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.
9447 Commits
32 Branches
582MB
Tree: 4440b1188e
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '4440b1188e'
${ noResults }
FFmpeg/libavcodec/i386/h264dsp_mmx.c

1512 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 STORE_DIFF_4P( p, t, z ) \

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

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

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

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

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

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

  58. 

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

  60. {

  61.     /* Load dct coeffs */

  62.     asm volatile(

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

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

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

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

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

  68. 

  69.     asm volatile(

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

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

  72. 

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

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

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

  76. 

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

  78. 

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

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

  81. 

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

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

  84. 

  85.     asm volatile(

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

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

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

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

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

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

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

  93.         : "+r"(dst)

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

  95.     );

  96. }

  97. 

  98. static inline void h264_idct8_1d(int16_t *block)

  99. {

  100.     asm volatile(

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

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

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

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

  105. 

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

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

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

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

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

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

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

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

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

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

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

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

  118. 

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

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

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

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

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

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

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

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

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

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

  129. 

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

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

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

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

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

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

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

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

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

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

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

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

  142. 

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

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

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

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

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

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

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

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

  151. 

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

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

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

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

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

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

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

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

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

  161.         :: "r"(block)

  162.     );

  163. }

  164. 

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

  166. {

  167.     int i;

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

  169. 

  170.     block[0] += 32;

  171. 

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

  173.         DECLARE_ALIGNED_8(uint64_t, tmp);

  174. 

  175.         h264_idct8_1d(block+4*i);

  176. 

  177.         asm volatile(

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

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

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

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

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

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

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

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

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

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

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

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

  190.             : "=m"(tmp)

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

  192.             : "memory"

  193.         );

  194.     }

  195. 

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

  197.         h264_idct8_1d(b2+4*i);

  198. 

  199.         asm volatile(

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

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

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

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

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

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

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

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

  208. 

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

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

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

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

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

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

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

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

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

  218.             : "memory"

  219.         );

  220.     }

  221. 

  222.     add_pixels_clamped_mmx(b2, dst, stride);

  223. }

  224. 

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

  226. {

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

  228.     asm volatile(

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

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

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

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

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

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

  235.         ::"r"(dc)

  236.     );

  237.     asm volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  258.     );

  259. }

  260. 

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

  262. {

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

  264.     int y;

  265.     asm volatile(

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

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

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

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

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

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

  272.         ::"r"(dc)

  273.     );

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

  275.     asm volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  296.     );

  297.     }

  298. }

  299. 

  300. 

  301. /***********************************/

  302. /* deblocking */

  303. 

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

  305. // clobbers: t

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

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

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

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

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

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

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

  313. 

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

  315. // clobbers: t

  316. #define DIFF_GT2_MMX(x,y,a,o,t)\

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

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

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

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

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

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

  323.     "pcmpeqb  "#t", "#o"  \n\t"\

  324. 

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

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

  327. // clobbers: mm4,mm6

  328. #define H264_DEBLOCK_MASK(alpha1, beta1) \

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

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

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

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

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

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

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

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

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

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

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

  340. 

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

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

  343. // clobbers: mm0,3-6

  344. #define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\

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

  346.         "pxor    %%mm2              , %%mm5 \n\t" /* p0^q0*/\

  347.         "pand    "#pb_01"           , %%mm5 \n\t" /* (p0^q0)&1*/\

  348.         "pcmpeqb %%mm4              , %%mm4 \n\t"\

  349.         "pxor    %%mm4              , %%mm3 \n\t"\

  350.         "pavgb   %%mm0              , %%mm3 \n\t" /* (p1 - q1 + 256)>>1*/\

  351.         "pavgb   "MANGLE(ff_pb_3)"  , %%mm3 \n\t" /*(((p1 - q1 + 256)>>1)+4)>>1 = 64+2+(p1-q1)>>2*/\

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

  353.         "pavgb   %%mm2              , %%mm4 \n\t" /* (q0 - p0 + 256)>>1*/\

  354.         "pavgb   %%mm5              , %%mm3 \n\t"\

  355.         "paddusb %%mm4              , %%mm3 \n\t" /* d+128+33*/\

  356.         "movq    "MANGLE(ff_pb_A1)" , %%mm6 \n\t"\

  357.         "psubusb %%mm3              , %%mm6 \n\t"\

  358.         "psubusb "MANGLE(ff_pb_A1)" , %%mm3 \n\t"\

  359.         "pminub  %%mm7              , %%mm6 \n\t"\

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

  361.         "psubusb %%mm6              , %%mm1 \n\t"\

  362.         "psubusb %%mm3              , %%mm2 \n\t"\

  363.         "paddusb %%mm3              , %%mm1 \n\t"\

  364.         "paddusb %%mm6              , %%mm2 \n\t"

  365. 

