tech-pkg archive

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

Re: BLAS in pkgsrc: present and future



On 2019-04-17 12:15, Jason Bacon wrote:
On 2019-03-27 11:23, Dr. Thomas Orgis wrote:
Am Tue, 26 Mar 2019 20:34:55 -0400
schrieb Greg Troxel <gdt%lexort.com@localhost>:

Separately from review and consensus, soon after 2019Q1 is cut is a good
time for possibly disruptive changes, since there is then a few months
to sort any issues.
Yeah, it is sensible to introduce changes after the release. It won't
make much of a difference to me, as I will run a patched pkgsrc anyway.
I am just working on reducing the patch set a bit.

Btw. I just pushed 1528ca8f05df088e5b74b959a3e1b9755a01f722 to wip for
installing the static libraries from the Netlib packages, too. This
needs hackery in the CMakeLists that I do not really like, but it seems
to work. I wonder if upstream would consider including this …

I am in fact looking forward to 2019Q1 being cut soon, as I will base my
installs on this coupled with a wip snapshot.

But waiting with the integration could in fact mean that we then can
just push patches as PRs to make packages use the BLAS stuff instead of
creating short-lived trivial copies of packages in wip. I would find
the latter rather annoying. Pushing patches to the CVS version would
fit my workflow, as I am collecting exactly those with my install
scripts.


Alrighty then,

Thomas

Quick heads-up:

===> Patching for plink2-0.0.0.20190202
===> Creating toolchain wrappers for plink2-0.0.0.20190202
ERROR: openblas>=0.3.5 is not installed; can't buildlink files.
*** Error code 1

Stop.
bmake.bin[1]: stopped in /usr/pkgsrc/wip/plink2
*** Error code 1

Stop.
bmake.bin: stopped in /usr/pkgsrc/wip/plink2

It builds fine with netlib, but this is what I get when setting BLAS_TYPE to openblas.
So this was caused by #including cblas/bl3 before mk/blas.bl3.

As currently configured, one must include

.include "../../wip/mk/blas.buildlink3.mk"
.include "../../wip/cblas/buildlink3.mk"

Is there a clear purpose behind the current cblas implementation, or have you not gotten around to refining it yet?

Home | Main Index | Thread Index | Old Index