| 1 | /* |
| 2 | * Generic fillrect for frame buffers with packed pixels of any depth. |
| 3 | * |
| 4 | * Copyright (C) 2000 James Simmons (jsimmons@linux-fbdev.org) |
| 5 | * |
| 6 | * This file is subject to the terms and conditions of the GNU General Public |
| 7 | * License. See the file COPYING in the main directory of this archive for |
| 8 | * more details. |
| 9 | * |
| 10 | * NOTES: |
| 11 | * |
| 12 | * The code for depths like 24 that don't have integer number of pixels per |
| 13 | * long is broken and needs to be fixed. For now I turned these types of |
| 14 | * mode off. |
| 15 | * |
| 16 | * Also need to add code to deal with cards endians that are different than |
| 17 | * the native cpu endians. I also need to deal with MSB position in the word. |
| 18 | * |
| 19 | */ |
| 20 | #include <linux/module.h> |
| 21 | #include <linux/string.h> |
| 22 | #include <linux/fb.h> |
| 23 | #include <asm/types.h> |
| 24 | #include "fb_draw.h" |
| 25 | |
| 26 | #if BITS_PER_LONG == 32 |
| 27 | # define FB_WRITEL fb_writel |
| 28 | # define FB_READL fb_readl |
| 29 | #else |
| 30 | # define FB_WRITEL fb_writeq |
| 31 | # define FB_READL fb_readq |
| 32 | #endif |
| 33 | |
| 34 | /* |
| 35 | * Aligned pattern fill using 32/64-bit memory accesses |
| 36 | */ |
| 37 | |
| 38 | static void |
| 39 | bitfill_aligned(struct fb_info *p, unsigned long __iomem *dst, int dst_idx, |
| 40 | unsigned long pat, unsigned n, int bits, u32 bswapmask) |
| 41 | { |
| 42 | unsigned long first, last; |
| 43 | |
| 44 | if (!n) |
| 45 | return; |
| 46 | |
| 47 | first = fb_shifted_pixels_mask_long(p, dst_idx, bswapmask); |
| 48 | last = ~fb_shifted_pixels_mask_long(p, (dst_idx+n) % bits, bswapmask); |
| 49 | |
| 50 | if (dst_idx+n <= bits) { |
| 51 | // Single word |
| 52 | if (last) |
| 53 | first &= last; |
| 54 | FB_WRITEL(comp(pat, FB_READL(dst), first), dst); |
| 55 | } else { |
| 56 | // Multiple destination words |
| 57 | |
| 58 | // Leading bits |
| 59 | if (first!= ~0UL) { |
| 60 | FB_WRITEL(comp(pat, FB_READL(dst), first), dst); |
| 61 | dst++; |
| 62 | n -= bits - dst_idx; |
| 63 | } |
| 64 | |
| 65 | // Main chunk |
| 66 | n /= bits; |
| 67 | while (n >= 8) { |
| 68 | FB_WRITEL(pat, dst++); |
| 69 | FB_WRITEL(pat, dst++); |
| 70 | FB_WRITEL(pat, dst++); |
| 71 | FB_WRITEL(pat, dst++); |
| 72 | FB_WRITEL(pat, dst++); |
| 73 | FB_WRITEL(pat, dst++); |
| 74 | FB_WRITEL(pat, dst++); |
| 75 | FB_WRITEL(pat, dst++); |
| 76 | n -= 8; |
| 77 | } |
| 78 | while (n--) |
| 79 | FB_WRITEL(pat, dst++); |
| 80 | |
| 81 | // Trailing bits |
| 82 | if (last) |
| 83 | FB_WRITEL(comp(pat, FB_READL(dst), last), dst); |
| 84 | } |
| 85 | } |
| 86 | |
| 87 | |
| 88 | /* |
| 89 | * Unaligned generic pattern fill using 32/64-bit memory accesses |
| 90 | * The pattern must have been expanded to a full 32/64-bit value |
| 91 | * Left/right are the appropriate shifts to convert to the pattern to be |
| 92 | * used for the next 32/64-bit word |
| 93 | */ |
| 94 | |
