n\t"\

  424.         "paddw %%"s2", %%"t2" \n\t"\

  425.         "paddw %%"s3", %%"t3" \n\t"

  426. 

  427. #define snow_vertical_compose_r2r_pmulhw(s0,s1,s2,s3,t0,t1,t2,t3)\

  428.         "pmulhw %%"s0", %%"t0" \n\t"\

  429.         "pmulhw %%"s1", %%"t1" \n\t"\

  430.         "pmulhw %%"s2", %%"t2" \n\t"\

  431.         "pmulhw %%"s3", %%"t3" \n\t"

  432. 

  433. #define snow_vertical_compose_sse2_move(s0,s1,s2,s3,t0,t1,t2,t3)\

  434.         "movdqa %%"s0", %%"t0" \n\t"\

  435.         "movdqa %%"s1", %%"t1" \n\t"\

  436.         "movdqa %%"s2", %%"t2" \n\t"\

  437.         "movdqa %%"s3", %%"t3" \n\t"

  438. 

  439. void ff_snow_vertical_compose97i_sse2(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){

  440.     x86_reg i = width;

  441. 

  442.     while(i & 0x1F)

  443.     {

  444.         i--;

  445.         b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;

  446.         b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;

  447.         b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;

  448.         b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;

  449.     }

  450.     i+=i;

  451. 

  452.          __asm__ volatile (

  453.         "jmp 2f                                      \n\t"

  454.         "1:                                          \n\t"

  455.         snow_vertical_compose_sse2_load("%4","xmm0","xmm2","xmm4","xmm6")

  456.         snow_vertical_compose_sse2_add("%6","xmm0","xmm2","xmm4","xmm6")

  457. 

  458. 

  459.         "pcmpeqw    %%xmm0, %%xmm0                   \n\t"

  460.         "pcmpeqw    %%xmm2, %%xmm2                   \n\t"

  461.         "paddw      %%xmm2, %%xmm2                   \n\t"

  462.         "paddw      %%xmm0, %%xmm2                   \n\t"

  463.         "psllw         $13, %%xmm2                   \n\t"

  464.         snow_vertical_compose_r2r_add("xmm0","xmm0","xmm0","xmm0","xmm1","xmm3","xmm5","xmm7")

  465.         snow_vertical_compose_r2r_pmulhw("xmm2","xmm2","xmm2","xmm2","xmm1","xmm3","xmm5","xmm7")

  466.         snow_vertical_compose_sse2_add("%5","xmm1","xmm3","xmm5","xmm7")

  467.         snow_vertical_compose_sse2_store("%5","xmm1","xmm3","xmm5","xmm7")

  468.         snow_vertical_compose_sse2_load("%4","xmm0","xmm2","xmm4","xmm6")

  469.         snow_vertical_compose_sse2_add("%3","xmm1","xmm3","xmm5","xmm7")

  470.         snow_vertical_compose_r2r_sub("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")

  471.         snow_vertical_compose_sse2_store("%4","xmm0","xmm2","xmm4","xmm6")

  472. 

  473.         "pcmpeqw %%xmm7, %%xmm7                      \n\t"

  474.         "pcmpeqw %%xmm5, %%xmm5                      \n\t"

  475.         "psllw $15, %%xmm7                           \n\t"

  476.         "psrlw $13, %%xmm5                           \n\t"

  477.         "paddw %%xmm7, %%xmm5                        \n\t"

  478.         snow_vertical_compose_r2r_add("xmm5","xmm5","xmm5","xmm5","xmm0","xmm2","xmm4","xmm6")

  479.         "movq   (%2,%%"REG_d"), %%xmm1        \n\t"

  480.         "movq  8(%2,%%"REG_d"), %%xmm3        \n\t"

  481.         "paddw %%xmm7, %%xmm1                        \n\t"

  482.         "paddw %%xmm7, %%xmm3                        \n\t"

  483.         "pavgw %%xmm1, %%xmm0                        \n\t"

  484.         "pavgw %%xmm3, %%xmm2                        \n\t"

  485.         "movq 16(%2,%%"REG_d"), %%xmm1        \n\t"

  486.         "movq 24(%2,%%"REG_d"), %%xmm3        \n\t"

  487.         "paddw %%xmm7, %%xmm1                        \n\t"

  488.         "paddw %%xmm7, %%xmm3                        \n\t"

  489.         "pavgw %%xmm1, %%xmm4                        \n\t"

  490.         "pavgw %%xmm3, %%xmm6                        \n\t"

  491.         snow_vertical_compose_r2r_sub("xmm7","xmm7","xmm7","xmm7","xmm0","xmm2","xmm4","xmm6")

  492.         snow_vertical_compose_sra("1","xmm0","xmm2","xmm4","xmm6")

  493.         snow_vertical_compose_sse2_add("%3","xmm0","xmm2","xmm4","xmm6")

  494. 

