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.
41838 Commits
32 Branches
759MB
Tree: 3058872c29
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '3058872c29'
${ noResults }
FFmpeg/libavcodec/x86/cavsdsp.c

559 lines
21KB
Raw Blame History 

  1. /*

  2.  * Chinese AVS video (AVS1-P2, JiZhun profile) decoder.

  3.  * Copyright (c) 2006  Stefan Gehrer <stefan.gehrer@gmx.de>

  4.  *

  5.  * MMX-optimized DSP functions, based on H.264 optimizations by

  6.  * Michael Niedermayer and Loren Merritt

  7.  *

  8.  * This file is part of Libav.

  9.  *

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

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

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

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

  14.  *

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

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

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

  18.  * Lesser General Public License for more details.

  19.  *

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

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

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

  23.  */

  24. 

  25. #include "libavutil/attributes.h"

  26. #include "libavutil/common.h"

  27. #include "libavutil/cpu.h"

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

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

  30. #include "libavcodec/cavsdsp.h"

  31. #include "libavcodec/idctdsp.h"

  32. #include "constants.h"

  33. #include "fpel.h"

  34. #include "idctdsp.h"

  35. #include "config.h"

  36. 

  37. #if HAVE_MMX_INLINE

  38. 

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

  40. #define SUMSUB_BA( a, b ) \

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

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

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

  44. 

  45. /*****************************************************************************

  46.  *

  47.  * inverse transform

  48.  *

  49.  ****************************************************************************/

  50. 

  51. static inline void cavs_idct8_1d(int16_t *block, uint64_t bias)

  52. {

  53.     __asm__ volatile(

  54.         "movq 112(%0), %%mm4  \n\t" /* mm4 = src7 */

  55.         "movq  16(%0), %%mm5  \n\t" /* mm5 = src1 */

  56.         "movq  80(%0), %%mm2  \n\t" /* mm2 = src5 */

  57.         "movq  48(%0), %%mm7  \n\t" /* mm7 = src3 */

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

  59.         "movq   %%mm5, %%mm3  \n\t"

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

  61.         "movq   %%mm7, %%mm1  \n\t"

  62. 

  63.         "paddw  %%mm4, %%mm4  \n\t" /* mm4 = 2*src7 */

  64.         "paddw  %%mm3, %%mm3  \n\t" /* mm3 = 2*src1 */

  65.         "paddw  %%mm6, %%mm6  \n\t" /* mm6 = 2*src5 */

  66.         "paddw  %%mm1, %%mm1  \n\t" /* mm1 = 2*src3 */

  67.         "paddw  %%mm4, %%mm0  \n\t" /* mm0 = 3*src7 */

  68.         "paddw  %%mm3, %%mm5  \n\t" /* mm5 = 3*src1 */

  69.         "paddw  %%mm6, %%mm2  \n\t" /* mm2 = 3*src5 */

  70.         "paddw  %%mm1, %%mm7  \n\t" /* mm7 = 3*src3 */

  71.         "psubw  %%mm4, %%mm5  \n\t" /* mm5 = 3*src1 - 2*src7 = a0 */

  72.         "paddw  %%mm6, %%mm7  \n\t" /* mm7 = 3*src3 + 2*src5 = a1 */

  73.         "psubw  %%mm2, %%mm1  \n\t" /* mm1 = 2*src3 - 3*src5 = a2 */

  74.         "paddw  %%mm0, %%mm3  \n\t" /* mm3 = 2*src1 + 3*src7 = a3 */

  75. 

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

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

  78.         "movq   %%mm3, %%mm0  \n\t"

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

  80.         SUMSUB_BA( %%mm7, %%mm5 )   /* mm7 = a0 + a1  mm5 = a0 - a1 */

  81.         "paddw  %%mm3, %%mm7  \n\t" /* mm7 = a0 + a1 + a3 */

  82.         "paddw  %%mm1, %%mm5  \n\t" /* mm5 = a0 - a1 + a2 */

  83.         "paddw  %%mm7, %%mm7  \n\t"

  84.         "paddw  %%mm5, %%mm5  \n\t"

  85.         "paddw  %%mm6, %%mm7  \n\t" /* mm7 = b4 */

  86.         "paddw  %%mm4, %%mm5  \n\t" /* mm5 = b5 */

  87. 

