未验证 提交 2d57158e 编写于 作者: Y Yan Chunwei 提交者: GitHub

fea/init tensorrt engine (#10003)

上级 64babc9a
develop 2.0.1-rocm-post Ligoml-patch-1 OliverLPH-patch-1 OliverLPH-patch-2 PaddlePM-patch-1 PaddlePM-patch-2 ZHUI-patch-1 add_default_att add_model_benchmark_ci add_some_yaml_config addfile all_new_design_exec ascendrc ascendrelease cherry_undefined_var compile_windows delete_2.0.1-rocm-post delete_add_default_att delete_all_new_design_exec delete_ascendrc delete_compile_windows delete_delete_addfile delete_disable_iterable_dataset_unittest delete_fix_dataloader_memory_leak delete_fix_imperative_dygraph_error delete_fix_retry_ci delete_fix_undefined_var delete_improve_sccache delete_incubate/lite delete_paddle_tiny_install delete_paralleltest delete_prv-disable-more-cache delete_revert-31068-fix_conv3d_windows delete_revert-31562-mean delete_revert-33630-bug-fix delete_revert-34159-add_npu_bce_logical_dev delete_revert-34910-spinlocks_for_allocator delete_revert-35069-revert-34910-spinlocks_for_allocator delete_revert-36057-dev/read_flags_in_ut dingjiaweiww-patch-1 disable_iterable_dataset_unittest dy2static enable_eager_model_test final_state_gen_python_c final_state_intermediate fix-numpy-issue fix_concat_slice fix_dataloader_memory_leak fix_imperative_dygraph_error fix_npu_ci fix_op_flops fix_retry_ci fix_rnn_docs fix_tensor_type fix_undefined_var fixiscan fixiscan1 fixiscan2 fixiscan3 github/fork/123malin/netifaces github/fork/123malin/tdm_abacus github/fork/AshburnLee/dev_unique github/fork/ForFishes/fix_memory_matmul github/fork/ForFishes/rm_fluid github/fork/LielinJiang/move-2.0-api github/fork/LielinJiang/visual-dl-cb github/fork/LiuChiachi/add-transformer-generate-square-subsequent-mask-api github/fork/LiuChiachi/fix-example-code-for-hapi-Model github/fork/LiuChiachi/remove-input-requirment-in-dygraph-Model github/fork/MrChengmo/fix_ps_profiler github/fork/MrChengmo/update_ps_heter github/fork/PWhiddy/patch-1 github/fork/Shixiaowei02/dev/save_load_upgrade github/fork/TCChenlong/fix_hapi github/fork/TCChenlong/fix_inden github/fork/Thunderbrook/xpu_slice github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var_2 github/fork/XieYunshen/disable_ut_test_parallel_executor_fetch_isolated_var_3 github/fork/XieYunshen/timeout_20S_ut github/fork/ZeyuChen/remove-nltk github/fork/arlesniak/arlesniak/selective__mkldnn_flags github/fork/baiyfbupt/code_doc_mig github/fork/chalsliu/set_timeout github/fork/chen-zhiyu/develop github/fork/chenwhql/ci/try_to_find_test_buffer_shared_memory_reuse_pass_error github/fork/chenwhql/dygraph/remove_scale_loss_and_apply_collective_grads github/fork/chenwhql/saveload/add_get_inference_program github/fork/chenwhql/saveload/remove_save_load_config github/fork/cryoco/pass-compatibility-trt github/fork/danleifeng/isempty_api2.0 github/fork/frankwhzhang/api_transfer github/fork/hbwx24/error_msg/cuda_kernel_error_msg github/fork/heavengate/cherry_yolo_box github/fork/heavengate/update_yolo_box github/fork/iclementine/rnn_fix github/fork/iducn/testestse github/fork/jczaja/prv-25537-fix github/fork/jeff41404/release/1.8 github/fork/jiweibo/api_2.0 github/fork/jiweibo/fix_lite_resnet50_test github/fork/juncaipeng/fix_doc_1 github/fork/lfchener/sample_code github/fork/littletomatodonkey/fix_reg_doc github/fork/liym27/dy2stat_update_assign_to_rc20 github/fork/luotao1/profiler_ut github/fork/mapingshuo/add_wait github/fork/mapingshuo/doc_2.0 github/fork/mapingshuo/zero-0.5 github/fork/miraiwk/dev github/fork/pangyoki/add-Categorical-class-branch github/fork/pangyoki/add-multinomial-op-branch github/fork/pangyoki/fix-test_distritbution-CI github/fork/qjing666/doublegrad github/fork/qjing666/fix_hdfs_download github/fork/sandyhouse/add_gather_etc