cast gpu kernel

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

@@ -0,0 +1,321 @@
+/**
+ * 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.
+ */
+
+#include "backend/kernel_compiler/gpu/arrays/cast_gpu_kernel.h"
+
+namespace mindspore {
+namespace kernel {
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt8), CastGpuKernel,
+                      int8_t, int8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt16), CastGpuKernel,
+                      int8_t, int16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt32), CastGpuKernel,
+                      int8_t, int32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeInt64), CastGpuKernel,
+                      int8_t, int64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeUInt8), CastGpuKernel,
+                      int8_t, uint8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeUInt16), CastGpuKernel,
+                      int8_t, uint16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeUInt32), CastGpuKernel,
+                      int8_t, uint32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeUInt64), CastGpuKernel,
+                      int8_t, uint64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeFloat32), CastGpuKernel,
+                      int8_t, float)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeFloat64), CastGpuKernel,
+                      int8_t, double)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeFloat16), CastGpuKernel,
+                      int8_t, half)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt8).AddOutputAttr(kNumberTypeBool), CastGpuKernel,
+                      int8_t, bool)
+
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt8), CastGpuKernel,
+                      int16_t, int8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt16), CastGpuKernel,
+                      int16_t, int16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt32), CastGpuKernel,
+                      int16_t, int32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeInt64), CastGpuKernel,
+                      int16_t, int64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeUInt8), CastGpuKernel,
+                      int16_t, uint8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeUInt16), CastGpuKernel,
+                      int16_t, uint16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeUInt32), CastGpuKernel,
+                      int16_t, uint32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeUInt64), CastGpuKernel,
+                      int16_t, uint64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeFloat32),
+                      CastGpuKernel, int16_t, float)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeFloat64),
+                      CastGpuKernel, int16_t, double)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeFloat16),
+                      CastGpuKernel, int16_t, half)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt16).AddOutputAttr(kNumberTypeBool), CastGpuKernel,
+                      int16_t, bool)
+
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt8), CastGpuKernel,
+                      int32_t, int8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt16), CastGpuKernel,
+                      int32_t, int16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt32), CastGpuKernel,
+                      int32_t, int32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeInt64), CastGpuKernel,
+                      int32_t, int64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeUInt8), CastGpuKernel,
+                      int32_t, uint8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeUInt16), CastGpuKernel,
+                      int32_t, uint16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeUInt32), CastGpuKernel,
+                      int32_t, uint32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeUInt64), CastGpuKernel,
+                      int32_t, uint64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeFloat32),
+                      CastGpuKernel, int32_t, float)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeFloat64),
+                      CastGpuKernel, int32_t, double)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeFloat16),
+                      CastGpuKernel, int32_t, half)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt32).AddOutputAttr(kNumberTypeBool), CastGpuKernel,
+                      int32_t, bool)
+
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt8), CastGpuKernel,
+                      int64_t, int8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt16), CastGpuKernel,
+                      int64_t, int16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt32), CastGpuKernel,
+                      int64_t, int32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeInt64), CastGpuKernel,
+                      int64_t, int64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeUInt8), CastGpuKernel,
+                      int64_t, uint8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeUInt16), CastGpuKernel,
+                      int64_t, uint16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeUInt32), CastGpuKernel,
+                      int64_t, uint32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeUInt64), CastGpuKernel,
+                      int64_t, uint64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeFloat32),
+                      CastGpuKernel, int64_t, float)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeFloat64),
+                      CastGpuKernel, int64_t, double)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeFloat16),
+                      CastGpuKernel, int64_t, half)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeInt64).AddOutputAttr(kNumberTypeBool), CastGpuKernel,
+                      int64_t, bool)
+
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeInt8), CastGpuKernel,
+                      uint8_t, int8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeInt16), CastGpuKernel,
+                      uint8_t, int16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeInt32), CastGpuKernel,
+                      uint8_t, int32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeInt64), CastGpuKernel,
+                      uint8_t, int64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt8), CastGpuKernel,
+                      uint8_t, uint8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt16), CastGpuKernel,
