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.
36452 Commits
32 Branches
748MB
Tree: 76e74e4831
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '76e74e4831'
${ noResults }
FFmpeg/libavcodec/x86/dsputilenc_mmx.c

1058 lines
30KB
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.  * This file is part of Libav.

  7.  *

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

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

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

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

  12.  *

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

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

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

  16.  * Lesser General Public License for more details.

  17.  *

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

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

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

  21.  *

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

  23.  */

  24. 

  25. #include "libavutil/cpu.h"

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

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

  28. #include "libavcodec/dsputil.h"

  29. #include "libavcodec/mpegvideo.h"

  30. #include "libavcodec/mathops.h"

  31. #include "dsputil_mmx.h"

  32. 

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

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

  35. void ff_diff_pixels_mmx(int16_t *block, const uint8_t *s1, const uint8_t *s2, int stride);

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

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

  38. 

  39. #if HAVE_INLINE_ASM

  40. 

  41. static int sse8_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  42.     int tmp;

  43.   __asm__ volatile (

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

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

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

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

  48.       "1:\n"

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

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

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

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

  53. 

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

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

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

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

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

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

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

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

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

  63. 

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

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

  66. 

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

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

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

  70. 

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

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

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

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

  75. 

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

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

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

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

  80. 

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

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

  83. 

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

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

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

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

  88. 

  89.       "decl %%ecx\n"

  90.       "jnz 1b\n"

  91. 

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

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

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

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

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

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

  98.       : "%ecx");

  99.     return tmp;

  100. }

  101. 

  102. static int sse16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  103.     int tmp;

  104.   __asm__ volatile (

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

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

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

  108.       "1:\n"

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

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

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

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

  113. 

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

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

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

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

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

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

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

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

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

  123. 

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

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

  126. 

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

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

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

  130. 

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

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

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

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

  135. 

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

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

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

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

  140. 

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

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

  143. 

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

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

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

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

  148. 

  149.       "decl %%ecx\n"

  150.       "jnz 1b\n"

  151. 

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

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

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

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

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

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

  158.       : "%ecx");

  159.     return tmp;

  160. }

  161. 

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

  163.     int tmp;

  164.   __asm__ volatile (

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

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

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

  168. 

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

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

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

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

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

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

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

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

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

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

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

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

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

  182. 

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

  184. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  210. 

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

  212.       "1:\n"

  213. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  239. 

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

  241. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  267. 

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

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

  270.       " jnz 1b\n"

  271. 

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

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

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

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

  276. 

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

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

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

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

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

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

  283.       : "%ecx");

  284.       return tmp;

  285. }

  286. 

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

  288.     int tmp;

  289.     uint8_t * pix= pix1;

  290.   __asm__ volatile (

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

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

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

  294. 

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

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

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

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

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

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

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

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

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

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

  305. 

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

  307. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  330. 

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

  332.       "1:\n"

  333. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  356. 

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

  358. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  381. 

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

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

  384.       " jnz 1b\n"

  385. 

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

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

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

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

  390. 

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

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

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

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

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

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

  397.       : "%ecx");

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

  399. }

  400. 

  401. static int nsse16_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  402.     MpegEncContext *c = p;

  403.     int score1, score2;

  404. 

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

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

  407.     score2= hf_noise16_mmx(pix1, line_size, h) - hf_noise16_mmx(pix2, line_size, h);

  408. 

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

  410.     else  return score1 + FFABS(score2)*8;

  411. }

  412. 

  413. static int nsse8_mmx(void *p, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  414.     MpegEncContext *c = p;

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

  416.     int score2= hf_noise8_mmx(pix1, line_size, h) - hf_noise8_mmx(pix2, line_size, h);

  417. 

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

  419.     else  return score1 + FFABS(score2)*8;

  420. }

  421. 

  422. static int vsad_intra16_mmx(void *v, uint8_t * pix, uint8_t * dummy, int line_size, int h) {

  423.     int tmp;

  424. 

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

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

  427. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  450. 

  451. 

