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.
5553 Commits
32 Branches
325MB
Tree: c17be81763
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c17be81763'
${ noResults }
FFmpeg/libavcodec/i386/h264dsp_mmx.c

1434 lines
55KB
Raw Blame History 

  1. /*

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

  3.  *

  4.  * This library is free software; you can redistribute it and/or

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

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

  7.  * version 2 of the License, or (at your option) any later version.

  8.  *

  9.  * This library is distributed in the hope that it will be useful,

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

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

  12.  * Lesser General Public License for more details.

  13.  *

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

  15.  * License along with this library; if not, write to the Free Software

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

  17.  */

  18. 

  19. 

  20. /***********************************/

  21. /* IDCT */

  22. 

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

  24. #define SUMSUB_BA( a, b ) \

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

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

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

  28. 

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

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

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

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

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

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

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

  36. 

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

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

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

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

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

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

  43. 

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

  45.     SUMSUB_BA  ( s02, d02 )\

  46.     SUMSUBD2_AB( s13, d13, t )\

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

  48. 

  49. #define SBUTTERFLY(a,b,t,n)\

  50.     "movq " #a ", " #t "                \n\t" /* abcd */\

  51.     "punpckl" #n " " #b ", " #a "       \n\t" /* aebf */\

  52.     "punpckh" #n " " #b ", " #t "       \n\t" /* cgdh */\

  53. 

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

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

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

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

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

  59. 

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

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

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

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

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

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

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

  67. 

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

  69. {

  70.     /* Load dct coeffs */

  71.     asm volatile(

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

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

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

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

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

  77. 

  78.     asm volatile(

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

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

  81. 

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

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

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

  85. 

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

  87. 

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

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

  90. 

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

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

  93. 

  94.     asm volatile(

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

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

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

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

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

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

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

  102.         : "+r"(dst)

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

  104.     );

  105. }

  106. 

  107. static inline void h264_idct8_1d(int16_t *block)

  108. {

  109.     asm volatile(

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

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

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

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

  114. 

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

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

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

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

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

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

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

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

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

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

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

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

  127. 

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

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

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

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

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

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

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

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

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

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

  138. 

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

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

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

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

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

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

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

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

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

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

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

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

  151. 

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

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

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

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

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

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

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

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

  160. 

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

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

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

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

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

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

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

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

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

  170.         :: "r"(block)

  171.     );

  172. }

  173. 

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

  175. {

  176.     int i;

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

  178. 

  179.     block[0] += 32;

  180. 

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

  182.         uint64_t tmp;

  183. 

  184.         h264_idct8_1d(block+4*i);

  185. 

  186.         asm volatile(

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

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

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

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

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

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

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

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

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

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

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

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

  199.             : "=m"(tmp)

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

  201.             : "memory"

  202.         );

  203.     }

  204. 

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

  206.         h264_idct8_1d(b2+4*i);

  207. 

  208.         asm volatile(

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

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

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

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

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

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

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

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

  217. 

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

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

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

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

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

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

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

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

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

  227.             : "memory"

  228.         );

  229.     }

  230. 

  231.     add_pixels_clamped_mmx(b2, dst, stride);

  232. }

  233. 

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

  235. {

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

  237.     asm volatile(

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

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

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

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

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

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

  244.         ::"r"(dc)

  245.     );

  246.     asm volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  267.     );

  268. }

  269. 

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

  271. {

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

  273.     int y;

  274.     asm volatile(

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

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

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

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

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

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

  281.         ::"r"(dc)

  282.     );

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

  284.     asm volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  305.     );

  306.     }

  307. }

  308. 

  309. 

  310. /***********************************/

  311. /* deblocking */

  312. 

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

  314. // clobbers: t

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

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

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

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

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

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

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

  322. 

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

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

  325. // clobbers: mm4,mm6

  326. #define H264_DEBLOCK_MASK(alpha1, beta1) \

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

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

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

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

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

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

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

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

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

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

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

  338. 

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

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

  341. // clobbers: mm0,3-6

  342. #define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  383. 

