prune.cc 16.8 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>

Y
Yang Yang 已提交
19
#include <algorithm>
20
#include <memory>
21
#include <queue>
Y
Yang Yang 已提交
22 23
#include <set>
#include <string>
24
#include <tuple>
K
Kexin Zhao 已提交
25
#include <unordered_map>
26
#include <unordered_set>
Y
Yang Yang 已提交
27 28
#include <vector>

29 30 31 32 33
#include "paddle/fluid/framework/block_desc.h"
#include "paddle/fluid/framework/op_desc.h"
#include "paddle/fluid/framework/op_proto_maker.h"
#include "paddle/fluid/framework/program_desc.h"

Y
Yang Yang 已提交
34 35 36
namespace paddle {
namespace framework {

37 38
const char kFeedOpType[] = "feed";
const char kFetchOpType[] = "fetch";
Y
Yang Yang 已提交
39

40 41 42 43
const char kRecurrent[] = "recurrent";
const char kStates[] = "states";
const char kExStates[] = "ex_states";

44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59
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
60 61 62 63 64 65 66 67 68 69
  for (auto& var : op_desc.outputs()) {
    for (auto& argu : var.arguments()) {
      if (dependent_vars.count(argu) != 0) {
        return true;
      }
    }
  }
  return false;
}

70
bool IsTarget(const proto::OpDesc& op_desc) {
Y
Yang Yang 已提交
71 72 73 74 75 76
  if (op_desc.has_is_target()) {
    return op_desc.is_target();
  }
  return false;
}

77 78 79 80 81 82 83 84
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
85
int GetSubBlockIndex(const proto::OpDesc& op_desc) {
86
  // The block index >= 0, so -1 is used to indicate "NotFound".
K
Kexin Zhao 已提交
87 88
  for (auto& attr : op_desc.attrs()) {
    if (attr.type() == proto::AttrType::BLOCK) {
89 90 91 92
      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
93 94 95 96 97
      return attr.block_idx();
    }
  }
  return -1;
}
Y
Yang Yang 已提交
98

99 100 101 102 103 104 105 106 107 108 109 110
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
111 112 113 114
bool HasSubBlock(const proto::OpDesc& op_desc) {
  return GetSubBlockIndex(op_desc) > 0;
}

115 116 117 118 119 120 121 122 123 124 125
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();
    }
  }
126 127 128 129
  // 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);
130 131
}

132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149
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);
    }
  }
}

150 151 152 153 154 155 156 157 158 159
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
160 161 162 163 164 165 166
// 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,
167
                std::unordered_set<std::string>* dependent_vars,
168 169
                const std::set<std::string> feed_var_names,
                std::map<int, int>* pruned_origin_block_id_map) {
Y
Yang Yang 已提交
170
  auto& block = input.blocks(block_id);
Y
Yang Yang 已提交
171 172 173 174
  auto& ops = block.ops();

  bool expect_feed = true;
  for (auto& op_desc : ops) {
175 176 177 178
    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
179 180 181 182 183 184
    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;
185 186 187
    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
188 189 190 191 192 193
    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;
194 195 196 197 198 199 200 201 202

    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
203 204
      for (auto& var : op_desc.inputs()) {
        for (auto& argu : var.arguments()) {
205 206 207
          if (feed_var_names.count(argu) == 0) {
            dependent_vars->insert(argu);
          }
Y
Yang Yang 已提交
208 209 210 211 212 213 214 215 216 217 218 219
        }
      }
      should_run.push_back(true);
    } else {
      should_run.push_back(false);
    }
  }

  // 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
220 221 222 223 224 225
  // 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);
226 227
  output_block->set_idx(output_block_id);
  output_block->set_parent_idx(parent_block_id);
K
Kexin Zhao 已提交
228

229 230
  (*pruned_origin_block_id_map)[output_block_id] = block_id;

K
Kexin Zhao 已提交
231
  auto* op_field = output_block->mutable_ops();
Y
Yang Yang 已提交
232 233 234
  op_field->Clear();
  for (size_t i = 0; i < should_run.size(); ++i) {
    if (should_run[i]) {
K
Kexin Zhao 已提交
235 236 237
      auto* op = op_field->Add();
      *op = input.blocks(block_id).ops(i);
      if (HasSubBlock(*op)) {
238
        VLOG(2) << "Pruning op which has sub block: " << op->type();
K
Kexin Zhao 已提交
239
        // create sub_block_dependent_vars here to help prune the sub block
240
        std::unordered_set<std::string> sub_block_dependent_vars;
241
        for (auto& var : op->inputs()) {
K
Kexin Zhao 已提交
242
          for (auto& argu : var.arguments()) {
243 244 245
            if (feed_var_names.count(argu) == 0) {
              sub_block_dependent_vars.insert(argu);
            }
K
Kexin Zhao 已提交
246 247
          }
        }
248
        for (auto& var : op->outputs()) {
K
Kexin Zhao 已提交
249
          for (auto& argu : var.arguments()) {
250 251 252
            if (feed_var_names.count(argu) == 0) {
              sub_block_dependent_vars.insert(argu);
            }
K
Kexin Zhao 已提交
253 254
          }
        }
255 256 257 258 259 260 261 262 263 264 265 266 267 268

