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.
735 Commits
32 Branches
850MB
Tree: 789bee1264
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '789bee1264'
${ noResults }
FFmpeg/libavcodec/i386/motion_est_mmx.c

369 lines
11KB
Raw Blame History 

  1. /*

  2.  * MMX optimized motion estimation

  3.  * Copyright (c) 2001 Gerard Lantau.

  4.  *

  5.  * This program is free software; you can redistribute it and/or modify

  6.  * it under the terms of the GNU General Public License as published by

  7.  * the Free Software Foundation; either version 2 of the License, or

  8.  * (at your option) any later version.

  9.  *

  10.  * This program is distributed in the hope that it will be useful,

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

  12.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

  13.  * GNU General Public License for more details.

  14.  *

  15.  * You should have received a copy of the GNU General Public License

  16.  * along with this program; if not, write to the Free Software

  17.  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.

  18.  *

  19.  * mostly by Michael Niedermayer <michaelni@gmx.at>

  20.  */

  21. #include "../dsputil.h"

  22. #include "../mangle.h"

  23. 

  24. static const __attribute__ ((aligned(8))) UINT64 round_tab[3]={

  25. 0x0000000000000000,

  26. 0x0001000100010001,

  27. 0x0002000200020002,

  28. };

  29. 

  30. static __attribute__ ((aligned(8))) uint64_t bone= 0x0101010101010101LL;

  31. 

  32. static inline void sad8_mmx(UINT8 *blk1, UINT8 *blk2, int stride, int h)

  33. {

  34.     int len= -(stride<<h);

  35.     asm volatile(

  36.         ".balign 16			\n\t"

  37.         "1:				\n\t"

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

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

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

  41.         "addl %3, %%eax			\n\t"

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

  43.         "psubusb %%mm4, %%mm0		\n\t"

  44.         "movq (%1, %%eax), %%mm1	\n\t"

  45.         "movq (%2, %%eax), %%mm3	\n\t"

  46.         "movq (%2, %%eax), %%mm5	\n\t"

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

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

  49.         "por %%mm2, %%mm0		\n\t"

  50.         "por %%mm1, %%mm3		\n\t"

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

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

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

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

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

  56.         "punpckhbw %%mm7, %%mm2		\n\t"

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

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

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

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

  61.         "addl %3, %%eax			\n\t"

  62.         " js 1b				\n\t"

  63.         : "+a" (len)

  64.         : "r" (blk1 - len), "r" (blk2 - len), "r" (stride)

  65.     );

  66. }

  67. 

  68. static inline void sad8_mmx2(UINT8 *blk1, UINT8 *blk2, int stride, int h)

  69. {

  70.     int len= -(stride<<h);

  71.     asm volatile(

  72.         ".balign 16			\n\t"

  73.         "1:				\n\t"

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

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

  76.         "psadbw %%mm2, %%mm0		\n\t"

  77.         "addl %3, %%eax			\n\t"

  78.         "movq (%1, %%eax), %%mm1	\n\t"

  79.         "movq (%2, %%eax), %%mm3	\n\t"

  80.         "psadbw %%mm1, %%mm3		\n\t"

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

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

  83.         "addl %3, %%eax			\n\t"

  84.         " js 1b				\n\t"

  85.         : "+a" (len)

  86.         : "r" (blk1 - len), "r" (blk2 - len), "r" (stride)

  87.     );

  88. }

  89. 

  90. static inline void sad8_2_mmx2(UINT8 *blk1a, UINT8 *blk1b, UINT8 *blk2, int stride, int h)

  91. {

  92.     int len= -(stride<<h);

  93.     asm volatile(

  94.         ".balign 16			\n\t"

  95.         "1:				\n\t"

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

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

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

  99.         "movq (%3, %%eax), %%mm2	\n\t"

  100.         "psadbw %%mm2, %%mm0		\n\t"

  101.         "addl %4, %%eax			\n\t"

  102.         "movq (%1, %%eax), %%mm1	\n\t"

  103.         "movq (%2, %%eax), %%mm3	\n\t"

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

  105.         "movq (%3, %%eax), %%mm1	\n\t"

  106.         "psadbw %%mm1, %%mm3		\n\t"

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

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

  109.         "addl %4, %%eax			\n\t"

  110.         " js 1b				\n\t"

  111.         : "+a" (len)

  112.         : "r" (blk1a - len), "r" (blk1b -len), "r" (blk2 - len), "r" (stride)

  113.     );

  114. }

  115. 