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

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

  386. // clobbers: q2, tmp, tc0

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

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

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

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

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

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

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

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

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

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

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

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

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

  400. 

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

  402. {

  403.     uint64_t tmp0;

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

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

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

  407. 

  408.     asm volatile(

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

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

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

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

  413.         H264_DEBLOCK_MASK(%6, %7)

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

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

  416. 

  417.         /* filter p1 */

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

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

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

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

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

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

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

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

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

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

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

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

  430. 

  431.         /* filter q1 */

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

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

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

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

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

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

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

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

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

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

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

  443. 

  444.         /* filter p0, q0 */

  445.         H264_DEBLOCK_P0_Q0(%8, %9)

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

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

  448. 

  449.         : "=m"(tmp0)

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

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

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

  453.     );

  454. }

  455. 

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

  457. {

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

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

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

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

  462. }

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

  464. {

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

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

  467.     uint8_t trans[8*8];

  468.     int i;

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

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

  471.             continue;

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

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

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

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

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

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

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

  479.     }

  480. }

  481. 

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

  483. {

  484.     asm volatile(

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

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

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

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

  489.         H264_DEBLOCK_MASK(%4, %5)

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

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

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

  493.         H264_DEBLOCK_P0_Q0(%6, %7)

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

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

  496. 

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

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

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

  500.     );

  501. }

  502. 

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

  504. {

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

  506. }

  507. 

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

  509. {

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

  511.     uint8_t trans[8*4];

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

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

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

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

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

  517. }

  518. 

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

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

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

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

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

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

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

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

  527. 

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

  529. {

  530.     asm volatile(

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

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

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

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

  535.         H264_DEBLOCK_MASK(%3, %4)

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  550.     );

  551. }

  552. 

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

  554. {

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

  556. }

  557. 

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

  559. {

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

  561.     uint8_t trans[8*4];

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

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

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

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

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

  567. }

  568. 

  569. 

  570. /***********************************/

  571. /* motion compensation */

  572. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  590. 

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

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

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

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

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

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

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

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

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

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

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

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

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

  604. 

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

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

  607.     int h=4;\

  608. \

  609.     asm volatile(\

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

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

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

  613.         "1:                         \n\t"\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  643.         : "memory"\

  644.     );\

  645. }\

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

  647.     int h=4;\

  648.     asm volatile(\

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

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

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

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

  653.     );\

  654.     do{\

  655.     asm volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  686.         : "memory"\

  687.     );\

  688.     }while(--h);\

  689. }\

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

  691.     src -= 2*srcStride;\

  692.     asm volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  713.          \

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

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

  716.         : "memory"\

  717.     );\

  718. }\

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

  720.     int h=4;\

  721.     int w=3;\

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

  723.     while(w--){\

  724.         asm volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  745.              \

  746.             : "+a"(src)\

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

  748.             : "memory"\

  749.         );\

  750.         tmp += 4;\

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

  752.     }\

  753.     tmp -= 3*4;\

  754.     asm volatile(\

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

  756.         "1:                         \n\t"\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  779.         : "memory"\

  780.     );\

  781. }\

  782. \

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

  784.     int h=8;\

  785.     asm volatile(\

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

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

  788.         "1:                         \n\t"\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  836.         : "memory"\

  837.     );\

  838. }\

  839. \

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

  841.     int h=8;\

  842.     asm volatile(\

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

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

  845.         :: "m"(ff_pw_5)\

  846.     );\

  847.     do{\

  848.     asm volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  897.           "m"(ff_pw_16)\

  898.         : "memory"\

  899.     );\

  900.     }while(--h);\

  901. }\

  902. \

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

  904.     int w= 2;\

  905.     src -= 2*srcStride;\

  906.     \

  907.     while(w--){\

  908.       asm volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  933.          \