github/fork/sandyhouse/add_send_recv_alltoall_etc github/fork/sandyhouse/pipeline_exe_run github/fork/seiriosPlus/feature/large_scale_kv_save_delta github/fork/seiriosPlus/fix/paddle_errors_fix github/fork/seiriosPlus/fix/paddle_op_errors github/fork/shangzhizhou/fix_test_activation_op_random_bug github/fork/smallv0221/yxp0924 github/fork/smallv0221/yxp0925 github/fork/swtkiwi/del-matplotlib github/fork/tianshuo78520a/kunlun_test github/fork/tianshuo78520a/update_dockerfile github/fork/wanghaoshuang/bert_fuse github/fork/wanghaoshuang/label_smooth github/fork/wanghuancoder/develop_CUDASynchronize github/fork/wanghuancoder/develop_Layer_doc github/fork/wanghuancoder/develop_ParameterList_doc github/fork/wanghuancoder/develop_Sequential_doc github/fork/wanghuancoder/develop_bilinear_tensor_product github/fork/wanghuancoder/develop_coverage_build_sh github/fork/wanghuancoder/develop_in_dynamic_mode_doc github/fork/wanghuancoder/develop_unique_name_doc github/fork/wangxicoding/fleet_meta_combine github/fork/wawltor/error_message_fix_5 github/fork/willthefrog/remove_l2_norm github/fork/windstamp/momentum_op github/fork/windstamp/mv_op_5 github/fork/windstamp/normal_api github/fork/wojtuss/wojtuss/fusion_gru_quantization github/fork/wojtuss/wojtuss/quantization-with-shift github/fork/wzzju/fix_err_info github/fork/wzzju/pure_fp16 github/fork/xiemoyuan/op_error_message github/fork/xiemoyuan/optimize_error_message github/fork/yaoxuefeng6/fix_doc github/fork/yaoxuefeng6/mod_dataset_v2 github/fork/yongqiangma/lod github/fork/ysh329/fix-clip-by-norm-error github/fork/ysh329/fix-error-clip-by-value github/fork/yukavio/error_info github/fork/zhangting2020/conv_filter_grad github/fork/zhangting2020/is_compile_with_cuda github/fork/zhangting2020/place_doc github/fork/zhangting2020/program github/fork/zhhsplendid/fix_any github/fork/zhhsplendid/refine_api2 github/fork/zhhsplendid/refine_api2_test github/fork/zhhsplendid/refine_api_test_ptb_lm github/fork/zhhsplendid/refine_api_test_resnet github/fork/zhhsplendid/refine_api_test_simnet github/fork/zhiqiu/dev/refine_initializer github/fork/zhiqiu/dev/remove_inplace_argument github/fork/zlsh80826/nvinfer_plugin_var_len_cuda11 improve_sccache incubate/infrt incubate/lite inplace_addto make_flag_adding_easier master move_embedding_to_phi move_histogram_to_pten move_sgd_to_phi move_slice_to_pten move_temporal_shift_to_phi move_yolo_box_to_phi npu_fix_alloc numel paddle_tiny_install paralleltest preln_ernie prv-disable-more-cache prv-md-even-more prv-onednn-2.5 pten_tensor_refactor release/0.12.0 release/0.13.0 release/0.14.0 release/0.15.0 release/1.0.0 release/1.1 release/1.2 release/1.3 release/1.4 release/1.5 release/1.6 release/1.7 release/1.8 release/2.0 release/2.0-alpha release/2.0-beta release/2.0-rc release/2.0-rc1 release/2.1 release/2.2 release/2.3 release/2.3-fc-ernie-fix release/2.4 release/lite-0.1 revert-24981-add_device_attr_for_regulization revert-26856-strategy_example2 revert-27520-disable_pr revert-31068-fix_conv3d_windows revert-31562-mean revert-32290-develop-hardlabel revert-33037-forci revert-33475-fix_cifar_label_dimension revert-33630-bug-fix revert-34159-add_npu_bce_logical_dev revert-34406-add_copy_from_tensor revert-34910-spinlocks_for_allocator revert-35069-revert-34910-spinlocks_for_allocator revert-36057-dev/read_flags_in_ut revert-36201-refine_fast_threaded_ssa_graph_executor revert-36985-add_license revert-37318-refactor_dygraph_to_eager revert-37926-eager_coreops_500 revert-37956-revert-37727-pylayer_support_tuple revert-38100-mingdong revert-38301-allocation_rearrange_pr revert-38703-numpy_bf16_package_reupload revert-38732-remove_useless_header_in_elementwise_mul_grad