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

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

  368. // clobbers: q2, tmp, tc0

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

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

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

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

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

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

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

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

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

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

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

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

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

  382. 

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

  384. {

  385.     DECLARE_ALIGNED_8(uint64_t, tmp0[2]);

  386. 

  387.     asm volatile(

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

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

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

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

  392.         H264_DEBLOCK_MASK(%6, %7)

  393. 

  394.         "movd      %5,    %%mm4    \n\t"

  395.         "punpcklbw %%mm4, %%mm4    \n\t"

  396.         "punpcklwd %%mm4, %%mm4    \n\t"

  397.         "pcmpeqb   %%mm3, %%mm3    \n\t"

  398.         "movq      %%mm4, %%mm6    \n\t"

  399.         "pcmpgtb   %%mm3, %%mm4    \n\t"

  400.         "movq      %%mm6, 8+%0     \n\t"

  401.         "pand      %%mm4, %%mm7    \n\t"

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

  403. 

  404.         /* filter p1 */

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

  406.         DIFF_GT2_MMX(%%mm1, %%mm3, %%mm5, %%mm6, %%mm4) // |p2-p0|>beta-1

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

  408.         "pand     8+%0,   %%mm7    \n\t" // mask & tc0

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

  410.         "psubb    %%mm6,  %%mm7    \n\t"

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

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

  413. 

  414.         /* filter q1 */

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

  416.         DIFF_GT2_MMX(%%mm2, %%mm4, %%mm5, %%mm6, %%mm3) // |q2-q0|>beta-1

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

  418.         "movq     8+%0,   %%mm5    \n\t" // can be merged with the and below but is slower then

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

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

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

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

  423. 

  424.         /* filter p0, q0 */

  425.         H264_DEBLOCK_P0_Q0(%8, unused)

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

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

  428. 

  429.         : "=m"(*tmp0)

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

  431.           "m"(*tmp0/*unused*/), "m"(*(uint32_t*)tc0), "m"(alpha1), "m"(beta1),

  432.           "m"(mm_bone)

  433.     );

  434. }

  435. 

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

  437. {

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

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

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

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

  442. }

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

  444. {

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

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

  447.     DECLARE_ALIGNED_8(uint8_t, trans[8*8]);

  448.     int i;

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

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

  451.             continue;

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

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

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

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

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

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

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

  459.     }

  460. }

  461. 

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

  463. {

  464.     asm volatile(

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

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

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

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

  469.         H264_DEBLOCK_MASK(%4, %5)

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

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

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

  473.         H264_DEBLOCK_P0_Q0(%6, %7)

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

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

  476. 

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

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

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

  480.     );

  481. }

  482. 

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

  484. {

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

  486. }

  487. 

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

  489. {

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

  491.     DECLARE_ALIGNED_8(uint8_t, trans[8*4]);

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

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

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

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

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

  497. }

  498. 

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

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

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

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

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

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

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

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

  507. 

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

  509. {

  510.     asm volatile(

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

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

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

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

  515.         H264_DEBLOCK_MASK(%3, %4)

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  530.     );

  531. }

  532. 

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

  534. {

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

  536. }

  537. 

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

  539. {

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

  541.     DECLARE_ALIGNED_8(uint8_t, trans[8*4]);

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

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

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

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

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

  547. }

  548. 

  549. 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],

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

  551.     int dir;

  552.     asm volatile(

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

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

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

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

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

  558.     );

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

  560.     // average filter time by 1.2%

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

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

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

  564.         DECLARE_ALIGNED_8(const uint64_t, mask_dir) = dir ? 0 : 0xffffffffffffffffULL;

  565.         int b_idx, edge, l;

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

  567.             asm volatile(

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

  569.                 ::"m"(mask_dir)

  570.             );

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

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

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

  574.                     asm volatile(

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

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

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

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

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

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

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

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

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

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

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

  586. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  602.                     );

  603.                 }

  604.             }

  605.             asm volatile(

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

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

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

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

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

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

  612.             );

  613.             asm volatile(

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

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

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

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

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

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

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

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

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

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

  624.                 ::"memory"

  625.             );

  626.         }

  627.         edges = 4;

  628.         step = 1;

  629.     }

  630.     asm volatile(

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

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

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

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

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

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

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

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

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

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

  641.         :"memory"

  642.     );

  643. }

  644. 