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

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

  936.         : "memory"\

  937.      );\

  938.      if(h==16){\

  939.         asm volatile(\

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

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

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

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

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

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

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

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

  948.             \

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

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

  951.            : "memory"\

  952.         );\

  953.      }\

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

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

  956.    }\

  957. }\

  958. 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){\

  959.     int h = size;\

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

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

  962.     while(w--){\

  963.         asm volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  980.             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*48)\

  981.             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\

  982.             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\

  983.             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\

  984.             QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\

  985.             QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\

  986.             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\

  987.             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\

  988.             : "+a"(src)\

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

  990.             : "memory"\

  991.         );\

  992.         if(size==16){\

  993.             asm volatile(\

  994.                 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1,  8*48)\

  995.                 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2,  9*48)\

  996.                 QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\

  997.                 QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\

  998.                 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\

  999.                 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\

  1000.                 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\

  1001.                 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\

  1002.                 : "+a"(src)\

  1003.                 : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\

  1004.                 : "memory"\

  1005.             );\

  1006.         }\

  1007.         tmp += 4;\

  1008.         src += 4 - (size+5)*srcStride;\

  1009.     }\

  1010.     tmp -= size+8;\

  1011.     w = size>>4;\

  1012.     do{\

  1013.     h = size;\

  1014.     asm volatile(\

  1015.         "movq %4, %%mm6             \n\t"\

  1016.         "1:                         \n\t"\

  1017.         "movq     (%0), %%mm0       \n\t"\

  1018.         "movq    8(%0), %%mm3       \n\t"\

  1019.         "movq    2(%0), %%mm1       \n\t"\

  1020.         "movq   10(%0), %%mm4       \n\t"\

  1021.         "paddw   %%mm4, %%mm0       \n\t"\

  1022.         "paddw   %%mm3, %%mm1       \n\t"\

  1023.         "paddw  18(%0), %%mm3       \n\t"\

  1024.         "paddw  16(%0), %%mm4       \n\t"\

  1025.         "movq    4(%0), %%mm2       \n\t"\

  1026.         "movq   12(%0), %%mm5       \n\t"\

  1027.         "paddw   6(%0), %%mm2       \n\t"\

  1028.         "paddw  14(%0), %%mm5       \n\t"\

  1029.         "psubw %%mm1, %%mm0         \n\t"\

  1030.         "psubw %%mm4, %%mm3         \n\t"\

  1031.         "psraw $2, %%mm0            \n\t"\

  1032.         "psraw $2, %%mm3            \n\t"\

  1033.         "psubw %%mm1, %%mm0         \n\t"\

  1034.         "psubw %%mm4, %%mm3         \n\t"\

  1035.         "paddsw %%mm2, %%mm0        \n\t"\

  1036.         "paddsw %%mm5, %%mm3        \n\t"\

  1037.         "psraw $2, %%mm0            \n\t"\

  1038.         "psraw $2, %%mm3            \n\t"\

  1039.         "paddw %%mm6, %%mm2         \n\t"\

  1040.         "paddw %%mm6, %%mm5         \n\t"\

  1041.         "paddw %%mm2, %%mm0         \n\t"\

  1042.         "paddw %%mm5, %%mm3         \n\t"\

  1043.         "psraw $6, %%mm0            \n\t"\

  1044.         "psraw $6, %%mm3            \n\t"\

  1045.         "packuswb %%mm3, %%mm0      \n\t"\

  1046.         OP(%%mm0, (%1),%%mm7, q)\

  1047.         "add $48, %0                \n\t"\

  1048.         "add %3, %1                 \n\t"\

  1049.         "decl %2                    \n\t"\

  1050.         " jnz 1b                    \n\t"\

  1051.         : "+a"(tmp), "+c"(dst), "+m"(h)\

  1052.         : "S"((long)dstStride), "m"(ff_pw_32)\

  1053.         : "memory"\

  1054.     );\

  1055.     tmp += 8 - size*24;\

  1056.     dst += 8 - size*dstStride;\

  1057.     }while(w--);\

  1058. }\

  1059. \

  1060. static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1061.     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\

  1062. }\

  1063. static void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1064.     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\