  495.         snow_vertical_compose_sra("2","xmm0","xmm2","xmm4","xmm6")

  496.         snow_vertical_compose_sse2_add("%3","xmm0","xmm2","xmm4","xmm6")

  497.         snow_vertical_compose_sse2_store("%3","xmm0","xmm2","xmm4","xmm6")

  498.         snow_vertical_compose_sse2_add("%1","xmm0","xmm2","xmm4","xmm6")

  499.         snow_vertical_compose_sse2_move("xmm0","xmm2","xmm4","xmm6","xmm1","xmm3","xmm5","xmm7")

  500.         snow_vertical_compose_sra("1","xmm0","xmm2","xmm4","xmm6")

  501.         snow_vertical_compose_r2r_add("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")

  502.         snow_vertical_compose_sse2_add("%2","xmm0","xmm2","xmm4","xmm6")

  503.         snow_vertical_compose_sse2_store("%2","xmm0","xmm2","xmm4","xmm6")

  504. 

  505.         "2:                                          \n\t"

  506.         "sub $64, %%"REG_d"                          \n\t"

  507.         "jge 1b                                      \n\t"

  508.         :"+d"(i)

  509.         :"r"(b0),"r"(b1),"r"(b2),"r"(b3),"r"(b4),"r"(b5));

  510. }

  511. 

  512. #define snow_vertical_compose_mmx_load_add(op,r,t0,t1,t2,t3)\

  513.         ""op" ("r",%%"REG_d"), %%"t0"   \n\t"\

  514.         ""op" 8("r",%%"REG_d"), %%"t1"  \n\t"\

  515.         ""op" 16("r",%%"REG_d"), %%"t2" \n\t"\

  516.         ""op" 24("r",%%"REG_d"), %%"t3" \n\t"

  517. 

  518. #define snow_vertical_compose_mmx_load(r,t0,t1,t2,t3)\

  519.         snow_vertical_compose_mmx_load_add("movq",r,t0,t1,t2,t3)

  520. 

  521. #define snow_vertical_compose_mmx_add(r,t0,t1,t2,t3)\

  522.         snow_vertical_compose_mmx_load_add("paddw",r,t0,t1,t2,t3)

  523. 

  524. #define snow_vertical_compose_mmx_store(w,s0,s1,s2,s3)\

  525.         "movq %%"s0", ("w",%%"REG_d")   \n\t"\

  526.         "movq %%"s1", 8("w",%%"REG_d")  \n\t"\

  527.         "movq %%"s2", 16("w",%%"REG_d") \n\t"\

  528.         "movq %%"s3", 24("w",%%"REG_d") \n\t"

  529. 

  530. #define snow_vertical_compose_mmx_move(s0,s1,s2,s3,t0,t1,t2,t3)\

  531.         "movq %%"s0", %%"t0" \n\t"\

  532.         "movq %%"s1", %%"t1" \n\t"\

  533.         "movq %%"s2", %%"t2" \n\t"\

  534.         "movq %%"s3", %%"t3" \n\t"

  535. 

  536. 

  537. void ff_snow_vertical_compose97i_mmx(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){

  538.     x86_reg i = width;

  539.     while(i & 15)

  540.     {

  541.         i--;

  542.         b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;

  543.         b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;

  544.         b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;

  545.         b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;

  546.     }

  547.     i+=i;

  548.     __asm__ volatile(

  549.         "jmp 2f                                      \n\t"

  550.         "1:                                          \n\t"

  551. 

