prune.cc 17.1 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserved.
Y
Yang Yang 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/prune.h"
Y
Yang Yang 已提交
16

17 18
#include <glog/logging.h>

19
#include <queue>
20 21
#include "paddle/fluid/framework/op_proto_maker.h"

Y
Yang Yang 已提交
22 23 24
namespace paddle {
namespace framework {

25 26
const char kFeedOpType[] = "feed";
const char kFetchOpType[] = "fetch";
Y
Yang Yang 已提交
27

28 29 30 31
const char kRecurrent[] = "recurrent";
const char kStates[] = "states";
const char kExStates[] = "ex_states";

32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47
bool HasDependentInputVar(
    const proto::OpDesc& op_desc,
    const std::unordered_set<std::string>& dependent_vars) {
  for (auto& var : op_desc.inputs()) {
    for (auto& argu : var.arguments()) {
      if (dependent_vars.count(argu) != 0) {
        return true;
      }
    }
  }
  return false;
}

bool HasDependentOutputVar(
    const proto::OpDesc& op_desc,
    const std::unordered_set<std::string>& dependent_vars) {
Y
Yang Yang 已提交
48 49 50 51 52 53 54 55 56 57
  for (auto& var : op_desc.outputs()) {
    for (auto& argu : var.arguments()) {
      if (dependent_vars.count(argu) != 0) {
        return true;
      }
    }
  }
  return false;
}

58
bool IsTarget(const proto::OpDesc& op_desc) {
Y
Yang Yang 已提交
59 60 61 62 63 64
  if (op_desc.has_is_target()) {
    return op_desc.is_target();
  }
  return false;
}

65 66 67 68 69 70 71 72
bool HasTrueTarget(const proto::OpDesc& op_desc) {
  return op_desc.has_is_target() && op_desc.is_target();
}

bool HasFalseTarget(const proto::OpDesc& op_desc) {
  return op_desc.has_is_target() && !op_desc.is_target();
}

K
Kexin Zhao 已提交
73
int GetSubBlockIndex(const proto::OpDesc& op_desc) {
74
  // The block index >= 0, so -1 is used to indicate "NotFound".
K
Kexin Zhao 已提交
75 76
  for (auto& attr : op_desc.attrs()) {
    if (attr.type() == proto::AttrType::BLOCK) {
77 78 79 80
      PADDLE_ENFORCE_EQ(attr.has_block_idx(), true,
                        platform::errors::NotFound(
                            "Attribute sub_block is not found in operator %s",
                            op_desc.type()));
K
Kexin Zhao 已提交
81 82 83 84 85
      return attr.block_idx();
    }
  }
  return -1;
}
Y
Yang Yang 已提交
86

87 88 89 90 91 92 93 94 95 96 97 98
void SetSubBlockIndex(proto::OpDesc* op_desc, int sub_idx) {
  for (auto& attr : *op_desc->mutable_attrs()) {
    if (attr.type() == proto::AttrType::BLOCK) {
      PADDLE_ENFORCE_EQ(attr.has_block_idx(), true,
                        platform::errors::NotFound(
                            "Attribute sub_block is not found in operator %s",
                            op_desc->type()));
      attr.set_block_idx(sub_idx);
    }
  }
}

K
Kexin Zhao 已提交
99 100 101 102
bool HasSubBlock(const proto::OpDesc& op_desc) {
  return GetSubBlockIndex(op_desc) > 0;
}

103 104 105 106 107 108 109 110 111 112 113
int GetOpRole(const proto::OpDesc& op_desc) {
  for (auto& attr : op_desc.attrs()) {
    if (attr.name() == OpProtoAndCheckerMaker::OpRoleAttrName()) {
      PADDLE_ENFORCE_EQ(
          attr.has_i(), true,
          platform::errors::NotFound("Attribute %s is empty in operator %s",
                                     OpProtoAndCheckerMaker::OpRoleAttrName(),
                                     op_desc.type()));
      return attr.i();
    }
  }
114 115 116 117
  // If attr op_role is not found, it may be operator created in c++ test, like
  // prune_test.cc. In that case, the op_role should be defaut value, which is
  // kNotSpecified.
  return static_cast<int>(OpRole::kNotSpecified);
118 119
}

