test_stride.py

# Copyright (c) 2023 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

import paddle
from paddle.pir_utils import DygraphPirGuard


def ref_view_as_real(x):
    return np.stack([x.real, x.imag], -1)


def ref_view_as_complex(x):
    real, imag = np.take(x, 0, axis=-1), np.take(x, 1, axis=-1)
    return real + 1j * imag


class TestStride(unittest.TestCase):
    def call_transpose(self):
        x_np = np.random.random(size=[2, 3, 4]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        x_transposed1 = paddle.transpose(x, perm=[1, 0, 2])
        x_np_transposed1 = x_np.transpose(1, 0, 2)
        self.assertTrue(np.allclose(x_transposed1.numpy(), x_np_transposed1))
        self.assertFalse(x_transposed1.is_contiguous())
        self.assertTrue(x._is_shared_buffer_with(x_transposed1))

        x_c = x_transposed1.contiguous()
        self.assertTrue(np.allclose(x_c.numpy(), x_np_transposed1))

        x_transposed2 = paddle.transpose(x_transposed1, perm=[2, 0, 1])
        x_np_transposed2 = x_np_transposed1.transpose(2, 0, 1)
        self.assertTrue(np.allclose(x_transposed2.numpy(), x_np_transposed2))
        self.assertFalse(x_transposed2.is_contiguous())

        y = x_transposed2 + 2
        y_np = x_np_transposed2 + 2
        self.assertTrue(np.allclose(y.numpy(), y_np))
        self.assertTrue(y.is_contiguous())
        self.assertFalse(x._is_shared_buffer_with(y))

    def call_diagonal(self):
        x_np = np.random.random(size=[2, 3, 4]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = paddle.diagonal(x)
        out2 = paddle.diagonal(x, offset=0, axis1=2, axis2=1)
        out3 = paddle.diagonal(x, offset=1, axis1=0, axis2=1)
        out4 = paddle.diagonal(x, offset=0, axis1=1, axis2=2)

        np_out = np.diagonal(x_np)
        np_out2 = np.diagonal(x_np, offset=0, axis1=2, axis2=1)
        np_out3 = np.diagonal(x_np, offset=1, axis1=0, axis2=1)
        np_out4 = np.diagonal(x_np, offset=0, axis1=1, axis2=2)

        self.assertTrue(np.allclose(out.numpy(), np_out))
        self.assertTrue(np.allclose(out2.numpy(), np_out2))
        self.assertTrue(np.allclose(out3.numpy(), np_out3))
        self.assertTrue(np.allclose(out4.numpy(), np_out4))

        self.assertFalse(out.is_contiguous())
        self.assertFalse(out2.is_contiguous())
        self.assertFalse(out3.is_contiguous())
        self.assertFalse(out4.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))
        self.assertTrue(x._is_shared_buffer_with(out2))
        self.assertTrue(x._is_shared_buffer_with(out3))
        self.assertTrue(x._is_shared_buffer_with(out4))

        out_c = out.contiguous()
        out2_c = out2.contiguous()
        out3_c = out3.contiguous()
        out4_c = out4.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))
        self.assertTrue(np.allclose(out2_c.numpy(), np_out2))
        self.assertTrue(np.allclose(out3_c.numpy(), np_out3))
        self.assertTrue(np.allclose(out4_c.numpy(), np_out4))

    def call_slice(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = x[1:10, 0:10, 0:10, 0:20]

        np_out = x_np[1:10, 0:10, 0:10, 0:20]

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

        self.assertTrue(out_c._is_shared_buffer_with(out))

    def call_strided_slice(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = x[1:10:2, 0:10:2, 0:10:2, 0:20:2]

        np_out = x_np[1:10:2, 0:10:2, 0:10:2, 0:20:2]

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertFalse(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

    def call_index_select(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = x[:, :, :, 5]
        np_out = x_np[:, :, :, 5]

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertFalse(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

    def call_reshape(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = paddle.reshape(x, [10, 100, 20])
        np_out = x_np.reshape(10, 100, 20)

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

        self.assertTrue(out_c._is_shared_buffer_with(out))

    def call_real(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('complex64')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = paddle.real(x)
        np_out = np.real(x_np)

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertFalse(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

    def call_imag(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('complex128')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = paddle.imag(x)
        np_out = np.imag(x_np)

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertFalse(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

    def call_as_real(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('complex128')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = paddle.as_real(x)
        np_out = ref_view_as_real(x_np)

