[WIP]Vendor in _dispatcher, _devicearray, mviewbuf C extension for CUDA-specific customization

.github/workflows/conda-python-build.yaml

@@ -104,7 +104,7 @@ jobs:
         if: "!cancelled()"
         uses: actions/upload-artifact@v4
         with:
-          name: conda-repo
+          name: conda-repo-py${{ matrix.PY_VER }}-${{ matrix.ARCH }}
           path: "/tmp/conda-bld-output"
       - name: Publish conda package
         if: inputs.upload_to_anaconda

.github/workflows/conda-python-tests.yaml

@@ -97,7 +97,9 @@ jobs:
           ref: ${{ inputs.sha }}
           fetch-depth: 0
       - uses: actions/download-artifact@v4
-        name: conda-repo
+        with:
+          name: conda-repo-py${{ matrix.PY_VER }}-${{ matrix.ARCH }}
+          path: conda-repo
       - name: Display structure of downloaded files
         run: ls -R
       - name: Standardize repository information

.github/workflows/docs-build.yaml

@@ -49,7 +49,7 @@ jobs:
     env:
       RAPIDS_ARTIFACTS_DIR: ${{ github.workspace }}/artifacts
     container:
-      image: rapidsai/ci-conda:latest
+      image: rapidsai/ci-conda:cuda13.0.0-ubuntu24.04-py3.13
       env:
         RAPIDS_BUILD_TYPE: ${{ inputs.build_type }}
     steps:
@@ -59,7 +59,9 @@ jobs:
           ref: ${{ inputs.sha }}
           fetch-depth: 0
       - uses: actions/download-artifact@v4
-        name: conda-repo
+        with:
+          name: conda-repo-py3.13-amd64
+          path: conda-repo
       - name: Display structure of downloaded files
         run: ls -R
       - name: Standardize repository information

.github/workflows/pr.yaml

@@ -22,6 +22,7 @@ jobs:
       - test-conda-ctypes-binding
       - test-simulator
       - build-wheels
+      - build-wheels-windows
       - test-wheels-windows
       - test-wheels
       - test-wheels-ctypes-binding
@@ -108,10 +109,11 @@ jobs:
       build_type: pull-request
       script: "ci/build_wheel.sh"
       matrix: ${{ needs.compute-matrix.outputs.BUILD_MATRIX }}
+  build-wheels-windows:
+    uses: ./.github/workflows/wheel-windows-build.yaml
   test-wheels-windows:
     needs:
-      - build-wheels
-      - compute-matrix
+      - build-wheels-windows
     uses: ./.github/workflows/wheel-windows-tests.yaml
   test-wheels:
     needs:
@@ -169,4 +171,4 @@ jobs:
       build_type: pull-request
       script: "ci/coverage_report.sh"
       matrix: ${{ needs.compute-matrix.outputs.TEST_MATRIX }}
-      matrix_filter: 'map(select(.ARCH == "amd64" and .CUDA_VER == "12.9.1" and .PY_VER == "3.11")) | .[0:1]'
+      matrix_filter: 'map(select(.ARCH == "amd64" and .CUDA_VER == "12.9.1" and .PY_VER == "3.12")) | .[0:1]'

.github/workflows/simulator-test.yaml

@@ -86,7 +86,9 @@ jobs:
           ref: ${{ inputs.sha }}
           fetch-depth: 0
       - uses: actions/download-artifact@v4
-        name: conda-repo
+        with:
+          name: conda-repo-py${{ matrix.PY_VER }}-${{ matrix.ARCH }}
+          path: conda-repo
       - name: Display structure of downloaded files
         run: ls -R
       - name: Standardize repository information

.github/workflows/wheel-windows-build.yaml

@@ -0,0 +1,55 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: BSD-2-Clause
+
+name: "CI: Build wheel on Windows"
+
+on:
+  workflow_call:
+    inputs:
+      sha:
+        type: string
+      repo:
+        type: string
+      script:
+        type: string
+        default: "./ci/build_wheel.ps1"
+
+jobs:
+  build:
+    name: ${{ matrix.PY_VER }}, windows
+    strategy:
+      fail-fast: false
+      matrix:
+        PY_VER:
+          - "3.10"
+          - "3.11"
+          - "3.12"
+          - "3.13"
+    runs-on: windows-2022
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          repository: ${{ inputs.repo }}
+          ref: ${{ inputs.sha }}
+          fetch-depth: 0
+
+      - name: Set up Python ${{ matrix.PY_VER }}
+        uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065  # v5.6.0
+        with:
+          python-version: ${{ matrix.PY_VER }}
+
+      - name: Set up MSVC
+        uses: ilammy/msvc-dev-cmd@v1  # TODO: ask admin to allow pinning commits
+
+      - name: Python build
+        run: |
+          ${{ inputs.script }}
+        env:
+          CUDA_VER: "12.8.0"
+
+      - name: Upload wheel
+        if: "!cancelled()"
+        uses: actions/upload-artifact@v4
+        with:
+          name: wheel-py${{ matrix.PY_VER }}-windows
+          path: ${{ env.wheel_path }}

