[build] Use static linking for CUDA build and allow dynamic AVX dispatch

Author: Breakthrough

.github/workflows/build-windows-cuda.yml

@@ -24,21 +24,27 @@ jobs:
       - name: 📥 Checkout code
         uses: actions/checkout@v4
 
-      - name: 🏗️ Install Build Dependencies
-        run: |
-          echo "Installing CMake and Ninja for building..."
-          choco install -y cmake ninja
+      #- name: 🏗️ Install Build Dependencies
+      #  run: |
+      #    echo "Installing CMake and Ninja for building..."
+      #    choco install -y cmake ninja
 
       - name: 🐍 Set up Python
         uses: actions/setup-python@v5
         with:
-          python-version: '3.13'
+          python-version: '3.12'
 
       - name: 📦 Install Python Build Requirements
         run: |
           python -m pip install --upgrade pip
           pip install numpy wheel
 
+      - name: Setup MSBuild.exe
+        uses: microsoft/[email protected]
+
+      - name: Setup NASM
+        uses: ilammy/setup-nasm@v1
+
       - name: 🔧 Install NVIDIA CUDA Toolkit
         run: |
           echo "Downloading CUDA Toolkit..."
@@ -63,17 +69,21 @@ jobs:
       - name: Clone opencv-python repository
         run: |
           git clone --depth 1 --recursive https://github.com/opencv/opencv-python.git
-          pip install numpy packaging scikit-build
+          pip install numpy packaging scikit-build cmake==3.24.2
 
         # - CUDA_ARCH_BIN: Specifies the CUDA architectures to build for.
         # - CUDA_ARCH_PTX: Specifies the virtual architecture for forward compatibility.
         # TODO: Set correct PTX versions below.
         # TODO: Validate the built wheel can be installed and run.
+        # TODO: Slim down the modules we build to only the ones that DVR-Scan needs to run:
+        # https://docs.opencv.org/4.x/db/d05/tutorial_config_reference.html#tutorial_config_reference_general_modules
       - name: 🛠️ Build opencv-python with CUDA
         run: |
           cd opencv-python
-          $CMAKE_ARGS = '-D CMAKE_BUILD_TYPE=RELEASE -D WITH_CUDA=ON -D CUDA_ARCH_BIN="6.0;6.1;7.0;7.5" -D CUDA_ARCH_PTX=7.5 -D OPENCV_ENABLE_NONFREE=ON -D BUILD_TESTS=OFF -D BUILD_PERF_TESTS=OFF -D BUILD_EXAMPLES=OFF'
+          $CMAKE_ARGS = '-D CMAKE_BUILD_TYPE=RELEASE -D BUILD_SHARED_LIBS=OFF -D WITH_CUDA=ON -D CUDA_ARCH_BIN="6.0;6.1;7.0;7.5" -D CUDA_ARCH_PTX=7.5 -D OPENCV_ENABLE_NONFREE=ON -D ENABLE_LTO=ON -D CPU_DISPATCH="AVX,AVX2" -D BUILD_TESTS=OFF -D BUILD_PERF_TESTS=OFF -D BUILD_EXAMPLES=OFF -D OPENCV_EXTRA_MODULES_PATH=opencv_contrib/modules'
+          $ENABLE_CONTRIB = 1
           pip wheel . --verbose
+        shell: pwsh
 
       - name: 📤 Upload Artifact
         uses: actions/[email protected]