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.
39979 Commits
32 Branches
821MB
Tree: 593886b5dd
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '593886b5dd'
${ noResults }
FFmpeg/libavcodec/x86/dsputilenc_mmx.c

1030 lines
31KB
Raw Blame History 

  1. /*

  2.  * MMX optimized DSP utils

  3.  * Copyright (c) 2000, 2001 Fabrice Bellard

  4.  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>

  5.  *

  6.  * MMX optimization by Nick Kurshev <nickols_k@mail.ru>

  7.  *

  8.  * This file is part of Libav.

  9.  *

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

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

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

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

  14.  *

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

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

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

  18.  * Lesser General Public License for more details.

  19.  *

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

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

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

  23.  */

  24. 

  25. #include "libavutil/attributes.h"

  26. #include "libavutil/cpu.h"

  27. #include "libavutil/x86/asm.h"

  28. #include "libavutil/x86/cpu.h"

  29. #include "libavcodec/dct.h"

  30. #include "libavcodec/dsputil.h"

  31. #include "libavcodec/mpegvideo.h"

  32. #include "dsputil_x86.h"

  33. 

  34. void ff_get_pixels_mmx(int16_t *block, const uint8_t *pixels, int line_size);

  35. void ff_get_pixels_sse2(int16_t *block, const uint8_t *pixels, int line_size);

  36. void ff_diff_pixels_mmx(int16_t *block, const uint8_t *s1, const uint8_t *s2,

  37.                         int stride);

  38. int ff_pix_sum16_mmx(uint8_t *pix, int line_size);

  39. int ff_pix_norm1_mmx(uint8_t *pix, int line_size);

  40. 

  41. #if HAVE_INLINE_ASM

  42. 

  43. static int sse8_mmx(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  44.                     int line_size, int h)

  45. {

  46.     int tmp;

  47. 

  48.     __asm__ volatile (

  49.         "movl         %4, %%ecx          \n"

  50.         "shr          $1, %%ecx          \n"

  51.         "pxor      %%mm0, %%mm0          \n" /* mm0 = 0 */

  52.         "pxor      %%mm7, %%mm7          \n" /* mm7 holds the sum */

  53.         "1:                              \n"

  54.         "movq       (%0), %%mm1          \n" /* mm1 = pix1[0][0 - 7] */

  55.         "movq       (%1), %%mm2          \n" /* mm2 = pix2[0][0 - 7] */

  56.         "movq   (%0, %3), %%mm3          \n" /* mm3 = pix1[1][0 - 7] */

  57.         "movq   (%1, %3), %%mm4          \n" /* mm4 = pix2[1][0 - 7] */

  58. 

  59.         /* todo: mm1-mm2, mm3-mm4 */

  60.         /* algo: subtract mm1 from mm2 with saturation and vice versa */

  61.         /*       OR the results to get absolute difference */

  62.         "movq      %%mm1, %%mm5          \n"

  63.         "movq      %%mm3, %%mm6          \n"

  64.         "psubusb   %%mm2, %%mm1          \n"

  65.         "psubusb   %%mm4, %%mm3          \n"

  66.         "psubusb   %%mm5, %%mm2          \n"

  67.         "psubusb   %%mm6, %%mm4          \n"

  68. 

  69.         "por       %%mm1, %%mm2          \n"

  70.         "por       %%mm3, %%mm4          \n"

  71. 

  72.         /* now convert to 16-bit vectors so we can square them */

  73.         "movq      %%mm2, %%mm1          \n"

  74.         "movq      %%mm4, %%mm3          \n"

  75. 

  76.         "punpckhbw %%mm0, %%mm2          \n"

  77.         "punpckhbw %%mm0, %%mm4          \n"

  78.         "punpcklbw %%mm0, %%mm1          \n" /* mm1 now spread over (mm1, mm2) */

  79.         "punpcklbw %%mm0, %%mm3          \n" /* mm4 now spread over (mm3, mm4) */

  80. 

  81.         "pmaddwd   %%mm2, %%mm2          \n"

  82.         "pmaddwd   %%mm4, %%mm4          \n"

  83.         "pmaddwd   %%mm1, %%mm1          \n"

  84.         "pmaddwd   %%mm3, %%mm3          \n"

  85. 

  86.         "lea (%0, %3, 2), %0             \n" /* pix1 += 2 * line_size */

  87.         "lea (%1, %3, 2), %1             \n" /* pix2 += 2 * line_size */

  88. 

  89.         "paddd     %%mm2, %%mm1          \n"

  90.         "paddd     %%mm4, %%mm3          \n"

  91.         "paddd     %%mm1, %%mm7          \n"

  92.         "paddd     %%mm3, %%mm7          \n"

  93. 

  94.         "decl      %%ecx                 \n"

  95.         "jnz       1b                    \n"

  96. 

  97.         "movq      %%mm7, %%mm1          \n"

  98.         "psrlq       $32, %%mm7          \n" /* shift hi dword to lo */

  99.         "paddd     %%mm7, %%mm1          \n"

  100.         "movd      %%mm1, %2             \n"

  101.         : "+r" (pix1), "+r" (pix2), "=r" (tmp)

  102.         : "r" ((x86_reg) line_size), "m" (h)

  103.         : "%ecx");

  104. 

  105.     return tmp;

  106. }

  107. 

  108. static int sse16_mmx(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  109.                      int line_size, int h)

  110. {

  111.     int tmp;

  112. 

  113.     __asm__ volatile (

  114.         "movl %4, %%ecx\n"

  115.         "pxor %%mm0, %%mm0\n"    /* mm0 = 0 */

  116.         "pxor %%mm7, %%mm7\n"    /* mm7 holds the sum */

  117.         "1:\n"

  118.         "movq (%0), %%mm1\n"     /* mm1 = pix1[0 -  7] */

  119.         "movq (%1), %%mm2\n"     /* mm2 = pix2[0 -  7] */

  120.         "movq 8(%0), %%mm3\n"    /* mm3 = pix1[8 - 15] */

  121.         "movq 8(%1), %%mm4\n"    /* mm4 = pix2[8 - 15] */

  122. 

