dist_model.cc

// Copyright (c) 2021 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/distributed/fleet_executor/dist_model.h"

#include <glog/logging.h>

#include <chrono>  // NOLINT

#include "paddle/fluid/distributed/fleet_executor/fleet_executor.h"
#include "paddle/fluid/distributed/fleet_executor/task_node.h"
#include "paddle/fluid/framework/block_desc.h"
#include "paddle/fluid/framework/feed_fetch_method.h"
#include "paddle/fluid/framework/naive_executor.h"
#include "paddle/fluid/framework/op_proto_maker.h"
#include "paddle/fluid/framework/program_desc.h"
#include "paddle/fluid/framework/scope.h"
#include "paddle/fluid/framework/tensor.h"

namespace paddle {
namespace distributed {

namespace {
bool IsPersistable(const framework::VarDesc *var) {
  if (var->Persistable() &&
      var->GetType() != framework::proto::VarType::FEED_MINIBATCH &&
      var->GetType() != framework::proto::VarType::FETCH_LIST &&
      var->GetType() != framework::proto::VarType::RAW) {
    return true;
  }
  return false;
}

bool LoadDataFromDistModelTensor(const DistModelTensor &input_data,
                                 framework::LoDTensor *input_tensor,
                                 const platform::Place &place) {
  VLOG(3) << "Loading data from DistModelTensor for " << input_data.name;
  framework::DDim dims = phi::make_ddim(input_data.shape);
  void *input_tensor_ptr;
  if (input_data.dtype == DistModelDataType::INT64) {
    input_tensor_ptr = input_tensor->mutable_data<int64_t>(dims, place);
  } else if (input_data.dtype == DistModelDataType::FLOAT32) {
    input_tensor_ptr = input_tensor->mutable_data<float>(dims, place);
  } else if (input_data.dtype == DistModelDataType::INT32) {
    input_tensor_ptr = input_tensor->mutable_data<int32_t>(dims, place);
  } else if (input_data.dtype == DistModelDataType::FLOAT16) {
    input_tensor_ptr = input_tensor->mutable_data<float16>(dims, place);
  } else {
    LOG(ERROR) << "unsupported feed type " << input_data.dtype;
    return false;
  }

  PADDLE_ENFORCE_NOT_NULL(
      input_tensor_ptr,
      paddle::platform::errors::Fatal(
          "LoDTensor creation failed. DistModel loaded data failed."));
  PADDLE_ENFORCE_NOT_NULL(input_data.data.data(),
                          paddle::platform::errors::InvalidArgument(
                              "DistModelTensor contains no data."));

  if (platform::is_cpu_place(place)) {
    VLOG(3) << "Loading data for CPU.";
    std::memcpy(static_cast<void *>(input_tensor_ptr), input_data.data.data(),
                input_data.data.length());
  } else if (platform::is_gpu_place(place)) {
    VLOG(3) << "Loading data for GPU.";
#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
    platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance();
    auto *dev_ctx =
        dynamic_cast<const platform::CUDADeviceContext *>(pool.Get(place));
    auto gpu_place = place;
    memory::Copy(gpu_place, static_cast<void *>(input_tensor_ptr),
                 platform::CPUPlace(), input_data.data.data(),
                 input_data.data.length(), dev_ctx->stream());
#else
    PADDLE_THROW(paddle::platform::errors::Fatal(
        "Paddle wasn't compiled with CUDA, but place is GPU."));
#endif
  } else {
    PADDLE_THROW(paddle::platform::errors::InvalidArgument(
        "DistModel only supports CPU and GPU."));
  }

  framework::LoD dst_lod;
  for (auto &src_lod : input_data.lod) {
    dst_lod.emplace_back(src_lod);
  }
  input_tensor->set_lod(dst_lod);
  return true;
}

