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.
1375 Commits
32 Branches
580MB
Tree: d7425f59d5
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'd7425f59d5'
${ noResults }
FFmpeg/libavcodec/i386/dsputil_mmx_avg.h

368 lines
11KB
Raw Blame History 

  1. /*

  2.  * DSP utils : average functions are compiled twice for 3dnow/mmx2

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

  4.  * Copyright (c) 2002 Michael Niedermayer

  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. /* XXX: we use explicit registers to avoid a gcc 2.95.2 register asm

  26.    clobber bug - now it will work with 2.95.2 and also with -fPIC

  27.  */

  28. static void DEF(put_pixels8_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  29. {

  30.     __asm __volatile(

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

  32. 	"1:				\n\t"

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

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

  35. 	PAVGB" 1(%1), %%mm0		\n\t"

  36. 	PAVGB" 1(%1, %3), %%mm1		\n\t"

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

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

  43. 	PAVGB" 1(%1), %%mm0		\n\t"

  44. 	PAVGB" 1(%1, %3), %%mm1		\n\t"

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

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

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

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

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

  50. 	"jnz 1b				\n\t"

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

  52. 	:"r" (line_size)

  53. 	:"%eax", "memory");

  54. }

  55. 

  56. static void DEF(put_pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  57. {

  58.     __asm __volatile(

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

  60. 	"1:				\n\t"

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

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

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

  64. 	"movq 8(%1, %3), %%mm3		\n\t"

  65. 	PAVGB" 1(%1), %%mm0		\n\t"

  66. 	PAVGB" 1(%1, %3), %%mm1		\n\t"

  67. 	PAVGB" 9(%1), %%mm2		\n\t"

  68. 	PAVGB" 9(%1, %3), %%mm3		\n\t"

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

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

  71. 	"movq %%mm2, 8(%2)		\n\t"

  72. 	"movq %%mm3, 8(%2, %3)		\n\t"

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

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

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

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

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

  78. 	"movq 8(%1, %3), %%mm3		\n\t"

  79. 	PAVGB" 1(%1), %%mm0		\n\t"

  80. 	PAVGB" 1(%1, %3), %%mm1		\n\t"

  81. 	PAVGB" 9(%1), %%mm2		\n\t"

  82. 	PAVGB" 9(%1, %3), %%mm3		\n\t"

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

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

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

  86. 	"movq %%mm2, 8(%2)		\n\t"

  87. 	"movq %%mm3, 8(%2, %3)		\n\t"

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

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

  90. 	"jnz 1b				\n\t"

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

  92. 	:"r" (line_size)

  93. 	:"%eax", "memory");

  94. }

  95.  

  96. /* GL: this function does incorrect rounding if overflow */

  97. static void DEF(put_no_rnd_pixels8_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  98. {

  99.     MOVQ_BONE(mm6);

  100.     __asm __volatile(

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

  102. 	"1:				\n\t"

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

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

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

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

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

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

  109. 	"psubusb %%mm6, %%mm2		\n\t"

  110. 	PAVGB" %%mm1, %%mm0		\n\t"

  111. 	PAVGB" %%mm3, %%mm2		\n\t"

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

  113. 	"movq %%mm2, (%2, %3)		\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. 	"addl %%eax, %2			\n\t"

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

  120. 	"psubusb %%mm6, %%mm0		\n\t"

  121. 	"psubusb %%mm6, %%mm2		\n\t"

  122. 	PAVGB" %%mm1, %%mm0		\n\t"

  123. 	PAVGB" %%mm3, %%mm2		\n\t"

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

  125. 	"movq %%mm2, (%2, %3)		\n\t"

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

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

  128. 	"jnz 1b				\n\t"

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

  130. 	:"r" (line_size)

  131. 	:"%eax", "memory");

  132. }

  133. 

