NvInferRuntime.h

转到此文件的文档。

/*
 * SPDX-FileCopyrightText: 版权所有 (c) 1993-2024 NVIDIA CORPORATION 及附属公司。保留所有权利。
 * SPDX-License-Identifier: Apache-2.0
 *
 * 根据 Apache License 2.0 版本（“许可证”）获得许可；
 * 除非遵守许可证，否则您不得使用此文件。
 * 您可以在以下网址获取许可证副本：
 *
 * https://apache.ac.cn/licenses/LICENSE-2.0
 *
 * 除非适用法律要求或以书面形式达成一致，否则
 * 根据许可证分发的软件以“现状”为基础分发，
 * 不附带任何形式的明示或暗示的保证或条件。
 * 请参阅许可证，以了解有关权限和
 * 许可证限制的具体语言。
 */
 
#ifndef NV_INFER_RUNTIME_H
#define NV_INFER_RUNTIME_H
 
 
#include "NvInferImpl.h"
#define NV_INFER_INTERNAL_INCLUDE 1
#include "NvInferPluginBase.h"
#undef NV_INFER_INTERNAL_INCLUDE
#include "NvInferRuntimeCommon.h"
 
namespace nvinfer1
{
 
class IExecutionContext;
class ICudaEngine;
class IPluginFactory;
class IEngineInspector;
 
 
class INoCopy
{
protected
 INoCopy() = default;
 virtual ~INoCopy() = default;
 INoCopy(INoCopy const& other) = delete;
 INoCopy& operator=(INoCopy const& other) = delete;
 INoCopy(INoCopy&& other) = delete;
 INoCopy& operator=(INoCopy&& other) = delete;
};
 
enum class EngineCapability : int32_t
{
 kSTANDARD = 0,
 
 kSAFETY = 1,
 
 kDLA_STANDALONE = 2,
};
 
namespace impl
{
template <>
struct EnumMaxImpl<EngineCapability>
{
 static constexpr int32_t kVALUE = 3;
};
} // namespace impl
 
class Weights
{
public
 DataType type;
 void const* values;
 int64_t count;
};
 
class IHostMemory : public INoCopy
{
public
 virtual ~IHostMemory() noexcept = default;
 
 void* data() const noexcept
    {
 return mImpl->data();
    }
 
 std::size_t size() const noexcept
    {
 return mImpl->size();
    }
 
 DataType type() const noexcept
    {
 return mImpl->type();
    }
 
protected
 apiv::VHostMemory* mImpl;
};
 
enum class DimensionOperation : int32_t
{
 kSUM = 0,
 kPROD = 1,
 kMAX = 2,
 kMIN = 3,
 kSUB = 4,
 kEQUAL = 5,
 kLESS = 6,
 kFLOOR_DIV = 7,
 kCEIL_DIV = 8
};
 
template <>
constexpr inline int32_t EnumMax<DimensionOperation>() noexcept
{
 return 9;
}
 
enum class TensorLocation : int32_t
{
 kDEVICE = 0,
 kHOST = 1,
};
 
namespace impl
{
template <>
struct EnumMaxImpl<TensorLocation>
{
 static constexpr int32_t kVALUE = 2;
};
} // namespace impl
 
class IDimensionExpr : public INoCopy
{
public
 bool isConstant() const noexcept
    {
 return mImpl->isConstant();
    }
 
 int64_t getConstantValue() const noexcept
    {
 return mImpl->getConstantValue();
    }
 
protected
 apiv::VDimensionExpr* mImpl;
 virtual ~IDimensionExpr() noexcept = default;
 
public
 bool isSizeTensor() const noexcept
    {
 return mImpl->isSizeTensor();
    }
};
 
class IExprBuilder : public INoCopy
{
public
 IDimensionExpr const* constant(int64_t value) noexcept
    {
 return mImpl->constant(value);
    }
 
 IDimensionExpr const* operation(
 DimensionOperation op, IDimensionExpr const& first, IDimensionExpr const& second) noexcept
    {
 return mImpl->operation(op, first, second);
    }
 
protected
 apiv::VExprBuilder* mImpl;
 virtual ~IExprBuilder() noexcept = default;
 
public
 IDimensionExpr const* declareSizeTensor(int32_t outputIndex, IDimensionExpr const& opt, IDimensionExpr const& upper)
    {
 return mImpl->declareSizeTensor(outputIndex, opt, upper);
    }
};
 
class DimsExprs
{
public
 int32_t nbDims;
 IDimensionExpr const* d[Dims::MAX_DIMS];
};
 
struct DynamicPluginTensorDesc
{
 PluginTensorDesc desc;
 
 Dims min;
 
 Dims max;
 
 Dims opt;
};
 
class TRT_DEPRECATED IPluginV2DynamicExt : public nvinfer1::IPluginV2Ext
{
public
 IPluginV2DynamicExt* clone() const noexcept override = 0;
 
 virtual DimsExprs getOutputDimensions(
 int32_t outputIndex, DimsExprs const* inputs, int32_t nbInputs, IExprBuilder& exprBuilder) noexcept = 0;
 
 static constexpr int32_t kFORMAT_COMBINATION_LIMIT = 100;
 
 virtual bool supportsFormatCombination(
 int32_t pos, PluginTensorDesc const* inOut, int32_t nbInputs, int32_t nbOutputs) noexcept = 0;
 
 virtual void configurePlugin(DynamicPluginTensorDesc const* in, int32_t nbInputs,
 DynamicPluginTensorDesc const* out, int32_t nbOutputs) noexcept = 0;
 
 virtual size_t getWorkspaceSize(PluginTensorDesc const* inputs, int32_t nbInputs, PluginTensorDesc const* outputs,
 int32_t nbOutputs) const noexcept = 0;
 
 virtual int32_t enqueue(PluginTensorDesc const* inputDesc, PluginTensorDesc const* outputDesc,
 void const* const* inputs, void* const* outputs, void* workspace, cudaStream_t stream) noexcept = 0;
 
protected
 int32_t getTensorRTVersion() const noexcept override
    {
 return (static_cast<int32_t>(PluginVersion::kV2_DYNAMICEXT) << 24 | (NV_TENSORRT_VERSION & 0xFFFFFF));
    }
 
 virtual ~IPluginV2DynamicExt() noexcept {}
 
private
 // Following are obsolete base class methods, and must not be implemented or used.
 
