analysis_predictor.cc 93.8 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>
18

19
#include <algorithm>
N
nhzlx 已提交
20
#include <fstream>
21
#include <memory>
22
#include <set>
23
#include <string>
24
#include <utility>
25
#include <vector>
26

W
Wilber 已提交
27
#include "paddle/fluid//platform/device/gpu/gpu_types.h"
28
#include "paddle/fluid/framework/feed_fetch_method.h"
29
#include "paddle/fluid/framework/feed_fetch_type.h"
30
#include "paddle/fluid/framework/generator.h"
Y
Yan Chunwei 已提交
31
#include "paddle/fluid/framework/ir/fuse_pass_base.h"
32
#include "paddle/fluid/framework/ir/pass.h"
33
#include "paddle/fluid/framework/naive_executor.h"
34
#include "paddle/fluid/framework/op_proto_maker.h"
35
#include "paddle/fluid/framework/operator.h"
36
#include "paddle/fluid/framework/scope.h"
J
JingZhuangzhuang 已提交
37
#include "paddle/fluid/framework/transfer_scope_cache.h"
Y
Yan Chunwei 已提交
38
#include "paddle/fluid/framework/var_type_traits.h"
39
#include "paddle/fluid/framework/version.h"
40
#include "paddle/fluid/inference/analysis/helper.h"
41
#include "paddle/fluid/inference/analysis/pass_result_info.h"
42
#include "paddle/fluid/inference/analysis/passes/convert_to_mixed_precision.h"
Y
Yan Chunwei 已提交
43
#include "paddle/fluid/inference/analysis/passes/memory_optimize_pass.h"
44
#include "paddle/fluid/inference/api/helper.h"
45
#include "paddle/fluid/inference/api/infer_context.h"
46
#include "paddle/fluid/inference/api/paddle_analysis_config.h"
47
#include "paddle/fluid/inference/api/paddle_inference_api.h"
L
luotao1 已提交
48
#include "paddle/fluid/inference/api/paddle_inference_pass.h"
W
Wilber 已提交
49
#include "paddle/fluid/inference/api/resource_manager.h"
50
#include "paddle/fluid/inference/utils/io_utils.h"
51
#include "paddle/fluid/inference/utils/model_utils.h"
52
#include "paddle/fluid/inference/utils/singleton.h"
53
#include "paddle/fluid/memory/memcpy.h"
54
#include "paddle/fluid/platform/cpu_helper.h"
55
#include "paddle/fluid/platform/device/gpu/gpu_info.h"
56
#include "paddle/fluid/platform/device_context.h"
57
#include "paddle/fluid/platform/place.h"
T
tensor-tang 已提交
58
#include "paddle/fluid/platform/profiler.h"
59
#include "paddle/phi/api/ext/op_meta_info.h"
60 61
#include "paddle/phi/common/backend.h"
#include "paddle/phi/common/data_type.h"
W
Wilber 已提交
62
#include "paddle/phi/common/place.h"
W
Wilber 已提交
63
#include "paddle/phi/core/enforce.h"
64 65
#include "paddle/utils/string/split.h"

66
#if defined(PADDLE_WITH_DISTRIBUTE) && defined(PADDLE_WITH_PSCORE)
67 68 69 70
#include "paddle/fluid/distributed/fleet_executor/fleet_executor.h"
#include "paddle/fluid/distributed/fleet_executor/fleet_executor_desc.pb.h"
#include "paddle/fluid/distributed/fleet_executor/task_node.h"
#endif
T
tensor-tang 已提交
71

72 73 74 75
#ifdef PADDLE_WITH_MKLML
#include "paddle/fluid/platform/dynload/mklml.h"
#endif

76 77 78 79
#ifdef PADDLE_WITH_MKLDNN
#include "paddle/fluid/inference/api/mkldnn_quantizer.h"
#endif

80 81 82 83
#ifdef PADDLE_WITH_ONNXRUNTIME
#include "paddle/fluid/inference/api/onnxruntime_predictor.h"
#endif

84
#ifdef PADDLE_WITH_TENSORRT
Y
Yan Chunwei 已提交
85
#include "paddle/fluid/inference/tensorrt/convert/op_converter.h"
86
#include "paddle/fluid/inference/tensorrt/helper.h"
87
#include "paddle/fluid/inference/tensorrt/trt_int8_calibrator.h"
Y
Yan Chunwei 已提交
88 89
#endif

90 91 92 93
#ifdef PADDLE_WITH_IPU
#include "paddle/fluid/platform/device/ipu/paddle_ipu_handler.h"
#endif

94 95
namespace paddle {

N
nhzlx 已提交
96
using inference::Singleton;
97
#ifdef PADDLE_WITH_TENSORRT
N
nhzlx 已提交
98 99
using inference::tensorrt::TRTCalibratorEngine;
using inference::tensorrt::TRTCalibratorEngineManager;
100
using inference::tensorrt::TRTInt8Calibrator;
N
nhzlx 已提交
101
#endif
102

103 104
int AnalysisPredictor::clone_num_ = 1;

105 106 107 108
namespace {
bool IsPersistable(const framework::VarDesc *var) {
  if (var->Persistable() &&
      var->GetType() != framework::proto::VarType::FEED_MINIBATCH &&
109 110
      var->GetType() != framework::proto::VarType::FETCH_LIST &&
      var->GetType() != framework::proto::VarType::RAW) {
111 112 113 114
    return true;
  }
  return false;
}
115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133

phi::DataType ConvertPrecision(AnalysisConfig::Precision precision) {
  switch (precision) {
    case AnalysisConfig::Precision::kFloat32:
      return phi::DataType::FLOAT32;
    case AnalysisConfig::Precision::kHalf:
      return phi::DataType::FLOAT16;
    case AnalysisConfig::Precision::kBf16:
      return phi::DataType::BFLOAT16;
    case AnalysisConfig::Precision::kInt8:
      return phi::DataType::INT8;
    default:
      PADDLE_THROW(paddle::platform::errors::InvalidArgument(
          "Paddle Inference not support precision. We now only support "
          "Float32, Half, Bfloat16 and Int8"));
      return phi::DataType::FLOAT32;
  }
}

134
phi::Backend ConvertBackend(paddle_infer::PlaceType backend) {
135
  switch (backend) {
136
    case paddle_infer::PlaceType::kGPU:
137 138
      // NOTE: phi also support phi::Backend::GPUDNN.
      return phi::Backend::GPU;
139
    case paddle_infer::PlaceType::kNPU:
140
      return phi::Backend::NPU;
141
    case paddle_infer::PlaceType::kXPU:
142
      return phi::Backend::XPU;
143
    case paddle_infer::PlaceType::kCPU:
144
      return phi::Backend::CPU;
145 146
    case paddle_infer::PlaceType::kIPU:
      return phi::Backend::IPU;
147 148 149 150 151 152 153
    default:
      PADDLE_THROW(paddle::platform::errors::InvalidArgument(
          "Paddle Inference not support backend, we now only support GPU, XPU, "
          "NPU and CPU."));
      return phi::Backend::CPU;
  }
}
154 155
}  // namespace

C
ccrrong 已提交
156
bool PaddleTensorToLoDTensor(const PaddleTensor &pt,
157
                             phi::DenseTensor *t,
158
                             const platform::Place &place) {
159
  framework::DDim ddim = phi::make_ddim(pt.shape);
160 161 162 163 164 165 166
  void *input_ptr;
  if (pt.dtype == PaddleDType::INT64) {
    input_ptr = t->mutable_data<int64_t>(ddim, place);
  } else if (pt.dtype == PaddleDType::FLOAT32) {
    input_ptr = t->mutable_data<float>(ddim, place);
  } else if (pt.dtype == PaddleDType::INT32) {
    input_ptr = t->mutable_data<int32_t>(ddim, place);
167 168
  } else if (pt.dtype == PaddleDType::FLOAT16) {
    input_ptr = t->mutable_data<float16>(ddim, place);
169 170 171 172
  } else {
    LOG(ERROR) << "unsupported feed type " << pt.dtype;
    return false;
  }
173 174 175
  // NOTE(Aurelius84): Some kernels support zero shape input
  // without memory holder, we should skip enforce logic.
  bool has_zero_dim = (phi::product(ddim) == 0);
176 177 178
  VLOG(3) << "Found zero dim: " << has_zero_dim
          << " from input with ddim: " << ddim;
  if (!has_zero_dim) {
179 180 181 182 183 184 185 186
    PADDLE_ENFORCE_NOT_NULL(
        input_ptr,
        paddle::platform::errors::Fatal(
            "Cannot convert to LoDTensor because LoDTensor creation failed."));
    PADDLE_ENFORCE_NOT_NULL(
        pt.data.data(),
        paddle::platform::errors::InvalidArgument(
            "The data contained in the input PaddleTensor is illegal."));
187 188 189 190 191
    PADDLE_ENFORCE_EQ(
        pt.data.length(),
        t->numel() * paddle::experimental::SizeOf(t->dtype()),
        paddle::platform::errors::InvalidArgument(
            "The data contained in the input PaddleTensor had wrong length."));
192
  }
193 194 195

  if (platform::is_cpu_place(place)) {
    // TODO(panyx0718): Init LoDTensor from existing memcpy to save a copy.
196 197 198 199
    if (input_ptr != nullptr) {
      std::memcpy(
          static_cast<void *>(input_ptr), pt.data.data(), pt.data.length());
    }
J
jianghaicheng 已提交
200 201
  } else if (platform::is_ipu_place(place)) {
#ifdef PADDLE_WITH_IPU
C
ccrrong 已提交
202 203
    std::memcpy(
        static_cast<void *>(input_ptr), pt.data.data(), pt.data.length());
J
jianghaicheng 已提交
204 205 206 207
#else
    PADDLE_THROW(paddle::platform::errors::Fatal(
        "Not compile with WITH_IPU, should not reach here."));
#endif
208
  } else if (platform::is_gpu_place(place)) {
C
ccrrong 已提交
209 210
    PADDLE_ENFORCE_EQ(platform::is_xpu_place(place),
                      false,
211 212
                      platform::errors::InvalidArgument(
                          "Only one choice can be made between CPU and XPU."));
213
#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
214
    platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance();
L
Leo Chen 已提交
215
    auto *dev_ctx = static_cast<const phi::GPUContext *>(pool.Get(place));
216
    auto dst_gpu_place = place;
C
ccrrong 已提交
217 218 219 220 221
    memory::Copy(dst_gpu_place,
                 static_cast<void *>(input_ptr),
                 platform::CPUPlace(),
                 pt.data.data(),
                 pt.data.length(),
222 223 224 225 226
                 dev_ctx->stream());
#else
    PADDLE_THROW(paddle::platform::errors::Fatal(
        "Not compile with CUDA, should not reach here."));
#endif
227 228
  } else if (platform::is_xpu_place(place)) {
#ifdef PADDLE_WITH_XPU
229
    auto dst_xpu_place = place;
C
ccrrong 已提交
230 231 232 233 234
    memory::Copy(dst_xpu_place,
                 static_cast<void *>(input_ptr),
                 platform::CPUPlace(),
                 pt.data.data(),
                 pt.data.length());
235 236 237 238 239 240 241
#else
    PADDLE_THROW(paddle::platform::errors::Fatal(
        "Not compile with XPU, should not reach here."));
#endif
  } else {
    PADDLE_THROW(paddle::platform::errors::InvalidArgument(
        "The analysis predictor supports CPU, GPU and XPU now."));
242 243 244 245 246 247 248 249 250 251
  }
  // TODO(Superjomn) Low performance, need optimization for heavy LoD copy.
  framework::LoD lod;
  for (auto &level : pt.lod) {
    lod.emplace_back(level);
  }
  t->set_lod(lod);
  return true;
}

Y
Yan Chunwei 已提交
252
bool AnalysisPredictor::Init(
253 254
    const std::shared_ptr<framework::Scope> &parent_scope,
    const std::shared_ptr<framework::ProgramDesc> &program) {
M
minqiyang 已提交
255
  VLOG(3) << "Predictor::init()";
256 257
  if (config_.with_profile_) {
    LOG(WARNING) << "Profiler is activated, which might affect the performance";
258 259
    auto tracking_device = config_.use_gpu() ? platform::ProfilerState::kAll
                                             : platform::ProfilerState::kCPU;
T
tensor-tang 已提交
260
    platform::EnableProfiler(tracking_device);
261
  } else {
262 263
    VLOG(2) << "Profiler is deactivated, and no profiling report will be "
               "generated.";
T
tensor-tang 已提交
264 265
  }

266 267 268 269
  if (!status_is_cloned_) {
    root_predictor_id_ = predictor_id_;
  }

270
  // no matter with or without MKLDNN
L
luotao1 已提交
271
  paddle::platform::SetNumThreads(config_.cpu_math_library_num_threads());
272

273 274 275
  if (!PrepareScope(parent_scope)) {
    return false;
  }
276 277 278

  InitPlace();

279 280 281 282 283 284 285
  if (!CreateExecutor()) {
    return false;
  }
  if (!PrepareProgram(program)) {
    return false;
  }

286 287 288
  // Get the feed_target_names and fetch_target_names
  PrepareFeedFetch();

289 290 291
  // Prepare executor, create local variables.
  if (!PrepareExecutor()) {
    return true;
Y
Yan Chunwei 已提交
292
  }
293

294 295 296 297 298 299 300 301 302 303 304 305 306
#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
  // TODO(inference): Now only gpu with external stream support private
  // device_context.
  if (config_.use_gpu_ && config_.use_external_stream_) {
    private_context_ = true;
  }
  if (private_context_) {
    if (!status_is_cloned_) {
      predictor_stream_ = config_.GetExecStream();
    }
    // NOTE: If the external_stream equals to global_device_contexts's stream,
    // then fallback.
    auto global_stream =
L
Leo Chen 已提交
307
        static_cast<phi::GPUContext *>(
308 309 310 311 312 313
            platform::DeviceContextPool::Instance().Get(place_))
            ->stream();
    if (predictor_stream_ != global_stream) {
      InitResourceManager(predictor_stream_);
      InitDeviceContexts();
    }
Y
Yan Chunwei 已提交
314
  }
315
#endif
316
  inference::DisplayMemoryInfo(place_, "Init predictor");
317 318
  return true;
}
319

