analysis_predictor.cc 18.2 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
// Copyright (c) 2018 PaddlePaddle Authors. All Rights Reserved.
//
// 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
Yan Chunwei 已提交
15
#include "paddle/fluid/inference/api/analysis_predictor.h"
16 17
#include <glog/logging.h>
#include <algorithm>
18
#include <memory>
19 20
#include <string>
#include <vector>
21
#include "paddle/fluid/framework/feed_fetch_method.h"
22
#include "paddle/fluid/framework/feed_fetch_type.h"
Y
Yan Chunwei 已提交
23
#include "paddle/fluid/framework/ir/fuse_pass_base.h"
24
#include "paddle/fluid/framework/ir/pass.h"
25
#include "paddle/fluid/framework/naive_executor.h"
26
#include "paddle/fluid/framework/scope.h"
27
#include "paddle/fluid/inference/api/helper.h"
28
#include "paddle/fluid/inference/api/paddle_inference_api.h"
L
luotao1 已提交
29
#include "paddle/fluid/inference/api/paddle_inference_pass.h"
30 31 32
#if PADDLE_WITH_TENSORRT
#include "paddle/fluid/inference/tensorrt/convert/op_converter.h"
#endif
33
#include "paddle/fluid/inference/utils/singleton.h"
34
#include "paddle/fluid/platform/cpu_helper.h"
T
tensor-tang 已提交
35 36 37
#include "paddle/fluid/platform/profiler.h"

DECLARE_bool(profile);
38 39 40

namespace paddle {

41 42
using contrib::AnalysisConfig;

43 44 45 46 47 48 49 50 51 52 53
namespace {
bool IsPersistable(const framework::VarDesc *var) {
  if (var->Persistable() &&
      var->GetType() != framework::proto::VarType::FEED_MINIBATCH &&
      var->GetType() != framework::proto::VarType::FETCH_LIST) {
    return true;
  }
  return false;
}
}  // namespace

Y
Yan Chunwei 已提交
54
bool AnalysisPredictor::Init(
55 56
    const std::shared_ptr<framework::Scope> &parent_scope,
    const std::shared_ptr<framework::ProgramDesc> &program) {
57
  VLOG(30) << "Predictor::init()";
T
tensor-tang 已提交
58 59 60 61 62 63 64 65 66 67
#if !defined(_WIN32)
  if (FLAGS_profile) {
    LOG(WARNING) << "Profiler is actived, might affect the performance";
    LOG(INFO) << "You can turn off by set gflags '-profile false'";
    auto tracking_device = config_.use_gpu ? platform::ProfilerState::kAll
                                           : platform::ProfilerState::kCPU;
    platform::EnableProfiler(tracking_device);
  }
#endif

68
  // no matter with or without MKLDNN
L
luotao1 已提交
69
  paddle::platform::SetNumThreads(config_.cpu_math_library_num_threads());
70

71 72 73 74 75 76 77 78 79 80 81 82 83
  if (!PrepareScope(parent_scope)) {
    return false;
  }
  if (!CreateExecutor()) {
    return false;
  }
  if (!PrepareProgram(program)) {
    return false;
  }

  // Prepare executor, create local variables.
  if (!PrepareExecutor()) {
    return true;
Y
Yan Chunwei 已提交
84
  }
85 86 87 88 89 90 91 92 93

  // Get the feed_target_names and fetch_target_names
  PrepareFeedFetch();

  return true;
}

bool AnalysisPredictor::PrepareScope(
    const std::shared_ptr<framework::Scope> &parent_scope) {
Y
Yan Chunwei 已提交
94
  if (parent_scope) {
95 96 97
    PADDLE_ENFORCE_NOT_NULL(
        parent_scope,
        "Both program and parent_scope should be set in Clone mode.");
Y
Yan Chunwei 已提交
98
    scope_ = parent_scope;
99
    status_is_cloned_ = true;
Y
Yan Chunwei 已提交
100 101 102
  } else {
    paddle::framework::InitDevices(false);
    scope_.reset(new paddle::framework::Scope());
103
    status_is_cloned_ = false;
Y
Yan Chunwei 已提交
104
  }
105 106 107 108 109
  sub_scope_ = &scope_->NewScope();
  return true;
}
bool AnalysisPredictor::PrepareProgram(
    const std::shared_ptr<framework::ProgramDesc> &program) {
110 111
  if (!program) {
    if (!LoadProgramDesc()) return false;
112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132