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

        self.assertTrue(out_c._is_shared_buffer_with(out))

    def call_as_complex(self):
        x_np = np.random.random(size=[10, 10, 10, 2]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = paddle.as_complex(x)
        np_out = ref_view_as_complex(x_np)

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

        self.assertTrue(out_c._is_shared_buffer_with(out))

    def call_flatten(self):
        x_np = np.random.random(size=[2, 3, 4, 4]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = paddle.flatten(x, start_axis=1, stop_axis=2)
        np_out = x_np.reshape(2, 12, 4)

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

        self.assertTrue(out_c._is_shared_buffer_with(out))

    def call_squeeze(self):
        x_np = np.random.random(size=[5, 1, 10]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = paddle.squeeze(x, axis=1)
        np_out = x_np.reshape(5, 10)

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

        self.assertTrue(out_c._is_shared_buffer_with(out))

    def call_unsqueeze(self):
        x_np = np.random.random(size=[5, 10]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = paddle.unsqueeze(x, axis=0)
        np_out = x_np.reshape(1, 5, 10)

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

        self.assertTrue(out_c._is_shared_buffer_with(out))

    def call_split(self):
        x_np = np.random.random(size=[3, 9, 5]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out0, out1, out2 = paddle.split(x, num_or_sections=3, axis=1)
        np_out0, np_out1, np_out2 = np.split(x_np, 3, 1)

        self.assertTrue(np.allclose(out0.numpy(), np_out0))
        self.assertTrue(np.allclose(out1.numpy(), np_out1))
        self.assertTrue(np.allclose(out2.numpy(), np_out2))

        self.assertFalse(out0.is_contiguous())
        self.assertFalse(out1.is_contiguous())
        self.assertFalse(out2.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out0))
        self.assertTrue(x._is_shared_buffer_with(out1))
        self.assertTrue(x._is_shared_buffer_with(out2))

        out0_c = out0.contiguous()
        out1_c = out1.contiguous()
        out2_c = out2.contiguous()

        self.assertTrue(np.allclose(out0_c.numpy(), np_out0))
        self.assertTrue(np.allclose(out1_c.numpy(), np_out1))
        self.assertTrue(np.allclose(out2_c.numpy(), np_out2))

    def call_split2(self):
        x_np = np.random.random(size=[3, 9, 5]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out0, out1, out2 = paddle.split(x, num_or_sections=[2, 3, 4], axis=1)
        out = np.split(x_np, [2, 5], 1)
        np_out0 = out[0]
        np_out1 = out[1]
        np_out2 = out[2]

        self.assertTrue(np.allclose(out0.numpy(), np_out0))
        self.assertTrue(np.allclose(out1.numpy(), np_out1))
        self.assertTrue(np.allclose(out2.numpy(), np_out2))

        self.assertFalse(out0.is_contiguous())
        self.assertFalse(out1.is_contiguous())
        self.assertFalse(out2.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out0))
        self.assertTrue(x._is_shared_buffer_with(out1))
        self.assertTrue(x._is_shared_buffer_with(out2))

        out0_c = out0.contiguous()
        out1_c = out1.contiguous()
        out2_c = out2.contiguous()

        self.assertTrue(np.allclose(out0_c.numpy(), np_out0))
        self.assertTrue(np.allclose(out1_c.numpy(), np_out1))
        self.assertTrue(np.allclose(out2_c.numpy(), np_out2))

    def call_split3(self):
        x_np = np.random.random(size=[9, 3, 5]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out0, out1, out2 = paddle.split(x, num_or_sections=3, axis=0)
        np_out0, np_out1, np_out2 = np.split(x_np, 3, 0)

        self.assertTrue(np.allclose(out0.numpy(), np_out0))
        self.assertTrue(np.allclose(out1.numpy(), np_out1))
        self.assertTrue(np.allclose(out2.numpy(), np_out2))

        self.assertTrue(out0.is_contiguous())
        self.assertTrue(out1.is_contiguous())
        self.assertTrue(out2.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out0))
        self.assertTrue(x._is_shared_buffer_with(out1))
        self.assertTrue(x._is_shared_buffer_with(out2))

        out0_c = out0.contiguous()
        out1_c = out1.contiguous()
        out2_c = out2.contiguous()

        self.assertTrue(np.allclose(out0_c.numpy(), np_out0))
        self.assertTrue(np.allclose(out1_c.numpy(), np_out1))
        self.assertTrue(np.allclose(out2_c.numpy(), np_out2))

