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.
5407 Commits
32 Branches
944MB
Tree: 2669777981
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '2669777981'
${ noResults }
FFmpeg/libavcodec/i386/h264dsp_mmx.c

1218 lines
48KB
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. void ff_h264_idct_add_mmx2(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. void ff_h264_idct_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)

  108. {

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

  110.     asm volatile(

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

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

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

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

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

  116.         "pmaxsw     %%mm7, %%mm0 \n\t"

  117.         "pmaxsw     %%mm7, %%mm1 \n\t"

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

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

  120.         ::"r"(dc)

  121.     );

  122.     asm volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  143.     );

  144. }

  145. 

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

  147. {

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

  149.     int y;

  150.     asm volatile(

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

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

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

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

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

  156.         "pmaxsw     %%mm7, %%mm0 \n\t"

  157.         "pmaxsw     %%mm7, %%mm1 \n\t"

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

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

  160.         ::"r"(dc)

  161.     );

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

  163.     asm volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  184.     );

  185.     }

  186. }

  187. 

  188. 

  189. /***********************************/

  190. /* deblocking */

  191. 

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

  193. // clobbers: t

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

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

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

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

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

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

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

  201. 

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

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

  204. // clobbers: mm4,mm6

  205. #define H264_DEBLOCK_MASK(alpha1, beta1) \

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

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

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

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

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

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

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

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

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

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

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

  217. 

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

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

  220. // clobbers: mm0,3-6

  221. #define H264_DEBLOCK_P0_Q0(pb_01, pb_3f)\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  262. 

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

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

  265. // clobbers: q2, tmp, tc0

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

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

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

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

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

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

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

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

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

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

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

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

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

  279. 

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

  281. {

  282.     uint64_t tmp0;

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

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

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

  286. 

  287.     asm volatile(

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

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

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

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

  292.         H264_DEBLOCK_MASK(%6, %7)

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

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

  295. 

  296.         /* filter p1 */

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

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

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

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

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

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

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

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

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

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

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

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

  309. 

  310.         /* filter q1 */

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

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

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

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

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

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

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

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

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

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

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

  322. 

  323.         /* filter p0, q0 */

  324.         H264_DEBLOCK_P0_Q0(%8, %9)

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

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

  327. 

  328.         : "=m"(tmp0)

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

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

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

  332.     );

  333. }

  334. 

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

  336. {

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

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

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

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

  341. }

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

  343. {

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

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

  346.     uint8_t trans[8*8];

  347.     int i;

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

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

  350.             continue;

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

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

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

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

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

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

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

  358.     }

  359. }

  360. 

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

  362. {

  363.     asm volatile(

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

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

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

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

  368.         H264_DEBLOCK_MASK(%4, %5)

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

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

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

  372.         H264_DEBLOCK_P0_Q0(%6, %7)

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

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

  375. 

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

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

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

  379.     );

  380. }

  381. 

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

  383. {

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

  385. }

  386. 

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

  388. {

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

  390.     uint8_t trans[8*4];

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

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

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

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

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

  396. }

  397. 

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

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

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

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

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

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

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

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

  406. 

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

  408. {

  409.     asm volatile(

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

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

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

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

  414.         H264_DEBLOCK_MASK(%3, %4)

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  429.     );

  430. }

  431. 

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

  433. {

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

  435. }

  436. 

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

  438. {

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

  440.     uint8_t trans[8*4];

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

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

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

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

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

  446. }

  447. 

  448. 

  449. /***********************************/

  450. /* motion compensation */

  451. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  469. 

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

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

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

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

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

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

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

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

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

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

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

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

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

  483. 

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

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

  486.     int h=4;\

  487. \

  488.     asm volatile(\

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

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

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

  492.         "1:                         \n\t"\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  522.         : "memory"\

  523.     );\

  524. }\

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

  526.     src -= 2*srcStride;\

  527.     asm volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  548.          \

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

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

  551.         : "memory"\

  552.     );\

  553. }\

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

  555.     int h=4;\

  556.     int w=3;\

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

  558.     while(w--){\

  559.         asm volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  580.              \

  581.             : "+a"(src)\

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

  583.             : "memory"\

  584.         );\

  585.         tmp += 4;\

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

  587.     }\

  588.     tmp -= 3*4;\

  589.     asm volatile(\

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

  591.         "1:                         \n\t"\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  614.         : "memory"\

  615.     );\

  616. }\

  617. \

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

  619.     int h=8;\

  620.     asm volatile(\

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

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

  623.         "1:                         \n\t"\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  671.         : "memory"\

  672.     );\

  673. }\

  674. \

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

  676.     int w= 2;\

  677.     src -= 2*srcStride;\

  678.     \

  679.     while(w--){\

  680.       asm volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  705.          \

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

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

  708.         : "memory"\

  709.      );\

  710.      if(h==16){\

  711.         asm volatile(\

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

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

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

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

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

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

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

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

  720.             \

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

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

  723.            : "memory"\

  724.         );\

  725.      }\

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

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

  728.    }\

  729. }\

  730. 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){\

  731.     int h = size;\

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

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

  734.     while(w--){\

  735.         asm volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  760.             : "+a"(src)\

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

  762.             : "memory"\

  763.         );\

  764.         if(size==16){\

  765.             asm volatile(\

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

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

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

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

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

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

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

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

  774.                 : "+a"(src)\

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

  776.                 : "memory"\

  777.             );\

  778.         }\

  779.         tmp += 4;\

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

  781.     }\

  782.     tmp -= size+8;\

  783.     w = size>>4;\

  784.     do{\

  785.     h = size;\

  786.     asm volatile(\

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

  788.         "1:                         \n\t"\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  825.         : "memory"\

  826.     );\

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

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

  829.     }while(w--);\

  830. }\

  831. \

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

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

  834. }\

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

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

  837.     OPNAME ## h264_qpel8or16_v_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\

  838. }\

  839. \

  840. static void OPNAME ## h264_qpel16_h_lowpass_ ## MMX(uint8_t *dst, uint8_t *src, int dstStride, int srcStride){\

  841.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\

  842.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  843.     src += 8*srcStride;\

  844.     dst += 8*dstStride;\

  845.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst  , src  , dstStride, srcStride);\

  846.     OPNAME ## h264_qpel8_h_lowpass_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  847. }\

  848. \

  849. static void OPNAME ## h264_qpel8_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\

  850.     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 8);\

  851. }\

  852. \

  853. static void OPNAME ## h264_qpel16_hv_lowpass_ ## MMX(uint8_t *dst, int16_t *tmp, uint8_t *src, int dstStride, int tmpStride, int srcStride){\

  854.     OPNAME ## h264_qpel8or16_hv_lowpass_ ## MMX(dst  , tmp  , src  , dstStride, tmpStride, srcStride, 16);\

  855. }\

  856. \

  857. static void OPNAME ## pixels4_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\

  858. {\

  859.     asm volatile(\

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

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

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

  863.         "movq    24(%1), %%mm1          \n\t"\

  864.         "paddw    %%mm6, %%mm0          \n\t"\

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

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

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

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

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

  870.         "packuswb %%mm7, %%mm0          \n\t"\

  871.         "packuswb %%mm7, %%mm1          \n\t"\

  872.         PAVGB"     (%0), %%mm0          \n\t"\

  873.         PAVGB"  (%0,%3), %%mm1          \n\t"\

  874.         OP(%%mm0, (%2),    %%mm4, d)\

  875.         OP(%%mm1, (%2,%4), %%mm5, d)\

  876.         "lea  (%0,%3,2), %0             \n\t"\

  877.         "lea  (%2,%4,2), %2             \n\t"\

  878.         "movq    48(%1), %%mm0          \n\t"\

  879.         "movq    72(%1), %%mm1          \n\t"\

  880.         "paddw    %%mm6, %%mm0          \n\t"\

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

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

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

  884.         "pmaxsw   %%mm7, %%mm0          \n\t"\

  885.         "pmaxsw   %%mm7, %%mm1          \n\t"\

  886.         "packuswb %%mm7, %%mm0          \n\t"\

  887.         "packuswb %%mm7, %%mm1          \n\t"\

  888.         PAVGB"     (%0), %%mm0          \n\t"\

  889.         PAVGB"  (%0,%3), %%mm1          \n\t"\

  890.         OP(%%mm0, (%2),    %%mm4, d)\

  891.         OP(%%mm1, (%2,%4), %%mm5, d)\

  892.         :"+a"(src8), "+c"(src16), "+d"(dst)\

  893.         :"S"((long)src8Stride), "D"((long)dstStride), "m"(ff_pw_16)\

  894.         :"memory");\

  895. }\

  896. static void OPNAME ## pixels8_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\

