remove OpenMP from Windows

ar4 · Jul 21, 2023 · 3e6de3c · 3e6de3c
1 parent 333f134
commit 3e6de3c
Show file tree

Hide file tree

Showing 2 changed files with 114 additions and 121 deletions.
diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
@@ -1,54 +1,54 @@
 name: Build and test
 on: push
 jobs:
-#    Linux-build:
-#        runs-on: ubuntu-latest
-#        container: quay.io/pypa/manylinux2014_x86_64
-#        steps:
-#            - name: Checkout
-#              uses: actions/checkout@v3
-#            - name: Install NVCC
-#              run: |
-#                yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel7/x86_64/cuda-rhel7.repo
-#                yum install -y cuda-nvcc-11-1-11.1.105-1 cuda-cudart-devel-11-1-11.1.74-1
-#            - name: Compile
-#              run: |
-#                PATH=$PATH:/usr/local/cuda-11.1/bin
-#                CUDA_HOME=/usr/local/cuda-11.1
-#                CUDA_ROOT=/usr/local/cuda-11.1
-#                CUDA_PATH=/usr/local/cuda-11.1
-#                CUDADIR=/usr/local/cuda-11.1
-#                LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-11.1/lib64
-#                cd src/deepwave
-#                cp /lib64/libgomp.so.1 .
-#                ./build_linux.sh
-#            - name: Archive built libraries
-#              uses: actions/upload-artifact@v3
-#              with:
-#                name: linux_libraries
-#                path: src/deepwave/*.so*
-#    MacOS-build:
-#        runs-on: macos-11
-#        steps:
-#            - name: Checkout
-#              uses: actions/checkout@v3
-#            - name: Set up Python
-#              uses: actions/setup-python@v3
-#            - name: Install dependencies
-#              run: |
-#                python -m pip install --upgrade pip
-#                python -m pip install torch
-#            - name: Compile
-#              run: |
-#                cd src/deepwave
-#                cp `python -c "import torch; print(torch.__path__[0])"`/lib/libiomp5.dylib .
-#                brew install libomp
-#                ./build_macos.sh
-#            - name: Archive built libraries
-#              uses: actions/upload-artifact@v3
-#              with:
-#                name: macos_libraries
-#                path: src/deepwave/*.dylib
+    Linux-build:
+        runs-on: ubuntu-latest
+        container: quay.io/pypa/manylinux2014_x86_64
+        steps:
+            - name: Checkout
+              uses: actions/checkout@v3
+            - name: Install NVCC
+              run: |
+                yum-config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel7/x86_64/cuda-rhel7.repo
+                yum install -y cuda-nvcc-11-1-11.1.105-1 cuda-cudart-devel-11-1-11.1.74-1
+            - name: Compile
+              run: |
+                PATH=$PATH:/usr/local/cuda-11.1/bin
+                CUDA_HOME=/usr/local/cuda-11.1
+                CUDA_ROOT=/usr/local/cuda-11.1
+                CUDA_PATH=/usr/local/cuda-11.1
+                CUDADIR=/usr/local/cuda-11.1
+                LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda-11.1/lib64
+                cd src/deepwave
+                cp /lib64/libgomp.so.1 .
+                ./build_linux.sh
+            - name: Archive built libraries
+              uses: actions/upload-artifact@v3
+              with:
+                name: linux_libraries
+                path: src/deepwave/*.so*
+    MacOS-build:
+        runs-on: macos-11
+        steps:
+            - name: Checkout
+              uses: actions/checkout@v3
+            - name: Set up Python
+              uses: actions/setup-python@v3
+            - name: Install dependencies
+              run: |
+                python -m pip install --upgrade pip
+                python -m pip install torch
+            - name: Compile
+              run: |
+                cd src/deepwave
+                cp `python -c "import torch; print(torch.__path__[0])"`/lib/libiomp5.dylib .
