[PTen]Add alias kernel name (#37881)

* add alias kernel name

* modify code as suggestions

paddle/fluid/framework/operator.cc

@@ -1760,7 +1760,8 @@ OpKernelType OperatorWithKernel::GetKernelTypeForVar(
 
 KernelSignature OperatorWithKernel::GetExpectedPtenKernelArgs(
     const ExecutionContext& ctx) const {
-  return KernelSignatureMap::Instance().Get(Type());
+  return KernelSignatureMap::Instance().Get(
+      pten::TransToPtenKernelName(Type()));
 }
 
 void OperatorWithKernel::BuildPtenKernelContext(

paddle/fluid/framework/pten_utils.cc

@@ -101,10 +101,10 @@ KernelSignatureMap& KernelSignatureMap::Instance() {
       if (pten::KernelFactory::Instance().HasCompatiblePtenKernel(op_type)) {
         KernelArgsNameMakerByOpProto maker(op_proto);
         VLOG(10) << "Register kernel signature for " << op_type;
-        auto success =
-            kernel_signature_map_->map_
-                .emplace(op_type, std::move(maker.GetKernelSignature()))
-                .second;
+        auto success = kernel_signature_map_->map_
+                           .emplace(pten::TransToPtenKernelName(op_type),
+                                    std::move(maker.GetKernelSignature()))
+                           .second;
         PADDLE_ENFORCE_EQ(
             success, true,
             platform::errors::PermissionDenied(

paddle/fluid/operators/elementwise/elementwise_op.h

@@ -144,26 +144,25 @@ class ElementwiseOp : public framework::OperatorWithKernel {
       const framework::ExecutionContext &ctx) const override {
     if (Type() == "elementwise_add") {
       if (ctx.InputVar("X")->IsType<framework::LoDTensor>()) {
-        return framework::KernelSignature("elementwise_add", {"X", "Y"},
-                                          {"axis"}, {"Out"});
+        return framework::KernelSignature("add", {"X", "Y"}, {"axis"}, {"Out"});
       }
     }
     if (Type() == "elementwise_sub") {
       if (ctx.InputVar("X")->IsType<framework::LoDTensor>()) {
-        return framework::KernelSignature("elementwise_sub", {"X", "Y"},
-                                          {"axis"}, {"Out"});
+        return framework::KernelSignature("subtract", {"X", "Y"}, {"axis"},
+                                          {"Out"});
       }
     }
     if (Type() == "elementwise_div") {
       if (ctx.InputVar("X")->IsType<framework::LoDTensor>()) {
-        return framework::KernelSignature("elementwise_div", {"X", "Y"},
-                                          {"axis"}, {"Out"});
+        return framework::KernelSignature("divide", {"X", "Y"}, {"axis"},
+                                          {"Out"});
       }
     }
     if (Type() == "elementwise_mul") {
       if (ctx.InputVar("X")->IsType<framework::LoDTensor>()) {
-        return framework::KernelSignature("elementwise_mul", {"X", "Y"},
-                                          {"axis"}, {"Out"});
+        return framework::KernelSignature("multiply", {"X", "Y"}, {"axis"},
+                                          {"Out"});
       }
     }
     return framework::KernelSignature("None", {"X"}, {}, {"Out"});

paddle/fluid/operators/fill_any_like_op.cc

@@ -50,7 +50,7 @@ class FillAnyLikeOp : public framework::OperatorWithKernel {
 
   framework::KernelSignature GetExpectedPtenKernelArgs(
       const framework::ExecutionContext &ctx) const override {
-    return framework::KernelSignature("fill_any_like", {}, {"value"}, {"Out"});
+    return framework::KernelSignature("full_like", {}, {"value"}, {"Out"});
   }
 };
 