  123.         /* todo: mm1-mm2, mm3-mm4 */

  124.         /* algo: subtract mm1 from mm2 with saturation and vice versa */

  125.         /*       OR the results to get absolute difference */

  126.         "movq %%mm1, %%mm5\n"

  127.         "movq %%mm3, %%mm6\n"

  128.         "psubusb %%mm2, %%mm1\n"

  129.         "psubusb %%mm4, %%mm3\n"

  130.         "psubusb %%mm5, %%mm2\n"

  131.         "psubusb %%mm6, %%mm4\n"

  132. 

  133.         "por %%mm1, %%mm2\n"

  134.         "por %%mm3, %%mm4\n"

  135. 

  136.         /* now convert to 16-bit vectors so we can square them */

  137.         "movq %%mm2, %%mm1\n"

  138.         "movq %%mm4, %%mm3\n"

  139. 

  140.         "punpckhbw %%mm0, %%mm2\n"

  141.         "punpckhbw %%mm0, %%mm4\n"

  142.         "punpcklbw %%mm0, %%mm1\n" /* mm1 now spread over (mm1, mm2) */

  143.         "punpcklbw %%mm0, %%mm3\n" /* mm4 now spread over (mm3, mm4) */

  144. 

  145.         "pmaddwd %%mm2, %%mm2\n"

  146.         "pmaddwd %%mm4, %%mm4\n"

  147.         "pmaddwd %%mm1, %%mm1\n"

  148.         "pmaddwd %%mm3, %%mm3\n"

  149. 

  150.         "add %3, %0\n"

  151.         "add %3, %1\n"

  152. 

  153.         "paddd %%mm2, %%mm1\n"

  154.         "paddd %%mm4, %%mm3\n"

  155.         "paddd %%mm1, %%mm7\n"

  156.         "paddd %%mm3, %%mm7\n"

  157. 

  158.         "decl %%ecx\n"

  159.         "jnz 1b\n"

  160. 

  161.         "movq %%mm7, %%mm1\n"

  162.         "psrlq $32, %%mm7\n"    /* shift hi dword to lo */

  163.         "paddd %%mm7, %%mm1\n"

  164.         "movd %%mm1, %2\n"

  165.         : "+r" (pix1), "+r" (pix2), "=r" (tmp)

  166.         : "r" ((x86_reg) line_size), "m" (h)

  167.         : "%ecx");

  168. 

  169.     return tmp;

  170. }

  171. 

  172. static int hf_noise8_mmx(uint8_t *pix1, int line_size, int h)

  173. {

  174.     int tmp;

  175. 

  176.     __asm__ volatile (

  177.         "movl %3, %%ecx\n"

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

  179.         "pxor %%mm6, %%mm6\n"

  180. 

  181.         "movq (%0), %%mm0\n"

  182.         "movq %%mm0, %%mm1\n"

  183.         "psllq $8, %%mm0\n"

  184.         "psrlq $8, %%mm1\n"

  185.         "psrlq $8, %%mm0\n"

  186.         "movq %%mm0, %%mm2\n"

  187.         "movq %%mm1, %%mm3\n"

  188.         "punpcklbw %%mm7, %%mm0\n"

  189.         "punpcklbw %%mm7, %%mm1\n"

  190.         "punpckhbw %%mm7, %%mm2\n"

  191.         "punpckhbw %%mm7, %%mm3\n"

  192.         "psubw %%mm1, %%mm0\n"

  193.         "psubw %%mm3, %%mm2\n"

  194. 

  195.         "add %2, %0\n"

  196. 

  197.         "movq (%0), %%mm4\n"

  198.         "movq %%mm4, %%mm1\n"

  199.         "psllq $8, %%mm4\n"

  200.         "psrlq $8, %%mm1\n"

  201.         "psrlq $8, %%mm4\n"

  202.         "movq %%mm4, %%mm5\n"

  203.         "movq %%mm1, %%mm3\n"

  204.         "punpcklbw %%mm7, %%mm4\n"

  205.         "punpcklbw %%mm7, %%mm1\n"

  206.         "punpckhbw %%mm7, %%mm5\n"

  207.         "punpckhbw %%mm7, %%mm3\n"

  208.         "psubw %%mm1, %%mm4\n"

  209.         "psubw %%mm3, %%mm5\n"

  210.         "psubw %%mm4, %%mm0\n"

  211.         "psubw %%mm5, %%mm2\n"

  212.         "pxor %%mm3, %%mm3\n"

  213.         "pxor %%mm1, %%mm1\n"

  214.         "pcmpgtw %%mm0, %%mm3\n\t"

  215.         "pcmpgtw %%mm2, %%mm1\n\t"

  216.         "pxor %%mm3, %%mm0\n"

  217.         "pxor %%mm1, %%mm2\n"

  218.         "psubw %%mm3, %%mm0\n"

  219.         "psubw %%mm1, %%mm2\n"

  220.         "paddw %%mm0, %%mm2\n"

  221.         "paddw %%mm2, %%mm6\n"

  222. 

  223.         "add %2, %0\n"

  224.         "1:\n"

  225. 

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

  227.         "movq %%mm0, %%mm1\n"

  228.         "psllq $8, %%mm0\n"

  229.         "psrlq $8, %%mm1\n"

  230.         "psrlq $8, %%mm0\n"

  231.         "movq %%mm0, %%mm2\n"

  232.         "movq %%mm1, %%mm3\n"

  233.         "punpcklbw %%mm7, %%mm0\n"

  234.         "punpcklbw %%mm7, %%mm1\n"

  235.         "punpckhbw %%mm7, %%mm2\n"

  236.         "punpckhbw %%mm7, %%mm3\n"

  237.         "psubw %%mm1, %%mm0\n"

  238.         "psubw %%mm3, %%mm2\n"

  239.         "psubw %%mm0, %%mm4\n"

  240.         "psubw %%mm2, %%mm5\n"

  241.         "pxor  %%mm3, %%mm3\n"

  242.         "pxor  %%mm1, %%mm1\n"

  243.         "pcmpgtw %%mm4, %%mm3\n\t"

  244.         "pcmpgtw %%mm5, %%mm1\n\t"

  245.         "pxor  %%mm3, %%mm4\n"

  246.         "pxor  %%mm1, %%mm5\n"

  247.         "psubw %%mm3, %%mm4\n"

  248.         "psubw %%mm1, %%mm5\n"

  249.         "paddw %%mm4, %%mm5\n"

  250.         "paddw %%mm5, %%mm6\n"

  251. 