  645. /***********************************/

  646. /* motion compensation */

  647. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  665. 

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

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

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

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

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

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

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

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

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

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

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

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

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

  679. 

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

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

  682.     int h=4;\

  683. \

  684.     asm volatile(\

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

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

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

  688.         "1:                         \n\t"\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  718.         : "memory"\

  719.     );\

  720. }\

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

  722.     int h=4;\

  723.     asm volatile(\

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

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

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

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

  728.     );\

  729.     do{\

  730.     asm volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  761.         : "memory"\

  762.     );\

  763.     }while(--h);\

  764. }\

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

  766.     src -= 2*srcStride;\

  767.     asm volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  788.          \

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

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

  791.         : "memory"\

  792.     );\

  793. }\

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

  795.     int h=4;\

  796.     int w=3;\

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

  798.     while(w--){\

  799.         asm volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  820.              \

  821.             : "+a"(src)\

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

  823.             : "memory"\

  824.         );\

  825.         tmp += 4;\

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

  827.     }\

  828.     tmp -= 3*4;\

  829.     asm volatile(\

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

  831.         "1:                         \n\t"\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  854.         : "memory"\

  855.     );\

  856. }\

  857. \

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

  859.     int h=8;\

  860.     asm volatile(\

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

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

  863.         "1:                         \n\t"\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  911.         : "memory"\

  912.     );\

  913. }\

  914. \

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

  916.     int h=8;\

  917.     asm volatile(\

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

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

  920.         :: "m"(ff_pw_5)\

  921.     );\

  922.     do{\

  923.     asm volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  972.           "m"(ff_pw_16)\

  973.         : "memory"\

  974.     );\

  975.     }while(--h);\

  976. }\

  977. \

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

  979.     int w= 2;\

  980.     src -= 2*srcStride;\

  981.     \

  982.     while(w--){\

  983.       asm volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  1000.         QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\

  1001.         QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\

  1002.         QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\

  1003.         QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\

  1004.         QPEL_H264V(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP)\

  1005.         QPEL_H264V(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP)\

  1006.         QPEL_H264V(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP)\

  1007.         QPEL_H264V(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP)\

  1008.          \

  1009.         : "+a"(src), "+c"(dst)\

  1010.         : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\

  1011.         : "memory"\

  1012.      );\

  1013.      if(h==16){\

  1014.         asm volatile(\

  1015.             QPEL_H264V(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP)\

  1016.             QPEL_H264V(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP)\

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

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

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

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

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

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

  1023.             \

  1024.            : "+a"(src), "+c"(dst)\

  1025.            : "S"((long)srcStride), "D"((long)dstStride), "m"(ff_pw_5), "m"(ff_pw_16)\

  1026.            : "memory"\

  1027.         );\

  1028.      }\

  1029.      src += 4-(h+5)*srcStride;\

  1030.      dst += 4-h*dstStride;\

  1031.    }\

  1032. }\

  1033. static av_noinline void OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride, int size){\