  88.         SUMSUB_BA( %%mm1, %%mm3 )   /* mm1 = a3 + a2  mm3 = a3 - a2 */

  89.         "psubw  %%mm1, %%mm4  \n\t" /* mm4 = a0 - a2 - a3 */

  90.         "movq   %%mm4, %%mm1  \n\t" /* mm1 = a0 - a2 - a3 */

  91.         "psubw  %%mm6, %%mm3  \n\t" /* mm3 = a3 - a2 - a1 */

  92.         "paddw  %%mm1, %%mm1  \n\t"

  93.         "paddw  %%mm3, %%mm3  \n\t"

  94.         "psubw  %%mm2, %%mm1  \n\t" /* mm1 = b7 */

  95.         "paddw  %%mm0, %%mm3  \n\t" /* mm3 = b6 */

  96. 

  97.         "movq  32(%0), %%mm2  \n\t" /* mm2 = src2 */

  98.         "movq  96(%0), %%mm6  \n\t" /* mm6 = src6 */

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

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

  101.         "psllw  $2,    %%mm4  \n\t" /* mm4 = 4*src2 */

  102.         "psllw  $2,    %%mm6  \n\t" /* mm6 = 4*src6 */

  103.         "paddw  %%mm4, %%mm2  \n\t" /* mm2 = 5*src2 */

  104.         "paddw  %%mm6, %%mm0  \n\t" /* mm0 = 5*src6 */

  105.         "paddw  %%mm2, %%mm2  \n\t"

  106.         "paddw  %%mm0, %%mm0  \n\t"

  107.         "psubw  %%mm0, %%mm4  \n\t" /* mm4 = 4*src2 - 10*src6 = a7 */

  108.         "paddw  %%mm2, %%mm6  \n\t" /* mm6 = 4*src6 + 10*src2 = a6 */

  109. 

  110.         "movq    (%0), %%mm2  \n\t" /* mm2 = src0 */

  111.         "movq  64(%0), %%mm0  \n\t" /* mm0 = src4 */

  112.         SUMSUB_BA( %%mm0, %%mm2 )   /* mm0 = src0+src4  mm2 = src0-src4 */

  113.         "psllw  $3,    %%mm0  \n\t"

  114.         "psllw  $3,    %%mm2  \n\t"

  115.         "paddw  %1,    %%mm0  \n\t" /* add rounding bias */

  116.         "paddw  %1,    %%mm2  \n\t" /* add rounding bias */

  117. 

  118.         SUMSUB_BA( %%mm6, %%mm0 )   /* mm6 = a4 + a6  mm0 = a4 - a6 */

  119.         SUMSUB_BA( %%mm4, %%mm2 )   /* mm4 = a5 + a7  mm2 = a5 - a7 */

  120.         SUMSUB_BA( %%mm7, %%mm6 )   /* mm7 = dst0  mm6 = dst7 */

  121.         SUMSUB_BA( %%mm5, %%mm4 )   /* mm5 = dst1  mm4 = dst6 */

  122.         SUMSUB_BA( %%mm3, %%mm2 )   /* mm3 = dst2  mm2 = dst5 */

  123.         SUMSUB_BA( %%mm1, %%mm0 )   /* mm1 = dst3  mm0 = dst4 */

  124.         :: "r"(block), "m"(bias)

  125.     );

  126. }

  127. 

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

  129.     "mov" #m " " #a ", " #t "         \n\t" /* abcd */\

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

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

  132. 

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

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

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

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

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

  138. 

  139. static void cavs_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)

  140. {

  141.     int i;

  142.     DECLARE_ALIGNED(8, int16_t, b2)[64];

  143. 

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

  145.         cavs_idct8_1d(block + 4 * i, ff_pw_4.a);

  146. 

