Add arm64 runner

.github/workflows/build-wheels.yml

@@ -109,6 +109,85 @@ jobs:
           path: wheelhouse/*py3-none-any.whl
           if-no-files-found: error
 
+  build_wheels_linux_arm64:
+    name: Build Linux ARM64 wheels
+    runs-on: ubuntu-22.04-arm
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Set up Python 3.12
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+
+      # Build CUDA wheel using cibuildwheel with manylinux container for ARM64
+      - name: Build CUDA wheel (manylinux ARM64)
+        uses: pypa/cibuildwheel@v2.22.0
+        env:
+          # Use manylinux_2_28 for ARM64
+          CIBW_MANYLINUX_AARCH64_IMAGE: quay.io/pypa/manylinux_2_28_aarch64
+
+          # Build for Python 3.10, 3.11, and 3.12 on ARM64
+          CIBW_BUILD: "cp310-manylinux_aarch64 cp311-manylinux_aarch64 cp312-manylinux_aarch64"
+          CIBW_SKIP: "*-musllinux_*"
+
+          # Native ARM64 runner - no emulation needed
+          CIBW_ARCHS_LINUX: aarch64
+
+          # Install CUDA 13.0 toolkit inside the container (using sbsa repo for ARM64)
+          CIBW_BEFORE_ALL_LINUX: |
+            # Install dnf plugins for config-manager command
+            dnf install -y dnf-plugins-core
+            # Add NVIDIA CUDA repository for RHEL 8 ARM64 (sbsa = server-based system architecture)
+            dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/rhel8/sbsa/cuda-rhel8.repo
+            dnf clean all
+            # Install ONLY the minimal CUDA packages needed for compilation (not the full toolkit)
+            dnf install -y cuda-nvcc-13-0 cuda-cudart-devel-13-0 libcublas-devel-13-0 libnvjitlink-devel-13-0
+            # Install Python development headers (fallback for CMake FindPython)
+            dnf install -y python3-devel
+            # Verify CUDA installation
+            /usr/local/cuda-13.0/bin/nvcc --version
+
+          # Set CUDA environment and install build dependencies
+          CIBW_ENVIRONMENT_LINUX: >
+            CUDA_HOME=/usr/local/cuda-13.0
+            PATH=/usr/local/cuda-13.0/bin:$PATH
+            LD_LIBRARY_PATH=/usr/local/cuda-13.0/lib64:$LD_LIBRARY_PATH
+            LIBRARY_PATH=/usr/local/cuda-13.0/lib64/stubs:$LIBRARY_PATH
+
+          # Install Python build dependencies
+          CIBW_BEFORE_BUILD: "pip install nanobind cmake setuptools wheel"
+
+          # Repair wheel for manylinux compliance, excluding CUDA runtime libs
+          # Users must have CUDA installed on their system
+          CIBW_REPAIR_WHEEL_COMMAND_LINUX: >
+            auditwheel repair -w {dest_dir} {wheel}
+            --exclude libcuda.so.1
+            --exclude libcudart.so.13
+            --exclude libnvrtc.so.13
+            --exclude libcublas.so.13
+            --exclude libcublasLt.so.13
+            --exclude libnvJitLink.so.13
+
+        with:
+          output-dir: wheelhouse
+
+      - name: List built wheels
+        shell: bash
+        run: |
+          ls -lh wheelhouse/
+          echo "Built wheels:"
+          ls -1 wheelhouse/*.whl
+
+      - name: Upload Linux ARM64 CUDA wheel
+        uses: actions/upload-artifact@v4
+        with:
+          name: wheels-linux-arm64-cuda
+          path: wheelhouse/*manylinux*.whl
+          if-no-files-found: error
+
   build_wheels_windows:
     name: Build Windows wheel (Python ${{ matrix.python-version }})
     runs-on: windows-2022
@@ -372,7 +451,7 @@ jobs:
 
   publish:
     name: Publish to PyPI
-    needs: [build_wheels_linux, build_wheels_windows, test_linux, test_windows]
+    needs: [build_wheels_linux, build_wheels_linux_arm64, build_wheels_windows, test_linux, test_windows]
     runs-on: ubuntu-latest
     if: github.event_name == 'push' && startsWith(github.ref, 'refs/tags/v')
     environment: pypi
@@ -384,6 +463,12 @@ jobs:
           name: wheels-linux-cuda
           path: dist/
 
+      - name: Download Linux ARM64 CUDA wheel
+        uses: actions/download-artifact@v4
+        with:
+          name: wheels-linux-arm64-cuda
+          path: dist/
+
       - name: Download Windows CUDA wheel (3.10)
         uses: actions/download-artifact@v4
         with: