scipy: BUG: large number of fails with macOS 15.1 using Accelerate

Describe your issue.

I’m seeing a large number of fails on macOS 15.1 (Sequoia, ARM M1) using Accelerate. I don’t see those fails with scipy-openblas32. This leads me to believe that there’s a problem with Accelerate, or how we’re using it.

I did this test in a new conda environment, did git clean -xdf, etc.

The test log is too long to insert here, you can find it in a gist here

@tylerjereddy, do you have any issues with Sequoia?

Reproducing Code Example

# conda create -c conda-forge -n dev312 python=3.12
# conda activate dev312
# pip install numpy cython ninja meson pythran pybind11 doit click rich_click pydevtool

python dev.py build --with-accelerate
python dev.py -n test -s linalg

Error message

N/A

SciPy/NumPy/Python version and system information

numpy=

Project version: 1.15.0.dev0+git20241110.b919b4a
C compiler for the host machine: cc (clang 16.0.0 "Apple clang version 16.0.0 (clang-1600.0.26.4)")
C linker for the host machine: cc ld64 classic  PROJECT:ld64
C++ compiler for the host machine: c++ (clang 16.0.0 "Apple clang version 16.0.0 (clang-1600.0.26.4)")
C++ linker for the host machine: c++ ld64 classic  PROJECT:ld64
Cython compiler for the host machine: cython (cython 3.0.11)
Host machine cpu family: aarch64
Host machine cpu: aarch64
Program python3 found: YES (/Users/andrew/miniforge3/envs/dev312/bin/python3.12)
Found pkg-config: YES (/usr/local/bin/pkg-config) 0.29.2
Run-time dependency python found: YES 3.12
Program cython found: YES (/Users/andrew/miniforge3/envs/dev312/bin/cython)
Compiler for C supports arguments -Wno-unused-but-set-variable: YES 
Compiler for C supports arguments -Wno-unused-function: YES 
Compiler for C supports arguments -Wno-conversion: YES 
Compiler for C supports arguments -Wno-misleading-indentation: YES 
Library m found: YES
Fortran compiler for the host machine: gfortran (gcc 13.2.0 "GNU Fortran (Homebrew GCC 13.2.0) 13.2.0")
Fortran linker for the host machine: gfortran ld64 classic  PROJECT:ld64
Compiler for Fortran supports arguments -Wno-conversion: YES 
Compiler for C supports link arguments -Wl,-ld_classic: YES 
Compiler for C supports link arguments -Wl,-dead_strip: YES 
Checking if "-Wl,--version-script" : links: NO 
Program tools/generate_f2pymod.py found: YES (/Users/andrew/miniforge3/envs/dev312/bin/python3.12 /Users/andrew/Documents/Andy/programming/scipy/tools/generate_f2pymod.py)
Program scipy/_build_utils/tempita.py found: YES (/Users/andrew/miniforge3/envs/dev312/bin/python3.12 /Users/andrew/Documents/Andy/programming/scipy/scipy/_build_utils/tempita.py)
Program pythran found: YES 0.17.0 0.17.0 (/Users/andrew/miniforge3/envs/dev312/bin/pythran)
Found CMake: /usr/local/bin/cmake (3.24.0)
WARNING: CMake Toolchain: Failed to determine CMake compilers state
Run-time dependency xsimd found: NO (tried pkgconfig, framework and cmake)
Run-time dependency threads found: YES
Library npymath found: YES
pybind11-config found: YES (/Users/andrew/miniforge3/envs/dev312/bin/pybind11-config) 2.13.6
Run-time dependency pybind11 found: YES 2.13.6
Program f2py found: YES (/Users/andrew/miniforge3/envs/dev312/bin/f2py)
Run-time dependency accelerate found: YES (/Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk/System/Library/Frameworks/Accelerate.framework)


(dev312) 192-168-1-105:scipy andrew$ pip list
Package            Version
------------------ -----------
attrs              24.2.0
beniget            0.4.2.post1
certifi            2024.8.30
charset-normalizer 3.4.0
click              8.1.7
cloudpickle        3.1.0
Cython             3.0.11
doit               0.36.0
gast               0.6.0
hypothesis         6.118.7
idna               3.10
importlib_metadata 8.5.0
iniconfig          2.0.0
markdown-it-py     3.0.0
mdurl              0.1.2
meson              1.5.2
ninja              1.11.1.1
numpy              2.1.3
packaging          24.2
pip                24.3.1
platformdirs       4.3.6
pluggy             1.5.0
ply                3.11
pooch              1.8.2
pybind11           2.13.6
pydevtool          0.3.0
Pygments           2.18.0
pytest             8.3.3
pythran            0.17.0
requests           2.32.3
rich               13.9.4
rich-click         1.8.3
scipy-openblas32   0.3.28.0.2
setuptools         75.3.0
sortedcontainers   2.4.0
typing_extensions  4.12.2
urllib3            2.2.3
wheel              0.45.0
zipp               3.21.0

About this issue

  • Original URL
  • State: closed
  • Created 2 years ago
  • Comments: 22 (18 by maintainers)

Most upvoted comments

macOS 15.2 beta 4 release notes highlight this issue

Resolved an issue impacting the accuracy of some linear algebra operations (*lartg). These functions have been restored to their 3.9.1 implementation. (139760965) (FB15782411)

Please let us know if there are any other issues @andyfaff , et al. Thank you again for the report.

@Developer-Ecosystem-Engineering I have submitted feedback via FB15782411. I’ve included the steps required to reproduce the problems, but not in the form of individually compilable programs. Rather, it gives the steps required to visualise the problems via running of the scipy test suite.

Hi!

Log a Feedback and feel free to @ us here. At a glance, seems like this is all complex128. Taking a deeper look, give us a few days thanks!

suspect that the macosx_14_0 wheels for SciPy 1.14.1 will show the same failures when run on macOS 15 (at least if it’s an Accelerate problem), could you confirm that?

The failures are present in the 1.14.1 wheels from PyPI.