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.
39526 Commits
32 Branches
307MB
Tree: 1b51b7ba00
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '1b51b7ba00'
${ noResults }
FFmpeg/libavcodec/x86/dsputilenc_mmx.c

1101 lines
34KB
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 "libavcodec/mathops.h"

  33. #include "dsputil_x86.h"

  34. 

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

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

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

  38.                         int stride);

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

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

  41. 

  42. #if HAVE_INLINE_ASM

  43. 

  44. static int sse8_mmx(void *v, uint8_t *pix1, uint8_t *pix2, 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(void *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(void *p, uint8_t *pix1, uint8_t *pix2,

  418.                       int line_size, int h)

  419. {

  420.     MpegEncContext *c = p;

  421.     int score1, score2;

  422. 

  423.     if (c)

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

  425.     else

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

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

  428.              hf_noise16_mmx(pix2, line_size, h);

  429. 

  430.     if (c)

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

  432.     else

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

  434. }

  435. 

  436. static int nsse8_mmx(void *p, uint8_t *pix1, uint8_t *pix2,

  437.                      int line_size, int h)

  438. {

  439.     MpegEncContext *c = p;

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

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

  442.                  hf_noise8_mmx(pix2, line_size, h);

  443. 

  444.     if (c)

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

  446.     else

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

  448. }

  449. 

  450. static int vsad_intra16_mmx(void *v, uint8_t *pix, uint8_t *dummy,

  451.                             int line_size, int h)

  452. {

  453.     int tmp;

  454. 

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

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

  457. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  480. 

  481. 

  482.     __asm__ volatile (

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

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

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

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

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

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

  489.         "jmp 2f\n"

  490.         "1:\n"

  491. 

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

  493.         "2:\n"

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

  495. 

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

  497.         "jnz 1b\n"

  498. 

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

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

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

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

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

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

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

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

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

  508.         : "%ecx");

  509. 

  510.     return tmp & 0xFFFF;

  511. }

  512. #undef SUM

  513. 

  514. static int vsad_intra16_mmxext(void *v, uint8_t *pix, uint8_t *dummy,

  515.                                int line_size, int h)

  516. {

  517.     int tmp;

  518. 

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

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

  521. 

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

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

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

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

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

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

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

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

  530. 

  531.     __asm__ volatile (

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

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

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

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

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

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

  538.         "jmp 2f\n"

  539.         "1:\n"

  540. 

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

  542.         "2:\n"

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

  544. 

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

  546.         "jnz 1b\n"

  547. 

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

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

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

  551.         : "%ecx");

  552. 

  553.     return tmp;

  554. }

  555. #undef SUM

  556. 

  557. static int vsad16_mmx(void *v, uint8_t *pix1, uint8_t *pix2,

  558.                       int line_size, int h)

  559. {

  560.     int tmp;

  561. 

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

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

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

  565. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  595. 

  596. 

  597.     __asm__ volatile (

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  613.         "jmp 2f\n"

  614.         "1:\n"

  615. 

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

  617.         "2:\n"

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

  619. 

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

  621.         "jnz 1b\n"

  622. 

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

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

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

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

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

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

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

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

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

  632.         : "%ecx");

  633. 

  634.     return tmp & 0x7FFF;

  635. }

  636. #undef SUM

  637. 

  638. static int vsad16_mmxext(void *v, uint8_t *pix1, uint8_t *pix2,

  639.                          int line_size, int h)

  640. {

  641.     int tmp;

  642. 

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

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

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

  646. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  662. 

  663.     __asm__ volatile (

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  679.         "jmp 2f\n"

  680.         "1:\n"

  681. 

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

  683.         "2:\n"

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

  685. 

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

  687.         "jnz 1b\n"

  688. 

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

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

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

  692.         : "%ecx");

  693. 

  694.     return tmp;

  695. }

  696. #undef SUM

  697. 

  698. static void diff_bytes_mmx(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w)

  699. {

  700.     x86_reg i = 0;

  701. 

