op_desc.cc 42.6 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
F
fengjiayi 已提交
2 3 4 5 6 7 8 9 10 11 12 13 14

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
Yi Wang 已提交
15
#include "paddle/fluid/framework/op_desc.h"
16

17
#include <string>
18

19
#include "glog/logging.h"
Y
Yi Wang 已提交
20
#include "paddle/fluid/framework/block_desc.h"
21
#include "paddle/fluid/framework/op_call_stack.h"
Y
yuyang18 已提交
22
#include "paddle/fluid/framework/op_proto_maker.h"
Y
Yi Wang 已提交
23 24
#include "paddle/fluid/framework/operator.h"
#include "paddle/fluid/framework/shape_inference.h"
M
minqiyang 已提交
25
#include "paddle/fluid/framework/var_type_inference.h"
26
#include "paddle/fluid/operators/ops_extra_info.h"
R
Ruibiao Chen 已提交
27
#include "paddle/utils/blank.h"
Y
Yu Yang 已提交
28

F
fengjiayi 已提交
29 30 31
namespace paddle {
namespace framework {

32 33
class CompileTimeInferShapeContext : public InferShapeContext {
 public:
Y
Yu Yang 已提交
34
  CompileTimeInferShapeContext(const OpDesc &op, const BlockDesc &block);
35 36 37 38 39

  bool HasInput(const std::string &name) const override;

  bool HasOutput(const std::string &name) const override;

40 41
  bool HasAttr(const std::string &name) const override;

42 43
  bool HasInputs(const std::string &name) const override;

44 45
  bool HasOutputs(const std::string &name,
                  bool allow_null = false) const override;
46 47 48

  AttrReader Attrs() const override;

H
hong 已提交
49
  std::vector<std::string> Inputs(const std::string &name) const override;
50

H
hong 已提交
51
  std::vector<std::string> Outputs(const std::string &name) const override;
52

53 54 55
  std::string GetInputNameByIdx(size_t idx) const override {
    auto &op_proto =
        paddle::framework::OpInfoMap::Instance().Get(op_.Type()).proto_;
56 57
    PADDLE_ENFORCE_LT(idx,
                      op_proto->inputs().size(),
58 59 60
                      platform::errors::OutOfRange(
                          "The index should be less than the size of inputs of "
                          "operator %s, but got index is %d and size is %d",
61 62 63
                          op_.Type(),
                          idx,
                          op_proto->inputs().size()));
64 65 66 67 68 69 70
    return op_proto->inputs()[idx].name();
  }

  std::string GetOutputNameByIdx(size_t idx) const override {
    auto &op_proto =
        paddle::framework::OpInfoMap::Instance().Get(op_.Type()).proto_;
    PADDLE_ENFORCE_LT(
71 72
        idx,
        op_proto->outputs().size(),
73 74 75
        platform::errors::OutOfRange(
            "The index should be less than the size of outputs of "
            "operator %s, but got index is %d and size is %d",
76 77 78
            op_.Type(),
            idx,
            op_proto->outputs().size()));
79 80 81
    return op_proto->outputs()[idx].name();
  }

82 83 84
  void ShareDim(const std::string &in,
                const std::string &out,
                size_t i = 0,
85
                size_t j = 0) override {
86 87
    PADDLE_ENFORCE_LT(i,
                      Inputs(in).size(),
88 89 90
                      platform::errors::InvalidArgument(
                          "The input variable index is out of range, expected "
                          "index less than %d, but received index is %d.",
91 92 93 94
                          Inputs(in).size(),
                          i));
    PADDLE_ENFORCE_LT(j,
                      Outputs(out).size(),
95 96 97
                      platform::errors::InvalidArgument(
                          "The output variable index is out of range, expected "
                          "index less than %d, but received index is %d.",
98 99
                          Outputs(out).size(),
                          j));
100

H
hong 已提交
101 102
    std::string input_n = Inputs(in)[i];
    std::string output_n = Outputs(out)[j];
103

104 105
    PADDLE_ENFORCE_NE(input_n,
                      framework::kEmptyVarName,
106 107
                      platform::errors::InvalidArgument(
                          "The input variable %s[%d] is empty.", in, i));
108 109
    PADDLE_ENFORCE_NE(output_n,
                      framework::kEmptyVarName,
110 111
                      platform::errors::InvalidArgument(
                          "The output variable %s[%d] is empty.", out, j));
112 113 114 115

    auto *in_var = block_.FindVarRecursive(input_n);
    auto *out_var = block_.FindVarRecursive(output_n);

116
    PADDLE_ENFORCE_EQ(
117 118
        in_var->GetType(),
        out_var->GetType(),
119 120 121
        platform::errors::InvalidArgument(
            "The type of input %s and output %s do not match. The input type "
            "is %s, output type is %s.",
122 123 124
            input_n,
            output_n,
            DataTypeToString(in_var->GetType()),
125
            DataTypeToString(out_var->GetType())));
126 127 128 129

    SetDim(output_n, GetDim(input_n));
  }

H
hong 已提交
130 131 132 133 134 135
  void ShareAllLoD(const std::string &in,
                   const std::string &out) const override {
    auto &in_var_names = op_.Input(in);
    auto &out_var_names = op_.Output(out);

    PADDLE_ENFORCE_EQ(
136 137
        in_var_names.size(),
        out_var_names.size(),
H
hong 已提交
138
        platform::errors::PreconditionNotMet(
T
tianshuo78520a 已提交
139
            "Op [%s]:  Input var number should be equal with output var number",
H
hong 已提交
140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157
            op_.Type()));

    for (size_t i = 0; i < in_var_names.size(); ++i) {
      if (out_var_names[i] == framework::kEmptyVarName) {
        continue;
      }

      auto *in_var = block_.FindVarRecursive(in_var_names[i]);
      auto *out_var = block_.FindVarRecursive(out_var_names[i]);
      if (in_var->GetType() != proto::VarType::LOD_TENSOR &&
          in_var->GetType() != proto::VarType::LOD_TENSOR_ARRAY) {
        VLOG(3) << "input " << in << " is not LoDTensor or LoDTensorArray.";
        return;
      }
      out_var->SetLoDLevel(in_var->GetLoDLevel());
    }
  }

158 159 160
  void ShareLoD(const std::string &in,
                const std::string &out,
                size_t i = 0,
Q
Qiao Longfei 已提交
161
                size_t j = 0) const override {
162 163
    PADDLE_ENFORCE_LT(i,
                      Inputs(in).size(),
164 165 166
                      platform::errors::InvalidArgument(
                          "The input variable index is out of range, expected "
                          "index less than %d, but received index is %d.",
167 168 169 170
                          Inputs(in).size(),
                          i));
    PADDLE_ENFORCE_LT(j,
                      Outputs(out).size(),
171 172 173
                      platform::errors::InvalidArgument(
                          "The output variable index is out of range, expected "
                          "index less than %d, but received index is %d.",
174 175 176 177
                          Outputs(out).size(),
                          j));
    PADDLE_ENFORCE_NE(Inputs(in)[i],
                      framework::kEmptyVarName,
178 179
                      platform::errors::InvalidArgument(
                          "The input variable %s[%d] is empty.", in, i));
180 181
    PADDLE_ENFORCE_NE(Outputs(out)[j],
                      framework::kEmptyVarName,
182 183
                      platform::errors::InvalidArgument(
                          "The output variable %s[%d] is empty.", out, j));
Q
Qiao Longfei 已提交
184 185
    auto *in_var = block_.FindVarRecursive(Inputs(in)[i]);
    auto *out_var = block_.FindVarRecursive(Outputs(out)[j]);
C
chengduo 已提交
186 187
    if (in_var->GetType() != proto::VarType::LOD_TENSOR &&
        in_var->GetType() != proto::VarType::LOD_TENSOR_ARRAY) {
188
      VLOG(3) << "input " << in << " is not LoDTensor or LoDTensorArray.";
X
fix  
Xin Pan 已提交
189 190
      return;
    }
191
    out_var->SetLoDLevel(in_var->GetLoDLevel());
Q
Qiao Longfei 已提交
192
  }
D
dzhwinter 已提交
193

