From dfe989cabfe42ffa73dac907e9ccda9ac4a09df7 Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Thu, 10 Apr 2025 16:24:53 -0400
Subject: [PATCH 01/19] Disable existing CI for PRs

---
 .github/workflows/build-and-test.yml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/.github/workflows/build-and-test.yml b/.github/workflows/build-and-test.yml
index 812a14117..f996ba79e 100644
--- a/.github/workflows/build-and-test.yml
+++ b/.github/workflows/build-and-test.yml
@@ -15,7 +15,7 @@ concurrency:
 on:
   push:
     branches:
-      - "pull-request/[0-9]+"
+      #- "pull-request/[0-9]+"
       - "main"
 
 jobs:

From 74c7048bb9ae179a1b429c18401e1f648d187517 Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Thu, 10 Apr 2025 16:25:26 -0400
Subject: [PATCH 02/19] Get the build CI working with python caching

---
 .github/workflows/build.yml | 185 ++++++++++++++++++++++++++++++++++++
 .github/workflows/ci.yml    |  32 +++++++
 2 files changed, 217 insertions(+)
 create mode 100644 .github/workflows/build.yml
 create mode 100644 .github/workflows/ci.yml

diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
new file mode 100644
index 000000000..d756220fe
--- /dev/null
+++ b/.github/workflows/build.yml
@@ -0,0 +1,185 @@
+on:
+  workflow_call:
+    inputs:
+      host-platform:
+        required: true
+        type: string
+      cuda-version:
+        type: string
+        default: "12.8.0"
+
+defaults:
+  run:
+    shell: bash --noprofile --norc -xeuo pipefail {0}
+
+permissions:
+  contents: read  # This is required for actions/checkout
+
+jobs:
+  build:
+    strategy:
+      fail-fast: false
+      matrix:
+        python-version:
+          - "3.9"
+          - "3.13"
+    name: py${{ matrix.python-version }}, ${{ inputs.cuda-version }}
+    runs-on: ${{ (inputs.host-platform == 'linux-64' && 'linux-amd64-cpu8') ||
+                 (inputs.host-platform == 'linux-aarch64' && 'linux-arm64-cpu8') ||
+                 (inputs.host-platform == 'win-64' && 'windows-2019') }}
+    steps:
+      - name: Checkout ${{ github.event.repository.name }}
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+
+      - name: Setup proxy cache
+        uses: nv-gha-runners/setup-proxy-cache@main
+        continue-on-error: true
+        # Skip the cache on Windows nodes outside of our org.
+        if: ${{ inputs.host-platform != 'win-64' }}
+
+      - name: Set up Python
+        if: ${{ startsWith(inputs.host-platform, 'linux') }}
+        id: setup-python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.12"
+
+      - name: Set up MSVC
+        if: ${{ startsWith(inputs.host-platform, 'win') }}
+        uses: ilammy/msvc-dev-cmd@v1
+      
+      - name: Set environment variables
+        run: |
+          PYTHON_VERSION_FORMATTED=$(echo '${{ matrix.python-version }}' | tr -d '.')
+          if [[ "${{ inputs.host-platform }}" == linux* ]]; then
+            CIBW_BUILD="cp${PYTHON_VERSION_FORMATTED}-manylinux*"
+            REPO_DIR=$(pwd)
+          elif [[ "${{ inputs.host-platform }}" == win* ]]; then
+            CIBW_BUILD="cp${PYTHON_VERSION_FORMATTED}-win_amd64"
+            PWD=$(pwd)
+            REPO_DIR=$(cygpath -w $PWD)
+          fi
+  
+          echo "CUDA_BINDINGS_PARALLEL_LEVEL=$(nproc)" >> $GITHUB_ENV
+          CUDA_CORE_ARTIFACT_BASENAME="cuda-core-python${PYTHON_VERSION_FORMATTED}-${{ inputs.host-platform }}"
+          echo "CUDA_CORE_ARTIFACT_BASENAME=${CUDA_CORE_ARTIFACT_BASENAME}" >> $GITHUB_ENV
+          echo "CUDA_CORE_ARTIFACT_NAME=${CUDA_CORE_ARTIFACT_BASENAME}-${{ github.sha }}" >> $GITHUB_ENV
+          echo "CUDA_CORE_ARTIFACTS_DIR=$(realpath "$REPO_DIR/cuda_core/dist")" >> $GITHUB_ENV
+          CUDA_BINDINGS_ARTIFACT_BASENAME="cuda-bindings-python${PYTHON_VERSION_FORMATTED}-cuda${{ inputs.cuda-version }}-${{ inputs.host-platform }}"
+          echo "CUDA_BINDINGS_ARTIFACT_BASENAME=${CUDA_BINDINGS_ARTIFACT_BASENAME}" >> $GITHUB_ENV
+          echo "CUDA_BINDINGS_ARTIFACT_NAME=${CUDA_BINDINGS_ARTIFACT_BASENAME}-${{ github.sha }}" >> $GITHUB_ENV
+          echo "CUDA_BINDINGS_ARTIFACTS_DIR=$(realpath "$REPO_DIR/cuda_bindings/dist")" >> $GITHUB_ENV
+          echo "CIBW_BUILD=${CIBW_BUILD}" >> $GITHUB_ENV
+      
+      - name: Dump environment
+        run: |
+          env
+      
+      - name: Build cuda.core wheel
+        uses: pypa/cibuildwheel@v2.22.0
+        env:
+          CIBW_BUILD: ${{ env.CIBW_BUILD }}
+          CIBW_ARCHS_LINUX: "native"
+          CIBW_BUILD_VERBOSITY: 1
+        with:
+          package-dir: ./cuda_core/
+          output-dir: ${{ env.CUDA_CORE_ARTIFACTS_DIR }}
+
+      - name: List the cuda.core artifacts directory
+        run: |
+          if [[ "${{ inputs.host-platform }}" == win* ]]; then
+            export CHOWN=chown
+          else
+            export CHOWN="sudo chown"
+          fi
+          $CHOWN -R $(whoami) ${{ env.CUDA_CORE_ARTIFACTS_DIR }}
+          ls -lahR ${{ env.CUDA_CORE_ARTIFACTS_DIR }}
+
+      - name: Check cuda.core wheel
+        run: |
+          pip install twine
+          twine check ${{ env.CUDA_CORE_ARTIFACTS_DIR }}/*.whl
+
+      - name: Upload cuda.core build artifacts
+        uses: actions/upload-artifact@v4
+        with:
+          name: ${{ env.CUDA_CORE_ARTIFACT_NAME }}
+          path: ${{ env.CUDA_CORE_ARTIFACTS_DIR }}/*.whl
+          if-no-files-found: error
+
+      - name: Set up mini CTK
+        uses: ./.github/actions/fetch_ctk
+        continue-on-error: false
+        with:
+          host-platform: ${{ inputs.host-platform }}
+          cuda-version: ${{ inputs.cuda-version }}
+
+      - name: Build cuda.bindings wheel
+        uses: pypa/cibuildwheel@v2.22.0
+        env:
+          CIBW_BUILD: ${{ env.CIBW_BUILD }}
+          CIBW_ARCHS_LINUX: "native"
+          CIBW_BUILD_VERBOSITY: 1
+          # CIBW mounts the host filesystem under /host
+          CIBW_ENVIRONMENT_LINUX: >
+            CUDA_PATH=/host/${{ env.CUDA_PATH }}
+            LIBRARY_PATH=/host/${{ env.CUDA_PATH }}/lib
+            CUDA_BINDINGS_PARALLEL_LEVEL=${{ env.CUDA_BINDINGS_PARALLEL_LEVEL }}
+          CIBW_ENVIRONMENT_WINDOWS: >
+            CUDA_HOME="$(cygpath -w ${{ env.CUDA_PATH }})"
+            LIB="${CUDA_HOME}\\lib\\x64;${LIB}"
+            CUDA_BINDINGS_PARALLEL_LEVEL=${{ env.CUDA_BINDINGS_PARALLEL_LEVEL }}
+        with:
+          package-dir: ./cuda_bindings/
+          output-dir: ${{ env.CUDA_BINDINGS_ARTIFACTS_DIR }}
+
+      - name: List the cuda.bindings artifacts directory
+        run: |
+          if [[ "${{ inputs.host-platform }}" == win* ]]; then
+            export CHOWN=chown
+          else
+            export CHOWN="sudo chown"
+          fi
+          $CHOWN -R $(whoami) ${{ env.CUDA_BINDINGS_ARTIFACTS_DIR }}
+          ls -lahR ${{ env.CUDA_BINDINGS_ARTIFACTS_DIR }}
+
+      - name: Check cuda.bindings wheel
+        run: |
+          twine check ${{ env.CUDA_BINDINGS_ARTIFACTS_DIR }}/*.whl
+
+      - name: Upload cuda.bindings build artifacts
+        uses: actions/upload-artifact@v4
+        with:
+          name: ${{ env.CUDA_BINDINGS_ARTIFACT_NAME }}
+          path: ${{ env.CUDA_BINDINGS_ARTIFACTS_DIR }}/*.whl
+          if-no-files-found: error
+
+      # We only need/want a single pure python wheel, pick linux-64 index 0.
+      - name: Build and check cuda-python wheel
+        if: ${{ strategy.job-index == 0 && inputs.host-platform == 'linux-64' }}
+        run: |
+          pushd cuda_python
+          pip wheel -v --no-deps .
+          twine check *.whl
+          popd
+
+      - name: List the cuda-python artifacts directory
+        if: ${{ strategy.job-index == 0 && inputs.host-platform == 'linux-64' }}
+        run: |
+          if [[ "${{ inputs.host-platform }}" == win* ]]; then
+            export CHOWN=chown
+          else
+            export CHOWN="sudo chown"
+          fi
+          $CHOWN -R $(whoami) cuda_python/*.whl
+          ls -lahR cuda_python
+
+      - name: Upload cuda-python build artifacts
+        if: ${{ strategy.job-index == 0 && inputs.host-platform == 'linux-64' }}
+        uses: actions/upload-artifact@v4
+        with:
+          name: cuda-python-wheel
+          path: cuda_python/*.whl
+          if-no-files-found: error
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
new file mode 100644
index 000000000..b4cfb000a
--- /dev/null
+++ b/.github/workflows/ci.yml
@@ -0,0 +1,32 @@
+name: "CI"
+
+concurrency:
+  group: ${{ github.workflow }}-${{
+           github.ref_name == 'main' && format('ci-main-build-test-{0}', github.run_id) ||
+           format('ci-pr-build-test-on-{0}-against-branch-{1}', github.event_name, github.ref_name)
+         }}
+  cancel-in-progress: true
+
+on:
+  push:
+    branches:
+      - "pull-request/[0-9]+"
+      - "main"
+
+jobs:
+  build:
+    strategy:
+      fail-fast: false
+      matrix:
+        host-platform:
+          - linux-64
+          - linux-aarch64
+          - win-64
+    name: Build ${{ matrix.host-platform }}
+    if: ${{ github.repository_owner == 'nvidia' }}
+    secrets: inherit
+    uses:
+      ./.github/workflows/build.yml
+    with:
+      host-platform: ${{ matrix.host-platform }}
+      cuda-version: "12.8.0"