        self.assertTrue(out0_c._is_shared_buffer_with(out0))
        self.assertTrue(out1_c._is_shared_buffer_with(out1))
        self.assertTrue(out2_c._is_shared_buffer_with(out2))

    def call_split4(self):
        x_np = np.random.random(size=[9, 3, 5]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out0, out1, out2 = paddle.split(x, num_or_sections=[2, 3, 4], axis=0)
        out = np.split(x_np, [2, 5], 0)
        np_out0 = out[0]
        np_out1 = out[1]
        np_out2 = out[2]

        self.assertTrue(np.allclose(out0.numpy(), np_out0))
        self.assertTrue(np.allclose(out1.numpy(), np_out1))
        self.assertTrue(np.allclose(out2.numpy(), np_out2))

        self.assertTrue(out0.is_contiguous())
        self.assertTrue(out1.is_contiguous())
        self.assertTrue(out2.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out0))
        self.assertTrue(x._is_shared_buffer_with(out1))
        self.assertTrue(x._is_shared_buffer_with(out2))

        out0_c = out0.contiguous()
        out1_c = out1.contiguous()
        out2_c = out2.contiguous()

        self.assertTrue(np.allclose(out0_c.numpy(), np_out0))
        self.assertTrue(np.allclose(out1_c.numpy(), np_out1))
        self.assertTrue(np.allclose(out2_c.numpy(), np_out2))

        self.assertTrue(out0_c._is_shared_buffer_with(out0))
        self.assertTrue(out1_c._is_shared_buffer_with(out1))
        self.assertTrue(out2_c._is_shared_buffer_with(out2))

    def call_chunk(self):
        x_np = np.random.random(size=[3, 9, 5]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out0, out1, out2 = paddle.chunk(x, chunks=3, axis=1)
        np_out0, np_out1, np_out2 = np.split(x_np, 3, 1)

        self.assertTrue(np.allclose(out0.numpy(), np_out0))
        self.assertTrue(np.allclose(out1.numpy(), np_out1))
        self.assertTrue(np.allclose(out2.numpy(), np_out2))

        self.assertFalse(out0.is_contiguous())
        self.assertFalse(out1.is_contiguous())
        self.assertFalse(out2.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out0))
        self.assertTrue(x._is_shared_buffer_with(out1))
        self.assertTrue(x._is_shared_buffer_with(out2))

        out0_c = out0.contiguous()
        out1_c = out1.contiguous()
        out2_c = out2.contiguous()

        self.assertTrue(np.allclose(out0_c.numpy(), np_out0))
        self.assertTrue(np.allclose(out1_c.numpy(), np_out1))
        self.assertTrue(np.allclose(out2_c.numpy(), np_out2))

    def call_unbind(self):
        x_np = np.random.random(size=[3, 9, 5]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out0, out1, out2 = paddle.unbind(x, axis=0)
        np_out0 = x_np[0, 0:100, 0:100]
        np_out1 = x_np[1, 0:100, 0:100]
        np_out2 = x_np[2, 0:100, 0:100]

        self.assertTrue(np.allclose(out0.numpy(), np_out0))
        self.assertTrue(np.allclose(out1.numpy(), np_out1))
        self.assertTrue(np.allclose(out2.numpy(), np_out2))

        self.assertTrue(out0.is_contiguous())
        self.assertTrue(out1.is_contiguous())
        self.assertTrue(out2.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out0))
        self.assertTrue(x._is_shared_buffer_with(out1))
        self.assertTrue(x._is_shared_buffer_with(out2))

        out0_c = out0.contiguous()
        out1_c = out1.contiguous()
        out2_c = out2.contiguous()

        self.assertTrue(np.allclose(out0_c.numpy(), np_out0))
        self.assertTrue(np.allclose(out1_c.numpy(), np_out1))
        self.assertTrue(np.allclose(out2_c.numpy(), np_out2))

        self.assertTrue(out0_c._is_shared_buffer_with(out0))
        self.assertTrue(out1_c._is_shared_buffer_with(out1))
        self.assertTrue(out2_c._is_shared_buffer_with(out2))

    def call_as_strided(self):
        x_np = np.random.random(size=[2, 4, 6]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = paddle.as_strided(x, [8, 6], [6, 1])
        np_out = x_np.reshape(8, 6)

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

        self.assertTrue(out_c._is_shared_buffer_with(out))

    def call_view(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = paddle.view(x, [10, 100, 20])
        np_out = x_np.reshape(10, 100, 20)