194
  int32_t GetLoDLevel(const std::string &in, size_t i = 0) const override {
195 196
    PADDLE_ENFORCE_LT(i,
                      Inputs(in).size(),
197 198 199
                      platform::errors::InvalidArgument(
                          "The input variable index is out of range, input "
                          "variable %s of operator %s only has %d elements.",
200 201 202 203 204
                          in,
                          op_.Type(),
                          Inputs(in).size()));
    PADDLE_ENFORCE_NE(Inputs(in)[i],
                      framework::kEmptyVarName,
205 206
                      platform::errors::InvalidArgument(
                          "The input variable %s[%d] of operator %s is empty.",
207 208 209
                          in,
                          i,
                          op_.Type()));
C
chengduo 已提交
210
    auto *in_var = block_.FindVarRecursive(Inputs(in)[i]);
211
    PADDLE_ENFORCE_NOT_NULL(
212 213 214 215 216 217
        in_var,
        platform::errors::NotFound(
            "The input variable %s[%d] of operator %s is not found.",
            in,
            i,
            op_.Type()));
218
    return in_var->GetLoDLevel();
C
chengduo 已提交
219 220
  }

221 222
  void SetLoDLevel(const std::string &out,
                   int32_t lod_level,
223
                   size_t j = 0) const override {
224 225
    PADDLE_ENFORCE_LT(j,
                      Outputs(out).size(),
226 227 228
                      platform::errors::InvalidArgument(
                          "The output variable index is out of range, output "
                          "variable %s of operator %s only has %d elements.",
229 230 231 232 233
                          out,
                          op_.Type(),
                          Outputs(out).size()));
    PADDLE_ENFORCE_NE(Outputs(out)[j],
                      framework::kEmptyVarName,
234 235
                      platform::errors::InvalidArgument(
                          "The output variable %s[%d] of operator %s is empty.",
236 237 238
                          out,
                          j,
                          op_.Type()));
239
    auto *out_var = block_.FindVarRecursive(Outputs(out)[j]);
240
    PADDLE_ENFORCE_NOT_NULL(
241 242 243 244 245 246
        out_var,
        platform::errors::NotFound(
            "The output variable %s[%d] of operator %s is not found.",
            out,
            j,
            op_.Type()));
247 248 249
    if (lod_level >= 0) {
      out_var->SetLoDLevel(lod_level);
    }
250 251
  }

C
Chen Weihang 已提交
252
  paddle::small_vector<InferShapeVarPtr, phi::kInputSmallVectorSize>
253
  GetInputVarPtrs(const std::string &name) const override {
254
    const std::vector<std::string> arg_names = Inputs(name);
C
Chen Weihang 已提交
255
    paddle::small_vector<InferShapeVarPtr, phi::kInputSmallVectorSize> res;
256
    res.reserve(arg_names.size());
257 258 259
    std::transform(arg_names.begin(),
                   arg_names.end(),
                   std::back_inserter(res),
260 261 262 263 264 265
                   [this](const std::string &name) {
                     return block_.FindVarRecursive(name);
                   });
    return res;
  }

C
Chen Weihang 已提交
266
  paddle::small_vector<InferShapeVarPtr, phi::kOutputSmallVectorSize>
267
  GetOutputVarPtrs(const std::string &name) const override {
268
    const std::vector<std::string> arg_names = Outputs(name);
C
Chen Weihang 已提交
269
    paddle::small_vector<InferShapeVarPtr, phi::kOutputSmallVectorSize> res;
270
    res.reserve(arg_names.size());
271 272 273
    std::transform(arg_names.begin(),
                   arg_names.end(),
                   std::back_inserter(res),
274 275 276 277 278 279
                   [this](const std::string &name) {
                     return block_.FindVarRecursive(name);
                   });
    return res;
  }

X
Xin Pan 已提交
280 281
  DDim GetInputDim(const std::string &name) const override {
    const std::vector<std::string> &arg_names = Inputs(name);
282 283
    PADDLE_ENFORCE_EQ(arg_names.size(),
                      1UL,
284 285 286
                      platform::errors::InvalidArgument(
                          "The input(%s) should hold only one element, but now "
                          "it holds %d elements.",
287 288
                          name,
                          arg_names.size()));
X
Xin Pan 已提交
289 290 291 292 293 294 295 296
    return this->GetDim(arg_names[0]);
  }

  std::vector<DDim> GetInputsDim(const std::string &name) const override {
    const std::vector<std::string> &arg_names = Inputs(name);
    return GetDims(arg_names);
  }

297 298
  bool IsRuntime() const override;

299 300
  bool IsRunMKLDNNKernel() const override;

301 302 303 304
  proto::VarType::Type GetInputVarType(const std::string &name) const override {
    return GetVarType(Inputs(name).at(0));
  }

X
Xin Pan 已提交
305 306 307 308 309 310 311 312 313 314
  std::vector<proto::VarType::Type> GetInputsVarType(
      const std::string &name) const override {
    return GetVarTypes(Inputs(name));
  }

  std::vector<proto::VarType::Type> GetOutputsVarType(
      const std::string &name) const override {
    return GetVarTypes(Outputs(name));
  }

X
Xin Pan 已提交
315
  void SetOutputDim(const std::string &name, const DDim &dim) override {
H
hong 已提交
316
    auto arg_names = Outputs(name);
317 318
    PADDLE_ENFORCE_EQ(arg_names.size(),
                      1UL,
319 320 321
                      platform::errors::InvalidArgument(
                          "The iutput(%s) should hold only one element, but "
                          "now it holds %d elements.",
322 323
                          name,
                          arg_names.size()));
X
Xin Pan 已提交
324 325 326 327 328
    SetDim(arg_names[0], dim);
  }

  void SetOutputsDim(const std::string &name,
                     const std::vector<DDim> &dims) override {
H
hong 已提交
329
    auto names = Outputs(name);
X
Xin Pan 已提交
330 331 332
    SetDims(names, dims);
  }

333 334 335 336 337 338 339 340
  const phi::ArgumentMappingFn *GetPhiArgumentMappingFn() const override {
    return phi::OpUtilsMap::Instance().GetArgumentMappingFn(op_.Type());
  }

  const phi::KernelSignature *GetPhiDefaultKernelSignature() const override {
    return &phi::DefaultKernelSignatureMap::Instance().Get(op_.Type());
  }

