Merge branch 'main' into reduce_cudart

Author: leofang

.github/workflows/build-and-test.yml

@@ -139,9 +139,11 @@ jobs:
           # CIBW mounts the host filesystem under /host
           CIBW_ENVIRONMENT_LINUX: >
             CUDA_PATH=/host/${{ env.CUDA_PATH }}
+            LIBRARY_PATH=/host/${{ env.CUDA_PATH }}/lib
             CUDA_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/

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

@@ -62,8 +62,16 @@ jobs:
           TEST_CUDA_MAJOR="$(cut -d '.' -f 1 <<< ${{ inputs.cuda-version }})"
           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 }})"
+            if [[ $BUILD_CUDA_MINOR != $TEST_CUDA_MINOR ]]; then
+              SKIP_CUDA_CORE_CYTHON_TEST=1
+            else
+              SKIP_CUDA_CORE_CYTHON_TEST=0
+            fi
           fi
 
           # make outputs from the previous job as env vars
@@ -77,6 +85,7 @@ jobs:
           echo "CUDA_BINDINGS_ARTIFACT_NAME=${CUDA_BINDINGS_ARTIFACT_BASENAME}-${{ github.sha }}" >> $GITHUB_ENV
           echo "CUDA_BINDINGS_ARTIFACTS_DIR=$(realpath "$REPO_DIR/cuda_bindings/dist")" >> $GITHUB_ENV
           echo "SKIP_CUDA_BINDINGS_TEST=${SKIP_CUDA_BINDINGS_TEST}" >> $GITHUB_ENV
+          echo "SKIP_CUDA_CORE_CYTHON_TEST=${SKIP_CUDA_CORE_CYTHON_TEST}" >> $GITHUB_ENV
 
       - name: Install dependencies
         uses: ./.github/actions/install_unix_deps
@@ -185,7 +194,7 @@ jobs:
 
           pushd ./cuda_bindings
           pip install -r requirements.txt
-          pytest -rxXs tests/
+          pytest -rxXs -v tests/
 
           # It is a bit convoluted to run the Cython tests against CTK wheels,
           # so let's just skip them.
@@ -196,9 +205,9 @@ jobs:
               # TODO: enable this once win-64 runners are up 
               exit 1
             fi  
-            pytest -rxXs tests/cython
-            popd
+            pytest -rxXs -v tests/cython
           fi
+          popd
 
       - name: Run cuda.core tests
         run: |
@@ -220,7 +229,22 @@ jobs:
 
           pushd ./cuda_core
           pip install -r "tests/requirements-cu${TEST_CUDA_MAJOR}.txt"
-          pytest -rxXs tests/
+          pytest -rxXs -v tests/
+
+          # It is a bit convoluted to run the Cython tests against CTK wheels,
+          # so let's just skip them. Also, currently our CI always installs the
+          # latest bindings (from either major version). This is not compatible
+          # with the test requirements.
+          if [[ "${{ inputs.local-ctk }}" == 1 && "${SKIP_CUDA_CORE_CYTHON_TEST}" == 0 ]]; then
+            pip install cython setuptools  # setuptools needed starting PY312
+            if [[ "${{ inputs.host-platform }}" == linux* ]]; then
+              bash tests/cython/build_tests.sh
+            elif [[ "${{ inputs.host-platform }}" == win* ]]; then
+              # TODO: enable this once win-64 runners are up
+              exit 1
+            fi
+            pytest -rxXs -v tests/cython
+          fi
           popd
 
       - name: Ensure cuda-python installable

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

@@ -186,8 +186,8 @@ jobs:
 
           Push-Location ./cuda_bindings
           pip install -r requirements.txt
-          pytest -rxXs tests/
-          # skip Cython tests for now
+          pytest -rxXs -v tests/
+          # skip Cython tests for now (NVIDIA/cuda-python#466)
           Pop-Location
 
       - name: Run cuda.core tests
@@ -210,7 +210,7 @@ jobs:
 
           Push-Location ./cuda_core
           pip install -r "tests/requirements-cu${TEST_CUDA_MAJOR}.txt"
-          pytest -rxXs tests/
+          pytest -rxXs -v tests/
           Pop-Location
 