revert-38959-Reduce_Grad revert-39143-adjust_empty revert-39227-move_trace_op_to_pten revert-39268-dev/remove_concat_fluid_kernel revert-40170-support_partial_grad revert-41056-revert-40727-move_some_activaion_to_phi revert-41065-revert-40993-mv_ele_floordiv_pow revert-41068-revert-40790-phi_new revert-41944-smaller_inference_api_test revert-42149-do-not-reset-default-stream-for-stream-safe-cuda-allocator revert-43155-fix_ut_tempfile revert-43882-revert-41944-smaller_inference_api_test revert-45808-phi/simplify_size_op revert-46827-deform_comment rocm_dev_0217 support_weight_transpose test_benchmark_ci test_feature_precision_test_c test_model_benchmark test_model_benchmark_ci zhiqiu-patch-1 v2.4.0-rc0 v2.3.2 v2.3.1 v2.3.0 v2.3.0-rc0 v2.2.2 v2.2.1 v2.2.0 v2.2.0-rc0 v2.2.0-bak0 v2.1.3 v2.1.2 v2.1.1 v2.1.0 v2.1.0-rc0 v2.0.2 v2.0.1 v2.0.0 v2.0.0-rc1 v2.0.0-rc0 v2.0.0-beta0 v2.0.0-alpha0 v1.8.5 v1.8.4 v1.8.3 v1.8.2 v1.8.1 v1.8.0 v1.7.2 v1.7.1 v1.7.0 v1.6.3 v1.6.2 v1.6.1 v1.6.0 v1.6.0-rc0 v1.5.2 v1.5.1 v1.5.0 v1.4.1 v1.4.0 v1.3.2 v1.3.1 v1.3.0 v1.2.1 v1.2.0 v1.1.0 v1.0.2 v1.0.1 v1.0.0 v1.0.0-rc0 v0.15.0 v0.15.0-rc0 v0.14.0 v0.13.0 v0.12.0 lite-v0.1
无相关合并请求
/* 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. */
#pragma once
#include "paddle/fluid/framework/framework.pb.h"
namespace paddle {
namespace inference {
/*
* EngineBase is the base class of all inference engines. An inference engine
* takes a paddle program as input, and outputs the result in fluid Tensor
* format. It can be used to optimize performance of computation sub-blocks, for
* example, break down the original block into sub-blocks and execute each
* sub-blocks in different engines.
*
* For example:
* When inference, the resnet50 model can put most of the model into subgraph
* and run it on a TensorRT engine.
*
* There are several engines such as TensorRT and other frameworks, so an
* EngineBase is put forward to give an unified interface for all the
* different engine implemention.
*/
class EngineBase {
public:
using DescType = ::paddle::framework::proto::BlockDesc;
// Build the model and do some preparation, for example, in TensorRT, run
// createInferBuilder, buildCudaEngine.
virtual void Build(const DescType& paddle_model) = 0;
// Execute the engine, that will run the inference network.
virtual void Execute(int batch_size) = 0;
virtual ~EngineBase() {}
}; // class EngineBase
} // namespace inference
} // namespace paddle
nv_test(test_tensorrt SRCS test_tensorrt.cc DEPS dynload_cuda device_context dynamic_loader)
if(WITH_TESTING)
nv_test(test_tensorrt SRCS test_tensorrt.cc DEPS dynload_cuda device_context dynamic_loader)
nv_test(test_tensorrt_engine SRCS test_engine.cc engine.cc DEPS dynload_cuda)
endif()
/* 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. */
#include "paddle/fluid/inference/tensorrt/engine.h"
#include <NvInfer.h>
#include <cuda.h>
#include <glog/logging.h>
#include "paddle/fluid/inference/tensorrt/helper.h"
#include "paddle/fluid/platform/enforce.h"
namespace paddle {
namespace inference {
namespace tensorrt {
void TensorRTEngine::Build(const DescType& paddle_model) {
PADDLE_ENFORCE(false, "not implemented");
}
void TensorRTEngine::Execute(int batch_size) {
infer_context_->enqueue(batch_size, buffers_.data(), *stream_, nullptr);
cudaStreamSynchronize(*stream_);
}
TensorRTEngine::~TensorRTEngine() {
// clean buffer
for (auto& buffer : buffers_) {
if (buffer != nullptr) {
PADDLE_ENFORCE_EQ(0, cudaFree(buffer));
buffer = nullptr;
}
}
}
void TensorRTEngine::FreezeNetwork() {
PADDLE_ENFORCE(infer_builder_ != nullptr,
"Call InitNetwork first to initialize network.");
PADDLE_ENFORCE(infer_network_ != nullptr,
"Call InitNetwork first to initialize network.");
// build engine.
