From 5d72c5af87c3dd054bd6aff170442d214edbfc6e Mon Sep 17 00:00:00 2001 From: Vineet <39617050+makaveli10@users.noreply.github.com> Date: Thu, 1 Apr 2021 17:51:29 +0530 Subject: [PATCH] Add DenseNet121 (#460) * add: densenet * update readme --- densenet/CMakeLists.txt | 36 +++ densenet/README.md | 38 +++ densenet/densenet121.cpp | 404 +++++++++++++++++++++++++++++++ densenet/logging.h | 507 +++++++++++++++++++++++++++++++++++++++ 4 files changed, 985 insertions(+) create mode 100644 densenet/CMakeLists.txt create mode 100644 densenet/README.md create mode 100644 densenet/densenet121.cpp create mode 100644 densenet/logging.h diff --git a/densenet/CMakeLists.txt b/densenet/CMakeLists.txt new file mode 100644 index 0000000..a6f55fc --- /dev/null +++ b/densenet/CMakeLists.txt @@ -0,0 +1,36 @@ +cmake_minimum_required(VERSION 3.10) + +# set the project name +project(densenet) + +add_definitions(-std=c++11) + +# get main project dir to include common files +get_filename_component(MAIN_DIR ../ ABSOLUTE) + +# When enabled the static version of the +# CUDA runtime library will be used in CUDA_LIBRARIES +option(CUDA_USE_STATIC_CUDA_RUNTIME OFF) + +# specify the C++ standard +set(CMAKE_CXX_STANDARD 11) +set(CMAKE_CXX_STANDARD_REQUIRED True) +set(CMAKE_BUILD_TYPE Debug) + +# include + +# include and link cuda +include_directories(/usr/local/cuda/include) +link_directories(/usr/local/cuda/lib64) + +# include and link tensorrt +include_directories(/usr/include/x86_64-linux-gnu) +link_directories(/usr/lib/x86_64-linux-gnu) + +# add the executable +add_executable(densenet ${PROJECT_SOURCE_DIR}/densenet121.cpp) + +target_link_libraries(densenet nvinfer) +target_link_libraries(densenet cudart) + +add_definitions(-O2 -pthread) \ No newline at end of file diff --git a/densenet/README.md b/densenet/README.md new file mode 100644 index 0000000..3282287 --- /dev/null +++ b/densenet/README.md @@ -0,0 +1,38 @@ +# Densenet121 + +The Pytorch implementation is [makaveli10/densenet](https://github.com/makaveli10/torchtrtz/tree/main/densenet). +The tensorrt implemenation is taken from [makaveli10/cpptensorrtz](https://github.com/makaveli10/cpptensorrtz/). + +## How to Run + +1. generate densenet121.wts from pytorch + +``` +git clone https://github.com/wang-xinyu/tensorrtx.git +git clone https://github.com/makaveli10/torchtrtz.git + +// go to torchtrtz/densenet +// Enter these two commands to create densenet121.wts + $ python models.py + $ python gen_trtwts.py +``` + +2. build densenet and run + +``` +// put densenet121.wts into tensorrtx/densenet +// go to tensorrtx/densenet +mkdir build +cd build +cmake .. +make +sudo ./densenet -s // serialize model to file i.e. 'LPRnet.engine' +sudo ./densenet -d // deserialize model and run inference +``` + +3. Verify output from [torch impl](https://github.com/makaveli10/torchtrtz/blob/main/densenet/README.md) +TensorRT output[:5]: +``` + [-0.587389, -0.329202, -1.83404, -1.89935, -0.928404] +``` + diff --git a/densenet/densenet121.cpp b/densenet/densenet121.cpp new file mode 100644 index 0000000..c32ec4e --- /dev/null +++ b/densenet/densenet121.cpp @@ -0,0 +1,404 @@ +#include "NvInfer.h" +#include "cuda_runtime_api.h" +#include "logging.h" +#include +#include +#include +#include +#include +#include +#include + +#define CHECK(status) \ + do\ + {\ + auto ret = (status);\ + if (ret != 0)\ + {\ + std::cerr << "Cuda failure: " << ret << std::endl;\ + abort();\ + }\ + } while (0) + +// stuff we know about the network and the input/output blobs +static const int INPUT_H = 224; +static const int INPUT_W = 224; +static const int OUTPUT_SIZE = 1000; + +const char* INPUT_BLOB_NAME = "data"; +const char* OUTPUT_BLOB_NAME = "prob"; + +using namespace nvinfer1; + +static Logger gLogger; + +// Load weights from files shared with TensorRT samples. +// TensorRT weight files have a simple space delimited format: +// [type] [size] +std::map loadWeights(const std::string file) +{ + std::cout << "Loading weights: " << file << std::endl; + std::map weightMap; + + // Open weights file + std::ifstream input(file); + assert(input.is_open() && "Unable to load weight file."); + + // Read number of weight blobs + int32_t count; + input >> count; + assert(count > 0 && "Invalid weight map file."); + + while (count--) + { + Weights wt{DataType::kFLOAT, nullptr, 0}; + uint32_t size; + + // Read name and type of blob + std::string name; + input >> name >> std::dec >> size; + wt.type = DataType::kFLOAT; + + // Load blob + uint32_t* val = reinterpret_cast(malloc(sizeof(val) * size)); + for (uint32_t x = 0, y = size; x < y; ++x) + { + input >> std::hex >> val[x]; + } + wt.values = val; + + wt.count = size; + weightMap[name] = wt; + } + + return weightMap; +} + +IScaleLayer* addBatchNorm2d(INetworkDefinition *network, std::map& weightMap, ITensor& input, std::string lname, float eps) { + float *gamma = (float*)weightMap[lname + ".weight"].values; + float *beta = (float*)weightMap[lname + ".bias"].values; + float *mean = (float*)weightMap[lname + ".running_mean"].values; + float *var = (float*)weightMap[lname + ".running_var"].values; + int len = weightMap[lname + ".running_var"].count; + std::cout << "len " << len << std::endl; + + float *scval = reinterpret_cast(malloc(sizeof(float) * len)); + for (int i = 0; i < len; i++) { + scval[i] = gamma[i] / sqrt(var[i] + eps); + } + Weights scale{DataType::kFLOAT, scval, len}; + + float *shval = reinterpret_cast(malloc(sizeof(float) * len)); + for (int i = 0; i < len; i++) { + shval[i] = beta[i] - mean[i] * gamma[i] / sqrt(var[i] + eps); + } + Weights shift{DataType::kFLOAT, shval, len}; + + float *pval = reinterpret_cast(malloc(sizeof(float) * len)); + for (int i = 0; i < len; i++) { + pval[i] = 1.0; + } + Weights power{DataType::kFLOAT, pval, len}; + + weightMap[lname + ".scale"] = scale; + weightMap[lname + ".shift"] = shift; + weightMap[lname + ".power"] = power; + IScaleLayer* scale_1 = network->addScale(input, ScaleMode::kCHANNEL, shift, scale, power); + assert(scale_1); + return scale_1; +} + +IConvolutionLayer* addDenseLayer(INetworkDefinition* network, ITensor* input, std::map& weightMap, std::string lname, float eps) +{ + // add Batchnorm + IScaleLayer* bn1 = addBatchNorm2d(network, weightMap, *input, lname + ".norm1", eps); + + // add relu + IActivationLayer* relu1 = network -> addActivation(*bn1->getOutput(0), ActivationType::kRELU); + assert(relu1); + + // add conv + Weights emptywts{DataType::kFLOAT, nullptr, 0}; + IConvolutionLayer* conv1 = network -> addConvolutionNd(*relu1->getOutput(0), 128, DimsHW{1, 1}, weightMap[lname + ".conv1.weight"], emptywts); + assert(conv1); + conv1 -> setStrideNd(DimsHW{1, 1}); + + // add Batchnorm + IScaleLayer* bn2 = addBatchNorm2d(network, weightMap, *conv1 -> getOutput(0), lname + ".norm2", eps); + + // add relu + IActivationLayer* relu2 = network -> addActivation(*bn2->getOutput(0), ActivationType::kRELU); + assert(relu2); + + // add conv + IConvolutionLayer* conv2 = network -> addConvolutionNd(*relu2->getOutput(0), 32, DimsHW{3, 3}, weightMap[lname + ".conv2.weight"], emptywts); + assert(conv2); + conv2 -> setStrideNd(DimsHW{1, 1}); + conv2 -> setPaddingNd(DimsHW{1, 1}); + return conv2; +} + + +IPoolingLayer* addTransition(INetworkDefinition* network, ITensor& input, std::map& weightMap, int outch, std::string lname, float eps) +{ + // add batch norm + IScaleLayer* bn1 = addBatchNorm2d(network, weightMap,input, lname + ".norm", eps); + + // add relu activation + IActivationLayer* relu1 = network -> addActivation(*bn1->getOutput(0), ActivationType::kRELU); + assert(relu1); + + // add convolution layer + // empty weights for no bias + Weights emptywts{DataType::kFLOAT, nullptr, 0}; + IConvolutionLayer* conv1 = network -> addConvolutionNd(*relu1->getOutput(0), outch, DimsHW{1, 1}, weightMap[lname + ".conv.weight"], emptywts); + assert(conv1); + conv1 -> setStrideNd(DimsHW{1, 