| 95 | static void |
| 96 | bitfill_unaligned(struct fb_info *p, unsigned long __iomem *dst, int dst_idx, |
| 97 | unsigned long pat, int left, int right, unsigned n, int bits) |
| 98 | { |
| 99 | unsigned long first, last; |
| 100 | |
| 101 | if (!n) |
| 102 | return; |
| 103 | |
| 104 | first = FB_SHIFT_HIGH(p, ~0UL, dst_idx); |
| 105 | last = ~(FB_SHIFT_HIGH(p, ~0UL, (dst_idx+n) % bits)); |
| 106 | |
| 107 | if (dst_idx+n <= bits) { |
| 108 | // Single word |
| 109 | if (last) |
| 110 | first &= last; |
| 111 | FB_WRITEL(comp(pat, FB_READL(dst), first), dst); |
| 112 | } else { |
| 113 | // Multiple destination words |
| 114 | // Leading bits |
| 115 | if (first) { |
| 116 | FB_WRITEL(comp(pat, FB_READL(dst), first), dst); |
| 117 | dst++; |
| 118 | pat = pat << left | pat >> right; |
| 119 | n -= bits - dst_idx; |
| 120 | } |
| 121 | |
| 122 | // Main chunk |
| 123 | n /= bits; |
| 124 | while (n >= 4) { |
| 125 | FB_WRITEL(pat, dst++); |
| 126 | pat = pat << left | pat >> right; |
| 127 | FB_WRITEL(pat, dst++); |
| 128 | pat = pat << left | pat >> right; |
| 129 | FB_WRITEL(pat, dst++); |
| 130 | pat = pat << left | pat >> right; |
| 131 | FB_WRITEL(pat, dst++); |
| 132 | pat = pat << left | pat >> right; |
| 133 | n -= 4; |
| 134 | } |
| 135 | while (n--) { |
| 136 | FB_WRITEL(pat, dst++); |
| 137 | pat = pat << left | pat >> right; |
| 138 | } |
| 139 | |
| 140 | // Trailing bits |
| 141 | if (last) |
| 142 | FB_WRITEL(comp(pat, FB_READL(dst), first), dst); |
| 143 | } |
| 144 | } |
| 145 | |
| 146 | /* |
| 147 | * Aligned pattern invert using 32/64-bit memory accesses |
| 148 | */ |
| 149 | static void |
| 150 | bitfill_aligned_rev(struct fb_info *p, unsigned long __iomem *dst, |
| 151 | int dst_idx, unsigned long pat, unsigned n, int bits, |
| 152 | u32 bswapmask) |
| 153 | { |
| 154 | unsigned long val = pat, dat; |
| 155 | unsigned long first, last; |
| 156 | |
| 157 | if (!n) |
| 158 | return; |
| 159 | |
| 160 | first = fb_shifted_pixels_mask_long(p, dst_idx, bswapmask); |
| 161 | last = ~fb_shifted_pixels_mask_long(p, (dst_idx+n) % bits, bswapmask); |
| 162 | |
| 163 | if (dst_idx+n <= bits) { |
| 164 | // Single word |
| 165 | if (last) |
| 166 | first &= last; |
| 167 | dat = FB_READL(dst); |
| 168 | FB_WRITEL(comp(dat ^ val, dat, first), dst); |
| 169 | } else { |
| 170 | // Multiple destination words |
| 171 | // Leading bits |
| 172 | if (first!=0UL) { |
| 173 | dat = FB_READL(dst); |
| 174 | FB_WRITEL(comp(dat ^ val, dat, first), dst); |
| 175 | dst++; |
| 176 | n -= bits - dst_idx; |
| 177 | } |
| 178 | |
| 179 | // Main chunk |
| 180 | n /= bits; |
| 181 | while (n >= 8) { |
| 182 | FB_WRITEL(FB_READL(dst) ^ val, dst); |
| 183 | dst++; |
| 184 | FB_WRITEL(FB_READL(dst) ^ val, dst); |
| 185 | dst++; |
| 186 | FB_WRITEL(FB_READL(dst) ^ val, dst); |
| 187 | dst++; |
| 188 | FB_WRITEL(FB_READL(dst) ^ val, dst); |
| 189 | dst++; |
| 190 | FB_WRITEL(FB_READL(dst) ^ val, dst); |
| 191 | dst++; |
