split_op_plugin.h 2.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13
// 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.
N
nhzlx 已提交
14 15 16

#pragma once

H
hjchen2 已提交
17
#include <thrust/device_vector.h>
N
nhzlx 已提交
18
#include <utility>
19
#include <vector>
20
#include "paddle/fluid/inference/tensorrt/plugin/trt_plugin.h"
N
nhzlx 已提交
21 22 23 24

namespace paddle {
namespace inference {
namespace tensorrt {
25
namespace plugin {
N
nhzlx 已提交
26 27

class SplitPlugin : public PluginTensorRT {
28
 public:
N
nhzlx 已提交
29
  SplitPlugin() {}
30
  SplitPlugin(int axis, std::vector<int> const &output_lengths)
H
hjchen2 已提交
31
      : axis_(axis), same_shape_(true), output_length_(output_lengths) {}
32 33 34 35 36 37 38 39 40 41 42

  SplitPlugin(void const *serial_data, size_t serial_length) {
    deserializeBase(serial_data, serial_length);
    DeserializeValue(&serial_data, &serial_length, &axis_);
    DeserializeValue(&serial_data, &serial_length, &output_length_);
  }

  SplitPlugin *clone() const override {
    return new SplitPlugin(axis_, output_length_);
  }

N
nhzlx 已提交
43
  const char *getPluginType() const override { return "split_plugin"; }
44 45 46 47 48 49 50 51
  int getNbOutputs() const override { return output_length_.size(); }
  nvinfer1::Dims getOutputDimensions(int index,
                                     const nvinfer1::Dims *input_dims,
                                     int num_inputs) override;

  int initialize() override;
  int enqueue(int batchSize, const void *const *inputs, void **outputs,
              void *workspace, cudaStream_t stream) override;
N
nhzlx 已提交
52 53

 protected:
54
  size_t getSerializationSize() override {
N
nhzlx 已提交
55 56
    return SerializedSize(getPluginType()) + SerializedSize(axis_) +
           SerializedSize(output_length_) + getBaseSerializationSize();
N
nhzlx 已提交
57 58
  }

59
  void serialize(void *buffer) override {
N
nhzlx 已提交
60
    SerializeValue(&buffer, getPluginType());
N
nhzlx 已提交
61
    serializeBase(buffer);
N
nhzlx 已提交
62 63
    SerializeValue(&buffer, axis_);
    SerializeValue(&buffer, output_length_);
N
nhzlx 已提交
64 65
  }

66
  int axis_;
H
hjchen2 已提交
67 68 69
  int outer_rows_;
  int inner_cols_;
  bool same_shape_;
70 71
  std::vector<int> output_length_;
  std::vector<int> segment_offsets_;
H
hjchen2 已提交
72 73
  thrust::device_vector<int> d_segment_offsets_;
  thrust::device_vector<float *> d_output_ptrs_;
N
nhzlx 已提交
74 75
};

76 77 78 79
}  // namespace plugin
}  // namespace tensorrt
}  // namespace inference
}  // namespace paddle