paddle/fluid/operators/fill_constant_op.cc

@@ -118,8 +118,7 @@ class FillConstantOp : public framework::OperatorWithKernel {
       value = str_value.empty() ? "value" : "str_value";
     }
     if (!ctx.OutputVar("Out")->IsType<framework::SelectedRows>()) {
-      return framework::KernelSignature("fill_constant", {}, {shape, value},
-                                        {"Out"});
+      return framework::KernelSignature("full", {}, {shape, value}, {"Out"});
     }
     return framework::KernelSignature("fill_constant.unregistered", {}, {}, {});
   }

paddle/fluid/operators/flatten_op.cc

@@ -337,11 +337,10 @@ class FlattenContiguousRangeOp : public framework::OperatorWithKernel {
   framework::KernelSignature GetExpectedPtenKernelArgs(
       const framework::ExecutionContext &ctx) const override {
     if (ctx.HasOutput("XShape")) {
-      return framework::KernelSignature("flatten_contiguous_range.mid", {"X"},
-                                        {"start_axis", "stop_axis"},
-                                        {"Out", "XShape"});
+      return framework::KernelSignature(
+          "flatten.mid", {"X"}, {"start_axis", "stop_axis"}, {"Out", "XShape"});
     } else {
-      return framework::KernelSignature("flatten_contiguous_range", {"X"},
+      return framework::KernelSignature("flatten", {"X"},
                                         {"start_axis", "stop_axis"}, {"Out"});
     }
   }

paddle/fluid/operators/reshape_op.cc

@@ -555,13 +555,13 @@ class Reshape2Op : public ReshapeOp {
       const framework::ExecutionContext &ctx) const override {
     auto multi_inputs = ctx.MultiInput<framework::Tensor>("ShapeTensor");
     if (multi_inputs.size() > 0) {
-      return framework::KernelSignature("reshape2.mulhost",
-                                        {"X", "ShapeTensor"}, {}, {"Out"});
+      return framework::KernelSignature("reshape.mulhost", {"X", "ShapeTensor"},
+                                        {}, {"Out"});
     } else if (ctx.HasInput("Shape")) {
-      return framework::KernelSignature("reshape2.host", {"X", "Shape"}, {},
+      return framework::KernelSignature("reshape.host", {"X", "Shape"}, {},
                                         {"Out"});
     } else {
-      return framework::KernelSignature("reshape2", {"X"}, {"shape"}, {"Out"});
+      return framework::KernelSignature("reshape", {"X"}, {"shape"}, {"Out"});
     }
   }
 };

paddle/pten/core/convert_utils.cc

@@ -12,7 +12,7 @@ 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 "paddle/pten/core/convert_utils.h"
-
+#include "paddle/pten/core/kernel_alias_name.h"
 // See Note [ Why still include the fluid headers? ]
 #include "paddle/fluid/platform/device/gpu/gpu_info.h"
 
@@ -270,4 +270,12 @@ std::string DataType2String(DataType dtype) {
   }
 }
 
+const std::string& TransToPtenKernelName(const std::string& fluid_op_name) {
+  if (kernel_alias_name_map.find(fluid_op_name) !=
+      kernel_alias_name_map.end()) {
+    return kernel_alias_name_map.at(fluid_op_name);
+  }
+  return fluid_op_name;
+}
+
 }  // namespace pten

paddle/pten/core/convert_utils.h

@@ -32,6 +32,8 @@ namespace pten {
 using DataType = paddle::experimental::DataType;
 using DataLayout = paddle::experimental::DataLayout;
 
