executor.cpp

/* 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 "framework/executor.h"
#include <algorithm>
#include <utility>
#include <vector>
#include "common/enforce.h"
#include "common/log.h"
#include "framework/framework.pb-c.h"
#include "framework/lod_tensor.h"
#include "framework/operator.h"
#include "framework/program/program-optimize/program_optimize.h"
#include "framework/program/program_desc.h"
#include "framework/program/var_desc.h"
#include "framework/scope.h"
#include "framework/tensor.h"
#include "memory/t_malloc.h"

#ifdef PADDLE_MOBILE_CL
#include "framework/cl/cl_image.h"
#endif

namespace paddle_mobile {
namespace framework {

#pragma mark - executor

template <typename Device, typename T>
Executor<Device, T>::Executor(const Program<Device> &program,
                              paddle_mobile::PaddleMobileConfigInternal config,
                              int batch_size, const bool use_optimize,
                              const bool lod_mode)
    : program_(program),
      batch_size_(batch_size),
      use_optimize_(use_optimize),
      lod_mode_(lod_mode),
      config_(config) {
  DLOG << "executor in lod mode: " << lod_mode_;

  Variable *variable_ptr = program_.scope->Var("batch_size");
  variable_ptr->SetValue<int>(batch_size);

  program_desc_ =
      use_optimize_ ? program_.optimizeProgram : program_.originProgram;
  PADDLE_MOBILE_ENFORCE(program_desc_ != nullptr,
                        "program_desc_ should not be nullptr");
  const auto &blocks = program_desc_->Blocks();
  ops_of_block_.resize(blocks.size());

  for (int i = 0; i < blocks.size(); ++i) {
    std::shared_ptr<BlockDesc> block_desc = blocks[i];
    std::vector<std::shared_ptr<OpDesc>> ops = block_desc->Ops();
    for (int j = 0; j < ops.size(); ++j) {
      std::shared_ptr<OpDesc> op_desc = ops[j];
      DLOG << "create op: " << op_desc->Type();
      auto op_handler = OpRegistry<Device>::CreateOp(
          op_desc->Type(), op_desc->GetInputs(), op_desc->GetOutputs(),
          op_desc->GetAttrMap(), program_.scope);
      // infer shape to reshape inputs and outputs before predict,
      // but for lod mode, it still need to infer shape in runtime
      if (!lod_mode) {
        op_handler->InferShape();
      }
      ops_of_block_[i].push_back(op_handler);
    }
  }

  if (program_.combined) {
    InitCombineMemory();
  } else {
    InitMemory();
  }

  int count = 0;
  for (int block_id = 0; block_id < ops_of_block_.size(); ++block_id) {
    for (auto &op_handler : ops_of_block_[block_id]) {
      DLOG << "Initialize op[" << count++ << "]: " << op_handler->Type();
      op_handler->Init();
      ops_list_.push_back(op_handler);
    }
  }
}

template <typename T>
static void LoadMemInternal(void **data, LoDTensor *tensor,
                            bool quant_uint8 = false) {
  char **data_buf = reinterpret_cast<char **>(data);
  int64_t size = tensor->numel();
  T *tensor_data = tensor->mutable_data<T>();
  if (quant_uint8) {
    // should be moved into operator init function
    float min_value;
    float max_value;
    memory::Copy(&min_value, *data_buf, sizeof(float));
    memory::Copy(&max_value, *data_buf + sizeof(float), sizeof(float));
    *data_buf += 2 * sizeof(float);
    const float factor = (max_value - min_value) / 255.0;
    const uint8_t *uint8_data = reinterpret_cast<uint8_t *>(*data_buf);
    for (int k = 0; k < size; ++k) {
      tensor_data[k] = uint8_data[k] * factor + min_value;
    }
    *data_buf += size * sizeof(uint8_t);
  } else {
    memory::Copy(tensor_data, *data_buf, size * sizeof(T));
    *data_buf += size * sizeof(T);
  }
}

