[Dune-devel] [GSoC2016] SuperLU v5 support. Ticket #1704

[Maikel Nadolski]

> Please don't use strong language, neither in code or in
> communication. Nevertheless, I - and probably most of the
> users of SuperLU - feel you pain…

It wont happen again, sorry.

> Great! I consider your junior job fulfilled. But I welcome
> if you try to polish the patch and get it included.

Ofcourse!? I have access to GitLab now!

> We prefer discussions in public. But if you feel it should
> stay private, you could choose private mail. In general,
> there are not that many GSoC students on our mailing list,
> so we can track who comes up with a cool or clever idea.

Ok, "be prepared". ;^)


Maikel


On 09.03.2016 22:32, Christoph Grüninger wrote:
> Hi Maikel,
> I liked reading your patch. Thanks for providing it! I am
> pretty sure we are going to include this patch after some
> minor revisions.
>
>> I made a test setting a macro HAVE_GLOBALLU_T_IN_UTIL which fails for
>> older versions, because they moved the structure
>> definition to match the function calls better i think. It
>> would be way better to test wether we can link against
>> solver functions having the correct signature instead
>> of this "hack" (taking 22 instead of 21 arguments). Because that is
>> semantically what we actually want to know. I
>> will probably do that.
>
> Your current approach is ok. That's what we already do to
> detect older versions.
> I'd like to get a shorter variable name and to include
> SUPERLU in it. What's about HAVE_SUPERLU_GLOBALLUT?
>
>> I have tested with SuperLU 4.3 and SuperLU 5.1.1 but it shouldn't be
>> much different with 5.0. I changed the
>> script for finding the library to only look for 5.1.1
>> and not 5.0 because i havent that version yet.
>
> You can include 5.1 and 5.0, too. I'll give them a try
> before merging this feature.
>
>> Does this mean the results are ok? I havent looked much at what the
>> tests do test.
>
> Yes, when they pass you are fine. That's the idea of tests,
> unless they fail, most people do not need to look into them.
>
> Quoting your code:
> +CHECK_C_SOURCE_COMPILES("
> +typedef int int_t; /* WTF this SuperLU... */
> +#include <supermatrix.h>
>
> Please don't use strong language, neither in code or in
> communication. Nevertheless, I - and probably most of the
> users of SuperLU - feel you pain…
>
> Great! I consider your junior job fulfilled. But I welcome
> if you try to polish the patch and get it included.
>
>> But I also looked at some of the projects for GSoC nad have some
>> questions. Shall i write to the mailing list or directly
>> to the mentors?
>
> We prefer discussions in public. But if you feel it should
> stay private, you could choose private mail. In general,
> there are not that many GSoC students on our mailing list,
> so we can track who comes up with a cool or clever idea.
>
> Thank you for the patch,
> Christoph
>

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 5498 bytes
Desc: S/MIME Cryptographic Signature
URL: <https://lists.dune-project.org/pipermail/dune-devel/attachments/20160310/7c6708f1/attachment.bin>