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.
10676 Commits
32 Branches
283MB
Tree: 73b46a95d2
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '73b46a95d2'
${ noResults }
FFmpeg/libavcodec/i386/snowdsp_mmx.c

893 lines
40KB
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 "avcodec.h"

  23. #include "snow.h"

  24. #include "x86_cpu.h"

  25. 

  26. void ff_snow_horizontal_compose97i_sse2(DWTELEM *b, int width){

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

  28.     // SSE2 code runs faster with pointers aligned on a 32-byte boundary.

  29.     DWTELEM temp_buf[(width>>1) + 4];

  30.     DWTELEM * const temp = temp_buf + 4 - (((int)temp_buf & 0xF) >> 2);

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

  32.     const int w_r= w2 - 1;

  33.     int i;

  34. 

  35.     { // Lift 0

  36.         DWTELEM * const ref = b + w2 - 1;

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

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

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

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

  41. 

  42.         i = 0;

  43.         asm volatile(

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

  45.             "pslld        $31, %%xmm7         \n\t"

  46.             "psrld        $29, %%xmm7         \n\t"

  47.         ::);

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

  49.             asm volatile(

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

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

  52.                 "movdqu  4(%1), %%xmm2        \n\t"

  53.                 "movdqu 20(%1), %%xmm6        \n\t"

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

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

  56.                 "movdqa %%xmm2, %%xmm0        \n\t"

  57.                 "movdqa %%xmm6, %%xmm4        \n\t"

  58.                 "paddd  %%xmm2, %%xmm2        \n\t"

  59.                 "paddd  %%xmm6, %%xmm6        \n\t"

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

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

  62.                 "paddd  %%xmm7, %%xmm2        \n\t"

  63.                 "paddd  %%xmm7, %%xmm6        \n\t"

  64.                 "psrad      $3, %%xmm2        \n\t"

  65.                 "psrad      $3, %%xmm6        \n\t"

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

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

  68.                 "psubd  %%xmm2, %%xmm0        \n\t"

  69.                 "psubd  %%xmm6, %%xmm4        \n\t"

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

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

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

  73.                 : "memory"

  74.             );

  75.         }

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

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

  78.     }

  79. 

  80.     { // Lift 1

  81.         DWTELEM * const dst = b+w2;

  82. 

  83.         i = 0;

  84.         for(; (((long)&dst[i]) & 0xF) && i<w_r; i++){

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

  86.         }

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

  88.             asm volatile(

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

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

  91.                 "movdqu  4(%1), %%xmm2        \n\t"

  92.                 "movdqu 20(%1), %%xmm6        \n\t"

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

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

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

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

  97.                 "psubd  %%xmm2, %%xmm0        \n\t"

  98.                 "psubd  %%xmm6, %%xmm4        \n\t"

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

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

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

  102.                 : "memory"

  103.             );

  104.         }

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

  106.     }

  107. 

  108.     { // Lift 2

  109.         DWTELEM * const ref = b+w2 - 1;

  110.         DWTELEM b_0 = b[0];

  111. 

  112.         i = 0;

  113.         asm volatile(

  114.             "pslld          $1, %%xmm7        \n\t"

  115.         ::);

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

  117.             asm volatile(

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

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

  120.                 "movdqu  4(%1), %%xmm0        \n\t"

  121.                 "movdqu 20(%1), %%xmm4        \n\t" //FIXME try aligned reads and shifts

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

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

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

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

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

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

  128.                 "psrad      $2, %%xmm0        \n\t"

  129.                 "psrad      $2, %%xmm4        \n\t"

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

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

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

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

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

  135.                 "paddd  %%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.         DWTELEM * const src = b+w2;

  148. 

