Antonino A. Daplas | 68648ed | 2007-05-08 00:38:57 -0700 | [diff] [blame^] | 1 | /* |
| 2 | * Generic Bit Block Transfer for frame buffers located in system RAM with |
| 3 | * packed pixels of any depth. |
| 4 | * |
| 5 | * Based almost entirely from cfbcopyarea.c (which is based almost entirely |
| 6 | * on Geert Uytterhoeven's copyarea routine) |
| 7 | * |
| 8 | * Copyright (C) 2007 Antonino Daplas <adaplas@pol.net> |
| 9 | * |
| 10 | * This file is subject to the terms and conditions of the GNU General Public |
| 11 | * License. See the file COPYING in the main directory of this archive for |
| 12 | * more details. |
| 13 | * |
| 14 | */ |
| 15 | #include <linux/module.h> |
| 16 | #include <linux/kernel.h> |
| 17 | #include <linux/string.h> |
| 18 | #include <linux/fb.h> |
| 19 | #include <linux/slab.h> |
| 20 | #include <asm/types.h> |
| 21 | #include <asm/io.h> |
| 22 | |
| 23 | /* |
| 24 | * Compose two values, using a bitmask as decision value |
| 25 | * This is equivalent to (a & mask) | (b & ~mask) |
| 26 | */ |
| 27 | |
| 28 | static inline unsigned long |
| 29 | comp(unsigned long a, unsigned long b, unsigned long mask) |
| 30 | { |
| 31 | return ((a ^ b) & mask) ^ b; |
| 32 | } |
| 33 | |
| 34 | /* |
| 35 | * Generic bitwise copy algorithm |
| 36 | */ |
| 37 | |
| 38 | static void |
| 39 | bitcpy(unsigned long *dst, int dst_idx, const unsigned long *src, |
| 40 | int src_idx, int bits, unsigned n) |
| 41 | { |
| 42 | unsigned long first, last; |
| 43 | int const shift = dst_idx-src_idx; |
| 44 | int left, right; |
| 45 | |
| 46 | first = FB_SHIFT_HIGH(~0UL, dst_idx); |
| 47 | last = ~(FB_SHIFT_HIGH(~0UL, (dst_idx+n) % bits)); |
| 48 | |
| 49 | if (!shift) { |
| 50 | /* Same alignment for source and dest */ |
| 51 | if (dst_idx+n <= bits) { |
| 52 | /* Single word */ |
| 53 | if (last) |
| 54 | first &= last; |
| 55 | *dst = comp(*src, *dst, first); |
| 56 | } else { |
| 57 | /* Multiple destination words */ |
| 58 | /* Leading bits */ |
| 59 | if (first != ~0UL) { |
| 60 | *dst = comp(*src, *dst, first); |
| 61 | dst++; |
| 62 | src++; |
| 63 | n -= bits - dst_idx; |
| 64 | } |
| 65 | |
| 66 | /* Main chunk */ |
| 67 | n /= bits; |
| 68 | while (n >= 8) { |
| 69 | *dst++ = *src++; |
| 70 | *dst++ = *src++; |
| 71 | *dst++ = *src++; |
| 72 | *dst++ = *src++; |
| 73 | *dst++ = *src++; |
| 74 | *dst++ = *src++; |
| 75 | *dst++ = *src++; |
| 76 | *dst++ = *src++; |
| 77 | n -= 8; |
| 78 | } |
| 79 | while (n--) |
| 80 | *dst++ = *src++; |
| 81 | |
| 82 | /* Trailing bits */ |
| 83 | if (last) |
| 84 | *dst = comp(*src, *dst, last); |
| 85 | } |
| 86 | } else { |
| 87 | unsigned long d0, d1; |
| 88 | int m; |
| 89 | |
| 90 | /* Different alignment for source and dest */ |
| 91 | right = shift & (bits - 1); |
| 92 | left = -shift & (bits - 1); |
| 93 | |
| 94 | if (dst_idx+n <= bits) { |
| 95 | /* Single destination word */ |
| 96 | if (last) |
| 97 | first &= last; |
| 98 | if (shift > 0) { |
| 99 | /* Single source word */ |
