New fb-based DDX, performance regressions relative to xfree86

[Timothy Normand Miller]

2009/12/15 Michel Dänzer <michel at daenzer.net>:

> If the old X server didn't have the damage layer yet, I suspect that
> might account for some of the overhead... Profiling with sysprof or
> oprofile might be interesting.

Yeah.  I've noticed that.  And I'm encountering something strange.
For wide PolyArc, I call miPolyArc, which decomposes the primitive
into spans.  If the rop is GXcopy, then my DDX's FillSpans gets
called.  If the rop is GXxor, it doesn't.  We're trying to figure out
why that might be happening.  In the mean time, can anyone give me
some insight into why my GC ops might get overridden?  I think
fbPolyArc is getting called by damage.


-- 
Timothy Normand Miller
http://www.cse.ohio-state.edu/~millerti
Open Graphics Project