/*
* SPDX-FileCopyrightText: Copyright (c) 1993-2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: Apache-2.0
*
* 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 <algorithm>
#include <chrono>
#include <fstream>
#include <iostream>
#include <iterator>
#include <map>
#include <random>
#include <set>
#include <string>
#include <unordered_map>
#include <vector>
#include "NvCaffeParser.h"
#include "NvInfer.h"
#include "NvOnnxParser.h"
#include "NvUffParser.h"
#include "ErrorRecorder.h"
#include "common.h"
#include "half.h"
#include "logger.h"
#include "sampleDevice.h"
#include "sampleEngines.h"
#include "sampleOptions.h"
#include "sampleUtils.h"
using namespace nvinfer1;
namespace sample
{
namespace
{
struct CaffeBufferShutter
{
~CaffeBufferShutter()
{
shutdownCaffeParser();
}
};
struct UffBufferShutter
{
~UffBufferShutter()
{
shutdownUffParser();
}
};
std::map<std::string, float> readScalesFromCalibrationCache(std::string const& calibrationFile)
{
std::map<std::string, float> tensorScales;
std::ifstream cache{calibrationFile};
if (!cache.is_open())
{
sample::gLogError << "[TRT] Can not open provided calibration cache file" << std::endl;
return tensorScales;
}
std::string line;
while (std::getline(cache, line))
{
auto colonPos = line.find_last_of(':');
if (colonPos != std::string::npos)
{
// Scales should be stored in calibration cache as 32-bit floating numbers encoded as 32-bit integers
int32_t scalesAsInt = std::stoi(line.substr(colonPos + 2, 8), nullptr, 16);
auto const tensorName = line.substr(0, colonPos);
tensorScales[tensorName] = *reinterpret_cast<float*>(&scalesAsInt);
}
}
cache.close();
return tensorScales;
}
} // namespace
nvinfer1::ICudaEngine* LazilyDeserializedEngine::get()
{
SMP_RETVAL_IF_FALSE(
!mIsSafe, "Safe mode is enabled, but trying to get standard engine!", nullptr, sample::gLogError);
if (mEngine == nullptr)
{
SMP_RETVAL_IF_FALSE(
!mEngineBlob.empty(), "Engine blob is empty. Nothing to deserialize!", nullptr, sample::gLogError);
using time_point = std::chrono::time_point<std::chrono::high_resolution_clock>;
using duration = std::chrono::duration<float>;
time_point const deserializeStartTime{std::chrono::high_resolution_clock::now()};
if (mLeanDLLPath.empty())
{
mRuntime.reset(createRuntime());
}
else
{
mParentRuntime.reset(createRuntime());
ASSERT(mParentRuntime.get() != nullptr);
mRuntime.reset(mParentRuntime->loadRuntime(mLeanDLLPath.c_str()));
}
ASSERT(mRuntime.get() != nullptr);
if (mVersionCompatible)
{
// Application needs to opt into allowing deserialization of engines with embedded lean runtime.
mRuntime->setEngineHostCodeAllowed(true);
}
if (!mTempdir.empty())
{
mRuntime->setTemporaryDirectory(mTempdir.c_str());
}
mRuntime->setTempfileControlFlags(mTempfileControls);
SMP_RETVAL_IF_FALSE(mRuntime != nullptr, "runtime creation failed", nullptr, sample::gLogError);
if (mDLACore != -1)
{
mRuntime->setDLACore(mDLACore);
}
mRuntime->setErrorRecorder(&gRecorder);
for (auto const& pluginPath : mDynamicPlugins)
{
mRuntime->getPluginRegistry().loadLibrary(pluginPath.c_str());
}
mEngine.reset(mRuntime->deserializeCudaEngine(mEngineBlob.data(), mEngineBlob.size()));
SMP_RETVAL_IF_FALSE(mEngine != nullptr, "Engine deserialization failed", nullptr, sample::gLogError);
time_point const deserializeEndTime{std::chrono::high_resolution_clock::now()};
sample::gLogInfo << "Engine deserialized in "
<< duration(deserializeEndTime - deserializeStartTime).count() << " sec." << std::endl;
}
return mEngine.get();
}
nvinfer1::ICudaEngine* LazilyDeserializedEngine::release()
{
return mEngine.release();
}
nvinfer1::safe::ICudaEngine* LazilyDeserializedEngine::getSafe()
{
SMP_RETVAL_IF_FALSE(
mIsSafe, "Safe mode is not enabled, but trying to get safe engine!", nullptr, sample::gLogError);
ASSERT(sample::hasSafeRuntime());
if (mSafeEngine == nullptr)
{
SMP_RETVAL_IF_FALSE(
!mEngineBlob.empty(), "Engine blob is empty. Nothing to deserialize!", nullptr, sample::gLogError);
SMP_RETVAL_IF_FALSE(
mDLACore == -1, "Safe DLA engine built with kDLA_STANDALONE should not be deserialized in TRT!", nullptr,
sample::gLogError);
using time_point = std::chrono::time_point<std::chrono::high_resolution_clock>;
using duration = std::chrono::duration<float>;
time_point const deserializeStartTime{std::chrono::high_resolution_clock::now()};
std::unique_ptr<safe::IRuntime> safeRuntime{sample::createSafeInferRuntime(sample::gLogger.getTRTLogger())};
SMP_RETVAL_IF_FALSE(safeRuntime != nullptr, "SafeRuntime creation failed", nullptr, sample::gLogError);
safeRuntime->setErrorRecorder(&gRecorder);
mSafeEngine.reset(
safeRuntime->deserializeCudaEngine(mEngineBlob.data(), mEngineBlob.size()));
SMP_RETVAL_IF_FALSE(mSafeEngine != nullptr, "SafeEngine deserialization failed", nullptr, sample::gLogError);
time_point const deserializeEndTime{std::chrono::high_resolution_clock::now()};
sample::gLogInfo << "SafeEngine deserialized in "
<< duration(deserializeEndTime - deserializeStartTime).count() << " sec." << std::endl;
}
return mSafeEngine.get();
}
void setTensorScalesFromCalibration(nvinfer1::INetworkDefinition& network, std::vector<IOFormat> const& inputFormats,
std::vector<IOFormat> const& outputFormats, std::string const& calibrationFile)
{
auto const tensorScales = readScalesFromCalibrationCache(calibrationFile);
bool const broadcastInputFormats = broadcastIOFormats(inputFormats, network.getNbInputs());
for (int32_t i = 0, n = network.getNbInputs(); i < n; ++i)
{
int32_t formatIdx = broadcastInputFormats ? 0 : i;
if (!inputFormats.empty() && inputFormats[formatIdx].first == DataType::kINT8)
{
auto* input = network.getInput(i);
auto const calibScale = tensorScales.at(input->getName());
input->setDynamicRange(-127 * calibScale, 127 * calibScale);
}
}
bool const broadcastOutputFormats = broadcastIOFormats(outputFormats, network.getNbInputs());
for (int32_t i = 0, n = network.getNbOutputs(); i < n; ++i)
{
int32_t formatIdx = broadcastOutputFormats ? 0 : i;
if (!outputFormats.empty() && outputFormats[formatIdx].first == DataType::kINT8)
{
auto* output = network.getOutput(i);
auto const calibScale = tensorScales.at(output->getName());
output->setDynamicRange(-127 * calibScale, 127 * calibScale);
}
}
}
//!
//! \brief Generate a network definition for a given model
//!