  552.         snow_vertical_compose_mmx_load("%4","mm1","mm3","mm5","mm7")

  553.         snow_vertical_compose_mmx_add("%6","mm1","mm3","mm5","mm7")

  554.         "pcmpeqw    %%mm0, %%mm0                     \n\t"

  555.         "pcmpeqw    %%mm2, %%mm2                     \n\t"

  556.         "paddw      %%mm2, %%mm2                     \n\t"

  557.         "paddw      %%mm0, %%mm2                     \n\t"

  558.         "psllw        $13, %%mm2                     \n\t"

  559.         snow_vertical_compose_r2r_add("mm0","mm0","mm0","mm0","mm1","mm3","mm5","mm7")

  560.         snow_vertical_compose_r2r_pmulhw("mm2","mm2","mm2","mm2","mm1","mm3","mm5","mm7")

  561.         snow_vertical_compose_mmx_add("%5","mm1","mm3","mm5","mm7")

  562.         snow_vertical_compose_mmx_store("%5","mm1","mm3","mm5","mm7")

  563.         snow_vertical_compose_mmx_load("%4","mm0","mm2","mm4","mm6")

  564.         snow_vertical_compose_mmx_add("%3","mm1","mm3","mm5","mm7")

  565.         snow_vertical_compose_r2r_sub("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")

  566.         snow_vertical_compose_mmx_store("%4","mm0","mm2","mm4","mm6")

  567.         "pcmpeqw %%mm7, %%mm7                        \n\t"

  568.         "pcmpeqw %%mm5, %%mm5                        \n\t"

  569.         "psllw $15, %%mm7                            \n\t"

  570.         "psrlw $13, %%mm5                            \n\t"

  571.         "paddw %%mm7, %%mm5                          \n\t"

  572.         snow_vertical_compose_r2r_add("mm5","mm5","mm5","mm5","mm0","mm2","mm4","mm6")

  573.         "movq   (%2,%%"REG_d"), %%mm1         \n\t"

  574.         "movq  8(%2,%%"REG_d"), %%mm3         \n\t"

  575.         "paddw %%mm7, %%mm1                          \n\t"

  576.         "paddw %%mm7, %%mm3                          \n\t"

  577.         "pavgw %%mm1, %%mm0                          \n\t"

  578.         "pavgw %%mm3, %%mm2                          \n\t"

  579.         "movq 16(%2,%%"REG_d"), %%mm1         \n\t"

  580.         "movq 24(%2,%%"REG_d"), %%mm3         \n\t"

  581.         "paddw %%mm7, %%mm1                          \n\t"

  582.         "paddw %%mm7, %%mm3                          \n\t"

  583.         "pavgw %%mm1, %%mm4                          \n\t"

  584.         "pavgw %%mm3, %%mm6                          \n\t"

  585.         snow_vertical_compose_r2r_sub("mm7","mm7","mm7","mm7","mm0","mm2","mm4","mm6")

  586.         snow_vertical_compose_sra("1","mm0","mm2","mm4","mm6")

  587.         snow_vertical_compose_mmx_add("%3","mm0","mm2","mm4","mm6")

  588. 

  589.         snow_vertical_compose_sra("2","mm0","mm2","mm4","mm6")

  590.         snow_vertical_compose_mmx_add("%3","mm0","mm2","mm4","mm6")

  591.         snow_vertical_compose_mmx_store("%3","mm0","mm2","mm4","mm6")

  592.         snow_vertical_compose_mmx_add("%1","mm0","mm2","mm4","mm6")

  593.         snow_vertical_compose_mmx_move("mm0","mm2","mm4","mm6","mm1","mm3","mm5","mm7")

  594.         snow_vertical_compose_sra("1","mm0","mm2","mm4","mm6")

  595.         snow_vertical_compose_r2r_add("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")

  596.         snow_vertical_compose_mmx_add("%2","mm0","mm2","mm4","mm6")

  597.         snow_vertical_compose_mmx_store("%2","mm0","mm2","mm4","mm6")

  598. 

  599.         "2:                                          \n\t"

  600.         "sub $32, %%"REG_d"                          \n\t"

  601.         "jge 1b                                      \n\t"

  602.         :"+d"(i)

  603.         :"r"(b0),"r"(b1),"r"(b2),"r"(b3),"r"(b4),"r"(b5));

  604. }

  605. #endif //HAVE_7REGS

  606. 

