[PATCH 17/41] fbdev: Add drawing functions for framebuffers in system RAM

[PATCH 17/41] fbdev: Add drawing functions for framebuffers in system RAM

[Antonino A. Daplas]

The generic drawing functions (cfbimgblt, cfbcopyarea, cfbfillrect) assume
that the framebuffer is in IO memory.  However, we have 3 drivers (hecubafb,
arcfb, and vfb) where the framebuffer is allocated from system RAM (via
vmalloc). Using _raw_read/write and family for these drivers (as used in
the cfb* functions) is illegal, especially in other platforms.

Create 3 new drawing functions, based almost entirely from the original
except that the framebuffer memory is assumed to be in system RAM.
These are named as sysimgblt, syscopyarea, and sysfillrect.

Signed-off-by: Antonino Daplas <adaplas@gmail.com>
---

 drivers/video/Kconfig       |   27 +++
 drivers/video/Makefile      |    3 
 drivers/video/syscopyarea.c |  388 ++++++++++++++++++++++++++++++++++++++++++
 drivers/video/sysfillrect.c |  400 +++++++++++++++++++++++++++++++++++++++++++
 drivers/video/sysimgblt.c   |  291 +++++++++++++++++++++++++++++++
 include/linux/fb.h          |    6 +
 6 files changed, 1115 insertions(+), 0 deletions(-)

diff --git a/drivers/video/Kconfig b/drivers/video/Kconfig
index 4f6ee88..234502d 100644
--- a/drivers/video/Kconfig
+++ b/drivers/video/Kconfig
@@ -95,6 +95,33 @@ config FB_CFB_IMAGEBLIT
 	  blitting. This is used by drivers that don't provide their own
 	  (accelerated) version.
 