  149.         i = 0;

  150.         for(; (((long)&temp[i]) & 0xF) && 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  4(%1), %%xmm2        \n\t"

  156.                 "movdqu 20(%1), %%xmm6        \n\t"

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

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

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

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

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

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

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

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

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

  166.                 "paddd  %%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 & 0x1E) != 0x1E; i-=2){

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

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

  182.         }

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

  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.                 "punpckldq   (%2), %%xmm0       \n\t"

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

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

  196.                 "punpckldq 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.                 "punpckhdq   (%2), %%xmm1       \n\t"

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

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

  204.                 "punpckhdq 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(DWTELEM *b, int width){

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

  218.     DWTELEM 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.         DWTELEM * 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.             "pcmpeqd    %%mm7, %%mm7         \n\t"

  230.             "pslld        $31, %%mm7         \n\t"

  231.             "psrld        $29, %%mm7         \n\t"

  232.            ::);

  233.         for(; i<w_l-3; i+=4){

  234.             asm volatile(

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

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

  237.                 "paddd   4(%1), %%mm2        \n\t"

  238.                 "paddd  12(%1), %%mm6        \n\t"

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

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

  241.                 "paddd   %%mm2, %%mm2        \n\t"

  242.                 "paddd   %%mm6, %%mm6        \n\t"

  243.                 "paddd   %%mm0, %%mm2        \n\t"

  244.                 "paddd   %%mm4, %%mm6        \n\t"

  245.                 "paddd   %%mm7, %%mm2        \n\t"

  246.                 "paddd   %%mm7, %%mm6        \n\t"

  247.                 "psrad      $3, %%mm2        \n\t"

  248.                 "psrad      $3, %%mm6        \n\t"

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

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

  251.                 "psubd   %%mm2, %%mm0        \n\t"

  252.                 "psubd   %%mm6, %%mm4        \n\t"

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

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

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

  256.                  : "memory"

  257.                );

  258.         }

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

  260.     }

  261. 

  262.     { // Lift 1

  263.         DWTELEM * const dst = b+w2;

  264. 

  265.         i = 0;

  266.         for(; i<w_r-3; i+=4){

  267.             asm volatile(

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

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

  270.                 "paddd   4(%1), %%mm2        \n\t"

  271.                 "paddd  12(%1), %%mm6        \n\t"

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

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

  274.                 "psubd   %%mm2, %%mm0        \n\t"

  275.                 "psubd   %%mm6, %%mm4        \n\t"

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

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

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

  279.                  : "memory"

  280.                );

  281.         }

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

  283.     }

  284. 

  285.     { // Lift 2

  286.         DWTELEM * const ref = b+w2 - 1;

  287. 

  288.         i = 1;

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

  290.         asm volatile(

  291.             "pslld          $1, %%mm7        \n\t"

  292.            ::);

  293.         for(; i<w_l-3; i+=4){

  294.             asm volatile(

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

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

  297.                 "paddd   4(%1), %%mm0        \n\t"

  298.                 "paddd  12(%1), %%mm4        \n\t"

  299.                 "paddd   %%mm7, %%mm0        \n\t"

  300.                 "paddd   %%mm7, %%mm4        \n\t"

  301.                 "psrad      $2, %%mm0        \n\t"

  302.                 "psrad      $2, %%mm4        \n\t"

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

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

  305.                 "paddd   %%mm1, %%mm0        \n\t"

  306.                 "paddd   %%mm5, %%mm4        \n\t"

  307.                 "psrad      $2, %%mm0        \n\t"

  308.                 "psrad      $2, %%mm4        \n\t"

  309.                 "paddd   %%mm1, %%mm0        \n\t"

  310.                 "paddd   %%mm5, %%mm4        \n\t"

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

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

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

  314.                  : "memory"

  315.                );

  316.         }

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

  318.     }

  319. 

  320.     { // Lift 3

  321.         DWTELEM * const src = b+w2;

  322.         i = 0;

  323. 

  324.         for(; i<w_r-3; i+=4){

  325.             asm volatile(

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

  327.                 "movq   12(%1), %%mm6        \n\t"

  328.                 "paddd    (%1), %%mm2        \n\t"

  329.                 "paddd   8(%1), %%mm6        \n\t"

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

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

  332.                 "paddd   %%mm2, %%mm0        \n\t"

  333.                 "paddd   %%mm6, %%mm4        \n\t"

  334.                 "psrad      $1, %%mm2        \n\t"

  335.                 "psrad      $1, %%mm6        \n\t"

  336.                 "paddd   %%mm0, %%mm2        \n\t"

  337.                 "paddd   %%mm4, %%mm6        \n\t"

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

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

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

  341.                  : "memory"

  342.                );

  343.         }

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

  345.     }

  346. 

  347.     {

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

  349. 

