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.
7023 Commits
32 Branches
303MB
Tree: 2b5269b51c
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '2b5269b51c'
${ noResults }
FFmpeg/libavcodec/i386/h264dsp_mmx.c

1525 lines
58KB
Raw Blame History 

  1. /*

  2.  * Copyright (c) 2004-2005 Michael Niedermayer, Loren Merritt

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. 

  22. /***********************************/

  23. /* IDCT */

  24. 

  25. /* in/out: mma=mma+mmb, mmb=mmb-mma */

  26. #define SUMSUB_BA( a, b ) \

  27.     "paddw "#b", "#a" \n\t"\

  28.     "paddw "#b", "#b" \n\t"\

  29.     "psubw "#a", "#b" \n\t"

  30. 

  31. #define SUMSUB_BADC( a, b, c, d ) \

  32.     "paddw "#b", "#a" \n\t"\

  33.     "paddw "#d", "#c" \n\t"\

  34.     "paddw "#b", "#b" \n\t"\

  35.     "paddw "#d", "#d" \n\t"\

  36.     "psubw "#a", "#b" \n\t"\

  37.     "psubw "#c", "#d" \n\t"

  38. 

  39. #define SUMSUBD2_AB( a, b, t ) \

  40.     "movq  "#b", "#t" \n\t"\

  41.     "psraw  $1 , "#b" \n\t"\

  42.     "paddw "#a", "#b" \n\t"\

  43.     "psraw  $1 , "#a" \n\t"\

  44.     "psubw "#t", "#a" \n\t"

  45. 

  46. #define IDCT4_1D( s02, s13, d02, d13, t ) \

  47.     SUMSUB_BA  ( s02, d02 )\

  48.     SUMSUBD2_AB( s13, d13, t )\

  49.     SUMSUB_BADC( d13, s02, s13, d02 )

  50. 

  51. #define TRANSPOSE4(a,b,c,d,t)\

  52.     SBUTTERFLY(a,b,t,wd) /* a=aebf t=cgdh */\

  53.     SBUTTERFLY(c,d,b,wd) /* c=imjn b=kolp */\

  54.     SBUTTERFLY(a,c,d,dq) /* a=aeim d=bfjn */\

  55.     SBUTTERFLY(t,b,c,dq) /* t=cgko c=dhlp */

  56. 

  57. #define STORE_DIFF_4P( p, t, z ) \

  58.     "psraw      $6,     "#p" \n\t"\

  59.     "movd       (%0),   "#t" \n\t"\

  60.     "punpcklbw "#z",    "#t" \n\t"\

  61.     "paddsw    "#t",    "#p" \n\t"\

  62.     "packuswb  "#z",    "#p" \n\t"\

  63.     "movd      "#p",    (%0) \n\t"

  64. 

  65. static void ff_h264_idct_add_mmx(uint8_t *dst, int16_t *block, int stride)

  66. {

  67.     /* Load dct coeffs */

  68.     asm volatile(

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

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

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

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

  73.     :: "r"(block) );

  74. 

  75.     asm volatile(

  76.         /* mm1=s02+s13  mm2=s02-s13  mm4=d02+d13  mm0=d02-d13 */

  77.         IDCT4_1D( %%mm2, %%mm1, %%mm0, %%mm3, %%mm4 )

  78. 

  79.         "movq      %0,    %%mm6 \n\t"

  80.         /* in: 1,4,0,2  out: 1,2,3,0 */

  81.         TRANSPOSE4( %%mm3, %%mm1, %%mm0, %%mm2, %%mm4 )

  82. 

  83.         "paddw     %%mm6, %%mm3 \n\t"

  84. 

  85.         /* mm2=s02+s13  mm3=s02-s13  mm4=d02+d13  mm1=d02-d13 */

  86.         IDCT4_1D( %%mm4, %%mm2, %%mm3, %%mm0, %%mm1 )

  87. 

  88.         "pxor %%mm7, %%mm7    \n\t"

  89.     :: "m"(ff_pw_32));

  90. 

  91.     asm volatile(

  92.     STORE_DIFF_4P( %%mm0, %%mm1, %%mm7)

  93.         "add %1, %0             \n\t"

  94.     STORE_DIFF_4P( %%mm2, %%mm1, %%mm7)

  95.         "add %1, %0             \n\t"

  96.     STORE_DIFF_4P( %%mm3, %%mm1, %%mm7)

  97.         "add %1, %0             \n\t"

  98.     STORE_DIFF_4P( %%mm4, %%mm1, %%mm7)

  99.         : "+r"(dst)

  100.         : "r" ((long)stride)

  101.     );

  102. }

  103. 

  104. static inline void h264_idct8_1d(int16_t *block)

  105. {

  106.     asm volatile(

  107.         "movq 112(%0), %%mm7  \n\t"

  108.         "movq  80(%0), %%mm5  \n\t"

  109.         "movq  48(%0), %%mm3  \n\t"

  110.         "movq  16(%0), %%mm1  \n\t"

  111. 

  112.         "movq   %%mm7, %%mm4  \n\t"

  113.         "movq   %%mm3, %%mm6  \n\t"

  114.         "movq   %%mm5, %%mm0  \n\t"

  115.         "movq   %%mm7, %%mm2  \n\t"

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

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

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

  119.         "psubw  %%mm6, %%mm2  \n\t"

  120.         "psubw  %%mm4, %%mm0  \n\t"

  121.         "psubw  %%mm3, %%mm2  \n\t"

  122.         "psubw  %%mm3, %%mm0  \n\t"

  123.         "paddw  %%mm1, %%mm2  \n\t"

  124. 

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

  126.         "movq   %%mm1, %%mm6  \n\t"

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

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

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

  130.         "paddw  %%mm1, %%mm6  \n\t"

  131.         "paddw  %%mm7, %%mm4  \n\t"

  132.         "paddw  %%mm5, %%mm6  \n\t"

  133.         "psubw  %%mm1, %%mm4  \n\t"

  134.         "paddw  %%mm3, %%mm6  \n\t"

  135. 

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

  137.         "movq   %%mm4, %%mm3  \n\t"

  138.         "movq   %%mm2, %%mm5  \n\t"

  139.         "movq   %%mm6, %%mm7  \n\t"

  140.         "psraw  $2,    %%mm6  \n\t"

  141.         "psraw  $2,    %%mm3  \n\t"

  142.         "psraw  $2,    %%mm5  \n\t"

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

  144.         "paddw  %%mm6, %%mm1  \n\t"

  145.         "paddw  %%mm2, %%mm3  \n\t"

  146.         "psubw  %%mm4, %%mm5  \n\t"

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

  148. 

  149.         "movq  32(%0), %%mm2  \n\t"

  150.         "movq  96(%0), %%mm6  \n\t"

  151.         "movq   %%mm2, %%mm4  \n\t"

  152.         "movq   %%mm6, %%mm0  \n\t"

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

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

  155.         "psubw  %%mm0, %%mm4  \n\t"

  156.         "paddw  %%mm2, %%mm6  \n\t"

  157. 

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

  159.         "movq  64(%0), %%mm0  \n\t"

  160.         SUMSUB_BA( %%mm0, %%mm2 )

  161.         SUMSUB_BA( %%mm6, %%mm0 )

  162.         SUMSUB_BA( %%mm4, %%mm2 )

  163.         SUMSUB_BA( %%mm7, %%mm6 )

  164.         SUMSUB_BA( %%mm5, %%mm4 )

  165.         SUMSUB_BA( %%mm3, %%mm2 )

  166.         SUMSUB_BA( %%mm1, %%mm0 )

  167.         :: "r"(block)

  168.     );

  169. }

  170. 

  171. static void ff_h264_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)

  172. {

  173.     int i;

  174.     int16_t __attribute__ ((aligned(8))) b2[64];

  175. 

  176.     block[0] += 32;

  177. 

  178.     for(i=0; i<2; i++){

  179.         uint64_t tmp;

  180. 

  181.         h264_idct8_1d(block+4*i);

  182. 

  183.         asm volatile(

  184.             "movq   %%mm7,    %0   \n\t"

  185.             TRANSPOSE4( %%mm0, %%mm2, %%mm4, %%mm6, %%mm7 )

  186.             "movq   %%mm0,  8(%1)  \n\t"

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

  188.             "movq   %%mm7, 40(%1)  \n\t"

  189.             "movq   %%mm4, 56(%1)  \n\t"

  190.             "movq    %0,    %%mm7  \n\t"

  191.             TRANSPOSE4( %%mm7, %%mm5, %%mm3, %%mm1, %%mm0 )

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

  193.             "movq   %%mm1, 16(%1)  \n\t"

  194.             "movq   %%mm0, 32(%1)  \n\t"

  195.             "movq   %%mm3, 48(%1)  \n\t"

  196.             : "=m"(tmp)

  197.             : "r"(b2+32*i)

  198.             : "memory"

  199.         );

  200.     }

  201. 

  202.     for(i=0; i<2; i++){

  203.         h264_idct8_1d(b2+4*i);

  204. 