  1034.     int h = size;\

  1035.     int w = (size+8)>>2;\

  1036.     src -= 2*srcStride+2;\

  1037.     while(w--){\

  1038.         asm volatile(\

  1039.             "pxor %%mm7, %%mm7      \n\t"\

  1040.             "movd (%0), %%mm0       \n\t"\

  1041.             "add %2, %0             \n\t"\

  1042.             "movd (%0), %%mm1       \n\t"\

  1043.             "add %2, %0             \n\t"\

  1044.             "movd (%0), %%mm2       \n\t"\

  1045.             "add %2, %0             \n\t"\

  1046.             "movd (%0), %%mm3       \n\t"\

  1047.             "add %2, %0             \n\t"\

  1048.             "movd (%0), %%mm4       \n\t"\

  1049.             "add %2, %0             \n\t"\

  1050.             "punpcklbw %%mm7, %%mm0 \n\t"\

  1051.             "punpcklbw %%mm7, %%mm1 \n\t"\

  1052.             "punpcklbw %%mm7, %%mm2 \n\t"\

  1053.             "punpcklbw %%mm7, %%mm3 \n\t"\

  1054.             "punpcklbw %%mm7, %%mm4 \n\t"\

  1055.             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 0*48)\

  1056.             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 1*48)\

  1057.             QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 2*48)\

  1058.             QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 3*48)\

  1059.             QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 4*48)\

  1060.             QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 5*48)\

  1061.             QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 6*48)\

  1062.             QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 7*48)\

  1063.             : "+a"(src)\

  1064.             : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\

  1065.             : "memory"\

  1066.         );\

  1067.         if(size==16){\

  1068.             asm volatile(\

  1069.                 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1,  8*48)\

  1070.                 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2,  9*48)\

  1071.                 QPEL_H264HV(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, 10*48)\

  1072.                 QPEL_H264HV(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, 11*48)\

  1073.                 QPEL_H264HV(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, 12*48)\

  1074.                 QPEL_H264HV(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, 13*48)\

  1075.                 QPEL_H264HV(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, 14*48)\

  1076.                 QPEL_H264HV(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, 15*48)\

  1077.                 : "+a"(src)\

  1078.                 : "c"(tmp), "S"((long)srcStride), "m"(ff_pw_5)\

  1079.                 : "memory"\

  1080.             );\

  1081.         }\

  1082.         tmp += 4;\

  1083.         src += 4 - (size+5)*srcStride;\

  1084.     }\

  1085.     tmp -= size+8;\

  1086.     w = size>>4;\

  1087.     do{\

  1088.     h = size;\

  1089.     asm volatile(\

  1090.         "movq %4, %%mm6             \n\t"\

  1091.         "1:                         \n\t"\

  1092.         "movq     (%0), %%mm0       \n\t"\

  1093.         "movq    8(%0), %%mm3       \n\t"\

  1094.         "movq    2(%0), %%mm1       \n\t"\

  1095.         "movq   10(%0), %%mm4       \n\t"\

  1096.         "paddw   %%mm4, %%mm0       \n\t"\

  1097.         "paddw   %%mm3, %%mm1       \n\t"\

  1098.         "paddw  18(%0), %%mm3       \n\t"\

  1099.         "paddw  16(%0), %%mm4       \n\t"\

  1100.         "movq    4(%0), %%mm2       \n\t"\

  1101.         "movq   12(%0), %%mm5       \n\t"\

  1102.         "paddw   6(%0), %%mm2       \n\t"\

  1103.         "paddw  14(%0), %%mm5       \n\t"\

  1104.         "psubw %%mm1, %%mm0         \n\t"\

  1105.         "psubw %%mm4, %%mm3         \n\t"\

  1106.         "psraw $2, %%mm0            \n\t"\

  1107.         "psraw $2, %%mm3            \n\t"\

  1108.         "psubw %%mm1, %%mm0         \n\t"\

  1109.         "psubw %%mm4, %%mm3         \n\t"\

  1110.         "paddsw %%mm2, %%mm0        \n\t"\

  1111.         "paddsw %%mm5, %%mm3        \n\t"\

  1112.         "psraw $2, %%mm0            \n\t"\

  1113.         "psraw $2, %%mm3            \n\t"\

  1114.         "paddw %%mm6, %%mm2         \n\t"\

  1115.         "paddw %%mm6, %%mm5         \n\t"\

  1116.         "paddw %%mm2, %%mm0         \n\t"\

  1117.         "paddw %%mm5, %%mm3         \n\t"\

  1118.         "psraw $6, %%mm0            \n\t"\

  1119.         "psraw $6, %%mm3            \n\t"\

  1120.         "packuswb %%mm3, %%mm0      \n\t"\

  1121.         OP(%%mm0, (%1),%%mm7, q)\

  1122.         "add $48, %0                \n\t"\

  1123.         "add %3, %1                 \n\t"\

  1124.         "decl %2                    \n\t"\

  1125.         " jnz 1b                    \n\t"\

  1126.         : "+a"(tmp), "+c"(dst), "+m"(h)\

  1127.         : "S"((long)dstStride), "m"(ff_pw_32)\

  1128.         : "memory"\

  1129.     );\

  1130.     tmp += 8 - size*24;\

  1131.     dst += 8 - size*dstStride;\

  1132.     }while(w--);\

  1133. }\

  1134. \

  1135. static void OPNAME ## h264_qpel8_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1136.     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 8);\

