EXA text performance regression (Re: Initial benchmark of Glucose/i965 with expedite)
cworth at cworth.org
Wed Oct 24 15:34:55 PDT 2007
On Tue, 23 Oct 2007 12:33:00 +0200, Michel Dänzer wrote:
> On Fri, 2007-10-19 at 14:56 -0700, Carl Worth wrote:
> > http://cworth.org/glucose/i965/first_look/
> Did you get a chance to profile the EXA text rendering slowness? It
> might be related to my recent changes allowing partial migration of
> pixmap contents. They save migration overhead but may introduce region
> calculation overhead in some situations. If this is confirmed to be the
> problem here, there are some possible countermeasures for the latter.
I haven't done any further profiling on this, no.
I would have liked to have done some "git bisect" work, but with the
various inter-dependent modules, (and the various API changes between
them), there's no easy way to find intermediate stages of the modules
that build together[*].
I do have the various modules tagged locally on my machine in the old
state that gave about 100k glyphs/second. As I update each to the
latest version I'll try to see if I can learn anything more about the
[*] I'd like to explore constructing a git "supermodule" of the
various xorg git repositories that would make this easier. This would
also give us a single place to tag _everything_ for releases, and it
would also allow for things like a single branch name to describe
complex cross-module work such as glucose.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Size: 189 bytes
Desc: not available
More information about the xorg