!8667 fix cast issue

From: @jonwe
Reviewed-by: @robingrosman,@tom__chen
Signed-off-by: @tom__chen

mindspore/ccsrc/backend/kernel_compiler/gpu/arrays/cast_gpu_kernel.h

@@ -0,0 +1,82 @@
+/**
+ * Copyright 2020 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.
+ * 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.
+ */
+
+#ifndef MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_CAST_GPU_KERNEL_H_
+#define MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_CAST_GPU_KERNEL_H_
+
+#include <vector>
+#include "backend/kernel_compiler/gpu/gpu_kernel.h"
+#include "backend/kernel_compiler/gpu/gpu_kernel_factory.h"
+#include "backend/kernel_compiler/gpu/cuda_impl/cast_impl.cuh"
+
+namespace mindspore {
+namespace kernel {
+template <typename S, typename T>
+class CastGpuKernel : public GpuKernel {
+ public:
+  CastGpuKernel() : input_size_(1), output_size_(1) {}
+  ~CastGpuKernel() = default;
+
+  const std::vector<size_t> &GetInputSizeList() const override { return input_size_list_; }
+  const std::vector<size_t> &GetOutputSizeList() const override { return output_size_list_; }
+  const std::vector<size_t> &GetWorkspaceSizeList() const override { return workspace_size_list_; }
+
+  bool Launch(const std::vector<AddressPtr> &inputs, const std::vector<AddressPtr> &,
+              const std::vector<AddressPtr> &outputs, void *stream_ptr) override {
+    S *input_addr = GetDeviceAddress<S>(inputs, 0);
+    T *output_addr = GetDeviceAddress<T>(outputs, 0);
+
+    Cast(input_size_, input_addr, output_addr, reinterpret_cast<cudaStream_t>(stream_ptr));
+    return true;
+  }
+  bool Init(const CNodePtr &kernel_node) override {
+    auto input_shapes = AnfAlgo::GetPrevNodeOutputInferShape(kernel_node, 0);
+    auto output_shapes = AnfAlgo::GetOutputInferShape(kernel_node, 0);
+    input_size_ = 1;
+    for (size_t i = 0; i < input_shapes.size(); i++) {
+      input_size_ *= input_shapes[i];
+    }
+
+    output_size_ = 1;
+    for (size_t j = 0; j < output_shapes.size(); j++) {
+      output_size_ *= output_shapes[j];
+    }
+
+    if (input_size_ != output_size_) {
+      MS_LOG(EXCEPTION) << "Input size is not equal to output size.";
+    }
+    InitSizeLists();
+    return true;
+  }
+
+ protected:
+  void InitSizeLists() override {
+    input_size_list_.push_back(input_size_ * sizeof(T));
+    output_size_list_.push_back(output_size_ * sizeof(T));
+  }
+
+ private:
+  int input_size_;
+  int output_size_;
+
+  std::vector<size_t> input_size_list_;
+  std::vector<size_t> output_size_list_;
+  std::vector<size_t> workspace_size_list_;
+};
+}  // namespace kernel
+}  // namespace mindspore
+
+#endif  // MINDSPORE_CCSRC_BACKEND_KERNEL_COMPILER_GPU_CAST_GPU_KERNEL_H_

mindspore/ccsrc/backend/kernel_compiler/gpu/cuda_impl/cast_impl.cuh

@@ -0,0 +1,25 @@
+/**
+ * Copyright 2020 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.
+ * 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.
+ */
+
+#ifndef MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_CAST_H_
+#define MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_CAST_H_
+
+#include <vector>
+#include "runtime/device/gpu/cuda_common.h"
+
+template <typename S, typename T>
+void Cast(const int input_size, const S *input_addr, T *output_addr, cudaStream_t stream);
+#endif  // MINDSPORE_CCSRC_KERNEL_GPU_CUDA_IMPL_CAST_H_