120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137
void AppendOpInputVarNames(const proto::OpDesc& op_desc,
                           std::unordered_set<std::string>* vars_set) {
  for (auto& var : op_desc.inputs()) {
    for (auto& arg : var.arguments()) {
      vars_set->emplace(arg);
    }
  }
}

void AppendOpOutputVarNames(const proto::OpDesc& op_desc,
                            std::unordered_set<std::string>* vars_set) {
  for (auto& var : op_desc.outputs()) {
    for (auto& arg : var.arguments()) {
      vars_set->emplace(arg);
    }
  }
}

138 139 140 141 142 143 144 145 146 147
int FindMapByValue(const std::map<int, int>& m, int val) {
  // The content in map should be >= 0, so -1 is used to indicate "NotFound".
  for (auto& pair : m) {
    if (pair.second == val) {
      return pair.first;
    }
  }
  return -1;
}

K
Kexin Zhao 已提交
148 149 150 151 152 153 154
// block_id is the idx of the current block in the input desc
// parent_block_id is the idx of the parent of the current block
// in the output desc, -1 means the current block is global block
// dependent_vars is passed recursively from the parent block to
// the child block to help pruning
void prune_impl(const proto::ProgramDesc& input, proto::ProgramDesc* output,
                int block_id, int parent_block_id,
155
                std::unordered_set<std::string>* dependent_vars,
156 157
                const std::set<std::string> feed_var_names,
                std::map<int, int>* pruned_origin_block_id_map) {
Y
Yang Yang 已提交
158
  auto& block = input.blocks(block_id);
Y
Yang Yang 已提交
159 160 161 162
  auto& ops = block.ops();

  bool expect_feed = true;
  for (auto& op_desc : ops) {
163 164 165 166
    PADDLE_ENFORCE_EQ(
        op_desc.type() != kFeedOpType || expect_feed, true,
        platform::errors::PreconditionNotMet(
            "All FeedOps are at the beginning of the ProgramDesc"));
Y
Yang Yang 已提交
167 168 169 170 171 172
    expect_feed = (op_desc.type() == kFeedOpType);
  }

  bool expect_fetch = true;
  for (auto op_iter = ops.rbegin(); op_iter != ops.rend(); ++op_iter) {
    auto& op_desc = *op_iter;
173 174 175
    PADDLE_ENFORCE_EQ(op_desc.type() != kFetchOpType || expect_fetch, true,
                      platform::errors::PreconditionNotMet(
                          "All FetchOps must at the end of the ProgramDesc"));
Y
Yang Yang 已提交
176 177 178 179 180 181
    expect_fetch = (op_desc.type() == kFetchOpType);
  }

  std::vector<bool> should_run;
  for (auto op_iter = ops.rbegin(); op_iter != ops.rend(); ++op_iter) {
    auto& op_desc = *op_iter;
182 183 184 185 186 187 188 189 190

    if (IsTarget(op_desc) ||
        (HasDependentOutputVar(op_desc, *dependent_vars) &&
         (GetOpRole(op_desc) & static_cast<int>(OpRole::kOptimize)) == 0)) {
      // NOTE(zhiqiu): since optimize op takes the trainable parameters as
      // inputs and output, it may introduce wrong dependency graph.
      // For train mode, the optimize op should be in targets, so is not need
      // and not right to mark optimize op by its outputs.
      // For eval / infer mode, there is no optimize op in program.
Y
Yang Yang 已提交
191 192
      for (auto& var : op_desc.inputs()) {
        for (auto& argu : var.arguments()) {
193 194 195
          if (feed_var_names.count(argu) == 0) {
            dependent_vars->insert(argu);
          }
Y
Yang Yang 已提交
196 197 198 199 200
        }
      }
      should_run.push_back(true);
    } else {
      should_run.push_back(false);
201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217
      // If the output of an op modifies feed vars, the op should not clip.
      // For example, in the transformer structure, the third parameter returned
      // by beam_search op is generally assigned to a feed var. Cutting the
      // assign op will cause an error.
      if (parent_block_id != -1) {
        bool flag = false;
        for (auto& var : op_desc.outputs()) {
          for (auto& argu : var.arguments()) {
            if (feed_var_names.count(argu)) {
              flag = true;
            }
          }
        }
        if (flag) {
          should_run.back() = true;
        }
      }
Y
Yang Yang 已提交
218 219 220 221 222 223 224
    }
  }

