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.
3135 Commits
32 Branches
837MB
Tree: 68951ecf0c
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '68951ecf0c'
${ noResults }
FFmpeg/libavcodec/i386/dsputil_mmx_rnd.h

572 lines
16KB
Raw Blame History 

  1. /*

  2.  * DSP utils mmx functions are compiled twice for rnd/no_rnd

  3.  * Copyright (c) 2000, 2001 Fabrice Bellard.

  4.  * Copyright (c) 2003-2004 Michael Niedermayer <michaelni@gmx.at>

  5.  *

  6.  * This library is free software; you can redistribute it and/or

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

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

  9.  * version 2 of the License, or (at your option) any later version.

  10.  *

  11.  * This library is distributed in the hope that it will be useful,

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

  17.  * License along with this library; if not, write to the Free Software

  18.  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA

  19.  *

  20.  * MMX optimization by Nick Kurshev <nickols_k@mail.ru>

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

  22.  * and improved by Zdenek Kabelac <kabi@users.sf.net>

  23.  */

  24. 

  25. // put_pixels

  26. static void DEF(put, pixels8_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  27. {

  28.     MOVQ_BFE(mm6);

  29.     __asm __volatile(

  30. 	"lea	(%3, %3), %%eax		\n\t"

  31. 	".balign 8			\n\t"

  32. 	"1:				\n\t"

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

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

  35. 	"movq	(%1, %3), %%mm2		\n\t"

  36. 	"movq	1(%1, %3), %%mm3	\n\t"

  37. 	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)

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

  39. 	"movq	%%mm5, (%2, %3)		\n\t"

  40. 	"addl	%%eax, %1		\n\t"

  41. 	"addl	%%eax, %2		\n\t"

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

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

  44. 	"movq	(%1, %3), %%mm2		\n\t"

  45. 	"movq	1(%1, %3), %%mm3	\n\t"

  46. 	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)

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

  48. 	"movq	%%mm5, (%2, %3)		\n\t"

  49. 	"addl	%%eax, %1		\n\t"

  50. 	"addl	%%eax, %2		\n\t"

  51. 	"subl	$4, %0			\n\t"

  52. 	"jnz	1b			\n\t"

  53. 	:"+g"(h), "+S"(pixels), "+D"(block)

  54. 	:"r"(line_size)

  55. 	:"eax", "memory");

  56. }

  57. 

  58. static void DEF(put, pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)

  59. {

  60.     MOVQ_BFE(mm6);

  61.     __asm __volatile(

  62. 	"testl $1, %0			\n\t"

  63.         " jz 1f				\n\t"

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

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

  66. 	"addl	%4, %1			\n\t"

  67.         "addl	$8, %2			\n\t"

  68. 	PAVGB(%%mm0, %%mm1, %%mm4, %%mm6)

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

  70. 	"addl	%5, %3			\n\t"

  71.         "decl	%0			\n\t"

  72. 	".balign 8			\n\t"

  73. 	"1:				\n\t"

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

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

  76. 	"addl	%4, %1			\n\t"

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

  78. 	"movq	8(%2), %%mm3		\n\t"

  79. 	"addl	%4, %1			\n\t"

  80. 	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)

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

  82. 	"addl	%5, %3			\n\t"

  83. 	"movq	%%mm5, (%3)		\n\t"

  84. 	"addl	%5, %3			\n\t"

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

  86. 	"movq	16(%2), %%mm1		\n\t"

  87. 	"addl	%4, %1			\n\t"

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

  89. 	"movq	24(%2), %%mm3		\n\t"

  90. 	"addl	%4, %1			\n\t"

  91. 	"addl	$32, %2			\n\t"

  92. 	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)

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

  94. 	"addl	%5, %3			\n\t"

  95. 	"movq	%%mm5, (%3)		\n\t"

  96. 	"addl	%5, %3			\n\t"

  97. 	"subl	$4, %0			\n\t"

  98. 	"jnz	1b			\n\t"

  99. #ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cant be used

  100.         :"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)

  101. #else

  102.         :"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)

  103. #endif

  104. 	:"S"(src1Stride), "D"(dstStride)

  105. 	:"memory");

  106. }

  107. 

