[Bug 91857] Mesa 10.6.3 linker is slow

bugzilla-daemon at freedesktop.org bugzilla-daemon at freedesktop.org
Sun Sep 13 00:08:43 PDT 2015


https://bugs.freedesktop.org/show_bug.cgi?id=91857

--- Comment #25 from Kenneth Graunke <kenneth at whitecape.org> ---
(In reply to Jordan Justen from comment #12) 
> Mesa 10.6.4 release build: 44 seconds
> Mesa ~master debug build: 267 seconds

Jordan, FWIW, linking in debug builds will be obscenely slow because we call
nir_validate_shader() after every optimization/lowering pass.  In release
builds, that is a no-op...so measuring debug builds isn't terribly meaningful.

I landed a few fixes in master that should help out a little...the front-end
time is a lot more reasonable now.  FS register allocation is still horridly
slow, however.  This is because of our stupid spilling loop.  It takes 132
iterations through the RA loop before it succeeds.

We can totally fix that, we just haven't made it a priority yet...most shaders
don't trip up this pathological case.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20150913/6bb8277b/attachment.html>


More information about the intel-3d-bugs mailing list