  607. #define snow_inner_add_yblock_sse2_header \

  608.     IDWTELEM * * dst_array = sb->line + src_y;\

  609.     x86_reg tmp;\

  610.     __asm__ volatile(\

  611.              "mov  %7, %%"REG_c"             \n\t"\

  612.              "mov  %6, %2                    \n\t"\

  613.              "mov  %4, %%"REG_S"             \n\t"\

  614.              "pxor %%xmm7, %%xmm7            \n\t" /* 0 */\

  615.              "pcmpeqd %%xmm3, %%xmm3         \n\t"\

  616.              "psllw $15, %%xmm3              \n\t"\

  617.              "psrlw $12, %%xmm3              \n\t" /* FRAC_BITS >> 1 */\

  618.              "1:                             \n\t"\

  619.              "mov %1, %%"REG_D"              \n\t"\

  620.              "mov (%%"REG_D"), %%"REG_D"     \n\t"\

  621.              "add %3, %%"REG_D"              \n\t"

  622. 

  623. #define snow_inner_add_yblock_sse2_start_8(out_reg1, out_reg2, ptr_offset, s_offset)\

  624.              "mov "PTR_SIZE"*"ptr_offset"(%%"REG_a"), %%"REG_d"; \n\t"\

  625.              "movq (%%"REG_d"), %%"out_reg1" \n\t"\

  626.              "movq (%%"REG_d", %%"REG_c"), %%"out_reg2" \n\t"\

  627.              "punpcklbw %%xmm7, %%"out_reg1" \n\t"\

  628.              "punpcklbw %%xmm7, %%"out_reg2" \n\t"\

  629.              "movq "s_offset"(%%"REG_S"), %%xmm0 \n\t"\

  630.              "movq "s_offset"+16(%%"REG_S"), %%xmm4 \n\t"\

  631.              "punpcklbw %%xmm7, %%xmm0       \n\t"\

  632.              "punpcklbw %%xmm7, %%xmm4       \n\t"\

  633.              "pmullw %%xmm0, %%"out_reg1"    \n\t"\

  634.              "pmullw %%xmm4, %%"out_reg2"    \n\t"

  635. 

  636. #define snow_inner_add_yblock_sse2_start_16(out_reg1, out_reg2, ptr_offset, s_offset)\

  637.              "mov "PTR_SIZE"*"ptr_offset"(%%"REG_a"), %%"REG_d"; \n\t"\

  638.              "movq (%%"REG_d"), %%"out_reg1" \n\t"\

  639.              "movq 8(%%"REG_d"), %%"out_reg2" \n\t"\

  640.              "punpcklbw %%xmm7, %%"out_reg1" \n\t"\

  641.              "punpcklbw %%xmm7, %%"out_reg2" \n\t"\

  642.              "movq "s_offset"(%%"REG_S"), %%xmm0 \n\t"\

  643.              "movq "s_offset"+8(%%"REG_S"), %%xmm4 \n\t"\

  644.              "punpcklbw %%xmm7, %%xmm0       \n\t"\

  645.              "punpcklbw %%xmm7, %%xmm4       \n\t"\

  646.              "pmullw %%xmm0, %%"out_reg1"    \n\t"\

  647.              "pmullw %%xmm4, %%"out_reg2"    \n\t"

  648. 

  649. #define snow_inner_add_yblock_sse2_accum_8(ptr_offset, s_offset) \

  650.              snow_inner_add_yblock_sse2_start_8("xmm2", "xmm6", ptr_offset, s_offset)\

  651.              "paddusw %%xmm2, %%xmm1         \n\t"\

  652.              "paddusw %%xmm6, %%xmm5         \n\t"

  653. 

  654. #define snow_inner_add_yblock_sse2_accum_16(ptr_offset, s_offset) \

  655.              snow_inner_add_yblock_sse2_start_16("xmm2", "xmm6", ptr_offset, s_offset)\

  656.              "paddusw %%xmm2, %%xmm1         \n\t"\

  657.              "paddusw %%xmm6, %%xmm5         \n\t"

  658. 

  659. #define snow_inner_add_yblock_sse2_end_common1\

  660.              "add $32, %%"REG_S"             \n\t"\

  661.              "add %%"REG_c", %0              \n\t"\

  662.              "add %%"REG_c", "PTR_SIZE"*3(%%"REG_a");\n\t"\

  663.              "add %%"REG_c", "PTR_SIZE"*2(%%"REG_a");\n\t"\

  664.              "add %%"REG_c", "PTR_SIZE"*1(%%"REG_a");\n\t"\

  665.              "add %%"REG_c", (%%"REG_a")     \n\t"

  666. 