  1137. }\

  1138. static av_noinline void OPNAME ## h264_qpel16_v_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1139.     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst  , src  , dstStride, srcStride, 16);\

  1140.     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\

  1141. }\

  1142. \

  1143. static av_noinline void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  1144.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\

  1145.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  1146.     src += 8*srcStride;\

  1147.     dst += 8*dstStride;\

  1148.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\

  1149.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  1150. }\

  1151. \

  1152. static av_noinline void OPNAME ## h264_qpel16_h_lowpass_l2_ ## MMX(uint8_t *dst, uint8_t *src, uint8_t *src2, int dstStride, int src2Stride){\

  1153.     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\

  1154.     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\

  1155.     src += 8*dstStride;\

  1156.     dst += 8*dstStride;\

  1157.     src2 += 8*src2Stride;\

  1158.     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst  , src  , src2  , dstStride, src2Stride);\

  1159.     OPNAME ## h264_qpel8_h_lowpass_l2_ ## MMX(dst+8, src+8, src2+8, dstStride, src2Stride);\

  1160. }\

  1161. \

  1162. static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\

  1163.     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 8);\

  1164. }\

  1165. \

  1166. static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\

  1167.     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 16);\

  1168. }\

  1169. \

  1170. static av_noinline void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\

  1171. {\

  1172.     asm volatile(\

  1173.         "movq       %5,  %%mm6          \n\t"\

  1174.         "movq      (%1), %%mm0          \n\t"\

  1175.         "movq    24(%1), %%mm1          \n\t"\

  1176.         "paddw    %%mm6, %%mm0          \n\t"\

  1177.         "paddw    %%mm6, %%mm1          \n\t"\

  1178.         "psraw      $5,  %%mm0          \n\t"\

  1179.         "psraw      $5,  %%mm1          \n\t"\

  1180.         "packuswb %%mm0, %%mm0          \n\t"\

  1181.         "packuswb %%mm1, %%mm1          \n\t"\

  1182.         PAVGB"     (%0), %%mm0          \n\t"\

  1183.         PAVGB"  (%0,%3), %%mm1          \n\t"\

  1184.         OP(%%mm0, (%2),    %%mm4, d)\

  1185.         OP(%%mm1, (%2,%4), %%mm5, d)\

  1186.         "lea  (%0,%3,2), %0             \n\t"\

  1187.         "lea  (%2,%4,2), %2             \n\t"\

  1188.         "movq    48(%1), %%mm0          \n\t"\

  1189.         "movq    72(%1), %%mm1          \n\t"\

  1190.         "paddw    %%mm6, %%mm0          \n\t"\

  1191.         "paddw    %%mm6, %%mm1          \n\t"\

  1192.         "psraw      $5,  %%mm0          \n\t"\

  1193.         "psraw      $5,  %%mm1          \n\t"\

  1194.         "packuswb %%mm0, %%mm0          \n\t"\

  1195.         "packuswb %%mm1, %%mm1          \n\t"\

  1196.         PAVGB"     (%0), %%mm0          \n\t"\

  1197.         PAVGB"  (%0,%3), %%mm1          \n\t"\

  1198.         OP(%%mm0, (%2),    %%mm4, d)\

  1199.         OP(%%mm1, (%2,%4), %%mm5, d)\

  1200.         :"+a"(src8), "+c"(src16), "+d"(dst)\

  1201.         :"S"((long)src8Stride), "D"((long)dstStride), "m"(ff_pw_16)\

  1202.         :"memory");\

  1203. }\

  1204. static av_noinline void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\