    // Optimize the program, and load parameters and modify them in the
    // scope_.
    // This will change the scope_ address.
    if (config_.enable_ir_optim) {
      status_ir_optim_enabled_ = true;
      OptimizeInferenceProgram();
    } else {
      // If the parent_scope is passed, we assert that the persistable variables
      // are already created, so just create the no persistable variables.

      // If not cloned, the parameters should be loaded
      // OptimizeInferenceProgram.
      // So in both cases, just the local variables are needed to load, not the
      // parematers.
      executor_->CreateVariables(*inference_program_, 0, true, sub_scope_);

      // Load parameters
      LOG(INFO) << "load parameters ";
      LoadParameters();
    }
Y
Yan Chunwei 已提交
133
  } else {
134 135
    // If the program is passed from external, no need to optimize it, this
    // logic is used in the clone scenario.
136 137
    inference_program_ = program;
  }
M
Michal Gallus 已提交
138

139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154
  executor_->CreateVariables(*inference_program_, 0, false, sub_scope_);

  return true;
}
bool AnalysisPredictor::CreateExecutor() {
  if (config_.use_gpu) {
    status_use_gpu_ = true;
    place_ = paddle::platform::CUDAPlace(config_.device);
  } else {
    place_ = paddle::platform::CPUPlace();
  }
  executor_.reset(new paddle::framework::NaiveExecutor(place_));
  return true;
}
bool AnalysisPredictor::PrepareExecutor() {
  executor_->Prepare(sub_scope_, *inference_program_, 0,
155 156
                     config_.use_feed_fetch_ops);

157
  PADDLE_ENFORCE_NOT_NULL(sub_scope_);
Y
Yan Chunwei 已提交
158

159 160 161
  return true;
}

L
luotao1 已提交
162
void AnalysisPredictor::SetMkldnnThreadID(int tid) {
L
luotao1 已提交
163 164 165 166 167 168 169
#ifdef PADDLE_WITH_MKLDNN
  platform::set_cur_thread_id(tid);
#else
  LOG(ERROR) << "Please compile with MKLDNN first to use MKLDNN";
#endif
}

170 171 172
bool AnalysisPredictor::Run(const std::vector<PaddleTensor> &inputs,
                            std::vector<PaddleTensor> *output_data,
                            int batch_size) {
173
  VLOG(30) << "Predictor::predict";
174 175 176 177 178 179
  inference::Timer timer;
  timer.tic();
  // set feed variable
  framework::Scope *scope = sub_scope_ ? sub_scope_ : scope_.get();
  if (!SetFeed(inputs, scope)) {
    LOG(ERROR) << "fail to set feed";
Y
Yan Chunwei 已提交
180
    return false;
181
  }
M
Michal Gallus 已提交
182

183 184 185
  // Run the inference program
  // if share variables, we need not create variables
  executor_->Run();
186

187 188 189 190
  // get fetch variable
  if (!GetFetch(output_data, scope)) {
    LOG(ERROR) << "fail to get fetches";
    return false;
T
tensor-tang 已提交
191
  }
192
  VLOG(30) << "predict cost: " << timer.toc() << "ms";
Y
Yan Chunwei 已提交
193 194 195 196

  // Fix TensorArray reuse not cleaned bug.
  tensor_array_batch_cleaner_.CollectTensorArrays(scope_.get());
  tensor_array_batch_cleaner_.ResetTensorArray();
197 198
  return true;
}
199

