Hi,
thanks for your response.
the defines are as follows:
NUMBER_OF_TRIAL_SHAPE_FUNCTIONS: 3
NUMBER_OF_TEST_SHAPE_FUNCTIONS: 3
NUMBER_OF_QUAD_POINTS: 6
The kernel is queued with a local size of (1, 16). If I set it to (1, 1)
the crash disappears. Also, I do not observe the crash with all types set
to single precision. Setting the number
of quadrature points to a power of 2 does not fix the segfault. When I
change the shape function constants from 3 to 4 the crash disappears. This
may hint to an alignment issue,
as you say. However, implementing the restriction that shape functions are
powers of two would be difficult, as we would have to padd the computations
until the nearest power of
two and then discard quite a few of the results again. Vectorization in our
code is achieved differently through manual workgroup level vectorization
(these kernels crash in the same way
as this one does).
Let me know if you need more information or experiments.
Regarding Centos compilation, this was done with LLVM 5 from the Redhat
developer toolkit. POCL compiles fine. But then the CL compiler gives back
error messages when just-in-time compiling kernels.
When I get to it I will have another look and raise a bug report if
necessary.
Best wishes
Timo
On Wed, 6 Jun 2018 at 10:05, Michal Babej <[email protected]>
wrote:
> Hello,
>
> > compile pocl on a centos machine, but had some compilation issues, so did
> > not investigate further there.
>
> Compilation issues compiling pocl, or one of its dependencies ? There is
> some reasonably old (i think 5.0) Clang+LLVM for Centos 7 somewhere, just
> FYI.
>
> > I used the disassembly hint for gdb. The crash is in line 310. The
> > associated gist is here:
> > https://gist.github.com/tbetcke/087b18ad7a238b0528d68805e5c46f14
>
> OK, so as i suspected it's likely a variant of this bug:
>
> https://github.com/pocl/pocl/issues/558
>
> I've tried to fix it but perhaps i missed something.
>
> > You can find the source code of the kernel here:
> > https://gist.github.com/tbetcke/c685568273f4df7250ed643524d6b175
>
> There's a bunch of #defines missing, and i suspect they're important (the
> NUMBER_OF_... defines). You could try setting these to nearest larger
> power-of-two and see if pocl still crashes. Also knowing your local sizes
> to enqueueNDRange would help.
>
> Regards,
> -- mb
>
>
> ------------------------------------------------------------------------------
> Check out the vibrant tech community on one of the world's most
> engaging tech sites, Slashdot.org! http://sdm.link/slashdot
> _______________________________________________
> pocl-devel mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/pocl-devel
>
--
Dr Timo Betcke
Reader in Mathematics
University College London
Department of Mathematics
E-Mail: [email protected]
Tel.: +44 (0) 20-3108-4068
Fax.: +44 (0) 20-7383-5519
------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
pocl-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/pocl-devel