//! \param[in] model Model options for this network
//! \param[in,out] network Network storing the parsed results
//! \param[in,out] err Error stream
//! \param[out] vcPluginLibrariesUsed If not nullptr, will be populated with paths to VC plugin libraries required by
//! the parsed network.
//!
//! \return Parser The parser used to initialize the network and that holds the weights for the network, or an invalid
//! parser (the returned parser converts to false if tested)
//!
//! Constant input dimensions in the model must not be changed in the corresponding
//! network definition, because its correctness may rely on the constants.
//!
//! \see Parser::operator bool()
//!
Parser modelToNetwork(const ModelOptions& model, nvinfer1::INetworkDefinition& network, std::ostream& err,
std::vector<std::string>* vcPluginLibrariesUsed)
{
sample::gLogInfo << "Start parsing network model." << std::endl;
auto const tBegin = std::chrono::high_resolution_clock::now();
Parser parser;
std::string const& modelName = model.baseModel.model;
switch (model.baseModel.format)
{
case ModelFormat::kCAFFE:
{
using namespace nvcaffeparser1;
parser.caffeParser.reset(sampleCreateCaffeParser());
CaffeBufferShutter bufferShutter;
auto const* const blobNameToTensor = parser.caffeParser->parse(
model.prototxt.c_str(), modelName.empty() ? nullptr : modelName.c_str(), network, DataType::kFLOAT);
if (!blobNameToTensor)
{
err << "Failed to parse caffe model or prototxt, tensors blob not found" << std::endl;
parser.caffeParser.reset();
break;
}
for (auto const& s : model.outputs)
{
if (blobNameToTensor->find(s.c_str()) == nullptr)
{
err << "Could not find output blob " << s << std::endl;
parser.caffeParser.reset();
break;
}
network.markOutput(*blobNameToTensor->find(s.c_str()));
}
break;
}
case ModelFormat::kUFF:
{
using namespace nvuffparser;
parser.uffParser.reset(sampleCreateUffParser());
UffBufferShutter bufferShutter;
for (auto const& s : model.uffInputs.inputs)
{
if (!parser.uffParser->registerInput(
s.first.c_str(), s.second, model.uffInputs.NHWC ? UffInputOrder::kNHWC : UffInputOrder::kNCHW))
{
err << "Failed to register input " << s.first << std::endl;
parser.uffParser.reset();
break;
}
}
for (auto const& s : model.outputs)
{
if (!parser.uffParser->registerOutput(s.c_str()))
{
err << "Failed to register output " << s << std::endl;
parser.uffParser.reset();
break;
}
}
if (!parser.uffParser->parse(model.baseModel.model.c_str(), network))
{
err << "Failed to parse uff file" << std::endl;
parser.uffParser.reset();
break;
}
break;
}
case ModelFormat::kONNX:
{
using namespace nvonnxparser;
parser.onnxParser.reset(createONNXParser(network));
if (!parser.onnxParser->parseFromFile(
model.baseModel.model.c_str(), static_cast<int>(sample::gLogger.getReportableSeverity())))
{
err << "Failed to parse onnx file" << std::endl;
parser.onnxParser.reset();
}
if (vcPluginLibrariesUsed && parser.onnxParser.get())
{
int64_t nbPluginLibs;
char const* const* pluginLibArray = parser.onnxParser->getUsedVCPluginLibraries(nbPluginLibs);
if (nbPluginLibs >= 0)
{
vcPluginLibrariesUsed->reserve(nbPluginLibs);
for (int64_t i = 0; i < nbPluginLibs; ++i)
{
sample::gLogInfo << "Using VC plugin library " << pluginLibArray[i] << std::endl;
vcPluginLibrariesUsed->emplace_back(std::string{pluginLibArray[i]});
}
}
else
{
sample::gLogWarning << "Failure to query VC plugin libraries required by parsed ONNX network"
<< std::endl;
}
}
break;
}
case ModelFormat::kANY: break;
}
auto const tEnd = std::chrono::high_resolution_clock::now();
float const parseTime = std::chrono::duration<float>(tEnd - tBegin).count();
sample::gLogInfo << "Finished parsing network model. Parse time: " << parseTime << std::endl;
return parser;
}
namespace
{
class RndInt8Calibrator : public nvinfer1::IInt8EntropyCalibrator2
{
public:
RndInt8Calibrator(int32_t batches, std::vector<int64_t>& elemCount, std::string const& cacheFile,
nvinfer1::INetworkDefinition const& network, std::ostream& err);
~RndInt8Calibrator() override
{
for (auto& elem : mInputDeviceBuffers)
{
cudaCheck(cudaFree(elem.second), mErr);
}
}
bool getBatch(void* bindings[], char const* names[], int32_t nbBindings) noexcept override;
int32_t getBatchSize() const noexcept override
{
return 1;
}
const void* readCalibrationCache(size_t& length) noexcept override;
void writeCalibrationCache(void const*, size_t) noexcept override {}
private:
int32_t mBatches{};
int32_t mCurrentBatch{};
std::string mCacheFile;
std::map<std::string, void*> mInputDeviceBuffers;
std::vector<char> mCalibrationCache;
std::ostream& mErr;
};
RndInt8Calibrator::RndInt8Calibrator(int32_t batches, std::vector<int64_t>& elemCount, std::string const& cacheFile,
INetworkDefinition const& network, std::ostream& err)
: mBatches(batches)
, mCurrentBatch(0)
, mCacheFile(cacheFile)
, mErr(err)
{
std::ifstream tryCache(cacheFile, std::ios::binary);
if (tryCache.good())
{
return;
}
std::default_random_engine generator;
std::uniform_real_distribution<float> distribution(-1.0F, 1.0F);
auto gen = [&generator, &distribution]() { return distribution(generator); };
for (int32_t i = 0; i < network.getNbInputs(); i++)
{
auto* input = network.getInput(i);
std::vector<float> rnd_data(elemCount[i]);
std::generate_n(rnd_data.begin(), elemCount[i], gen);
void* data;
cudaCheck(cudaMalloc(&data, elemCount[i] * sizeof(float)), mErr);
cudaCheck(cudaMemcpy(data, rnd_data.data(), elemCount[i] * sizeof(float), cudaMemcpyHostToDevice), mErr);
mInputDeviceBuffers.insert(std::make_pair(input->getName(), data));
}
}
bool RndInt8Calibrator::getBatch(void* bindings[], char const* names[], int32_t nbBindings) noexcept
{
if (mCurrentBatch >= mBatches)
{
return false;
}
for (int32_t i = 0; i < nbBindings; ++i)
{
bindings[i] = mInputDeviceBuffers[names[i]];
}
++mCurrentBatch;
return true;
}
const void* RndInt8Calibrator::readCalibrationCache(size_t& length) noexcept
{
mCalibrationCache.clear();
std::ifstream input(mCacheFile, std::ios::binary);
input >> std::noskipws;
if (input.good())
{
std::copy(
std::istream_iterator<char>(input), std::istream_iterator<char>(), std::back_inserter(mCalibrationCache));
}
length = mCalibrationCache.size();
return !mCalibrationCache.empty() ? mCalibrationCache.data() : nullptr;
}
bool setTensorDynamicRange(INetworkDefinition const& network, float inRange = 2.0F, float outRange = 4.0F)
{
// Ensure that all layer inputs have a dynamic range.
for (int32_t l = 0; l < network.getNbLayers(); l++)
{
auto* layer = network.getLayer(l);
for (int32_t i = 0; i < layer->getNbInputs(); i++)
{
ITensor* input{layer->getInput(i)};
// Optional inputs are nullptr here and are from RNN layers.
