/xbmc/visualizations/Goom/goom2k4-0/src/xmmx.c

http://github.com/xbmc/xbmc · C · 390 lines · 244 code · 59 blank · 87 comment · 36 complexity · 0b35e33c4a9cf5dc7777a2b85c4589d4 MD5 · raw file 

    

  1. #ifdef HAVE_MMX
    2. 

  2. 

  3. /* a definir pour avoir exactement le meme resultat que la fonction C
    4. 

  4.  * (un chouillat plus lent).. mais la difference est assez peu notable.
    5. 

  5.  */
    6. 

  6. // #define STRICT_COMPAT
    7. 

  7. 

  8. #define BUFFPOINTNB 16
    9. 

  9. #define BUFFPOINTMASK 0xffff
    10. 

  10. #define BUFFINCR 0xff
    11. 

  11. 

  12. #define sqrtperte 16
    13. 

  13. /* faire : a % sqrtperte <=> a & pertemask*/
    14. 

  14. #define PERTEMASK 0xf
    15. 

  15. /* faire : a / sqrtperte <=> a >> PERTEDEC*/
    16. 

  16. #define PERTEDEC 4
    17. 

  17. 

  18. 

  19. /*#define MMX_TRACE*/
    20. 

  20. #include "mmx.h"
    21. 

  21. /*#include "xmmx.h"*/
    22. 

  22. #include "goom_graphic.h"
    23. 

  23. 

  24. int xmmx_supported (void) {
    25. 

  25. 	return (mm_support()&0x8)>>3;
    26. 

  26. }
    27. 

  27. 

  28. void zoom_filter_xmmx (int prevX, int prevY,
    29. 

  29.                        Pixel *expix1, Pixel *expix2,
    30. 

  30.                        int *lbruS, int *lbruD, int buffratio,
    31. 

  31.                        int precalCoef[16][16])
    32. 

  32. {
    33. 

  33. 	int bufsize = prevX * prevY; /* taille du buffer */
    34. 

  34. 	volatile int loop;                    /* variable de boucle */
    35. 

  35. 

  36. 	mmx_t *brutS = (mmx_t*)lbruS; /* buffer de transformation source */
    37. 

  37. 	mmx_t *brutD = (mmx_t*)lbruD; /* buffer de transformation dest */
    38. 

  38. 

  39. 	volatile mmx_t prevXY;
    40. 

  40. 	volatile mmx_t ratiox;
    41. 

  41. 	/*	volatile mmx_t interpix; */
    42. 

  42. 

  43. 	expix1[0].val=expix1[prevX-1].val=expix1[prevX*prevY-1].val=expix1[prevX*prevY-prevX].val=0;
    44. 

  44. 

  45. 	prevXY.ud[0] = (prevX-1)<<PERTEDEC;
    46. 

  46. 	prevXY.ud[1] = (prevY-1)<<PERTEDEC;
    47. 

  47. 

  48. 	ratiox.d[0] = buffratio;
    49. 

  49. 	ratiox.d[1] = buffratio;
    50. 

  50. 

  51.   asm volatile
    52. 

  52.     ("\n\t movq  %[ratio], %%mm6"
    53. 

  53.      "\n\t pslld $16,      %%mm6" /* mm6 = [rat16=buffratio<<16 | rat16=buffratio<<16] */
    54. 

  54.      "\n\t pxor  %%mm7,    %%mm7" /* mm7 = 0 */
    55. 

  55.      ::[ratio]"m"(ratiox));
    56. 

  56. 

  57. 	loop=0;
    58. 

  58. 

  59. 	/*
    60. 

  60. 	 * NOTE : mm6 et mm7 ne sont pas modifies dans la boucle.
    61. 

  61. 	 */
    62. 

  62. 	while (loop < bufsize)
    63. 

  63. 	{
    64. 

  64. 		/* Thread #1
    65. 

  65. 		 * pre :  mm6 = [rat16|rat16]
    66. 

  66. 		 * post : mm0 = S + ((D-S)*rat16 format [X|Y]
    67. 

  67. 		 * modified = mm0,mm1,mm2
    68. 

  68. 		 */
    69. 

  69. 

  70.     asm volatile ("#1 \n\t movq 0(%[brutS]), %%mm0"
    71. 

  71.        "#1 \n\t movq 0(%[brutD]), %%mm1"
    72. 

  72.        "#1 \n\t psubd   %%mm0, %%mm1"  /* mm1 = D - S */
    73. 

