backward_test.cc 34.6 KB
Newer Older
Y
Yu Yang 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.

   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. */

Y
Yu Yang 已提交
15
#include "paddle/framework/backward.h"
D
dongzhihong 已提交
16

Y
Yu Yang 已提交
17
#include <gtest/gtest.h>
18 19
#include "paddle/framework/block_desc.h"
#include "paddle/framework/op_desc.h"
Y
Yu Yang 已提交
20
#include "paddle/framework/op_registry.h"
21
#include "paddle/framework/var_desc.h"
Y
Yan Chunwei 已提交
22
#include "paddle/operators/net_op.h"
Y
Yu Yang 已提交
23

Y
Yang Yu 已提交
24
USE_NO_KERNEL_OP(fill_constant);
Q
QI JUN 已提交
25

Y
Yu Yang 已提交
26 27 28
namespace paddle {
namespace framework {

D
dongzhihong 已提交
29 30
using DeviceContext = platform::DeviceContext;

Q
Qiao Longfei 已提交
31 32 33 34 35 36 37 38 39 40 41 42 43 44
class NoneOp : public framework::OperatorWithKernel {
 public:
  using framework::OperatorWithKernel::OperatorWithKernel;

 protected:
  void InferShape(framework::InferShapeContext *ctx) const override {}
};

template <typename Place, typename T>
class NoneKernel : public framework::OpKernel<T> {
 public:
  void Compute(const framework::ExecutionContext &context) const override {}
};

Y
Yu Yang 已提交
45
class RowWiseAddOpMaker : public OpProtoAndCheckerMaker {
Y
Yu Yang 已提交
46
 public:
Y
Yu Yang 已提交
47
  RowWiseAddOpMaker(OpProto *proto, OpAttrChecker *op_checker)
Y
Yu Yang 已提交
48
      : OpProtoAndCheckerMaker(proto, op_checker) {
49 50 51
    AddInput("X", "Input X of Add");
    AddInput("b", "Bias of Add");
    AddOutput("Out", "Out of Add");
Y
Yu Yang 已提交
52 53 54 55
    AddComment("Add Op");
  }
};

56 57 58 59 60
class RowWiseAddGradMaker : public SingleGradOpDescMaker {
 public:
  using SingleGradOpDescMaker::SingleGradOpDescMaker;

 protected:
Y
Yu Yang 已提交
61 62 63 64 65 66 67
  std::unique_ptr<OpDescBind> Apply() const override {
    auto grad_op = new OpDescBind();
    grad_op->SetInput(GradVarName("Out"), OutputGrad("Out"));
    grad_op->SetOutput(GradVarName("X"), InputGrad("X"));
    grad_op->SetOutput(GradVarName("b"), InputGrad("b"));
    grad_op->SetType("rowwise_add_grad");
    return std::unique_ptr<OpDescBind>(grad_op);
68 69 70
  }
};

Y
Yu Yang 已提交
71 72 73 74
class MulOpMaker : public OpProtoAndCheckerMaker {
 public:
  MulOpMaker(OpProto *proto, OpAttrChecker *op_checker)
      : OpProtoAndCheckerMaker(proto, op_checker) {
Y
Yu Yang 已提交
75 76
    AddInput("X", "A");
    AddInput("Y", "B");
Y
Yu Yang 已提交
77
    AddOutput("Out", "Out");
F
fengjiayi 已提交
78 79
    AddAttr<int>("x_num_col_dims", "").SetDefault(1).EqualGreaterThan(1);
    AddAttr<int>("y_num_col_dims", "").SetDefault(1).EqualGreaterThan(1);
Y
Yu Yang 已提交
80 81 82 83 84 85 86 87 88
    AddComment("Mul");
  }
};

class SigmoidOpMaker : public OpProtoAndCheckerMaker {
 public:
  SigmoidOpMaker(OpProto *proto, OpAttrChecker *op_checker)
      : OpProtoAndCheckerMaker(proto, op_checker) {
    AddInput("X", "X");
Y
Yu Yang 已提交
89
    AddOutput("Out", "Y");
Y
Yu Yang 已提交
90 91 92 93
    AddComment("Sigmoid");
  }
};

D
dongzhihong 已提交
94 95 96 97 98
class NoGradOpMaker : public OpProtoAndCheckerMaker {
 public:
  NoGradOpMaker(OpProto *proto, OpAttrChecker *op_checker)
      : OpProtoAndCheckerMaker(proto, op_checker) {
    AddInput("X", "X input");
Y
Yu Yang 已提交
99
    AddOutput("Out", "Y output");
D
dongzhihong 已提交
100 101 102 103
    AddComment("NoGradOp, same input output. no Grad");
  }
};

D
dongzhihong 已提交
104
class FcOp : public operators::NetOp {
Y
Yu Yang 已提交
105
 public:
Y
Yu Yang 已提交
106 107
  FcOp(const std::string &type, const VariableNameMap &inputs,
       const VariableNameMap &outputs, const AttributeMap &attrs)
Y
Yu Yang 已提交
108
      : NetOp(type, inputs, outputs, attrs) {
Y
Yiqun Liu 已提交
109 110 111
    AppendOp(OpRegistry::CreateOp(
        "mul", {{"X", {Input("X")}}, {"Y", {Input("W")}}},
        {{"Out", {Output("mul_result")}}}, AttributeMap{}));
112
    auto input_b = Inputs("b");
Y
Yu Yang 已提交
113
    std::string before_act = "mul_result";
114
    if (input_b.size() != 0) {
Y
Yu Yang 已提交
115
      AppendOp(OpRegistry::CreateOp(
116
          "rowwise_add", {{"X", {Output("mul_result")}}, {"b", {input_b[0]}}},
Y
Yiqun Liu 已提交
117
          {{"Out", {Output("add_result")}}}, AttributeMap{}));
Y
Yu Yang 已提交
118 119 120
      before_act = "add_result";
    } else {
      auto out_varname = Output("add_result");
121 122
      if (out_varname != kEmptyVarName) {
        this->Rename(out_varname, kEmptyVarName);
Y
Yu Yang 已提交
123
      }
Y
Yu Yang 已提交
124
    }
Y
Yu Yang 已提交
125

