> int over_n_dst_ps (int alpha, int color, int dst_color)
> {
> int q0 = 255 - alpha;
> int q1 = color * alpha + 0x80;
> return (dst_color * q0 + q1) >> 8;
> }
double over_n_dst_float (int alpha, int color, int dst_color)
{
return (dst_color * (0xff - alpha) + color * alpha)/255.0;
}
...
try vraddhn instead of vaddhn:
int over_n_dst_ps (int alpha, int color, int dst_color)
{
int q0 = 0xff - alpha;
int q1 = color * alpha + 0x80;
int u = dst_color * q0 + q1;
return (u + (u >> 8)) >> 8;
}
does not generate any asserts and
stddev_pixman = 0.405459
stddev_ps = 0.287262
As you can see my math isn't much different from pixman; it only doesn't suffer
from that precision loss in premultiply step of the source color.
_______________________________________________
Pixman mailing list
[email protected]
http://lists.freedesktop.org/mailman/listinfo/pixman