  73.        "#1 \n\t movq    %%mm1, %%mm2" /* mm2 = D - S */
    74. 

  74.        "#1 \n\t pslld     $16, %%mm1"
    75. 

  75.        "#1 \n\t pmullw  %%mm6, %%mm2"
    76. 

  76.        "#1 \n\t pmulhuw %%mm6, %%mm1"
    77. 

  77.        "#1 \n\t pslld   $16,   %%mm0"
    78. 

  78.        "#1 \n\t paddd   %%mm2, %%mm1"  /* mm1 = (D - S) * buffratio >> 16 */
    79. 

  79. 

  80.        "#1 \n\t paddd   %%mm1, %%mm0"  /* mm0 = S + mm1 */
    81. 

  81.        "#1 \n\t psrld   $16,   %%mm0"
    82. 

  82.        :
    83. 

  83.        :[brutS] "r" (&brutS[loop]) ,[brutD] "r" (&brutD[loop])
    84. 

  84.          );                      /* mm0 = S */
    85. 

  85. 

  86. 		/*
    87. 

  87. 		 * pre : mm0 : position vector on screen
    88. 

  88. 		 *       prevXY : coordinate of the lower-right point on screen
    89. 

  89. 		 * post : clipped mm0
    90. 

  90. 		 * modified : mm0,mm1,mm2
    91. 

  91. 		 */
    92. 

  92.     asm volatile
    93. 

  93.       ("#1 \n\t movq %[prevXY], %%mm1"
    94. 

  94.        "#1 \n\t pcmpgtd %%mm0,  %%mm1"
    95. 

  95.        /* mm0 en X contient (idem pour Y) :
    96. 

  96.         *   1111 si prevXY > px
    97. 

  97.         *   0000 si prevXY <= px */
    98. 

  98. #ifdef STRICT_COMPAT
    99. 

  99.        "#1 \n\t movq      %%mm1, %%mm2"
    100. 

  100.        "#1 \n\t punpckhdq %%mm2, %%mm2"
    101. 

  101.        "#1 \n\t punpckldq %%mm1, %%mm1"
    102. 

  102.        "#1 \n\t pand      %%mm2, %%mm0"
    103. 

  103. #endif
    104. 

  104. 

  105.        "#1 \n\t pand %%mm1, %%mm0" /* on met a zero la partie qui deborde */
    106. 

  106.         ::[prevXY]"m"(prevXY));
    107. 

  107. 

  108. 		/* Thread #2
    109. 

  109. 		 * pre :  mm0 : clipped position on screen
    110. 

  110. 		 *
    111. 

  111. 		 * post : mm3 : coefs for this position
    112. 

  112. 		 *        mm1 : X vector [0|X]
    113. 

  113. 		 *
    114. 

  114. 		 * modif : eax,esi
    115. 

  115. 		 */
    116. 

  116. 		__asm__ __volatile__ (
    117. 

  117. 			"#2 \n\t movd %%mm0,%%esi"
    118. 

  118. 			"#2 \n\t movq %%mm0,%%mm1"
    119. 

  119. 

  120. 			"#2 \n\t andl $15,%%esi"
    121. 

  121. 			"#2 \n\t psrlq $32,%%mm1"
    122. 

  122. 

  123. 			"#2 \n\t shll $6,%%esi"
    124. 

  124. 			"#2 \n\t movd %%mm1,%%eax"
    125. 

  125. 

  126. 			"#2 \n\t addl %[precalCoef],%%esi"
    127. 

  127. 			"#2 \n\t andl $15,%%eax"
    128. 

  128. 

  129. 			"#2 \n\t movd (%%esi,%%eax,4),%%mm3"
    130. 

  130. 			::[precalCoef]"g"(precalCoef):"eax","esi");
    131. 

  131. 

  132. 		/*
    133. 

  133. 		 * extraction des coefficients... (Thread #3)
    134. 

  134. 		 *
    135. 

  135. 		 * pre : coef dans mm3
    136. 

  136. 		 *
    137. 

  137. 		 * post : coef extraits dans mm3 (c1 & c2)
    138. 

  138. 		 *                        et mm4 (c3 & c4)
    139. 

  139. 		 *
    140. 

  140. 		 * modif : mm5
    141. 

  141. 		 */
    142. 

  142. 

  143. 		/* (Thread #4)
    144. 

  144. 		 * pre : mm0 : Y pos [*|Y]
    145. 

