Menu
▾
▴
Re: [Linux-fbdev-devel] Re: [PATCH]: cfb_imageblit() fix: handle widths not divisible by 8
|
From: Antonino D. <ad...@po...> - 2003-01-15 11:57:16
|
On Wed, 2003-01-15 at 17:28, Geert Uytterhoeven wrote:
> On 15 Jan 2003, Antonino Daplas wrote:
> > On Wed, 2003-01-15 at 08:26, James Simmons wrote:
> > > Applied.
> > >
> > > > c. Fix for fast_imageblit() so it always refer to mask tables in 32-bits
> > > > which should make it work for 64-bit machines.
> > >
> > > Ug. I rather try yo take advantge of using the full 64 bits of data to
> > > pass across the bus. What I was think is treat the 64 bit case as two 32
> > > bit cases. The 64 bit data comes in and we run the data twice at tabs[].
> > >
> > Hi James,
> >
> > Yes, I was trying to find a way to make fast_imageblit() be fast for all
> > machine architectures. With the patch attached, there's
> > fast_imageblit32() and fast_imageblit64(). fast_imageblit32() is
> > probably slower than fast_imageblit64 on 64-bit machines and, on the
> > other hand, fast_imageblit64() is 20% slower on 32-bit machines, but is
> > probably faster on 64-bit and higher machines. So, the only way I can
> > think of doing this on all machine architectures is to have them go
> > separate paths.
>
> Can't you merge fast_imageblit32() and fast_imageblit64() a bit more (with some
> #ifdef's), and just call the result fast_imageblit()? Then the definition of
> FAST_IMAGEBLIT can go away.
>
> u32 is the same as unsigned long if BITS_PER_LONG == 32.
>
That's true. I don't want to do the merge before you people have seen
it. Anyway, here's an updated one.
Tony
diff -Naur linux-2.5.56-fbdev/drivers/video/cfbimgblt.c linux/drivers/video/cfbimgblt.c
--- linux-2.5.56-fbdev/drivers/video/cfbimgblt.c 2003-01-15 01:56:47.000000000 +0000
+++ linux/drivers/video/cfbimgblt.c 2003-01-15 11:43:53.000000000 +0000
@@ -73,14 +73,6 @@
0x00000000, 0xffffffff
};
-#if BITS_PER_LONG == 32
-#define FB_WRITEL fb_writel
-#define FB_READL fb_readl
-#else
-#define FB_WRITEL fb_writeq
-#define FB_READL fb_readq
-#endif
-
#if defined (__BIG_ENDIAN)
#define LEFT_POS(bpp) (BITS_PER_LONG - bpp)
#define LEFT_POS32(bpp) (32 - bpp)
@@ -95,6 +87,28 @@
#define SHIFT_LOW(val, bits) ((val) >> (bits))
#endif
+#if BITS_PER_LONG == 32
+#define FB_WRITEL fb_writel
+#define FB_READL fb_readl
+#define DECLARE_FASTPATH {}
+#define INIT_FASTPATH {}
+#define FASTPATH fb_writel((end_mask & eorx)^bgx, dst++)
+#else
+#define FB_WRITEL fb_writeq
+#define FB_READL fb_readq
+#define DECLARE_FASTPATH unsigned long val, bpl
+#define INIT_FASTPATH { val = 0; bpl = 0; }
+#define FASTPATH { \
+ val |= SHIFT_HIGH((end_mask & eorx)^bgx, bpl); \
+ bpl += 32; \
+ bpl &= BITS_PER_LONG - 1; \
+ if (!bpl) { \
+ FB_WRITEL(val, dst++); \
+ val = 0; \
+ } \
+}
+#endif
+
static inline void color_imageblit(struct fb_image *image, struct fb_info *p,
u8 *dst1, unsigned long start_index,
unsigned long pitch_index)
@@ -242,10 +256,11 @@
u32 bit_mask, end_mask, eorx, shift;
u32 fgx = fgcolor, bgx = bgcolor, bpp = p->var.bits_per_pixel;
u32 ppw = 32/bpp, spitch = (image->width + 7)/8;
- u32 *dst;
u32 *tab = NULL;
+ unsigned long *dst;
char *s = image->data, *src;
-
+ DECLARE_FASTPATH;
+
switch (bpp) {
case 8:
tab = cfb_tab8;
@@ -270,18 +285,19 @@
k = image->width/ppw;
for (i = image->height; i--; ) {
- dst = (u32 *) dst1; shift = 8; src = s;
+ dst = (unsigned long *) dst1; shift = 8; src = s;
+ INIT_FASTPATH;
for (j = k; j--; ) {
shift -= ppw;
end_mask = tab[(*src >> shift) & bit_mask];
- fb_writel((end_mask & eorx)^bgx, dst++);
+ FASTPATH;
if (!shift) { shift = 8; src++; }
}
dst1 += p->fix.line_length;
s += spitch;
}
}
-
+
void cfb_imageblit(struct fb_info *p, struct fb_image *image)
{
int x2, y2, vxres, vyres;
@@ -331,7 +347,7 @@
if (BITS_PER_LONG % bpp == 0 && !start_index &&
!pitch_index && bpp >= 8 && bpp <= 32 &&
- (image->width & (32/bpp-1)) == 0)
+ (image->width & (BITS_PER_LONG/bpp-1)) == 0)
fast_imageblit(image, p, dst1, fgcolor, bgcolor);
else
slow_imageblit(image, p, dst1, fgcolor, bgcolor,
|