  252.         "add %2, %0\n"

  253. 

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

  255.         "movq      %%mm4, %%mm1\n"

  256.         "psllq $8, %%mm4\n"

  257.         "psrlq $8, %%mm1\n"

  258.         "psrlq $8, %%mm4\n"

  259.         "movq      %%mm4, %%mm5\n"

  260.         "movq      %%mm1, %%mm3\n"

  261.         "punpcklbw %%mm7, %%mm4\n"

  262.         "punpcklbw %%mm7, %%mm1\n"

  263.         "punpckhbw %%mm7, %%mm5\n"

  264.         "punpckhbw %%mm7, %%mm3\n"

  265.         "psubw     %%mm1, %%mm4\n"

  266.         "psubw     %%mm3, %%mm5\n"

  267.         "psubw     %%mm4, %%mm0\n"

  268.         "psubw     %%mm5, %%mm2\n"

  269.         "pxor      %%mm3, %%mm3\n"

  270.         "pxor      %%mm1, %%mm1\n"

  271.         "pcmpgtw   %%mm0, %%mm3\n\t"

  272.         "pcmpgtw   %%mm2, %%mm1\n\t"

  273.         "pxor      %%mm3, %%mm0\n"

  274.         "pxor      %%mm1, %%mm2\n"

  275.         "psubw     %%mm3, %%mm0\n"

  276.         "psubw     %%mm1, %%mm2\n"

  277.         "paddw     %%mm0, %%mm2\n"

  278.         "paddw     %%mm2, %%mm6\n"

  279. 

  280.         "add  %2, %0\n"

  281.         "subl $2, %%ecx\n"

  282.         " jnz 1b\n"

  283. 

  284.         "movq      %%mm6, %%mm0\n"

  285.         "punpcklwd %%mm7, %%mm0\n"

  286.         "punpckhwd %%mm7, %%mm6\n"

  287.         "paddd     %%mm0, %%mm6\n"

  288. 

  289.         "movq  %%mm6, %%mm0\n"

  290.         "psrlq $32,   %%mm6\n"

  291.         "paddd %%mm6, %%mm0\n"

  292.         "movd  %%mm0, %1\n"

  293.         : "+r" (pix1), "=r" (tmp)

  294.         : "r" ((x86_reg) line_size), "g" (h - 2)

  295.         : "%ecx");

  296. 

  297.     return tmp;

  298. }

  299. 

  300. static int hf_noise16_mmx(uint8_t *pix1, int line_size, int h)

  301. {

  302.     int tmp;

  303.     uint8_t *pix = pix1;

  304. 

  305.     __asm__ volatile (

  306.         "movl %3, %%ecx\n"

  307.         "pxor %%mm7, %%mm7\n"

  308.         "pxor %%mm6, %%mm6\n"

  309. 

  310.         "movq (%0), %%mm0\n"

  311.         "movq 1(%0), %%mm1\n"

  312.         "movq %%mm0, %%mm2\n"

  313.         "movq %%mm1, %%mm3\n"

  314.         "punpcklbw %%mm7, %%mm0\n"

  315.         "punpcklbw %%mm7, %%mm1\n"

  316.         "punpckhbw %%mm7, %%mm2\n"

  317.         "punpckhbw %%mm7, %%mm3\n"

  318.         "psubw %%mm1, %%mm0\n"

  319.         "psubw %%mm3, %%mm2\n"

  320. 

  321.         "add %2, %0\n"

  322. 

  323.         "movq (%0), %%mm4\n"

  324.         "movq 1(%0), %%mm1\n"

  325.         "movq %%mm4, %%mm5\n"

  326.         "movq %%mm1, %%mm3\n"

  327.         "punpcklbw %%mm7, %%mm4\n"

  328.         "punpcklbw %%mm7, %%mm1\n"

  329.         "punpckhbw %%mm7, %%mm5\n"

  330.         "punpckhbw %%mm7, %%mm3\n"

  331.         "psubw %%mm1, %%mm4\n"

  332.         "psubw %%mm3, %%mm5\n"

  333.         "psubw %%mm4, %%mm0\n"

  334.         "psubw %%mm5, %%mm2\n"

  335.         "pxor %%mm3, %%mm3\n"

  336.         "pxor %%mm1, %%mm1\n"

  337.         "pcmpgtw %%mm0, %%mm3\n\t"

  338.         "pcmpgtw %%mm2, %%mm1\n\t"

  339.         "pxor %%mm3, %%mm0\n"

  340.         "pxor %%mm1, %%mm2\n"

  341.         "psubw %%mm3, %%mm0\n"

  342.         "psubw %%mm1, %%mm2\n"

  343.         "paddw %%mm0, %%mm2\n"

  344.         "paddw %%mm2, %%mm6\n"

  345. 

  346.         "add %2, %0\n"

  347.         "1:\n"

  348. 

  349.         "movq (%0), %%mm0\n"

  350.         "movq 1(%0), %%mm1\n"

  351.         "movq %%mm0, %%mm2\n"

  352.         "movq %%mm1, %%mm3\n"

  353.         "punpcklbw %%mm7, %%mm0\n"

  354.         "punpcklbw %%mm7, %%mm1\n"

  355.         "punpckhbw %%mm7, %%mm2\n"

  356.         "punpckhbw %%mm7, %%mm3\n"

  357.         "psubw %%mm1, %%mm0\n"

  358.         "psubw %%mm3, %%mm2\n"

  359.         "psubw %%mm0, %%mm4\n"

  360.         "psubw %%mm2, %%mm5\n"

  361.         "pxor %%mm3, %%mm3\n"

  362.         "pxor %%mm1, %%mm1\n"

  363.         "pcmpgtw %%mm4, %%mm3\n\t"

  364.         "pcmpgtw %%mm5, %%mm1\n\t"

  365.         "pxor %%mm3, %%mm4\n"

  366.         "pxor %%mm1, %%mm5\n"

  367.         "psubw %%mm3, %%mm4\n"

  368.         "psubw %%mm1, %%mm5\n"

  369.         "paddw %%mm4, %%mm5\n"

  370.         "paddw %%mm5, %%mm6\n"

  371. 

