Release to PyPI

.github/workflows/ci-cmake_tests.yml

@@ -1,8 +1,5 @@
 name: Tests and Codecov
 on:
-  pull_request:
-    branches:
-      - master
   workflow_dispatch:
 jobs:
   BuildAndTest:

.github/workflows/clang-format-check.yml

@@ -1,6 +1,5 @@
 name: clang-format Check
 on:
-  pull_request:
   workflow_dispatch:
 jobs:
   formatting-check:

.github/workflows/conda_build.yml

@@ -4,9 +4,6 @@ on:
   push:
     branches:
       - master
-  pull_request:
-    branches:
-      - master
   workflow_dispatch:
 
 jobs:

.github/workflows/release_pypi.yml

@@ -0,0 +1,78 @@
+name: Release to PyPI
+
+on:
+  pull_request:
+  workflow_dispatch:
+
+jobs:
+  BuildWheel:
+    name: BuildWheel-${{ matrix.os }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [ubuntu-latest, macos-14, macos-15-intel]
+        # os: [ubuntu-latest]
+    defaults:
+      run:
+        shell: bash -l {0}
+
+    steps:
+    - uses: actions/checkout@v4
+      with:
+        fetch-depth: 0
+
+    # When re-running a PR job, GitHub Actions uses the merge commit created at
+    # the time of the original run, not a fresh merge with the latest target branch.
+    # This step ensures we always test against the most recent target branch.
+    - name: Merge with latest target branch (pull_request only)
+      if: github.event_name == 'pull_request'
+      run: |
+        git fetch origin ${{ github.event.pull_request.base.ref }}
+        git merge --no-edit origin/${{ github.event.pull_request.base.ref }}
+
+    - name: Cache Ccache Directory (pull_request)
+      if: ${{ github.event_name == 'pull_request' }}
+      uses: actions/cache@v4
+      with:
+        path: |
+          ~/.ccache
+        key: ccache-${{ runner.os }}-${{ github.event.pull_request.head.ref }}-${{ github.sha }}
+        restore-keys: |
+          ccache-${{ runner.os }}-${{ github.event.pull_request.head.ref }}-
+          ccache-${{ runner.os }}-${{ github.event.pull_request.base.ref }}-
+
+    - name: Set Ccache Directory
+      run: |
+        echo "Start building---------------------------------"
+        export CCACHE_DIR="${HOME}/.ccache"
+
+    - name: Setup macOS
+      if: matrix.os == 'macos-15-intel' || matrix.os == 'macos-14'
+      run: |
+        if [[ ${{ matrix.os }} == 'macos-15-intel' ]]; then
+          echo "MACOSX_DEPLOYMENT_TARGET=15.0" >> "$GITHUB_ENV"
+        elif [[ ${{ matrix.os }} == 'macos-14' ]]; then
+          echo "MACOSX_DEPLOYMENT_TARGET=14.0" >> "$GITHUB_ENV"
+        fi
+
+    - name: Build Wheels
+      uses: pypa/cibuildwheel@v3.3.0
+      env:
+        CMAKE_C_COMPILER_LAUNCHER: ccache
+        CMAKE_CXX_COMPILER_LAUNCHER: ccache
+        CMAKE_CUDA_COMPILER_LAUNCHER: ccache
+        # DO NOT enable this line because the script for conda build is not
+        # executed in a interactive shell, and "~" will not be expanded to the
+        # home directory. Export CCACHE_DIR in the run command instead.
+        # CCACHE_DIR: ~/.ccache
+        CCACHE_MAXSIZE: 1G  # The limit of actions/cache is 10GB
+        # Pass environment variables into cibuildwheel build environments
+        # This is only needed on Linux becuase the wheels are built in docker
+        # images only on Linux.
+        CIBW_ENVIRONMENT_PASS_LINUX: CMAKE_C_COMPILER_LAUNCHER CMAKE_CXX_COMPILER_LAUNCHER CMAKE_CUDA_COMPILER_LAUNCHER CCACHE_MAXSIZE
+
+    - uses: actions/upload-artifact@v4
+      with:
+        name: cibw-wheels-${{ matrix.os }}-${{ strategy.job-index }}
+        path: ./wheelhouse/*.whl

