Updates Linux container and CUDA version

This commit updates Linux builds from the manylinux2014 container to the
manylinux_2_28 container. The manylinux2014 container uses GCC 10.2,
which produces code that runs substantially slower than more recent
versions (~4 times in my tests). Code compiled on manylinux_2_28 should
still run on most systems, so portability will not be greatly reduced.

This commit also changes the version of the CUDA Toolkit used on the
Linux builds from 11.6 to 12.4. This matches the Windows builds (where
the more recent version was forced by the older version no longer being
supported), and should result in the inclusion of code compiled for more
recent GPU architectures.

Author: ar4

.github/workflows/build.yml

@@ -9,7 +9,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-latest, windows-2022, macos-13, macos-14]
+        os: [ubuntu-latest] #, windows-2022, macos-13, macos-14]
 
     steps:
       - uses: actions/checkout@v5
@@ -25,14 +25,13 @@ jobs:
         shell: bash
 
       - name: Build wheels
-        uses: pypa/cibuildwheel@v3.1.4
+        uses: pypa/cibuildwheel@v3.3.0
         env:
-          # For Linux, install CUDA in the container. Note that manylinux2014
+          # For Linux, install CUDA in the container. Note that manylinux_2_28
           # is specified in pyproject.toml
           CIBW_BEFORE_ALL_LINUX: >
-            yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel7/x86_64/cuda-rhel7.repo &&
-            yum install -y cuda-nvcc-11-8 cuda-cudart-devel-11-8
-          CIBW_ENVIRONMENT_LINUX: PATH="$PATH:/usr/local/cuda-11.8/bin" CUDA_PATH="/usr/local/cuda-11.8/bin" CUDA_HOME="/usr/local/cuda-11.8" CUDA_ROOT="/usr/local/cuda-11.8" LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda-11.8/lib64" CUDACXX="/usr/local/cuda-11.8/bin/nvcc"
+            curl -sL https://developer.download.nvidia.com/compute/cuda/repos/rhel8/x86_64/cuda-rhel8.repo -o /etc/yum.repos.d/cuda-rhel8.repo && dnf install -y cuda-nvcc-12-4 cuda-cudart-devel-12-4 gcc-toolset-13 && scl enable gcc-toolset-13 bash
+          CIBW_ENVIRONMENT_LINUX: PATH="$PATH:/usr/local/cuda-12.4/bin" CUDA_PATH="/usr/local/cuda-12.4/bin" CUDA_HOME="/usr/local/cuda-12.4" CUDA_ROOT="/usr/local/cuda-12.4" LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda-12.4/lib64" CUDACXX="/usr/local/cuda-12.4/bin/nvcc" NVCC_PREPEND_FLAGS="--allow-unsupported-compiler" CMAKE_CUDA_FLAGS="--allow-unsupported-compiler"
           # For Windows, install CUDA
           CIBW_BEFORE_ALL_WINDOWS: >
             nuget install intelopenmp.devel.win -DirectDownload -NonInteractive &&
@@ -50,15 +49,15 @@ jobs:
           name: cibw-wheels-${{ matrix.os }}-${{ strategy.job-index }}
           path: ./wheelhouse/*.whl
 
-  build_sdist:
-    name: Build source distribution
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v5
-
-      - name: Build sdist
-        run: pipx run build --sdist
-
-      - uses: actions/upload-artifact@v4
-        with:
-          path: dist/*.tar.gz
+            #  build_sdist:
+            #    name: Build source distribution
+            #    runs-on: ubuntu-latest
+            #    steps:
+            #      - uses: actions/checkout@v5
+            #
+            #      - name: Build sdist
+            #        run: pipx run build --sdist
+            #
+            #      - uses: actions/upload-artifact@v4
+            #        with:
+            #          path: dist/*.tar.gz

CMakeLists.txt

@@ -3,12 +3,16 @@
 cmake_minimum_required(VERSION 3.18)
 # If scikit-build-core sets SKBUILD_PROJECT_NAME / VERSION, prefer those.
 if(DEFINED SKBUILD_PROJECT_NAME)
-    project(${SKBUILD_PROJECT_NAME} LANGUAGES C)
+	project(${SKBUILD_PROJECT_NAME} LANGUAGES C CXX)
 else()
-    project(deepwave LANGUAGES C)
+	project(deepwave LANGUAGES C CXX)
 endif()
 
 # Detect and enable CUDA if present
+#if(CMAKE_C_COMPILER_ID STREQUAL "GNU" AND CMAKE_C_COMPILER_VERSION VERSION_GREATER_EQUAL "14.0")
+#  message(STATUS "GCC ${CMAKE_C_COMPILER_VERSION} detected – enabling nvcc unsupported compiler override")
+#  set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -allow-unsupported-compiler")
+#endif()
 find_package(CUDAToolkit)
 if(CUDAToolkit_FOUND)
     enable_language(CUDA)

pyproject.toml

@@ -34,8 +34,8 @@ build = "cp311-*"
 skip = ["*musllinux*"]
 
 # Use an older manylinux image for better compatibility
-manylinux-x86_64-image = "manylinux2014"
-manylinux-aarch64-image = "manylinux2014"
+manylinux-x86_64-image = "manylinux_2_28"
+manylinux-aarch64-image = "manylinux_2_28"
 
 build-verbosity = 1