  702.     __asm__ volatile (

  703.         "1:                             \n\t"

  704.         "movq  (%2, %0), %%mm0          \n\t"

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

  706.         "psubb %%mm0, %%mm1             \n\t"

  707.         "movq %%mm1, (%3, %0)           \n\t"

  708.         "movq 8(%2, %0), %%mm0          \n\t"

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

  710.         "psubb %%mm0, %%mm1             \n\t"

  711.         "movq %%mm1, 8(%3, %0)          \n\t"

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

  713.         "cmp %4, %0                     \n\t"

  714.         " jb 1b                         \n\t"

  715.         : "+r" (i)

  716.         : "r" (src1), "r" (src2), "r" (dst), "r" ((x86_reg) w - 15));

  717. 

  718.     for (; i < w; i++)

  719.         dst[i + 0] = src1[i + 0] - src2[i + 0];

  720. }

  721. 

  722. static void sub_hfyu_median_prediction_mmxext(uint8_t *dst, const uint8_t *src1,

  723.                                               const uint8_t *src2, int w,

  724.                                               int *left, int *left_top)

  725. {

  726.     x86_reg i = 0;

  727.     uint8_t l, lt;

  728. 

  729.     __asm__ volatile (

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

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

  732.         "1:                             \n\t"

  733.         "movq  (%1, %0), %%mm1          \n\t" // T

  734.         "movq  -1(%2, %0), %%mm2        \n\t" // L

  735.         "movq  (%2, %0), %%mm3          \n\t" // X

  736.         "movq %%mm2, %%mm4              \n\t" // L

  737.         "psubb %%mm0, %%mm2             \n\t"

  738.         "paddb %%mm1, %%mm2             \n\t" // L + T - LT

  739.         "movq %%mm4, %%mm5              \n\t" // L

  740.         "pmaxub %%mm1, %%mm4            \n\t" // max(T, L)

  741.         "pminub %%mm5, %%mm1            \n\t" // min(T, L)

  742.         "pminub %%mm2, %%mm4            \n\t"

  743.         "pmaxub %%mm1, %%mm4            \n\t"

  744.         "psubb %%mm4, %%mm3             \n\t" // dst - pred

  745.         "movq %%mm3, (%3, %0)           \n\t"

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

  747.         "movq -1(%1, %0), %%mm0         \n\t" // LT

  748.         "cmp %4, %0                     \n\t"

  749.         " jb 1b                         \n\t"

  750.         : "+r" (i)

  751.         : "r" (src1), "r" (src2), "r" (dst), "r" ((x86_reg) w));

  752. 

  753.     l  = *left;

  754.     lt = *left_top;

  755. 

  756.     dst[0] = src2[0] - mid_pred(l, src1[0], (l + src1[0] - lt) & 0xFF);

  757. 

  758.     *left_top = src1[w - 1];

  759.     *left     = src2[w - 1];

  760. }

  761. 

  762. #define MMABS_MMX(a,z)                          \

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

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

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

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

  767. 

  768. #define MMABS_MMXEXT(a, z)                      \

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

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

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

  772. 

  773. #define MMABS_SSSE3(a,z)                        \

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

  775. 

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

  777.     MMABS(a,z)                                  \

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

  779. 

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

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

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

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

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

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

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

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

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

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

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

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

  792. 

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

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

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

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

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

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

  799. 

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

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

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

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

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

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

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

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

  808. 

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

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

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

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

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

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

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

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

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

  818. 

  819. #define DCT_SAD_MMX                             \

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

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

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

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

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

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

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

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

  828. 

  829. #define DCT_SAD_SSE2                            \

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

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

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

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

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

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

  836. 

  837. #define DCT_SAD_FUNC(cpu)                           \

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

  839. {                                                   \

  840.     int sum;                                        \

  841.     __asm__ volatile (                              \

  842.         DCT_SAD                                     \

  843.         :"=r"(sum)                                  \

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

  845.     return sum & 0xFFFF;                            \

  846. }

  847. 

  848. #define DCT_SAD         DCT_SAD_MMX

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

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

  851. DCT_SAD_FUNC(mmx)

  852. #undef MMABS

  853. #undef HSUM

  854. 

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

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

  857. DCT_SAD_FUNC(mmxext)

  858. #undef HSUM

  859. #undef DCT_SAD

  860. 

