error using libframe

[Zhiyong Zhang]

Greeting, 

I encountered the following errors running nwchem compiled with libflame: 

libflame: src/lapack/red/tridiagut/front/flamec/FLA_Tridiag_UT_internal.c (line 26):

libflame: Function or conditional branch/case not yet implemented.

libflame: Aborting.

Program received signal SIGABRT: Process abort signal.

Any ideas what could be wrong here? 

Thanks!

Zhiyong

[Field G. Van Zee]

Zhiyong,

Looks like you're trying to execute the "upper" storage case of
tridiagonal reduction, which is not implemented.

Just curious: how you are using libflame?

Field

> --
> You received this message because you are subscribed to the Google
> Groups "libflame-discuss" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to libflame-discu...@googlegroups.com
> <mailto:libflame-discu...@googlegroups.com>.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/libflame-discuss/3e55c433-29a5-4a7c-9e19-c4e21c940377n%40googlegroups.com
> <https://groups.google.com/d/msgid/libflame-discuss/3e55c433-29a5-4a7c-9e19-c4e21c940377n%40googlegroups.com?utm_medium=email&utm_source=footer>.

[Zhiyong Zhang]

Thanks Field. I am trying to link nwchem, a chemistry software, against libflame. These turn out to be related to the more stringent checks of variables in gcc/10.1.0. The current development version of libflame fixes (maybe bypassed) that issue but there are still many failures in the test cases of the installation.

Zhiyong

Sent via the Samsung Galaxy Note8, an AT&T 5G Evolution capable smartphone

Get Outlook for Android

---

From: libflame...@googlegroups.com <libflame...@googlegroups.com> on behalf of Field G. Van Zee <fi...@cs.utexas.edu>
Sent: Monday, September 21, 2020 1:45:50 PM
To: libflame...@googlegroups.com <libflame...@googlegroups.com>
Subject: Re: [libflame-discuss] error using libframe

 

To unsubscribe from this group and stop receiving emails from it, send an email to libflame-discu...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/libflame-discuss/12ea6e93-7391-39fa-37f3-eca89d7b3863%40cs.utexas.edu.

[Field G. Van Zee]

I'm a little confused. Why would the version of gcc matter in terms of
which branches in libflame get executed?

Also, what do you mean by the current development version of libflame?

Field

On 9/21/20 4:00 PM, Zhiyong Zhang wrote:
> Thanks Field. I am trying to link nwchem, a chemistry software, against
> libflame. These turn out to be related to the more stringent checks of
> variables in gcc/10.1.0. The current development version of libflame
> fixes (maybe bypassed) that issue but there are still many failures in
> the test cases of the installation.
>
> Zhiyong
>
> Sent via the Samsung Galaxy Note8, an AT&T 5G Evolution capable smartphone

> Get Outlook for Android <https://aka.ms/ghei36>
>
> ------------------------------------------------------------------------
> *From:* libflame...@googlegroups.com

> <libflame...@googlegroups.com> on behalf of Field G. Van Zee
> <fi...@cs.utexas.edu>

> *Sent:* Monday, September 21, 2020 1:45:50 PM
> *To:* libflame...@googlegroups.com <libflame...@googlegroups.com>
> *Subject:* Re: [libflame-discuss] error using libframe

> https://groups.google.com/d/msgid/libflame-discuss/BYAPR02MB51411529B2877BE998275653CE3A0%40BYAPR02MB5141.namprd02.prod.outlook.com
> <https://groups.google.com/d/msgid/libflame-discuss/BYAPR02MB51411529B2877BE998275653CE3A0%40BYAPR02MB5141.namprd02.prod.outlook.com?utm_medium=email&utm_source=footer>.

[Zhiyong Zhang]

Hi Field, 

Here is a more detailed description of the situation. I was trying to link nwchem 7.0.0 against scalapack 2.1, libflame 2.2.4 and blis 2.2.4 on AMD rome with openmpi/4.0.5 and gcc/10.1.0. The compilation and linking was successful but I encountered the error originally reported: 

libflame: src/lapack/red/tridiagut/front/flamec/FLA_Tridiag_UT_internal.c (line 26):

libflame: Function or conditional branch/case not yet implemented.

libflame: Aborting.

The scalapack I used was compiled with a different version of openmpi so I tried to install it with the same version of openmpi for nwchem. I encountered a couple of issues with scalapack installation that looked like not related to libflame. The version of scalapack I was trying to install is 2.2.4, the most recent version. The compiler version is gcc/10.1.0. Starting from 10.1.0 and it enforces implicit function parameter mismatches. So I used the development version of the scalapack for the installation. I mistakenly stated it is libflame in the previous reply. Sorry for the confusion. 

So the issue with nwchem with libflame is then the error of  Function or conditional branch/case not yet implemented that I encountered when I tried to run nwchem. That is the failure encountered when running nwchem with libflame. I haven't digged in the code to figure out how that particular function of libflame is being used though. There might be other usage of libflame functions in nwchem that is not yet implemented. In that case what are the solutions? Or as these functions are not implemented in libflame, is it still possible to use libflame for nwchem? 

Zhiyong

[Field G. Van Zee]

Zhiyong,

Thank you for this background information.

In theory, it may be straightforward to implement the upper case of
tridiagonal reduction in terms of the lower case by first inducing a
transposition of the matrix, computing the result, and then transposing
the results. (I could still be mistaken.) However, as you point out,
this may not be the only thing that isn't implemented in libflame. And
unfortunately, we don't have the resources to dig into this at the
moment. (We also don't have much, if any, experience with ScaLAPACK.)

Perhaps others in the community will step up to offer support.

Best of luck,
Field

> > variables in gcc/10.1.0. <http://10.1.0.> The current development

> https://groups.google.com/d/msgid/libflame-discuss/5bdefcae-858e-469e-a4bc-411537c8f3f9n%40googlegroups.com
> <https://groups.google.com/d/msgid/libflame-discuss/5bdefcae-858e-469e-a4bc-411537c8f3f9n%40googlegroups.com?utm_medium=email&utm_source=footer>.