template <typename Device, typename T>
void Executor<Device, T>::LoadMemory(void **data,
                                     const std::shared_ptr<VarDesc> var_desc,
                                     LoDTensor *tensor) {
  char **data_buf = reinterpret_cast<char **>(data);
  // version
  uint32_t version = *(reinterpret_cast<uint32_t *>(*data_buf));
  *data_buf += sizeof(uint32_t);
  // lod information
  // uint64_t lod_level = *(reinterpret_cast<uint64_t *>(*data_buf));
  uint64_t lod_level = 0;
  memory::Copy(&lod_level, *data_buf, sizeof(uint64_t));
  *data_buf += sizeof(uint64_t);

  auto *lod = tensor->mutable_lod();
  lod->resize(lod_level);
  for (uint64_t i = 0; i < lod_level; ++i) {
    uint64_t size = *(reinterpret_cast<uint64_t *>(*data_buf));
    *data_buf += sizeof(uint64_t);
    std::vector<size_t> tmp_dim(size / sizeof(size_t));
    memory::Copy(tmp_dim.data(), *data_buf, size);
    (*lod)[i] = std::move(tmp_dim);
    *data_buf += size;
  }
  // tensor version
  uint32_t tensor_version = *(reinterpret_cast<uint32_t *>(*data_buf));
  *data_buf += sizeof(uint32_t);
  // tensor desc size
  int32_t tensor_desc_size = *(reinterpret_cast<int32_t *>(*data_buf));
  *data_buf += sizeof(int32_t);
  // skip tensor desc
  *data_buf += tensor_desc_size;

  const TensorDesc &tensor_desc = var_desc->Tensor_desc();
  tensor->Resize(make_ddim(tensor_desc.Dims()));
  // parse tensor from stream
  switch (tensor_desc.DataType()) {
    case VARTYPE_TYPE_FP32:
      LoadMemInternal<float>(reinterpret_cast<void **>(data_buf), tensor,
                             program_.quantification);
      break;
    case VARTYPE_TYPE_INT8:
      LoadMemInternal<int8_t>(reinterpret_cast<void **>(data_buf), tensor);
      break;
    case VARTYPE_TYPE_INT32:
      LoadMemInternal<int>(reinterpret_cast<void **>(data_buf), tensor);
      break;
    default:
      LOG(kLOG_ERROR) << "data type is not supported";
  }
}

template <typename Device, typename T>
void Executor<Device, T>::InitMemory() {
  for (const auto &block : program_desc_->Blocks()) {
    for (const auto &var_desc : block->Vars()) {
      auto var = program_.scope->Var(var_desc->Name());
      auto tensor = var->template GetMutable<LoDTensor>();
      if (var_desc->Persistable()) {
        if (var_desc->Name() == "feed" || var_desc->Name() == "fetch") {
          continue;
        }
        char *origin_data =
            ReadFileToBuff(program_.model_path + "/" + var_desc->Name());
        char *data = origin_data;
        LoadMemory(reinterpret_cast<void **>(&data), var_desc, tensor);
        delete[] origin_data;
      } else {
        if (var_desc->Type() == VARTYPE_TYPE_LOD_TENSOR) {
          varInputMemory(var_desc, var, tensor);
        }
      }
    }
  }
}

template <typename Device, typename T>
void Executor<Device, T>::InitCombineMemory() {
  char *origin_data = nullptr;
  bool self_alloc = false;
  if (program_.combined_params_buf && program_.combined_params_len) {
    origin_data = reinterpret_cast<char *>(
        const_cast<uint8_t *>(program_.combined_params_buf));
  } else {
    self_alloc = true;
    origin_data = ReadFileToBuff(program_.para_path);
  }
  PADDLE_MOBILE_ENFORCE(origin_data != nullptr, "data == nullptr");
  char *data = origin_data;
  for (const auto &block : program_desc_->Blocks()) {
    for (const auto &var_desc : block->Vars()) {
      auto var = program_.scope->Var(var_desc->Name());
      auto tensor = var->template GetMutable<LoDTensor>();
      if (var_desc->Persistable()) {
        if (var_desc->Name() == "feed" || var_desc->Name() == "fetch") {
          continue;
        }

        DLOG << " init combine memory persistable: " << var_desc->Name();

