Add some basic test coverage for cuda.core.experimental (#153)

Add basic test coverage for cuda.core.experimental

cuda_core/cuda/core/experimental/_program.py

@@ -14,9 +14,9 @@ class Program:
     _supported_target_type = ("ptx", "cubin", "ltoir", )
 
     def __init__(self, code, code_type):
+        self._handle = None
         if code_type not in self._supported_code_type:
             raise NotImplementedError
-        self._handle = None
 
         if code_type.lower() == "c++":
             if not isinstance(code, str):

cuda_core/tests/conftest.py

@@ -0,0 +1,16 @@
+# Copyright 2024 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
+# this software. Any use, reproduction, disclosure, or distribution of
+# this software and related documentation outside the terms of the EULA
+# is strictly prohibited.
+
+from cuda.core.experimental._device import Device
+import pytest
+
+@pytest.fixture(scope="module")
+def init_cuda():
+    device = Device()
+    device.set_current()
+

cuda_core/tests/example_tests/test_basic_examples.py

@@ -0,0 +1,25 @@
+# Copyright 2024 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
+# this software. Any use, reproduction, disclosure, or distribution of
+# this software and related documentation outside the terms of the EULA
+# is strictly prohibited.
+
+# If we have subcategories of examples in the future, this file can be split along those lines
+
+from .utils import run_example
+import os
+import glob
+import pytest
+
+samples_path = os.path.join(
+    os.path.dirname(__file__), '..', '..', 'examples')
+sample_files = glob.glob(samples_path+'**/*.py', recursive=True)
+@pytest.mark.parametrize(
+    'example', sample_files
+)
+class TestExamples:
+    def test_example(self, example):
+        filename = os.path.basename(example)
+        run_example(samples_path, example)

cuda_core/tests/example_tests/utils.py

@@ -0,0 +1,54 @@
+# Copyright 2024 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
+# this software. Any use, reproduction, disclosure, or distribution of
+# this software and related documentation outside the terms of the EULA
+# is strictly prohibited.
+
+from cuda import cuda
+import gc
+import os
+import sys
+import pytest
+import cupy as cp
+
+class SampleTestError(Exception):
+    pass
+
+def parse_python_script(filepath):
+    if not filepath.endswith('.py'):
+        raise ValueError(f"{filepath} not supported")
+    with open(filepath, "r", encoding='utf-8') as f:
+        script = f.read()
+    return script
+
+
+def run_example(samples_path, filename, env=None):
+    fullpath = os.path.join(samples_path, filename)
+    script = parse_python_script(fullpath)
+    try:
+        old_argv = sys.argv
+        sys.argv = [fullpath]
+        old_sys_path = sys.path.copy()
+        sys.path.append(samples_path)
+        exec(script, env if env else {})
+    except ImportError as e:
+        # for samples requiring any of optional dependencies
+        for m in ('cupy',):
+            if f"No module named '{m}'" in str(e):
+                pytest.skip(f'{m} not installed, skipping related tests')
+                break
+        else:
+            raise
+    except Exception as e:
+            msg = "\n"
+            msg += f'Got error ({filename}):\n'
+            msg += str(e)
+            raise SampleTestError(msg) from e
+    finally:
+        sys.path = old_sys_path
+        sys.argv = old_argv
+        # further reduce the memory watermark
+        gc.collect()
+        cp.get_default_memory_pool().free_all_blocks()

cuda_core/tests/test_device.py

@@ -0,0 +1,67 @@
+# Copyright 2024 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
+# this software. Any use, reproduction, disclosure, or distribution of
+# this software and related documentation outside the terms of the EULA
+# is strictly prohibited.
+
+from cuda import cuda, cudart
+from cuda.core.experimental._device import Device
+from cuda.core.experimental._utils import handle_return, ComputeCapability, CUDAError, \
+                             precondition
+import pytest
+
+def test_device_repr():
+    device = Device(0)
+    assert str(device).startswith('<Device 0')
+
+def test_device_alloc(init_cuda):
+    device = Device()
+    buffer = device.allocate(1024)
+    device.sync()
+    assert buffer.handle != 0
+    assert buffer.size == 1024
+    assert buffer.device_id == 0
+
+def test_device_set_current():
+    device = Device()
+    device.set_current()
+
+def test_device_create_stream():
+    device = Device()
+    stream = device.create_stream()
+    assert stream is not None
+    assert stream.handle
+
+def test_pci_bus_id():
+    device = Device()
+    bus_id = handle_return(cudart.cudaDeviceGetPCIBusId(13, device.device_id))
+    assert device.pci_bus_id == bus_id[:12].decode()
+
+def test_uuid():
+    device = Device()
+    driver_ver = handle_return(cuda.cuDriverGetVersion())
+    if driver_ver >= 11040:
+        uuid = handle_return(cuda.cuDeviceGetUuid_v2(device.device_id))
+    else:
+        uuid = handle_return(cuda.cuDeviceGetUuid(device.device_id))
+    uuid = uuid.bytes.hex()
+    expected_uuid = f"{uuid[:8]}-{uuid[8:12]}-{uuid[12:16]}-{uuid[16:20]}-{uuid[20:]}"
+    assert device.uuid == expected_uuid
+
+def test_name():
+    device = Device()
+    name = handle_return(cuda.cuDeviceGetName(128, device.device_id))
+    name = name.split(b'\0')[0]
+    assert device.name == name.decode()
+
+def test_compute_capability():
+    device = Device()
+    major = handle_return(cudart.cudaDeviceGetAttribute(
+        cudart.cudaDeviceAttr.cudaDevAttrComputeCapabilityMajor, device.device_id))
+    minor = handle_return(cudart.cudaDeviceGetAttribute(
+        cudart.cudaDeviceAttr.cudaDevAttrComputeCapabilityMinor, device.device_id))
+    expected_cc = ComputeCapability(major, minor)
+    assert device.compute_capability == expected_cc
+