  372.         "add %2, %0\n"

  373. 

  374.         "movq (%0), %%mm4\n"

  375.         "movq 1(%0), %%mm1\n"

  376.         "movq %%mm4, %%mm5\n"

  377.         "movq %%mm1, %%mm3\n"

  378.         "punpcklbw %%mm7, %%mm4\n"

  379.         "punpcklbw %%mm7, %%mm1\n"

  380.         "punpckhbw %%mm7, %%mm5\n"

  381.         "punpckhbw %%mm7, %%mm3\n"

  382.         "psubw %%mm1, %%mm4\n"

  383.         "psubw %%mm3, %%mm5\n"

  384.         "psubw %%mm4, %%mm0\n"

  385.         "psubw %%mm5, %%mm2\n"

  386.         "pxor %%mm3, %%mm3\n"

  387.         "pxor %%mm1, %%mm1\n"

  388.         "pcmpgtw %%mm0, %%mm3\n\t"

  389.         "pcmpgtw %%mm2, %%mm1\n\t"

  390.         "pxor %%mm3, %%mm0\n"

  391.         "pxor %%mm1, %%mm2\n"

  392.         "psubw %%mm3, %%mm0\n"

  393.         "psubw %%mm1, %%mm2\n"

  394.         "paddw %%mm0, %%mm2\n"

  395.         "paddw %%mm2, %%mm6\n"

  396. 

  397.         "add %2, %0\n"

  398.         "subl $2, %%ecx\n"

  399.         " jnz 1b\n"

  400. 

  401.         "movq %%mm6, %%mm0\n"

  402.         "punpcklwd %%mm7, %%mm0\n"

  403.         "punpckhwd %%mm7, %%mm6\n"

  404.         "paddd %%mm0, %%mm6\n"

  405. 

  406.         "movq %%mm6, %%mm0\n"

  407.         "psrlq $32, %%mm6\n"

  408.         "paddd %%mm6, %%mm0\n"

  409.         "movd %%mm0, %1\n"

  410.         : "+r" (pix1), "=r" (tmp)

  411.         : "r" ((x86_reg) line_size), "g" (h - 2)

  412.         : "%ecx");

  413. 

  414.     return tmp + hf_noise8_mmx(pix + 8, line_size, h);

  415. }

  416. 

  417. static int nsse16_mmx(MpegEncContext *c, uint8_t *pix1, uint8_t *pix2,

  418.                       int line_size, int h)

  419. {

  420.     int score1, score2;

  421. 

  422.     if (c)

  423.         score1 = c->dsp.sse[0](c, pix1, pix2, line_size, h);

  424.     else

  425.         score1 = sse16_mmx(c, pix1, pix2, line_size, h);

  426.     score2 = hf_noise16_mmx(pix1, line_size, h) -

  427.              hf_noise16_mmx(pix2, line_size, h);

  428. 

  429.     if (c)

  430.         return score1 + FFABS(score2) * c->avctx->nsse_weight;

  431.     else

  432.         return score1 + FFABS(score2) * 8;

  433. }

  434. 

  435. static int nsse8_mmx(MpegEncContext *c, uint8_t *pix1, uint8_t *pix2,

  436.                      int line_size, int h)

  437. {

  438.     int score1 = sse8_mmx(c, pix1, pix2, line_size, h);

  439.     int score2 = hf_noise8_mmx(pix1, line_size, h) -

  440.                  hf_noise8_mmx(pix2, line_size, h);

  441. 

  442.     if (c)

  443.         return score1 + FFABS(score2) * c->avctx->nsse_weight;

  444.     else

  445.         return score1 + FFABS(score2) * 8;

  446. }

  447. 

  448. static int vsad_intra16_mmx(MpegEncContext *v, uint8_t *pix, uint8_t *dummy,

  449.                             int line_size, int h)

  450. {

  451.     int tmp;

  452. 

  453.     assert((((int) pix) & 7) == 0);

  454.     assert((line_size & 7) == 0);

  455. 

  456. #define SUM(in0, in1, out0, out1)               \

  457.     "movq (%0), %%mm2\n"                        \

  458.     "movq 8(%0), %%mm3\n"                       \

  459.     "add %2,%0\n"                               \

  460.     "movq %%mm2, " #out0 "\n"                   \

  461.     "movq %%mm3, " #out1 "\n"                   \

  462.     "psubusb " #in0 ", %%mm2\n"                 \

  463.     "psubusb " #in1 ", %%mm3\n"                 \

  464.     "psubusb " #out0 ", " #in0 "\n"             \

  465.     "psubusb " #out1 ", " #in1 "\n"             \

  466.     "por %%mm2, " #in0 "\n"                     \

  467.     "por %%mm3, " #in1 "\n"                     \

  468.     "movq " #in0 ", %%mm2\n"                    \

  469.     "movq " #in1 ", %%mm3\n"                    \

  470.     "punpcklbw %%mm7, " #in0 "\n"               \

  471.     "punpcklbw %%mm7, " #in1 "\n"               \

  472.     "punpckhbw %%mm7, %%mm2\n"                  \

  473.     "punpckhbw %%mm7, %%mm3\n"                  \

  474.     "paddw " #in1 ", " #in0 "\n"                \

  475.     "paddw %%mm3, %%mm2\n"                      \

  476.     "paddw %%mm2, " #in0 "\n"                   \

  477.     "paddw " #in0 ", %%mm6\n"

  478. 

  479. 

  480.     __asm__ volatile (

  481.         "movl    %3, %%ecx\n"

  482.         "pxor %%mm6, %%mm6\n"

  483.         "pxor %%mm7, %%mm7\n"

  484.         "movq  (%0), %%mm0\n"

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

  486.         "add %2, %0\n"

  487.         "jmp 2f\n"

  488.         "1:\n"

  489. 

  490.         SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  491.         "2:\n"

  492.         SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  493. 

  494.         "subl $2, %%ecx\n"

  495.         "jnz 1b\n"

  496. 

