add reader & infer op for general server

examples/demo-serving/op/general_infer_op.cpp

+// Copyright (c) 2020 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 "examples/demo-serving/op/general_infer_op.h"
+#include <algorithm>
+#include <iostream>
+#include <memory>
+#include <sstream>
+#include "core/predictor/framework/infer.h"
+#include "core/predictor/framework/memory.h"
+#include "core/predictor/framework/resource.h"
+#include "examples/demo-serving/op/general_reader_op.h"
+
+namespace baidu {
+namespace paddle_serving {
+namespace serving {
+
+using baidu::paddle_serving::predictor::MempoolWrapper;
+using baidu::paddle_serving::predictor::general_model::Tensor;
+using baidu::paddle_serving::predictor::general_model::Response;
+using baidu::paddle_serving::predictor::general_model::FetchInst;
+using baidu::paddle_serving::predictor::InferManager;
+
+int GeneralInferOp::inference() {
+  const GeneralReaderOutput *reader_out =
+      get_depend_argument<GeneralReaderOutput>("general_reader_op");
+  if (!reader_out) {
+    LOG(ERROR) << "Failed mutable depended argument, op:"
+               << "general_reader_op";
+    return -1;
+  }
+
+  int reader_status = reader_out->reader_status;
+  if (reader_status != 0) {
+    LOG(ERROR) << "Read request wrong.";
+    return -1;
+  }
+
+  const TensorVector *in = &reader_out->tensor_vector;
+  TensorVector *out = butil::get_object<TensorVector>();
+  int batch_size = (*in)[0].shape[0];
+  // infer
+  if (InferManager::instance().infer(GENERAL_MODEL_NAME, in, out, batch_size)) {
+    LOG(ERROR) << "Failed do infer in fluid model: " << GENERAL_MODEL_NAME;
+    return -1;
+  }
+
+  Response *res = mutable_data<Response>();
+
+  for (int i = 0; i < batch_size; ++i) {
+    FetchInst *fetch_inst = res->add_insts();
+    for (int j = 0; j < out->size(); ++j) {
+      Tensor *tensor = fetch_inst->add_tensor_array();
+      tensor->set_elem_type(1);
+      if (out->at(j).lod.size() == 1) {
+        tensor->add_shape(-1);
+      } else {
+        for (int k = 1; k < out->at(j).shape.size(); ++k) {
+          tensor->add_shape(out->at(j).shape[k]);
+        }
+      }
+    }
+  }
+
+  for (int i = 0; i < out->size(); ++i) {
+    float *data_ptr = static_cast<float *>(out->at(i).data.data());
+    int cap = 1;
+    for (int j = 1; j < out->at(i).shape.size(); ++j) {
+      cap *= out->at(i).shape[j];
+    }
+    if (out->at(i).lod.size() == 1) {
+      for (int j = 0; j < batch_size; ++j) {
+        for (int k = out->at(i).lod[0][j]; k < out->at(i).lod[0][j + 1]; k++) {
+          res->mutable_insts(j)->mutable_tensor_array(i)->add_data(
+              reinterpret_cast<char *>(&(data_ptr[k])), sizeof(float));
+        }
+      }
+    } else {
+      for (int j = 0; j < batch_size; ++j) {
+        for (int k = j * cap; k < (j + 1) * cap; ++k) {
+          res->mutable_insts(j)->mutable_tensor_array(i)->add_data(
+              reinterpret_cast<char *>(&(data_ptr[k])), sizeof(float));
+        }
+      }
+    }
+  }
+  /*
+      for (size_t i = 0; i < in->size(); ++i) {
+        (*in)[i].shape.clear();
+      }
+      in->clear();
+      butil::return_object<TensorVector>(in);
+
+      for (size_t i = 0; i < out->size(); ++i) {
+        (*out)[i].shape.clear();
+      }
+      out->clear();
+      butil::return_object<TensorVector>(out);
+    }
+  */
+  return 0;
+}
+DEFINE_OP(GeneralInferOp);
+
+}  // namespace serving
+}  // namespace paddle_serving
+}  // namespace baidu

examples/demo-serving/op/general_infer_op.h

+// 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 <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 "examples/demo-serving/general_model_service.pb.h"
+
+namespace baidu {
+namespace paddle_serving {
+namespace serving {
+
+static const char* GENERAL_MODEL_NAME = "general_model";
+
+class GeneralInferOp
+    : public baidu::paddle_serving::predictor::OpWithChannel<
+          baidu::paddle_serving::predictor::general_model::Response> {
+ public:
+  typedef std::vector<paddle::PaddleTensor> TensorVector;
+
+  DECLARE_OP(GeneralInferOp);
+
+  int inference();
+};
+
+}  // namespace serving
+}  // namespace paddle_serving
+}  // namespace baidu