| 100 | *dst = comp(*src >> right, *dst, first); |
| 101 | } else if (src_idx+n <= bits) { |
| 102 | /* Single source word */ |
| 103 | *dst = comp(*src << left, *dst, first); |
| 104 | } else { |
| 105 | /* 2 source words */ |
| 106 | d0 = *src++; |
| 107 | d1 = *src; |
| 108 | *dst = comp(d0 << left | d1 >> right, *dst, |
| 109 | first); |
| 110 | } |
| 111 | } else { |
| 112 | /* Multiple destination words */ |
| 113 | /** We must always remember the last value read, |
| 114 | because in case SRC and DST overlap bitwise (e.g. |
| 115 | when moving just one pixel in 1bpp), we always |
| 116 | collect one full long for DST and that might |
| 117 | overlap with the current long from SRC. We store |
| 118 | this value in 'd0'. */ |
| 119 | d0 = *src++; |
| 120 | /* Leading bits */ |
| 121 | if (shift > 0) { |
| 122 | /* Single source word */ |
| 123 | *dst = comp(d0 >> right, *dst, first); |
| 124 | dst++; |
| 125 | n -= bits - dst_idx; |
| 126 | } else { |
| 127 | /* 2 source words */ |
| 128 | d1 = *src++; |
| 129 | *dst = comp(d0 << left | *dst >> right, *dst, first); |
| 130 | d0 = d1; |
| 131 | dst++; |
| 132 | n -= bits - dst_idx; |
| 133 | } |
| 134 | |
| 135 | /* Main chunk */ |
| 136 | m = n % bits; |
| 137 | n /= bits; |
| 138 | while (n >= 4) { |
| 139 | d1 = *src++; |
| 140 | *dst++ = d0 << left | d1 >> right; |
| 141 | d0 = d1; |
| 142 | d1 = *src++; |
| 143 | *dst++ = d0 << left | d1 >> right; |
| 144 | d0 = d1; |
| 145 | d1 = *src++; |
| 146 | *dst++ = d0 << left | d1 >> right; |
| 147 | d0 = d1; |
| 148 | d1 = *src++; |
| 149 | *dst++ = d0 << left | d1 >> right; |
| 150 | d0 = d1; |
| 151 | n -= 4; |
| 152 | } |
| 153 | while (n--) { |
| 154 | d1 = *src++; |
| 155 | *dst++ = d0 << left | d1 >> right; |
| 156 | d0 = d1; |
| 157 | } |
| 158 | |
| 159 | /* Trailing bits */ |
| 160 | if (last) { |
| 161 | if (m <= right) { |
| 162 | /* Single source word */ |
| 163 | *dst = comp(d0 << left, *dst, last); |
| 164 | } else { |
| 165 | /* 2 source words */ |
| 166 | d1 = *src; |
| 167 | *dst = comp(d0 << left | d1 >> right, |
| 168 | *dst, last); |
| 169 | } |
| 170 | } |
| 171 | } |
| 172 | } |
| 173 | } |
| 174 | |
| 175 | /* |
| 176 | * Generic bitwise copy algorithm, operating backward |
| 177 | */ |
| 178 | |
| 179 | static void |
| 180 | bitcpy_rev(unsigned long *dst, int dst_idx, const unsigned long *src, |
| 181 | int src_idx, int bits, unsigned n) |
| 182 | { |
| 183 | unsigned long first, last; |
| 184 | int shift; |
| 185 | |
| 186 | dst += (n-1)/bits; |
| 187 | src += (n-1)/bits; |
| 188 | if ((n-1) % bits) { |
| 189 | dst_idx += (n-1) % bits; |
| 190 | dst += dst_idx >> (ffs(bits) - 1); |
| 191 | dst_idx &= bits - 1; |
| 192 | src_idx += (n-1) % bits; |
| 193 | src += src_idx >> (ffs(bits) - 1); |
| 194 | src_idx &= bits - 1; |
| 195 | } |
| 196 | |
| 197 | shift = dst_idx-src_idx; |
| 198 | |
| 199 | first = FB_SHIFT_LOW(~0UL, bits - 1 - dst_idx); |
| 200 | last = ~(FB_SHIFT_LOW(~0UL, bits - 1 - ((dst_idx-n) % bits))); |