  1065.     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\

  1066. }\

  1067. \

  1068. static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1069.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\

  1070.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  1071.     src += 8*srcStride;\

  1072.     dst += 8*dstStride;\

  1073.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\

  1074.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  1075. }\

  1076. \

  1077. static void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\

  1078.     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\

  1079.     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\

  1080.     src += 8*dstStride;\

  1081.     dst += 8*dstStride;\

  1082.     src2 += 8*src2Stride;\

  1083.     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\

  1084.     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\

  1085. }\

  1086. \

  1087. static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\

  1088.     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 8);\

  1089. }\

  1090. \

  1091. static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\

  1092.     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 16);\

  1093. }\

  1094. \

  1095. static void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\

  1096. {\

  1097.     asm volatile(\

  1098.         "movq       %5,  %%mm6          \n\t"\

  1099.         "movq      (%1), %%mm0          \n\t"\

  1100.         "movq    24(%1), %%mm1          \n\t"\

  1101.         "paddw    %%mm6, %%mm0          \n\t"\

  1102.         "paddw    %%mm6, %%mm1          \n\t"\

  1103.         "psraw      $5,  %%mm0          \n\t"\

  1104.         "psraw      $5,  %%mm1          \n\t"\

  1105.         "packuswb %%mm0, %%mm0          \n\t"\

  1106.         "packuswb %%mm1, %%mm1          \n\t"\

  1107.         PAVGB"     (%0), %%mm0          \n\t"\

  1108.         PAVGB"  (%0,%3), %%mm1          \n\t"\

  1109.         OP(%%mm0, (%2),    %%mm4, d)\

  1110.         OP(%%mm1, (%2,%4), %%mm5, d)\

  1111.         "lea  (%0,%3,2), %0             \n\t"\

  1112.         "lea  (%2,%4,2), %2             \n\t"\

  1113.         "movq    48(%1), %%mm0          \n\t"\

  1114.         "movq    72(%1), %%mm1          \n\t"\

  1115.         "paddw    %%mm6, %%mm0          \n\t"\

  1116.         "paddw    %%mm6, %%mm1          \n\t"\

  1117.         "psraw      $5,  %%mm0          \n\t"\

  1118.         "psraw      $5,  %%mm1          \n\t"\

  1119.         "packuswb %%mm0, %%mm0          \n\t"\

  1120.         "packuswb %%mm1, %%mm1          \n\t"\

  1121.         PAVGB"     (%0), %%mm0          \n\t"\

  1122.         PAVGB"  (%0,%3), %%mm1          \n\t"\

  1123.         OP(%%mm0, (%2),    %%mm4, d)\

  1124.         OP(%%mm1, (%2,%4), %%mm5, d)\

  1125.         :"+a"(src8), "+c"(src16), "+d"(dst)\

  1126.         :"S"((long)src8Stride), "D"((long)dstStride), "m"(ff_pw_16)\

  1127.         :"memory");\

  1128. }\

  1129. static void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\

  1130. {\

  1131.     asm volatile(\

  1132.         "movq       %0,  %%mm6          \n\t"\

  1133.         ::"m"(ff_pw_16)\

  1134.         );\

  1135.     while(h--){\

  1136.     asm volatile(\

  1137.         "movq      (%1), %%mm0          \n\t"\

  1138.         "movq     8(%1), %%mm1          \n\t"\

  1139.         "paddw    %%mm6, %%mm0          \n\t"\

  1140.         "paddw    %%mm6, %%mm1          \n\t"\

  1141.         "psraw      $5,  %%mm0          \n\t"\

  1142.         "psraw      $5,  %%mm1          \n\t"\

  1143.         "packuswb %%mm1, %%mm0          \n\t"\

  1144.         PAVGB"     (%0), %%mm0          \n\t"\

  1145.         OP(%%mm0, (%2), %%mm5, q)\

  1146.         ::"a"(src8), "c"(src16), "d"(dst)\

  1147.         :"memory");\

  1148.         src8 += src8Stride;\

  1149.         src16 += 24;\

  1150.         dst += dstStride;\

  1151.     }\

  1152. }\

  1153. static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\

  1154. {\

  1155.     OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\

  1156.     OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\

  1157. }\

  1158. 