  452.   __asm__ volatile (

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

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

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

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

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

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

  459.       "jmp 2f\n"

  460.       "1:\n"

  461. 

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

  463.       "2:\n"

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

  465. 

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

  467.       "jnz 1b\n"

  468. 

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

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

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

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

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

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

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

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

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

  478.       : "%ecx");

  479.     return tmp & 0xFFFF;

  480. }

  481. #undef SUM

  482. 

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

  484.                                int line_size, int h)

  485. {

  486.     int tmp;

  487. 

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

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

  490. 

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

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

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

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

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

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

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

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

  499. 

  500.   __asm__ volatile (

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

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

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

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

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

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

  507.       "jmp 2f\n"

  508.       "1:\n"

  509. 

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

  511.       "2:\n"

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

  513. 

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

  515.       "jnz 1b\n"

  516. 

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

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

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

  520.       : "%ecx");

  521.     return tmp;

  522. }

  523. #undef SUM

  524. 

  525. static int vsad16_mmx(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h) {

  526.     int tmp;

  527. 

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

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

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

  531. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  561. 

  562. 

  563.   __asm__ volatile (

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  579.       "jmp 2f\n"

  580.       "1:\n"

  581. 

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

  583.       "2:\n"

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

  585. 

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

  587.       "jnz 1b\n"

  588. 

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

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

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

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

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

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

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

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

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

  598.       : "%ecx");

  599.     return tmp & 0x7FFF;

  600. }

  601. #undef SUM

  602. 

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

  604.                          int line_size, int h)

  605. {

  606.     int tmp;

  607. 

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

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

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

  611. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  627. 

  628.   __asm__ volatile (

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  644.       "jmp 2f\n"

  645.       "1:\n"

  646. 

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

  648.       "2:\n"

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

  650. 

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

  652.       "jnz 1b\n"

  653. 

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

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

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

  657.       : "%ecx");

  658.     return tmp;

  659. }

  660. #undef SUM

  661. 

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

  663.     x86_reg i=0;

  664.     __asm__ volatile(

  665.         "1:                             \n\t"

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

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

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

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

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

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

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

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

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

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

  676.         " jb 1b                         \n\t"

  677.         : "+r" (i)

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

  679.     );

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

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

  682. }

  683. 

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

  685.                                               const uint8_t *src2, int w,

  686.                                               int *left, int *left_top)

  687. {

  688.     x86_reg i=0;

  689.     uint8_t l, lt;

  690. 

  691.     __asm__ volatile(

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

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

  694.         "1:                             \n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  711.         " jb 1b                         \n\t"

  712.         : "+r" (i)

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

  714.     );

  715. 

  716.     l= *left;

  717.     lt= *left_top;

  718. 

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

  720. 

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

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

  723. }

  724. 

  725. #define MMABS_MMX(a,z)\

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

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

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

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

  730. 

  731. #define MMABS_MMXEXT(a, z)                 \

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

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

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

  735. 

  736. #define MMABS_SSSE3(a,z)\

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

  738. 

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

  740.     MMABS(a,z)\

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

  742. 

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

  744.  * about 100k on extreme inputs. But that's very unlikely to occur in natural video,

  745.  * and it's even more unlikely to not have any alternative mvs/modes with lower cost. */

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

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

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

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

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

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

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

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

  754. 

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

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

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

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

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

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

  761. 

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

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

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

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

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

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

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

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

  770. 

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

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

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

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

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

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

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

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

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

  780. 

  781. #define DCT_SAD_MMX\

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

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

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

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

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

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

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

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

  790. 

  791. #define DCT_SAD_SSE2\

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

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

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

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

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

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

  798. 

  799. #define DCT_SAD_FUNC(cpu) \

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

  801.     int sum;\

  802.     __asm__ volatile(\

  803.         DCT_SAD\

  804.         :"=r"(sum)\

  805.         :"r"(block)\

  806.     );\

  807.     return sum&0xFFFF;\

  808. }

  809. 

  810. #define DCT_SAD       DCT_SAD_MMX

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

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

  813. DCT_SAD_FUNC(mmx)

  814. #undef MMABS

  815. #undef HSUM

  816. 