320
void AnalysisPredictor::InitPlace() {
321
  if (config_.use_gpu()) {
C
ccrrong 已提交
322 323
    PADDLE_ENFORCE_EQ(config_.use_xpu(),
                      false,
324 325
                      platform::errors::InvalidArgument(
                          "Only one choice can be made between CPU and XPU."));
326
    place_ = paddle::platform::CUDAPlace(config_.gpu_device_id());
327
#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
328
    if (config_.thread_local_stream_enabled()) {
W
Wilber 已提交
329 330
      LOG_FIRST_N(WARNING, 1) << "We will remove this interface in the future. "
                                 "Please use config.SetExecStream instead.";
331 332
    }
#endif
333
  } else if (config_.use_xpu()) {
334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356
    if (config_.lite_engine_enabled()) {
#ifdef LITE_SUBGRAPH_WITH_XPU
      // Currently, Paddle-Lite's XPU user interface only supports the transfer
      // of Host data pointers. If it is currently used as a subgraph, execution
      // efficiency will be sacrificed, so it is temporarily set to cpu place.
      // And, the current lite engine of xpu must execute all parts of the
      // model.
      place_ = paddle::platform::CPUPlace();
#else
      PADDLE_THROW(platform::errors::Unavailable(
          "You tried to use an XPU lite engine, but Paddle was not compiled "
          "with it."));
#endif  // LITE_SUBGRAPH_WITH_XPU
    } else {
#ifdef PADDLE_WITH_XPU
      place_ = paddle::platform::XPUPlace(config_.xpu_device_id());
#else
      PADDLE_THROW(platform::errors::Unavailable(
          "You tried to use XPU forward propagation (inference without lite "
          "engine), but Paddle was not compiled "
          "with WITH_XPU."));
#endif  // PADDLE_WITH_XPU
    }
W
Wilber 已提交
357 358 359 360 361 362 363 364
  } else if (config_.use_npu()) {
#ifdef PADDLE_WITH_ASCEND_CL
    place_ = paddle::platform::NPUPlace(config_.npu_device_id());
#else
    PADDLE_THROW(platform::errors::Unavailable(
        "You tried to use NPU forward propagation, but Paddle was not compiled "
        "with WITH_ASCEND_CL."));
#endif
365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380
  } else if (config_.NNAdapter().use_nnadapter) {
    if (config_.lite_engine_enabled()) {
      place_ = paddle::platform::CPUPlace();
#ifndef LITE_SUBGRAPH_WITH_NNADAPTER
      PADDLE_THROW(
          platform::errors::Unavailable("You tried to use an NNAdapter lite "
                                        "engine, but Paddle was not compiled "
                                        "with it."));
#endif  // LITE_SUBGRAPH_WITH_NNADAPTER
    } else {
      PADDLE_THROW(
          platform::errors::Unavailable("You tried to use NNadapter forward "
                                        "propagation (inference without lite "
                                        "engine), but Paddle was not compiled "
                                        "with LITE_WITH_NNADAPTER."));
    }
J
jianghaicheng 已提交
381 382 383 384 385 386 387
  } else if (config_.use_ipu()) {
#ifdef PADDLE_WITH_IPU
    place_ = paddle::platform::IPUPlace();
#else
    PADDLE_THROW(platform::errors::Unavailable(
        "You tried to use IPU forward propagation, but Paddle was not compiled "
        "with WITH_IPU."));
388 389 390 391 392 393 394 395 396
#endif
  } else if (config_.use_custom_device()) {
#ifdef PADDLE_WITH_CUSTOM_DEVICE
    place_ = paddle::platform::CustomPlace(config_.custom_device_type());
#else
    PADDLE_THROW(platform::errors::Unavailable(
        "You tried to use CustomDevice forward propagation, but Paddle was not "
        "compiled "
        "with WITH_CUSTOM_DEVICE."));
J
jianghaicheng 已提交
397
#endif
398 399 400
  } else {
    place_ = paddle::platform::CPUPlace();
  }
401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417
}

void AnalysisPredictor::InitResourceManager(void *stream) {
#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
  predictor_stream_ =
      ResourceManager::Instance().InitGPUResource(place_, stream);
#endif
}

void AnalysisPredictor::InitDeviceContexts() {
// Init GPUContext.
#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
  if (place_.GetType() == phi::AllocationType::GPU) {
    device_contexts_.emplace(
        place_, std::async(std::launch::deferred, [=] {
          auto *gpu_resource =
              ResourceManager::Instance().GetGPUResource(predictor_stream_);
W
Wilber 已提交
418
          auto *gpu_context = new InferGPUContext(place_);
419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434
          gpu_context->SetAllocator(
              memory::allocation::AllocatorFacade::Instance()
                  .GetAllocator(place_, gpu_resource->GetStream())
                  .get());
          gpu_context->SetPinnedAllocator(
              memory::allocation::AllocatorFacade::Instance()
                  .GetAllocator(paddle::platform::CUDAPinnedPlace())
                  .get());
          gpu_context->SetHostAllocator(
              memory::allocation::AllocatorFacade::Instance()
                  .GetAllocator(platform::CPUPlace())
                  .get());
          gpu_context->SetZeroAllocator(
              memory::allocation::AllocatorFacade::Instance()
                  .GetZeroAllocator(place_)
                  .get());
435 436 437 438
          gpu_context->SetHostZeroAllocator(
              memory::allocation::AllocatorFacade::Instance()
                  .GetZeroAllocator(platform::CPUPlace())
                  .get());
439 440 441 442 443
          gpu_context->SetGenerator(
              framework::DefaultCUDAGenerator(place_.GetDeviceId()).get());
          gpu_context->SetHostGenerator(framework::DefaultCPUGenerator().get());

          gpu_context->SetStream(gpu_resource->GetStream());
444
          gpu_context->SetBlasHandle(gpu_resource->GetBlasHandleCreator());
445
          gpu_context->SetBlasTensorCoreHandle(
446 447 448 449 450 451 452 453
              gpu_resource->GetBlasTensorCoreHandleCreator());
          gpu_context->SetBlasTF32Handle(
              gpu_resource->GetBlasTF32TensorCoreHandleCreator());
          gpu_context->SetDnnHandle(gpu_resource->GetDnnHandleCreator());
          gpu_context->SetSolverHandle(
              gpu_resource->GetSolverDnHandleCreator());
          gpu_context->SetSparseHandle(gpu_resource->GetSparseHandleCreator());
          gpu_context->SetEigenDevice(gpu_resource->GetGpuEigenDeviceCreator());
454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547
          gpu_context->SetComputeCapability(
              gpu_resource->GetGpuComputeCapability());
          gpu_context->SetMaxThreadsPerBlock(
              gpu_resource->GetGpuMaxThreadsPerBlock());
          gpu_context->SetMaxThreadsPerMultiProcessor(
              gpu_resource->GetGpuMaxThreadsPerMp());
          gpu_context->SetMaxGridDimSize(gpu_resource->GetGpuMaxGridDimSize());
          gpu_context->SetMultiProcessors(
              gpu_resource->GetGPUMultiProcessors());
          gpu_context->SetDriverVersion(gpu_resource->GetGpuDriverVersion());
          gpu_context->SetRuntimeVersion(gpu_resource->GetGpuRuntimeVersion());
          VLOG(1) << "thread id is " << std::this_thread::get_id()
                  << ", stream id is "
                  << reinterpret_cast<void *>(gpu_resource->GetStream())
                  << ", allotor ptr is "
                  << reinterpret_cast<void *>(
                         memory::allocation::AllocatorFacade::Instance()
                             .GetAllocator(place_, gpu_resource->GetStream())
                             .get());
          return std::unique_ptr<phi::DeviceContext>(gpu_context);
        }));
  }
#endif
  // TODO(Inference): Support other backends.
}

void *AnalysisPredictor::GetExecStream() const {
#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
  if (place_.GetType() == phi::AllocationType::GPU) {
    if (private_context_) {
      return predictor_stream_;
    } else {
      paddle::platform::DeviceContextPool &pool =
          paddle::platform::DeviceContextPool::Instance();
      return reinterpret_cast<const phi::GPUContext *>(pool.Get(place_))
          ->stream();
    }
  } else {
    return nullptr;
  }
  return nullptr;
#else
  // TODO(inference): Support other backends.
  return nullptr;
#endif
}

const void *AnalysisPredictor::GetDeviceContexts() const {
  if (private_context_) {
    return &device_contexts_;
  } else {
    paddle::platform::DeviceContextPool &pool =
        paddle::platform::DeviceContextPool::Instance();
    const auto &dev_ctxs = pool.device_contexts();
    return &dev_ctxs;
  }
}

bool AnalysisPredictor::PrepareScope(
    const std::shared_ptr<framework::Scope> &parent_scope) {
  if (parent_scope) {
    PADDLE_ENFORCE_NOT_NULL(
        parent_scope,
        platform::errors::PreconditionNotMet(
            "Both program and parent_scope should be set in Clone mode."));
    scope_ = parent_scope;
    status_is_cloned_ = true;
  } else {
    paddle::framework::InitDevices();
    paddle::framework::InitDefaultKernelSignatureMap();
    // TODO(wilber): we need to release memory occupied by weights.
    scope_.reset(new paddle::framework::Scope());
    status_is_cloned_ = false;
  }
  sub_scope_ = &scope_->NewScope();
  return true;
}

bool AnalysisPredictor::PrepareProgram(
    const std::shared_ptr<framework::ProgramDesc> &program) {
  if (!program) {
    if (!LoadProgramDesc()) return false;
    // If not cloned, the parameters should be loaded.
    // If config_.ir_optim() is True, parameters is loaded in
    // OptimizeInferenceProgram(), but other persistable variables
    // (like RAW type var) are not created in scope.
    // If config_.ir_optim() is False, parameters is loaded in LoadParameters(),
    // still need to create other persistable variables.
    // So in both case, create persistable variables at first.
    executor_->CreateVariables(*inference_program_, 0, true, sub_scope_);

    // if enable_ir_optim_ is false,
    // the analysis pass(op fuse, graph analysis, trt subgraph, mkldnn etc) will
    // not be executed.
548 549
    model_precision_ =
        paddle::inference::GetModelPrecision(*inference_program_);
550 551 552 553 554
    OptimizeInferenceProgram();
  } else {
    // If the program is passed from external, no need to optimize it, this
    // logic is used in the clone scenario.
    inference_program_ = program;
555 556 557 558 559
    if (config_.apply_optim_) {
      VLOG(3)
          << "apply_optim is enabled, will call OptimizeInferenceProgram().";
      OptimizeInferenceProgram();
    }
560 561 562 563 564 565 566 567
  }

  executor_->CreateVariables(*inference_program_, 0, false, sub_scope_);

  return true;
}

bool AnalysisPredictor::CreateExecutor() {
568 569 570
  executor_.reset(new paddle::framework::NaiveExecutor(place_));
  return true;
}
W
wenbin 已提交
571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589

static bool IsPrepareDataOptTargetOp(framework::OpDesc *op) {
  // here is prepare data optimization related bad cases:
  // let's assume an op behind conditional_block and if conditional_block
  // chooses branch 1, the op need to call prepare data. else the op don't need
  // to call prepare data. In running, if predictor chooses branch 2, then
  // optimization takes effect, later issue is followed if predictor chooses
  // branch 1, because the op lost chance to prepare data.
  std::vector<std::string> op_type = {"conditional_block_infer",
                                      "select_input"};
  for (const auto &type : op_type) {
    if (op->Type() == type) {
      return true;
    }
  }
  return false;
}

static void DisablePrepareDataOpt(
C
ccrrong 已提交
590 591
    std::shared_ptr<framework::ProgramDesc> inference_program,
    int block,
W
wenbin 已提交
592 593 594 595 596 597 598 599 600
    bool pre_disable_opt) {
  bool disable_opt = false;
  auto &infer_block = inference_program->Block(block);
  for (auto *op : infer_block.AllOps()) {
    if (disable_opt || pre_disable_opt) {
      op->SetAttr("inference_force_prepare_data", true);
    }
    if (op->HasAttr("sub_block")) {
      int blockID = op->GetBlockAttrId("sub_block");
C
ccrrong 已提交
601 602
      DisablePrepareDataOpt(
          inference_program, blockID, disable_opt || pre_disable_opt);
W
wenbin 已提交
603 604
    }
    // disable prepare data if unfriendly op is found
W
wenbin 已提交
605 606 607
    if (!disable_opt) {
      disable_opt = IsPrepareDataOptTargetOp(op);
    }
W
wenbin 已提交
608 609 610
  }
}

611
bool AnalysisPredictor::PrepareExecutor() {
612
#if defined(PADDLE_WITH_DISTRIBUTE) && defined(PADDLE_WITH_PSCORE)
613 614 615 616 617
  if (config_.dist_config().use_dist_model()) {
    VLOG(3) << "use_dist_model is enabled, will init FleetExecutor.";
    return PrepareFleetExecutor();
  }
#endif
W
wenbin 已提交
618 619
  DisablePrepareDataOpt(inference_program_, 0, false);

C
ccrrong 已提交
620 621
  executor_->Prepare(
      sub_scope_, *inference_program_, 0, config_.use_feed_fetch_ops_);
622

623 624 625 626 627 628 629 630 631
  if (config_.enable_memory_optim_) {
    auto *pass_res_info =
        inference::analysis::PassResultInfoForRuntime::Instance();
    auto reuse_table =
        pass_res_info->Get<std::unordered_map<std::string, std::string>>(
            root_predictor_id_, "memory_optimize_pass");
    executor_->MakeReusePlan(reuse_table);
  }

632 633 634
  PADDLE_ENFORCE_NOT_NULL(sub_scope_,
                          platform::errors::PreconditionNotMet(
                              "The sub_scope should not be nullptr."));
Y
Yan Chunwei 已提交
635

636 637 638
  return true;
}

639
#if defined(PADDLE_WITH_DISTRIBUTE) && defined(PADDLE_WITH_PSCORE)
640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675
bool AnalysisPredictor::PrepareFleetExecutor() {
  VLOG(3) << "AnalysisPredictor::PrepareFleetExecutor()";
  if (config_.dist_config().nranks() > 1 && !CommInit()) {
    return false;
  }
  task_node_.reset(new distributed::TaskNode(inference_program_.get(),
                                             config_.dist_config().rank()));
  // With auto cut, there is no concept of pp, no need to add dependency.
  task_node_->SetType("Compute");
  task_node_->Init(config_.use_feed_fetch_ops_enabled());
  executor_desc_ = distributed::FleetExecutorDesc();
  executor_desc_.set_cur_rank(config_.dist_config().rank());
  std::unordered_map<int64_t, int64_t> id_to_rank;
  for (int i = 0; i < config_.dist_config().nranks(); ++i) {
    distributed::RankInfo *rank_info = executor_desc_.add_cluster_info();
    rank_info->set_rank(i);
    rank_info->set_ip_port(config_.dist_config().trainer_endpoints()[i]);
    id_to_rank.insert({i, i});
  }
  fleet_exe_.reset(new distributed::FleetExecutor(executor_desc_));
  // NOTE: Vars of feed fetch ops are not persistable,
  // which will result in that those vars will be created in
  // the subscope (microscope) in fleet executor. This will
  // cause that the GetInputTensor/GetOutputTensor funct
  // in analysis predictor cannot find those vars in the scope
  // returned by the DistModel, since DistModel only return the
  // root scope. So, those vars must  to be created in the root
  // scope instead of in the microscope
  std::vector<std::string> feed_fetch_vars;
  for (auto pair : idx2feeds_) {
    feed_fetch_vars.emplace_back(pair.second);
  }
  for (auto pair : idx2fetches_) {
    feed_fetch_vars.emplace_back(pair.second);
  }
  fleet_exe_->Init(config_.dist_config().carrier_id(),
C
ccrrong 已提交
676 677 678 679 680 681 682
                   *(inference_program_.get()),
                   scope_.get(),
                   place_,
                   1,
                   {task_node_.get()},
                   id_to_rank,
                   feed_fetch_vars);
683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718
  return true;
}