+config FB_SYS_FILLRECT
+	tristate
+	depends on FB
+	default n
+	---help---
+	  Include the sys_fillrect function for generic software rectangle
+	  filling. This is used by drivers that don't provide their own
+	  (accelerated) version and the framebuffer is in system RAM.
+
+config FB_SYS_COPYAREA
+	tristate
+	depends on FB
+	default n
+	---help---
+	  Include the sys_copyarea function for generic software area copying.
+	  This is used by drivers that don't provide their own (accelerated)
+	  version and the framebuffer is in system RAM.
+
+config FB_SYS_IMAGEBLIT
+	tristate
+	depends on FB
+	default n
+	---help---
+	  Include the sys_imageblit function for generic software image
+	  blitting. This is used by drivers that don't provide their own
+	  (accelerated) version and the framebuffer is in system RAM.
+
 config FB_DEFERRED_IO
 	bool
 	depends on FB
diff --git a/drivers/video/Makefile b/drivers/video/Makefile
index 3ff0a70..a245cc7 100644
--- a/drivers/video/Makefile
+++ b/drivers/video/Makefile
@@ -18,6 +18,9 @@ obj-y				  += backlight/ display/
 obj-$(CONFIG_FB_CFB_FILLRECT)  += cfbfillrect.o
 obj-$(CONFIG_FB_CFB_COPYAREA)  += cfbcopyarea.o
 obj-$(CONFIG_FB_CFB_IMAGEBLIT) += cfbimgblt.o
+obj-$(CONFIG_FB_SYS_FILLRECT)  += sysfillrect.o
+obj-$(CONFIG_FB_SYS_COPYAREA)  += syscopyarea.o
+obj-$(CONFIG_FB_SYS_IMAGEBLIT) += sysimgblt.o
 obj-$(CONFIG_FB_SVGALIB)       += svgalib.o
 obj-$(CONFIG_FB_MACMODES)      += macmodes.o
 obj-$(CONFIG_FB_DDC)           += fb_ddc.o
diff --git a/drivers/video/syscopyarea.c b/drivers/video/syscopyarea.c
new file mode 100644
index 0000000..e348893
--- /dev/null
+++ b/drivers/video/syscopyarea.c
@@ -0,0 +1,388 @@
+/*
+ *  Generic Bit Block Transfer for frame buffers located in system RAM with
+ *  packed pixels of any depth.
+ *
+ *  Based almost entirely from cfbcopyarea.c (which is based almost entirely
+ *  on Geert Uytterhoeven's copyarea routine)
+ *
+ *      Copyright (C)  2007 Antonino Daplas <adaplas@pol.net>
+ *
+ *  This file is subject to the terms and conditions of the GNU General Public
+ *  License.  See the file COPYING in the main directory of this archive for
+ *  more details.
+ *
+ */
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/string.h>
+#include <linux/fb.h>
+#include <linux/slab.h>
+#include <asm/types.h>
+#include <asm/io.h>
+
+    /*
+     *  Compose two values, using a bitmask as decision value
+     *  This is equivalent to (a & mask) | (b & ~mask)
+     */
+
+static inline unsigned long
+comp(unsigned long a, unsigned long b, unsigned long mask)
+{
+    return ((a ^ b) & mask) ^ b;
+}
+
+    /*
+     *  Generic bitwise copy algorithm
+     */
+
+static void
+bitcpy(unsigned long *dst, int dst_idx, const unsigned long *src,
+	int src_idx, int bits, unsigned n)
+{
+	unsigned long first, last;
+	int const shift = dst_idx-src_idx;
+	int left, right;
+
+	first = FB_SHIFT_HIGH(~0UL, dst_idx);
+	last = ~(FB_SHIFT_HIGH(~0UL, (dst_idx+n) % bits));
+
+	if (!shift) {
+		/* Same alignment for source and dest */
+		if (dst_idx+n <= bits) {
+			/* Single word */
+			if (last)
+				first &= last;
+			*dst = comp(*src, *dst, first);
+		} else {
+			/* Multiple destination words */
+			/* Leading bits */
+ 			if (first != ~0UL) {
+				*dst = comp(*src, *dst, first);
+				dst++;
+				src++;
+				n -= bits - dst_idx;
+			}
+
+			/* Main chunk */
+			n /= bits;
+			while (n >= 8) {
+				*dst++ = *src++;
+				*dst++ = *src++;
+				*dst++ = *src++;
+				*dst++ = *src++;
+				*dst++ = *src++;
+				*dst++ = *src++;
+				*dst++ = *src++;
+				*dst++ = *src++;
+				n -= 8;
+			}
+			while (n--)
+				*dst++ = *src++;
+
+			/* Trailing bits */
+			if (last)
+				*dst = comp(*src, *dst, last);
+		}
+	} else {
+		unsigned long d0, d1;
+		int m;
+
+		/* Different alignment for source and dest */
+		right = shift & (bits - 1);
+		left = -shift & (bits - 1);
+
+		if (dst_idx+n <= bits) {
+			/* Single destination word */
+			if (last)
+				first &= last;
+			if (shift > 0) {
+				/* Single source word */
+				*dst = comp(*src >> right, *dst, first);
+			} else if (src_idx+n <= bits) {
+				/* Single source word */
+				*dst = comp(*src << left, *dst, first);
+			} else {
+				/* 2 source words */
+				d0 = *src++;
+				d1 = *src;
+				*dst = comp(d0 << left | d1 >> right, *dst,
+					    first);
+			}
+		} else {
+			/* Multiple destination words */
+			/** We must always remember the last value read,
+			    because in case SRC and DST overlap bitwise (e.g.
+			    when moving just one pixel in 1bpp), we always
+			    collect one full long for DST and that might
+			    overlap with the current long from SRC. We store
+			    this value in 'd0'. */
+			d0 = *src++;
+			/* Leading bits */
+			if (shift > 0) {
+				/* Single source word */
+				*dst = comp(d0 >> right, *dst, first);