  897. {\

  898.     asm volatile(\

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

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

  901.         ::"m"(ff_pw_16)\

  902.         );\

  903.     while(h--){\

  904.     asm volatile(\

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

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

  907.         "paddw    %%mm6, %%mm0          \n\t"\

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

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

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

  911.         "pmaxsw   %%mm7, %%mm0          \n\t"\

  912.         "pmaxsw   %%mm7, %%mm1          \n\t"\

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

  914.         PAVGB"     (%0), %%mm0          \n\t"\

  915.         OP(%%mm0, (%2), %%mm5, q)\

  916.         ::"a"(src8), "c"(src16), "d"(dst)\

  917.         :"memory");\

  918.         src8 += src8Stride;\

  919.         src16 += 24;\

  920.         dst += dstStride;\

  921.     }\

  922. }\

  923. static void OPNAME ## pixels16_l2_shift5_ ## MMX(uint8_t *dst, int16_t *src16, uint8_t *src8, int dstStride, int src8Stride, int h)\

  924. {\

  925.     OPNAME ## pixels8_l2_shift5_ ## MMX(dst  , src16  , src8  , dstStride, src8Stride, h);\

  926.     OPNAME ## pixels8_l2_shift5_ ## MMX(dst+8, src16+8, src8+8, dstStride, src8Stride, h);\

  927. }\

  928. 

  929. 

  930. #define H264_MC(OPNAME, SIZE, MMX) \

  931. static void OPNAME ## h264_qpel ## SIZE ## _mc00_ ## MMX (uint8_t *dst, uint8_t *src, int stride){\

  932.     OPNAME ## pixels ## SIZE ## _mmx(dst, src, stride, SIZE);\

  933. }\

  934. \

  935. static void OPNAME ## h264_qpel ## SIZE ## _mc10_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  936.     uint64_t temp[SIZE*SIZE/8];\

  937.     uint8_t * const half= (uint8_t*)temp;\

  938.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(half, src, SIZE, stride);\

  939.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\

  940. }\

  941. \

  942. static void OPNAME ## h264_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  943.     OPNAME ## h264_qpel ## SIZE ## _h_lowpass_ ## MMX(dst, src, stride, stride);\

  944. }\

  945. \

  946. static void OPNAME ## h264_qpel ## SIZE ## _mc30_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  947.     uint64_t temp[SIZE*SIZE/8];\

  948.     uint8_t * const half= (uint8_t*)temp;\

  949.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(half, src, SIZE, stride);\

  950.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+1, half, stride, stride, SIZE);\

  951. }\

  952. \

  953. static void OPNAME ## h264_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  954.     uint64_t temp[SIZE*SIZE/8];\

  955.     uint8_t * const half= (uint8_t*)temp;\

  956.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\

  957.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src, half, stride, stride, SIZE);\

  958. }\

  959. \

  960. static void OPNAME ## h264_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  961.     OPNAME ## h264_qpel ## SIZE ## _v_lowpass_ ## MMX(dst, src, stride, stride);\

  962. }\

  963. \

  964. static void OPNAME ## h264_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  965.     uint64_t temp[SIZE*SIZE/8];\

  966.     uint8_t * const half= (uint8_t*)temp;\

  967.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(half, src, SIZE, stride);\

  968.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, src+stride, half, stride, stride, SIZE);\

  969. }\

  970. \

  971. static void OPNAME ## h264_qpel ## SIZE ## _mc11_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  972.     uint64_t temp[SIZE*SIZE/4];\

  973.     uint8_t * const halfH= (uint8_t*)temp;\

  974.     uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\

  975.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\

  976.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\

  977.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\

  978. }\

  979. \

  980. static void OPNAME ## h264_qpel ## SIZE ## _mc31_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  981.     uint64_t temp[SIZE*SIZE/4];\

  982.     uint8_t * const halfH= (uint8_t*)temp;\

  983.     uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\

  984.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\

  985.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\

  986.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\

  987. }\

  988. \

  989. static void OPNAME ## h264_qpel ## SIZE ## _mc13_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  990.     uint64_t temp[SIZE*SIZE/4];\