  // since we are traversing the ProgramDesc in reverse order
  // we reverse the should_run vector
  std::reverse(should_run.begin(), should_run.end());

K
Kexin Zhao 已提交
225 226 227 228 229 230
  // copy the current block from input to output
  auto* block_field = output->mutable_blocks();
  *block_field->Add() = input.blocks(block_id);

  int output_block_id = output->blocks_size() - 1;
  auto* output_block = output->mutable_blocks(output_block_id);
231 232
  output_block->set_idx(output_block_id);
  output_block->set_parent_idx(parent_block_id);
K
Kexin Zhao 已提交
233

234 235
  (*pruned_origin_block_id_map)[output_block_id] = block_id;

K
Kexin Zhao 已提交
236
  auto* op_field = output_block->mutable_ops();
Y
Yang Yang 已提交
237 238 239
  op_field->Clear();
  for (size_t i = 0; i < should_run.size(); ++i) {
    if (should_run[i]) {
K
Kexin Zhao 已提交
240 241 242
      auto* op = op_field->Add();
      *op = input.blocks(block_id).ops(i);
      if (HasSubBlock(*op)) {
243
        VLOG(2) << "Pruning op which has sub block: " << op->type();
K
Kexin Zhao 已提交
244
        // create sub_block_dependent_vars here to help prune the sub block
245
        std::unordered_set<std::string> sub_block_dependent_vars;
246
        for (auto& var : op->inputs()) {
K
Kexin Zhao 已提交
247
          for (auto& argu : var.arguments()) {
248 249 250
            if (feed_var_names.count(argu) == 0) {
              sub_block_dependent_vars.insert(argu);
            }
K
Kexin Zhao 已提交
251 252
          }
        }
253
        for (auto& var : op->outputs()) {
K
Kexin Zhao 已提交
254
          for (auto& argu : var.arguments()) {
255 256 257
            if (feed_var_names.count(argu) == 0) {
              sub_block_dependent_vars.insert(argu);
            }
K
Kexin Zhao 已提交
258 259
          }
        }
260 261 262 263 264 265 266 267 268 269 270 271 272 273

        // Recurrent op's states are also dependent vars
        if (op->type() == kRecurrent) {
          auto& attributes = op->attrs();
          for (auto& attr : attributes) {
            if (attr.name() == kStates || attr.name() == kExStates) {
              for (auto& argu : attr.strings()) {
                if (feed_var_names.count(argu) == 0) {
                  sub_block_dependent_vars.insert(argu);
                }
              }
            }
          }
        }
K
Kexin Zhao 已提交
274 275 276
        // GetSubBlockIndex(*op) is the idx of the sub_block in the input desc
        // output_block_id is the idx of the current block in the output desc
        prune_impl(input, output, GetSubBlockIndex(*op), output_block_id,
277 278
                   &sub_block_dependent_vars, feed_var_names,
                   pruned_origin_block_id_map);
K
Kexin Zhao 已提交
279
      }
Y
Yang Yang 已提交
280 281
    }
  }
K
Kexin Zhao 已提交
282

K
Kexin Zhao 已提交
283 284 285
  // remove the VarDescs in BlockDesc that are not referenced in
  // the pruned OpDescs
  std::unordered_map<std::string, proto::VarDesc> var_map;
K
Kexin Zhao 已提交
286
  auto* var_field = output->mutable_blocks(output_block_id)->mutable_vars();
K
Kexin Zhao 已提交
287 288
  for (const auto& var : *var_field) {
    var_map[var.name()] = var;
K
Kexin Zhao 已提交
289 290
  }

291
  std::set<std::string> var_names;
K
Kexin Zhao 已提交
292 293
  for (const auto& op : *op_field) {
    auto& input_field = op.inputs();
K
Kexin Zhao 已提交
294 295
    for (auto& input_var : input_field) {
      for (auto& arg : input_var.arguments()) {
296 297 298
        if (var_map.count(arg) != 0) {
          var_names.insert(arg);
        }
K
Kexin Zhao 已提交
299 300 301
      }
    }
    auto& output_field = op.outputs();
K
Kexin Zhao 已提交
302 303
    for (auto& output_var : output_field) {
      for (auto& arg : output_var.arguments()) {
304 305 306
        if (var_map.count(arg) != 0) {
          var_names.insert(arg);
        }
K
Kexin Zhao 已提交
307 308 309
      }
    }
  }
310 311 312 313 314

  var_field->Clear();
  for (const auto& name : var_names) {
    *var_field->Add() = var_map[name];
  }
Y
Yang Yang 已提交
315
}
Y
Yang Yang 已提交
316

317
// TODO(fengjiayi): Prune() could be inplaced to avoid unnecessary copies
318 319 320
std::map<int, int> Prune(const proto::ProgramDesc& input,
                         const std::set<std::string>& feed_var_names,
                         proto::ProgramDesc* output) {
321
  std::unordered_set<std::string> dependent_vars;
K
fix bug  
Kexin Zhao 已提交
322
  output->clear_blocks();
323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341
  std::map<int, int> pruned_origin_block_id_map;
  prune_impl(input, output, 0, -1, &dependent_vars, feed_var_names,
             &pruned_origin_block_id_map);
  // update subblock idx
  for (int i = 0; i < output->blocks_size(); i++) {
    auto* pruned = output->mutable_blocks(i);
    auto* ops = pruned->mutable_ops();
    for (auto op_iter = ops->rbegin(); op_iter != ops->rend(); ++op_iter) {
      auto& op_desc = *op_iter;
      if (HasSubBlock(op_desc)) {
        int origin_sub_idx = GetSubBlockIndex(op_desc);
        auto sub_idx =
            FindMapByValue(pruned_origin_block_id_map, origin_sub_idx);
        PADDLE_ENFORCE_NE(sub_idx, -1,
                          platform::errors::NotFound(
                              "The origin sub block id should be found in "
                              "pruned_progin_block_id_map"));
        SetSubBlockIndex(&op_desc, sub_idx);
      }
342 343
    }
  }
344
  return pruned_origin_block_id_map;
345 346
}

347
void PruneBackwardImpl(proto::BlockDesc* origin, proto::BlockDesc* pruned) {
348 349 350
  std::unordered_set<std::string> op_input_vars;
  std::unordered_set<std::string> op_output_vars;

351 352
  // Step 1. Mark backward, optimize and lrsched ops in the block
  auto* ops = origin->mutable_ops();
353 354
  for (auto op_iter = ops->begin(); op_iter != ops->end(); ++op_iter) {
    auto& op_desc = *op_iter;
355 356 357 358
    auto op_role = GetOpRole(op_desc);
    if (op_role & static_cast<int>(OpRole::kOptimize) ||
        op_role & static_cast<int>(OpRole::kBackward) ||
        op_role & static_cast<int>(OpRole::kLRSched)) {
359 360 361 362
      op_desc.set_is_target(false);
    }
  }

363 364 365 366 367
  // Step 2. Copy the forward ops which have not been set false target to new
  // ProgramDesc
  // Note: The proto::ProgramDesc doesn't have interface
  //       to remove op and var
  auto* op_field = pruned->mutable_ops();
368 369
  op_field->Clear();
  for (auto op_iter = ops->begin(); op_iter != ops->end(); ++op_iter) {
370
    if (!HasFalseTarget(*op_iter)) {
371
      auto* op = op_field->Add();
372 373
      AppendOpInputVarNames(*op_iter, &op_input_vars);
      AppendOpOutputVarNames(*op_iter, &op_output_vars);
374 375 376 377
      *op = *op_iter;
    }
  }

378
  // Step 3. Copy the forward vars to new ProgramDesc,
379
  // construct all var's map before clear
380 381
  auto* origin_vars = origin->mutable_vars();
  auto* pruned_vars = pruned->mutable_vars();
382
  std::unordered_map<std::string, proto::VarDesc> var_map;
383
  for (const auto& var : *origin_vars) {
384 385
    var_map[var.name()] = var;
  }
386 387
  pruned_vars->Clear();

388 389 390 391
  std::unordered_set<std::string> var_names;
  var_names.insert(op_input_vars.begin(), op_input_vars.end());
  var_names.insert(op_output_vars.begin(), op_output_vars.end());
  for (const auto& name : var_names) {
392
    if (var_map.count(name)) {
393 394
      // NOTE(zhiqiu): For operator in a conditional block, the related vars
      // may not exist in current block, but in its futher block.
395 396
      *pruned_vars->Add() = var_map[name];
    }
397
  }
398
}  // namespace framework
399