        LoadMemory(reinterpret_cast<void **>(&data), var_desc, tensor);
      } else {
        if (var_desc->Type() == VARTYPE_TYPE_LOD_TENSOR) {
          DLOG << " init combine memory no persistable in lod: "
               << var_desc->Name();
          varInputMemory(var_desc, var, tensor);
        } else {
          DLOG << " init combine memory no persistable: " << var_desc->Name();
        }
      }
    }
  }
  if (self_alloc) {
    delete[] origin_data;
  }
  LOG(kLOG_INFO) << "init combine memory finish";
}

template <typename Device, typename T>
void Executor<Device, T>::InitNoPersistableMemory(
    const Tensor &input_tensor) {
  for (const auto &block : program_desc_->Blocks()) {
    for (const auto &var_desc : block->Vars()) {
      auto var = program_.scope->Var(var_desc->Name());
      auto tensor = var->template GetMutable<LoDTensor>();
      if (var_desc->Persistable()) {
        if (var_desc->Name() == "feed" || var_desc->Name() == "fetch") {
          continue;
        }
      } else {
        if (var_desc->Type() == VARTYPE_TYPE_LOD_TENSOR) {
          DDim tensor_dim = tensor->dims();
          DDim new_dim =
              make_ddim({tensor_dim[0], tensor_dim[1], input_tensor.dims()[2],
                         input_tensor.dims()[3]});
          tensor->Resize(new_dim);
          tensor->template mutable_data<T>();
        }
      }
    }
  }

  std::shared_ptr<LoDTensor> output = GetOutput("fetch");
  output->Resize(input_tensor.dims());
  output->mutable_data<T>();
}

template <typename Device, typename T>
bool Executor<Device, T>::varInputMemory(
    const std::shared_ptr<VarDesc> &var_desc, Variable *var,
    LoDTensor *tensor) const {
#ifdef PADDLE_MOBILE_FPGA
  tensor->init(typeid(float));
  return true;
#endif
  auto type = var_desc->Tensor_desc().DataType();
  switch (type) {
    case VARTYPE_TYPE_FP32:
      tensor->mutable_data<float>();
      break;
    case VARTYPE_TYPE_INT8:
      tensor->mutable_data<int8_t>();
      break;
    case VARTYPE_TYPE_INT32:
      tensor->mutable_data<int32_t>();
      break;
    case VARTYPE_TYPE_INT64:
      tensor->mutable_data<int64_t>();
      break;
    default:
      break;
  }
  bool is_mute_match =
      (type == VARTYPE_TYPE_FP32) || (type == VARTYPE_TYPE_INT8) ||
      (type == VARTYPE_TYPE_INT32) || (type == VARTYPE_TYPE_INT64);
  PADDLE_MOBILE_ENFORCE(is_mute_match, "got unhandled data type : %d", type);
  return is_mute_match;
}

template <typename Device, typename T>
PMStatus Executor<Device, T>::Predict(
    const std::vector<std::pair<std::string, Tensor>> &inputs) {
  for (const auto &input : inputs) {
    SetInput(input.second, input.first);
  }
  return this->Predict();
}

template <typename Device, typename T>
PMStatus Executor<Device, T>::Predict(
    const std::vector<std::pair<std::string, LoDTensor>> &inputs) {
  for (const auto &input : inputs) {
    SetInput(input.second, input.first);
  }
  return this->Predict();
}

template <typename Device, typename T>
std::vector<T> Executor<Device, T>::Predict(const std::vector<T> &input,
                                            const std::vector<int64_t> &dims) {
  Tensor feed_tensor(input, make_ddim(dims));
  SetInput(feed_tensor, "feed");
  std::vector<T> output;
  if (this->Predict() == PMSuccess) {
    const auto output_tensor = GetOutput("fetch");
    output.resize(output_tensor->numel());
    memcpy(output.data(), output_tensor->template data<T>(),
           output.size() * sizeof(T));
  }
  return output;
}

template <typename Device, typename T>
void Executor<Device, T>::SetInput(const Tensor &input,
                                   const std::string &var_name) {
  auto *target_var = program_.scope->FindVar(var_name);
  PADDLE_MOBILE_ENFORCE(target_var != nullptr, "Variable %s is not exist",
                        var_name.c_str());

  auto *target_tensor = target_var->template GetMutable<LoDTensor>();

  if (config_.load_when_predict) {
    if (input_dim_last_ != input.dims()) {
      InitNoPersistableMemory(input);
      input_dim_last_ = input.dims();
    }
  }