  134. static void DEF(put_pixels8_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  135. {

  136.     __asm __volatile(

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

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

  139. 	"subl %3, %2			\n\t"

  140. 	"1:				\n\t"

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

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

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

  144. 	PAVGB" %%mm1, %%mm0		\n\t"

  145. 	PAVGB" %%mm2, %%mm1		\n\t"

  146. 	"movq %%mm0, (%2, %3)		\n\t"

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

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

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

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

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

  152. 	PAVGB" %%mm1, %%mm2		\n\t"

  153. 	PAVGB" %%mm0, %%mm1		\n\t"

  154. 	"movq %%mm2, (%2, %3)		\n\t"

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

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

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

  158. 	"jnz 1b				\n\t"

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

  160. 	:"r" (line_size)

  161. 	:"%eax", "memory");

  162. }

  163. 

  164. /* GL: this function does incorrect rounding if overflow */

  165. static void DEF(put_no_rnd_pixels8_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  166. {

  167.     MOVQ_BONE(mm6);

  168.     __asm __volatile(

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

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

  171. 	"subl %3, %2			\n\t"

  172. 	"1:				\n\t"

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

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

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

  176. 	"psubusb %%mm6, %%mm1		\n\t"

  177. 	PAVGB" %%mm1, %%mm0		\n\t"

  178. 	PAVGB" %%mm2, %%mm1		\n\t"

  179. 	"movq %%mm0, (%2, %3)		\n\t"

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

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

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

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

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

  185. 	"psubusb %%mm6, %%mm1		\n\t"

  186. 	PAVGB" %%mm1, %%mm2		\n\t"

  187. 	PAVGB" %%mm0, %%mm1		\n\t"

  188. 	"movq %%mm2, (%2, %3)		\n\t"

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

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

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

  192. 	"jnz 1b				\n\t"

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

  194. 	:"r" (line_size)

  195. 	:"%eax", "memory");

  196. }

  197. 

  198. static void DEF(avg_pixels8)(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  199. {

  200.     __asm __volatile(

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

  202. 	"1:				\n\t"

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

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

  205. 	PAVGB" (%1), %%mm0		\n\t"

  206. 	PAVGB" (%1, %3), %%mm1		\n\t"

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

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

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

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

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

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

  213. 	PAVGB" (%1), %%mm0		\n\t"

  214. 	PAVGB" (%1, %3), %%mm1		\n\t"

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

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

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

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

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

  220. 	"jnz 1b				\n\t"

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

  222. 	:"r" (line_size)

  223. 	:"%eax", "memory");

  224. }

  225. 

  226. static void DEF(avg_pixels8_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  227. {

  228.     __asm __volatile(

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

  230. 	"1:				\n\t"

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

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

  233. 	PAVGB" 1(%1), %%mm0		\n\t"

  234. 	PAVGB" 1(%1, %3), %%mm2		\n\t"

  235. 	PAVGB" (%2), %%mm0		\n\t"

  236. 	PAVGB" (%2, %3), %%mm2		\n\t"

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

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

  239. 	"movq %%mm2, (%2, %3)		\n\t"

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

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

  242. 	PAVGB" 1(%1), %%mm0		\n\t"

  243. 	PAVGB" 1(%1, %3), %%mm2		\n\t"

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

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

  246. 	PAVGB" (%2), %%mm0		\n\t"

  247. 	PAVGB" (%2, %3), %%mm2		\n\t"

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

  249. 	"movq %%mm2, (%2, %3)		\n\t"

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

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

  252. 	"jnz 1b				\n\t"

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

  254. 	:"r" (line_size)

  255. 	:"%eax", "memory");

  256. }

  257. 

  258. static void DEF(avg_pixels8_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  259. {

  260.     __asm __volatile(

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

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

  263. 	"subl %3, %2			\n\t"

  264. 	"1:				\n\t"

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

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

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

  268. 	PAVGB" %%mm1, %%mm0		\n\t"

  269. 	PAVGB" %%mm2, %%mm1		\n\t"

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

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

  272. 	PAVGB" %%mm3, %%mm0		\n\t"

  273. 	PAVGB" %%mm4, %%mm1		\n\t"

  274. 	"movq %%mm0, (%2, %3)		\n\t"

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

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

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

  278. 	PAVGB" %%mm1, %%mm2		\n\t"

  279. 	PAVGB" %%mm0, %%mm1		\n\t"

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

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

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

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

  284. 	PAVGB" %%mm3, %%mm2		\n\t"

  285. 	PAVGB" %%mm4, %%mm1		\n\t"

  286. 	"movq %%mm2, (%2, %3)		\n\t"

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

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

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

  290. 	"jnz 1b				\n\t"

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

  292. 	:"r" (line_size)

  293. 	:"%eax", "memory");

  294. }

  295. 

  296. // Note this is not correctly rounded, but this function is only used for b frames so it doesnt matter 

  297. static void DEF(avg_pixels8_xy2)(UINT8 *block, const UINT8 *pixels, int line_size, int h)

  298. {

  299.     MOVQ_BONE(mm6);

  300.     __asm __volatile(

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

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

  303. 	PAVGB" 1(%1), %%mm0		\n\t"

  304. 	".balign 8			\n\t"

  305. 	"1:				\n\t"

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

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

  308. 	"psubusb %%mm6, %%mm2		\n\t"

  309. 	PAVGB" 1(%1, %3), %%mm1		\n\t"

  310. 	PAVGB" 1(%1, %%eax), %%mm2	\n\t"

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

  312. 	PAVGB" %%mm1, %%mm0		\n\t"

  313. 	PAVGB" %%mm2, %%mm1		\n\t"

  314. 	PAVGB" (%2), %%mm0		\n\t"

  315. 	PAVGB" (%2, %3), %%mm1		\n\t"

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

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

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

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

  320. 	PAVGB" 1(%1, %3), %%mm1		\n\t"

  321. 	PAVGB" 1(%1, %%eax), %%mm0	\n\t"

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

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

  324. 	PAVGB" %%mm1, %%mm2		\n\t"

  325. 	PAVGB" %%mm0, %%mm1		\n\t"

  326. 	PAVGB" (%2), %%mm2		\n\t"

  327. 	PAVGB" (%2, %3), %%mm1		\n\t"

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

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

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

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

  332. 	"jnz 1b				\n\t"

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

  334. 	:"r" (line_size)

  335. 	:"%eax",  "memory");

  336. }

  337. 

  338. //FIXME the following could be optimized too ...

  339. static void DEF(put_no_rnd_pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h){

  340.     DEF(put_no_rnd_pixels8_x2)(block  , pixels  , line_size, h);

  341.     DEF(put_no_rnd_pixels8_x2)(block+8, pixels+8, line_size, h);

  342. }

  343. static void DEF(put_pixels16_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h){

  344.     DEF(put_pixels8_y2)(block  , pixels  , line_size, h);

  345.     DEF(put_pixels8_y2)(block+8, pixels+8, line_size, h);

  346. }

  347. static void DEF(put_no_rnd_pixels16_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h){

  348.     DEF(put_no_rnd_pixels8_y2)(block  , pixels  , line_size, h);

  349.     DEF(put_no_rnd_pixels8_y2)(block+8, pixels+8, line_size, h);

  350. }

  351. static void DEF(avg_pixels16)(UINT8 *block, const UINT8 *pixels, int line_size, int h){

  352.     DEF(avg_pixels8)(block  , pixels  , line_size, h);

  353.     DEF(avg_pixels8)(block+8, pixels+8, line_size, h);

  354. }

  355. static void DEF(avg_pixels16_x2)(UINT8 *block, const UINT8 *pixels, int line_size, int h){

  356.     DEF(avg_pixels8_x2)(block  , pixels  , line_size, h);

  357.     DEF(avg_pixels8_x2)(block+8, pixels+8, line_size, h);

  358. }

  359. static void DEF(avg_pixels16_y2)(UINT8 *block, const UINT8 *pixels, int line_size, int h){

  360.     DEF(avg_pixels8_y2)(block  , pixels  , line_size, h);

  361.     DEF(avg_pixels8_y2)(block+8, pixels+8, line_size, h);

  362. }

  363. static void DEF(avg_pixels16_xy2)(UINT8 *block, const UINT8 *pixels, int line_size, int h){

  364.     DEF(avg_pixels8_xy2)(block  , pixels  , line_size, h);

  365.     DEF(avg_pixels8_xy2)(block+8, pixels+8, line_size, h);

  366. }