adam_op.h 12.6 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
2 3 4 5 6 7 8 9 10 11 12 13 14 15

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
Y
Yang Yu 已提交
16
#include <math.h>  // for sqrt in CPU and CUDA
17
#include <Eigen/Dense>
S
sneaxiy 已提交
18
#include <vector>
Y
Yi Wang 已提交
19 20
#include "paddle/fluid/framework/op_registry.h"
#include "paddle/fluid/operators/detail/safe_ref.h"
S
sneaxiy 已提交
21
#include "paddle/fluid/operators/math/algorithm.h"
Y
Yi Wang 已提交
22 23
#include "paddle/fluid/operators/math/selected_rows_functor.h"
#include "paddle/fluid/platform/for_range.h"
24 25 26 27

namespace paddle {
namespace operators {

T
wip  
typhoonzero 已提交
28 29
namespace scatter = paddle::operators::math::scatter;

30 31 32 33 34 35
struct GPUAdam;
struct CPUAdam;

template <typename T, typename Flavour>
struct AdamFunctor;

Y
Yang Yu 已提交
36
template <typename T>
37
struct AdamFunctor<T, GPUAdam> {
Y
Yang Yu 已提交
38 39 40 41 42 43 44 45 46 47 48 49 50
  T beta1_;
  T beta2_;
  T epsilon_;

  const T* beta1_pow_;
  const T* beta2_pow_;
  const T* moment1_;
  T* moment1_out_;
  const T* moment2_;
  T* moment2_out_;
  const T* lr_;
  const T* grad_;
  const T* param_;
Y
Yang Yu 已提交
51
  T* param_out_;
Y
Yang Yu 已提交
52 53 54

  AdamFunctor(T beta1, T beta2, T epsilon, const T* beta1_pow,
              const T* beta2_pow, const T* mom1, T* mom1_out, const T* mom2,
Y
Yang Yu 已提交
55 56
              T* mom2_out, const T* lr, const T* grad, const T* param,
              T* param_out)
Y
Yang Yu 已提交
57 58 59 60 61 62 63 64 65 66 67
      : beta1_(beta1),
        beta2_(beta2),
        epsilon_(epsilon),
        beta1_pow_(beta1_pow),
        beta2_pow_(beta2_pow),
        moment1_(mom1),
        moment1_out_(mom1_out),
        moment2_(mom2),
        moment2_out_(mom2_out),
        lr_(lr),
        grad_(grad),
Y
Yang Yu 已提交
68 69
        param_(param),
        param_out_(param_out) {}
Y
Yang Yu 已提交
70

Y
Yang Yu 已提交
71
  inline HOSTDEVICE void operator()(size_t i) const {
Y
Yang Yu 已提交
72 73 74 75 76 77 78
    // Merge all memory access together.
    T g = grad_[i];
    T mom1 = moment1_[i];
    T mom2 = moment2_[i];
    T lr = *lr_;
    T beta1_pow = *beta1_pow_;
    T beta2_pow = *beta2_pow_;
Y
Yang Yu 已提交
79
    T p = param_[i];
Y
Yang Yu 已提交
80 81

    // Calculation
Y
Yang Yu 已提交
82
    lr *= sqrt(1 - beta2_pow) / (1 - beta1_pow);
83

Y
Yang Yu 已提交
84 85
    mom1 = beta1_ * mom1 + (1 - beta1_) * g;
    mom2 = beta2_ * mom2 + (1 - beta2_) * g * g;
Y
Yang Yu 已提交
86
    p -= lr * (mom1 / (sqrt(mom2) + epsilon_));
Y
Yang Yu 已提交
87 88 89 90

    // Write back to global memory
    moment1_out_[i] = mom1;
    moment2_out_[i] = mom2;
Y
Yang Yu 已提交
91
    param_out_[i] = p;
Y
Yang Yu 已提交
92 93 94
  }
};

95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159
template <typename T>
struct AdamFunctor<T, CPUAdam> {
  T beta1_;
  T beta2_;
  T epsilon_;

  const T* beta1_pow_;
  const T* beta2_pow_;
  const T* moment1_;
  T* moment1_out_;
  const T* moment2_;
  T* moment2_out_;
  const T* lr_;
  const T* grad_;
  const T* param_;
  T* param_out_;