  target_tensor->Resize(input.dims());
  target_tensor->ShareDataWith(input);
}

template <typename Device, typename T>
void Executor<Device, T>::SetInput(const LoDTensor &input,
                                   const std::string &var_name) {
  auto *target_var = program_.scope->FindVar(var_name);
  PADDLE_MOBILE_ENFORCE(target_var != nullptr, "Variable %s is not exist",
                        var_name.c_str());
  auto *target_tensor = target_var->template GetMutable<LoDTensor>();

  if (config_.load_when_predict) {
    if (input_dim_last_ != input.dims()) {
      InitNoPersistableMemory(*target_tensor);
      input_dim_last_ = input.dims();
    }
  }

  target_tensor->Resize(input.dims());
  target_tensor->ShareDataWith(input);
  target_tensor->set_lod(input.lod());
}

template <typename Device, typename T>
PMStatus Executor<Device, T>::Predict() {
#ifdef PADDLE_MOBILE_PROFILE
  std::vector<ProfInfo> profile(ops_list_.size());
  struct timespec ts;
  int op_index = 0;
#endif
  for (auto &block : ops_of_block_) {
    for (auto &op_handler : block) {
#ifdef PADDLE_MOBILE_PROFILE
      clock_gettime(CLOCK_MONOTONIC, &ts);
      profile[op_index].runBegin = (uint64_t)ts.tv_sec * 1e9 + ts.tv_nsec;
#endif
      if (lod_mode_) {
        op_handler->InferShape();
      }
      op_handler->Run();
#ifdef PADDLE_MOBILE_PROFILE
      clock_gettime(CLOCK_MONOTONIC, &ts);
      profile[op_index].runEnd = (uint64_t)ts.tv_sec * 1e9 + ts.tv_nsec;
      ++op_index;
#endif
    }
  }
#ifdef PADDLE_MOBILE_PROFILE
  std::unordered_map<std::string, uint64_t> _tp;
  for (int i = 0; i < profile.size(); i++) {
    const auto &pInfo = profile[i];
    uint64_t timeCost = pInfo.runEnd - pInfo.runBegin;
    if (ops_list_[i]->Type() == "conv2d" ||
        ops_list_[i]->Type() == "depthwise_conv2d") {
      auto inputs = ops_list_[i]->Inputs();
      auto *filter =
          GetVarValue<LoDTensor>("Filter", inputs, *(program_.scope));
      int kernel_size = filter->dims()[2];
      _tp[ops_list_[i]->Type() + "_" + std::to_string(kernel_size)] += timeCost;
    } else {
      _tp[ops_list_[i]->Type()] += timeCost;
    }
  }
  printf("====================[ profile ]======================\n");
  typedef std::pair<std::string, uint64_t> prof_t;
  std::vector<prof_t> _tv(_tp.begin(), _tp.end());
  uint64_t _ptotal = 0;
  for (auto const &p : _tv) {
    _ptotal += p.second;
  }
  auto compf = [](const prof_t &a, const prof_t &b) {
    return a.second > b.second;
  };
  std::sort(_tv.begin(), _tv.end(), compf);
  _tv.push_back(std::make_pair("total", _ptotal));
  for (auto const &p : _tv) {
    printf("%-16s\t%-10.0f\t%-2.4f\n", p.first.c_str(),
           static_cast<float>(p.second),
           static_cast<float>(p.second) / _ptotal * 100.0);
  }
  printf("====================[---------]======================\n");
#endif
  return PMSuccess;
}

template <typename Device, typename T>
std::shared_ptr<LoDTensor> Executor<Device, T>::GetOutput(
    const std::string &var_name) {
  auto *target_var = program_.scope->FindVar(var_name);
  PADDLE_MOBILE_ENFORCE(target_var != nullptr, "Variable %s is not exist",
                        var_name.c_str());
  auto *output_tensor = target_var->template GetMutable<LoDTensor>();
  return std::make_shared<LoDTensor>(*output_tensor);
}

