OpenBLAS package

[vbraun]

Provide an OpenBLAS optional package

Tarball is on the mirrors

Upstream: Fixed upstream, but not in a stable release.

CC: @kiwifb

Component: packages: optional

Author: Volker Braun

Branch/Commit: 4a2b4dc

Reviewer: Jeroen Demeyer

Issue created by migration from https://trac.sagemath.org/ticket/20129

[jdemeyer]

comment:1

Duplicate of #20096

[vbraun]

Branch: u/vbraun/openblas_package

[vbraun]

comment:3

No its not

---

New commits:

fd198fa

OpenBLAS optional package

[vbraun]

Commit: fd198fa

[vbraun]

Author: Volker Braun

[jdemeyer]

comment:5

Can you rename COMPILE_ARGS to OPENBLAS_CONFIGURE? I know it's not using configure, but that would make the environment variable compatible with similar variables in Sage like ATLAS_CONFIGURE, PARI_CONFIGURE,...

[jdemeyer]

comment:6

And also remove this: COMPILE_ARGS=*. It's a *feature'' that users can set this.

[jdemeyer]

comment:7

You should add a dependencies file.

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

8e22428

Rename variable

[sagetrac-git]

Changed commit from fd198fa to 8e22428

[jdemeyer]

comment:10

1. Add a spkg-check which runs $MAKE tests.

2. Apply this patch for powerpc or use version 0.2.16.rc1

[kiwifb]

comment:11

I'll add something to Jeroen's last comment, if we are only going to install dynamic libraries (NO_STATIC=1) may be we shouldn't build them either - I don't think that will shave much time of the build process though.

[kiwifb]

comment:12

A note on tests, the lapack testsuite is currently run as part of the build process. The blas part is tested with make tests, I am not sure how to disable lapack test suite and run it later.

[vbraun]

comment:13

• I don't have access to a power machine, so I'm not going to add random patches.

• I don't think you can avoid building the static library, NO_STATIC is really only referenced in Makefile.install

[sagetrac-git]

Changed commit from 8e22428 to 6b9fab6

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

6b9fab6

Add spkg-check

[sagetrac-git]

Changed commit from 6b9fab6 to f48e69b

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. This was a forced push. New commits:

f48e69b

Add spkg-check

[kiwifb]

comment:17

Replying to @vbraun:

• I don't have access to a power machine, so I'm not going to add random patches.

Well Jeroen (power8) and I (power6, power7) do. If you don't include this, we'll need a follow up ticket. And we are both ready to review such patch.

• I don't think you can avoid building the static library, NO_STATIC is really only referenced in Makefile.install

I checked little and I agree it is in the too hard basket.

[jdemeyer]

comment:18

Replying to @vbraun:

• I don't have access to a power machine, so I'm not going to add random patches.

It's not at all a random patch. I found an issue on my POWER8, looked on github and saw that this issue was fixed by the patch I mentioned.

[vbraun]

comment:19

Replying to @kiwifb:

Well Jeroen (power8) and I (power6, power7) do.

But I can't test it, nor do I think its a priority for this ticket. There are a couple of pieces missing before you can build Sage with OpenBLAS at all. But if you don't have the 10 seconds it takes to create a followup ticket when you actually need it feel free to commit it here. Just test it because I can't.

[jdemeyer]

comment:20

Replying to @vbraun:

when you actually need it feel free to commit it here.

Sounds good. If I do that, will you review that patch-that-you-cannot-test then?

[jdemeyer]

Changed branch from u/vbraun/openblas_package to u/jdemeyer/openblas_package

[jdemeyer]

Changed commit from f48e69b to 4a2b4dc

[jdemeyer]

New commits:

759612b	Minor fixes to spkg-install
4a2b4dc	Add patch for powerpc

[jdemeyer]

comment:23

With this branch:

openblas-0.2.15
====================================================
Setting up build directory for openblas-0.2.15
mv: cannot stat 'OpenBLAS-0.2.15*': No such file or directory
Finished set up
****************************************************
Host system:
Linux sardonis 3.19.0-15-generic #15-Ubuntu SMP Thu Apr 16 23:32:13 UTC 2015 ppc64le ppc64le ppc64le GNU/Linux
****************************************************
C compiler: gcc
C compiler version:
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/home2/jdemeyer/local/libexec/gcc/powerpc64le-unknown-linux-gnu/5.3.0/lto-wrapper
Target: powerpc64le-unknown-linux-gnu
Configured with: ../gcc-5.3.0/configure --prefix=/home2/jdemeyer/local --enable-languages=c,c++,fortran MAKE=make
Thread model: posix
gcc version 5.3.0 (GCC) 
****************************************************
patching file c_check
Building OpenBLAS: make 
[.......]
 cblas_zhpr2  PASSED THE COMPUTATIONAL TESTS (   577 CALLS)
 cblas_zhpr2  PASSED THE COMPUTATIONAL TESTS (   577 CALLS)

 END OF TESTS
make[3]: Leaving directory '/home2/jdemeyer/sage-check/local/var/tmp/sage/build/openblas-0.2.15/xianyi-OpenBLAS-afedc8e/ctest'
make[2]: Leaving directory '/home2/jdemeyer/sage-check/local/var/tmp/sage/build/openblas-0.2.15/xianyi-OpenBLAS-afedc8e'

real    0m4.521s
user    0m8.894s
sys     0m3.871s
Deleting temporary build directory
/home2/jdemeyer/sage-check/local/var/tmp/sage/build/openblas-0.2.15
Finished installing openblas-0.2.15.spkg

[jdemeyer]

Upstream: Fixed upstream, but not in a stable release.

[vbraun]

Reviewer: Jeroen Demeyer

[vbraun]

comment:25

I guess you didn't have the 10 seconds to open a followup ticket...

[vbraun]

Reviewer: Jeroen Demeyer

[vbraun]

comment:26

I guess you didn't have the 10 seconds to open a followup ticket...

[jdemeyer]

comment:27

Replying to @vbraun:

I guess you didn't have the 10 seconds to open a followup ticket...

I never understood why you wanted a follow-up ticket for this...

I think it's completely normal to review a ticket by saying "X doesn't work, but after applying patch Y it does work, so please apply patch Y".

[vbraun]

comment:28

You fail to mention that it doesn't work on an exotic architecture on which Sage currently doesn't work. Porting to said architecture shouldn't be mashed into every other ticket just because you are interested in it.

[jdemeyer]

comment:29

Replying to @vbraun:

Porting to said architecture shouldn't be mashed into every other ticket just because you are interested in it.

I obviously disagree with this... if a trivial accepted-by-upstream patch can be added, I don't see the problem.

And I don't see how it's relevant if Sage currently works or not on said architecture. Shouldn't we try to support as much architectures as possible?

[jdemeyer]

comment:30

For the record, Sage mostly works on powerpc64le after applying #19719. There are some pexpect issues, but I guess those might be OS bugs unrelated to Sage.

[vbraun]

comment:31

Replying to @jdemeyer:

Shouldn't we try to support as much architectures as possible?

Yes, but ideally by having one ticket per issue. And not delaying unrelated tickets until we can also stuff some porting effort in there.

[kiwifb]

comment:32

I'll be a pain in the ass and point out that we support power7 and that this patch concerns power7 as well.

[vbraun]

Changed branch from u/jdemeyer/openblas_package to 4a2b4dc