std::string DistModelDTypeToString(DistModelDataType dtype) {
  switch (dtype) {
    case DistModelDataType::FLOAT32:
      return "float32";
    case DistModelDataType::FLOAT16:
      return "float16";
    case DistModelDataType::INT64:
      return "int64";
    case DistModelDataType::INT32:
      return "int32";
    case DistModelDataType::INT8:
      return "int8";
  }
  return "NOT SUPPORT DTYPE";
}

class DistModelTimer {
 public:
  void tic() { tic_time = std::chrono::high_resolution_clock::now(); }
  double toc() {
    std::chrono::high_resolution_clock::time_point toc_time =
        std::chrono::high_resolution_clock::now();
    std::chrono::duration<double> time_elapse =
        std::chrono::duration_cast<std::chrono::duration<double>>(toc_time -
                                                                  tic_time);
    double time_elapse_in_ms =
        static_cast<double>(time_elapse.count()) * 1000.0;
    return time_elapse_in_ms;
  }

 private:
  std::chrono::high_resolution_clock::time_point tic_time;
};

}  // namespace

bool DistModel::Init() {
  carrier_id_ = "inference";
  bool init_method = (!config_.model_dir.empty() || config_.program_desc);
  PADDLE_ENFORCE_EQ(init_method, true,
                    platform::errors::InvalidArgument(
                        "One of model dir or program desc must be provided to "
                        "dist model inference."));
  if (config_.program_desc) {
    PADDLE_ENFORCE_NOT_NULL(
        config_.scope, platform::errors::InvalidArgument(
                           "Scope must be provided to dist model inference if "
                           "program desc has been provided."));
  }
  if (!PreparePlace()) {
    return false;
  }
  if (!config_.program_desc) {
    if (config_.scope) {
      LOG(WARNING) << "The provided scope will be ignored if model dir has "
                      "also been provided.";
    }
    if (!PrepareScope()) {
      return false;
    }
    if (!PrepareProgram()) {
      return false;
    }
  } else {
    program_.reset(config_.program_desc);
    scope_.reset(config_.scope);
  }
  if (!PrepareFeedAndFetch()) {
    return false;
  }
  if (config_.nranks > 1 && !CommInit()) {
    return false;
  }
  if (!PrepareFleetExe()) {
    return false;
  }
  return true;
}

bool DistModel::PreparePlace() {
  if (config_.place == "GPU") {
    place_ = paddle::platform::CUDAPlace(config_.device_id);
  } else if (config_.place == "CPU") {
    place_ = paddle::platform::CPUPlace();
  } else {
    PADDLE_THROW(platform::errors::InvalidArgument(
        "Place must be choosen from GPU or CPU, but got %s.", config_.place));
  }
  return true;
}

bool DistModel::CommInit() {
  std::unique_ptr<framework::ProgramDesc> comm_init_program(
      new framework::ProgramDesc());
  framework::BlockDesc *comm_init_block = comm_init_program->MutableBlock(0);
  std::vector<int64_t> &ring_ids =
      config_.rank_to_ring_ids_[config_.local_rank];
  int64_t order = 0;
  std::string var_name_base = "comm_init_";
  for (int64_t ring_id : ring_ids) {
    VLOG(3) << "Init comm for ring id: " << ring_id;
    int64_t ranks_in_group = config_.ring_id_to_ranks_[ring_id].size();
    int64_t rank_in_group = 0;
    std::vector<int64_t> &ranks = config_.ring_id_to_ranks_[ring_id];
    for (int64_t rank : ranks) {
      if (config_.local_rank == rank) {
        break;
      }
      rank_in_group += 1;
    }
    std::vector<std::string> peer_endpoints;
    for (int64_t rank : ranks) {
      if (config_.local_rank == rank) {
        continue;
      }
      peer_endpoints.emplace_back(config_.trainer_endpoints[rank]);
    }
    InsertCommOp(var_name_base + std::to_string(order), ranks_in_group,
                 rank_in_group, peer_endpoints, comm_init_block, ring_id);
    order += 1;
  }
  framework::NaiveExecutor e(place_);
  e.CreateVariables(*comm_init_program, 0, true, scope_.get());
  e.Prepare(scope_.get(), *comm_init_program, 0, false);
  e.Run();
  VLOG(3) << "Comm init successful.";
  return true;
}