  145. 		 *       mm1 : X pos [*|X]
    146. 

  146. 		 *
    147. 

  147. 		 * post : mm0 : expix1[position]
    148. 

  148. 		 *        mm2 : expix1[position+largeur]
    149. 

  149. 		 *
    150. 

  150. 		 * modif : eax, esi
    151. 

  151. 		 */
    152. 

  152. 		__asm__ __volatile__ (
    153. 

  153.       "#2 \n\t psrld $4, %%mm0"
    154. 

  154.       "#2 \n\t psrld $4, %%mm1"      /* PERTEDEC = $4 */
    155. 

  155. 

  156.       "#4 \n\t movd %%mm1,%%eax"
    157. 

  157. 			"#3 \n\t movq %%mm3,%%mm5" 
    158. 

  158. 

  159. 			"#4 \n\t mull %[prevX]"
    160. 

  160. 			"#4 \n\t movd %%mm0,%%esi"
    161. 

  161. 

  162.       "#3 \n\t punpcklbw %%mm5, %%mm3"
    163. 

  163. 			"#4 \n\t addl %%esi, %%eax"
    164. 

  164. 

  165.       "#3 \n\t movq %%mm3, %%mm4"     
    166. 

  166.       "#3 \n\t movq %%mm3, %%mm5"     
    167. 

  167. 

  168.       "#4 \n\t movl %[expix1], %%esi"
    169. 

  169.       "#3 \n\t punpcklbw %%mm5, %%mm3"
    170. 

  170. 

  171.       "#4 \n\t movq (%%esi,%%eax,4),%%mm0"
    172. 

  172.       "#3 \n\t punpckhbw %%mm5, %%mm4"
    173. 

  173. 

  174.       "#4 \n\t addl %[prevX],%%eax"
    175. 

  175.       "#4 \n\t movq (%%esi,%%eax,4),%%mm2"
    176. 

  176. 

  177. 			:
    178. 

  178.       : [expix1] "g"(expix1)
    179. 

  179.       , [prevX]  "g"(prevX)
    180. 

  180.       :"eax","esi"
    181. 

  181. 		);
    182. 

  182. 

  183. 		/*
    184. 

  184. 		 * pre :       mm0 : expix1[position]
    185. 

  185. 		 *             mm2 : expix1[position+largeur]
    186. 

  186. 		 *       mm3 & mm4 : coefs
    187. 

  187. 		 */
    188. 

  188. 

  189. 		/* recopie des deux premiers pixels dans mm0 et mm1 */
    190. 

  190. 		movq_r2r (mm0, mm1);            /* b1-v1-r1-a1-b2-v2-r2-a2 */
    191. 

  191. 

  192. 		/* depackage du premier pixel */
    193. 

  193. 		punpcklbw_r2r (mm7, mm0);       /* 00-b2-00-v2-00-r2-00-a2 */
    194. 

  194. 

  195. 		/* extraction des coefficients... */
    196. 

  196. 

  197. 		movq_r2r (mm3, mm5);      /* c2-c2-c2-c2-c1-c1-c1-c1 */
    198. 

  198. 

  199. 		/*^en parrallele^*/ /* depackage du 2ieme pixel */
    200. 

  200. 		/*^*/ punpckhbw_r2r (mm7, mm1); /* 00-b1-00-v1-00-r1-00-a1 */
    201. 

  201. 

  202. 		punpcklbw_r2r (mm7, mm5);	/* 00-c1-00-c1-00-c1-00-c1 */
    203. 

  203. 		punpckhbw_r2r (mm7, mm3);	/* 00-c2-00-c2-00-c2-00-c2 */
    204. 

  204. 

  205. 		/* multiplication des pixels par les coefficients */
    206. 

  206. 		pmullw_r2r (mm5, mm0);		/* c1*b2-c1*v2-c1*r2-c1*a2 */
    207. 

  207. 		pmullw_r2r (mm3, mm1);		/* c2*b1-c2*v1-c2*r1-c2*a1 */
    208. 

  208. 		paddw_r2r (mm1, mm0);
    209. 

  209. 

  210. 		/* ...extraction des 2 derniers coefficients */
    211. 

  211. 		movq_r2r (mm4, mm5);			/* c4-c4-c4-c4-c3-c3-c3-c3 */
    212. 

  212. 		punpcklbw_r2r (mm7, mm4);	/* 00-c3-00-c3-00-c3-00-c3 */
    213. 