1}); + + // add pooling + IPoolingLayer* pool1 = network->addPoolingNd(*conv1->getOutput(0), PoolingType::kAVERAGE, DimsHW{2, 2}); + assert(pool1); + pool1 -> setStrideNd(DimsHW{2, 2}); + pool1 -> setPaddingNd(DimsHW{0,0}); + return pool1; +} + + +IConcatenationLayer* addDenseBlock(INetworkDefinition* network, ITensor* input, std::map& weightMap, int numDenseLayers, std::string lname, float eps) +{ + IConvolutionLayer* c{nullptr}; + IConcatenationLayer* concat{nullptr}; + ITensor* inputTensors[numDenseLayers+1]; + inputTensors[0] = input; + + c = addDenseLayer(network, input, weightMap, lname + ".denselayer" + std::to_string(1), eps); + int i; + for(i=1; i getOutput(0); + concat = network -> addConcatenation(inputTensors, i+1); + assert(concat); + c = addDenseLayer(network, concat->getOutput(0), weightMap, lname + ".denselayer" + std::to_string(i+1), eps); + } + inputTensors[numDenseLayers] = c -> getOutput(0); + concat = network -> addConcatenation(inputTensors, numDenseLayers+1); + assert(concat); + return concat; +} + + +/** + * Uses the TensorRT API to create the network engine. +**/ +ICudaEngine* createEngine(unsigned int maxBatchSize, IBuilder* builder, IBuilderConfig* config, DataType dt) +{ + // Initialize NetworkDefinition + INetworkDefinition* network = builder -> createNetworkV2(0U); + + auto data = network -> addInput(INPUT_BLOB_NAME, dt, Dims3{3, INPUT_H, INPUT_W}); + assert(data); + + std::map weightMap = loadWeights("../densenet121.wts"); + Weights emptywts{DataType::kFLOAT, nullptr, 0}; + auto conv0 = network -> addConvolutionNd(*data, 64, DimsHW{7, 7}, weightMap["features.conv0.weight"], emptywts); + assert(conv0); + conv0 -> setStrideNd(DimsHW{2, 2}); + conv0 -> setPaddingNd(DimsHW{3, 3}); + + auto norm0 = addBatchNorm2d(network, weightMap, *conv0 -> getOutput(0), "features.norm0", 1e-5); + + auto relu0 = network -> addActivation(*norm0 -> getOutput(0), ActivationType::kRELU); + assert(relu0); + + auto pool0 = network -> addPoolingNd(*relu0 -> getOutput(0), PoolingType::kMAX, DimsHW{3, 3}); + assert(pool0); + pool0 -> setStrideNd(DimsHW{2, 2}); + pool0 -> setPaddingNd(DimsHW{1, 1}); + + auto dense1 = addDenseBlock(network, pool0 -> getOutput(0), weightMap, 6, "features.denseblock1", 1e-5); + auto transition1 = addTransition(network, *dense1 -> getOutput(0), weightMap, 128, "features.transition1", 1e-5); + + auto dense2 = addDenseBlock(network, transition1 -> getOutput(0), weightMap, 12, "features.denseblock2", 1e-5); + auto transition2 = addTransition(network, *dense2 -> getOutput(0), weightMap, 256, "features.transition2", 1e-5); + + auto dense3 = addDenseBlock(network, transition2 -> getOutput(0), weightMap, 24, "features.denseblock3", 1e-5); + auto transition3 = addTransition(network, *dense3 -> getOutput(0), weightMap, 512, "features.transition3", 1e-5); + + auto dense4 = addDenseBlock(network, transition3 -> getOutput(0), weightMap, 16, "features.denseblock4", 1e-5); + + auto bn5 = addBatchNorm2d(network, weightMap, *dense4 -> getOutput(0), "features.norm5", 1e-5); + auto relu5 = network -> addActivation(*bn5 -> getOutput(0), ActivationType::kRELU); + + // adaptive average pool => pytorch (F.adaptive_avg_pool2d(input, (1, 1))) + auto pool5 = network -> addPoolingNd(*relu5 -> getOutput(0), PoolingType::kAVERAGE, DimsHW{7,7}); + + auto fc1 = network -> addFullyConnected(*pool5 -> getOutput(0), 1000, weightMap["classifier.weight"], weightMap["classifier.bias"]); + assert(fc1); + + // set ouput blob name + fc1 -> getOutput(0) -> setName(OUTPUT_BLOB_NAME); + std::cout << "set name out" << std::endl; + + // mark the output + network -> markOutput(*fc1 -> getOutput(0)); + + // set batchsize and workspace size + builder -> setMaxBatchSize(maxBatchSize); + config -> setMaxWorkspaceSize(1 << 28); // 256 MiB + + // build engine + ICudaEngine* engine = builder -> buildEngineWithConfig(*network, *config); + std::cout << "build out" << std::endl; + + // destroy + network -> destroy(); + + // fere host mem + for(auto& mem: weightMap) + { + free((void*)(mem.second.values)); + } + + return engine; +} + + +void APIToModel(unsigned int maxBatchSize, IHostMemory** modelStream) +{ + // Create builder + IBuilder* builder = createInferBuilder(gLogger); + IBuilderConfig* config = builder->createBuilderConfig(); + + // Create model to populate the network, then set the outputs and create an engine + ICudaEngine* engine = createEngine(maxBatchSize, builder, config, DataType::kFLOAT); + assert(engine != nullptr); + + // Serialize the engine + (*modelStream) = engine->serialize(); + + // Close everything down + engine->destroy(); + builder->destroy(); + config->destroy(); +} + +/** + * Performs inference on the given input and + * writes the output from device to host memory. +**/ +void doInference(IExecutionContext& context, float* input, float* output, int batchSize) +{ + const ICudaEngine& engine = context.getEngine(); + + // Pointers to input and output device buffers to pass to engine. + // Engine requires exactly IEngine::getNbBindings() number of buffers. + assert(engine.getNbBindings() == 2); + void* buffers[2]; + + // In order to bind the buffers, we need to know the names of the input and output tensors. + // Note that indices are guaranteed to be less than IEngine::getNbBindings() + const int inputIndex = engine.getBindingIndex(INPUT_BLOB_NAME); + const int outputIndex = engine.getBindingIndex(OUTPUT_BLOB_NAME); + + // Create GPU buffers on device + CHECK(cudaMalloc(&buffers[inputIndex], batchSize * 3 * INPUT_H * INPUT_W * sizeof(float))); + CHECK(cudaMalloc(&buffers[outputIndex], batchSize * OUTPUT_SIZE * sizeof(float))); + + // Create stream + cudaStream_t stream; + CHECK(cudaStreamCreate(&stream)); + + // DMA input batch data to device, infer on the batch asynchronously, and DMA output back to host + CHECK(cudaMemcpyAsync(buffers[inputIndex], input, batchSize * 3 * INPUT_H * INPUT_W * sizeof(float), cudaMemcpyHostToDevice, stream)); + context.enqueue(batchSize, buffers, stream, nullptr); + CHECK(cudaMemcpyAsync(output, buffers[outputIndex], batchSize * OUTPUT_SIZE * sizeof(float), cudaMemcpyDeviceToHost, stream)); + cudaStreamSynchronize(stream); + + // Release stream and buffers + cudaStreamDestroy(stream); + CHECK(cudaFree(buffers[inputIndex])); + CHECK(cudaFree(buffers[outputIndex])); +} + +int main(int argc, char** argv) +{ + if (argc != 2) { + std::cerr << "arguments not right!" << std::endl; + std::cerr << "./densenet -s // serialize model to plan file" << std::endl; + std::cerr << "./densenet -d // deserialize plan file and run inference" << std::endl; + return -1; + } + + // create a model using the API directly and serialize it to a stream + char *trtModelStream{nullptr}; + size_t size{0}; + + if (std::string(argv[1]) == "-s") { + IHostMemory* modelStream{nullptr}; + APIToModel(1, &modelStream); + assert(modelStream != nullptr); + + std::ofstream p("densenet.engine", std::ios::binary); + if (!p) + { + std::cerr << "could not open plan output file" << std::endl; + return -1; + } + p.write(reinterpret_cast(modelStream->data()), modelStream->size()); + modelStream->destroy(); + return 1; + } else if (std::string(argv[1]) == "-d") { + std::ifstream file("densenet.engine", std::ios::binary); + if (file.good()) { + file.seekg(0, file.end); + size = file.tellg(); + file.seekg(0, file.beg); + trtModelStream = new char[size]; + assert(trtModelStream); + file.read(trtModelStream, size); + file.close(); + } + } else { + return -1; + } + + + // Subtract mean from image + static float data[3 * INPUT_H * INPUT_W]; + for (int i = 0; i < 3 * INPUT_H * INPUT_W; i++) + data[i] = 1.0; + + IRuntime* runtime = createInferRuntime(gLogger); + assert(runtime != nullptr); + ICudaEngine* engine = runtime->deserializeCudaEngine(trtModelStream, size, nullptr); + assert(engine != nullptr); + IExecutionContext* context = engine->createExecutionContext(); + assert(context != nullptr); + delete[] trtModelStream; + + // Run inference + static float prob[OUTPUT_SIZE]; + for (int i = 0; i < 100; i++) { + auto start = std::chrono::system_clock::now(); + doInference(*context, data, prob, 1); + auto end = std::chrono::system_clock::now(); + std::cout << std::chrono::duration_cast(end - start).count() << "ms" << std::endl; + } + + // Destroy the engine + context->destroy(); + engine->destroy(); + runtime->destroy(); + + // Print histogram of the output distribution + std::cout << "\nOutput:\n\n"; + for (unsigned int i = 0; i < OUTPUT_SIZE; i++) + { + std::cout << prob[i] << ", "; + if (i % 10 == 0) std::cout << i / 10 << std::endl; + } + std::cout << std::endl; + + return 0; +} diff --git a/densenet/logging.h b/densenet/logging.h new file mode 100644 index 0000000..0b1eca3 --- /dev/null +++ b/densenet/logging.h @@ -0,0 +1,507 @@ +/* + * Copyright (c) 2021, NVIDIA CORPORATION. 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. + */ + +#ifndef TENSORRT_LOGGING_H +#define TENSORRT_LOGGING_H + +#include "NvInferRuntimeCommon.h" +#include +#include +#include +#include +#include +#include +#include + +using Severity = nvinfer1::ILogger::Severity; + +class LogStreamConsumerBuffer : public std::stringbuf +{ +public: + LogStreamConsumerBuffer(std::ostream& stream, const std::string& prefix, bool shouldLog) + : mOutput(stream) + , mPrefix(prefix) + , mShouldLog(shouldLog) + { + } + + LogStreamConsumerBuffer(LogStreamConsumerBuffer&& other) + : mOutput(other.mOutput) + , mPrefix(other.mPrefix) + , mShouldLog(other.mShouldLog) + { + } + + ~LogStreamConsumerBuffer() + { + // std::streambuf::pbase() gives a pointer to the beginning of the buffered part of the output sequence + // std::streambuf::pptr() gives a pointer to the current position of the output sequence + // if the pointer to the beginning is not equal to the pointer to the current position, + // call putOutput() to log the output to the stream + if (pbase() != pptr()) + { + putOutput(); + } + } + + // synchronizes the stream buffer and returns 0 on success + // synchronizing the stream buffer consists of inserting the buffer contents into the stream, + // resetting the buffer and flushing the stream + virtual int sync() + { + putOutput(); + return 0; + } + + void putOutput() + { + if (mShouldLog) + { + // prepend timestamp + std::time_t timestamp = std::time(nullptr); + tm* tm_local = std::localtime(×tamp); + std::cout << "["; + std::cout << std::setw(2) << std::setfill('0') << 1 + tm_local->tm_mon << "/"; + std::cout << std::setw(2) << std::setfill('0') << tm_local->tm_mday << "/"; + std::cout << std::setw(4) << std::setfill('0') << 1900 + tm_local->tm_year << "-"; + std::cout << std::setw(2) << std::setfill('0') << tm_local->tm_hour << ":"; + std::cout << std::setw(2) << std::setfill('0') << tm_local->tm_min << ":"; + std::cout << std::setw(2) << std::setfill('0') << tm_local->tm_sec << "] "; + // std::stringbuf::str() gets the string contents of the buffer + // insert the buffer contents pre-appended by the appropriate prefix into the stream + mOutput << mPrefix << str(); + // set the buffer to empty + str(""); + // flush the stream + mOutput.flush(); + } + } + + void setShouldLog(bool shouldLog) + { + mShouldLog = shouldLog; + } + +private: + std::ostream& mOutput; + std::string mPrefix; + bool mShouldLog; +}; + +//! +//! \class LogStreamConsumerBase +//! \brief Convenience object used to initialize LogStreamConsumerBuffer before std::ostream in LogStreamConsumer +//! +class LogStreamConsumerBase +{ +public: + LogStreamConsumerBase(std::ostream& stream, const std::string& prefix, bool shouldLog) + : mBuffer(stream, prefix, shouldLog) + { + } + +protected: + LogStreamConsumerBuffer mBuffer; +}; + +//! +//! \class LogStreamConsumer +//! \brief Convenience object used to facilitate use of C++ stream syntax when logging messages. +//! Order of base classes is