+                      uint8_t, uint16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt32), CastGpuKernel,
+                      uint8_t, uint32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeUInt64), CastGpuKernel,
+                      uint8_t, uint64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeFloat32),
+                      CastGpuKernel, uint8_t, float)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeFloat64),
+                      CastGpuKernel, uint8_t, double)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeFloat16),
+                      CastGpuKernel, uint8_t, half)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt8).AddOutputAttr(kNumberTypeBool), CastGpuKernel,
+                      uint8_t, bool)
+
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeInt8), CastGpuKernel,
+                      uint16_t, int8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeInt16), CastGpuKernel,
+                      uint16_t, int16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeInt32), CastGpuKernel,
+                      uint16_t, int32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeInt64), CastGpuKernel,
+                      uint16_t, int64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeUInt8), CastGpuKernel,
+                      uint16_t, uint8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeUInt16),
+                      CastGpuKernel, uint16_t, uint16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeUInt32),
+                      CastGpuKernel, uint16_t, uint32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeUInt64),
+                      CastGpuKernel, uint16_t, uint64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeFloat32),
+                      CastGpuKernel, uint16_t, float)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeFloat64),
+                      CastGpuKernel, uint16_t, double)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeFloat16),
+                      CastGpuKernel, uint16_t, half)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt16).AddOutputAttr(kNumberTypeBool), CastGpuKernel,
+                      uint16_t, bool)
+
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeInt8), CastGpuKernel,
+                      uint32_t, int8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeInt16), CastGpuKernel,
+                      uint32_t, int16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeInt32), CastGpuKernel,
+                      uint32_t, int32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeInt64), CastGpuKernel,
+                      uint32_t, int64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeUInt8), CastGpuKernel,
+                      uint32_t, uint8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeUInt16),
+                      CastGpuKernel, uint32_t, uint16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeUInt32),
+                      CastGpuKernel, uint32_t, uint32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeUInt64),
+                      CastGpuKernel, uint32_t, uint64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeFloat32),
+                      CastGpuKernel, uint32_t, float)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeFloat64),
+                      CastGpuKernel, uint32_t, double)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeFloat16),
+                      CastGpuKernel, uint32_t, half)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt32).AddOutputAttr(kNumberTypeBool), CastGpuKernel,
+                      uint32_t, bool)
+
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeInt8), CastGpuKernel,
+                      uint64_t, int8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeInt16), CastGpuKernel,
+                      uint64_t, int16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeInt32), CastGpuKernel,
+                      uint64_t, int32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeInt64), CastGpuKernel,
+                      uint64_t, int64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeUInt8), CastGpuKernel,
+                      uint64_t, uint8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeUInt16),
+                      CastGpuKernel, uint64_t, uint16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeUInt32),
+                      CastGpuKernel, uint64_t, uint32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeUInt64),
+                      CastGpuKernel, uint64_t, uint64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeFloat32),
+                      CastGpuKernel, uint64_t, float)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeFloat64),
+                      CastGpuKernel, uint64_t, double)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeFloat16),
+                      CastGpuKernel, uint64_t, half)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeUInt64).AddOutputAttr(kNumberTypeBool), CastGpuKernel,
+                      uint64_t, bool)
+
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeInt8), CastGpuKernel,
+                      half, int8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeInt16),
+                      CastGpuKernel, half, int16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeInt32),
+                      CastGpuKernel, half, int32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeInt64),
+                      CastGpuKernel, half, int64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeUInt8),
+                      CastGpuKernel, half, uint8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeUInt16),
+                      CastGpuKernel, half, uint16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeUInt32),
+                      CastGpuKernel, half, uint32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeUInt64),
+                      CastGpuKernel, half, uint64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat32),
+                      CastGpuKernel, half, float)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat64),
+                      CastGpuKernel, half, double)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeFloat16),
+                      CastGpuKernel, half, half)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat16).AddOutputAttr(kNumberTypeBool), CastGpuKernel,
+                      half, bool)
+
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeInt8), CastGpuKernel,
+                      float, int8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeInt16),