  205.         asm volatile(

  206.             "psraw     $6, %%mm7  \n\t"

  207.             "psraw     $6, %%mm6  \n\t"

  208.             "psraw     $6, %%mm5  \n\t"

  209.             "psraw     $6, %%mm4  \n\t"

  210.             "psraw     $6, %%mm3  \n\t"

  211.             "psraw     $6, %%mm2  \n\t"

  212.             "psraw     $6, %%mm1  \n\t"

  213.             "psraw     $6, %%mm0  \n\t"

  214. 

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

  216.             "movq   %%mm5,  16(%0)  \n\t"

  217.             "movq   %%mm3,  32(%0)  \n\t"

  218.             "movq   %%mm1,  48(%0)  \n\t"

  219.             "movq   %%mm0,  64(%0)  \n\t"

  220.             "movq   %%mm2,  80(%0)  \n\t"

  221.             "movq   %%mm4,  96(%0)  \n\t"

  222.             "movq   %%mm6, 112(%0)  \n\t"

  223.             :: "r"(b2+4*i)

  224.             : "memory"

  225.         );

  226.     }

  227. 

  228.     add_pixels_clamped_mmx(b2, dst, stride);

  229. }

  230. 

  231. static void ff_h264_idct_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)

  232. {

  233.     int dc = (block[0] + 32) >> 6;

  234.     asm volatile(

  235.         "movd          %0, %%mm0 \n\t"

  236.         "pshufw $0, %%mm0, %%mm0 \n\t"

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

  238.         "psubw      %%mm0, %%mm1 \n\t"

  239.         "packuswb   %%mm0, %%mm0 \n\t"

  240.         "packuswb   %%mm1, %%mm1 \n\t"

  241.         ::"r"(dc)

  242.     );

  243.     asm volatile(

  244.         "movd          %0, %%mm2 \n\t"

  245.         "movd          %1, %%mm3 \n\t"

  246.         "movd          %2, %%mm4 \n\t"

  247.         "movd          %3, %%mm5 \n\t"

  248.         "paddusb    %%mm0, %%mm2 \n\t"

  249.         "paddusb    %%mm0, %%mm3 \n\t"

  250.         "paddusb    %%mm0, %%mm4 \n\t"

  251.         "paddusb    %%mm0, %%mm5 \n\t"

  252.         "psubusb    %%mm1, %%mm2 \n\t"

  253.         "psubusb    %%mm1, %%mm3 \n\t"

  254.         "psubusb    %%mm1, %%mm4 \n\t"

  255.         "psubusb    %%mm1, %%mm5 \n\t"

  256.         "movd       %%mm2, %0    \n\t"

  257.         "movd       %%mm3, %1    \n\t"

  258.         "movd       %%mm4, %2    \n\t"

  259.         "movd       %%mm5, %3    \n\t"

  260.         :"+m"(*(uint32_t*)(dst+0*stride)),

  261.          "+m"(*(uint32_t*)(dst+1*stride)),

  262.          "+m"(*(uint32_t*)(dst+2*stride)),

  263.          "+m"(*(uint32_t*)(dst+3*stride))

  264.     );

  265. }

  266. 

  267. static void ff_h264_idct8_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)

  268. {

  269.     int dc = (block[0] + 32) >> 6;

  270.     int y;

  271.     asm volatile(

  272.         "movd          %0, %%mm0 \n\t"

  273.         "pshufw $0, %%mm0, %%mm0 \n\t"

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

  275.         "psubw      %%mm0, %%mm1 \n\t"

  276.         "packuswb   %%mm0, %%mm0 \n\t"

  277.         "packuswb   %%mm1, %%mm1 \n\t"

  278.         ::"r"(dc)

  279.     );

  280.     for(y=2; y--; dst += 4*stride){

  281.     asm volatile(

  282.         "movq          %0, %%mm2 \n\t"

  283.         "movq          %1, %%mm3 \n\t"

  284.         "movq          %2, %%mm4 \n\t"

  285.         "movq          %3, %%mm5 \n\t"

  286.         "paddusb    %%mm0, %%mm2 \n\t"

  287.         "paddusb    %%mm0, %%mm3 \n\t"

  288.         "paddusb    %%mm0, %%mm4 \n\t"

  289.         "paddusb    %%mm0, %%mm5 \n\t"

  290.         "psubusb    %%mm1, %%mm2 \n\t"

  291.         "psubusb    %%mm1, %%mm3 \n\t"

  292.         "psubusb    %%mm1, %%mm4 \n\t"

  293.         "psubusb    %%mm1, %%mm5 \n\t"

  294.         "movq       %%mm2, %0    \n\t"

  295.         "movq       %%mm3, %1    \n\t"

  296.         "movq       %%mm4, %2    \n\t"

  297.         "movq       %%mm5, %3    \n\t"

  298.         :"+m"(*(uint64_t*)(dst+0*stride)),

  299.          "+m"(*(uint64_t*)(dst+1*stride)),

  300.          "+m"(*(uint64_t*)(dst+2*stride)),

  301.          "+m"(*(uint64_t*)(dst+3*stride))

  302.     );

  303.     }

  304. }

  305. 

  306. 

  307. /***********************************/

  308. /* deblocking */

  309. 

  310. // out: o = |x-y|>a

  311. // clobbers: t

  312. #define DIFF_GT_MMX(x,y,a,o,t)\

  313.     "movq     "#y", "#t"  \n\t"\

  314.     "movq     "#x", "#o"  \n\t"\

  315.     "psubusb  "#x", "#t"  \n\t"\

  316.     "psubusb  "#y", "#o"  \n\t"\

  317.     "por      "#t", "#o"  \n\t"\

  318.     "psubusb  "#a", "#o"  \n\t"

  319. 

  320. // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1

  321. // out: mm5=beta-1, mm7=mask

  322. // clobbers: mm4,mm6

  323. #define H264_DEBLOCK_MASK(alpha1, beta1) \

  324.     "pshufw $0, "#alpha1", %%mm4 \n\t"\

  325.     "pshufw $0, "#beta1 ", %%mm5 \n\t"\

  326.     "packuswb  %%mm4, %%mm4      \n\t"\

  327.     "packuswb  %%mm5, %%mm5      \n\t"\

  328.     DIFF_GT_MMX(%%mm1, %%mm2, %%mm4, %%mm7, %%mm6) /* |p0-q0| > alpha-1 */\

  329.     DIFF_GT_MMX(%%mm0, %%mm1, %%mm5, %%mm4, %%mm6) /* |p1-p0| > beta-1 */\

  330.     "por       %%mm4, %%mm7      \n\t"\

  331.     DIFF_GT_MMX(%%mm3, %%mm2, %%mm5, %%mm4, %%mm6) /* |q1-q0| > beta-1 */\

  332.     "por       %%mm4, %%mm7      \n\t"\

  333.     "pxor      %%mm6, %%mm6      \n\t"\

  334.     "pcmpeqb   %%mm6, %%mm7      \n\t"

  335. 

  336. // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask)

  337. // out: mm1=p0' mm2=q0'

  338. // clobbers: mm0,3-6

  339. #define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\

  340.         /* a = q0^p0^((p1-q1)>>2) */\

  341.         "movq    %%mm0, %%mm4  \n\t"\

  342.         "psubb   %%mm3, %%mm4  \n\t"\

  343.         "psrlw   $2,    %%mm4  \n\t"\

  344.         "pxor    %%mm1, %%mm4  \n\t"\

  345.         "pxor    %%mm2, %%mm4  \n\t"\

  346.         /* b = p0^(q1>>2) */\

  347.         "psrlw   $2,    %%mm3  \n\t"\

  348.         "pand "#pb_3f", %%mm3  \n\t"\

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

  350.         "pxor    %%mm3, %%mm5  \n\t"\

  351.         /* c = q0^(p1>>2) */\

  352.         "psrlw   $2,    %%mm0  \n\t"\

  353.         "pand "#pb_3f", %%mm0  \n\t"\

  354.         "movq    %%mm2, %%mm6  \n\t"\

  355.         "pxor    %%mm0, %%mm6  \n\t"\

  356.         /* d = (c^b) & ~(b^a) & 1 */\

  357.         "pxor    %%mm5, %%mm6  \n\t"\

  358.         "pxor    %%mm4, %%mm5  \n\t"\

  359.         "pandn   %%mm6, %%mm5  \n\t"\

  360.         "pand "#pb_01", %%mm5  \n\t"\

  361.         /* delta = (avg(q0, p1>>2) + (d&a))

  362.          *       - (avg(p0, q1>>2) + (d&~a)) */\

  363.         "pavgb   %%mm2, %%mm0  \n\t"\

  364.         "pand    %%mm5, %%mm4  \n\t"\

  365.         "paddusb %%mm4, %%mm0  \n\t"\

  366.         "pavgb   %%mm1, %%mm3  \n\t"\

  367.         "pxor    %%mm5, %%mm4  \n\t"\

  368.         "paddusb %%mm4, %%mm3  \n\t"\

  369.         /* p0 += clip(delta, -tc0, tc0)

  370.          * q0 -= clip(delta, -tc0, tc0) */\

  371.         "movq    %%mm0, %%mm4  \n\t"\

  372.         "psubusb %%mm3, %%mm0  \n\t"\

  373.         "psubusb %%mm4, %%mm3  \n\t"\

  374.         "pminub  %%mm7, %%mm0  \n\t"\

  375.         "pminub  %%mm7, %%mm3  \n\t"\

  376.         "paddusb %%mm0, %%mm1  \n\t"\

  377.         "paddusb %%mm3, %%mm2  \n\t"\

  378.         "psubusb %%mm3, %%mm1  \n\t"\

  379.         "psubusb %%mm0, %%mm2  \n\t"

  380. 

