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.
40948 Commits
32 Branches
1.0GB
Tree: bc75b64cff
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'bc75b64cff'
${ noResults }
FFmpeg/libavcodec/x86/cavsdsp.c

561 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.         DECLARE_ALIGNED(8, uint64_t, tmp);

  146. 

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

  148. 

  149.         __asm__ volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  170.             : "=m"(tmp)

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

  172.             : "memory"

  173.         );

  174.     }

  175. 

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

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

  178. 

  179.         __asm__ volatile(

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  197.             : "memory"

  198.         );

  199.     }

  200. 

  201.     ff_add_pixels_clamped_mmx(b2, dst, stride);

  202. }

  203. 

  204. #endif /* HAVE_MMX_INLINE */

  205. 

  206. #if (HAVE_MMXEXT_INLINE || HAVE_AMD3DNOW_INLINE)

  207. 

  208. /*****************************************************************************

  209.  *

  210.  * motion compensation

  211.  *

  212.  ****************************************************************************/

  213. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  238. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  254. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  279. 

  280. 

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

  282.     int w= 2;\

  283.     src -= 2*srcStride;\

  284.     \

  285.     while(w--){\

  286.       __asm__ volatile(\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  311.         \

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

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

  314.         : "memory"\

  315.      );\

  316.      if(h==16){\

  317.         __asm__ volatile(\

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

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

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

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

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

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

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

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

  326.             \

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

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

  329.            : "memory"\

  330.         );\

  331.      }\

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

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

  334.    }

  335. 

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

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

  338.     int h=8;\

  339.     __asm__ volatile(\

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

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

  342.         "1:                         \n\t"\

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  380.         : "memory"\

  381.     );\

  382. }\

  383. \

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

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

  386. }\

  387. \

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

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

  390. }\

  391. \

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

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

  394. }\

  395. \

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

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

  398. }\

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

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

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

  402. }\

  403. \

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

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

  406. }\

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

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

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

  410. }\

  411. \

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

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

  414. }\

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

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

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

  418. }\

  419. \

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

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

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

  423.     src += 8*srcStride;\

  424.     dst += 8*dstStride;\

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

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

  427. }\

  428. 

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

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

  431. {\

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

  433. }\

  434. \

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

  436. {\

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

  438. }\

  439. \

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

  441. {\

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

  443. }\

  444. \

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

  446. {\

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

  448. }\

  449. 

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

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

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

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

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

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

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

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

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

  459. 

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

  461. 

  462. #if HAVE_MMX_INLINE

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

  464.                                     ptrdiff_t stride)

  465. {

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

  467. }

  468. 

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

  470.                                     ptrdiff_t stride)

  471. {

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

  473. }

  474. 

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

  476.                                      ptrdiff_t stride)

  477. {

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

  479. }

  480. 

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

  482.                                      ptrdiff_t stride)

  483. {

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

  485. }

  486. 

  487. static av_cold void cavsdsp_init_mmx(CAVSDSPContext *c,

  488.                                      AVCodecContext *avctx)

  489. {

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

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

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

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

  494. 

  495.     c->cavs_idct8_add = cavs_idct8_add_mmx;

  496.     c->idct_perm      = FF_IDCT_PERM_TRANSPOSE;

  497. }

  498. #endif /* HAVE_MMX_INLINE */

  499. 

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

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

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

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

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

  505. 

  506. #if HAVE_MMXEXT_INLINE

  507. QPEL_CAVS(put_,        PUT_OP, mmxext)

  508. QPEL_CAVS(avg_, AVG_MMXEXT_OP, mmxext)

  509. 

  510. CAVS_MC(put_,  8, mmxext)

  511. CAVS_MC(put_, 16, mmxext)

  512. CAVS_MC(avg_,  8, mmxext)

  513. CAVS_MC(avg_, 16, mmxext)

  514. 

  515. static av_cold void cavsdsp_init_mmxext(CAVSDSPContext *c,

  516.                                         AVCodecContext *avctx)

  517. {

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

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

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

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

  522. }

  523. #endif /* HAVE_MMXEXT_INLINE */

  524. 

  525. #if HAVE_AMD3DNOW_INLINE

  526. QPEL_CAVS(put_,       PUT_OP, 3dnow)

  527. QPEL_CAVS(avg_, AVG_3DNOW_OP, 3dnow)

  528. 

  529. CAVS_MC(put_, 8, 3dnow)

  530. CAVS_MC(put_, 16,3dnow)

  531. CAVS_MC(avg_, 8, 3dnow)

  532. CAVS_MC(avg_, 16,3dnow)

  533. 

  534. static av_cold void cavsdsp_init_3dnow(CAVSDSPContext *c,

  535.                                        AVCodecContext *avctx)

  536. {

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

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

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

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

  541. }

  542. #endif /* HAVE_AMD3DNOW_INLINE */

  543. 

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

  545. {

  546. #if HAVE_MMX_INLINE

  547.     int cpu_flags = av_get_cpu_flags();

  548. 

  549.     if (INLINE_MMX(cpu_flags))

  550.         cavsdsp_init_mmx(c, avctx);

  551. #endif /* HAVE_MMX_INLINE */

  552. #if HAVE_AMD3DNOW_INLINE

  553.     if (INLINE_AMD3DNOW(cpu_flags))

  554.         cavsdsp_init_3dnow(c, avctx);

  555. #endif /* HAVE_AMD3DNOW_INLINE */

  556. #if HAVE_MMXEXT_INLINE

  557.     if (INLINE_MMXEXT(cpu_flags))

  558.         cavsdsp_init_mmxext(c, avctx);

  559. #endif /* HAVE_MMXEXT_INLINE */

  560. }