#ifdef PADDLE_MOBILE_FPGA
template <typename Device, typename T>
void Executor<Device, T>::InjectVariable(const Tensor &t,
                                         std::string var_name) {
  Variable *g_feed_value = program_.scope->Var(var_name);
  Tensor *feed_tensor = g_feed_value->GetMutable<LoDTensor>();
  feed_tensor->Resize(t.dims());
  feed_tensor->ShareDataWith(t);
}

template <typename Device, typename T>
void Executor<Device, T>::FeedData(const Tensor &t) {
  InjectVariable(t, "feed");
}

template <typename Device, typename T>
std::shared_ptr<Tensor> Executor<Device, T>::FetchResult(int id) {
  auto &ops = ops_of_block_[0];

  PADDLE_MOBILE_ENFORCE(id < (int)ops.size(), "Index out of range");
  auto op = id < 0 ? ops[ops.size() - 1] : ops[id];
  auto output_map = op->Outputs();
  std::vector<std::string> out_keys = op->GetOutKeys();
  PADDLE_MOBILE_ENFORCE(!out_keys.empty(), "this op contains no output");
  auto *output_tensor =
      GetVarValue<LoDTensor>(out_keys[0], output_map, *(program_.scope));
  return std::make_shared<Tensor>(Tensor(*output_tensor));
}

template <typename Device, typename T>
void Executor<Device, T>::Predict_From_To(int start, int end) {
  auto &ops = ops_of_block_[0];
  end = end < 0 ? static_cast<int>(ops.size()) : end;
  PADDLE_MOBILE_ENFORCE(start >= 0 && start < end && end <= ops.size(),
                        "start or end parameter is wrong");

#ifdef PADDLE_MOBILE_PROFILE
  std::vector<ProfInfo> profile(ops.size());
#endif
  for (int i = start; i < end; i++) {
#ifdef PADDLE_MOBILE_PROFILE
    struct timespec ts;
    clock_gettime(CLOCK_MONOTONIC, &ts);
    profile[i].runBegin = (uint64_t)ts.tv_sec * 1e9 + ts.tv_nsec;
#endif
    DLOG << "Running op: " << i << "  " << ops[i]->Type();
    ops[i]->Run();

#ifdef PADDLE_MOBILE_PROFILE
    clock_gettime(CLOCK_MONOTONIC, &ts);
    profile[i].runEnd = (uint64_t)ts.tv_sec * 1e9 + ts.tv_nsec;
#endif
  }
}

template <typename Device, typename T>
void Executor<Device, T>::Predict_From(int start) {
  Predict_From_To(start);
}

template <typename Device, typename T>
void Executor<Device, T>::Predict_To(int end) {
  Predict_From_To(0, end);
}
#endif

#ifdef PADDLE_MOBILE_CL
template <>
void Executor<GPU_CL, float>::InitNoPersistableMemory(
    const Tensor &input_tensor) {
  DLOG << "CL InitNoPersistableMemory ";
  for (const auto &block : program_desc_->Blocks()) {
    for (const auto &var_desc : block->Vars()) {
      auto var = program_.scope->Var(var_desc->Name());

      auto cl_image = var->template GetMutable<CLImage>();

      if (var_desc->Persistable()) {
        if (var_desc->Name() == "feed" || var_desc->Name() == "fetch") {
          continue;
        }
      } else {
        if (var_desc->Type() == VARTYPE_TYPE_LOD_TENSOR) {
          cl_context context = program_.scope->GetCLScpoe()->Context();
          cl_command_queue command_queue =
              program_.scope->GetCLScpoe()->CommandQueue();

          DDim tensor_dim = cl_image->dims();
          DDim new_dim =
              make_ddim({tensor_dim[0], tensor_dim[1], input_tensor.dims()[2],
                         input_tensor.dims()[3]});
          cl_image->Resize(new_dim);
          cl_image->InitEmptyImage(context, command_queue, new_dim);
        }
      }
    }
  }
  std::shared_ptr<LoDTensor> output = GetOutput("fetch");
  output->Resize(input_tensor.dims());
  output->mutable_data<float>();
}
template <>
void Executor<GPU_CL, float>::SetInput(const Tensor &input,
                                       const std::string &var_name) {
  auto *target_var = program_.scope->FindVar(var_name);
  PADDLE_MOBILE_ENFORCE(target_var != nullptr, "Variable %s is not exist",
                        var_name.c_str());

