[petsc-dev] building with MKL

[Jeff Hammond]

On Sun, Jul 1, 2018 at 10:43 AM Victor Eijkhout <eijkhout at tacc.utexas.edu>
wrote:

>
>
> On Jul 1, 2018, at 12:30 PM, Jeff Hammond <jeff.science at gmail.com> wrote:
>
> If you really want to do this, then replace COMMON with CORE to specialize
> for SKX. There’s not point to using COMMON if you’ve got the MIC path
> already.
>
>
> We advocate CORE on our userguides, but the Intel compiler crashes
> reliably with that in certain cases. I think in particular Intel has never
> figured out how complex numbers work.
>

Is PETSc using C99 _Complex or C++ std::complex?

I haven’t heard any complaints about how Intel compilers treat ISO-standard
complex types. I’ll see what’s in the ticket system now.

If the compiler crashes it’s always on the petsc complex mode. I think I
> have submitted tickets about that, so maybe it’s fixed in update 3.
>

Version 17 I assume?


> I have also seen cases where CORE gives numerical problems and replacing
> by COMMON fixed them. Sorry, that was a user ticket and having solved it I
> didn’t bother to submit an Intel support ticket.
>

AVX-512DQ shouldn’t change numerical results so this is an indirect effect
and worthy of creating a compiler bug report if you can isolate the issue.


> The support website is a nightmare anyway, so I’m not overly motivated to
> submit support tickets in the first place.
>

I know that website sucks, but I have filed 50+ bug reports that way in the
past 4+ years, because it is the single most effective way to get positive
results.

If you send me single-file preprocessed (Linux preferred, but Mac works)
MCVEs, I’ll put them straight into the internal ticket system.

Jeff

> --
Jeff Hammond
jeff.science at gmail.com
http://jeffhammond.github.io/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20180701/9ee099bb/attachment.html>