+const std::string& TransToPtenKernelName(const std::string& fluid_op_name);
+
 Backend TransToPtenBackend(const paddle::platform::Place& place);
 DataType TransToPtenDataType(
     const paddle::framework::proto::VarType::Type& dtype);

paddle/pten/core/kernel_alias_name.h

+/* Copyright (c) 2021 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. */
+
+// TODO(yuanrisheng): this file may need to be removed
+#pragma once
+
+namespace pten {
+
+// the key is kernel_name in fluid, the value is the kernel_name in pten
+// the key is sorted by key's alphabet
+const std::unordered_map<std::string, std::string> kernel_alias_name_map = {
+    {"elementwise_add", "add"},
+    {"elementwise_div", "divide"},
+    {"elementwise_mul", "muliply"},
+    {"elementwise_sub", "subtract"},
+    {"fill_any_like", "full_like"},
+    {"fill_constant", "full"},
+    {"flatten_contiguous_range", "flatten"},
+    // {"matmul_v2", "matmul"},
+    {"reduce_mean", "mean"},
+    {"reduce_sum", "sum"},
+    {"reshape2", "reshape"},
+    // fluid kernel "mean/reshape/matmul/flatten/sum" should be deprecated
+    {"flatten", "deprecated"},
+    // {"matmul", "deprecated"},
+    {"mean", "deprecated"},
+    {"reshape", "deprecated"},
+    {"sum", "deprecated"}};
+
+}  // namespace pten

paddle/pten/core/kernel_factory.h

@@ -23,6 +23,7 @@
 #include "paddle/pten/common/backend.h"
 #include "paddle/pten/common/data_type.h"
 #include "paddle/pten/common/layout.h"
+#include "paddle/pten/core/convert_utils.h"
 #include "paddle/pten/core/kernel_def.h"
 
 // See Note [ Why still include the fluid headers? ]
@@ -269,7 +270,7 @@ class KernelFactory {
   }
 
   bool HasCompatiblePtenKernel(const std::string& op_type) const {
-    return compatible_op_types_.count(op_type) > 0;
+    return compatible_op_types_.count(TransToPtenKernelName(op_type)) > 0;
   }
 
   const Kernel& SelectKernelOrThrowError(const KernelName& kernel_name,

paddle/pten/kernels/cpu/creation.cc

@@ -63,7 +63,7 @@ void FillConstant(const CPUContext& dev_ctx,
 
 PT_REGISTER_MODULE(CreationCPU);
 
-PT_REGISTER_KERNEL("fill_any_like",
+PT_REGISTER_KERNEL("full_like",
                    CPU,
                    ANY,
                    pten::FillAnyLike,
@@ -74,7 +74,7 @@ PT_REGISTER_KERNEL("fill_any_like",
                    bool,
                    paddle::platform::float16) {}
 
-PT_REGISTER_KERNEL("fill_constant",
+PT_REGISTER_KERNEL("full",
                    CPU,
                    ANY,
                    pten::FillConstant,

paddle/pten/kernels/cpu/manipulation.cc

@@ -135,7 +135,7 @@ PT_REGISTER_MODULE(ManipulationCPU);
 
 // TODO(yuanrisheng): "flatten_contiguous_range" is compatible with old kernel
 // architecture, kernel_name should be "flatten".
-PT_REGISTER_KERNEL("flatten_contiguous_range",
+PT_REGISTER_KERNEL("flatten",
                    CPU,
                    ANY,
                    pten::Flatten,
@@ -146,7 +146,7 @@ PT_REGISTER_KERNEL("flatten_contiguous_range",
                    int,
                    int64_t) {}
 
-PT_REGISTER_KERNEL("flatten_contiguous_range.mid",
+PT_REGISTER_KERNEL("flatten.mid",
                    CPU,
                    ANY,
                    pten::FlattenWithXShape,
@@ -176,32 +176,29 @@ PT_REGISTER_KERNEL("cast",
 
 // TODO(yuanrisheng): "reshape2" is compatible with old kernel
 // architecture, kernel_name should be "reshape".
-PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2",
+PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape",
                                 CPU,
                                 ANY,
                                 pten::ReshapeFromVectorVal) {}
 
-PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2.mid",
+PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape.mid",
                                 CPU,
                                 ANY,
                                 pten::ReshapeFromVectorValWithXShape) {}
 
-PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2.host",
-                                CPU,
-                                ANY,
-                                pten::ReshapeFromDT) {
+PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape.host", CPU, ANY, pten::ReshapeFromDT) {
   kernel->InputAt(1).SetBackend(pten::Backend::CPU);
   kernel->InputAt(1).SetDataType(paddle::experimental::DataType::INT32);
 }
 
-PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2.host.mid",
+PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape.host.mid",
                                 CPU,
                                 ANY,
                                 pten::ReshapeFromDTWithXShape) {
   kernel->InputAt(1).SetBackend(pten::Backend::CPU);
   kernel->InputAt(1).SetDataType(paddle::experimental::DataType::INT32);
 }
-PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2.mulhost",
+PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape.mulhost",
                                 CPU,
                                 ANY,
                                 pten::ReshapeFromVectorDT) {
@@ -209,7 +206,7 @@ PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2.mulhost",
   kernel->InputAt(1).SetDataType(paddle::experimental::DataType::INT32);
 }
 
-PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2.mulhost.mid",
+PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape.mulhost.mid",
                                 CPU,
                                 ANY,
                                 pten::ReshapeFromVectorDTWithXShape) {

paddle/pten/kernels/cpu/math.cc

@@ -116,7 +116,7 @@ using complex128 = ::paddle::platform::complex<double>;
 // using bfloat16 = ::paddle::platform::bfloat16;
 
 PT_REGISTER_KERNEL("sign", CPU, ANY, pten::Sign, float, double) {}
-PT_REGISTER_KERNEL("reduce_mean", CPU, ANY, pten::Mean, float, double, bool) {}
+PT_REGISTER_KERNEL("mean", CPU, ANY, pten::Mean, float, double, bool) {}
 PT_REGISTER_KERNEL("scale",
                    CPU,
                    ANY,
@@ -130,7 +130,7 @@ PT_REGISTER_KERNEL("scale",
                    int,
                    int64_t) {}
 
-PT_REGISTER_KERNEL("elementwise_add",
+PT_REGISTER_KERNEL("add",
                    CPU,
                    ANY,
                    pten::ElementwiseAdd,
@@ -140,7 +140,7 @@ PT_REGISTER_KERNEL("elementwise_add",
                    int64_t,
                    complex64,
                    complex128) {}
-PT_REGISTER_KERNEL("elementwise_sub",
+PT_REGISTER_KERNEL("subtract",
                    CPU,
                    ANY,
                    pten::ElementwiseSub,
@@ -150,7 +150,7 @@ PT_REGISTER_KERNEL("elementwise_sub",
                    int64_t,
                    complex64,
                    complex128) {}
-PT_REGISTER_KERNEL("elementwise_div",
+PT_REGISTER_KERNEL("divide",
                    CPU,
                    ANY,
                    pten::ElementwiseDiv,
@@ -160,7 +160,7 @@ PT_REGISTER_KERNEL("elementwise_div",
                    int64_t,
                    complex64,
                    complex128) {}
-PT_REGISTER_KERNEL("elementwise_mul",
+PT_REGISTER_KERNEL("multiply",
                    CPU,
                    ANY,
                    pten::ElementwiseMul,
@@ -172,7 +172,7 @@ PT_REGISTER_KERNEL("elementwise_mul",
                    complex64,
                    complex128) {}
 
-PT_REGISTER_KERNEL("reduce_sum",
+PT_REGISTER_KERNEL("sum",
                    CPU,
                    ANY,
                    pten::Sum,

paddle/pten/kernels/cuda/creation.cu

@@ -64,7 +64,7 @@ void FillConstant(const CUDAContext& dev_ctx,
 
 PT_REGISTER_MODULE(CreationCUDA);
 
-PT_REGISTER_KERNEL("fill_any_like",
+PT_REGISTER_KERNEL("full_like",
                    CUDA,
                    ANY,
                    pten::FillAnyLike,
@@ -75,7 +75,7 @@ PT_REGISTER_KERNEL("fill_any_like",
                    bool,
                    paddle::platform::float16) {}
 