341
 protected:
X
Xin Pan 已提交
342 343 344 345 346
  std::vector<proto::VarType::Type> GetVarTypes(
      const std::vector<std::string> &names) const {
    std::vector<proto::VarType::Type> retv;
    retv.resize(names.size());
    std::transform(
347 348 349 350 351
        names.begin(),
        names.end(),
        retv.begin(),
        std::bind(std::mem_fn(&CompileTimeInferShapeContext::GetVarType),
                  this,
X
Xin Pan 已提交
352 353 354 355 356
                  std::placeholders::_1));
    return retv;
  }

  proto::VarType::Type GetVarType(const std::string &name) const;
Q
Qiao Longfei 已提交
357

X
Xin Pan 已提交
358 359
  DDim GetDim(const std::string &name) const {
    auto var = block_.FindVarRecursive(name);
360 361
    PADDLE_ENFORCE_NOT_NULL(
        var, platform::errors::NotFound("Variable %s is not found.", name));
X
Xin Pan 已提交
362 363 364
    DDim res;
    try {
      auto shape = var->GetShape();
365
      res = shape.empty() ? phi::make_ddim({0UL}) : phi::make_ddim(shape);
X
Xin Pan 已提交
366 367 368 369 370 371 372 373 374 375 376
    } catch (...) {
      VLOG(5) << "GetDim of variable " << name << " error";
      std::rethrow_exception(std::current_exception());
    }
    return res;
  }

  std::vector<DDim> GetDims(const std::vector<std::string> &names) const {
    std::vector<DDim> ret;
    ret.reserve(names.size());
    std::transform(
377 378 379
        names.begin(),
        names.end(),
        std::back_inserter(ret),
X
Xin Pan 已提交
380 381 382
        [this](const std::string &name) { return this->GetDim(name); });
    return ret;
  }
383

X
Xin Pan 已提交
384 385 386 387 388
  void SetDim(const std::string &name, const DDim &dim);

  void SetDims(const std::vector<std::string> &names,
               const std::vector<DDim> &dims) {
    size_t length = names.size();
389 390
    PADDLE_ENFORCE_EQ(length,
                      dims.size(),
391 392 393
                      platform::errors::InvalidArgument(
                          "The input variables number(%d) and input dimensions "
                          "number(%d) do not match.",
394 395
                          length,
                          dims.size()));
X
Xin Pan 已提交
396 397 398 399 400 401 402
    for (size_t i = 0; i < length; ++i) {
      if (names[i] == framework::kEmptyVarName) {
        continue;
      }
      SetDim(names[i], dims[i]);
    }
  }
403

F
fengjiayi 已提交
404 405 406 407
  std::vector<DDim> GetRepeatedDims(const std::string &name) const override;

  void SetRepeatedDims(const std::string &name,
                       const std::vector<DDim> &dims) override;
F
fengjiayi 已提交
408

Y
Yu Yang 已提交
409 410
  const OpDesc &op_;
  const BlockDesc &block_;
411 412
};

413 414 415 416 417 418 419
static void InitRuntimeAttributeMapByOpExtraInfo(const std::string &op_type,
                                                 AttributeMap *runtime_attrs) {
  const auto &extra_attr_map =
      operators::ExtraInfoUtils::Instance().GetExtraAttrsMap(op_type);
  runtime_attrs->insert(extra_attr_map.begin(), extra_attr_map.end());
}

420 421 422 423
OpDesc::OpDesc(const std::string &type,
               const VariableNameMap &inputs,
               const VariableNameMap &outputs,
               const AttributeMap &attrs) {
424
  desc_.set_type(type);
F
fengjiayi 已提交
425 426 427
  inputs_ = inputs;
  outputs_ = outputs;
  attrs_ = attrs;
F
Fix bug  
fengjiayi 已提交
428
  need_update_ = true;
L
liuwei1031 已提交
429
  block_ = nullptr;
430
  InitRuntimeAttributeMapByOpExtraInfo(type, &runtime_attrs_);
F
fengjiayi 已提交
431 432
}

433 434 435 436 437 438
OpDesc::OpDesc(const OpDesc &other) {
  CopyFrom(other);
  block_ = other.block_;
  need_update_ = true;
}

X
Xin Pan 已提交
439 440 441 442
OpDesc::OpDesc(const OpDesc &other, BlockDesc *block) {
  CopyFrom(other);
  block_ = block;
  need_update_ = true;
443 444 445
  for (auto &iter : attrs_) {
    UpdateVarAttr(iter.first, iter.second);
  }
X
Xin Pan 已提交
446 447
}

448
void OpDesc::CopyFrom(const OpDesc &op_desc) {
F
fengjiayi 已提交
449 450 451 452
  desc_.set_type(op_desc.Type());
  inputs_ = op_desc.inputs_;
  outputs_ = op_desc.outputs_;
  attrs_ = op_desc.attrs_;
453 454
  runtime_attrs_ = op_desc.runtime_attrs_;
  // The record of original_id_ is only for auto parallel.
455
  original_id_ = op_desc.original_id_;
456 457 458
  if (op_desc.dist_attr_) {
    dist_attr_.reset(new OperatorDistAttr(*op_desc.dist_attr_));
  }
F
fengjiayi 已提交
459 460 461
  need_update_ = true;
}

F
fengjiayi 已提交
462
OpDesc::OpDesc(const proto::OpDesc &desc, BlockDesc *block)
463 464 465 466
    : desc_(desc), need_update_(false) {
  // restore inputs_
  int input_size = desc_.inputs_size();
  for (int i = 0; i < input_size; ++i) {
467
    const proto::OpDesc::Var &var = desc_.inputs(i);
468 469 470 471 472 473 474 475 476 477
    std::vector<std::string> &args = inputs_[var.parameter()];
    int argu_size = var.arguments_size();
    args.reserve(argu_size);
    for (int j = 0; j < argu_size; ++j) {
      args.push_back(var.arguments(j));
    }
  }
  // restore outputs_
  int output_size = desc_.outputs_size();
  for (int i = 0; i < output_size; ++i) {
478
    const proto::OpDesc::Var &var = desc_.outputs(i);
479 480 481 482 483 484 485 486
    std::vector<std::string> &args = outputs_[var.parameter()];
    int argu_size = var.arguments_size();
    args.reserve(argu_size);
    for (int j = 0; j < argu_size; ++j) {
      args.push_back(var.arguments(j));
    }
  }
  // restore attrs_
487
  InitRuntimeAttributeMapByOpExtraInfo(desc.type(), &runtime_attrs_);
488
  for (const proto::OpDesc::Attr &attr : desc_.attrs()) {
489
    const std::string &attr_name = attr.name();
490
    // The sub_block referred to by the BLOCK attr hasn't been added
491 492 493 494 495 496 497
    // to ProgramDesc class yet, we skip setting BLOCK/BLOCKS/VAR/VARS attr
    // here.
    auto attr_type = attr.type();
    if (attr_type != proto::AttrType::BLOCK &&
        attr_type != proto::AttrType::BLOCKS &&
        attr_type != proto::AttrType::VAR &&
        attr_type != proto::AttrType::VARS) {
498 499 500 501 502 503
      auto iter = runtime_attrs_.find(attr_name);
      if (iter == runtime_attrs_.end()) {
        attrs_[attr_name] = GetAttrValue(attr);
      } else {
        iter->second = GetAttrValue(attr);
      }
504
    }
505
  }
506
  this->block_ = block;
507 508
}