 void configurePlugin(Dims const*, int32_t, Dims const*, int32_t, DataType const*, DataType const*, bool const*,
 bool const*, PluginFormat, int32_t) noexcept override final
    {
    }
 
 bool supportsFormat(DataType, PluginFormat) const noexcept override final
    {
 return false;
    }
 
 Dims getOutputDimensions(int32_t, Dims const*, int32_t) noexcept override final
    {
 return Dims{-1, {}};
    }
 
 TRT_DEPRECATED bool isOutputBroadcastAcrossBatch(int32_t, bool const*, int32_t) const noexcept override final
    {
 return false;
    }
 
 TRT_DEPRECATED bool canBroadcastInputAcrossBatch(int32_t) const noexcept override final
    {
 return true;
    }
 
 size_t getWorkspaceSize(int32_t) const noexcept override final
    {
 return 0;
    }
 
 int32_t enqueue(int32_t, void const* const*, void* const*, void*, cudaStream_t) noexcept override final
    {
 return 1;
    }
};
 
namespace v_1_0
{
class IStreamReader : public IVersionedInterface
{
public
 ~IStreamReader() override = default;
 IStreamReader() = default;
 
 InterfaceInfo getInterfaceInfo() const noexcept override
    {
 return InterfaceInfo{"IStreamReader", 1, 0};
    }
 
 virtual int64_t read(void* destination, int64_t nbBytes) = 0;
 
protected
 IStreamReader(IStreamReader const&) = default;
 IStreamReader(IStreamReader&&) = default;
 IStreamReader& operator=(IStreamReader const&) & = default;
 IStreamReader& operator=(IStreamReader&&) & = default;
};
} // namespace v_1_0
 
using IStreamReader = v_1_0::IStreamReader;
 
enum class SeekPosition : int32_t
{
 kSET = 0,
 
 kCUR = 1,
 
 kEND = 2,
};
 
namespace v_1_0
{
class IStreamReaderV2 : public IVersionedInterface
{
public
 ~IStreamReaderV2() override = default;
 IStreamReaderV2() = default;
 
 InterfaceInfo getInterfaceInfo() const noexcept override
    {
 return InterfaceInfo{"IStreamReaderV2", 1, 0};
    }
 
 virtual int64_t read(void* destination, int64_t nbBytes, cudaStream_t stream) noexcept = 0;
 
 virtual bool seek(int64_t offset, SeekPosition where) noexcept = 0;
 
protected
 IStreamReaderV2(IStreamReaderV2 const&) = default;
 IStreamReaderV2(IStreamReaderV2&&) = default;
 IStreamReaderV2& operator=(IStreamReaderV2 const&) & = default;
 IStreamReaderV2& operator=(IStreamReaderV2&&) & = default;
};
} // 命名空间 v_1_0
 
using IStreamReaderV2 = v_1_0::IStreamReaderV2;
 
class IPluginResourceContext
{
public
 virtual IGpuAllocator* getGpuAllocator() const noexcept = 0;
 
 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
 virtual ~IPluginResourceContext() noexcept = default;
 
protected
 IPluginResourceContext() = default;
 IPluginResourceContext(IPluginResourceContext const&) = default;
 IPluginResourceContext(IPluginResourceContext&&) = default;
 IPluginResourceContext& operator=(IPluginResourceContext const&) & = default;
 IPluginResourceContext& operator=(IPluginResourceContext&&) & = default;
};
 
namespace v_1_0
{
class IPluginV3OneCore : public IPluginCapability
{
public
 InterfaceInfo getInterfaceInfo() const noexcept override
    {
 return InterfaceInfo{"PLUGIN_V3ONE_CORE", 1, 0};
    }
 
 virtual AsciiChar const* getPluginName() const noexcept = 0;
 
 virtual AsciiChar const* getPluginVersion() const noexcept = 0;
 
 virtual AsciiChar const* getPluginNamespace() const noexcept = 0;
};
 
class IPluginV3OneBuild : public IPluginCapability
{
public
 static constexpr int32_t kDEFAULT_FORMAT_COMBINATION_LIMIT = 100;
 
 InterfaceInfo getInterfaceInfo() const noexcept override
    {
 return InterfaceInfo{"PLUGIN_V3ONE_BUILD", 1, 0};
    }
 
 virtual int32_t configurePlugin(DynamicPluginTensorDesc const* in, int32_t nbInputs,
 DynamicPluginTensorDesc const* out, int32_t nbOutputs) noexcept = 0;
 
 virtual int32_t getOutputDataTypes(
 DataType* outputTypes, int32_t nbOutputs, const DataType* inputTypes, int32_t nbInputs) const noexcept = 0;
 
 virtual int32_t getOutputShapes(DimsExprs const* inputs, int32_t nbInputs, DimsExprs const* shapeInputs,
 int32_t nbShapeInputs, DimsExprs* outputs, int32_t nbOutputs, IExprBuilder& exprBuilder) noexcept = 0;
 
 virtual bool supportsFormatCombination(
 int32_t pos, DynamicPluginTensorDesc const* inOut, int32_t nbInputs, int32_t nbOutputs) noexcept = 0;
 
 virtual int32_t getNbOutputs() const noexcept = 0;
 
 virtual size_t getWorkspaceSize(DynamicPluginTensorDesc const* inputs, int32_t nbInputs,
 DynamicPluginTensorDesc const* outputs, int32_t nbOutputs) const noexcept
    {
 return 0;
    }
 
 virtual int32_t getValidTactics(int32_t* tactics, int32_t nbTactics) noexcept
    {
 return 0;
    }
 
