gelu_op.cc 2.7 KB
Newer Older
P
Pei Yang 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
/* 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"
#include "paddle/fluid/inference/tensorrt/plugin/gelu_op_plugin.h"

namespace paddle {
namespace inference {
namespace tensorrt {

22 23 24
/*
 * Gelu converter from fluid to tensorRT.
 */
P
Pei Yang 已提交
25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45
/*
 * Gelu converter from fluid to tensorRT.
 */
class GeluOpConverter : public OpConverter {
 public:
  void operator()(const framework::proto::OpDesc& op,
                  const framework::Scope& scope, bool test_mode) override {
    VLOG(4) << "convert fluid gelu op to tensorrt gelu layer";

    framework::OpDesc op_desc(op, nullptr);
    // Declare inputs
    int input_num = op_desc.Input("X").size();
    PADDLE_ENFORCE_EQ(input_num, 1,
                      platform::errors::InvalidArgument(
                          "gelu op has only 1 input, but got %d", input_num));
    auto* input = engine_->GetITensor(op_desc.Input("X")[0]);
    // Get output
    size_t output_num = op_desc.Output("Out").size();
    PADDLE_ENFORCE_EQ(output_num, 1,
                      platform::errors::InvalidArgument(
                          "gelu op has only 1 output, but got %d", output_num));
46 47 48 49

    nvinfer1::ILayer* layer = nullptr;
    if (engine_->with_dynamic_shape()) {
#if IS_TRT_VERSION_GE(6000)
S
Shang Zhizhou 已提交
50 51 52 53
      bool with_fp16 =
          engine_->WithFp16() && !engine_->disable_trt_plugin_fp16();
      plugin::GeluPluginDynamic* plugin =
          new plugin::GeluPluginDynamic(with_fp16);
54 55 56 57 58 59 60
      layer = engine_->AddPluginV2(&input, input_num, plugin);
#else
      PADDLE_THROW(platform::errors::Fatal(
          "You are running the TRT Dynamic Shape mode, need to confirm that "
          "your TRT version is no less than 6.0"));
#endif
    } else {
S
Shang Zhizhou 已提交
61 62 63
      bool with_fp16 =
          engine_->WithFp16() && !engine_->disable_trt_plugin_fp16();
      plugin::GeluPlugin* plugin = new plugin::GeluPlugin(with_fp16);
64
      layer = engine_->AddPlugin(&input, input_num, plugin);
P
Pei Yang 已提交
65 66 67 68 69 70 71 72 73 74 75
    }
    auto output_name = op_desc.Output("Out")[0];
    RreplenishLayerAndOutput(layer, "gelu", {output_name}, test_mode);
  }
};

}  // namespace tensorrt
}  // namespace inference
}  // namespace paddle

REGISTER_TRT_OP_CONVERTER(gelu, GeluOpConverter);