data_balance_op_handle.cc 5.8 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
// 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/framework/details/data_balance_op_handle.h"
#include <algorithm>
#include "paddle/fluid/framework/details/container_cast.h"

namespace paddle {
namespace framework {
namespace details {

P
peizhilin 已提交
23
#if defined(PADDLE_WITH_CUDA) && !defined(_WIN32)
F
fengjiayi 已提交
24
DataBalanceOpHandle::DataBalanceOpHandle(
X
Xin Pan 已提交
25
    ir::Node *node, const std::vector<Scope *> &local_scopes,
F
fengjiayi 已提交
26 27
    const std::vector<platform::Place> &places,
    const platform::NCCLContextMap *ctxs)
X
Xin Pan 已提交
28
    : OpHandleBase(node), local_scopes_(local_scopes), places_(places) {
F
fengjiayi 已提交
29 30
  if (ctxs) {
    for (auto &p : places_) {
C
chengduo 已提交
31
      this->SetDeviceContext(p, ctxs->DevCtx(p));
F
fengjiayi 已提交
32 33 34 35
    }
  }
}
#else
36
DataBalanceOpHandle::DataBalanceOpHandle(
X
Xin Pan 已提交
37
    ir::Node *node, const std::vector<Scope *> &local_scopes,
38
    const std::vector<platform::Place> &places)
X
Xin Pan 已提交
39
    : OpHandleBase(node), local_scopes_(local_scopes), places_(places) {}
F
fengjiayi 已提交
40
#endif
41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64

std::string DataBalanceOpHandle::Name() const { return "data balance"; }

std::vector<std::array<int, 3>> DataBalanceOpHandle::GetBalancePlan(
    const std::vector<int> &device_sizes) {
  int device_num = device_sizes.size();
  int total_size = 0;
  int empty_num = 0;
  std::vector<std::array<int, 2>> size_device_vec;
  size_device_vec.reserve(device_num);
  for (int i = 0; i < device_num; ++i) {
    if (device_sizes[i] == 0) {
      ++empty_num;
    }
    total_size += device_sizes[i];
    size_device_vec.push_back({{device_sizes[i], i}});
  }
  std::vector<std::array<int, 3>> res;
  if (empty_num == 0) {
    // No need to do data balance.
    return res;
  }
  if (total_size < device_num) {
    // No enough data.
65
    PADDLE_THROW_EOF();
66 67 68 69 70 71 72 73 74 75
  }
  std::sort(size_device_vec.begin(), size_device_vec.end(),
            [](const std::array<int, 2> &a, const std::array<int, 2> &b) {
              return a[0] > b[0];
            });
  int expected_device_size = total_size / device_num;
  int src_idx = 0;
  for (int dst_idx = device_num - empty_num; dst_idx < device_num; ++dst_idx) {
    if (size_device_vec[src_idx][0] <= expected_device_size) {
      ++src_idx;
F
fengjiayi 已提交
76 77 78
      PADDLE_ENFORCE_LT(
          src_idx, device_num - empty_num,
          "In current srategy an empty tensor should not be copy source.");
79 80 81 82 83 84 85 86 87 88
    }
    size_device_vec[src_idx][0] -= expected_device_size;
    size_device_vec[dst_idx][0] += expected_device_size;
    res.push_back({{size_device_vec[src_idx][1], size_device_vec[dst_idx][1],
                    expected_device_size}});
  }
  return res;
}

void DataBalanceOpHandle::RunImpl() {
Y
yuyang18 已提交
89 90 91
  PADDLE_ENFORCE_GT(places_.size(), 1,
                    "Data balance can only be enabled when the number of "
                    "places to run larger than 1.");
C
chengduo 已提交
92 93
  auto in_var_handles = DynamicCast<VarHandle>(this->Inputs());
  auto out_var_handles = DynamicCast<VarHandle>(this->Outputs());
94 95 96 97 98 99
  PADDLE_ENFORCE(in_var_handles.size() % places_.size() == 0);
  PADDLE_ENFORCE_EQ(
      in_var_handles.size(), out_var_handles.size(),
      "The NoDummyInputSize and NoDummyOutputSize should be equal.");
  int data_num = in_var_handles.size() / places_.size();
  WaitInputVarGenerated();
F
fengjiayi 已提交
100
  std::vector<std::vector<LoDTensor *>> lod_tensors(data_num);
101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117
  std::vector<int> device_sizes;
  for (int i = 0; i < static_cast<int>(in_var_handles.size()); ++i) {
    PADDLE_ENFORCE_EQ(in_var_handles[i]->name_, out_var_handles[i]->name_,
                      "The name of input and output should be equal.");
    int place_idx = i / data_num;
    int data_idx = i % data_num;
    auto *local_scope =
        local_scopes_[place_idx]->FindVar(kLocalExecScopeName)->Get<Scope *>();
    auto *tensor_var = local_scope->FindVar(in_var_handles[i]->name_);
    PADDLE_ENFORCE(tensor_var->IsType<LoDTensor>());
    auto *tensor = tensor_var->GetMutable<LoDTensor>();
    lod_tensors[data_idx].push_back(tensor);
    int ins_size =
        tensor->lod().empty() ? tensor->dims()[0] : tensor->NumElements();
    if (data_idx == 0) {
      device_sizes.emplace_back(ins_size);
    } else {
F
fengjiayi 已提交
118 119 120
      PADDLE_ENFORCE_EQ(
          ins_size, device_sizes.at(place_idx),
          "All data on the same device shall have the same batch size.");
121 122 123
    }
  }
  const auto &balance_plan = GetBalancePlan(device_sizes);
F
fengjiayi 已提交
124

125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154
  for (const auto &trans : balance_plan) {
    for (int data_idx = 0; data_idx < data_num; ++data_idx) {
      LoDTensor *src_tensor = lod_tensors[data_idx][trans[0]];
      LoDTensor *dst_tensor = lod_tensors[data_idx][trans[1]];
      int trans_ins_size = trans[2];
      LoD src_lod = src_tensor->lod();
      int src_ins_size =
          src_lod.empty() ? src_tensor->dims()[0] : src_tensor->NumElements();
      int cut_point = src_ins_size - trans_ins_size;
      if (!src_lod.empty()) {
        for (auto &level : src_lod) {
          cut_point = level[cut_point];
        }
      }
      TensorCopySync(src_tensor->Slice(cut_point, src_tensor->dims()[0]),
                     dst_tensor->place(), dst_tensor);
      src_tensor->ShareDataWith(src_tensor->Slice(0, cut_point));
      if (!src_lod.empty()) {
        dst_tensor->set_lod(SliceInLevel(
            src_lod, 0, src_ins_size - trans_ins_size, src_ins_size));
        src_tensor->set_lod(
            SliceInLevel(src_lod, 0, 0, src_ins_size - trans_ins_size));
      }
    }
  }
}

}  // namespace details
}  // namespace framework
}  // namespace paddle