Test cuda_core examples

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

@@ -100,8 +100,8 @@ jobs:
         uses: ./.github/actions/install_unix_deps
         continue-on-error: false
         with:
-          # for artifact fetching, graphics libs
-          dependencies: "jq wget libgl1 libegl1"
+          # for artifact fetching, graphics libs, g++ required for cffi in example
+          dependencies: "jq wget libgl1 libegl1 g++"
           dependent_exes: "jq wget"
 
       - name: Set environment variables

cuda_core/examples/cuda_graphs.py

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # SPDX-License-Identifier: Apache-2.0
 
@@ -10,6 +10,10 @@
 #
 # ################################################################################
 
+# /// script
+# dependencies = ["cuda_bindings", "cuda_core", "nvidia-cuda-nvrtc", "cupy-cuda13x"]
+# ///
+
 import sys
 import time
 
@@ -121,6 +125,9 @@ def main():
         end_time = time.time()
 
         graph_execution_time = end_time - start_time
+        if graph_execution_time == 0.0:
+            print("Graph execution time is too fast to measure accurately.")
+            graph_execution_time = 1e-9  # Assign a small value to avoid division by zero in speedup calculation
         print(f"Graph execution time: {graph_execution_time:.6f} seconds")
 
         # Verify results

cuda_core/examples/gl_interop_plasma.py

@@ -53,9 +53,10 @@
 # effect popular in the demoscene). The window title shows the current FPS.
 # Close the window or press Escape to exit.
 #
-# Requirements
-# ============
-#   pip install pyglet
+
+# /// script
+# dependencies = ["cuda_bindings", "cuda_core>0.6.0", "pyglet"]
+# ///
 
 import ctypes
 import sys

cuda_core/examples/jit_lto_fractal.py

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # SPDX-License-Identifier: Apache-2.0
 
@@ -12,6 +12,10 @@
 #
 # ################################################################################
 
+# /// script
+# dependencies = ["cuda_bindings", "cuda_core", "nvidia-cuda-nvrtc", "cupy-cuda13x"]
+# ///
+
 import argparse
 import sys
 

cuda_core/examples/memory_ops.py

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # SPDX-License-Identifier: Apache-2.0
 
@@ -10,6 +10,10 @@
 #
 # ################################################################################
 
+# /// script
+# dependencies = ["cuda_bindings", "cuda_core", "nvidia-cuda-nvrtc", "cupy-cuda13x"]
+# ///
+
 import sys
 
 import cupy as cp

cuda_core/examples/pytorch_example.py

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # SPDX-License-Identifier: Apache-2.0
 
@@ -9,6 +9,10 @@
 #
 # ################################################################################
 
+# /// script
+# dependencies = ["cuda_bindings", "cuda_core", "torch"]
+# ///
+
 import sys
 
 import torch

cuda_core/examples/saxpy.py

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # SPDX-License-Identifier: Apache-2.0
 
@@ -10,8 +10,17 @@
 #
 # ################################################################################
 
+# /// script
+# dependencies = ["cuda_bindings", "cuda_core", "nvidia-cuda-nvrtc", "cupy-cuda13x"]
+# ///
+
+
 import sys
 
+from cuda import pathfinder
+
+print(pathfinder.load_nvidia_dynamic_lib("nvrtc"))
+
 import cupy as cp
 
 from cuda.core import Device, LaunchConfig, Program, ProgramOptions, launch

cuda_core/examples/show_device_properties.py

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # SPDX-License-Identifier: Apache-2.0
 
@@ -9,6 +9,10 @@
 #
 # ################################################################################
 
+# /// script
+# dependencies = ["cuda_bindings", "cuda_core"]
+# ///
+
 import sys
 
 from cuda.core import Device, system

cuda_core/examples/simple_multi_gpu_example.py

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # SPDX-License-Identifier: Apache-2.0
 
@@ -9,6 +9,10 @@
 #
 # ################################################################################
 
+# /// script
+# dependencies = ["cuda_bindings", "cuda_core", "cupy-cuda13x"]
+# ///
+
 import sys
 
 import cupy as cp

cuda_core/examples/strided_memory_view_cpu.py

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # SPDX-License-Identifier: Apache-2.0
 
@@ -9,6 +9,10 @@
 #
 # ################################################################################
 
+# /// script
+# dependencies = ["cuda_bindings", "cuda_core", "cffi", "setuptools"]
+# ///
+
 import importlib
 import string
 import sys

cuda_core/examples/strided_memory_view_gpu.py

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # SPDX-License-Identifier: Apache-2.0
 
@@ -9,6 +9,10 @@
 #
 # ################################################################################
 
+# /// script
+# dependencies = ["cuda_bindings", "cuda_core", "nvidia-cuda-nvrtc", "cupy-cuda13x"]
+# ///
+
 import string
 import sys
 

cuda_core/examples/thread_block_cluster.py

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # SPDX-License-Identifier: Apache-2.0
 
@@ -10,6 +10,10 @@
 #
 # ################################################################################
 
+# /// script
+# dependencies = ["cuda_bindings", "cuda_core"]
+# ///
+
 import os
 import sys
 

cuda_core/examples/tma_tensor_map.py

@@ -22,6 +22,10 @@
 #
 # ################################################################################
 
+# /// script
+# dependencies = ["cuda_bindings", "cuda_core>0.6.0", "cupy-cuda13x"]
+# ///
+
 import os
 import sys
 

cuda_core/examples/vector_add.py

@@ -1,4 +1,4 @@
-# SPDX-FileCopyrightText: Copyright (c) 2024-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 #
 # SPDX-License-Identifier: Apache-2.0
 
@@ -9,6 +9,10 @@
 #
 # ################################################################################
 