  1205. {\

  1206.     asm volatile(\

  1207.         "movq       %0,  %%mm6          \n\t"\

  1208.         ::"m"(ff_pw_16)\

  1209.         );\

  1210.     while(h--){\

  1211.     asm volatile(\

  1212.         "movq      (%1), %%mm0          \n\t"\

  1213.         "movq     8(%1), %%mm1          \n\t"\

  1214.         "paddw    %%mm6, %%mm0          \n\t"\

  1215.         "paddw    %%mm6, %%mm1          \n\t"\

  1216.         "psraw      $5,  %%mm0          \n\t"\

  1217.         "psraw      $5,  %%mm1          \n\t"\

  1218.         "packuswb %%mm1, %%mm0          \n\t"\

  1219.         PAVGB"     (%0), %%mm0          \n\t"\

  1220.         OP(%%mm0, (%2), %%mm5, q)\

  1221.         ::"a"(src8), "c"(src16), "d"(dst)\

  1222.         :"memory");\

  1223.         src8 += src8Stride;\

  1224.         src16 += 24;\

  1225.         dst += dstStride;\

  1226.     }\

  1227. }\

  1228. static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\

  1229. {\

  1230.     OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\

  1231.     OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\

  1232. }\

  1233. 

  1234. 

  1235. #define H264_MC(OPNAME, SIZE, MMX) \

  1236. static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  1237.     OPNAME ## pixels ## SIZE ## _mmx(dst, src, stride, SIZE);\

  1238. }\

  1239. \

  1240. static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1241.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src, stride, stride);\

  1242. }\

  1243. \

  1244. static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1245.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\

  1246. }\

  1247. \

  1248. static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1249.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, src+1, stride, stride);\

  1250. }\

  1251. \

  1252. static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1253.     uint64_t temp[SIZE*SIZE/8];\

  1254.     uint8_t * const half= (uint8_t*)temp;\

  1255.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\

  1256.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\

  1257. }\

  1258. \

  1259. static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1260.     OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\

  1261. }\

  1262. \

  1263. static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1264.     uint64_t temp[SIZE*SIZE/8];\

  1265.     uint8_t * const half= (uint8_t*)temp;\

  1266.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\

  1267.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, half, stride, stride, SIZE);\

  1268. }\

  1269. \

  1270. static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1271.     uint64_t temp[SIZE*SIZE/8];\

  1272.     uint8_t * const halfV= (uint8_t*)temp;\

  1273.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\

  1274.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfV, stride, SIZE);\

  1275. }\

  1276. \

  1277. static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1278.     uint64_t temp[SIZE*SIZE/8];\

  1279.     uint8_t * const halfV= (uint8_t*)temp;\

  1280.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\

  1281.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfV, stride, SIZE);\

  1282. }\

  1283. \

  1284. static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1285.     uint64_t temp[SIZE*SIZE/8];\

  1286.     uint8_t * const halfV= (uint8_t*)temp;\

  1287.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\

  1288.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfV, stride, SIZE);\

  1289. }\

  1290. \

  1291. static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1292.     uint64_t temp[SIZE*SIZE/8];\

  1293.     uint8_t * const halfV= (uint8_t*)temp;\

  1294.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\

  1295.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfV, stride, SIZE);\

  1296. }\

  1297. \

  1298. static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1299.     uint64_t temp[SIZE*(SIZE<8?12:24)/4];\

  1300.     int16_t * const tmp= (int16_t*)temp;\

  1301.     OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, tmp, src, stride, SIZE, stride);\

  1302. }\

  1303. \

  1304. static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1305.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\

  1306.     uint8_t * const halfHV= (uint8_t*)temp;\

  1307.     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE/2;\

  1308.     assert((int)temp & 7 == 0);\

  1309.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\

  1310.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src, halfHV, stride, SIZE);\

  1311. }\

  1312. \

  1313. static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1314.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\

  1315.     uint8_t * const halfHV= (uint8_t*)temp;\

  1316.     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE/2;\

  1317.     assert((int)temp & 7 == 0);\

  1318.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\

  1319.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_l2_ ## MMX(dst, src+stride, halfHV, stride, SIZE);\

  1320. }\

  1321. \

  1322. static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1323.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\

  1324.     int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\

  1325.     uint8_t * const halfHV= ((uint8_t*)temp);\

  1326.     assert((int)temp & 7 == 0);\

  1327.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\

  1328.     OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\

  1329. }\

  1330. \

  1331. static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1332.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\

  1333.     int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\

  1334.     uint8_t * const halfHV= ((uint8_t*)temp);\

  1335.     assert((int)temp & 7 == 0);\

  1336.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\

  1337.     OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\

  1338. }\

  1339. 

  1340. 

