Summary: | math/atlas fails with lapack 3.10.0 lapacklib not valid make target during LAPACK build phase | ||||||
---|---|---|---|---|---|---|---|
Product: | Ports & Packages | Reporter: | alt2600 | ||||
Component: | Individual Port(s) | Assignee: | Thierry Thomas <thierry> | ||||
Status: | Closed FIXED | ||||||
Severity: | Affects Only Me | CC: | thierry | ||||
Priority: | --- | Keywords: | needs-qa | ||||
Version: | Latest | Flags: | thierry:
merge-quarterly+
|
||||
Hardware: | Any | ||||||
OS: | Any | ||||||
See Also: | https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=247485 | ||||||
Attachments: |
|
Description
alt2600
2021-07-11 23:30:39 UTC
^Triage: Request support from Fortran team. Can you help with this Thierry? @Reporter Can you include a full build log (as an attachment, compressed if necessary) please. Created attachment 226407 [details]
Full Atlas Build log
See attached
(In reply to alt2600 from comment #2) I just noticed this when I was checking status. I'm not sure why it is saying make failed in /root, I didn't start make from /root, I was in the port directory. A commit in branch main references this bug: URL: https://cgit.FreeBSD.org/ports/commit/?id=75752cd50be8de2ff11026bf3a536bcad6245a8f commit 75752cd50be8de2ff11026bf3a536bcad6245a8f Author: Thierry Thomas <thierry@FreeBSD.org> AuthorDate: 2021-07-21 20:48:14 +0000 Commit: Thierry Thomas <thierry@FreeBSD.org> CommitDate: 2021-07-21 20:48:14 +0000 math/atlas: unbreak after math/lapack + various fixes The latest upgrade of math/lapack broke ATLAS, due to the lack of the post-target, but anyways it was broken in many others ways: - gmake is required (the fixes ifdef / #ifdef are not sufficient) - -fPIC is needed everywhere - fix EXTRA_FFLAGS - remove a conflict with atlas-devel. PR: 257122 Reported by: alt2600 (at) icloud.com MFH: 2021Q3 math/atlas/Makefile | 29 ++++++++++++++--------------- 1 file changed, 14 insertions(+), 15 deletions(-) A commit in branch main references this bug: URL: https://cgit.FreeBSD.org/ports/commit/?id=b7eec409d5664a73725c392b83e4f2741c5ad33f commit b7eec409d5664a73725c392b83e4f2741c5ad33f Author: Thierry Thomas <thierry@FreeBSD.org> AuthorDate: 2021-07-21 20:38:05 +0000 Commit: Thierry Thomas <thierry@FreeBSD.org> CommitDate: 2021-07-21 20:44:31 +0000 math/lapack: fix post-patch target for math/atlas When upgrading this port, I previously merged the post-patch with the pre-configure target, and it was a mistake. The patch target is used from math/atlas, and make.inc is needed. PR: 257122 Reported by: alt2600 (at) icloud.com MFH: 2021Q3 math/lapack/Makefile | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) Indeed, my previous upgrade of math/lapack broke ATLAS: the post-patch target of lapack is required, because atlas uses the patch target of lapack and nothing more. But this was not sufficient, and atlas was still broken in several other ways… I hope that everything is now fixed, but let me know if you encounter other problem. Next step: upgrade to the latest release! Note: I have not used this port for many years, since I need OpenBLAS. Just out of curiosity, what is your use case for this beast? (In reply to Thierry Thomas from comment #6) It is still considered marginally faster then OpenBLAS, and in the past the dev notes for some packages such as SciPy or maybe Numpy seem to suggest it doesn't build properly when openblas is configured to use openmpi, which is our default due to some threading issue in python, but i seem to have deleted that bookmark. but form a site.cfg.example for SciPy I have this which seems to suggest this was an issue with Python 2.7 and no longer a problem perhaps. # OpenBLAS # -------- # OpenBLAS is another open source optimized