.github/workflows/wheel-windows-tests.yaml

@@ -15,24 +15,16 @@ on:
         default: "./ci/test_wheel.ps1"
 
 jobs:
-  compute-matrix:
-    runs-on: ubuntu-latest
-    outputs:
-      MATRIX: ${{ steps.compute-matrix.outputs.MATRIX }}
-    steps:
-      - name: Compute Python Test Matrix
-        id: compute-matrix
-        run: |
-          set -eo pipefail
-          export TEST_MATRIX="{ ARCH: 'amd64', PY_VER: '3.12', CUDA_VER: '12.8.0' }"
-          MATRIX=$(yq -n -o json 'env(TEST_MATRIX)' | jq -c)
-          echo "MATRIX=${MATRIX}" | tee --append "${GITHUB_OUTPUT}"
   tests:
-    name: ${{ matrix.CUDA_VER }}, ${{ matrix.PY_VER }}, ${{ matrix.ARCH }}, windows
-    needs: compute-matrix
+    name: ${{ matrix.PY_VER }}, windows
     strategy:
       fail-fast: false
-      matrix: ${{ fromJSON('{"ARCH":["amd64"],"PY_VER":["3.12"],"CUDA_VER":["12.8.0"]}') }}
+      matrix:
+        PY_VER:
+          - "3.10"
+          - "3.11"
+          - "3.12"
+          - "3.13"
     runs-on: "cuda-python-windows-gpu-github"
     steps:
       - uses: actions/checkout@v4
@@ -49,7 +41,9 @@ jobs:
         run: nvidia-smi
 
       - uses: actions/download-artifact@v4
-        name: wheel
+        with:
+          name: wheel-py${{ matrix.PY_VER }}-windows
+          path: wheel
 
       - name: Display structure of downloaded files
         run: Get-ChildItem -Recurse
@@ -63,4 +57,4 @@ jobs:
         run: |
           ${{ inputs.script }}
         env:
-          CUDA_VER: ${{ matrix.CUDA_VER }}
+          CUDA_VER: "12.8.0"

.github/workflows/wheels-build.yaml

@@ -152,11 +152,11 @@ jobs:
         if: "!cancelled()"
         uses: actions/upload-artifact@v4
         with:
-          name: sdist
+          name: sdist-py${{ matrix.PY_VER }}-${{ matrix.ARCH }}
           path: ${{ env.sdist_path }}
       - name: Upload wheel
         if: "!cancelled()"
         uses: actions/upload-artifact@v4
         with:
-          name: wheel
+          name: wheel-py${{ matrix.PY_VER }}-${{ matrix.ARCH }}
           path: ${{ env.wheel_path }}

.github/workflows/wheels-test.yaml

@@ -115,7 +115,9 @@ jobs:
         fetch-depth: 0 # unshallow fetch for setuptools-scm
         persist-credentials: false
     - uses: actions/download-artifact@v4
-      name: wheel
+      with:
+        name: wheel-py${{ matrix.PY_VER }}-${{ matrix.ARCH }}
+        path: wheel
     - name: Display structure of downloaded files
       run: ls -R
     - name: Standardize repository information