+				dst++;
+				n -= bits - dst_idx;
+			} else {
+				/* 2 source words */
+				d1 = *src++;
+				*dst = comp(d0 << left | *dst >> right, *dst, first);
+				d0 = d1;
+				dst++;
+				n -= bits - dst_idx;
+			}
+
+			/* Main chunk */
+			m = n % bits;
+			n /= bits;
+			while (n >= 4) {
+				d1 = *src++;
+				*dst++ = d0 << left | d1 >> right;
+				d0 = d1;
+				d1 = *src++;
+				*dst++ = d0 << left | d1 >> right;
+				d0 = d1;
+				d1 = *src++;
+				*dst++ = d0 << left | d1 >> right;
+				d0 = d1;
+				d1 = *src++;
+				*dst++ = d0 << left | d1 >> right;
+				d0 = d1;
+				n -= 4;
+			}
+			while (n--) {
+				d1 = *src++;
+				*dst++ = d0 << left | d1 >> right;
+				d0 = d1;
+			}
+
+			/* Trailing bits */
+			if (last) {
+				if (m <= right) {
+					/* Single source word */
+					*dst = comp(d0 << left, *dst, last);
+				} else {
+					/* 2 source words */
+ 					d1 = *src;
+					*dst = comp(d0 << left | d1 >> right,
+						    *dst, last);
+				}
+			}
+		}
+	}
+}
+
+    /*
+     *  Generic bitwise copy algorithm, operating backward
+     */
+
+static void
+bitcpy_rev(unsigned long *dst, int dst_idx, const unsigned long *src,
+	   int src_idx, int bits, unsigned n)
+{
+	unsigned long first, last;
+	int shift;
+
+	dst += (n-1)/bits;
+	src += (n-1)/bits;
+	if ((n-1) % bits) {
+		dst_idx += (n-1) % bits;
+		dst += dst_idx >> (ffs(bits) - 1);
+		dst_idx &= bits - 1;
+		src_idx += (n-1) % bits;
+		src += src_idx >> (ffs(bits) - 1);
+		src_idx &= bits - 1;
+	}
+
+	shift = dst_idx-src_idx;
+
+	first = FB_SHIFT_LOW(~0UL, bits - 1 - dst_idx);
+	last = ~(FB_SHIFT_LOW(~0UL, bits - 1 - ((dst_idx-n) % bits)));
+
+	if (!shift) {
+		/* Same alignment for source and dest */
+		if ((unsigned long)dst_idx+1 >= n) {
+			/* Single word */
+			if (last)
+				first &= last;
+			*dst = comp(*src, *dst, first);
+		} else {
+			/* Multiple destination words */
+
+			/* Leading bits */
+			if (first != ~0UL) {
+				*dst = comp(*src, *dst, first);
+				dst--;
+				src--;
+				n -= dst_idx+1;
+			}
+
+			/* Main chunk */
+			n /= bits;
+			while (n >= 8) {
+				*dst-- = *src--;
+				*dst-- = *src--;
+				*dst-- = *src--;
+				*dst-- = *src--;
+				*dst-- = *src--;
+				*dst-- = *src--;
+				*dst-- = *src--;
+				*dst-- = *src--;
+				n -= 8;
+			}
+			while (n--)
+				*dst-- = *src--;
+			/* Trailing bits */
+			if (last)
+				*dst = comp(*src, *dst, last);
+		}
+	} else {
+		/* Different alignment for source and dest */
+
+		int const left = -shift & (bits-1);
+		int const right = shift & (bits-1);
+
+		if ((unsigned long)dst_idx+1 >= n) {
+			/* Single destination word */
+			if (last)
+				first &= last;
+			if (shift < 0) {
+				/* Single source word */
+				*dst = comp(*src << left, *dst, first);
+			} else if (1+(unsigned long)src_idx >= n) {
+				/* Single source word */
+				*dst = comp(*src >> right, *dst, first);
+			} else {
+				/* 2 source words */
+				*dst = comp(*src >> right | *(src-1) << left,
+					    *dst, first);
+			}
+		} else {
+			/* Multiple destination words */
+			/** We must always remember the last value read,
+			    because in case SRC and DST overlap bitwise (e.g.
+			    when moving just one pixel in 1bpp), we always
+			    collect one full long for DST and that might
+			    overlap with the current long from SRC. We store
+			    this value in 'd0'. */
+			unsigned long d0, d1;
+			int m;
+
+			d0 = *src--;
+			/* Leading bits */
+			if (shift < 0) {
+				/* Single source word */
+				*dst = comp(d0 << left, *dst, first);
+			} else {
+				/* 2 source words */
+				d1 = *src--;
+				*dst = comp(d0 >> right | d1 << left, *dst,
+					    first);
+				d0 = d1;
+			}
+			dst--;
+			n -= dst_idx+1;
+
+			/* Main chunk */
+			m = n % bits;
+			n /= bits;
+			while (n >= 4) {
+				d1 = *src--;
+				*dst-- = d0 >> right | d1 << left;
+				d0 = d1;
+				d1 = *src--;
+				*dst-- = d0 >> right | d1 << left;
+				d0 = d1;
+				d1 = *src--;
+				*dst-- = d0 >> right | d1 << left;
+				d0 = d1;
+				d1 = *src--;
+				*dst-- = d0 >> right | d1 << left;
+				d0 = d1;
+				n -= 4;
+			}
+			while (n--) {
+				d1 = *src--;
+				*dst-- = d0 >> right | d1 << left;
+				d0 = d1;
+			}
+
+			/* Trailing bits */
+			if (last) {
+				if (m <= left) {
+					/* Single source word */
+					*dst = comp(d0 >> right, *dst, last);
+				} else {
+					/* 2 source words */
+					d1 = *src;
+					*dst = comp(d0 >> right | d1 << left,
+						    *dst, last);
+				}
+			}
+		}
+	}
+}
+
+void sys_copyarea(struct fb_info *p, const struct fb_copyarea *area)
+{
+	u32 dx = area->dx, dy = area->dy, sx = area->sx, sy = area->sy;
+	u32 height = area->height, width = area->width;
+	unsigned long const bits_per_line = p->fix.line_length*8u;
+	unsigned long *dst = NULL, *src = NULL;
+	int bits = BITS_PER_LONG, bytes = bits >> 3;