400
std::tuple<framework::ProgramDesc, std::map<int, int>> PruneBackward(
401 402 403 404
    const framework::ProgramDesc& origin) {
  // Copy original ProgramDesc, origin can't be change
  framework::ProgramDesc origin_clone(origin);

405 406 407 408 409 410 411 412
  // Step 1. check if the program contains grad loss operator.
  // If not, the program need no pruning.
  bool has_loss_grad_op = false;
  std::queue<int> block_contains_loss;
  std::queue<int> block_contains_loss_grad;
  for (size_t i = 0; i < origin_clone.Size(); i++) {
    auto block_ops = origin_clone.Block(i).AllOps();
    for (auto op : block_ops) {
413 414
      int op_role = BOOST_GET_MUTABLE(
          int, op->GetAttr(OpProtoAndCheckerMaker::OpRoleAttrName()));
415 416 417 418 419 420
      if (op_role == (static_cast<int>(OpRole::kBackward) |
                      static_cast<int>(OpRole::kLoss))) {
        op->SetIsTarget(false);
        has_loss_grad_op = true;
        break;
      }
421 422 423
    }
  }

424 425 426 427 428 429 430 431
  std::map<int, int> pruned_progin_block_id_map;
  if (!has_loss_grad_op) {
    // No pruning, fast return a copy of the origin ProgramDesc with an empty
    // map, means default mapped, i.e.{0:0, 1:1, ..., n:n}.
    return std::make_tuple(framework::ProgramDesc(origin_clone),
                           pruned_progin_block_id_map);
  }

432 433
  proto::ProgramDesc pruned_desc;
  pruned_desc.clear_blocks();
434

435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465
  // Step 2. Prune backward for each block.
  for (size_t i = 0; i < origin_clone.Size(); i++) {
    auto pruned = proto::BlockDesc();
    auto origin = origin_clone.Proto()->mutable_blocks(i);

    PruneBackwardImpl(origin, &pruned);
    // If pruned block contains no operator, it means the block is a
    // backward block and should be pruned.
    // Else, add the block to pruned_desc and update its id & parent_id.
    if (pruned.ops_size() > 0) {
      auto* block_field = pruned_desc.mutable_blocks();
      *block_field->Add() = pruned;

      auto pruned_block_id = pruned_desc.blocks_size() - 1;
      pruned_progin_block_id_map[pruned_block_id] = origin->idx();
      auto* pruned_block = pruned_desc.mutable_blocks(pruned_block_id);
      pruned_block->set_idx(pruned_block_id);

      if (origin->parent_idx() == -1) {
        pruned_block->set_parent_idx(-1);
      } else {
        auto parent_idx =
            FindMapByValue(pruned_progin_block_id_map, origin->parent_idx());
        PADDLE_ENFORCE_NE(parent_idx, -1,
                          platform::errors::NotFound(
                              "The origin parent block id is not found in "
                              "pruned_progin_block_id_map"));
        pruned_block->set_parent_idx(parent_idx);
      }
    }
  }
466

467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490
  // Step 3. Update subblock attribute for conditional operator.
  // This should be performed after all blocks pruned.
  for (int i = 0; i < pruned_desc.blocks_size(); i++) {
    auto* pruned = pruned_desc.mutable_blocks(i);
    auto* ops = pruned->mutable_ops();
    for (auto op_iter = ops->begin(); op_iter != ops->end(); ++op_iter) {
      auto& op_desc = *op_iter;
      if (HasSubBlock(op_desc)) {
        int origin_sub_idx = GetSubBlockIndex(op_desc);
        auto sub_idx =
            FindMapByValue(pruned_progin_block_id_map, origin_sub_idx);
        PADDLE_ENFORCE_NE(sub_idx, -1,
                          platform::errors::NotFound(
                              "The origin sub block id is not found in "
                              "pruned_progin_block_id_map"));
        SetSubBlockIndex(&op_desc, sub_idx);
      }
    }
  }

  // Step 4. Return a tuple
  return std::make_tuple(framework::ProgramDesc(pruned_desc),
                         pruned_progin_block_id_map);
}  // namespace framework
491

Y
Yang Yang 已提交
492 493
}  // namespace framework
}  // namespace paddle