LogStreamConsumerBase and then std::ostream. +//! This is because the LogStreamConsumerBase class is used to initialize the LogStreamConsumerBuffer member field +//! in LogStreamConsumer and then the address of the buffer is passed to std::ostream. +//! This is necessary to prevent the address of an uninitialized buffer from being passed to std::ostream. +//! Please do not change the order of the parent classes. +//! +class LogStreamConsumer : protected LogStreamConsumerBase, public std::ostream +{ +public: + //! \brief Creates a LogStreamConsumer which logs messages with level severity. + //! Reportable severity determines if the messages are severe enough to be logged. + LogStreamConsumer(Severity reportableSeverity, Severity severity) + : LogStreamConsumerBase(severityOstream(severity), severityPrefix(severity), severity <= reportableSeverity) + , std::ostream(&mBuffer) // links the stream buffer with the stream + , mShouldLog(severity <= reportableSeverity) + , mSeverity(severity) + { + } + + LogStreamConsumer(LogStreamConsumer&& other) + : LogStreamConsumerBase(severityOstream(other.mSeverity), severityPrefix(other.mSeverity), other.mShouldLog) + , std::ostream(&mBuffer) // links the stream buffer with the stream + , mShouldLog(other.mShouldLog) + , mSeverity(other.mSeverity) + { + } + + void setReportableSeverity(Severity reportableSeverity) + { + mShouldLog = mSeverity <= reportableSeverity; + mBuffer.setShouldLog(mShouldLog); + } + +private: + static std::ostream& severityOstream(Severity severity) + { + return severity >= Severity::kINFO ? std::cout : std::cerr; + } + + static std::string severityPrefix(Severity severity) + { + switch (severity) + { + case Severity::kINTERNAL_ERROR: return "[F] "; + case Severity::kERROR: return "[E] "; + case Severity::kWARNING: return "[W] "; + case Severity::kINFO: return "[I] "; + case Severity::kVERBOSE: return "[V] "; + default: assert(0); return ""; + } + } + + bool mShouldLog; + Severity mSeverity; +}; + +//! \class Logger +//! +//! \brief Class which manages logging of TensorRT tools and samples +//! +//! \details This class provides a common interface for TensorRT tools and samples to log information to the console, +//! and supports logging two types of messages: +//! +//! - Debugging messages with an associated severity (info, warning, error, or internal error/fatal) +//! - Test pass/fail messages +//! +//! The advantage of having all samples use this class for logging as opposed to emitting directly to stdout/stderr is +//! that the logic for controlling the verbosity and formatting of sample output is centralized in one location. +//! +//! In the future, this class could be extended to support dumping test results to a file in some standard format +//! (for example, JUnit XML), and providing additional metadata (e.g. timing the duration of a test run). +//! +//! TODO: For backwards compatibility with existing samples, this class inherits directly from the nvinfer1::ILogger +//! interface, which is problematic since there isn't a clean separation between messages coming from the TensorRT +//! library and messages coming from the sample. +//! +//! In the future (once all samples are updated to use Logger::getTRTLogger() to access the ILogger) we can refactor the +//! class to eliminate the inheritance and instead make the nvinfer1::ILogger implementation a member of the Logger +//! object. + +class Logger : public nvinfer1::ILogger +{ +public: + Logger(Severity severity = Severity::kWARNING) + : mReportableSeverity(severity) + { + } + + //! + //! \enum TestResult + //! \brief Represents the state of a given test + //! + enum class TestResult + { + kRUNNING, //!< The test is running + kPASSED, //!< The test passed + kFAILED, //!< The test failed + kWAIVED //!