509 510 511 512 513 514 515 516 517
// Explicitly implement the assign operator, Since the added
// unique_ptr data member does not have the implicit assign operator.
OpDesc &OpDesc::operator=(const OpDesc &other) {
  CopyFrom(other);
  block_ = other.block_;
  need_update_ = true;
  return *this;
}

Y
Yu Yang 已提交
518
proto::OpDesc *OpDesc::Proto() {
519
  Flush();
520
  return &desc_;
F
fengjiayi 已提交
521 522
}

Y
Yu Yang 已提交
523
const std::vector<std::string> &OpDesc::Input(const std::string &name) const {
F
fengjiayi 已提交
524
  auto it = inputs_.find(name);
525
  PADDLE_ENFORCE_NE(
526 527 528 529
      it,
      inputs_.end(),
      platform::errors::NotFound(
          "Input %s cannot be found in operator %s.", name, Type()));
F
fengjiayi 已提交
530 531 532
  return it->second;
}

533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555
std::vector<std::string> OpDesc::Input(const std::string &name,
                                       bool with_attr_var) const {
  // Attribute with VarDesc type will consider as Input
  if (with_attr_var) {
    auto it = attrs_.find(name);
    if (it != attrs_.end() && HasAttrVar(it->second))
      return AttrVarNames(it->second);
  }
  return this->Input(name);
}

VariableNameMap OpDesc::Inputs(bool with_attr_var) const {
  if (!with_attr_var) {
    return inputs_;
  }
  VariableNameMap res = inputs_;
  for (auto &attr : FilterAttrVar(attrs_)) {
    res[attr.first] = AttrVarNames(attr.second);
  }
  return res;
}

std::vector<std::string> OpDesc::InputArgumentNames(bool with_attr_var) const {
F
Update  
fengjiayi 已提交
556
  std::vector<std::string> retv;
557
  for (auto &ipt : this->Inputs(with_attr_var)) {
F
Update  
fengjiayi 已提交
558 559 560 561 562
    retv.insert(retv.end(), ipt.second.begin(), ipt.second.end());
  }
  return retv;
}

Y
Yu Yang 已提交
563 564
void OpDesc::SetInput(const std::string &param_name,
                      const std::vector<std::string> &args) {
F
fengjiayi 已提交
565 566 567 568
  need_update_ = true;
  inputs_[param_name] = args;
}

Y
Yu Yang 已提交
569
const std::vector<std::string> &OpDesc::Output(const std::string &name) const {
F
fengjiayi 已提交
570
  auto it = outputs_.find(name);
571
  PADDLE_ENFORCE_NE(
572 573 574 575
      it,
      outputs_.end(),
      platform::errors::NotFound(
          "Output %s cannot be found in operator %s.", name, Type()));
F
fengjiayi 已提交
576 577 578
  return it->second;
}

579 580 581 582
bool OpDesc::HasOutput(const std::string &name) const {
  return outputs_.find(name) != outputs_.end();
}

Y
Yu Yang 已提交
583
std::vector<std::string> OpDesc::OutputArgumentNames() const {
F
Update  
fengjiayi 已提交
584 585 586 587 588 589 590
  std::vector<std::string> retv;
  for (auto &ipt : this->outputs_) {
    retv.insert(retv.end(), ipt.second.begin(), ipt.second.end());
  }
  return retv;
}

Y
Yu Yang 已提交
591 592
void OpDesc::SetOutput(const std::string &param_name,
                       const std::vector<std::string> &args) {
F
fengjiayi 已提交
593 594 595 596
  need_update_ = true;
  this->outputs_[param_name] = args;
}

597 598 599 600 601
void OpDesc::RemoveOutput(const std::string &name) {
  outputs_.erase(name);
  need_update_ = true;
}

602 603 604 605 606
void OpDesc::RemoveInput(const std::string &name) {
  inputs_.erase(name);
  need_update_ = true;
}

607 608 609 610 611 612 613 614 615 616
bool OpDesc::HasProtoAttr(const std::string &name) const {
  auto &op_info = OpInfoMap::Instance();
  if (op_info.Has(desc_.type())) {
    auto op_info_ptr = op_info.Get(desc_.type());
    if (op_info_ptr.HasOpProtoAndChecker()) {
      const proto::OpProto &proto = op_info_ptr.Proto();
      for (int i = 0; i != proto.attrs_size(); ++i) {
        const proto::OpProto::Attr &attr = proto.attrs(i);
        if (attr.name() == name) {
          return true;
L
luotao1 已提交
617 618
        }
      }
L
luotao1 已提交
619 620 621 622 623
    }
  }
  return false;
}

624 625 626 627
proto::AttrType OpDesc::GetAttrType(const std::string &name,
                                    bool with_attr_var) const {
  auto attr = this->GetAttr(name, with_attr_var);
  return static_cast<proto::AttrType>(attr.index() - 1);
F
fengjiayi 已提交
628 629
}

630
std::vector<std::string> OpDesc::AttrNames(bool with_attr_var) const {
F
fengjiayi 已提交
631 632 633
  std::vector<std::string> retv;
  retv.reserve(attrs_.size());
  for (auto &attr : attrs_) {
634
    if (!with_attr_var && HasAttrVar(attr.second)) continue;
F
fengjiayi 已提交
635 636 637 638 639
    retv.push_back(attr.first);
  }
  return retv;
}

640 641
bool OpDesc::HasAttr(const std::string &name, bool with_attr_var) const {
  auto iter = attrs_.find(name);
642 643 644 645 646 647 648
  bool is_found = true;
  if (iter == attrs_.end()) {
    iter = runtime_attrs_.find(name);
    if (iter == runtime_attrs_.end()) {
      is_found = false;
    }
  }
649 650 651 652 653 654
  if (with_attr_var) {
    return is_found;
  }
  return is_found && !HasAttrVar(iter->second);
}

655 656
void OpDesc::RemoveAttr(const std::string &name) {
  attrs_.erase(name);
657
  runtime_attrs_.erase(name);
658 659 660
  need_update_ = true;
}