CMakeLists.txt

@@ -191,6 +191,14 @@ if(USE_CUDA)
   set(CMAKE_CUDA_USE_RESPONSE_FILE_FOR_OBJECTS 0)
 endif()
 
+
+# Print all environment variables for debugging
+message(STATUS "==================== Environment Variables ====================")
+execute_process(COMMAND ${CMAKE_COMMAND} -E environment
+                OUTPUT_VARIABLE ENV_OUTPUT)
+message(STATUS "${ENV_OUTPUT}")
+message(STATUS "================================================================")
+
 # C++ uses link-time optimization anyway; this enables additionally -flto=auto,
 # for parallel compilation
 # It cannot enable on MacOS since it causes building errors when linking the library libcytnx.a.

pyproject.toml

@@ -43,6 +43,10 @@ cmake.args = [
     # conflicts against the build process of HPTT.
     "-G Unix Makefiles",
 ]
+# Force the build directory to be "build" instead of a random temp folder. This helps
+# ccache caches the compiler outputs successfully.
+build-dir = "build/{wheel_tag}"
+# cp39-cp39-macosx_14_0_arm64
 
 [tool.scikit-build.metadata.version]
 provider = "scikit_build_core.metadata.regex"
@@ -53,3 +57,31 @@ set\(\w+?VERSION_MINOR\s+(?P<minor>\d+)\)
 set\(\w+?VERSION_PATCH\s+(?P<patch>\d+)\)
 '''
 result = "{major}.{minor}.{patch}"
+
+[tool.cibuildwheel]
+# The image used to build wheels for x86_64 on Linux is manylinux_2_28, which supports
+# older version of Linux than manylinux_2_34. However, manylinux_2_28 is AlmaLinux 8
+# based and AlmaLinux 8 ships Boost v1.66. The Boost's CMake configuration files are only
+# available since v1.70, so AlmaLinux 9 based manylinux_2_34 is picked here.
+# There is a caveat for manylinux_2_34. manylinux_2_34 assume the machines use x86-64-v2
+# architecture which is an extension of x86_64, so the wheels built with manylinux_2_34
+# may not support older machines. See https://github.com/pypa/manylinux?tab=readme-ov-file#manylinux_2_34-almalinux-9-based---alpha
+# for more information.
+manylinux-x86_64-image = "manylinux_2_34"
+before-build = "python ./tools/cibuildwheel_before_build.py"
+
+[tool.cibuildwheel.linux]
+# before-all = "apt-get install -y arpack boost ccache libomp openblas"
+before-all = "bash ./tools/cibuildwheel_before_all.sh"
+# To include LAPACKE header.
+environment = { CMAKE_PREFIX_PATH = "/usr/include/openblas", CCACHE_CONFIGPATH = "~/ccache.conf" }
+
+[tool.cibuildwheel.macos]
+before-all = "brew update; brew install arpack boost ccache libomp openblas"
+environment = { CMAKE_PREFIX_PATH = "$(brew --prefix arpack):$(brew --prefix boost):$(brew --prefix libomp):$(brew --prefix openblas)", PATH = "$(brew --prefix ccache)/libexec:$PATH", CCACHE_CONFIGPATH = "~/ccache.conf" }
+
+# [tool.cibuildwheel]
+# before-build = "bash {project}/tools/cibuildwheel_before_build.sh"
+# # environment = { CMAKE_PREFIX_PATH = "$(brew --prefix openblas)", PATH = "$(brew --prefix ccache)/libexec:$PATH" }
+# # Alternative static configuration for macOS:
+# environment = { CMAKE_PREFIX_PATH = "/opt/homebrew/opt/openblas", PATH = "/opt/homebrew/opt/ccache/libexec:$PATH" }

tools/cibuildwheel_before_all.sh

@@ -0,0 +1,18 @@
+set -xe
+
+yum search lapack
+yum --showduplicates list "lapack*"
+# yum install -y arpack-devel boost-devel ccache lapack-devel libomp-devel openblas-devel
+yum install -y arpack-devel boost-devel ccache libomp-devel openblas-devel
+# On Red Hat/CentOS systems (which manylinux images are based on), the lapack-devel
+# package typically provides the Fortran bindings and miss the C-interface header, so
+# use lapacke.h shipped with openblas instead.
+ls /usr/include
+ln -s /usr/include/openblas/lapacke.h /usr/include/lapacke.h
+ln -s /usr/include/openblas/lapack.h /usr/include/lapack.h
+ln -s /usr/include/openblas/lapacke_mangling.h /usr/include/lapacke_mangling.h
+
+ls /usr/include/openblas
+#  boost1.78-devel
+# ls /usr/include/boost | head
+# ls /usr/lib64/libboost_*.so