Y
Yu Yang 已提交
126
    AppendOp(OpRegistry::CreateOp("sigmoid", {{"X", {Output(before_act)}}},
Y
Yiqun Liu 已提交
127
                                  {{"Out", {Output("Out")}}}, AttributeMap{}));
Y
Yu Yang 已提交
128 129 130 131 132 133 134 135 136 137 138
    CompleteAddOp(false);
  }
};

class FcOpMaker : public OpProtoAndCheckerMaker {
 public:
  FcOpMaker(OpProto *proto, OpAttrChecker *op_checker)
      : OpProtoAndCheckerMaker(proto, op_checker) {
    AddInput("X", "x");
    AddInput("W", "w");
    AddInput("b", "b");
Y
Yu Yang 已提交
139 140
    AddOutput("mul_result", "").AsIntermediate();
    AddOutput("add_result", "").AsIntermediate();
Y
Yu Yang 已提交
141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160
    AddOutput("Out", "");
    AddComment("");
  }
};

class ManyOutputOpMaker : public OpProtoAndCheckerMaker {
 public:
  ManyOutputOpMaker(OpProto *proto, OpAttrChecker *op_checker)
      : OpProtoAndCheckerMaker(proto, op_checker) {
    AddInput("x", "x");
    AddOutput("y", "y");
    AddOutput("z", "z");
    AddComment("");
  }
};

class FillZeroOpMaker : public OpProtoAndCheckerMaker {
 public:
  FillZeroOpMaker(OpProto *proto, OpAttrChecker *op_checker)
      : OpProtoAndCheckerMaker(proto, op_checker) {
D
dangqingqing 已提交
161 162
    AddInput("X", "x");
    AddOutput("Y", "out");
Y
Yu Yang 已提交
163 164 165
    AddComment("");
  }
};
Y
Yu Yang 已提交
166

D
dongzhihong 已提交
167
class SumOpMaker : public framework::OpProtoAndCheckerMaker {
Y
Yu Yang 已提交
168
 public:
D
dongzhihong 已提交
169
  SumOpMaker(framework::OpProto *proto, framework::OpAttrChecker *op_checker)
Y
Yu Yang 已提交
170
      : OpProtoAndCheckerMaker(proto, op_checker) {
Y
Yu Yang 已提交
171 172
    AddInput("X", "the input tensors of sum operator.").AsDuplicable();
    AddOutput("Out", "the output tensor of sum operator.");
Y
Yu Yang 已提交
173 174 175
    AddComment("");
  }
};
D
dongzhihong 已提交
176

F
fengjiayi 已提交
177 178 179 180 181 182 183 184 185 186 187 188
class MultInOutOpMaker : public OpProtoAndCheckerMaker {
 public:
  MultInOutOpMaker(OpProto *proto, OpAttrChecker *op_checker)
      : OpProtoAndCheckerMaker(proto, op_checker) {
    AddInput("X", "x");
    AddInput("H", "h");
    AddOutput("Y", "y");
    AddOutput("Z", "z");
    AddComment("");
  }
};

189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227
class MinusGradOpDescMaker : public GradOpDescMakerBase {
 public:
  using GradOpDescMakerBase::GradOpDescMakerBase;

  std::vector<std::unique_ptr<OpDescBind>> operator()() const override {
    std::vector<std::unique_ptr<OpDescBind>> retv;
    auto x_g = InputGrad("X");
    if (!x_g.empty()) {
      auto *op_desc = new OpDescBind();
      op_desc->SetType("scale");
      op_desc->SetInput("X", OutputGrad("Out"));
      op_desc->SetOutput("Out", x_g);
      op_desc->SetAttr("scale", 1.0f);
      retv.emplace_back(op_desc);
    }

    auto y_g = InputGrad("Y");
    if (!y_g.empty()) {
      auto *op_desc = new OpDescBind();
      op_desc->SetType("scale");
      op_desc->SetInput("X", OutputGrad("Out"));
      op_desc->SetOutput("Out", y_g);
      op_desc->SetAttr("scale", -1.0f);
      retv.emplace_back(op_desc);
    }
    return retv;
  }
};

class MinusOpMaker : public OpProtoAndCheckerMaker {
 public:
  MinusOpMaker(OpProto *proto, OpAttrChecker *op_checker)
      : OpProtoAndCheckerMaker(proto, op_checker) {
    AddInput("X", "");
    AddInput("Y", "");
    AddOutput("Out", "");
    AddComment("minus for unittest");
  }
};
Y
Yu Yang 已提交
228 229 230 231
}  // namespace framework
}  // namespace paddle

namespace f = paddle::framework;
D
dongzhihong 已提交
232
namespace ops = paddle::operators;
Y
Yu Yang 已提交
233
using EnforceNotMet = paddle::platform::EnforceNotMet;
Q
Qiao Longfei 已提交
234 235
// rowwise_add
REGISTER_OPERATOR(rowwise_add, f::NoneOp, f::RowWiseAddOpMaker,
236
                  f::RowWiseAddGradMaker);
Q
Qiao Longfei 已提交
237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261
REGISTER_OP_CPU_KERNEL(rowwise_add,
                       f::NoneKernel<paddle::platform::CPUPlace, float>);
REGISTER_OPERATOR(rowwise_add_grad, f::NoneOp);
REGISTER_OP_CPU_KERNEL(rowwise_add_grad,
                       f::NoneKernel<paddle::platform::CPUPlace, float>);
// mul
REGISTER_OP(mul, f::NoneOp, f::MulOpMaker, mul_grad, f::NoneOp);
REGISTER_OP_CPU_KERNEL(mul, f::NoneKernel<paddle::platform::CPUPlace, float>);
REGISTER_OP_CPU_KERNEL(mul_grad,
                       f::NoneKernel<paddle::platform::CPUPlace, float>);
// sigmoid
REGISTER_OP(sigmoid, f::NoneOp, f::SigmoidOpMaker, sigmoid_grad, f::NoneOp);
REGISTER_OP_CPU_KERNEL(sigmoid,
                       f::NoneKernel<paddle::platform::CPUPlace, float>);
REGISTER_OP_WITHOUT_GRADIENT(nograd, f::NoneOp, f::NoGradOpMaker);
// fill_zeros_like
REGISTER_OP_WITHOUT_GRADIENT(fill_zeros_like, f::NoneOp, f::FillZeroOpMaker);
REGISTER_OP_CPU_KERNEL(fill_zeros_like,
                       f::NoneKernel<paddle::platform::CPUPlace, float>);
// sum
REGISTER_OP(sum, f::NoneOp, f::SumOpMaker, sum_grad, f::NoneOp);
REGISTER_OP_CPU_KERNEL(sum, f::NoneKernel<paddle::platform::CPUPlace, float>);
REGISTER_OP_CPU_KERNEL(sum_grad,
                       f::NoneKernel<paddle::platform::CPUPlace, float>);
// fc
F
fengjiayi 已提交
262
REGISTER_OP_WITHOUT_GRADIENT(fc, f::FcOp, f::FcOpMaker);
Q
Qiao Longfei 已提交
263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278
// many_output_op
REGISTER_OP(many_output_op, f::NoneOp, f::ManyOutputOpMaker,
            many_output_op_grad, f::NoneOp);
// mult_in_out
REGISTER_OP(mult_in_out, f::NoneOp, f::MultInOutOpMaker, mult_in_out_grad,
            f::NoneOp);
REGISTER_OP_CPU_KERNEL(mult_in_out,
                       f::NoneKernel<paddle::platform::CPUPlace, float>);
REGISTER_OP_CPU_KERNEL(mult_in_out_grad,
                       f::NoneKernel<paddle::platform::CPUPlace, float>);
// minus
REGISTER_OPERATOR(minus, f::NoneOp, f::MinusOpMaker, f::MinusGradOpDescMaker);
REGISTER_OP_CPU_KERNEL(minus, f::NoneKernel<paddle::platform::CPUPlace, float>);
// scale
REGISTER_OPERATOR(scale, f::NoneOp);
REGISTER_OP_CPU_KERNEL(scale, f::NoneKernel<paddle::platform::CPUPlace, float>);
Y
Yu Yang 已提交
279

