def test_cuda_stream_wait_event_and_record_event(self):
if paddle.is_compiled_with_cuda():
s1 = cuda.Stream(0)
tensor1 = paddle.to_tensor(paddle.rand([1000, 1000]))
tensor2 = paddle.matmul(tensor1, tensor1)
e1 = cuda.Event(False, False, False)
s1.record_event(e1)
s2 = cuda.Stream(0)
s2.wait_event(e1)
s2.synchronize()
self.assertTrue(e1.query() and s1.query() and s2.query())
def func_setUp(self):
self.empty = paddle.to_tensor(np.array([], dtype="int64"),
place=paddle.CPUPlace())
data = np.random.randn(100, 50, 50).astype("float32")
self.src = paddle.to_tensor(data, place=paddle.CUDAPinnedPlace())
self.dst = paddle.empty(shape=[100, 50, 50], dtype="float32")
self.index = paddle.to_tensor(np.array([1, 3, 5, 7, 9],
dtype="int64")).cpu()
self.buffer = paddle.empty(shape=[50, 50, 50],
dtype="float32").pin_memory()
self.stream = cuda.Stream()
def setUp(self):
self.src = paddle.rand(shape=[100, 50, 50, 5], dtype="float32")
self.dst = paddle.empty(
shape=[200, 50, 50, 5], dtype="float32").pin_memory()
self.stream = cuda.Stream()