  350.         for (; (i & 0xE) != 0xE; i-=2){

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

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

  353.         }

  354.         for (i-=14; i>=0; i-=16){

  355.             asm volatile(

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

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

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

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

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

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

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

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

  364.                 "punpckldq   (%2), %%mm0       \n\t"

  365.                 "punpckldq  8(%2), %%mm2       \n\t"

  366.                 "punpckldq 16(%2), %%mm4       \n\t"

  367.                 "punpckldq 24(%2), %%mm6       \n\t"

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

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

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

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

  372.                 "punpckhdq   (%2), %%mm1       \n\t"

  373.                 "punpckhdq  8(%2), %%mm3       \n\t"

  374.                 "punpckhdq 16(%2), %%mm5       \n\t"

  375.                 "punpckhdq 24(%2), %%mm7       \n\t"

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

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

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

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

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

  381.                  : "memory"

  382.                );

  383.         }

  384.     }

  385. }

  386. 

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

  388.         ""op" (%%"r",%%"REG_d",4), %%"t0"      \n\t"\

  389.         ""op" 16(%%"r",%%"REG_d",4), %%"t1"    \n\t"\

  390.         ""op" 32(%%"r",%%"REG_d",4), %%"t2"    \n\t"\

  391.         ""op" 48(%%"r",%%"REG_d",4), %%"t3"    \n\t"

  392. 

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

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

  395. 

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

  397.         snow_vertical_compose_sse2_load_add("paddd",r,t0,t1,t2,t3)

  398. 

  399. #define snow_vertical_compose_sse2_sub(s0,s1,s2,s3,t0,t1,t2,t3)\

  400.         "psubd %%"s0", %%"t0" \n\t"\

  401.         "psubd %%"s1", %%"t1" \n\t"\

  402.         "psubd %%"s2", %%"t2" \n\t"\

  403.         "psubd %%"s3", %%"t3" \n\t"

  404. 

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

  406.         "movdqa %%"s0", (%%"w",%%"REG_d",4)      \n\t"\

  407.         "movdqa %%"s1", 16(%%"w",%%"REG_d",4)    \n\t"\

  408.         "movdqa %%"s2", 32(%%"w",%%"REG_d",4)    \n\t"\

  409.         "movdqa %%"s3", 48(%%"w",%%"REG_d",4)    \n\t"

  410. 

  411. #define snow_vertical_compose_sse2_sra(n,t0,t1,t2,t3)\

  412.         "psrad $"n", %%"t0" \n\t"\

  413.         "psrad $"n", %%"t1" \n\t"\

  414.         "psrad $"n", %%"t2" \n\t"\

  415.         "psrad $"n", %%"t3" \n\t"

  416. 

  417. #define snow_vertical_compose_sse2_r2r_add(s0,s1,s2,s3,t0,t1,t2,t3)\

  418.         "paddd %%"s0", %%"t0" \n\t"\

  419.         "paddd %%"s1", %%"t1" \n\t"\

  420.         "paddd %%"s2", %%"t2" \n\t"\

  421.         "paddd %%"s3", %%"t3" \n\t"

  422. 

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

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

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

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

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

  428. 

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

  430.     long i = width;

  431. 

  432.     while(i & 0xF)

  433.     {

  434.         i--;

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

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

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

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

  439.     }

  440. 

  441.          asm volatile (

  442.         "jmp 2f                                      \n\t"

  443.         "1:                                          \n\t"

  444. 

  445.         "mov %6, %%"REG_a"                           \n\t"

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

  447. 

  448.         snow_vertical_compose_sse2_load(REG_S,"xmm0","xmm2","xmm4","xmm6")

  449.         snow_vertical_compose_sse2_add(REG_a,"xmm0","xmm2","xmm4","xmm6")

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

  451.         snow_vertical_compose_sse2_r2r_add("xmm0","xmm2","xmm4","xmm6","xmm0","xmm2","xmm4","xmm6")

  452.         snow_vertical_compose_sse2_r2r_add("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")

  453. 

  454.         "pcmpeqd %%xmm1, %%xmm1                      \n\t"

  455.         "pslld $31, %%xmm1                           \n\t"

  456.         "psrld $29, %%xmm1                           \n\t"

  457.         "mov %5, %%"REG_a"                           \n\t"

  458. 