-PT_REGISTER_KERNEL("fill_constant",
+PT_REGISTER_KERNEL("full",
                    CUDA,
                    ANY,
                    pten::FillConstant,

paddle/pten/kernels/cuda/manipulation.cu

@@ -135,7 +135,7 @@ PT_REGISTER_MODULE(ManipulationCUDA);
 using float16 = paddle::platform::float16;
 // TODO(yuanrisheng): "flatten_contiguous_range" is compatible with old kernel
 // architecture, kernel_name should be "flatten".
-PT_REGISTER_KERNEL("flatten_contiguous_range",
+PT_REGISTER_KERNEL("flatten",
                    CUDA,
                    ANY,
                    pten::Flatten,
@@ -147,7 +147,7 @@ PT_REGISTER_KERNEL("flatten_contiguous_range",
                    int,
                    int64_t) {}
 
-PT_REGISTER_KERNEL("flatten_contiguous_range.mid",
+PT_REGISTER_KERNEL("flatten.mid",
                    CUDA,
                    ANY,
                    pten::FlattenWithXShape,
@@ -184,17 +184,17 @@ PTEN_REGISTER_CAST_CUDA_BASE_TYPE(cast, paddle::platform::bfloat16)
 PTEN_REGISTER_CAST_CUDA_BASE_TYPE(cast)
 #endif
 
-PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2",
+PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape",
                                 CUDA,
                                 ANY,
                                 pten::ReshapeFromVectorVal) {}
 
-PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2.mid",
+PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape.mid",
                                 CUDA,
                                 ANY,
                                 pten::ReshapeFromVectorValWithXShape) {}
 
-PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2.host",
+PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape.host",
                                 CUDA,
                                 ANY,
                                 pten::ReshapeFromDT) {
@@ -202,7 +202,7 @@ PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2.host",
   kernel->InputAt(1).SetDataType(paddle::experimental::DataType::INT32);
 }
 
-PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2.host.mid",
+PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape.host.mid",
                                 CUDA,
                                 ANY,
                                 pten::ReshapeFromDTWithXShape) {
@@ -210,7 +210,7 @@ PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2.host.mid",
   kernel->InputAt(1).SetDataType(paddle::experimental::DataType::INT32);
 }
 
-PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2.mulhost",
+PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape.mulhost",
                                 CUDA,
                                 ANY,
                                 pten::ReshapeFromVectorDT) {
@@ -218,7 +218,7 @@ PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2.mulhost",
   kernel->InputAt(1).SetDataType(paddle::experimental::DataType::INT32);
 }
 
-PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape2.mulhost.mid",
+PT_REGISTER_KERNEL_WITH_NO_TYPE("reshape.mulhost.mid",
                                 CUDA,
                                 ANY,
                                 pten::ReshapeFromVectorDTWithXShape) {

paddle/pten/kernels/cuda/math.cu

@@ -119,7 +119,7 @@ using complex64 = ::paddle::platform::complex<float>;
 using complex128 = ::paddle::platform::complex<double>;
 