+	int dst_idx = 0, src_idx = 0, rev_copy = 0;
+
+	if (p->state != FBINFO_STATE_RUNNING)
+		return;
+
+	/* if the beginning of the target area might overlap with the end of
+	the source area, be have to copy the area reverse. */
+	if ((dy == sy && dx > sx) || (dy > sy)) {
+		dy += height;
+		sy += height;
+		rev_copy = 1;
+	}
+
+	/* split the base of the framebuffer into a long-aligned address and
+	   the index of the first bit */
+	dst = src = (unsigned long *)((unsigned long)p->screen_base &
+				      ~(bytes-1));
+	dst_idx = src_idx = 8*((unsigned long)p->screen_base & (bytes-1));
+	/* add offset of source and target area */
+	dst_idx += dy*bits_per_line + dx*p->var.bits_per_pixel;
+	src_idx += sy*bits_per_line + sx*p->var.bits_per_pixel;
+
+	if (p->fbops->fb_sync)
+		p->fbops->fb_sync(p);
+
+	if (rev_copy) {
+		while (height--) {
+			dst_idx -= bits_per_line;
+			src_idx -= bits_per_line;
+			dst += dst_idx >> (ffs(bits) - 1);
+			dst_idx &= (bytes - 1);
+			src += src_idx >> (ffs(bits) - 1);
+			src_idx &= (bytes - 1);
+			bitcpy_rev(dst, dst_idx, src, src_idx, bits,
+				width*p->var.bits_per_pixel);
+		}
+	} else {
+		while (height--) {
+			dst += dst_idx >> (ffs(bits) - 1);
+			dst_idx &= (bytes - 1);
+			src += src_idx >> (ffs(bits) - 1);
+			src_idx &= (bytes - 1);
+			bitcpy(dst, dst_idx, src, src_idx, bits,
+				width*p->var.bits_per_pixel);
+			dst_idx += bits_per_line;
+			src_idx += bits_per_line;
+		}
+	}
+}
+
+EXPORT_SYMBOL(sys_copyarea);
+
+MODULE_AUTHOR("Antonino Daplas <adaplas@pol.net>");
+MODULE_DESCRIPTION("Generic copyarea (sys-to-sys)");
+MODULE_LICENSE("GPL");
+
diff --git a/drivers/video/sysfillrect.c b/drivers/video/sysfillrect.c
new file mode 100644
index 0000000..10de707
--- /dev/null
+++ b/drivers/video/sysfillrect.c
@@ -0,0 +1,400 @@
+/*
+ *  Generic fillrect for frame buffers in system RAM with packed pixels of
+ *  any depth.
+ *
+ *  Based almost entirely from cfbfillrect.c (which is based almost entirely
+ *  on Geert Uytterhoeven's fillrect routine)
+ *
+ *      Copyright (C)  2007 Antonino Daplas <adaplas@pol.net>
+ *
+ *  This file is subject to the terms and conditions of the GNU General Public
+ *  License.  See the file COPYING in the main directory of this archive for
+ *  more details.
+ */
+#include <linux/module.h>
+#include <linux/string.h>
+#include <linux/fb.h>
+#include <asm/types.h>
+
+    /*
+     *  Compose two values, using a bitmask as decision value
+     *  This is equivalent to (a & mask) | (b & ~mask)
+     */
+
+static inline unsigned long
+comp(unsigned long a, unsigned long b, unsigned long mask)
+{
+    return ((a ^ b) & mask) ^ b;
+}
+
+    /*
+     *  Create a pattern with the given pixel's color
+     */
+
+#if BITS_PER_LONG == 64
+static inline unsigned long
+pixel_to_pat( u32 bpp, u32 pixel)
+{
+	switch (bpp) {
+	case 1:
+		return 0xfffffffffffffffful*pixel;
+	case 2:
+		return 0x5555555555555555ul*pixel;
+	case 4:
+		return 0x1111111111111111ul*pixel;
+	case 8:
+		return 0x0101010101010101ul*pixel;
+	case 12:
+		return 0x0001001001001001ul*pixel;
+	case 16:
+		return 0x0001000100010001ul*pixel;
+	case 24:
+		return 0x0000000001000001ul*pixel;
+	case 32:
+		return 0x0000000100000001ul*pixel;
+	default:
+		panic("pixel_to_pat(): unsupported pixelformat\n");
+    }
+}
+#else
+static inline unsigned long
+pixel_to_pat( u32 bpp, u32 pixel)
+{
+	switch (bpp) {
+	case 1:
+		return 0xfffffffful*pixel;
+	case 2:
+		return 0x55555555ul*pixel;
+	case 4:
+		return 0x11111111ul*pixel;
+	case 8:
+		return 0x01010101ul*pixel;
+	case 12:
+		return 0x00001001ul*pixel;
+	case 16:
+		return 0x00010001ul*pixel;
+	case 24:
+		return 0x00000001ul*pixel;
+	case 32:
+		return 0x00000001ul*pixel;
+	default:
+		panic("pixel_to_pat(): unsupported pixelformat\n");
+    }
+}
+#endif
+
+    /*
+     *  Aligned pattern fill using 32/64-bit memory accesses
+     */
+
+static void
+bitfill_aligned(unsigned long *dst, int dst_idx, unsigned long pat,
+		unsigned n, int bits)
+{
+	unsigned long first, last;
+
+	if (!n)
+		return;
+
+	first = FB_SHIFT_HIGH(~0UL, dst_idx);
+	last = ~(FB_SHIFT_HIGH(~0UL, (dst_idx+n) % bits));
+
+	if (dst_idx+n <= bits) {
+		/* Single word */
+		if (last)
+			first &= last;
+		*dst = comp(pat, *dst, first);
+	} else {
+		/* Multiple destination words */
+
+		/* Leading bits */
+ 		if (first!= ~0UL) {
+			*dst = comp(pat, *dst, first);
+			dst++;
+			n -= bits - dst_idx;
+		}
+
+		/* Main chunk */
+		n /= bits;
+		while (n >= 8) {
+			*dst++ = pat;
+			*dst++ = pat;
+			*dst++ = pat;
+			*dst++ = pat;
+			*dst++ = pat;
+			*dst++ = pat;
+			*dst++ = pat;
+			*dst++ = pat;
+			n -= 8;
+		}
+		while (n--)
+			*dst++ = pat;
+		/* Trailing bits */
+		if (last)
+			*dst = comp(pat, *dst, last);
+	}
+}
+
+
+    /*
+     *  Unaligned generic pattern fill using 32/64-bit memory accesses
+     *  The pattern must have been expanded to a full 32/64-bit value
+     *  Left/right are the appropriate shifts to convert to the pattern to be
+     *  used for the next 32/64-bit word
+     */
+
+static void
+bitfill_unaligned(unsigned long *dst, int dst_idx, unsigned long pat,
+		  int left, int right, unsigned n, int bits)
+{
+	unsigned long first, last;
+
+	if (!n)
+		return;
+
+	first = FB_SHIFT_HIGH(~0UL, dst_idx);
+	last = ~(FB_SHIFT_HIGH(~0UL, (dst_idx+n) % bits));
+
+	if (dst_idx+n <= bits) {
+		/* Single word */
+		if (last)
+			first &= last;
+		*dst = comp(pat, *dst, first);
+	} else {
+		/* Multiple destination words */
+		/* Leading bits */
+		if (first) {
+			*dst = comp(pat, *dst, first);
+			dst++;
+			pat = pat << left | pat >> right;
+			n -= bits - dst_idx;
+		}
+
+		/* Main chunk */
+		n /= bits;
+		while (n >= 4) {
+			*dst++ = pat;
+			pat = pat << left | pat >> right;
+			*dst++ = pat;
+			pat = pat << left | pat >> right;
+			*dst++ = pat;
+			pat = pat << left | pat >> right;
+			*dst++ = pat;
+			pat = pat << left | pat >> right;
+			n -= 4;
+		}
+		while (n--) {
+			*dst++ = pat;
+			pat = pat << left | pat >> right;
+		}
+
+		/* Trailing bits */
+		if (last)
+			*dst = comp(pat, *dst, first);
+	}
+}
+
+    /*
+     *  Aligned pattern invert using 32/64-bit memory accesses
+     */
+static void
+bitfill_aligned_rev(unsigned long *dst, int dst_idx, unsigned long pat,
+		    unsigned n, int bits)
+{
+	unsigned long val = pat;
+	unsigned long first, last;
+
+	if (!n)
+		return;
+
+	first = FB_SHIFT_HIGH(~0UL, dst_idx);
+	last = ~(FB_SHIFT_HIGH(~0UL, (dst_idx+n) % bits));
+
+	if (dst_idx+n <= bits) {
+		/* Single word */
+		if (last)
+			first &= last;
+		*dst = comp(*dst ^ val, *dst, first);
+	} else {
+		/* Multiple destination words */
+		/* Leading bits */
+		if (first!=0UL) {
+			*dst = comp(*dst ^ val, *dst, first);
+			dst++;
+			n -= bits - dst_idx;
+		}
+
+		/* Main chunk */
+		n /= bits;
+		while (n >= 8) {
+			*dst++ ^= val;
+			*dst++ ^= val;
+			*dst++ ^= val;
+			*dst++ ^= val;
+			*dst++ ^= val;
+			*dst++ ^= val;
+			*dst++ ^= val;
+			*dst++ ^= val;
+			n -= 8;
+		}
+		while (n--)
+			*dst++ ^= val;
+		/* Trailing bits */
+		if (last)
+			*dst = comp(*dst ^ val, *dst, last);
+	}
+}
+
+
+    /*
+     *  Unaligned generic pattern invert using 32/64-bit memory accesses
+     *  The pattern must have been expanded to a full 32/64-bit value
+     *  Left/right are the appropriate shifts to convert to the pattern to be
+     *  used for the next 32/64-bit word
+     */
+
+static void
+bitfill_unaligned_rev(unsigned long *dst, int dst_idx, unsigned long pat,
+			int left, int right, unsigned n, int bits)
+{
+	unsigned long first, last;
+
+	if (!n)
+		return;
+
+	first = FB_SHIFT_HIGH(~0UL, dst_idx);
+	last = ~(FB_SHIFT_HIGH(~0UL, (dst_idx+n) % bits));
+
+	if (dst_idx+n <= bits) {
+		/* Single word */
+		if (last)
+			first &= last;
+		*dst = comp(*dst ^ pat, *dst, first);
+	} else {
+		/* Multiple destination words */
+
+		/* Leading bits */
+		if (first != 0UL) {
+			*dst = comp(*dst ^ pat, *dst, first);
+			dst++;
+			pat = pat << left | pat >> right;
+			n -= bits - dst_idx;
+		}
+
+		/* Main chunk */
+		n /= bits;
+		while (n >= 4) {
+			*dst++ ^= pat;
+			pat = pat << left | pat >> right;
+			*dst++ ^= pat;
+			pat = pat << left | pat >> right;
+			*dst++ ^= pat;
+			pat = pat << left | pat >> right;
+			*dst++ ^= pat;
+			pat = pat << left | pat >> right;
+			n -= 4;
+		}
+		while (n--) {
+			*dst ^= pat;
+			pat = pat << left | pat >> right;
+		}
+
+		/* Trailing bits */
+		if (last)
+			*dst = comp(*dst ^ pat, *dst, last);
+	}
+}
+
+void sys_fillrect(struct fb_info *p, const struct fb_fillrect *rect)
+{
+	unsigned long pat, fg;
+	unsigned long width = rect->width, height = rect->height;
+	int bits = BITS_PER_LONG, bytes = bits >> 3;
+	u32 bpp = p->var.bits_per_pixel;
+	unsigned long *dst;
+	int dst_idx, left;
+
+	if (p->state != FBINFO_STATE_RUNNING)
+		return;
+
+	if (p->fix.visual == FB_VISUAL_TRUECOLOR ||
+	    p->fix.visual == FB_VISUAL_DIRECTCOLOR )
+		fg = ((u32 *) (p->pseudo_palette))[rect->color];
+	else
+		fg = rect->color;
+
+	pat = pixel_to_pat( bpp, fg);
+
+	dst = (unsigned long *)((unsigned long)p->screen_base & ~(bytes-1));
+	dst_idx = ((unsigned long)p->screen_base & (bytes - 1))*8;
+	dst_idx += rect->dy*p->fix.line_length*8+rect->dx*bpp;
+	/* FIXME For now we support 1-32 bpp only */
+	left = bits % bpp;
+	if (p->fbops->fb_sync)
+		p->fbops->fb_sync(p);
+	if (!left) {
+		void (*fill_op32)(unsigned long *dst, int dst_idx,
+		                  unsigned long pat, unsigned n, int bits) =
+			NULL;
+
+		switch (rect->rop) {
+		case ROP_XOR:
+			fill_op32 = bitfill_aligned_rev;
+			break;
+		case ROP_COPY:
+			fill_op32 = bitfill_aligned;