From c251fffe5feac5ca97fcffb500d3a66dd874af85 Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Thu, 17 Apr 2025 14:06:29 -0400
Subject: [PATCH 03/19] Add in some testing matrix updates (simple version)

---
 .github/workflows/ci.yml               | 25 +++++++-
 .github/workflows/test-wheel-linux.yml | 86 +++++++++++++++++++++-----
 2 files changed, 93 insertions(+), 18 deletions(-)

diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index b4cfb000a..4317c6c30 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -21,7 +21,7 @@ jobs:
         host-platform:
           - linux-64
           - linux-aarch64
-          - win-64
+          #- win-64
     name: Build ${{ matrix.host-platform }}
     if: ${{ github.repository_owner == 'nvidia' }}
     secrets: inherit
@@ -30,3 +30,26 @@ jobs:
     with:
       host-platform: ${{ matrix.host-platform }}
       cuda-version: "12.8.0"
+
+  test-linux:
+    strategy:
+      fail-fast: false
+      # TODO: add driver version here
+      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:
+      - build
+    secrets: inherit
+    uses:
+      ./.github/workflows/test-wheel-linux.yml
+    with:
+      build-type: pull-request
+      host-platform: ${{ matrix.host-platform }}
+      local-ctk: 1
+      build-ctk-ver: 12.8.0
diff --git a/.github/workflows/test-wheel-linux.yml b/.github/workflows/test-wheel-linux.yml
index 2761f8c40..f5eca7f4c 100644
--- a/.github/workflows/test-wheel-linux.yml
+++ b/.github/workflows/test-wheel-linux.yml
@@ -7,32 +7,84 @@ name: "CI: Test wheels"
 on:
   workflow_call:
     inputs:
-      host-platform:
+      build-type:
         type: string
         required: true
-      python-version:
+      host-platform:
         type: string
         required: true
       build-ctk-ver:
         type: string
         required: true
-      cuda-version:
-        type: string
-        required: true
       local-ctk:
         type: string
         required: true
-      runner:
+      matrix_filter:
         type: string
-        required: true
+        default: "."
 
 jobs:
+  compute-matrix:
+    runs-on: ubuntu-latest
+    env:
+      BUILD_TYPE: ${{ inputs.build-type }}
+      ARCH: ${{ (inputs.host-platform == 'linux-64' && 'amd64') ||
+                (inputs.host-platform == 'linux-aarch64' && 'arm64') }}
+    outputs:
+      MATRIX: ${{ steps.compute-matrix.outputs.MATRIX }}
+    steps:
+      - name: Validate Test Type
+        run: |
+          if [[ "$BUILD_TYPE" != "pull-request" ]] && [[ "$BUILD_TYPE" != "nightly" ]] && [[ "$BUILD_TYPE" != "branch" ]]; then
+              echo "Invalid build type! Must be one of 'nightly', 'pull-request', or 'branch'."
+              exit 1
+          fi
+      - name: Compute Python Test Matrix
+        id: compute-matrix
+        run: |
+          set -eo pipefail
+          # Please keep the matrices sorted in ascending order by the following:
+          #
+          #     [PY_VER, CUDA_VER, LINUX_VER, GPU, DRIVER]
+          #
+          gpu="l4"
+          if [[ "${ARCH}" == "arm64" ]]; then
+            gpu="a100"
+          fi
+          export MATRICES="
+            pull-request:
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LINUX_VER: 'rockylinux8', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.0.1', LINUX_VER: 'ubuntu24.04', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LINUX_VER: 'ubuntu22.04', GPU: ${gpu}, DRIVER: 'latest' }
+            nightly:
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LINUX_VER: 'rockylinux8', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LINUX_VER: 'ubuntu22.04', GPU: ${gpu}, DRIVER: 'latest' }
+          "
+
+          # Use the nightly matrix for branch tests
+          MATRIX_TYPE="${BUILD_TYPE}"
+          if [[ "${MATRIX_TYPE}" == "branch" ]]; then
+            MATRIX_TYPE="nightly"
+          fi
+          export MATRIX_TYPE
+          TEST_MATRIX=$(yq -n 'env(MATRICES) | .[strenv(MATRIX_TYPE)]')
+          export TEST_MATRIX
+
+          MATRIX="$(
+            yq -n -o json 'env(TEST_MATRIX)' | \
+            jq -c '${{ inputs.matrix_filter }} | if (. | length) > 0 then {include: .} else "Error: Empty matrix\n" | halt_error(1) end'
+          )"
+
+          echo "MATRIX=${MATRIX}" | tee --append "${GITHUB_OUTPUT}"
+
   test:
+    needs: compute-matrix
+    strategy:
+      fail-fast: false
+      matrix: ${{ fromJSON(needs.compute-matrix.outputs.MATRIX) }}
+    runs-on: "linux-${{ matrix.ARCH }}-gpu-${{ matrix.GPU }}-${{ matrix.DRIVER }}-1"
     # The build stage could fail but we want the CI to keep moving.
     if: ${{ github.repository_owner == 'nvidia' && !cancelled() }}
-    runs-on: ${{ (inputs.runner == 'default' && inputs.host-platform == 'linux-64' && 'linux-amd64-gpu-v100-latest-1') ||
-                 (inputs.runner == 'default' && inputs.host-platform == 'linux-aarch64' && 'linux-arm64-gpu-a100-latest-1') ||
-                 (inputs.runner == 'H100' && 'linux-amd64-gpu-h100-latest-1') }}
     # Our self-hosted runners require a container
     # TODO: use a different (nvidia?) container
     container:
@@ -62,7 +114,7 @@ jobs:
 
       - name: Set environment variables
         run: |
-          PYTHON_VERSION_FORMATTED=$(echo '${{ inputs.python-version }}' | tr -d '.')
+          PYTHON_VERSION_FORMATTED=$(echo '${{ matrix.PY_VER }}' | tr -d '.')
           if [[ "${{ inputs.host-platform }}" == linux* ]]; then
             REPO_DIR=$(pwd)
           elif [[ "${{ inputs.host-platform }}" == win* ]]; then
@@ -71,14 +123,14 @@ jobs:
           fi
 
           BUILD_CUDA_MAJOR="$(cut -d '.' -f 1 <<< ${{ inputs.build-ctk-ver }})"
-          TEST_CUDA_MAJOR="$(cut -d '.' -f 1 <<< ${{ inputs.cuda-version }})"
+          TEST_CUDA_MAJOR="$(cut -d '.' -f 1 <<< ${{ matrix.CUDA_VER }})"
           if [[ $BUILD_CUDA_MAJOR != $TEST_CUDA_MAJOR ]]; then
             SKIP_CUDA_BINDINGS_TEST=1
             SKIP_CUDA_CORE_CYTHON_TEST=0
           else
             SKIP_CUDA_BINDINGS_TEST=0
             BUILD_CUDA_MINOR="$(cut -d '.' -f 2 <<< ${{ inputs.build-ctk-ver }})"
-            TEST_CUDA_MINOR="$(cut -d '.' -f 2 <<< ${{ inputs.cuda-version }})"
+            TEST_CUDA_MINOR="$(cut -d '.' -f 2 <<< ${{ matrix.CUDA_VER }})"
             if [[ $BUILD_CUDA_MINOR != $TEST_CUDA_MINOR ]]; then
               SKIP_CUDA_CORE_CYTHON_TEST=1
             else
@@ -179,10 +231,10 @@ jobs:
           pwd
           ls -lahR $CUDA_CORE_ARTIFACTS_DIR
 