bool AnalysisPredictor::CommInit() {
  std::map<int64_t, std::vector<int64_t>> ring_id_to_ranks{};
  std::map<int64_t, std::vector<int64_t>> rank_to_ring_ids{};
  if (!LoadConverterConfig(&ring_id_to_ranks, &rank_to_ring_ids)) {
    VLOG(3) << "Load converter config failed, DistModel init failed.";
    return false;
  }
  std::unique_ptr<framework::ProgramDesc> comm_init_program(
      new framework::ProgramDesc());
  framework::BlockDesc *comm_init_block = comm_init_program->MutableBlock(0);
  std::vector<int64_t> &ring_ids =
      rank_to_ring_ids[config_.dist_config().rank()];
  int64_t order = 0;
  std::string var_name_base = "comm_init_";
  for (int64_t ring_id : ring_ids) {
    VLOG(3) << "Init comm for ring id: " << ring_id;
    int64_t ranks_in_group = ring_id_to_ranks[ring_id].size();
    int64_t rank_in_group = 0;
    std::vector<int64_t> &ranks = ring_id_to_ranks[ring_id];
    for (int64_t rank : ranks) {
      if (config_.dist_config().rank() == rank) {
        break;
      }
      rank_in_group += 1;
    }
    std::vector<std::string> peer_endpoints;
    for (int64_t rank : ranks) {
      if (config_.dist_config().rank() == rank) {
        continue;
      }
      peer_endpoints.emplace_back(
          config_.dist_config().trainer_endpoints()[rank]);
    }
C
ccrrong 已提交
719 720 721 722 723 724
    InsertCommOp(var_name_base + std::to_string(order),
                 ranks_in_group,
                 rank_in_group,
                 peer_endpoints,
                 comm_init_block,
                 ring_id);
725 726 727 728 729 730 731 732 733 734 735
    order += 1;
  }
  framework::NaiveExecutor e(place_);
  e.CreateVariables(*comm_init_program, 0, true, scope_.get());
  e.Prepare(scope_.get(), *comm_init_program, 0, false);
  e.Run();
  VLOG(3) << "Comm init successful.";
  return true;
}

void AnalysisPredictor::InsertCommOp(
C
ccrrong 已提交
736 737 738 739 740
    std::string tmp_var_name,
    int nranks,
    int rank,
    const std::vector<std::string> &peer_endpoints,
    framework::BlockDesc *block,
741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796
    int ring_id) {
  /*
   * tmp_var_name: the var name for var comm_id
   * nranks: number of total ranks
   * rank: the rank of local rank in the comm group
   * peer_endpoints: peer's endpoints
   * block: the block where to insert the comm ops
   * ring_id: the ring_id to be inited
   */
  const std::string &endpoint = config_.dist_config().current_endpoint();
  std::stringstream ss;
  ss << "Init comm with tmp var: " << tmp_var_name
     << ". The ring id is: " << ring_id << ". The group has: " << nranks
     << " ranks. Current rank in the group is: " << rank
     << ". The endpoint is: " << endpoint << ". Peer endpoints are: ";
  for (auto ep : peer_endpoints) {
    ss << ep << ", ";
  }
  VLOG(3) << ss.str();
  if (config_.use_gpu()) {
    framework::VarDesc *new_var = block->Var(tmp_var_name);
    new_var->SetType(framework::proto::VarType::RAW);
    new_var->SetPersistable(true);
    framework::OpDesc *gen_nccl_id_op = block->AppendOp();
    gen_nccl_id_op->SetType("c_gen_nccl_id");
    gen_nccl_id_op->SetOutput("Out", {tmp_var_name});
    gen_nccl_id_op->SetAttr("rank", rank);
    gen_nccl_id_op->SetAttr("endpoint",
                            config_.dist_config().current_endpoint());
    gen_nccl_id_op->SetAttr("other_endpoints", peer_endpoints);
    gen_nccl_id_op->SetAttr("ring_id", ring_id);
    gen_nccl_id_op->SetAttr("op_role",
                            static_cast<int>(framework::OpRole::kForward));
    gen_nccl_id_op->CheckAttrs();
    framework::OpDesc *comm_init_op = block->AppendOp();
    comm_init_op->SetType("c_comm_init");
    comm_init_op->SetInput("X", {tmp_var_name});
    comm_init_op->SetAttr("rank", rank);
    comm_init_op->SetAttr("nranks", nranks);
    comm_init_op->SetAttr("ring_id", ring_id);
    comm_init_op->SetAttr("op_role",
                          static_cast<int>(framework::OpRole::kForward));
    comm_init_op->CheckAttrs();
  } else {
    LOG(WARNING) << "DistModelInf doesn't init comm.";
    // TODO(fleet exe dev): comm init for more devices
  }
}

bool AnalysisPredictor::LoadConverterConfig(
    std::map<int64_t, std::vector<int64_t>> *ring_id_to_ranks,
    std::map<int64_t, std::vector<int64_t>> *rank_to_ring_ids) {
  VLOG(3) << "Going to load converter config from: "
          << config_.dist_config().comm_init_config() << "\n";
  std::ifstream fin(config_.dist_config().comm_init_config(), std::ios::in);
  PADDLE_ENFORCE_EQ(
C
ccrrong 已提交
797 798
      static_cast<bool>(fin.is_open()),
      true,
799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870
      platform::errors::NotFound(
          "Cannot open file %s, please confirm whether the file is normal.",
          config_.dist_config().comm_init_config()));
  std::string line;
  bool ring_to_rank{true};
  // Reading config from file, the config file should like these format
  //  [ring_id -> ranks]
  //  0,0,1,2,3
  //  1,0,1
  //  2,2,3
  //  21,0,1
  //  22,1,2
  //  23,2,3
  //  [rank -> ring_ids]
  //  0,0,1,21
  //  1,0,1,21,22
  //  2,0,2,22,23
  //  3,0,2,23
  while (std::getline(fin, line)) {
    std::vector<std::string> one_line = paddle::string::Split(line, ',');
    if (one_line.size() == 1) {
      // start a new section of the config
      if (line == "[ring_id -> ranks]") {
        ring_to_rank = true;
      } else if (line == "[rank -> ring_ids]") {
        ring_to_rank = false;
      }
    } else {
      // parse key - values pairs in one section
      int64_t key = std::stoll(one_line[0]);
      for (size_t i = 1; i < one_line.size(); ++i) {
        int64_t val = std::stoll(one_line[i]);
        if (ring_to_rank) {
          if (ring_id_to_ranks->find(key) == ring_id_to_ranks->end()) {
            ring_id_to_ranks->insert({key, std::vector<int64_t>()});
          }
          ring_id_to_ranks->at(key).emplace_back(val);
        } else {
          if (rank_to_ring_ids->find(key) == rank_to_ring_ids->end()) {
            rank_to_ring_ids->insert({key, std::vector<int64_t>()});
          }
          rank_to_ring_ids->at(key).emplace_back(val);
        }
        // NOTE: add more configuration sections here
      }
    }
  }
  std::stringstream ss;
  ss << "Loaded the following converter config:\n";
  ss << "ring_id_to_ranks:\n";
  for (auto pair : *ring_id_to_ranks) {
    int64_t key = pair.first;
    ss << "\t" << key << "\t->\t";
    for (auto value : pair.second) {
      ss << value << "\t";
    }
    ss << "\n";
  }
  ss << "rank_to_ring_ids:\n";
  for (auto pair : *rank_to_ring_ids) {
    int64_t key = pair.first;
    ss << "\t" << key << "\t->\t";
    for (auto value : pair.second) {
      ss << value << "\t";
    }
    ss << "\n";
  }
  VLOG(3) << ss.str();
  return true;
}
#endif

871 872
void AnalysisPredictor::MkldnnPreSet(const std::vector<PaddleTensor> &inputs) {
#ifdef PADDLE_WITH_MKLDNN
W
Wilber 已提交
873 874 875 876 877 878 879 880 881 882 883 884
  std::vector<std::vector<int>> inputs_shape;
  for (size_t i = 0; i < inputs.size(); ++i) {
    inputs_shape.emplace_back(inputs[i].shape);
  }
  MkldnnPreSet(inputs_shape);
#endif
}

void AnalysisPredictor::MkldnnPreSet(
    const std::vector<std::vector<int>> &inputs_shape) {
#ifdef PADDLE_WITH_MKLDNN
  VLOG(2) << "AnalysisPredictor::ZeroCopyRun get_cur_mkldnn_session_id="
885
          << phi::OneDNNContext::tls().get_cur_mkldnn_session_id();
886 887 888
  // In cache clearing mode.
  if (config_.mkldnn_cache_capacity_ > 0) {
    VLOG(2) << "In mkldnn cache clear mode.";
889 890
    phi::OneDNNContext::tls().set_cur_mkldnn_session_id(
        phi::OneDNNContextThreadLocals::kMKLDNNSessionID_CacheClearing);
891 892
    // Set current_input_shape for caching dynamic shape.
    std::stringstream ss;
W
Wilber 已提交
893 894 895
    for (size_t i = 0; i < inputs_shape.size(); ++i) {
      for (size_t j = 0; j < inputs_shape[i].size(); ++j) {
        ss << inputs_shape[i][j] << "-";
896 897 898
      }
    }
    VLOG(2) << "Set input shape=" << ss.str();
899
    phi::OneDNNContext::tls().set_cur_input_shape_str(ss.str());
900
  }
901
  phi::OneDNNContext::tls().set_cur_input_shape_cache_capacity(
902 903
      config_.mkldnn_cache_capacity_);

904 905 906 907 908 909
#endif
}

void AnalysisPredictor::MkldnnPostReset() {
#ifdef PADDLE_WITH_MKLDNN
  // In cache clearing mode.
910
  if (config_.mkldnn_cache_capacity_ > 0 &&
911
      static_cast<phi::OneDNNContext *>(
912 913
          (&platform::DeviceContextPool::Instance())->Get(platform::CPUPlace()))
              ->GetCachedObjectsNumber() > 0) {
914
    if (VLOG_IS_ON(2)) {
915
      auto shape_blob_size = static_cast<phi::OneDNNContext *>(
916 917 918 919 920 921
                                 (&platform::DeviceContextPool::Instance())
                                     ->Get(platform::CPUPlace()))
                                 ->GetShapeBlobSize();
      CHECK_LE(shape_blob_size,
               static_cast<size_t>(config_.mkldnn_cache_capacity_));
    }
922 923 924
    // We cannot reset to the default cache settings
    // as there maybe CopyToCPU method used and oneDNN
    // primitives are used there so cache would grow
925 926 927 928
  }
#endif
}

929 930 931
bool AnalysisPredictor::Run(const std::vector<PaddleTensor> &inputs,
                            std::vector<PaddleTensor> *output_data,
                            int batch_size) {
932
  paddle::platform::SetNumThreads(config_.cpu_math_library_num_threads());
933 934 935
#ifdef PADDLE_WITH_MKLDNN
  if (config_.use_mkldnn_) MkldnnPreSet(inputs);
#endif
M
minqiyang 已提交
936
  VLOG(3) << "Predictor::predict";
937 938 939 940
  inference::Timer timer;
  timer.tic();
  // set feed variable
  framework::Scope *scope = sub_scope_ ? sub_scope_ : scope_.get();
C
ccrrong 已提交
941 942 943
  PADDLE_ENFORCE_NOT_NULL(
      scope,
      platform::errors::PreconditionNotMet("The scope should not be nullptr."));
944 945
  if (!SetFeed(inputs, scope)) {
    LOG(ERROR) << "fail to set feed";
Y
Yan Chunwei 已提交
946
    return false;
947
  }
M
Michal Gallus 已提交
948

949 950 951 952 953 954 955 956 957
#ifdef PADDLE_WITH_TENSORRT
  if (config_.tensorrt_engine_enabled()) {
    inference::tensorrt::TensorRTEngine::predictor_id_per_thread =
        predictor_id_;
    VLOG(3) << "thread_local var predictor_id in TensorRTEngine is set to: "
            << inference::tensorrt::TensorRTEngine::predictor_id_per_thread;
  }
#endif

958 959 960
  // Run the inference program
  // if share variables, we need not create variables
  executor_->Run();
961

962 963 964 965
  // get fetch variable
  if (!GetFetch(output_data, scope)) {
    LOG(ERROR) << "fail to get fetches";
    return false;
T
tensor-tang 已提交
966
  }
Y
Yan Chunwei 已提交
967

M
minqiyang 已提交
968
  VLOG(3) << "predict cost: " << timer.toc() << "ms";
Y
Yan Chunwei 已提交
969

Y
Yan Chunwei 已提交
970 971 972 973 974
  // All the containers in the scope will be hold in inference, but the
  // operators assume that the container will be reset after each batch.
  // Here is a bugfix, collect all the container variables, and reset then to a
  // bool; the next time, the operator will call MutableData and construct a new
  // container again, so that the container will be empty for each batch.
975 976 977
  if (sub_scope_) {
    tensor_array_batch_cleaner_.CollectNoTensorVars(sub_scope_);
  }
Y
Yan Chunwei 已提交
978
  tensor_array_batch_cleaner_.ResetNoTensorVars();
979 980 981 982

  // recover the cpu_math_library_num_threads to 1, in order to avoid thread
  // conflict when integrating it into deployment service.
  paddle::platform::SetNumThreads(1);
983 984
#ifdef PADDLE_WITH_MKLDNN
  if (config_.use_mkldnn_) MkldnnPostReset();
T
Tao Luo 已提交
985
#endif
986
#if defined(PADDLE_WITH_MKLML)
T
Tao Luo 已提交
987 988 989 990
  // Frees unused memory allocated by the Intel® MKL Memory Allocator to
  // avoid memory leak. See:
  // https://software.intel.com/en-us/mkl-developer-reference-c-mkl-free-buffers
  platform::dynload::MKL_Free_Buffers();
991
#endif
992 993
  return true;
}
994

