# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. import unittest import numpy as np from op_test import get_device_place from utils import dygraph_guard import paddle from paddle import base @unittest.skipIf( paddle.core.is_compiled_with_xpu(), "xpu does not support dlpack", ) class TestDLPack(unittest.TestCase): def test_dlpack_dygraph(self): if paddle.is_compiled_with_cuda(): with dygraph_guard(): tensor = paddle.to_tensor(np.array([1, 2, 3, 4]).astype("int")) dlpack_v1 = paddle.utils.dlpack.to_dlpack(tensor) out_from_dlpack_v1 = paddle.utils.dlpack.from_dlpack(dlpack_v1) dlpack_v2 = tensor.__dlpack__() out_from_dlpack_v2 = paddle.from_dlpack(dlpack_v2) self.assertTrue( isinstance( out_from_dlpack_v1, paddle.base.core.eager.Tensor ) ) self.assertTrue( isinstance( out_from_dlpack_v2, paddle.base.core.eager.Tensor ) ) self.assertEqual( str(tensor.place), str(out_from_dlpack_v1.place) ) self.assertEqual( str(tensor.place), str(out_from_dlpack_v2.place) ) np.testing.assert_array_equal( out_from_dlpack_v1.numpy(), np.array([1, 2, 3, 4]).astype("int"), ) np.testing.assert_array_equal( out_from_dlpack_v2.numpy(), np.array([1, 2, 3, 4]).astype("int"), ) def test_dlpack_tensor_larger_than_2dim(self): if paddle.is_compiled_with_cuda(): with dygraph_guard(): numpy_data = np.random.randn(4, 5, 6) t = paddle.to_tensor(numpy_data) dlpack_v1 = paddle.utils.dlpack.to_dlpack(t) dlpack_v2 = t.__dlpack__() out_v1 = paddle.utils.dlpack.from_dlpack(dlpack_v1) out_v2 = paddle.from_dlpack(dlpack_v2) self.assertEqual(str(t.place), str(out_v1.place)) self.assertEqual(str(t.place), str(out_v2.place)) np.testing.assert_allclose( numpy_data, out_v1.numpy(), rtol=1e-05 ) np.testing.assert_allclose( numpy_data, out_v2.numpy(), rtol=1e-05 ) def test_dlpack_dtype_and_place_consistency(self): with dygraph_guard(): dtypes = [ "float16", "float32", "float64", "int8", "int16", "int32", "int64", "uint8", "bool", ] places = [paddle.CPUPlace()] if paddle.device.is_compiled_with_cuda(): places.append(get_device_place()) dtypes.append("bfloat16") data = np.ones((2, 3, 4)) for place in places: for dtype in dtypes: x = paddle.to_tensor(data, dtype=dtype, place=place) dlpack_v1 = paddle.utils.dlpack.to_dlpack(x) o_v1 = paddle.utils.dlpack.from_dlpack(dlpack_v1) dlpack_v2 = x.__dlpack__() o_v2 = paddle.from_dlpack(dlpack_v2) self.assertEqual(x.dtype, o_v1.dtype) self.assertEqual(x.dtype, o_v2.dtype) np.testing.assert_allclose( x.numpy(), o_v1.numpy(), rtol=1e-05 ) np.testing.assert_allclose( x.numpy(), o_v2.numpy(), rtol=1e-05 ) self.assertEqual(str(x.place), str(o_v1.place)) self.assertEqual(str(x.place), str(o_v2.place)) complex_dtypes = ["complex64", "complex128"] for place in places: for dtype in complex_dtypes: x = paddle.to_tensor( [[1 + 6j, 2 + 5j, 3 + 4j], [4 + 3j, 5 + 2j, 6 + 1j]], dtype=dtype, place=place, ) dlpack_v1 = paddle.utils.dlpack.to_dlpack(x) o_v1 = paddle.utils.dlpack.from_dlpack(dlpack_v1) dlpack_v2 = x.