-      - name: Set up Python ${{ inputs.python-version }}
+      - name: Set up Python ${{ matrix.PY_VER }}
         uses: actions/setup-python@v5
         with:
-          python-version: ${{ inputs.python-version }}
+          python-version: ${{ matrix.PY_VER }}
         env:
           # we use self-hosted runners on which setup-python behaves weirdly...
           AGENT_TOOLSDIRECTORY: "/opt/hostedtoolcache"
@@ -193,7 +245,7 @@ jobs:
         continue-on-error: false
         with:
           host-platform: ${{ inputs.host-platform }}
-          cuda-version: ${{ inputs.cuda-version }}
+          cuda-version: ${{ matrix.CUDA_VER }}
 
       - name: Set up latest cuda_sanitizer_api
         if: ${{ env.SETUP_SANITIZER == '1' }}
@@ -261,7 +313,7 @@ jobs:
             fi
             popd
           fi
-          TEST_CUDA_MAJOR="$(cut -d '.' -f 1 <<< ${{ inputs.cuda-version }})"
+          TEST_CUDA_MAJOR="$(cut -d '.' -f 1 <<< ${{ matrix.CUDA_VER }})"
           pushd "${CUDA_CORE_ARTIFACTS_DIR}"
           pip install $(ls *.whl)["cu${TEST_CUDA_MAJOR}"]
           popd

From 5f771f6b7835c3db4b33ea633236b2473a18d31b Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Fri, 18 Apr 2025 10:17:05 -0400
Subject: [PATCH 04/19] Tweak labeling and CTK install moves to the matrix

---
 .github/workflows/build.yml            |  2 +-
 .github/workflows/ci.yml               |  3 +-
 .github/workflows/test-wheel-linux.yml | 43 +++++++++++++++-----------
 3 files changed, 27 insertions(+), 21 deletions(-)

diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
index d756220fe..bf2fc99a9 100644
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -23,7 +23,7 @@ jobs:
         python-version:
           - "3.9"
           - "3.13"
-    name: py${{ matrix.python-version }}, ${{ inputs.cuda-version }}
+    name: py${{ matrix.python-version }}
     runs-on: ${{ (inputs.host-platform == 'linux-64' && 'linux-amd64-cpu8') ||
                  (inputs.host-platform == 'linux-aarch64' && 'linux-arm64-cpu8') ||
                  (inputs.host-platform == 'win-64' && 'windows-2019') }}
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 4317c6c30..e4c0f2979 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -22,7 +22,7 @@ jobs:
           - linux-64
           - linux-aarch64
           #- win-64
-    name: Build ${{ matrix.host-platform }}
+    name: Build ${{ matrix.host-platform }}, CUDA 12.8.0
     if: ${{ github.repository_owner == 'nvidia' }}
     secrets: inherit
     uses:
@@ -51,5 +51,4 @@ jobs:
     with:
       build-type: pull-request
       host-platform: ${{ matrix.host-platform }}
-      local-ctk: 1
       build-ctk-ver: 12.8.0
diff --git a/.github/workflows/test-wheel-linux.yml b/.github/workflows/test-wheel-linux.yml
index f5eca7f4c..3275bee7e 100644
--- a/.github/workflows/test-wheel-linux.yml
+++ b/.github/workflows/test-wheel-linux.yml
@@ -16,9 +16,6 @@ on:
       build-ctk-ver:
         type: string
         required: true
-      local-ctk:
-        type: string
-        required: true
       matrix_filter:
         type: string
         default: "."
@@ -43,22 +40,31 @@ jobs:
         id: compute-matrix
         run: |
           set -eo pipefail
-          # Please keep the matrices sorted in ascending order by the following:
-          #
-          #     [PY_VER, CUDA_VER, LINUX_VER, GPU, DRIVER]
-          #
+
+          # Set a default GPU based upon architecture.
           gpu="l4"
           if [[ "${ARCH}" == "arm64" ]]; then
             gpu="a100"
           fi
+          # Add a special entry for the H100 runner on amd64.
+          special_runner=""
+          if [[ "${ARCH}" == "amd64" ]]; then
+            special_runner="- { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: 'H100', DRIVER: 'latest' }"
+          fi
+
+          # Please keep the matrices sorted in ascending order by the following:
+          #
+          #     [PY_VER, CUDA_VER, LOCAL_CTK, GPU, DRIVER]
+          #
           export MATRICES="
             pull-request:
-              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LINUX_VER: 'rockylinux8', GPU: ${gpu}, DRIVER: 'earliest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.0.1', LINUX_VER: 'ubuntu24.04', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LINUX_VER: 'ubuntu22.04', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              ${special_runner}
             nightly:
-              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LINUX_VER: 'rockylinux8', GPU: ${gpu}, DRIVER: 'earliest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LINUX_VER: 'ubuntu22.04', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
           "
 
           # Use the nightly matrix for branch tests
@@ -78,6 +84,7 @@ jobs:
           echo "MATRIX=${MATRIX}" | tee --append "${GITHUB_OUTPUT}"
 
   test:
+    name: py${{ matrix.PY_VER }}, ${{ matrix.CUDA_VER }}, ${{ (matrix.LOCAL_CTK == '1' && 'local') || 'wheels' }}, GPU ${{ matrix.GPU }}
     needs: compute-matrix
     strategy:
       fail-fast: false
@@ -240,7 +247,7 @@ jobs:
           AGENT_TOOLSDIRECTORY: "/opt/hostedtoolcache"
 
       - name: Set up mini CTK
-        if: ${{ inputs.local-ctk == '1' }}
+        if: ${{ matrix.LOCAL_CTK == '1' }}
         uses: ./.github/actions/fetch_ctk
         continue-on-error: false
         with:
@@ -275,7 +282,7 @@ jobs:
         if: ${{ env.SKIP_CUDA_BINDINGS_TEST == '0' }}
         run: |
           pushd "${CUDA_BINDINGS_ARTIFACTS_DIR}"
-          if [[ "${{ inputs.local-ctk }}" == 1 ]]; then
+          if [[ "${{ matrix.LOCAL_CTK }}" == 1 ]]; then
             ls $CUDA_PATH
             pip install *.whl
           else
@@ -289,7 +296,7 @@ jobs:
 
           # It is a bit convoluted to run the Cython tests against CTK wheels,
           # so let's just skip them.
-          if [[ "${{ inputs.local-ctk }}" == 1 ]]; then
+          if [[ "${{ matrix.LOCAL_CTK }}" == 1 ]]; then
             if [[ "${{ inputs.host-platform }}" == linux* ]]; then
               bash tests/cython/build_tests.sh
             elif [[ "${{ inputs.host-platform }}" == win* ]]; then
@@ -306,7 +313,7 @@ jobs:
           # If mismatch: cuda.bindings is installed from the backport branch.
           if [[ "${SKIP_CUDA_BINDINGS_TEST}" == 1 ]]; then
             pushd "${CUDA_BINDINGS_ARTIFACTS_DIR}"
-            if [[ "${{ inputs.local-ctk }}" == 1 ]]; then
+            if [[ "${{ matrix.LOCAL_CTK }}" == 1 ]]; then
               pip install *.whl
             else
               pip install $(ls *.whl)[all]
@@ -326,7 +333,7 @@ jobs:
           # so let's just skip them. Also, currently our CI always installs the
           # latest bindings (from either major version). This is not compatible
           # with the test requirements.
-          if [[ "${{ inputs.local-ctk }}" == 1 && "${SKIP_CUDA_CORE_CYTHON_TEST}" == 0 ]]; then
+          if [[ "${{ matrix.LOCAL_CTK }}" == 1 && "${SKIP_CUDA_CORE_CYTHON_TEST}" == 0 ]]; then
             pip install cython setuptools  # setuptools needed starting PY312
             if [[ "${{ inputs.host-platform }}" == linux* ]]; then
               bash tests/cython/build_tests.sh
@@ -340,7 +347,7 @@ jobs:
 
       - name: Ensure cuda-python installable
         run: |
-          if [[ "${{ inputs.local-ctk }}" == 1 ]]; then
+          if [[ "${{ matrix.LOCAL_CTK }}" == 1 ]]; then
             pip install cuda_python*.whl
           else
             pip install $(ls cuda_python*.whl)[all]

From 2b77ca3117345cc9177b9edca9b37c66210b6831 Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Wed, 23 Apr 2025 15:20:15 -0400
Subject: [PATCH 05/19] Add back in Windows build and test