  147.         __asm__ volatile(

  148.             "psraw     $3, %%mm7  \n\t"

  149.             "psraw     $3, %%mm6  \n\t"

  150.             "psraw     $3, %%mm5  \n\t"

  151.             "psraw     $3, %%mm4  \n\t"

  152.             "psraw     $3, %%mm3  \n\t"

  153.             "psraw     $3, %%mm2  \n\t"

  154.             "psraw     $3, %%mm1  \n\t"

  155.             "psraw     $3, %%mm0  \n\t"

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

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

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

  159.             "movq   %%mm6, 24(%0)  \n\t"

  160.             "movq   %%mm7, 40(%0)  \n\t"

  161.             "movq   %%mm4, 56(%0)  \n\t"

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

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

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

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

  166.             "movq   %%mm0, 32(%0)  \n\t"

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

  168.             :

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

  170.             : "memory"

  171.         );

  172.     }

  173. 

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

  175.         cavs_idct8_1d(b2+4*i, ff_pw_64.a);

  176. 

  177.         __asm__ volatile(

  178.             "psraw     $7, %%mm7  \n\t"

  179.             "psraw     $7, %%mm6  \n\t"

  180.             "psraw     $7, %%mm5  \n\t"

  181.             "psraw     $7, %%mm4  \n\t"

  182.             "psraw     $7, %%mm3  \n\t"

  183.             "psraw     $7, %%mm2  \n\t"

  184.             "psraw     $7, %%mm1  \n\t"

  185.             "psraw     $7, %%mm0  \n\t"

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

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

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

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

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

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

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

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

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

  195.             : "memory"

  196.         );

  197.     }

  198. 

  199.     ff_add_pixels_clamped_mmx(b2, dst, stride);

  200. }

  201. 

  202. #endif /* HAVE_MMX_INLINE */

  203. 

  204. #if (HAVE_MMXEXT_INLINE || HAVE_AMD3DNOW_INLINE)

  205. 

  206. /*****************************************************************************

  207.  *

  208.  * motion compensation

  209.  *

  210.  ****************************************************************************/

  211. 

  212. /* vertical filter [-1 -2 96 42 -7  0]  */

  213. #define QPEL_CAVSV1(A,B,C,D,E,F,OP,MUL2) \

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

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

  216.         "pmullw %5, %%mm6           \n\t"\

  217.         "movq "#D", %%mm7           \n\t"\

  218.         "pmullw "MANGLE(MUL2)", %%mm7\n\t"\

  219.         "psllw $3, "#E"             \n\t"\

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

  221.         "psraw $3, "#E"             \n\t"\

  222.         "paddw %%mm7, %%mm6         \n\t"\

  223.         "paddw "#E", %%mm6          \n\t"\

  224.         "paddw "#B", "#B"           \n\t"\

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

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

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

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

  229.         "psraw $1, "#B"             \n\t"\

  230.         "psubw "#A", %%mm6          \n\t"\

  231.         "paddw %4, %%mm6            \n\t"\

  232.         "psraw $7, %%mm6            \n\t"\

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

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

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

  236. 

  237. /* vertical filter [ 0 -1  5  5 -1  0]  */

  238. #define QPEL_CAVSV2(A,B,C,D,E,F,OP,MUL2) \

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

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

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

  242.         "pmullw %5, %%mm6           \n\t"\

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

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

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

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

  247.         "paddw %4, %%mm6            \n\t"\

  248.         "psraw $3, %%mm6            \n\t"\

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

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

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

  252. 

  253. /* vertical filter [ 0 -7 42 96 -2 -1]  */

  254. #define QPEL_CAVSV3(A,B,C,D,E,F,OP,MUL2) \

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

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

  257.         "pmullw "MANGLE(MUL2)", %%mm6\n\t"\

  258.         "movq "#D", %%mm7           \n\t"\

  259.         "pmullw %5, %%mm7           \n\t"\

  260.         "psllw $3, "#B"             \n\t"\

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

  262.         "psraw $3, "#B"             \n\t"\

  263.         "paddw %%mm7, %%mm6         \n\t"\

  264.         "paddw "#B", %%mm6          \n\t"\

  265.         "paddw "#E", "#E"           \n\t"\

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

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

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

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

  270.         "psraw $1, "#E"             \n\t"\

  271.         "psubw "#F", %%mm6          \n\t"\

  272.         "paddw %4, %%mm6            \n\t"\

  273.         "psraw $7, %%mm6            \n\t"\

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

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

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

  277. 

