Merge remote-tracking branch 'upstream/main' into cuda.core.system-events

Author: mdboom

.github/workflows/ci.yml

@@ -40,6 +40,7 @@ jobs:
     runs-on: ubuntu-latest
     outputs:
       skip: ${{ steps.get-should-skip.outputs.skip }}
+      doc-only: ${{ steps.get-should-skip.outputs.doc_only }}
     steps:
       - name: Checkout repository
         uses: actions/checkout@8e8c483db84b4bee98b60c0593521ed34d9990e8  # v6.0.1
@@ -50,11 +51,16 @@ jobs:
         run: |
           set -euxo pipefail
           if ${{ startsWith(github.ref_name, 'pull-request/') }}; then
-            skip="$(gh pr view "$(grep -Po '(\d+)$' <<< '${{ github.ref_name }}')" --json title --jq '.title | contains("[no-ci]")')"
+            pr_number="$(grep -Po '(\d+)$' <<< '${{ github.ref_name }}')"
+            pr_title="$(gh pr view "${pr_number}" --json title --jq '.title')"
+            skip="$(echo "${pr_title}" | grep -q '\[no-ci\]' && echo true || echo false)"
+            doc_only="$(echo "${pr_title}" | grep -q '\[doc-only\]' && echo true || echo false)"
           else
             skip=false
+            doc_only=false
           fi
           echo "skip=${skip}" >> "$GITHUB_OUTPUT"
+          echo "doc_only=${doc_only}" >> "$GITHUB_OUTPUT"
 
   # WARNING: make sure all of the build jobs are in sync
   build-linux-64:
@@ -86,7 +92,7 @@ jobs:
         host-platform:
           - linux-aarch64
     name: Build ${{ matrix.host-platform }}, CUDA ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
-    if: ${{ github.repository_owner == 'nvidia' && !fromJSON(needs.should-skip.outputs.skip) }}
+    if: ${{ github.repository_owner == 'nvidia' && !fromJSON(needs.should-skip.outputs.skip) && !fromJSON(needs.should-skip.outputs.doc-only) }}
     secrets: inherit
     uses: ./.github/workflows/build-wheel.yml
     with:
@@ -105,7 +111,7 @@ jobs:
         host-platform:
           - win-64
     name: Build ${{ matrix.host-platform }}, CUDA ${{ needs.ci-vars.outputs.CUDA_BUILD_VER }}
-    if: ${{ github.repository_owner == 'nvidia' && !fromJSON(needs.should-skip.outputs.skip) }}
+    if: ${{ github.repository_owner == 'nvidia' && !fromJSON(needs.should-skip.outputs.skip) && !fromJSON(needs.should-skip.outputs.doc-only) }}
     secrets: inherit
     uses: ./.github/workflows/build-wheel.yml
     with:
@@ -121,11 +127,12 @@ jobs:
         host-platform:
           - linux-64
     name: Test ${{ matrix.host-platform }}
-    if: ${{ github.repository_owner == 'nvidia' }}
+    if: ${{ github.repository_owner == 'nvidia' && !fromJSON(needs.should-skip.outputs.doc-only) }}
     permissions:
       contents: read  # This is required for actions/checkout
     needs:
       - ci-vars
+      - should-skip
       - build-linux-64
     secrets: inherit
     uses: ./.github/workflows/test-wheel-linux.yml
@@ -142,6 +149,8 @@ jobs:
         host-platform:
           - linux-aarch64
     name: Test ${{ matrix.host-platform }}
+    # Note: No doc-only check needed here - if build-linux-aarch64 is skipped,
+    # this job is automatically skipped due to the dependency.
     if: ${{ github.repository_owner == 'nvidia' }}
     permissions:
       contents: read  # This is required for actions/checkout
@@ -162,6 +171,8 @@ jobs:
         host-platform:
           - win-64
     name: Test ${{ matrix.host-platform }}
+    # Note: No doc-only check needed here - if build-windows is skipped,
+    # this job is automatically skipped due to the dependency.
     if: ${{ github.repository_owner == 'nvidia' }}
     permissions:
       contents: read  # This is required for actions/checkout
@@ -196,6 +207,7 @@ jobs:
     if: always()
     runs-on: ubuntu-latest
     needs:
+      - should-skip
       - test-linux-64
       - test-linux-aarch64
       - test-windows
@@ -219,11 +231,18 @@ jobs:
           # failing job(s) will timeout causing a cancellation here and the
           # build to succeed which we don't want (originally this was just
           # 'exit 0')
-          if ${{ needs.test-linux-64.result == 'cancelled' ||
-                 needs.test-linux-aarch64.result == 'cancelled' ||
-                 needs.test-windows.result == 'cancelled' ||
-                 needs.doc.result == 'cancelled' }}; then
+          #
+          # Note: When [doc-only] is in PR title, test jobs are intentionally
+          # skipped and should not cause failure.
+          doc_only=${{ needs.should-skip.outputs.doc-only }}
+          if ${{ needs.doc.result == 'cancelled' }}; then
             exit 1
-          else
-            exit 0
           fi
+          if [[ "${doc_only}" != "true" ]]; then
+            if ${{ needs.test-linux-64.result == 'cancelled' ||
+                   needs.test-linux-aarch64.result == 'cancelled' ||
+                   needs.test-windows.result == 'cancelled' }}; then
+              exit 1
+            fi
+          fi
+          exit 0