  459.         snow_vertical_compose_sse2_r2r_add("xmm1","xmm1","xmm1","xmm1","xmm0","xmm2","xmm4","xmm6")

  460.         snow_vertical_compose_sse2_sra("3","xmm0","xmm2","xmm4","xmm6")

  461.         snow_vertical_compose_sse2_load(REG_a,"xmm1","xmm3","xmm5","xmm7")

  462.         snow_vertical_compose_sse2_sub("xmm0","xmm2","xmm4","xmm6","xmm1","xmm3","xmm5","xmm7")

  463.         snow_vertical_compose_sse2_store(REG_a,"xmm1","xmm3","xmm5","xmm7")

  464.         "mov %3, %%"REG_c"                           \n\t"

  465.         snow_vertical_compose_sse2_load(REG_S,"xmm0","xmm2","xmm4","xmm6")

  466.         snow_vertical_compose_sse2_add(REG_c,"xmm1","xmm3","xmm5","xmm7")

  467.         snow_vertical_compose_sse2_sub("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")

  468.         snow_vertical_compose_sse2_store(REG_S,"xmm0","xmm2","xmm4","xmm6")

  469.         "mov %2, %%"REG_a"                           \n\t"

  470.         snow_vertical_compose_sse2_add(REG_a,"xmm0","xmm2","xmm4","xmm6")

  471.         snow_vertical_compose_sse2_sra("2","xmm0","xmm2","xmm4","xmm6")

  472.         snow_vertical_compose_sse2_add(REG_c,"xmm0","xmm2","xmm4","xmm6")

  473. 

  474.         "pcmpeqd %%xmm1, %%xmm1                      \n\t"

  475.         "pslld $31, %%xmm1                           \n\t"

  476.         "psrld $30, %%xmm1                           \n\t"

  477.         "mov %1, %%"REG_S"                           \n\t"

  478. 

  479.         snow_vertical_compose_sse2_r2r_add("xmm1","xmm1","xmm1","xmm1","xmm0","xmm2","xmm4","xmm6")

  480.         snow_vertical_compose_sse2_sra("2","xmm0","xmm2","xmm4","xmm6")

  481.         snow_vertical_compose_sse2_add(REG_c,"xmm0","xmm2","xmm4","xmm6")

  482.         snow_vertical_compose_sse2_store(REG_c,"xmm0","xmm2","xmm4","xmm6")

  483.         snow_vertical_compose_sse2_add(REG_S,"xmm0","xmm2","xmm4","xmm6")

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

  485.         snow_vertical_compose_sse2_sra("1","xmm0","xmm2","xmm4","xmm6")

  486.         snow_vertical_compose_sse2_r2r_add("xmm1","xmm3","xmm5","xmm7","xmm0","xmm2","xmm4","xmm6")

  487.         snow_vertical_compose_sse2_add(REG_a,"xmm0","xmm2","xmm4","xmm6")

  488.         snow_vertical_compose_sse2_store(REG_a,"xmm0","xmm2","xmm4","xmm6")

  489. 

  490.         "2:                                          \n\t"

  491.         "sub $16, %%"REG_d"                          \n\t"

  492.         "jge 1b                                      \n\t"

  493.         :"+d"(i)

  494.         :

  495.         "m"(b0),"m"(b1),"m"(b2),"m"(b3),"m"(b4),"m"(b5):

  496.         "%"REG_a"","%"REG_S"","%"REG_c"");

  497. }

  498. 

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

  500.         ""op" (%%"r",%%"REG_d",4), %%"t0"   \n\t"\

  501.         ""op" 8(%%"r",%%"REG_d",4), %%"t1"  \n\t"\

  502.         ""op" 16(%%"r",%%"REG_d",4), %%"t2" \n\t"\

  503.         ""op" 24(%%"r",%%"REG_d",4), %%"t3" \n\t"

  504. 

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

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

  507. 

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

  509.         snow_vertical_compose_mmx_load_add("paddd",r,t0,t1,t2,t3)

  510. 

  511. #define snow_vertical_compose_mmx_sub(s0,s1,s2,s3,t0,t1,t2,t3)\

  512.         snow_vertical_compose_sse2_sub(s0,s1,s2,s3,t0,t1,t2,t3)

  513. 