  497.         "movq  %%mm6, %%mm0\n"

  498.         "psrlq $32,   %%mm6\n"

  499.         "paddw %%mm6, %%mm0\n"

  500.         "movq  %%mm0, %%mm6\n"

  501.         "psrlq $16,   %%mm0\n"

  502.         "paddw %%mm6, %%mm0\n"

  503.         "movd  %%mm0, %1\n"

  504.         : "+r" (pix), "=r" (tmp)

  505.         : "r" ((x86_reg) line_size), "m" (h)

  506.         : "%ecx");

  507. 

  508.     return tmp & 0xFFFF;

  509. }

  510. #undef SUM

  511. 

  512. static int vsad_intra16_mmxext(MpegEncContext *v, uint8_t *pix, uint8_t *dummy,

  513.                                int line_size, int h)

  514. {

  515.     int tmp;

  516. 

  517.     assert((((int) pix) & 7) == 0);

  518.     assert((line_size & 7) == 0);

  519. 

  520. #define SUM(in0, in1, out0, out1)               \

  521.     "movq (%0), " #out0 "\n"                    \

  522.     "movq 8(%0), " #out1 "\n"                   \

  523.     "add %2, %0\n"                              \

  524.     "psadbw " #out0 ", " #in0 "\n"              \

  525.     "psadbw " #out1 ", " #in1 "\n"              \

  526.     "paddw " #in1 ", " #in0 "\n"                \

  527.     "paddw " #in0 ", %%mm6\n"

  528. 

  529.     __asm__ volatile (

  530.         "movl %3, %%ecx\n"

  531.         "pxor %%mm6, %%mm6\n"

  532.         "pxor %%mm7, %%mm7\n"

  533.         "movq (%0), %%mm0\n"

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

  535.         "add %2, %0\n"

  536.         "jmp 2f\n"

  537.         "1:\n"

  538. 

  539.         SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  540.         "2:\n"

  541.         SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  542. 

  543.         "subl $2, %%ecx\n"

  544.         "jnz 1b\n"

  545. 

  546.         "movd %%mm6, %1\n"

  547.         : "+r" (pix), "=r" (tmp)

  548.         : "r" ((x86_reg) line_size), "m" (h)

  549.         : "%ecx");

  550. 

  551.     return tmp;

  552. }

  553. #undef SUM

  554. 

  555. static int vsad16_mmx(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  556.                       int line_size, int h)

  557. {

  558.     int tmp;

  559. 

  560.     assert((((int) pix1) & 7) == 0);

  561.     assert((((int) pix2) & 7) == 0);

  562.     assert((line_size & 7) == 0);

  563. 

  564. #define SUM(in0, in1, out0, out1)       \

  565.     "movq (%0), %%mm2\n"                \

  566.     "movq (%1), " #out0 "\n"            \

  567.     "movq 8(%0), %%mm3\n"               \

  568.     "movq 8(%1), " #out1 "\n"           \

  569.     "add %3, %0\n"                      \

  570.     "add %3, %1\n"                      \

  571.     "psubb " #out0 ", %%mm2\n"          \

  572.     "psubb " #out1 ", %%mm3\n"          \

  573.     "pxor %%mm7, %%mm2\n"               \

  574.     "pxor %%mm7, %%mm3\n"               \

  575.     "movq %%mm2, " #out0 "\n"           \

  576.     "movq %%mm3, " #out1 "\n"           \

  577.     "psubusb " #in0 ", %%mm2\n"         \

  578.     "psubusb " #in1 ", %%mm3\n"         \

  579.     "psubusb " #out0 ", " #in0 "\n"     \

  580.     "psubusb " #out1 ", " #in1 "\n"     \

  581.     "por %%mm2, " #in0 "\n"             \

  582.     "por %%mm3, " #in1 "\n"             \

  583.     "movq " #in0 ", %%mm2\n"            \

  584.     "movq " #in1 ", %%mm3\n"            \

  585.     "punpcklbw %%mm7, " #in0 "\n"       \

  586.     "punpcklbw %%mm7, " #in1 "\n"       \

  587.     "punpckhbw %%mm7, %%mm2\n"          \

  588.     "punpckhbw %%mm7, %%mm3\n"          \

  589.     "paddw " #in1 ", " #in0 "\n"        \

  590.     "paddw %%mm3, %%mm2\n"              \

  591.     "paddw %%mm2, " #in0 "\n"           \

  592.     "paddw " #in0 ", %%mm6\n"

  593. 

  594. 

  595.     __asm__ volatile (

  596.         "movl %4, %%ecx\n"

  597.         "pxor %%mm6, %%mm6\n"

  598.         "pcmpeqw %%mm7, %%mm7\n"

  599.         "psllw $15, %%mm7\n"

  600.         "packsswb %%mm7, %%mm7\n"

  601.         "movq (%0), %%mm0\n"

  602.         "movq (%1), %%mm2\n"

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

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

  605.         "add %3, %0\n"

  606.         "add %3, %1\n"

  607.         "psubb %%mm2, %%mm0\n"

  608.         "psubb %%mm3, %%mm1\n"

  609.         "pxor %%mm7, %%mm0\n"

  610.         "pxor %%mm7, %%mm1\n"

  611.         "jmp 2f\n"

  612.         "1:\n"

  613. 

  614.         SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  615.         "2:\n"

  616.         SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  617. 

  618.         "subl $2, %%ecx\n"

  619.         "jnz 1b\n"

  620. 

  621.         "movq %%mm6, %%mm0\n"

  622.         "psrlq $32, %%mm6\n"

  623.         "paddw %%mm6, %%mm0\n"

  624.         "movq %%mm0, %%mm6\n"

  625.         "psrlq $16, %%mm0\n"

  626.         "paddw %%mm6, %%mm0\n"

  627.         "movd %%mm0, %2\n"

  628.         : "+r" (pix1), "+r" (pix2), "=r" (tmp)

  629.         : "r" ((x86_reg) line_size), "m" (h)

  630.         : "%ecx");

  631. 

  632.     return tmp & 0x7FFF;

  633. }

  634. #undef SUM

  635. 

  636. static int vsad16_mmxext(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  637.                          int line_size, int h)

  638. {

  639.     int tmp;

  640. 

  641.     assert((((int) pix1) & 7) == 0);

  642.     assert((((int) pix2) & 7) == 0);

  643.     assert((line_size & 7) == 0);

  644. 

