Hi Michal,
I just tried out your gist. It is still crashing in vmovaps. Below you find
the disassembled instructions around the crash:
-------------------------
0x00007fff87d7c983 <+1331>: xor r8d,r8d
0x00007fff87d7c986 <+1334>: nop WORD PTR cs:[rax+rax*1+0x0]
0x00007fff87d7c990 <+1344>: mov rax,r8
0x00007fff87d7c993 <+1347>: shl rax,0x4
0x00007fff87d7c997 <+1351>: vmovapd xmm21,XMMWORD PTR [rdx+rax*1]
0x00007fff87d7c99e <+1358>: vpermilpd xmm30,xmm21,0x1
0x00007fff87d7c9a5 <+1365>: vmovlpd QWORD PTR [r15],xmm21
0x00007fff87d7c9ab <+1371>: vmovhpd QWORD PTR [r15+0x8],xmm21
=> 0x00007fff87d7c9b2 <+1378>: vmovaps XMMWORD PTR [rsi],xmm20
0x00007fff87d7c9b8 <+1384>: vbroadcastsd ymm0,xmm21
0x00007fff87d7c9be <+1390>: vfmadd213pd ymm0,ymm9,ymm10
0x00007fff87d7c9c3 <+1395>: vpermpd ymm22,ymm21,0xd5
0x00007fff87d7c9ca <+1402>: vfmadd213pd ymm22,ymm12,ymm0
0x00007fff87d7c9d0 <+1408>: vmovapd xmm6,XMMWORD PTR [rsi]
0x00007fff87d7c9d4 <+1412>: vxorpd xmm0,xmm0,xmm0
0x00007fff87d7c9d8 <+1416>: mov r10,0xffffffffffffffd0
0x00007fff87d7c9df <+1423>: nop
--------------------------
Just let me know how I can help further. If you want to run the offending
code yourself it is quite simple. Just use
git clone https://bitbucket.org/bemppsolutions/bempp-cl.git
This is the development code. It depends on Python 3.6, pyopencl and numba.
The actual script file to run to cause the crash is in the following gist:
https://gist.github.com/tbetcke/19d4a3a289e83353fb332587ebb3b72d
Just add the bempp-cl directory to your PYTHONPATH and run the script with
python test_p1.py
Best wishes
Timo
On Wed, 6 Jun 2018 at 15:27, Michal Babej <[email protected]>
wrote:
> On Wed, 6 Jun 2018 10:52:20 +0100, Timo Betcke <[email protected]>
> wrote:
>
> > 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.
>
> Yep i the problem was triggered by "shapeIntegral" array of arrays. Here
> is an unclean patch:
>
> https://gist.github.com/franz/8a5a26b051f5b1d5dbec551a3236084a
>
> against pocl git master. It seems to fix the crash with your kernel, but i
> have to properly test it and clean it up.
>
> > However, implementing the restriction that shape functions are
> > powers of two would be difficult, as we would have to padd the
> computations
>
> this is clearly a bug in pocl, and should be fixed there.
>
> 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
