Fix the numpy ptests by doing the following:
- Add meson to ptest RDEPENDS in the recipe;
- Add python3-unittest-automake-output as a ptest RDEPENDS;
- Convert run-ptest to a shell script that sets PYTEST_DEBUG_TEMPROOT to
a directory inside the same path that contains the script, create that
directory, and then invoke the tests with `pytest --automake` pointed
at the numpy path in site-packages (copying the tests into the normal
PTEST_DIRECTORY seems to cause module import breakages in some tests).
This also includes skipping two problematic tests which require a C
compiler and use up a lot of space, respectively;
- Set 'IMAGE_ROOTFS_EXTRA_SPACE = "3048576"' for python3-numpy in
core-image-ptest.bb;
- Also set 'QB_MEM:virtclass-mcextend-python3-numpy = "-m 4096"' in
core-image-ptest.bb;
- Move python3-numpy from the PTESTS_PROBLEMS list to the PTESTS_SLOW
one.
Results on qemux86-64:
Testsuite summary
DURATION: 87
END: /usr/lib/python3-numpy/ptest
2025-04-29T17:35
STOP: ptest-runner
TOTAL: 1 FAIL: 0
Note that many of the skipped tests are due to the absence of
python3-mypy (which is currently in meta-python) and of a Fortran
compiler.
(From OE-Core rev: 7f0b5e8faa1b246531ac425c99a629eb344b21e2)
Signed-off-by: Trevor Gamblin <tgamblin@baylibre.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Add a svml option, disabled by default
(it's avx-512 only, and uses randomly generated
intermediate .o file names, which makes the installed
library non-reproducible).
(From OE-Core rev: a3ce94f1fa79502d8b41de93f9a27d8af2cdc787)
Signed-off-by: Alexander Kanavin <alex@linutronix.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
- This release contains numerous API changes and bug fixes.
Changelog: https://github.com/numpy/numpy/releases
- Notably, the build backend has switched to mesonpy since 1.26.4.
- Patch 0001-numpy-core-Define-RISCV-32-support.patch had to be updated
to change the paths for the target files. It has now been merged
upstream, but isn't included in any release yet. PR for reference:
https://github.com/numpy/numpy/pull/17780
- Builds OK and seems to be compatible with current piglit and pandas
(meta-python), which was not the case for the previous upgrade attempt
to numpy 2.0.1.
- For numpy, add 'pkgconfig' to inherit to avoid errors like this at
do_compile:
| Found Pkg-config: NO
| Run-time dependency python found: YES 3.12
| Has header "Python.h" with dependency python: NO
|
| ../numpy-2.0.0/meson.build:44:2: ERROR: Problem encountered: Cannot compile `Python.h`. Perhaps you need to install python-dev|python-devel
|
| A full log can be found at /home/tgamblin/workspace/yocto/poky/build/tmp/work/core2-64-poky-linux/python3-numpy/2.0.0/build/meson-logs/meson-log.txt
- Mmodify the FILES:${PN}-staticdev line to fix a QA issue.
- This is now reproducible as of oe-core commit: 917df5ed022f
License-Update: Change copyright year to 2024
buildall-qemu log:
BUILDALL-QEMU LOG FOR python3-numpy
START TIME: 2024-11-20_14:02:08
HOSTNAME: megalith
HOST OS: Debian GNU/Linux 12 (bookworm)
HOST KERNEL: 6.1.0-27-amd64
===============
BUILD RESULTS:
[glibc]
FAIL: qemuloongarch64
PASS: qemuriscv32
PASS: qemuarmv5
PASS: qemuppc
PASS: qemumips64
PASS: qemuriscv64
PASS: qemuarm64
PASS: qemuarm
PASS: qemux86-64
PASS: qemuppc64
PASS: qemux86
PASS: qemumips
[musl]
FAIL: qemuloongarch64
FAIL: qemuriscv32
PASS: qemuarmv5
PASS: qemuppc
PASS: qemumips64
PASS: qemuriscv64
PASS: qemuarm64
PASS: qemuarm
PASS: qemux86-64
PASS: qemuppc64
PASS: qemux86
PASS: qemumips
===============
PASSED: 21
FAILED: 3
(From OE-Core rev: fd656aaf7fbed3115e3b4251ff76257ee8f19c82)
Signed-off-by: Trevor Gamblin <tgamblin@baylibre.com>
Signed-off-by: Mathieu Dubois-Briand <mathieu.dubois-briand@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Changelog: https://github.com/numpy/numpy/releases/tag/v1.26.4
Upgrade to pull in a variety of bug fixes.
The same number of ptests (6) fail as in 1.26.3, so no regression there.
(From OE-Core rev: 0314d2b43701984b5648bce9ca9aaeba5f768b78)
Signed-off-by: Trevor Gamblin <tgamblin@baylibre.com>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This was supposed to always be the case from upstream but was exposed
by the 64-bit-time QA checker when file function scanning was enabled.
(From OE-Core rev: 3d556e59f658ac29615fb7a14b6ea48533122ff6)
Signed-off-by: Ola x Nilsson <olani@axis.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Add a patch to fix the reproducibility issue being seen during testing. The issue
is from a poor choice of regex against the compiler flags, being triggered by
different path names.
Drop the previous attempt at fixing this as it wasn't that.