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

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

  819. DCT_SAD_FUNC(mmxext)

  820. #undef HSUM

  821. #undef DCT_SAD

  822. 

  823. #define DCT_SAD       DCT_SAD_SSE2

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

  825. DCT_SAD_FUNC(sse2)

  826. #undef MMABS

  827. 

  828. #if HAVE_SSSE3_INLINE

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

  830. DCT_SAD_FUNC(ssse3)

  831. #undef MMABS

  832. #endif

  833. #undef HSUM

  834. #undef DCT_SAD

  835. 

  836. static int ssd_int8_vs_int16_mmx(const int8_t *pix1, const int16_t *pix2, int size){

  837.     int sum;

  838.     x86_reg i=size;

  839.     __asm__ volatile(

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

  841.         "1: \n"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  856.         "jg 1b \n"

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

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

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

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

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

  862.         :"r"(pix1), "r"(pix2)

  863.     );

  864.     return sum;

  865. }

  866. 

  867. #define PHADDD(a, t)\

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

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

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

  871. /*

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

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

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

  875.  */

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

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

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

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

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

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

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

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

  884. #define SET_RND MOVQ_WONE

  885. #define SCALE_OFFSET 1

  886. 

  887. #include "dsputil_qns_template.c"

  888. 

  889. #undef DEF

  890. #undef SET_RND

  891. #undef SCALE_OFFSET

  892. #undef PMULHRW

  893. 

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

  895. #define SET_RND(x)

  896. #define SCALE_OFFSET 0

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

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

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

  900. 

  901. #include "dsputil_qns_template.c"

  902. 

  903. #undef DEF

  904. #undef SET_RND

  905. #undef SCALE_OFFSET

  906. #undef PMULHRW

  907. 

  908. #if HAVE_SSSE3_INLINE

  909. #undef PHADDD

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

  911. #define SET_RND(x)

  912. #define SCALE_OFFSET -1

  913. #define PHADDD(a, t)\

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

  915.     "paddd "#t", "#a"                 \n\t" /* faster than phaddd on core2 */

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

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

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

  919. 

  920. #include "dsputil_qns_template.c"

  921. 

  922. #undef DEF

  923. #undef SET_RND

  924. #undef SCALE_OFFSET

  925. #undef PMULHRW

  926. #undef PHADDD

  927. #endif /* HAVE_SSSE3_INLINE */

  928. 

  929. #endif /* HAVE_INLINE_ASM */

  930. 

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

  932. 

  933. #define hadamard_func(cpu) \

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

  935.                               int stride, int h); \

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

  937.                               int stride, int h);

  938. 

  939. hadamard_func(mmx)

  940. hadamard_func(mmxext)

  941. hadamard_func(sse2)

  942. hadamard_func(ssse3)

  943. 

  944. void ff_dsputilenc_init_mmx(DSPContext* c, AVCodecContext *avctx)

  945. {

  946.     int mm_flags = av_get_cpu_flags();

  947.     int bit_depth = avctx->bits_per_raw_sample;

  948. 

  949. #if HAVE_YASM

  950.     if (EXTERNAL_MMX(mm_flags)) {

  951.         if (bit_depth <= 8)

  952.             c->get_pixels = ff_get_pixels_mmx;

  953.         c->diff_pixels = ff_diff_pixels_mmx;

  954.         c->pix_sum = ff_pix_sum16_mmx;

  955. 

  956.         c->pix_norm1 = ff_pix_norm1_mmx;

  957.     }

  958.     if (EXTERNAL_SSE2(mm_flags))

  959.         if (bit_depth <= 8)

  960.             c->get_pixels = ff_get_pixels_sse2;

  961. #endif /* HAVE_YASM */

  962. 