  667. #define snow_inner_add_yblock_sse2_end_common2\

  668.              "jnz 1b                         \n\t"\

  669.              :"+m"(dst8),"+m"(dst_array),"=&r"(tmp)\

  670.              :\

  671.              "rm"((x86_reg)(src_x<<1)),"m"(obmc),"a"(block),"m"(b_h),"m"(src_stride):\

  672.              "%"REG_c"","%"REG_S"","%"REG_D"","%"REG_d"");

  673. 

  674. #define snow_inner_add_yblock_sse2_end_8\

  675.              "sal $1, %%"REG_c"              \n\t"\

  676.              "add $"PTR_SIZE"*2, %1          \n\t"\

  677.              snow_inner_add_yblock_sse2_end_common1\

  678.              "sar $1, %%"REG_c"              \n\t"\

  679.              "sub $2, %2                     \n\t"\

  680.              snow_inner_add_yblock_sse2_end_common2

  681. 

  682. #define snow_inner_add_yblock_sse2_end_16\

  683.              "add $"PTR_SIZE"*1, %1          \n\t"\

  684.              snow_inner_add_yblock_sse2_end_common1\

  685.              "dec %2                         \n\t"\

  686.              snow_inner_add_yblock_sse2_end_common2

  687. 

  688. static void inner_add_yblock_bw_8_obmc_16_bh_even_sse2(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h,

  689.                       int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){

  690. snow_inner_add_yblock_sse2_header

  691. snow_inner_add_yblock_sse2_start_8("xmm1", "xmm5", "3", "0")

  692. snow_inner_add_yblock_sse2_accum_8("2", "8")

  693. snow_inner_add_yblock_sse2_accum_8("1", "128")

  694. snow_inner_add_yblock_sse2_accum_8("0", "136")

  695. 

  696.              "mov %0, %%"REG_d"              \n\t"

  697.              "movdqa (%%"REG_D"), %%xmm0     \n\t"

  698.              "movdqa %%xmm1, %%xmm2          \n\t"

  699. 

  700.              "punpckhwd %%xmm7, %%xmm1       \n\t"

  701.              "punpcklwd %%xmm7, %%xmm2       \n\t"

  702.              "paddd %%xmm2, %%xmm0           \n\t"

  703.              "movdqa 16(%%"REG_D"), %%xmm2   \n\t"

  704.              "paddd %%xmm1, %%xmm2           \n\t"

  705.              "paddd %%xmm3, %%xmm0           \n\t"

  706.              "paddd %%xmm3, %%xmm2           \n\t"

  707. 

  708.              "mov %1, %%"REG_D"              \n\t"

  709.              "mov "PTR_SIZE"(%%"REG_D"), %%"REG_D";\n\t"

  710.              "add %3, %%"REG_D"              \n\t"

  711. 

  712.              "movdqa (%%"REG_D"), %%xmm4     \n\t"

  713.              "movdqa %%xmm5, %%xmm6          \n\t"

  714.              "punpckhwd %%xmm7, %%xmm5       \n\t"

  715.              "punpcklwd %%xmm7, %%xmm6       \n\t"

  716.              "paddd %%xmm6, %%xmm4           \n\t"

  717.              "movdqa 16(%%"REG_D"), %%xmm6   \n\t"

  718.              "paddd %%xmm5, %%xmm6           \n\t"

  719.              "paddd %%xmm3, %%xmm4           \n\t"

  720.              "paddd %%xmm3, %%xmm6           \n\t"

  721. 

  722.              "psrad $8, %%xmm0               \n\t" /* FRAC_BITS. */

  723.              "psrad $8, %%xmm2               \n\t" /* FRAC_BITS. */

  724.              "packssdw %%xmm2, %%xmm0        \n\t"

  725.              "packuswb %%xmm7, %%xmm0        \n\t"

  726.              "movq %%xmm0, (%%"REG_d")       \n\t"

  727. 

  728.              "psrad $8, %%xmm4               \n\t" /* FRAC_BITS. */

  729.              "psrad $8, %%xmm6               \n\t" /* FRAC_BITS. */

  730.              "packssdw %%xmm6, %%xmm4        \n\t"

  731.              "packuswb %%xmm7, %%xmm4        \n\t"

  732.              "movq %%xmm4, (%%"REG_d",%%"REG_c");\n\t"

  733. snow_inner_add_yblock_sse2_end_8

  734. }

  735. 