280
TEST(Backward, simple_op_not_need_grad) {
Y
Yiqun Liu 已提交
281 282 283
  auto fwd =
      f::OpRegistry::CreateOp("rowwise_add", {{"X", {"x"}}, {"b", {"b"}}},
                              {{"Out", {"out"}}}, f::AttributeMap{});
284 285 286 287 288 289 290
  ASSERT_NE(fwd, nullptr);
  auto gop = f::Backward(*fwd, {"x"});
  ASSERT_EQ(gop->Output(f::GradVarName("X")), f::kEmptyVarName);

  auto no_input_gop = f::Backward(*fwd, {"x", "b"});
  ASSERT_NE(no_input_gop, nullptr);
  ASSERT_TRUE(no_input_gop->IsNetOp());
Y
Yu Yang 已提交
291
  ASSERT_EQ(0UL, static_cast<ops::NetOp *>(no_input_gop.get())->ops_.size());
292 293 294 295 296 297 298 299
}

TEST(Backward, net_fc_backward_normal) {
  std::shared_ptr<f::OperatorBase> fwd =
      f::OpRegistry::CreateOp("fc", {{"X", {"x"}}, {"W", {"w"}}, {"b", {"b"}}},
                              {{"mul_result", {"mul_res"}},
                               {"add_result", {"add_re"}},
                               {"Out", {"out"}}},
Y
Yiqun Liu 已提交
300
                              f::AttributeMap{});
301
  ASSERT_NE(fwd, nullptr);
Y
Yiqun Liu 已提交
302 303
  std::shared_ptr<f::OperatorBase> gop =
      f::Backward(*fwd, std::unordered_set<std::string>{});
304 305 306 307 308 309 310 311
  ASSERT_TRUE(gop->IsNetOp());
  auto net = static_cast<ops::NetOp *>(gop.get());

  ASSERT_NO_THROW(net->DebugString());

  ASSERT_EQ(3UL, net->ops_.size());

  f::OperatorBase &d_sigmoid = *net->ops_[0];
Q
qiaolongfei 已提交
312
  ASSERT_EQ("sigmoid_grad", d_sigmoid.Type());
313 314

  f::OperatorBase &d_add = *net->ops_[1];
Q
qiaolongfei 已提交
315
  ASSERT_EQ("rowwise_add_grad", d_add.Type());
316 317

  f::OperatorBase &d_mul = *net->ops_[2];
Q
qiaolongfei 已提交
318
  ASSERT_EQ("mul_grad", d_mul.Type());
319 320 321 322 323 324 325 326
}

TEST(Backward, net_fc_backward_not_have_b) {
  std::shared_ptr<f::OperatorBase> fwd =
      f::OpRegistry::CreateOp("fc", {{"X", {"x"}}, {"W", {"w"}}, {"b", {}}},
                              {{"mul_result", {"mul_res"}},
                               {"add_result", {"add_res"}},
                               {"Out", {"tmp"}}},
Y
Yiqun Liu 已提交
327
                              f::AttributeMap{});
328
  ASSERT_NE(fwd, nullptr);
Y
Yiqun Liu 已提交
329 330
  std::shared_ptr<f::OperatorBase> gop =
      f::Backward(*fwd, std::unordered_set<std::string>{});
331 332 333 334 335 336 337 338
  ASSERT_TRUE(gop->IsNetOp());
  auto net = static_cast<ops::NetOp *>(gop.get());

  ASSERT_NO_THROW(net->DebugString());

  ASSERT_EQ(2UL, net->ops_.size());

  f::OperatorBase &d_sigmoid = *net->ops_[0];
Q
qiaolongfei 已提交
339
  ASSERT_EQ("sigmoid_grad", d_sigmoid.Type());
340 341

  f::OperatorBase &d_mul = *net->ops_[1];
Q
qiaolongfei 已提交
342
  ASSERT_EQ("mul_grad", d_mul.Type());
343 344 345 346
}

TEST(Backward, net_input_of_network_not_need_grad) {
  ops::NetOp net;
Y
Yu Yang 已提交
347
  net.AppendOp(f::OpRegistry::CreateOp(
348 349 350 351
      "fc", {{"X", {"x"}}, {"W", {"W1"}}, {"b", {"b1"}}},
      {{"mul_result", {"mul_tmp_0"}},
       {"add_result", {"add_tmp_0"}},
       {"Out", {"hidden0"}}},
Y
Yiqun Liu 已提交
352
      f::AttributeMap{}));
Y
Yu Yang 已提交
353
  net.AppendOp(f::OpRegistry::CreateOp(
354 355 356 357
      "fc", {{"X", {"hidden0"}}, {"W", {"W2"}}, {"b", {"b2"}}},
      {{"mul_result", {"mul_tmp_1"}},
       {"add_result", {"add_tmp_1"}},
       {"Out", {"hidden1"}}},
Y
Yiqun Liu 已提交
358
      f::AttributeMap{}));
359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385
  net.CompleteAddOp();
  auto bwd = Backward(net, {"x"});  // x@GRAD is not need.
  ASSERT_TRUE(bwd->IsNetOp());
  auto bwd_net = static_cast<ops::NetOp *>(bwd.get());

  auto output_vars = bwd_net->OutputVars(true);
  std::unordered_set<std::string> all_outputs =
      std::unordered_set<std::string>(output_vars.begin(), output_vars.end());
  all_outputs.erase(f::kEmptyVarName);

  for (auto &out : {"W1", "b1", "hidden0", "W2", "b2"}) {
    ASSERT_NE(all_outputs.find(f::GradVarName(out)), all_outputs.end());
  }

  // Not Generated X
  ASSERT_EQ(all_outputs.find(f::GradVarName("X")), all_outputs.end());

  ASSERT_EQ(2UL, bwd_net->ops_.size());
  ASSERT_TRUE(bwd_net->ops_[1]->IsNetOp());
  auto first_fc_grad = static_cast<ops::NetOp *>(bwd_net->ops_[1].get());
  ASSERT_EQ(3UL, first_fc_grad->ops_.size());
  ASSERT_EQ(f::kEmptyVarName,
            first_fc_grad->ops_[2]->Output(f::GradVarName("X")));
}

TEST(Backward, net_shared_weight) {
  ops::NetOp net;
Y
Yu Yang 已提交
386
  net.AppendOp(f::OpRegistry::CreateOp("mul", {{"X", {"x"}}, {"Y", {"w"}}},
Y
Yiqun Liu 已提交
387
                                       {{"Out", {"out"}}}, f::AttributeMap{}));
Y
Yu Yang 已提交
388
  net.AppendOp(f::OpRegistry::CreateOp("mul", {{"X", {"out"}}, {"Y", {"w"}}},
Y
Yiqun Liu 已提交
389 390
                                       {{"Out", {"FinalOut"}}},
                                       f::AttributeMap{}));
391 392
  net.CompleteAddOp();