  991.     uint8_t * const halfH= (uint8_t*)temp;\

  992.     uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\

  993.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\

  994.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src, SIZE, stride);\

  995.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\

  996. }\

  997. \

  998. static void OPNAME ## h264_qpel ## SIZE ## _mc33_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  999.     uint64_t temp[SIZE*SIZE/4];\

  1000.     uint8_t * const halfH= (uint8_t*)temp;\

  1001.     uint8_t * const halfV= ((uint8_t*)temp) + SIZE*SIZE;\

  1002.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\

  1003.     put_h264_qpel ## SIZE ## _v_lowpass_ ## MMX(halfV, src+1, SIZE, stride);\

  1004.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfV, stride, SIZE, SIZE);\

  1005. }\

  1006. \

  1007. static void OPNAME ## h264_qpel ## SIZE ## _mc22_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1008.     uint64_t temp[SIZE*(SIZE<8?12:24)/4];\

  1009.     int16_t * const tmp= (int16_t*)temp;\

  1010.     OPNAME ## h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(dst, tmp, src, stride, SIZE, stride);\

  1011. }\

  1012. \

  1013. static void OPNAME ## h264_qpel ## SIZE ## _mc21_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1014.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/4];\

  1015.     uint8_t * const halfH= (uint8_t*)temp;\

  1016.     uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\

  1017.     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\

  1018.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src, SIZE, stride);\

  1019.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\

  1020.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfHV, stride, SIZE, SIZE);\

  1021. }\

  1022. \

  1023. static void OPNAME ## h264_qpel ## SIZE ## _mc23_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1024.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/4];\

  1025.     uint8_t * const halfH= (uint8_t*)temp;\

  1026.     uint8_t * const halfHV= ((uint8_t*)temp) + SIZE*SIZE;\

  1027.     int16_t * const tmp= ((int16_t*)temp) + SIZE*SIZE;\

  1028.     put_h264_qpel ## SIZE ## _h_lowpass_ ## MMX(halfH, src + stride, SIZE, stride);\

  1029.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, tmp, src, SIZE, SIZE, stride);\

  1030.     OPNAME ## pixels ## SIZE ## _l2_ ## MMX(dst, halfH, halfHV, stride, SIZE, SIZE);\

  1031. }\

  1032. \

  1033. static void OPNAME ## h264_qpel ## SIZE ## _mc12_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1034.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\

  1035.     int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\

  1036.     uint8_t * const halfHV= ((uint8_t*)temp);\

  1037.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\

  1038.     OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+2, halfHV, stride, SIZE, SIZE);\

  1039. }\

  1040. \

  1041. static void OPNAME ## h264_qpel ## SIZE ## _mc32_ ## MMX(uint8_t *dst, uint8_t *src, int stride){\

  1042.     uint64_t temp[SIZE*(SIZE<8?12:24)/4 + SIZE*SIZE/8];\

  1043.     int16_t * const halfV= ((int16_t*)temp) + SIZE*SIZE/2;\

  1044.     uint8_t * const halfHV= ((uint8_t*)temp);\

  1045.     put_h264_qpel ## SIZE ## _hv_lowpass_ ## MMX(halfHV, halfV, src, SIZE, SIZE, stride);\

  1046.     OPNAME ## pixels ## SIZE ## _l2_shift5_ ## MMX(dst, halfV+3, halfHV, stride, SIZE, SIZE);\

  1047. }\

  1048. 

  1049. 

  1050. #define PUT_OP(a,b,temp, size) "mov" #size " " #a ", " #b "    \n\t"

  1051. #define AVG_3DNOW_OP(a,b,temp, size) \

  1052. "mov" #size " " #b ", " #temp "   \n\t"\

  1053. "pavgusb " #temp ", " #a "        \n\t"\

  1054. "mov" #size " " #a ", " #b "      \n\t"

  1055. #define AVG_MMX2_OP(a,b,temp, size) \

  1056. "mov" #size " " #b ", " #temp "   \n\t"\

  1057. "pavgb " #temp ", " #a "          \n\t"\

  1058. "mov" #size " " #a ", " #b "      \n\t"

  1059. 

  1060. #define PAVGB "pavgusb"

  1061. QPEL_H264(put_,       PUT_OP, 3dnow)

  1062. QPEL_H264(avg_, AVG_3DNOW_OP, 3dnow)

  1063. #undef PAVGB

  1064. #define PAVGB "pavgb"

  1065. QPEL_H264(put_,       PUT_OP, mmx2)

  1066. QPEL_H264(avg_,  AVG_MMX2_OP, mmx2)

  1067. #undef PAVGB

  1068. 