void DistModel::InsertCommOp(std::string tmp_var_name, int nranks, int rank,
                             const std::vector<std::string> &peer_endpoints,
                             framework::BlockDesc *block, int ring_id) {
  /*
   * tmp_var_name: the var name for var comm_id
   * nranks: number of total ranks
   * rank: the rank of local rank in the comm group
   * peer_endpoints: peer's endpoints
   * block: the block where to insert the comm ops
   * ring_id: the ring_id to be inited
   */
  std::string &endpoint = config_.current_endpoint;
  std::stringstream ss;
  ss << "Init comm with tmp var: " << tmp_var_name
     << ". The ring id is: " << ring_id << ". The group has: " << nranks
     << " ranks. Current rank in the group is: " << rank
     << ". The endpoint is: " << endpoint << ". Peer endpoints are: ";
  for (auto ep : peer_endpoints) {
    ss << ep << ", ";
  }
  VLOG(3) << ss.str();
  if (config_.place == "GPU") {
    framework::VarDesc *new_var = block->Var(tmp_var_name);
    new_var->SetType(framework::proto::VarType::RAW);
    new_var->SetPersistable(true);
    framework::OpDesc *gen_nccl_id_op = block->AppendOp();
    gen_nccl_id_op->SetType("c_gen_nccl_id");
    gen_nccl_id_op->SetOutput("Out", {tmp_var_name});
    gen_nccl_id_op->SetAttr("rank", rank);
    gen_nccl_id_op->SetAttr("endpoint", config_.current_endpoint);
    gen_nccl_id_op->SetAttr("other_endpoints", peer_endpoints);
    gen_nccl_id_op->SetAttr("ring_id", ring_id);
    gen_nccl_id_op->SetAttr("op_role",
                            static_cast<int>(framework::OpRole::kForward));
    gen_nccl_id_op->CheckAttrs();
    framework::OpDesc *comm_init_op = block->AppendOp();
    comm_init_op->SetType("c_comm_init");
    comm_init_op->SetInput("X", {tmp_var_name});
    comm_init_op->SetAttr("rank", rank);
    comm_init_op->SetAttr("nranks", nranks);
    comm_init_op->SetAttr("ring_id", ring_id);
    comm_init_op->SetAttr("op_role",
                          static_cast<int>(framework::OpRole::kForward));
    comm_init_op->CheckAttrs();
  } else {
    LOG(WARNING) << "DistModelInf doesn't init comm.";
    // TODO(fleet exe dev): comm init for more devices
  }
}

bool DistModel::PrepareScope() {
  scope_.reset(new framework::Scope());
  return true;
}

bool DistModel::PrepareProgram() {
  if (!LoadProgram()) {
    return false;
  }
  if (!LoadParameters()) {
    return false;
  }
  return true;
}

bool DistModel::LoadProgram() {
  VLOG(3) << "Loading program from " << config_.model_dir;
  PADDLE_ENFORCE_NE(
      config_.model_dir, "",
      platform::errors::InvalidArgument("Model dir must be provided."));
  std::string model_path = config_.model_dir + ".pdmodel";
  framework::proto::ProgramDesc program_proto;
  std::string pb_content;
  // Read binary
  std::ifstream fin(model_path, std::ios::in | std::ios::binary);
  PADDLE_ENFORCE_EQ(
      static_cast<bool>(fin.is_open()), true,
      platform::errors::NotFound(
          "Cannot open file %s, please confirm whether the file is normal.",
          model_path));
  fin.seekg(0, std::ios::end);
  pb_content.resize(fin.tellg());
  fin.seekg(0, std::ios::beg);
  fin.read(&(pb_content.at(0)), pb_content.size());
  fin.close();
  program_proto.ParseFromString(pb_content);
  VLOG(5) << pb_content;
  program_.reset(new framework::ProgramDesc(program_proto));
  return true;
}