  116. static inline void sad8_4_mmx2(UINT8 *blk1, UINT8 *blk2, int stride, int h)

  117. { //FIXME reuse src

  118.     int len= -(stride<<h);

  119.     asm volatile(

  120.         ".balign 16			\n\t"

  121.         "movq "MANGLE(bone)", %%mm5	\n\t"

  122.         "1:				\n\t" 

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

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

  125.         "movq 1(%1, %%eax), %%mm1	\n\t"

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

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

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

  129.         "psubusb %%mm5, %%mm3		\n\t"

  130.         "pavgb %%mm3, %%mm0		\n\t"

  131.         "movq (%3, %%eax), %%mm2	\n\t"

  132.         "psadbw %%mm2, %%mm0		\n\t"

  133.         "addl %4, %%eax			\n\t"

  134.         "movq (%1, %%eax), %%mm1	\n\t"

  135.         "movq (%2, %%eax), %%mm3	\n\t"

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

  137.         "movq 1(%2, %%eax), %%mm4	\n\t"

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

  139.         "pavgb %%mm4, %%mm2		\n\t"

  140.         "psubusb %%mm5, %%mm2		\n\t"

  141.         "pavgb %%mm1, %%mm2		\n\t"

  142.         "movq (%3, %%eax), %%mm1	\n\t"

  143.         "psadbw %%mm1, %%mm2		\n\t"

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

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

  146.         "addl %4, %%eax			\n\t"

  147.         " js 1b				\n\t"

  148.         : "+a" (len)

  149.         : "r" (blk1 - len), "r" (blk1 - len + stride), "r" (blk2 - len), "r" (stride)

  150.     );

  151. }

  152. 

  153. static inline void sad8_2_mmx(UINT8 *blk1a, UINT8 *blk1b, UINT8 *blk2, int stride, int h)

  154. {

  155.     int len= -(stride<<h);

  156.     asm volatile(

  157.         ".balign 16			\n\t"

  158.         "1:				\n\t"

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

  160.         "movq (%2, %%eax), %%mm1	\n\t"

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

  162.         "movq (%2, %%eax), %%mm3	\n\t"

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

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

  165.         "punpckhbw %%mm7, %%mm2		\n\t"

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

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

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

  169.         "movq (%3, %%eax), %%mm4	\n\t" 

  170.         "movq (%3, %%eax), %%mm2	\n\t"

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

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

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

  174.         "psrlw $1, %%mm3		\n\t"

  175.         "packuswb %%mm3, %%mm1		\n\t"

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

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

  178.         "por %%mm4, %%mm1		\n\t"

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

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

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

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

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

  184.         "addl %4, %%eax			\n\t"

  185.         " js 1b				\n\t"

  186.         : "+a" (len)

  187.         : "r" (blk1a - len), "r" (blk1b -len), "r" (blk2 - len), "r" (stride)

  188.     );

  189. }

  190. 

  191. static inline void sad8_4_mmx(UINT8 *blk1, UINT8 *blk2, int stride, int h)

  192. {

  193.     int len= -(stride<<h);

  194.     asm volatile(

  195.         ".balign 16			\n\t"

  196.         "1:				\n\t"

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

  198.         "movq (%2, %%eax), %%mm1	\n\t"

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

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

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

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

  203.         "punpckhbw %%mm7, %%mm4		\n\t"

  204.         "punpckhbw %%mm7, %%mm2		\n\t"

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

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

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

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

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

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

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

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

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

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

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

  216.         "punpckhbw %%mm7, %%mm4		\n\t"

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

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

  219.         "movq (%3, %%eax), %%mm3	\n\t" 

  220.         "movq (%3, %%eax), %%mm4	\n\t" 

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

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

  223.         "psrlw $2, %%mm2		\n\t"

  224.         "psrlw $2, %%mm1		\n\t"

  225.         "packuswb %%mm1, %%mm2		\n\t"

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

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

  228.         "por %%mm3, %%mm2		\n\t"

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

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

  231.         "punpckhbw %%mm7, %%mm2		\n\t"

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

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

  234.         "addl %4, %%eax			\n\t"

  235.         " js 1b				\n\t"

  236.         : "+a" (len)

  237.         : "r" (blk1 - len), "r" (blk1 -len + stride), "r" (blk2 - len), "r" (stride)

  238.     );

  239. }

  240. 

  241. static inline int sum_mmx()

  242. {

  243.     int ret;

  244.     asm volatile(

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

  246.         "psrlq $32, %%mm6		\n\t"

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

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

  249.         "psrlq $16, %%mm6		\n\t"

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

  251.         "movd %%mm6, %0			\n\t"

  252.         : "=r" (ret)

  253.     );

  254.     return ret&0xFFFF;

  255. }

  256. 

  257. static inline int sum_mmx2()

  258. {

  259.     int ret;

  260.     asm volatile(

  261.         "movd %%mm6, %0			\n\t"

  262.         : "=r" (ret)

  263.     );

  264.     return ret;

  265. }

  266. 

