update tests

cuda_core/tests/test_context.py

@@ -0,0 +1,9 @@
+from cuda.core.experimental._context import Context
+
+def test_context_initialization():
+    try:
+        context = Context()
+    except NotImplementedError as e:
+        assert True
+    else:
+        assert False, "Expected NotImplementedError was not raised"

cuda_core/tests/test_device.py

@@ -0,0 +1,66 @@
+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
+
+@pytest.fixture(scope='module')
+def init_cuda():
+    Device().set_current()
+
+def test_device_initialization():
+    device = Device()
+    assert device is not None
+
+def test_device_repr():
+    device = Device()
+    assert str(device).startswith('<Device 0')
+
+def test_device_alloc():
+    device = Device()
+    device.set_current()
+    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
+
+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 +1,38 @@
+from cuda import cuda
+from cuda.core.experimental._event import  EventOptions, Event
+from cuda.core.experimental._utils import handle_return
+
+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_is_ipc_supported():
+    options = EventOptions(support_ipc=True)
+    try:
+        event = Event._init(options)
+    except NotImplementedError:
+        assert True
+    else:
+        assert False
+
+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 +1,77 @@
+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
+from cuda import cuda
+
+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():
+    try:
+        LaunchConfig(grid=0, block=1)
+    except ValueError:
+        assert True
+    else:
+        assert False
+
+    try:
+        LaunchConfig(grid=(0, 1), block=1)
+    except ValueError:
+        assert True
+    else:
+        assert False
+
+    try:
+        LaunchConfig(grid=(1, 1, 1), block=0)
+    except ValueError:
+        assert True
+    else:
+        assert False
+
+    try:
+        LaunchConfig(grid=(1, 1, 1), block=(0, 1))
+    except ValueError:
+        assert True
+    else:
+        assert False
+
+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
+
+    try:
+        LaunchConfig(grid=(1, 1, 1), block=(1, 1, 1), stream="invalid_stream", shmem_size=0)
+    except ValueError:
+        assert True
+    else:
+        assert False
+
+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

