Re: [Buildroot] [PATCH] host-python-numpy: correct implied host-lapack/host-blas dependency.

["Yann E. MORIN" <yann.morin.1998@free.fr>]

Graeme, All,

+James

On 2024-01-29 12:40 -0800, gsmecher@threespeedlogic.com spake thusly:
> From: Graeme Smecher <gsmecher@threespeedlogic.com>
> 
> Before this patch, host-python-numpy attempted (and failed) to detect
> OpenBLAS as follows:
> 
>     buildroot$ make
>     >>> host-python-numpy 1.25.0 Configuring
>     [...]
>     Run-time dependency openblas found: NO (tried pkgconfig and cmake)
>     WARNING: CMake Toolchain: Failed to determine CMake compilers state
>     Run-time dependency openblas found: YES 0.3.21
>     Dependency openblas found: YES 0.3.21 (cached)
>     Program _build_utils/process_src_template.py found: YES (/path/to/buildroot/output/host/bin/python3 /path/to/buildroot/output/build/host-python-numpy-1.25.0/numpy/_build_utils/process_src_template.py)
>     Program _build_utils/tempita.py found: YES (/path/to/buildroot/output/build/host-python-numpy-1.25.0/numpy/_build_utils/tempita.py)
> 
>     output/build/host-python-numpy-1.25.0/numpy/meson.build:201:44: ERROR: Could not get cmake variable and no default provided for <CMakeDependency OpenBLAS: True None>
>     [...]
> 
> Two obvious ways to correct this:
> 
> - Add some kind of host-lapack / host-openblas dependency to
>   host-python-numpy, or
> - Tell host-python-numpy not to look for (or rely on) external code.
> 
> Because the host numpy environment shouldn't be doing any numerical
> heavy lifting, we opt for the latter here.
> 
> Signed-off-by: Graeme Smecher <gsmecher@threespeedlogic.com>

Good commit log, but there is a missing piece to it, see below...

James: as far as I understand, this was the situation before the bump to
1.25.0 and the switch to meson, because we did not have the BLAS=None
and LAPACK=None options before that either, but had no such failure
report.

So I guess the defaults changed with the switch to meson, which might
explain why we did not notice before?

If so, we need to add that info to the commit log, to decide whether
this needs backporting to 2023.11, and what to do about 2023.02 LTS that
does not have 1.25.

Regards,
Yann E. MORIN.

> ---
>  package/python-numpy/python-numpy.mk | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/package/python-numpy/python-numpy.mk b/package/python-numpy/python-numpy.mk
> index b52e0da17d..cc5dfc9224 100644
> --- a/package/python-numpy/python-numpy.mk
> +++ b/package/python-numpy/python-numpy.mk
> @@ -36,6 +36,10 @@ else
>  PYTHON_NUMPY_CONF_OPTS += -Dblas=""
>  endif
>  
> +# Rather than add a host-blas or host-lapack dependencies, just use unoptimized,
> +# in-tree code.
> +HOST_PYTHON_NUMPY_CONF_OPTS = -Dblas="" -Dlapack=""
> +
>  # Fixup the npymath.ini prefix path with actual target staging area where
>  # numpy core was built. Without this, target builds using numpy distutils
>  # extensions like python-scipy, python-numba cannot find -lnpymath since
> -- 
> 2.39.2
> 

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 561 099 427 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'
_______________________________________________
buildroot mailing list
buildroot@buildroot.org
https://lists.buildroot.org/mailman/listinfo/buildroot