bool DistModel::LoadParameters() {
  VLOG(3) << "Loading parameters from " << config_.model_dir;
  PADDLE_ENFORCE_NOT_NULL(program_.get(),
                          platform::errors::PreconditionNotMet(
                              "The program should be loaded first."));
  const auto &global_block = program_->MutableBlock(0);

  // create a temporary program to load parameters.

  std::unique_ptr<framework::ProgramDesc> load_program(
      new framework::ProgramDesc());
  framework::BlockDesc *load_block = load_program->MutableBlock(0);
  std::vector<std::string> params;

  for (auto *var : global_block->AllVars()) {
    if (IsPersistable(var)) {
      VLOG(3) << "persistable variable's name: " << var->Name();
      framework::VarDesc *new_var = load_block->Var(var->Name());
      new_var->SetShape(var->GetShape());
      new_var->SetDataType(var->GetDataType());
      new_var->SetType(var->GetType());
      new_var->SetLoDLevel(var->GetLoDLevel());
      new_var->SetPersistable(true);
      params.push_back(new_var->Name());
      // NOTE: if the params are stored in different files, 'load' op should be
      // added here
    }
  }

  std::string param_path = config_.model_dir + ".pdiparams";
  // sort paramlist to have consistent ordering
  std::sort(params.begin(), params.end());
  // append just the load_combine op
  framework::OpDesc *op = load_block->AppendOp();
  op->SetType("load_combine");
  op->SetOutput("Out", params);
  op->SetAttr("file_path", {param_path});
  op->CheckAttrs();

  framework::NaiveExecutor e(place_);
  // Create all persistable variables in root scope to load them from ckpt.
  // Other non-persistable variables will be created in the micro scope
  // managed by fleet executor.
  e.CreateVariables(*program_, 0, true, scope_.get());
  e.Prepare(scope_.get(), *load_program, 0, false);
  e.Run();
  VLOG(3) << "After loading there are " << scope_->LocalVarNames().size()
          << " vars.";

  return true;
}

bool DistModel::PrepareFleetExe() {
  task_node_.reset(new TaskNode(program_.get(), config_.local_rank));
  // With auto cut, there is no concept of pp, no need to add dependency.
  task_node_->SetType("Compute");
  task_node_->Init();
  executor_desc_ = FleetExecutorDesc();
  executor_desc_.set_cur_rank(config_.local_rank);
  std::unordered_map<int64_t, int64_t> id_to_rank;
  for (int i = 0; i < config_.nranks; ++i) {
    RankInfo *rank_info = executor_desc_.add_cluster_info();
    rank_info->set_rank(i);
    rank_info->set_ip_port(config_.trainer_endpoints[i]);
    id_to_rank.insert({i, i});
  }
  fleet_exe.reset(new FleetExecutor(executor_desc_));
  fleet_exe->Init(carrier_id_, *(program_.get()), scope_.get(), place_, 1,
                  {task_node_.get()}, id_to_rank);
  return true;
}