cuda_core/tests/test_memory.py

@@ -0,0 +1,199 @@
+# FILE: test_memory.py
+
+from cuda.core.experimental._memory import Buffer, MemoryResource
+from cuda.core.experimental._device import Device
+from cuda import cuda
+from cuda.core.experimental._utils import handle_return
+import ctypes
+
+@pytest.fixture(scope='module')
+def init_cuda():
+    Device().set_current()
+
+class DummyDeviceMemoryResource(MemoryResource):
+    def __init__(self, device):
+        self.device = device
+        pass
+
+    def allocate(self, size, stream=None) -> Buffer:
+        ptr = handle_return(cuda.cuMemAlloc(size))
+        return Buffer(ptr=ptr, size=size, mr=self)
+
+    def deallocate(self, ptr, size, stream=None):
+        cuda.cuMemFree(ptr)
+
+    @property
+    def is_device_accessible(self) -> bool:
+        return True
+
+    @property
+    def is_host_accessible(self) -> bool:
+        return False
+
+    @property
+    def device_id(self) -> int:
+        return 0
+
+class DummyHostMemoryResource(MemoryResource):
+    def __init__(self):
+        pass
+
+    def allocate(self, size, stream=None) -> Buffer:
+        # Allocate a ctypes buffer of size `size`
+        ptr = (ctypes.c_byte * size)()
+        return Buffer(ptr=ptr, size=size, mr=self)
+
+    def deallocate(self, ptr, size, stream=None):
+        #the memory is deallocated per the ctypes deallocation at garbage collection time
+        pass
+
+    @property
+    def is_device_accessible(self) -> bool:
+        return False
+
+    @property
+    def is_host_accessible(self) -> bool:
+        return True
+
+    @property
+    def device_id(self) -> int:
+        raise RuntimeError("the pinned memory resource is not bound to any GPU")
+
+class DummyUnifiedMemoryResource(MemoryResource):
+    def __init__(self, device):
+        self.device = device
+        pass
+
+    def allocate(self, size, stream=None) -> Buffer:
+        ptr = handle_return(cuda.cuMemAllocManaged(size, cuda.CUmemAttach_flags.CU_MEM_ATTACH_GLOBAL.value))
+        return Buffer(ptr=ptr, size=size, mr=self)
+
+    def deallocate(self, ptr, size, stream=None):
+        cuda.cuMemFree(ptr)
+
+    @property
+    def is_device_accessible(self) -> bool:
+        return True
+
+    @property
+    def is_host_accessible(self) -> bool:
+        return True
+
+    @property
+    def device_id(self) -> int:
+        return 0
+
+class DummyPinnedMemoryResource(MemoryResource):
+    def __init__(self, device):
+        self.device = device
+        pass
+
+    def allocate(self, size, stream=None) -> Buffer:
+        ptr = handle_return(cuda.cuMemAllocHost(size))
+        return Buffer(ptr=ptr, size=size, mr=self)
+
+    def deallocate(self, ptr, size, stream=None):
+        cuda.cuMemFreeHost(ptr)
+
+    @property
+    def is_device_accessible(self) -> bool:
+        return True
+
+    @property
+    def is_host_accessible(self) -> bool:
+        return True
+
+    @property
+    def device_id(self) -> int:
+        raise RuntimeError("the pinned memory resource is not bound to any GPU")
+
+def buffer_initialization(dummy_mr : MemoryResource):
+    buffer = dummy_mr.allocate(size=1024)
+    assert buffer.handle != 0
+    assert buffer.size == 1024
+    assert buffer.memory_resource == dummy_mr
+    assert buffer.is_device_accessible == dummy_mr.is_device_accessible
+    assert buffer.is_host_accessible == dummy_mr.is_host_accessible
+    dummy_mr.deallocate(buffer.handle, buffer.size)
+
+def test_buffer_initialization():
+    device = Device()
+    device.set_current()
+    buffer_initialization(DummyDeviceMemoryResource(device))
+    buffer_initialization(DummyHostMemoryResource())
+    buffer_initialization(DummyUnifiedMemoryResource(device))
+    buffer_initialization(DummyPinnedMemoryResource(device))
+
+def buffer_copy_to(dummy_mr : MemoryResource, device : Device, check = False):
+    src_buffer = dummy_mr.allocate(size=1024)
+    dst_buffer = dummy_mr.allocate(size=1024)
+    stream = device.create_stream()
+
+    if check:
+        src_ptr = ctypes.cast(src_buffer.handle, ctypes.POINTER(ctypes.c_byte))
+        for i in range(1024):
+            src_ptr[i] = ctypes.c_byte(i)
+
+    src_buffer.copy_to(dst_buffer, stream=stream)
+    device.sync()
+
+    if check:
+        dst_ptr = ctypes.cast(dst_buffer.handle, ctypes.POINTER(ctypes.c_byte))
+
+        for i in range(10):
+            assert dst_ptr[i] == src_ptr[i]
+
+    dummy_mr.deallocate(src_buffer.handle, src_buffer.size)
+    dummy_mr.deallocate(dst_buffer.handle, dst_buffer.size)
+
+def test_buffer_copy_to():
+    device = Device()
+    device.set_current()
+    buffer_copy_to(DummyDeviceMemoryResource(device), device)
+    buffer_copy_to(DummyUnifiedMemoryResource(device), device)
+    buffer_copy_to(DummyPinnedMemoryResource(device), device, check = True)
+
+def buffer_copy_from(dummy_mr : MemoryResource, device, check = False):
+    src_buffer = dummy_mr.allocate(size=1024)
+    dst_buffer = dummy_mr.allocate(size=1024)
+    stream = device.create_stream()
+
+    if check:
+        src_ptr = ctypes.cast(src_buffer.handle, ctypes.POINTER(ctypes.c_byte))
+        for i in range(1024):
+            src_ptr[i] = ctypes.c_byte(i)
+
+    dst_buffer.copy_from(src_buffer, stream=stream)
+    device.sync()
+
+    if check:
+        dst_ptr = ctypes.cast(dst_buffer.handle, ctypes.POINTER(ctypes.c_byte))
+
+        for i in range(10):
+            assert dst_ptr[i] == src_ptr[i]
+
+    dummy_mr.deallocate(src_buffer.handle, src_buffer.size)
+    dummy_mr.deallocate(dst_buffer.handle, dst_buffer.size)
+
+def test_buffer_copy_from():
+    device = Device()
+    device.set_current()
+    buffer_copy_from(DummyDeviceMemoryResource(device), device)
+    buffer_copy_from(DummyUnifiedMemoryResource(device), device)
+    buffer_copy_from(DummyPinnedMemoryResource(device), device, check = True)
+
+def buffer_close(dummy_mr : MemoryResource):
+    buffer = dummy_mr.allocate(size=1024)
+    buffer.close()
+    assert buffer.handle == 0
+    assert buffer.memory_resource == None
+
+def test_buffer_close():
+    device = Device()
+    device.set_current()
+    buffer_close(DummyDeviceMemoryResource(device))
+    buffer_close(DummyHostMemoryResource())
+    buffer_close(DummyUnifiedMemoryResource(device))
+    buffer_close(DummyPinnedMemoryResource(device))
+
+test_buffer_copy_to()