  213. 		punpckhbw_r2r (mm7, mm5);	/* 00-c4-00-c4-00-c4-00-c4 */
    214. 

  214. 

  215. 		/* recuperation des 2 derniers pixels */
    216. 

  216. 		movq_r2r (mm2, mm1);
    217. 

  217. 

  218. 		/* depackage des pixels */
    219. 

  219. 		punpcklbw_r2r (mm7, mm1);
    220. 

  220. 		punpckhbw_r2r (mm7, mm2);
    221. 

  221. 

  222. 		/* multiplication pas les coeffs */
    223. 

  223. 		pmullw_r2r (mm4, mm1);
    224. 

  224. 		pmullw_r2r (mm5, mm2);
    225. 

  225. 

  226. 		/* ajout des valeurs obtenues ? la valeur finale */
    227. 

  227. 		paddw_r2r (mm1, mm0);
    228. 

  228. 		paddw_r2r (mm2, mm0);
    229. 

  229. 

  230. 		/* division par 256 = 16+16+16+16, puis repackage du pixel final */
    231. 

  231. 		psrlw_i2r (8, mm0);
    232. 

  232. 		packuswb_r2r (mm7, mm0);
    233. 

  233. 

  234. 		movd_r2m (mm0,expix2[loop]);
    235. 

  235. 

  236. 		++loop;
    237. 

  237. 	}
    238. 

  238. 	__asm__ __volatile__ ("emms\n");
    239. 

  239. }
    240. 

  240. 

  241. #define DRAWMETHOD_PLUS_XMMX(_out,_backbuf,_col) \
    242. 

  242. { \
    243. 

  243. 	movd_m2r(_backbuf, mm0); \
    244. 

  244. 	paddusb_m2r(_col, mm0); \
    245. 

  245. 	movd_r2m(mm0, _out); \
    246. 

  246. }
    247. 

  247. 

  248. #define DRAWMETHOD DRAWMETHOD_PLUS_XMMX(*p,*p,col)
    249. 

  249. 

  250. void draw_line_xmmx (Pixel *data, int x1, int y1, int x2, int y2, int col, int screenx, int screeny)
    251. 

  251. {
    252. 

  252. 	int x, y, dx, dy, yy, xx;
    253. 

  253. 	Pixel *p;
    254. 

  254. 

  255. 	if ((y1 < 0) || (y2 < 0) || (x1 < 0) || (x2 < 0) || (y1 >= screeny) || (y2 >= screeny) || (x1 >= screenx) || (x2 >= screenx))
    256. 

  256. 		goto end_of_line;
    257. 

  257. 

  258. 	dx = x2 - x1;
    259. 

  259. 	dy = y2 - y1;
    260. 

  260. 	if (x1 >= x2) {
    261. 

  261. 		int tmp;
    262. 

  262. 

  263. 		tmp = x1;
    264. 

  264. 		x1 = x2;
    265. 

  265. 		x2 = tmp;
    266. 

  266. 		tmp = y1;
    267. 

  267. 		y1 = y2;
    268. 

  268. 		y2 = tmp;
    269. 

  269. 		dx = x2 - x1;
    270. 

  270. 		dy = y2 - y1;
    271. 

  271. 	}
    272. 

  272. 

  273. 	/* vertical line */
    274. 

  274. 	if (dx == 0) {
    275. 

  275. 		if (y1 < y2) {
    276. 

  276. 			p = &(data[(screenx * y1) + x1]);
    277. 

  277. 			for (y = y1; y <= y2; y++) {
    278. 

  278. 				DRAWMETHOD;
    279. 

  279. 				p += screenx;
    280. 

  280. 			}
    281. 

  281. 		}
    282. 

  282. 		else {
    283. 

  283. 			p = &(data[(screenx * y2) + x1]);
    284. 

  284. 			for (y = y2; y <= y1; y++) {
    285. 

  285. 				DRAWMETHOD;
    286. 

  286. 				p += screenx;
    287. 

  287. 			}
    288. 

  288. 		}
    289. 

  289. 		goto end_of_line;
    290. 

  290. 	}
    291. 

  291. 	/* horizontal line */
    292. 

  292. 	if (dy == 0) {
    293. 

  293. 		if (x1 < x2) {
    294. 

  294. 			p = &(data[(screenx * y1) + x1]);
    295. 

  295. 			for (x = x1; x <= x2; x++) {
    296. 