  278. 

  279. #define QPEL_CAVSVNUM(VOP,OP,ADD,MUL1,MUL2)\

  280.     int w= 2;\

  281.     src -= 2*srcStride;\

  282.     \

  283.     while(w--){\

  284.       __asm__ volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  301.         VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\

  302.         VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\

  303.         VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\

  304.         VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\

  305.         VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, MUL2)\

  306.         VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, MUL2)\

  307.         VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\

  308.         VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\

  309.         \

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

  311.         : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride), "m"(ADD), "m"(MUL1)\

  312.         : "memory"\

  313.      );\

  314.      if(h==16){\

  315.         __asm__ volatile(\

  316.             VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\

  317.             VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\

  318.             VOP(%%mm4, %%mm5, %%mm0, %%mm1, %%mm2, %%mm3, OP, MUL2)\

  319.             VOP(%%mm5, %%mm0, %%mm1, %%mm2, %%mm3, %%mm4, OP, MUL2)\

  320.             VOP(%%mm0, %%mm1, %%mm2, %%mm3, %%mm4, %%mm5, OP, MUL2)\

  321.             VOP(%%mm1, %%mm2, %%mm3, %%mm4, %%mm5, %%mm0, OP, MUL2)\

  322.             VOP(%%mm2, %%mm3, %%mm4, %%mm5, %%mm0, %%mm1, OP, MUL2)\

  323.             VOP(%%mm3, %%mm4, %%mm5, %%mm0, %%mm1, %%mm2, OP, MUL2)\

  324.             \

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

  326.            : "S"((x86_reg)srcStride), "r"((x86_reg)dstStride), "m"(ADD),  "m"(MUL1)\

  327.            : "memory"\

  328.         );\

  329.      }\

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

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

  332.    }

  333. 

  334. #define QPEL_CAVS(OPNAME, OP, MMX)\

  335. static void OPNAME ## cavs_qpel8_h_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\

  336.     int h=8;\

  337.     __asm__ volatile(\

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

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

  340.         "1:                         \n\t"\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  367.         "paddw %%mm5, %%mm1         \n\t"\

  368.         "psraw $3, %%mm0            \n\t"\

  369.         "psraw $3, %%mm1            \n\t"\

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

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

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

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

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

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

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

  377.         : "d"((x86_reg)srcStride), "S"((x86_reg)dstStride), "m"(ff_pw_5), "m"(ff_pw_4)\

  378.         : "memory"\

  379.     );\

  380. }\

  381. \

  382. static inline void OPNAME ## cavs_qpel8or16_v1_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h){\

  383.   QPEL_CAVSVNUM(QPEL_CAVSV1,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \

  384. }\

  385. \

  386. static inline void OPNAME ## cavs_qpel8or16_v2_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h){\

  387.   QPEL_CAVSVNUM(QPEL_CAVSV2,OP,ff_pw_4,ff_pw_5,ff_pw_5)         \

  388. }\

  389. \

  390. static inline void OPNAME ## cavs_qpel8or16_v3_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h){\

  391.   QPEL_CAVSVNUM(QPEL_CAVSV3,OP,ff_pw_64,ff_pw_96,ff_pw_42)      \

  392. }\

  393. \

  394. static void OPNAME ## cavs_qpel8_v1_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\

  395.     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 8);\

  396. }\

  397. static void OPNAME ## cavs_qpel16_v1_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\

  398.     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst  , src  , dstStride, srcStride, 16);\

  399.     OPNAME ## cavs_qpel8or16_v1_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\

  400. }\

  401. \

  402. static void OPNAME ## cavs_qpel8_v2_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\

  403.     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 8);\

  404. }\

  405. static void OPNAME ## cavs_qpel16_v2_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\

  406.     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst  , src  , dstStride, srcStride, 16);\

