[Bug 90056] Unigine Valley regression since radeon/llvm: Run LLVM's instruction combining pass

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=90056

--- Comment #12 from Tom Stellard <tstellar at gmail.com> ---
(In reply to Andy Furniss from comment #10)
> (In reply to Grigori Goronzy from comment #9)
> > Created attachment 115409 [details] [review] [review]
> > Full shader
> > 
> > That does not help either. Seems to break the SSA somehow. Full shader that
> > triggers the bug attached.
> > 
> > The patch I posted earlier help with the the reduced testcase, but not with
> > the full shader. Both undef incoming values in phi nodes and undef branch
> > conditions cause problems in different ways.
> 
> With the second patch + valley I get -
> 
> valley_x64: LiveVariables.cpp:114: void
> llvm::LiveVariables::MarkVirtRegAliveInBlock(llvm::LiveVariables::VarInfo &,
> llvm::MachineBasicBlock *, llvm::MachineBasicBlock *,
> std::vector<MachineBasicBlock *> &): Assertion `MBB != &MF->front() &&
> "Can't find reaching def for virtreg"' failed.

Did test with only the patch from comment #8 or did you test with both patches?

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/dri-devel/attachments/20150429/11659f30/attachment.html>