  108. static void DEF(put, pixels16_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  109. {

  110.     MOVQ_BFE(mm6);

  111.     __asm __volatile(

  112. 	"lea	(%3, %3), %%eax		\n\t"

  113. 	".balign 8			\n\t"

  114. 	"1:				\n\t"

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

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

  117. 	"movq	(%1, %3), %%mm2		\n\t"

  118. 	"movq	1(%1, %3), %%mm3	\n\t"

  119. 	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)

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

  121. 	"movq	%%mm5, (%2, %3)		\n\t"

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

  123. 	"movq	9(%1), %%mm1		\n\t"

  124. 	"movq	8(%1, %3), %%mm2	\n\t"

  125. 	"movq	9(%1, %3), %%mm3	\n\t"

  126. 	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)

  127. 	"movq	%%mm4, 8(%2)		\n\t"

  128. 	"movq	%%mm5, 8(%2, %3)	\n\t"

  129. 	"addl	%%eax, %1		\n\t"

  130. 	"addl	%%eax, %2		\n\t"

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

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

  133. 	"movq	(%1, %3), %%mm2		\n\t"

  134. 	"movq	1(%1, %3), %%mm3	\n\t"

  135. 	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)

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

  137. 	"movq	%%mm5, (%2, %3)		\n\t"

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

  139. 	"movq	9(%1), %%mm1		\n\t"

  140. 	"movq	8(%1, %3), %%mm2	\n\t"

  141. 	"movq	9(%1, %3), %%mm3	\n\t"

  142. 	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)

  143. 	"movq	%%mm4, 8(%2)		\n\t"

  144. 	"movq	%%mm5, 8(%2, %3)	\n\t"

  145. 	"addl	%%eax, %1		\n\t"

  146. 	"addl	%%eax, %2		\n\t"

  147. 	"subl	$4, %0			\n\t"

  148. 	"jnz	1b			\n\t"

  149. 	:"+g"(h), "+S"(pixels), "+D"(block)

  150. 	:"r"(line_size)

  151. 	:"eax", "memory");

  152. }

  153. 

  154. static void DEF(put, pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)

  155. {

  156.     MOVQ_BFE(mm6);

  157.     __asm __volatile(

  158. 	"testl $1, %0			\n\t"

  159.         " jz 1f				\n\t"

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

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

  162. 	"movq	8(%1), %%mm2		\n\t"

  163. 	"movq	8(%2), %%mm3		\n\t"

  164. 	"addl	%4, %1			\n\t"

  165. 	"addl	$16, %2			\n\t"

  166. 	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)

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

  168. 	"movq	%%mm5, 8(%3)		\n\t"

  169. 	"addl	%5, %3			\n\t"

  170. 	"decl	%0			\n\t"

  171. 	".balign 8			\n\t"

  172. 	"1:				\n\t"

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

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

  175. 	"movq	8(%1), %%mm2		\n\t"

  176. 	"movq	8(%2), %%mm3		\n\t"

  177. 	"addl	%4, %1			\n\t"

  178. 	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)

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

  180. 	"movq	%%mm5, 8(%3)		\n\t"

  181. 	"addl	%5, %3			\n\t"

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

  183. 	"movq	16(%2), %%mm1		\n\t"

  184. 	"movq	8(%1), %%mm2		\n\t"

  185. 	"movq	24(%2), %%mm3		\n\t"

  186. 	"addl	%4, %1			\n\t"

  187. 	PAVGBP(%%mm0, %%mm1, %%mm4,   %%mm2, %%mm3, %%mm5)

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

  189. 	"movq	%%mm5, 8(%3)		\n\t"

  190. 	"addl	%5, %3			\n\t"

  191. 	"addl	$32, %2			\n\t"

  192. 	"subl	$2, %0			\n\t"

  193. 	"jnz	1b			\n\t"

  194. #ifdef PIC //Note "+bm" and "+mb" are buggy too (with gcc 3.2.2 at least) and cant be used

  195. 	:"+m"(h), "+a"(src1), "+c"(src2), "+d"(dst)

  196. #else

  197. 	:"+b"(h), "+a"(src1), "+c"(src2), "+d"(dst)

  198. #endif

  199. 	:"S"(src1Stride), "D"(dstStride)

  200. 	:"memory"); 

  201. }

  202. 