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

        self.assertTrue(out_c._is_shared_buffer_with(out))

    def call_view2(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = paddle.view(x, "uint8")
        np_out = x_np.view(np.uint8)

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

        self.assertTrue(out_c._is_shared_buffer_with(out))

    def call_view3(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('float32')
        x = paddle.to_tensor(x_np)
        np.testing.assert_allclose(x.numpy(), x_np)

        # shape inference
        out = paddle.view(x, [10, 100, -1])
        np_out = x_np.reshape(10, 100, 20)

        np.testing.assert_allclose(out.numpy(), np_out)

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        np.testing.assert_allclose(out_c.numpy(), np_out)

        self.assertTrue(out_c._is_shared_buffer_with(out))

    def call_view4(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('float32')
        x = paddle.to_tensor(x_np)
        np.testing.assert_allclose(x.numpy(), x_np)

        out = paddle.view(x, paddle.uint8)
        np_out = x_np.view(np.uint8)

        np.testing.assert_allclose(out.numpy(), np_out)

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        np.testing.assert_allclose(out_c.numpy(), np_out)

        self.assertTrue(out_c._is_shared_buffer_with(out))

    # dim4 -> dim2
    def call_view5(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('float32')
        x = paddle.to_tensor(x_np)
        np.testing.assert_allclose(x.numpy(), x_np)

        # shape inference
        out = paddle.view(x, [1000, -1])
        np_out = x_np.reshape(1000, 20)

        np.testing.assert_allclose(out.numpy(), np_out)

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        np.testing.assert_allclose(out_c.numpy(), np_out)

        self.assertTrue(out_c._is_shared_buffer_with(out))

    # dim4 -> dim1
    def call_view6(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('float32')
        x = paddle.to_tensor(x_np)
        np.testing.assert_allclose(x.numpy(), x_np)

        # shape inference
        out = paddle.view(x, [-1])
        np_out = x_np.reshape(20000)

        np.testing.assert_allclose(out.numpy(), np_out)

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        np.testing.assert_allclose(out_c.numpy(), np_out)

        self.assertTrue(out_c._is_shared_buffer_with(out))

    # dim4 -> dim5
    def call_view7(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('float32')
        x = paddle.to_tensor(x_np)
        np.testing.assert_allclose(x.numpy(), x_np)

        # shape inference
        out = paddle.view(x, [10, 10, 10, 10, -1])
        np_out = x_np.reshape(10, 10, 10, 10, 2)

        np.testing.assert_allclose(out.numpy(), np_out)

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        np.testing.assert_allclose(out_c.numpy(), np_out)

        self.assertTrue(out_c._is_shared_buffer_with(out))

    # test invoke of view_reshape_grad for high order derivative
    def call_view8(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('float32')
        x = paddle.to_tensor(x_np)
        x.stop_gradient = False
        y = x.view([10, -1]).tanh()
        dx = paddle.grad(y, x, create_graph=True)[0]
        np.testing.assert_allclose(
            dx.numpy(), (1 - y**2).reshape(x.shape).numpy(), 1e-6, 1e-6
        )
        dxx = paddle.grad(dx, x, create_graph=True)[0]
        np.testing.assert_allclose(
            dxx.numpy(),
            (-2 * y * (1 - y**2)).reshape(x.shape).numpy(),
            1e-6,
            1e-6,
        )
        dxxx = paddle.grad(dxx, x, create_graph=True)[0]
        np.testing.assert_allclose(
            dxxx.numpy(),
            (-2 * (1 - y**2) ** 2 + -2 * y * (-2 * y) * (1 - y**2))
            .reshape(x.shape)
            .numpy(),
            1e-6,
            1e-6,
        )

    def call_view_as(self):
        x_np = np.random.random(size=[10, 10, 10, 20]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        np_out = x_np.reshape(10, 100, 20)
        tmp = paddle.to_tensor(np_out)
        out = paddle.view_as(x, tmp)