+                      CastGpuKernel, float, int16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeInt32),
+                      CastGpuKernel, float, int32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeInt64),
+                      CastGpuKernel, float, int64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeUInt8),
+                      CastGpuKernel, float, uint8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeUInt16),
+                      CastGpuKernel, float, uint16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeUInt32),
+                      CastGpuKernel, float, uint32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeUInt64),
+                      CastGpuKernel, float, uint64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat32),
+                      CastGpuKernel, float, float)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat64),
+                      CastGpuKernel, float, double)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeFloat16),
+                      CastGpuKernel, float, half)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat32).AddOutputAttr(kNumberTypeBool), CastGpuKernel,
+                      float, bool)
+
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeInt8), CastGpuKernel,
+                      double, int8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeInt16),
+                      CastGpuKernel, double, int16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeInt32),
+                      CastGpuKernel, double, int32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeInt64),
+                      CastGpuKernel, double, int64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeUInt8),
+                      CastGpuKernel, double, uint8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeUInt16),
+                      CastGpuKernel, double, uint16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeUInt32),
+                      CastGpuKernel, double, uint32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeUInt64),
+                      CastGpuKernel, double, uint64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeFloat32),
+                      CastGpuKernel, double, float)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeFloat64),
+                      CastGpuKernel, double, double)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeFloat16),
+                      CastGpuKernel, double, half)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeFloat64).AddOutputAttr(kNumberTypeBool), CastGpuKernel,
+                      double, bool)
+
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeInt8), CastGpuKernel,
+                      bool, int8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeInt16), CastGpuKernel,
+                      bool, int16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeInt32), CastGpuKernel,
+                      bool, int32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeInt64), CastGpuKernel,
+                      bool, int64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeUInt8), CastGpuKernel,
+                      bool, uint8_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeUInt16), CastGpuKernel,
+                      bool, uint16_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeUInt32), CastGpuKernel,
+                      bool, uint32_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeUInt64), CastGpuKernel,
+                      bool, uint64_t)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeFloat32), CastGpuKernel,
+                      bool, float)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeFloat64), CastGpuKernel,
+                      bool, double)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeFloat16), CastGpuKernel,
+                      bool, half)
+MS_REG_GPU_KERNEL_TWO(Cast, KernelAttr().AddInputAttr(kNumberTypeBool).AddOutputAttr(kNumberTypeBool), CastGpuKernel,
+                      bool, bool)
+}  // namespace kernel
+}  // namespace mindspore

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.cu

@@ -0,0 +1,262 @@
+/**
+ * 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.
+ */
+
+#include <vector>
+#include <iostream>
+
+#include "backend/kernel_compiler/gpu/cuda_impl/cast_impl.cuh"
+#include "runtime/device/gpu/cuda_common.h"
+
+// Generic cast
+template <typename S, typename T>
+__device__ __forceinline__ void CastBase(const S *input_addr, T *output_addr) {
+  *output_addr = static_cast<T>((*input_addr));
+}
+
+// half --> integer
+__device__ __forceinline__ void CastBase(const half *input_addr, uint64_t *output_addr) {
+  *output_addr = __half2ull_rn((*input_addr));
+}
+
+__device__ __forceinline__ void CastBase(const half *input_addr, int64_t *output_addr) {
+  *output_addr = __half2ll_rn((*input_addr));
+}
+
+__device__ __forceinline__ void CastBase(const half *input_addr, uint32_t *output_addr) {
+  *output_addr = __half2uint_rn((*input_addr));
+}
+
+__device__ __forceinline__ void CastBase(const half *input_addr, int32_t *output_addr) {
+  *output_addr = __half2int_rn((*input_addr));
+}
+
+__device__ __forceinline__ void CastBase(const half *input_addr, uint16_t *output_addr) {
+  *output_addr = __half2ushort_rn((*input_addr));
+}
+
+__device__ __forceinline__ void CastBase(const half *input_addr, int16_t *output_addr) {
+  *output_addr = __half2short_rn((*input_addr));
+}
+
+__device__ __forceinline__ void CastBase(const half *input_addr, uint8_t *output_addr) {
+  *output_addr = static_cast<uint8_t>(__half2ushort_rn((*input_addr)));
+}
+
+__device__ __forceinline__ void CastBase(const half *input_addr, int8_t *output_addr) {
+  *output_addr = static_cast<int8_t>(__half2short_rn((*input_addr)));
+}
+
+// integer --> half
+__device__ __forceinline__ void CastBase(const uint64_t *input_addr, half *output_addr) {
+  *output_addr = __ull2half_rn((*input_addr));
+}
+
+__device__ __forceinline__ void CastBase(const int64_t *input_addr, half *output_addr) {
+  *output_addr = __ll2half_rn((*input_addr));
+}
+
+__device__ __forceinline__ void CastBase(const uint32_t *input_addr, half *output_addr) {
+  *output_addr = __uint2half_rn((*input_addr));
+}
+
+__device__ __forceinline__ void CastBase(const int32_t *input_addr, half *output_addr) {
+  *output_addr = __int2half_rn((*input_addr));
+}
+
+__device__ __forceinline__ void CastBase(const uint16_t *input_addr, half *output_addr) {
+  *output_addr = __ushort2half_rn((*input_addr));
+}
+