       - name: Ensure cuda-python installable

.gitignore

@@ -17,6 +17,11 @@ __pycache__/
 # CUDA Python specific (auto-generated)
 cuda_bindings/cuda/bindings/_bindings/cydriver.pxd
 cuda_bindings/cuda/bindings/_bindings/cydriver.pyx
+cuda_bindings/cuda/bindings/_bindings/cyruntime.pxi
+cuda_bindings/cuda/bindings/_bindings/cyruntime.pxd
+cuda_bindings/cuda/bindings/_bindings/cyruntime.pyx
+cuda_bindings/cuda/bindings/_bindings/cyruntime_ptds.pxd
+cuda_bindings/cuda/bindings/_bindings/cyruntime_ptds.pyx
 cuda_bindings/cuda/bindings/_bindings/cynvrtc.pxd
 cuda_bindings/cuda/bindings/_bindings/cynvrtc.pyx
 cuda_bindings/cuda/bindings/_internal/nvjitlink.pyx

README.md

@@ -4,8 +4,8 @@ CUDA Python is the home for accessing NVIDIA’s CUDA platform from Python. It c
 
 * [cuda.core](https://nvidia.github.io/cuda-python/cuda-core/latest): Pythonic access to CUDA Runtime and other core functionalities
 * [cuda.bindings](https://nvidia.github.io/cuda-python/cuda-bindings/latest): Low-level Python bindings to CUDA C APIs
-* [cuda.cooperative](https://nvidia.github.io/cccl/cuda_cooperative/): Pythonic exposure of CUB cooperative algorithms
-* [cuda.parallel](https://nvidia.github.io/cccl/cuda_parallel/): Pythonic exposure of Thrust parallel algorithms
+* [cuda.cooperative](https://nvidia.github.io/cccl/cuda_cooperative/): A Python package for easy access to highly efficient and customizable parallel algorithms, like `sort`, `scan`, `reduce`, `transform`, etc.
+* [cuda.parallel](https://nvidia.github.io/cccl/cuda_parallel/): A Python package providing CUB's reusable block-wide and warp-wide primitives for use within Numba CUDA kernels
 
 For access to NVIDIA CPU & GPU Math Libraries, please refer to [nvmath-python](https://docs.nvidia.com/cuda/nvmath-python/latest).
 

cuda_bindings/MANIFEST.in

@@ -2,3 +2,4 @@ recursive-include cuda/ *.pyx *.pxd
 # at least with setuptools 75.0.0 this folder was added erroneously
 # to the payload, causing file copying to the build environment failed
 exclude cuda/bindings cuda?bindings
+exclude cuda/bindings/_bindings cuda?bindings?_bindings

cuda_bindings/README.md

@@ -95,10 +95,10 @@ To run these samples:
 * `python -m pytest tests/cython/` against editable installations
 * `pytest tests/cython/` against installed packages
 
-### Benchmark (WIP)
+### Benchmark
 
-Benchmarks were used for performance analysis during initial release of CUDA Python. Today they need to be updated the 12.x toolkit and are work in progress.
+Allows for analyzing binding performance using plugin [pytest-benchmark](https://github.com/ionelmc/pytest-benchmark).
 
-The intended way to run these benchmarks was:
-* `python -m pytest --benchmark-only benchmark/` against editable installations
-* `pytest --benchmark-only benchmark/` against installed packages
+To run these benchmarks:
+* `python -m pytest --benchmark-only benchmarks/` against editable installations
+* `pytest --benchmark-only benchmarks/` against installed packages

cuda_bindings/benchmarks/__init__.py

@@ -1,4 +1,4 @@
-# Copyright 2021-2024 NVIDIA Corporation.  All rights reserved.
+# Copyright 2021-2025 NVIDIA Corporation.  All rights reserved.
 #
 # Please refer to the NVIDIA end user license agreement (EULA) associated
 # with this source code for terms and conditions that govern your use of
@@ -25,7 +25,7 @@ def ASSERT_DRV(err):
         raise RuntimeError(f"Unknown error type: {err}")
 
 
-@pytest.fixture
+@pytest.fixture(scope="function")
 def init_cuda():
     # Initialize
     (err,) = cuda.cuInit(0)
@@ -47,7 +47,7 @@ def init_cuda():
     ASSERT_DRV(err)
 
 
-@pytest.fixture
+@pytest.fixture(scope="function")
 def load_module():
     module = None
 

cuda_bindings/benchmarks/perf_test_utils.py renamed to cuda_bindings/benchmarks/conftest.py

@@ -1,4 +1,4 @@
-# Copyright 2021-2024 NVIDIA Corporation.  All rights reserved.
+# Copyright 2021-2025 NVIDIA Corporation.  All rights reserved.
 #
 # Please refer to the NVIDIA end user license agreement (EULA) associated
 # with this source code for terms and conditions that govern your use of