[Bug 276789] math/openblas: Update to 0.3.26

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=276789

Daniel Engberg <diizzy@FreeBSD.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |diizzy@FreeBSD.org

--- Comment #4 from Daniel Engberg <diizzy@FreeBSD.org> ---
Created attachment 248175
  --> https://bugs.freebsd.org/bugzilla/attachment.cgi?id=248175&action=edit
OpenBLAS WIP

https://gitlab.archlinux.org/archlinux/packaging/packages/openblas/-/blob/main/PKGBUILD?ref_type=heads

I got a bit inspired by Arch's port of OpenBLAS which seemed a bit cleaner.
It's a wip and as I don't know the inner workings of OpenBLAS it's a bit hard
to work on making sure everything works as expected. It did get a bit cleaner
than before.

- Use upstream release tarball
- aarch64 and amd64 appears to work as excepted (incl make test) and in
Poudriere
- I haven't tested the remaining platforms however they should also work from
what I can tell
- Simplify arch optimizaion(s), I'm not a fan of manually doing this at all
since we have CPUTYPE but due to how OpenBLAS builds I guess this is about as
good as it gets. Just provade a "baseline" switch or full blast to simplify
options in general.

Things that needs to worked on,
- Test armv7, i386 and powerpc* - I doubt anyone does any serious work on these
platforms at least for now (on FreeBSD )so we don't need to do
micro-optimization. As long as it builds and runs it should be fine for now at
least.
- Connect the benchmark build(?)
- Add description for POWER8 and POWER9, seems to be the only "real" platforms
supported/usable
- Remove old code in Makefile

-- 
You are receiving this mail because:
You are the assignee for the bug.