995 996
bool AnalysisPredictor::SetFeed(const std::vector<PaddleTensor> &inputs,
                                framework::Scope *scope) {
M
minqiyang 已提交
997
  VLOG(3) << "Predictor::set_feed";
998 999 1000 1001 1002 1003 1004 1005 1006 1007
  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) {
1008
    phi::DenseTensor *input = &feed_tensors_[i];
1009
    if (!PaddleTensorToLoDTensor(inputs[i], input, place_)) {
1010 1011 1012
      return false;
    }
    int idx = -1;
1013
    if (config_.specify_input_name_) {
T
tensor-tang 已提交
1014 1015
      auto name = inputs[i].name;
      if (feed_names_.find(name) == feed_names_.end()) {
T
tensor-tang 已提交
1016 1017
        LOG(ERROR) << "feed names from program do not have name: [" << name
                   << "] from specified input";
T
tensor-tang 已提交
1018 1019
      }
      idx = feed_names_[name];
1020
    } else {
R
Ruibiao Chen 已提交
1021
      idx = PADDLE_GET_CONST(int, feeds_[i]->GetAttr("col"));
1022
    }
1023
    framework::SetFeedVariable(scope, *input, "feed", idx);
1024 1025 1026 1027 1028
  }
  return true;
}

template <typename T>
1029
void AnalysisPredictor::GetFetchOne(const phi::DenseTensor &fetch,
1030 1031
                                    PaddleTensor *output) {
  // set shape.
1032
  auto shape = phi::vectorize(fetch.dims());
1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049
  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) {
M
minqiyang 已提交
1050
  VLOG(3) << "Predictor::get_fetch";
Y
Yan Chunwei 已提交
1051 1052
  outputs->resize(fetches_.size());
  for (size_t i = 0; i < fetches_.size(); ++i) {
R
Ruibiao Chen 已提交
1053
    int idx = PADDLE_GET_CONST(int, fetches_[i]->GetAttr("col"));
1054
    PADDLE_ENFORCE_EQ(
C
ccrrong 已提交
1055 1056
        static_cast<size_t>(idx),
        i,
1057
        platform::errors::InvalidArgument(
C
ccrrong 已提交
1058 1059
            "Fetch op's col attr(%d) should be equal to the index(%d)",
            idx,
1060
            i));
1061
    framework::FetchType &fetch_var =
1062
        framework::GetFetchVariable(*scope, "fetch", idx);
1063
    auto &fetch = PADDLE_GET(phi::DenseTensor, fetch_var);
1064
    auto type = framework::TransToProtoVarType(fetch.dtype());
1065
    auto output = &(outputs->at(i));
Y
Yan Chunwei 已提交
1066
    output->name = fetches_[idx]->Input("X")[0];
Y
Yu Yang 已提交
1067
    if (type == framework::proto::VarType::FP32) {
1068 1069
      GetFetchOne<float>(fetch, output);
      output->dtype = PaddleDType::FLOAT32;
Y
Yu Yang 已提交
1070
    } else if (type == framework::proto::VarType::INT64) {
1071 1072
      GetFetchOne<int64_t>(fetch, output);
      output->dtype = PaddleDType::INT64;
1073 1074 1075
    } else if (type == framework::proto::VarType::INT32) {
      GetFetchOne<int32_t>(fetch, output);
      output->dtype = PaddleDType::INT32;
1076 1077 1078
    } else if (type == framework::proto::VarType::FP16) {
      GetFetchOne<float16>(fetch, output);
      output->dtype = PaddleDType::FLOAT16;
1079
    } else {
1080 1081
      LOG(ERROR) << "unknown type, only support float32, float16, int64 and "
                    "int32 now.";
1082 1083
    }
  }
Y
Yan Chunwei 已提交
1084 1085
  return true;
}
1086

1087
void AnalysisPredictor::PrepareArgument() {
1088
  argument_.SetUseGPU(config_.use_gpu());
1089
  argument_.SetUseFcPadding(config_.use_fc_padding());
1090
  argument_.SetGPUDeviceId(config_.gpu_device_id());
1091
  argument_.SetEnableIrOptim(config_.enable_ir_optim_);
1092
  argument_.SetEnableMemoryOptim(config_.enable_memory_optim());
T
Tao Luo 已提交
1093
  argument_.SetModelFromMemory(config_.model_from_memory_);
Y
Yan Chunwei 已提交
1094
  // Analyze inference_program
1095
  argument_.SetPredictorID(predictor_id_);
1096
  argument_.SetRootPredictorID(root_predictor_id_);
1097
  argument_.SetOptimCacheDir(config_.opt_cache_dir_);
1098 1099
  if (!config_.model_dir().empty()) {
    argument_.SetModelDir(config_.model_dir());
T
Tao Luo 已提交
1100
  } else {
C
ccrrong 已提交
1101 1102
    PADDLE_ENFORCE_EQ(config_.prog_file().empty(),
                      false,
1103 1104
                      platform::errors::PreconditionNotMet(
                          "Either model_dir or prog_file should be set."));
N
nhzlx 已提交
1105

1106 1107
    argument_.SetModelProgramPath(config_.prog_file());
    argument_.SetModelParamsPath(config_.params_file());
Y
Yan Chunwei 已提交
1108
  }
1109 1110
  // For JITLayer
  argument_.SetSkipLoadParams(config_.skip_load_params_);
1111

1112
  argument_.SetTensorRtPrecisionMode(config_.tensorrt_precision_mode_);
1113
  argument_.SetTensorRtUseOSS(config_.trt_use_varseqlen_);
1114
  argument_.SetTensorRtWithInterleaved(config_.trt_with_interleaved_);
1115 1116
  argument_.SetTensorRtTransformerPosid(config_.tensorrt_transformer_posid_);
  argument_.SetTensorRtTransformerMaskid(config_.tensorrt_transformer_maskid_);
1117 1118 1119 1120 1121
  argument_.SetMinInputShape(config_.min_input_shape_);
  argument_.SetMaxInputShape(config_.max_input_shape_);
  argument_.SetOptimInputShape(config_.optim_input_shape_);
  argument_.SetTensorRtTunedDynamicShape(
      config_.tuned_tensorrt_dynamic_shape());
1122
  if (config_.use_gpu() && config_.tensorrt_engine_enabled()) {
Y
Yan Chunwei 已提交
1123
    LOG(INFO) << "TensorRT subgraph engine is enabled";
1124 1125 1126
    argument_.SetUseTensorRT(true);
    argument_.SetTensorRtWorkspaceSize(config_.tensorrt_workspace_size_);
    argument_.SetTensorRtMaxBatchSize(config_.tensorrt_max_batchsize_);
1127
    argument_.SetTensorRtMinSubgraphSize(config_.tensorrt_min_subgraph_size_);
1128
    argument_.SetTensorRtDisabledOPs(config_.trt_disabled_ops_);
1129 1130
    argument_.SetTensorRtUseDLA(config_.trt_use_dla_);
    argument_.SetTensorRtDLACore(config_.trt_dla_core_);
N
nhzlx 已提交
1131
    argument_.SetTensorRtUseStaticEngine(config_.trt_use_static_engine_);
1132
    argument_.SetTensorRtUseCalibMode(config_.trt_use_calib_mode_);
1133
    argument_.SetCloseTrtPluginFp16(config_.disable_trt_plugin_fp16_);
1134 1135 1136
    argument_.SetTensorRtShapeRangeInfoPath(config_.shape_range_info_path());
    argument_.SetTensorRtAllowBuildAtRuntime(
        config_.trt_allow_build_at_runtime());
1137
    argument_.SetTensorRtUseInspector(config_.trt_use_inspector_);
1138
    argument_.SetTrtEngineMemorySharing(config_.trt_engine_memory_sharing());
W
Wojciech Uss 已提交
1139
  }
1140

D
denglin-github 已提交
1141 1142 1143 1144
  if (config_.dlnne_enabled()) {
    LOG(INFO) << "Dlnne subgraph is enabled";
    argument_.SetUseDlnne(true);
    argument_.SetDlnneMinSubgraphSize(config_.dlnne_min_subgraph_size_);
D
denglin-github 已提交
1145 1146 1147 1148 1149 1150 1151 1152
    argument_.SetDlnneMaxBatchSize(config_.dlnne_max_batchsize_);
    argument_.SetDlnneUseStaticBatch(config_.dlnne_use_static_batch_);
    argument_.SetDlnneWeightShareMode(config_.dlnne_weight_share_mode_);
    argument_.SetDlnneDisableNodesByOutputs(
        config_.dlnne_disable_nodes_by_outputs_);
    argument_.SetDlnneInputShapeDict(config_.dlnne_input_shape_dict_);
    argument_.SetDlnneUseCalibMode(config_.dlnne_use_calib_mode_);
    argument_.SetDlnnePrecisionMode(config_.dlnne_precision_mode_);
D
denglin-github 已提交
1153 1154
  }

石晓伟 已提交
1155
  if (config_.lite_engine_enabled()) {
W
Wilber 已提交
1156 1157
    argument_.SetCpuMathLibraryNumThreads(
        config_.cpu_math_library_num_threads());
石晓伟 已提交
1158 1159 1160
    argument_.SetLitePrecisionMode(config_.lite_precision_mode_);
    argument_.SetLitePassesFilter(config_.lite_passes_filter_);
    argument_.SetLiteOpsFilter(config_.lite_ops_filter_);
1161 1162 1163
    argument_.SetLiteZeroCopy(config_.lite_zero_copy_);
    argument_.SetUseXpu(config_.use_xpu_);
    argument_.SetXpuL3WorkspaceSize(config_.xpu_l3_workspace_size_);
W
Wilber 已提交
1164 1165 1166 1167 1168
    argument_.SetXpuLocked(config_.xpu_locked_);
    argument_.SetXpuAutotune(config_.xpu_autotune_);
    argument_.SetXpuAutotuneFile(config_.xpu_autotune_file_);
    argument_.SetXpuPrecision(config_.xpu_precision_);
    argument_.SetXpuAdaptiveSeqlen(config_.xpu_adaptive_seqlen_);
1169
    argument_.SetXpuDeviceId(config_.xpu_device_id_);
1170
    argument_.SetXpuEnableMultiStream(config_.xpu_enable_multi_stream_);
1171
    argument_.SetUseOpenCL(config_.use_opencl_);
1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191
    // NNAdapter related
    argument_.SetUseNNAdapter(config_.NNAdapter().use_nnadapter);
    argument_.SetNNAdapterDeviceNames(
        config_.NNAdapter().nnadapter_device_names);
    argument_.SetNNAdapterContextProperties(
        config_.NNAdapter().nnadapter_context_properties);
    argument_.SetNNAdapterModelCacheDir(
        config_.NNAdapter().nnadapter_model_cache_dir);
    argument_.SetNNAdapterSubgraphPartitionConfigBuffer(
        config_.NNAdapter().nnadapter_subgraph_partition_config_buffer);
    argument_.SetNNAdapterSubgraphPartitionConfigPath(
        config_.NNAdapter().nnadapter_subgraph_partition_config_path);
    std::vector<std::string> buffer_keys;
    std::vector<std::vector<char>> buffer_vals;
    for (auto it : config_.NNAdapter().nnadapter_model_cache_buffers) {
      buffer_keys.emplace_back(it.first);
      buffer_vals.emplace_back(it.second);
    }
    argument_.SetNNAdapterModelCacheToken(buffer_keys);
    argument_.SetNNAdapterModelCacheBuffer(buffer_vals);
石晓伟 已提交
1192 1193 1194
    LOG(INFO) << "Lite subgraph engine is enabled";
  }

1195
#ifdef PADDLE_WITH_IPU
J
jianghaicheng 已提交
1196 1197
  argument_.SetUseIpu(config_.use_ipu_);
  argument_.SetIpuDeviceNum(config_.ipu_device_num());
1198
  argument_.SetIpuMicroBatchSize(config_.ipu_micro_batch_size_);
J
jianghaicheng 已提交
1199 1200
  argument_.SetIpuEnablePipelining(config_.ipu_enable_pipelining_);
  argument_.SetIpuBatchesPerStep(config_.ipu_batches_per_step_);
1201 1202 1203 1204 1205
  argument_.SetIpuEnableFp16(config_.ipu_enable_fp16_);
  argument_.SetIpuReplicaNum(config_.ipu_replica_num_);
  argument_.SetIpuAvailableMemoryProportion(
      config_.ipu_available_memory_proportion_);
  argument_.SetIpuEnableHalfPartial(config_.ipu_enable_half_partial_);
1206 1207
  argument_.SetIpuEnableModelRuntimeExecutor(
      config_.ipu_enable_model_runtime_executor_);
1208 1209
  argument_.SetIpuCustomOpsInfo(config_.ipu_custom_ops_info_);
  argument_.SetIpuCustomPatterns(config_.ipu_custom_patterns_);
1210
#endif
J
jianghaicheng 已提交
1211

1212 1213 1214
  argument_.SetUseNpu(config_.use_npu_);
  argument_.SetNPUDeviceId(config_.npu_device_id());

1215
  if (config_.use_mkldnn_) {
Y
Yan Chunwei 已提交
1216
    LOG(INFO) << "MKLDNN is enabled";
1217 1218 1219
    argument_.SetMKLDNNEnabledOpTypes(config_.mkldnn_enabled_op_types_);
  }

1220 1221 1222 1223 1224 1225 1226 1227
#ifdef PADDLE_WITH_MKLDNN
  if (config_.mkldnn_quantizer_enabled()) {
    LOG(INFO) << "Quantization is enabled";
    argument_.SetQuantizeEnabledOpTypes(
        config_.mkldnn_quantizer_config()->enabled_op_types());
    argument_.SetQuantizeExcludedOpIds(
        config_.mkldnn_quantizer_config()->excluded_op_ids());
  }
1228 1229 1230 1231
  if (config_.use_mkldnn_bfloat16_) {
    LOG(INFO) << "Bfloat16 is enabled";
    argument_.SetBfloat16EnabledOpTypes(config_.bfloat16_enabled_op_types_);
  }
B
baoachun 已提交
1232 1233 1234 1235 1236 1237 1238

  if (config_.use_mkldnn_int8_) {
    LOG(INFO) << "Int8 is enabled";
    argument_.SetQuantizeEnabledOpTypes(config_.quantize_enabled_op_types_);
    argument_.SetQuantizeExcludedOpIds(config_.quantize_excluded_op_ids_);
    argument_.SetQuantVarScales({});
  }
1239 1240
#endif

1241
  auto *pass_builder = config_.pass_builder();
1242 1243 1244 1245
  if (model_precision_ != phi::DataType::FLOAT32) {
    LOG(INFO) << "Model is mixed precision type with " << model_precision_
              << ", we will use a new PassStrategy. Note that only the GPU "
                 "backend is supported for now.";
1246 1247
    pass_builder->ClearPasses();
    const auto &deleted_passes = pass_builder->GetAllDeletedPasses();
1248 1249
    if (config_.tensorrt_engine_enabled()) {
      for (const auto &pass : kTrtLowerPrecisionPasses) {
1250 1251
        if (deleted_passes.count(pass)) continue;
        pass_builder->AppendPass(pass);
1252 1253 1254
      }
    } else if (config_.use_gpu()) {
      for (const auto &pass : kGpuLowerPrecisionPasses) {
1255 1256
        if (deleted_passes.count(pass)) continue;
        pass_builder->AppendPass(pass);
1257 1258 1259
      }
    }
  }
1260 1261 1262
  if (config_.ir_debug_) {
    pass_builder->TurnOnDebug();
  }
Y
Yan Chunwei 已提交
1263
  if (!config_.ir_optim()) {
1264
    argument_.SetEnableIrOptim(false);
Y
Yan Chunwei 已提交
1265 1266
    LOG(INFO) << "ir_optim is turned off, no IR pass will be executed";
  }
1267
  argument_.SetDisableLogs(config_.glog_info_disabled());
1268 1269
  argument_.SetIrAnalysisPasses(pass_builder->AllPasses());
  argument_.SetAnalysisPasses(pass_builder->AnalysisPasses());
1270
  argument_.SetScopeNotOwned(scope_.get());
1271

1272
  // mixed precison.
1273
  argument_.SetModelPrecision(static_cast<int>(model_precision_));
1274
  argument_.SetMixedBlackList(config_.mixed_black_list_);
1275 1276 1277 1278 1279
}

// NOTE All the members in AnalysisConfig should be copied to Argument.
void AnalysisPredictor::OptimizeInferenceProgram() {
  PrepareArgument();
1280 1281 1282 1283 1284 1285 1286 1287 1288 1289