| 201 | |
| 202 | if (!shift) { |
| 203 | /* Same alignment for source and dest */ |
| 204 | if ((unsigned long)dst_idx+1 >= n) { |
| 205 | /* Single word */ |
| 206 | if (last) |
| 207 | first &= last; |
| 208 | *dst = comp(*src, *dst, first); |
| 209 | } else { |
| 210 | /* Multiple destination words */ |
| 211 | |
| 212 | /* Leading bits */ |
| 213 | if (first != ~0UL) { |
| 214 | *dst = comp(*src, *dst, first); |
| 215 | dst--; |
| 216 | src--; |
| 217 | n -= dst_idx+1; |
| 218 | } |
| 219 | |
| 220 | /* Main chunk */ |
| 221 | n /= bits; |
| 222 | while (n >= 8) { |
| 223 | *dst-- = *src--; |
| 224 | *dst-- = *src--; |
| 225 | *dst-- = *src--; |
| 226 | *dst-- = *src--; |
| 227 | *dst-- = *src--; |
| 228 | *dst-- = *src--; |
| 229 | *dst-- = *src--; |
| 230 | *dst-- = *src--; |
| 231 | n -= 8; |
| 232 | } |
| 233 | while (n--) |
| 234 | *dst-- = *src--; |
| 235 | /* Trailing bits */ |
| 236 | if (last) |
| 237 | *dst = comp(*src, *dst, last); |
| 238 | } |
| 239 | } else { |
| 240 | /* Different alignment for source and dest */ |
| 241 | |
| 242 | int const left = -shift & (bits-1); |
| 243 | int const right = shift & (bits-1); |
| 244 | |
| 245 | if ((unsigned long)dst_idx+1 >= n) { |
| 246 | /* Single destination word */ |
| 247 | if (last) |
| 248 | first &= last; |
| 249 | if (shift < 0) { |
| 250 | /* Single source word */ |
| 251 | *dst = comp(*src << left, *dst, first); |
| 252 | } else if (1+(unsigned long)src_idx >= n) { |
| 253 | /* Single source word */ |
| 254 | *dst = comp(*src >> right, *dst, first); |
| 255 | } else { |
| 256 | /* 2 source words */ |
| 257 | *dst = comp(*src >> right | *(src-1) << left, |
| 258 | *dst, first); |
| 259 | } |
| 260 | } else { |
| 261 | /* Multiple destination words */ |
| 262 | /** We must always remember the last value read, |
| 263 | because in case SRC and DST overlap bitwise (e.g. |
| 264 | when moving just one pixel in 1bpp), we always |
| 265 | collect one full long for DST and that might |
| 266 | overlap with the current long from SRC. We store |
| 267 | this value in 'd0'. */ |
| 268 | unsigned long d0, d1; |
| 269 | int m; |
| 270 | |
| 271 | d0 = *src--; |
| 272 | /* Leading bits */ |
| 273 | if (shift < 0) { |
| 274 | /* Single source word */ |
| 275 | *dst = comp(d0 << left, *dst, first); |
| 276 | } else { |
| 277 | /* 2 source words */ |
| 278 | d1 = *src--; |
| 279 | *dst = comp(d0 >> right | d1 << left, *dst, |
| 280 | first); |
| 281 | d0 = d1; |
| 282 | } |
| 283 | dst--; |
| 284 | n -= dst_idx+1; |
| 285 | |
| 286 | /* Main chunk */ |
| 287 | m = n % bits; |
| 288 | n /= bits; |
| 289 | while (n >= 4) { |
| 290 | d1 = *src--; |
| 291 | *dst-- = d0 >> right | d1 << left; |
| 292 | d0 = d1; |
| 293 | d1 = *src--; |
| 294 | *dst-- = d0 >> right | d1 << left; |
| 295 | d0 = d1; |
| 296 | d1 = *src--; |
| 297 | *dst-- = d0 >> right | d1 << left; |
| 298 | d0 = d1; |
| 299 | d1 = *src--; |
| 300 | *dst-- = d0 >> right | d1 << left; |
