1 /**
2 * Copyright 2021 Huawei Technologies Co., Ltd
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 * http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17 #include "src/delegate/tensorrt/op/activation_tensorrt.h"
18
19 namespace mindspore::lite {
IsSupport(const schema::Primitive * primitive,const std::vector<mindspore::MSTensor> & in_tensors,const std::vector<mindspore::MSTensor> & out_tensors)20 int ActivationTensorRT::IsSupport(const schema::Primitive *primitive,
21 const std::vector<mindspore::MSTensor> &in_tensors,
22 const std::vector<mindspore::MSTensor> &out_tensors) {
23 if (!IsShapeKnown()) {
24 MS_LOG(ERROR) << "Unsupported input tensor unknown shape: " << op_name_;
25 return RET_ERROR;
26 }
27 if (in_tensors.size() != 1) {
28 MS_LOG(ERROR) << "Unsupported input tensor size, size is " << in_tensors.size();
29 return RET_ERROR;
30 }
31 if (out_tensors.size() != 1) {
32 MS_LOG(ERROR) << "Unsupported output tensor size, size is " << out_tensors.size();
33 return RET_ERROR;
34 }
35 auto activation_op = this->op_primitive_->value_as_Activation();
36 if (activation_op == nullptr) {
37 MS_LOG(ERROR) << "op convert failed";
38 return RET_ERROR;
39 }
40 this->action_code_ = ConvertActivationType(activation_op->activation_type()).activation_type;
41 if (this->action_code_ == nvinfer1::ActivationType::kRELU &&
42 activation_op->activation_type() != schema::ActivationType_RELU) {
43 MS_LOG(ERROR) << "Unsupported op action type for TensorRT: " << activation_op->activation_type();
44 return RET_ERROR;
45 }
46 return RET_OK;
47 }
AddInnerOp(nvinfer1::INetworkDefinition * network)48 int ActivationTensorRT::AddInnerOp(nvinfer1::INetworkDefinition *network) {
49 if (network == nullptr) {
50 MS_LOG(ERROR) << "network is invalid";
51 return RET_ERROR;
52 }
53 auto activation_op = this->op_primitive_->value_as_Activation();
54 if (activation_op == nullptr) {
55 MS_LOG(ERROR) << "op convert failed";
56 return RET_ERROR;
57 }
58 float alpha = activation_op->alpha();
59
60 nvinfer1::IActivationLayer *activation_layer = ActivationTensorRT::AddActivation(
61 network, activation_op->activation_type(), alpha, tensorrt_in_tensors_[0].trt_tensor_);
62 if (activation_layer == nullptr) {
63 MS_LOG(ERROR) << "add activation op failed for TensorRT.";
64 return RET_ERROR;
65 }
66
67 activation_layer->setName(op_name_.c_str());
68 activation_layer->getOutput(0)->setName((op_name_ + "_output").c_str());
69 this->AddInnerOutTensors(ITensorHelper{activation_layer->getOutput(0), tensorrt_in_tensors_[0].format_});
70
71 return RET_OK;
72 }
AddActivation(nvinfer1::INetworkDefinition * network,schema::ActivationType activation_type,float alpha,nvinfer1::ITensor * trt_in_tensor)73 nvinfer1::IActivationLayer *ActivationTensorRT::AddActivation(nvinfer1::INetworkDefinition *network,
74 schema::ActivationType activation_type, float alpha,
75 nvinfer1::ITensor *trt_in_tensor) {
76 // Just some action_code correct, unfind code is set to default relu. need double check.
77 lite::ActivationParams action_param = ConvertActivationType(activation_type);
78 if (action_param.activation_type == nvinfer1::ActivationType::kRELU &&
79 activation_type != schema::ActivationType_RELU) {
80 MS_LOG(ERROR) << "Unsupported op action type for TensorRT: " << activation_type;
81 return nullptr;
82 }
83 nvinfer1::IActivationLayer *activation_layer = network->addActivation(*trt_in_tensor, action_param.activation_type);
84 if (activation_layer == nullptr) {
85 MS_LOG(ERROR) << "add activation op failed for TensorRT.";
86 return nullptr;
87 }
88
89 if (action_param.has_alpha) {
90 activation_layer->setAlpha(alpha);
91 }
92
93 if (action_param.has_beta) {
94 activation_layer->setBeta(action_param.beta);
95 }
96
97 return activation_layer;
98 }
99 } // namespace mindspore::lite
100