 PT_REGISTER_KERNEL("sign", CUDA, ANY, pten::Sign, float, double, float16) {}
-PT_REGISTER_KERNEL("reduce_mean", CUDA, ANY, pten::Mean, float, double, bool) {}
+PT_REGISTER_KERNEL("mean", CUDA, ANY, pten::Mean, float, double, bool) {}
 PT_REGISTER_KERNEL("scale",
                    CUDA,
                    ANY,
@@ -132,7 +132,7 @@ PT_REGISTER_KERNEL("scale",
                    int16_t,
                    int,
                    int64_t) {}
-PT_REGISTER_KERNEL("elementwise_add",
+PT_REGISTER_KERNEL("add",
                    CUDA,
                    ANY,
                    pten::ElementwiseAdd,
@@ -143,7 +143,7 @@ PT_REGISTER_KERNEL("elementwise_add",
                    float16,
                    complex64,
                    complex128) {}
-PT_REGISTER_KERNEL("elementwise_sub",
+PT_REGISTER_KERNEL("subtract",
                    CUDA,
                    ANY,
                    pten::ElementwiseSub,
@@ -154,7 +154,7 @@ PT_REGISTER_KERNEL("elementwise_sub",
                    float16,
                    complex64,
                    complex128) {}
-PT_REGISTER_KERNEL("elementwise_div",
+PT_REGISTER_KERNEL("divide",
                    CUDA,
                    ANY,
                    pten::ElementwiseDiv,
@@ -165,7 +165,7 @@ PT_REGISTER_KERNEL("elementwise_div",
                    float16,
                    complex64,
                    complex128) {}
-PT_REGISTER_KERNEL("elementwise_mul",
+PT_REGISTER_KERNEL("multiply",
                    CUDA,
                    ANY,
                    pten::ElementwiseMul,
@@ -177,7 +177,7 @@ PT_REGISTER_KERNEL("elementwise_mul",
                    float16,
                    complex64,
                    complex128) {}
-PT_REGISTER_KERNEL("reduce_sum",
+PT_REGISTER_KERNEL("sum",
                    CUDA,
                    ANY,
                    pten::Sum,

python/paddle/utils/code_gen/api.yaml

@@ -5,7 +5,7 @@
     func : ElementwiseInferMeta
     param : [x, y, -1]
   kernel :
-    func : elementwise_add
+    func : add
     param : [x, y, -1]
 
 - api : cast
@@ -25,7 +25,7 @@
     func : ElementwiseInferMeta
     param : [x, y, -1]
   kernel :
-    func : elementwise_div
+    func : divide
     param : [x, y, -1]
 
 - api : dot
@@ -42,7 +42,7 @@
   infer_meta : 
     func : FlattenInferMeta
   kernel : 
-    func : flatten_contiguous_range
+    func : flatten
 
 - api : full
   args : (const ScalarArray& shape, const Scalar& value, DataType dtype=DataType::FLOAT32, Backend place=Backend::CPU, DataLayout layout=DataLayout::NCHW)
@@ -51,7 +51,7 @@
     func : FullInferMeta
     param : [shape, dtype, layout]
   kernel : 
-    func : fill_constant
+    func : full
     param : [shape, value]
     data_type : dtype
     backend : place
@@ -64,7 +64,7 @@
     func : FullLikeInferMeta
     param : [x, dtype, layout]
   kernel : 
-    func : fill_any_like
+    func : full_like
     param : [x, value]
     data_type : dtype > x
     backend : place > x
@@ -84,7 +84,7 @@
   infer_meta : 
     func : ReduceInferMeta
   kernel : 
-    func : reduce_mean
+    func : mean
     param : [x, axis, keep_dim, false, x.dtype(), DataType::UNDEFINED]
 
 - api : multiply
@@ -94,7 +94,7 @@
     func : ElementwiseInferMeta
     param : [x, y, -1]
   kernel :
-    func : elementwise_mul
+    func : multiply
     param : [x, y, -1]
 
 - api : ones_like
@@ -108,7 +108,7 @@
   infer_meta : 
     func : InferMetaFromVecValue
   kernel : 
-    func : reshape2
+    func : reshape
 
 - api : scale
   args : (const Tensor& x, const Scalar& scale, float bias, bool bias_after_scale)
@@ -126,7 +126,7 @@
     func : ElementwiseInferMeta
     param : [x, y, -1]
   kernel :
-    func : elementwise_sub
+    func : subtract
     param : [x, y, -1]
 
 - api : sum
@@ -135,7 +135,7 @@
   infer_meta : 
     func : ReduceInferMeta
   kernel : 
-    func : reduce_sum
+    func : sum
     param : [x, axis, keep_dim, false, x.dtype(), DataType::UNDEFINED]
 
 - api : zeros_like