 virtual int32_t getNbTactics() noexcept
    {
 return 0;
    }
 
 virtual char const* getTimingCacheID() noexcept
    {
 return nullptr;
    }
 
 virtual int32_t getFormatCombinationLimit() noexcept
    {
 return kDEFAULT_FORMAT_COMBINATION_LIMIT;
    }
 
 virtual char const* getMetadataString() noexcept
    {
 return nullptr;
    }
};
 
class IPluginV3OneRuntime : public IPluginCapability
{
public
 InterfaceInfo getInterfaceInfo() const noexcept override
    {
 return InterfaceInfo{"PLUGIN_V3ONE_RUNTIME", 1, 0};
    }
 
 virtual int32_t setTactic(int32_t tactic) noexcept
    {
 return 0;
    }
 
 virtual int32_t onShapeChange(
 PluginTensorDesc const* in, int32_t nbInputs, PluginTensorDesc const* out, int32_t nbOutputs) noexcept = 0;
 
 virtual int32_t enqueue(PluginTensorDesc const* inputDesc, PluginTensorDesc const* outputDesc,
 void const* const* inputs, void* const* outputs, void* workspace, cudaStream_t stream) noexcept = 0;
 
 virtual IPluginV3* attachToContext(IPluginResourceContext* context) noexcept = 0;
 
 
 virtual PluginFieldCollection const* getFieldsToSerialize() noexcept = 0;
};
} // 命名空间 v_1_0
 
namespace v_2_0
{
 
class IPluginV3OneBuild : public v_1_0::IPluginV3OneBuild
{
public
 InterfaceInfo getInterfaceInfo() const noexcept override
    {
 return InterfaceInfo{"PLUGIN_V3ONE_BUILD", 2, 0};
    }
 
 virtual int32_t getAliasedInput(int32_t outputIndex) noexcept
    {
 return -1;
    }
};
 
} // 命名空间 v_2_0
 
using IPluginV3OneCore = v_1_0::IPluginV3OneCore;
 
using IPluginV3OneBuild = v_1_0::IPluginV3OneBuild;
 
using IPluginV3OneRuntime = v_1_0::IPluginV3OneRuntime;
 
using IPluginV3OneBuildV2 = v_2_0::IPluginV3OneBuild;
 
namespace v_1_0
{
class IProfiler
{
public
 virtual void reportLayerTime(char const* layerName, float ms) noexcept = 0;
 
 virtual ~IProfiler() noexcept {}
};
} // 命名空间 v_1_0
 
using IProfiler = v_1_0::IProfiler;
 
enum class WeightsRole : int32_t
{
 kKERNEL = 0,
 kBIAS = 1,
 kSHIFT = 2,
 kSCALE = 3,
 kCONSTANT = 4,
 kANY = 5,
};
 
template <>
constexpr inline int32_t EnumMax<WeightsRole>() noexcept
{
 return 6;
}
 
enum class DeviceType : int32_t
{
 kGPU = 0,
 kDLA = 1,
};
 
template <>
constexpr inline int32_t EnumMax<DeviceType>() noexcept
{
 return 2;
}
 
enum class TempfileControlFlag : int32_t
{
 kALLOW_IN_MEMORY_FILES = 0,
 
 kALLOW_TEMPORARY_FILES = 1,
};
 
template <>
constexpr inline int32_t EnumMax<TempfileControlFlag>() noexcept
{
 return 2;
}
 
using TempfileControlFlags = uint32_t;
 
enum class TensorFormat : int32_t
{
 kLINEAR = 0,
 
 kCHW2 = 1,
 
 kHWC8 = 2,
 
 kCHW4 = 3,
 
 kCHW16 = 4,
 
 kCHW32 = 5,
 
 kDHWC8 = 6,
 
 kCDHW32 = 7,
 
 kHWC = 8,
 
 kDLA_LINEAR = 9,
 
 kDLA_HWC4 = 10,
 
 kHWC16 = 11,
 
 kDHWC = 12
};
 
namespace impl
{
template <>
struct EnumMaxImpl<TensorFormat>
{
 static constexpr int32_t kVALUE = 13;
};
} // 命名空间 impl
 
enum class AllocatorFlag : int32_t
{
 kRESIZABLE = 0,
};
 
namespace impl
{
template <>
struct EnumMaxImpl<AllocatorFlag>
{
 static constexpr int32_t kVALUE = 1;
};
} // 命名空间 impl
 
using AllocatorFlags = uint32_t;
 
 
class ILogger
{
public
 enum class Severity : int32_t
    {
 kINTERNAL_ERROR = 0,
 kERROR = 1,
 kWARNING = 2,
 kINFO = 3,
 kVERBOSE = 4,
    };
 
 virtual void log(Severity severity, AsciiChar const* msg) noexcept = 0;
 
 ILogger() = default;
 virtual ~ILogger() = default;
 
protected
 // @cond SuppressDoxyWarnings
 ILogger(ILogger const&) = default;
 ILogger(ILogger&&) = default;
 ILogger& operator=(ILogger const&) & = default;
 ILogger& operator=(ILogger&&) & = default;
 // @endcond
};
 
namespace impl
{
template <>
struct EnumMaxImpl<ILogger::Severity>
{
 static constexpr int32_t kVALUE = 5;
};
} // 命名空间 impl
 
namespace v_1_0
{
 
class IGpuAllocator : public IVersionedInterface
{
public
 TRT_DEPRECATED virtual void* allocate(
 uint64_t const size, uint64_t const alignment, AllocatorFlags const flags) noexcept = 0;
 
 ~IGpuAllocator() override = default;
 IGpuAllocator() = default;
 
 virtual void* reallocate(void* const /*baseAddr*/, uint64_t /*alignment*/, uint64_t /*newSize*/) noexcept
    {
 return nullptr;
    }
 