infer_builder_->setMaxBatchSize(max_batch_);
infer_builder_->setMaxWorkspaceSize(max_workspace_);
infer_engine_.reset(infer_builder_->buildCudaEngine(*infer_network_));
PADDLE_ENFORCE(infer_engine_ != nullptr, "build cuda engine failed!");
infer_context_.reset(infer_engine_->createExecutionContext());
// allocate GPU buffers.
buffers_.resize(buffer_sizes_.size(), nullptr);
for (auto& item : buffer_sizes_) {
if (item.second == 0) {
auto slot_offset = infer_engine_->getBindingIndex(item.first.c_str());
item.second = kDataTypeSize[static_cast<int>(
infer_engine_->getBindingDataType(slot_offset))] *
AccumDims(infer_engine_->getBindingDimensions(slot_offset));
}
PADDLE_ENFORCE_EQ(0, cudaMalloc(&buffer(item.first), item.second));
}
}
nvinfer1::ITensor* TensorRTEngine::DeclareInput(const std::string& name,
nvinfer1::DataType dtype,
const nvinfer1::Dims& dim) {
PADDLE_ENFORCE_EQ(0, buffer_sizes_.count(name), "duplicate input name %s",
name);
PADDLE_ENFORCE(infer_network_ != nullptr, "should initnetwork first");
auto* input = infer_network_->addInput(name.c_str(), dtype, dim);
PADDLE_ENFORCE(input, "infer network add input %s failed", name);
buffer_sizes_[name] = kDataTypeSize[static_cast<int>(dtype)] * AccumDims(dim);
return input;
}
void TensorRTEngine::DeclareOutput(const nvinfer1::ILayer* layer, int offset,
const std::string& name) {
PADDLE_ENFORCE_EQ(0, buffer_sizes_.count(name), "duplicate output name %s",
name);
auto* output = layer->getOutput(offset);
PADDLE_ENFORCE(output != nullptr);
output->setName(name.c_str());
infer_network_->markOutput(*output);
// output buffers' size can only be decided latter, set zero here to mark this
// and will reset latter.
buffer_sizes_[name] = 0;
}
void* TensorRTEngine::GetOutputInGPU(const std::string& name) {
return buffer(name);
}
void TensorRTEngine::GetOutputInCPU(const std::string& name, void* dst,
size_t max_size) {
// determine data size
auto it = buffer_sizes_.find(name);
PADDLE_ENFORCE(it != buffer_sizes_.end());
PADDLE_ENFORCE_GT(it->second, 0);
PADDLE_ENFORCE_GE(max_size, it->second);
PADDLE_ENFORCE_EQ(0, cudaMemcpyAsync(dst, buffer(name), it->second,
cudaMemcpyDeviceToHost, *stream_));
}
void*& TensorRTEngine::buffer(const std::string& name) {
PADDLE_ENFORCE(infer_engine_ != nullptr, "call FreezeNetwork first.");
auto it = buffer_sizes_.find(name);
PADDLE_ENFORCE(it != buffer_sizes_.end());
auto slot_offset = infer_engine_->getBindingIndex(name.c_str());
return buffers_[slot_offset];
}
void TensorRTEngine::SetInputFromCPU(const std::string& name, void* data,
size_t size) {
void* buf = buffer(name);
PADDLE_ENFORCE_EQ(
0, cudaMemcpyAsync(buf, data, size, cudaMemcpyHostToDevice, *stream_));
}
} // namespace tensorrt
} // namespace inference
} // namespace paddle
/* 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. */
#pragma once
#include <NvInfer.h>
#include <memory>
#include <unordered_map>
#include "paddle/fluid/inference/engine.h"
#include "paddle/fluid/inference/tensorrt/helper.h"
namespace paddle {
namespace inference {
namespace tensorrt {
/*
* TensorRT Engine.