ci/build_wheel.ps1

@@ -0,0 +1,33 @@
+# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: BSD-2-Clause
+
+$ErrorActionPreference = 'Stop'
+Set-StrictMode -Version Latest
+
+function rapids-logger {
+    param (
+        [Parameter(Mandatory=$true)]
+        [string]$Text
+    )
+
+    # Determine padding and box width
+    $padding = 2
+    $boxWidth = $Text.Length + ($padding * 2)
+    $topBottom = '+' + ('-' * $boxWidth) + '+'
+    $middle = '|' + (' ' * $padding) + $Text + (' ' * $padding) + '|'
+
+    # Print the box in green
+    Write-Host $topBottom -ForegroundColor Green
+    Write-Host $middle    -ForegroundColor Green
+    Write-Host $topBottom -ForegroundColor Green
+}
+
+rapids-logger "Install build package"
+python -m pip install build
+
+rapids-logger "Build sdist and wheel"
+python -m build .
+
+$wheel_path = Resolve-Path dist\numba_cuda*.whl | Select-Object -ExpandProperty Path
+echo "Wheel path: $wheel_path"
+echo "wheel_path=$wheel_path" >> $env:GITHUB_ENV

ci/matrix.yml

@@ -5,23 +5,30 @@
 #     [ARCH, PY_VER, CUDA_VER, LINUX_VER, GPU, DRIVER, DEPENDENCIES]
 #
 build-matrix:
+  - { ARCH: 'amd64', PY_VER: '3.10', CUDA_VER: '12.8.0', LINUX_VER: 'rockylinux8' }
+  - { ARCH: 'amd64', PY_VER: '3.11', CUDA_VER: '12.8.0', LINUX_VER: 'rockylinux8' }
+  - { ARCH: 'amd64', PY_VER: '3.12', CUDA_VER: '12.8.0', LINUX_VER: 'rockylinux8' }
   - { ARCH: 'amd64', PY_VER: '3.13', CUDA_VER: '12.8.0', LINUX_VER: 'rockylinux8' }
+  - { ARCH: 'arm64', PY_VER: '3.10', CUDA_VER: '12.8.0', LINUX_VER: 'rockylinux8' }
+  - { ARCH: 'arm64', PY_VER: '3.11', CUDA_VER: '12.8.0', LINUX_VER: 'rockylinux8' }
+  - { ARCH: 'arm64', PY_VER: '3.12', CUDA_VER: '12.8.0', LINUX_VER: 'rockylinux8' }
+  - { ARCH: 'arm64', PY_VER: '3.13', CUDA_VER: '12.8.0', LINUX_VER: 'rockylinux8' }
 simulator-matrix:
   - { ARCH: 'amd64', PY_VER: '3.12', CUDA_VER: '12.8.0', LINUX_VER: 'rockylinux8' }
 # We test "oldest" dependencies with the oldest supported Python version and
 # the second-newest Python version.
 test-matrix:
-  - { CUDA_VER: '12.0.1', ARCH: 'amd64', PY_VER: '3.9',  LINUX_VER: 'rockylinux8', GPU: 'l4',   DRIVER: 'earliest', DEPENDENCIES: 'oldest' }
-  - { CUDA_VER: '12.0.1', ARCH: 'amd64', PY_VER: '3.9',  LINUX_VER: 'rockylinux8', GPU: 'l4',   DRIVER: 'earliest', DEPENDENCIES: 'latest' }
-  - { CUDA_VER: '12.2.2', ARCH: 'amd64', PY_VER: '3.10', LINUX_VER: 'ubuntu22.04', GPU: 'l4',   DRIVER: 'latest',   DEPENDENCIES: 'latest' }
-  - { CUDA_VER: '12.9.1', ARCH: 'amd64', PY_VER: '3.11', LINUX_VER: 'ubuntu24.04', GPU: 'l4',   DRIVER: 'latest',   DEPENDENCIES: 'latest' }
+  - { CUDA_VER: '12.0.1', ARCH: 'amd64', PY_VER: '3.10',  LINUX_VER: 'rockylinux8', GPU: 'l4',   DRIVER: 'earliest', DEPENDENCIES: 'oldest' }
+  - { CUDA_VER: '12.0.1', ARCH: 'amd64', PY_VER: '3.11',  LINUX_VER: 'rockylinux8', GPU: 'l4',   DRIVER: 'earliest', DEPENDENCIES: 'latest' }