+			break;
+		default:
+			printk( KERN_ERR "cfb_fillrect(): unknown rop, "
+				"defaulting to ROP_COPY\n");
+			fill_op32 = bitfill_aligned;
+			break;
+		}
+		while (height--) {
+			dst += dst_idx >> (ffs(bits) - 1);
+			dst_idx &= (bits - 1);
+			fill_op32(dst, dst_idx, pat, width*bpp, bits);
+			dst_idx += p->fix.line_length*8;
+		}
+	} else {
+		int right;
+		int r;
+		int rot = (left-dst_idx) % bpp;
+		void (*fill_op)(unsigned long *dst, int dst_idx,
+		                unsigned long pat, int left, int right,
+		                unsigned n, int bits) = NULL;
+
+		/* rotate pattern to correct start position */
+		pat = pat << rot | pat >> (bpp-rot);
+
+		right = bpp-left;
+		switch (rect->rop) {
+		case ROP_XOR:
+			fill_op = bitfill_unaligned_rev;
+			break;
+		case ROP_COPY:
+			fill_op = bitfill_unaligned;
+			break;
+		default:
+			printk(KERN_ERR "cfb_fillrect(): unknown rop, "
+				"defaulting to ROP_COPY\n");
+			fill_op = bitfill_unaligned;
+			break;
+		}
+		while (height--) {
+			dst += dst_idx >> (ffs(bits) - 1);
+			dst_idx &= (bits - 1);
+			fill_op(dst, dst_idx, pat, left, right,
+				width*bpp, bits);
+			r = (p->fix.line_length*8) % bpp;
+			pat = pat << (bpp-r) | pat >> r;
+			dst_idx += p->fix.line_length*8;
+		}
+	}
+}
+
+EXPORT_SYMBOL(sys_fillrect);
+
+MODULE_AUTHOR("Antonino Daplas <adaplas@pol.net>");
+MODULE_DESCRIPTION("Generic fill rectangle (sys-to-sys)");
+MODULE_LICENSE("GPL");
diff --git a/drivers/video/sysimgblt.c b/drivers/video/sysimgblt.c
new file mode 100644
index 0000000..bd7e7e9
--- /dev/null
+++ b/drivers/video/sysimgblt.c
@@ -0,0 +1,291 @@
+/*
+ *  Generic 1-bit or 8-bit source to 1-32 bit destination expansion
+ *  for frame buffer located in system RAM with packed pixels of any depth.
+ *
+ *  Based almost entirely on cfbimgblt.c
+ *
+ *      Copyright (C)  April 2007 Antonino Daplas <adaplas@pol.net>
+ *
+ *  This file is subject to the terms and conditions of the GNU General Public
+ *  License.  See the file COPYING in the main directory of this archive for
+ *  more details.
+ */
+#include <linux/module.h>
+#include <linux/string.h>
+#include <linux/fb.h>
+#include <asm/types.h>
+
+#define DEBUG
+
+#ifdef DEBUG
+#define DPRINTK(fmt, args...) printk(KERN_DEBUG "%s: " fmt,__FUNCTION__,## args)
+#else
+#define DPRINTK(fmt, args...)
+#endif
+
+static const u32 cfb_tab8[] = {
+#if defined(__BIG_ENDIAN)
+    0x00000000,0x000000ff,0x0000ff00,0x0000ffff,
+    0x00ff0000,0x00ff00ff,0x00ffff00,0x00ffffff,
+    0xff000000,0xff0000ff,0xff00ff00,0xff00ffff,
+    0xffff0000,0xffff00ff,0xffffff00,0xffffffff
+#elif defined(__LITTLE_ENDIAN)
+    0x00000000,0xff000000,0x00ff0000,0xffff0000,
+    0x0000ff00,0xff00ff00,0x00ffff00,0xffffff00,
+    0x000000ff,0xff0000ff,0x00ff00ff,0xffff00ff,
+    0x0000ffff,0xff00ffff,0x00ffffff,0xffffffff
+#else
+#error FIXME: No endianness??
+#endif
+};
+
+static const u32 cfb_tab16[] = {
+#if defined(__BIG_ENDIAN)
+    0x00000000, 0x0000ffff, 0xffff0000, 0xffffffff
+#elif defined(__LITTLE_ENDIAN)
+    0x00000000, 0xffff0000, 0x0000ffff, 0xffffffff
+#else
+#error FIXME: No endianness??
+#endif
+};
+
+static const u32 cfb_tab32[] = {
+	0x00000000, 0xffffffff
+};
+
+static void color_imageblit(const struct fb_image *image, struct fb_info *p,
+			    void *dst1, u32 start_index, u32 pitch_index)
+{
+	/* Draw the penguin */
+	u32 *dst, *dst2;
+	u32 color = 0, val, shift;
+	int i, n, bpp = p->var.bits_per_pixel;
+	u32 null_bits = 32 - bpp;
+	u32 *palette = (u32 *) p->pseudo_palette;
+	const u8 *src = image->data;
+
+	dst2 = dst1;
+	for (i = image->height; i--; ) {
+		n = image->width;
+		dst = dst1;
+		shift = 0;
+		val = 0;
+
+		if (start_index) {
+			u32 start_mask = ~(FB_SHIFT_HIGH(~(u32)0,
+							 start_index));
+			val = *dst & start_mask;
+			shift = start_index;
+		}
+		while (n--) {
+			if (p->fix.visual == FB_VISUAL_TRUECOLOR ||
+			    p->fix.visual == FB_VISUAL_DIRECTCOLOR )
+				color = palette[*src];
+			else
+				color = *src;
+			color <<= FB_LEFT_POS(bpp);
+			val |= FB_SHIFT_HIGH(color, shift);
+			if (shift >= null_bits) {
+				*dst++ = val;
+
+				val = (shift == null_bits) ? 0 :
+					FB_SHIFT_LOW(color, 32 - shift);
+			}
+			shift += bpp;
+			shift &= (32 - 1);
+			src++;
+		}
+		if (shift) {
+			u32 end_mask = FB_SHIFT_HIGH(~(u32)0, shift);
+
+			*dst &= end_mask;
+			*dst |= val;
+		}
+		dst1 += p->fix.line_length;
+		if (pitch_index) {
+			dst2 += p->fix.line_length;
+			dst1 = (u8 *)((long)dst2 & ~(sizeof(u32) - 1));
+
+			start_index += pitch_index;
+			start_index &= 32 - 1;
+		}
+	}
+}
+
+static void slow_imageblit(const struct fb_image *image, struct fb_info *p,
+				  void *dst1, u32 fgcolor, u32 bgcolor,
+				  u32 start_index, u32 pitch_index)
+{
+	u32 shift, color = 0, bpp = p->var.bits_per_pixel;
+	u32 *dst, *dst2;