  1159. 

  1160. #define H264_MC(OPNAME, SIZE, MMX) \

  1161. static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  1162.     OPNAME ## pixels ## SIZE ## _mmx(dst, src, stride, SIZE);\

  1163. }\

  1164. \

  1165. static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1166.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\

  1167. }\

  1168. \

  1169. static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1170.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\

  1171. }\

  1172. \

  1173. static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1174.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\

  1175. }\

  1176. \

  1177. static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1178.     uint64_t temp[SIZE*SIZE/8];\

  1179.     uint8_t * const half= (uint8_t*)temp;\

  1180.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\

  1181.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\

  1182. }\

  1183. \

  1184. static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1185.     OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\

  1186. }\

  1187. \

  1188. static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1189.     uint64_t temp[SIZE*SIZE/8];\

  1190.     uint8_t * const half= (uint8_t*)temp;\

  1191.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\

  1192.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, half, stride, stride, SIZE);\

  1193. }\

  1194. \

  1195. static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1196.     uint64_t temp[SIZE*SIZE/8];\

  1197.     uint8_t * const halfV= (uint8_t*)temp;\

  1198.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\

  1199.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfV, stride, SIZE);\

  1200. }\

  1201. \

  1202. static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1203.     uint64_t temp[SIZE*SIZE/8];\

  1204.     uint8_t * const halfV= (uint8_t*)temp;\

  1205.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\

  1206.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfV, stride, SIZE);\

  1207. }\

  1208. \

  1209. static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1210.     uint64_t temp[SIZE*SIZE/8];\

  1211.     uint8_t * const halfV= (uint8_t*)temp;\

  1212.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\

  1213.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfV, stride, SIZE);\

  1214. }\

  1215. \

  1216. static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1217.     uint64_t temp[SIZE*SIZE/8];\

  1218.     uint8_t * const halfV= (uint8_t*)temp;\

  1219.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\

  1220.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfV, stride, SIZE);\

  1221. }\

  1222. \

  1223. static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1224.     uint64_t temp[SIZE*(SIZE<8?12:24)/4];\

  1225.     int16_t * const tmp= (int16_t*)temp;\

  1226.     OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, tmp, src, stride, SIZE, stride);\

  1227. }\

  1228. \

  1229. static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1230.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\

  1231.     uint8_t * const halfHV= (uint8_t*)temp;\

  1232.     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE/2;\

  1233.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\

  1234.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\

  1235. }\

  1236. \

  1237. static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1238.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\

  1239.     uint8_t * const halfHV= (uint8_t*)temp;\

  1240.     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE/2;\

  1241.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\

  1242.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\

  1243. }\

  1244. \

  1245. static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1246.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\

  1247.     int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\

  1248.     uint8_t * const halfHV= ((uint8_t*)temp);\

  1249.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\

  1250.     OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\

  1251. }\

  1252. \

  1253. static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1254.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\

  1255.     int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\

  1256.     uint8_t * const halfHV= ((uint8_t*)temp);\

  1257.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\

  1258.     OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\

  1259. }\

  1260. 

  1261. 

  1262. #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "    \n\t"

  1263. #define AVG_3DNOW_OP(a,b,temp, size) \

  1264. "mov" #size " " #b ", " #temp "   \n\t"\

  1265. "pavgusb " #temp ", " #a "        \n\t"\

  1266. "mov" #size " " #a ", " #b "      \n\t"

  1267. #define AVG_MMX2_OP(a,b,temp, size) \

  1268. "mov" #size " " #b ", " #temp "   \n\t"\

  1269. "pavgb " #temp ", " #a "          \n\t"\

  1270. "mov" #size " " #a ", " #b "      \n\t"

  1271. 