Y
Yiqun Liu 已提交
393
  auto bwd = f::Backward(net, std::unordered_set<std::string>{});
394 395 396
  ASSERT_TRUE(bwd->IsNetOp());
  auto bwd_net = static_cast<ops::NetOp *>(bwd.get());
  ASSERT_EQ(3UL, bwd_net->ops_.size());
D
dongzhihong 已提交
397
  ASSERT_EQ("sum", bwd_net->ops_[2]->Type());
398 399 400
}

TEST(Backward, op_all_input_are_not_need) {
Y
Yiqun Liu 已提交
401 402 403
  auto fwd =
      f::OpRegistry::CreateOp("rowwise_add", {{"X", {"x"}}, {"b", {"b"}}},
                              {{"Out", {"out"}}}, f::AttributeMap{});
404 405 406 407 408 409 410
  auto backward = f::Backward(*fwd, {"x", "b"});
  ASSERT_TRUE(backward->IsNetOp());
  auto net = static_cast<ops::NetOp *>(backward.get());
  ASSERT_TRUE(net->ops_.empty());
}

TEST(Backward, op_all_output_are_not_need) {
Y
Yiqun Liu 已提交
411 412 413
  auto fwd =
      f::OpRegistry::CreateOp("rowwise_add", {{"X", {"x"}}, {"b", {"b"}}},
                              {{"Out", {"out"}}}, f::AttributeMap{});
414 415 416 417 418 419 420
  auto backward = f::Backward(*fwd, {"out"});
  ASSERT_TRUE(backward->IsNetOp());
  auto net = static_cast<ops::NetOp *>(backward.get());
  ASSERT_TRUE(net->ops_.empty());
}

TEST(Backward, op_part_of_output_are_not_need) {
Y
Yiqun Liu 已提交
421 422 423
  auto fwd =
      f::OpRegistry::CreateOp("many_output_op", {{"x", {"X"}}},
                              {{"y", {"Y"}}, {"z", {"Z"}}}, f::AttributeMap{});
424 425 426 427 428 429
  auto backward = f::Backward(*fwd, {"Z"});
  ASSERT_TRUE(backward->IsNetOp());
  auto net = static_cast<ops::NetOp *>(backward.get());
  ASSERT_EQ(net->ops_.size(), 2UL);

  auto &fill_zero = *net->ops_[0];
Q
qiaolongfei 已提交
430
  ASSERT_EQ("fill_zeros_like", fill_zero.Type());
D
dangqingqing 已提交
431 432
  ASSERT_EQ(1UL, fill_zero.Inputs("X").size());
  ASSERT_EQ("Z", fill_zero.Input("X"));
F
fengjiayi 已提交
433 434
  ASSERT_EQ(1UL, fill_zero.Outputs("Out").size());
  ASSERT_EQ(std::string("Z") + f::kZeroVarSuffix, fill_zero.Output("Out"));
435 436

  auto &d_many_out = *net->ops_[1];
Q
qiaolongfei 已提交
437 438
  ASSERT_EQ("many_output_op_grad", d_many_out.Type());
  ASSERT_EQ(1UL + 2UL + 2UL, d_many_out.Inputs().size());  // I/O/OG
439 440 441 442 443 444 445 446
  ASSERT_EQ(std::string("Z") + f::kZeroVarSuffix,
            d_many_out.Input(f::GradVarName("z")));
  ASSERT_EQ(f::GradVarName("Y"), d_many_out.Input(f::GradVarName("y")));
  ASSERT_EQ(f::GradVarName("X"), d_many_out.Output(f::GradVarName("x")));
}

TEST(Backward, op_part_of_input_are_not_need) {
  auto fwd = f::OpRegistry::CreateOp("mul", {{"X", {"a"}}, {"Y", {"b"}}},
Y
Yiqun Liu 已提交
447
                                     {{"Out", {"out"}}}, f::AttributeMap{});
448 449
  auto backward = f::Backward(*fwd, {"a"});
  auto &grad_mul = *backward;
Q
qiaolongfei 已提交
450 451 452
  ASSERT_EQ(grad_mul.Type(), "mul_grad");
  ASSERT_EQ(grad_mul.Inputs().size(), 2UL + 1UL + 1UL);
  ASSERT_EQ(grad_mul.Outputs().size(), 2UL);
453 454 455 456 457 458 459 460 461 462
  ASSERT_EQ(grad_mul.Output(f::GradVarName("X")), f::kEmptyVarName);
  ASSERT_EQ(grad_mul.Output(f::GradVarName("Y")), f::GradVarName("b"));
  ASSERT_EQ(grad_mul.Input(f::GradVarName("Out")), f::GradVarName("out"));
  ASSERT_EQ(grad_mul.Input("X"), "a");
  ASSERT_EQ(grad_mul.Input("Y"), "b");
  ASSERT_EQ(grad_mul.Input("Out"), "out");
}

TEST(Backward, linear_net_intermediate_variable_has_no_grad) {
  ops::NetOp net;
Y
Yu Yang 已提交
463
  net.AppendOp(f::OpRegistry::CreateOp(
464 465 466 467
      "fc", {{"X", {"x1"}}, {"W", {"w1"}}, {"b", {"b1"}}},
      {{"mul_result", {"mul_out1"}},
       {"add_result", {"add_out1"}},
       {"Out", {"out1"}}},
Y
Yiqun Liu 已提交
468
      f::AttributeMap{}));
Y
Yu Yang 已提交
469
  net.AppendOp(f::OpRegistry::CreateOp(
470 471 472 473
      "fc", {{"X", {"out1"}}, {"W", {"w2"}}, {"b", {"b2"}}},
      {{"mul_result", {"mul_out2"}},
       {"add_result", {"tmp_out2"}},
       {"Out", {"out2"}}},
Y
Yiqun Liu 已提交
474
      f::AttributeMap{}));
Y
Yu Yang 已提交
475
  net.AppendOp(f::OpRegistry::CreateOp(
476 477 478 479
      "fc", {{"X", {"out2"}}, {"W", {"w3"}}, {"b", {"b3"}}},
      {{"mul_result", {"mul_out3"}},
       {"add_result", {"tmp_out3"}},
       {"Out", {"out3"}}},
Y
Yiqun Liu 已提交
480
      f::AttributeMap{}));
481 482 483 484 485 486 487
  net.CompleteAddOp();

  auto backward = f::Backward(net, {"mul_out2", "tmp_out2", "out2"});
  ASSERT_TRUE(backward->IsNetOp());
  auto bwd_net = static_cast<ops::NetOp *>(backward.get());
  ASSERT_EQ(bwd_net->ops_.size(), 3UL);
  auto &grad_fc = *bwd_net->ops_[0];
Y
Yu Yang 已提交
488 489

  const char *all = paddle::operators::NetOp::kAll;
Q
qiaolongfei 已提交
490
  EXPECT_EQ(grad_fc.Inputs(all).size(),
491 492 493
            2UL       /* external input number */
                + 1UL /* external output number*/
                + 1UL /* number of gradient of external output*/
494 495
                + 2UL /* internal variable number*/
            );
Q
qiaolongfei 已提交
496
  EXPECT_EQ(grad_fc.Outputs(all).size(),
497
            2UL       /* input number of mul*/
498 499 500
                + 2UL /* input number of rowwise_add*/
                + 1UL /* input number of sigmod */
                - 1UL /* out2 is not needed*/);
Q
qiaolongfei 已提交
501 502 503 504
  EXPECT_EQ(bwd_net->ops_[1]->Inputs(all).size(), 0UL);
  EXPECT_EQ(bwd_net->ops_[1]->Outputs(all).size(), 0UL);
  EXPECT_EQ(bwd_net->ops_[2]->Inputs(all).size(), 0UL);
  EXPECT_EQ(bwd_net->ops_[2]->Outputs(all).size(), 0UL);
505
}
506 507