  203. static void DEF(put, pixels8_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  204. {

  205.     MOVQ_BFE(mm6);

  206.     __asm __volatile(

  207. 	"lea (%3, %3), %%eax		\n\t"

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

  209. 	".balign 8			\n\t"

  210. 	"1:				\n\t"

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

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

  213. 	PAVGBP(%%mm1, %%mm0, %%mm4,   %%mm2, %%mm1, %%mm5)

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

  215. 	"movq	%%mm5, (%2, %3)		\n\t"

  216. 	"addl	%%eax, %1		\n\t"

  217. 	"addl	%%eax, %2		\n\t"

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

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

  220. 	PAVGBP(%%mm1, %%mm2, %%mm4,   %%mm0, %%mm1, %%mm5)

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

  222. 	"movq	%%mm5, (%2, %3)		\n\t"

  223. 	"addl	%%eax, %1		\n\t"

  224. 	"addl	%%eax, %2		\n\t"

  225. 	"subl	$4, %0			\n\t"

  226. 	"jnz	1b			\n\t"

  227. 	:"+g"(h), "+S"(pixels), "+D"(block)

  228. 	:"r"(line_size)

  229. 	:"eax", "memory");

  230. }

  231. 

  232. static void DEF(put, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  233. {

  234.     MOVQ_ZERO(mm7);

  235.     SET_RND(mm6); // =2 for rnd  and  =1 for no_rnd version

  236.     __asm __volatile(

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

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

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

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

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

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

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

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

  245. 	"paddusw %%mm0, %%mm4		\n\t"

  246. 	"paddusw %%mm1, %%mm5		\n\t"

  247. 	"xorl	%%eax, %%eax		\n\t"

  248. 	"addl	%3, %1			\n\t"

  249. 	".balign 8      		\n\t"

  250. 	"1:				\n\t"

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

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

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

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

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

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

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

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

  259. 	"paddusw %%mm2, %%mm0	 	\n\t"

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

  261. 	"paddusw %%mm6, %%mm4		\n\t"

  262. 	"paddusw %%mm6, %%mm5		\n\t"

  263. 	"paddusw %%mm0, %%mm4		\n\t"

  264. 	"paddusw %%mm1, %%mm5		\n\t"

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

  266. 	"psrlw	$2, %%mm5		\n\t"

  267. 	"packuswb  %%mm5, %%mm4		\n\t"

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

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

  270. 

  271. 	"movq	(%1, %%eax), %%mm2	\n\t" // 0 <-> 2   1 <-> 3

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

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

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

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

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

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

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

  279. 	"paddusw %%mm2, %%mm4	 	\n\t"

  280. 	"paddusw %%mm3, %%mm5		\n\t"

  281. 	"paddusw %%mm6, %%mm0		\n\t"

  282. 	"paddusw %%mm6, %%mm1		\n\t"

  283. 	"paddusw %%mm4, %%mm0		\n\t"

  284. 	"paddusw %%mm5, %%mm1		\n\t"

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

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

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

  288. 	"movq	%%mm0, (%2, %%eax)	\n\t"

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

  290. 

  291. 	"subl	$2, %0			\n\t"

  292. 	"jnz	1b			\n\t"

  293. 	:"+g"(h), "+S"(pixels)

  294. 	:"D"(block), "r"(line_size)

  295. 	:"eax", "memory");

  296. }

  297. 

  298. // avg_pixels

  299. // in case more speed is needed - unroling would certainly help

  300. static void DEF(avg, pixels8)(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  301. {

  302.     MOVQ_BFE(mm6);

  303.     JUMPALIGN();

  304.     do {

  305. 	__asm __volatile(

  306. 	     "movq  %0, %%mm0		\n\t"

  307. 	     "movq  %1, %%mm1		\n\t"

  308. 	     PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)

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

  310. 	     :"+m"(*block)

  311. 	     :"m"(*pixels)

  312. 	     :"memory");

  313. 	pixels += line_size;

  314. 	block += line_size;

  315.     }

  316.     while (--h);

  317. }

  318. 

  319. static void DEF(avg, pixels16)(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  320. {

  321.     MOVQ_BFE(mm6);

  322.     JUMPALIGN();

  323.     do {

  324. 	__asm __volatile(

  325. 	     "movq  %0, %%mm0		\n\t"

  326. 	     "movq  %1, %%mm1		\n\t"

  327. 	     PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)

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

  329. 	     "movq  8%0, %%mm0		\n\t"

  330. 	     "movq  8%1, %%mm1		\n\t"

  331. 	     PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)

  332. 	     "movq  %%mm2, 8%0		\n\t"

  333. 	     :"+m"(*block)

  334. 	     :"m"(*pixels)

  335. 	     :"memory");

  336. 	pixels += line_size;

  337. 	block += line_size;

  338.     }

  339.     while (--h);

  340. }

  341. 