  645. #define SUM(in0, in1, out0, out1)               \

  646.     "movq (%0), " #out0 "\n"                    \

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

  648.     "movq 8(%0), " #out1 "\n"                   \

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

  650.     "add %3, %0\n"                              \

  651.     "add %3, %1\n"                              \

  652.     "psubb %%mm2, " #out0 "\n"                  \

  653.     "psubb %%mm3, " #out1 "\n"                  \

  654.     "pxor %%mm7, " #out0 "\n"                   \

  655.     "pxor %%mm7, " #out1 "\n"                   \

  656.     "psadbw " #out0 ", " #in0 "\n"              \

  657.     "psadbw " #out1 ", " #in1 "\n"              \

  658.     "paddw " #in1 ", " #in0 "\n"                \

  659.     "paddw " #in0 ", %%mm6\n    "

  660. 

  661.     __asm__ volatile (

  662.         "movl %4, %%ecx\n"

  663.         "pxor %%mm6, %%mm6\n"

  664.         "pcmpeqw %%mm7, %%mm7\n"

  665.         "psllw $15, %%mm7\n"

  666.         "packsswb %%mm7, %%mm7\n"

  667.         "movq (%0), %%mm0\n"

  668.         "movq (%1), %%mm2\n"

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

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

  671.         "add %3, %0\n"

  672.         "add %3, %1\n"

  673.         "psubb %%mm2, %%mm0\n"

  674.         "psubb %%mm3, %%mm1\n"

  675.         "pxor %%mm7, %%mm0\n"

  676.         "pxor %%mm7, %%mm1\n"

  677.         "jmp 2f\n"

  678.         "1:\n"

  679. 

  680.         SUM(%%mm4, %%mm5, %%mm0, %%mm1)

  681.         "2:\n"

  682.         SUM(%%mm0, %%mm1, %%mm4, %%mm5)

  683. 

  684.         "subl $2, %%ecx\n"

  685.         "jnz 1b\n"

  686. 

  687.         "movd %%mm6, %2\n"

  688.         : "+r" (pix1), "+r" (pix2), "=r" (tmp)

  689.         : "r" ((x86_reg) line_size), "m" (h)

  690.         : "%ecx");

  691. 

  692.     return tmp;

  693. }

  694. #undef SUM

  695. 

  696. #define MMABS_MMX(a,z)                          \

  697.     "pxor "    #z ", " #z "             \n\t"   \

  698.     "pcmpgtw " #a ", " #z "             \n\t"   \

  699.     "pxor "    #z ", " #a "             \n\t"   \

  700.     "psubw "   #z ", " #a "             \n\t"

  701. 

  702. #define MMABS_MMXEXT(a, z)                      \

  703.     "pxor "    #z ", " #z "             \n\t"   \

  704.     "psubw "   #a ", " #z "             \n\t"   \

  705.     "pmaxsw "  #z ", " #a "             \n\t"

  706. 

  707. #define MMABS_SSSE3(a,z)                        \

  708.     "pabsw "   #a ", " #a "             \n\t"

  709. 

  710. #define MMABS_SUM(a,z, sum)                     \

  711.     MMABS(a,z)                                  \

  712.     "paddusw " #a ", " #sum "           \n\t"

  713. 

  714. /* FIXME: HSUM_* saturates at 64k, while an 8x8 hadamard or dct block can get

  715.  * up to about 100k on extreme inputs. But that's very unlikely to occur in

  716.  * natural video, and it's even more unlikely to not have any alternative

  717.  * mvs/modes with lower cost. */

  718. #define HSUM_MMX(a, t, dst)                     \

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

  720.     "psrlq      $32, " #a "             \n\t"   \

  721.     "paddusw " #t ", " #a "             \n\t"   \

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

  723.     "psrlq      $16, " #a "             \n\t"   \

  724.     "paddusw " #t ", " #a "             \n\t"   \

  725.     "movd    " #a ", " #dst "           \n\t"   \

  726. 

  727. #define HSUM_MMXEXT(a, t, dst)                  \

  728.     "pshufw   $0x0E, " #a ", " #t "     \n\t"   \

  729.     "paddusw " #t ", " #a "             \n\t"   \

  730.     "pshufw   $0x01, " #a ", " #t "     \n\t"   \

  731.     "paddusw " #t ", " #a "             \n\t"   \

  732.     "movd    " #a ", " #dst "           \n\t"   \

  733. 

  734. #define HSUM_SSE2(a, t, dst)                    \

  735.     "movhlps " #a ", " #t "             \n\t"   \

  736.     "paddusw " #t ", " #a "             \n\t"   \

  737.     "pshuflw  $0x0E, " #a ", " #t "     \n\t"   \

  738.     "paddusw " #t ", " #a "             \n\t"   \

  739.     "pshuflw  $0x01, " #a ", " #t "     \n\t"   \

  740.     "paddusw " #t ", " #a "             \n\t"   \

  741.     "movd    " #a ", " #dst "           \n\t"   \

  742. 

  743. #define DCT_SAD4(m, mm, o)                      \

  744.     "mov"#m" "#o" +  0(%1), " #mm "2    \n\t"   \

  745.     "mov"#m" "#o" + 16(%1), " #mm "3    \n\t"   \

  746.     "mov"#m" "#o" + 32(%1), " #mm "4    \n\t"   \

  747.     "mov"#m" "#o" + 48(%1), " #mm "5    \n\t"   \

  748.     MMABS_SUM(mm ## 2, mm ## 6, mm ## 0)        \

  749.     MMABS_SUM(mm ## 3, mm ## 7, mm ## 1)        \

  750.     MMABS_SUM(mm ## 4, mm ## 6, mm ## 0)        \

  751.     MMABS_SUM(mm ## 5, mm ## 7, mm ## 1)        \

  752. 

  753. #define DCT_SAD_MMX                             \

  754.     "pxor    %%mm0, %%mm0               \n\t"   \

  755.     "pxor    %%mm1, %%mm1               \n\t"   \

  756.     DCT_SAD4(q, %%mm, 0)                        \

  757.     DCT_SAD4(q, %%mm, 8)                        \

  758.     DCT_SAD4(q, %%mm, 64)                       \

  759.     DCT_SAD4(q, %%mm, 72)                       \

  760.     "paddusw %%mm1, %%mm0               \n\t"   \

  761.     HSUM(%%mm0, %%mm1, %0)

  762. 