Y
Yu Yang 已提交
661
void OpDesc::SetAttr(const std::string &name, const Attribute &v) {
662 663 664 665 666 667 668 669
  AttributeMap *attrs_ptr = &(this->attrs_);

  const auto &extra_attr_map =
      operators::ExtraInfoUtils::Instance().GetExtraAttrsMap(Type());
  auto extra_attr_iter = extra_attr_map.find(name);
  if (extra_attr_iter != extra_attr_map.end()) {
    attrs_ptr = &(this->runtime_attrs_);
  }
M
minqiyang 已提交
670 671 672
  // NOTICE(minqiyang): pybind11 will take the empty list in python as
  // the std::vector<int> type in C++; so we have to change the attr's type
  // here if we meet this issue
R
Ruibiao Chen 已提交
673
  proto::AttrType attr_type = static_cast<proto::AttrType>(v.index() - 1);
M
minqiyang 已提交
674
  if (attr_type == proto::AttrType::INTS &&
R
Ruibiao Chen 已提交
675
      PADDLE_GET_CONST(std::vector<int>, v).size() == 0u) {
M
minqiyang 已提交
676
    // Find current attr via attr name and set the correct attribute value
M
minqiyang 已提交
677
    const proto::OpProto::Attr &attr = GetProtoAttr(name);
M
minqiyang 已提交
678 679
    switch (attr.type()) {
      case proto::AttrType::BOOLEANS: {
M
minqiyang 已提交
680 681
        VLOG(11) << "SetAttr: " << Type() << ", " << name
                 << " from INTS to BOOLEANS";
682
        attrs_ptr->operator[](name) = std::vector<bool>();
M
minqiyang 已提交
683 684 685
        break;
      }
      case proto::AttrType::INTS: {
M
minqiyang 已提交
686 687
        VLOG(11) << "SetAttr: " << Type() << ", " << name
                 << " from INTS to INTS";
688
        attrs_ptr->operator[](name) = std::vector<int>();
M
minqiyang 已提交
689 690
        break;
      }
691
      case proto::AttrType::LONGS: {
M
minqiyang 已提交
692 693
        VLOG(11) << "SetAttr: " << Type() << ", " << name
                 << " from LONGS to LONGS";
694
        attrs_ptr->operator[](name) = std::vector<int64_t>();
695 696
        break;
      }
M
minqiyang 已提交
697
      case proto::AttrType::FLOATS: {
M
minqiyang 已提交
698 699
        VLOG(11) << "SetAttr: " << Type() << ", " << name
                 << " from INTS to FLOATS";
700
        attrs_ptr->operator[](name) = std::vector<float>();
M
minqiyang 已提交
701 702
        break;
      }
703 704 705 706 707 708
      case proto::AttrType::FLOAT64S: {
        VLOG(11) << "SetAttr: " << Type() << ", " << name
                 << " from INTS to FLOAT64S";
        this->attrs_[name] = std::vector<double>();
        break;
      }
M
minqiyang 已提交
709
      case proto::AttrType::STRINGS: {
M
minqiyang 已提交
710 711
        VLOG(11) << "SetAttr: " << Type() << ", " << name
                 << " from INTS to STRINGS";
712
        attrs_ptr->operator[](name) = std::vector<std::string>();
M
minqiyang 已提交
713 714 715
        break;
      }
      case proto::AttrType::BLOCKS: {
M
minqiyang 已提交
716 717
        VLOG(11) << "SetAttr: " << Type() << ", " << name
                 << " from INTS to BLOCKS";
718
        attrs_ptr->operator[](name) = std::vector<BlockDesc *>();
M
minqiyang 已提交
719 720
        return;
      }
M
minqiyang 已提交
721
      default:
722 723
        PADDLE_THROW(platform::errors::Unimplemented(
            "Unsupported attribute type (code %d).", attr.type()));
M
minqiyang 已提交
724
    }
M
minqiyang 已提交
725 726
    need_update_ = true;
    return;
M
minqiyang 已提交
727 728
  }

729
  // In order to set bool attr properly
730 731 732 733 734 735 736 737 738 739 740 741 742 743
  if (attr_type == proto::AttrType::INT) {
    if (HasProtoAttr(name) &&
        GetProtoAttr(name).type() == proto::AttrType::BOOLEAN) {
      attrs_ptr->operator[](name) = static_cast<bool>(PADDLE_GET_CONST(int, v));
      need_update_ = true;
      return;
    }
    if (extra_attr_iter != extra_attr_map.end() &&
        static_cast<proto::AttrType>(extra_attr_iter->second.index() - 1) ==
            proto::AttrType::BOOLEAN) {
      attrs_ptr->operator[](name) = static_cast<bool>(PADDLE_GET_CONST(int, v));
      need_update_ = true;
      return;
    }
744 745
  }

746
  attrs_ptr->operator[](name) = v;
F
fengjiayi 已提交
747 748 749
  need_update_ = true;
}

750 751 752 753 754 755 756 757 758 759
void OpDesc::SetVarAttr(const std::string &name, VarDesc *var) {
  this->attrs_[name] = var;
  need_update_ = true;
}

void OpDesc::SetVarsAttr(const std::string &name, std::vector<VarDesc *> vars) {
  this->attrs_[name] = vars;
  need_update_ = true;
}

A
Abhinav Arora 已提交
760 761
void OpDesc::SetBlockAttr(const std::string &name, BlockDesc *block) {
  this->attrs_[name] = block;
F
fengjiayi 已提交
762
  need_update_ = true;
F
fengjiayi 已提交
763 764
}

765 766 767 768 769 770
void OpDesc::SetBlocksAttr(const std::string &name,
                           std::vector<BlockDesc *> blocks) {
  this->attrs_[name] = blocks;
  need_update_ = true;
}

Y
Yu Yang 已提交
771
void OpDesc::SetAttrMap(
F
fengjiayi 已提交
772 773 774 775 776
    const std::unordered_map<std::string, Attribute> &attr_map) {
  attrs_ = attr_map;
  need_update_ = true;
}

777 778 779 780 781 782
void OpDesc::SetRuntimeAttrMap(
    const std::unordered_map<std::string, Attribute> &attr_map) {
  runtime_attrs_ = attr_map;
  need_update_ = true;
}

783
Attribute OpDesc::GetAttr(const std::string &name, bool with_attr_var) const {
F
fengjiayi 已提交
784
  auto it = attrs_.find(name);
785 786 787
  if (it == attrs_.end()) {
    it = runtime_attrs_.find(name);
  }
788
  PADDLE_ENFORCE_NE(
789 790
      it,
      attrs_.end(),
791
      platform::errors::NotFound("Attribute %s is not found.", name));
792 793 794 795 796 797
  if (!with_attr_var) {
    PADDLE_ENFORCE_EQ(
        HasAttrVar(it->second),
        false,
        platform::errors::NotFound("Attribute %s is not found.", name));
  }
F
fengjiayi 已提交
798 799 800
  return it->second;
}

M
minqiyang 已提交
801 802 803
const proto::OpProto::Attr &OpDesc::GetProtoAttr(
    const std::string &name) const {
  const proto::OpProto &proto = OpInfoMap::Instance().Get(Type()).Proto();
M
minqiyang 已提交
804 805 806 807 808 809 810
  for (int i = 0; i != proto.attrs_size(); ++i) {
    const proto::OpProto::Attr &attr = proto.attrs(i);
    if (attr.name() == name) {
      return attr;
    }
  }

811 812
  PADDLE_THROW(platform::errors::NotFound(
      "Attribute %s is not found in proto %s.", name, proto.type()));
M
minqiyang 已提交
813 814
}

Y
yuyang18 已提交
815
Attribute OpDesc::GetNullableAttr(const std::string &name) const {
Y
Fix bug  
yuyang18 已提交
816 817 818 819
  auto it = attrs_.find(name);
  if (it != attrs_.end()) {
    return it->second;
  } else {
Y
yuyang18 已提交
820
    return Attribute();
Y
Fix bug  
yuyang18 已提交
821 822 823
  }
}

