libomp buildbots

Hi all,

There are currently two bots configured to build and test libomp with clang
and gcc [1][2].
However the clang build is currently always failing - could someone look
after this and fix the issue?

This been ongoing for almost one month now (failing since Feb 25th) - have I
missed a mail from the buildbots or are they only supposed to send to the
committer if the state changes?

Moreover I would like to propose to add buildbot configurations for building
and testing libomp with OMPT support. I have just committed the first four
tests and plan to add more so that we don't regress some basic features.
I don't really know what the process is here. I have found
https://llvm.org/svn/llvm-project/zorg/trunk/ and made the necessary changes
already locally, at least what I think is needed. Can someone give me hints
here?

Cheers,
Jonas

[1] http://lab.llvm.org:8011/builders/libomp-clang-x86_64-linux-debian/
[2] http://lab.llvm.org:8011/builders/libomp-gcc-x86_64-linux-debian/

Hi all,

There are currently two bots configured to build and test libomp with clang
and gcc [1][2].
However the clang build is currently always failing - could someone look
after this and fix the issue?

This been ongoing for almost one month now (failing since Feb 25th) - have I
missed a mail from the buildbots or are they only supposed to send to the
committer if the state changes?

Hi,

To me it looks like the job is trying to build LLVM and Clang in '-O0
-g' mode, and the machine does not have enough RAM per core to run the
linker. (The machine has plenty of RAM in general, but it also has
many cores.)

Could we switch the job to build in Release+Assert?

Moreover I would like to propose to add buildbot configurations for building
and testing libomp with OMPT support. I have just committed the first four
tests and plan to add more so that we don't regress some basic features.
I don't really know what the process is here. I have found
https://llvm.org/svn/llvm-project/zorg/trunk/ and made the necessary changes
already locally, at least what I think is needed. Can someone give me hints
here?

Yes, that's the right repository. Please feel free to add more
(reasonably-sized) jobs to my buildbot. It would be best if the
libomp and OMPT jobs did not try to build the compiler like they try
to do now. That would leave more time for other jobs.

Dmitri

From: Dmitri Gribenko [mailto:gribozavr@gmail.com]
Sent: Tuesday, March 22, 2016 4:12 PM
To: Hahnfeld, Jonas
Cc: LLVM-OpenMP (openmp-dev@lists.llvm.org); gkistanova@gmail.com
Subject: Re: libomp buildbots

> Hi all,
>
> There are currently two bots configured to build and test libomp with
> clang and gcc [1][2].
> However the clang build is currently always failing - could someone
> look after this and fix the issue?
>
> This been ongoing for almost one month now (failing since Feb 25th) -
> have I missed a mail from the buildbots or are they only supposed to
> send to the committer if the state changes?

Hi,

To me it looks like the job is trying to build LLVM and Clang in '-O0 -g'
mode,
and the machine does not have enough RAM per core to run the linker. (The
machine has plenty of RAM in general, but it also has many cores.)

Could we switch the job to build in Release+Assert?

> Moreover I would like to propose to add buildbot configurations for
> building and testing libomp with OMPT support. I have just committed
> the first four tests and plan to add more so that we don't regress some
basic features.
> I don't really know what the process is here. I have found
> https://llvm.org/svn/llvm-project/zorg/trunk/ and made the necessary
> changes already locally, at least what I think is needed. Can someone
> give me hints here?

Yes, that's the right repository. Please feel free to add more
(reasonably-sized) jobs to my buildbot. It would be best if the libomp and
OMPT jobs did not try to build the compiler like they try to do now. That
would leave more time for other jobs.

Hi,

So should we use a pre-compiled version? In the configuration there is some
reference to /home/llvmbb/bin/clang-latest/bin - does that contain a
reasonably recent version of clang?

Jonas

I think that would be best.

$ /home/llvmbb/bin/clang-latest/bin/clang --version
clang version 3.9.0 (trunk 259021)

If it is not recent enough, I will update it.