+                brew install libomp
+                ./build_macos.sh
+            - name: Archive built libraries
+              uses: actions/upload-artifact@v3
+              with:
+                name: macos_libraries
+                path: src/deepwave/*.dylib
     Windows-build:
         runs-on: windows-2019
         defaults:
@@ -59,62 +59,56 @@ jobs:
               uses: actions/checkout@v3
             - name: Set up Python
               uses: actions/setup-python@v3
-#            - name: Install NVCC
-#              run: |
-#                curl https://developer.download.nvidia.com/compute/cuda/11.1.1/network_installers/cuda_11.1.1_win10_network.exe -o cuda_11.1.1_win10_network.exe
-#                chmod +x ./cuda_11.1.1_win10_network.exe
-#                ./cuda_11.1.1_win10_network.exe -s nvcc_11.1 cudart_11.1
-#                echo "CUDA_PATH=C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v11.1" >> $GITHUB_ENV
-#                echo "C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v11.1\\bin" >> $GITHUB_PATH
+            - name: Install NVCC
+              run: |
+                curl https://developer.download.nvidia.com/compute/cuda/11.1.1/network_installers/cuda_11.1.1_win10_network.exe -o cuda_11.1.1_win10_network.exe
+                chmod +x ./cuda_11.1.1_win10_network.exe
+                ./cuda_11.1.1_win10_network.exe -s nvcc_11.1 cudart_11.1
+                echo "CUDA_PATH=C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v11.1" >> $GITHUB_ENV
+                echo "C:\\Program Files\\NVIDIA GPU Computing Toolkit\\CUDA\\v11.1\\bin" >> $GITHUB_PATH
             - name: Setup MSVC
               uses: ilammy/msvc-dev-cmd@v1
-            - name: Install dependencies
-              run: |
-                python -m pip install --upgrade pip
-                python -m pip install torch
             - name: Compile
               run: |
                 cd src/deepwave
-                cp `python -c "import torch; print(torch.__path__[0])"`"\\lib\\libiomp5md.dll" .
-                cp `python -c "import torch; print(torch.__path__[0])"`"\\lib\\libiompstubs5md.dll" .
                 ./build_windows.sh
             - name: Archive built libraries
               uses: actions/upload-artifact@v3
               with:
                 name: windows_libraries
                 path: src/deepwave/*.dll
-#    Test:
-#        strategy:
-#            matrix:
-#                os: [ubuntu-latest, macos-latest, windows-latest]
-#            fail-fast: false
-#        runs-on: ${{ matrix.os }}
-#        needs: [Linux-build, MacOS-build, Windows-build]
-#        steps:
-#            - name: Checkout
-#              uses: actions/checkout@v3
-#            - name: Download built Linux libraries
-#              uses: actions/download-artifact@v3
-#              with:
-#                name: linux_libraries
-#                path: src/deepwave/
-#            - name: Download built MacOS libraries
-#              uses: actions/download-artifact@v3
-#              with:
-#                name: macos_libraries
-#                path: src/deepwave/
-#            - name: Download built Windows libraries
-#              uses: actions/download-artifact@v3
-#              with:
-#                name: windows_libraries
-#                path: src/deepwave/
-#            - name: Set up Python
-#              uses: actions/setup-python@v3
-#            - name: Install dependencies
-#              run: |
-#                python -m pip install --upgrade pip
-#                python -m pip install pytest scipy
-#                python -m pip install .
-#            - name: Test with pytest
-#              run: |
-#                pytest
+    Test:
+        strategy:
+            matrix:
+                os: [ubuntu-latest, macos-latest, windows-latest]
+            fail-fast: false
+        runs-on: ${{ matrix.os }}
+        needs: [Linux-build, MacOS-build, Windows-build]
+        steps:
+            - name: Checkout
+              uses: actions/checkout@v3
+            - name: Download built Linux libraries
+              uses: actions/download-artifact@v3
+              with:
+                name: linux_libraries
+                path: src/deepwave/
+            - name: Download built MacOS libraries
+              uses: actions/download-artifact@v3
+              with:
+                name: macos_libraries
+                path: src/deepwave/
+            - name: Download built Windows libraries
+              uses: actions/download-artifact@v3
+              with:
+                name: windows_libraries
+                path: src/deepwave/
+            - name: Set up Python
+              uses: actions/setup-python@v3
+            - name: Install dependencies
+              run: |
+                python -m pip install --upgrade pip
+                python -m pip install pytest scipy
+                python -m pip install .
