backport ft fix

.github/workflows/build-wheel.yml

@@ -30,6 +30,9 @@ jobs:
           - "3.11"
           - "3.12"
           - "3.13"
+          - "3.13t"
+          - "3.14"
+          - "3.14t"
     name: py${{ matrix.python-version }}
     runs-on: ${{ (inputs.host-platform == 'linux-64' && 'linux-amd64-cpu8') ||
                  (inputs.host-platform == 'linux-aarch64' && 'linux-arm64-cpu8') ||
@@ -116,6 +119,7 @@ jobs:
           CIBW_REPAIR_WHEEL_COMMAND_WINDOWS: "delvewheel repair --namespace-pkg cuda -w {dest_dir} {wheel}"
           CIBW_ENVIRONMENT: >
             CUDA_PYTHON_PARALLEL_LEVEL=${{ env.CUDA_PYTHON_PARALLEL_LEVEL }}
+          CIBW_ENABLE: "cpython-freethreading"
         with:
           package-dir: ./cuda_core/
           output-dir: ${{ env.CUDA_CORE_ARTIFACTS_DIR }}
@@ -157,14 +161,13 @@ jobs:
           # CIBW mounts the host filesystem under /host
           CIBW_ENVIRONMENT_LINUX: >
             CUDA_PATH=/host/${{ env.CUDA_PATH }}
-            LIBRARY_PATH=/host/${{ env.CUDA_PATH }}/lib
             CUDA_PYTHON_PARALLEL_LEVEL=${{ env.CUDA_PYTHON_PARALLEL_LEVEL }}
           CIBW_ENVIRONMENT_WINDOWS: >
             CUDA_HOME="$(cygpath -w ${{ env.CUDA_PATH }})"
-            LIB="${CUDA_HOME}\\lib\\x64;${LIB}"
             CUDA_PYTHON_PARALLEL_LEVEL=${{ env.CUDA_PYTHON_PARALLEL_LEVEL }}
           CIBW_BEFORE_BUILD_WINDOWS: "pip install delvewheel"
           CIBW_REPAIR_WHEEL_COMMAND_WINDOWS: "delvewheel repair --namespace-pkg cuda -w {dest_dir} {wheel}"
+          CIBW_ENABLE: "cpython-freethreading"
         with:
           package-dir: ./cuda_bindings/
           output-dir: ${{ env.CUDA_BINDINGS_ARTIFACTS_DIR }}
@@ -224,6 +227,12 @@ jobs:
         with:
           # workaround for actions/runner-images#12377 (the cached 3.13.4 is buggy on Windows)
           python-version: ${{ matrix.python-version == '3.13' && '3.13.5' || matrix.python-version }}
+          # allow prereleases only fetches prereleases for unreleased versions of Python
+          allow-prereleases: true
+
+      - name: verify free-threaded build
+        if: endsWith(matrix.python-verison, 't')
+        run: python -c 'import sys; assert not sys._is_gil_enabled()'
 
       - name: Set up Python include paths
         run: |

.github/workflows/ci.yml

@@ -14,7 +14,7 @@ on:
   push:
     branches:
       - "pull-request/[0-9]+"
-      - "main"
+      - "12.9.x"
 
 jobs:
   ci-vars:
@@ -23,7 +23,7 @@ jobs:
       CUDA_BUILD_VER: ${{ steps.get-vars.outputs.cuda_build_ver }}
     steps:
       - name: Checkout repository
-        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683  # v4.2.2
+        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8  # v5.0.0
         with:
           fetch-depth: 0
       - name: Get CUDA build version
@@ -32,89 +32,153 @@ jobs:
           cuda_build_ver=$(jq -r .cuda.build.version ci/versions.json)
           echo "cuda_build_ver=$cuda_build_ver" >> $GITHUB_OUTPUT
 
-  build:
+  should-skip:
+    runs-on: ubuntu-latest
+    outputs:
+      skip: ${{ steps.get-should-skip.outputs.skip }}
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@08c6903cd8c0fde910a37f88322edcfb5dd907a8  # v5.0.0
+      - name: Compute whether to skip builds and tests
+        id: get-should-skip
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          set -euxo pipefail
+          if ${{ startsWith(github.ref_name, 'pull-request/') }}; then
+            skip="$(gh pr view "$(grep -Po '(\d+)$' <<< '${{ github.ref_name }}')" --json title --jq '.title | contains("[no-ci]")')"
+          else
+            skip=false
+          fi
+          echo "skip=${skip}" >> "$GITHUB_OUTPUT"
+
+  # WARNING: make sure all of the build jobs are in sync
+  build-linux-64:
     needs:
       - ci-vars
+      - should-skip
     strategy:
       fail-fast: false
       matrix:
         host-platform:
           - linux-64
+    name: Build ${{ matrix.host-platform }}, CUDA ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
+    if: ${{ github.repository_owner == 'nvidia' && !fromJSON(needs.should-skip.outputs.skip) }}
+    secrets: inherit
+    uses: ./.github/workflows/build-wheel.yml
+    with:
+      host-platform: ${{ matrix.host-platform }}
+      cuda-version: ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
+
+  # WARNING: make sure all of the build jobs are in sync
+  build-linux-aarch64:
+    needs:
+      - ci-vars
+      - should-skip
+    strategy:
+      fail-fast: false
+      matrix:
+        host-platform:
           - linux-aarch64
+    name: Build ${{ matrix.host-platform }}, CUDA ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
+    if: ${{ github.repository_owner == 'nvidia' && !fromJSON(needs.should-skip.outputs.skip) }}
+    secrets: inherit
+    uses: ./.github/workflows/build-wheel.yml
+    with:
+      host-platform: ${{ matrix.host-platform }}
+      cuda-version: ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
+
+  # WARNING: make sure all of the build jobs are in sync
+  build-windows:
+    needs:
+      - ci-vars
+      - should-skip
+    strategy:
+      fail-fast: false
+      matrix:
+        host-platform:
           - win-64
     name: Build ${{ matrix.host-platform }}, CUDA ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
-    if: ${{ github.repository_owner == 'nvidia' }}
+    if: ${{ github.repository_owner == 'nvidia' && !fromJSON(needs.should-skip.outputs.skip) }}
     secrets: inherit
     uses: ./.github/workflows/build-wheel.yml
     with:
       host-platform: ${{ matrix.host-platform }}
       cuda-version: ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
 