if (input && !input->dynamicRangeIsSet())
{
// Concat should propagate dynamic range from outputs to inputs to avoid
// Re-quantization during the concatenation
auto dynRange = (layer->getType() == LayerType::kCONCATENATION) ? outRange : inRange;
if (!input->setDynamicRange(-dynRange, dynRange))
{
return false;
}
}
}
for (int32_t o = 0; o < layer->getNbOutputs(); o++)
{
ITensor* output{layer->getOutput(o)};
// Optional outputs are nullptr here and are from RNN layers.
if (output && !output->dynamicRangeIsSet())
{
// Pooling must have the same input and output dynamic range.
if (layer->getType() == LayerType::kPOOLING)
{
if (!output->setDynamicRange(-inRange, inRange))
{
return false;
}
}
else
{
if (!output->setDynamicRange(-outRange, outRange))
{
return false;
}
}
}
}
}
return true;
}
void setLayerPrecisions(INetworkDefinition& network, LayerPrecisions const& layerPrecisions)
{
bool const hasGlobalPrecision{layerPrecisions.find("*") != layerPrecisions.end()};
auto const globalPrecision = hasGlobalPrecision ? layerPrecisions.at("*") : nvinfer1::DataType::kFLOAT;
bool hasLayerPrecisionSkipped{false};
for (int32_t layerIdx = 0; layerIdx < network.getNbLayers(); ++layerIdx)
{
auto* layer = network.getLayer(layerIdx);
auto const layerName = layer->getName();
if (layerPrecisions.find(layer->getName()) != layerPrecisions.end())
{
layer->setPrecision(layerPrecisions.at(layer->getName()));
}
else if (hasGlobalPrecision)
{
// We should not set the layer precision if its default precision is INT32 or Bool.
if (layer->getPrecision() == nvinfer1::DataType::kINT32
|| layer->getPrecision() == nvinfer1::DataType::kBOOL)
{
hasLayerPrecisionSkipped = true;
sample::gLogVerbose << "Skipped setting precision for layer " << layerName << " because the "
<< " default layer precision is INT32 or Bool." << std::endl;
continue;
}
// We should not set the constant layer precision if its weights are in INT32.
if (layer->getType() == nvinfer1::LayerType::kCONSTANT
&& static_cast<IConstantLayer*>(layer)->getWeights().type == nvinfer1::DataType::kINT32)
{
hasLayerPrecisionSkipped = true;
sample::gLogVerbose << "Skipped setting precision for layer " << layerName << " because this "
<< "constant layer has INT32 weights." << std::endl;
continue;
}
// We should not set the layer precision if the layer operates on a shape tensor.
if (layer->getNbInputs() >= 1 && layer->getInput(0)->isShapeTensor())
{
hasLayerPrecisionSkipped = true;
sample::gLogVerbose << "Skipped setting precision for layer " << layerName << " because this layer "
<< "operates on a shape tensor." << std::endl;
continue;
}
if (layer->getNbInputs() >= 1 && layer->getInput(0)->getType() == nvinfer1::DataType::kINT32
&& layer->getNbOutputs() >= 1 && layer->getOutput(0)->getType() == nvinfer1::DataType::kINT32)
{
hasLayerPrecisionSkipped = true;
sample::gLogVerbose << "Skipped setting precision for layer " << layerName << " because this "
<< "layer has INT32 input and output." << std::endl;
continue;
}
// All heuristics passed. Set the layer precision.
layer->setPrecision(globalPrecision);
}
}
if (hasLayerPrecisionSkipped)
{
sample::gLogInfo << "Skipped setting precisions for some layers. Check verbose logs for more details."
<< std::endl;
}
}
void setLayerOutputTypes(INetworkDefinition& network, LayerOutputTypes const& layerOutputTypes)
{
bool const hasGlobalOutputType{layerOutputTypes.find("*") != layerOutputTypes.end()};
auto const globalOutputType = hasGlobalOutputType ? layerOutputTypes.at("*").at(0) : nvinfer1::DataType::kFLOAT;
bool hasLayerOutputTypeSkipped{false};
for (int32_t layerIdx = 0; layerIdx < network.getNbLayers(); ++layerIdx)
{
auto* layer = network.getLayer(layerIdx);
auto const layerName = layer->getName();
auto const nbOutputs = layer->getNbOutputs();
if (layerOutputTypes.find(layer->getName()) != layerOutputTypes.end())
{
auto const& outputTypes = layerOutputTypes.at(layer->getName());
bool const isBroadcast = (outputTypes.size() == 1);
if (!isBroadcast && static_cast<int32_t>(outputTypes.size()) != nbOutputs)
{
sample::gLogError << "Layer " << layerName << " has " << nbOutputs << " outputs but "
<< outputTypes.size() << " output types are given in --layerOutputTypes flag."
<< std::endl;
throw std::invalid_argument("Invalid --layerOutputTypes flag.");
}
for (int32_t outputIdx = 0; outputIdx < nbOutputs; ++outputIdx)
{
layer->setOutputType(outputIdx, outputTypes.at(isBroadcast ? 0 : outputIdx));
}
}
else if (hasGlobalOutputType)
{
// We should not set the layer output types if its default precision is INT32 or Bool.
if (layer->getPrecision() == nvinfer1::DataType::kINT32
|| layer->getPrecision() == nvinfer1::DataType::kBOOL)
{
hasLayerOutputTypeSkipped = true;
sample::gLogVerbose << "Skipped setting output types for layer " << layerName << " because the "
<< " default layer precision is INT32 or Bool." << std::endl;
continue;
}
// We should not set the constant layer output types if its weights are in INT32.
if (layer->getType() == nvinfer1::LayerType::kCONSTANT
&& static_cast<IConstantLayer*>(layer)->getWeights().type == nvinfer1::DataType::kINT32)
{
hasLayerOutputTypeSkipped = true;
sample::gLogVerbose << "Skipped setting output types for layer " << layerName << " because this "
<< "constant layer has INT32 weights." << std::endl;
continue;
}
for (int32_t outputIdx = 0; outputIdx < nbOutputs; ++outputIdx)
{
// We should not set the output type if the output is a shape tensor.
if (layer->getOutput(0)->isShapeTensor())
{
hasLayerOutputTypeSkipped = true;
sample::gLogVerbose << "Skipped setting output type for output " << outputIdx << " of layer "
<< layerName << " because it is a shape tensor." << std::endl;
continue;
}
layer->setOutputType(outputIdx, globalOutputType);
}
}
}
if (hasLayerOutputTypeSkipped)
{
sample::gLogInfo << "Skipped setting output types for some layers. Check verbose logs for more details."