  381. // in: mm0=p1 mm1=p0 mm2=q0 mm3=q1 mm7=(tc&mask) %8=mm_bone

  382. // out: (q1addr) = clip( (q2+((p0+q0+1)>>1))>>1, q1-tc0, q1+tc0 )

  383. // clobbers: q2, tmp, tc0

  384. #define H264_DEBLOCK_Q1(p1, q2, q2addr, q1addr, tc0, tmp)\

  385.         "movq     %%mm1,  "#tmp"   \n\t"\

  386.         "pavgb    %%mm2,  "#tmp"   \n\t"\

  387.         "pavgb    "#tmp", "#q2"    \n\t" /* avg(p2,avg(p0,q0)) */\

  388.         "pxor   "q2addr", "#tmp"   \n\t"\

  389.         "pand     %8,     "#tmp"   \n\t" /* (p2^avg(p0,q0))&1 */\

  390.         "psubusb  "#tmp", "#q2"    \n\t" /* (p2+((p0+q0+1)>>1))>>1 */\

  391.         "movq     "#p1",  "#tmp"   \n\t"\

  392.         "psubusb  "#tc0", "#tmp"   \n\t"\

  393.         "paddusb  "#p1",  "#tc0"   \n\t"\

  394.         "pmaxub   "#tmp", "#q2"    \n\t"\

  395.         "pminub   "#tc0", "#q2"    \n\t"\

  396.         "movq     "#q2",  "q1addr" \n\t"

  397. 

  398. static inline void h264_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)

  399. {

  400.     uint64_t tmp0;

  401.     uint64_t tc = (uint8_t)tc0[1]*0x01010000 | (uint8_t)tc0[0]*0x0101;

  402.     // with luma, tc0=0 doesn't mean no filtering, so we need a separate input mask

  403.     uint32_t mask[2] = { (tc0[0]>=0)*0xffffffff, (tc0[1]>=0)*0xffffffff };

  404. 

  405.     asm volatile(

  406.         "movq    (%1,%3), %%mm0    \n\t" //p1

  407.         "movq    (%1,%3,2), %%mm1  \n\t" //p0

  408.         "movq    (%2),    %%mm2    \n\t" //q0

  409.         "movq    (%2,%3), %%mm3    \n\t" //q1

  410.         H264_DEBLOCK_MASK(%6, %7)

  411.         "pand     %5,     %%mm7    \n\t"

  412.         "movq     %%mm7,  %0       \n\t"

  413. 

  414.         /* filter p1 */

  415.         "movq     (%1),   %%mm3    \n\t" //p2

  416.         DIFF_GT_MMX(%%mm1, %%mm3, %%mm5, %%mm6, %%mm4) // |p2-p0|>beta-1

  417.         "pandn    %%mm7,  %%mm6    \n\t"

  418.         "pcmpeqb  %%mm7,  %%mm6    \n\t"

  419.         "pand     %%mm7,  %%mm6    \n\t" // mask & |p2-p0|<beta

  420.         "pshufw  $80, %4, %%mm4    \n\t"

  421.         "pand     %%mm7,  %%mm4    \n\t" // mask & tc0

  422.         "movq     %8,     %%mm7    \n\t"

  423.         "pand     %%mm6,  %%mm7    \n\t" // mask & |p2-p0|<beta & 1

  424.         "pand     %%mm4,  %%mm6    \n\t" // mask & |p2-p0|<beta & tc0

  425.         "paddb    %%mm4,  %%mm7    \n\t" // tc++

  426.         H264_DEBLOCK_Q1(%%mm0, %%mm3, "(%1)", "(%1,%3)", %%mm6, %%mm4)

  427. 

  428.         /* filter q1 */

  429.         "movq    (%2,%3,2), %%mm4  \n\t" //q2

  430.         DIFF_GT_MMX(%%mm2, %%mm4, %%mm5, %%mm6, %%mm3) // |q2-q0|>beta-1

  431.         "pandn    %0,     %%mm6    \n\t"

  432.         "pcmpeqb  %0,     %%mm6    \n\t"

  433.         "pand     %0,     %%mm6    \n\t"

  434.         "pshufw  $80, %4, %%mm5    \n\t"

  435.         "pand     %%mm6,  %%mm5    \n\t"

  436.         "pand     %8,     %%mm6    \n\t"

  437.         "paddb    %%mm6,  %%mm7    \n\t"

  438.         "movq    (%2,%3), %%mm3    \n\t"

  439.         H264_DEBLOCK_Q1(%%mm3, %%mm4, "(%2,%3,2)", "(%2,%3)", %%mm5, %%mm6)

  440. 

  441.         /* filter p0, q0 */

  442.         H264_DEBLOCK_P0_Q0(%8, %9)

  443.         "movq      %%mm1, (%1,%3,2) \n\t"

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

  445. 

  446.         : "=m"(tmp0)

  447.         : "r"(pix-3*stride), "r"(pix), "r"((long)stride),

  448.           "m"(tc), "m"(*(uint64_t*)mask), "m"(alpha1), "m"(beta1),

  449.           "m"(mm_bone), "m"(ff_pb_3F)

  450.     );

  451. }

  452. 

  453. static void h264_v_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)

  454. {

  455.     if((tc0[0] & tc0[1]) >= 0)

  456.         h264_loop_filter_luma_mmx2(pix, stride, alpha-1, beta-1, tc0);

  457.     if((tc0[2] & tc0[3]) >= 0)

  458.         h264_loop_filter_luma_mmx2(pix+8, stride, alpha-1, beta-1, tc0+2);

  459. }

  460. static void h264_h_loop_filter_luma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)

  461. {

  462.     //FIXME: could cut some load/stores by merging transpose with filter

  463.     // also, it only needs to transpose 6x8

  464.     uint8_t trans[8*8];

  465.     int i;

  466.     for(i=0; i<2; i++, pix+=8*stride, tc0+=2) {

  467.         if((tc0[0] & tc0[1]) < 0)

  468.             continue;

  469.         transpose4x4(trans,       pix-4,          8, stride);

  470.         transpose4x4(trans  +4*8, pix,            8, stride);

  471.         transpose4x4(trans+4,     pix-4+4*stride, 8, stride);

  472.         transpose4x4(trans+4+4*8, pix  +4*stride, 8, stride);

  473.         h264_loop_filter_luma_mmx2(trans+4*8, 8, alpha-1, beta-1, tc0);

  474.         transpose4x4(pix-2,          trans  +2*8, stride, 8);

  475.         transpose4x4(pix-2+4*stride, trans+4+2*8, stride, 8);

  476.     }

  477. }

  478. 

  479. static inline void h264_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha1, int beta1, int8_t *tc0)

  480. {

  481.     asm volatile(

  482.         "movq    (%0),    %%mm0     \n\t" //p1

  483.         "movq    (%0,%2), %%mm1     \n\t" //p0

  484.         "movq    (%1),    %%mm2     \n\t" //q0

  485.         "movq    (%1,%2), %%mm3     \n\t" //q1

  486.         H264_DEBLOCK_MASK(%4, %5)

  487.         "movd      %3,    %%mm6     \n\t"

  488.         "punpcklbw %%mm6, %%mm6     \n\t"

  489.         "pand      %%mm6, %%mm7     \n\t" // mm7 = tc&mask

  490.         H264_DEBLOCK_P0_Q0(%6, %7)

  491.         "movq      %%mm1, (%0,%2)   \n\t"

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

  493. 

  494.         :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),

  495.            "r"(*(uint32_t*)tc0),

  496.            "m"(alpha1), "m"(beta1), "m"(mm_bone), "m"(ff_pb_3F)

  497.     );

  498. }

  499. 

  500. static void h264_v_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)

  501. {

  502.     h264_loop_filter_chroma_mmx2(pix, stride, alpha-1, beta-1, tc0);

  503. }

  504. 

  505. static void h264_h_loop_filter_chroma_mmx2(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0)

  506. {

  507.     //FIXME: could cut some load/stores by merging transpose with filter

  508.     uint8_t trans[8*4];

  509.     transpose4x4(trans, pix-2, 8, stride);

  510.     transpose4x4(trans+4, pix-2+4*stride, 8, stride);

  511.     h264_loop_filter_chroma_mmx2(trans+2*8, 8, alpha-1, beta-1, tc0);

  512.     transpose4x4(pix-2, trans, stride, 8);

  513.     transpose4x4(pix-2+4*stride, trans+4, stride, 8);

  514. }

  515. 

  516. // p0 = (p0 + q1 + 2*p1 + 2) >> 2

  517. #define H264_FILTER_CHROMA4(p0, p1, q1, one) \

  518.     "movq    "#p0", %%mm4  \n\t"\

  519.     "pxor    "#q1", %%mm4  \n\t"\

  520.     "pand   "#one", %%mm4  \n\t" /* mm4 = (p0^q1)&1 */\

  521.     "pavgb   "#q1", "#p0"  \n\t"\

  522.     "psubusb %%mm4, "#p0"  \n\t"\

  523.     "pavgb   "#p1", "#p0"  \n\t" /* dst = avg(p1, avg(p0,q1) - ((p0^q1)&1)) */\

  524. 