tools/cibuildwheel_before_build.py

@@ -0,0 +1,22 @@
+import os
+import pathlib
+import sys
+
+import packaging.tags
+tag = f'cp{sys.version_info.major}{sys.version_info.minor}'
+
+ccache_config_path = os.getenv('CCACHE_CONFIGPATH')
+if not ccache_config_path:
+    raise RuntimeError('The CCACHE_CONFIGPATH environment variable must be set.')
+
+print("ccache_config_path:", ccache_config_path)
+print("ccache_config_path:", pathlib.Path(ccache_config_path).absolute())
+
+
+wheel_tag = str(next(packaging.tags.sys_tags()))
+print("wheel_tag:", wheel_tag)
+
+with open(ccache_config_path, 'w') as f:
+    f.writelines([
+        f'base_dir = /project/build/{wheel_tag}'
+    ])

tools/cibuildwheel_before_build.sh

@@ -0,0 +1,55 @@
+set -xe
+
+# Install OpenBLAS
+# python -m pip install scipy-openblas64
+# blas_lib_dir=$(python -c "import scipy_openblas64; print(scipy_openblas64.get_lib_dir())")
+# blas_lib_name=$(python -c "import scipy_openblas64; print(scipy_openblas64.get_library())")
+# # This handles different extensions (.so on Linux, .dylib on macOS)
+# blas_lib_file=$(find "$blas_lib_dir" -name "*${blas_lib_name}.*" -type f | head -n 1)
+# export BLAS_LIBRARIES="$blas_lib_file"
+
+# # Verify BLAS_LIBRARIES points to an existing file
+# if [ -f "$BLAS_LIBRARIES" ]; then
+#     echo "✓ BLAS_LIBRARIES points to existing file: $BLAS_LIBRARIES"
+# else
+#     echo "✗ ERROR: BLAS_LIBRARIES does not point to a valid file: $BLAS_LIBRARIES"
+#     exit 1
+# fi
+
+# BLA_SIZEOF_INTEGER=8 matches the ILP64 openblas64_ interface
+# that scipy-openblas64 provides; this avoids CMake thinking
+# it found a 32-bit BLAS.
+# export BLA_SIZEOF_INTEGER=8
+
+
+uname -s
+# Set up Homebrew for Linux
+# Homebrew was removed from PATH on Linux:
+# https://github.com/actions/runner-images/issues/6283
+if [[ "$(uname -s)" == "Linux" ]]; then
+    # Find linuxbrew from root folder recursively
+    # Common locations: /home/linuxbrew/.linuxbrew, /home/*/linuxbrew/.linuxbrew
+    # brew_find=$(find / -name "*linuxbrew*" -type d 2>/dev/null)
+    # linuxbrew_bin=$(echo $brew_find | head -n 1)
+    # if [ -n "$linuxbrew_bin" ] && [ -d "$linuxbrew_bin/bin" ]; then
+    #     echo "Found linuxbrew at: $linuxbrew_bin"
+    #     export PATH="$linuxbrew_bin/bin:$PATH"
+    # else
+    #     echo "Warning: linuxbrew not found, trying default path"
+    #     export PATH="/home/linuxbrew/.linuxbrew/bin:$PATH"
+    # fi
+    export PATH="/host/home/linuxbrew/.linuxbrew/bin:$PATH"
+fi
+
+# Install boost, ccache and arpack-ng
+# All Ubuntu and macOS GitHub action runners have Homebrew installed.
+brew install boost ccache arpack openblas
+
+brew --prefix openblas
+ls $(brew --prefix openblas)
+brew --prefix ccache
+ls $(brew --prefix ccache)
+brew --prefix arpack
+ls $(brew --prefix arpack)
+brew --prefix boost
+ls $(brew --prefix boost)