<< std::endl;
}
}
void setLayerDeviceTypes(
INetworkDefinition const& network, IBuilderConfig& config, LayerDeviceTypes const& layerDeviceTypes)
{
for (int32_t layerIdx = 0; layerIdx < network.getNbLayers(); ++layerIdx)
{
auto* layer = network.getLayer(layerIdx);
auto const layerName = layer->getName();
if (layerDeviceTypes.find(layerName) != layerDeviceTypes.end())
{
DeviceType const deviceType = layerDeviceTypes.at(layerName);
config.setDeviceType(layer, deviceType);
}
}
}
void setMemoryPoolLimits(IBuilderConfig& config, BuildOptions const& build)
{
auto const roundToBytes = [](double const sizeInMB) { return static_cast<size_t>(sizeInMB * (1 << 20)); };
if (build.workspace >= 0)
{
config.setMemoryPoolLimit(MemoryPoolType::kWORKSPACE, roundToBytes(build.workspace));
}
if (build.dlaSRAM >= 0)
{
config.setMemoryPoolLimit(MemoryPoolType::kDLA_MANAGED_SRAM, roundToBytes(build.dlaSRAM));
}
if (build.dlaLocalDRAM >= 0)
{
config.setMemoryPoolLimit(MemoryPoolType::kDLA_LOCAL_DRAM, roundToBytes(build.dlaLocalDRAM));
}
if (build.dlaGlobalDRAM >= 0)
{
config.setMemoryPoolLimit(MemoryPoolType::kDLA_GLOBAL_DRAM, roundToBytes(build.dlaGlobalDRAM));
}
}
void setPreviewFeatures(IBuilderConfig& config, BuildOptions const& build)
{
auto const setFlag = [&](PreviewFeature feat) {
int32_t featVal = static_cast<int32_t>(feat);
if (build.previewFeatures.find(featVal) != build.previewFeatures.end())
{
config.setPreviewFeature(feat, build.previewFeatures.at(featVal));
}
};
setFlag(PreviewFeature::kFASTER_DYNAMIC_SHAPES_0805);
setFlag(PreviewFeature::kDISABLE_EXTERNAL_TACTIC_SOURCES_FOR_CORE_0805);
setFlag(PreviewFeature::kPROFILE_SHARING_0806);
}
} // namespace
bool setupNetworkAndConfig(BuildOptions const& build, SystemOptions const& sys, IBuilder& builder,
INetworkDefinition& network, IBuilderConfig& config, std::unique_ptr<nvinfer1::IInt8Calibrator>& calibrator,
std::ostream& err, std::vector<std::vector<int8_t>>& sparseWeights)
{
IOptimizationProfile* profile{nullptr};
if (build.maxBatch)
{
builder.setMaxBatchSize(build.maxBatch);
}
else
{
profile = builder.createOptimizationProfile();
}
bool hasDynamicShapes{false};
bool broadcastInputFormats = broadcastIOFormats(build.inputFormats, network.getNbInputs());
if (profile)
{
// Check if the provided input tensor names match the input tensors of the engine.
// Throw an error if the provided input tensor names cannot be found because it implies a potential typo.
for (auto const& shape : build.shapes)
{
bool tensorNameFound{false};
for (int32_t i = 0; i < network.getNbInputs(); ++i)
{
if (network.getInput(i)->getName() == shape.first)
{
tensorNameFound = true;
break;
}
}
if (!tensorNameFound)
{
sample::gLogError << "Cannot find input tensor with name \"" << shape.first << "\" in the network "
<< "inputs! Please make sure the input tensor names are correct." << std::endl;
return false;
}
}
}
for (uint32_t i = 0, n = network.getNbInputs(); i < n; i++)
{
// Set formats and data types of inputs
auto* input = network.getInput(i);
if (!build.inputFormats.empty())
{
int inputFormatIndex = broadcastInputFormats ? 0 : i;
input->setType(build.inputFormats[inputFormatIndex].first);
input->setAllowedFormats(build.inputFormats[inputFormatIndex].second);
}
else
{
switch (input->getType())
{
case DataType::kINT32:
case DataType::kBOOL:
case DataType::kHALF:
case DataType::kUINT8:
// Leave these as is.
break;
case DataType::kFLOAT:
case DataType::kINT8:
// User did not specify a floating-point format. Default to kFLOAT.
input->setType(DataType::kFLOAT);
break;
case DataType::kFP8: ASSERT(!"FP8 is not supported"); break;
}
input->setAllowedFormats(1U << static_cast<int>(TensorFormat::kLINEAR));
}
if (profile)
{
auto const dims = input->getDimensions();
auto const isScalar = dims.nbDims == 0;
auto const isDynamicInput = std::any_of(dims.d, dims.d + dims.nbDims, [](int32_t dim) { return dim == -1; })
|| input->isShapeTensor();
if (isDynamicInput)
{
hasDynamicShapes = true;
auto shape = build.shapes.find(input->getName());
ShapeRange shapes{};
// If no shape is provided, set dynamic dimensions to 1.
if (shape == build.shapes.end())
{
constexpr int DEFAULT_DIMENSION = 1;
std::vector<int> staticDims;
if (input->isShapeTensor())
{
if (isScalar)
{
staticDims.push_back(1);
}
else
{
staticDims.resize(dims.d[0]);
std::fill(staticDims.begin(), staticDims.end(), DEFAULT_DIMENSION);
}
}
else
{
staticDims.resize(dims.nbDims);
std::transform(dims.d, dims.d + dims.nbDims, staticDims.begin(),
[&](int dimension) { return dimension > 0 ? dimension : DEFAULT_DIMENSION; });
}
sample::gLogWarning << "Dynamic dimensions required for input: " << input->getName()
<< ", but no shapes were provided. Automatically overriding shape to: "
<< staticDims << std::endl;
std::fill(shapes.begin(), shapes.end(), staticDims);
}
else
{
shapes = shape->second;
}
std::vector<int> profileDims{};
if (input->isShapeTensor())
{
profileDims = shapes[static_cast<size_t>(OptProfileSelector::kMIN)];
SMP_RETVAL_IF_FALSE(profile->setShapeValues(input->getName(), OptProfileSelector::kMIN,
profileDims.data(), static_cast<int>(profileDims.size())),
"Error in set shape values MIN", false, err);
profileDims = shapes[static_cast<size_t>(OptProfileSelector::kOPT)];
SMP_RETVAL_IF_FALSE(profile->setShapeValues(input->getName(), OptProfileSelector::kOPT,
profileDims.data(), static_cast<int>(profileDims.size())),
"Error in set shape values OPT", false, err);
profileDims = shapes[static_cast<size_t>(OptProfileSelector::kMAX)];
SMP_RETVAL_IF_FALSE(profile->setShapeValues(input->getName(), OptProfileSelector::kMAX,
profileDims.data(), static_cast<int>(profileDims.size())),
"Error in set shape values MAX", false, err);
}
else
{
profileDims = shapes[static_cast<size_t>(OptProfileSelector::kMIN)];
SMP_RETVAL_IF_FALSE(
profile->setDimensions(input->getName(), OptProfileSelector::kMIN, toDims(profileDims)),
"Error in set dimensions to profile MIN", false, err);
profileDims = shapes[static_cast<size_t>(OptProfileSelector::kOPT)];
SMP_RETVAL_IF_FALSE(
profile->setDimensions(input->getName(), OptProfileSelector::kOPT, toDims(profileDims)),
"Error in set dimensions to profile OPT", false, err);
profileDims = shapes[static_cast<size_t>(OptProfileSelector::kMAX)];
SMP_RETVAL_IF_FALSE(
profile->setDimensions(input->getName(), OptProfileSelector::kMAX, toDims(profileDims)),
"Error in set dimensions to profile MAX", false, err);
}
}
}
}
for (uint32_t i = 0, n = network.getNbOutputs(); i < n; i++)
{
auto* output = network.getOutput(i);
if (profile)
{
auto const dims = output->getDimensions();
// A shape tensor output with known static dimensions may have dynamic shape values inside it.