 TRT_DEPRECATED virtual bool deallocate(void* const memory) noexcept = 0;
 
 virtual void* allocateAsync(
 uint64_t const size, uint64_t const alignment, AllocatorFlags const flags, cudaStream_t /*stream*/) noexcept
    {
 return allocate(size, alignment, flags);
    }
 virtual bool deallocateAsync(void* const memory, cudaStream_t /*stream*/) noexcept
    {
 return deallocate(memory);
    }
 
 InterfaceInfo getInterfaceInfo() const noexcept override
    {
 return {"IGpuAllocator", 1, 0};
    }
 
protected
 // @cond SuppressDoxyWarnings
 IGpuAllocator(IGpuAllocator const&) = default;
 IGpuAllocator(IGpuAllocator&&) = default;
 IGpuAllocator& operator=(IGpuAllocator const&) & = default;
 IGpuAllocator& operator=(IGpuAllocator&&) & = default;
 // @endcond
};
 
} // namespace v_1_0
 
using IGpuAllocator = v_1_0::IGpuAllocator;
 
class IRuntime : public INoCopy
{
public
 virtual ~IRuntime() noexcept = default;
 
 void setDLACore(int32_t dlaCore) noexcept
    {
 mImpl->setDLACore(dlaCore);
    }
 
 int32_t getDLACore() const noexcept
    {
 return mImpl->getDLACore();
    }
 
 int32_t getNbDLACores() const noexcept
    {
 return mImpl->getNbDLACores();
    }
 
 void setGpuAllocator(IGpuAllocator* allocator) noexcept
    {
 mImpl->setGpuAllocator(allocator);
    }
 
    //
 void setErrorRecorder(IErrorRecorder* recorder) noexcept
    {
 mImpl->setErrorRecorder(recorder);
    }
 
 IErrorRecorder* getErrorRecorder() const noexcept
    {
 return mImpl->getErrorRecorder();
    }
 
 ICudaEngine* deserializeCudaEngine(void const* blob, std::size_t size) noexcept
    {
 return mImpl->deserializeCudaEngine(blob, size);
    }
 
 TRT_DEPRECATED ICudaEngine* deserializeCudaEngine(IStreamReader& streamReader)
    {
 return mImpl->deserializeCudaEngine(streamReader);
    }
 
 ICudaEngine* deserializeCudaEngine(IStreamReaderV2& streamReader)
    {
 return mImpl->deserializeCudaEngineV2(streamReader);
    }
 
 ILogger* getLogger() const noexcept
    {
 return mImpl->getLogger();
    }
 
 bool setMaxThreads(int32_t maxThreads) noexcept
    {
 return mImpl->setMaxThreads(maxThreads);
    }
 
 int32_t getMaxThreads() const noexcept
    {
 return mImpl->getMaxThreads();
    }
 
 void setTemporaryDirectory(char const* path) noexcept
    {
 return mImpl->setTemporaryDirectory(path);
    }
 
 char const* getTemporaryDirectory() const noexcept
    {
 return mImpl->getTemporaryDirectory();
    }
 
 void setTempfileControlFlags(TempfileControlFlags flags) noexcept
    {
 return mImpl->setTempfileControlFlags(flags);
    }
 
 TempfileControlFlags getTempfileControlFlags() const noexcept
    {
 return mImpl->getTempfileControlFlags();
    }
 
 IPluginRegistry& getPluginRegistry() noexcept
    {
 return mImpl->getPluginRegistry();
    }
 
 IRuntime* loadRuntime(char const* path) noexcept
    {
 return mImpl->loadRuntime(path);
    }
 
 void setEngineHostCodeAllowed(bool allowed) noexcept
    {
 return mImpl->setEngineHostCodeAllowed(allowed);
    }
 
 bool getEngineHostCodeAllowed() const noexcept
    {
 return mImpl->getEngineHostCodeAllowed();
    }
 
protected
 apiv::VRuntime* mImpl;
};
 
class IRefitter : public INoCopy
{
public
 virtual ~IRefitter() noexcept = default;
 
 bool setWeights(char const* layerName, WeightsRole role, Weights weights) noexcept
    {
 return mImpl->setWeights(layerName, role, weights);
    }
 
 bool refitCudaEngine() noexcept
    {
 return mImpl->refitCudaEngine();
    }
 
 int32_t getMissing(int32_t size, char const** layerNames, WeightsRole* roles) noexcept
    {
 return mImpl->getMissing(size, layerNames, roles);
    }
 
 int32_t getAll(int32_t size, char const** layerNames, WeightsRole* roles) noexcept
    {
 return mImpl->getAll(size, layerNames, roles);
    }
 
 TRT_DEPRECATED bool setDynamicRange(char const* tensorName, float min, float max) noexcept
    {
 return mImpl->setDynamicRange(tensorName, min, max);
    }
 
 TRT_DEPRECATED float getDynamicRangeMin(char const* tensorName) const noexcept
    {
 return mImpl->getDynamicRangeMin(tensorName);
    }
 
 TRT_DEPRECATED float getDynamicRangeMax(char const* tensorName) const noexcept
    {
 return mImpl->getDynamicRangeMax(tensorName);
    }
 
 TRT_DEPRECATED int32_t getTensorsWithDynamicRange(int32_t size, char const** tensorNames) const noexcept
    {
 return mImpl->getTensorsWithDynamicRange(size, tensorNames);
    }
 
    //
 void setErrorRecorder(IErrorRecorder* recorder) noexcept
    {
 mImpl->setErrorRecorder(recorder);
    }
 
 IErrorRecorder* getErrorRecorder() const noexcept
    {
 return mImpl->getErrorRecorder();
    }
 
 bool setNamedWeights(char const* name, Weights weights) noexcept
    {
 return mImpl->setNamedWeights(name, weights);
    }
 
 int32_t getMissingWeights(int32_t size, char const** weightsNames) noexcept
    {
 return mImpl->getMissingWeights(size, weightsNames);
    }
 
 int32_t getAllWeights(int32_t size, char const** weightsNames) noexcept
    {
 return mImpl->getAllWeights(size, weightsNames);
    }
 