  267. #define PIX_SAD(suf)\

  268. int pix_abs8x8_ ## suf(UINT8 *blk2, UINT8 *blk1, int stride)\

  269. {\

  270.     asm volatile("pxor %%mm7, %%mm7		\n\t"\

  271.                  "pxor %%mm6, %%mm6		\n\t":);\

  272. \

  273.     sad8_ ## suf(blk1, blk2, stride, 3);\

  274. \

  275.     return sum_ ## suf();\

  276. }\

  277. \

  278. int pix_abs8x8_x2_ ## suf(UINT8 *blk2, UINT8 *blk1, int stride)\

  279. {\

  280.     asm volatile("pxor %%mm7, %%mm7		\n\t"\

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

  282.                  "movq %0, %%mm5		\n\t"\

  283.                  :: "m"(round_tab[1]) \

  284.                  );\

  285. \

  286.     sad8_2_ ## suf(blk1, blk2+1, blk2, stride, 3);\

  287. \

  288.     return sum_ ## suf();\

  289. }\

  290. \

  291. int pix_abs8x8_y2_ ## suf(UINT8 *blk2, UINT8 *blk1, int stride)\

  292. {\

  293.     asm volatile("pxor %%mm7, %%mm7		\n\t"\

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

  295.                  "movq %0, %%mm5		\n\t"\

  296.                  :: "m"(round_tab[1]) \

  297.                  );\

  298. \

  299.     sad8_2_ ## suf(blk1, blk1+stride, blk2, stride, 3);\

  300. \

  301.     return sum_ ## suf();\

  302. }\

  303. \

  304. int pix_abs8x8_xy2_ ## suf(UINT8 *blk2, UINT8 *blk1, int stride)\

  305. {\

  306.     asm volatile("pxor %%mm7, %%mm7		\n\t"\

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

  308.                  "movq %0, %%mm5		\n\t"\

  309.                  :: "m"(round_tab[2]) \

  310.                  );\

  311. \

  312.     sad8_4_ ## suf(blk1, blk2, stride, 3);\

  313. \

  314.     return sum_ ## suf();\

  315. }\

  316. \

  317. int pix_abs16x16_ ## suf(UINT8 *blk2, UINT8 *blk1, int stride)\

  318. {\

  319.     asm volatile("pxor %%mm7, %%mm7		\n\t"\

  320.                  "pxor %%mm6, %%mm6		\n\t":);\

  321. \

  322.     sad8_ ## suf(blk1  , blk2  , stride, 4);\

  323.     sad8_ ## suf(blk1+8, blk2+8, stride, 4);\

  324. \

  325.     return sum_ ## suf();\

  326. }\

  327. int pix_abs16x16_x2_ ## suf(UINT8 *blk2, UINT8 *blk1, int stride)\

  328. {\

  329.     asm volatile("pxor %%mm7, %%mm7		\n\t"\

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

  331.                  "movq %0, %%mm5		\n\t"\

  332.                  :: "m"(round_tab[1]) \

  333.                  );\

  334. \

  335.     sad8_2_ ## suf(blk1  , blk1+1, blk2  , stride, 4);\

  336.     sad8_2_ ## suf(blk1+8, blk1+9, blk2+8, stride, 4);\

  337. \

  338.     return sum_ ## suf();\

  339. }\

  340. int pix_abs16x16_y2_ ## suf(UINT8 *blk2, UINT8 *blk1, int stride)\

  341. {\

  342.     asm volatile("pxor %%mm7, %%mm7		\n\t"\

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

  344.                  "movq %0, %%mm5		\n\t"\

  345.                  :: "m"(round_tab[1]) \

  346.                  );\

  347. \

  348.     sad8_2_ ## suf(blk1  , blk1+stride,  blk2  , stride, 4);\

  349.     sad8_2_ ## suf(blk1+8, blk1+stride+8,blk2+8, stride, 4);\

  350. \

  351.     return sum_ ## suf();\

  352. }\

  353. int pix_abs16x16_xy2_ ## suf(UINT8 *blk2, UINT8 *blk1, int stride)\

  354. {\

  355.     asm volatile("pxor %%mm7, %%mm7		\n\t"\

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

  357.                  "movq %0, %%mm5		\n\t"\

  358.                  :: "m"(round_tab[2]) \

  359.                  );\

  360. \

  361.     sad8_4_ ## suf(blk1  , blk2  , stride, 4);\

  362.     sad8_4_ ## suf(blk1+8, blk2+8, stride, 4);\

  363. \

  364.     return sum_ ## suf();\

  365. }\

  366. 

  367. PIX_SAD(mmx)

  368. PIX_SAD(mmx2)