-  test-linux:
+  # WARNING: make sure both Linux test jobs are in sync
+  test-linux-64:
     strategy:
       fail-fast: false
       matrix:
         host-platform:
           - linux-64
-          - linux-aarch64
     name: Test ${{ matrix.host-platform }}
     if: ${{ github.repository_owner == 'nvidia' }}
     permissions:
       contents: read  # This is required for actions/checkout
     needs:
       - ci-vars
-      - build
+      - build-linux-64
     secrets: inherit
     uses: ./.github/workflows/test-wheel-linux.yml
     with:
       build-type: pull-request
       host-platform: ${{ matrix.host-platform }}
       build-ctk-ver: ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
+      matrix_filter: "map(select([.CUDA_VER // empty | split(\".\")[] | tonumber] as $v | ($v[0] < 13)))"
 
-  test-windows:
+  # WARNING: make sure both Linux test jobs are in sync
+  test-linux-aarch64:
     strategy:
       fail-fast: false
       matrix:
         host-platform:
-          - win-64
+          - linux-aarch64
     name: Test ${{ matrix.host-platform }}
     if: ${{ github.repository_owner == 'nvidia' }}
     permissions:
       contents: read  # This is required for actions/checkout
     needs:
       - ci-vars
-      - build
+      - build-linux-aarch64
     secrets: inherit
-    uses: ./.github/workflows/test-wheel-windows.yml
+    uses: ./.github/workflows/test-wheel-linux.yml
     with:
       build-type: pull-request
       host-platform: ${{ matrix.host-platform }}
       build-ctk-ver: ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
+      matrix_filter: "map(select([.CUDA_VER // empty | split(\".\")[] | tonumber] as $v | ($v[0] < 13)))"
 
-  doc:
-    name: Docs
+  test-windows:
+    strategy:
+      fail-fast: false
+      matrix:
+        host-platform:
+          - win-64
+    name: Test ${{ matrix.host-platform }}
     if: ${{ github.repository_owner == 'nvidia' }}
-    # Sets permissions of the GITHUB_TOKEN to allow deployment to GitHub Pages
     permissions:
-      id-token: write
-      contents: write
-      pull-requests: write
+      contents: read  # This is required for actions/checkout
     needs:
       - ci-vars
-      - build
+      - build-windows
     secrets: inherit
-    uses: ./.github/workflows/build-docs.yml
+    uses: ./.github/workflows/test-wheel-windows.yml
     with:
+      build-type: pull-request
+      host-platform: ${{ matrix.host-platform }}
       build-ctk-ver: ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
+      matrix_filter: "map(select([.CUDA_VER // empty | split(\".\")[] | tonumber] as $v | ($v[0] < 13)))"
 
   checks:
     name: Check job status
-    permissions:
-      checks: read
+    runs-on: ubuntu-latest
     needs:
-      - build
-      - test-linux
+      - test-linux-64
+      - test-linux-aarch64
       - test-windows
-      - doc
-    secrets: inherit
-    uses: ./.github/workflows/status-check.yml
+      # we don't build docs on the backport branch, see #809
+    steps:
+      - name: Exit
+        run: exit 0

.github/workflows/guess_latest.sh

@@ -6,18 +6,27 @@
 # URL to search
 URL="https://developer.download.nvidia.com/compute/cuda/redist/"
 
+# Ensure exactly one argument is provided
+if [ "$#" -ne 1 ]; then
+    echo "Usage: $0 <CUDA_major_version>"
+    exit 1
+fi
+
+# Accept major version as the first argument
+MAJOR_VERSION="$1"
+
 # Fetch the directory listing and extract the latest version number
 get_latest_version() {
     # Get the HTML content of the page
     local html_content=$(wget -q -O - "$URL")
 
     # Extract links matching the pattern redistrib_?.?.?.json
-    local files=$(echo "$html_content" | grep -oP 'redistrib_[0-9]+\.[0-9]+\.[0-9]+\.json' | cut -d'"' -f2)
+    local files=$(echo "$html_content" | grep -oP "redistrib_${MAJOR_VERSION}\.[0-9]+\.[0-9]+\.json" | cut -d'"' -f2)
 
     # If files were found, extract the version numbers and find the latest
     if [ -n "$files" ]; then
         # Extract just the version numbers using regex
-        local versions=$(echo "$files" | grep -oP 'redistrib_\K[0-9]+\.[0-9]+\.[0-9]+(?=\.json)')
+        local versions=$(echo "$files" | grep -oP "redistrib_\K${MAJOR_VERSION}\.[0-9]+\.[0-9]+(?=\.json)")
 
         # Sort the versions and get the latest
         local latest_version=$(echo "$versions" | sort -V | tail -n 1)