cuda_core/tests/test_event.py

@@ -0,0 +1,39 @@
+# Copyright 2024 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
+# this software. Any use, reproduction, disclosure, or distribution of
+# this software and related documentation outside the terms of the EULA
+# is strictly prohibited.
+
+from cuda import cuda
+from cuda.core.experimental._event import  EventOptions, Event
+from cuda.core.experimental._utils import handle_return
+from cuda.core.experimental._device import Device
+import pytest
+
+def test_is_timing_disabled():
+    options = EventOptions(enable_timing=False)
+    event = Event._init(options)
+    assert event.is_timing_disabled == True
+
+def test_is_sync_busy_waited():
+    options = EventOptions(busy_waited_sync=True)
+    event = Event._init(options)
+    assert event.is_sync_busy_waited == True
+
+def test_sync():
+    options = EventOptions()
+    event = Event._init(options)
+    event.sync()
+    assert event.is_done == True
+
+def test_is_done():
+    options = EventOptions()
+    event = Event._init(options)
+    assert event.is_done == True
+
+def test_handle():
+    options = EventOptions()
+    event = Event._init(options)
+    assert isinstance(event.handle, int)

cuda_core/tests/test_launcher.py

@@ -0,0 +1,66 @@
+# Copyright 2024 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
+# this software. Any use, reproduction, disclosure, or distribution of
+# this software and related documentation outside the terms of the EULA
+# is strictly prohibited.
+
+from cuda import cuda
+from cuda.core.experimental._launcher import LaunchConfig
+from cuda.core.experimental._stream import Stream
+from cuda.core.experimental._device import Device
+from cuda.core.experimental._utils import handle_return
+import pytest
+
+def test_launch_config_init():
+    config = LaunchConfig(grid=(1, 1, 1), block=(1, 1, 1), stream=None, shmem_size=0)
+    assert config.grid == (1, 1, 1)
+    assert config.block == (1, 1, 1)
+    assert config.stream is None
+    assert config.shmem_size == 0
+
+    config = LaunchConfig(grid=(2, 2, 2), block=(2, 2, 2), stream=Device().create_stream(), shmem_size=1024)
+    assert config.grid == (2, 2, 2)
+    assert config.block == (2, 2, 2)
+    assert isinstance(config.stream, Stream)
+    assert config.shmem_size == 1024
+
+def test_launch_config_cast_to_3_tuple():
+    config = LaunchConfig(grid=1, block=1)
+    assert config._cast_to_3_tuple(1) == (1, 1, 1)
+    assert config._cast_to_3_tuple((1, 2)) == (1, 2, 1)
+    assert config._cast_to_3_tuple((1, 2, 3)) == (1, 2, 3)
+
+    # Edge cases
+    assert config._cast_to_3_tuple(999) == (999, 1, 1)
+    assert config._cast_to_3_tuple((999, 888)) == (999, 888, 1)
+    assert config._cast_to_3_tuple((999, 888, 777)) == (999, 888, 777)
+
+def test_launch_config_invalid_values():
+    with pytest.raises(ValueError):
+        LaunchConfig(grid=0, block=1)
+
+    with pytest.raises(ValueError):
+        LaunchConfig(grid=(0, 1), block=1)
+
+    with pytest.raises(ValueError):
+        LaunchConfig(grid=(1, 1, 1), block=0)
+
+    with pytest.raises(ValueError):
+        LaunchConfig(grid=(1, 1, 1), block=(0, 1))
+
+def test_launch_config_stream():
+    stream = Device().create_stream()
+    config = LaunchConfig(grid=(1, 1, 1), block=(1, 1, 1), stream=stream, shmem_size=0)
+    assert config.stream == stream
+
+    with pytest.raises(ValueError):
+        LaunchConfig(grid=(1, 1, 1), block=(1, 1, 1), stream="invalid_stream", shmem_size=0)
+
+def test_launch_config_shmem_size():
+    config = LaunchConfig(grid=(1, 1, 1), block=(1, 1, 1), stream=None, shmem_size=2048)
+    assert config.shmem_size == 2048
+
+    config = LaunchConfig(grid=(1, 1, 1), block=(1, 1, 1), stream=None)
+    assert config.shmem_size == 0