---
 .github/workflows/build.yml              |  1 +
 .github/workflows/ci.yml                 | 23 +++++-
 .github/workflows/test-wheel-windows.yml | 92 ++++++++++++++++++------
 3 files changed, 93 insertions(+), 23 deletions(-)

diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
index bf2fc99a9..d002ea830 100644
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -22,6 +22,7 @@ jobs:
       matrix:
         python-version:
           - "3.9"
+          - "3.12"
           - "3.13"
     name: py${{ matrix.python-version }}
     runs-on: ${{ (inputs.host-platform == 'linux-64' && 'linux-amd64-cpu8') ||
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index e4c0f2979..4f3d77ca9 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -21,7 +21,7 @@ jobs:
         host-platform:
           - linux-64
           - linux-aarch64
-          #- win-64
+          - win-64
     name: Build ${{ matrix.host-platform }}, CUDA 12.8.0
     if: ${{ github.repository_owner == 'nvidia' }}
     secrets: inherit
@@ -34,7 +34,6 @@ jobs:
   test-linux:
     strategy:
       fail-fast: false
-      # TODO: add driver version here
       matrix:
         host-platform:
           - linux-64
@@ -52,3 +51,23 @@ jobs:
       build-type: pull-request
       host-platform: ${{ matrix.host-platform }}
       build-ctk-ver: 12.8.0
+
+  test-windows:
+    strategy:
+      fail-fast: false
+      matrix:
+        host-platform:
+          - win-64
+    name: Test ${{ matrix.host-platform }}
+    if: ${{ github.repository_owner == 'nvidia' }}
+    permissions:
+      contents: read  # This is required for actions/checkout
+    needs:
+      - build
+    secrets: inherit
+    uses:
+      ./.github/workflows/test-wheel-windows.yml
+    with:
+      build-type: pull-request
+      host-platform: ${{ matrix.host-platform }}
+      build-ctk-ver: 12.8.0
diff --git a/.github/workflows/test-wheel-windows.yml b/.github/workflows/test-wheel-windows.yml
index c741fe161..a1820e890 100644
--- a/.github/workflows/test-wheel-windows.yml
+++ b/.github/workflows/test-wheel-windows.yml
@@ -7,30 +7,80 @@ name: "CI: Test wheels"
 on:
   workflow_call:
     inputs:
-      host-platform:
+      build-type:
         type: string
         required: true
-      python-version:
+      host-platform:
         type: string
         required: true
       build-ctk-ver:
         type: string
         required: true
-      cuda-version:
+      matrix_filter:
         type: string
-        required: true
-      local-ctk:
-        type: string
-        required: true
-      runner:
-        type: string
-        required: true
+        default: "."
 
 jobs:
+  compute-matrix:
+    runs-on: ubuntu-latest
+    env:
+      BUILD_TYPE: ${{ inputs.build-type }}
+      ARCH: ${{ (inputs.host-platform == 'win-64' && 'amd64') }}
+    outputs:
+      MATRIX: ${{ steps.compute-matrix.outputs.MATRIX }}
+    steps:
+      - name: Validate Test Type
+        run: |
+          if [[ "$BUILD_TYPE" != "pull-request" ]] && [[ "$BUILD_TYPE" != "nightly" ]] && [[ "$BUILD_TYPE" != "branch" ]]; then
+              echo "Invalid build type! Must be one of 'nightly', 'pull-request', or 'branch'."
+              exit 1
+          fi
+      - name: Compute Python Test Matrix
+        id: compute-matrix
+        run: |
+          set -eo pipefail
+
+          # Please keep the matrices sorted in ascending order by the following:
+          #
+          #     [PY_VER, CUDA_VER, LOCAL_CTK]
+          #
+          export MATRICES="
+            pull-request:
+              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '11.8.0', LOCAL_CTK: '0' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '11.8.0', LOCAL_CTK: '1' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '12.8.0', LOCAL_CTK: '0' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '12.8.0', LOCAL_CTK: '1' }
+            nightly:
+              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '11.8.0', LOCAL_CTK: '0' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '11.8.0', LOCAL_CTK: '1' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '12.8.0', LOCAL_CTK: '0' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '12.8.0', LOCAL_CTK: '1' }
+          "
+
+          # Use the nightly matrix for branch tests
+          MATRIX_TYPE="${BUILD_TYPE}"
+          if [[ "${MATRIX_TYPE}" == "branch" ]]; then
+            MATRIX_TYPE="nightly"
+          fi
+          export MATRIX_TYPE
+          TEST_MATRIX=$(yq -n 'env(MATRICES) | .[strenv(MATRIX_TYPE)]')
+          export TEST_MATRIX
+
+          MATRIX="$(
+            yq -n -o json 'env(TEST_MATRIX)' | \
+            jq -c '${{ inputs.matrix_filter }} | if (. | length) > 0 then {include: .} else "Error: Empty matrix\n" | halt_error(1) end'
+          )"
+
+          echo "MATRIX=${MATRIX}" | tee --append "${GITHUB_OUTPUT}"
+
   test:
     # The build stage could fail but we want the CI to keep moving.
+    needs: compute-matrix
+    strategy:
+      fail-fast: false
+      matrix: ${{ fromJSON(needs.compute-matrix.outputs.MATRIX) }}
     if: ${{ github.repository_owner == 'nvidia' && !cancelled() }}
-    runs-on: ${{ (inputs.runner == 'default' && inputs.host-platform == 'win-64' && 'cuda-python-windows-gpu-github') }}
+    runs-on: 'cuda-python-windows-gpu-github'
     steps:
       - name: Checkout ${{ github.event.repository.name }}
         uses: actions/checkout@v4
@@ -46,11 +96,11 @@ jobs:
 
       - name: Set environment variables
         run: |
-          $PYTHON_VERSION_FORMATTED = '${{ inputs.python-version }}' -replace '\.'
+          $PYTHON_VERSION_FORMATTED = '${{ matrix.PY_VER }}' -replace '\.'
           $REPO_DIR = $PWD.Path
 
           $BUILD_CUDA_MAJOR = '${{ inputs.build-ctk-ver }}' -split '\.' | Select-Object -First 1
-          $TEST_CUDA_MAJOR = '${{ inputs.cuda-version }}' -split '\.' | Select-Object -First 1
+          $TEST_CUDA_MAJOR = '${{ matrix.CUDA_VER }}' -split '\.' | Select-Object -First 1
           if ($BUILD_CUDA_MAJOR -ne $TEST_CUDA_MAJOR) {
             $SKIP_CUDA_BINDINGS_TEST = 1
           } else {
@@ -174,24 +224,24 @@ jobs:
           Get-Location
           Get-ChildItem -Recurse -Force $env:CUDA_CORE_ARTIFACTS_DIR | Select-Object Mode, LastWriteTime, Length, FullName
 
-      - name: Set up Python ${{ inputs.python-version }}
+      - name: Set up Python ${{ matrix.PY_VER }}
         uses: actions/setup-python@v5
         with:
-          python-version: ${{ inputs.python-version }}
+          python-version: ${{ matrix.PY_VER }}
 
       - name: Set up mini CTK
-        if: ${{ inputs.local-ctk == '1' }}
+        if: ${{ matrix.LOCAL_CTK == '1' }}
         uses: ./.github/actions/fetch_ctk
         continue-on-error: false
         with:
           host-platform: ${{ inputs.host-platform }}
-          cuda-version: ${{ inputs.cuda-version }}
+          cuda-version: ${{ matrix.CUDA_VER }}
 
       - name: Run cuda.bindings tests
         if: ${{ env.SKIP_CUDA_BINDINGS_TEST == '0' }}
         run: |
           Push-Location $env:CUDA_BINDINGS_ARTIFACTS_DIR
-          if ('${{ inputs.local-ctk }}' -eq '1') {
+          if ('${{ matrix.LOCAL_CTK }}' -eq '1') {
             Get-ChildItem $env:CUDA_PATH
             echo $PATH
             pip install (Get-ChildItem -Filter *.whl).FullName
@@ -212,14 +262,14 @@ jobs:
           # If mismatch: cuda.bindings is installed from the backport branch.
           if ($env:SKIP_CUDA_BINDINGS_TEST -eq '1') {
             Push-Location $env:CUDA_BINDINGS_ARTIFACTS_DIR
-            if ('${{ inputs.local-ctk }}' -eq '1') {
+            if ('${{ matrix.LOCAL_CTK }}' -eq '1') {
               pip install (Get-ChildItem -Filter *.whl).FullName
             } else {
               pip install "$((Get-ChildItem -Filter *.whl).FullName)[all]"
             }
             Pop-Location
           }
-          $TEST_CUDA_MAJOR = '${{ inputs.cuda-version }}' -split '\.' | Select-Object -First 1
+          $TEST_CUDA_MAJOR = '${{ matrix.CUDA_VER }}' -split '\.' | Select-Object -First 1
           Push-Location $env:CUDA_CORE_ARTIFACTS_DIR
           pip install "$((Get-ChildItem -Filter *.whl).FullName)[cu${TEST_CUDA_MAJOR}]"
           Pop-Location
@@ -231,7 +281,7 @@ jobs:
 
       - name: Ensure cuda-python installable
         run: |
-          if ('${{ inputs.local-ctk }}' -eq '1') {
+          if ('${{ matrix.LOCAL_CTK }}' -eq '1') {
             pip install (Get-ChildItem -Filter cuda_python*.whl).FullName
           } else {
             pip install "$((Get-ChildItem -Filter cuda_python*.whl).FullName)[all]"