auto const isDynamicOutput = std::any_of(dims.d, dims.d + dims.nbDims, [](int32_t dim) { return dim == -1; })
|| output->isShapeTensor();
if (isDynamicOutput)
{
hasDynamicShapes = true;
}
}
}
if (!hasDynamicShapes && !build.shapes.empty())
{
sample::gLogError << "Static model does not take explicit shapes since the shape of inference tensors will be "
"determined by the model itself"
<< std::endl;
return false;
}
if (profile && hasDynamicShapes)
{
SMP_RETVAL_IF_FALSE(profile->isValid(), "Required optimization profile is invalid", false, err);
SMP_RETVAL_IF_FALSE(
config.addOptimizationProfile(profile) != -1, "Error in add optimization profile", false, err);
}
bool broadcastOutputFormats = broadcastIOFormats(build.outputFormats, network.getNbOutputs(), false);
for (uint32_t i = 0, n = network.getNbOutputs(); i < n; i++)
{
// Set formats and data types of outputs
auto* output = network.getOutput(i);
if (!build.outputFormats.empty())
{
int outputFormatIndex = broadcastOutputFormats ? 0 : i;
output->setType(build.outputFormats[outputFormatIndex].first);
output->setAllowedFormats(build.outputFormats[outputFormatIndex].second);
}
else
{
output->setAllowedFormats(1U << static_cast<int>(TensorFormat::kLINEAR));
}
}
setMemoryPoolLimits(config, build);
setPreviewFeatures(config, build);
if (build.heuristic)
{
config.setFlag(BuilderFlag::kENABLE_TACTIC_HEURISTIC);
}
config.setBuilderOptimizationLevel(build.builderOptimizationLevel);
if (build.timingCacheMode == TimingCacheMode::kDISABLE)
{
config.setFlag(BuilderFlag::kDISABLE_TIMING_CACHE);
}
if (!build.tf32)
{
config.clearFlag(BuilderFlag::kTF32);
}
if (build.refittable)
{
config.setFlag(BuilderFlag::kREFIT);
}
if (build.versionCompatible)
{
config.setFlag(BuilderFlag::kVERSION_COMPATIBLE);
}
std::vector<char const*> pluginPaths;
for (auto const& pluginPath : sys.setPluginsToSerialize)
{
sample::gLogVerbose << "Setting plugin to serialize: " << pluginPath << std::endl;
pluginPaths.push_back(pluginPath.c_str());
}
if (!pluginPaths.empty())
{
config.setPluginsToSerialize(pluginPaths.data(), pluginPaths.size());
}
if (build.excludeLeanRuntime)
{
config.setFlag(BuilderFlag::kEXCLUDE_LEAN_RUNTIME);
}
if (build.sparsity != SparsityFlag::kDISABLE)
{
config.setFlag(BuilderFlag::kSPARSE_WEIGHTS);
if (build.sparsity == SparsityFlag::kFORCE)
{
sparsify(network, sparseWeights);
}
}
config.setProfilingVerbosity(build.profilingVerbosity);
config.setMinTimingIterations(build.minTiming);
config.setAvgTimingIterations(build.avgTiming);
if (build.fp16)
{
config.setFlag(BuilderFlag::kFP16);
}
if (build.int8)
{
config.setFlag(BuilderFlag::kINT8);
}
SMP_RETVAL_IF_FALSE(!(build.int8 && build.fp8),
"FP8 and INT8 precisions have been specified", false, err);
if (build.fp8)
{
config.setFlag(BuilderFlag::kFP8);
}
if (build.int8 && !build.fp16)
{
sample::gLogInfo
<< "FP32 and INT8 precisions have been specified - more performance might be enabled by additionally "
"specifying --fp16 or --best"
<< std::endl;
}
auto isInt8 = [](const IOFormat& format) { return format.first == DataType::kINT8; };
auto int8IO = std::count_if(build.inputFormats.begin(), build.inputFormats.end(), isInt8)
+ std::count_if(build.outputFormats.begin(), build.outputFormats.end(), isInt8);
auto hasQDQLayers = [](INetworkDefinition& network) {
// Determine if our network has QDQ layers.
auto const nbLayers = network.getNbLayers();
for (int32_t i = 0; i < nbLayers; i++)
{
auto const& layer = network.getLayer(i);
if (layer->getType() == LayerType::kQUANTIZE || layer->getType() == LayerType::kDEQUANTIZE)
{
return true;
}
}
return false;
};
if (!hasQDQLayers(network) && (build.int8 || int8IO) && build.calibration.empty())
{
// Explicitly set int8 scales if no calibrator is provided and if I/O tensors use int8,
// because auto calibration does not support this case.
SMP_RETVAL_IF_FALSE(setTensorDynamicRange(network), "Error in set tensor dynamic range.", false, err);
}
else if (build.int8)
{
if (!hasQDQLayers(network) && int8IO)
{
try
{
// Set dynamic ranges of int8 inputs / outputs to match scales loaded from calibration cache
// TODO http://nvbugs/3262234 Change the network validation so that this workaround can be removed
setTensorScalesFromCalibration(network, build.inputFormats, build.outputFormats, build.calibration);
}
catch (std::exception&)
{
sample::gLogError
<< "Int8IO was specified but impossible to read tensor scales from provided calibration cache file"
<< std::endl;
return false;
}
}
IOptimizationProfile* profileCalib{nullptr};
if (!build.shapesCalib.empty())
{
profileCalib = builder.createOptimizationProfile();
for (uint32_t i = 0, n = network.getNbInputs(); i < n; i++)
{
auto* input = network.getInput(i);
Dims profileDims{};
auto shape = build.shapesCalib.find(input->getName());
ShapeRange shapesCalib{};
shapesCalib = shape->second;
profileDims = toDims(shapesCalib[static_cast<size_t>(OptProfileSelector::kOPT)]);
// Here we check only kMIN as all profileDims are the same.
SMP_RETVAL_IF_FALSE(
profileCalib->setDimensions(input->getName(), OptProfileSelector::kMIN, profileDims),
"Error in set dimensions to calibration profile OPT", false, err);
profileCalib->setDimensions(input->getName(), OptProfileSelector::kOPT, profileDims);
profileCalib->setDimensions(input->getName(), OptProfileSelector::kMAX, profileDims);
}
SMP_RETVAL_IF_FALSE(profileCalib->isValid(), "Calibration profile is invalid", false, err);
SMP_RETVAL_IF_FALSE(
config.setCalibrationProfile(profileCalib), "Error in set calibration profile", false, err);
}
std::vector<int64_t> elemCount{};
for (int i = 0; i < network.getNbInputs(); i++)
{
auto* input = network.getInput(i);
auto const dims = input->getDimensions();
auto const isDynamicInput
= std::any_of(dims.d, dims.d + dims.nbDims, [](int32_t dim) { return dim == -1; });
if (profileCalib)
{
elemCount.push_back(volume(profileCalib->getDimensions(input->getName(), OptProfileSelector::kOPT)));
}
else if (profile && isDynamicInput)
{
elemCount.push_back(volume(profile->getDimensions(input->getName(), OptProfileSelector::kOPT)));
}
else
{
elemCount.push_back(volume(input->getDimensions()));
}
}
calibrator.reset(new RndInt8Calibrator(1, elemCount, build.calibration, network, err));
config.setInt8Calibrator(calibrator.get());
}
if (build.directIO)
{
config.setFlag(BuilderFlag::kDIRECT_IO);
}
switch (build.precisionConstraints)
{
case PrecisionConstraints::kNONE:
// It's the default for TensorRT.