| 192 | FB_WRITEL(FB_READL(dst) ^ val, dst); |
| 193 | dst++; |
| 194 | FB_WRITEL(FB_READL(dst) ^ val, dst); |
| 195 | dst++; |
| 196 | FB_WRITEL(FB_READL(dst) ^ val, dst); |
| 197 | dst++; |
| 198 | n -= 8; |
| 199 | } |
| 200 | while (n--) { |
| 201 | FB_WRITEL(FB_READL(dst) ^ val, dst); |
| 202 | dst++; |
| 203 | } |
| 204 | // Trailing bits |
| 205 | if (last) { |
| 206 | dat = FB_READL(dst); |
| 207 | FB_WRITEL(comp(dat ^ val, dat, last), dst); |
| 208 | } |
| 209 | } |
| 210 | } |
| 211 | |
| 212 | |
| 213 | /* |
| 214 | * Unaligned generic pattern invert using 32/64-bit memory accesses |
| 215 | * The pattern must have been expanded to a full 32/64-bit value |
| 216 | * Left/right are the appropriate shifts to convert to the pattern to be |
| 217 | * used for the next 32/64-bit word |
| 218 | */ |
| 219 | |
| 220 | static void |
| 221 | bitfill_unaligned_rev(struct fb_info *p, unsigned long __iomem *dst, |
| 222 | int dst_idx, unsigned long pat, int left, int right, |
| 223 | unsigned n, int bits) |
| 224 | { |
| 225 | unsigned long first, last, dat; |
| 226 | |
| 227 | if (!n) |
| 228 | return; |
| 229 | |
| 230 | first = FB_SHIFT_HIGH(p, ~0UL, dst_idx); |
| 231 | last = ~(FB_SHIFT_HIGH(p, ~0UL, (dst_idx+n) % bits)); |
| 232 | |
| 233 | if (dst_idx+n <= bits) { |
| 234 | // Single word |
| 235 | if (last) |
| 236 | first &= last; |
| 237 | dat = FB_READL(dst); |
| 238 | FB_WRITEL(comp(dat ^ pat, dat, first), dst); |
| 239 | } else { |
| 240 | // Multiple destination words |
| 241 | |
| 242 | // Leading bits |
| 243 | if (first != 0UL) { |
| 244 | dat = FB_READL(dst); |
| 245 | FB_WRITEL(comp(dat ^ pat, dat, first), dst); |
| 246 | dst++; |
| 247 | pat = pat << left | pat >> right; |
| 248 | n -= bits - dst_idx; |
| 249 | } |
| 250 | |
| 251 | // Main chunk |
| 252 | n /= bits; |
| 253 | while (n >= 4) { |
| 254 | FB_WRITEL(FB_READL(dst) ^ pat, dst); |
| 255 | dst++; |
| 256 | pat = pat << left | pat >> right; |
| 257 | FB_WRITEL(FB_READL(dst) ^ pat, dst); |
| 258 | dst++; |
| 259 | pat = pat << left | pat >> right; |
| 260 | FB_WRITEL(FB_READL(dst) ^ pat, dst); |
| 261 | dst++; |
| 262 | pat = pat << left | pat >> right; |
| 263 | FB_WRITEL(FB_READL(dst) ^ pat, dst); |
| 264 | dst++; |
| 265 | pat = pat << left | pat >> right; |
| 266 | n -= 4; |
| 267 | } |
| 268 | while (n--) { |
| 269 | FB_WRITEL(FB_READL(dst) ^ pat, dst); |
| 270 | dst++; |
| 271 | pat = pat << left | pat >> right; |
| 272 | } |
| 273 | |
| 274 | // Trailing bits |
| 275 | if (last) { |
| 276 | dat = FB_READL(dst); |
| 277 | FB_WRITEL(comp(dat ^ pat, dat, last), dst); |
| 278 | } |
| 279 | } |
| 280 | } |
| 281 | |
| 282 | void cfb_fillrect(struct fb_info *p, const struct fb_fillrect *rect) |
| 283 | { |
| 284 | unsigned long pat, fg; |
| 285 | unsigned long width = rect->width, height = rect->height; |
| 286 | int bits = BITS_PER_LONG, bytes = bits >> 3; |
| 287 | u32 bpp = p->var.bits_per_pixel; |
| 288 | unsigned long __iomem *dst; |
| 289 | int dst_idx, left; |