  AdamFunctor(T beta1, T beta2, T epsilon, const T* beta1_pow,
              const T* beta2_pow, const T* mom1, T* mom1_out, const T* mom2,
              T* mom2_out, const T* lr, const T* grad, const T* param,
              T* param_out)
      : beta1_(beta1),
        beta2_(beta2),
        epsilon_(epsilon),
        beta1_pow_(beta1_pow),
        beta2_pow_(beta2_pow),
        moment1_(mom1),
        moment1_out_(mom1_out),
        moment2_(mom2),
        moment2_out_(mom2_out),
        lr_(lr),
        grad_(grad),
        param_(param),
        param_out_(param_out) {}

  void operator()(size_t numel) const {
    Eigen::Map<const Eigen::Array<T, 1, Eigen::Dynamic>> g{
        grad_, static_cast<Eigen::Index>(numel)};
    Eigen::Map<const Eigen::Array<T, 1, Eigen::Dynamic>> mom1{
        moment1_, static_cast<Eigen::Index>(numel)};
    Eigen::Map<const Eigen::Array<T, 1, Eigen::Dynamic>> mom2{
        moment2_, static_cast<Eigen::Index>(numel)};
    Eigen::Map<const Eigen::Array<T, 1, Eigen::Dynamic>> param{
        param_, static_cast<Eigen::Index>(numel)};

    Eigen::Map<Eigen::Array<T, 1, Eigen::Dynamic>> param_out{
        param_out_, static_cast<Eigen::Index>(numel)};
    Eigen::Map<Eigen::Array<T, 1, Eigen::Dynamic>> moment1_out{
        moment1_out_, static_cast<Eigen::Index>(numel)};
    Eigen::Map<Eigen::Array<T, 1, Eigen::Dynamic>> moment2_out{
        moment2_out_, static_cast<Eigen::Index>(numel)};

    T lr = *lr_;
    T beta1_pow = *beta1_pow_;
    T beta2_pow = *beta2_pow_;

    // Calculation
    lr *= sqrt(1 - beta2_pow) / (1 - beta1_pow);

    moment1_out = beta1_ * mom1 + (1 - beta1_) * g;
    moment2_out = beta2_ * mom2 + (1 - beta2_) * g * g;
    param_out = param - lr * (moment1_out / (moment2_out.sqrt() + epsilon_));
  }
};

T
wip  
typhoonzero 已提交
160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178
template <typename T>
struct SparseAdamFunctor {
  T beta1_;
  T beta2_;
  T epsilon_;

  const T* beta1_pow_;
  const T* beta2_pow_;
  const T* moment1_;
  T* moment1_out_;
  const T* moment2_;
  T* moment2_out_;
  const T* lr_;
  const T* grad_;
  const T* param_;
  T* param_out_;

  const int64_t* rows_;
  int64_t row_numel_;
S
sneaxiy 已提交
179
  int64_t row_count_;
Q
Qiao Longfei 已提交
180
  bool sparse_mode_;
T
wip  
typhoonzero 已提交
181 182 183 184 185

  SparseAdamFunctor(T beta1, T beta2, T epsilon, const T* beta1_pow,
                    const T* beta2_pow, const T* mom1, T* mom1_out,
                    const T* mom2, T* mom2_out, const T* lr, const T* grad,
                    const T* param, T* param_out, const int64_t* rows,
Q
Qiao Longfei 已提交
186
                    int64_t row_numel, int64_t row_count, bool sparse_mode)
T
wip  
typhoonzero 已提交
187 188 189 190 191 192 193 194 195 196 197 198 199 200
      : beta1_(beta1),
        beta2_(beta2),
        epsilon_(epsilon),
        beta1_pow_(beta1_pow),
        beta2_pow_(beta2_pow),
        moment1_(mom1),
        moment1_out_(mom1_out),
        moment2_(mom2),
        moment2_out_(mom2_out),
        lr_(lr),
        grad_(grad),
        param_(param),
        param_out_(param_out),
        rows_(rows),
S
sneaxiy 已提交
201
        row_numel_(row_numel),
Q
Qiao Longfei 已提交
202 203
        row_count_(row_count),
        sparse_mode_(sparse_mode) {}
S
sneaxiy 已提交
204

Q
Qiao Longfei 已提交
205
  inline HOSTDEVICE void sparse_update(size_t i, T g) const {
S
sneaxiy 已提交
206 207 208 209
    // The following code is the same as dense
    T mom1 = moment1_[i];
    T mom2 = moment2_[i];
    T lr = *lr_;
T
typhoonzero 已提交
210 211
    T beta1_pow = *beta1_pow_;
    T beta2_pow = *beta2_pow_;
S
sneaxiy 已提交
212 213 214 215 216 217 218 219 220 221 222 223 224
    T p = param_[i];

