Moved CUDA environment variables to GitHub Action so that they will not interfere with users building themselves, added -fPIC to C compile, and added OpenMP link to overall library

Author: ar4

.github/workflows/build.yml

@@ -28,6 +28,7 @@ jobs:
             export CUDA_ROOT=/usr/local/cuda-11.8
             export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-11.8/lib64
             export CUDACXX=/usr/local/cuda-11.8/bin/nvcc
+          CIBW_ENVIRONMENT_LINUX: PATH="$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"
           # For Windows, install CUDA
           CIBW_BEFORE_ALL_WINDOWS: |
             cd src/deepwave
@@ -40,6 +41,7 @@ jobs:
             ./cuda_12.4.0_windows_network.exe -s nvcc_12.4 cudart_12.4
             echo "CUDA_PATH=C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4" >> $GITHUB_ENV
             echo "C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\bin" >> $GITHUB_PATH
+          CIBW_ENVIRONMENT_WINDOWS: CUDA_PATH="C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4" PATH="$PATH:C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\bin"
           CUDAToolkit_ROOT: ${{ env.CUDA_PATH }}
 
       - uses: actions/upload-artifact@v4

CMakeLists.txt

@@ -89,9 +89,6 @@ else()
     message(STATUS "AVX2 is not supported.")
 endif()
 
-# Compiler optimization flags
-
-
 # Add warning flags for GCC/Clang
 add_compile_options(
     $<$<C_COMPILER_ID:GNU,Clang>:-Wall>
@@ -107,9 +104,9 @@ endif()
 
 # Prepare lists that will collect the object files
 set(DEEPWAVE_OBJECTS)
-set(SCALAR_ACCURACIES 2 4 6 8)
-set(ELASTIC_ACCURACIES 2 4)
-set(DTYPES float double)
+set(SCALAR_ACCURACIES 2 )#4 6 8)
+set(ELASTIC_ACCURACIES 2 )#4)
+set(DTYPES float )#double)
 
 # --- CPU object libraries ---
 # All CPU object libraries will link to our unified OpenMP target if it was configured.
@@ -130,8 +127,8 @@ foreach(ACCURACY ${SCALAR_ACCURACIES})
 
         if(CMAKE_BUILD_TYPE MATCHES Release)
             if(CMAKE_C_COMPILER_ID MATCHES "GNU" OR CMAKE_C_COMPILER_ID MATCHES "Clang" OR CMAKE_C_COMPILER_ID MATCHES "Intel")
-                target_compile_options(scalar_${ACCURACY}_${DTYPE}_obj PRIVATE -Ofast)
-                target_compile_options(scalar_born_${ACCURACY}_${DTYPE}_obj PRIVATE -Ofast)
+		target_compile_options(scalar_${ACCURACY}_${DTYPE}_obj PRIVATE -Ofast -fPIC)
+		target_compile_options(scalar_born_${ACCURACY}_${DTYPE}_obj PRIVATE -Ofast -fPIC)
             elseif(CMAKE_C_COMPILER_ID MATCHES "MSVC")
                 target_compile_options(scalar_${ACCURACY}_${DTYPE}_obj PRIVATE /O2 /fp:fast)
                 target_compile_options(scalar_born_${ACCURACY}_${DTYPE}_obj PRIVATE /O2 /fp:fast)
@@ -160,7 +157,7 @@ foreach(ACCURACY ${ELASTIC_ACCURACIES})
 
         if(CMAKE_BUILD_TYPE MATCHES Release)
             if(CMAKE_C_COMPILER_ID MATCHES "GNU" OR CMAKE_C_COMPILER_ID MATCHES "Clang" OR CMAKE_C_COMPILER_ID MATCHES "Intel")
-                target_compile_options(elastic_${ACCURACY}_${DTYPE}_obj PRIVATE -Ofast)
+		target_compile_options(elastic_${ACCURACY}_${DTYPE}_obj PRIVATE -Ofast -fPIC)
             elseif(CMAKE_C_COMPILER_ID MATCHES "MSVC")
                 target_compile_options(elastic_${ACCURACY}_${DTYPE}_obj PRIVATE /O2 /fp:fast)
             endif()
@@ -233,6 +230,10 @@ set_target_properties(deepwave_C PROPERTIES
     WINDOWS_EXPORT_ALL_SYMBOLS ON
 )
 
+if(OPENMP_CONFIGURED)
+    target_link_libraries(deepwave_C PRIVATE Deepwave_OpenMP_Interface)
+endif()
+
 if(CUDAToolkit_FOUND)
     target_link_libraries(deepwave_C PRIVATE CUDA::cudart)
 endif()

pyproject.toml

@@ -51,11 +51,11 @@ test-requires = "pytest scipy torch"
 test-command = "pytest {project}/tests"
 # Don't bundle CUDA libraries with the wheel
 repair-wheel-command = "auditwheel repair -w {dest_dir} {wheel} --exclude libcuda.so.1 --exclude 'libcudart.so.*'"
-environment = { 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" }
+#environment = { 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" }
 
 [tool.cibuildwheel.windows]
 test-command = "pytest {project}/tests/test_scalar.py"
-environment = { CUDA_PATH="C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4", PATH="$PATH:C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\bin" }
+#environment = { CUDA_PATH="C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4", PATH="$PATH:C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v12.4\\bin" }
 
 [tool.cibuildwheel.macos]
 test-command = "pytest {project}/tests/test_scalar.py"