  1341. #define AVG_3DNOW_OP(a,b,temp, size) \

  1342. "mov" #size " " #b ", " #temp "   \n\t"\

  1343. "pavgusb " #temp ", " #a "        \n\t"\

  1344. "mov" #size " " #a ", " #b "      \n\t"

  1345. #define AVG_MMX2_OP(a,b,temp, size) \

  1346. "mov" #size " " #b ", " #temp "   \n\t"\

  1347. "pavgb " #temp ", " #a "          \n\t"\

  1348. "mov" #size " " #a ", " #b "      \n\t"

  1349. 

  1350. #define PAVGB "pavgusb"

  1351. QPEL_H264(put_,       PUT_OP, 3dnow)

  1352. QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)

  1353. #undef PAVGB

  1354. #define PAVGB "pavgb"

  1355. QPEL_H264(put_,       PUT_OP, mmx2)

  1356. QPEL_H264(avg_,  AVG_MMX2_OP, mmx2)

  1357. #undef PAVGB

  1358. 

  1359. H264_MC(put_, 4, 3dnow)

  1360. H264_MC(put_, 8, 3dnow)

  1361. H264_MC(put_, 16,3dnow)

  1362. H264_MC(avg_, 4, 3dnow)

  1363. H264_MC(avg_, 8, 3dnow)

  1364. H264_MC(avg_, 16,3dnow)

  1365. H264_MC(put_, 4, mmx2)

  1366. H264_MC(put_, 8, mmx2)

  1367. H264_MC(put_, 16,mmx2)

  1368. H264_MC(avg_, 4, mmx2)

  1369. H264_MC(avg_, 8, mmx2)

  1370. H264_MC(avg_, 16,mmx2)

  1371. 

  1372. 

  1373. #define H264_CHROMA_OP(S,D)

  1374. #define H264_CHROMA_OP4(S,D,T)

  1375. #define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx

  1376. #define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_mmx

  1377. #define H264_CHROMA_MC2_TMPL put_h264_chroma_mc2_mmx2

  1378. #define H264_CHROMA_MC8_MV0 put_pixels8_mmx

  1379. #include "dsputil_h264_template_mmx.c"

  1380. #undef H264_CHROMA_OP

  1381. #undef H264_CHROMA_OP4

  1382. #undef H264_CHROMA_MC8_TMPL

  1383. #undef H264_CHROMA_MC4_TMPL

  1384. #undef H264_CHROMA_MC2_TMPL

  1385. #undef H264_CHROMA_MC8_MV0

  1386. 

  1387. #define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"

  1388. #define H264_CHROMA_OP4(S,D,T) "movd  " #S ", " #T " \n\t"\

  1389.                                "pavgb " #T ", " #D " \n\t"

  1390. #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2

  1391. #define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_mmx2

  1392. #define H264_CHROMA_MC2_TMPL avg_h264_chroma_mc2_mmx2

  1393. #define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2

  1394. #include "dsputil_h264_template_mmx.c"

  1395. #undef H264_CHROMA_OP

  1396. #undef H264_CHROMA_OP4

  1397. #undef H264_CHROMA_MC8_TMPL

  1398. #undef H264_CHROMA_MC4_TMPL

  1399. #undef H264_CHROMA_MC2_TMPL

  1400. #undef H264_CHROMA_MC8_MV0

  1401. 

  1402. #define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"

  1403. #define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\

  1404.                                "pavgusb " #T ", " #D " \n\t"

  1405. #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow

  1406. #define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_3dnow

  1407. #define H264_CHROMA_MC8_MV0 avg_pixels8_3dnow

  1408. #include "dsputil_h264_template_mmx.c"

  1409. #undef H264_CHROMA_OP

  1410. #undef H264_CHROMA_OP4

  1411. #undef H264_CHROMA_MC8_TMPL

  1412. #undef H264_CHROMA_MC4_TMPL

  1413. #undef H264_CHROMA_MC8_MV0

  1414. 

  1415. /***********************************/

  1416. /* weighted prediction */

  1417. 

