/xbmc/visualizations/Goom/goom2k4-0/src/xmmx.c
C | 390 lines | 244 code | 59 blank | 87 comment | 36 complexity | 0b35e33c4a9cf5dc7777a2b85c4589d4 MD5 | raw file
1 2#ifdef HAVE_MMX 3 4/* a definir pour avoir exactement le meme resultat que la fonction C 5 * (un chouillat plus lent).. mais la difference est assez peu notable. 6 */ 7// #define STRICT_COMPAT 8 9#define BUFFPOINTNB 16 10#define BUFFPOINTMASK 0xffff 11#define BUFFINCR 0xff 12 13#define sqrtperte 16 14/* faire : a % sqrtperte <=> a & pertemask*/ 15#define PERTEMASK 0xf 16/* faire : a / sqrtperte <=> a >> PERTEDEC*/ 17#define PERTEDEC 4 18 19 20/*#define MMX_TRACE*/ 21#include "mmx.h" 22/*#include "xmmx.h"*/ 23#include "goom_graphic.h" 24 25int xmmx_supported (void) { 26 return (mm_support()&0x8)>>3; 27} 28 29void zoom_filter_xmmx (int prevX, int prevY, 30 Pixel *expix1, Pixel *expix2, 31 int *lbruS, int *lbruD, int buffratio, 32 int precalCoef[16][16]) 33{ 34 int bufsize = prevX * prevY; /* taille du buffer */ 35 volatile int loop; /* variable de boucle */ 36 37 mmx_t *brutS = (mmx_t*)lbruS; /* buffer de transformation source */ 38 mmx_t *brutD = (mmx_t*)lbruD; /* buffer de transformation dest */ 39 40 volatile mmx_t prevXY; 41 volatile mmx_t ratiox; 42 /* volatile mmx_t interpix; */ 43 44 expix1[0].val=expix1[prevX-1].val=expix1[prevX*prevY-1].val=expix1[prevX*prevY-prevX].val=0; 45 46 prevXY.ud[0] = (prevX-1)<<PERTEDEC; 47 prevXY.ud[1] = (prevY-1)<<PERTEDEC; 48 49 ratiox.d[0] = buffratio; 50 ratiox.d[1] = buffratio; 51 52 asm volatile 53 ("\n\t movq %[ratio], %%mm6" 54 "\n\t pslld $16, %%mm6" /* mm6 = [rat16=buffratio<<16 | rat16=buffratio<<16] */ 55 "\n\t pxor %%mm7, %%mm7" /* mm7 = 0 */ 56 ::[ratio]"m"(ratiox)); 57 58 loop=0; 59 60 /* 61 * NOTE : mm6 et mm7 ne sont pas modifies dans la boucle. 62 */ 63 while (loop < bufsize) 64 { 65 /* Thread #1 66 * pre : mm6 = [rat16|rat16] 67 * post : mm0 = S + ((D-S)*rat16 format [X|Y] 68 * modified = mm0,mm1,mm2 69 */ 70 71 asm volatile ("#1 \n\t movq 0(%[brutS]), %%mm0" 72 "#1 \n\t movq 0(%[brutD]), %%mm1" 73 "#1 \n\t psubd %%mm0, %%mm1" /* mm1 = D - S */ 74 "#1 \n\t movq %%mm1, %%mm2" /* mm2 = D - S */ 75 "#1 \n\t pslld $16, %%mm1" 76 "#1 \n\t pmullw %%mm6, %%mm2" 77 "#1 \n\t pmulhuw %%mm6, %%mm1" 78 "#1 \n\t pslld $16, %%mm0" 79 "#1 \n\t paddd %%mm2, %%mm1" /* mm1 = (D - S) * buffratio >> 16 */ 80 81 "#1 \n\t paddd %%mm1, %%mm0" /* mm0 = S + mm1 */ 82 "#1 \n\t psrld $16, %%mm0" 83 : 84 :[brutS] "r" (&brutS[loop]) ,[brutD] "r" (&brutD[loop]) 85 ); /* mm0 = S */ 86 87 /* 88 * pre : mm0 : position vector on screen 89 * prevXY : coordinate of the lower-right point on screen 90 * post : clipped mm0 91 * modified : mm0,mm1,mm2 92 */ 93 asm volatile 94 ("#1 \n\t movq %[prevXY], %%mm1" 95 "#1 \n\t pcmpgtd %%mm0, %%mm1" 96 /* mm0 en X contient (idem pour Y) : 97 * 1111 si prevXY > px 98 * 0000 si prevXY <= px */ 99#ifdef STRICT_COMPAT 100 "#1 \n\t movq %%mm1, %%mm2" 101 "#1 \n\t punpckhdq %%mm2, %%mm2" 102 "#1 \n\t punpckldq %%mm1, %%mm1" 103 "#1 \n\t pand %%mm2, %%mm0" 104#endif 105 106 "#1 \n\t pand %%mm1, %%mm0" /* on met a zero la partie qui deborde */ 107 ::[prevXY]"m"(prevXY)); 108 109 /* Thread #2 110 * pre : mm0 : clipped position on screen 111 * 112 * post : mm3 : coefs for this position 113 * mm1 : X vector [0|X] 114 * 115 * modif : eax,esi 116 */ 117 __asm__ __volatile__ ( 118 "#2 \n\t movd %%mm0,%%esi" 119 "#2 \n\t movq %%mm0,%%mm1" 120 121 "#2 \n\t andl $15,%%esi" 122 "#2 \n\t psrlq $32,%%mm1" 123 124 "#2 \n\t shll $6,%%esi" 125 "#2 \n\t movd %%mm1,%%eax" 126 127 "#2 \n\t addl %[precalCoef],%%esi" 128 "#2 \n\t andl $15,%%eax" 129 130 "#2 \n\t movd (%%esi,%%eax,4),%%mm3" 131 ::[precalCoef]"g"(precalCoef):"eax","esi"); 132 133 /* 134 * extraction des coefficients... (Thread #3) 135 * 136 * pre : coef dans mm3 137 * 138 * post : coef extraits dans mm3 (c1 & c2) 139 * et mm4 (c3 & c4) 140 * 141 * modif : mm5 142 */ 143 144 /* (Thread #4) 145 * pre : mm0 : Y pos [*|Y] 146 * mm1 : X pos [*|X] 147 * 148 * post : mm0 : expix1[position] 149 * mm2 : expix1[position+largeur] 150 * 151 * modif : eax, esi 152 */ 153 __asm__ __volatile__ ( 154 "#2 \n\t psrld $4, %%mm0" 155 "#2 \n\t psrld $4, %%mm1" /* PERTEDEC = $4 */ 156 157 "#4 \n\t movd %%mm1,%%eax" 158 "#3 \n\t movq %%mm3,%%mm5" 159 160 "#4 \n\t mull %[prevX]" 161 "#4 \n\t movd %%mm0,%%esi" 162 163 "#3 \n\t punpcklbw %%mm5, %%mm3" 164 "#4 \n\t addl %%esi, %%eax" 165 166 "#3 \n\t movq %%mm3, %%mm4" 167 "#3 \n\t movq %%mm3, %%mm5" 168 169 "#4 \n\t movl %[expix1], %%esi" 170 "#3 \n\t punpcklbw %%mm5, %%mm3" 171 172 "#4 \n\t movq (%%esi,%%eax,4),%%mm0" 173 "#3 \n\t punpckhbw %%mm5, %%mm4" 174 175 "#4 \n\t addl %[prevX],%%eax" 176 "#4 \n\t movq (%%esi,%%eax,4),%%mm2" 177 178 : 179 : [expix1] "g"(expix1) 180 , [prevX] "g"(prevX) 181 :"eax","esi" 182 ); 183 184 /* 185 * pre : mm0 : expix1[position] 186 * mm2 : expix1[position+largeur] 187 * mm3 & mm4 : coefs 188 */ 189 190 /* recopie des deux premiers pixels dans mm0 et mm1 */ 191 movq_r2r (mm0, mm1); /* b1-v1-r1-a1-b2-v2-r2-a2 */ 192 193 /* depackage du premier pixel */ 194 punpcklbw_r2r (mm7, mm0); /* 00-b2-00-v2-00-r2-00-a2 */ 195 196 /* extraction des coefficients... */ 197 198 movq_r2r (mm3, mm5); /* c2-c2-c2-c2-c1-c1-c1-c1 */ 199 200 /*^en parrallele^*/ /* depackage du 2ieme pixel */ 201 /*^*/ punpckhbw_r2r (mm7, mm1); /* 00-b1-00-v1-00-r1-00-a1 */ 202 203 punpcklbw_r2r (mm7, mm5); /* 00-c1-00-c1-00-c1-00-c1 */ 204 punpckhbw_r2r (mm7, mm3); /* 00-c2-00-c2-00-c2-00-c2 */ 205 206 /* multiplication des pixels par les coefficients */ 207 pmullw_r2r (mm5, mm0); /* c1*b2-c1*v2-c1*r2-c1*a2 */ 208 pmullw_r2r (mm3, mm1); /* c2*b1-c2*v1-c2*r1-c2*a1 */ 209 paddw_r2r (mm1, mm0); 210 211 /* ...extraction des 2 derniers coefficients */ 212 movq_r2r (mm4, mm5); /* c4-c4-c4-c4-c3-c3-c3-c3 */ 213 punpcklbw_r2r (mm7, mm4); /* 00-c3-00-c3-00-c3-00-c3 */ 214 punpckhbw_r2r (mm7, mm5); /* 00-c4-00-c4-00-c4-00-c4 */ 215 216 /* recuperation des 2 derniers pixels */ 217 movq_r2r (mm2, mm1); 218 219 /* depackage des pixels */ 220 punpcklbw_r2r (mm7, mm1); 221 punpckhbw_r2r (mm7, mm2); 222 223 /* multiplication pas les coeffs */ 224 pmullw_r2r (mm4, mm1); 225 pmullw_r2r (mm5, mm2); 226 227 /* ajout des valeurs obtenues ? la