  763. #define DCT_SAD_SSE2                            \

  764.     "pxor    %%xmm0, %%xmm0             \n\t"   \

  765.     "pxor    %%xmm1, %%xmm1             \n\t"   \

  766.     DCT_SAD4(dqa, %%xmm, 0)                     \

  767.     DCT_SAD4(dqa, %%xmm, 64)                    \

  768.     "paddusw %%xmm1, %%xmm0             \n\t"   \

  769.     HSUM(%%xmm0, %%xmm1, %0)

  770. 

  771. #define DCT_SAD_FUNC(cpu)                           \

  772. static int sum_abs_dctelem_ ## cpu(int16_t *block)  \

  773. {                                                   \

  774.     int sum;                                        \

  775.     __asm__ volatile (                              \

  776.         DCT_SAD                                     \

  777.         :"=r"(sum)                                  \

  778.         :"r"(block));                               \

  779.     return sum & 0xFFFF;                            \

  780. }

  781. 

  782. #define DCT_SAD         DCT_SAD_MMX

  783. #define HSUM(a, t, dst) HSUM_MMX(a, t, dst)

  784. #define MMABS(a, z)     MMABS_MMX(a, z)

  785. DCT_SAD_FUNC(mmx)

  786. #undef MMABS

  787. #undef HSUM

  788. 

  789. #define HSUM(a, t, dst) HSUM_MMXEXT(a, t, dst)

  790. #define MMABS(a, z)     MMABS_MMXEXT(a, z)

  791. DCT_SAD_FUNC(mmxext)

  792. #undef HSUM

  793. #undef DCT_SAD

  794. 

  795. #define DCT_SAD         DCT_SAD_SSE2

  796. #define HSUM(a, t, dst) HSUM_SSE2(a, t, dst)

  797. DCT_SAD_FUNC(sse2)

  798. #undef MMABS

  799. 

  800. #if HAVE_SSSE3_INLINE

  801. #define MMABS(a, z)     MMABS_SSSE3(a, z)

  802. DCT_SAD_FUNC(ssse3)

  803. #undef MMABS

  804. #endif

  805. #undef HSUM

  806. #undef DCT_SAD

  807. 

  808. static int ssd_int8_vs_int16_mmx(const int8_t *pix1, const int16_t *pix2,

  809.                                  int size)

  810. {

  811.     int sum;

  812.     x86_reg i = size;

  813. 

  814.     __asm__ volatile (

  815.         "pxor %%mm4, %%mm4 \n"

  816.         "1: \n"

  817.         "sub $8, %0 \n"

  818.         "movq (%2, %0), %%mm2 \n"

  819.         "movq (%3, %0, 2), %%mm0 \n"

  820.         "movq 8(%3, %0, 2), %%mm1 \n"

  821.         "punpckhbw %%mm2, %%mm3 \n"

  822.         "punpcklbw %%mm2, %%mm2 \n"

  823.         "psraw $8, %%mm3 \n"

  824.         "psraw $8, %%mm2 \n"

  825.         "psubw %%mm3, %%mm1 \n"

  826.         "psubw %%mm2, %%mm0 \n"

  827.         "pmaddwd %%mm1, %%mm1 \n"

  828.         "pmaddwd %%mm0, %%mm0 \n"

  829.         "paddd %%mm1, %%mm4 \n"

  830.         "paddd %%mm0, %%mm4 \n"

  831.         "jg 1b \n"

  832.         "movq %%mm4, %%mm3 \n"

  833.         "psrlq $32, %%mm3 \n"

  834.         "paddd %%mm3, %%mm4 \n"

  835.         "movd %%mm4, %1 \n"

  836.         : "+r" (i), "=r" (sum)

  837.         : "r" (pix1), "r" (pix2));

  838. 

  839.     return sum;

  840. }

  841. 

  842. #define PHADDD(a, t)                            \

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

  844.     "psrlq    $32, " #a "               \n\t"   \

  845.     "paddd " #t ", " #a "               \n\t"

  846. 

  847. /*

  848.  * pmulhw:   dst[0 - 15] = (src[0 - 15] * dst[0 - 15])[16 - 31]

  849.  * pmulhrw:  dst[0 - 15] = (src[0 - 15] * dst[0 - 15] + 0x8000)[16 - 31]

  850.  * pmulhrsw: dst[0 - 15] = (src[0 - 15] * dst[0 - 15] + 0x4000)[15 - 30]

  851.  */

  852. #define PMULHRW(x, y, s, o)                     \

  853.     "pmulhw " #s ", " #x "              \n\t"   \

  854.     "pmulhw " #s ", " #y "              \n\t"   \

  855.     "paddw  " #o ", " #x "              \n\t"   \

  856.     "paddw  " #o ", " #y "              \n\t"   \

  857.     "psraw      $1, " #x "              \n\t"   \

  858.     "psraw      $1, " #y "              \n\t"

  859. #define DEF(x) x ## _mmx

  860. #define SET_RND MOVQ_WONE

  861. #define SCALE_OFFSET 1

  862. 

  863. #include "dsputil_qns_template.c"

  864. 

  865. #undef DEF

  866. #undef SET_RND

  867. #undef SCALE_OFFSET

  868. #undef PMULHRW

  869. 

  870. #define DEF(x) x ## _3dnow

  871. #define SET_RND(x)

  872. #define SCALE_OFFSET 0

  873. #define PMULHRW(x, y, s, o)                     \

  874.     "pmulhrw " #s ", " #x "             \n\t"   \

  875.     "pmulhrw " #s ", " #y "             \n\t"

  876. 

  877. #include "dsputil_qns_template.c"

  878. 

  879. #undef DEF

  880. #undef SET_RND

  881. #undef SCALE_OFFSET

  882. #undef PMULHRW

  883. 

  884. #if HAVE_SSSE3_INLINE

  885. #undef PHADDD

  886. #define DEF(x) x ## _ssse3

  887. #define SET_RND(x)

  888. #define SCALE_OFFSET -1

  889. 

