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.
1847 Commits
32 Branches
293MB
Tree: a9badb51bf
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'a9badb51bf'
${ noResults }
FFmpeg/libavcodec/i386/dsputil_mmx_rnd.h

571 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.  *

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

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

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

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

  9.  *

  10.  * This library 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 GNU

  13.  * Lesser General Public License for more details.

  14.  *

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

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

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

  18.  *

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

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

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

  22.  */

  23. 

  24. // put_pixels

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

  26. {

  27.     MOVQ_BFE(mm6);

  28.     __asm __volatile(

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

  30. 	".balign 8			\n\t"

  31. 	"1:				\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  51. 	"jnz	1b			\n\t"

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

  53. 	:"r"(line_size)

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

  55. }

  56. 

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

  58. {

  59.     MOVQ_BFE(mm6);

  60.     __asm __volatile(

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

  62.         " jz 1f				\n\t"

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

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

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

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

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

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

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

  70.         "decl	%0			\n\t"

  71. 	".balign 8			\n\t"

  72. 	"1:				\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  97. 	"jnz	1b			\n\t"

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

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

  100. #else

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

  102. #endif

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

  104. 	:"memory");

  105. }

  106. 

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

  108. {

  109.     MOVQ_BFE(mm6);

  110.     __asm __volatile(

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

  112. 	".balign 8			\n\t"

  113. 	"1:				\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  147. 	"jnz	1b			\n\t"

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

  149. 	:"r"(line_size)

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

  151. }

  152. 

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

  154. {

  155.     MOVQ_BFE(mm6);

  156.     __asm __volatile(

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

  158.         " jz 1f				\n\t"

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

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

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

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

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

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

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

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

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

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

  169. 	"decl	%0			\n\t"

  170. 	".balign 8			\n\t"

  171. 	"1:				\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  192. 	"jnz	1b			\n\t"

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

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

  195. #else

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

  197. #endif

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

  199. 	:"memory"); 

  200. }

  201. 

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

  203. {

  204.     MOVQ_BFE(mm6);

  205.     __asm __volatile(

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

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

  208. 	".balign 8			\n\t"

  209. 	"1:				\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  225. 	"jnz	1b			\n\t"

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

  227. 	:"r"(line_size)

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

  229. }

  230. 

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

  232. {

  233.     MOVQ_ZERO(mm7);

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

  235.     __asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

  248. 	".balign 8      		\n\t"

  249. 	"1:				\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  269. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  289. 

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

  291. 	"jnz	1b			\n\t"

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

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

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

  295. }

  296. 

  297. // avg_pixels

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

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

  300. {

  301.     MOVQ_BFE(mm6);

  302.     JUMPALIGN();

  303.     do {

  304. 	__asm __volatile(

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

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

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

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

  309. 	     :"+m"(*block)

  310. 	     :"m"(*pixels)

  311. 	     :"memory");

  312. 	pixels += line_size;

  313. 	block += line_size;

  314.     }

  315.     while (--h);

  316. }

  317. 

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

  319. {

  320.     MOVQ_BFE(mm6);

  321.     JUMPALIGN();

  322.     do {

  323. 	__asm __volatile(

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

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

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

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

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

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

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

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

  332. 	     :"+m"(*block)

  333. 	     :"m"(*pixels)

  334. 	     :"memory");

  335. 	pixels += line_size;

  336. 	block += line_size;

  337.     }

  338.     while (--h);

  339. }

  340. 

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

  342. {

  343.     MOVQ_BFE(mm6);

  344.     JUMPALIGN();

  345.     do {

  346. 	__asm __volatile(

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

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

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

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

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

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

  353. 	    :"+m"(*block)

  354. 	    :"m"(*pixels)

  355. 	    :"memory");

  356. 	pixels += line_size;

  357. 	block += line_size;

  358.     } while (--h);

  359. }

  360. 

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

  362. {

  363.     MOVQ_BFE(mm6);

  364.     JUMPALIGN();

  365.     do {

  366. 	__asm __volatile(

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

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

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

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

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

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

  373. 	    :"+m"(*dst)

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

  375. 	    :"memory");

  376. 	dst += dstStride;

  377.         src1 += src1Stride;

  378.         src2 += 8;

  379.     } while (--h);

  380. }

  381. 

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

  383. {

  384.     MOVQ_BFE(mm6);

  385.     JUMPALIGN();

  386.     do {

  387. 	__asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

  400. 	    :"+m"(*block)

  401. 	    :"m"(*pixels)

  402. 	    :"memory");

  403. 	pixels += line_size;

  404. 	block += line_size;

  405.     } while (--h);

  406. }

  407. 

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

  409. {

  410.     MOVQ_BFE(mm6);

  411.     JUMPALIGN();

  412.     do {

  413. 	__asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

  426. 	    :"+m"(*dst)

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

  428. 	    :"memory");

  429. 	dst += dstStride;

  430.         src1 += src1Stride;

  431.         src2 += 16;

  432.     } while (--h);

  433. }

  434. 

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

  436. {

  437.     MOVQ_BFE(mm6);

  438.     __asm __volatile(

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

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

  441. 	".balign 8			\n\t"

  442. 	"1:				\n\t"

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

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

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

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

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

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

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

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

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

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

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

  454. 

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

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

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

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

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

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

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

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

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

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

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

  466. 

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

  468. 	"jnz	1b			\n\t"

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

  470. 	:"r"(line_size)

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

  472. }

  473. 

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

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

  476. {

  477.     MOVQ_ZERO(mm7);

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

  479.     __asm __volatile(

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

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

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

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

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

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

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

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

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

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

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

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

  492. 	".balign 8			\n\t"

  493. 	"1:				\n\t"

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  517. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  541. 

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

  543. 	"jnz	1b			\n\t"

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

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

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

  547. }

  548. 

  549. //FIXME optimize

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

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

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

  553. }

  554. 

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

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

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

  558. }

  559. 

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

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

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

  563. }

  564. 

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

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

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

  568. }

  569.