[petsc-users] Possible bug PETSc+Complex+CUDA

[Junchao Zhang]

Hello, Jorge,
  The problem is fixed in petsc master and you need to add
--download-thrust to your configure options to use this feature.
  Thanks for reporting.
--Junchao Zhang


On Fri, May 29, 2020 at 7:52 AM Jorge Chiva Segura <
jorge.chiva.segura at gmail.com> wrote:

> Dear all,
>
> I just wanted to add what we found in case that it can help to solve this
> problem:
>
> * Examples ex2.c, ex11.c and ex32.c under src/ksp/ksp/tutorials all of
> them seg-fault with "-mat_type mpiaijcusparse" or "-vec_type mpicuda" with
> CUDA 9.2 and 10.2.
>    Examples ex2.c and ex32.c work fine if PETSc scalar type is set to real
> instead of complex. PETSc has been compiled with gcc-6.4.0
>
> * Here:
>   https://dynamite.readthedocs.io/en/latest/tips.html#gpu-support
>   it is mentioned that there is some problem for CUDA versions 8 and
> above.
>
> * It seems that the same problem was mentioned in this thread:
>
> https://lists.mcs.anl.gov/mailman/htdig/petsc-users/2019-February/037748.html
>
> Please, let me know if you need any extra information.
>
> Thank you very much for your help.
>
> Best regards,
>
> Jorge
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20200616/436b0654/attachment.html>