+__device__ __forceinline__ void CastBase(const int16_t *input_addr, half *output_addr) {
+  *output_addr = __short2half_rn((*input_addr));
+}
+
+__device__ __forceinline__ void CastBase(const uint8_t *input_addr, half *output_addr) {
+  *output_addr = __ushort2half_rn(static_cast<uint16_t>(*input_addr));
+}
+
+__device__ __forceinline__ void CastBase(const int8_t *input_addr, half *output_addr) {
+  *output_addr = __short2half_rn(static_cast<int16_t>(*input_addr));
+}
+
+// Cast
+template <typename S, typename T>
+__global__ void CastKernel(const int input_size, const S *input_addr, T *output_addr) {
+  for (size_t pos = blockIdx.x * blockDim.x + threadIdx.x; pos < input_size; pos += blockDim.x * gridDim.x) {
+    CastBase(input_addr + pos, output_addr + pos);
+  }
+}
+
+template <typename S, typename T>
+void Cast(const int input_size, const S *input_addr, T *output_addr, cudaStream_t stream) {
+  CastKernel<<<GET_BLOCKS(input_size), GET_THREADS, 0, stream>>>(input_size, input_addr, output_addr);
+}
+
+template void Cast(const int input_size, const int8_t *input_addr, int8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int8_t *input_addr, int16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int8_t *input_addr, int32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int8_t *input_addr, int64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int8_t *input_addr, uint8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int8_t *input_addr, uint16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int8_t *input_addr, uint32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int8_t *input_addr, uint64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int8_t *input_addr, float *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int8_t *input_addr, double *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int8_t *input_addr, half *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int8_t *input_addr, bool *output_addr, cudaStream_t stream);
+
+template void Cast(const int input_size, const int16_t *input_addr, int8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int16_t *input_addr, int16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int16_t *input_addr, int32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int16_t *input_addr, int64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int16_t *input_addr, uint8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int16_t *input_addr, uint16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int16_t *input_addr, uint32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int16_t *input_addr, uint64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int16_t *input_addr, float *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int16_t *input_addr, double *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int16_t *input_addr, half *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int16_t *input_addr, bool *output_addr, cudaStream_t stream);
+
+template void Cast(const int input_size, const int32_t *input_addr, int8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int32_t *input_addr, int16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int32_t *input_addr, int32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int32_t *input_addr, int64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int32_t *input_addr, uint8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int32_t *input_addr, uint16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int32_t *input_addr, uint32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int32_t *input_addr, uint64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int32_t *input_addr, float *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int32_t *input_addr, double *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int32_t *input_addr, half *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int32_t *input_addr, bool *output_addr, cudaStream_t stream);
+
+template void Cast(const int input_size, const int64_t *input_addr, int8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int64_t *input_addr, int16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int64_t *input_addr, int32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int64_t *input_addr, int64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int64_t *input_addr, uint8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int64_t *input_addr, uint16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int64_t *input_addr, uint32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int64_t *input_addr, uint64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int64_t *input_addr, float *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int64_t *input_addr, double *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int64_t *input_addr, half *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const int64_t *input_addr, bool *output_addr, cudaStream_t stream);
+
+template void Cast(const int input_size, const uint8_t *input_addr, int8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint8_t *input_addr, int16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint8_t *input_addr, int32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint8_t *input_addr, int64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint8_t *input_addr, uint8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint8_t *input_addr, uint16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint8_t *input_addr, uint32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint8_t *input_addr, uint64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint8_t *input_addr, float *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint8_t *input_addr, double *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint8_t *input_addr, half *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint8_t *input_addr, bool *output_addr, cudaStream_t stream);
+
+template void Cast(const int input_size, const uint16_t *input_addr, int8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint16_t *input_addr, int16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint16_t *input_addr, int32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint16_t *input_addr, int64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint16_t *input_addr, uint8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint16_t *input_addr, uint16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint16_t *input_addr, uint32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint16_t *input_addr, uint64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint16_t *input_addr, float *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint16_t *input_addr, double *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint16_t *input_addr, half *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint16_t *input_addr, bool *output_addr, cudaStream_t stream);