examples/demo-serving/op/general_reader_op.cpp

+// 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.
+
+#include "examples/demo-serving/op/general_reader_op.h"
+#include <algorithm>
+#include <iostream>
+#include <memory>
+#include <sstream>
+#include "core/predictor/framework/infer.h"
+#include "core/predictor/framework/memory.h"
+
+namespace baidu {
+namespace paddle_serving {
+namespace serving {
+
+using baidu::paddle_serving::predictor::MempoolWrapper;
+using baidu::paddle_serving::predictor::general_model::Tensor;
+using baidu::paddle_serving::predictor::general_model::Request;
+using baidu::paddle_serving::predictor::general_model::FeedInst;
+using baidu::paddle_serving::predictor::PaddleGeneralModelConfig;
+
+int conf_check(const Request *req,
+               const std::shared_ptr<PaddleGeneralModelConfig> &model_config) {
+  int var_num = req->insts(0).tensor_array_size();
+  if (var_num != model_config->_feed_type.size()) {
+    LOG(ERROR) << "feed var number not match.";
+    return -1;
+  }
+  for (int i = 0; i < var_num; ++i) {
+    if (model_config->_feed_type[i] !=
+        req->insts(0).tensor_array(i).elem_type()) {
+      LOG(ERROR) << "feed type not match.";
+      return -1;
+    }
+    if (model_config->_feed_shape[i].size() ==
+        req->insts(0).tensor_array(i).shape_size()) {
+      for (int j = 0; j < model_config->_feed_shape[i].size(); ++j) {
+        req->insts(0).tensor_array(i).shape(j);
+        if (model_config->_feed_shape[i][j] !=
+            req->insts(0).tensor_array(i).shape(j)) {
+          LOG(ERROR) << "feed shape not match.";
+          return -1;
+        }
+      }
+    } else {
+      LOG(ERROR) << "feed shape not match.";
+      return -1;
+    }
+  }
+  return 0;
+}
+
+int GeneralReaderOp::inference() {
+  // reade request from client
+  const Request *req = dynamic_cast<const Request *>(get_request_message());
+
+  int batch_size = req->insts_size();
+  int input_var_num = 0;
+
+  std::vector<int64_t> elem_type;
+  std::vector<int64_t> elem_size;
+  std::vector<int64_t> capacity;
+
+  GeneralReaderOutput *res = mutable_data<GeneralReaderOutput>();
+  TensorVector *in = &res->tensor_vector;
+
+  if (!res) {
+    LOG(ERROR) << "Failed get op tls reader object output";
+  }
+  if (batch_size <= 0) {
+    res->reader_status = -1;
+    return 0;
+  }
+
+  int var_num = req->insts(0).tensor_array_size();
+  VLOG(3) << "var num: " << var_num;
+  // read config
+
+  LOG(INFO) << "start to call load general model_conf op";
+  baidu::paddle_serving::predictor::Resource &resource =
+      baidu::paddle_serving::predictor::Resource::instance();
+
+  LOG(INFO) << "get resource pointer done.";
+  std::shared_ptr<PaddleGeneralModelConfig> model_config =
+      resource.get_general_model_config();
+
+  LOG(INFO) << "print general model config done.";
+
+  // check
+  res->reader_status = conf_check(req, model_config);
+  if (res->reader_status != 0) {
+    LOG(INFO) << "model conf of server:";
+    resource.print_general_model_config(model_config);
+    return 0;
+  }
+  // package tensor
+
+  elem_type.resize(var_num);
+  elem_size.resize(var_num);
+  capacity.resize(var_num);
+  paddle::PaddleTensor lod_tensor;
+  for (int i = 0; i < var_num; ++i) {
+    elem_type[i] = req->insts(0).tensor_array(i).elem_type();
+    VLOG(3) << "var[" << i << "] has elem type: " << elem_type[i];
+    if (elem_type[i] == 0) {  // int64
+      elem_size[i] = sizeof(int64_t);
+      lod_tensor.dtype = paddle::PaddleDType::INT64;
+    } else {
+      elem_size[i] = sizeof(float);
+      lod_tensor.dtype = paddle::PaddleDType::FLOAT32;
+    }
+
+    if (req->insts(0).tensor_array(i).shape(0) == -1) {
+      lod_tensor.lod.resize(1);
+      lod_tensor.lod[0].push_back(0);
+      VLOG(3) << "var[" << i << "] is lod_tensor";
+    } else {
+      lod_tensor.shape.push_back(batch_size);
+      capacity[i] = 1;
+      for (int k = 0; k < req->insts(0).tensor_array(i).shape_size(); ++k) {