*
* There are two alternative ways to use it, one is to build from a paddle
* protobuf model, another way is to manully construct the network.
*/
class TensorRTEngine : public EngineBase {
public:
// Weight is model parameter.
class Weight {
public:
Weight(nvinfer1::DataType dtype, void* value, int num_elem) {
w_.type = dtype;
w_.values = value;
w_.count = num_elem;
}
const nvinfer1::Weights& get() { return w_; }
private:
nvinfer1::Weights w_;
};
TensorRTEngine(int max_batch, int max_workspace, cudaStream_t* stream,
nvinfer1::ILogger& logger = NaiveLogger::Global())
: max_batch_(max_batch),
max_workspace_(max_workspace),
stream_(stream),
logger_(logger) {}
virtual ~TensorRTEngine();
// TODO(Superjomn) implement it later when graph segmentation is supported.
virtual void Build(const DescType& paddle_model) override;
virtual void Execute(int batch_size) override;
// Initialize the inference network, so that TensorRT layers can add to this
// network.
void InitNetwork() {
infer_builder_.reset(createInferBuilder(logger_));
infer_network_.reset(infer_builder_->createNetwork());
}
// After finishing adding ops, freeze this network and creates the executation
// environment.
void FreezeNetwork();
// Add an input and set its name, data type and dimention.
nvinfer1::ITensor* DeclareInput(const std::string& name,
nvinfer1::DataType dtype,
const nvinfer1::Dims& dim);
// Set the offset-th output from a layer as the network's output, and set its
// name.
void DeclareOutput(const nvinfer1::ILayer* layer, int offset,
const std::string& name);
// GPU memory address for an ITensor with specific name. One can operate on
// these memory directly for acceleration, for example, output the converted
// data directly to the buffer to save data copy overhead.
// NOTE this should be used after calling `FreezeNetwork`.
void*& buffer(const std::string& name);
// Fill an input from CPU memory with name and size.
void SetInputFromCPU(const std::string& name, void* data, size_t size);
// TODO(Superjomn) is this method necessary given that buffer(xxx) can be
// accessed directly. Fill an input from GPU memory with name and size.
void SetInputFromGPU(const std::string& name, void* data, size_t size);
// Get an output called name, the output of tensorrt is in GPU, so this method
// will just return the output's GPU memory address.
void* GetOutputInGPU(const std::string& name);
// LOW EFFICENCY! Get output to CPU, this will trigger a memory copy from GPU
// to CPU.
void GetOutputInCPU(const std::string& name, void* dst, size_t max_size);
nvinfer1::ICudaEngine* engine() { return infer_engine_.get(); }
nvinfer1::INetworkDefinition* network() { return infer_network_.get(); }
private:
// the max batch size
int max_batch_;
// the max memory size the engine uses
int max_workspace_;
cudaStream_t* stream_;
nvinfer1::ILogger& logger_;
std::vector<void*> buffers_;
// max data size for the buffers.
std::unordered_map<std::string /*name*/, size_t /*max size*/> buffer_sizes_;
// TensorRT related internal members
template <typename T>
struct Destroyer {
void operator()(T* x) { x->destroy(); }
};
template <typename T>
using infer_ptr = std::unique_ptr<T, Destroyer<T>>;
infer_ptr<nvinfer1::IBuilder> infer_builder_;
infer_ptr<nvinfer1::INetworkDefinition> infer_network_;
infer_ptr<nvinfer1::ICudaEngine> infer_engine_;
infer_ptr<nvinfer1::IExecutionContext> infer_context_;
}; // class TensorRTEngine
// Add an layer__ into engine__ with args ARGS.
// For example:
// TRT_ENGINE_ADD_LAYER(xxx, FullyConnected, input, dim, weights, bias)
//
// Reference
// https://docs.nvidia.com/deeplearning/sdk/tensorrt-developer-guide/index.html#charRNN_define_network
//
// will add a fully connected layer into the engine.
// TensorRT has too many layers, so that is not wise to add member functions for
// them, and an macro like this is more extensible when underlying TensorRT
// library add new layer supports.