    // Calculation
    lr *= sqrt(1 - beta2_pow) / (1 - beta1_pow);

    mom1 = beta1_ * mom1 + (1 - beta1_) * g;
    mom2 = beta2_ * mom2 + (1 - beta2_) * g * g;
    p -= lr * (mom1 / (sqrt(mom2) + epsilon_));

    // Write back to global memory
    moment1_out_[i] = mom1;
    moment2_out_[i] = mom2;
    param_out_[i] = p;
T
wip  
typhoonzero 已提交
225
  }
Q
Qiao Longfei 已提交
226 227 228 229 230 231 232

  inline HOSTDEVICE void operator()(size_t i) const {
    auto row_idx =
        math::BinarySearch<int64_t>(rows_, row_count_, i / row_numel_);
    T g = row_idx >= 0 ? grad_[row_idx * row_numel_ + i % row_numel_] : 0;
    sparse_update(i, g);
  }
T
wip  
typhoonzero 已提交
233 234
};

Q
QI JUN 已提交
235
template <typename DeviceContext, typename T>
236 237 238
class AdamOpKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext& ctx) const override {
C
chengduo 已提交
239 240 241 242 243 244
    const auto* param_var = ctx.InputVar("Param");
    PADDLE_ENFORCE(param_var->IsType<framework::LoDTensor>(),
                   "The Var(%s)'s type should be LoDTensor, "
                   "but the received is %s",
                   ctx.Inputs("Param").front(), param_var->Type().name());

Y
Yang Yu 已提交
245 246
    using paddle::framework::LoDTensor;
    using paddle::operators::detail::Ref;
247

Q
Qiao Longfei 已提交
248
    bool sparse_mode = ctx.Attr<bool>("sparse_mode");
249 250 251
    T beta1 = static_cast<T>(ctx.Attr<float>("beta1"));
    T beta2 = static_cast<T>(ctx.Attr<float>("beta2"));
    T epsilon = static_cast<T>(ctx.Attr<float>("epsilon"));
Y
Yang Yu 已提交
252
    auto& param = Ref(ctx.Input<LoDTensor>("Param"), "Must set Param");
T
wip  
typhoonzero 已提交
253 254
    // auto& grad = Ref(ctx.Input<LoDTensor>("Grad"), "Must set Grad");
    auto* grad_var = ctx.InputVar("Grad");
Y
Yang Yu 已提交
255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271
    auto& mom1 = Ref(ctx.Input<LoDTensor>("Moment1"), "Must set Moment1");
    auto& mom2 = Ref(ctx.Input<LoDTensor>("Moment2"), "Must set Moment2");
    auto& lr =
        Ref(ctx.Input<LoDTensor>("LearningRate"), "Must set LearningRate");

    auto& beta1_pow =
        Ref(ctx.Input<LoDTensor>("Beta1Pow"), "Must set Beta1Pow");
    auto& beta2_pow =
        Ref(ctx.Input<LoDTensor>("Beta2Pow"), "Must set Beta2Pow");

    auto& param_out =
        Ref(ctx.Output<LoDTensor>("ParamOut"), "Must set ParamOut");
    auto& mom1_out =
        Ref(ctx.Output<LoDTensor>("Moment1Out"), "Must set Moment1Out");
    auto& mom2_out =
        Ref(ctx.Output<LoDTensor>("Moment2Out"), "Must set Moment1Out");

T
wip  
typhoonzero 已提交
272 273
    if (grad_var->IsType<framework::LoDTensor>()) {
      auto& grad = Ref(ctx.Input<LoDTensor>("Grad"), "Must set Grad");
274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301

      if (platform::is_cpu_place(ctx.GetPlace())) {
        AdamFunctor<T, CPUAdam> functor(
            beta1, beta2, epsilon, beta1_pow.template data<T>(),
            beta2_pow.template data<T>(), mom1.template data<T>(),
            mom1_out.template mutable_data<T>(ctx.GetPlace()),
            mom2.template data<T>(),
            mom2_out.template mutable_data<T>(ctx.GetPlace()),
            lr.template data<T>(), grad.template data<T>(),
            param.template data<T>(),
            param_out.template mutable_data<T>(ctx.GetPlace()));
        functor(param.numel());
      } else if (platform::is_gpu_place(ctx.GetPlace())) {
        AdamFunctor<T, GPUAdam> functor(
            beta1, beta2, epsilon, beta1_pow.template data<T>(),
            beta2_pow.template data<T>(), mom1.template data<T>(),
            mom1_out.template mutable_data<T>(ctx.GetPlace()),
            mom2.template data<T>(),
            mom2_out.template mutable_data<T>(ctx.GetPlace()),
            lr.template data<T>(), grad.template data<T>(),
            param.template data<T>(),
            param_out.template mutable_data<T>(ctx.GetPlace()));