  525. static inline void h264_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha1, int beta1)

  526. {

  527.     asm volatile(

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

  529.         "movq    (%0,%2), %%mm1     \n\t"

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

  531.         "movq    (%1,%2), %%mm3     \n\t"

  532.         H264_DEBLOCK_MASK(%3, %4)

  533.         "movq    %%mm1,   %%mm5     \n\t"

  534.         "movq    %%mm2,   %%mm6     \n\t"

  535.         H264_FILTER_CHROMA4(%%mm1, %%mm0, %%mm3, %5) //p0'

  536.         H264_FILTER_CHROMA4(%%mm2, %%mm3, %%mm0, %5) //q0'

  537.         "psubb   %%mm5,   %%mm1     \n\t"

  538.         "psubb   %%mm6,   %%mm2     \n\t"

  539.         "pand    %%mm7,   %%mm1     \n\t"

  540.         "pand    %%mm7,   %%mm2     \n\t"

  541.         "paddb   %%mm5,   %%mm1     \n\t"

  542.         "paddb   %%mm6,   %%mm2     \n\t"

  543.         "movq    %%mm1,   (%0,%2)   \n\t"

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

  545.         :: "r"(pix-2*stride), "r"(pix), "r"((long)stride),

  546.            "m"(alpha1), "m"(beta1), "m"(mm_bone)

  547.     );

  548. }

  549. 

  550. static void h264_v_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)

  551. {

  552.     h264_loop_filter_chroma_intra_mmx2(pix, stride, alpha-1, beta-1);

  553. }

  554. 

  555. static void h264_h_loop_filter_chroma_intra_mmx2(uint8_t *pix, int stride, int alpha, int beta)

  556. {

  557.     //FIXME: could cut some load/stores by merging transpose with filter

  558.     uint8_t trans[8*4];

  559.     transpose4x4(trans, pix-2, 8, stride);

  560.     transpose4x4(trans+4, pix-2+4*stride, 8, stride);

  561.     h264_loop_filter_chroma_intra_mmx2(trans+2*8, 8, alpha-1, beta-1);

  562.     transpose4x4(pix-2, trans, stride, 8);

  563.     transpose4x4(pix-2+4*stride, trans+4, stride, 8);

  564. }

  565. 

  566. static void h264_loop_filter_strength_mmx2( int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],

  567.                                             int bidir, int edges, int step, int mask_mv0, int mask_mv1 ) {

  568.     int dir;

  569.     asm volatile(

  570.         "pxor %%mm7, %%mm7 \n\t"

  571.         "movq %0, %%mm6 \n\t"

  572.         "movq %1, %%mm5 \n\t"

  573.         "movq %2, %%mm4 \n\t"

  574.         ::"m"(ff_pb_1), "m"(ff_pb_3), "m"(ff_pb_7)

  575.     );

  576.     // could do a special case for dir==0 && edges==1, but it only reduces the

  577.     // average filter time by 1.2%

  578.     for( dir=1; dir>=0; dir-- ) {

  579.         const int d_idx = dir ? -8 : -1;

  580.         const int mask_mv = dir ? mask_mv1 : mask_mv0;

  581.         const uint64_t mask_dir = dir ? 0 : 0xffffffffffffffffULL;

  582.         int b_idx, edge, l;

  583.         for( b_idx=12, edge=0; edge<edges; edge+=step, b_idx+=8*step ) {

  584.             asm volatile(

  585.                 "pand %0, %%mm0 \n\t"

  586.                 ::"m"(mask_dir)

  587.             );

  588.             if(!(mask_mv & edge)) {

  589.                 asm volatile("pxor %%mm0, %%mm0 \n\t":);

  590.                 for( l = bidir; l >= 0; l-- ) {

  591.                     asm volatile(

  592.                         "movd %0, %%mm1 \n\t"

  593.                         "punpckldq %1, %%mm1 \n\t"

  594.                         "movq %%mm1, %%mm2 \n\t"

  595.                         "psrlw $7, %%mm2 \n\t"

  596.                         "pand %%mm6, %%mm2 \n\t"

  597.                         "por %%mm2, %%mm1 \n\t" // ref_cache with -2 mapped to -1

  598.                         "punpckldq %%mm1, %%mm2 \n\t"

  599.                         "pcmpeqb %%mm2, %%mm1 \n\t"

  600.                         "paddb %%mm6, %%mm1 \n\t"

  601.                         "punpckhbw %%mm7, %%mm1 \n\t" // ref[b] != ref[bn]

  602.                         "por %%mm1, %%mm0 \n\t"

  603. 

  604.                         "movq %2, %%mm1 \n\t"

  605.                         "movq %3, %%mm2 \n\t"

  606.                         "psubw %4, %%mm1 \n\t"

  607.                         "psubw %5, %%mm2 \n\t"

  608.                         "packsswb %%mm2, %%mm1 \n\t"

  609.                         "paddb %%mm5, %%mm1 \n\t"

  610.                         "pminub %%mm4, %%mm1 \n\t"

  611.                         "pcmpeqb %%mm4, %%mm1 \n\t" // abs(mv[b] - mv[bn]) >= limit

  612.                         "por %%mm1, %%mm0 \n\t"

  613.                         ::"m"(ref[l][b_idx]),

  614.                           "m"(ref[l][b_idx+d_idx]),

  615.                           "m"(mv[l][b_idx][0]),

  616.                           "m"(mv[l][b_idx+2][0]),

  617.                           "m"(mv[l][b_idx+d_idx][0]),

  618.                           "m"(mv[l][b_idx+d_idx+2][0])

  619.                     );

  620.                 }

  621.             }

  622.             asm volatile(

  623.                 "movd %0, %%mm1 \n\t"

  624.                 "por  %1, %%mm1 \n\t"

  625.                 "punpcklbw %%mm7, %%mm1 \n\t"

  626.                 "pcmpgtw %%mm7, %%mm1 \n\t" // nnz[b] || nnz[bn]

  627.                 ::"m"(nnz[b_idx]),

  628.                   "m"(nnz[b_idx+d_idx])

  629.             );

  630.             asm volatile(

  631.                 "pcmpeqw %%mm7, %%mm0 \n\t"

  632.                 "pcmpeqw %%mm7, %%mm0 \n\t"

  633.                 "psrlw $15, %%mm0 \n\t" // nonzero -> 1

  634.                 "psrlw $14, %%mm1 \n\t"

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

  636.                 "por %%mm1, %%mm2 \n\t"

  637.                 "psrlw $1, %%mm1 \n\t"

  638.                 "pandn %%mm2, %%mm1 \n\t"

  639.                 "movq %%mm1, %0 \n\t"

  640.                 :"=m"(*bS[dir][edge])

  641.                 ::"memory"

  642.             );

  643.         }

  644.         edges = 4;

  645.         step = 1;

  646.     }

  647.     asm volatile(

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

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

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

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

  652.         TRANSPOSE4(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4)

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

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

  655.         "movq %%mm4, 16(%0) \n\t"

  656.         "movq %%mm2, 24(%0) \n\t"

  657.         ::"r"(bS[0])

  658.         :"memory"

  659.     );

  660. }

  661. 

  662. /***********************************/

  663. /* motion compensation */

  664. 

  665. #define QPEL_H264V(A,B,C,D,E,F,OP)\

  666.         "movd (%0), "#F"            \n\t"\

  667.         "movq "#C", %%mm6           \n\t"\

  668.         "paddw "#D", %%mm6          \n\t"\

  669.         "psllw $2, %%mm6            \n\t"\

  670.         "psubw "#B", %%mm6          \n\t"\

  671.         "psubw "#E", %%mm6          \n\t"\

  672.         "pmullw %4, %%mm6           \n\t"\

  673.         "add %2, %0                 \n\t"\

  674.         "punpcklbw %%mm7, "#F"      \n\t"\

  675.         "paddw %5, "#A"             \n\t"\

  676.         "paddw "#F", "#A"           \n\t"\

  677.         "paddw "#A", %%mm6          \n\t"\

  678.         "psraw $5, %%mm6            \n\t"\

  679.         "packuswb %%mm6, %%mm6      \n\t"\

  680.         OP(%%mm6, (%1), A, d)\

  681.         "add %3, %1                 \n\t"

  682. 

  683. #define QPEL_H264HV(A,B,C,D,E,F,OF)\

  684.         "movd (%0), "#F"            \n\t"\

  685.         "movq "#C", %%mm6           \n\t"\

  686.         "paddw "#D", %%mm6          \n\t"\

  687.         "psllw $2, %%mm6            \n\t"\

  688.         "psubw "#B", %%mm6          \n\t"\

  689.         "psubw "#E", %%mm6          \n\t"\

  690.         "pmullw %3, %%mm6           \n\t"\

  691.         "add %2, %0                 \n\t"\

  692.         "punpcklbw %%mm7, "#F"      \n\t"\

  693.         "paddw "#F", "#A"           \n\t"\

  694.         "paddw "#A", %%mm6          \n\t"\

  695.         "movq %%mm6, "#OF"(%1)      \n\t"

  696. 