#ifdef PADDLE_WITH_TENSORRT
  if (config_.tensorrt_engine_enabled()) {
    inference::tensorrt::TensorRTEngine::predictor_id_per_thread =
        predictor_id_;
    VLOG(3) << "thread_local var predictor_id in TensorRTEngine is set to: "
            << inference::tensorrt::TensorRTEngine::predictor_id_per_thread;
  }
#endif

1290 1291
  Analyzer().Run(&argument_);

1292
  PADDLE_ENFORCE_EQ(
C
ccrrong 已提交
1293 1294
      argument_.scope_valid(),
      true,
1295
      platform::errors::InvalidArgument("The argument scope should be valid."));
1296 1297
  VLOG(5) << "to prepare executor";
  ARGUMENT_CHECK_FIELD((&argument_), ir_analyzed_program);
Y
Yan Chunwei 已提交
1298
  inference_program_.reset(
1299 1300 1301 1302
      new framework::ProgramDesc(argument_.ir_analyzed_program()),
      [](framework::ProgramDesc *prog) {
// Note, please do NOT use any member variables, because member variables may
// have been destructed in multiple threads.
1303
#ifdef PADDLE_WITH_TENSORRT
W
Wilber 已提交
1304 1305 1306 1307
        auto &block = prog->Block(0);
        for (auto &op_desc : block.AllOps()) {
          if (op_desc->Type() == "tensorrt_engine") {
            std::string engine_key =
R
Ruibiao Chen 已提交
1308
                PADDLE_GET_CONST(std::string, op_desc->GetAttr("engine_key"));
W
Wilber 已提交
1309
            int engine_predictor_id =
R
Ruibiao Chen 已提交
1310
                PADDLE_GET_CONST(int, op_desc->GetAttr("predictor_id"));
W
Wilber 已提交
1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321
            std::string engine_name =
                engine_key + std::to_string(engine_predictor_id);
            if (paddle::inference::Singleton<
                    inference::tensorrt::TRTEngineManager>::Global()
                    .Has(engine_name)) {
              paddle::inference::Singleton<
                  inference::tensorrt::TRTEngineManager>::Global()
                  .DeleteKey(engine_name);
            }
          }
        }
1322 1323 1324
#endif
        delete prog;
      });
1325 1326 1327 1328
  // The config and argument take a lot of storage,
  // when the predictor settings are complete, we release these stores.
  argument_.PartiallyRelease();
  config_.PartiallyRelease();
1329
  LOG(INFO) << "======= optimize end =======";
Y
Yan Chunwei 已提交
1330
}
1331 1332

template <>
1333 1334 1335
std::unique_ptr<PaddlePredictor>
CreatePaddlePredictor<AnalysisConfig, PaddleEngineKind::kAnalysis>(
    const AnalysisConfig &config) {
W
Wilber 已提交
1336 1337
  // TODO(NHZlX): Should add the link to the doc of
  // paddle_infer::CreatePredictor<paddle_infer::Config>
P
Pei Yang 已提交
1338 1339 1340 1341
  if (config.glog_info_disabled()) {
    FLAGS_logtostderr = 1;
    FLAGS_minloglevel = 2;  // GLOG_ERROR
  }
M
minqiyang 已提交
1342
  VLOG(3) << "create AnalysisConfig";
1343
  PADDLE_ENFORCE_EQ(
C
ccrrong 已提交
1344 1345
      config.is_valid(),
      true,
1346 1347
      platform::errors::InvalidArgument(
          "Note: Each config can only be used for one predictor."));
1348

1349 1350 1351 1352
  // Register custom operators compiled by the user.
  // This function can only be executed once per process.
  static std::once_flag custom_operators_registered;
  std::call_once(custom_operators_registered,
1353
                 []() { inference::RegisterAllCustomOperator(); });
1354

1355
  if (config.use_gpu()) {
1356 1357 1358 1359 1360 1361
    static std::once_flag gflags_initialized;
    static bool process_level_allocator_enabled;

    std::call_once(gflags_initialized, [&]() {
      std::vector<std::string> gflags;
      PADDLE_ENFORCE_GE(
C
ccrrong 已提交
1362 1363
          config.memory_pool_init_size_mb(),
          0.f,
1364 1365 1366
          platform::errors::InvalidArgument(
              "The size of memory pool should be greater than 0."));
      PADDLE_ENFORCE_GE(
C
ccrrong 已提交
1367 1368
          config.gpu_device_id(),
          0,
1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381
          platform::errors::InvalidArgument(
              "Invalid device id (%d). The device id should be greater than 0.",
              config.gpu_device_id()));
      gflags.push_back("dummy");

      float fraction_of_gpu_memory = config.fraction_of_gpu_memory_for_pool();
      if (fraction_of_gpu_memory > 0.95f) {
        LOG(ERROR)
            << "Allocate too much memory for the GPU memory pool, assigned "
            << config.memory_pool_init_size_mb() << " MB";
        LOG(ERROR) << "Try to shink the value by setting "
                      "AnalysisConfig::EnableGpu(...)";
      }
1382

1383 1384 1385 1386 1387 1388 1389
      if (fraction_of_gpu_memory >= 0.0f || fraction_of_gpu_memory <= 0.95f) {
        std::string flag = "--fraction_of_gpu_memory_to_use=" +
                           std::to_string(fraction_of_gpu_memory);
        VLOG(3) << "set flag: " << flag;
        gflags.push_back(flag);
      }

1390 1391 1392 1393 1394 1395 1396 1397 1398
      // TODO(Shixiaowei02): Add a mandatory scheme to use the thread local
      // allocator when multi-stream is enabled.
      if (config.thread_local_stream_enabled()) {
        gflags.push_back("--allocator_strategy=thread_local");
        process_level_allocator_enabled = false;
      } else {
        process_level_allocator_enabled = true;
      }

W
Wilber 已提交
1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409
      // support set flags from enviorment.
      const platform::ExportedFlagInfoMap &env_map =
          platform::GetExportedFlagInfoMap();
      std::ostringstream os;
      os << "--tryfromenv=";
      for (auto &pair : env_map) {
        os << pair.second.name << ",";
      }
      auto tryfromenv_str = os.str();
      gflags.push_back(os.str().substr(0, tryfromenv_str.size() - 1));

1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424
      if (framework::InitGflags(gflags)) {
        VLOG(3) << "The following gpu analysis configurations only take effect "
                   "for the first predictor: ";
        for (size_t i = 1; i < gflags.size(); ++i) {
          VLOG(3) << gflags[i];
        }
      } else {
        LOG(WARNING) << "The one-time configuration of analysis predictor "
                        "failed, which may be due to native predictor called "
                        "first and its configurations taken effect.";
      }
    });

    if (config.thread_local_stream_enabled() &&
        process_level_allocator_enabled) {
1425 1426 1427 1428 1429 1430
      PADDLE_THROW(platform::errors::Fatal(
          "When binding threads and streams, the use of "
          "process-level allocators will result in undefined result "
          "errors due to memory asynchronous operations."
          "The thread and stream binding configuration of all "
          "predictors should be the same in a single process."));
1431 1432 1433 1434
    }
  }

  std::unique_ptr<PaddlePredictor> predictor(new AnalysisPredictor(config));
1435 1436
  // Each config can only be used for one predictor.
  config.SetInValid();
1437 1438
  auto predictor_p = dynamic_cast<AnalysisPredictor *>(predictor.get());

1439 1440 1441 1442
#ifdef PADDLE_WITH_TENSORRT
  paddle::framework::ir::patterns::KeyCounter::Instance().CleanCounter();
#endif

1443 1444 1445 1446 1447
  if (!predictor_p->Init(nullptr)) {
    return nullptr;
  }

  if (config.mkldnn_quantizer_enabled() && !predictor_p->MkldnnQuantize()) {
1448 1449
    return nullptr;
  }
1450

G
Gabor Buella 已提交
1451
  return predictor;
1452 1453
}

1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465
bool AnalysisPredictor::MkldnnQuantize() {
#if PADDLE_WITH_MKLDNN
  if (!mkldnn_quantizer_)
    mkldnn_quantizer_ = new AnalysisPredictor::MkldnnQuantizer(
        *this, config_.mkldnn_quantizer_config());
  return mkldnn_quantizer_->Quantize();
#else
  LOG(ERROR) << "Please compile with MKLDNN first to use MkldnnQuantizer";
  return false;
#endif
}

1466
void AnalysisPredictor::PrepareFeedFetch() {
1467 1468 1469
  PADDLE_ENFORCE_NOT_NULL(sub_scope_,
                          platform::errors::InvalidArgument(
                              "The sub_scope should not be nullptr."));
1470
  CreateFeedFetchVar(sub_scope_);
1471 1472
  for (auto *op : inference_program_->Block(0).AllOps()) {
    if (op->Type() == "feed") {
R
Ruibiao Chen 已提交
1473
      int idx = PADDLE_GET_CONST(int, op->GetAttr("col"));
1474 1475 1476 1477 1478
      if (feeds_.size() <= static_cast<size_t>(idx)) {
        feeds_.resize(idx + 1);
      }
      feeds_[idx] = op;
      feed_names_[op->Output("Out")[0]] = idx;
N
nhzlx 已提交
1479
      idx2feeds_[idx] = op->Output("Out")[0];
1480
    } else if (op->Type() == "fetch") {
R
Ruibiao Chen 已提交
1481
      int idx = PADDLE_GET_CONST(int, op->GetAttr("col"));
Y
Yan Chunwei 已提交
1482 1483
      if (fetches_.size() <= static_cast<size_t>(idx)) {
        fetches_.resize(idx + 1);
1484
      }
Y
Yan Chunwei 已提交
1485
      fetches_[idx] = op;
N
nhzlx 已提交
1486
      idx2fetches_[idx] = op->Input("X")[0];
1487 1488 1489 1490
    }
  }
}

1491
void AnalysisPredictor::CreateFeedFetchVar(framework::Scope *scope) {
C
ccrrong 已提交
1492 1493 1494
  PADDLE_ENFORCE_NOT_NULL(
      scope,
      platform::errors::InvalidArgument("The scope should not be nullptr."));
1495
  auto *var = scope->Var("feed");
1496
  var->GetMutable<framework::FeedList>();
1497
  var = scope->Var("fetch");
1498
  var->GetMutable<framework::FetchList>();
1499 1500
}

N
nhzlx 已提交
1501 1502 1503 1504 1505 1506 1507 1508
std::vector<std::string> AnalysisPredictor::GetInputNames() {
  std::vector<std::string> input_names;
  for (auto &item : idx2feeds_) {
    input_names.push_back(item.second);
  }
  return input_names;
}

1509 1510 1511 1512 1513 1514
std::map<std::string, std::vector<int64_t>>
AnalysisPredictor::GetInputTensorShape() {
  std::map<std::string, std::vector<int64_t>> input_shapes;
  std::vector<std::string> names = GetInputNames();
  for (std::string name : names) {
    auto *var = inference_program_->Block(0).FindVar(name);
C
ccrrong 已提交
1515 1516 1517
    PADDLE_ENFORCE_NOT_NULL(
        var,
        platform::errors::PreconditionNotMet("Input %s does not exist.", name));
1518 1519 1520 1521 1522
    input_shapes[name] = var->GetShape();
  }
  return input_shapes;
}

1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553
std::map<std::string, paddle_infer::DataType>
AnalysisPredictor::GetInputTypes() {
  std::map<std::string, paddle_infer::DataType> input_type;
  std::vector<std::string> names = GetInputNames();
  for (const auto &name : names) {
    auto *var = inference_program_->Block(0).FindVar(name);
    PADDLE_ENFORCE_NOT_NULL(
        var,
        platform::errors::PreconditionNotMet(
            "Input %s does not exist inference_program_.", name));
    auto dtype = var->GetDataType();
    if (dtype == paddle::framework::proto::VarType::FP32) {
      input_type[name] = paddle_infer::DataType::FLOAT32;
    } else if (dtype == paddle::framework::proto::VarType::FP16) {
      input_type[name] = paddle_infer::DataType::FLOAT16;
    } else if (dtype == paddle::framework::proto::VarType::INT64) {
      input_type[name] = paddle_infer::DataType::INT64;
    } else if (dtype == paddle::framework::proto::VarType::INT32) {
      input_type[name] = paddle_infer::DataType::INT32;
    } else if (dtype == paddle::framework::proto::VarType::UINT8) {
      input_type[name] = paddle_infer::DataType::UINT8;
    } else if (dtype == paddle::framework::proto::VarType::INT8) {
      input_type[name] = paddle_infer::DataType::INT8;
    } else {
      PADDLE_THROW(paddle::platform::errors::Unimplemented(
          "Unsupported data type `%s` when get input dtype ", dtype));
    }
  }
  return input_type;
}

N
nhzlx 已提交
1554 1555 1556 1557 1558 1559 1560 1561
std::vector<std::string> AnalysisPredictor::GetOutputNames() {
  std::vector<std::string> output_names;
  for (auto &item : idx2fetches_) {
    output_names.push_back(item.second);
  }
  return output_names;
}

