tech-pkg archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Pull in pkg/55999 to un-break math/openblas with gcc10 for 2021Q1?



Another point in my series of things that I wouldn't like to ship
broken with 2021Q1:

	http://gnats.netbsd.org/cgi-bin/query-pr-single.pl?number=55999

I opened that one a month ago. Openblas needs a bump to 0.3.10 as in
wip/openblas to work with gcc10. I quote myself:

commit 86af17db8526e629c2c02c6af1f1ce7db6f6ba6d
Author: Dr. Thomas Orgis <thomas.orgis%uni-hamburg.de@localhost>
Date:   Thu Nov 12 12:44:39 2020 +0100

    openblas: version 0.3.10
    
    This updated fixes the build with gcc 10 (segfault in cblat1 test).
    
    I did not go all the way to the current 0.3.12, as that would
    need some hacking of chosen compiler flags. 0.3.13 should be
    the next one.
    
    This commit also fixes the ARCH → ARCH_ sed to change all
    occurences on a line.


Also, there are fixes on top:

commit 3c6284cba90280bc367cf4d1d8252ae4d6e92e76
Author: Jason Bacon <bacon%NetBSD.org@localhost>
Date:   Thu Feb 25 11:56:13 2021 -0600

    openblas: Update ONLY_FOR_PLATFORMS documentation

commit 8071bf28f3ffc95af046ff3eaaac6983f4f70035
Author: Jason Bacon <bacon%NetBSD.org@localhost>
Date:   Thu Feb 25 11:51:32 2021 -0600

    openblas*: Successful build on NetBSD

commit 056e3d5c972a4b286e8755dbee323a9951855165
Author: Dr. Thomas Orgis <thomas.orgis%uni-hamburg.de@localhost>
Date:   Wed Feb 24 18:40:17 2021 +0100

    openblas: flags from environment again to un-break PICy build
    
    The bug that prompted us to force the compiler flags in the
    make arguments is fixed. Now we got a different one: The logic
    that decides to add -fPIC where needed is overridden when doing
    this, resulting in relocation errors (strangely, not with every
    toolchain).
    
    So let's remove that again and take FFLAGS and friends from the
    environment again.


I suggest greenlighting pulling in the current state of wip/openblas,
along with the proposed fixes of math/blas and math/py-numpy. Then all
can together be tested by bulk builds. It's a related set of packages.

I have a somewhat asocial rationale to push for the inclusion of these
latest versions and patches: This is what I test, or Jason, for that
matter, especially since it is about fixing build issues. The version
currently scheduled 2021Q1 is just bitrotting.


Alrighty then,

Thomas

-- 
Dr. Thomas Orgis
HPC @ Universität Hamburg


Home | Main Index | Thread Index | Old Index