  697. #define QPEL_H264(OPNAME, OP, MMX)\

  698. static void OPNAME ## h264_qpel4_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  699.     int h=4;\

  700. \

  701.     asm volatile(\

  702.         "pxor %%mm7, %%mm7          \n\t"\

  703.         "movq %5, %%mm4             \n\t"\

  704.         "movq %6, %%mm5             \n\t"\

  705.         "1:                         \n\t"\

  706.         "movd  -1(%0), %%mm1        \n\t"\

  707.         "movd    (%0), %%mm2        \n\t"\

  708.         "movd   1(%0), %%mm3        \n\t"\

  709.         "movd   2(%0), %%mm0        \n\t"\

  710.         "punpcklbw %%mm7, %%mm1     \n\t"\

  711.         "punpcklbw %%mm7, %%mm2     \n\t"\

  712.         "punpcklbw %%mm7, %%mm3     \n\t"\

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

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

  715.         "paddw %%mm3, %%mm2         \n\t"\

  716.         "movd  -2(%0), %%mm0        \n\t"\

  717.         "movd   3(%0), %%mm3        \n\t"\

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

  719.         "punpcklbw %%mm7, %%mm3     \n\t"\

  720.         "paddw %%mm3, %%mm0         \n\t"\

  721.         "psllw $2, %%mm2            \n\t"\

  722.         "psubw %%mm1, %%mm2         \n\t"\

  723.         "pmullw %%mm4, %%mm2        \n\t"\

  724.         "paddw %%mm5, %%mm0         \n\t"\

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

  726.         "psraw $5, %%mm0            \n\t"\

  727.         "packuswb %%mm0, %%mm0      \n\t"\

  728.         OP(%%mm0, (%1),%%mm6, d)\

  729.         "add %3, %0                 \n\t"\

  730.         "add %4, %1                 \n\t"\

  731.         "decl %2                    \n\t"\

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

  733.         : "+a"(src), "+c"(dst), "+m"(h)\

  734.         : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\

  735.         : "memory"\

  736.     );\

  737. }\

  738. static void OPNAME ## h264_qpel4_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\

  739.     int h=4;\

  740.     asm volatile(\

  741.         "pxor %%mm7, %%mm7          \n\t"\

  742.         "movq %0, %%mm4             \n\t"\

  743.         "movq %1, %%mm5             \n\t"\

  744.         :: "m"(ff_pw_5), "m"(ff_pw_16)\

  745.     );\

  746.     do{\

  747.     asm volatile(\

  748.         "movd  -1(%0), %%mm1        \n\t"\

  749.         "movd    (%0), %%mm2        \n\t"\

  750.         "movd   1(%0), %%mm3        \n\t"\

  751.         "movd   2(%0), %%mm0        \n\t"\

  752.         "punpcklbw %%mm7, %%mm1     \n\t"\

  753.         "punpcklbw %%mm7, %%mm2     \n\t"\

  754.         "punpcklbw %%mm7, %%mm3     \n\t"\

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

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

  757.         "paddw %%mm3, %%mm2         \n\t"\

  758.         "movd  -2(%0), %%mm0        \n\t"\

  759.         "movd   3(%0), %%mm3        \n\t"\

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

  761.         "punpcklbw %%mm7, %%mm3     \n\t"\

  762.         "paddw %%mm3, %%mm0         \n\t"\

  763.         "psllw $2, %%mm2            \n\t"\

  764.         "psubw %%mm1, %%mm2         \n\t"\

  765.         "pmullw %%mm4, %%mm2        \n\t"\

  766.         "paddw %%mm5, %%mm0         \n\t"\

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

  768.         "movd   (%2), %%mm3         \n\t"\

  769.         "psraw $5, %%mm0            \n\t"\

  770.         "packuswb %%mm0, %%mm0      \n\t"\

  771.         PAVGB" %%mm3, %%mm0         \n\t"\

  772.         OP(%%mm0, (%1),%%mm6, d)\

  773.         "add %4, %0                 \n\t"\

  774.         "add %4, %1                 \n\t"\

  775.         "add %3, %2                 \n\t"\

  776.         : "+a"(src), "+c"(dst), "+d"(src2)\

  777.         : "D"((long)src2Stride), "S"((long)dstStride)\

  778.         : "memory"\

  779.     );\

  780.     }while(--h);\

  781. }\

  782. static void OPNAME ## h264_qpel4_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  783.     src -= 2*srcStride;\

  784.     asm volatile(\

  785.         "pxor %%mm7, %%mm7          \n\t"\

  786.         "movd (%0), %%mm0           \n\t"\

  787.         "add %2, %0                 \n\t"\

  788.         "movd (%0), %%mm1           \n\t"\

  789.         "add %2, %0                 \n\t"\

  790.         "movd (%0), %%mm2           \n\t"\

  791.         "add %2, %0                 \n\t"\

  792.         "movd (%0), %%mm3           \n\t"\

  793.         "add %2, %0                 \n\t"\

  794.         "movd (%0), %%mm4           \n\t"\

  795.         "add %2, %0                 \n\t"\

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

  797.         "punpcklbw %%mm7, %%mm1     \n\t"\

  798.         "punpcklbw %%mm7, %%mm2     \n\t"\

  799.         "punpcklbw %%mm7, %%mm3     \n\t"\

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

  801.         QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\

  802.         QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\

  803.         QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\

  804.         QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\

  805.          \

  806.         : "+a"(src), "+c"(dst)\

  807.         : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\

  808.         : "memory"\

  809.     );\

  810. }\

  811. static void OPNAME ## h264_qpel4_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\

  812.     int h=4;\

  813.     int w=3;\

  814.     src -= 2*srcStride+2;\

  815.     while(w--){\

  816.         asm volatile(\

  817.             "pxor %%mm7, %%mm7      \n\t"\

  818.             "movd (%0), %%mm0       \n\t"\

  819.             "add %2, %0             \n\t"\

  820.             "movd (%0), %%mm1       \n\t"\

  821.             "add %2, %0             \n\t"\

  822.             "movd (%0), %%mm2       \n\t"\

  823.             "add %2, %0             \n\t"\

  824.             "movd (%0), %%mm3       \n\t"\

  825.             "add %2, %0             \n\t"\

  826.             "movd (%0), %%mm4       \n\t"\

  827.             "add %2, %0             \n\t"\

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

  829.             "punpcklbw %%mm7, %%mm1 \n\t"\

  830.             "punpcklbw %%mm7, %%mm2 \n\t"\

  831.             "punpcklbw %%mm7, %%mm3 \n\t"\

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

  833.             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*8*3)\

  834.             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*8*3)\

  835.             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*8*3)\

  836.             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*8*3)\

  837.              \

  838.             : "+a"(src)\

  839.             : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\

  840.             : "memory"\

  841.         );\

  842.         tmp += 4;\

  843.         src += 4 - 9*srcStride;\

  844.     }\

  845.     tmp -= 3*4;\

  846.     asm volatile(\

  847.         "movq %4, %%mm6             \n\t"\

  848.         "1:                         \n\t"\

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

  850.         "paddw  10(%0), %%mm0       \n\t"\

  851.         "movq    2(%0), %%mm1       \n\t"\

  852.         "paddw   8(%0), %%mm1       \n\t"\

  853.         "movq    4(%0), %%mm2       \n\t"\

  854.         "paddw   6(%0), %%mm2       \n\t"\

  855.         "psubw %%mm1, %%mm0         \n\t"/*a-b   (abccba)*/\

  856.         "psraw $2, %%mm0            \n\t"/*(a-b)/4 */\

  857.         "psubw %%mm1, %%mm0         \n\t"/*(a-b)/4-b */\

  858.         "paddsw %%mm2, %%mm0        \n\t"\

  859.         "psraw $2, %%mm0            \n\t"/*((a-b)/4-b+c)/4 */\

  860.         "paddw %%mm6, %%mm2         \n\t"\

  861.         "paddw %%mm2, %%mm0         \n\t"/*(a-5*b+20*c)/16 +32 */\

  862.         "psraw $6, %%mm0            \n\t"\

  863.         "packuswb %%mm0, %%mm0      \n\t"\

  864.         OP(%%mm0, (%1),%%mm7, d)\

  865.         "add $24, %0                \n\t"\

  866.         "add %3, %1                 \n\t"\

  867.         "decl %2                    \n\t"\

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

  869.         : "+a"(tmp), "+c"(dst), "+m"(h)\

  870.         : "S"((long)dstStride), "m"(ff_pw_32)\

  871.         : "memory"\

  872.     );\

  873. }\

  874. \

  875. static void OPNAME ## h264_qpel8_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  876.     int h=8;\