 ILogger* getLogger() const noexcept
    {
 return mImpl->getLogger();
    }
 
 bool setMaxThreads(int32_t maxThreads) noexcept
    {
 return mImpl->setMaxThreads(maxThreads);
    }
 
 int32_t getMaxThreads() const noexcept
    {
 return mImpl->getMaxThreads();
    }
 
 bool setNamedWeights(char const* name, Weights weights, TensorLocation location) noexcept
    {
 return mImpl->setNamedWeightsWithLocation(name, weights, location);
    }
 
 Weights getNamedWeights(char const* weightsName) const noexcept
    {
 return mImpl->getNamedWeights(weightsName);
    }
 
 TensorLocation getWeightsLocation(char const* weightsName) const noexcept
    {
 return mImpl->getWeightsLocation(weightsName);
    }
 
 bool unsetNamedWeights(char const* weightsName) noexcept
    {
 return mImpl->unsetNamedWeights(weightsName);
    }
 
 void setWeightsValidation(bool weightsValidation) noexcept
    {
 return mImpl->setWeightsValidation(weightsValidation);
    }
 
 bool getWeightsValidation() const noexcept
    {
 return mImpl->getWeightsValidation();
    }
 
 bool refitCudaEngineAsync(cudaStream_t stream) noexcept
    {
 return mImpl->refitCudaEngineAsync(stream);
    }
 
 Weights getWeightsPrototype(char const* weightsName) const noexcept
    {
 return mImpl->getWeightsPrototype(weightsName);
    }
 
protected
 apiv::VRefitter* mImpl;
};
 
enum class OptProfileSelector : int32_t
{
 kMIN = 0,
 kOPT = 1,
 kMAX = 2
};
 
template <>
constexpr inline int32_t EnumMax<OptProfileSelector>() noexcept
{
 return 3;
}
 
class IOptimizationProfile : public INoCopy
{
public
 bool setDimensions(char const* inputName, OptProfileSelector select, Dims const& dims) noexcept
    {
 return mImpl->setDimensions(inputName, select, dims);
    }
 
 Dims getDimensions(char const* inputName, OptProfileSelector select) const noexcept
    {
 return mImpl->getDimensions(inputName, select);
    }
 
 bool setShapeValues(
 char const* inputName, OptProfileSelector select, int32_t const* values, int32_t nbValues) noexcept
    {
 return mImpl->setShapeValues(inputName, select, values, nbValues);
    }
 
 int32_t getNbShapeValues(char const* inputName) const noexcept
    {
 return mImpl->getNbShapeValues(inputName);
    }
 
 int32_t const* getShapeValues(char const* inputName, OptProfileSelector select) const noexcept
    {
 return mImpl->getShapeValues(inputName, select);
    }
 
 bool setExtraMemoryTarget(float target) noexcept
    {
 return mImpl->setExtraMemoryTarget(target);
    }
 
 float getExtraMemoryTarget() const noexcept
    {
 return mImpl->getExtraMemoryTarget();
    }
 
 bool isValid() const noexcept
    {
 return mImpl->isValid();
    }
 
protected
 apiv::VOptimizationProfile* mImpl;
 virtual ~IOptimizationProfile() noexcept = default;
};
 
enum class TacticSource : int32_t
{
 kCUBLAS TRT_DEPRECATED_ENUM = 0,
 
 kCUBLAS_LT TRT_DEPRECATED_ENUM = 1,
 
 kCUDNN TRT_DEPRECATED_ENUM = 2,
 
 kEDGE_MASK_CONVOLUTIONS = 3,
 
 kJIT_CONVOLUTIONS = 4,
};
 
template <>
constexpr inline int32_t EnumMax<TacticSource>() noexcept
{
 return 5;
} 
 
using TacticSources = uint32_t;
 
enum class ProfilingVerbosity : int32_t
{
 kLAYER_NAMES_ONLY = 0,
 kNONE = 1,
 kDETAILED = 2,
};
 
template <>
constexpr inline int32_t EnumMax<ProfilingVerbosity>() noexcept
{
 return 3;
}
 
using SerializationFlags = uint32_t;
 
enum class SerializationFlag : int32_t
{
 kEXCLUDE_WEIGHTS = 0,
 kEXCLUDE_LEAN_RUNTIME = 1,
};
 
template <>
constexpr inline int32_t EnumMax<SerializationFlag>() noexcept
{
 return 2;
}
 
class ISerializationConfig : public INoCopy
{
public
 virtual ~ISerializationConfig() noexcept = default;
 
 bool setFlags(SerializationFlags serializationFlags) noexcept
    {
 return mImpl->setFlags(serializationFlags);
    }
 
 SerializationFlags getFlags() const noexcept
    {
 return mImpl->getFlags();
    }
 
 bool clearFlag(SerializationFlag serializationFlag) noexcept
    {
 return mImpl->clearFlag(serializationFlag);
    }
 
 bool setFlag(SerializationFlag serializationFlag) noexcept
    {
 return mImpl->setFlag(serializationFlag);
    }
 
 bool getFlag(SerializationFlag serializationFlag) const noexcept
    {
 return mImpl->getFlag(serializationFlag);
    }
 
protected
 apiv::VSerializationConfig* mImpl;
};
 
enum class ExecutionContextAllocationStrategy : int32_t
{
 kSTATIC = 0,
 kON_PROFILE_CHANGE = 1,
 kUSER_MANAGED = 2,
};
 
template <>
constexpr inline int32_t EnumMax<ExecutionContextAllocationStrategy>() noexcept
{
 return 3;
}
 
class ICudaEngine : public INoCopy
{
public
 virtual ~ICudaEngine() noexcept = default;
 
 Dims getTensorShape(char const* tensorName) const noexcept
    {
 return mImpl->getTensorShape(tensorName);
    }
 