From d2d9def453e0a77810b791744a589de4e23d5e86 Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Wed, 23 Apr 2025 15:58:40 -0400
Subject: [PATCH 06/19] Use better naming for the windows test runs

---
 .github/workflows/test-wheel-windows.yml | 1 +
 1 file changed, 1 insertion(+)

diff --git a/.github/workflows/test-wheel-windows.yml b/.github/workflows/test-wheel-windows.yml
index a1820e890..2114c7cf4 100644
--- a/.github/workflows/test-wheel-windows.yml
+++ b/.github/workflows/test-wheel-windows.yml
@@ -74,6 +74,7 @@ jobs:
           echo "MATRIX=${MATRIX}" | tee --append "${GITHUB_OUTPUT}"
 
   test:
+    name: py${{ matrix.PY_VER }}, ${{ matrix.CUDA_VER }}, ${{ (matrix.LOCAL_CTK == '1' && 'local') || 'wheels' }}
     # The build stage could fail but we want the CI to keep moving.
     needs: compute-matrix
     strategy:

From 2a11961013adc741cf4006113a6b27a6da779180 Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Thu, 24 Apr 2025 10:28:33 -0400
Subject: [PATCH 07/19] Get CUDA build version from JSON, docs and checks, full
 matrix

---
 .github/workflows/build.yml            |  2 +
 .github/workflows/ci.yml               | 62 ++++++++++++++++++++++----
 .github/workflows/test-wheel-linux.yml | 50 ++++++++++++++++++++-
 versions.json                          |  7 +++
 4 files changed, 111 insertions(+), 10 deletions(-)
 create mode 100644 versions.json

diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
index d002ea830..5f0cfb7d5 100644
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -22,6 +22,8 @@ jobs:
       matrix:
         python-version:
           - "3.9"
+          - "3.10"
+          - "3.11"
           - "3.12"
           - "3.13"
     name: py${{ matrix.python-version }}
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 4f3d77ca9..546a46d23 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -1,10 +1,7 @@
 name: "CI"
 
 concurrency:
-  group: ${{ github.workflow }}-${{
-           github.ref_name == 'main' && format('ci-main-build-test-{0}', github.run_id) ||
-           format('ci-pr-build-test-on-{0}-against-branch-{1}', github.event_name, github.ref_name)
-         }}
+  group: ${{ github.workflow }}-${{ github.ref }}-${{ github.event_name }}
   cancel-in-progress: true
 
 on:
@@ -14,7 +11,24 @@ on:
       - "main"
 
 jobs:
+  ci-vars:
+    runs-on: ubuntu-latest
+    outputs:
+      CUDA_BUILD_VER: ${{ steps.get-vars.outputs.cuda_build_ver }}
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+      - name: Get CUDA build version
+        id: get-vars
+        run: |
+          cuda_build_ver=$(jq -r .cuda.build.version versions.json)
+          echo "cuda_build_ver=$cuda_build_ver" >> $GITHUB_OUTPUT
+
   build:
+    needs:
+      - ci-vars
     strategy:
       fail-fast: false
       matrix:
@@ -22,14 +36,14 @@ jobs:
           - linux-64
           - linux-aarch64
           - win-64
-    name: Build ${{ matrix.host-platform }}, CUDA 12.8.0
+    name: Build ${{ matrix.host-platform }}, CUDA ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
     if: ${{ github.repository_owner == 'nvidia' }}
     secrets: inherit
     uses:
       ./.github/workflows/build.yml
     with:
       host-platform: ${{ matrix.host-platform }}
-      cuda-version: "12.8.0"
+      cuda-version: ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
 
   test-linux:
     strategy:
@@ -43,6 +57,7 @@ jobs:
     permissions:
       contents: read  # This is required for actions/checkout
     needs:
+      - ci-vars
       - build
     secrets: inherit
     uses:
@@ -50,7 +65,7 @@ jobs:
     with:
       build-type: pull-request
       host-platform: ${{ matrix.host-platform }}
-      build-ctk-ver: 12.8.0
+      build-ctk-ver: ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
 
   test-windows:
     strategy:
@@ -63,6 +78,7 @@ jobs:
     permissions:
       contents: read  # This is required for actions/checkout
     needs:
+      - ci-vars
       - build
     secrets: inherit
     uses:
@@ -70,4 +86,34 @@ jobs:
     with:
       build-type: pull-request
       host-platform: ${{ matrix.host-platform }}
-      build-ctk-ver: 12.8.0
+      build-ctk-ver: ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
+
+  doc:
+    name: Docs
+    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
+    needs:
+      - ci-vars
+      - build
+    secrets: inherit
+    uses:
+      ./.github/workflows/build-docs.yml
+    with:
+      build-ctk-ver: ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
+
+  checks:
+    name: Check job status
+    permissions:
+      checks: read
+    needs:
+      - build
+      - test-linux
+      - test-windows
+      - doc
+    secrets: inherit
+    uses:
+      ./.github/workflows/status-check.yml
diff --git a/.github/workflows/test-wheel-linux.yml b/.github/workflows/test-wheel-linux.yml
index 3275bee7e..e13dafb1f 100644
--- a/.github/workflows/test-wheel-linux.yml
+++ b/.github/workflows/test-wheel-linux.yml
@@ -58,13 +58,59 @@ jobs:
           #
           export MATRICES="
             pull-request:
-              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
               ${special_runner}
             nightly:
-              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              ${special_runner}
           "
 
           # Use the nightly matrix for branch tests
diff --git a/versions.json b/versions.json
new file mode 100644
index 000000000..eb1e19f2f
--- /dev/null
+++ b/versions.json
@@ -0,0 +1,7 @@
+{
+  "cuda": {
+    "build": {
+      "version": "12.8.0"
+    }
+  }
+}

From 231ce39f090191333a3cf6c4efec225b4a172e6a Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Thu, 24 Apr 2025 11:21:49 -0400
Subject: [PATCH 08/19] Use the latest driver for CUDA 12.8.0 testing

---
 .github/workflows/test-wheel-linux.yml | 20 ++++++++++----------
 1 file changed, 10 insertions(+), 10 deletions(-)

diff --git a/.github/workflows/test-wheel-linux.yml b/.github/workflows/test-wheel-linux.yml
index e13dafb1f..16024c787 100644
--- a/.github/workflows/test-wheel-linux.yml
+++ b/.github/workflows/test-wheel-linux.yml
@@ -62,54 +62,54 @@ jobs:
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               ${special_runner}
             nightly:
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               ${special_runner}
           "
 

From 3c293313270cc0c281833498857e81dcc9ef5563 Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Thu, 24 Apr 2025 16:20:10 -0400
Subject: [PATCH 09/19] Trim down the matrix, remove old workflow

---
 .github/workflows/test-wheel-linux.yml | 13 -------------
 1 file changed, 13 deletions(-)

diff --git a/.github/workflows/test-wheel-linux.yml b/.github/workflows/test-wheel-linux.yml
index 16024c787..23cf64ebe 100644
--- a/.github/workflows/test-wheel-linux.yml
+++ b/.github/workflows/test-wheel-linux.yml
@@ -59,30 +59,17 @@ jobs:
           export MATRICES="
             pull-request:
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               ${special_runner}
             nightly:
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }

From ca3f18ee8d999089ba0fc07410677c40f0aa0861 Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Thu, 24 Apr 2025 17:13:29 -0400
Subject: [PATCH 10/19] Tweak the matrix for more coverage

Add a note on combinations that are known not to work together.
---
 .github/workflows/test-wheel-linux.yml | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

diff --git a/.github/workflows/test-wheel-linux.yml b/.github/workflows/test-wheel-linux.yml
index 23cf64ebe..8378222e6 100644
--- a/.github/workflows/test-wheel-linux.yml
+++ b/.github/workflows/test-wheel-linux.yml
@@ -56,18 +56,20 @@ jobs:
           #
           #     [PY_VER, CUDA_VER, LOCAL_CTK, GPU, DRIVER]
           #
+          # Note that DRIVER: `earliest` does not work with CUDA 12.8.0 and LOCAL_CTK: 0 does not work with CUDA 12.0.1.
+          #
           export MATRICES="
             pull-request:
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
               ${special_runner}

From ef440a5af457b1bd9debd8f14f6ac5521b1f2b94 Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Mon, 5 May 2025 12:41:23 -0400
Subject: [PATCH 11/19] Suggestion from Leo to retain original workflow name

---
 .github/workflows/ci.yml | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 546a46d23..3e64cbd75 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -1,4 +1,5 @@
-name: "CI"
+# Note: This name is referred to in the test job, so make sure any changes are sync'd up!
+name: "CI: Build and test"
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}-${{ github.event_name }}

From e5286d0bef093bebe2c79d3b978f56d3e4e9f6db Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Mon, 5 May 2025 12:49:29 -0400
Subject: [PATCH 12/19] Rename build.yml and addd comment on action