  877.     asm volatile(\

  878.         "pxor %%mm7, %%mm7          \n\t"\

  879.         "movq %5, %%mm6             \n\t"\

  880.         "1:                         \n\t"\

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

  882.         "movq   1(%0), %%mm2        \n\t"\

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

  884.         "movq %%mm2, %%mm3          \n\t"\

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

  886.         "punpckhbw %%mm7, %%mm1     \n\t"\

  887.         "punpcklbw %%mm7, %%mm2     \n\t"\

  888.         "punpckhbw %%mm7, %%mm3     \n\t"\

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

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

  891.         "psllw $2, %%mm0            \n\t"\

  892.         "psllw $2, %%mm1            \n\t"\

  893.         "movq   -1(%0), %%mm2       \n\t"\

  894.         "movq    2(%0), %%mm4       \n\t"\

  895.         "movq %%mm2, %%mm3          \n\t"\

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

  897.         "punpcklbw %%mm7, %%mm2     \n\t"\

  898.         "punpckhbw %%mm7, %%mm3     \n\t"\

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

  900.         "punpckhbw %%mm7, %%mm5     \n\t"\

  901.         "paddw %%mm4, %%mm2         \n\t"\

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

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

  904.         "psubw %%mm5, %%mm1         \n\t"\

  905.         "pmullw %%mm6, %%mm0        \n\t"\

  906.         "pmullw %%mm6, %%mm1        \n\t"\

  907.         "movd   -2(%0), %%mm2       \n\t"\

  908.         "movd    7(%0), %%mm5       \n\t"\

  909.         "punpcklbw %%mm7, %%mm2     \n\t"\

  910.         "punpcklbw %%mm7, %%mm5     \n\t"\

  911.         "paddw %%mm3, %%mm2         \n\t"\

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

  913.         "movq %6, %%mm5             \n\t"\

  914.         "paddw %%mm5, %%mm2         \n\t"\

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

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

  917.         "paddw %%mm4, %%mm1         \n\t"\

  918.         "psraw $5, %%mm0            \n\t"\

  919.         "psraw $5, %%mm1            \n\t"\

  920.         "packuswb %%mm1, %%mm0      \n\t"\

  921.         OP(%%mm0, (%1),%%mm5, q)\

  922.         "add %3, %0                 \n\t"\

  923.         "add %4, %1                 \n\t"\

  924.         "decl %2                    \n\t"\

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

  926.         : "+a"(src), "+c"(dst), "+m"(h)\

  927.         : "d"((long)srcStride), "S"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\

  928.         : "memory"\

  929.     );\

  930. }\

  931. \

  932. static void OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\

  933.     int h=8;\

  934.     asm volatile(\

  935.         "pxor %%mm7, %%mm7          \n\t"\

  936.         "movq %0, %%mm6             \n\t"\

  937.         :: "m"(ff_pw_5)\

  938.     );\

  939.     do{\

  940.     asm volatile(\

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

  942.         "movq   1(%0), %%mm2        \n\t"\

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

  944.         "movq %%mm2, %%mm3          \n\t"\

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

  946.         "punpckhbw %%mm7, %%mm1     \n\t"\

  947.         "punpcklbw %%mm7, %%mm2     \n\t"\

  948.         "punpckhbw %%mm7, %%mm3     \n\t"\

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

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

  951.         "psllw $2, %%mm0            \n\t"\

  952.         "psllw $2, %%mm1            \n\t"\

  953.         "movq   -1(%0), %%mm2       \n\t"\

  954.         "movq    2(%0), %%mm4       \n\t"\

  955.         "movq %%mm2, %%mm3          \n\t"\

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

  957.         "punpcklbw %%mm7, %%mm2     \n\t"\

  958.         "punpckhbw %%mm7, %%mm3     \n\t"\

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

  960.         "punpckhbw %%mm7, %%mm5     \n\t"\

  961.         "paddw %%mm4, %%mm2         \n\t"\

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

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

  964.         "psubw %%mm5, %%mm1         \n\t"\

  965.         "pmullw %%mm6, %%mm0        \n\t"\

  966.         "pmullw %%mm6, %%mm1        \n\t"\

  967.         "movd   -2(%0), %%mm2       \n\t"\

  968.         "movd    7(%0), %%mm5       \n\t"\

  969.         "punpcklbw %%mm7, %%mm2     \n\t"\

  970.         "punpcklbw %%mm7, %%mm5     \n\t"\

  971.         "paddw %%mm3, %%mm2         \n\t"\

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

  973.         "movq %5, %%mm5             \n\t"\

  974.         "paddw %%mm5, %%mm2         \n\t"\

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

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

  977.         "paddw %%mm4, %%mm1         \n\t"\

  978.         "psraw $5, %%mm0            \n\t"\

  979.         "psraw $5, %%mm1            \n\t"\

  980.         "movq (%2), %%mm4           \n\t"\

  981.         "packuswb %%mm1, %%mm0      \n\t"\

  982.         PAVGB" %%mm4, %%mm0         \n\t"\

  983.         OP(%%mm0, (%1),%%mm5, q)\

  984.         "add %4, %0                 \n\t"\

  985.         "add %4, %1                 \n\t"\

  986.         "add %3, %2                 \n\t"\

  987.         : "+a"(src), "+c"(dst), "+d"(src2)\

  988.         : "D"((long)src2Stride), "S"((long)dstStride),\

  989.           "m"(ff_pw_16)\

  990.         : "memory"\

  991.     );\

  992.     }while(--h);\

  993. }\

  994. \

  995. static inline void OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h){\

  996.     int w= 2;\

  997.     src -= 2*srcStride;\

  998.     \

  999.     while(w--){\

  1000.       asm volatile(\

  1001.         "pxor %%mm7, %%mm7          \n\t"\

  1002.         "movd (%0), %%mm0           \n\t"\

  1003.         "add %2, %0                 \n\t"\

  1004.         "movd (%0), %%mm1           \n\t"\

  1005.         "add %2, %0                 \n\t"\

  1006.         "movd (%0), %%mm2           \n\t"\

  1007.         "add %2, %0                 \n\t"\

  1008.         "movd (%0), %%mm3           \n\t"\

  1009.         "add %2, %0                 \n\t"\

  1010.         "movd (%0), %%mm4           \n\t"\

  1011.         "add %2, %0                 \n\t"\

  1012.         "punpcklbw %%mm7, %%mm0     \n\t"\

  1013.         "punpcklbw %%mm7, %%mm1     \n\t"\

  1014.         "punpcklbw %%mm7, %%mm2     \n\t"\

  1015.         "punpcklbw %%mm7, %%mm3     \n\t"\

  1016.         "punpcklbw %%mm7, %%mm4     \n\t"\

  1017.         QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\

  1018.         QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\

  1019.         QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\

  1020.         QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\

  1021.         QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\

  1022.         QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\

  1023.         QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\

  1024.         QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\

  1025.          \

  1026.         : "+a"(src), "+c"(dst)\

  1027.         : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\

  1028.         : "memory"\

  1029.      );\

  1030.      if(h==16){\

  1031.         asm volatile(\

  1032.             QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\

  1033.             QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\

  1034.             QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\

  1035.             QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\

  1036.             QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\

  1037.             QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\

  1038.             QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\

  1039.             QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\

  1040.             \

  1041.            : "+a"(src), "+c"(dst)\

  1042.            : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\

  1043.            : "memory"\

  1044.         );\

  1045.      }\

  1046.      src += 4-(h+5)*srcStride;\

  1047.      dst += 4-h*dstStride;\

  1048.    }\

  1049. }\

  1050. static inline void OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\

  1051.     int h = size;\

  1052.     int w = (size+8)>>2;\

  1053.     src -= 2*srcStride+2;\

  1054.     while(w--){\

  1055.         asm volatile(\

  1056.             "pxor %%mm7, %%mm7      \n\t"\

  1057.             "movd (%0), %%mm0       \n\t"\

  1058.             "add %2, %0             \n\t"\

  1059.             "movd (%0), %%mm1       \n\t"\

  1060.             "add %2, %0             \n\t"\

  1061.             "movd (%0), %%mm2       \n\t"\

  1062.             "add %2, %0             \n\t"\

  1063.             "movd (%0), %%mm3       \n\t"\

  1064.             "add %2, %0             \n\t"\

  1065.             "movd (%0), %%mm4       \n\t"\

  1066.             "add %2, %0             \n\t"\

  1067.             "punpcklbw %%mm7, %%mm0 \n\t"\

  1068.             "punpcklbw %%mm7, %%mm1 \n\t"\

  1069.             "punpcklbw %%mm7, %%mm2 \n\t"\

  1070.             "punpcklbw %%mm7, %%mm3 \n\t"\

  1071.             "punpcklbw %%mm7, %%mm4 \n\t"\

  1072.             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*48)\

  1073.             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\

  1074.             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\

  1075.             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\

  1076.             QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\

  1077.             QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\

  1078.             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\

  1079.             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\

  1080.             : "+a"(src)\

  1081.             : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\

  1082.             : "memory"\

  1083.         );\

  1084.         if(size==16){\

  1085.             asm volatile(\

  1086.                 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1,  8*48)\

  1087.                 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2,  9*48)\