        platform::ForRange<DeviceContext> for_range(
            static_cast<const DeviceContext&>(ctx.device_context()),
            param.numel());
        for_range(functor);
      }
T
wip  
typhoonzero 已提交
302 303 304
    } else if (grad_var->IsType<framework::SelectedRows>()) {
      auto& grad =
          Ref(ctx.Input<framework::SelectedRows>("Grad"), "Must set Grad");
305
      if (grad.rows().size() == 0) {
M
minqiyang 已提交
306
        VLOG(3) << "grad row size is 0!!";
307 308
        return;
      }
S
sneaxiy 已提交
309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334

      std::vector<int64_t> cpu_rows(grad.rows().begin(), grad.rows().end());
      bool is_strict_sorted = true;
      for (size_t i = 1; i < cpu_rows.size(); ++i) {
        if (cpu_rows[i - 1] >= cpu_rows[i]) {
          is_strict_sorted = false;
          break;
        }
      }

      const framework::SelectedRows* grad_merge_ptr;
      if (is_strict_sorted) {
        grad_merge_ptr = &grad;
      } else {
        // merge duplicated rows if any.
        // The rows of grad_merge have been sorted inside MergeAdd functor
        scatter::MergeAdd<DeviceContext, T> merge_func;
        auto* grad_merge_var = const_cast<framework::Scope&>(ctx.scope())
                                   .Var()
                                   ->GetMutable<framework::SelectedRows>();
        merge_func(ctx.template device_context<DeviceContext>(), grad,
                   grad_merge_var);
        grad_merge_ptr = grad_merge_var;
      }

      auto& grad_merge = *grad_merge_ptr;
T
wip  
typhoonzero 已提交
335
      auto& grad_tensor = grad_merge.value();
T
wip  
typhoonzero 已提交
336
      const T* grad_data = grad_tensor.template data<T>();
S
sneaxiy 已提交
337 338
      const int64_t* rows = nullptr;
// When compiled without CUDA, the CUDAData() interface should not be
339 340
// provided.
#if defined(PADDLE_WITH_CUDA)
D
dzhwinter 已提交
341
      if (platform::is_gpu_place(ctx.GetPlace())) {
S
sneaxiy 已提交
342
        rows = grad_merge.rows().CUDAData(ctx.GetPlace());
D
dzhwinter 已提交
343
      } else {
344
#endif
S
sneaxiy 已提交
345
        rows = grad_merge.rows().data();
346 347

#if defined(PADDLE_WITH_CUDA)
D
dzhwinter 已提交
348
      }
349
#endif
T
wip  
typhoonzero 已提交
350
      auto row_numel = grad_tensor.numel() / grad_merge.rows().size();
T
wip  
typhoonzero 已提交
351 352 353 354 355 356 357 358

      SparseAdamFunctor<T> functor(
          beta1, beta2, epsilon, beta1_pow.template data<T>(),
          beta2_pow.template data<T>(), mom1.template data<T>(),
          mom1_out.template mutable_data<T>(ctx.GetPlace()),
          mom2.template data<T>(),
          mom2_out.template mutable_data<T>(ctx.GetPlace()),
          lr.template data<T>(), grad_data, param.template data<T>(),
S
sneaxiy 已提交
359
          param_out.template mutable_data<T>(ctx.GetPlace()), rows, row_numel,
Q
Qiao Longfei 已提交
360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375
          grad_merge.rows().size(), sparse_mode);
      if (sparse_mode) {
        size_t row_count = grad_merge.rows().size();
        for (size_t row_index = 0; row_index < row_count; ++row_index) {
          for (size_t offset = 0; offset < row_numel; ++offset) {
            size_t i = rows[row_index] * row_numel + offset;
            T g = grad_data[row_index * row_numel + offset];
            functor.sparse_update(i, g);
          }
        }
      } else {
        platform::ForRange<DeviceContext> for_range(
            static_cast<const DeviceContext&>(ctx.device_context()),
            param.numel());
        for_range(functor);
      }
T
wip  
typhoonzero 已提交
376 377 378
    } else {
      PADDLE_THROW("Variable type not supported by adam_op");
    }
379 380 381 382 383
  }
};

}  // namespace operators
}  // namespace paddle