200 201
bool AnalysisPredictor::SetFeed(const std::vector<PaddleTensor> &inputs,
                                framework::Scope *scope) {
202
  VLOG(30) << "Predictor::set_feed";
203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266
  if (inputs.size() != feeds_.size()) {
    LOG(ERROR) << "wrong feed input size, need " << feeds_.size() << " but get "
               << inputs.size();
    return false;
  }

  // Cache the inputs memory for better concurrency performance.
  feed_tensors_.resize(inputs.size());

  for (size_t i = 0; i < inputs.size(); ++i) {
    auto &input = feed_tensors_[i];
    framework::DDim ddim = framework::make_ddim(inputs[i].shape);
    void *input_ptr;
    if (inputs[i].dtype == PaddleDType::INT64) {
      input_ptr = input.mutable_data<int64_t>(ddim, platform::CPUPlace());
    } else if (inputs[i].dtype == PaddleDType::FLOAT32) {
      input_ptr = input.mutable_data<float>(ddim, platform::CPUPlace());
    } else {
      LOG(ERROR) << "unsupported feed type " << inputs[i].dtype;
      return false;
    }

    // TODO(panyx0718): Init LoDTensor from existing memcpy to save a copy.
    std::memcpy(static_cast<void *>(input_ptr), inputs[i].data.data(),
                inputs[i].data.length());
    // TODO(Superjomn) Low performance, need optimization for heavy LoD copy.
    framework::LoD lod;
    for (auto &level : inputs[i].lod) {
      lod.emplace_back(level);
    }
    input.set_lod(lod);
    int idx = -1;
    if (config_.specify_input_name) {
      idx = feed_names_[inputs[i].name];
    } else {
      idx = boost::get<int>(feeds_[i]->GetAttr("col"));
    }
    framework::SetFeedVariable(scope, input, "feed", idx);
  }
  return true;
}

template <typename T>
void AnalysisPredictor::GetFetchOne(const framework::LoDTensor &fetch,
                                    PaddleTensor *output) {
  // set shape.
  auto shape = framework::vectorize(fetch.dims());
  output->shape.assign(shape.begin(), shape.end());
  // set data.
  const T *data = fetch.data<T>();
  int num_elems = inference::VecReduceToInt(shape);
  output->data.Resize(num_elems * sizeof(T));
  // The fetched tensor output by fetch op, should always in CPU memory, so just
  // copy.
  memcpy(output->data.data(), data, num_elems * sizeof(T));
  // set lod
  output->lod.clear();
  for (auto &level : fetch.lod()) {
    output->lod.emplace_back(level.begin(), level.end());
  }
}

bool AnalysisPredictor::GetFetch(std::vector<PaddleTensor> *outputs,
                                 framework::Scope *scope) {
267
  VLOG(30) << "Predictor::get_fetch";
268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285
  outputs->resize(fetchs_.size());
  for (size_t i = 0; i < fetchs_.size(); ++i) {
    int idx = boost::get<int>(fetchs_[i]->GetAttr("col"));
    PADDLE_ENFORCE((size_t)idx == i);
    framework::LoDTensor &fetch =
        framework::GetFetchVariable(*scope, "fetch", idx);
    auto type = fetch.type();
    auto output = &(outputs->at(i));
    if (type == typeid(float)) {
      GetFetchOne<float>(fetch, output);
      output->dtype = PaddleDType::FLOAT32;
    } else if (type == typeid(int64_t)) {
      GetFetchOne<int64_t>(fetch, output);
      output->dtype = PaddleDType::INT64;
    } else {
      LOG(ERROR) << "unknown type, only support float32 and int64 now.";
    }
  }
Y
Yan Chunwei 已提交
286 287
  return true;
}
288