G
gongweibao 已提交
824 825
std::vector<int> OpDesc::GetBlocksAttrIds(const std::string &name) const {
  auto it = attrs_.find(name);
826
  PADDLE_ENFORCE_NE(
827 828
      it,
      attrs_.end(),
829 830
      platform::errors::NotFound(
          "Attribute `%s` is not found in operator `%s`.", name, desc_.type()));
R
Ruibiao Chen 已提交
831
  auto blocks = PADDLE_GET_CONST(std::vector<BlockDesc *>, it->second);
G
gongweibao 已提交
832 833 834 835 836 837 838 839 840 841

  std::vector<int> ids;
  for (auto n : blocks) {
    ids.push_back(n->ID());
  }

  return ids;
}

int OpDesc::GetBlockAttrId(const std::string &name) const {
F
fengjiayi 已提交
842
  auto it = attrs_.find(name);
843
  PADDLE_ENFORCE_NE(
844 845
      it,
      attrs_.end(),
846 847
      platform::errors::NotFound(
          "Attribute `%s` is not found in operator `%s`.", name, desc_.type()));
R
Ruibiao Chen 已提交
848
  return PADDLE_GET_CONST(BlockDesc *, it->second)->ID();
F
fengjiayi 已提交
849 850
}

Y
Yu Yang 已提交
851
const std::unordered_map<std::string, Attribute> &OpDesc::GetAttrMap() const {
F
fengjiayi 已提交
852 853 854
  return attrs_;
}

855 856
const AttributeMap &OpDesc::GetRuntimeAttrMap() const { return runtime_attrs_; }

Y
Yu Yang 已提交
857
void OpDesc::Rename(const std::string &old_name, const std::string &new_name) {
Y
Yancey1989 已提交
858 859
  RenameInput(old_name, new_name);
  RenameOutput(old_name, new_name);
F
fengjiayi 已提交
860 861 862
  need_update_ = true;
}

Y
Yu Yang 已提交
863 864
void OpDesc::RenameOutput(const std::string &old_name,
                          const std::string &new_name) {
Y
Yang Yang(Tony) 已提交
865
  for (auto &output : outputs_) {
866 867
    std::replace(
        output.second.begin(), output.second.end(), old_name, new_name);
Y
Yang Yang(Tony) 已提交
868
  }
Y
yuyang18 已提交
869 870 871

  auto it = attrs_.find(framework::OpProtoAndCheckerMaker::OpRoleVarAttrName());
  if (it != attrs_.end()) {
R
Ruibiao Chen 已提交
872
    auto &op_vars = PADDLE_GET(std::vector<std::string>, it->second);
Y
yuyang18 已提交
873 874 875
    std::replace(op_vars.begin(), op_vars.end(), old_name, new_name);
  }

Y
Yang Yang(Tony) 已提交
876 877 878
  need_update_ = true;
}

Y
Yu Yang 已提交
879 880
void OpDesc::RenameInput(const std::string &old_name,
                         const std::string &new_name) {
Y
Yang Yang(Tony) 已提交
881 882 883
  for (auto &input : inputs_) {
    std::replace(input.second.begin(), input.second.end(), old_name, new_name);
  }
Y
Yancey1989 已提交
884 885 886

  auto it = attrs_.find(framework::OpProtoAndCheckerMaker::OpRoleVarAttrName());
  if (it != attrs_.end()) {
R
Ruibiao Chen 已提交
887
    auto &op_vars = PADDLE_GET(std::vector<std::string>, it->second);
Y
Yancey1989 已提交
888 889 890
    std::replace(op_vars.begin(), op_vars.end(), old_name, new_name);
  }

Y
Yang Yang(Tony) 已提交
891 892 893
  need_update_ = true;
}

894
struct SetAttrDescVisitor {
895 896
  explicit SetAttrDescVisitor(proto::OpDesc::Attr *attr) : attr_(attr) {}
  mutable proto::OpDesc::Attr *attr_;
Y
Yu Yang 已提交
897 898
  void operator()(int v) const { attr_->set_i(v); }
  void operator()(float v) const { attr_->set_f(v); }
899
  void operator()(double v) const { attr_->set_float64(v); }
Y
Yu Yang 已提交
900
  void operator()(const std::string &v) const { attr_->set_s(v); }
Q
QI JUN 已提交
901 902 903 904 905 906 907

  // Please refer to https://github.com/PaddlePaddle/Paddle/issues/7162
  template <class T,
            class = typename std::enable_if<std::is_same<bool, T>::value>::type>
  void operator()(T b) const {
    attr_->set_b(b);
  }
Y
Yu Yang 已提交
908 909 910 911 912 913 914 915 916 917 918 919 920

  void operator()(const std::vector<int> &v) const {
    VectorToRepeated(v, attr_->mutable_ints());
  }
  void operator()(const std::vector<float> &v) const {
    VectorToRepeated(v, attr_->mutable_floats());
  }
  void operator()(const std::vector<std::string> &v) const {
    VectorToRepeated(v, attr_->mutable_strings());
  }
  void operator()(const std::vector<bool> &v) const {
    VectorToRepeated(v, attr_->mutable_bools());
  }
921 922 923 924 925 926 927 928 929 930 931 932 933

  void operator()(const std::vector<VarDesc *> &v) const {
    std::vector<std::string> var_names;
    for (auto var : v) {
      var_names.emplace_back(var->Name());
    }
    VectorToRepeated(var_names, attr_->mutable_vars_name());
  }

  void operator()(const VarDesc *desc) const {
    attr_->set_var_name(desc->Name());
  }

934 935 936
  void operator()(const std::vector<BlockDesc *> &v) const {
    std::vector<int> blocks_idx;
    for (auto blk : v) {
T
tangwei12 已提交
937
      blocks_idx.push_back(blk->ID());
938 939 940
    }
    VectorToRepeated(blocks_idx, attr_->mutable_blocks_idx());
  }
T
tangwei12 已提交
941 942 943

  void operator()(BlockDesc *desc) const { attr_->set_block_idx(desc->ID()); }

944
  void operator()(int64_t v) const { attr_->set_l(v); }
T
tangwei12 已提交
945 946 947 948 949

  void operator()(const std::vector<int64_t> &v) const {
    VectorToRepeated(v, attr_->mutable_longs());
  }

950 951 952 953
  void operator()(const std::vector<double> &v) const {
    VectorToRepeated(v, attr_->mutable_float64s());
  }

R
Ruibiao Chen 已提交
954
  void operator()(paddle::blank) const {
955 956 957 958
    PADDLE_THROW(platform::errors::Unavailable(
        "Unsupported calling method of SetAttrDescVisitor object for "
        "`boosst::blank` type."));
  }
Y
Yu Yang 已提交
959 960
};

Y
Yu Yang 已提交
961
void OpDesc::Flush() {
L
Leo Chen 已提交
962 963
  VLOG(4) << "Flush "
          << " " << Type() << " " << need_update_;
F
fengjiayi 已提交
964
  if (need_update_) {
965
    this->desc_.mutable_inputs()->Clear();
F
fengjiayi 已提交
966
    for (auto &ipt : inputs_) {
967
      auto *input = desc_.add_inputs();
F
fengjiayi 已提交
968 969 970 971
      input->set_parameter(ipt.first);
      VectorToRepeated(ipt.second, input->mutable_arguments());
    }

972
    this->desc_.mutable_outputs()->Clear();
F
fengjiayi 已提交
973
    for (auto &opt : outputs_) {
974
      auto *output = desc_.add_outputs();
F
fengjiayi 已提交
975 976 977 978
      output->set_parameter(opt.first);
      VectorToRepeated(opt.second, output->mutable_arguments());
    }