  963. #if HAVE_INLINE_ASM

  964.     if (mm_flags & AV_CPU_FLAG_MMX) {

  965.         const int dct_algo = avctx->dct_algo;

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

  967.             (dct_algo==FF_DCT_AUTO || dct_algo==FF_DCT_MMX)) {

  968.             if(mm_flags & AV_CPU_FLAG_SSE2){

  969.                 c->fdct = ff_fdct_sse2;

  970.             } else if (mm_flags & AV_CPU_FLAG_MMXEXT) {

  971.                 c->fdct = ff_fdct_mmxext;

  972.             }else{

  973.                 c->fdct = ff_fdct_mmx;

  974.             }

  975.         }

  976. 

  977. 

  978.         c->diff_bytes= diff_bytes_mmx;

  979.         c->sum_abs_dctelem= sum_abs_dctelem_mmx;

  980. 

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

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

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

  984. 

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

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

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

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

  989.         }

  990. 

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

  992.             c->try_8x8basis= try_8x8basis_mmx;

  993.         }

  994.         c->add_8x8basis= add_8x8basis_mmx;

  995. 

  996.         c->ssd_int8_vs_int16 = ssd_int8_vs_int16_mmx;

  997. 

  998.         if (mm_flags & AV_CPU_FLAG_MMXEXT) {

  999.             c->sum_abs_dctelem = sum_abs_dctelem_mmxext;

  1000.             c->vsad[4]         = vsad_intra16_mmxext;

  1001. 

  1002.             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  1003.                 c->vsad[0] = vsad16_mmxext;

  1004.             }

  1005. 

  1006.             c->sub_hfyu_median_prediction = sub_hfyu_median_prediction_mmxext;

  1007.         }

  1008. 

  1009.         if(mm_flags & AV_CPU_FLAG_SSE2){

  1010.             c->sum_abs_dctelem= sum_abs_dctelem_sse2;

  1011.         }

  1012. 

  1013. #if HAVE_SSSE3_INLINE

  1014.         if(mm_flags & AV_CPU_FLAG_SSSE3){

  1015.             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  1016.                 c->try_8x8basis= try_8x8basis_ssse3;

  1017.             }

  1018.             c->add_8x8basis= add_8x8basis_ssse3;

  1019.             c->sum_abs_dctelem= sum_abs_dctelem_ssse3;

  1020.         }

  1021. #endif

  1022. 

  1023.         if(mm_flags & AV_CPU_FLAG_3DNOW){

  1024.             if(!(avctx->flags & CODEC_FLAG_BITEXACT)){

  1025.                 c->try_8x8basis= try_8x8basis_3dnow;

  1026.             }

  1027.             c->add_8x8basis= add_8x8basis_3dnow;

  1028.         }

  1029.     }

  1030. #endif /* HAVE_INLINE_ASM */

  1031. 

  1032.     if (EXTERNAL_MMX(mm_flags)) {

  1033.         c->hadamard8_diff[0] = ff_hadamard8_diff16_mmx;

  1034.         c->hadamard8_diff[1] = ff_hadamard8_diff_mmx;

  1035. 

  1036.         if (EXTERNAL_MMXEXT(mm_flags)) {

  1037.             c->hadamard8_diff[0] = ff_hadamard8_diff16_mmxext;

  1038.             c->hadamard8_diff[1] = ff_hadamard8_diff_mmxext;

  1039.         }

  1040. 

  1041.         if (EXTERNAL_SSE2(mm_flags)) {

  1042.             c->sse[0] = ff_sse16_sse2;

  1043. 

  1044. #if HAVE_ALIGNED_STACK

  1045.             c->hadamard8_diff[0] = ff_hadamard8_diff16_sse2;

  1046.             c->hadamard8_diff[1] = ff_hadamard8_diff_sse2;

  1047. #endif

  1048.         }

  1049. 

  1050.         if (EXTERNAL_SSSE3(mm_flags) && HAVE_ALIGNED_STACK) {

  1051.             c->hadamard8_diff[0] = ff_hadamard8_diff16_ssse3;

  1052.             c->hadamard8_diff[1] = ff_hadamard8_diff_ssse3;

  1053.         }

  1054.     }

  1055. 

  1056.     ff_dsputil_init_pix_mmx(c, avctx);

  1057. }