__dlpack__() o_v2 = paddle.from_dlpack(dlpack_v2) self.assertEqual(x.dtype, o_v1.dtype) self.assertEqual(x.dtype, o_v2.dtype) np.testing.assert_allclose( x.numpy(), o_v1.numpy(), rtol=1e-05 ) np.testing.assert_allclose( x.numpy(), o_v2.numpy(), rtol=1e-05 ) self.assertEqual(str(x.place), str(o_v1.place)) self.assertEqual(str(x.place), str(o_v2.place)) def test_dlpack_deletion(self): # See Paddle issue 47171 with dygraph_guard(): places = [base.CPUPlace()] if paddle.is_compiled_with_cuda(): places.append(get_device_place()) for place in places: for _ in range(4): a = paddle.rand(shape=[3, 5], dtype="float32").to( device=place ) dlpack_v1 = paddle.utils.dlpack.to_dlpack(a) dlpack_v2 = a.__dlpack__() b1 = paddle.utils.dlpack.from_dlpack(dlpack_v1) b2 = paddle.from_dlpack(dlpack_v2) self.assertEqual(str(a.place), str(b1.place)) self.assertEqual(str(a.place), str(b2.place)) def test_to_dlpack_for_loop(self): # See Paddle issue 50120 with dygraph_guard(): places = [base.CPUPlace()] if paddle.is_compiled_with_cuda(): places.append(get_device_place()) for place in places: for _ in range(4): x = paddle.rand([3, 5]).to(device=place) dlpack_v1 = paddle.utils.dlpack.to_dlpack(x) dlpack_v2 = x.__dlpack__() def test_to_dlpack_modification(self): # See Paddle issue 50120 with dygraph_guard(): places = [base.CPUPlace()] if paddle.is_compiled_with_cuda(): places.append(get_device_place()) for place in places: for _ in range(4): x = paddle.rand([3, 5]).to(device=place) dlpack_v1 = paddle.utils.dlpack.to_dlpack(x) dlpack_v2 = x.__dlpack__() y1 = paddle.utils.dlpack.from_dlpack(dlpack_v1) y2 = paddle.from_dlpack(dlpack_v2) y1[1:2, 2:5] = 2.0 y2[1:2, 2:5] = 2.0 np.testing.assert_allclose(x.numpy(), y1.numpy()) np.testing.assert_allclose(x.numpy(), y2.numpy()) self.assertEqual(str(x.place), str(y1.place)) self.assertEqual(str(x.place), str(y2.place)) def test_to_dlpack_data_ptr_consistency(self): # See Paddle issue 50120 with dygraph_guard(): places = [base.CPUPlace()] if paddle.is_compiled_with_cuda(): places.append(get_device_place()) for place in places: for _ in range(4): x = paddle.rand([3, 5]).to(device=place) dlpack_v1 = paddle.utils.dlpack.to_dlpack(x) dlpack_v2 = x.__dlpack__() y1 = paddle.utils.dlpack.from_dlpack(dlpack_v1) y2 = paddle.from_dlpack(dlpack_v2) self.assertEqual(x.data_ptr(), y1.data_ptr()) self.assertEqual(x.data_ptr(), y2.data_ptr()) self.assertEqual(str(x.place), str(y1.place)) self.assertEqual(str(x.place), str(y2.place)) def test_to_dlpack_strides_consistency(self): with dygraph_guard(): places = [base.CPUPlace()] if paddle.is_compiled_with_cuda(): places.append(get_device_place()) for place in places: for _ in range(4): x = paddle.rand([10, 10]).to(device=place) x_strided = x[::2, ::2] dlpack_v1 = paddle.utils.dlpack.to_dlpack(x_strided) dlpack_v2 = x_strided.__dlpack__() y1 = paddle.utils.dlpack.from_dlpack(dlpack_v1) y2 = paddle.from_dlpack(dlpack_v2) self.assertEqual(x_strided.strides, y1.strides) self.assertEqual(x_strided.strides, y2.strides) self.assertEqual(str(x_strided.place), str(y1.place)) self.assertEqual(str(x_strided.place), str(y2.place)) np.testing.assert_equal(x_strided.numpy(), y1.numpy()) np.testing.assert_equal(x_strided.numpy(), y2.numpy()) def test_to_dlpack_from_zero_dim(self): with dygraph_guard(): places = [base.CPUPlace()] if paddle.is_compiled_with_cuda(): places.append(get_device_place()) for place in places: for _ in range(4): x = paddle.to_tensor(1.0, place=place) dlpack_v1 = paddle.utils.dlpack.to_dlpack(x) dlpack_v2 = x.