| 290 | |
| 291 | if (p->state != FBINFO_STATE_RUNNING) |
| 292 | return; |
| 293 | |
| 294 | if (p->fix.visual == FB_VISUAL_TRUECOLOR || |
| 295 | p->fix.visual == FB_VISUAL_DIRECTCOLOR ) |
| 296 | fg = ((u32 *) (p->pseudo_palette))[rect->color]; |
| 297 | else |
| 298 | fg = rect->color; |
| 299 | |
| 300 | pat = pixel_to_pat( bpp, fg); |
| 301 | |
| 302 | dst = (unsigned long __iomem *)((unsigned long)p->screen_base & ~(bytes-1)); |
| 303 | dst_idx = ((unsigned long)p->screen_base & (bytes - 1))*8; |
| 304 | dst_idx += rect->dy*p->fix.line_length*8+rect->dx*bpp; |
| 305 | /* FIXME For now we support 1-32 bpp only */ |
| 306 | left = bits % bpp; |
| 307 | if (p->fbops->fb_sync) |
| 308 | p->fbops->fb_sync(p); |
| 309 | if (!left) { |
| 310 | u32 bswapmask = fb_compute_bswapmask(p); |
| 311 | void (*fill_op32)(struct fb_info *p, |
| 312 | unsigned long __iomem *dst, int dst_idx, |
| 313 | unsigned long pat, unsigned n, int bits, |
| 314 | u32 bswapmask) = NULL; |
| 315 | |
| 316 | switch (rect->rop) { |
| 317 | case ROP_XOR: |
| 318 | fill_op32 = bitfill_aligned_rev; |
| 319 | break; |
| 320 | case ROP_COPY: |
| 321 | fill_op32 = bitfill_aligned; |
| 322 | break; |
| 323 | default: |
| 324 | printk( KERN_ERR "cfb_fillrect(): unknown rop, defaulting to ROP_COPY\n"); |
| 325 | fill_op32 = bitfill_aligned; |
| 326 | break; |
| 327 | } |
| 328 | while (height--) { |
| 329 | dst += dst_idx >> (ffs(bits) - 1); |
| 330 | dst_idx &= (bits - 1); |
| 331 | fill_op32(p, dst, dst_idx, pat, width*bpp, bits, |
| 332 | bswapmask); |
| 333 | dst_idx += p->fix.line_length*8; |
| 334 | } |
| 335 | } else { |
| 336 | int right; |
| 337 | int r; |
| 338 | int rot = (left-dst_idx) % bpp; |
| 339 | void (*fill_op)(struct fb_info *p, unsigned long __iomem *dst, |
| 340 | int dst_idx, unsigned long pat, int left, |
| 341 | int right, unsigned n, int bits) = NULL; |
| 342 | |
| 343 | /* rotate pattern to correct start position */ |
| 344 | pat = pat << rot | pat >> (bpp-rot); |
| 345 | |
| 346 | right = bpp-left; |
| 347 | switch (rect->rop) { |
| 348 | case ROP_XOR: |
| 349 | fill_op = bitfill_unaligned_rev; |
| 350 | break; |
| 351 | case ROP_COPY: |
| 352 | fill_op = bitfill_unaligned; |
| 353 | break; |
| 354 | default: |
| 355 | printk( KERN_ERR "cfb_fillrect(): unknown rop, defaulting to ROP_COPY\n"); |
| 356 | fill_op = bitfill_unaligned; |
| 357 | break; |
| 358 | } |
| 359 | while (height--) { |
| 360 | dst += dst_idx >> (ffs(bits) - 1); |
| 361 | dst_idx &= (bits - 1); |
| 362 | fill_op(p, dst, dst_idx, pat, left, right, |
| 363 | width*bpp, bits); |
| 364 | r = (p->fix.line_length*8) % bpp; |
| 365 | pat = pat << (bpp-r) | pat >> r; |
| 366 | dst_idx += p->fix.line_length*8; |
| 367 | } |
| 368 | } |
| 369 | } |
| 370 | |
| 371 | EXPORT_SYMBOL(cfb_fillrect); |
| 372 | |
| 373 | MODULE_AUTHOR("James Simmons <jsimmons@users.sf.net>"); |
| 374 | MODULE_DESCRIPTION("Generic software accelerated fill rectangle"); |
| 375 | MODULE_LICENSE("GPL"); |