valeur finale */ 228 paddw_r2r (mm1, mm0); 229 paddw_r2r (mm2, mm0); 230 231 /* division par 256 = 16+16+16+16, puis repackage du pixel final */ 232 psrlw_i2r (8, mm0); 233 packuswb_r2r (mm7, mm0); 234 235 movd_r2m (mm0,expix2[loop]); 236 237 ++loop; 238 } 239 __asm__ __volatile__ ("emms\n"); 240} 241 242#define DRAWMETHOD_PLUS_XMMX(_out,_backbuf,_col) \ 243{ \ 244 movd_m2r(_backbuf, mm0); \ 245 paddusb_m2r(_col, mm0); \ 246 movd_r2m(mm0, _out); \ 247} 248 249#define DRAWMETHOD DRAWMETHOD_PLUS_XMMX(*p,*p,col) 250 251void draw_line_xmmx (Pixel *data, int x1, int y1, int x2, int y2, int col, int screenx, int screeny) 252{ 253 int x, y, dx, dy, yy, xx; 254 Pixel *p; 255 256 if ((y1 < 0) || (y2 < 0) || (x1 < 0) || (x2 < 0) || (y1 >= screeny) || (y2 >= screeny) || (x1 >= screenx) || (x2 >= screenx)) 257 goto end_of_line; 258 259 dx = x2 - x1; 260 dy = y2 - y1; 261 if (x1 >= x2) { 262 int tmp; 263 264 tmp = x1; 265 x1 = x2; 266 x2 = tmp; 267 tmp = y1; 268 y1 = y2; 269 y2 = tmp; 270 dx = x2 - x1; 271 dy = y2 - y1; 272 } 273 274 /* vertical line */ 275 if (dx == 0) { 276 if (y1 < y2) { 277 p = &(data[(screenx * y1) + x1]); 278 for (y = y1; y <= y2; y++) { 279 DRAWMETHOD; 280 p += screenx; 281 } 282 } 283 else { 284 p = &(data[(screenx * y2) + x1]); 285 for (y = y2; y <= y1; y++) { 286 DRAWMETHOD; 287 p += screenx; 288 } 289 } 290 goto end_of_line; 291 } 292 /* horizontal line */ 293 if (dy == 0) { 294 if (x1 < x2) { 295 p = &(data[(screenx * y1) + x1]); 296 for (x = x1; x <= x2; x++) { 297 DRAWMETHOD; 298 p++; 299 } 300 goto end_of_line; 301 } 302 else { 303 p = &(data[(screenx * y1) + x2]); 304 for (x = x2; x <= x1; x++) { 305 DRAWMETHOD; 306 p++; 307 } 308 goto end_of_line; 309 } 310 } 311 /* 1 */ 312 /* \ */ 313 /* \ */ 314 /* 2 */ 315 if (y2 > y1) { 316 /* steep */ 317 if (dy > dx) { 318 dx = ((dx << 16) / dy); 319 x = x1 << 16; 320 for (y = y1; y <= y2; y++) { 321 xx = x >> 16; 322 p = &(data[(screenx * y) + xx]); 323 DRAWMETHOD; 324 if (xx < (screenx - 1)) { 325 p++; 326 /* DRAWMETHOD; */ 327 } 328 x += dx; 329 } 330 goto end_of_line; 331 } 332 /* shallow */ 333 else { 334 dy = ((dy << 16) / dx); 335 y = y1 << 16; 336 for (x = x1; x <= x2; x++) { 337 yy = y >> 16; 338 p = &(data[(screenx * yy) + x]); 339 DRAWMETHOD; 340 if (yy < (screeny - 1)) { 341 p += screeny; 342 /* DRAWMETHOD; */ 343 } 344 y += dy; 345 } 346 } 347 } 348 /* 2 */ 349 /* / */ 350 /* / */ 351 /* 1 */ 352 else { 353 /* steep */ 354 if (-dy > dx) { 355 dx = ((dx << 16) / -dy); 356 x = (x1 + 1) << 16; 357 for (y = y1; y >= y2; y--) { 358 xx = x >> 16; 359 p = &(data[(screenx * y) + xx]); 360 DRAWMETHOD; 361 if (xx < (screenx - 1)) { 362 p--; 363 /* DRAWMETHOD; */ 364 } 365 x += dx; 366 } 367 goto end_of_line; 368 } 369 /* shallow */ 370 else { 371 dy = ((dy << 16) / dx); 372 y = y1 << 16; 373 for (x = x1; x <= x2; x++) { 374 yy = y >> 16; 375 p = &(data[(screenx * yy) + x]); 376 DRAWMETHOD; 377 if (yy < (screeny - 1)) { 378 p += screeny; 379 /* DRAWMETHOD; */ 380 } 381 y += dy; 382 } 383 goto end_of_line; 384 } 385 } 386end_of_line: 387 __asm__ __volatile__ ("emms\n"); 388} 389 390#endif