 DataType getTensorDataType(char const* tensorName) const noexcept
    {
 return mImpl->getTensorDataType(tensorName);
    }
 
 int32_t getNbLayers() const noexcept
    {
 return mImpl->getNbLayers();
    }
 
 IHostMemory* serialize() const noexcept
    {
 return mImpl->serialize();
    }
 
 IExecutionContext* createExecutionContext(
 ExecutionContextAllocationStrategy strategy = ExecutionContextAllocationStrategy::kSTATIC) noexcept
    {
 return mImpl->createExecutionContext(strategy);
    }
 
 TensorLocation getTensorLocation(char const* tensorName) const noexcept
    {
 return mImpl->getTensorLocation(tensorName);
    }
 
 bool isShapeInferenceIO(char const* tensorName) const noexcept
    {
 return mImpl->isShapeInferenceIO(tensorName);
    }
 
 TensorIOMode getTensorIOMode(char const* tensorName) const noexcept
    {
 return mImpl->getTensorIOMode(tensorName);
    }
 
 TRT_DEPRECATED IExecutionContext* createExecutionContextWithoutDeviceMemory() noexcept
    {
 return mImpl->createExecutionContextWithoutDeviceMemory();
    }
 
 TRT_DEPRECATED size_t getDeviceMemorySize() const noexcept
    {
 return mImpl->getDeviceMemorySize();
    }
 
 TRT_DEPRECATED size_t getDeviceMemorySizeForProfile(int32_t profileIndex) const noexcept
    {
 return mImpl->getDeviceMemorySizeForProfile(profileIndex);
    }
 
 int64_t getDeviceMemorySizeV2() const noexcept
    {
 return mImpl->getDeviceMemorySizeV2();
    }
 
 int64_t getDeviceMemorySizeForProfileV2(int32_t profileIndex) const noexcept
    {
 return mImpl->getDeviceMemorySizeForProfileV2(profileIndex);
    }
 
 bool isRefittable() const noexcept
    {
 return mImpl->isRefittable();
    }
 
 int32_t getTensorBytesPerComponent(char const* tensorName) const noexcept
    {
 return mImpl->getTensorBytesPerComponent(tensorName);
    }
 
 int32_t getTensorBytesPerComponent(char const* tensorName, int32_t profileIndex) const noexcept
    {
 return mImpl->getTensorBytesPerComponentV2(tensorName, profileIndex);
    }
 
 int32_t getTensorComponentsPerElement(char const* tensorName) const noexcept
    {
 return mImpl->getTensorComponentsPerElement(tensorName);
    }
 
 int32_t getTensorComponentsPerElement(char const* tensorName, int32_t profileIndex) const noexcept
    {
 return mImpl->getTensorComponentsPerElementV2(tensorName, profileIndex);
    }
 
 TensorFormat getTensorFormat(char const* tensorName) const noexcept
    {
 return mImpl->getTensorFormat(tensorName);
    }
 
 TensorFormat getTensorFormat(char const* tensorName, int32_t profileIndex) const noexcept
    {
 return mImpl->getTensorFormatV2(tensorName, profileIndex);
    }
 
 char const* getTensorFormatDesc(char const* tensorName) const noexcept
    {
 return mImpl->getTensorFormatDesc(tensorName);
    }
 
 char const* getTensorFormatDesc(char const* tensorName, int32_t profileIndex) const noexcept
    {
 return mImpl->getTensorFormatDescV2(tensorName, profileIndex);
    }
 
 int32_t getTensorVectorizedDim(char const* tensorName) const noexcept
    {
 return mImpl->getTensorVectorizedDim(tensorName);
    }
 
 int32_t getTensorVectorizedDim(char const* tensorName, int32_t profileIndex) const noexcept
    {
 return mImpl->getTensorVectorizedDimV2(tensorName, profileIndex);
    }
 
 char const* getName() const noexcept
    {
 return mImpl->getName();
    }
 
 int32_t getNbOptimizationProfiles() const noexcept
    {
 return mImpl->getNbOptimizationProfiles();
    }
 
 Dims getProfileShape(char const* tensorName, int32_t profileIndex, OptProfileSelector select) const noexcept
    {
 return mImpl->getProfileShape(tensorName, profileIndex, select);
    }
 
 int32_t const* getProfileTensorValues(char const* tensorName, int32_t profileIndex, OptProfileSelector select) const
 noexcept
    {
 return mImpl->getProfileTensorValues(tensorName, profileIndex, select);
    }
 
 EngineCapability getEngineCapability() const noexcept
    {
 return mImpl->getEngineCapability();
    }
 
 void setErrorRecorder(IErrorRecorder* recorder) noexcept
    {
 return mImpl->setErrorRecorder(recorder);
    }
 
 IErrorRecorder* getErrorRecorder() const noexcept
    {
 return mImpl->getErrorRecorder();
    }
 
 TRT_DEPRECATED bool hasImplicitBatchDimension() const noexcept
    {
 return mImpl->hasImplicitBatchDimension();
    }
 
 TacticSources getTacticSources() const noexcept
    {
 return mImpl->getTacticSources();
    }
 
 ProfilingVerbosity getProfilingVerbosity() const noexcept
    {
 return mImpl->getProfilingVerbosity();
    }
 
 IEngineInspector* createEngineInspector() const noexcept
    {
 return mImpl->createEngineInspector();
    }
 
 int32_t getNbIOTensors() const noexcept
    {
 return mImpl->getNbIOTensors();
    }
 
 char const* getIOTensorName(int32_t index) const noexcept
    {
 return mImpl->getIOTensorName(index);
    }
 
 HardwareCompatibilityLevel getHardwareCompatibilityLevel() const noexcept
    {
 return mImpl->getHardwareCompatibilityLevel();
    }
 
 int32_t getNbAuxStreams() const noexcept
    {
 return mImpl->getNbAuxStreams();
    }
 
 ISerializationConfig* createSerializationConfig() noexcept
    {
 return mImpl->createSerializationConfig();
    }
 