  1272. #define PAVGB "pavgusb"

  1273. QPEL_H264(put_,       PUT_OP, 3dnow)

  1274. QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)

  1275. #undef PAVGB

  1276. #define PAVGB "pavgb"

  1277. QPEL_H264(put_,       PUT_OP, mmx2)

  1278. QPEL_H264(avg_,  AVG_MMX2_OP, mmx2)

  1279. #undef PAVGB

  1280. 

  1281. H264_MC(put_, 4, 3dnow)

  1282. H264_MC(put_, 8, 3dnow)

  1283. H264_MC(put_, 16,3dnow)

  1284. H264_MC(avg_, 4, 3dnow)

  1285. H264_MC(avg_, 8, 3dnow)

  1286. H264_MC(avg_, 16,3dnow)

  1287. H264_MC(put_, 4, mmx2)

  1288. H264_MC(put_, 8, mmx2)

  1289. H264_MC(put_, 16,mmx2)

  1290. H264_MC(avg_, 4, mmx2)

  1291. H264_MC(avg_, 8, mmx2)

  1292. H264_MC(avg_, 16,mmx2)

  1293. 

  1294. 

  1295. #define H264_CHROMA_OP(S,D)

  1296. #define H264_CHROMA_OP4(S,D,T)

  1297. #define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx

  1298. #define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_mmx

  1299. #define H264_CHROMA_MC2_TMPL put_h264_chroma_mc2_mmx2

  1300. #define H264_CHROMA_MC8_MV0 put_pixels8_mmx

  1301. #include "dsputil_h264_template_mmx.c"

  1302. #undef H264_CHROMA_OP

  1303. #undef H264_CHROMA_OP4

  1304. #undef H264_CHROMA_MC8_TMPL

  1305. #undef H264_CHROMA_MC4_TMPL

  1306. #undef H264_CHROMA_MC2_TMPL

  1307. #undef H264_CHROMA_MC8_MV0

  1308. 

  1309. #define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"

  1310. #define H264_CHROMA_OP4(S,D,T) "movd  " #S ", " #T " \n\t"\

  1311.                                "pavgb " #T ", " #D " \n\t"

  1312. #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2

  1313. #define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_mmx2

  1314. #define H264_CHROMA_MC2_TMPL avg_h264_chroma_mc2_mmx2

  1315. #define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2

  1316. #include "dsputil_h264_template_mmx.c"

  1317. #undef H264_CHROMA_OP

  1318. #undef H264_CHROMA_OP4

  1319. #undef H264_CHROMA_MC8_TMPL

  1320. #undef H264_CHROMA_MC4_TMPL

  1321. #undef H264_CHROMA_MC2_TMPL

  1322. #undef H264_CHROMA_MC8_MV0

  1323. 

  1324. #define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"

  1325. #define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\

  1326.                                "pavgusb " #T ", " #D " \n\t"

  1327. #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow

  1328. #define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_3dnow

  1329. #define H264_CHROMA_MC8_MV0 avg_pixels8_3dnow

  1330. #include "dsputil_h264_template_mmx.c"

  1331. #undef H264_CHROMA_OP

  1332. #undef H264_CHROMA_OP4

  1333. #undef H264_CHROMA_MC8_TMPL

  1334. #undef H264_CHROMA_MC4_TMPL

  1335. #undef H264_CHROMA_MC8_MV0

  1336. 

  1337. /***********************************/

  1338. /* weighted prediction */

  1339. 

  1340. static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)

  1341. {

  1342.     int x, y;

  1343.     offset <<= log2_denom;

  1344.     offset += (1 << log2_denom) >> 1;

  1345.     asm volatile(

  1346.         "movd    %0, %%mm4        \n\t"

  1347.         "movd    %1, %%mm5        \n\t"

  1348.         "movd    %2, %%mm6        \n\t"

  1349.         "pshufw  $0, %%mm4, %%mm4 \n\t"

  1350.         "pshufw  $0, %%mm5, %%mm5 \n\t"

  1351.         "pxor    %%mm7, %%mm7     \n\t"

  1352.         :: "g"(weight), "g"(offset), "g"(log2_denom)