  736. static void inner_add_yblock_bw_16_obmc_32_sse2(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h,

  737.                       int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){

  738. snow_inner_add_yblock_sse2_header

  739. snow_inner_add_yblock_sse2_start_16("xmm1", "xmm5", "3", "0")

  740. snow_inner_add_yblock_sse2_accum_16("2", "16")

  741. snow_inner_add_yblock_sse2_accum_16("1", "512")

  742. snow_inner_add_yblock_sse2_accum_16("0", "528")

  743. 

  744.              "mov %0, %%"REG_d"              \n\t"

  745.              "psrlw $4, %%xmm1               \n\t"

  746.              "psrlw $4, %%xmm5               \n\t"

  747.              "paddw   (%%"REG_D"), %%xmm1    \n\t"

  748.              "paddw 16(%%"REG_D"), %%xmm5    \n\t"

  749.              "paddw %%xmm3, %%xmm1           \n\t"

  750.              "paddw %%xmm3, %%xmm5           \n\t"

  751.              "psraw $4, %%xmm1               \n\t" /* FRAC_BITS. */

  752.              "psraw $4, %%xmm5               \n\t" /* FRAC_BITS. */

  753.              "packuswb %%xmm5, %%xmm1        \n\t"

  754. 

  755.              "movdqu %%xmm1, (%%"REG_d")       \n\t"

  756. 

  757. snow_inner_add_yblock_sse2_end_16

  758. }

  759. 

  760. #define snow_inner_add_yblock_mmx_header \

  761.     IDWTELEM * * dst_array = sb->line + src_y;\

  762.     x86_reg tmp;\

  763.     __asm__ volatile(\

  764.              "mov  %7, %%"REG_c"             \n\t"\

  765.              "mov  %6, %2                    \n\t"\

  766.              "mov  %4, %%"REG_S"             \n\t"\

  767.              "pxor %%mm7, %%mm7              \n\t" /* 0 */\

  768.              "pcmpeqd %%mm3, %%mm3           \n\t"\

  769.              "psllw $15, %%mm3               \n\t"\

  770.              "psrlw $12, %%mm3               \n\t" /* FRAC_BITS >> 1 */\

  771.              "1:                             \n\t"\

  772.              "mov %1, %%"REG_D"              \n\t"\

  773.              "mov (%%"REG_D"), %%"REG_D"     \n\t"\

  774.              "add %3, %%"REG_D"              \n\t"

  775. 

  776. #define snow_inner_add_yblock_mmx_start(out_reg1, out_reg2, ptr_offset, s_offset, d_offset)\

  777.              "mov "PTR_SIZE"*"ptr_offset"(%%"REG_a"), %%"REG_d"; \n\t"\

  778.              "movd "d_offset"(%%"REG_d"), %%"out_reg1" \n\t"\

  779.              "movd "d_offset"+4(%%"REG_d"), %%"out_reg2" \n\t"\

  780.              "punpcklbw %%mm7, %%"out_reg1" \n\t"\

  781.              "punpcklbw %%mm7, %%"out_reg2" \n\t"\

  782.              "movd "s_offset"(%%"REG_S"), %%mm0 \n\t"\

  783.              "movd "s_offset"+4(%%"REG_S"), %%mm4 \n\t"\

  784.              "punpcklbw %%mm7, %%mm0       \n\t"\

  785.              "punpcklbw %%mm7, %%mm4       \n\t"\

  786.              "pmullw %%mm0, %%"out_reg1"    \n\t"\

  787.              "pmullw %%mm4, %%"out_reg2"    \n\t"

  788. 

  789. #define snow_inner_add_yblock_mmx_accum(ptr_offset, s_offset, d_offset) \

  790.              snow_inner_add_yblock_mmx_start("mm2", "mm6", ptr_offset, s_offset, d_offset)\

  791.              "paddusw %%mm2, %%mm1         \n\t"\

  792.              "paddusw %%mm6, %%mm5         \n\t"

  793. 