---
 .github/workflows/{build.yml => build-wheel.yml} | 2 ++
 .github/workflows/ci.yml                         | 2 +-
 2 files changed, 3 insertions(+), 1 deletion(-)
 rename .github/workflows/{build.yml => build-wheel.yml} (98%)

diff --git a/.github/workflows/build.yml b/.github/workflows/build-wheel.yml
similarity index 98%
rename from .github/workflows/build.yml
rename to .github/workflows/build-wheel.yml
index 5f0cfb7d5..c2c016cdc 100644
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build-wheel.yml
@@ -47,6 +47,8 @@ jobs:
         id: setup-python
         uses: actions/setup-python@v5
         with:
+          # WAR: setup-python is not relocatable...
+          # see https://github.com/actions/setup-python/issues/871
           python-version: "3.12"
 
       - name: Set up MSVC
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 3e64cbd75..9b9064e8f 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -41,7 +41,7 @@ jobs:
     if: ${{ github.repository_owner == 'nvidia' }}
     secrets: inherit
     uses:
-      ./.github/workflows/build.yml
+      ./.github/workflows/build-wheel.yml
     with:
       host-platform: ${{ matrix.host-platform }}
       cuda-version: ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}

From 35716986c6de388bd51aa6743a2d47d0aafefdfd Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Mon, 5 May 2025 12:49:59 -0400
Subject: [PATCH 13/19] Adjust bash default for the workflow

---
 .github/workflows/test-wheel-linux.yml | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/.github/workflows/test-wheel-linux.yml b/.github/workflows/test-wheel-linux.yml
index 8378222e6..899bafb95 100644
--- a/.github/workflows/test-wheel-linux.yml
+++ b/.github/workflows/test-wheel-linux.yml
@@ -20,6 +20,10 @@ on:
         type: string
         default: "."
 
+defaults:
+  run:
+    shell: bash --noprofile --norc -xeuo pipefail {0}
+
 jobs:
   compute-matrix:
     runs-on: ubuntu-latest
@@ -39,8 +43,6 @@ jobs:
       - name: Compute Python Test Matrix
         id: compute-matrix
         run: |
-          set -eo pipefail
-
           # Set a default GPU based upon architecture.
           gpu="l4"
           if [[ "${ARCH}" == "arm64" ]]; then
@@ -134,9 +136,6 @@ jobs:
       image: ubuntu:22.04
       env:
         NVIDIA_VISIBLE_DEVICES: ${{ env.NVIDIA_VISIBLE_DEVICES }}
-    defaults:
-      run:
-        shell: bash --noprofile --norc -xeuo pipefail {0}
     steps:
       - name: Ensure GPU is working
         run: nvidia-smi

From 4d1738b5fa9a0cdcee648f45f06a6dfec007f429 Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Mon, 5 May 2025 12:54:00 -0400
Subject: [PATCH 14/19] Adjust the bash default for workflows

---
 .github/workflows/test-wheel-windows.yml | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/test-wheel-windows.yml b/.github/workflows/test-wheel-windows.yml
index 2114c7cf4..4dadf1b11 100644
--- a/.github/workflows/test-wheel-windows.yml
+++ b/.github/workflows/test-wheel-windows.yml
@@ -23,6 +23,9 @@ on:
 jobs:
   compute-matrix:
     runs-on: ubuntu-latest
+    defaults:
+      run:
+        shell: bash --noprofile --norc -xeuo pipefail {0}
     env:
       BUILD_TYPE: ${{ inputs.build-type }}
       ARCH: ${{ (inputs.host-platform == 'win-64' && 'amd64') }}
@@ -38,8 +41,6 @@ jobs:
       - name: Compute Python Test Matrix
         id: compute-matrix
         run: |
-          set -eo pipefail
-
           # Please keep the matrices sorted in ascending order by the following:
           #
           #     [PY_VER, CUDA_VER, LOCAL_CTK]

From 453531004be068418edcb3de77621a9c18719c5e Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Mon, 5 May 2025 13:41:00 -0400
Subject: [PATCH 15/19] Move the JSON file, keep names unique

---
 .github/workflows/ci.yml          | 5 +++--
 versions.json => ci/versions.json | 0
 2 files changed, 3 insertions(+), 2 deletions(-)
 rename versions.json => ci/versions.json (100%)

diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 9b9064e8f..dd26b8cd5 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -1,5 +1,6 @@
 # Note: This name is referred to in the test job, so make sure any changes are sync'd up!
-name: "CI: Build and test"
+# Further this is referencing a run in the backport branch to fetch old bindings.
+name: "CI"
 
 concurrency:
   group: ${{ github.workflow }}-${{ github.ref }}-${{ github.event_name }}
@@ -24,7 +25,7 @@ jobs:
       - name: Get CUDA build version
         id: get-vars
         run: |
-          cuda_build_ver=$(jq -r .cuda.build.version versions.json)
+          cuda_build_ver=$(jq -r .cuda.build.version ci/versions.json)
           echo "cuda_build_ver=$cuda_build_ver" >> $GITHUB_OUTPUT
 
   build:
diff --git a/versions.json b/ci/versions.json
similarity index 100%
rename from versions.json
rename to ci/versions.json

From ab1a327d042ac8e944ccd77a71d035cee425104f Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Mon, 5 May 2025 15:40:58 -0400
Subject: [PATCH 16/19] 12.0.1 won't with with CUDA 11.x drivers

Also fix the conditional install of git for windows based on local CTK.
---
 .github/workflows/test-wheel-linux.yml   | 2 +-
 .github/workflows/test-wheel-windows.yml | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/.github/workflows/test-wheel-linux.yml b/.github/workflows/test-wheel-linux.yml
index 899bafb95..bcc24f459 100644
--- a/.github/workflows/test-wheel-linux.yml
+++ b/.github/workflows/test-wheel-linux.yml
@@ -69,7 +69,7 @@ jobs:
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
diff --git a/.github/workflows/test-wheel-windows.yml b/.github/workflows/test-wheel-windows.yml
index 4dadf1b11..8f694e5dd 100644
--- a/.github/workflows/test-wheel-windows.yml
+++ b/.github/workflows/test-wheel-windows.yml
@@ -152,7 +152,7 @@ jobs:
 
       - name: Install Git for Windows
         # the GPU runner image does not have Git Bash pre-installed...
-        if: ${{ inputs.local-ctk == '1' }}
+        if: ${{ matrix.LOCAL_CTK == '1' }}
         env:
           # doesn't seem there's an easy way to avoid hard-coding it?
           GFW_EXE_URL: https://github.com/git-for-windows/git/releases/download/v2.49.0.windows.1/PortableGit-2.49.0-64-bit.7z.exe
@@ -167,7 +167,7 @@ jobs:
 
       - name: Install zstd
         # the GPU runner image does not have zstd pre-installed... and it's needed by actions/cache
-        if: ${{ inputs.local-ctk == '1' }}
+        if: ${{ matrix.LOCAL_CTK == '1' }}
         env:
           # doesn't seem there's an easy way to avoid hard-coding it?
           ZSTD_URL: https://github.com/facebook/zstd/releases/download/v1.5.7/zstd-v1.5.7-win64.zip

From 2e811f3ec4b765a6ba74203989f968e28df82912 Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Mon, 5 May 2025 17:38:41 -0400
Subject: [PATCH 17/19] Remove `build-and-test.yml`

This is replaced by `ci.yml`.
---
 .github/workflows/build-and-test.yml | 324 ---------------------------
 1 file changed, 324 deletions(-)
 delete mode 100644 .github/workflows/build-and-test.yml