1562 1563
std::unique_ptr<ZeroCopyTensor> AnalysisPredictor::GetInputTensor(
    const std::string &name) {
1564
  framework::Scope *scope;
1565
#if defined(PADDLE_WITH_DISTRIBUTE) && defined(PADDLE_WITH_PSCORE)
1566 1567 1568
  if (config_.dist_config().use_dist_model()) {
    scope = scope_.get();
  } else {
1569
    scope = executor_->GetScope();
1570 1571
  }
#else
1572
  scope = executor_->GetScope();
1573
#endif
1574
  PADDLE_ENFORCE_NOT_NULL(
1575
      scope->FindVar(name),
1576
      platform::errors::PreconditionNotMet(
1577
          "The variable named %s is not found in the scope of the executor.",
1578
          name));
1579 1580
  std::unique_ptr<ZeroCopyTensor> res(new ZeroCopyTensor(
      static_cast<void *>(scope), this->GetDeviceContexts()));
1581 1582
  res->input_or_output_ = true;
  res->SetName(name);
N
nhzlx 已提交
1583 1584
  if (platform::is_cpu_place(place_)) {
    res->SetPlace(PaddlePlace::kCPU);
J
jianghaicheng 已提交
1585 1586 1587 1588
  } else if (platform::is_ipu_place(place_)) {
    // Currently, IPUPlace's tensor copy between cpu and ipu has been set in
    // IpuBackend.
    res->SetPlace(PaddlePlace::kCPU);
1589
  } else if (platform::is_xpu_place(place_)) {
1590 1591 1592 1593 1594 1595 1596 1597
    if (config_.lite_engine_enabled()) {
      // Currently, Paddle-Lite's XPU user interface only supports the transfer
      // of host data pointers. If it is currently used as a subgraph, execution
      // efficiency will be sacrificed, so it is temporarily set to cpu place.
      // And, the current lite engine of xpu must execute all parts of the
      // model.
      res->SetPlace(PaddlePlace::kCPU);
    } else {
1598
      auto xpu_place = place_;
1599 1600
      res->SetPlace(PaddlePlace::kXPU, xpu_place.GetDeviceId());
    }
W
Wilber 已提交
1601
  } else if (platform::is_npu_place(place_)) {
1602
    auto npu_place = place_;
W
Wilber 已提交
1603
    res->SetPlace(PaddlePlace::kNPU, npu_place.GetDeviceId());
1604 1605 1606 1607 1608 1609
  } else if (platform::is_custom_place(place_)) {
    auto custom_place = place_;
    auto paddleplace = static_cast<PaddlePlace>(
        static_cast<size_t>(PaddlePlace::kCUSTOM) +
        phi::GetOrRegisterGlobalDeviceTypeId(place_.GetDeviceType()));
    res->SetPlace(paddleplace, custom_place.GetDeviceId());
N
nhzlx 已提交
1610
  } else {
1611
    auto gpu_place = place_;
N
nhzlx 已提交
1612 1613
    res->SetPlace(PaddlePlace::kGPU, gpu_place.GetDeviceId());
  }
1614 1615 1616 1617 1618
  return res;
}

std::unique_ptr<ZeroCopyTensor> AnalysisPredictor::GetOutputTensor(
    const std::string &name) {
1619
  framework::Scope *scope;
1620
#if defined(PADDLE_WITH_DISTRIBUTE) && defined(PADDLE_WITH_PSCORE)
1621 1622 1623
  if (config_.dist_config().use_dist_model()) {
    scope = scope_.get();
  } else {
1624
    scope = executor_->GetScope();
1625 1626
  }
#else
1627
  scope = executor_->GetScope();
1628
#endif
1629
  PADDLE_ENFORCE_NOT_NULL(
1630
      scope->FindVar(name),
1631
      platform::errors::PreconditionNotMet(
1632
          "The variable named %s is not found in the scope of the executor.",
1633
          name));
1634 1635
  std::unique_ptr<ZeroCopyTensor> res(new ZeroCopyTensor(
      static_cast<void *>(scope), this->GetDeviceContexts()));
1636 1637
  res->input_or_output_ = false;
  res->SetName(name);
N
nhzlx 已提交
1638 1639
  if (platform::is_cpu_place(place_)) {
    res->SetPlace(PaddlePlace::kCPU);
J
jianghaicheng 已提交
1640 1641 1642 1643
  } else if (platform::is_ipu_place(place_)) {
    // Currently, IPUPlace's tensor copy between cpu and ipu has been set in
    // IpuBackend.
    res->SetPlace(PaddlePlace::kCPU);
1644
  } else if (platform::is_xpu_place(place_)) {
1645 1646 1647 1648 1649 1650 1651 1652
    if (config_.lite_engine_enabled()) {
      // Currently, Paddle-Lite's XPU user interface only supports the transfer
      // of host data pointers. If it is currently used as a subgraph, execution
      // efficiency will be sacrificed, so it is temporarily set to cpu place.
      // And, the current lite engine of xpu must execute all parts of the
      // model.
      res->SetPlace(PaddlePlace::kCPU);
    } else {
1653
      auto xpu_place = place_;
1654 1655
      res->SetPlace(PaddlePlace::kXPU, xpu_place.GetDeviceId());
    }
W
Wilber 已提交
1656
  } else if (platform::is_npu_place(place_)) {
1657
    auto npu_place = place_;
W
Wilber 已提交
1658
    res->SetPlace(PaddlePlace::kNPU, npu_place.GetDeviceId());
1659 1660 1661 1662 1663 1664
  } else if (platform::is_custom_place(place_)) {
    auto custom_place = place_;
    auto paddleplace = static_cast<PaddlePlace>(
        static_cast<size_t>(PaddlePlace::kCUSTOM) +
        phi::GetOrRegisterGlobalDeviceTypeId(place_.GetDeviceType()));
    res->SetPlace(paddleplace, custom_place.GetDeviceId());
N
nhzlx 已提交
1665
  } else {
1666
    auto gpu_place = place_;
N
nhzlx 已提交
1667 1668
    res->SetPlace(PaddlePlace::kGPU, gpu_place.GetDeviceId());
  }
1669 1670 1671 1672
  return res;
}

bool AnalysisPredictor::ZeroCopyRun() {
1673
  inference::DisplayMemoryInfo(place_, "before run");
1674
#if defined(PADDLE_WITH_DISTRIBUTE) && defined(PADDLE_WITH_PSCORE)
1675 1676 1677 1678 1679 1680 1681 1682 1683 1684
  if (config_.dist_config().use_dist_model()) {
    VLOG(3) << "ZeroCopyRun will use the fleet executor.";
    inference::Timer timer;
    timer.tic();
    fleet_exe_->Run(config_.dist_config().carrier_id());
    VLOG(3) << "Fleet executor inf runs once use: "
            << std::to_string(timer.toc()) << "ms";
    return true;
  }
#endif
1685 1686 1687
  if (private_context_) {
    paddle::platform::DeviceContextPool::SetDeviceContexts(&device_contexts_);
  }
1688
  paddle::platform::SetNumThreads(config_.cpu_math_library_num_threads());
W
Wilber 已提交
1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699
#ifdef PADDLE_WITH_MKLDNN
  if (config_.use_mkldnn_) {
    std::vector<std::vector<int>> shape_vector;
    auto names = GetInputNames();
    for (size_t i = 0; i < names.size(); ++i) {
      auto in_tensor = GetInputTensor(names[i]);
      shape_vector.emplace_back(in_tensor->shape());
    }
    MkldnnPreSet(shape_vector);
  }
#endif
1700 1701 1702 1703 1704 1705 1706 1707 1708 1709

#ifdef PADDLE_WITH_TENSORRT
  if (config_.tensorrt_engine_enabled()) {
    inference::tensorrt::TensorRTEngine::predictor_id_per_thread =
        predictor_id_;
    VLOG(3) << "thread_local var predictor_id in TensorRTEngine is set to: "
            << inference::tensorrt::TensorRTEngine::predictor_id_per_thread;
  }
#endif

1710
  executor_->Run();
1711
  inference::DisplayMemoryInfo(place_, "after run");
1712 1713 1714 1715 1716

  if (config_.shape_range_info_collected()) {
    CollectShapeRangeInfo();
  }

Y
Yan Chunwei 已提交
1717
  // Fix TensorArray reuse not cleaned bug.
Y
Yan Chunwei 已提交
1718
  tensor_array_batch_cleaner_.CollectTensorArrays(sub_scope_);
Y
Yan Chunwei 已提交
1719
  tensor_array_batch_cleaner_.ResetTensorArray();
1720 1721 1722 1723

  // recover the cpu_math_library_num_threads to 1, in order to avoid thread
  // conflict when integrating it into deployment service.
  paddle::platform::SetNumThreads(1);
1724 1725 1726
  if (private_context_) {
    paddle::platform::DeviceContextPool::SetDeviceContexts(nullptr);
  }
W
Wilber 已提交
1727 1728 1729
#ifdef PADDLE_WITH_MKLDNN
  if (config_.use_mkldnn_) MkldnnPostReset();
#endif
1730
#if defined(PADDLE_WITH_MKLML)
T
Tao Luo 已提交
1731 1732 1733 1734 1735
  // Frees unused memory allocated by the Intel® MKL Memory Allocator to
  // avoid memory leak. See:
  // https://software.intel.com/en-us/mkl-developer-reference-c-mkl-free-buffers
  platform::dynload::MKL_Free_Buffers();
#endif
1736 1737 1738
  return true;
}

W
Wilber 已提交
1739 1740
#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
bool AnalysisPredictor::ExpRunWithExternalStream(const gpuStream_t stream) {
W
Wilber 已提交
1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765
  if (!private_context_) {
    PADDLE_THROW(platform::errors::Fatal(
        "Please use config.SetExecStream to init gpu resources, and then we "
        "will bind gpu resources to execution stream."));
  }

  if (stream != predictor_stream_) {
#ifdef PADDLE_WITH_HIP
    hipStreamSynchronize(static_cast<gpuStream_t>(predictor_stream_));
#else
    cudaStreamSynchronize(static_cast<gpuStream_t>(predictor_stream_));
#endif
    ResourceManager::Instance().GpuResourceReBindStream(predictor_stream_,
                                                        stream);
    predictor_stream_ = stream;

    auto *dev_ctxs = reinterpret_cast<const std::map<
        phi::Place,
        std::shared_future<std::unique_ptr<phi::DeviceContext>>> *>(
        this->GetDeviceContexts());
    auto *dev_ctx =
        static_cast<InferGPUContext *>(dev_ctxs->at(place_).get().get());
    dev_ctx->SetStream(stream);
  }

W
Wilber 已提交
1766 1767 1768 1769
  return ZeroCopyRun();
}
#endif

1770 1771 1772 1773 1774 1775
void AnalysisPredictor::CollectShapeRangeInfo() {
  // if use gpu, sync first.
  if (config_.use_gpu()) {
#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
    paddle::platform::DeviceContextPool &pool =
        paddle::platform::DeviceContextPool::Instance();
1776
    auto gpu_place = place_;
L
Leo Chen 已提交
1777
    auto *dev_ctx = static_cast<const phi::GPUContext *>(pool.Get(gpu_place));
1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788
#ifdef PADDLE_WITH_HIP
    hipStreamSynchronize(dev_ctx->stream());
#else
    cudaStreamSynchronize(dev_ctx->stream());
#endif
#endif
  }

  std::vector<std::string> var_names = sub_scope_->LocalVarNames();
  for (const auto &name : var_names) {
    auto *var = sub_scope_->GetVar(name);
1789
    if (!var->IsType<phi::DenseTensor>()) {
1790 1791
      continue;
    }
1792 1793
    auto tensor = var->Get<phi::DenseTensor>();
    framework::DDim dim = tensor.dims();
1794 1795 1796
    std::vector<int32_t> shape(dim.size());
    for (size_t i = 0; i < shape.size(); ++i) shape[i] = dim[i];
    shape_info_[name].emplace_back(shape);
1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824

    // We need collect value range for shape tensor for Paddle-TRT's use.
    // To be noticed, this method to identify all shape tensors is based on
    // assumption that all shape tensors in the model have numbers <= 7.
    // This is a simple method to identify all shape tensors with some
    // mistakes, but it doesn't matter.
    auto is_shape_tensor = tensor.numel() <= 7 && tensor.numel() >= 1;
    if (tensor.dtype() == paddle::experimental::DataType::INT32 &&
        is_shape_tensor) {
      std::vector<int> int32_host(tensor.numel());
      if (tensor.place() == platform::CPUPlace()) {
        paddle::memory::Copy(platform::CPUPlace(),
                             int32_host.data(),
                             platform::CPUPlace(),
                             tensor.data<int>(),
                             tensor.numel() * sizeof(int));
      } else if (tensor.place() == platform::CUDAPlace()) {
#if defined(PADDLE_WITH_CUDA)
        paddle::memory::Copy(platform::CPUPlace(),
                             int32_host.data(),
                             platform::CUDAPlace(),
                             tensor.data<int>(),
                             tensor.numel() * sizeof(int),
                             nullptr);
#endif
      }
      shape_tensor_value_[name].emplace_back(int32_host);
    }
1825 1826 1827 1828 1829 1830 1831
  }
}

void AnalysisPredictor::StatisticShapeRangeInfo() {
  std::map<std::string, std::vector<int32_t>> min_shapes;
  std::map<std::string, std::vector<int32_t>> max_shapes;
  std::map<std::string, std::vector<int32_t>> opt_shapes;
1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870
  std::map<std::string, std::vector<int32_t>> min_values;
  std::map<std::string, std::vector<int32_t>> max_values;
  std::map<std::string, std::vector<int32_t>> opt_values;

  auto extract_min_max_opt =
      [](std::map<std::string, std::vector<int32_t>> &min_data,
         decltype(min_data) max_data,
         decltype(min_data) opt_data,
         decltype(shape_info_) shape_data) {
        for (auto it : shape_data) {
          auto name = it.first;
          auto shapes = it.second;

          std::vector<int32_t> min_shape(shapes[0].begin(), shapes[0].end());
          std::vector<int32_t> max_shape(shapes[0].begin(), shapes[0].end());
          std::vector<int32_t> opt_shape(shapes[0].begin(), shapes[0].end());

          auto ShapeMaxFreq =
              [](const std::map<int32_t, int32_t> &m) -> int32_t {
            std::vector<std::pair<int32_t, int32_t>> counter;
            for (auto &it : m) counter.push_back(it);
            std::sort(counter.begin(),
                      counter.end(),
                      [](std::pair<int32_t, int32_t> &a,
                         std::pair<int32_t, int32_t> &b) {
                        return a.second > b.second;
                      });
            return counter[0].first;
          };

          for (size_t d = 0; d < shapes[0].size(); ++d) {
            std::map<int32_t, int32_t> counter;
            for (size_t i = 0; i < shapes.size(); ++i) {
              counter[shapes[i][d]] += 1;
              if (shapes[i][d] < min_shape[d]) min_shape[d] = shapes[i][d];
              if (shapes[i][d] > max_shape[d]) max_shape[d] = shapes[i][d];
            }
            opt_shape[d] = ShapeMaxFreq(counter);
          }
1871

1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886
          min_data[name] = min_shape;
          max_data[name] = max_shape;
          opt_data[name] = opt_shape;
        }
      };
  extract_min_max_opt(min_shapes, max_shapes, opt_shapes, shape_info_);
  extract_min_max_opt(min_values, max_values, opt_values, shape_tensor_value_);

  inference::SerializeShapeRangeInfo(config_.shape_range_info_path(),
                                     min_shapes,
                                     max_shapes,
                                     opt_shapes,
                                     min_values,
                                     max_values,
                                     opt_values);
1887 1888
}

1889 1890
bool AnalysisPredictor::LoadProgramDesc() {
  // Initialize the inference program
1891
  std::string filename;
1892 1893
  if (!config_.model_dir().empty()) {
    filename = config_.model_dir() + "/__model__";
1894
  } else if (!config_.prog_file().empty()) {
1895 1896 1897
    // 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`.
1898
    filename = config_.prog_file();
1899
  } else {
1900
    if (config_.model_dir().empty() && config_.prog_file().empty()) {
1901 1902 1903 1904
      LOG(ERROR)
          << "Either model_dir or (prog_file, param_file) should be set.";
      return false;
    }
1905
    LOG(ERROR) << string::Sprintf(
C
ccrrong 已提交
1906 1907
        "not valid model path '%s' or program path '%s'.",
        config_.model_dir(),
1908
        config_.params_file());
1909 1910
    return false;
  }
1911 1912 1913

  // Create ProgramDesc
  framework::proto::ProgramDesc proto;
T
Tao Luo 已提交
1914
  if (!config_.model_from_memory()) {
T
Tao Luo 已提交
1915 1916 1917
    std::string pb_content;
    // Read binary
    std::ifstream fin(filename, std::ios::in | std::ios::binary);
1918
    PADDLE_ENFORCE_EQ(
C
ccrrong 已提交
1919 1920
        static_cast<bool>(fin.is_open()),
        true,
1921 1922 1923
        platform::errors::NotFound(
            "Cannot open file %s, please confirm whether the file is normal.",
            filename));
T
Tao Luo 已提交
1924 1925 1926 1927 1928 1929 1930 1931
    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();

    proto.ParseFromString(pb_content);
  } else {
1932
    proto.ParseFromString(config_.prog_file());
T
Tao Luo 已提交
1933
  }
1934 1935 1936 1937 1938 1939
  inference_program_.reset(new framework::ProgramDesc(proto));
  return true;
}

bool AnalysisPredictor::LoadParameters() {
  PADDLE_ENFORCE_NOT_NULL(inference_program_.get(),
1940 1941
                          platform::errors::PreconditionNotMet(
                              "The inference program should be loaded first."));
T
Tao Luo 已提交
1942