+            - name: Test with pytest
+              run: |
+                pytest
diff --git a/src/deepwave/build_windows.sh b/src/deepwave/build_windows.sh
@@ -2,8 +2,7 @@
 
 set -e
 
-DW_OMP_NAME=libiompstubs5md.dll
-CFLAGS="-Wall -DDW_USE_OPENMP -O2 -fp:fast -arch:AVX2 -openmp"
+CFLAGS="-Wall -O2 -fp:fast -arch:AVX2"
 CUDAFLAGS="--restrict --use_fast_math -O3 -gencode=arch=compute_52,code=sm_52, -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=compute_80"
 cl $CFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=float -c scalar.c -Foscalar_cpu_iso_2_float.obj
 cl $CFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=float -c scalar.c -Foscalar_cpu_iso_4_float.obj
@@ -25,26 +24,26 @@ cl $CFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=float -c elastic.c -Foelastic_cpu_iso_2_fl
 cl $CFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=float -c elastic.c -Foelastic_cpu_iso_4_float.obj
 cl $CFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=double -c elastic.c -Foelastic_cpu_iso_2_double.obj
 cl $CFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=double -c elastic.c -Foelastic_cpu_iso_4_double.obj
-cl $CFLAGS -LD scalar_born_cpu_iso_2_float.obj scalar_born_cpu_iso_4_float.obj scalar_born_cpu_iso_6_float.obj scalar_born_cpu_iso_8_float.obj scalar_born_cpu_iso_2_double.obj scalar_born_cpu_iso_4_double.obj scalar_born_cpu_iso_6_double.obj scalar_born_cpu_iso_8_double.obj scalar_cpu_iso_2_float.obj scalar_cpu_iso_4_float.obj scalar_cpu_iso_6_float.obj scalar_cpu_iso_8_float.obj scalar_cpu_iso_2_double.obj scalar_cpu_iso_4_double.obj scalar_cpu_iso_6_double.obj scalar_cpu_iso_8_double.obj elastic_cpu_iso_2_float.obj elastic_cpu_iso_4_float.obj elastic_cpu_iso_2_double.obj elastic_cpu_iso_4_double.obj -nodefaultlib:vcomp $DW_OMP_NAME -Felibdeepwave_cpu_windows_x86_64.dll
-#nvcc $CUDAFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=float -c scalar.cu -o scalar_cuda_iso_2_float.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=float -c scalar.cu -o scalar_cuda_iso_4_float.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=6 -DDW_DTYPE=float -c scalar.cu -o scalar_cuda_iso_6_float.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=8 -DDW_DTYPE=float -c scalar.cu -o scalar_cuda_iso_8_float.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=double -c scalar.cu -o scalar_cuda_iso_2_double.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=double -c scalar.cu -o scalar_cuda_iso_4_double.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=6 -DDW_DTYPE=double -c scalar.cu -o scalar_cuda_iso_6_double.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=8 -DDW_DTYPE=double -c scalar.cu -o scalar_cuda_iso_8_double.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=float -c scalar_born.cu -o scalar_born_cuda_iso_2_float.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=float -c scalar_born.cu -o scalar_born_cuda_iso_4_float.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=6 -DDW_DTYPE=float -c scalar_born.cu -o scalar_born_cuda_iso_6_float.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=8 -DDW_DTYPE=float -c scalar_born.cu -o scalar_born_cuda_iso_8_float.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=double -c scalar_born.cu -o scalar_born_cuda_iso_2_double.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=double -c scalar_born.cu -o scalar_born_cuda_iso_4_double.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=6 -DDW_DTYPE=double -c scalar_born.cu -o scalar_born_cuda_iso_6_double.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=8 -DDW_DTYPE=double -c scalar_born.cu -o scalar_born_cuda_iso_8_double.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=float -c elastic.cu -o elastic_cuda_iso_2_float.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=float -c elastic.cu -o elastic_cuda_iso_4_float.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=double -c elastic.cu -o elastic_cuda_iso_2_double.obj
-#nvcc $CUDAFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=double -c elastic.cu -o elastic_cuda_iso_4_double.obj