diff --git a/.github/workflows/build-and-test.yml b/.github/workflows/build-and-test.yml
deleted file mode 100644
index f996ba79e..000000000
--- a/.github/workflows/build-and-test.yml
+++ /dev/null
@@ -1,324 +0,0 @@
-# Copyright (c) 2024-2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
-#
-# SPDX-License-Identifier: Apache-2.0
-
-# Note: This name is referred to in the test job, so make sure any changes are sync'd up!
-name: "CI: Build and test"
-
-concurrency:
-  group: ${{ github.workflow }}-${{
-           github.ref_name == 'main' && format('ci-main-build-test-{0}', github.run_id) ||
-           format('ci-pr-build-test-on-{0}-against-branch-{1}', github.event_name, github.ref_name)
-         }}
-  cancel-in-progress: true
-
-on:
-  push:
-    branches:
-      #- "pull-request/[0-9]+"
-      - "main"
-
-jobs:
-  build:
-    strategy:
-      fail-fast: false
-      matrix:
-        host-platform:
-          - linux-64
-          - linux-aarch64
-          - win-64
-        python-version:
-          - "3.13"
-          - "3.12"
-          - "3.11"
-          - "3.10"
-          - "3.9"
-        cuda-version:
-          # Note: this is for build-time only.
-          - "12.9.0"
-    name: Build (${{ matrix.host-platform }}, Python ${{ matrix.python-version }}, CUDA ${{ matrix.cuda-version }})
-    if: ${{ github.repository_owner == 'nvidia' }}
-    permissions:
-      contents: read  # This is required for actions/checkout
-    runs-on: ${{ (matrix.host-platform == 'linux-64' && 'linux-amd64-cpu8') ||
-                 (matrix.host-platform == 'linux-aarch64' && 'linux-arm64-cpu8') ||
-                 (matrix.host-platform == 'win-64' && 'windows-2019') }}
-               #  (matrix.host-platform == 'win-64' && 'windows-amd64-cpu8') }}
-    outputs:
-      BUILD_CTK_VER: ${{ steps.pass_env.outputs.CUDA_VERSION }}
-    defaults:
-      run:
-        shell: bash --noprofile --norc -xeuo pipefail {0}
-    steps:
-      - name: Checkout ${{ github.event.repository.name }}
-        uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-
-      # WAR: setup-python is not relocatable...
-      # see https://github.com/actions/setup-python/issues/871
-      - name: Set up Python ${{ matrix.python-version }}
-        if: ${{ startsWith(matrix.host-platform, 'linux') }}
-        id: setup-python
-        uses: actions/setup-python@v5
-        with:
-          python-version: "3.12"
-  
-      - name: Set up MSVC
-        if: ${{ startsWith(matrix.host-platform, 'win') }}
-        uses: ilammy/msvc-dev-cmd@v1
-  
-      - name: Set environment variables
-        run: |
-          PYTHON_VERSION_FORMATTED=$(echo '${{ matrix.python-version }}' | tr -d '.')
-          if [[ "${{ matrix.host-platform }}" == linux* ]]; then
-            CIBW_BUILD="cp${PYTHON_VERSION_FORMATTED}-manylinux*"
-            REPO_DIR=$(pwd)
-          elif [[ "${{ matrix.host-platform }}" == win* ]]; then
-            CIBW_BUILD="cp${PYTHON_VERSION_FORMATTED}-win_amd64"
-            PWD=$(pwd)
-            REPO_DIR=$(cygpath -w $PWD)
-          fi
-  
-          echo "CUDA_BINDINGS_PARALLEL_LEVEL=$(nproc)" >> $GITHUB_ENV
-          CUDA_CORE_ARTIFACT_BASENAME="cuda-core-python${PYTHON_VERSION_FORMATTED}-${{ matrix.host-platform }}"
-          echo "CUDA_CORE_ARTIFACT_BASENAME=${CUDA_CORE_ARTIFACT_BASENAME}" >> $GITHUB_ENV
-          echo "CUDA_CORE_ARTIFACT_NAME=${CUDA_CORE_ARTIFACT_BASENAME}-${{ github.sha }}" >> $GITHUB_ENV
-          echo "CUDA_CORE_ARTIFACTS_DIR=$(realpath "$REPO_DIR/cuda_core/dist")" >> $GITHUB_ENV
-          CUDA_BINDINGS_ARTIFACT_BASENAME="cuda-bindings-python${PYTHON_VERSION_FORMATTED}-cuda${{ matrix.cuda-version }}-${{ matrix.host-platform }}"
-          echo "CUDA_BINDINGS_ARTIFACT_BASENAME=${CUDA_BINDINGS_ARTIFACT_BASENAME}" >> $GITHUB_ENV
-          echo "CUDA_BINDINGS_ARTIFACT_NAME=${CUDA_BINDINGS_ARTIFACT_BASENAME}-${{ github.sha }}" >> $GITHUB_ENV
-          echo "CUDA_BINDINGS_ARTIFACTS_DIR=$(realpath "$REPO_DIR/cuda_bindings/dist")" >> $GITHUB_ENV
-          echo "CIBW_BUILD=${CIBW_BUILD}" >> $GITHUB_ENV
-
-      - name: Dump environment
-        run: |
-          env
-
-      - name: Build cuda.core wheel
-        uses: pypa/cibuildwheel@v2.22.0
-        env:
-          CIBW_BUILD: ${{ env.CIBW_BUILD }}
-          CIBW_ARCHS_LINUX: "native"
-          CIBW_BUILD_VERBOSITY: 1
-        with:
-          package-dir: ./cuda_core/
-          output-dir: ${{ env.CUDA_CORE_ARTIFACTS_DIR }}
-
-      - name: List the cuda.core artifacts directory
-        run: |
-          if [[ "${{ matrix.host-platform }}" == win* ]]; then
-            export CHOWN=chown
-          else
-            export CHOWN="sudo chown"
-          fi
-          $CHOWN -R $(whoami) ${{ env.CUDA_CORE_ARTIFACTS_DIR }}
-          ls -lahR ${{ env.CUDA_CORE_ARTIFACTS_DIR }}
-
-      - name: Check cuda.core wheel
-        run: |
-          pip install twine
-          twine check ${{ env.CUDA_CORE_ARTIFACTS_DIR }}/*.whl
-
-      - name: Upload cuda.core build artifacts
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ env.CUDA_CORE_ARTIFACT_NAME }}
-          path: ${{ env.CUDA_CORE_ARTIFACTS_DIR }}/*.whl
-          if-no-files-found: error
-
-      - name: Set up mini CTK
-        uses: ./.github/actions/fetch_ctk
-        continue-on-error: false
-        with:
-          host-platform: ${{ matrix.host-platform }}
-          cuda-version: ${{ matrix.cuda-version }}
-
-      - name: Build cuda.bindings wheel
-        uses: pypa/cibuildwheel@v2.22.0
-        env:
-          CIBW_BUILD: ${{ env.CIBW_BUILD }}
-          CIBW_ARCHS_LINUX: "native"
-          CIBW_BUILD_VERBOSITY: 1
-          # CIBW mounts the host filesystem under /host
-          CIBW_ENVIRONMENT_LINUX: >
-            CUDA_PATH=/host/${{ env.CUDA_PATH }}
-            LIBRARY_PATH=/host/${{ env.CUDA_PATH }}/lib
-            CUDA_BINDINGS_PARALLEL_LEVEL=${{ env.CUDA_BINDINGS_PARALLEL_LEVEL }}
-          CIBW_ENVIRONMENT_WINDOWS: >
-            CUDA_HOME=${{ env.CUDA_PATH }}
-            LIB="${CUDA_HOME}\\lib\\x64;${LIB}"
-            CUDA_BINDINGS_PARALLEL_LEVEL=${{ env.CUDA_BINDINGS_PARALLEL_LEVEL }}
-        with:
-          package-dir: ./cuda_bindings/
-          output-dir: ${{ env.CUDA_BINDINGS_ARTIFACTS_DIR }}
-
-      - name: List the cuda.bindings artifacts directory
-        run: |
-          if [[ "${{ matrix.host-platform }}" == win* ]]; then
-            export CHOWN=chown
-          else
-            export CHOWN="sudo chown"
-          fi
-          $CHOWN -R $(whoami) ${{ env.CUDA_BINDINGS_ARTIFACTS_DIR }}
-          ls -lahR ${{ env.CUDA_BINDINGS_ARTIFACTS_DIR }}
-
-      - name: Check cuda.bindings wheel
-        run: |
-          twine check ${{ env.CUDA_BINDINGS_ARTIFACTS_DIR }}/*.whl
-
-      - name: Upload cuda.bindings build artifacts
-        uses: actions/upload-artifact@v4
-        with:
-          name: ${{ env.CUDA_BINDINGS_ARTIFACT_NAME }}
-          path: ${{ env.CUDA_BINDINGS_ARTIFACTS_DIR }}/*.whl
-          if-no-files-found: error
-
-      # upload-artifact's "overwrite: true" option has a race condition among parallel
-      # jobs, so we let job 0 do the work
-      - name: Build and check cuda-python wheel
-        if: ${{ strategy.job-index == 0 }}
-        run: |
-          pushd cuda_python
-          pip wheel -v --no-deps .
-          twine check *.whl
-          popd
-
-      - name: List the cuda-python artifacts directory
-        if: ${{ strategy.job-index == 0 }}
-        run: |
-          if [[ "${{ matrix.host-platform }}" == win* ]]; then
-            export CHOWN=chown
-          else
-            export CHOWN="sudo chown"
-          fi
-          $CHOWN -R $(whoami) cuda_python/*.whl
-          ls -lahR cuda_python
-
-      - name: Upload cuda-python build artifacts
-        if: ${{ strategy.job-index == 0 }}
-        uses: actions/upload-artifact@v4
-        with:
-          name: cuda-python-wheel
-          path: cuda_python/*.whl
-          if-no-files-found: error
-
-      - name: Pass environment variables to the next runner
-        id: pass_env
-        run: |
-          echo "CUDA_VERSION=${{ matrix.cuda-version }}" >> $GITHUB_OUTPUT
-
-  test-linux:
-    strategy:
-      fail-fast: false
-      # TODO: add driver version here
-      matrix:
-        host-platform:
-          - linux-64
-          - linux-aarch64
-        python-version:
-          - "3.13"
-          - "3.12"
-          - "3.11"
-          - "3.10"
-          - "3.9"
-        cuda-version:
-          # Note: this is for test-time only.
-          - "12.9.0"
-          - "12.0.1"
-          - "11.8.0"
-        local-ctk:
-          - 1  # use mini CTK
-          - 0  # use CTK wheels
-        runner:
-          - default
-        exclude:
-          # To test this combo would require nontrivial installation steps.
-          - cuda-version: "12.0.1"
-            local-ctk: 0
-        include:
-          - host-platform: linux-64
-            python-version: "3.12"
-            cuda-version: "12.9.0"
-            local-ctk: 1
-            runner: H100
-    name: Test (${{ matrix.host-platform }}, Python ${{ matrix.python-version }}, CUDA ${{ matrix.cuda-version }}, Runner ${{ matrix.runner }}, ${{ (matrix.local-ctk == '1' && 'local CTK') || 'CTK wheels' }})
-    if: ${{ github.repository_owner == 'nvidia' }}
-    permissions:
-      contents: read  # This is required for actions/checkout
-    needs:
-      - build
-    secrets: inherit
-    uses:
-      ./.github/workflows/test-wheel-linux.yml
-    with:
-      host-platform: ${{ matrix.host-platform }}
-      python-version: ${{ matrix.python-version }}
-      build-ctk-ver: ${{ needs.build.outputs.BUILD_CTK_VER }}
-      cuda-version: ${{ matrix.cuda-version }}
-      local-ctk: ${{ matrix.local-ctk}}
-      runner: ${{ matrix.runner }}
-
-  test-windows:
-    strategy:
-      fail-fast: false
-      # TODO: add driver version here
-      matrix:
-        host-platform:
-          - win-64
-        python-version:
-          - "3.12"
-        cuda-version:
-          # Note: this is for test-time only.
-          - "12.9.0"
-          - "11.8.0"
-        local-ctk:
-          - 1  # use mini CTK
-          - 0  # use CTK wheels
-        runner:
-          - default
-    name: Test (${{ matrix.host-platform }}, Python ${{ matrix.python-version }}, CUDA ${{ matrix.cuda-version }}, Runner ${{ matrix.runner }}, ${{ (matrix.local-ctk == '1' && 'local CTK') || 'CTK wheels' }})
-    if: ${{ github.repository_owner == 'nvidia' }}
-    permissions:
-      contents: read  # This is required for actions/checkout
-    needs:
-      - build
-    secrets: inherit
-    uses:
-      ./.github/workflows/test-wheel-windows.yml
-    with:
-      host-platform: ${{ matrix.host-platform }}
-      python-version: ${{ matrix.python-version }}
-      build-ctk-ver: ${{ needs.build.outputs.BUILD_CTK_VER }}
-      cuda-version: ${{ matrix.cuda-version }}
-      local-ctk: ${{ matrix.local-ctk}}
-      runner: ${{ matrix.runner }}
-
-  doc:
-    name: Docs
-    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
-    needs:
-      - build
-    secrets: inherit
-    uses:
-      ./.github/workflows/build-docs.yml
-    with:
-      build-ctk-ver: ${{ needs.build.outputs.BUILD_CTK_VER }}
-
-  checks:
-    name: Check job status
-    permissions:
-      checks: read
-    needs:
-      - build
-      - test-linux
-      - test-windows
-      - doc
-    secrets: inherit
-    uses:
-      ./.github/workflows/status-check.yml