  auto *target_tensor = target_var->template GetMutable<LoDTensor>();
  DLOG << "config_.load_when_predict   " << config_.load_when_predict;
  DLOG << "target_tensor->IsInitialized() " << target_tensor->IsInitialized();
  DLOG << "target_tensor->dims()   " << target_tensor->dims();
  DLOG << "input.dims()   " << input.dims();
  DLOG << "input_dim_last_   " << input_dim_last_;
  if (config_.load_when_predict) {
    if (input_dim_last_ != input.dims()) {
      DLOG << "SetInput ---- > resize1";
      target_tensor->Resize(input.dims());
      target_tensor->mutable_data<float>();
      InitNoPersistableMemory(*target_tensor);
    }
  } else {
    DLOG << "SetInput ---- > resize2";
    target_tensor->Resize(input.dims());
    DLOG << "SetInput ---- > ShareDataWith";
  }
  target_tensor->ShareDataWith(input);
  auto &dim = input.dims();
  input_dim_last_ = static_cast<DDim>(dim);
}

template <typename Device, typename T>
void Executor<Device, T>::LoadMemory(const VarDesc var_desc, float *tensorInput,
                                     char **data) {}

template <>
void Executor<GPU_CL, float>::LoadMemory(const VarDesc var_desc,
                                         float *tensorInput, char **data) {
  // 1. version
  uint32_t version = *reinterpret_cast<uint32_t *>(*data);

  (*data) += sizeof(uint32_t);

  // 2 Lod information
  uint64_t *lod_level_ptr = new uint64_t();
  memcpy(lod_level_ptr, (*data), sizeof(uint64_t));
  uint64_t lod_level = *lod_level_ptr;
  delete lod_level_ptr;
  (*data) += sizeof(uint64_t);

  for (uint64_t i = 0; i < lod_level; ++i) {
    uint64_t size = *reinterpret_cast<uint64_t *>(*data);
    (*data) += sizeof(uint64_t);
    std::vector<size_t> tmp(size / sizeof(size_t));

    for (int k = 0; k < tmp.size(); ++k) {
      tmp[k] = *reinterpret_cast<size_t *>(*data);
      (*data) += sizeof(size_t);
    }
  }

  // 3. tensor version
  uint32_t tensor_version = *reinterpret_cast<uint32_t *>(*data);
  (*data) += sizeof(uint32_t);

  // 4. tensor desc
  int32_t size = *reinterpret_cast<int32_t *>(*data);
  (*data) += sizeof(int32_t);

  std::unique_ptr<char[]> buf(new char[size]);
  for (int m = 0; m < size; ++m) {
    buf.get()[m] = (*data)[m];
  }
  (*data) += (sizeof(char) * size);

  const TensorDesc &desc = var_desc.Tensor_desc();
  int memory_size = 1;
  for (auto l : desc.Dims()) {
    memory_size *= l;
  }

  void *memory = nullptr;
  int type_size = 4;
  memory = tensorInput;
  if (program_.quantification) {
    float min_value;
    float max_value;

    memcpy(&min_value, *data, sizeof(float));
    memcpy(&max_value, *data + sizeof(float), sizeof(float));
    *data += 2 * sizeof(float);
    const float factor = (max_value - min_value) / 255.0;
    uint8_t *uint8_data = reinterpret_cast<uint8_t *>(*data);
    for (int k = 0; k < memory_size; ++k) {
      static_cast<float *>(memory)[k] = uint8_data[k] * factor + min_value;
    }
    *data += (memory_size * sizeof(uint8_t));
  } else {
    for (int n = 0; n < memory_size; n++) {
      float value;
      memcpy(&value, *data + n * type_size, type_size);
      if (value < 1e-30 && value > -1e-30) {
        static_cast<float *>(memory)[n] = 0.0;
      } else {
        static_cast<float *>(memory)[n] = value;
      }
    }
    (*data) += (sizeof(char) * memory_size * type_size);
  }
}

template <>
void Executor<GPU_CL, float>::InitMemory() {
  for (const auto &block : program_desc_->Blocks()) {
    for (const auto &var_desc : block->Vars()) {
      auto var = program_.scope->Var(var_desc->Name());
      if (var_desc->Persistable()) {
        CLImage *cl_image = nullptr;
        if (var_desc->Name() == "feed" || var_desc->Name() == "fetch") {
          var->template GetMutable<LoDTensor>();
          continue;
        } else {
          cl_image = var->template GetMutable<CLImage>();
        }