  1069. H264_MC(put_, 4, 3dnow)

  1070. H264_MC(put_, 8, 3dnow)

  1071. H264_MC(put_, 16,3dnow)

  1072. H264_MC(avg_, 4, 3dnow)

  1073. H264_MC(avg_, 8, 3dnow)

  1074. H264_MC(avg_, 16,3dnow)

  1075. H264_MC(put_, 4, mmx2)

  1076. H264_MC(put_, 8, mmx2)

  1077. H264_MC(put_, 16,mmx2)

  1078. H264_MC(avg_, 4, mmx2)

  1079. H264_MC(avg_, 8, mmx2)

  1080. H264_MC(avg_, 16,mmx2)

  1081. 

  1082. 

  1083. #define H264_CHROMA_OP(S,D)

  1084. #define H264_CHROMA_OP4(S,D,T)

  1085. #define H264_CHROMA_MC8_TMPL put_h264_chroma_mc8_mmx

  1086. #define H264_CHROMA_MC4_TMPL put_h264_chroma_mc4_mmx

  1087. #define H264_CHROMA_MC8_MV0 put_pixels8_mmx

  1088. #include "dsputil_h264_template_mmx.c"

  1089. #undef H264_CHROMA_OP

  1090. #undef H264_CHROMA_OP4

  1091. #undef H264_CHROMA_MC8_TMPL

  1092. #undef H264_CHROMA_MC4_TMPL

  1093. #undef H264_CHROMA_MC8_MV0

  1094. 

  1095. #define H264_CHROMA_OP(S,D) "pavgb " #S ", " #D " \n\t"

  1096. #define H264_CHROMA_OP4(S,D,T) "movd  " #S ", " #T " \n\t"\

  1097.                                "pavgb " #T ", " #D " \n\t"

  1098. #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_mmx2

  1099. #define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_mmx2

  1100. #define H264_CHROMA_MC8_MV0 avg_pixels8_mmx2

  1101. #include "dsputil_h264_template_mmx.c"

  1102. #undef H264_CHROMA_OP

  1103. #undef H264_CHROMA_OP4

  1104. #undef H264_CHROMA_MC8_TMPL

  1105. #undef H264_CHROMA_MC4_TMPL

  1106. #undef H264_CHROMA_MC8_MV0

  1107. 

  1108. #define H264_CHROMA_OP(S,D) "pavgusb " #S ", " #D " \n\t"

  1109. #define H264_CHROMA_OP4(S,D,T) "movd " #S ", " #T " \n\t"\

  1110.                                "pavgusb " #T ", " #D " \n\t"

  1111. #define H264_CHROMA_MC8_TMPL avg_h264_chroma_mc8_3dnow

  1112. #define H264_CHROMA_MC4_TMPL avg_h264_chroma_mc4_3dnow

  1113. #define H264_CHROMA_MC8_MV0 avg_pixels8_3dnow

  1114. #include "dsputil_h264_template_mmx.c"

  1115. #undef H264_CHROMA_OP

  1116. #undef H264_CHROMA_OP4

  1117. #undef H264_CHROMA_MC8_TMPL

  1118. #undef H264_CHROMA_MC4_TMPL

  1119. #undef H264_CHROMA_MC8_MV0

  1120. 

  1121. /***********************************/

  1122. /* weighted prediction */

  1123. 

  1124. static inline void ff_h264_weight_WxH_mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset, int w, int h)

  1125. {

  1126.     int x, y;

  1127.     offset <<= log2_denom;

  1128.     offset += (1 << log2_denom) >> 1;

  1129.     asm volatile(

  1130.         "movd    %0, %%mm4        \n\t"

  1131.         "movd    %1, %%mm5        \n\t"

  1132.         "movd    %2, %%mm6        \n\t"

  1133.         "pshufw  $0, %%mm4, %%mm4 \n\t"

  1134.         "pshufw  $0, %%mm5, %%mm5 \n\t"

  1135.         "pxor    %%mm7, %%mm7     \n\t"

  1136.         :: "g"(weight), "g"(offset), "g"(log2_denom)

  1137.     );