From c5eb72ceaf157babfca3176f7c681f329e784507 Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Tue, 6 May 2025 10:34:52 -0400
Subject: [PATCH 18/19] Make `cuda-verion` required, add missing
 copright/license

Also pin actions to SHAs (skip ilammy/msvc-dev-cmd for now awaiting
update to the allow list).
---
 .github/workflows/build-wheel.yml | 10 +++++++---
 .github/workflows/ci.yml          |  4 ++++
 2 files changed, 11 insertions(+), 3 deletions(-)

diff --git a/.github/workflows/build-wheel.yml b/.github/workflows/build-wheel.yml
index c2c016cdc..e776c70ec 100644
--- a/.github/workflows/build-wheel.yml
+++ b/.github/workflows/build-wheel.yml
@@ -1,3 +1,7 @@
+# Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+#
+# SPDX-License-Identifier: Apache-2.0
+
 on:
   workflow_call:
     inputs:
@@ -5,8 +9,8 @@ on:
         required: true
         type: string
       cuda-version:
+        required: true
         type: string
-        default: "12.8.0"
 
 defaults:
   run:
@@ -83,7 +87,7 @@ jobs:
           env
       
       - name: Build cuda.core wheel
-        uses: pypa/cibuildwheel@v2.22.0
+        uses: pypa/cibuildwheel@faf86a6ed7efa889faf6996aa23820831055001a  # v2.23.3
         env:
           CIBW_BUILD: ${{ env.CIBW_BUILD }}
           CIBW_ARCHS_LINUX: "native"
@@ -122,7 +126,7 @@ jobs:
           cuda-version: ${{ inputs.cuda-version }}
 
       - name: Build cuda.bindings wheel
-        uses: pypa/cibuildwheel@v2.22.0
+        uses: pypa/cibuildwheel@faf86a6ed7efa889faf6996aa23820831055001a  # v2.23.3
         env:
           CIBW_BUILD: ${{ env.CIBW_BUILD }}
           CIBW_ARCHS_LINUX: "native"
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index dd26b8cd5..9cd892c2f 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -1,3 +1,7 @@
+# Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
+#
+# SPDX-License-Identifier: Apache-2.0
+
 # Note: This name is referred to in the test job, so make sure any changes are sync'd up!
 # Further this is referencing a run in the backport branch to fetch old bindings.
 name: "CI"

From 959321a1fa67462141e1bee0cedd12d00f8ea272 Mon Sep 17 00:00:00 2001
From: "Marcus D. Hanwell" <mhanwell@nvidia.com>
Date: Tue, 6 May 2025 16:40:34 -0400
Subject: [PATCH 19/19] Move the build and tests to CUDA 12.9.0

---
 .github/workflows/test-wheel-linux.yml   | 34 ++++++++++++------------
 .github/workflows/test-wheel-windows.yml |  8 +++---
 ci/versions.json                         |  2 +-
 3 files changed, 22 insertions(+), 22 deletions(-)

diff --git a/.github/workflows/test-wheel-linux.yml b/.github/workflows/test-wheel-linux.yml
index bcc24f459..5b6fb43e5 100644
--- a/.github/workflows/test-wheel-linux.yml
+++ b/.github/workflows/test-wheel-linux.yml
@@ -51,56 +51,56 @@ jobs:
           # Add a special entry for the H100 runner on amd64.
           special_runner=""
           if [[ "${ARCH}" == "amd64" ]]; then
-            special_runner="- { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: 'H100', DRIVER: 'latest' }"
+            special_runner="- { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.9.0', LOCAL_CTK: '1', GPU: 'H100', DRIVER: 'latest' }"
           fi
 
           # Please keep the matrices sorted in ascending order by the following:
           #
           #     [PY_VER, CUDA_VER, LOCAL_CTK, GPU, DRIVER]
           #
-          # Note that DRIVER: `earliest` does not work with CUDA 12.8.0 and LOCAL_CTK: 0 does not work with CUDA 12.0.1.
+          # Note that DRIVER: `earliest` does not work with CUDA 12.9.0 and LOCAL_CTK: 0 does not work with CUDA 12.0.1.
           #
           export MATRICES="
             pull-request:
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.9.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'earliest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.9.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.9.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.9.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.9.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
               ${special_runner}
             nightly:
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.9.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.9',  CUDA_VER: '12.9.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.9.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.10', CUDA_VER: '12.9.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.9.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.11', CUDA_VER: '12.9.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.9.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12', CUDA_VER: '12.9.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'earliest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '11.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.0.1', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
-              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.8.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.9.0', LOCAL_CTK: '0', GPU: ${gpu}, DRIVER: 'latest' }
+              - { ARCH: ${ARCH}, PY_VER: '3.13', CUDA_VER: '12.9.0', LOCAL_CTK: '1', GPU: ${gpu}, DRIVER: 'latest' }
               ${special_runner}
           "
 
diff --git a/.github/workflows/test-wheel-windows.yml b/.github/workflows/test-wheel-windows.yml
index 8f694e5dd..818fb903f 100644
--- a/.github/workflows/test-wheel-windows.yml
+++ b/.github/workflows/test-wheel-windows.yml
@@ -49,13 +49,13 @@ jobs:
             pull-request:
               - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '11.8.0', LOCAL_CTK: '0' }
               - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '11.8.0', LOCAL_CTK: '1' }
-              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '12.8.0', LOCAL_CTK: '0' }
-              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '12.8.0', LOCAL_CTK: '1' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '12.9.0', LOCAL_CTK: '0' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '12.9.0', LOCAL_CTK: '1' }
             nightly:
               - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '11.8.0', LOCAL_CTK: '0' }
               - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '11.8.0', LOCAL_CTK: '1' }
-              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '12.8.0', LOCAL_CTK: '0' }
-              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '12.8.0', LOCAL_CTK: '1' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '12.9.0', LOCAL_CTK: '0' }
+              - { ARCH: ${ARCH}, PY_VER: '3.12',  CUDA_VER: '12.9.0', LOCAL_CTK: '1' }
           "
 
           # Use the nightly matrix for branch tests
diff --git a/ci/versions.json b/ci/versions.json
index eb1e19f2f..5608eeb1d 100644
--- a/ci/versions.json
+++ b/ci/versions.json
@@ -1,7 +1,7 @@
 {
   "cuda": {
     "build": {
-      "version": "12.8.0"
+      "version": "12.9.0"
     }
   }
 }