TEST(Backward, simple_single_op) {
508
  f::ProgramDescBind program;
509
  f::BlockDescBind *block = program.MutableBlock(0);
Q
Qiao Longfei 已提交
510

511 512 513 514 515 516
  f::OpDescBind *op = block->AppendOp();
  op->SetType("rowwise_add");
  op->SetInput("X", {"x"});
  op->SetInput("b", {"b"});
  op->SetOutput("Out", {"out"});

517
  auto target = f::VarDescBind("out");
518
  target.SetShape({1});
Y
Yiqun Liu 已提交
519 520
  auto var_to_grad =
      AppendBackward(program, target, std::unordered_set<std::string>{});
521

522 523 524 525 526
  ASSERT_EQ(block->AllOps().size(), 3UL);
  f::OpDescBind *fill_op = block->AllOps()[1];
  EXPECT_EQ(fill_op->Type(), "fill_constant");

  f::OpDescBind *grad_op = block->AllOps()[2];
527 528 529 530 531 532 533 534 535
  EXPECT_EQ(grad_op->Type(), "rowwise_add_grad");
  ASSERT_EQ(grad_op->InputNames().size(), 1UL);
  ASSERT_EQ(grad_op->OutputNames().size(), 2UL);
  EXPECT_EQ(grad_op->Input(f::GradVarName("Out")),
            std::vector<std::string>({f::GradVarName("out")}));
  EXPECT_EQ(grad_op->Output(f::GradVarName("X")),
            std::vector<std::string>({f::GradVarName("x")}));
  EXPECT_EQ(grad_op->Output(f::GradVarName("b")),
            std::vector<std::string>({f::GradVarName("b")}));
F
fengjiayi 已提交
536

Q
Qiao Longfei 已提交
537
  EXPECT_EQ(var_to_grad.size(), 3UL);
F
fengjiayi 已提交
538 539 540 541 542
  EXPECT_EQ(var_to_grad.at("b"), f::GradVarInfo(f::GradVarName("b"), 0, 2));
  EXPECT_EQ(var_to_grad.at("x"), f::GradVarInfo(f::GradVarName("x"), 0, 2));

  EXPECT_TRUE(block->HasVar(f::GradVarName("b")));
  EXPECT_TRUE(block->HasVar(f::GradVarName("x")));
543 544
}

F
fengjiayi 已提交
545
TEST(Backward, default_attribute) {
546
  f::ProgramDescBind program;
547
  f::BlockDescBind *block = program.MutableBlock(0);
F
fengjiayi 已提交
548 549 550 551 552
  f::OpDescBind *op = block->AppendOp();
  op->SetType("mul");
  op->SetInput("X", {"x"});
  op->SetInput("Y", {"y"});
  op->SetOutput("Out", {"out"});
553
  op->CheckAttrs();
F
fengjiayi 已提交
554

555
  auto target = f::VarDescBind("out");
556
  target.SetShape({1});
Y
Yiqun Liu 已提交
557
  AppendBackward(program, target, std::unordered_set<std::string>{});
F
fengjiayi 已提交
558

559
  ASSERT_EQ(block->AllOps().size(), 3UL);
F
fengjiayi 已提交
560 561 562
  EXPECT_EQ(boost::get<int>(op->GetAttr("x_num_col_dims")), 1);
  EXPECT_EQ(boost::get<int>(op->GetAttr("y_num_col_dims")), 1);

563 564 565 566
  f::OpDescBind *fill_op = block->AllOps()[1];
  EXPECT_EQ(fill_op->Type(), "fill_constant");

  f::OpDescBind *grad_op = block->AllOps()[2];
F
fengjiayi 已提交
567 568 569 570 571
  ASSERT_EQ(grad_op->Type(), "mul_grad");
  EXPECT_EQ(boost::get<int>(grad_op->GetAttr("x_num_col_dims")), 1);
  EXPECT_EQ(boost::get<int>(grad_op->GetAttr("y_num_col_dims")), 1);
}

572
TEST(Backward, simple_mult_op) {
573
  f::ProgramDescBind program;
574
  f::BlockDescBind *block = program.MutableBlock(0);
575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592
  f::OpDescBind *op1 = block->AppendOp();
  op1->SetType("rowwise_add");
  op1->SetInput("X", {"x1"});
  op1->SetInput("b", {"b1"});
  op1->SetOutput("Out", {"out1"});

  f::OpDescBind *op2 = block->AppendOp();
  op2->SetType("mul");
  op2->SetInput("X", {"out1"});
  op2->SetInput("Y", {"y2"});
  op2->SetOutput("Out", {"out2"});

  f::OpDescBind *op3 = block->AppendOp();
  op3->SetType("rowwise_add");
  op3->SetInput("X", {"out2"});
  op3->SetInput("b", {"b3"});
  op3->SetOutput("Out", {"out3"});

593
  auto target = f::VarDescBind("out3");
594
  target.SetShape({1});
595
  size_t forward_len = block->AllOps().size();
Y
Yiqun Liu 已提交
596 597
  auto var_to_grad =
      AppendBackward(program, target, std::unordered_set<std::string>{});
598

599 600 601 602 603
  ASSERT_EQ(block->AllOps().size(), 6UL + 1);
  f::OpDescBind *fill_op = block->AllOps()[forward_len];
  EXPECT_EQ(fill_op->Type(), "fill_constant");

  f::OpDescBind *grad_op1 = block->AllOps()[6];
604 605 606 607 608 609 610 611 612 613
  EXPECT_EQ(grad_op1->Type(), "rowwise_add_grad");
  ASSERT_EQ(grad_op1->InputNames().size(), 1UL);
  ASSERT_EQ(grad_op1->OutputNames().size(), 2UL);
  EXPECT_EQ(grad_op1->Input(f::GradVarName("Out")),
            std::vector<std::string>({f::GradVarName("out1")}));
  EXPECT_EQ(grad_op1->Output(f::GradVarName("X")),
            std::vector<std::string>({f::GradVarName("x1")}));
  EXPECT_EQ(grad_op1->Output(f::GradVarName("b")),
            std::vector<std::string>({f::GradVarName("b1")}));