979
    this->desc_.mutable_attrs()->Clear();
980 981 982 983 984 985 986 987 988
    auto set_attr_desc = [this](const std::string &attr_name,
                                const Attribute &attr) -> void {
      auto *attr_desc = desc_.add_attrs();
      attr_desc->set_name(attr_name);
      attr_desc->set_type(static_cast<proto::AttrType>(attr.index() - 1));
      SetAttrDescVisitor visitor(attr_desc);
      paddle::visit(visitor, attr);
    };

L
Leo Chen 已提交
989 990 991 992 993 994 995
    std::vector<std::pair<std::string, Attribute>> sorted_attrs{attrs_.begin(),
                                                                attrs_.end()};
    std::sort(
        sorted_attrs.begin(),
        sorted_attrs.end(),
        [](std::pair<std::string, Attribute> a,
           std::pair<std::string, Attribute> b) { return a.first < b.first; });
996

L
Leo Chen 已提交
997
    for (auto &attr : sorted_attrs) {
998 999 1000 1001
      set_attr_desc(attr.first, attr.second);
    }
    for (auto &attr : runtime_attrs_) {
      set_attr_desc(attr.first, attr.second);
F
fengjiayi 已提交
1002 1003 1004 1005 1006
    }

    need_update_ = false;
  }
}
Y
Yu Yang 已提交
1007

Y
Yu Yang 已提交
1008
void OpDesc::CheckAttrs() {
1009 1010
  PADDLE_ENFORCE_EQ(Type().empty(),
                    false,
1011 1012
                    platform::errors::PreconditionNotMet(
                        "CheckAttrs() can not be called before type is set."));
Y
Yu Yang 已提交
1013 1014 1015 1016 1017 1018
  auto *checker = OpInfoMap::Instance().Get(Type()).Checker();
  if (checker == nullptr) {
    // checker is not configured. That operator could be generated by Paddle,
    // not by users.
    return;
  }
1019
  VLOG(10) << "begin to check attribute of " << Type();
T
tangwei12 已提交
1020
  checker->Check(&attrs_);
1021 1022 1023 1024 1025 1026 1027
  const auto &extra_attr_checkers =
      operators::ExtraInfoUtils::Instance().GetExtraAttrsChecker(Type());
  if (!extra_attr_checkers.empty()) {
    for (const auto &extra_checker : extra_attr_checkers) {
      extra_checker(&runtime_attrs_, false);
    }
  }
F
fengjiayi 已提交
1028 1029
}

H
hong 已提交
1030
void OpDesc::InferShape(const BlockDesc &block) {
1031 1032
  try {
    VLOG(3) << "CompileTime infer shape on " << Type();
H
hong 已提交
1033
    auto &op_info = OpInfoMap::Instance().Get(this->Type());
1034
    this->CheckAttrs();
H
hong 已提交
1035
    auto &infer_shape = op_info.infer_shape_;
1036
    PADDLE_ENFORCE_EQ(
1037 1038
        static_cast<bool>(infer_shape),
        true,
1039 1040
        platform::errors::NotFound(
            "Operator %s's infer_shape is not registered.", this->Type()));
1041 1042 1043 1044 1045
    CompileTimeInferShapeContext ctx(*this, block);
    if (VLOG_IS_ON(10)) {
      std::ostringstream sout;
      auto inames = this->InputArgumentNames();
      sout << " From [";
1046 1047
      std::copy(inames.begin(),
                inames.end(),
1048 1049 1050
                std::ostream_iterator<std::string>(sout, ", "));
      sout << "] to [";
      auto onames = this->OutputArgumentNames();
1051 1052
      std::copy(onames.begin(),
                onames.end(),
1053 1054 1055 1056 1057
                std::ostream_iterator<std::string>(sout, ", "));
      sout << "]";
      VLOG(10) << sout.str();
    }
    infer_shape(&ctx);
1058
  } catch (platform::EnforceNotMet &exception) {
1059
    framework::AppendErrorOpHint(Type(), &exception);
1060 1061 1062 1063
    throw std::move(exception);
  } catch (...) {
    std::rethrow_exception(std::current_exception());
  }
Y
Yu Yang 已提交
1064 1065
}

Y
Yu Yang 已提交
1066
void OpDesc::InferVarType(BlockDesc *block) const {
X
Xin Pan 已提交
1067 1068
  // There are a few places that var type can be set.
  // When VarDesc is created, default set to LOD_TENSOR.
T
tianshuo78520a 已提交
1069
  // When output variable is created, default is default set to LOD_TENSOR.
X
Xin Pan 已提交
1070 1071
  // We limit here to be the only place that operator defines its customized
  // var type inference. Hence, we don't do any "default" setting here.
Y
Yu Yang 已提交
1072 1073
  auto &info = OpInfoMap::Instance().Get(this->Type());
  if (info.infer_var_type_) {
M
minqiyang 已提交
1074
    InferVarTypeContext context(this, block);
M
minqiyang 已提交
1075
    info.infer_var_type_(&context);
Y
Yu Yang 已提交
1076 1077 1078
  }
}

1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092
OperatorDistAttr *OpDesc::MutableDistAttr() {
  if (dist_attr_) {
    return dist_attr_.get();
  } else {
    dist_attr_.reset(new OperatorDistAttr(*this));
    return dist_attr_.get();
  }
}

void OpDesc::SetDistAttr(const OperatorDistAttr &dist_attr) {
  MutableDistAttr();
  *dist_attr_ = dist_attr;
}

1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137
void OpDesc::UpdateVarAttr(const std::string &name, const Attribute &attr) {
  auto attr_type = static_cast<proto::AttrType>(attr.index() - 1);
  auto type = GetAttrType(name, true);
  if (type == proto::AttrType::VAR) {
    PADDLE_ENFORCE_EQ(
        attr_type,
        type,
        platform::errors::InvalidArgument(
            "Required attr.type == proto::AttrType::VAR, but received %s",
            attr_type));
    auto *var_desc = PADDLE_GET_CONST(VarDesc *, attr);
    VLOG(3) << "Update AttrVar " << name << " with " << var_desc->Name();
    attrs_[name] = FindVarRecursive(var_desc->Name());
  } else if (type == proto::AttrType::VARS) {
    PADDLE_ENFORCE_EQ(
        attr_type,
        type,
        platform::errors::InvalidArgument(
            "Required attr.type == proto::AttrType::VARS, but received %s",
            attr_type));
    auto vars_desc = PADDLE_GET_CONST(std::vector<VarDesc *>, attr);
    std::vector<VarDesc *> new_val;
    for (auto &var_desc : vars_desc) {
      VLOG(3) << "Update AttrVars " << name << " with " << var_desc->Name();
      new_val.emplace_back(FindVarRecursive(var_desc->Name()));
    }
    attrs_[name] = std::move(new_val);
  }
}

VarDesc *OpDesc::FindVarRecursive(const std::string &name) {
  auto *cur_block = block_;
  while (cur_block != nullptr && cur_block->ID() >= 0) {
    auto *var = block_->FindVar(name);
    if (var != nullptr) {
      return var;
    }
    cur_block = cur_block->ParentBlock();
  }
  PADDLE_THROW(platform::errors::NotFound(
      "Not found Var(%s) from Block(%d) back into global Block.",
      name,
      block_->ID()));
}

