[Mesa-dev] [PIGLIT,radeonsi] crash in spec/glsl-1.50/execution/geometry/max-input-components – who's bug is it?

[Michel Dänzer]

On 23.03.2014 04:53, Kai Wasserbäch wrote:
> Dear Mesa devs,
> I'm not sure whether this is a bug in Mesa, LLVM or in eglibc. The crash happens
> in _int_malloc, but since that is certainly one of the more often used
> functions, I'm not yet convinced, the fault lies indeed with eglibc.
> 
> Therefore I'm attaching the full backtrace of the crash in
> spec/glsl-1.50/execution/geometry/max-input-components (it takes a very long
> time until the crash actually happens, Piglit recorded an execution time of
> 1538.0506579875946) and hope you can point me to the right bug tracker.
> 
> I'm unable to tell, whether this is a regression or not, since today was the
> first time I was able to run a full Piglit quick test, without crashing my X on
> this machine with the radeonsi.

It's not a regression. If you build LLVM with assertions enabled, you get:

shader_runner:
/home/daenzer/src/llvm-git/llvm/lib/CodeGen/RegAllocGreedy.cpp:2268:
unsigned int
{anonymous}::RAGreedy::selectOrSplitImpl(llvm::LiveInterval&,
llvm::SmallVectorImpl<unsigned int>&,
{anonymous}::RAGreedy::SmallVirtRegSet&, unsigned int): Assertion
`NewVRegs.empty() && "Cannot append to existing NewVRegs"' failed.

So this is an LLVM issue. It might be worth testing if Tom's register
spilling patches help.


-- 
Earthling Michel Dänzer            |                  http://www.amd.com
Libre software enthusiast          |                Mesa and X developer

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 234 bytes
Desc: OpenPGP digital signature
URL: <http://lists.freedesktop.org/archives/mesa-dev/attachments/20140415/e6f1f064/attachment.sig>