 IHostMemory* serializeWithConfig(ISerializationConfig& config) const noexcept
    {
 return mImpl->serializeWithConfig(config);
    }
 
 TRT_DEPRECATED bool setWeightStreamingBudget(int64_t gpuMemoryBudget) noexcept
    {
 return mImpl->setWeightStreamingBudget(gpuMemoryBudget);
    }
 
 TRT_DEPRECATED int64_t getWeightStreamingBudget() const noexcept
    {
 return mImpl->getWeightStreamingBudget();
    }
 
 TRT_DEPRECATED int64_t getMinimumWeightStreamingBudget() const noexcept
    {
 return mImpl->getMinimumWeightStreamingBudget();
    }
 
 int64_t getStreamableWeightsSize() const noexcept
    {
 return mImpl->getStreamableWeightsSize();
    }
 
 bool setWeightStreamingBudgetV2(int64_t gpuMemoryBudget) noexcept
    {
 return mImpl->setWeightStreamingBudgetV2(gpuMemoryBudget);
    }
 
 int64_t getWeightStreamingBudgetV2() const noexcept
    {
 return mImpl->getWeightStreamingBudgetV2();
    }
 
 int64_t getWeightStreamingAutomaticBudget() const noexcept
    {
 return mImpl->getWeightStreamingAutomaticBudget();
    }
 
 int64_t getWeightStreamingScratchMemorySize() const noexcept
    {
 return mImpl->getWeightStreamingScratchMemorySize();
    }
 
 bool isDebugTensor(char const* name) const noexcept
    {
 return mImpl->isDebugTensor(name);
    }
 
protected
 apiv::VCudaEngine* mImpl;
};
 
namespace v_1_0
{
class IOutputAllocator : public IVersionedInterface
{
public
 InterfaceInfo getInterfaceInfo() const noexcept override
    {
 return {"IOutputAllocator", 1, 0};
    }
 
 TRT_DEPRECATED virtual void* reallocateOutput(
 char const* tensorName, void* currentMemory, uint64_t size, uint64_t alignment) noexcept
    {
 return nullptr;
    }
 
 virtual void* reallocateOutputAsync(
 char const* tensorName, void* currentMemory, uint64_t size, uint64_t alignment, cudaStream_t /*stream*/)
    {
 return reallocateOutput(tensorName, currentMemory, size, alignment);
    }
 
 virtual void notifyShape(char const* tensorName, Dims const& dims) noexcept = 0;
};
} // namespace v_1_0
 
using IOutputAllocator = v_1_0::IOutputAllocator;
 
namespace v_1_0
{
class IDebugListener : public IVersionedInterface
{
public
 InterfaceInfo getInterfaceInfo() const noexcept override
    {
 return {"IDebugListener", 1, 0};
    }
 
 virtual bool processDebugTensor(void const* addr, TensorLocation location, DataType type, Dims const& shape,
 char const* name, cudaStream_t stream)
        = 0;
 
 ~IDebugListener() override = default;
};
} // namespace v_1_0 命名空间 v_1_0
 
using IDebugListener = v_1_0::IDebugListener;
 
class IExecutionContext : public INoCopy
{
public
 virtual ~IExecutionContext() noexcept = default;
 
 void setDebugSync(bool sync) noexcept
    {
 mImpl->setDebugSync(sync);
    }
 
 bool getDebugSync() const noexcept
    {
 return mImpl->getDebugSync();
    }
 
 void setProfiler(IProfiler* profiler) noexcept
    {
 mImpl->setProfiler(profiler);
    }
 
 IProfiler* getProfiler() const noexcept
    {
 return mImpl->getProfiler();
    }
 
 ICudaEngine const& getEngine() const noexcept
    {
 return mImpl->getEngine();
    }
 
 void setName(char const* name) noexcept
    {
 mImpl->setName(name);
    }
 
 char const* getName() const noexcept
    {
 return mImpl->getName();
    }
 
 void setDeviceMemory(void* memory) noexcept
    {
 mImpl->setDeviceMemory(memory);
    }
 
 void setDeviceMemoryV2(void* memory, int64_t size) noexcept
    {
 return mImpl->setDeviceMemoryV2(memory, size);
    }
 
 Dims getTensorStrides(char const* tensorName) const noexcept
    {
 return mImpl->getTensorStrides(tensorName);
    }
 
public
 int32_t getOptimizationProfile() const noexcept
    {
 return mImpl->getOptimizationProfile();
    }
 
 bool setInputShape(char const* tensorName, Dims const& dims) noexcept
    {
 return mImpl->setInputShape(tensorName, dims);
    }
 
 Dims getTensorShape(char const* tensorName) const noexcept
    {
 return mImpl->getTensorShape(tensorName);
    }
 
 bool allInputDimensionsSpecified() const noexcept
    {
 return mImpl->allInputDimensionsSpecified();
    }
 
 TRT_DEPRECATED bool allInputShapesSpecified() const noexcept
    {
 return mImpl->allInputShapesSpecified();
    }
 
 void setErrorRecorder(IErrorRecorder* recorder) noexcept
    {
 mImpl->setErrorRecorder(recorder);
    }
 
