engine.h 6.4 KB
Newer Older
Y
Yan Chunwei 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
/* 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. */

#pragma once

#include <NvInfer.h>
#include <memory>
19
#include <string>
Y
Yan Chunwei 已提交
20
#include <unordered_map>
21
#include <vector>
Y
Yan Chunwei 已提交
22 23
#include "paddle/fluid/inference/engine.h"
#include "paddle/fluid/inference/tensorrt/helper.h"
24
#include "paddle/fluid/inference/utils/singleton.h"
Y
Yan Chunwei 已提交
25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40

namespace paddle {
namespace inference {
namespace tensorrt {

/*
 * TensorRT Engine.
 *
 * There are two alternative ways to use it, one is  to build from a paddle
 * protobuf model, another way is to manully construct the network.
 */
class TensorRTEngine : public EngineBase {
 public:
  // Weight is model parameter.
  class Weight {
   public:
41
    Weight(nvinfer1::DataType dtype, void* value, size_t num_elem) {
Y
Yan Chunwei 已提交
42 43 44 45 46 47
      w_.type = dtype;
      w_.values = value;
      w_.count = num_elem;
    }
    const nvinfer1::Weights& get() { return w_; }

48 49
    std::vector<int64_t> dims;

Y
Yan Chunwei 已提交
50 51 52 53 54 55 56 57 58 59 60 61 62 63
   private:
    nvinfer1::Weights w_;
  };

  TensorRTEngine(int max_batch, int max_workspace, cudaStream_t* stream,
                 nvinfer1::ILogger& logger = NaiveLogger::Global())
      : max_batch_(max_batch),
        max_workspace_(max_workspace),
        stream_(stream),
        logger_(logger) {}

  virtual ~TensorRTEngine();

  // TODO(Superjomn) implement it later when graph segmentation is supported.
64
  void Build(const DescType& paddle_model) override;
Y
Yan Chunwei 已提交
65

66
  void Execute(int batch_size) override;
Y
Yan Chunwei 已提交
67 68 69 70

  // Initialize the inference network, so that TensorRT layers can add to this
  // network.
  void InitNetwork() {
71
    infer_builder_.reset(createInferBuilder(&logger_));
Y
Yan Chunwei 已提交
72 73 74 75 76 77 78 79 80 81 82 83 84 85
    infer_network_.reset(infer_builder_->createNetwork());
  }
  // After finishing adding ops, freeze this network and creates the executation
  // environment.
  void FreezeNetwork();

  // Add an input and set its name, data type and dimention.
  nvinfer1::ITensor* DeclareInput(const std::string& name,
                                  nvinfer1::DataType dtype,
                                  const nvinfer1::Dims& dim);
  // Set the offset-th output from a layer as the network's output, and set its
  // name.
  void DeclareOutput(const nvinfer1::ILayer* layer, int offset,
                     const std::string& name);
L
Luo Tao 已提交
86 87
  // Set the itensor_map_[name] as the network's output, and set its name.
  void DeclareOutput(const std::string& name);
Y
Yan Chunwei 已提交
88 89 90 91 92

  // GPU memory address for an ITensor with specific name. One can operate on
  // these memory directly for acceleration, for example, output the converted
  // data directly to the buffer to save data copy overhead.
  // NOTE this should be used after calling `FreezeNetwork`.
Y
Yan Chunwei 已提交
93 94 95
  Buffer& buffer(const std::string& name) override;