  1088.                 QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\

  1089.                 QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\

  1090.                 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\

  1091.                 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\

  1092.                 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\

  1093.                 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\

  1094.                 : "+a"(src)\

  1095.                 : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\

  1096.                 : "memory"\

  1097.             );\

  1098.         }\

  1099.         tmp += 4;\

  1100.         src += 4 - (size+5)*srcStride;\

  1101.     }\

  1102.     tmp -= size+8;\

  1103.     w = size>>4;\

  1104.     do{\

  1105.     h = size;\

  1106.     asm volatile(\

  1107.         "movq %4, %%mm6             \n\t"\

  1108.         "1:                         \n\t"\

  1109.         "movq     (%0), %%mm0       \n\t"\

  1110.         "movq    8(%0), %%mm3       \n\t"\

  1111.         "movq    2(%0), %%mm1       \n\t"\

  1112.         "movq   10(%0), %%mm4       \n\t"\

  1113.         "paddw   %%mm4, %%mm0       \n\t"\

  1114.         "paddw   %%mm3, %%mm1       \n\t"\

  1115.         "paddw  18(%0), %%mm3       \n\t"\

  1116.         "paddw  16(%0), %%mm4       \n\t"\

  1117.         "movq    4(%0), %%mm2       \n\t"\

  1118.         "movq   12(%0), %%mm5       \n\t"\

  1119.         "paddw   6(%0), %%mm2       \n\t"\

  1120.         "paddw  14(%0), %%mm5       \n\t"\

  1121.         "psubw %%mm1, %%mm0         \n\t"\

  1122.         "psubw %%mm4, %%mm3         \n\t"\

  1123.         "psraw $2, %%mm0            \n\t"\

  1124.         "psraw $2, %%mm3            \n\t"\

  1125.         "psubw %%mm1, %%mm0         \n\t"\

  1126.         "psubw %%mm4, %%mm3         \n\t"\

  1127.         "paddsw %%mm2, %%mm0        \n\t"\

  1128.         "paddsw %%mm5, %%mm3        \n\t"\

  1129.         "psraw $2, %%mm0            \n\t"\

  1130.         "psraw $2, %%mm3            \n\t"\

  1131.         "paddw %%mm6, %%mm2         \n\t"\

  1132.         "paddw %%mm6, %%mm5         \n\t"\

  1133.         "paddw %%mm2, %%mm0         \n\t"\

  1134.         "paddw %%mm5, %%mm3         \n\t"\

  1135.         "psraw $6, %%mm0            \n\t"\

  1136.         "psraw $6, %%mm3            \n\t"\

  1137.         "packuswb %%mm3, %%mm0      \n\t"\

  1138.         OP(%%mm0, (%1),%%mm7, q)\

  1139.         "add $48, %0                \n\t"\

  1140.         "add %3, %1                 \n\t"\

  1141.         "decl %2                    \n\t"\

  1142.         " jnz 1b                    \n\t"\

  1143.         : "+a"(tmp), "+c"(dst), "+m"(h)\

  1144.         : "S"((long)dstStride), "m"(ff_pw_32)\

  1145.         : "memory"\

  1146.     );\

  1147.     tmp += 8 - size*24;\

  1148.     dst += 8 - size*dstStride;\

  1149.     }while(w--);\

  1150. }\

  1151. \

  1152. static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1153.     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\

  1154. }\

  1155. static void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1156.     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\

  1157.     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\

  1158. }\

  1159. \

  1160. static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1161.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\

  1162.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  1163.     src += 8*srcStride;\

  1164.     dst += 8*dstStride;\

  1165.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\

  1166.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  1167. }\

  1168. \

  1169. static void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\

  1170.     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\

  1171.     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\

  1172.     src += 8*dstStride;\

  1173.     dst += 8*dstStride;\

  1174.     src2 += 8*src2Stride;\

  1175.     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\

  1176.     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\

  1177. }\

  1178. \

  1179. static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\

  1180.     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 8);\

  1181. }\

  1182. \

  1183. static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\

  1184.     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 16);\

  1185. }\

  1186. \

  1187. static void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\

  1188. {\

  1189.     asm volatile(\

  1190.         "movq       %5,  %%mm6          \n\t"\

  1191.         "movq      (%1), %%mm0          \n\t"\

  1192.         "movq    24(%1), %%mm1          \n\t"\

  1193.         "paddw    %%mm6, %%mm0          \n\t"\

  1194.         "paddw    %%mm6, %%mm1          \n\t"\

  1195.         "psraw      $5,  %%mm0          \n\t"\

  1196.         "psraw      $5,  %%mm1          \n\t"\

  1197.         "packuswb %%mm0, %%mm0          \n\t"\

  1198.         "packuswb %%mm1, %%mm1          \n\t"\

  1199.         PAVGB"     (%0), %%mm0          \n\t"\

  1200.         PAVGB"  (%0,%3), %%mm1          \n\t"\

  1201.         OP(%%mm0, (%2),    %%mm4, d)\

  1202.         OP(%%mm1, (%2,%4), %%mm5, d)\

  1203.         "lea  (%0,%3,2), %0             \n\t"\

  1204.         "lea  (%2,%4,2), %2             \n\t"\

  1205.         "movq    48(%1), %%mm0          \n\t"\

  1206.         "movq    72(%1), %%mm1          \n\t"\

  1207.         "paddw    %%mm6, %%mm0          \n\t"\

  1208.         "paddw    %%mm6, %%mm1          \n\t"\

  1209.         "psraw      $5,  %%mm0          \n\t"\

  1210.         "psraw      $5,  %%mm1          \n\t"\

  1211.         "packuswb %%mm0, %%mm0          \n\t"\

  1212.         "packuswb %%mm1, %%mm1          \n\t"\

  1213.         PAVGB"     (%0), %%mm0          \n\t"\

  1214.         PAVGB"  (%0,%3), %%mm1          \n\t"\

  1215.         OP(%%mm0, (%2),    %%mm4, d)\

  1216.         OP(%%mm1, (%2,%4), %%mm5, d)\

  1217.         :"+a"(src8), "+c"(src16), "+d"(dst)\

  1218.         :"S"((long)src8Stride), "D"((long)dstStride), "m"(ff_pw_16)\

  1219.         :"memory");\

  1220. }\

  1221. static void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\

  1222. {\

  1223.     asm volatile(\

  1224.         "movq       %0,  %%mm6          \n\t"\

  1225.         ::"m"(ff_pw_16)\

  1226.         );\

  1227.     while(h--){\

  1228.     asm volatile(\

  1229.         "movq      (%1), %%mm0          \n\t"\

  1230.         "movq     8(%1), %%mm1          \n\t"\

  1231.         "paddw    %%mm6, %%mm0          \n\t"\

  1232.         "paddw    %%mm6, %%mm1          \n\t"\

  1233.         "psraw      $5,  %%mm0          \n\t"\

  1234.         "psraw      $5,  %%mm1          \n\t"\

  1235.         "packuswb %%mm1, %%mm0          \n\t"\

  1236.         PAVGB"     (%0), %%mm0          \n\t"\

  1237.         OP(%%mm0, (%2), %%mm5, q)\

  1238.         ::"a"(src8), "c"(src16), "d"(dst)\

  1239.         :"memory");\

  1240.         src8 += src8Stride;\

  1241.         src16 += 24;\

  1242.         dst += dstStride;\

  1243.     }\

  1244. }\

  1245. static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\

  1246. {\

  1247.     OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\

  1248.     OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\

  1249. }\

  1250. 

  1251. 

  1252. #define H264_MC(OPNAME, SIZE, MMX) \

  1253. static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  1254.     OPNAME ## pixels ## SIZE ## _mmx(dst, src, stride, SIZE);\

  1255. }\

  1256. \

  1257. static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1258.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\

  1259. }\

  1260. \

  1261. static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1262.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\

  1263. }\

  1264. \

  1265. static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1266.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\

  1267. }\

  1268. \

  1269. static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1270.     uint64_t temp[SIZE*SIZE/8];\

  1271.     uint8_t * const half= (uint8_t*)temp;\

  1272.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\

  1273.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\

  1274. }\

  1275. \

  1276. static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1277.     OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\

  1278. }\

  1279. \

  1280. static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1281.     uint64_t temp[SIZE*SIZE/8];\

  1282.     uint8_t * const half= (uint8_t*)temp;\

  1283.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\

  1284.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, half, stride, stride, SIZE);\

  1285. }\

  1286. \

  1287. static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1288.     uint64_t temp[SIZE*SIZE/8];\

  1289.     uint8_t * const halfV= (uint8_t*)temp;\

  1290.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\

  1291.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfV, stride, SIZE);\

  1292. }\

  1293. \

  1294. static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1295.     uint64_t temp[SIZE*SIZE/8];\

  1296.     uint8_t * const halfV= (uint8_t*)temp;\

  1297.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\

  1298.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfV, stride, SIZE);\

  1299. }\

  1300. \

  1301. static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1302.     uint64_t temp[SIZE*SIZE/8];\

  1303.     uint8_t * const halfV= (uint8_t*)temp;\

  1304.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\

  1305.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfV, stride, SIZE);\

  1306. }\

  1307. \

  1308. static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1309.     uint64_t temp[SIZE*SIZE/8];\

  1310.     uint8_t * const halfV= (uint8_t*)temp;\

  1311.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\

  1312.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfV, stride, SIZE);\

  1313. }\

  1314. \

  1315. static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1316.     uint64_t temp[SIZE*(SIZE<8?12:24)/4];\

  1317.     int16_t * const tmp= (int16_t*)temp;\

  1318.     OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, tmp, src, stride, SIZE, stride);\

  1319. }\

  1320. \

  1321. static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1322.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\

  1323.     uint8_t * const halfHV= (uint8_t*)temp;\

  1324.     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE/2;\

  1325.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\

  1326.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\

  1327. }\

  1328. \

  1329. static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1330.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\

  1331.     uint8_t * const halfHV= (uint8_t*)temp;\

  1332.     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE/2;\

  1333.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\

  1334.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\

  1335. }\

  1336. \

  1337. static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1338.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\

  1339.     int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\

  1340.     uint8_t * const halfHV= ((uint8_t*)temp);\

  1341.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\

  1342.     OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\

  1343. }\

  1344. \

  1345. static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1346.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\

  1347.     int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\

  1348.     uint8_t * const halfHV= ((uint8_t*)temp);\

  1349.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\

  1350.     OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\

  1351. }\

  1352. 