-#nvcc $CUDAFLAGS -shared scalar_born_cuda_iso_2_float.obj scalar_born_cuda_iso_4_float.obj scalar_born_cuda_iso_6_float.obj scalar_born_cuda_iso_8_float.obj scalar_born_cuda_iso_2_double.obj scalar_born_cuda_iso_4_double.obj scalar_born_cuda_iso_6_double.obj scalar_born_cuda_iso_8_double.obj scalar_cuda_iso_2_float.obj scalar_cuda_iso_4_float.obj scalar_cuda_iso_6_float.obj scalar_cuda_iso_8_float.obj scalar_cuda_iso_2_double.obj scalar_cuda_iso_4_double.obj scalar_cuda_iso_6_double.obj scalar_cuda_iso_8_double.obj elastic_cuda_iso_2_float.obj elastic_cuda_iso_4_float.obj elastic_cuda_iso_2_double.obj elastic_cuda_iso_4_double.obj -o libdeepwave_cuda_windows_x86_64.dll
-#rm *.obj
+cl $CFLAGS -LD scalar_born_cpu_iso_2_float.obj scalar_born_cpu_iso_4_float.obj scalar_born_cpu_iso_6_float.obj scalar_born_cpu_iso_8_float.obj scalar_born_cpu_iso_2_double.obj scalar_born_cpu_iso_4_double.obj scalar_born_cpu_iso_6_double.obj scalar_born_cpu_iso_8_double.obj scalar_cpu_iso_2_float.obj scalar_cpu_iso_4_float.obj scalar_cpu_iso_6_float.obj scalar_cpu_iso_8_float.obj scalar_cpu_iso_2_double.obj scalar_cpu_iso_4_double.obj scalar_cpu_iso_6_double.obj scalar_cpu_iso_8_double.obj elastic_cpu_iso_2_float.obj elastic_cpu_iso_4_float.obj elastic_cpu_iso_2_double.obj elastic_cpu_iso_4_double.obj -Felibdeepwave_cpu_windows_x86_64.dll
+nvcc $CUDAFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=float -c scalar.cu -o scalar_cuda_iso_2_float.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=float -c scalar.cu -o scalar_cuda_iso_4_float.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=6 -DDW_DTYPE=float -c scalar.cu -o scalar_cuda_iso_6_float.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=8 -DDW_DTYPE=float -c scalar.cu -o scalar_cuda_iso_8_float.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=double -c scalar.cu -o scalar_cuda_iso_2_double.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=double -c scalar.cu -o scalar_cuda_iso_4_double.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=6 -DDW_DTYPE=double -c scalar.cu -o scalar_cuda_iso_6_double.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=8 -DDW_DTYPE=double -c scalar.cu -o scalar_cuda_iso_8_double.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=float -c scalar_born.cu -o scalar_born_cuda_iso_2_float.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=float -c scalar_born.cu -o scalar_born_cuda_iso_4_float.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=6 -DDW_DTYPE=float -c scalar_born.cu -o scalar_born_cuda_iso_6_float.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=8 -DDW_DTYPE=float -c scalar_born.cu -o scalar_born_cuda_iso_8_float.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=double -c scalar_born.cu -o scalar_born_cuda_iso_2_double.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=double -c scalar_born.cu -o scalar_born_cuda_iso_4_double.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=6 -DDW_DTYPE=double -c scalar_born.cu -o scalar_born_cuda_iso_6_double.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=8 -DDW_DTYPE=double -c scalar_born.cu -o scalar_born_cuda_iso_8_double.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=float -c elastic.cu -o elastic_cuda_iso_2_float.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=float -c elastic.cu -o elastic_cuda_iso_4_float.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=2 -DDW_DTYPE=double -c elastic.cu -o elastic_cuda_iso_2_double.obj
+nvcc $CUDAFLAGS -DDW_ACCURACY=4 -DDW_DTYPE=double -c elastic.cu -o elastic_cuda_iso_4_double.obj
+nvcc $CUDAFLAGS -shared scalar_born_cuda_iso_2_float.obj scalar_born_cuda_iso_4_float.obj scalar_born_cuda_iso_6_float.obj scalar_born_cuda_iso_8_float.obj scalar_born_cuda_iso_2_double.obj scalar_born_cuda_iso_4_double.obj scalar_born_cuda_iso_6_double.obj scalar_born_cuda_iso_8_double.obj scalar_cuda_iso_2_float.obj scalar_cuda_iso_4_float.obj scalar_cuda_iso_6_float.obj scalar_cuda_iso_8_float.obj scalar_cuda_iso_2_double.obj scalar_cuda_iso_4_double.obj scalar_cuda_iso_6_double.obj scalar_cuda_iso_8_double.obj elastic_cuda_iso_2_float.obj elastic_cuda_iso_4_float.obj elastic_cuda_iso_2_double.obj elastic_cuda_iso_4_double.obj -o libdeepwave_cuda_windows_x86_64.dll
+rm *.obj