  890. #define PHADDD(a, t)                            \

  891.     "pshufw $0x0E, " #a ", " #t "       \n\t"   \

  892.     /* faster than phaddd on core2 */           \

  893.     "paddd " #t ", " #a "               \n\t"

  894. 

  895. #define PMULHRW(x, y, s, o)                     \

  896.     "pmulhrsw " #s ", " #x "            \n\t"   \

  897.     "pmulhrsw " #s ", " #y "            \n\t"

  898. 

  899. #include "dsputil_qns_template.c"

  900. 

  901. #undef DEF

  902. #undef SET_RND

  903. #undef SCALE_OFFSET

  904. #undef PMULHRW

  905. #undef PHADDD

  906. #endif /* HAVE_SSSE3_INLINE */

  907. 

  908. #endif /* HAVE_INLINE_ASM */

  909. 

  910. int ff_sse16_sse2(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  911.                   int line_size, int h);

  912. 

  913. #define hadamard_func(cpu)                                              \

  914.     int ff_hadamard8_diff_ ## cpu(MpegEncContext *s, uint8_t *src1,     \

  915.                                   uint8_t *src2, int stride, int h);    \

  916.     int ff_hadamard8_diff16_ ## cpu(MpegEncContext *s, uint8_t *src1,   \

  917.                                     uint8_t *src2, int stride, int h);

  918. 

  919. hadamard_func(mmx)

  920. hadamard_func(mmxext)

  921. hadamard_func(sse2)

  922. hadamard_func(ssse3)

  923. 

  924. av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx,

  925.                                     unsigned high_bit_depth)

  926. {

  927.     int cpu_flags = av_get_cpu_flags();

  928.     const int dct_algo = avctx->dct_algo;

  929. 

  930.     if (EXTERNAL_MMX(cpu_flags)) {

  931.         if (!high_bit_depth)

  932.             c->get_pixels = ff_get_pixels_mmx;

  933.         c->diff_pixels = ff_diff_pixels_mmx;

  934.         c->pix_sum     = ff_pix_sum16_mmx;

  935.         c->pix_norm1   = ff_pix_norm1_mmx;

  936.     }

  937. 

  938.     if (EXTERNAL_SSE2(cpu_flags))

  939.         if (!high_bit_depth)

  940.             c->get_pixels = ff_get_pixels_sse2;

  941. 

  942. #if HAVE_INLINE_ASM

  943.     if (INLINE_MMX(cpu_flags)) {

  944.         if (!high_bit_depth &&

  945.             (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX))

  946.             c->fdct = ff_fdct_mmx;

  947. 

  948.         c->sum_abs_dctelem = sum_abs_dctelem_mmx;

  949. 

  950.         c->sse[0]  = sse16_mmx;

  951.         c->sse[1]  = sse8_mmx;

  952.         c->vsad[4] = vsad_intra16_mmx;

  953. 

  954.         c->nsse[0] = nsse16_mmx;

  955.         c->nsse[1] = nsse8_mmx;

  956.         if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {

  957.             c->vsad[0]      = vsad16_mmx;

  958.             c->try_8x8basis = try_8x8basis_mmx;

  959.         }

  960.         c->add_8x8basis = add_8x8basis_mmx;

  961. 

  962.         c->ssd_int8_vs_int16 = ssd_int8_vs_int16_mmx;

  963.     }

  964. 

  965.     if (INLINE_AMD3DNOW(cpu_flags)) {

  966.         if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {

  967.             c->try_8x8basis = try_8x8basis_3dnow;

  968.         }

  969.         c->add_8x8basis = add_8x8basis_3dnow;

  970.     }

  971. 

  972.     if (INLINE_MMXEXT(cpu_flags)) {

  973.         if (!high_bit_depth &&

  974.             (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX))

  975.             c->fdct = ff_fdct_mmxext;

  976. 

  977.         c->sum_abs_dctelem = sum_abs_dctelem_mmxext;

  978.         c->vsad[4]         = vsad_intra16_mmxext;

  979. 

  980.         if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {

  981.             c->vsad[0] = vsad16_mmxext;

  982.         }

  983.     }

  984. 

  985.     if (INLINE_SSE2(cpu_flags)) {

  986.         if (!high_bit_depth &&

  987.             (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX))

  988.             c->fdct = ff_fdct_sse2;

  989. 

  990.         c->sum_abs_dctelem = sum_abs_dctelem_sse2;

  991.     }

  992. 

  993. #if HAVE_SSSE3_INLINE

  994.     if (INLINE_SSSE3(cpu_flags)) {

  995.         if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {

  996.             c->try_8x8basis = try_8x8basis_ssse3;

  997.         }

  998.         c->add_8x8basis    = add_8x8basis_ssse3;

  999.         c->sum_abs_dctelem = sum_abs_dctelem_ssse3;

  1000.     }

  1001. #endif

  1002. #endif /* HAVE_INLINE_ASM */

  1003. 

  1004.     if (EXTERNAL_MMX(cpu_flags)) {

  1005.         c->hadamard8_diff[0] = ff_hadamard8_diff16_mmx;

  1006.         c->hadamard8_diff[1] = ff_hadamard8_diff_mmx;

  1007.     }

  1008. 

  1009.     if (EXTERNAL_MMXEXT(cpu_flags)) {

  1010.         c->hadamard8_diff[0] = ff_hadamard8_diff16_mmxext;

  1011.         c->hadamard8_diff[1] = ff_hadamard8_diff_mmxext;

  1012.     }

  1013. 

  1014.     if (EXTERNAL_SSE2(cpu_flags)) {

  1015.         c->sse[0] = ff_sse16_sse2;

  1016. 

  1017. #if HAVE_ALIGNED_STACK

  1018.         c->hadamard8_diff[0] = ff_hadamard8_diff16_sse2;

  1019.         c->hadamard8_diff[1] = ff_hadamard8_diff_sse2;

  1020. #endif

  1021.     }

  1022. 

  1023.     if (EXTERNAL_SSSE3(cpu_flags) && HAVE_ALIGNED_STACK) {

  1024.         c->hadamard8_diff[0] = ff_hadamard8_diff16_ssse3;

  1025.         c->hadamard8_diff[1] = ff_hadamard8_diff_ssse3;

  1026.     }

  1027. 

  1028.     ff_dsputil_init_pix_mmx(c, avctx);

  1029. }