+  - { CUDA_VER: '12.2.2', ARCH: 'amd64', PY_VER: '3.11', LINUX_VER: 'ubuntu22.04', GPU: 'l4',   DRIVER: 'latest',   DEPENDENCIES: 'latest' }
+  - { CUDA_VER: '12.9.1', ARCH: 'amd64', PY_VER: '3.12', LINUX_VER: 'ubuntu24.04', GPU: 'l4',   DRIVER: 'latest',   DEPENDENCIES: 'latest' }
   - { CUDA_VER: '13.0.0', ARCH: 'amd64', PY_VER: '3.12', LINUX_VER: 'ubuntu22.04', GPU: 'l4',   DRIVER: 'latest',   DEPENDENCIES: 'oldest' }
-  - { CUDA_VER: '13.0.0', ARCH: 'amd64', PY_VER: '3.12', LINUX_VER: 'ubuntu22.04', GPU: 'l4',   DRIVER: 'latest',   DEPENDENCIES: 'latest' }
+  - { CUDA_VER: '13.0.0', ARCH: 'amd64', PY_VER: '3.13', LINUX_VER: 'ubuntu22.04', GPU: 'l4',   DRIVER: 'latest',   DEPENDENCIES: 'latest' }
   - { CUDA_VER: '13.0.0', ARCH: 'amd64', PY_VER: '3.13', LINUX_VER: 'ubuntu24.04', GPU: 'l4',   DRIVER: 'latest',   DEPENDENCIES: 'latest' }
-  - { CUDA_VER: '12.0.1', ARCH: 'arm64', PY_VER: '3.9',  LINUX_VER: 'rockylinux8', GPU: 'a100', DRIVER: 'earliest', DEPENDENCIES: 'oldest' }
-  - { CUDA_VER: '12.0.1', ARCH: 'arm64', PY_VER: '3.9',  LINUX_VER: 'rockylinux8', GPU: 'a100', DRIVER: 'earliest', DEPENDENCIES: 'latest' }
-  - { CUDA_VER: '12.2.2', ARCH: 'arm64', PY_VER: '3.10', LINUX_VER: 'ubuntu22.04', GPU: 'a100', DRIVER: 'latest',   DEPENDENCIES: 'latest' }
-  - { CUDA_VER: '12.9.1', ARCH: 'arm64', PY_VER: '3.11', LINUX_VER: 'ubuntu24.04', GPU: 'a100', DRIVER: 'latest',   DEPENDENCIES: 'latest' }
+  - { CUDA_VER: '12.0.1', ARCH: 'arm64', PY_VER: '3.10',  LINUX_VER: 'rockylinux8', GPU: 'a100', DRIVER: 'earliest', DEPENDENCIES: 'oldest' }
+  - { CUDA_VER: '12.0.1', ARCH: 'arm64', PY_VER: '3.11',  LINUX_VER: 'rockylinux8', GPU: 'a100', DRIVER: 'earliest', DEPENDENCIES: 'latest' }
+  - { CUDA_VER: '12.2.2', ARCH: 'arm64', PY_VER: '3.11', LINUX_VER: 'ubuntu22.04', GPU: 'a100', DRIVER: 'latest',   DEPENDENCIES: 'latest' }
+  - { CUDA_VER: '12.9.1', ARCH: 'arm64', PY_VER: '3.12', LINUX_VER: 'ubuntu24.04', GPU: 'a100', DRIVER: 'latest',   DEPENDENCIES: 'latest' }
   - { CUDA_VER: '13.0.0', ARCH: 'arm64', PY_VER: '3.12', LINUX_VER: 'ubuntu22.04', GPU: 'a100', DRIVER: 'latest',   DEPENDENCIES: 'oldest' }
-  - { CUDA_VER: '13.0.0', ARCH: 'arm64', PY_VER: '3.12', LINUX_VER: 'ubuntu22.04', GPU: 'a100', DRIVER: 'latest',   DEPENDENCIES: 'latest' }
+  - { CUDA_VER: '13.0.0', ARCH: 'arm64', PY_VER: '3.13', LINUX_VER: 'ubuntu22.04', GPU: 'a100', DRIVER: 'latest',   DEPENDENCIES: 'latest' }
   - { CUDA_VER: '13.0.0', ARCH: 'arm64', PY_VER: '3.13', LINUX_VER: 'ubuntu24.04', GPU: 'a100', DRIVER: 'latest',   DEPENDENCIES: 'latest' }

conda/recipes/numba-cuda/meta.yaml

@@ -17,7 +17,6 @@ source:
   path: ../../..
 
 build:
-  noarch: python
   script:
     - {{ PYTHON }} -m pip install . -vv
 
@@ -26,6 +25,7 @@ requirements:
     - python
     - pip
     - setuptools
+    - numpy >=2.1.0
   run:
     - python
     - numba >=0.59.1

numba_cuda/numba/cuda/cext/__init__.py

@@ -0,0 +1,2 @@
+# SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: BSD-2-Clause