289
// NOTE All the members in AnalysisConfig should be copied to Argument.
Y
Yan Chunwei 已提交
290
void AnalysisPredictor::OptimizeInferenceProgram() {
291 292 293
  status_program_optimized_ = true;

  argument_.SetUseGPU(config_.use_gpu);
S
superjomn 已提交
294
  argument_.SetGPUDeviceId(config_.device);
Y
Yan Chunwei 已提交
295 296
  // Analyze inference_program
  if (!config_.model_dir.empty()) {
297
    argument_.SetModelDir(config_.model_dir);
Y
Yan Chunwei 已提交
298 299 300 301 302
  } else {
    PADDLE_ENFORCE(
        !config_.param_file.empty(),
        "Either model_dir or (param_file, prog_file) should be set.");
    PADDLE_ENFORCE(!config_.prog_file.empty());
303 304
    argument_.SetModelProgramPath(config_.prog_file);
    argument_.SetModelParamsPath(config_.param_file);
Y
Yan Chunwei 已提交
305
  }
306

307 308 309 310
  if (config_.use_gpu && config_.use_tensorrt_) {
    argument_.SetUseTensorRT(true);
    argument_.SetTensorRtWorkspaceSize(config_.tensorrt_workspace_size_);
    argument_.SetTensorRtMaxBatchSize(config_.tensorrt_max_batchsize_);
W
Wojciech Uss 已提交
311
  }
312

313 314 315 316 317 318 319 320 321
  auto passes = config_.pass_builder()->AllPasses();
  if (!config_.enable_ir_optim) passes.clear();
  argument_.SetIrAnalysisPasses(passes);
  argument_.SetScopeNotOwned(const_cast<framework::Scope *>(scope_.get()));
  Analyzer().Run(&argument_);

  PADDLE_ENFORCE(argument_.scope_valid());
  VLOG(5) << "to prepare executor";
  ARGUMENT_CHECK_FIELD((&argument_), ir_analyzed_program);
Y
Yan Chunwei 已提交
322
  inference_program_.reset(
323
      new framework::ProgramDesc(argument_.ir_analyzed_program()));
324
  LOG(INFO) << "== optimize end ==";
Y
Yan Chunwei 已提交
325
}
326 327

template <>
328 329
std::unique_ptr<PaddlePredictor> CreatePaddlePredictor<
    AnalysisConfig, PaddleEngineKind::kAnalysis>(const AnalysisConfig &config) {
330
  VLOG(30) << "create AnalysisConfig";
331 332 333 334 335 336 337 338 339 340 341 342 343
  if (config.use_gpu) {
    // 1. GPU memeroy
    PADDLE_ENFORCE_GT(
        config.fraction_of_gpu_memory, 0.f,
        "fraction_of_gpu_memory in the config should be set to range (0., 1.]");
    PADDLE_ENFORCE_GE(config.device, 0, "Invalid device id %d", config.device);
    std::vector<std::string> flags;
    if (config.fraction_of_gpu_memory >= 0.0f ||
        config.fraction_of_gpu_memory <= 0.95f) {
      flags.push_back("dummpy");
      std::string flag = "--fraction_of_gpu_memory_to_use=" +
                         std::to_string(config.fraction_of_gpu_memory);
      flags.push_back(flag);
344
      VLOG(30) << "set flag: " << flag;
345 346 347 348 349
      framework::InitGflags(flags);
    }
  }

  std::unique_ptr<PaddlePredictor> predictor(new AnalysisPredictor(config));
350
  if (!dynamic_cast<AnalysisPredictor *>(predictor.get())->Init(nullptr)) {
351 352
    return nullptr;
  }
353
  return std::move(predictor);
354 355
}

356
void AnalysisPredictor::PrepareFeedFetch() {
357 358
  PADDLE_ENFORCE_NOT_NULL(sub_scope_);
  CreateFeedFetchVar(sub_scope_);
359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376
  for (auto *op : inference_program_->Block(0).AllOps()) {
    if (op->Type() == "feed") {
      int idx = boost::get<int>(op->GetAttr("col"));
      if (feeds_.size() <= static_cast<size_t>(idx)) {
        feeds_.resize(idx + 1);
      }
      feeds_[idx] = op;
      feed_names_[op->Output("Out")[0]] = idx;
    } else if (op->Type() == "fetch") {
      int idx = boost::get<int>(op->GetAttr("col"));
      if (fetchs_.size() <= static_cast<size_t>(idx)) {
        fetchs_.resize(idx + 1);
      }
      fetchs_[idx] = op;
    }
  }
}