  1418. static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)

  1419. {

  1420.     int x, y;

  1421.     offset <<= log2_denom;

  1422.     offset += (1 << log2_denom) >> 1;

  1423.     asm volatile(

  1424.         "movd    %0, %%mm4        \n\t"

  1425.         "movd    %1, %%mm5        \n\t"

  1426.         "movd    %2, %%mm6        \n\t"

  1427.         "pshufw  $0, %%mm4, %%mm4 \n\t"

  1428.         "pshufw  $0, %%mm5, %%mm5 \n\t"

  1429.         "pxor    %%mm7, %%mm7     \n\t"

  1430.         :: "g"(weight), "g"(offset), "g"(log2_denom)

  1431.     );

  1432.     for(y=0; y<h; y+=2){

  1433.         for(x=0; x<w; x+=4){

  1434.             asm volatile(

  1435.                 "movd      %0,    %%mm0 \n\t"

  1436.                 "movd      %1,    %%mm1 \n\t"

  1437.                 "punpcklbw %%mm7, %%mm0 \n\t"

  1438.                 "punpcklbw %%mm7, %%mm1 \n\t"

  1439.                 "pmullw    %%mm4, %%mm0 \n\t"

  1440.                 "pmullw    %%mm4, %%mm1 \n\t"

  1441.                 "paddsw    %%mm5, %%mm0 \n\t"

  1442.                 "paddsw    %%mm5, %%mm1 \n\t"

  1443.                 "psraw     %%mm6, %%mm0 \n\t"

  1444.                 "psraw     %%mm6, %%mm1 \n\t"

  1445.                 "packuswb  %%mm7, %%mm0 \n\t"

  1446.                 "packuswb  %%mm7, %%mm1 \n\t"

  1447.                 "movd      %%mm0, %0    \n\t"

  1448.                 "movd      %%mm1, %1    \n\t"

  1449.                 : "+m"(*(uint32_t*)(dst+x)),

  1450.                   "+m"(*(uint32_t*)(dst+x+stride))

  1451.             );

  1452.         }

  1453.         dst += 2*stride;

  1454.     }

  1455. }

  1456. 

  1457. 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)

  1458. {

  1459.     int x, y;

  1460.     offset = ((offset + 1) | 1) << log2_denom;

  1461.     asm volatile(

  1462.         "movd    %0, %%mm3        \n\t"

  1463.         "movd    %1, %%mm4        \n\t"

  1464.         "movd    %2, %%mm5        \n\t"

  1465.         "movd    %3, %%mm6        \n\t"

  1466.         "pshufw  $0, %%mm3, %%mm3 \n\t"

  1467.         "pshufw  $0, %%mm4, %%mm4 \n\t"

  1468.         "pshufw  $0, %%mm5, %%mm5 \n\t"

  1469.         "pxor    %%mm7, %%mm7     \n\t"

  1470.         :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)

  1471.     );

  1472.     for(y=0; y<h; y++){

  1473.         for(x=0; x<w; x+=4){

  1474.             asm volatile(

  1475.                 "movd      %0,    %%mm0 \n\t"

  1476.                 "movd      %1,    %%mm1 \n\t"

  1477.                 "punpcklbw %%mm7, %%mm0 \n\t"

  1478.                 "punpcklbw %%mm7, %%mm1 \n\t"

  1479.                 "pmullw    %%mm3, %%mm0 \n\t"

  1480.                 "pmullw    %%mm4, %%mm1 \n\t"

  1481.                 "paddsw    %%mm1, %%mm0 \n\t"

  1482.                 "paddsw    %%mm5, %%mm0 \n\t"

  1483.                 "psraw     %%mm6, %%mm0 \n\t"

  1484.                 "packuswb  %%mm0, %%mm0 \n\t"

  1485.                 "movd      %%mm0, %0    \n\t"

  1486.                 : "+m"(*(uint32_t*)(dst+x))

  1487.                 :  "m"(*(uint32_t*)(src+x))

  1488.             );

  1489.         }

  1490.         src += stride;

  1491.         dst += stride;

  1492.     }

  1493. }

  1494. 

  1495. #define H264_WEIGHT(W,H) \

  1496. 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){ \

  1497.     ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \

  1498. } \

  1499. static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \

  1500.     ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \

  1501. }

  1502. 

  1503. H264_WEIGHT(16,16)

  1504. H264_WEIGHT(16, 8)

  1505. H264_WEIGHT( 8,16)

  1506. H264_WEIGHT( 8, 8)

  1507. H264_WEIGHT( 8, 4)

  1508. H264_WEIGHT( 4, 8)

  1509. H264_WEIGHT( 4, 4)

  1510. H264_WEIGHT( 4, 2)