        // 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
269 270 271
        // 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,
272 273
                   &sub_block_dependent_vars, feed_var_names,
                   pruned_origin_block_id_map);
K
Kexin Zhao 已提交
274
      }
Y
Yang Yang 已提交
275 276
    }
  }
K
Kexin Zhao 已提交
277

K
Kexin Zhao 已提交
278 279 280
  // 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
281
  auto* var_field = output->mutable_blocks(output_block_id)->mutable_vars();
K
Kexin Zhao 已提交
282 283
  for (const auto& var : *var_field) {
    var_map[var.name()] = var;
K
Kexin Zhao 已提交
284 285
  }

286
  std::set<std::string> var_names;
K
Kexin Zhao 已提交
287 288
  for (const auto& op : *op_field) {
    auto& input_field = op.inputs();
K
Kexin Zhao 已提交
289 290
    for (auto& input_var : input_field) {
      for (auto& arg : input_var.arguments()) {
291 292 293
        if (var_map.count(arg) != 0) {
          var_names.insert(arg);
        }
K
Kexin Zhao 已提交
294 295 296
      }
    }
    auto& output_field = op.outputs();
K
Kexin Zhao 已提交
297 298
    for (auto& output_var : output_field) {
      for (auto& arg : output_var.arguments()) {
299 300 301
        if (var_map.count(arg) != 0) {
          var_names.insert(arg);
        }
K
Kexin Zhao 已提交
302 303 304
      }
    }
  }
305 306 307 308 309

  var_field->Clear();
  for (const auto& name : var_names) {
    *var_field->Add() = var_map[name];
  }
Y
Yang Yang 已提交
310
}
Y
Yang Yang 已提交
311

312
// TODO(fengjiayi): Prune() could be inplaced to avoid unnecessary copies
313 314 315
std::map<int, int> Prune(const proto::ProgramDesc& input,
                         const std::set<std::string>& feed_var_names,
                         proto::ProgramDesc* output) {
316
  std::unordered_set<std::string> dependent_vars;
K
fix bug  
Kexin Zhao 已提交
317
  output->clear_blocks();
318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336
  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);
      }
337 338
    }
  }
339
  return pruned_origin_block_id_map;
340 341
}

342
void PruneBackwardImpl(proto::BlockDesc* origin, proto::BlockDesc* pruned) {
343 344 345
  std::unordered_set<std::string> op_input_vars;
  std::unordered_set<std::string> op_output_vars;

346 347
  // Step 1. Mark backward, optimize and lrsched ops in the block
  auto* ops = origin->mutable_ops();
348 349
  for (auto op_iter = ops->begin(); op_iter != ops->end(); ++op_iter) {
    auto& op_desc = *op_iter;
350 351 352 353
    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)) {
354 355 356 357
      op_desc.set_is_target(false);
    }
  }

358 359 360 361 362
  // 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();
363 364
  op_field->Clear();
  for (auto op_iter = ops->begin(); op_iter != ops->end(); ++op_iter) {
365
    if (!HasFalseTarget(*op_iter)) {
366
      auto* op = op_field->Add();
367 368
      AppendOpInputVarNames(*op_iter, &op_input_vars);
      AppendOpOutputVarNames(*op_iter, &op_output_vars);
369 370 371 372
      *op = *op_iter;
    }
  }

373
  // Step 3. Copy the forward vars to new ProgramDesc,
374
  // construct all var's map before clear
375 376
  auto* origin_vars = origin->mutable_vars();
  auto* pruned_vars = pruned->mutable_vars();
377
  std::unordered_map<std::string, proto::VarDesc> var_map;
378
  for (const auto& var : *origin_vars) {
379 380
    var_map[var.name()] = var;
  }
381 382
  pruned_vars->Clear();

383 384 385 386
  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) {
387
    if (var_map.count(name)) {
388 389
      // NOTE(zhiqiu): For operator in a conditional block, the related vars
      // may not exist in current block, but in its futher block.
390 391
      *pruned_vars->Add() = var_map[name];
    }
392
  }
393
}  // namespace framework
394

395
std::tuple<framework::ProgramDesc, std::map<int, int>> PruneBackward(
396 397 398 399
    const framework::ProgramDesc& origin) {
  // Copy original ProgramDesc, origin can't be change
  framework::ProgramDesc origin_clone(origin);

400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415
  // 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) {
      int op_role = boost::get<int>(
          op->GetAttr(OpProtoAndCheckerMaker::OpRoleAttrName()));
      if (op_role == (static_cast<int>(OpRole::kBackward) |
                      static_cast<int>(OpRole::kLoss))) {
        op->SetIsTarget(false);
        has_loss_grad_op = true;
        break;
      }
416 417 418
    }
  }

419 420 421 422 423 424 425 426
  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);
  }

427 428
  proto::ProgramDesc pruned_desc;
  pruned_desc.clear_blocks();
429

430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460
  // 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);
      }
    }
  }
461

462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485
  // 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
486

Y
Yang Yang 已提交
487 488
}  // namespace framework
}  // namespace paddle