377 378 379 380 381 382 383 384
void AnalysisPredictor::CreateFeedFetchVar(framework::Scope *scope) {
  PADDLE_ENFORCE_NOT_NULL(scope);
  auto *var = scope->Var("feed");
  var->GetMutable<framework::FeedFetchList>();
  var = scope->Var("fetch");
  var->GetMutable<framework::FeedFetchList>();
}

385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406
std::unique_ptr<ZeroCopyTensor> AnalysisPredictor::GetInputTensor(
    const std::string &name) {
  PADDLE_ENFORCE(executor_->scope()->FindVar(name), "no name called %s", name);
  std::unique_ptr<ZeroCopyTensor> res(
      new ZeroCopyTensor(static_cast<void *>(executor_->scope())));
  res->input_or_output_ = true;
  res->SetName(name);
  return res;
}

std::unique_ptr<ZeroCopyTensor> AnalysisPredictor::GetOutputTensor(
    const std::string &name) {
  PADDLE_ENFORCE(executor_->scope()->FindVar(name), "no name called %s", name);
  std::unique_ptr<ZeroCopyTensor> res(
      new ZeroCopyTensor(static_cast<void *>(executor_->scope())));
  res->input_or_output_ = false;
  res->SetName(name);
  return res;
}

bool AnalysisPredictor::ZeroCopyRun() {
  executor_->Run();
Y
Yan Chunwei 已提交
407 408 409
  // Fix TensorArray reuse not cleaned bug.
  tensor_array_batch_cleaner_.CollectTensorArrays(scope_.get());
  tensor_array_batch_cleaner_.ResetTensorArray();
410 411 412 413 414
  return true;
}

bool AnalysisPredictor::LoadProgramDesc() {
  // Initialize the inference program
415
  std::string filename;
416
  if (!config_.model_dir.empty()) {
417
    filename = config_.model_dir + "/__model__";
418 419 420 421
  } else if (!config_.prog_file.empty() && !config_.param_file.empty()) {
    // All parameters are saved in a single file.
    // The file names should be consistent with that used
    // in Python API `fluid.io.save_inference_model`.
422
    filename = config_.prog_file;
423
  } else {
424 425 426 427 428
    if (config_.model_dir.empty() && config_.prog_file.empty()) {
      LOG(ERROR)
          << "Either model_dir or (prog_file, param_file) should be set.";
      return false;
    }
429 430 431 432 433
    LOG(ERROR) << string::Sprintf(
        "not valid model path '%s' or program path '%s'.", config_.model_dir,
        config_.param_file);
    return false;
  }
434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500

  std::string pb_content;
  // Read binary
  std::ifstream fin(filename, std::ios::in | std::ios::binary);
  PADDLE_ENFORCE(static_cast<bool>(fin), "Cannot open file %s", filename);
  fin.seekg(0, std::ios::end);

  pb_content.resize(fin.tellg());
  fin.seekg(0, std::ios::beg);
  fin.read(&(pb_content.at(0)), pb_content.size());
  fin.close();

  // Create ProgramDesc
  framework::proto::ProgramDesc proto;
  proto.ParseFromString(pb_content);
  inference_program_.reset(new framework::ProgramDesc(proto));
  return true;
}

bool AnalysisPredictor::LoadParameters() {
  PADDLE_ENFORCE_NOT_NULL(inference_program_.get(),
                          "The inference program should be loaded first.");
  const auto &global_block = inference_program_->MutableBlock(0);

  // create a temporary program to load parameters.