614
  f::OpDescBind *grad_op2 = block->AllOps()[5];
615 616 617 618 619 620 621 622 623 624 625 626 627
  EXPECT_EQ(grad_op2->Type(), "mul_grad");
  ASSERT_EQ(grad_op2->InputNames().size(), 4UL);
  ASSERT_EQ(grad_op2->OutputNames().size(), 2UL);
  EXPECT_EQ(grad_op2->Input("X"), std::vector<std::string>({"out1"}));
  EXPECT_EQ(grad_op2->Input("Y"), std::vector<std::string>({"y2"}));
  EXPECT_EQ(grad_op2->Input("Out"), std::vector<std::string>({"out2"}));
  EXPECT_EQ(grad_op2->Input(f::GradVarName("Out")),
            std::vector<std::string>({f::GradVarName("out2")}));
  EXPECT_EQ(grad_op2->Output(f::GradVarName("X")),
            std::vector<std::string>({f::GradVarName("out1")}));
  EXPECT_EQ(grad_op2->Output(f::GradVarName("Y")),
            std::vector<std::string>({f::GradVarName("y2")}));

628
  f::OpDescBind *grad_op3 = block->AllOps()[4];
629 630 631 632 633 634 635 636 637
  EXPECT_EQ(grad_op3->Type(), "rowwise_add_grad");
  ASSERT_EQ(grad_op3->InputNames().size(), 1UL);
  ASSERT_EQ(grad_op3->OutputNames().size(), 2UL);
  EXPECT_EQ(grad_op3->Input(f::GradVarName("Out")),
            std::vector<std::string>({f::GradVarName("out3")}));
  EXPECT_EQ(grad_op3->Output(f::GradVarName("X")),
            std::vector<std::string>({f::GradVarName("out2")}));
  EXPECT_EQ(grad_op3->Output(f::GradVarName("b")),
            std::vector<std::string>({f::GradVarName("b3")}));
F
fengjiayi 已提交
638

Q
Qiao Longfei 已提交
639
  EXPECT_EQ(var_to_grad.size(), 7UL);
F
fengjiayi 已提交
640 641 642 643 644 645 646 647 648 649 650 651 652 653 654
  EXPECT_EQ(var_to_grad.at("x1"), f::GradVarInfo(f::GradVarName("x1"), 0, 6));
  EXPECT_EQ(var_to_grad.at("b1"), f::GradVarInfo(f::GradVarName("b1"), 0, 6));
  EXPECT_EQ(var_to_grad.at("out1"),
            f::GradVarInfo(f::GradVarName("out1"), 0, 5));
  EXPECT_EQ(var_to_grad.at("y2"), f::GradVarInfo(f::GradVarName("y2"), 0, 5));
  EXPECT_EQ(var_to_grad.at("out2"),
            f::GradVarInfo(f::GradVarName("out2"), 0, 4));
  EXPECT_EQ(var_to_grad.at("b3"), f::GradVarInfo(f::GradVarName("b3"), 0, 4));

  EXPECT_TRUE(block->HasVar(f::GradVarName("x1")));
  EXPECT_TRUE(block->HasVar(f::GradVarName("b1")));
  EXPECT_TRUE(block->HasVar(f::GradVarName("out1")));
  EXPECT_TRUE(block->HasVar(f::GradVarName("y2")));
  EXPECT_TRUE(block->HasVar(f::GradVarName("out2")));
  EXPECT_TRUE(block->HasVar(f::GradVarName("b3")));
F
fengjiayi 已提交
655 656 657
}

TEST(Backward, intermedia_var_no_grad) {
658
  f::ProgramDescBind program;
659
  f::BlockDescBind *block = program.MutableBlock(0);
F
fengjiayi 已提交
660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683
  f::OpDescBind *op1 = block->AppendOp();
  op1->SetType("rowwise_add");
  op1->SetInput("X", {"x1"});
  op1->SetInput("b", {"b1"});
  op1->SetOutput("Out", {"out1"});

  f::OpDescBind *op2 = block->AppendOp();
  op2->SetType("mul");
  op2->SetInput("X", {"x2"});
  op2->SetInput("Y", {"y2"});
  op2->SetOutput("Out", {"out2"});

  f::OpDescBind *op3 = block->AppendOp();
  op3->SetType("rowwise_add");
  op3->SetInput("X", {"out2"});
  op3->SetInput("b", {"b3"});
  op3->SetOutput("Out", {"out3"});

  f::OpDescBind *op4 = block->AppendOp();
  op4->SetType("mul");
  op4->SetInput("X", {"out1"});
  op4->SetInput("Y", {"out3"});
  op4->SetOutput("Out", {"out4"});

684
  auto target = f::VarDescBind("out4");
685
  target.SetShape({1});
686
  size_t forward_len = block->AllOps().size();
F
fengjiayi 已提交
687
  auto var_to_grad = AppendBackward(program, target, {"out3"});
F
fengjiayi 已提交
688

689 690 691 692 693
  ASSERT_EQ(block->AllOps().size(), 7UL);
  f::OpDescBind *fill_op = block->AllOps()[forward_len];
  EXPECT_EQ(fill_op->Type(), "fill_constant");

  f::OpDescBind *grad_op1 = block->AllOps()[6];
F
fengjiayi 已提交
694 695 696 697 698 699 700 701 702 703
  EXPECT_EQ(grad_op1->Type(), "rowwise_add_grad");
  ASSERT_EQ(grad_op1->InputNames().size(), 1UL);
  ASSERT_EQ(grad_op1->OutputNames().size(), 2UL);
  EXPECT_EQ(grad_op1->Input(f::GradVarName("Out")),
            std::vector<std::string>({f::GradVarName("out1")}));
  EXPECT_EQ(grad_op1->Output(f::GradVarName("X")),
            std::vector<std::string>({f::GradVarName("x1")}));
  EXPECT_EQ(grad_op1->Output(f::GradVarName("b")),
            std::vector<std::string>({f::GradVarName("b1")}));

704
  f::OpDescBind *grad_op4 = block->AllOps()[5];
F
fengjiayi 已提交
705 706 707 708 709 710 711 712 713 714
  EXPECT_EQ(grad_op4->Type(), "mul_grad");
  ASSERT_EQ(grad_op4->InputNames().size(), 4UL);
  ASSERT_EQ(grad_op4->OutputNames().size(), 2UL);
  EXPECT_EQ(grad_op4->Input("X"), std::vector<std::string>({"out1"}));
  EXPECT_EQ(grad_op4->Input("Y"), std::vector<std::string>({"out3"}));
  EXPECT_EQ(grad_op4->Input("Out"), std::vector<std::string>({"out4"}));
  EXPECT_EQ(grad_op4->Input(f::GradVarName("Out")),
            std::vector<std::string>({f::GradVarName("out4")}));
  EXPECT_EQ(grad_op4->Output(f::GradVarName("X")),
            std::vector<std::string>({f::GradVarName("out1")}));
715
  EXPECT_EQ(grad_op4->Output(f::GradVarName("Y")), std::vector<std::string>());
F
fengjiayi 已提交
716

