`make pyprof` runs super slow on `macos-15` GA runner

[yungyuc]

PR #634 temporarily turns off make pyprof step in build_macos-15_Release because one or more profiling scripts takes super long time, like hours:

Run make pyprof
  make pyprof
  shell: /bin/bash -e {0}
  env:
    QT_DEBUG_PLUGINS: 1
    PIP_BREAK_SYSTEM_PACKAGES: 1
    AQT_CONFIG: thirdparty/aqt_settings.ini
    PKG_CONFIG_PATH: /Users/runner/work/modmesh/Qt/6.8.1/macos/lib/pkgconfig
    QT_ROOT_DIR: /Users/runner/work/modmesh/Qt/6.8.1/macos
    QT_PLUGIN_PATH: /Users/runner/work/modmesh/Qt/6.8.1/macos/plugins
    QML2_IMPORT_PATH: /Users/runner/work/modmesh/Qt/6.8.1/macos/qml
cmake --build build/dev314 --target _modmesh_py VERBOSE= -j
[ 96%] Built target modmesh_primary
[100%] Built target _modmesh
[100%] Built target _modmesh_py
/opt/homebrew/bin/python3 profiling/profile_arithmetic_simd.py > profiling/results/profile_arithmetic_simd.output
/Users/runner/work/modmesh/modmesh/profiling/profile_arithmetic_simd.py:111: RuntimeWarning: divide by zero encountered in divide
  return np.divide(src1, src2)
/opt/homebrew/bin/python3 profiling/profile_arithmetic.py > profiling/results/profile_arithmetic.output
/opt/homebrew/bin/python3 profiling/profile_median.py > profiling/results/profile_median.output
/opt/homebrew/bin/python3 profiling/profile_statistic_ops.py > profiling/results/profile_statistic_ops.output
Matplotlib is building the font cache; this may take a moment.
/opt/homebrew/bin/python3 profiling/profile_take_along_axis.py > profiling/results/profile_take_along_axis.output
Error: The operation was canceled.

@KHLee529 suggested:

From the error message it seems like that profile_take_along_axis.py is the script that cause infinite run. Could you please try to first remove this script by renaming it to any name without prefix profile_?

We need to fix the bad profiling script (or SIMD code).

[tychuang1211]

The workflow appears to stuck at the line profile_take_along_axis_simd(test_sa, idx_sa). Switching to macos-15-intel resolves the issue, so I suspect the SIMD code may not be working correctly on arm64. I’ll try to pinpoint the issue, but unfortunately I don’t have access to a Mac myself.

[yungyuc]

The workflow appears to stuck at the line profile_take_along_axis_simd(test_sa, idx_sa). Switching to macos-15-intel resolves the issue, so I suspect the SIMD code may not be working correctly on arm64. I’ll try to pinpoint the issue, but unfortunately I don’t have access to a Mac myself.

The SIMD code is only with Neon and should be running on arm. It is not executed on x86. Please discuss with @KHLee529 .

[nccvector]

Addressing in #641