+
+template void Cast(const int input_size, const uint32_t *input_addr, int8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint32_t *input_addr, int16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint32_t *input_addr, int32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint32_t *input_addr, int64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint32_t *input_addr, uint8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint32_t *input_addr, uint16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint32_t *input_addr, uint32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint32_t *input_addr, uint64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint32_t *input_addr, float *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint32_t *input_addr, double *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint32_t *input_addr, half *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint32_t *input_addr, bool *output_addr, cudaStream_t stream);
+
+template void Cast(const int input_size, const uint64_t *input_addr, int8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint64_t *input_addr, int16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint64_t *input_addr, int32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint64_t *input_addr, int64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint64_t *input_addr, uint8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint64_t *input_addr, uint16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint64_t *input_addr, uint32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint64_t *input_addr, uint64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint64_t *input_addr, float *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint64_t *input_addr, double *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint64_t *input_addr, half *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const uint64_t *input_addr, bool *output_addr, cudaStream_t stream);
+
+template void Cast(const int input_size, const half *input_addr, int8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const half *input_addr, int16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const half *input_addr, int32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const half *input_addr, int64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const half *input_addr, uint8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const half *input_addr, uint16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const half *input_addr, uint32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const half *input_addr, uint64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const half *input_addr, float *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const half *input_addr, double *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const half *input_addr, half *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const half *input_addr, bool *output_addr, cudaStream_t stream);
+
+template void Cast(const int input_size, const float *input_addr, int8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const float *input_addr, int16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const float *input_addr, int32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const float *input_addr, int64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const float *input_addr, uint8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const float *input_addr, uint16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const float *input_addr, uint32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const float *input_addr, uint64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const float *input_addr, float *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const float *input_addr, double *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const float *input_addr, half *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const float *input_addr, bool *output_addr, cudaStream_t stream);
+
+template void Cast(const int input_size, const double *input_addr, int8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const double *input_addr, int16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const double *input_addr, int32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const double *input_addr, int64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const double *input_addr, uint8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const double *input_addr, uint16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const double *input_addr, uint32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const double *input_addr, uint64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const double *input_addr, float *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const double *input_addr, double *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const double *input_addr, half *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const double *input_addr, bool *output_addr, cudaStream_t stream);
+
+template void Cast(const int input_size, const bool *input_addr, int8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const bool *input_addr, int16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const bool *input_addr, int32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const bool *input_addr, int64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const bool *input_addr, uint8_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const bool *input_addr, uint16_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const bool *input_addr, uint32_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const bool *input_addr, uint64_t *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const bool *input_addr, float *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const bool *input_addr, double *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const bool *input_addr, half *output_addr, cudaStream_t stream);
+template void Cast(const int input_size, const bool *input_addr, bool *output_addr, cudaStream_t stream);

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_