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

  515.         "movq %%"s0", (%%"w",%%"REG_d",4)   \n\t"\

  516.         "movq %%"s1", 8(%%"w",%%"REG_d",4)  \n\t"\

  517.         "movq %%"s2", 16(%%"w",%%"REG_d",4) \n\t"\

  518.         "movq %%"s3", 24(%%"w",%%"REG_d",4) \n\t"

  519. 

  520. #define snow_vertical_compose_mmx_sra(n,t0,t1,t2,t3)\

  521.         snow_vertical_compose_sse2_sra(n,t0,t1,t2,t3)

  522. 

  523. #define snow_vertical_compose_mmx_r2r_add(s0,s1,s2,s3,t0,t1,t2,t3)\

  524.         snow_vertical_compose_sse2_r2r_add(s0,s1,s2,s3,t0,t1,t2,t3)

  525. 

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

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

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

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

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

  531. 

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

  533.     long i = width;

  534.     while(i & 0x7)

  535.     {

  536.         i--;

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

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

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

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

  541.     }

  542. 

  543.     asm volatile(

  544.         "jmp 2f                                      \n\t"

  545.         "1:                                          \n\t"

  546. 

  547.         "mov %6, %%"REG_a"                           \n\t"

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

  549. 

  550.         snow_vertical_compose_mmx_load(REG_S,"mm0","mm2","mm4","mm6")

  551.         snow_vertical_compose_mmx_add(REG_a,"mm0","mm2","mm4","mm6")

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

  553.         snow_vertical_compose_mmx_r2r_add("mm0","mm2","mm4","mm6","mm0","mm2","mm4","mm6")

  554.         snow_vertical_compose_mmx_r2r_add("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")

  555. 

  556.         "pcmpeqd %%mm1, %%mm1                        \n\t"

  557.         "pslld $31, %%mm1                            \n\t"

  558.         "psrld $29, %%mm1                            \n\t"

  559.         "mov %5, %%"REG_a"                           \n\t"

  560. 

  561.         snow_vertical_compose_mmx_r2r_add("mm1","mm1","mm1","mm1","mm0","mm2","mm4","mm6")

  562.         snow_vertical_compose_mmx_sra("3","mm0","mm2","mm4","mm6")

  563.         snow_vertical_compose_mmx_load(REG_a,"mm1","mm3","mm5","mm7")

  564.         snow_vertical_compose_mmx_sub("mm0","mm2","mm4","mm6","mm1","mm3","mm5","mm7")

  565.         snow_vertical_compose_mmx_store(REG_a,"mm1","mm3","mm5","mm7")

  566.         "mov %3, %%"REG_c"                           \n\t"

  567.         snow_vertical_compose_mmx_load(REG_S,"mm0","mm2","mm4","mm6")

  568.         snow_vertical_compose_mmx_add(REG_c,"mm1","mm3","mm5","mm7")

  569.         snow_vertical_compose_mmx_sub("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")

  570.         snow_vertical_compose_mmx_store(REG_S,"mm0","mm2","mm4","mm6")

  571.         "mov %2, %%"REG_a"                           \n\t"

  572.         snow_vertical_compose_mmx_add(REG_a,"mm0","mm2","mm4","mm6")

  573.         snow_vertical_compose_mmx_sra("2","mm0","mm2","mm4","mm6")

  574.         snow_vertical_compose_mmx_add(REG_c,"mm0","mm2","mm4","mm6")

  575. 

  576.         "pcmpeqd %%mm1, %%mm1                        \n\t"

  577.         "pslld $31, %%mm1                            \n\t"

  578.         "psrld $30, %%mm1                            \n\t"

  579.         "mov %1, %%"REG_S"                           \n\t"

  580. 

  581.         snow_vertical_compose_mmx_r2r_add("mm1","mm1","mm1","mm1","mm0","mm2","mm4","mm6")

  582.         snow_vertical_compose_mmx_sra("2","mm0","mm2","mm4","mm6")

  583.         snow_vertical_compose_mmx_add(REG_c,"mm0","mm2","mm4","mm6")

  584.         snow_vertical_compose_mmx_store(REG_c,"mm0","mm2","mm4","mm6")

  585.         snow_vertical_compose_mmx_add(REG_S,"mm0","mm2","mm4","mm6")

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

  587.         snow_vertical_compose_mmx_sra("1","mm0","mm2","mm4","mm6")

  588.         snow_vertical_compose_mmx_r2r_add("mm1","mm3","mm5","mm7","mm0","mm2","mm4","mm6")

  589.         snow_vertical_compose_mmx_add(REG_a,"mm0","mm2","mm4","mm6")

  590.         snow_vertical_compose_mmx_store(REG_a,"mm0","mm2","mm4","mm6")

  591. 