break;
case PrecisionConstraints::kOBEY:
config.setFlag(BuilderFlag::kOBEY_PRECISION_CONSTRAINTS);
break;
case PrecisionConstraints::kPREFER: config.setFlag(BuilderFlag::kPREFER_PRECISION_CONSTRAINTS); break;
}
if (!build.layerPrecisions.empty() && build.precisionConstraints != PrecisionConstraints::kNONE)
{
setLayerPrecisions(network, build.layerPrecisions);
}
if (!build.layerOutputTypes.empty() && build.precisionConstraints != PrecisionConstraints::kNONE)
{
setLayerOutputTypes(network, build.layerOutputTypes);
}
if (!build.layerDeviceTypes.empty())
{
setLayerDeviceTypes(network, config, build.layerDeviceTypes);
}
if (build.safe)
{
config.setEngineCapability(sys.DLACore != -1 ? EngineCapability::kDLA_STANDALONE : EngineCapability::kSAFETY);
}
if (build.restricted)
{
config.setFlag(BuilderFlag::kSAFETY_SCOPE);
}
if (sys.DLACore != -1)
{
if (sys.DLACore < builder.getNbDLACores())
{
config.setDefaultDeviceType(DeviceType::kDLA);
config.setDLACore(sys.DLACore);
config.setFlag(BuilderFlag::kPREFER_PRECISION_CONSTRAINTS);
if (sys.fallback)
{
config.setFlag(BuilderFlag::kGPU_FALLBACK);
}
else
{
// Reformatting runs on GPU, so avoid I/O reformatting.
config.setFlag(BuilderFlag::kDIRECT_IO);
}
if (!build.int8)
{
config.setFlag(BuilderFlag::kFP16);
}
}
else
{
err << "Cannot create DLA engine, " << sys.DLACore << " not available" << std::endl;
return false;
}
}
if (build.enabledTactics || build.disabledTactics)
{
TacticSources tacticSources = config.getTacticSources();
tacticSources |= build.enabledTactics;
tacticSources &= ~build.disabledTactics;
config.setTacticSources(tacticSources);
}
config.setHardwareCompatibilityLevel(build.hardwareCompatibilityLevel);
if (build.maxAuxStreams != defaultMaxAuxStreams)
{
config.setMaxAuxStreams(build.maxAuxStreams);
}
return true;
}
//!
//! \brief Create a serialized engine for a network defintion
//!
//! \return Whether the engine creation succeeds or fails.
//!
bool networkToSerializedEngine(
BuildOptions const& build, SystemOptions const& sys, IBuilder& builder, BuildEnvironment& env, std::ostream& err)
{
std::unique_ptr<IBuilderConfig> config{builder.createBuilderConfig()};
std::unique_ptr<nvinfer1::IInt8Calibrator> calibrator;
std::vector<std::vector<int8_t>> sparseWeights;
SMP_RETVAL_IF_FALSE(config != nullptr, "Config creation failed", false, err);
SMP_RETVAL_IF_FALSE(
setupNetworkAndConfig(build, sys, builder, *env.network, *config, calibrator, err, sparseWeights),
"Network And Config setup failed", false, err);
std::unique_ptr<ITimingCache> timingCache{nullptr};
// Try to load cache from file. Create a fresh cache if the file doesn't exist
if (build.timingCacheMode == TimingCacheMode::kGLOBAL)
{
std::vector<char> loadedCache = samplesCommon::loadTimingCacheFile(build.timingCacheFile);
timingCache.reset(config->createTimingCache(static_cast<const void*>(loadedCache.data()), loadedCache.size()));
SMP_RETVAL_IF_FALSE(timingCache != nullptr, "TimingCache creation failed", false, err);
config->setTimingCache(*timingCache, false);
}
// CUDA stream used for profiling by the builder.
auto profileStream = samplesCommon::makeCudaStream();
SMP_RETVAL_IF_FALSE(profileStream != nullptr, "Cuda stream creation failed", false, err);
config->setProfileStream(*profileStream);
std::unique_ptr<IHostMemory> serializedEngine{builder.buildSerializedNetwork(*env.network, *config)};
SMP_RETVAL_IF_FALSE(serializedEngine != nullptr, "Engine could not be created from network", false, err);
env.engine.setBlob(serializedEngine->data(), serializedEngine->size());
if (build.safe && build.consistency)
{
checkSafeEngine(serializedEngine->data(), serializedEngine->size());
}
if (build.timingCacheMode == TimingCacheMode::kGLOBAL)
{
auto timingCache = config->getTimingCache();
samplesCommon::updateTimingCacheFile(build.timingCacheFile, timingCache);
}
return true;
}
//!
//! \brief Parse a given model, create a network and an engine.
//!
bool modelToBuildEnv(
ModelOptions const& model, BuildOptions const& build, SystemOptions& sys, BuildEnvironment& env, std::ostream& err)
{
env.builder.reset(createBuilder());
SMP_RETVAL_IF_FALSE(env.builder != nullptr, "Builder creation failed", false, err);
env.builder->setErrorRecorder(&gRecorder);
auto networkFlags
= (build.maxBatch) ? 0U : 1U << static_cast<uint32_t>(nvinfer1::NetworkDefinitionCreationFlag::kEXPLICIT_BATCH);
for (auto const& pluginPath : sys.dynamicPlugins)
{
env.builder->getPluginRegistry().loadLibrary(pluginPath.c_str());
}
env.network.reset(env.builder->createNetworkV2(networkFlags));
std::vector<std::string> vcPluginLibrariesUsed;
SMP_RETVAL_IF_FALSE(env.network != nullptr, "Network creation failed", false, err);
env.parser = modelToNetwork(model, *env.network, err, build.versionCompatible ? &vcPluginLibrariesUsed : nullptr);
SMP_RETVAL_IF_FALSE(env.parser.operator bool(), "Parsing model failed", false, err);
if (build.versionCompatible && !sys.ignoreParsedPluginLibs && !vcPluginLibrariesUsed.empty())
{
sample::gLogInfo << "The following plugin libraries were identified by the parser as required for a "
"version-compatible engine:"
<< std::endl;
for (auto const& lib : vcPluginLibrariesUsed)
{
sample::gLogInfo << " " << lib << std::endl;
}
if (!build.excludeLeanRuntime)
{
sample::gLogInfo << "These libraries will be added to --setPluginsToSerialize since --excludeLeanRuntime "
"was not specified."
<< std::endl;
std::copy(vcPluginLibrariesUsed.begin(), vcPluginLibrariesUsed.end(),
std::back_inserter(sys.setPluginsToSerialize));
}
sample::gLogInfo << "These libraries will be added to --dynamicPlugins for use at inference time." << std::endl;
std::copy(vcPluginLibrariesUsed.begin(), vcPluginLibrariesUsed.end(), std::back_inserter(sys.dynamicPlugins));
// Implicitly-added plugins from ONNX parser should be loaded into plugin registry as well.
for (auto const& pluginPath : vcPluginLibrariesUsed)
{
env.builder->getPluginRegistry().loadLibrary(pluginPath.c_str());
}
sample::gLogInfo << "Use --ignoreParsedPluginLibs to disable this behavior." << std::endl;
}
SMP_RETVAL_IF_FALSE(
networkToSerializedEngine(build, sys, *env.builder, env, err), "Building engine failed", false, err);
return true;
}
namespace
{
std::pair<std::vector<std::string>, std::vector<WeightsRole>> getLayerWeightsRolePair(IRefitter& refitter)
{
// Get number of refittable items.
auto const nbAll = refitter.getAll(0, nullptr, nullptr);
std::vector<char const*> layerNames(nbAll);
// Allocate buffers for the items and get them.
std::vector<nvinfer1::WeightsRole> weightsRoles(nbAll);
refitter.getAll(nbAll, layerNames.data(), weightsRoles.data());
std::vector<std::string> layerNameStrs(nbAll);
std::transform(layerNames.begin(), layerNames.end(), layerNameStrs.begin(), [](char const* name) {
if (name == nullptr)
{
return std::string{};
}
return std::string{name};
});
return {layerNameStrs, weightsRoles};
}
std::pair<std::vector<std::string>, std::vector<WeightsRole>> getMissingLayerWeightsRolePair(IRefitter& refitter)
{
// Get number of refittable items.