1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962
  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);

1963
      if (!config_.params_file().empty()) {
1964 1965 1966 1967 1968 1969
        params.push_back(new_var->Name());
      } else {
        // append_op
        framework::OpDesc *op = load_block->AppendOp();
        op->SetType("load");
        op->SetOutput("Out", {new_var->Name()});
1970
        op->SetAttr("file_path", {config_.model_dir() + "/" + new_var->Name()});
1971 1972 1973 1974 1975
        op->CheckAttrs();
      }
    }
  }

1976
  if (!config_.params_file().empty()) {
1977 1978 1979 1980 1981 1982
    // 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);
1983
    op->SetAttr("file_path", {config_.params_file()});
1984 1985 1986 1987
    op->CheckAttrs();
  }

  // Use NaiveExecutor to Load parameters.
S
superjomn 已提交
1988
  framework::NaiveExecutor e(place_);
1989 1990 1991 1992
  e.Prepare(scope_.get(), *load_program, 0, false);
  e.Run();
  VLOG(3) << "get " << scope_->LocalVarNames().size() << " vars after load";

1993 1994
  return true;
}
1995

1996 1997 1998 1999 2000
uint64_t AnalysisPredictor::TryShrinkMemory() {
  ClearIntermediateTensor();
  return paddle::memory::Release(place_);
}

2001 2002 2003 2004 2005 2006 2007 2008
void AnalysisPredictor::ClearIntermediateTensor() {
  PADDLE_ENFORCE_NOT_NULL(inference_program_.get(),
                          platform::errors::PreconditionNotMet(
                              "The inference program should be loaded first."));
  const auto &global_block = inference_program_->MutableBlock(0);
  for (auto *var : global_block->AllVars()) {
    if (!IsPersistable(var)) {
      const std::string name = var->Name();
2009
      auto *variable = executor_->GetScope()->FindVar(name);
2010
      if (variable != nullptr && variable->IsType<phi::DenseTensor>() &&
2011 2012
          name != "feed" && name != "fetch") {
        VLOG(3) << "Clear Intermediate Tensor: " << name;
2013
        auto *t = variable->GetMutable<phi::DenseTensor>();
2014 2015 2016 2017 2018 2019
        t->clear();
      }
    }
  }
}

2020
#ifdef PADDLE_WITH_TENSORRT
N
nhzlx 已提交
2021
bool AnalysisPredictor::SaveTrtCalibToDisk() {
C
ccrrong 已提交
2022 2023
  PADDLE_ENFORCE_EQ(config_.tensorrt_engine_enabled(),
                    true,
2024 2025
                    platform::errors::PreconditionNotMet(
                        "This func can be invoked only in trt mode"));
N
nhzlx 已提交
2026 2027 2028
  auto &block = inference_program_->Block(0);
  for (auto &op_desc : block.AllOps()) {
    if (op_desc->Type() == "tensorrt_engine") {
R
Ruibiao Chen 已提交
2029
      std::string engine_name = PADDLE_GET_CONST(
2030
          std::string, op_desc->GetAttr("calibration_engine_key"));
N
nhzlx 已提交
2031
      if (!Singleton<TRTCalibratorEngineManager>::Global().Has(engine_name)) {
N
nhzlx 已提交
2032 2033 2034 2035
        LOG(ERROR) << "You should run the predictor(with trt) on the real data "
                      "to generate calibration info";
        return false;
      }
N
nhzlx 已提交
2036 2037
      TRTCalibratorEngine *calib_engine =
          Singleton<TRTCalibratorEngineManager>::Global().Get(engine_name);
N
nhzlx 已提交
2038
      LOG(INFO) << "Wait for calib threads done.";
N
nhzlx 已提交
2039
      calib_engine->calib_->waitAndSetDone();
N
nhzlx 已提交
2040 2041
      LOG(INFO) << "Generating TRT Calibration table data, this may cost a lot "
                   "of time...";
N
nhzlx 已提交
2042 2043 2044
      calib_engine->thr_->join();
      std::string calibration_table_data =
          calib_engine->calib_->getCalibrationTableAsString();
N
nhzlx 已提交
2045

N
nhzlx 已提交
2046
      if (calibration_table_data.empty()) {
N
nhzlx 已提交
2047 2048 2049
        LOG(ERROR) << "the calibration table is empty.";
        return false;
      }
N
nhzlx 已提交
2050

N
nhzlx 已提交
2051 2052 2053 2054 2055
      std::string model_opt_cache_dir =
          argument_.Has("model_dir")
              ? argument_.model_dir()
              : inference::analysis::GetDirRoot(argument_.model_program_path());

N
nhzlx 已提交
2056
      std::string calibration_table_data_path =
N
nhzlx 已提交
2057 2058 2059 2060
          inference::analysis::GetTrtCalibPath(
              inference::analysis::GetOrCreateModelOptCacheDir(
                  model_opt_cache_dir),
              engine_name);
N
nhzlx 已提交
2061 2062 2063 2064 2065

      std::ofstream ofile(calibration_table_data_path, std::ios::out);
      LOG(INFO) << "Write Paddle-TRT INT8 calibration table data to file "
                << calibration_table_data_path;
      ofile << calibration_table_data;
N
nhzlx 已提交
2066 2067 2068 2069
      ofile.close();
    }
  }
  // Free all calibrator resources.
N
nhzlx 已提交
2070
  Singleton<TRTCalibratorEngineManager>::Global().DeleteALL();
N
nhzlx 已提交
2071 2072
  return true;
}
N
nhzlx 已提交
2073
#endif
N
nhzlx 已提交
2074

2075
AnalysisPredictor::~AnalysisPredictor() {
2076
#ifdef PADDLE_WITH_TENSORRT
N
nhzlx 已提交
2077
  if (config_.tensorrt_engine_enabled() &&
N
nhzlx 已提交
2078 2079
      config_.tensorrt_precision_mode_ == AnalysisConfig::Precision::kInt8 &&
      Singleton<TRTCalibratorEngineManager>::Global().Has()) {
N
nhzlx 已提交
2080 2081
    SaveTrtCalibToDisk();
  }
N
nhzlx 已提交
2082
#endif
2083
  if (config_.with_profile_) {
2084 2085 2086 2087
    platform::DisableProfiler(platform::EventSortingKey::kTotal,
                              "./profile.log");
  }
  if (sub_scope_) {
J
JingZhuangzhuang 已提交
2088 2089 2090 2091 2092 2093 2094 2095 2096
    if (framework::global_transfer_scope_key().find(sub_scope_) !=
        framework::global_transfer_scope_key().end()) {
      auto scope_key_set = framework::global_transfer_scope_key()[sub_scope_];
      for (auto iter = scope_key_set.begin(); iter != scope_key_set.end();
           iter++) {
        framework::global_transfer_data_cache().erase(*iter);
      }
      framework::global_transfer_scope_key().erase(sub_scope_);
    }
2097 2098
    scope_->DeleteScope(sub_scope_);
  }
Y
Yan Chunwei 已提交
2099

2100 2101 2102 2103 2104 2105
#if PADDLE_WITH_MKLDNN
  if (mkldnn_quantizer_) {
    delete mkldnn_quantizer_;
    mkldnn_quantizer_ = nullptr;
  }
#endif
2106

2107 2108 2109
  if (config_.shape_range_info_collected()) {
    StatisticShapeRangeInfo();
  }
2110 2111 2112 2113 2114
#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
  if (predictor_stream_ != nullptr) {
    ResourceManager::Instance().DestroyGPUResource(predictor_stream_);
  }
#endif
W
Wilber 已提交
2115 2116 2117
  if (place_.GetType() != phi::AllocationType::UNDEFINED) {
    memory::Release(place_);
  }
2118
  device_contexts_.clear();
2119 2120 2121 2122 2123 2124 2125

#ifdef PADDLE_WITH_TENSORRT
  if (config_.trt_engine_memory_sharing()) {
    inference::Singleton<inference::tensorrt::TRTEngineManager>::Global()
        .releaseContextMemory(predictor_id_);
  }
#endif
2126 2127
}

2128
std::unique_ptr<PaddlePredictor> AnalysisPredictor::Clone(void *stream) {
Y
Yan Chunwei 已提交
2129
  std::lock_guard<std::mutex> lk(clone_mutex_);
2130
  auto *x = new AnalysisPredictor(config_);
2131
  x->status_is_cloned_ = true;
2132
  x->root_predictor_id_ = this->root_predictor_id_;
2133 2134 2135 2136 2137 2138 2139 2140 2141 2142
  if (config_.use_external_stream_ && stream == nullptr) {
    PADDLE_THROW(platform::errors::InvalidArgument(
        "config has been configured to use external stream, but the Clone "
        "function has not received a valid stream parameter."));
  } else if (!config_.use_external_stream_ && stream != nullptr) {
    PADDLE_THROW(platform::errors::InvalidArgument(
        "config has not been configured to use external stream, but the Clone "
        "function has received a stream parameter."));
  }
  x->predictor_stream_ = stream;
2143
  x->Init(scope_, inference_program_);
2144
#ifdef PADDLE_WITH_TENSORRT
2145
  x->executor_->ResetTrtOps(++AnalysisPredictor::clone_num_);
2146
#endif
2147 2148 2149
  return std::unique_ptr<PaddlePredictor>(x);
}

2150
std::string AnalysisPredictor::GetSerializedProgram() const {
Y
Yan Chunwei 已提交
2151 2152 2153
  return inference_program_->Proto()->SerializeAsString();
}

2154 2155 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192
// Add SaveOptimModel
void AnalysisPredictor::SaveOptimModel(const std::string &dir) {
  // save model
  std::string model_name = dir + "/model";
  std::ofstream outfile;
  outfile.open(model_name, std::ios::out | std::ios::binary);
  std::string inference_prog_desc = GetSerializedProgram();
  outfile << inference_prog_desc;
  // save params
  framework::ProgramDesc save_program;
  auto *save_block = save_program.MutableBlock(0);

  const framework::ProgramDesc &main_program = program();
  const framework::BlockDesc &global_block = main_program.Block(0);
  std::vector<std::string> save_var_list;
  for (framework::VarDesc *var : global_block.AllVars()) {
    if (IsPersistable(var)) {
      framework::VarDesc *new_var = save_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);

      save_var_list.push_back(new_var->Name());
    }
  }
  std::sort(save_var_list.begin(), save_var_list.end());
  auto *op = save_block->AppendOp();
  op->SetType("save_combine");
  op->SetInput("X", save_var_list);
  op->SetAttr("file_path", dir + "/params");
  op->CheckAttrs();

  platform::CPUPlace place;
  framework::Executor exe(place);
  exe.Run(save_program, scope(), 0, true, true);
}

2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213
void AnalysisPredictor::RegisterOutputHook(const Exp_OutputHookFunc &hookfunc) {
  static std::once_flag register_hook_flag;
  std::call_once(register_hook_flag, [this] {
    executor_->RegisterOutputHook([this](framework::OperatorBase *op) {
      for (auto &output : op->Outputs()) {
        for (auto &var_name : output.second) {
          auto *var = this->sub_scope_->FindVar(var_name);
          if (!var || !var->IsType<phi::DenseTensor>()) continue;
          auto dense_tensor = var->Get<phi::DenseTensor>();
          if (!dense_tensor.initialized()) continue;
          auto tensor = this->GetOutputTensor(var_name);
          for (auto &hookfunc : this->hookfuncs_) {
            hookfunc(op->Type(), var_name, *tensor);
          }
        }
      }
    });
  });
  hookfuncs_.push_back(hookfunc);
}

Y
Yan Chunwei 已提交
2214
template <>
2215 2216
std::unique_ptr<PaddlePredictor> CreatePaddlePredictor<AnalysisConfig>(
    const AnalysisConfig &config) {
W
Wilber 已提交
2217
  LOG(WARNING) << "Deprecated. Please use CreatePredictor instead.";
2218 2219
  return CreatePaddlePredictor<AnalysisConfig, PaddleEngineKind::kAnalysis>(
      config);
Y
Yan Chunwei 已提交
2220 2221
}