  407.     OPNAME ## cavs_qpel8or16_v2_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\

  408. }\

  409. \

  410. static void OPNAME ## cavs_qpel8_v3_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\

  411.     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 8);\

  412. }\

  413. static void OPNAME ## cavs_qpel16_v3_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\

  414.     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst  , src  , dstStride, srcStride, 16);\

  415.     OPNAME ## cavs_qpel8or16_v3_ ## MMX(dst+8, src+8, dstStride, srcStride, 16);\

  416. }\

  417. \

  418. static void OPNAME ## cavs_qpel16_h_ ## MMX(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride){\

  419.     OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\

  420.     OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  421.     src += 8*srcStride;\

  422.     dst += 8*dstStride;\

  423.     OPNAME ## cavs_qpel8_h_ ## MMX(dst  , src  , dstStride, srcStride);\

  424.     OPNAME ## cavs_qpel8_h_ ## MMX(dst+8, src+8, dstStride, srcStride);\

  425. }\

  426. 

  427. #define CAVS_MC(OPNAME, SIZE, MMX) \

  428. static void OPNAME ## cavs_qpel ## SIZE ## _mc20_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  429. {\

  430.     OPNAME ## cavs_qpel ## SIZE ## _h_ ## MMX(dst, src, stride, stride);\

  431. }\

  432. \

  433. static void OPNAME ## cavs_qpel ## SIZE ## _mc01_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  434. {\

  435.     OPNAME ## cavs_qpel ## SIZE ## _v1_ ## MMX(dst, src, stride, stride);\

  436. }\

  437. \

  438. static void OPNAME ## cavs_qpel ## SIZE ## _mc02_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  439. {\

  440.     OPNAME ## cavs_qpel ## SIZE ## _v2_ ## MMX(dst, src, stride, stride);\

  441. }\

  442. \

  443. static void OPNAME ## cavs_qpel ## SIZE ## _mc03_ ## MMX(uint8_t *dst, const uint8_t *src, ptrdiff_t stride)\

  444. {\

  445.     OPNAME ## cavs_qpel ## SIZE ## _v3_ ## MMX(dst, src, stride, stride);\

  446. }\

  447. 

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

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

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

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

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

  453. #define AVG_MMXEXT_OP(a, b, temp, size) \

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

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

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

  457. 

  458. #endif /* (HAVE_MMXEXT_INLINE || HAVE_AMD3DNOW_INLINE) */

  459. 

  460. #if HAVE_MMX_INLINE

  461. static void put_cavs_qpel8_mc00_mmx(uint8_t *dst, const uint8_t *src,

  462.                                     ptrdiff_t stride)

  463. {

  464.     ff_put_pixels8_mmx(dst, src, stride, 8);

  465. }

  466. 

  467. static void avg_cavs_qpel8_mc00_mmx(uint8_t *dst, const uint8_t *src,

  468.                                     ptrdiff_t stride)

  469. {

  470.     ff_avg_pixels8_mmx(dst, src, stride, 8);

  471. }

  472. 

  473. static void put_cavs_qpel16_mc00_mmx(uint8_t *dst, const uint8_t *src,

  474.                                      ptrdiff_t stride)

  475. {

  476.     ff_put_pixels16_mmx(dst, src, stride, 16);

  477. }

  478. 

  479. static void avg_cavs_qpel16_mc00_mmx(uint8_t *dst, const uint8_t *src,

  480.                                      ptrdiff_t stride)

  481. {

  482.     ff_avg_pixels16_mmx(dst, src, stride, 16);

  483. }

  484. 

  485. static av_cold void cavsdsp_init_mmx(CAVSDSPContext *c,

  486.                                      AVCodecContext *avctx)

  487. {

  488.     c->put_cavs_qpel_pixels_tab[0][0] = put_cavs_qpel16_mc00_mmx;

  489.     c->put_cavs_qpel_pixels_tab[1][0] = put_cavs_qpel8_mc00_mmx;

  490.     c->avg_cavs_qpel_pixels_tab[0][0] = avg_cavs_qpel16_mc00_mmx;

  491.     c->avg_cavs_qpel_pixels_tab[1][0] = avg_cavs_qpel8_mc00_mmx;

  492. 