  1353.     );

  1354.     for(y=0; y<h; y+=2){

  1355.         for(x=0; x<w; x+=4){

  1356.             asm volatile(

  1357.                 "movd      %0,    %%mm0 \n\t"

  1358.                 "movd      %1,    %%mm1 \n\t"

  1359.                 "punpcklbw %%mm7, %%mm0 \n\t"

  1360.                 "punpcklbw %%mm7, %%mm1 \n\t"

  1361.                 "pmullw    %%mm4, %%mm0 \n\t"

  1362.                 "pmullw    %%mm4, %%mm1 \n\t"

  1363.                 "paddsw    %%mm5, %%mm0 \n\t"

  1364.                 "paddsw    %%mm5, %%mm1 \n\t"

  1365.                 "psraw     %%mm6, %%mm0 \n\t"

  1366.                 "psraw     %%mm6, %%mm1 \n\t"

  1367.                 "packuswb  %%mm7, %%mm0 \n\t"

  1368.                 "packuswb  %%mm7, %%mm1 \n\t"

  1369.                 "movd      %%mm0, %0    \n\t"

  1370.                 "movd      %%mm1, %1    \n\t"

  1371.                 : "+m"(*(uint32_t*)(dst+x)),

  1372.                   "+m"(*(uint32_t*)(dst+x+stride))

  1373.             );

  1374.         }

  1375.         dst += 2*stride;

  1376.     }

  1377. }

  1378. 

  1379. 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)

  1380. {

  1381.     int x, y;

  1382.     offset = ((offset + 1) | 1) << log2_denom;

  1383.     asm volatile(

  1384.         "movd    %0, %%mm3        \n\t"

  1385.         "movd    %1, %%mm4        \n\t"

  1386.         "movd    %2, %%mm5        \n\t"

  1387.         "movd    %3, %%mm6        \n\t"

  1388.         "pshufw  $0, %%mm3, %%mm3 \n\t"

  1389.         "pshufw  $0, %%mm4, %%mm4 \n\t"

  1390.         "pshufw  $0, %%mm5, %%mm5 \n\t"

  1391.         "pxor    %%mm7, %%mm7     \n\t"

  1392.         :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)

  1393.     );

  1394.     for(y=0; y<h; y++){

  1395.         for(x=0; x<w; x+=4){

  1396.             asm volatile(

  1397.                 "movd      %0,    %%mm0 \n\t"

  1398.                 "movd      %1,    %%mm1 \n\t"

  1399.                 "punpcklbw %%mm7, %%mm0 \n\t"

  1400.                 "punpcklbw %%mm7, %%mm1 \n\t"

  1401.                 "pmullw    %%mm3, %%mm0 \n\t"

  1402.                 "pmullw    %%mm4, %%mm1 \n\t"

  1403.                 "paddsw    %%mm1, %%mm0 \n\t"

  1404.                 "paddsw    %%mm5, %%mm0 \n\t"

  1405.                 "psraw     %%mm6, %%mm0 \n\t"

  1406.                 "packuswb  %%mm0, %%mm0 \n\t"

  1407.                 "movd      %%mm0, %0    \n\t"

  1408.                 : "+m"(*(uint32_t*)(dst+x))

  1409.                 :  "m"(*(uint32_t*)(src+x))

  1410.             );

  1411.         }

  1412.         src += stride;

  1413.         dst += stride;

  1414.     }

  1415. }

  1416. 

  1417. #define H264_WEIGHT(W,H) \

  1418. 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){ \

  1419.     ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \

  1420. } \

  1421. static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \

  1422.     ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \

  1423. }

  1424. 

  1425. H264_WEIGHT(16,16)

  1426. H264_WEIGHT(16, 8)

  1427. H264_WEIGHT( 8,16)

  1428. H264_WEIGHT( 8, 8)

  1429. H264_WEIGHT( 8, 4)

  1430. H264_WEIGHT( 4, 8)

  1431. H264_WEIGHT( 4, 4)

  1432. H264_WEIGHT( 4, 2)