1138
CompileTimeInferShapeContext::CompileTimeInferShapeContext(
Y
Yu Yang 已提交
1139
    const OpDesc &op, const BlockDesc &block)
1140 1141 1142
    : op_(op), block_(block) {}

bool CompileTimeInferShapeContext::HasInput(const std::string &name) const {
1143 1144
  auto inputs = op_.Inputs(/*with_attr_var=*/true);
  if (inputs.find(name) == inputs.end()) {
1145 1146
    return false;
  }
1147 1148
  const std::vector<std::string> &input_names =
      op_.Input(name, /*with_attr_var=*/true);
1149 1150 1151 1152
  auto length = input_names.size();
  if (length == 0) {
    return false;
  }
1153
  PADDLE_ENFORCE_EQ(
1154 1155
      length,
      1UL,
1156 1157
      platform::errors::InvalidArgument("Input(%s) should have only one value, "
                                        "but it has %d values now.",
1158 1159
                                        name,
                                        length));
1160 1161 1162 1163
  return block_.HasVarRecursive(input_names[0]);
}

bool CompileTimeInferShapeContext::HasOutput(const std::string &name) const {
1164 1165 1166
  if (op_.Outputs().find(name) == op_.Outputs().end()) {
    return false;
  }
1167 1168 1169 1170 1171
  const std::vector<std::string> &output_names = op_.Output(name);
  auto length = output_names.size();
  if (length == 0) {
    return false;
  }
1172 1173
  PADDLE_ENFORCE_EQ(length,
                    1UL,
1174 1175 1176
                    platform::errors::InvalidArgument(
                        "Output(%s) should have only one value, "
                        "but it has %d values now.",
1177 1178
                        name,
                        length));
1179 1180 1181
  return block_.HasVarRecursive(output_names[0]);
}

1182
bool CompileTimeInferShapeContext::HasAttr(const std::string &name) const {
1183
  return op_.HasAttr(name, /*with_attr_var=*/false);
1184 1185
}

1186
bool CompileTimeInferShapeContext::HasInputs(const std::string &name) const {
1187 1188
  auto inputs = op_.Inputs(/*with_attr_var=*/true);
  if (inputs.find(name) == inputs.end()) {
1189 1190
    return false;
  }
1191 1192
  const std::vector<std::string> &input_names =
      op_.Input(name, /*with_attr_var=*/true);
1193 1194 1195 1196 1197 1198 1199 1200 1201
  if (input_names.empty()) {
    return false;
  }
  for (auto &input : input_names) {
    if (!block_.HasVarRecursive(input)) return false;
  }
  return true;
}

1202 1203
bool CompileTimeInferShapeContext::HasOutputs(const std::string &name,
                                              bool allow_null) const {
1204 1205 1206
  if (op_.Outputs().find(name) == op_.Outputs().end()) {
    return false;
  }
1207 1208 1209 1210
  const std::vector<std::string> &output_names = op_.Output(name);
  if (output_names.empty()) {
    return false;
  }
1211 1212 1213 1214 1215 1216 1217 1218 1219 1220
  if (allow_null) {
    for (auto &output : output_names) {
      if (block_.HasVarRecursive(output)) return true;
    }
    return false;
  } else {
    for (auto &output : output_names) {
      if (!block_.HasVarRecursive(output)) return false;
    }
    return true;
1221 1222 1223 1224
  }
}

AttrReader CompileTimeInferShapeContext::Attrs() const {
1225
  return AttrReader(op_.GetAttrMap(), op_.GetRuntimeAttrMap());
1226 1227
}

H
hong 已提交
1228
std::vector<std::string> CompileTimeInferShapeContext::Inputs(
1229
    const std::string &name) const {
1230
  return op_.Input(name, /*with_attr_var=*/true);
1231 1232
}

H
hong 已提交
1233
std::vector<std::string> CompileTimeInferShapeContext::Outputs(
1234 1235 1236 1237
    const std::string &name) const {
  return op_.Output(name);
}

F
fengjiayi 已提交
1238
std::vector<DDim> CompileTimeInferShapeContext::GetRepeatedDims(
F
fengjiayi 已提交
1239 1240
    const std::string &name) const {
  auto var = block_.FindVarRecursive(name);
1241 1242
  PADDLE_ENFORCE_NOT_NULL(
      var, platform::errors::NotFound("Variable %s is not found.", name));
F
fengjiayi 已提交
1243 1244 1245 1246
  std::vector<DDim> res;
  try {
    auto shapes = var->GetShapes();
    for (const auto &s : shapes) {
1247
      res.push_back(s.empty() ? phi::make_ddim({0UL}) : phi::make_ddim(s));
F
fengjiayi 已提交
1248 1249
    }
  } catch (...) {
M
minqiyang 已提交
1250
    VLOG(5) << "GetRepeatedDim of variable " << name << " error.";
F
fengjiayi 已提交
1251 1252 1253
    std::rethrow_exception(std::current_exception());
  }
  return res;
1254 1255 1256 1257
}

void CompileTimeInferShapeContext::SetDim(const std::string &name,
                                          const DDim &dim) {
F
fengjiayi 已提交
1258
  block_.FindVarRecursive(name)->SetShape(vectorize(dim));
1259
}
F
fengjiayi 已提交
1260 1261 1262 1263

void CompileTimeInferShapeContext::SetRepeatedDims(
    const std::string &name, const std::vector<DDim> &dims) {
  auto var = block_.FindVarRecursive(name);
1264 1265
  PADDLE_ENFORCE_NOT_NULL(
      var, platform::errors::NotFound("Variable %s is not found.", name));
F
fengjiayi 已提交
1266
  std::vector<std::vector<int64_t>> dim_vec(dims.size());
1267
  std::transform(dims.begin(), dims.end(), dim_vec.begin(), phi::vectorize<>);
F
fengjiayi 已提交
1268
  var->SetShapes(dim_vec);
1269
}
F
fengjiayi 已提交
1270

1271 1272
bool CompileTimeInferShapeContext::IsRuntime() const { return false; }

1273 1274
bool CompileTimeInferShapeContext::IsRunMKLDNNKernel() const { return false; }

1275
proto::VarType::Type CompileTimeInferShapeContext::GetVarType(
1276 1277 1278
    const std::string &name) const {
  return block_.FindVarRecursive(name)->GetType();
}
1279

1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295
std::vector<std::string> AttrVarNames(const Attribute &attr) {
  std::vector<std::string> vars_name;
  if (IsAttrVar(attr)) {
    vars_name.emplace_back(PADDLE_GET_CONST(VarDesc *, attr)->Name());
  } else if (IsAttrVars(attr)) {
    for (auto &iter : PADDLE_GET_CONST(std::vector<VarDesc *>, attr)) {
      vars_name.emplace_back(iter->Name());
    }
  } else {
    PADDLE_THROW(platform::errors::Unimplemented(
        "Unsupported Attribute value type `%s` for AttrVarNames",
        platform::demangle(attr.type().name())));
  }
  return vars_name;
}

F
fengjiayi 已提交
1296 1297
}  // namespace framework
}  // namespace paddle