+	u32 val, pitch = p->fix.line_length;
+	u32 null_bits = 32 - bpp;
+	u32 spitch = (image->width+7)/8;
+	const u8 *src = image->data, *s;
+	u32 i, j, l;
+
+	dst2 = dst1;
+	fgcolor <<= FB_LEFT_POS(bpp);
+	bgcolor <<= FB_LEFT_POS(bpp);
+
+	for (i = image->height; i--; ) {
+		shift = val = 0;
+		l = 8;
+		j = image->width;
+		dst = dst1;
+		s = src;
+
+		/* write leading bits */
+		if (start_index) {
+			u32 start_mask = ~(FB_SHIFT_HIGH(~(u32)0,start_index));
+			val = *dst & start_mask;
+			shift = start_index;
+		}
+
+		while (j--) {
+			l--;
+			color = (*s & (1 << l)) ? fgcolor : bgcolor;
+			val |= FB_SHIFT_HIGH(color, shift);
+
+			/* Did the bitshift spill bits to the next long? */
+			if (shift >= null_bits) {
+				*dst++ = val;
+				val = (shift == null_bits) ? 0 :
+					FB_SHIFT_LOW(color,32 - shift);
+			}
+			shift += bpp;
+			shift &= (32 - 1);
+			if (!l) { l = 8; s++; };
+		}
+
+		/* write trailing bits */
+ 		if (shift) {
+			u32 end_mask = FB_SHIFT_HIGH(~(u32)0, shift);
+
+			*dst &= end_mask;
+			*dst |= val;
+		}
+
+		dst1 += pitch;
+		src += spitch;
+		if (pitch_index) {
+			dst2 += pitch;
+			dst1 = (u8 *)((long)dst2 & ~(sizeof(u32) - 1));
+			start_index += pitch_index;
+			start_index &= 32 - 1;
+		}
+
+	}
+}
+
+/*
+ * fast_imageblit - optimized monochrome color expansion
+ *
+ * Only if:  bits_per_pixel == 8, 16, or 32
+ *           image->width is divisible by pixel/dword (ppw);
+ *           fix->line_legth is divisible by 4;
+ *           beginning and end of a scanline is dword aligned
+ */
+static void fast_imageblit(const struct fb_image *image, struct fb_info *p,
+				  void *dst1, u32 fgcolor, u32 bgcolor)
+{
+	u32 fgx = fgcolor, bgx = bgcolor, bpp = p->var.bits_per_pixel;
+	u32 ppw = 32/bpp, spitch = (image->width + 7)/8;
+	u32 bit_mask, end_mask, eorx, shift;
+	const char *s = image->data, *src;
+	u32 *dst;
+	const u32 *tab = NULL;
+	int i, j, k;
+
+	switch (bpp) {
+	case 8:
+		tab = cfb_tab8;
+		break;
+	case 16:
+		tab = cfb_tab16;
+		break;
+	case 32:
+	default:
+		tab = cfb_tab32;
+		break;
+	}
+
+	for (i = ppw-1; i--; ) {
+		fgx <<= bpp;
+		bgx <<= bpp;
+		fgx |= fgcolor;
+		bgx |= bgcolor;
+	}
+
+	bit_mask = (1 << ppw) - 1;
+	eorx = fgx ^ bgx;
+	k = image->width/ppw;
+
+	for (i = image->height; i--; ) {
+		dst = dst1;
+		shift = 8;
+		src = s;
+
+		for (j = k; j--; ) {
+			shift -= ppw;
+			end_mask = tab[(*src >> shift) & bit_mask];
+			*dst++ = (end_mask & eorx) ^ bgx;
+			if (!shift) {
+				shift = 8;
+				src++;
+			}
+		}
+		dst1 += p->fix.line_length;
+		s += spitch;
+	}
+}
+
+void sys_imageblit(struct fb_info *p, const struct fb_image *image)
+{
+	u32 fgcolor, bgcolor, start_index, bitstart, pitch_index = 0;
+	u32 bpl = sizeof(u32), bpp = p->var.bits_per_pixel;
+	u32 width = image->width;
+	u32 dx = image->dx, dy = image->dy;
+	void *dst1;
+
+	if (p->state != FBINFO_STATE_RUNNING)
+		return;
+
+	bitstart = (dy * p->fix.line_length * 8) + (dx * bpp);
+	start_index = bitstart & (32 - 1);
+	pitch_index = (p->fix.line_length & (bpl - 1)) * 8;
+
+	bitstart /= 8;
+	bitstart &= ~(bpl - 1);
+	dst1 = (void __force *)p->screen_base + bitstart;
+
+	if (p->fbops->fb_sync)
+		p->fbops->fb_sync(p);
+
+	if (image->depth == 1) {
+		if (p->fix.visual == FB_VISUAL_TRUECOLOR ||
+		    p->fix.visual == FB_VISUAL_DIRECTCOLOR) {
+			fgcolor = ((u32*)(p->pseudo_palette))[image->fg_color];
+			bgcolor = ((u32*)(p->pseudo_palette))[image->bg_color];
+		} else {
+			fgcolor = image->fg_color;
+			bgcolor = image->bg_color;
+		}
+
+		if (32 % bpp == 0 && !start_index && !pitch_index &&
+		    ((width & (32/bpp-1)) == 0) &&
+		    bpp >= 8 && bpp <= 32)
+			fast_imageblit(image, p, dst1, fgcolor, bgcolor);
+		else
+			slow_imageblit(image, p, dst1, fgcolor, bgcolor,
+					start_index, pitch_index);
+	} else
+		color_imageblit(image, p, dst1, start_index, pitch_index);
+}
+
+EXPORT_SYMBOL(sys_imageblit);
+
+MODULE_AUTHOR("Antonino Daplas <adaplas@pol.net>");
+MODULE_DESCRIPTION("1-bit/8-bit to 1-32 bit color expansion (sys-to-sys)");
+MODULE_LICENSE("GPL");
+
diff --git a/include/linux/fb.h b/include/linux/fb.h
index 0de4260..c4e1b3d 100644
--- a/include/linux/fb.h
+++ b/include/linux/fb.h
@@ -895,6 +895,12 @@ extern int fb_blank(struct fb_info *info
 extern void cfb_fillrect(struct fb_info *info, const struct fb_fillrect *rect); 
 extern void cfb_copyarea(struct fb_info *info, const struct fb_copyarea *area); 
 extern void cfb_imageblit(struct fb_info *info, const struct fb_image *image);
+/*
+ * Drawing operations where framebuffer is in system RAM
+ */
+extern void sys_fillrect(struct fb_info *info, const struct fb_fillrect *rect);
+extern void sys_copyarea(struct fb_info *info, const struct fb_copyarea *area);
+extern void sys_imageblit(struct fb_info *info, const struct fb_image *image);
 