        char *origin_data =
            ReadFileToBuff(program_.model_path + "/" + var_desc->Name());
        char *data = origin_data;
        cl_context context = program_.scope->GetCLScpoe()->Context();
        const TensorDesc &desc = var_desc->Tensor_desc();
        int numel = 1;
        for (auto l : desc.Dims()) {
          numel *= l;
        }
        DLOG << var_desc->Name();
        float *tensorInput = static_cast<float *>(
            paddle_mobile::memory::Alloc(sizeof(float) * numel));
        LoadMemory(*var_desc, tensorInput, &data);

        DDim ddim = make_ddim(desc.Dims());

        // has not init
        cl_image->SetTensorData(tensorInput, ddim);

        delete origin_data;
        paddle_mobile::memory::Free(tensorInput);
      } else {
        if (var_desc->Type() == VARTYPE_TYPE_LOD_TENSOR) {
          auto cl_image = var->template GetMutable<CLImage>();
          cl_context context = program_.scope->GetCLScpoe()->Context();
          cl_command_queue command_queue =
              program_.scope->GetCLScpoe()->CommandQueue();

          const TensorDesc &desc = var_desc->Tensor_desc();
          //          DDim ddim = make_ddim(desc.Dims());
          DDim ddim = cl_image->dims();
          DLOG << var_desc->Name();
          cl_image->InitEmptyImage(context, command_queue, ddim);
        }
      }
    }
  }
}

template <>
void Executor<GPU_CL, float>::InitCombineMemory() {
  DLOG << "CL InitCombineMemory---- "
       << "config_.load_when_predict: " << config_.load_when_predict;
  char *origin_data = nullptr;
  bool self_alloc = false;
  if (program_.combined_params_buf && program_.combined_params_len) {
    LOG(kLOG_INFO) << "use outter memory";
    origin_data = reinterpret_cast<char *>(program_.combined_params_buf);
  } else {
    LOG(kLOG_INFO) << " begin init combine memory";
    self_alloc = true;
    origin_data = ReadFileToBuff(program_.para_path);
  }
  PADDLE_MOBILE_ENFORCE(origin_data != nullptr, "origin_data==nullptr!!!");
  float *data = reinterpret_cast<float *>(origin_data);

  for (const auto &block : program_desc_->Blocks()) {
    for (const auto &var_desc : block->Vars()) {
      auto var = program_.scope->Var(var_desc->Name());
      if (var_desc->Persistable()) {
        CLImage *cl_image = nullptr;
        if (var_desc->Name() == "feed" || var_desc->Name() == "fetch") {
          var->template GetMutable<LoDTensor>();
          continue;
        } else {
          cl_image = var->template GetMutable<CLImage>();
        }

        cl_context context = program_.scope->GetCLScpoe()->Context();

        const TensorDesc &desc = var_desc->Tensor_desc();
        DDim ddim = make_ddim(desc.Dims());

        int numel = 1;
        for (int i = 0; i < ddim.size(); i++) {
          numel = numel * ddim[i];
        }
        float *tensorInput = static_cast<float *>(
            paddle_mobile::memory::Alloc(sizeof(float) * numel));
        LoadMemory(*var_desc, tensorInput, &origin_data);

        // has not init
        cl_image->SetTensorData(tensorInput, ddim);

        paddle_mobile::memory::Free(tensorInput);
      } else {
        auto cl_image = var->template GetMutable<CLImage>();
        cl_context context = program_.scope->GetCLScpoe()->Context();
        cl_command_queue command_queue =
            program_.scope->GetCLScpoe()->CommandQueue();
        const TensorDesc &desc = var_desc->Tensor_desc();
        DDim ddim = cl_image->dims();
        //  DDim ddim = make_ddim(desc.Dims());
        cl_image->InitEmptyImage(context, command_queue, ddim);
      }
    }
  }
  if (self_alloc) {
    delete data;
  }
  LOG(kLOG_INFO) << " end init combine memory ";
}

#endif

template class Executor<CPU, float>;

template class Executor<FPGA, float>;

template class Executor<GPU_CL, float>;

template class Executor<GPU_MALI, float>;

}  // namespace framework
}  // namespace paddle_mobile