< The test was waived + }; + + //! + //! \brief Forward-compatible method for retrieving the nvinfer::ILogger associated with this Logger + //! \return The nvinfer1::ILogger associated with this Logger + //! + //! TODO Once all samples are updated to use this method to register the logger with TensorRT, + //! we can eliminate the inheritance of Logger from ILogger + //! + nvinfer1::ILogger& getTRTLogger() + { + return *this; + } + + //! + //! \brief Implementation of the nvinfer1::ILogger::log() virtual method + //! + //! Note samples should not be calling this function directly; it will eventually go away once we eliminate the + //! inheritance from nvinfer1::ILogger + //! + void log(Severity severity, const char* msg) override + { + LogStreamConsumer(mReportableSeverity, severity) << "[TRT] " << std::string(msg) << std::endl; + } + + //! + //! \brief Method for controlling the verbosity of logging output + //! + //! \param severity The logger will only emit messages that have severity of this level or higher. + //! + void setReportableSeverity(Severity severity) + { + mReportableSeverity = severity; + } + + //! + //! \brief Opaque handle that holds logging information for a particular test + //! + //! This object is an opaque handle to information used by the Logger to print test results. + //! The sample must call Logger::defineTest() in order to obtain a TestAtom that can be used + //! with Logger::reportTest{Start,End}(). + //! + class TestAtom + { + public: + TestAtom(TestAtom&&) = default; + + private: + friend class Logger; + + TestAtom(bool started, const std::string& name, const std::string& cmdline) + : mStarted(started) + , mName(name) + , mCmdline(cmdline) + { + } + + bool mStarted; + std::string mName; + std::string mCmdline; + }; + + //! + //! \brief Define a test for logging + //! + //! \param[in] name The name of the test. This should be a string starting with + //! "TensorRT" and containing dot-separated strings containing + //! the characters [A-Za-z0-9_]. + //! For example, "TensorRT.sample_googlenet" + //! \param[in] cmdline The command line used to reproduce the test + // + //! \return a TestAtom that can be used in Logger::reportTest{Start,End}(). + //! + static TestAtom defineTest(const std::string& name, const std::string& cmdline) + { + return TestAtom(false, name, cmdline); + } + + //! + //! \brief A convenience overloaded version of defineTest() that accepts an array of command-line arguments + //! as input + //! + //! \param[in] name The name of the test + //! \param[in] argc The number of command-line arguments + //! \param[in] argv The array of command-line arguments (given as C strings) + //! + //! \return a TestAtom that can be used in Logger::reportTest{Start,End}(). + static TestAtom defineTest(const std::string& name, int argc, char const* const* argv) + { + auto cmdline = genCmdlineString(argc, argv); + return defineTest(name, cmdline); + } + + //! + //! \brief Report that a test has started. + //! + //! \pre reportTestStart() has not been called yet for the given testAtom + //! + //! \param[in] testAtom The handle to the test that has started + //! + static void reportTestStart(TestAtom& testAtom) + { + reportTestResult(testAtom, TestResult::kRUNNING); + assert(!testAtom.mStarted); + testAtom.mStarted = true; + } + + //! + //! \brief Report that a test has ended. + //! + //! \pre reportTestStart() has been called for the given testAtom + //! + //! \param[in] testAtom The handle to the test that has ended + //! \param[in] result The result of the test. Should be one of TestResult::kPASSED, + //! TestResult::kFAILED, TestResult::kWAIVED + //! + static void reportTestEnd(const TestAtom& testAtom, TestResult result) + { + assert(result != TestResult::kRUNNING); + assert(testAtom.mStarted); + reportTestResult(testAtom, result); + } + + static int reportPass(const TestAtom& testAtom) + { + reportTestEnd(testAtom, TestResult::kPASSED); + return EXIT_SUCCESS; + } + + static int reportFail(const TestAtom& testAtom) + { + reportTestEnd(testAtom, TestResult::kFAILED); + return EXIT_FAILURE; + } + + static