  1138.     for(y=0; y<h; y+=2){

  1139.         for(x=0; x<w; x+=4){

  1140.             asm volatile(

  1141.                 "movd      %0,    %%mm0 \n\t"

  1142.                 "movd      %1,    %%mm1 \n\t"

  1143.                 "punpcklbw %%mm7, %%mm0 \n\t"

  1144.                 "punpcklbw %%mm7, %%mm1 \n\t"

  1145.                 "pmullw    %%mm4, %%mm0 \n\t"

  1146.                 "pmullw    %%mm4, %%mm1 \n\t"

  1147.                 "paddsw    %%mm5, %%mm0 \n\t"

  1148.                 "paddsw    %%mm5, %%mm1 \n\t"

  1149.                 "psraw     %%mm6, %%mm0 \n\t"

  1150.                 "psraw     %%mm6, %%mm1 \n\t"

  1151.                 "packuswb  %%mm7, %%mm0 \n\t"

  1152.                 "packuswb  %%mm7, %%mm1 \n\t"

  1153.                 "movd      %%mm0, %0    \n\t"

  1154.                 "movd      %%mm1, %1    \n\t"

  1155.                 : "+m"(*(uint32_t*)(dst+x)),

  1156.                   "+m"(*(uint32_t*)(dst+x+stride))

  1157.             );

  1158.         }

  1159.         dst += 2*stride;

  1160.     }

  1161. }

  1162. 

  1163. 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)

  1164. {

  1165.     int x, y;

  1166.     offset = ((offset + 1) | 1) << log2_denom;

  1167.     asm volatile(

  1168.         "movd    %0, %%mm3        \n\t"

  1169.         "movd    %1, %%mm4        \n\t"

  1170.         "movd    %2, %%mm5        \n\t"

  1171.         "movd    %3, %%mm6        \n\t"

  1172.         "pshufw  $0, %%mm3, %%mm3 \n\t"

  1173.         "pshufw  $0, %%mm4, %%mm4 \n\t"

  1174.         "pshufw  $0, %%mm5, %%mm5 \n\t"

  1175.         "pxor    %%mm7, %%mm7     \n\t"

  1176.         :: "g"(weightd), "g"(weights), "g"(offset), "g"(log2_denom+1)

  1177.     );

  1178.     for(y=0; y<h; y++){

  1179.         for(x=0; x<w; x+=4){

  1180.             asm volatile(

  1181.                 "movd      %0,    %%mm0 \n\t"

  1182.                 "movd      %1,    %%mm1 \n\t"

  1183.                 "punpcklbw %%mm7, %%mm0 \n\t"

  1184.                 "punpcklbw %%mm7, %%mm1 \n\t"

  1185.                 "pmullw    %%mm3, %%mm0 \n\t"

  1186.                 "pmullw    %%mm4, %%mm1 \n\t"

  1187.                 "paddsw    %%mm1, %%mm0 \n\t"

  1188.                 "paddsw    %%mm5, %%mm0 \n\t"

  1189.                 "psraw     %%mm6, %%mm0 \n\t"

  1190.                 "packuswb  %%mm0, %%mm0 \n\t"

  1191.                 "movd      %%mm0, %0    \n\t"

  1192.                 : "+m"(*(uint32_t*)(dst+x))

  1193.                 :  "m"(*(uint32_t*)(src+x))

  1194.             );

  1195.         }

  1196.         src += stride;

  1197.         dst += stride;

  1198.     }

  1199. }

  1200. 

  1201. #define H264_WEIGHT(W,H) \

  1202. 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){ \

  1203.     ff_h264_biweight_WxH_mmx2(dst, src, stride, log2_denom, weightd, weights, offset, W, H); \

  1204. } \

  1205. static void ff_h264_weight_ ## W ## x ## H ## _mmx2(uint8_t *dst, int stride, int log2_denom, int weight, int offset){ \

  1206.     ff_h264_weight_WxH_mmx2(dst, stride, log2_denom, weight, offset, W, H); \

  1207. }

  1208. 

  1209. H264_WEIGHT(16,16)

  1210. H264_WEIGHT(16, 8)

  1211. H264_WEIGHT( 8,16)

  1212. H264_WEIGHT( 8, 8)

  1213. H264_WEIGHT( 8, 4)

  1214. H264_WEIGHT( 4, 8)

  1215. H264_WEIGHT( 4, 4)

  1216. H264_WEIGHT( 4, 2)