auto const nbMissing = refitter.getMissing(0, nullptr, nullptr);
std::vector<char const*> layerNames(nbMissing);
// Allocate buffers for the items and get them.
std::vector<nvinfer1::WeightsRole> weightsRoles(nbMissing);
refitter.getMissing(nbMissing, layerNames.data(), weightsRoles.data());
std::vector<std::string> layerNameStrs(nbMissing);
std::transform(layerNames.begin(), layerNames.end(), layerNameStrs.begin(), [](char const* name) {
if (name == nullptr)
{
return std::string{};
}
return std::string{name};
});
return {layerNameStrs, weightsRoles};
}
} // namespace
bool loadEngineToBuildEnv(std::string const& engine, bool enableConsistency, BuildEnvironment& env, std::ostream& err)
{
std::ifstream engineFile(engine, std::ios::binary);
SMP_RETVAL_IF_FALSE(engineFile.good(), "", false, err << "Error opening engine file: " << engine);
engineFile.seekg(0, std::ifstream::end);
int64_t fsize = engineFile.tellg();
engineFile.seekg(0, std::ifstream::beg);
std::vector<uint8_t> engineBlob(fsize);
engineFile.read(reinterpret_cast<char*>(engineBlob.data()), fsize);
SMP_RETVAL_IF_FALSE(engineFile.good(), "", false, err << "Error loading engine file: " << engine);
if (enableConsistency)
{
checkSafeEngine(engineBlob.data(), fsize);
}
env.engine.setBlob(engineBlob.data(), engineBlob.size());
return true;
}
void dumpRefittable(nvinfer1::ICudaEngine& engine)
{
std::unique_ptr<IRefitter> refitter{createRefitter(engine)};
if (refitter == nullptr)
{
sample::gLogError << "Failed to create a refitter." << std::endl;
return;
}
auto const& layerWeightsRolePair = getLayerWeightsRolePair(*refitter);
auto const& layerNames = layerWeightsRolePair.first;
auto const& weightsRoles = layerWeightsRolePair.second;
auto const nbAll = layerWeightsRolePair.first.size();
for (size_t i = 0; i < nbAll; ++i)
{
sample::gLogInfo << layerNames[i] << " " << weightsRoles[i] << std::endl;
}
}
ICudaEngine* loadEngine(std::string const& engine, int32_t DLACore, std::ostream& err)
{
BuildEnvironment env(/* isSafe */ false, /* versionCompatible */ false, DLACore, "", getTempfileControlDefaults());
return loadEngineToBuildEnv(engine, false, env, err) ? env.engine.release() : nullptr;
}
bool saveEngine(const ICudaEngine& engine, std::string const& fileName, std::ostream& err)
{
std::ofstream engineFile(fileName, std::ios::binary);
if (!engineFile)
{
err << "Cannot open engine file: " << fileName << std::endl;
return false;
}
std::unique_ptr<IHostMemory> serializedEngine{engine.serialize()};
if (serializedEngine == nullptr)
{
err << "Engine serialization failed" << std::endl;
return false;
}
engineFile.write(static_cast<char*>(serializedEngine->data()), serializedEngine->size());
return !engineFile.fail();
}
bool getEngineBuildEnv(
const ModelOptions& model, BuildOptions const& build, SystemOptions& sys, BuildEnvironment& env, std::ostream& err)
{
bool createEngineSuccess{false};
if (build.load)
{
createEngineSuccess = loadEngineToBuildEnv(build.engine, build.safe && build.consistency, env, err);
}
else
{
createEngineSuccess = modelToBuildEnv(model, build, sys, env, err);
}
SMP_RETVAL_IF_FALSE(createEngineSuccess, "Failed to create engine from model or file.", false, err);
if (build.save)
{
std::ofstream engineFile(build.engine, std::ios::binary);
engineFile.write(reinterpret_cast<char const*>(env.engine.getBlob().data()), env.engine.getBlob().size());
SMP_RETVAL_IF_FALSE(!engineFile.fail(), "Saving engine to file failed.", false, err);
}
return true;
}
// There is not a getWeightsName API, so we need to use WeightsRole.
std::vector<std::pair<WeightsRole, Weights>> getAllRefitWeightsForLayer(const ILayer& l)
{
switch (l.getType())
{
case LayerType::kCONSTANT:
{
auto const& layer = static_cast<const nvinfer1::IConstantLayer&>(l);
auto const weights = layer.getWeights();
switch (weights.type)
{
case DataType::kFLOAT:
case DataType::kHALF:
case DataType::kINT8:
case DataType::kINT32: return {std::make_pair(WeightsRole::kCONSTANT, weights)};
case DataType::kBOOL:
case DataType::kUINT8:
case DataType::kFP8:
// Refit not supported for these types.
break;
}
break;
}
case LayerType::kCONVOLUTION:
{
auto const& layer = static_cast<const nvinfer1::IConvolutionLayer&>(l);
return {std::make_pair(WeightsRole::kKERNEL, layer.getKernelWeights()),
std::make_pair(WeightsRole::kBIAS, layer.getBiasWeights())};
}
case LayerType::kDECONVOLUTION:
{
auto const& layer = static_cast<const nvinfer1::IDeconvolutionLayer&>(l);
return {std::make_pair(WeightsRole::kKERNEL, layer.getKernelWeights()),
std::make_pair(WeightsRole::kBIAS, layer.getBiasWeights())};
}
case LayerType::kFULLY_CONNECTED:
{
auto const& layer = static_cast<const nvinfer1::IFullyConnectedLayer&>(l);
return {std::make_pair(WeightsRole::kKERNEL, layer.getKernelWeights()),
std::make_pair(WeightsRole::kBIAS, layer.getBiasWeights())};
}
case LayerType::kSCALE:
{
auto const& layer = static_cast<const nvinfer1::IScaleLayer&>(l);
return {std::make_pair(WeightsRole::kSCALE, layer.getScale()),
std::make_pair(WeightsRole::kSHIFT, layer.getShift())};
}
case LayerType::kACTIVATION:
case LayerType::kASSERTION:
case LayerType::kCAST:
case LayerType::kCONCATENATION:
case LayerType::kCONDITION:
case LayerType::kCONDITIONAL_INPUT:
case LayerType::kCONDITIONAL_OUTPUT:
case LayerType::kDEQUANTIZE:
case LayerType::kEINSUM:
case LayerType::kELEMENTWISE:
case LayerType::kFILL:
case LayerType::kGATHER:
case LayerType::kGRID_SAMPLE:
case LayerType::kIDENTITY:
case LayerType::kITERATOR:
case LayerType::kLOOP_OUTPUT:
case LayerType::kLRN:
case LayerType::kMATRIX_MULTIPLY:
case LayerType::kNMS:
case LayerType::kNON_ZERO:
case LayerType::kNORMALIZATION:
case LayerType::kONE_HOT:
case LayerType::kPADDING:
case LayerType::kPARAMETRIC_RELU:
case LayerType::kPLUGIN:
case LayerType::kPLUGIN_V2:
case LayerType::kPOOLING:
case LayerType::kQUANTIZE:
case LayerType::kRAGGED_SOFTMAX:
case LayerType::kRECURRENCE:
case LayerType::kREDUCE:
case LayerType::kRESIZE:
case LayerType::kREVERSE_SEQUENCE:
case LayerType::kRNN_V2:
case LayerType::kSCATTER:
case LayerType::kSELECT:
case LayerType::kSHAPE:
case LayerType::kSHUFFLE:
case LayerType::kSLICE:
case LayerType::kSOFTMAX:
case LayerType::kTOPK:
case LayerType::kTRIP_LIMIT:
case LayerType::kUNARY: return {};
}
return {};
}
bool timeRefit(INetworkDefinition const& network, nvinfer1::ICudaEngine& engine, bool multiThreading)
{
using time_point = std::chrono::time_point<std::chrono::steady_clock>;
using durationMs = std::chrono::duration<float, std::milli>;
auto const nbLayers = network.getNbLayers();
std::unique_ptr<IRefitter> refitter{createRefitter(engine)};
// Set max threads that can be used by refitter.