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertTrue(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

        self.assertTrue(out_c._is_shared_buffer_with(out))

    def call_unfold(self):
        x_np = np.random.random(size=[9]).astype('float32')
        x = paddle.to_tensor(x_np)
        self.assertTrue(np.allclose(x.numpy(), x_np))

        out = paddle.unfold(x, 0, 2, 4)
        np_out = np.stack((x_np[0:2], x_np[4:6]))

        self.assertTrue(np.allclose(out.numpy(), np_out))

        self.assertFalse(out.is_contiguous())

        self.assertTrue(x._is_shared_buffer_with(out))

        out_c = out.contiguous()

        self.assertTrue(np.allclose(out_c.numpy(), np_out))

    def call_stride(self):
        self.call_transpose()
        self.call_diagonal()
        self.call_slice()
        self.call_strided_slice()
        self.call_index_select()
        self.call_reshape()
        self.call_real()
        self.call_imag()
        self.call_as_real()
        self.call_as_complex()
        self.call_flatten()
        self.call_squeeze()
        self.call_unsqueeze()
        # self.call_split()
        # self.call_split2()
        # self.call_split3()
        # self.call_split4()
        # self.call_chunk()
        self.call_unbind()
        self.call_as_strided()
        self.call_view()
        self.call_view2()
        self.call_view3()
        self.call_view4()
        self.call_view5()
        self.call_view6()
        self.call_view7()
        self.call_view_as()
        self.call_unfold()


class TestStrideCPU(TestStride):
    def test_stride_cpu(self):
        paddle.set_device('cpu')
        self.call_stride()


@unittest.skipIf(
    not paddle.base.core.is_compiled_with_cuda(),
    "core is not compiled with CUDA",
)
class TestStrideGPU(TestStride):
    def test_stride_gpu(self):
        paddle.set_device('gpu')
        self.call_stride()


class TestToStaticCheck(unittest.TestCase):

    def test_error(self):
        @paddle.jit.to_static(full_graph=True)
        def func1():
            x_np = np.random.random(size=[2, 3, 4]).astype('float32')
            x = paddle.to_tensor(x_np)
            y = paddle.transpose(x, perm=[1, 0, 2])
            z = paddle.ones([3, 2, 4])
            y.add_(z)

        self.assertRaises(ValueError, func1)

        @paddle.jit.to_static(full_graph=True)
        def func2():
            x_np = np.random.random(size=[2, 3, 4]).astype('float32')
            x = paddle.to_tensor(x_np)
            y = paddle.transpose(x, perm=[1, 0, 2])
            z = paddle.ones([2, 3, 4])
            x.add_(z)

        self.assertRaises(ValueError, func2)

    def test_error_with_program(self):
        with DygraphPirGuard():

            @paddle.jit.to_static(full_graph=True)
            def func1():
                x = paddle.ones((4, 3)) * 2
                y = paddle.transpose(x, [1, 0])
                z = paddle.ones((3,))
                paddle.tensor.manipulation.fill_diagonal_tensor_(y, z)

            self.assertRaises(ValueError, func1)

            @paddle.jit.to_static(full_graph=True)
            def func2():
                x = paddle.ones((4, 3)) * 2
                y = paddle.transpose(x, [1, 0])
                z = paddle.ones((3,))
                paddle.tensor.manipulation.fill_diagonal_tensor_(x, z)

            self.assertRaises(ValueError, func2)

    def test_no_error_with_program(self):
        with DygraphPirGuard():

            @paddle.jit.to_static(full_graph=True)
            def func1():
                x = paddle.ones((4, 3)) * 2
                y = paddle.transpose(x, [1, 0])
                yy = paddle.assign(y)
                z = paddle.ones((3,))
                paddle.tensor.manipulation.fill_diagonal_tensor_(yy, z)

            func1()

            @paddle.jit.to_static(full_graph=True)
            def func2():
                x = paddle.ones((4, 3)) * 2
                y = paddle.transpose(x, [1, 0])
                xx = paddle.assign(x)
                z = paddle.ones((3,))
                paddle.tensor.manipulation.fill_diagonal_tensor_(xx, z)

            func2()

    def test_no_error(self):

        @paddle.jit.to_static(full_graph=True)
        def func1():
            x_np = np.random.random(size=[2, 3, 4]).astype('float32')
            x = paddle.to_tensor(x_np)
            z = paddle.ones([3, 2, 4])
            y = paddle.transpose(x, perm=[1, 0, 2])
            yy = paddle.assign(y)
            yy.add_(z)

        func1()

        @paddle.jit.to_static(full_graph=True)
        def func2():
            x_np = np.random.random(size=[2, 3, 4]).astype('float32')
            x = paddle.to_tensor(x_np)
            y = paddle.transpose(x, perm=[1, 0, 2])
            xx = paddle.assign(x)
            z = paddle.ones([2, 3, 4])
            xx.add_(z)

        func2()


if __name__ == '__main__':
    unittest.main()