提交 7382f986 编写于 作者: N nhzlx

1. set ut batch > 1 2. readd the mul op(utest will be added later)

上级 4e377f8e
/* Copyright (c) 2018 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. */
#include "paddle/fluid/inference/tensorrt/convert/op_converter.h"
namespace paddle {
namespace inference {
namespace tensorrt {
/*
* MulOp, IMatrixMultiplyLayer in TRT. This Layer doesn't has weights.
*/
class MulOpConverter : public OpConverter {
public:
void operator()(const framework::proto::OpDesc& op,
const framework::Scope& scope, bool test_mode) override {
VLOG(4) << "convert a fluid mul op to tensorrt mul layer without bias";
framework::OpDesc op_desc(op, nullptr);
// Declare inputs
auto* input1 = engine_->GetITensor(op_desc.Input("X")[0]);
auto* input2 = engine_->GetITensor(op_desc.Input("Y")[0]);
// Both the input1 and input2 do not need transpose.
auto* layer = TRT_ENGINE_ADD_LAYER(
engine_, MatrixMultiply, *const_cast<nvinfer1::ITensor*>(input1), false,
*const_cast<nvinfer1::ITensor*>(input2), false);
auto output_name = op_desc.Output("Out")[0];
engine_->SetITensor(output_name, layer->getOutput(0));
if (test_mode) { // the test framework can not determine which is the
// output, so place the declaration inside.
engine_->DeclareOutput(output_name);
}
}
};
} // namespace tensorrt
} // namespace inference
} // namespace paddle
USE_OP(mul);
// TODO(xingzhaolong): change the name to mul then
REGISTER_TRT_OP_CONVERTER(mul_temp, MulOpConverter);
...@@ -23,7 +23,7 @@ namespace tensorrt { ...@@ -23,7 +23,7 @@ namespace tensorrt {
TEST(ReluOpConverter, main) { TEST(ReluOpConverter, main) {
framework::Scope scope; framework::Scope scope;
std::unordered_set<std::string> parameters; std::unordered_set<std::string> parameters;
TRTConvertValidation validator(1, parameters, scope, 1000); TRTConvertValidation validator(10, parameters, scope, 1000);
validator.DeclInputVar("relu-X", nvinfer1::Dims2(10, 6)); validator.DeclInputVar("relu-X", nvinfer1::Dims2(10, 6));
validator.DeclOutputVar("relu-Out", nvinfer1::Dims2(10, 6)); validator.DeclOutputVar("relu-Out", nvinfer1::Dims2(10, 6));
......
...@@ -23,7 +23,7 @@ namespace tensorrt { ...@@ -23,7 +23,7 @@ namespace tensorrt {
TEST(fc_op, test) { TEST(fc_op, test) {
std::unordered_set<std::string> parameters({"mul-Y"}); std::unordered_set<std::string> parameters({"mul-Y"});
framework::Scope scope; framework::Scope scope;
TRTConvertValidation validator(1, parameters, scope, 1000); TRTConvertValidation validator(10, parameters, scope, 1000);
validator.DeclInputVar("mul-X", nvinfer1::Dims4(1, 10, 1, 1)); validator.DeclInputVar("mul-X", nvinfer1::Dims4(1, 10, 1, 1));
validator.DeclParamVar("mul-Y", nvinfer1::Dims2(10, 2)); validator.DeclParamVar("mul-Y", nvinfer1::Dims2(10, 2));
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册