  794. #define snow_inner_add_yblock_mmx_mix(read_offset, write_offset)\

  795.              "mov %0, %%"REG_d"              \n\t"\

  796.              "psrlw $4, %%mm1                \n\t"\

  797.              "psrlw $4, %%mm5                \n\t"\

  798.              "paddw "read_offset"(%%"REG_D"), %%mm1 \n\t"\

  799.              "paddw "read_offset"+8(%%"REG_D"), %%mm5 \n\t"\

  800.              "paddw %%mm3, %%mm1             \n\t"\

  801.              "paddw %%mm3, %%mm5             \n\t"\

  802.              "psraw $4, %%mm1                \n\t"\

  803.              "psraw $4, %%mm5                \n\t"\

  804.              "packuswb %%mm5, %%mm1          \n\t"\

  805.              "movq %%mm1, "write_offset"(%%"REG_d") \n\t"

  806. 

  807. #define snow_inner_add_yblock_mmx_end(s_step)\

  808.              "add $"s_step", %%"REG_S"             \n\t"\

  809.              "add %%"REG_c", "PTR_SIZE"*3(%%"REG_a");\n\t"\

  810.              "add %%"REG_c", "PTR_SIZE"*2(%%"REG_a");\n\t"\

  811.              "add %%"REG_c", "PTR_SIZE"*1(%%"REG_a");\n\t"\

  812.              "add %%"REG_c", (%%"REG_a")     \n\t"\

  813.              "add $"PTR_SIZE"*1, %1          \n\t"\

  814.              "add %%"REG_c", %0              \n\t"\

  815.              "dec %2                         \n\t"\

  816.              "jnz 1b                         \n\t"\

  817.              :"+m"(dst8),"+m"(dst_array),"=&r"(tmp)\

  818.              :\

  819.              "rm"((x86_reg)(src_x<<1)),"m"(obmc),"a"(block),"m"(b_h),"m"(src_stride):\

  820.              "%"REG_c"","%"REG_S"","%"REG_D"","%"REG_d"");

  821. 

  822. static void inner_add_yblock_bw_8_obmc_16_mmx(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h,

  823.                       int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){

  824. snow_inner_add_yblock_mmx_header

  825. snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "0", "0")

  826. snow_inner_add_yblock_mmx_accum("2", "8", "0")

  827. snow_inner_add_yblock_mmx_accum("1", "128", "0")

  828. snow_inner_add_yblock_mmx_accum("0", "136", "0")

  829. snow_inner_add_yblock_mmx_mix("0", "0")

  830. snow_inner_add_yblock_mmx_end("16")

  831. }

  832. 

  833. static void inner_add_yblock_bw_16_obmc_32_mmx(const uint8_t *obmc, const x86_reg obmc_stride, uint8_t * * block, int b_w, x86_reg b_h,

  834.                       int src_x, int src_y, x86_reg src_stride, slice_buffer * sb, int add, uint8_t * dst8){

  835. snow_inner_add_yblock_mmx_header

  836. snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "0", "0")

  837. snow_inner_add_yblock_mmx_accum("2", "16", "0")

  838. snow_inner_add_yblock_mmx_accum("1", "512", "0")

  839. snow_inner_add_yblock_mmx_accum("0", "528", "0")

  840. snow_inner_add_yblock_mmx_mix("0", "0")

  841. 

  842. snow_inner_add_yblock_mmx_start("mm1", "mm5", "3", "8", "8")

  843. snow_inner_add_yblock_mmx_accum("2", "24", "8")

  844. snow_inner_add_yblock_mmx_accum("1", "520", "8")

  845. snow_inner_add_yblock_mmx_accum("0", "536", "8")

  846. snow_inner_add_yblock_mmx_mix("16", "8")

  847. snow_inner_add_yblock_mmx_end("32")

  848. }

  849. 

  850. void ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,

  851.                            int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){

  852. 

  853.     if (b_w == 16)

  854.         inner_add_yblock_bw_16_obmc_32_sse2(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);

  855.     else if (b_w == 8 && obmc_stride == 16) {

  856.         if (!(b_h & 1))

  857.             inner_add_yblock_bw_8_obmc_16_bh_even_sse2(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);

  858.         else

  859.             inner_add_yblock_bw_8_obmc_16_mmx(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);

  860.     } else

  861.          ff_snow_inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);

  862. }

  863. 

  864. void ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,

  865.                           int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){

  866.     if (b_w == 16)

  867.         inner_add_yblock_bw_16_obmc_32_mmx(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);

  868.     else if (b_w == 8 && obmc_stride == 16)

  869.         inner_add_yblock_bw_8_obmc_16_mmx(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);

  870.     else

  871.         ff_snow_inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);

  872. }