if (multiThreading && !refitter->setMaxThreads(10))
{
sample::gLogError << "Failed to set max threads to refitter." << std::endl;
return false;
}
auto const& layerWeightsRolePair = getLayerWeightsRolePair(*refitter);
// We use std::string instead of char const* since we can have copies of layer names.
std::set<std::pair<std::string, WeightsRole>> layerRoleSet;
auto const& layerNames = layerWeightsRolePair.first;
auto const& weightsRoles = layerWeightsRolePair.second;
std::transform(layerNames.begin(), layerNames.end(), weightsRoles.begin(),
std::inserter(layerRoleSet, layerRoleSet.begin()),
[](std::string const& layerName, WeightsRole const role) { return std::make_pair(layerName, role); });
auto const isRefittable = [&layerRoleSet](char const* layerName, WeightsRole const role) {
return layerRoleSet.find(std::make_pair(layerName, role)) != layerRoleSet.end();
};
auto const setWeights = [&] {
for (int32_t i = 0; i < nbLayers; i++)
{
auto const layer = network.getLayer(i);
auto const roleWeightsVec = getAllRefitWeightsForLayer(*layer);
for (auto const& roleWeights : roleWeightsVec)
{
if (isRefittable(layer->getName(), roleWeights.first))
{
bool const success = refitter->setWeights(layer->getName(), roleWeights.first, roleWeights.second);
if (!success)
{
return false;
}
}
}
}
return true;
};
auto const reportMissingWeights = [&] {
auto const& missingPair = getMissingLayerWeightsRolePair(*refitter);
auto const& layerNames = missingPair.first;
auto const& weightsRoles = missingPair.second;
for (size_t i = 0; i < layerNames.size(); ++i)
{
sample::gLogError << "Missing (" << layerNames[i] << ", " << weightsRoles[i] << ") for refitting."
<< std::endl;
}
return layerNames.empty();
};
// Warm up and report missing weights
bool const success = setWeights() && reportMissingWeights() && refitter->refitCudaEngine();
if (!success)
{
return false;
}
constexpr int32_t loop = 5;
time_point const refitStartTime{std::chrono::steady_clock::now()};
{
for (int32_t l = 0; l < loop; l++)
{
bool const success = setWeights() && refitter->refitCudaEngine();
if (!success)
{
return false;
}
}
}
time_point const refitEndTime{std::chrono::steady_clock::now()};
sample::gLogInfo << "Engine refitted"
<< " in " << durationMs(refitEndTime - refitStartTime).count() / loop << " ms." << std::endl;
return true;
}
namespace
{
void* initSafeRuntime()
{
void* handle{nullptr};
#if !defined(_WIN32)
std::string const dllName{samplesCommon::isDebug() ? "libnvinfer_safe_debug.so.8" : "libnvinfer_safe.so.8"};
#if SANITIZER_BUILD
handle = dlopen(dllName.c_str(), RTLD_LAZY | RTLD_NODELETE);
#else
handle = dlopen(dllName.c_str(), RTLD_LAZY);
#endif
#endif
return handle;
}
void* initConsistencyCheckerLibrary()
{
void* handle{nullptr};
#if !defined(_WIN32)
std::string const dllName{samplesCommon::isDebug() ? "libnvinfer_checker_debug.so.8" : "libnvinfer_checker.so.8"};
#if SANITIZER_BUILD
handle = dlopen(dllName.c_str(), RTLD_LAZY | RTLD_NODELETE);
#else
handle = dlopen(dllName.c_str(), RTLD_LAZY);
#endif
#endif
return handle;
}
#if !defined(_WIN32)
struct DllDeleter
{
void operator()(void* handle)
{
if (handle != nullptr)
{
dlclose(handle);
}
}
};
const std::unique_ptr<void, DllDeleter> safeRuntimeLibrary{initSafeRuntime()};
const std::unique_ptr<void, DllDeleter> consistencyCheckerLibrary{initConsistencyCheckerLibrary()};
#endif
} // namespace
bool hasSafeRuntime()
{
bool ret{false};
#if !defined(_WIN32)
ret = (safeRuntimeLibrary != nullptr);
#endif
return ret;
}
nvinfer1::safe::IRuntime* createSafeInferRuntime(nvinfer1::ILogger& logger) noexcept
{
nvinfer1::safe::IRuntime* runtime{nullptr};
#if !defined(_WIN32)
constexpr char symbolName[] = "_ZN8nvinfer14safe18createInferRuntimeERNS_7ILoggerE";
typedef nvinfer1::safe::IRuntime* (*CreateInferRuntimeFn)(nvinfer1::ILogger & logger);
if (hasSafeRuntime())
{
auto createFn = reinterpret_cast<CreateInferRuntimeFn>(dlsym(safeRuntimeLibrary.get(), symbolName));
if (createFn != nullptr)
{
runtime = createFn(logger);
}
}
#endif
return runtime;
}
bool hasConsistencyChecker()
{
bool ret{false};
#if !defined(_WIN32)
ret = (consistencyCheckerLibrary != nullptr);
#endif
return ret;
}
nvinfer1::consistency::IConsistencyChecker* createConsistencyChecker(
nvinfer1::ILogger& logger, void const* serializedEngine, int32_t const engineSize) noexcept
{
nvinfer1::consistency::IConsistencyChecker* checker{nullptr};
if (serializedEngine == nullptr || engineSize == 0)
{
return checker;
}
#if !defined(_WIN32)
constexpr char symbolName[] = "createConsistencyChecker_INTERNAL";
typedef nvinfer1::consistency::IConsistencyChecker* (*CreateCheckerFn)(
nvinfer1::ILogger * logger, void const* data, size_t size, uint32_t version);
if (hasSafeRuntime())
{
auto createFn = reinterpret_cast<CreateCheckerFn>(dlsym(consistencyCheckerLibrary.get(), symbolName));
if (createFn != nullptr)
{
checker = createFn(&logger, serializedEngine, engineSize, NV_TENSORRT_VERSION);
}
}
#endif
return checker;
}
bool checkSafeEngine(void const* serializedEngine, int32_t const engineSize)
{
if (!hasConsistencyChecker())
{
sample::gLogError << "Cannot perform consistency check because the checker is not loaded.." << std::endl;
return false;
}
auto checker = std::unique_ptr<nvinfer1::consistency::IConsistencyChecker>(
createConsistencyChecker(sample::gLogger.getTRTLogger(), serializedEngine, engineSize));
if (checker.get() == nullptr)
{
sample::gLogError << "Failed to create consistency checker." << std::endl;
return false;
}
sample::gLogInfo << "Start consistency checking." << std::endl;
if (!checker->validate())
{
sample::gLogError << "Consistency validation failed." << std::endl;
return false;
}
sample::gLogInfo << "Consistency validation passed." << std::endl;
return true;
}
} // namespace sample