如何开发一个新的General Op?

// Copyright (c) 2019 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 <string>
#include <vector>
#ifdef BCLOUD
#ifdef WITH_GPU
#include "paddle/paddle_inference_api.h"
#else
#include "paddle/fluid/inference/api/paddle_inference_api.h"
#endif
#else
#include "paddle_inference_api.h"  // NOLINT
#endif
#include "core/general-server/general_model_service.pb.h"
#include "core/general-server/op/general_infer_helper.h"

namespace baidu {
namespace paddle_serving {
namespace serving {

class GeneralInferOp
    : public baidu::paddle_serving::predictor::OpWithChannel<GeneralBlob> {
 public:
  typedef std::vector<paddle::PaddleTensor> TensorVector;

  DECLARE_OP(GeneralInferOp);

  int inference();

};

}  // namespace serving
}  // namespace paddle_serving
}  // namespace baidu

struct GeneralBlob {
  std::vector<paddle::PaddleTensor> tensor_vector;
  int64_t time_stamp[20];
  int p_size = 0;

  int _batch_size;

  void Clear() {
    size_t tensor_count = tensor_vector.size();
    for (size_t ti = 0; ti < tensor_count; ++ti) {
      tensor_vector[ti].shape.clear();
    }
    tensor_vector.clear();
  }

  int SetBatchSize(int batch_size) { _batch_size = batch_size; }

  int GetBatchSize() const { return _batch_size; }
  std::string ShortDebugString() const { return "Not implemented!"; }
};

int GeneralInferOp::inference() {
  VLOG(2) << "Going to run inference";
  const GeneralBlob *input_blob = get_depend_argument<GeneralBlob>(pre_name());
  VLOG(2) << "Get precedent op name: " << pre_name();
  GeneralBlob *output_blob = mutable_data<GeneralBlob>();

  if (!input_blob) {
    LOG(ERROR) << "Failed mutable depended argument, op:" << pre_name();
    return -1;
  }

  const TensorVector *in = &input_blob->tensor_vector;
  TensorVector *out = &output_blob->tensor_vector;
  int batch_size = input_blob->GetBatchSize();
  VLOG(2) << "input batch size: " << batch_size;

  output_blob->SetBatchSize(batch_size);

  VLOG(2) << "infer batch size: " << batch_size;

  Timer timeline;
  int64_t start = timeline.TimeStampUS();
  timeline.Start();

  if (InferManager::instance().infer(engine_name().c_str(), in, out, batch_size)) {
    LOG(ERROR) << "Failed do infer in fluid model: " << engine_name().c_str();
    return -1;
  }

  int64_t end = timeline.TimeStampUS();
  CopyBlobInfo(input_blob, output_blob);
  AddBlobInfo(output_blob, start);
  AddBlobInfo(output_blob, end);
  return 0;
}
DEFINE_OP(GeneralInferOp);

self.op_dict = {
            "general_infer": "GeneralInferOp",
            "general_reader": "GeneralReaderOp",
            "general_response": "GeneralResponseOp",
            "general_text_reader": "GeneralTextReaderOp",
            "general_text_response": "GeneralTextResponseOp",
            "general_single_kv": "GeneralSingleKVOp",
            "general_dist_kv": "GeneralDistKVOp"
        }

default_engine_types = [
                'GeneralInferOp',
                'GeneralDistKVInferOp',
                'GeneralDistKVQuantInferOp',
                'GeneralDetectionOp',
            ]