implementation of BLAS and Lapack # and can be seen as an alternative to Atlas. To build Scipy against OpenBLAS # instead of Atlas, use this section instead of the above, adjusting as needed # for your configuration (in the following example we installed OpenBLAS with # ``make install PREFIX=/opt/OpenBLAS``. # OpenBLAS is generically installed as a shared library, to force the OpenBLAS # library linked to also be used at runtime you can utilize the # runtime_library_dirs variable. # # **Warning**: OpenBLAS, by default, is built in multithreaded mode. Due to the # way Python's multiprocessing is implemented, a multithreaded OpenBLAS can # cause programs using both to hang as soon as a worker process is forked on # POSIX systems (Linux, Mac). # This is fixed in Openblas 0.2.9 for the pthread build, the OpenMP build using # GNU openmp is as of gcc-4.9 not fixed yet. # Python 3.4 will introduce a new feature in multiprocessing, called the # "forkserver", which solves this problem. For older versions, make sure # OpenBLAS is built using pthreads or use Python threads instead of # multiprocessing. # (This problem does not exist with multithreaded ATLAS.) # # https://docs.python.org/3.4/library/multiprocessing.html#contexts-and-start-methods # https://github.com/xianyi/OpenBLAS/issues/294 Not that our SciPy supports atlas right now. I do not like that many ports ignore BLAS selection and force OpenBLAS for instance when the default is netlib. In reality there doesn't seem to be any reason why we cannot use all of them other then they prefix into conflicting install prefixes, whereas in other systems they are setup to use be non-conflicting. I've been using this for near 15 years now, I don't see a compelling reason to change right now. Given openmpi is broken in tree when libfabric is installed, and qmake needs to have AVX fixes to fix non-AVX architectures due to a change made in qt5-3d rebasing my system is currently impossible anyway. I'll give the new updates a try and open new issue if it fails to install. A commit in branch 2021Q3 references this bug: URL: https://cgit.FreeBSD.org/ports/commit/?id=7e1d6c0f6c1b935413eb5ecc689a60a5d236af80 commit 7e1d6c0f6c1b935413eb5ecc689a60a5d236af80 Author: Thierry Thomas <thierry@FreeBSD.org> AuthorDate: 2021-07-21 20:48:14 +0000 Commit: Thierry Thomas <thierry@FreeBSD.org> CommitDate: 2021-07-25 16:02:35 +0000 math/atlas: unbreak after math/lapack + various fixes The latest upgrade of math/lapack broke ATLAS, due to the lack of the post-target, but anyways it was broken in many others ways: - gmake is required (the fixes ifdef / #ifdef are not sufficient) - -fPIC is needed everywhere - fix EXTRA_FFLAGS - remove a conflict with atlas-devel. PR: 257122 Reported by: alt2600 (at) icloud.com MFH: 2021Q3 (cherry picked from commit 75752cd50be8de2ff11026bf3a536bcad6245a8f) math/atlas/Makefile | 29 ++++++++++++++--------------- 1 file changed, 14 insertions(+), 15 deletions(-) A commit in branch 2021Q3 references this bug: URL: https://cgit.FreeBSD.org/ports/commit/?id=a36e5a3440e59bd0e2a42f16d0ca5b820be234d9 commit a36e5a3440e59bd0e2a42f16d0ca5b820be234d9 Author: Thierry Thomas <thierry@FreeBSD.org> AuthorDate: 2021-07-21 20:38:05 +0000 Commit: Thierry Thomas <thierry@FreeBSD.org> CommitDate: 2021-07-25 16:01:48 +0000 math/lapack: fix post-patch target for math/atlas When upgrading this port, I previously merged the post-patch with the pre-configure target, and it was a mistake. The patch target is used from math/atlas, and make.inc is needed. PR: 257122 Reported by: alt2600 (at) icloud.com MFH: 2021Q3 (cherry picked from commit b7eec409d5664a73725c392b83e4f2741c5ad33f) math/lapack/Makefile | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) |