+        int dim = req->insts(0).tensor_array(i).shape(k);
+        VLOG(3) << "shape for var[" << i << "]: " << dim;
+        capacity[i] *= dim;
+        lod_tensor.shape.push_back(dim);
+      }
+      VLOG(3) << "var[" << i << "] is tensor, capacity: " << capacity[i];
+    }
+    if (i == 0) {
+      lod_tensor.name = "words";
+    } else {
+      lod_tensor.name = "label";
+    }
+    in->push_back(lod_tensor);
+  }
+
+  for (int i = 0; i < var_num; ++i) {
+    if (in->at(i).lod.size() == 1) {
+      for (int j = 0; j < batch_size; ++j) {
+        const Tensor &tensor = req->insts(j).tensor_array(i);
+        int data_len = tensor.data_size();
+        VLOG(3) << "tensor size for var[" << i << "]: " << tensor.data_size();
+        int cur_len = in->at(i).lod[0].back();
+        VLOG(3) << "current len: " << cur_len;
+        in->at(i).lod[0].push_back(cur_len + data_len);
+        VLOG(3) << "new len: " << cur_len + data_len;
+      }
+      in->at(i).data.Resize(in->at(i).lod[0].back() * elem_size[i]);
+      in->at(i).shape = {in->at(i).lod[0].back(), 1};
+      VLOG(3) << "var[" << i
+              << "] is lod_tensor and len=" << in->at(i).lod[0].back();
+    } else {
+      in->at(i).data.Resize(batch_size * capacity[i] * elem_size[i]);
+      VLOG(3) << "var[" << i
+              << "] is tensor and capacity=" << batch_size * capacity[i];
+    }
+  }
+
+  for (int i = 0; i < var_num; ++i) {
+    if (elem_type[i] == 0) {
+      int64_t *dst_ptr = static_cast<int64_t *>(in->at(i).data.data());
+      int offset = 0;
+      for (int j = 0; j < batch_size; ++j) {
+        for (int k = 0; k < req->insts(j).tensor_array(i).data_size(); ++k) {
+          dst_ptr[offset + k] =
+              *(const int64_t *)req->insts(j).tensor_array(i).data(k).c_str();
+        }
+        if (in->at(i).lod.size() == 1) {
+          offset = in->at(i).lod[0][j + 1];
+        } else {
+          offset += capacity[i];
+        }
+      }
+    } else {
+      float *dst_ptr = static_cast<float *>(in->at(i).data.data());
+      int offset = 0;
+      for (int j = 0; j < batch_size; ++j) {
+        for (int k = 0; k < req->insts(j).tensor_array(i).data_size(); ++k) {
+          dst_ptr[offset + k] =
+              *(const float *)req->insts(j).tensor_array(i).data(k).c_str();
+        }
+        if (in->at(i).lod.size() == 1) {
+          offset = in->at(i).lod[0][j + 1];
+        } else {
+          offset += capacity[i];
+        }
+      }
+    }
+  }
+
+  VLOG(3) << "read data from client success";
+  // print request
+  std::ostringstream oss;
+  int64_t *example = reinterpret_cast<int64_t *>((*in)[0].data.data());
+  for (int i = 0; i < 10; i++) {
+    oss << *(example + i) << " ";
+  }
+  VLOG(3) << "head element of first feed var : " << oss.str();
+  //
+  return 0;
+}
+DEFINE_OP(GeneralReaderOp);
+}  // namespace serving
+}  // namespace paddle_serving
+}  // namespace baidu

examples/demo-serving/op/general_reader_op.h

+// 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 <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 <string>
+#include "core/predictor/framework/resource.h"
+#include "examples/demo-serving/general_model_service.pb.h"
+#include "examples/demo-serving/load_general_model_service.pb.h"
+
+namespace baidu {
+namespace paddle_serving {
+namespace serving {
+
+struct GeneralReaderOutput {
+  std::vector<paddle::PaddleTensor> tensor_vector;
+  int reader_status = 0;
+
+  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();
+  }
+  std::string ShortDebugString() const { return "Not implemented!"; }
+};
+
+class GeneralReaderOp : public baidu::paddle_serving::predictor::OpWithChannel<
+                            GeneralReaderOutput> {
+ public:
+  typedef std::vector<paddle::PaddleTensor> TensorVector;
+
+  DECLARE_OP(GeneralReaderOp);
+
+  int inference();
+};
+
+}  // namespace serving
+}  // namespace paddle_serving
+}  // namespace baidu