2222
}  // namespace paddle
2223

2224
#ifdef PADDLE_WITH_TENSORRT
2225
USE_TRT_CONVERTER(elementwise_add_weight);
S
shentanyue 已提交
2226 2227 2228
USE_TRT_CONVERTER(elementwise_sub_weight);
USE_TRT_CONVERTER(elementwise_mul_weight);
USE_TRT_CONVERTER(elementwise_div_weight);
2229 2230
USE_TRT_CONVERTER(elementwise_min_weight);
USE_TRT_CONVERTER(elementwise_max_weight);
S
shentanyue 已提交
2231
USE_TRT_CONVERTER(elementwise_pow_weight);
W
wenbin 已提交
2232
USE_TRT_CONVERTER(elementwise_floordiv_weight);
2233 2234 2235 2236 2237 2238 2239
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);
W
wenbin 已提交
2240
USE_TRT_CONVERTER(elementwise_floordiv_tensor);
2241 2242 2243 2244 2245 2246
USE_TRT_CONVERTER(less_than);
USE_TRT_CONVERTER(greater_than);
USE_TRT_CONVERTER(logical_or);
USE_TRT_CONVERTER(logical_xor);
USE_TRT_CONVERTER(logical_and);
USE_TRT_CONVERTER(less_equal);
2247
USE_TRT_CONVERTER(transpose);
2248
USE_TRT_CONVERTER(transpose2);
2249
USE_TRT_CONVERTER(flatten);
2250
USE_TRT_CONVERTER(flatten_contiguous_range);
2251
USE_TRT_CONVERTER(matmul);
2252
USE_TRT_CONVERTER(matmul_v2);
2253
USE_TRT_CONVERTER(bmm);
G
gem5 已提交
2254
USE_TRT_CONVERTER(rsqrt);
2255 2256
USE_TRT_CONVERTER(conv2d);
USE_TRT_CONVERTER(relu);
Z
zhupengyang 已提交
2257 2258
USE_TRT_CONVERTER(exp);
USE_TRT_CONVERTER(log);
2259 2260 2261 2262 2263 2264 2265 2266 2267
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);
2268 2269
USE_TRT_CONVERTER(hard_sigmoid);
USE_TRT_CONVERTER(hard_swish);
2270
USE_TRT_CONVERTER(split);
2271
USE_TRT_CONVERTER(fill_any_like);
2272 2273
USE_TRT_CONVERTER(prelu);
USE_TRT_CONVERTER(conv2d_transpose);
H
hjchen2 已提交
2274
USE_TRT_CONVERTER(leaky_relu);
2275
USE_TRT_CONVERTER(shuffle_channel);
2276
USE_TRT_CONVERTER(where);
2277
USE_TRT_CONVERTER(swish);
L
LielinJiang 已提交
2278
USE_TRT_CONVERTER(silu);
2279
USE_TRT_CONVERTER(group_norm);
2280
USE_TRT_CONVERTER(instance_norm);
P
Pei Yang 已提交
2281 2282 2283
USE_TRT_CONVERTER(layer_norm);
USE_TRT_CONVERTER(gelu);
USE_TRT_CONVERTER(multihead_matmul);
2284
USE_TRT_CONVERTER(multihead_matmul_roformer);
2285
USE_TRT_CONVERTER(skip_layernorm);
2286
USE_TRT_CONVERTER(slice);
2287
USE_TRT_CONVERTER(scale);
2288
USE_TRT_CONVERTER(stack);
P
Pei Yang 已提交
2289
USE_TRT_CONVERTER(clip);
2290
USE_TRT_CONVERTER(gather);
2291
USE_TRT_CONVERTER(anchor_generator);
Z
zlsh80826 已提交
2292
USE_TRT_CONVERTER(yolo_box);
2293
USE_TRT_CONVERTER(yolo_box_head);
2294
USE_TRT_CONVERTER(arg_max);
2295
USE_TRT_CONVERTER(roi_align);
2296
USE_TRT_CONVERTER(affine_channel);
Z
zlsh80826 已提交
2297
USE_TRT_CONVERTER(multiclass_nms);
2298
USE_TRT_CONVERTER(multiclass_nms3);
2299
USE_TRT_CONVERTER(nearest_interp);
2300
USE_TRT_CONVERTER(nearest_interp_v2);
2301
USE_TRT_CONVERTER(bilinear_interp_v2);
W
Wangzheee 已提交
2302
USE_TRT_CONVERTER(reshape);
2303
USE_TRT_CONVERTER(reshape2);
2304 2305
USE_TRT_CONVERTER(reduce_sum);
USE_TRT_CONVERTER(gather_nd);
W
wenbin 已提交
2306
USE_TRT_CONVERTER(reduce_mean);
W
wenbin 已提交
2307
USE_TRT_CONVERTER(tile);
W
wenbin 已提交
2308 2309
USE_TRT_CONVERTER(conv3d);
USE_TRT_CONVERTER(conv3d_transpose);
W
wangxinxin08 已提交
2310
USE_TRT_CONVERTER(mish);
W
wangxinxin08 已提交
2311
USE_TRT_CONVERTER(deformable_conv);
F
feng_shuai 已提交
2312
USE_TRT_CONVERTER(pool3d)
2313 2314
#ifdef _WIN32
#else
2315
USE_TRT_CONVERTER(fused_preln_embedding_eltwise_layernorm)
2316 2317
USE_TRT_CONVERTER(fused_embedding_eltwise_layernorm);
#endif
2318
USE_TRT_CONVERTER(preln_skip_layernorm)
2319 2320
USE_TRT_CONVERTER(preln_residual_bias)
USE_TRT_CONVERTER(c_allreduce_sum)
F
feng_shuai 已提交
2321
USE_TRT_CONVERTER(roll)
F
feng_shuai 已提交
2322
USE_TRT_CONVERTER(strided_slice)
Z
zhoutianzi666 已提交
2323 2324
USE_TRT_CONVERTER(rnn)
USE_TRT_CONVERTER(fill_constant_batch_size_like)
2325
USE_TRT_CONVERTER(transformer_input_convert)
C
ccrrong 已提交
2326
USE_TRT_CONVERTER(cast)
2327 2328
USE_TRT_CONVERTER(recover_padding)
USE_TRT_CONVERTER(remove_padding)
C
ccrrong 已提交
2329
USE_TRT_CONVERTER(equal);
2330 2331
USE_TRT_CONVERTER(top_k)
USE_TRT_CONVERTER(top_k_v2)
2332
USE_TRT_CONVERTER(range)
2333 2334
USE_TRT_CONVERTER(squeeze2)
USE_TRT_CONVERTER(unsqueeze2)
2335 2336
USE_TRT_CONVERTER(sum)
USE_TRT_CONVERTER(shape)
2337
USE_TRT_CONVERTER(fill_constant)
2338
USE_TRT_CONVERTER(fused_token_prune)
2339
USE_TRT_CONVERTER(celu)
W
wenbin 已提交
2340
USE_TRT_CONVERTER(layernorm_shift_partition)
W
Wang Bojun 已提交
2341
USE_TRT_CONVERTER(reverse_roll)
W
wenbin 已提交
2342
USE_TRT_CONVERTER(preln_layernorm_shift_partition)
W
Wang Bojun 已提交
2343
USE_TRT_CONVERTER(merge_layernorm)
W
wenbin 已提交
2344
USE_TRT_CONVERTER(skip_merge_layernorm)
W
weishengying 已提交
2345 2346
USE_TRT_CONVERTER(generic_plugin_creater)
USE_TRT_CONVERTER(custom_plugin_creater)
2347 2348
USE_TRT_CONVERTER(tanh_shrink)
USE_TRT_CONVERTER(logsigmoid)
2349
USE_TRT_CONVERTER(lookup_table)
2350
USE_TRT_CONVERTER(expand_v2)
2351
USE_TRT_CONVERTER(take_along_axis)
2352 2353 2354 2355
#if PADDLE_WITH_CUSPARSELT && IS_TRT_VERSION_GE(8000)
USE_TRT_CONVERTER(sparse_fc)
USE_TRT_CONVERTER(sparse_multihead_matmul)
#endif
2356
#endif
W
Wilber 已提交
2357 2358 2359 2360 2361 2362

namespace paddle_infer {

Predictor::Predictor(const Config &config) {
  const_cast<Config *>(&config)->SwitchUseFeedFetchOps(false);
  // The second parameter indicates that the discard log is not printed
2363 2364 2365 2366 2367 2368 2369 2370 2371 2372
  if (config.use_onnxruntime()) {
#ifdef PADDLE_WITH_ONNXRUNTIME
    if (config.use_gpu()) {
      LOG(WARNING) << "The current ONNXRuntime backend doesn't support GPU,"
                      "and it falls back to use Paddle Inference.";
    } else if (!paddle::CheckConvertToONNX(config)) {
      LOG(WARNING)
          << "Paddle2ONNX do't support convert the Model, fall back to using "
             "Paddle Inference.";
    } else {
C
ccrrong 已提交
2373 2374 2375 2376
      predictor_ =
          paddle::CreatePaddlePredictor<Config,
                                        paddle::PaddleEngineKind::kONNXRuntime>(
              config);
2377 2378 2379 2380 2381 2382 2383 2384 2385
      return;
    }
#else
    LOG(WARNING)
        << "The onnxruntime backend isn't enabled,"
           " and please re-compile Paddle with WITH_ONNXRUNTIME option,"
           "fall back to using Paddle Inference.";
#endif
  }
C
ccrrong 已提交
2386 2387 2388 2389
  predictor_ =
      paddle::CreatePaddlePredictor<Config,
                                    paddle::PaddleEngineKind::kAnalysis>(
          config);
W
Wilber 已提交
2390 2391 2392 2393 2394
}

std::vector<std::string> Predictor::GetInputNames() {
  return predictor_->GetInputNames();
}
2395 2396 2397 2398

std::map<std::string, DataType> Predictor::GetInputTypes() {
  return predictor_->GetInputTypes();
}
W
Wilber 已提交
2399 2400

std::unique_ptr<Tensor> Predictor::GetInputHandle(const std::string &name) {
2401
  return predictor_->GetInputTensor(name);
W
Wilber 已提交
2402 2403 2404 2405 2406 2407 2408
}

std::vector<std::string> Predictor::GetOutputNames() {
  return predictor_->GetOutputNames();
}

std::unique_ptr<Tensor> Predictor::GetOutputHandle(const std::string &name) {
2409
  return predictor_->GetOutputTensor(name);
W
Wilber 已提交
2410 2411 2412 2413
}

bool Predictor::Run() { return predictor_->ZeroCopyRun(); }

2414 2415
std::unique_ptr<Predictor> Predictor::Clone(void *stream) {
  auto analysis_pred = predictor_->Clone(stream);
W
Wilber 已提交
2416 2417 2418 2419 2420 2421 2422 2423
  std::unique_ptr<Predictor> pred(new Predictor(std::move(analysis_pred)));
  return pred;
}

void Predictor::ClearIntermediateTensor() {
  predictor_->ClearIntermediateTensor();
}

2424 2425
uint64_t Predictor::TryShrinkMemory() { return predictor_->TryShrinkMemory(); }

2426 2427 2428 2429
void Predictor::RegisterOutputHook(const Exp_OutputHookFunc &hookfunc) {
  predictor_->RegisterOutputHook(hookfunc);
}

2430 2431
void *Predictor::GetExecStream() const { return predictor_->GetExecStream(); }

W
Wilber 已提交
2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449
int GetNumBytesOfDataType(DataType dtype) {
  switch (dtype) {
    case DataType::FLOAT32:
      return sizeof(float);
    case DataType::INT64:
      return sizeof(int64_t);
    case DataType::INT32:
      return sizeof(int32_t);
    case DataType::UINT8:
      return sizeof(uint8_t);
    default:
      assert(false);
      return -1;
  }
}

std::string GetVersion() { return paddle::get_version(); }

2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465
std::tuple<int, int, int> GetTrtCompileVersion() {
#ifdef PADDLE_WITH_TENSORRT
  return paddle::inference::tensorrt::GetTrtCompileVersion();
#else
  return std::tuple<int, int, int>{0, 0, 0};
#endif
}

std::tuple<int, int, int> GetTrtRuntimeVersion() {
#ifdef PADDLE_WITH_TENSORRT
  return paddle::inference::tensorrt::GetTrtRuntimeVersion();
#else
  return std::tuple<int, int, int>{0, 0, 0};
#endif
}

W
Wilber 已提交
2466 2467 2468 2469
std::string UpdateDllFlag(const char *name, const char *value) {
  return paddle::UpdateDllFlag(name, value);
}

2470 2471 2472 2473 2474
void ConvertToMixedPrecision(const std::string &model_file,
                             const std::string &params_file,
                             const std::string &mixed_model_file,
                             const std::string &mixed_params_file,
                             PrecisionType mixed_precision,
2475
                             paddle_infer::PlaceType backend,
2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489
                             bool keep_io_types,
                             std::unordered_set<std::string> black_list) {
  auto phi_backend = paddle::ConvertBackend(backend);
  auto phi_precision = paddle::ConvertPrecision(mixed_precision);
  paddle::inference::analysis::ConvertToMixedPrecision(model_file,
                                                       params_file,
                                                       mixed_model_file,
                                                       mixed_params_file,
                                                       phi_precision,
                                                       phi_backend,
                                                       keep_io_types,
                                                       black_list);
}

W
Wilber 已提交
2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500
}  // namespace paddle_infer

namespace paddle_infer {
std::shared_ptr<Predictor> CreatePredictor(const Config &config) {  // NOLINT
  std::shared_ptr<Predictor> predictor(new Predictor(config));
  return predictor;
}

namespace services {
PredictorPool::PredictorPool(const Config &config, size_t size) {
  PADDLE_ENFORCE_GE(
C
ccrrong 已提交
2501 2502
      size,
      1UL,
W
Wilber 已提交
2503 2504 2505 2506 2507 2508 2509 2510
      paddle::platform::errors::InvalidArgument(
          "The predictor pool size should be greater than 1, but it's (%d)",
          size));
  Config copy_config(config);
  main_pred_.reset(new Predictor(config));
  for (size_t i = 0; i < size - 1; i++) {
    if (config.tensorrt_engine_enabled()) {
      Config config_tmp(copy_config);
2511
      preds_.emplace_back(new Predictor(config_tmp));
W
Wilber 已提交
2512
    } else {
2513
      preds_.emplace_back(main_pred_->Clone());
W
Wilber 已提交
2514 2515 2516 2517 2518 2519
    }
  }
}

Predictor *PredictorPool::Retrive(size_t idx) {
  PADDLE_ENFORCE_LT(
C
ccrrong 已提交
2520 2521
      idx,
      preds_.size() + 1,
W
Wilber 已提交
2522
      paddle::platform::errors::InvalidArgument(
C
ccrrong 已提交
2523 2524
          "There are (%d) predictors in the pool, but the idx is (%d)",
          idx,
W
Wilber 已提交
2525 2526 2527 2528 2529 2530 2531
          preds_.size() + 1));
  if (idx == 0) {
    return main_pred_.get();
  }
  return preds_[idx - 1].get();
}
}  // namespace services
W
Wilber 已提交
2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551

namespace experimental {

// Note: Can only be used under thread_local semantics.
bool InternalUtils::RunWithExternalStream(paddle_infer::Predictor *p,
                                          cudaStream_t stream) {
#ifdef PADDLE_WITH_CUDA
  auto pred = dynamic_cast<paddle::AnalysisPredictor *>(p->predictor_.get());
  return pred->ExpRunWithExternalStream(stream);
#endif
  return false;
}
bool InternalUtils::RunWithExternalStream(paddle_infer::Predictor *p,
                                          hipStream_t stream) {
#ifdef PADDLE_WITH_HIP
  auto pred = dynamic_cast<paddle::AnalysisPredictor *>(p->predictor_.get());
  return pred->ExpRunWithExternalStream(stream);
#endif
  return false;
}
W
Wilber 已提交
2552

2553 2554 2555 2556 2557 2558
void InternalUtils::UpdateConfigInterleaved(paddle_infer::Config *c,
                                            bool with_interleaved) {
#ifdef PADDLE_WITH_CUDA
  c->trt_with_interleaved_ = with_interleaved;
#endif
}
W
Wilber 已提交
2559

2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573
void InternalUtils::SetTransformerPosid(
    paddle_infer::Config *c, const std::string &tensorrt_transformer_posid) {
#ifdef PADDLE_WITH_CUDA
  c->tensorrt_transformer_posid_ = tensorrt_transformer_posid;
#endif
}

void InternalUtils::SetTransformerMaskid(
    paddle_infer::Config *c, const std::string &tensorrt_transformer_maskid) {
#ifdef PADDLE_WITH_CUDA
  c->tensorrt_transformer_maskid_ = tensorrt_transformer_maskid;
#endif
}

W
Wilber 已提交
2574 2575 2576 2577 2578
void InternalUtils::SyncStream(paddle_infer::Predictor *p) {
#ifdef PADDLE_WITH_CUDA
  auto *pred = dynamic_cast<paddle::AnalysisPredictor *>(p->predictor_.get());
  paddle::platform::DeviceContextPool &pool =
      paddle::platform::DeviceContextPool::Instance();
L
Leo Chen 已提交
2579
  auto *dev_ctx = reinterpret_cast<phi::GPUContext *>(pool.Get(pred->place_));
W
Wilber 已提交
2580 2581 2582 2583 2584 2585 2586 2587 2588
  cudaStreamSynchronize(dev_ctx->stream());
#endif
}
void InternalUtils::SyncStream(cudaStream_t stream) {
#ifdef PADDLE_WITH_CUDA
  cudaStreamSynchronize(stream);
#endif
}

W
Wilber 已提交
2589
}  // namespace experimental
W
Wilber 已提交
2590
}  // namespace paddle_infer