.github/workflows/test-wheel-linux.yml

@@ -269,17 +269,15 @@ jobs:
       - name: Ensure cuda-python installable
         run: |
           if [[ "${{ matrix.LOCAL_CTK }}" == 1 ]]; then
-            pip install cuda_python*.whl
+            pip install --only-binary=:all: cuda_python*.whl
           else
-            pip install $(ls cuda_python*.whl)[all]
+            pip install --only-binary=:all: $(ls cuda_python*.whl)[all]
           fi
 
       - name: Install cuda.pathfinder extra wheels for testing
         run: |
           set -euo pipefail
           pushd cuda_pathfinder
-          # Install pathfinder from the pre-built wheel, since building from
-          # source won't work in this context (we don't have a git checkout).
           pip install --only-binary=:all: -v ./*.whl --group "test-cu${TEST_CUDA_MAJOR}"
           pip list
           popd

.github/workflows/test-wheel-windows.yml

@@ -236,16 +236,16 @@ jobs:
       - name: Ensure cuda-python installable
         run: |
           if ('${{ matrix.LOCAL_CTK }}' -eq '1') {
-            pip install (Get-ChildItem -Filter cuda_python*.whl).FullName
+            pip install --only-binary=:all: (Get-ChildItem -Filter cuda_python*.whl).FullName
           } else {
-            pip install "$((Get-ChildItem -Filter cuda_python*.whl).FullName)[all]"
+            pip install --only-binary=:all: "$((Get-ChildItem -Filter cuda_python*.whl).FullName)[all]"
           }
 
       - name: Install cuda.pathfinder extra wheels for testing
         shell: bash --noprofile --norc -xeuo pipefail {0}
         run: |
           pushd cuda_pathfinder
-          pip install --only-binary=:all: -v . --group "test-cu${TEST_CUDA_MAJOR}"
+          pip install --only-binary=:all: -v ./*.whl --group "test-cu${TEST_CUDA_MAJOR}"
           pip list
           popd
 

README.md

@@ -13,6 +13,7 @@ CUDA Python is the home for accessing NVIDIA’s CUDA platform from Python. It c
 * [nvshmem4py](https://docs.nvidia.com/nvshmem/api/api/language_bindings/python/index.html): Pythonic interface to the NVSHMEM library, enabling Python applications to leverage NVSHMEM's high-performance PGAS (Partitioned Global Address Space) programming model for GPU-accelerated computing
 * [Nsight Python](https://docs.nvidia.com/nsight-python/index.html): Python kernel profiling interface that automates performance analysis across multiple kernel configurations using NVIDIA Nsight Tools
 * [CUPTI Python](https://docs.nvidia.com/cupti-python/): Python APIs for creation of profiling tools that target CUDA Python applications via the CUDA Profiling Tools Interface (CUPTI)
+* [Accelerated Computing Hub](https://github.com/NVIDIA/accelerated-computing-hub): Open-source learning materials related to GPU computing. You will find user guides, tutorials, and other works freely available for all learners interested in GPU computing.
 
 CUDA Python is currently undergoing an overhaul to improve existing and introduce new components. All of the previously available functionality from the `cuda-python` package will continue to be available, please refer to the [cuda.bindings](https://nvidia.github.io/cuda-python/cuda-bindings/latest) documentation for installation guide and further detail.
 

cuda_bindings/tests/nvml/test_pynvml.py

@@ -136,13 +136,19 @@ def test_device_get_p2p_status(handles, index):
 
 def test_device_get_power_usage(ngpus, handles):
     for i in range(ngpus):
-        power_mwatts = nvml.device_get_power_usage(handles[i])
+        try:
+            power_mwatts = nvml.device_get_power_usage(handles[i])
+        except nvml.NotSupportedError:
+            pytest.skip("device_get_power_usage not supported")
         assert power_mwatts >= 0.0
 
 
 def test_device_get_total_energy_consumption(ngpus, handles):
     for i in range(ngpus):
-        energy_mjoules1 = nvml.device_get_total_energy_consumption(handles[i])
+        try:
+            energy_mjoules1 = nvml.device_get_total_energy_consumption(handles[i])
+        except nvml.NotSupportedError:
+            pytest.skip("device_get_total_energy_consumption not supported")
         for j in range(10):  # idle for 150 ms
             time.sleep(0.015)  # and check for increase every 15 ms
             energy_mjoules2 = nvml.device_get_total_energy_consumption(handles[i])