  493.     c->cavs_idct8_add = cavs_idct8_add_mmx;

  494.     c->idct_perm      = FF_IDCT_PERM_TRANSPOSE;

  495. }

  496. #endif /* HAVE_MMX_INLINE */

  497. 

  498. #define DSPFUNC(PFX, IDX, NUM, EXT)                                                       \

  499.     c->PFX ## _cavs_qpel_pixels_tab[IDX][ 2] = PFX ## _cavs_qpel ## NUM ## _mc20_ ## EXT; \

  500.     c->PFX ## _cavs_qpel_pixels_tab[IDX][ 4] = PFX ## _cavs_qpel ## NUM ## _mc01_ ## EXT; \

  501.     c->PFX ## _cavs_qpel_pixels_tab[IDX][ 8] = PFX ## _cavs_qpel ## NUM ## _mc02_ ## EXT; \

  502.     c->PFX ## _cavs_qpel_pixels_tab[IDX][12] = PFX ## _cavs_qpel ## NUM ## _mc03_ ## EXT; \

  503. 

  504. #if HAVE_MMXEXT_INLINE

  505. QPEL_CAVS(put_,        PUT_OP, mmxext)

  506. QPEL_CAVS(avg_, AVG_MMXEXT_OP, mmxext)

  507. 

  508. CAVS_MC(put_,  8, mmxext)

  509. CAVS_MC(put_, 16, mmxext)

  510. CAVS_MC(avg_,  8, mmxext)

  511. CAVS_MC(avg_, 16, mmxext)

  512. 

  513. static av_cold void cavsdsp_init_mmxext(CAVSDSPContext *c,

  514.                                         AVCodecContext *avctx)

  515. {

  516.     DSPFUNC(put, 0, 16, mmxext);

  517.     DSPFUNC(put, 1,  8, mmxext);

  518.     DSPFUNC(avg, 0, 16, mmxext);

  519.     DSPFUNC(avg, 1,  8, mmxext);

  520. }

  521. #endif /* HAVE_MMXEXT_INLINE */

  522. 

  523. #if HAVE_AMD3DNOW_INLINE

  524. QPEL_CAVS(put_,       PUT_OP, 3dnow)

  525. QPEL_CAVS(avg_, AVG_3DNOW_OP, 3dnow)

  526. 

  527. CAVS_MC(put_, 8, 3dnow)

  528. CAVS_MC(put_, 16,3dnow)

  529. CAVS_MC(avg_, 8, 3dnow)

  530. CAVS_MC(avg_, 16,3dnow)

  531. 

  532. static av_cold void cavsdsp_init_3dnow(CAVSDSPContext *c,

  533.                                        AVCodecContext *avctx)

  534. {

  535.     DSPFUNC(put, 0, 16, 3dnow);

  536.     DSPFUNC(put, 1,  8, 3dnow);

  537.     DSPFUNC(avg, 0, 16, 3dnow);

  538.     DSPFUNC(avg, 1,  8, 3dnow);

  539. }

  540. #endif /* HAVE_AMD3DNOW_INLINE */

  541. 

  542. av_cold void ff_cavsdsp_init_x86(CAVSDSPContext *c, AVCodecContext *avctx)

  543. {

  544. #if HAVE_MMX_INLINE

  545.     int cpu_flags = av_get_cpu_flags();

  546. 

  547.     if (INLINE_MMX(cpu_flags))

  548.         cavsdsp_init_mmx(c, avctx);

  549. #endif /* HAVE_MMX_INLINE */

  550. #if HAVE_AMD3DNOW_INLINE

  551.     if (INLINE_AMD3DNOW(cpu_flags))

  552.         cavsdsp_init_3dnow(c, avctx);

  553. #endif /* HAVE_AMD3DNOW_INLINE */

  554. #if HAVE_MMXEXT_INLINE

  555.     if (INLINE_MMXEXT(cpu_flags))

  556.         cavsdsp_init_mmxext(c, avctx);

  557. #endif /* HAVE_MMXEXT_INLINE */

  558. }