| 301 | d0 = d1; |
| 302 | n -= 4; |
| 303 | } |
| 304 | while (n--) { |
| 305 | d1 = *src--; |
| 306 | *dst-- = d0 >> right | d1 << left; |
| 307 | d0 = d1; |
| 308 | } |
| 309 | |
| 310 | /* Trailing bits */ |
| 311 | if (last) { |
| 312 | if (m <= left) { |
| 313 | /* Single source word */ |
| 314 | *dst = comp(d0 >> right, *dst, last); |
| 315 | } else { |
| 316 | /* 2 source words */ |
| 317 | d1 = *src; |
| 318 | *dst = comp(d0 >> right | d1 << left, |
| 319 | *dst, last); |
| 320 | } |
| 321 | } |
| 322 | } |
| 323 | } |
| 324 | } |
| 325 | |
| 326 | void sys_copyarea(struct fb_info *p, const struct fb_copyarea *area) |
| 327 | { |
| 328 | u32 dx = area->dx, dy = area->dy, sx = area->sx, sy = area->sy; |
| 329 | u32 height = area->height, width = area->width; |
| 330 | unsigned long const bits_per_line = p->fix.line_length*8u; |
| 331 | unsigned long *dst = NULL, *src = NULL; |
| 332 | int bits = BITS_PER_LONG, bytes = bits >> 3; |
| 333 | int dst_idx = 0, src_idx = 0, rev_copy = 0; |
| 334 | |
| 335 | if (p->state != FBINFO_STATE_RUNNING) |
| 336 | return; |
| 337 | |
| 338 | /* if the beginning of the target area might overlap with the end of |
| 339 | the source area, be have to copy the area reverse. */ |
| 340 | if ((dy == sy && dx > sx) || (dy > sy)) { |
| 341 | dy += height; |
| 342 | sy += height; |
| 343 | rev_copy = 1; |
| 344 | } |
| 345 | |
| 346 | /* split the base of the framebuffer into a long-aligned address and |
| 347 | the index of the first bit */ |
| 348 | dst = src = (unsigned long *)((unsigned long)p->screen_base & |
| 349 | ~(bytes-1)); |
| 350 | dst_idx = src_idx = 8*((unsigned long)p->screen_base & (bytes-1)); |
| 351 | /* add offset of source and target area */ |
| 352 | dst_idx += dy*bits_per_line + dx*p->var.bits_per_pixel; |
| 353 | src_idx += sy*bits_per_line + sx*p->var.bits_per_pixel; |
| 354 | |
| 355 | if (p->fbops->fb_sync) |
| 356 | p->fbops->fb_sync(p); |
| 357 | |
| 358 | if (rev_copy) { |
| 359 | while (height--) { |
| 360 | dst_idx -= bits_per_line; |
| 361 | src_idx -= bits_per_line; |
| 362 | dst += dst_idx >> (ffs(bits) - 1); |
| 363 | dst_idx &= (bytes - 1); |
| 364 | src += src_idx >> (ffs(bits) - 1); |
| 365 | src_idx &= (bytes - 1); |
| 366 | bitcpy_rev(dst, dst_idx, src, src_idx, bits, |
| 367 | width*p->var.bits_per_pixel); |
| 368 | } |
| 369 | } else { |
| 370 | while (height--) { |
| 371 | dst += dst_idx >> (ffs(bits) - 1); |
| 372 | dst_idx &= (bytes - 1); |
| 373 | src += src_idx >> (ffs(bits) - 1); |
| 374 | src_idx &= (bytes - 1); |
| 375 | bitcpy(dst, dst_idx, src, src_idx, bits, |
| 376 | width*p->var.bits_per_pixel); |
| 377 | dst_idx += bits_per_line; |
| 378 | src_idx += bits_per_line; |
| 379 | } |
| 380 | } |
| 381 | } |
| 382 | |
| 383 | EXPORT_SYMBOL(sys_copyarea); |
| 384 | |
| 385 | MODULE_AUTHOR("Antonino Daplas <adaplas@pol.net>"); |
| 386 | MODULE_DESCRIPTION("Generic copyarea (sys-to-sys)"); |
| 387 | MODULE_LICENSE("GPL"); |
| 388 | |