  592.         "2:                                          \n\t"

  593.         "sub $8, %%"REG_d"                           \n\t"

  594.         "jge 1b                                      \n\t"

  595.         :"+d"(i)

  596.         :

  597.         "m"(b0),"m"(b1),"m"(b2),"m"(b3),"m"(b4),"m"(b5):

  598.         "%"REG_a"","%"REG_S"","%"REG_c"");

  599. }

  600. 

  601. #define snow_inner_add_yblock_sse2_header \

  602.     DWTELEM * * dst_array = sb->line + src_y;\

  603.     long tmp;\

  604.     asm volatile(\

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

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

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

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

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

  610.              "pslld $31, %%xmm3              \n\t"\

  611.              "psrld $24, %%xmm3              \n\t" /* FRAC_BITS >> 1 */\

  612.              "1:                             \n\t"\

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

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

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

  616. 

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

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

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

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

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

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

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

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

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

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

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

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

  629. 

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

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

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

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

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

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

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

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

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

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

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

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

  642. 

  643. #define snow_inner_add_yblock_sse2_accum_8(ptr_offset, s_offset) \

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

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

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

  647. 

  648. #define snow_inner_add_yblock_sse2_accum_16(ptr_offset, s_offset) \

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

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

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

  652. 

  653. #define snow_inner_add_yblock_sse2_end_common1\

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

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

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

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

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

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

  660. 

  661. #define snow_inner_add_yblock_sse2_end_common2\

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

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

  664.              :\

  665.              "rm"((long)(src_x<<2)),"m"(obmc),"a"(block),"m"((long)b_h),"m"((long)src_stride):\

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

  667. 

  668. #define snow_inner_add_yblock_sse2_end_8\

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

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

  671.              snow_inner_add_yblock_sse2_end_common1\

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

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

  674.              snow_inner_add_yblock_sse2_end_common2

  675. 

  676. #define snow_inner_add_yblock_sse2_end_16\

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

  678.              snow_inner_add_yblock_sse2_end_common1\

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

  680.              snow_inner_add_yblock_sse2_end_common2

  681. 

  682. static void inner_add_yblock_bw_8_obmc_16_bh_even_sse2(const uint8_t *obmc, const long obmc_stride, uint8_t * * block, int b_w, long b_h,

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

  684. snow_inner_add_yblock_sse2_header

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

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

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

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

  689. 

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

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

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

  693. 

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

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

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

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

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

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

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

  701. 

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

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

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

  705. 

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

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

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

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

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

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

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

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

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

  715. 

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

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

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

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

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

  721. 

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

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

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

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

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

  727. snow_inner_add_yblock_sse2_end_8

  728. }

  729. 

  730. static void inner_add_yblock_bw_16_obmc_32_sse2(const uint8_t *obmc, const long obmc_stride, uint8_t * * block, int b_w, long b_h,

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

  732. snow_inner_add_yblock_sse2_header

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

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

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

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

  737. 

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

  739.              "movdqa %%xmm1, %%xmm0          \n\t"

  740.              "movdqa %%xmm5, %%xmm4          \n\t"

  741.              "punpcklwd %%xmm7, %%xmm0       \n\t"

  742.              "paddd (%%"REG_D"), %%xmm0      \n\t"

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

  744.              "paddd 16(%%"REG_D"), %%xmm1    \n\t"

  745.              "punpcklwd %%xmm7, %%xmm4       \n\t"

  746.              "paddd 32(%%"REG_D"), %%xmm4    \n\t"

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

  748.              "paddd 48(%%"REG_D"), %%xmm5    \n\t"

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

  750.              "paddd %%xmm3, %%xmm1           \n\t"

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

  752.              "paddd %%xmm3, %%xmm5           \n\t"

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

  754.              "psrad $8, %%xmm1               \n\t" /* FRAC_BITS. */

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

  756.              "psrad $8, %%xmm5               \n\t" /* FRAC_BITS. */

  757. 