  cudaStream_t* stream() { return stream_; }
Y
Yan Chunwei 已提交
96 97

  // Fill an input from CPU memory with name and size.
98
  void SetInputFromCPU(const std::string& name, const void* data, size_t size);
Y
Yan Chunwei 已提交
99 100
  // TODO(Superjomn) is this method necessary given that buffer(xxx) can be
  // accessed directly. Fill an input from GPU memory with name and size.
101
  void SetInputFromGPU(const std::string& name, const void* data, size_t size);
Y
Yan Chunwei 已提交
102
  // Get an output called name, the output of tensorrt is in GPU, so this method
103
  // Return the output's GPU memory address without copy.
Y
Yan Chunwei 已提交
104
  void* GetOutputInGPU(const std::string& name);
105 106
  // Copy data into dst inside the GPU device.
  void GetOutputInGPU(const std::string& name, void* dst, size_t max_size);
Y
Yan Chunwei 已提交
107 108 109
  // LOW EFFICENCY! Get output to CPU, this will trigger a memory copy from GPU
  // to CPU.
  void GetOutputInCPU(const std::string& name, void* dst, size_t max_size);
L
Luo Tao 已提交
110 111 112 113
  // Fill an ITensor into map itensor_map_.
  void SetITensor(const std::string& name, nvinfer1::ITensor* tensor);
  // Get an ITensor called name.
  nvinfer1::ITensor* GetITensor(const std::string& name);
Y
Yan Chunwei 已提交
114 115 116 117 118 119 120 121 122 123 124 125

  nvinfer1::ICudaEngine* engine() { return infer_engine_.get(); }
  nvinfer1::INetworkDefinition* network() { return infer_network_.get(); }

 private:
  // the max batch size
  int max_batch_;
  // the max memory size the engine uses
  int max_workspace_;
  cudaStream_t* stream_;
  nvinfer1::ILogger& logger_;

Y
Yan Chunwei 已提交
126
  std::vector<Buffer> buffers_;
Y
Yan Chunwei 已提交
127 128
  // max data size for the buffers.
  std::unordered_map<std::string /*name*/, size_t /*max size*/> buffer_sizes_;
L
Luo Tao 已提交
129 130
  std::unordered_map<std::string /*name*/, nvinfer1::ITensor* /*ITensor*/>
      itensor_map_;
Y
Yan Chunwei 已提交
131 132 133 134

  // TensorRT related internal members
  template <typename T>
  struct Destroyer {
135 136 137 138 139
    void operator()(T* x) {
      if (x) {
        x->destroy();
      }
    }
Y
Yan Chunwei 已提交
140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162
  };
  template <typename T>
  using infer_ptr = std::unique_ptr<T, Destroyer<T>>;
  infer_ptr<nvinfer1::IBuilder> infer_builder_;
  infer_ptr<nvinfer1::INetworkDefinition> infer_network_;
  infer_ptr<nvinfer1::ICudaEngine> infer_engine_;
  infer_ptr<nvinfer1::IExecutionContext> infer_context_;
};  // class TensorRTEngine

// Add an layer__ into engine__ with args ARGS.
// For example:
//   TRT_ENGINE_ADD_LAYER(xxx, FullyConnected, input, dim, weights, bias)
//
// Reference
// https://docs.nvidia.com/deeplearning/sdk/tensorrt-developer-guide/index.html#charRNN_define_network
//
// will add a fully connected layer into the engine.
// TensorRT has too many layers, so that is not wise to add member functions for
// them, and an macro like this is more extensible when underlying TensorRT
// library add new layer supports.
#define TRT_ENGINE_ADD_LAYER(engine__, layer__, ARGS...) \
  engine__->network()->add##layer__(ARGS);

163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183
/*
 * Helper to control the TensorRT engine's creation and deletion.
 */
class TRT_EngineManager {
 public:
  TensorRTEngine* Create(int max_batch, int max_workspace,
                         cudaStream_t* stream) {
    engines_.emplace_back(new TensorRTEngine(max_batch, max_workspace, stream));
    return engines_.back().get();
  }

  void DeleteALl() {
    for (auto& ptr : engines_) {
      ptr.reset(nullptr);
    }
  }

 private:
  std::vector<std::unique_ptr<TensorRTEngine>> engines_;
};

Y
Yan Chunwei 已提交
184 185 186
}  // namespace tensorrt
}  // namespace inference
}  // namespace paddle