  861. #define DCT_SAD         DCT_SAD_SSE2

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

  863. DCT_SAD_FUNC(sse2)

  864. #undef MMABS

  865. 

  866. #if HAVE_SSSE3_INLINE

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

  868. DCT_SAD_FUNC(ssse3)

  869. #undef MMABS

  870. #endif

  871. #undef HSUM

  872. #undef DCT_SAD

  873. 

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

  875.                                  int size)

  876. {

  877.     int sum;

  878.     x86_reg i = size;

  879. 

  880.     __asm__ volatile (

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

  882.         "1: \n"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  897.         "jg 1b \n"

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

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

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

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

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

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

  904. 

  905.     return sum;

  906. }

  907. 

  908. #define PHADDD(a, t)                            \

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

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

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

  912. 

  913. /*

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

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

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

  917.  */

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

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

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

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

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

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

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

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

  926. #define SET_RND MOVQ_WONE

  927. #define SCALE_OFFSET 1

  928. 

  929. #include "dsputil_qns_template.c"

  930. 

  931. #undef DEF

  932. #undef SET_RND

  933. #undef SCALE_OFFSET

  934. #undef PMULHRW

  935. 

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

  937. #define SET_RND(x)

  938. #define SCALE_OFFSET 0

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

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

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

  942. 

  943. #include "dsputil_qns_template.c"

  944. 

  945. #undef DEF

  946. #undef SET_RND

  947. #undef SCALE_OFFSET

  948. #undef PMULHRW

  949. 

  950. #if HAVE_SSSE3_INLINE

  951. #undef PHADDD

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

  953. #define SET_RND(x)

  954. #define SCALE_OFFSET -1

  955. 

  956. #define PHADDD(a, t)                            \

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

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

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

  960. 

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

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

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

  964. 

  965. #include "dsputil_qns_template.c"

  966. 

  967. #undef DEF

  968. #undef SET_RND

  969. #undef SCALE_OFFSET

  970. #undef PMULHRW

  971. #undef PHADDD

  972. #endif /* HAVE_SSSE3_INLINE */

  973. 

  974. #endif /* HAVE_INLINE_ASM */

  975. 

  976. int ff_sse16_sse2(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h);

  977. 

  978. #define hadamard_func(cpu)                                              \

  979. int ff_hadamard8_diff_ ## cpu(void *s, uint8_t *src1, uint8_t *src2,    \

  980.                               int stride, int h);                       \

  981. int ff_hadamard8_diff16_ ## cpu(void *s, uint8_t *src1, uint8_t *src2,  \

  982.                                 int stride, int h);

  983. 

  984. hadamard_func(mmx)

  985. hadamard_func(mmxext)

  986. hadamard_func(sse2)

  987. hadamard_func(ssse3)

  988. 

  989. av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx)

  990. {

  991.     int cpu_flags = av_get_cpu_flags();

  992.     const int dct_algo = avctx->dct_algo;

  993. 

  994. #if HAVE_YASM

  995.     int bit_depth = avctx->bits_per_raw_sample;

  996. 

  997.     if (EXTERNAL_MMX(cpu_flags)) {

  998.         if (bit_depth <= 8)

  999.             c->get_pixels = ff_get_pixels_mmx;

  1000.         c->diff_pixels = ff_diff_pixels_mmx;

  1001.         c->pix_sum     = ff_pix_sum16_mmx;

  1002.         c->pix_norm1   = ff_pix_norm1_mmx;

  1003.     }

  1004. 

  1005.     if (EXTERNAL_SSE2(cpu_flags))

  1006.         if (bit_depth <= 8)

  1007.             c->get_pixels = ff_get_pixels_sse2;

  1008. #endif /* HAVE_YASM */

  1009. 

  1010. #if HAVE_INLINE_ASM

  1011.     if (INLINE_MMX(cpu_flags)) {

  1012.         if (avctx->bits_per_raw_sample <= 8 &&

  1013.             (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX))

  1014.             c->fdct = ff_fdct_mmx;

  1015. 

  1016.         c->diff_bytes      = diff_bytes_mmx;

  1017.         c->sum_abs_dctelem = sum_abs_dctelem_mmx;

  1018. 