  342. static void DEF(avg, pixels8_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  343. {

  344.     MOVQ_BFE(mm6);

  345.     JUMPALIGN();

  346.     do {

  347. 	__asm __volatile(

  348. 	    "movq  %1, %%mm0		\n\t"

  349. 	    "movq  1%1, %%mm1		\n\t"

  350. 	    "movq  %0, %%mm3		\n\t"

  351. 	    PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)

  352. 	    PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)

  353. 	    "movq  %%mm0, %0		\n\t"

  354. 	    :"+m"(*block)

  355. 	    :"m"(*pixels)

  356. 	    :"memory");

  357. 	pixels += line_size;

  358. 	block += line_size;

  359.     } while (--h);

  360. }

  361. 

  362. static void DEF(avg, pixels8_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)

  363. {

  364.     MOVQ_BFE(mm6);

  365.     JUMPALIGN();

  366.     do {

  367. 	__asm __volatile(

  368. 	    "movq  %1, %%mm0		\n\t"

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

  370. 	    "movq  %0, %%mm3		\n\t"

  371. 	    PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)

  372. 	    PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)

  373. 	    "movq  %%mm0, %0		\n\t"

  374. 	    :"+m"(*dst)

  375. 	    :"m"(*src1), "m"(*src2)

  376. 	    :"memory");

  377. 	dst += dstStride;

  378.         src1 += src1Stride;

  379.         src2 += 8;

  380.     } while (--h);

  381. }

  382. 

  383. static void DEF(avg, pixels16_x2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  384. {

  385.     MOVQ_BFE(mm6);

  386.     JUMPALIGN();

  387.     do {

  388. 	__asm __volatile(

  389. 	    "movq  %1, %%mm0		\n\t"

  390. 	    "movq  1%1, %%mm1		\n\t"

  391. 	    "movq  %0, %%mm3		\n\t"

  392. 	    PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)

  393. 	    PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)

  394. 	    "movq  %%mm0, %0		\n\t"

  395. 	    "movq  8%1, %%mm0		\n\t"

  396. 	    "movq  9%1, %%mm1		\n\t"

  397. 	    "movq  8%0, %%mm3		\n\t"

  398. 	    PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)

  399. 	    PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)

  400. 	    "movq  %%mm0, 8%0		\n\t"

  401. 	    :"+m"(*block)

  402. 	    :"m"(*pixels)

  403. 	    :"memory");

  404. 	pixels += line_size;

  405. 	block += line_size;

  406.     } while (--h);

  407. }

  408. 

  409. static void DEF(avg, pixels16_l2)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int dstStride, int src1Stride, int h)

  410. {

  411.     MOVQ_BFE(mm6);

  412.     JUMPALIGN();

  413.     do {

  414. 	__asm __volatile(

  415. 	    "movq  %1, %%mm0		\n\t"

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

  417. 	    "movq  %0, %%mm3		\n\t"

  418. 	    PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)

  419. 	    PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)

  420. 	    "movq  %%mm0, %0		\n\t"

  421. 	    "movq  8%1, %%mm0		\n\t"

  422. 	    "movq  8%2, %%mm1		\n\t"

  423. 	    "movq  8%0, %%mm3		\n\t"

  424. 	    PAVGB(%%mm0, %%mm1, %%mm2, %%mm6)

  425. 	    PAVGB(%%mm3, %%mm2, %%mm0, %%mm6)

  426. 	    "movq  %%mm0, 8%0		\n\t"

  427. 	    :"+m"(*dst)

  428. 	    :"m"(*src1), "m"(*src2)

  429. 	    :"memory");

  430. 	dst += dstStride;

  431.         src1 += src1Stride;

  432.         src2 += 16;

  433.     } while (--h);

  434. }

  435. 

  436. static void DEF(avg, pixels8_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  437. {

  438.     MOVQ_BFE(mm6);

  439.     __asm __volatile(

  440. 	"lea	(%3, %3), %%eax		\n\t"

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

  442. 	".balign 8			\n\t"

  443. 	"1:				\n\t"

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

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

  446. 	PAVGBP(%%mm1, %%mm0, %%mm4,   %%mm2, %%mm1, %%mm5)

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

  448. 	PAVGB(%%mm3, %%mm4, %%mm0, %%mm6)

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

  450. 	PAVGB(%%mm3, %%mm5, %%mm1, %%mm6)

  451. 	"movq	%%mm0, (%2)		\n\t"

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

  453. 	"addl	%%eax, %1		\n\t"

  454. 	"addl	%%eax, %2		\n\t"

  455. 