Q
Qiao Longfei 已提交
717
  EXPECT_EQ(var_to_grad.size(), 4UL);
F
fengjiayi 已提交
718 719 720 721 722 723 724 725
  EXPECT_EQ(var_to_grad.at("x1"), f::GradVarInfo(f::GradVarName("x1"), 0, 6));
  EXPECT_EQ(var_to_grad.at("b1"), f::GradVarInfo(f::GradVarName("b1"), 0, 6));
  EXPECT_EQ(var_to_grad.at("out1"),
            f::GradVarInfo(f::GradVarName("out1"), 0, 5));

  EXPECT_TRUE(block->HasVar(f::GradVarName("x1")));
  EXPECT_TRUE(block->HasVar(f::GradVarName("b1")));
  EXPECT_TRUE(block->HasVar(f::GradVarName("out1")));
F
fengjiayi 已提交
726 727 728
}

TEST(Backward, var_no_grad) {
729
  f::ProgramDescBind program;
730
  f::BlockDescBind *block = program.MutableBlock(0);
F
fengjiayi 已提交
731 732 733 734 735 736 737 738 739 740 741 742 743 744
  f::OpDescBind *op1 = block->AppendOp();
  op1->SetType("mult_in_out");
  op1->SetInput("X", {"x1"});
  op1->SetInput("H", {"h1"});
  op1->SetOutput("Y", {"y1"});
  op1->SetOutput("Z", {"z1"});

  f::OpDescBind *op2 = block->AppendOp();
  op2->SetType("mult_in_out");
  op2->SetInput("X", {"y1"});
  op2->SetInput("H", {"z1"});
  op2->SetOutput("Y", {"y2"});
  op2->SetOutput("Z", {"z2"});

745
  auto target = f::VarDescBind("z2");
746
  target.SetShape({1});
747
  size_t forward_len = block->AllOps().size();
F
fengjiayi 已提交
748
  auto var_to_grad = AppendBackward(program, target, {"z1"});
F
fengjiayi 已提交
749

750 751 752 753 754
  ASSERT_EQ(block->AllOps().size(), 6UL);
  f::OpDescBind *fill_op = block->AllOps()[forward_len];
  EXPECT_EQ(fill_op->Type(), "fill_constant");

  f::OpDescBind *grad_op2 = block->AllOps()[3];
F
fengjiayi 已提交
755 756 757 758 759 760 761 762 763 764 765 766 767
  ASSERT_EQ(grad_op2->Type(), "mult_in_out_grad");
  ASSERT_EQ(grad_op2->InputNames().size(), 6UL);
  ASSERT_EQ(grad_op2->OutputNames().size(), 2UL);
  EXPECT_EQ(grad_op2->Input("X"), std::vector<std::string>({"y1"}));
  EXPECT_EQ(grad_op2->Input("H"), std::vector<std::string>({"z1"}));
  EXPECT_EQ(grad_op2->Input("Y"), std::vector<std::string>({"y2"}));
  EXPECT_EQ(grad_op2->Input("Z"), std::vector<std::string>({"z2"}));
  EXPECT_EQ(grad_op2->Input(f::GradVarName("Y")),
            std::vector<std::string>({f::GradVarName("y2")}));
  EXPECT_EQ(grad_op2->Input(f::GradVarName("Z")),
            std::vector<std::string>({f::GradVarName("z2")}));
  EXPECT_EQ(grad_op2->Output(f::GradVarName("X")),
            std::vector<std::string>({f::GradVarName("y1")}));
768
  EXPECT_EQ(grad_op2->Output(f::GradVarName("H")), std::vector<std::string>());
F
fengjiayi 已提交
769

770
  f::OpDescBind *fill_zero_op = block->AllOps()[4];
F
fengjiayi 已提交
771 772 773 774
  ASSERT_EQ(fill_zero_op->Type(), "fill_zeros_like");
  ASSERT_EQ(fill_zero_op->InputNames().size(), 1UL);
  ASSERT_EQ(fill_zero_op->OutputNames().size(), 1UL);
  EXPECT_EQ(fill_zero_op->Input("X"), std::vector<std::string>({"z1"}));
F
fengjiayi 已提交
775
  EXPECT_EQ(fill_zero_op->Output("Out"),
F
fengjiayi 已提交
776 777
            std::vector<std::string>({std::string("z1") + f::kZeroVarSuffix}));

778
  f::OpDescBind *grad_op1 = block->AllOps()[5];
F
fengjiayi 已提交
779 780 781 782 783 784 785 786 787 788 789 790 791 792 793
  ASSERT_EQ(grad_op1->Type(), "mult_in_out_grad");
  ASSERT_EQ(grad_op1->InputNames().size(), 6UL);
  ASSERT_EQ(grad_op1->OutputNames().size(), 2UL);
  EXPECT_EQ(grad_op1->Input("X"), std::vector<std::string>({"x1"}));
  EXPECT_EQ(grad_op1->Input("H"), std::vector<std::string>({"h1"}));
  EXPECT_EQ(grad_op1->Input("Y"), std::vector<std::string>({"y1"}));
  EXPECT_EQ(grad_op1->Input("Z"), std::vector<std::string>({"z1"}));
  EXPECT_EQ(grad_op1->Input(f::GradVarName("Y")),
            std::vector<std::string>({f::GradVarName("y1")}));
  EXPECT_EQ(grad_op1->Input(f::GradVarName("Z")),
            std::vector<std::string>({std::string("z1") + f::kZeroVarSuffix}));
  EXPECT_EQ(grad_op1->Output(f::GradVarName("X")),
            std::vector<std::string>({f::GradVarName("x1")}));
  EXPECT_EQ(grad_op1->Output(f::GradVarName("H")),
            std::vector<std::string>({f::GradVarName("h1")}));
F
fengjiayi 已提交
794

Q
Qiao Longfei 已提交
795
  EXPECT_EQ(var_to_grad.size(), 4UL);
F
fengjiayi 已提交
796 797 798 799 800 801 802
  EXPECT_EQ(var_to_grad.at("y1"), f::GradVarInfo(f::GradVarName("y1"), 0, 3));
  EXPECT_EQ(var_to_grad.at("x1"), f::GradVarInfo(f::GradVarName("x1"), 0, 5));
  EXPECT_EQ(var_to_grad.at("h1"), f::GradVarInfo(f::GradVarName("h1"), 0, 5));

  EXPECT_TRUE(block->HasVar(f::GradVarName("y1")));
  EXPECT_TRUE(block->HasVar(f::GradVarName("x1")));
  EXPECT_TRUE(block->HasVar(f::GradVarName("h1")));
F
fengjiayi 已提交
803 804 805
}