  758.              "packssdw %%xmm1, %%xmm0        \n\t"

  759.              "packssdw %%xmm5, %%xmm4        \n\t"

  760.              "packuswb %%xmm4, %%xmm0        \n\t"

  761. 

  762.              "movdqu %%xmm0, (%%"REG_d")       \n\t"

  763. 

  764. snow_inner_add_yblock_sse2_end_16

  765. }

  766. 

  767. #define snow_inner_add_yblock_mmx_header \

  768.     DWTELEM * * dst_array = sb->line + src_y;\

  769.     long tmp;\

  770.     asm volatile(\

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

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

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

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

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

  776.              "pslld $31, %%mm3               \n\t"\

  777.              "psrld $24, %%mm3               \n\t" /* FRAC_BITS >> 1 */\

  778.              "1:                             \n\t"\

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

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

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

  782. 

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

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

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

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

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

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

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

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

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

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

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

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

  795. 

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

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

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

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

  800. 

  801. #define snow_inner_add_yblock_mmx_mix(read_offset, write_offset)\

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

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

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

  805.              "punpcklwd %%mm7, %%mm0         \n\t"\

  806.              "paddd "read_offset"(%%"REG_D"), %%mm0 \n\t"\

  807.              "punpckhwd %%mm7, %%mm1         \n\t"\

  808.              "paddd "read_offset"+8(%%"REG_D"), %%mm1 \n\t"\

  809.              "punpcklwd %%mm7, %%mm4         \n\t"\

  810.              "paddd "read_offset"+16(%%"REG_D"), %%mm4 \n\t"\

  811.              "punpckhwd %%mm7, %%mm5         \n\t"\

  812.              "paddd "read_offset"+24(%%"REG_D"), %%mm5 \n\t"\

  813.              "paddd %%mm3, %%mm0             \n\t"\

  814.              "paddd %%mm3, %%mm1             \n\t"\

  815.              "paddd %%mm3, %%mm4             \n\t"\

  816.              "paddd %%mm3, %%mm5             \n\t"\

  817.              "psrad $8, %%mm0                \n\t"\

  818.              "psrad $8, %%mm1                \n\t"\

  819.              "psrad $8, %%mm4                \n\t"\

  820.              "psrad $8, %%mm5                \n\t"\

  821. \

  822.              "packssdw %%mm1, %%mm0          \n\t"\

  823.              "packssdw %%mm5, %%mm4          \n\t"\

  824.              "packuswb %%mm4, %%mm0          \n\t"\

  825.              "movq %%mm0, "write_offset"(%%"REG_d") \n\t"

  826. 

  827. #define snow_inner_add_yblock_mmx_end(s_step)\

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

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

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

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

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

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

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

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

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

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

  838.              :\

  839.              "rm"((long)(src_x<<2)),"m"(obmc),"a"(block),"m"((long)b_h),"m"((long)src_stride):\

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

  841. 

  842. static void inner_add_yblock_bw_8_obmc_16_mmx(const uint8_t *obmc, const long obmc_stride, uint8_t * * block, int b_w, long b_h,

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

  844. snow_inner_add_yblock_mmx_header

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

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

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

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

  849. snow_inner_add_yblock_mmx_mix("0", "0")

  850. snow_inner_add_yblock_mmx_end("16")

  851. }

  852. 

  853. static void inner_add_yblock_bw_16_obmc_32_mmx(const uint8_t *obmc, const long obmc_stride, uint8_t * * block, int b_w, long b_h,

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

  855. snow_inner_add_yblock_mmx_header

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

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

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

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

  860. snow_inner_add_yblock_mmx_mix("0", "0")

  861. 

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

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

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

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

  866. snow_inner_add_yblock_mmx_mix("32", "8")

  867. snow_inner_add_yblock_mmx_end("32")

  868. }

  869. 

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

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

  872. 

  873.     if (b_w == 16)

  874.         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);

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

  876.         if (!(b_h & 1))

  877.             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);

  878.         else

  879.             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);

  880.     } else

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

  882. }

  883. 

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

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

  886.     if (b_w == 16)

  887.         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);

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

  889.         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);

  890.     else

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

  892. }