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

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

  458. 	PAVGBP(%%mm1, %%mm2, %%mm4,   %%mm0, %%mm1, %%mm5)

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

  460. 	PAVGB(%%mm3, %%mm4, %%mm2, %%mm6)

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

  462. 	PAVGB(%%mm3, %%mm5, %%mm1, %%mm6)

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

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

  465. 	"addl	%%eax, %1		\n\t"

  466. 	"addl	%%eax, %2		\n\t"

  467. 

  468. 	"subl	$4, %0			\n\t"

  469. 	"jnz	1b			\n\t"

  470. 	:"+g"(h), "+S"(pixels), "+D"(block)

  471. 	:"r"(line_size)

  472. 	:"eax", "memory");

  473. }

  474. 

  475. // this routine is 'slightly' suboptimal but mostly unused

  476. static void DEF(avg, pixels8_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h)

  477. {

  478.     MOVQ_ZERO(mm7);

  479.     SET_RND(mm6); // =2 for rnd  and  =1 for no_rnd version

  480.     __asm __volatile(

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

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

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

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

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

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

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

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

  489. 	"paddusw %%mm0, %%mm4		\n\t"

  490. 	"paddusw %%mm1, %%mm5		\n\t"

  491. 	"xorl	%%eax, %%eax		\n\t"

  492. 	"addl	%3, %1			\n\t"

  493. 	".balign 8			\n\t"

  494. 	"1:				\n\t"

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

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

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

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

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

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

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

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

  503. 	"paddusw %%mm2, %%mm0	 	\n\t"

  504. 	"paddusw %%mm3, %%mm1		\n\t"

  505. 	"paddusw %%mm6, %%mm4		\n\t"

  506. 	"paddusw %%mm6, %%mm5		\n\t"

  507. 	"paddusw %%mm0, %%mm4		\n\t"

  508. 	"paddusw %%mm1, %%mm5		\n\t"

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

  510. 	"psrlw	$2, %%mm5		\n\t"

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

  512. 	"packuswb  %%mm5, %%mm4		\n\t"

  513. 		"pcmpeqd %%mm2, %%mm2	\n\t"

  514. 		"paddb %%mm2, %%mm2	\n\t"

  515. 		PAVGB(%%mm3, %%mm4, %%mm5, %%mm2)

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

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

  518. 

  519. 	"movq	(%1, %%eax), %%mm2	\n\t" // 0 <-> 2   1 <-> 3

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

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

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

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

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

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

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

  527. 	"paddusw %%mm2, %%mm4	 	\n\t"

  528. 	"paddusw %%mm3, %%mm5		\n\t"

  529. 	"paddusw %%mm6, %%mm0		\n\t"

  530. 	"paddusw %%mm6, %%mm1		\n\t"

  531. 	"paddusw %%mm4, %%mm0		\n\t"

  532. 	"paddusw %%mm5, %%mm1		\n\t"

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

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

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

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

  537. 		"pcmpeqd %%mm2, %%mm2	\n\t"

  538. 		"paddb %%mm2, %%mm2	\n\t"

  539. 		PAVGB(%%mm3, %%mm0, %%mm1, %%mm2)

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

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

  542. 

  543. 	"subl	$2, %0			\n\t"

  544. 	"jnz	1b			\n\t"

  545. 	:"+g"(h), "+S"(pixels)

  546. 	:"D"(block), "r"(line_size)

  547. 	:"eax", "memory");

  548. }

  549. 

  550. //FIXME optimize

  551. static void DEF(put, pixels16_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){

  552.     DEF(put, pixels8_y2)(block  , pixels  , line_size, h);

  553.     DEF(put, pixels8_y2)(block+8, pixels+8, line_size, h);

  554. }

  555. 

  556. static void DEF(put, pixels16_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){

  557.     DEF(put, pixels8_xy2)(block  , pixels  , line_size, h);

  558.     DEF(put, pixels8_xy2)(block+8, pixels+8, line_size, h);

  559. }

  560. 

  561. static void DEF(avg, pixels16_y2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){

  562.     DEF(avg, pixels8_y2)(block  , pixels  , line_size, h);

  563.     DEF(avg, pixels8_y2)(block+8, pixels+8, line_size, h);

  564. }

  565. 

  566. static void DEF(avg, pixels16_xy2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){

  567.     DEF(avg, pixels8_xy2)(block  , pixels  , line_size, h);

  568.     DEF(avg, pixels8_xy2)(block+8, pixels+8, line_size, h);

  569. }

  570.