__dlpack__() y1 = paddle.utils.dlpack.from_dlpack(dlpack_v1) y2 = paddle.from_dlpack(dlpack_v2) self.assertEqual(x.data_ptr(), y1.data_ptr()) self.assertEqual(x.data_ptr(), y2.data_ptr()) self.assertEqual(str(x.place), str(y1.place)) self.assertEqual(str(x.place), str(y2.place)) self.assertEqual(y1.shape, []) self.assertEqual(y2.shape, []) self.assertEqual(y1.numel().item(), 1) self.assertEqual(y2.numel().item(), 1) np.testing.assert_array_equal(x.numpy(), y1.numpy()) np.testing.assert_array_equal(x.numpy(), y2.numpy()) def test_to_dlpack_from_zero_size(self): with dygraph_guard(): places = [base.CPUPlace()] if paddle.is_compiled_with_cuda(): places.append(get_device_place()) for place in places: for _ in range(4): x = paddle.zeros([0, 10]).to(device=place) self.assertEqual(x.strides, [10, 1]) dlpack_v1 = paddle.utils.dlpack.to_dlpack(x) dlpack_v2 = x.__dlpack__() y1 = paddle.utils.dlpack.from_dlpack(dlpack_v1) y2 = paddle.from_dlpack(dlpack_v2) self.assertEqual(x.data_ptr(), y1.data_ptr()) self.assertEqual(x.data_ptr(), y2.data_ptr()) self.assertEqual(str(x.place), str(y1.place)) self.assertEqual(str(x.place), str(y2.place)) self.assertEqual(y1.strides, [10, 1]) self.assertEqual(y2.strides, [10, 1]) self.assertEqual(y1.shape, [0, 10]) self.assertEqual(y2.shape, [0, 10]) self.assertEqual(y1.numel().item(), 0) self.assertEqual(y2.numel().item(), 0) np.testing.assert_array_equal(x.numpy(), y1.numpy()) np.testing.assert_array_equal(x.numpy(), y2.numpy()) def test_dlpack_with_custom_stream(self): if not (paddle.is_compiled_with_cuda()): self.skipTest("Test requires CUDA support.") with dygraph_guard(): paddle.set_device('gpu:0') s1 = paddle.device.Stream() s2 = paddle.device.Stream() e = paddle.device.Event() s2.wait_event(e) x = paddle.to_tensor([1, 2, 3], dtype='float32') s1.synchronize() dlpack_capsule = x.__dlpack__(stream=s1.stream_base.raw_stream) y = paddle.from_dlpack(dlpack_capsule) np.testing.assert_array_equal(x.numpy(), y.numpy()) self.assertTrue(s1.query(), "Stream s1 did not complete all tasks.") self.assertTrue(s2.query(), "Stream s2 did not complete all tasks.") def test_dlpack_with_custom_stream_error(self): if not (paddle.is_compiled_with_cuda()): self.skipTest("Test requires CUDA support.") with dygraph_guard(): x = paddle.to_tensor([1, 2, 3], dtype='float32') with self.assertRaisesRegex( TypeError, "stream must be an integer or None." ): dlpack_capsule = x.__dlpack__(stream=object()) with self.assertRaisesRegex( ValueError, "For CUDA, stream=0 is ambiguityous" ): dlpack_capsule = x.__dlpack__(stream=0) with self.assertRaisesRegex( ValueError, "For CUDA, stream=2 means per-thread default stream, which is not supported.", ): dlpack_capsule = x.__dlpack__(stream=2) @unittest.skipIf( paddle.core.is_compiled_with_xpu(), "xpu does not support dlpack", ) class TestRaiseError(unittest.TestCase): def test_dlpack_invalid_sparse(self): sparse_tensor = paddle.sparse.sparse_coo_tensor( indices=[[0]], values=[1], shape=[3] ) with self.assertRaises(BufferError): sparse_tensor.__dlpack__() def test_dlpack_requires_grad(self): tensor_with_grad = paddle.to_tensor( [1.0, 2.0, 3.0], stop_gradient=False ) with self.assertRaises(BufferError): tensor_with_grad.__dlpack__() if __name__ == "__main__": unittest.main()