  1019.         c->sse[0]  = sse16_mmx;

  1020.         c->sse[1]  = sse8_mmx;

  1021.         c->vsad[4] = vsad_intra16_mmx;

  1022. 

  1023.         c->nsse[0] = nsse16_mmx;

  1024.         c->nsse[1] = nsse8_mmx;

  1025.         if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {

  1026.             c->vsad[0]      = vsad16_mmx;

  1027.             c->try_8x8basis = try_8x8basis_mmx;

  1028.         }

  1029.         c->add_8x8basis = add_8x8basis_mmx;

  1030. 

  1031.         c->ssd_int8_vs_int16 = ssd_int8_vs_int16_mmx;

  1032.     }

  1033. 

  1034.     if (INLINE_AMD3DNOW(cpu_flags)) {

  1035.         if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {

  1036.             c->try_8x8basis = try_8x8basis_3dnow;

  1037.         }

  1038.         c->add_8x8basis = add_8x8basis_3dnow;

  1039.     }

  1040. 

  1041.     if (INLINE_MMXEXT(cpu_flags)) {

  1042.         if (avctx->bits_per_raw_sample <= 8 &&

  1043.             (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX))

  1044.             c->fdct = ff_fdct_mmxext;

  1045. 

  1046.         c->sum_abs_dctelem = sum_abs_dctelem_mmxext;

  1047.         c->vsad[4]         = vsad_intra16_mmxext;

  1048. 

  1049.         if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {

  1050.             c->vsad[0] = vsad16_mmxext;

  1051.         }

  1052. 

  1053.         c->sub_hfyu_median_prediction = sub_hfyu_median_prediction_mmxext;

  1054.     }

  1055. 

  1056.     if (INLINE_SSE2(cpu_flags)) {

  1057.         if (avctx->bits_per_raw_sample <= 8 &&

  1058.             (dct_algo == FF_DCT_AUTO || dct_algo == FF_DCT_MMX))

  1059.             c->fdct = ff_fdct_sse2;

  1060. 

  1061.         c->sum_abs_dctelem = sum_abs_dctelem_sse2;

  1062.     }

  1063. 

  1064. #if HAVE_SSSE3_INLINE

  1065.     if (INLINE_SSSE3(cpu_flags)) {

  1066.         if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {

  1067.             c->try_8x8basis = try_8x8basis_ssse3;

  1068.         }

  1069.         c->add_8x8basis    = add_8x8basis_ssse3;

  1070.         c->sum_abs_dctelem = sum_abs_dctelem_ssse3;

  1071.     }

  1072. #endif

  1073. #endif /* HAVE_INLINE_ASM */

  1074. 

  1075.     if (EXTERNAL_MMX(cpu_flags)) {

  1076.         c->hadamard8_diff[0] = ff_hadamard8_diff16_mmx;

  1077.         c->hadamard8_diff[1] = ff_hadamard8_diff_mmx;

  1078.     }

  1079. 

  1080.     if (EXTERNAL_MMXEXT(cpu_flags)) {

  1081.         c->hadamard8_diff[0] = ff_hadamard8_diff16_mmxext;

  1082.         c->hadamard8_diff[1] = ff_hadamard8_diff_mmxext;

  1083.     }

  1084. 

  1085.     if (EXTERNAL_SSE2(cpu_flags)) {

  1086.         c->sse[0] = ff_sse16_sse2;

  1087. 

  1088. #if HAVE_ALIGNED_STACK

  1089.         c->hadamard8_diff[0] = ff_hadamard8_diff16_sse2;

  1090.         c->hadamard8_diff[1] = ff_hadamard8_diff_sse2;

  1091. #endif

  1092.     }

  1093. 

  1094.     if (EXTERNAL_SSSE3(cpu_flags) && HAVE_ALIGNED_STACK) {

  1095.         c->hadamard8_diff[0] = ff_hadamard8_diff16_ssse3;

  1096.         c->hadamard8_diff[1] = ff_hadamard8_diff_ssse3;

  1097.     }

  1098. 

  1099.     ff_dsputil_init_pix_mmx(c, avctx);

  1100. }