On Sun, 3 Dec 2000, Andreas Beck wrote:
> > > To catch all cases you need a loop,
>
> > A loop over what? Over the rgb-compontents? Or over different colordepth?
>
> Over the width of the expanded version, with a stepsize determined by the
> significant bits of the to-be-expanded one.
>
> > > which is why I am worrying about performance.
> > What about unrolling the loop?
>
> You can't do that generically. You'd have to introduce lots of special cases
> for the different color schemes, then.
How about moving this stuff into linear_*/color.c ?
Then you'd have neither switch()'es for special cases nor a loop (if it is
unrolled).
> The algorithm for expanding a right-aligned "value" of length "nowlegth"
> to "exlength" is:
>
> res=0;
> for(a=exlength-nowlength;a>-nowlength;a-=nowlength)
> if (a>=0)
> res|=(value<<a);
> else
> res|=(value>>(-a));
If I got you right, it should _basically_ like the attached patch. I know, that
the patch is still wrong, because I tested it. So _don't_ commit it into CVS,
please!!
Christoph Egger
E-Mail: [EMAIL PROTECTED]
--- 20001126/degas/lib/libggi/default/color/color.c Mon Aug 30 07:17:36 1999
+++ degas/lib/libggi/default/color/color.c Sun Dec 3 14:36:24 2000
@@ -162,14 +162,32 @@
int GGI_color_TRUE_unmappixel(ggi_visual *vis, ggi_pixel pixel, ggi_color *col)
{
color_truepriv *priv = vis->colorpriv;
+ int a;
+
+#define EXPAND(a, exlength, nowlength, value, res) \
+ res = 0; \
+ for (a = exlength - nowlength; a > -nowlength; a -= nowlength) { \
+ if (a >= 0) { \
+ res |= (value << a); \
+ } else { \
+ res |= (value >> (-a)); \
+ } \
+ }
+
+
+ EXPAND(a, 16, 16 - priv->red_unmap, pixel & priv->red_mask, col->r);
+ EXPAND(a, 16, 16 - priv->green_unmap, pixel & priv->green_mask, col->g);
+ EXPAND(a, 16, 16 - priv->blue_unmap, pixel & priv->blue_mask, col->b);
+
+#if 0
col->r = DOSHIFT(pixel & priv->red_mask,
priv->red_unmap);
col->g = DOSHIFT(pixel & priv->green_mask,
priv->green_unmap);
col->b = DOSHIFT(pixel & priv->blue_mask,
priv->blue_unmap);
-
+#endif
return 0;
}