  1353. 

  1354. #define AVG_3DNOW_OP(a,b,temp, size) \

  1355. "mov" #size " " #b ", " #temp "   \n\t"\

  1356. "pavgusb " #temp ", " #a "        \n\t"\

  1357. "mov" #size " " #a ", " #b "      \n\t"

  1358. #define AVG_MMX2_OP(a,b,temp, size) \

  1359. "mov" #size " " #b ", " #temp "   \n\t"\

  1360. "pavgb " #temp ", " #a "          \n\t"\

  1361. "mov" #size " " #a ", " #b "      \n\t"

  1362. 

  1363. #define PAVGB "pavgusb"

  1364. QPEL_H264(put_,       PUT_OP, 3dnow)

  1365. QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)

  1366. #undef PAVGB

  1367. #define PAVGB "pavgb"

  1368. QPEL_H264(put_,       PUT_OP, mmx2)

  1369. QPEL_H264(avg_,  AVG_MMX2_OP, mmx2)

  1370. #undef PAVGB

  1371. 

  1372. H264_MC(put_, 4, 3dnow)

  1373. H264_MC(put_, 8, 3dnow)

  1374. H264_MC(put_, 16,3dnow)

  1375. H264_MC(avg_, 4, 3dnow)

  1376. H264_MC(avg_, 8, 3dnow)

  1377. H264_MC(avg_, 16,3dnow)

  1378. H264_MC(put_, 4, mmx2)

  1379. H264_MC(put_, 8, mmx2)

  1380. H264_MC(put_, 16,mmx2)

  1381. H264_MC(avg_, 4, mmx2)

  1382. H264_MC(avg_, 8, mmx2)

  1383. H264_MC(avg_, 16,mmx2)

  1384. 

  1385. 

  1386. #define H264_CHROMA_OP(S,D)

  1387. #define H264_CHROMA_OP4(S,D,T)

  1388. #define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx

  1389. #define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_mmx

  1390. #define H264_CHROMA_MC2_TMPL put_h264_chroma_mc2_mmx2

  1391. #define H264_CHROMA_MC8_MV0 put_pixels8_mmx

  1392. #include "dsputil_h264_template_mmx.c"

  1393. #undef H264_CHROMA_OP

  1394. #undef H264_CHROMA_OP4

  1395. #undef H264_CHROMA_MC8_TMPL

  1396. #undef H264_CHROMA_MC4_TMPL

  1397. #undef H264_CHROMA_MC2_TMPL

  1398. #undef H264_CHROMA_MC8_MV0

  1399. 

  1400. #define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"

  1401. #define H264_CHROMA_OP4(S,D,T) "movd  " #S ", " #T " \n\t"\

  1402.                                "pavgb " #T ", " #D " \n\t"

  1403. #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2

  1404. #define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_mmx2

  1405. #define H264_CHROMA_MC2_TMPL avg_h264_chroma_mc2_mmx2

  1406. #define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2

  1407. #include "dsputil_h264_template_mmx.c"

  1408. #undef H264_CHROMA_OP

  1409. #undef H264_CHROMA_OP4

  1410. #undef H264_CHROMA_MC8_TMPL

  1411. #undef H264_CHROMA_MC4_TMPL

  1412. #undef H264_CHROMA_MC2_TMPL

  1413. #undef H264_CHROMA_MC8_MV0

  1414. 

  1415. #define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"

  1416. #define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\

  1417.                                "pavgusb " #T ", " #D " \n\t"

  1418. #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow

  1419. #define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_3dnow

  1420. #define H264_CHROMA_MC8_MV0 avg_pixels8_3dnow

  1421. #include "dsputil_h264_template_mmx.c"

  1422. #undef H264_CHROMA_OP

  1423. #undef H264_CHROMA_OP4

  1424. #undef H264_CHROMA_MC8_TMPL

  1425. #undef H264_CHROMA_MC4_TMPL

  1426. #undef H264_CHROMA_MC8_MV0

  1427. 

  1428. /***********************************/

  1429. /* weighted prediction */

  1430. 

  1431. static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)

  1432. {

  1433.     int x, y;

  1434.     offset <<= log2_denom;

  1435.     offset += (1 << log2_denom) >> 1;

  1436.     asm volatile(

  1437.         "movd    %0, %%mm4        \n\t"

  1438.         "movd    %1, %%mm5        \n\t"

  1439.         "movd    %2, %%mm6        \n\t"

  1440.         "pshufw  $0, %%mm4, %%mm4 \n\t"

  1441.         "pshufw  $0, %%mm5, %%mm5 \n\t"

  1442.         "pxor    %%mm7, %%mm7     \n\t"

  1443.         :: "g"(weight), "g"(offset), "g"(log2_denom)

  1444.     );

  1445.     for(y=0; y<h; y+=2){

  1446.         for(x=0; x<w; x+=4){

  1447.             asm volatile(

  1448.                 "movd      %0,    %%mm0 \n\t"

  1449.                 "movd      %1,    %%mm1 \n\t"

  1450.                 "punpcklbw %%mm7, %%mm0 \n\t"

  1451.                 "punpcklbw %%mm7, %%mm1 \n\t"

  1452.                 "pmullw    %%mm4, %%mm0 \n\t"

  1453.                 "pmullw    %%mm4, %%mm1 \n\t"

  1454.                 "paddsw    %%mm5, %%mm0 \n\t"

  1455.                 "paddsw    %%mm5, %%mm1 \n\t"

  1456.                 "psraw     %%mm6, %%mm0 \n\t"

  1457.                 "psraw     %%mm6, %%mm1 \n\t"

  1458.                 "packuswb  %%mm7, %%mm0 \n\t"

  1459.                 "packuswb  %%mm7, %%mm1 \n\t"

  1460.                 "movd      %%mm0, %0    \n\t"

  1461.                 "movd      %%mm1, %1    \n\t"

  1462.                 : "+m"(*(uint32_t*)(dst+x)),

  1463.                   "+m"(*(uint32_t*)(dst+x+stride))

  1464.             );

  1465.         }

  1466.         dst += 2*stride;

  1467.     }

  1468. }

  1469. 

  1470. static inline void ff_h264_biweight_WxH_mmx2(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset, int w, int h)

  1471. {

  1472.     int x, y;

  1473.     offset = ((offset + 1) | 1) << log2_denom;

  1474.     asm volatile(

  1475.         "movd    %0, %%mm3        \n\t"

  1476.         "movd    %1, %%mm4        \n\t"

  1477.         "movd    %2, %%mm5        \n\t"

  1478.         "movd    %3, %%mm6        \n\t"

  1479.         "pshufw  $0, %%mm3, %%mm3 \n\t"

  1480.         "pshufw  $0, %%mm4, %%mm4 \n\t"

  1481.         "pshufw  $0, %%mm5, %%mm5 \n\t"

  1482.         "pxor    %%mm7, %%mm7     \n\t"

  1483.         :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)

  1484.     );

  1485.     for(y=0; y<h; y++){

  1486.         for(x=0; x<w; x+=4){

  1487.             asm volatile(

  1488.                 "movd      %0,    %%mm0 \n\t"

  1489.                 "movd      %1,    %%mm1 \n\t"

  1490.                 "punpcklbw %%mm7, %%mm0 \n\t"

  1491.                 "punpcklbw %%mm7, %%mm1 \n\t"

  1492.                 "pmullw    %%mm3, %%mm0 \n\t"

  1493.                 "pmullw    %%mm4, %%mm1 \n\t"

  1494.                 "paddsw    %%mm1, %%mm0 \n\t"

  1495.                 "paddsw    %%mm5, %%mm0 \n\t"

  1496.                 "psraw     %%mm6, %%mm0 \n\t"

  1497.                 "packuswb  %%mm0, %%mm0 \n\t"

  1498.                 "movd      %%mm0, %0    \n\t"

  1499.                 : "+m"(*(uint32_t*)(dst+x))

  1500.                 :  "m"(*(uint32_t*)(src+x))

  1501.             );

  1502.         }

  1503.         src += stride;

  1504.         dst += stride;

  1505.     }

  1506. }

  1507. 

  1508. #define H264_WEIGHT(W,H) \

  1509. static void ff_h264_biweight_ ## W ## x ## H ## _mmx2(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset){ \

  1510.     ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \

  1511. } \

  1512. static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \

  1513.     ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \

  1514. }

  1515. 

  1516. H264_WEIGHT(16,16)

  1517. H264_WEIGHT(16, 8)

  1518. H264_WEIGHT( 8,16)

  1519. H264_WEIGHT( 8, 8)

  1520. H264_WEIGHT( 8, 4)

  1521. H264_WEIGHT( 4, 8)

  1522. H264_WEIGHT( 4, 4)

  1523. H264_WEIGHT( 4, 2)