From bd70512806b2abfe01f3be7ecd3b24faebb7cabc Mon Sep 17 00:00:00 2001
From: caifubi <caifubi1@huawei.com>
Date: Tue, 6 Jul 2021 15:52:43 +0800
Subject: [PATCH] fix TensorCopySlices bprop

---
 .../ops/_grad_experimental/grad_inner_ops.py  |  5 ++--
 .../test_aicpu_ops/test_tensor_copy_slices.py | 29 +++++++++++++++++++
 2 files changed, 32 insertions(+), 2 deletions(-)

diff --git a/mindspore/ops/_grad_experimental/grad_inner_ops.py b/mindspore/ops/_grad_experimental/grad_inner_ops.py
index 47550a4a62f..ff84e8ffd65 100644
--- a/mindspore/ops/_grad_experimental/grad_inner_ops.py
+++ b/mindspore/ops/_grad_experimental/grad_inner_ops.py
@@ -18,6 +18,7 @@
 from .._grad.grad_base import bprop_getters
 from ..operations import _inner_ops as inner
 from .. import functional as F
+from ..composite.multitype_ops.zeros_like_impl import zeros_like
 
 @bprop_getters.register(inner.TensorCopySlices)
 def get_bprop_tensor_copy_slices(self):
@@ -25,8 +26,8 @@ def get_bprop_tensor_copy_slices(self):
     tensor_copy_slices = inner.TensorCopySlices()
 
     def bprop(x, update, begin, end, stride, out, dout):
-        x_grad = tensor_copy_slices(dout, F.zeros_like(update), begin, end, stride)
+        x_grad = tensor_copy_slices(dout, zeros_like(update), begin, end, stride)
         update_grad = F.strided_slice(dout, begin, end, stride)
-        return x_grad, update_grad, F.zeros_like(begin), F.zeros_like(end), F.zeros_like(stride)
+        return x_grad, update_grad, zeros_like(begin), zeros_like(end), zeros_like(stride)
 
     return bprop
diff --git a/tests/st/ops/ascend/test_aicpu_ops/test_tensor_copy_slices.py b/tests/st/ops/ascend/test_aicpu_ops/test_tensor_copy_slices.py
index 305f7ec50e3..d3d34749df0 100644
--- a/tests/st/ops/ascend/test_aicpu_ops/test_tensor_copy_slices.py
+++ b/tests/st/ops/ascend/test_aicpu_ops/test_tensor_copy_slices.py
@@ -18,7 +18,11 @@ import pytest
 import mindspore.context as context
 import mindspore.nn as nn
 from mindspore import Tensor
+from mindspore import Parameter
+import mindspore.common.dtype as mstype
 from mindspore.ops.operations import _inner_ops
+from mindspore import ops as P
+from mindspore.ops import composite as C
 
 class Net(nn.Cell):
     def __init__(self):
@@ -28,6 +32,16 @@ class Net(nn.Cell):
     def construct(self, input_x, update, begin, end, strides):
         return self.copy_slices(input_x, update, begin, end, strides)
 
+class GradNet(nn.Cell):
+    def __init__(self):
+        super(GradNet, self).__init__()
+        self.input_x = Parameter(Tensor([[2, 3, 4, 5], [2, 3, 4, 5]], mstype.float32))
+        self.relu = P.ReLU()
+
+    def construct(self, input_y):
+        self.input_x[0:1] /= input_y
+        return self.relu(self.input_x)
+
 def convert_begin_end_strides_to_slice(begin, end, strides):
     result = []
     for x, y, z in zip(begin, end, strides):
@@ -60,6 +74,21 @@ def test_tensor_copy_slices():
     test_tensor_copy_slices_net_many_dtype((10, 10, 10), (5, 10), (0, 5, 0), (1, 10, 10), (1, 1, 1,), support_dtype)
     test_tensor_copy_slices_net_many_dtype((10, 10, 10), (5, 10), (9, 5,), (10, 10,), (1, 1,), support_dtype)
 
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_tensor_copy_slices_bprop():
+    input_y = Tensor(2.0, mstype.float32)
+    net = GradNet()
+    net.set_grad()
+    grad_net = C.GradOperation(get_all=True, sens_param=True)(net)
+
+    output = net(input_y)
+    grad_output = grad_net(input_y, output)
+    assert np.allclose(grad_output[0].asnumpy(), np.array([-6.75]))
+
+
 @pytest.mark.level0
 @pytest.mark.platform_arm_ascend_training
 @pytest.mark.platform_x86_ascend_training