  296. 				DRAWMETHOD;
    297. 

  297. 				p++;
    298. 

  298. 			}
    299. 

  299. 			goto end_of_line;
    300. 

  300. 		}
    301. 

  301. 		else {
    302. 

  302. 			p = &(data[(screenx * y1) + x2]);
    303. 

  303. 			for (x = x2; x <= x1; x++) {
    304. 

  304. 				DRAWMETHOD;
    305. 

  305. 				p++;
    306. 

  306. 			}
    307. 

  307. 			goto end_of_line;
    308. 

  308. 		}
    309. 

  309. 	}
    310. 

  310. 	/* 1    */
    311. 

  311. 	/*  \   */
    312. 

  312. 	/*   \  */
    313. 

  313. 	/*    2 */
    314. 

  314. 	if (y2 > y1) {
    315. 

  315. 		/* steep */
    316. 

  316. 		if (dy > dx) {
    317. 

  317. 			dx = ((dx << 16) / dy);
    318. 

  318. 			x = x1 << 16;
    319. 

  319. 			for (y = y1; y <= y2; y++) {
    320. 

  320. 				xx = x >> 16;
    321. 

  321. 				p = &(data[(screenx * y) + xx]);
    322. 

  322. 				DRAWMETHOD;
    323. 

  323. 				if (xx < (screenx - 1)) {
    324. 

  324. 					p++;
    325. 

  325. 					/* DRAWMETHOD; */
    326. 

  326. 				}
    327. 

  327. 				x += dx;
    328. 

  328. 			}
    329. 

  329. 			goto end_of_line;
    330. 

  330. 		}
    331. 

  331. 		/* shallow */
    332. 

  332. 		else {
    333. 

  333. 			dy = ((dy << 16) / dx);
    334. 

  334. 			y = y1 << 16;
    335. 

  335. 			for (x = x1; x <= x2; x++) {
    336. 

  336. 				yy = y >> 16;
    337. 

  337. 				p = &(data[(screenx * yy) + x]);
    338. 

  338. 				DRAWMETHOD;
    339. 

  339. 				if (yy < (screeny - 1)) {
    340. 

  340. 					p += screeny;
    341. 

  341. 					/* DRAWMETHOD; */
    342. 

  342. 				}
    343. 

  343. 				y += dy;
    344. 

  344. 			}
    345. 

  345. 		}
    346. 

  346. 	}
    347. 

  347. 	/*    2 */
    348. 

  348. 	/*   /  */
    349. 

  349. 	/*  /   */
    350. 

  350. 	/* 1    */
    351. 

  351. 	else {
    352. 

  352. 		/* steep */
    353. 

  353. 		if (-dy > dx) {
    354. 

  354. 			dx = ((dx << 16) / -dy);
    355. 

  355. 			x = (x1 + 1) << 16;
    356. 

  356. 			for (y = y1; y >= y2; y--) {
    357. 

  357. 				xx = x >> 16;
    358. 

  358. 				p = &(data[(screenx * y) + xx]);
    359. 

  359. 				DRAWMETHOD;
    360. 

  360. 				if (xx < (screenx - 1)) {
    361. 

  361. 					p--;
    362. 

  362. 					/* DRAWMETHOD; */
    363. 

  363. 				}
    364. 

  364. 				x += dx;
    365. 

  365. 			}
    366. 

  366. 			goto end_of_line;
    367. 

  367. 		}
    368. 

  368. 		/* shallow */
    369. 

  369. 		else {
    370. 

  370. 			dy = ((dy << 16) / dx);
    371. 

  371. 			y = y1 << 16;
    372. 

  372. 			for (x = x1; x <= x2; x++) {
    373. 

  373. 				yy = y >> 16;
    374. 

  374. 				p = &(data[(screenx * yy) + x]);
    375. 

  375. 				DRAWMETHOD;
    376. 

  376. 				if (yy < (screeny - 1)) {
    377. 

  377. 					p += screeny;
    378. 

  378. 					/* DRAWMETHOD; */
    379. 

  379. 				}
    380. 

  380. 				y += dy;
    381. 

  381. 			}
    382. 

  382. 			goto end_of_line;
    383. 

  383. 		}
    384. 

  384. 	}
    385. 

  385. end_of_line:
    386. 

  386. 	__asm__ __volatile__ ("emms\n"); 
    387. 

  387. }
    388. 

  388. 

  389. #endif
    390.