bool DistModel::PrepareFeedAndFetch() {
  for (auto *op : program_->Block(0).AllOps()) {
    if (op->Type() == "feed") {
      VLOG(3) << "feed op with feed var: " << op->Output("Out")[0];
      int idx = BOOST_GET_CONST(int, op->GetAttr("col"));
      if (feeds_.size() <= static_cast<size_t>(idx)) {
        feeds_.resize(idx + 1);
      }
      feeds_[idx] = op;
      std::string var_name = op->Output("Out")[0];
      feed_names_[var_name] = idx;
      idx_to_feeds_[idx] = var_name;
      framework::VarDesc *real_var = program_->Block(0).FindVar(var_name);
      if (!real_var) {
        LOG(ERROR)
            << "The output of feed ops [" << var_name
            << "] cannot be found in the program. Check the inference program.";
        return false;
      }
      if (real_var->GetDataType() == framework::proto::VarType::FP32) {
        feeds_to_dtype_.insert({var_name, DistModelDataType::FLOAT32});
      } else if (real_var->GetDataType() == framework::proto::VarType::INT32) {
        feeds_to_dtype_.insert({var_name, DistModelDataType::INT32});
      } else if (real_var->GetDataType() == framework::proto::VarType::INT64) {
        feeds_to_dtype_.insert({var_name, DistModelDataType::INT64});
      } else if (real_var->GetDataType() == framework::proto::VarType::FP16) {
        feeds_to_dtype_.insert({var_name, DistModelDataType::FLOAT16});
      } else {
        LOG(ERROR) << "Don't support feed var dtype for: "
                   << real_var->GetDataType();
        return false;
      }
    } else if (op->Type() == "fetch") {
      VLOG(3) << "fetch op with fetch var: " << op->Input("X")[0];
      int idx = BOOST_GET_CONST(int, op->GetAttr("col"));
      if (fetches_.size() <= static_cast<size_t>(idx)) {
        fetches_.resize(idx + 1);
      }
      fetches_[idx] = op;
      idx_to_fetches_[idx] = op->Input("X")[0];
    }
  }

  if (feeds_.size() == 0) {
    LOG(ERROR) << "No feed ops in the inf program, please check the program.";
    return false;
  }
  if (fetches_.size() == 0) {
    LOG(ERROR) << "No fetch op in the inf program, please check the program.";
    return false;
  }
  return true;
}

bool DistModel::FeedData(const std::vector<DistModelTensor> &input_data,
                         framework::Scope *scope) {
  VLOG(3) << "DistModel is feeding data.";
  if (input_data.size() != feeds_.size()) {
    LOG(ERROR) << "Should provide " << feeds_.size() << " feeds, but got "
               << input_data.size() << " data.";
    return false;
  }
  feed_tensors_.resize(feeds_.size());
  for (size_t i = 0; i < input_data.size(); ++i) {
    // feed each data separately
    framework::LoDTensor *input_tensor = &(feed_tensors_[i]);
    if (!LoadDataFromDistModelTensor(input_data[i], input_tensor, place_)) {
      LOG(ERROR) << "Fail to load data from tensor " << input_data[i].name;
      return false;
    }
    std::string target_name = input_data[i].name;
    if (feed_names_.find(target_name) == feed_names_.end()) {
      LOG(ERROR) << "The input name [" << target_name
                 << "] cannot be found in the program."
                 << " DistModel loads data failed.";
      return false;
    }
    if (input_data[i].dtype != feeds_to_dtype_[target_name]) {
      LOG(ERROR) << "Feed var [" << target_name << "] expected dtype is: "
                 << DistModelDTypeToString(feeds_to_dtype_[target_name])
                 << ". But received dtype is: "
                 << DistModelDTypeToString(input_data[i].dtype) << ".";
      return false;
    }
    int feed_idx = feed_names_[target_name];
    framework::SetFeedVariable(scope, *input_tensor, "feed", feed_idx);
  }
  return true;
}

