diff --git a/mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/array_reduce_gpu_kernel.cc b/mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/array_reduce_gpu_kernel.cc
index 9af961647b..589dbee97f 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/array_reduce_gpu_kernel.cc
+++ b/mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/array_reduce_gpu_kernel.cc
@@ -1,5 +1,5 @@
 /**
- * Copyright 2019 Huawei Technologies Co., Ltd
+ * Copyright 2019-2021 Huawei Technologies Co., Ltd
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -18,6 +18,8 @@
 
 namespace mindspore {
 namespace kernel {
+MS_REG_GPU_KERNEL_ONE(ReduceMax, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeFloat64),
+                      ArrayReduceGpuKernel, double)
 MS_REG_GPU_KERNEL_ONE(ReduceMax, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
                       ArrayReduceGpuKernel, float)
 MS_REG_GPU_KERNEL_ONE(ReduceMax, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
@@ -26,6 +28,8 @@ MS_REG_GPU_KERNEL_ONE(ReduceMean, KernelAttr().AddInputAttr(kNumberTypeFloat32).
                       ArrayReduceGpuKernel, float)
 MS_REG_GPU_KERNEL_ONE(ReduceMean, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
                       ArrayReduceGpuKernel, half)
+MS_REG_GPU_KERNEL_ONE(ReduceSum, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeFloat64),
+                      ArrayReduceGpuKernel, double)
 MS_REG_GPU_KERNEL_ONE(ReduceSum, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
                       ArrayReduceGpuKernel, float)
 MS_REG_GPU_KERNEL_ONE(ReduceSum, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
diff --git a/mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/array_reduce_gpu_kernel.h b/mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/array_reduce_gpu_kernel.h
index 4ca2fb8a4a..db2a88a927 100644
--- a/mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/array_reduce_gpu_kernel.h
+++ b/mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/array_reduce_gpu_kernel.h
@@ -1,5 +1,5 @@
 /**
- * Copyright 2019-2020 Huawei Technologies Co., Ltd
+ * Copyright 2019-2021 Huawei Technologies Co., Ltd
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -14,8 +14,8 @@
  * limitations under the License.
  */
 
-#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_ARRAYREDUCE_GPU_KERNEL_H_
-#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_ARRAYREDUCE_GPU_KERNEL_H_
+#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_ARRAYS_ARRAY_REDUCE_GPU_KERNEL_H_
+#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_ARRAYS_ARRAY_REDUCE_GPU_KERNEL_H_
 
 #include <map>
 #include <string>
@@ -291,4 +291,4 @@ class ArrayReduceGpuKernel : public GpuKernel {
 }  // namespace kernel
 }  // namespace mindspore
 
-#endif  // MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_ARRAYREDUCE_GPU_KERNEL_H_
+#endif  // MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_ARRAYS_ARRAY_REDUCE_GPU_KERNEL_H_
diff --git a/tests/st/ops/gpu/test_reduce_max_op.py b/tests/st/ops/gpu/test_reduce_max_op.py
index 968a0353e9..46943b4bbb 100644
--- a/tests/st/ops/gpu/test_reduce_max_op.py
+++ b/tests/st/ops/gpu/test_reduce_max_op.py
@@ -1,4 +1,4 @@
-# Copyright 2019 Huawei Technologies Co., Ltd
+# Copyright 2019-2021 Huawei Technologies Co., Ltd
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -209,3 +209,26 @@ def test_reduce_max_dynamic():
 
     np.testing.assert_almost_equal(output_1.asnumpy(), expect_1)
     np.testing.assert_almost_equal(output_2.asnumpy(), expect_2)
+
+class ReduceMaxTypeNet(nn.Cell):
+    def __init__(self, nptype):
+        super(ReduceMaxTypeNet, self).__init__()
+        self.x0 = Tensor(x0.astype(nptype))
+        self.axis0 = axis0
+        self.keep_dims0 = keep_dims0
+
+    def construct(self):
+        return P.ReduceMax(self.keep_dims0)(self.x0, self.axis0)
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_reduce_max_float64():
+    context.set_context(mode=context.GRAPH_MODE, device_target="GPU")
+    net = ReduceMaxTypeNet(np.float64)
+    output = net()
+    expect = np.max(x0, axis=axis0, keepdims=keep_dims0).astype(np.float64)
+    diff = abs(output.asnumpy() - expect)
+    error = np.ones(shape=expect.shape) * 1.0e-5
+    assert np.all(diff < error)
+    assert output.shape == expect.shape
diff --git a/tests/st/ops/gpu/test_reduce_sum_op.py b/tests/st/ops/gpu/test_reduce_sum_op.py
index d89216969e..708deb2184 100644
--- a/tests/st/ops/gpu/test_reduce_sum_op.py
+++ b/tests/st/ops/gpu/test_reduce_sum_op.py
@@ -1,4 +1,4 @@
-# Copyright 2019 Huawei Technologies Co., Ltd
+# Copyright 2019-2021 Huawei Technologies Co., Ltd
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -301,3 +301,26 @@ def test_reduce_sum_dynamic():
 
     np.testing.assert_almost_equal(output_1.asnumpy(), expect_1)
     np.testing.assert_almost_equal(output_2.asnumpy(), expect_2)
+
+class ReduceSumTypeNet(nn.Cell):
+    def __init__(self, nptype):
+        super(ReduceSumTypeNet, self).__init__()
+        self.x0 = Tensor(x0.astype(nptype))
+        self.axis0 = axis0
+        self.keep_dims0 = keep_dims0
+
+    def construct(self):
+        return P.ReduceSum(self.keep_dims0)(self.x0, self.axis0)
+
+@pytest.mark.level0
+@pytest.mark.platform_x86_gpu_training
+@pytest.mark.env_onecard
+def test_reduce_sum_float64():
+    context.set_context(mode=context.GRAPH_MODE, device_target="GPU")
+    net = ReduceSumTypeNet(np.float64)
+    output = net()
+    expect = np.sum(x0, axis=axis0, keepdims=keep_dims0).astype(np.float64)
+    diff = abs(output.asnumpy() - expect)
+    error = np.ones(shape=expect.shape) * 1.0e-5
+    assert np.all(diff < error)
+    assert output.shape == expect.shape