#define TRT_ENGINE_ADD_LAYER(engine__, layer__, ARGS...) \
engine__->network()->add##layer__(ARGS);
} // namespace tensorrt
} // namespace inference
} // namespace paddle
/* 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. */
#pragma once
#include <NvInfer.h>
#include <cuda.h>
#include <glog/logging.h>
#include "paddle/fluid/platform/dynload/tensorrt.h"
#include "paddle/fluid/platform/enforce.h"
namespace paddle {
namespace inference {
namespace tensorrt {
namespace dy = paddle::platform::dynload;
static size_t AccumDims(nvinfer1::Dims dims) {
size_t num = dims.nbDims == 0 ? 0 : 1;
for (int i = 0; i < dims.nbDims; i++) {
PADDLE_ENFORCE_GT(dims.d[i], 0);
num *= dims.d[i];
}
return num;
}
// TensorRT data type to size
const int kDataTypeSize[] = {
4, // kFLOAT
2, // kHALF
1, // kINT8
4 // kINT32
};
// The following two API are implemented in TensorRT's header file, cannot load
// from the dynamic library. So create our own implementation and directly
// trigger the method from the dynamic library.
static nvinfer1::IBuilder* createInferBuilder(nvinfer1::ILogger& logger) {
return static_cast<nvinfer1::IBuilder*>(
dy::createInferBuilder_INTERNAL(&logger, NV_TENSORRT_VERSION));
}
static nvinfer1::IRuntime* createInferRuntime(nvinfer1::ILogger& logger) {
return static_cast<nvinfer1::IRuntime*>(
dy::createInferRuntime_INTERNAL(&logger, NV_TENSORRT_VERSION));
}
// A logger for create TensorRT infer builder.
class NaiveLogger : public nvinfer1::ILogger {
public:
void log(nvinfer1::ILogger::Severity severity, const char* msg) override {
switch (severity) {
case Severity::kINFO:
LOG(INFO) << msg;
break;
case Severity::kWARNING:
LOG(WARNING) << msg;
break;
case Severity::kINTERNAL_ERROR:
case Severity::kERROR:
LOG(ERROR) << msg;
break;
default:
break;
}
}
static nvinfer1::ILogger& Global() {
static nvinfer1::ILogger* x = new NaiveLogger;
return *x;
}
virtual ~NaiveLogger() override {}
};
} // namespace tensorrt
} // namespace inference
} // namespace paddle
/* 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. */
#include "paddle/fluid/inference/tensorrt/engine.h"
#include <cuda.h>
#include <cuda_runtime_api.h>
#include <glog/logging.h>
#include <gtest/gtest.h>
#include "paddle/fluid/platform/enforce.h"
namespace paddle {
namespace inference {
namespace tensorrt {
class TensorRTEngineTest : public ::testing::Test {
protected:
void SetUp() override {
ASSERT_EQ(0, cudaStreamCreate(&stream_));
engine_ = new TensorRTEngine(1, 1 << 10, &stream_);
engine_->InitNetwork();
}
void TearDown() override {
delete engine_;
cudaStreamDestroy(stream_);
}
protected:
TensorRTEngine* engine_;
cudaStream_t stream_;
};
TEST_F(TensorRTEngineTest, add_layer) {
const int size = 1;
float raw_weight[size] = {2.}; // Weight in CPU memory.
float raw_bias[size] = {3.};
LOG(INFO) << "create weights";
TensorRTEngine::Weight weight(nvinfer1::DataType::kFLOAT, raw_weight, size);
TensorRTEngine::Weight bias(nvinfer1::DataType::kFLOAT, raw_bias, size);
auto* x = engine_->DeclareInput("x", nvinfer1::DataType::kFLOAT,
nvinfer1::DimsCHW{1, 1, 1});
auto* fc_layer = TRT_ENGINE_ADD_LAYER(engine_, FullyConnected, *x, size,
weight.get(), bias.get());
PADDLE_ENFORCE(fc_layer != nullptr);
engine_->DeclareOutput(fc_layer, 0, "y");
LOG(INFO) << "freeze network";
engine_->FreezeNetwork();
ASSERT_EQ(engine_->engine()->getNbBindings(), 2);
// fill in real data
float x_v = 1234;
engine_->SetInputFromCPU("x", (void*)&x_v, 1 * sizeof(float));
LOG(INFO) << "to execute";
engine_->Execute(1);
LOG(INFO) << "to get output";
// void* y_v =
float y_cpu;
engine_->GetOutputInCPU("y", &y_cpu, sizeof(float));
LOG(INFO) << "to checkout output";
ASSERT_EQ(y_cpu, x_v * 2 + 3);
}
} // namespace tensorrt
} // namespace inference
} // namespace paddle
/* 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
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
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. */
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. */
#include <glog/logging.h>
#include <gtest/gtest.h>
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册
反馈
建议
客服 返回
顶部