(From OE-Core rev: c4d1dc5e33734b94835e7f5e0e1746d4a6542b55)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
We've seen reproducibility failures where it appears an extra -O3 compiler flag ends
up in the multiarray library compilation. This can only really have come through
extra_info for BLASS support since it only affects just this library. Rather than
try and track down exactly where this came from in a rabbit warren of code, just
disable this since we don't have any of the dependencies anyway.
(From OE-Core rev: 1d07be514ccacedb4d1ac48e0fdd3a36ae098698)
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
* remove patch applied upstream
* FWIW: this version still fails to build with DEBUG_BUILD AND gcc-13 as
shown in:
http://errors.yoctoproject.org/Errors/Details/689841/
In file included from TOPDIR/tmp-glibc/work/core2-64-oe-linux/python3-numpy/1.24.1-r0/recipe-sysroot-native/usr/lib/x86_64-oe-linux/gcc/x86_64-oe-linux/13.0.1/include/immintrin.h:57,
from TOPDIR/tmp-glibc/work/core2-64-oe-linux/python3-numpy/1.24.1-r0/numpy-1.24.1/numpy/distutils/checks/cpu_avx512_knl.c:14:
In function '_mm512_mask_prefetch_i64scatter_pd',
inlined from 'main' at TOPDIR/tmp-glibc/work/core2-64-oe-linux/python3-numpy/1.24.1-r0/numpy-1.24.1/numpy/distutils/checks/cpu_avx512_knl.c:23:5:
TOPDIR/tmp-glibc/work/core2-64-oe-linux/python3-numpy/1.24.1-r0/recipe-sysroot-native/usr/lib/x86_64-oe-linux/gcc/x86_64-oe-linux/13.0.1/include/avx512pfintrin.h:180:3: error: 'base' may be used uninitialized [-Werror=maybe-uninitialized]
180 | __builtin_ia32_scatterpfqpd (__mask, (__v8di) __index, __addr, __scale,
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
181 | __hint);
| ~~~~~~~
<built-in>: In function 'main':
<built-in>: note: by argument 3 of type 'const void *' to '__builtin_ia32_scatterpfqpd' declared here
TOPDIR/tmp-glibc/work/core2-64-oe-linux/python3-numpy/1.24.1-r0/numpy-1.24.1/numpy/distutils/checks/cpu_avx512_knl.c:18:9: note: 'base' declared here
18 | int base[128];
| ^~~~
In file included from TOPDIR/tmp-glibc/work/core2-64-oe-linux/python3-numpy/1.24.1-r0/recipe-sysroot-native/usr/lib/x86_64-oe-linux/gcc/x86_64-oe-linux/13.0.1/include/immintrin.h:53,
from numpy/core/src/umath/simd.inc.src:25:
TOPDIR/tmp-glibc/work/core2-64-oe-linux/python3-numpy/1.24.1-r0/recipe-sysroot-native/usr/lib/x86_64-oe-linux/gcc/x86_64-oe-linux/13.0.1/include/avx512fintrin.h: In function 'AVX512F_square_CFLOAT':
TOPDIR/tmp-glibc/work/core2-64-oe-linux/python3-numpy/1.24.1-r0/recipe-sysroot-native/usr/lib/x86_64-oe-linux/gcc/x86_64-oe-linux/13.0.1/include/avx512fintrin.h:314:1: error: inlining failed in call to 'always_inline' '_mm512_setzero_ps': target specific option mismatch
314 | _mm512_setzero_ps (void)
| ^~~~~~~~~~~~~~~~~
numpy/core/src/umath/simd.inc.src:977:20: note: called from here
977 | @vtype@ zeros = _mm512_setzero_@vsuffix@();
| ^~~~~~~~~~~~~~~~~~~
numpy/core/src/umath/simd.inc.src:596:1: error: inlining failed in call to 'always_inline' 'avx512_get_full_load_mask_ps': target specific option mismatch
596 | avx512_get_full_load_mask_ps(void)
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
It will be probably resolved in next release from 1.25 as whole simd.inc.src was removed in:
640e85017a
but this PR https://github.com/numpy/numpy/pull/21056 wasn't backported
to maintenance/1.24.x and probably isn't worth backporting now as
neither DEBUG_BUILD is enabled by default nor gcc-13 merged in oe-core,
and the changes don't backport cleanly.
(From OE-Core rev: 8596678667797971559aed962b1c204266032186)
Signed-off-by: Martin Jansa <Martin.Jansa@gmail.com>
Signed-off-by: Alexandre Belloni <alexandre.belloni@bootlin.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
Replace an ad hoc fix via patch with an upstream one.
(From OE-Core rev: 991f8617b9ca9680b3f81cb215e3b39095260e7c)
Signed-off-by: Alexander Kanavin <alex@linutronix.de>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
This recipe had been moved out from python directory since 2016[1] in
order to share patches between python2 and python3. But now there is no
reason to keep it in its own directory as we only keep python3-nump.
Move it back to python directory.
[1] https://git.openembedded.org/openembedded-core/commit/?id=9bffe2f9fb4ce6c0b265f27e5b484fbe076c6349
(From OE-Core rev: e5702b8ebd16949c736b8c2f18bf35c0d64d80bd)
Signed-off-by: Yi Zhao <yi.zhao@windriver.com>
Signed-off-by: Richard Purdie <richard.purdie@linuxfoundation.org>