+# /// script
+# dependencies = ["cuda_bindings", "cuda_core", "nvidia-cuda-nvrtc", "cupy-cuda13x"]
+# ///
+
 import cupy as cp
 
 from cuda.core import Device, LaunchConfig, Program, ProgramOptions, launch

cuda_core/pyproject.toml

@@ -56,7 +56,7 @@ cu12 = ["cuda-bindings[all]==12.*"]
 cu13 = ["cuda-bindings[all]==13.*"]
 
 [dependency-groups]
-test = ["cython>=3.2,<3.3", "setuptools", "pytest>=6.2.4", "pytest-benchmark", "pytest-randomly", "pytest-repeat", "pytest-rerunfailures", "cloudpickle", "psutil"]
+test = ["cython>=3.2,<3.3", "setuptools", "pytest>=6.2.4", "pytest-benchmark", "pytest-randomly", "pytest-repeat", "pytest-rerunfailures", "cloudpickle", "psutil", "cffi"]
 ml-dtypes = ["ml-dtypes>=0.5.4,<0.6.0"]
 test-cu12 = [ {include-group = "ml-dtypes" }, {include-group = "test" }, "cupy-cuda12x; python_version < '3.14'", "cuda-toolkit[cudart]==12.*"]  # runtime headers needed by CuPy
 test-cu13 = [ {include-group = "ml-dtypes" }, {include-group = "test" }, "cupy-cuda13x; python_version < '3.14'", "cuda-toolkit[cudart]==13.*"]  # runtime headers needed by CuPy

cuda_core/tests/example_tests/test_basic_examples.py

@@ -1,24 +1,109 @@
-# SPDX-FileCopyrightText: Copyright (c) 2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-FileCopyrightText: Copyright (c) 2024-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
 # SPDX-License-Identifier: Apache-2.0
 
 # If we have subcategories of examples in the future, this file can be split along those lines
 
 import glob
+import importlib.metadata
 import os
+import platform
+import re
+import subprocess
+import sys
 
 import pytest
 
-from cuda.core import Device
+from cuda.core import Device, system
+
+
+def has_compute_capability_9_or_higher() -> bool:
+    return Device().compute_capability >= (9, 0)
+
+
+def has_multiple_devices() -> bool:
+    return system.get_num_devices() >= 2
+
+
+def has_display() -> bool:
+    # We assume that we don't want to open any windows during testing,
+    # so we always return False
+    return False
+
+
+def is_not_windows() -> bool:
+    return sys.platform != "win32"
+
+
+def is_x86_64() -> bool:
+    return platform.machine() == "x86_64"
+
+
+def has_cuda_path() -> bool:
+    return os.environ.get("CUDA_PATH", os.environ.get("CUDA_HOME")) is not None
+
+
+# Specific system requirements for each of the examples.
+
+
+SYSTEM_REQUIREMENTS = {
+    "gl_interop_plasma.py": has_display,
+    "pytorch_example.py": lambda: (
+        has_compute_capability_9_or_higher() and is_x86_64()
+    ),  # PyTorch only provides CUDA support for x86_64
+    "saxpy.py": has_compute_capability_9_or_higher,
+    "simple_multi_gpu_example.py": has_multiple_devices,
+    "strided_memory_view_cpu.py": is_not_windows,
+    "thread_block_cluster.py": lambda: has_compute_capability_9_or_higher() and has_cuda_path(),
+    "tma_tensor_map.py": has_cuda_path,
+}
 
-from .utils import run_example
 
 samples_path = os.path.join(os.path.dirname(__file__), "..", "..", "examples")
-sample_files = glob.glob(samples_path + "**/*.py", recursive=True)
+sample_files = [os.path.basename(x) for x in glob.glob(samples_path + "**/*.py", recursive=True)]
+
+
+def has_package_requirements_or_skip(example):
+    with open(example, encoding="utf-8") as f:
+        content = f.read()
+
+    # The canonical regex as defined in PEP 723
+    pep723 = re.search(r"(?m)^# /// (?P<type>[a-zA-Z0-9-]+)$\s(?P<content>(^#(| .*)$\s)+)^# ///$", content)
+    if not pep723:
+        return
+
+    metadata = {}
+    for line in pep723.group("content").splitlines():
+        line = line.lstrip("# ").rstrip()
+        if not line:
+            continue
+        key, value = line.split("=", 1)
+        key = key.strip()
+        value = value.strip()
+        metadata[key] = value
+
+    if "dependencies" in metadata:
+        dependencies = eval(metadata["dependencies"])  # noqa: S307
+        for dependency in dependencies:
+            name = re.match("[a-zA-Z0-9_-]+", dependency)
+            try:
+                importlib.metadata.distribution(name.string)
+            except importlib.metadata.PackageNotFoundError:
+                pytest.skip(f"Skipping {example} due to missing package requirement: {name}")
 
 
 @pytest.mark.parametrize("example", sample_files)
-class TestExamples:
-    def test_example(self, example, deinit_cuda):
-        run_example(samples_path, example)
-        if Device().device_id != 0:
-            Device(0).set_current()
+def test_example(example):
+    example_path = os.path.join(samples_path, example)
+    has_package_requirements_or_skip(example_path)
+
+    system_requirement = SYSTEM_REQUIREMENTS.get(example, lambda: True)
+    if not system_requirement():
+        pytest.skip(f"Skipping {example} due to unmet system requirement")
+
+    process = subprocess.run([sys.executable, example_path], capture_output=True)  # noqa: S603
+    if process.returncode != 0:
+        if process.stdout:
+            print(process.stdout.decode())
+        if process.stderr:
+            print(process.stderr.decode(), file=sys.stderr)
+        raise AssertionError(f"`{example}` failed ({process.returncode})")