bool DistModel::FetchResults(std::vector<DistModelTensor> *output_data,
                             framework::Scope *scope) {
  VLOG(3) << "DistModel is fetch results.";
  output_data->resize(fetches_.size());
  for (size_t i = 0; i < fetches_.size(); ++i) {
    int idx = BOOST_GET_CONST(int, fetches_[i]->GetAttr("col"));
    VLOG(3) << "Fetching data for [" << idx_to_fetches_[idx] << "]";
    PADDLE_ENFORCE_EQ(
        static_cast<size_t>(idx), i,
        platform::errors::InvalidArgument(
            "Fetch op's col attr(%d) should be equal to the index(%d)", idx,
            i));
    framework::FetchType &fetch_var =
        framework::GetFetchVariable(*scope, "fetch", idx);
    auto &fetch = BOOST_GET(framework::LoDTensor, fetch_var);
    auto type = framework::TransToProtoVarType(fetch.dtype());
    auto output = &(output_data->at(i));
    output->name = idx_to_fetches_[idx];
    bool rst = false;
    if (type == framework::proto::VarType::FP32) {
      rst = FetchResult<float>(fetch, output);
      output->dtype = DistModelDataType::FLOAT32;
    } else if (type == framework::proto::VarType::INT64) {
      rst = FetchResult<int64_t>(fetch, output);
      output->dtype = DistModelDataType::INT64;
    } else if (type == framework::proto::VarType::INT32) {
      rst = FetchResult<int32_t>(fetch, output);
      output->dtype = DistModelDataType::INT32;
    } else if (type == framework::proto::VarType::FP16) {
      rst = FetchResult<float16>(fetch, output);
      output->dtype = DistModelDataType::FLOAT16;
    } else {
      LOG(ERROR) << "DistModel meets unknown fetch data type. DistModel only "
                    "supports float32, float16, int64 and int32 fetch type "
                    "for now.";
    }
    if (!rst) {
      LOG(ERROR) << "DistModel fails to fetch result " << idx_to_fetches_[idx];
      return false;
    }
  }
  return true;
}

template <typename T>
bool DistModel::FetchResult(const framework::LoDTensor &fetch,
                            DistModelTensor *output_data) {
  auto shape = phi::vectorize(fetch.dims());
  output_data->shape.assign(shape.begin(), shape.end());
  const T *data = fetch.data<T>();
  int64_t num_elems = fetch.numel();
  output_data->data.Resize(num_elems * sizeof(T));
  // The output of fetch op is always on the cpu, no need switch on place
  memcpy(output_data->data.data(), data, num_elems * sizeof(T));
  output_data->lod.clear();
  for (auto &level : fetch.lod()) {
    output_data->lod.emplace_back(level.begin(), level.end());
  }
  return true;
}

bool DistModel::Run(const std::vector<DistModelTensor> &input_data,
                    std::vector<DistModelTensor> *output_data) {
  VLOG(3) << "DistModel run for once.";

  DistModelTimer timer;
  timer.tic();
  double feed_elapse = 0;
  double fleet_exe_elapse = 0;
  double fetch_elapse = 0;

  if (!FeedData(input_data, scope_.get())) {
    LOG(ERROR) << "DistModel failed at feeding data.";
    return false;
  }
  if (config_.enable_timer) {
    feed_elapse = timer.toc();
    LOG(INFO) << "Finish loading data, cost " << feed_elapse << "ms.";
  } else {
    VLOG(3) << "Finish loading data.";
  }

  fleet_exe->Run(carrier_id_);
  if (config_.enable_timer) {
    fleet_exe_elapse = timer.toc();
    LOG(INFO) << "Finish FleetExe running, cost "
              << fleet_exe_elapse - feed_elapse << "ms.";
  } else {
    VLOG(3) << "Finish FleetExe running.";
  }

  if (!FetchResults(output_data, scope_.get())) {
    LOG(ERROR) << "DistModel failed at fetching result.";
    return false;
  }
  if (config_.enable_timer) {
    fetch_elapse = timer.toc();
    LOG(INFO) << "Finish fetching data, cost "
              << fetch_elapse - fleet_exe_elapse << "ms.";
    LOG(INFO) << "DistModel finish inf, cost " << fetch_elapse << "ms";
  } else {
    VLOG(3) << "Finish fetching data.";
    VLOG(3) << "DistModel finish inf.";
  }
  return true;
}

}  // namespace distributed
}  // namespace paddle