[Bug 104271] i965: Timeout in dEQP-GLES31.functional.ssbo.layout.random.all_shared_buffer.5

bugzilla-daemon at freedesktop.org bugzilla-daemon at freedesktop.org
Fri Dec 15 21:23:07 UTC 2017


https://bugs.freedesktop.org/show_bug.cgi?id=104271

--- Comment #5 from Francisco Jerez <currojerez at riseup.net> ---
IMHO the ultimate reason for the regression is that the dataflow propagation
logic of the copy propagation pass has a serious run-time complexity problem in
acyclic CFGs.  The patch this was bisected to is merely increasing the amount
of work done in each iteration which substantially exacerbates the problem.  A
proper fix would involve reworking the copy propagation dataflow logic to
reduce the polynomial degree of its run-time complexity in the acyclic case. 
Working on it -- It will likely give you better performance than before my
undef handling patch.

Updating importance to high-priority enhancement since this is purely a
(largely pre-existing) compile-time performance issue, spec conformance is
unaffected by it.

-- 
You are receiving this mail because:
You are the QA Contact for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/intel-3d-bugs/attachments/20171215/5639172c/attachment.html>


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