 /* drivers/video/fbmem.c */
 extern int register_framebuffer(struct fb_info *fb_info);



-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

Re: [PATCH 17/41] fbdev: Add drawing functions forframebuffers in system RAM

[James Simmons]

> 	Hi Tony,
> 
> > The generic drawing functions (cfbimgblt, cfbcopyarea, 
> > cfbfillrect) assume
> > that the framebuffer is in IO memory.  However, we have 3 
> > drivers (hecubafb,
> > arcfb, and vfb) where the framebuffer is allocated from 
> > system RAM (via
> > vmalloc). Using _raw_read/write and family for these drivers 
> > (as used in
> > the cfb* functions) is illegal, especially in other platforms.
> 
> This is also true for ps3fb. I'll update the support for that once this
> patch has hit mainline.

Bloat. It would be much easier to replace the FB_WRITEL and FB_READL 
macros. Last year I was working on a general solution that I wanted to 
submit but I have been pushing other patches I considered higher priority.
Perhaps I should move them up the que.


-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

Re: [PATCH 17/41] fbdev: Add drawing functions forframebuffers in system RAM

[Geert Uytterhoeven]

On Wed, 2 May 2007, James Simmons wrote:
> > > The generic drawing functions (cfbimgblt, cfbcopyarea, 
> > > cfbfillrect) assume
> > > that the framebuffer is in IO memory.  However, we have 3 
> > > drivers (hecubafb,
> > > arcfb, and vfb) where the framebuffer is allocated from 
> > > system RAM (via
> > > vmalloc). Using _raw_read/write and family for these drivers 
> > > (as used in
> > > the cfb* functions) is illegal, especially in other platforms.
> > 
> > This is also true for ps3fb. I'll update the support for that once this
> > patch has hit mainline.
> 
> Bloat. It would be much easier to replace the FB_WRITEL and FB_READL 
> macros. Last year I was working on a general solution that I wanted to 
> submit but I have been pushing other patches I considered higher priority.
> Perhaps I should move them up the que.

That's true: just replace those macros and recompile...

Gr{oetje,eeting}s,

						Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
							    -- Linus Torvalds

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

Re: [PATCH 17/41] fbdev: Add drawing functions forframebuffers in system RAM

[Antonino A. Daplas]

On Wed, 2007-05-02 at 21:22 +0100, James Simmons wrote:
> > 	Hi Tony,
> > 
> > > The generic drawing functions (cfbimgblt, cfbcopyarea, 
> > > cfbfillrect) assume
> > > that the framebuffer is in IO memory.  However, we have 3 
> > > drivers (hecubafb,
> > > arcfb, and vfb) where the framebuffer is allocated from 
> > > system RAM (via
> > > vmalloc). Using _raw_read/write and family for these drivers 
> > > (as used in
> > > the cfb* functions) is illegal, especially in other platforms.
> > 
> > This is also true for ps3fb. I'll update the support for that once this
> > patch has hit mainline.
> 
> Bloat.

Only the source gets larger.  Only drivers that need it will compile it.

>  It would be much easier to replace the FB_WRITEL and FB_READL 
> macros.

Of course it's not as simple as replacing FB_WRITEL and FB_READL. Think
very carefully. 

Hint1: #define FB_MAX 32
Hint2: sparse

Tony  



-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

Re: [PATCH 17/41] fbdev: Add drawing functions forframebuffers in system RAM

[James Simmons]

> > > 	Hi Tony,
> > > 
> > > > The generic drawing functions (cfbimgblt, cfbcopyarea, 
> > > > cfbfillrect) assume
> > > > that the framebuffer is in IO memory.  However, we have 3 
> > > > drivers (hecubafb,
> > > > arcfb, and vfb) where the framebuffer is allocated from 
> > > > system RAM (via
> > > > vmalloc). Using _raw_read/write and family for these drivers 
> > > > (as used in
> > > > the cfb* functions) is illegal, especially in other platforms.
> > > 
> > > This is also true for ps3fb. I'll update the support for that once this
> > > patch has hit mainline.
> > 
> > Bloat.
> 
> Only the source gets larger.  Only drivers that need it will compile it.
> 
> >  It would be much easier to replace the FB_WRITEL and FB_READL 
> > macros.
> 
> Of course it's not as simple as replacing FB_WRITEL and FB_READL. Think
> very carefully. 
> 
> Hint1: #define FB_MAX 32
> Hint2: sparse

Yeap. fbwrite/fbread are not flexable enough. The is way I put hooks into
struct fb_pixmap for writing/reading the framebuffer.

-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/

Re: [PATCH 17/41] fbdev: Add drawing functions forframebuffers in system RAM

[Antonino A. Daplas]

On Wed, 2007-05-02 at 21:55 +0100, James Simmons wrote:
> > > > 	Hi Tony,
> > > > 
> > > > > The generic drawing functions (cfbimgblt, cfbcopyarea, 
> > > > > cfbfillrect) assume
> > > > > that the framebuffer is in IO memory.  However, we have 3 
> > > > > drivers (hecubafb,
> > > > > arcfb, and vfb) where the framebuffer is allocated from 
> > > > > system RAM (via
> > > > > vmalloc). Using _raw_read/write and family for these drivers 
> > > > > (as used in
> > > > > the cfb* functions) is illegal, especially in other platforms.
> > > > 
> > > > This is also true for ps3fb. I'll update the support for that once this
> > > > patch has hit mainline.
> > > 
> > > Bloat.
> > 
> > Only the source gets larger.  Only drivers that need it will compile it.
> > 
> > >  It would be much easier to replace the FB_WRITEL and FB_READL 
> > > macros.
> > 
> > Of course it's not as simple as replacing FB_WRITEL and FB_READL. Think
> > very carefully. 
> > 
> > Hint1: #define FB_MAX 32
> > Hint2: sparse
> 
> Yeap. fbwrite/fbread are not flexable enough. The is way I put hooks into
> struct fb_pixmap for writing/reading the framebuffer.

I did try that too, but the drawing functions are part of the innermost
loop, in an extremely hot path.  Replacing those macros with function
pointers will drop the performance considerably.  Remember the
memcpy/for-loop replacement?  The speed degradation was significant and
it's not even in the hottest path.

Secondly, how would you reconcile sparse annotations?

Tony



-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/