TEST(Backward, shared_var) {
806
  f::ProgramDescBind program;
807
  f::BlockDescBind *block = program.MutableBlock(0);
F
fengjiayi 已提交
808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825
  f::OpDescBind *op1 = block->AppendOp();
  op1->SetType("rowwise_add");
  op1->SetInput("X", {"x1"});
  op1->SetInput("b", {"b1"});
  op1->SetOutput("Out", {"out1"});

  f::OpDescBind *op2 = block->AppendOp();
  op2->SetType("mul");
  op2->SetInput("X", {"out1"});
  op2->SetInput("Y", {"y2"});
  op2->SetOutput("Out", {"out2"});

  f::OpDescBind *op3 = block->AppendOp();
  op3->SetType("rowwise_add");
  op3->SetInput("X", {"out1"});
  op3->SetInput("b", {"b3"});
  op3->SetOutput("Out", {"out3"});

826
  auto target = f::VarDescBind("out3");
827
  target.SetShape({1});
828
  size_t forward_len = block->AllOps().size();
Y
Yiqun Liu 已提交
829 830
  auto var_to_grad =
      AppendBackward(program, target, std::unordered_set<std::string>{});
F
fengjiayi 已提交
831

832 833 834 835 836
  ASSERT_EQ(block->AllOps().size(), 8UL);
  f::OpDescBind *fill_op = block->AllOps()[forward_len];
  EXPECT_EQ(fill_op->Type(), "fill_constant");

  f::OpDescBind *grad_op3 = block->AllOps()[4];
F
fengjiayi 已提交
837 838 839 840 841 842 843 844 845 846
  ASSERT_EQ(grad_op3->Type(), "rowwise_add_grad");
  ASSERT_EQ(grad_op3->InputNames().size(), 1UL);
  ASSERT_EQ(grad_op3->OutputNames().size(), 2UL);
  EXPECT_EQ(grad_op3->Input(f::GradVarName("Out")),
            std::vector<std::string>({f::GradVarName("out3")}));
  EXPECT_EQ(grad_op3->Output(f::GradVarName("X")),
            std::vector<std::string>({f::GradVarName("out1") + "@RENAME@0"}));
  EXPECT_EQ(grad_op3->Output(f::GradVarName("b")),
            std::vector<std::string>({f::GradVarName("b3")}));

847
  f::OpDescBind *grad_op4 = block->AllOps()[5];
F
fengjiayi 已提交
848 849 850 851 852 853 854 855 856 857 858 859 860
  ASSERT_EQ(grad_op4->Type(), "mul_grad");
  ASSERT_EQ(grad_op4->InputNames().size(), 4UL);
  ASSERT_EQ(grad_op4->OutputNames().size(), 2UL);
  EXPECT_EQ(grad_op4->Input("X"), std::vector<std::string>({"out1"}));
  EXPECT_EQ(grad_op4->Input("Y"), std::vector<std::string>({"y2"}));
  EXPECT_EQ(grad_op4->Input("Out"), std::vector<std::string>({"out2"}));
  EXPECT_EQ(grad_op4->Input(f::GradVarName("Out")),
            std::vector<std::string>({f::GradVarName("out2")}));
  EXPECT_EQ(grad_op4->Output(f::GradVarName("X")),
            std::vector<std::string>({f::GradVarName("out1") + "@RENAME@1"}));
  EXPECT_EQ(grad_op4->Output(f::GradVarName("Y")),
            std::vector<std::string>({f::GradVarName("y2")}));

861
  f::OpDescBind *sum_op = block->AllOps()[6];
F
fengjiayi 已提交
862 863 864 865 866 867 868 869 870
  ASSERT_EQ(sum_op->Type(), "sum");
  ASSERT_EQ(sum_op->InputNames().size(), 1UL);
  ASSERT_EQ(sum_op->OutputNames().size(), 1UL);
  EXPECT_EQ(sum_op->Input("X"),
            std::vector<std::string>({f::GradVarName("out1") + "@RENAME@0",
                                      f::GradVarName("out1") + "@RENAME@1"}));
  EXPECT_EQ(sum_op->Output("Out"),
            std::vector<std::string>({f::GradVarName("out1")}));

871
  f::OpDescBind *grad_op1 = block->AllOps()[7];
F
fengjiayi 已提交
872 873 874 875 876 877 878 879 880
  ASSERT_EQ(grad_op1->Type(), "rowwise_add_grad");
  ASSERT_EQ(grad_op1->InputNames().size(), 1UL);
  ASSERT_EQ(grad_op1->OutputNames().size(), 2UL);
  EXPECT_EQ(grad_op1->Input(f::GradVarName("Out")),
            std::vector<std::string>({f::GradVarName("out1")}));
  EXPECT_EQ(grad_op1->Output(f::GradVarName("X")),
            std::vector<std::string>({f::GradVarName("x1")}));
  EXPECT_EQ(grad_op1->Output(f::GradVarName("b")),
            std::vector<std::string>({f::GradVarName("b1")}));
F
fengjiayi 已提交
881

Q
Qiao Longfei 已提交
882
  EXPECT_EQ(var_to_grad.size(), 6UL);
F
fengjiayi 已提交
883 884 885 886 887 888 889 890 891 892 893 894
  EXPECT_EQ(var_to_grad.at("b3"), f::GradVarInfo(f::GradVarName("b3"), 0, 4));
  EXPECT_EQ(var_to_grad.at("y2"), f::GradVarInfo(f::GradVarName("y2"), 0, 5));
  EXPECT_EQ(var_to_grad.at("out1"),
            f::GradVarInfo(f::GradVarName("out1"), 0, 6));
  EXPECT_EQ(var_to_grad.at("x1"), f::GradVarInfo(f::GradVarName("x1"), 0, 7));
  EXPECT_EQ(var_to_grad.at("b1"), f::GradVarInfo(f::GradVarName("b1"), 0, 7));

  EXPECT_TRUE(block->HasVar(f::GradVarName("b3")));
  EXPECT_TRUE(block->HasVar(f::GradVarName("y2")));
  EXPECT_TRUE(block->HasVar(f::GradVarName("out1")));
  EXPECT_TRUE(block->HasVar(f::GradVarName("x1")));
  EXPECT_TRUE(block->HasVar(f::GradVarName("b1")));
895 896 897
}

TEST(Backward, half_backward) {
898
  f::ProgramDescBind program;
899
  f::BlockDescBind *block = program.MutableBlock(0);
900 901 902 903 904 905
  auto *op1 = block->AppendOp();
  op1->SetType("minus");
  op1->SetInput("X", {"a"});
  op1->SetInput("Y", {"b"});
  op1->SetOutput("Out", {"out"});

906
  auto target = f::VarDescBind("out");
907
  target.SetShape({1});
908
  size_t forward_len = block->AllOps().size();
F
fengjiayi 已提交
909
  auto var_to_grad = AppendBackward(program, target, {"b"});
910 911
  f::OpDescBind *fill_op = block->AllOps()[forward_len];
  EXPECT_EQ(fill_op->Type(), "fill_constant");
912
  auto ops = block->AllOps();
913
  ASSERT_EQ(3UL, ops.size());
F
fengjiayi 已提交
914

Q
Qiao Longfei 已提交
915
  EXPECT_EQ(var_to_grad.size(), 2UL);
F
fengjiayi 已提交
916 917
  EXPECT_EQ(var_to_grad.at("a"),
            f::GradVarInfo(f::GradVarName("a"), 0, forward_len + 1));
918
}