  std::unique_ptr<framework::ProgramDesc> load_program(
      new framework::ProgramDesc());
  framework::BlockDesc *load_block = load_program->MutableBlock(0);
  std::vector<std::string> params;

  for (auto *var : global_block->AllVars()) {
    if (IsPersistable(var)) {
      VLOG(3) << "persistable variable's name: " << var->Name();

      framework::VarDesc *new_var = load_block->Var(var->Name());
      new_var->SetShape(var->GetShape());
      new_var->SetDataType(var->GetDataType());
      new_var->SetType(var->GetType());
      new_var->SetLoDLevel(var->GetLoDLevel());
      new_var->SetPersistable(true);

      if (!config_.param_file.empty()) {
        params.push_back(new_var->Name());
      } else {
        // append_op
        framework::OpDesc *op = load_block->AppendOp();
        op->SetType("load");
        op->SetOutput("Out", {new_var->Name()});
        op->SetAttr("file_path", {config_.model_dir + "/" + new_var->Name()});
        op->CheckAttrs();
      }
    }
  }

  if (!config_.param_file.empty()) {
    // sort paramlist to have consistent ordering
    std::sort(params.begin(), params.end());
    // append just the load_combine op
    framework::OpDesc *op = load_block->AppendOp();
    op->SetType("load_combine");
    op->SetOutput("Out", params);
    op->SetAttr("file_path", {config_.param_file});
    op->CheckAttrs();
  }

  // Use NaiveExecutor to Load parameters.
S
superjomn 已提交
501
  framework::NaiveExecutor e(place_);
502 503 504 505
  e.Prepare(scope_.get(), *load_program, 0, false);
  e.Run();
  VLOG(3) << "get " << scope_->LocalVarNames().size() << " vars after load";

506 507
  return true;
}
508 509 510 511 512 513 514 515 516 517 518 519 520

AnalysisPredictor::~AnalysisPredictor() {
#if !defined(_WIN32)
  if (FLAGS_profile) {
    platform::DisableProfiler(platform::EventSortingKey::kTotal,
                              "./profile.log");
  }
#endif
  if (sub_scope_) {
    scope_->DeleteScope(sub_scope_);
  }
}

521 522 523 524 525 526
std::unique_ptr<PaddlePredictor> AnalysisPredictor::Clone() {
  auto *x = new AnalysisPredictor(config_);
  x->Init(scope_, inference_program_);
  return std::unique_ptr<PaddlePredictor>(x);
}

Y
Yan Chunwei 已提交
527 528
template <>
std::unique_ptr<PaddlePredictor> CreatePaddlePredictor<contrib::AnalysisConfig>(
529
    const contrib::AnalysisConfig &config) {
Y
Yan Chunwei 已提交
530 531 532 533
  return CreatePaddlePredictor<contrib::AnalysisConfig,
                               PaddleEngineKind::kAnalysis>(config);
}

534
}  // namespace paddle
535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556

#if PADDLE_WITH_TENSORRT
USE_TRT_CONVERTER(elementwise_add_weight);
USE_TRT_CONVERTER(elementwise_add_tensor);
USE_TRT_CONVERTER(elementwise_sub_tensor);
USE_TRT_CONVERTER(elementwise_div_tensor);
USE_TRT_CONVERTER(elementwise_mul_tensor);
USE_TRT_CONVERTER(elementwise_max_tensor);
USE_TRT_CONVERTER(elementwise_min_tensor);
USE_TRT_CONVERTER(elementwise_pow_tensor);
USE_TRT_CONVERTER(mul);
USE_TRT_CONVERTER(conv2d);
USE_TRT_CONVERTER(relu);
USE_TRT_CONVERTER(sigmoid);
USE_TRT_CONVERTER(tanh);
USE_TRT_CONVERTER(fc);
USE_TRT_CONVERTER(pool2d);
USE_TRT_CONVERTER(softmax);
USE_TRT_CONVERTER(batch_norm);
USE_TRT_CONVERTER(concat);
USE_TRT_CONVERTER(dropout);
USE_TRT_CONVERTER(pad);
557
USE_TRT_CONVERTER(split);
558 559
USE_TRT_CONVERTER(prelu);
USE_TRT_CONVERTER(conv2d_transpose);
H
hjchen2 已提交
560
USE_TRT_CONVERTER(leaky_relu);
561
#endif