Dmitri

From: Dmitri Gribenko [mailto:gribozavr@gmail.com]
Sent: Tuesday, March 22, 2016 4:52 PM
To: Hahnfeld, Jonas
Cc: LLVM-OpenMP (openmp-dev@lists.llvm.org); gkistanova@gmail.com
Subject: Re: libomp buildbots

> Hi,
>
> So should we use a pre-compiled version? In the configuration there is
> some reference to /home/llvmbb/bin/clang-latest/bin - does that
> contain a reasonably recent version of clang?

I think that would be best.

$ /home/llvmbb/bin/clang-latest/bin/clang --version clang version 3.9.0
(trunk 259021)

If it is not recent enough, I will update it.

Looks ok to me, CC'ing some Intel folks...
Do you also have llvm-lit and FileCheck there?

Jonas

No -- it does not look like a default installation of Clang would
install those. If you have a suggestion of how to install Clang
differently -- I'm all ears.

Dmitri

Hi,
I don't think this a good idea to not use latest clang compiler. There
might be some changes, that may have bugs. Currently these bugs may be
revealed during testing of OpenMP runtime on libomp buildbots.
But definitely we must switch to Release+Asserts version of build.

Best regards,
Alexey Bataev

From: Alexey Bataev [mailto:a.bataev@hotmail.com]
Sent: Wednesday, March 23, 2016 6:12 AM
To: Dmitri Gribenko
Cc: LLVM-OpenMP (openmp-dev@lists.llvm.org); gkistanova@gmail.com;
Peyton, Jonathan L (jonathan.l.peyton@intel.com); Hahnfeld, Jonas
Subject: Re: libomp buildbots

Hi,
I don't think this a good idea to not use latest clang compiler. There might be
some changes, that may have bugs. Currently these bugs may be revealed
during testing of OpenMP runtime on libomp buildbots.
But definitely we must switch to Release+Asserts version of build.

If we want to use it like this, it would be a good idea to actually trigger a libomp build when something gets committed to clang. I think right now, only commits to openmp do that which would make it difficult to spot the error if clang broke...

Cheers,
Jonas

I wouldn't be opposed to that. My current concern is that libomp is
duplicating the work (building the compiler) that other jobs running
on the same host do anyway.

Dmitri

Agree, but still it is better to build the compiler from scratch to
catch possible regressions in clang. Currently there are no other
buildbots that execute tests from libomp testsuite. Or openmp runtime
and its tests must be added to other buidbots

Best regards,
Alexey Bataev

From: "Alexey Bataev via Openmp-dev" <openmp-dev@lists.llvm.org>
To: "Dmitri Gribenko" <gribozavr@gmail.com>
Cc: gkistanova@gmail.com, "LLVM-OpenMP (openmp-dev@lists.llvm.org)" <openmp-dev@lists.llvm.org>
Sent: Wednesday, March 23, 2016 2:19:01 AM
Subject: Re: [Openmp-dev] libomp buildbots

Agree, but still it is better to build the compiler from scratch to
catch possible regressions in clang. Currently there are no other
buildbots that execute tests from libomp testsuite. Or openmp runtime
and its tests must be added to other buidbots

+1

-Hal

Agree, but still it is better to build the compiler from scratch to
catch possible regressions in clang. Currently there are no other
buildbots that execute tests from libomp testsuite. Or openmp runtime
and its tests must be added to other buidbots

Best regards,
Alexey Bataev

Software Engineer
Intel Compiler Team

Aren't there any buildbots which perform 3-stage bootstraps with file
comparisons? Wouldn't it be simpler to just add openmp to the projects
sources on such a buildbot and adjust the test suite run there to
include libomp? That approach would insure that libomp is being built
with the same compiler compiler for a given revision.

Hi,

I see that the build type has been switched to Release+Asserts.
Is there any ongoing plan about triggering libomp builds by clang commits or integrating libomp into the other clang buildbots?

Cheers
Jonas