int reportWaive(const TestAtom& testAtom) + { + reportTestEnd(testAtom, TestResult::kWAIVED); + return EXIT_SUCCESS; + } + + static int reportTest(const TestAtom& testAtom, bool pass) + { + return pass ? reportPass(testAtom) : reportFail(testAtom); + } + + Severity getReportableSeverity() const + { + return mReportableSeverity; + } + +private: + //! + //! \brief returns an appropriate string for prefixing a log message with the given severity + //! + static const char* severityPrefix(Severity severity) + { + switch (severity) + { + case Severity::kINTERNAL_ERROR: return "[F] "; + case Severity::kERROR: return "[E] "; + case Severity::kWARNING: return "[W] "; + case Severity::kINFO: return "[I] "; + case Severity::kVERBOSE: return "[V] "; + default: assert(0); return ""; + } + } + + //! + //! \brief returns an appropriate string for prefixing a test result message with the given result + //! + static const char* testResultString(TestResult result) + { + switch (result) + { + case TestResult::kRUNNING: return "RUNNING"; + case TestResult::kPASSED: return "PASSED"; + case TestResult::kFAILED: return "FAILED"; + case TestResult::kWAIVED: return "WAIVED"; + default: assert(0); return ""; + } + } + + //! + //! \brief returns an appropriate output stream (cout or cerr) to use with the given severity + //! + static std::ostream& severityOstream(Severity severity) + { + return severity >= Severity::kINFO ? std::cout : std::cerr; + } + + //! + //! \brief method that implements logging test results + //! + static void reportTestResult(const TestAtom& testAtom, TestResult result) + { + severityOstream(Severity::kINFO) << "&&&& " << testResultString(result) << " " << testAtom.mName << " # " + << testAtom.mCmdline << std::endl; + } + + //! + //! \brief generate a command line string from the given (argc, argv) values + //! + static std::string genCmdlineString(int argc, char const* const* argv) + { + std::stringstream ss; + for (int i = 0; i < argc; i++) + { + if (i > 0) + { + ss << " "; + } + ss << argv[i]; + } + return ss.str(); + } + + Severity mReportableSeverity; +}; + +namespace +{ + +//! +//! \brief produces a LogStreamConsumer object that can be used to log messages of severity kVERBOSE +//! +//! Example usage: +//! +//! LOG_VERBOSE(logger) << "hello world" << std::endl; +//! +inline LogStreamConsumer LOG_VERBOSE(const Logger& logger) +{ + return LogStreamConsumer(logger.getReportableSeverity(), Severity::kVERBOSE); +} + +//! +//! \brief produces a LogStreamConsumer object that can be used to log messages of severity kINFO +//! +//! Example usage: +//! +//! LOG_INFO(logger) << "hello world" << std::endl; +//! +inline LogStreamConsumer LOG_INFO(const Logger& logger) +{ + return LogStreamConsumer(logger.getReportableSeverity(), Severity::kINFO); +} + +//! +//! \brief produces a LogStreamConsumer object that can be used to log messages of severity kWARNING +//! +//! Example usage: +//! +//! LOG_WARN(logger) << "hello world" << std::endl; +//! +inline LogStreamConsumer LOG_WARN(const Logger& logger) +{ + return LogStreamConsumer(logger.getReportableSeverity(), Severity::kWARNING); +} + +//! +//! \brief produces a LogStreamConsumer object that can be used to log messages of severity kERROR +//! +//! Example usage: +//! +//! LOG_ERROR(logger) << "hello world" << std::endl; +//! +inline LogStreamConsumer LOG_ERROR(const Logger& logger) +{ + return LogStreamConsumer(logger.getReportableSeverity(), Severity::kERROR); +} + +//! +//! \brief produces a LogStreamConsumer object that can be used to log messages of severity kINTERNAL_ERROR +//! ("fatal" severity) +//! +//! Example usage: +//! +//! LOG_FATAL(logger) << "hello world" << std::endl; +//! +inline LogStreamConsumer LOG_FATAL(const Logger& logger) +{ + return LogStreamConsumer(logger.getReportableSeverity(), Severity::kINTERNAL_ERROR); +} + +} // anonymous namespace + +#endif // TENSORRT_LOGGING_H \ No newline at end of file