 IErrorRecorder* getErrorRecorder() const noexcept
    {
 return mImpl->getErrorRecorder();
    }
 
 bool executeV2(void* const* bindings) noexcept
    {
 return mImpl->executeV2(bindings);
    }
 
 bool setOptimizationProfileAsync(int32_t profileIndex, cudaStream_t stream) noexcept
    {
 return mImpl->setOptimizationProfileAsync(profileIndex, stream);
    }
 
 void setEnqueueEmitsProfile(bool enqueueEmitsProfile) noexcept
    {
 mImpl->setEnqueueEmitsProfile(enqueueEmitsProfile);
    }
 
 bool getEnqueueEmitsProfile() const noexcept
    {
 return mImpl->getEnqueueEmitsProfile();
    }
 
 bool reportToProfiler() const noexcept
    {
 return mImpl->reportToProfiler();
    }
 
 bool setTensorAddress(char const* tensorName, void* data) noexcept
    {
 return mImpl->setTensorAddress(tensorName, data);
    }
 
 void const* getTensorAddress(char const* tensorName) const noexcept
    {
 return mImpl->getTensorAddress(tensorName);
    }
 
 bool setOutputTensorAddress(char const* tensorName, void* data) noexcept
    {
 return mImpl->setOutputTensorAddress(tensorName, data);
    }
 
 bool setInputTensorAddress(char const* tensorName, void const* data) noexcept
    {
 return mImpl->setInputTensorAddress(tensorName, data);
    }
 
 void* getOutputTensorAddress(char const* tensorName) const noexcept
    {
 return mImpl->getOutputTensorAddress(tensorName);
    }
 
 int32_t inferShapes(int32_t nbMaxNames, char const** tensorNames) noexcept
    {
 return mImpl->inferShapes(nbMaxNames, tensorNames);
    }
 
 size_t updateDeviceMemorySizeForShapes() noexcept
    {
 return mImpl->updateDeviceMemorySizeForShapes();
    }
 
 bool setInputConsumedEvent(cudaEvent_t event) noexcept
    {
 return mImpl->setInputConsumedEvent(event);
    }
 
 cudaEvent_t getInputConsumedEvent() const noexcept
    {
 return mImpl->getInputConsumedEvent();
    }
 
 bool setOutputAllocator(char const* tensorName, IOutputAllocator* outputAllocator) noexcept
    {
 return mImpl->setOutputAllocator(tensorName, outputAllocator);
    }
 
 IOutputAllocator* getOutputAllocator(char const* tensorName) const noexcept
    {
 return mImpl->getOutputAllocator(tensorName);
    }
 
 int64_t getMaxOutputSize(char const* tensorName) const noexcept
    {
 return mImpl->getMaxOutputSize(tensorName);
    }
 
 bool setTemporaryStorageAllocator(IGpuAllocator* allocator) noexcept
    {
 return mImpl->setTemporaryStorageAllocator(allocator);
    }
 
 IGpuAllocator* getTemporaryStorageAllocator() const noexcept
    {
 return mImpl->getTemporaryStorageAllocator();
    }
 
 bool enqueueV3(cudaStream_t stream) noexcept
    {
 return mImpl->enqueueV3(stream);
    }
 
 void setPersistentCacheLimit(size_t size) noexcept
    {
 mImpl->setPersistentCacheLimit(size);
    }
 
 size_t getPersistentCacheLimit() const noexcept
    {
 return mImpl->getPersistentCacheLimit();
    }
 
 bool setNvtxVerbosity(ProfilingVerbosity verbosity) noexcept
    {
 return mImpl->setNvtxVerbosity(verbosity);
    }
 
 ProfilingVerbosity getNvtxVerbosity() const noexcept
    {
 return mImpl->getNvtxVerbosity();
    }
 
 void setAuxStreams(cudaStream_t* auxStreams, int32_t nbStreams) noexcept
    {
 mImpl->setAuxStreams(auxStreams, nbStreams);
    }
 
 bool setDebugListener(IDebugListener* listener) noexcept
    {
 return mImpl->setDebugListener(listener);
    }
 
 IDebugListener* getDebugListener() noexcept
    {
 return mImpl->getDebugListener();
    }
 
 bool setTensorDebugState(char const* name, bool flag) noexcept
    {
 return mImpl->setTensorDebugState(name, flag);
    }
 
 bool setAllTensorsDebugState(bool flag) noexcept
    {
 return mImpl->setAllTensorsDebugState(flag);
    }
 
 bool getDebugState(char const* name) const noexcept
    {
 return mImpl->getDebugState(name);
    }
 
protected
 apiv::VExecutionContext* mImpl;
}; // class IExecutionContext 类 IExecutionContext
 
enum class LayerInformationFormat : int32_t
{
 kONELINE = 0,
 kJSON = 1,
};
 
template <>
constexpr inline int32_t EnumMax<LayerInformationFormat>() noexcept
{
 return 2;
}
 
class IEngineInspector : public INoCopy
{
public
 virtual ~IEngineInspector() noexcept = default;
 
 bool setExecutionContext(IExecutionContext const* context) noexcept
    {
 return mImpl->setExecutionContext(context);
    }
 
 IExecutionContext const* getExecutionContext() const noexcept
    {
 return mImpl->getExecutionContext();
    }
 
 char const* getLayerInformation(int32_t layerIndex, LayerInformationFormat format) const noexcept
    {
 return mImpl->getLayerInformation(layerIndex, format);
    }
 
 char const* getEngineInformation(LayerInformationFormat format) const noexcept
    {
 return mImpl->getEngineInformation(format);
    }
 
 void setErrorRecorder(IErrorRecorder* recorder) noexcept
    {
 mImpl->setErrorRecorder(recorder);
    }
 
 IErrorRecorder* getErrorRecorder() const noexcept
    {
 return mImpl->getErrorRecorder();
    }
 
protected
 apiv::VEngineInspector* mImpl;
}; // class IEngineInspector 类 IEngineInspector
 
} // namespace nvinfer1 命名空间 nvinfer1
 
extern "C" TENSORRTAPI void* createInferRuntime_INTERNAL(void* logger, int32_t version) noexcept;
 
extern "C" TENSORRTAPI void* createInferRefitter_INTERNAL(void* engine, void* logger, int32_t version) noexcept;
 
extern "C" TENSORRTAPI nvinfer1::IPluginRegistry* getPluginRegistry() noexcept;
 
extern "C" TENSORRTAPI nvinfer1::ILogger* getLogger() noexcept;
 
namespace nvinfer1
{
namespace // unnamed namespace avoids linkage surprises when linking objects built with different versions of this  未命名的命名空间避免了在链接使用此头文件不同版本构建的对象时出现链接意外。
 // header.
{
inline IRuntime* createInferRuntime(ILogger& logger) noexcept
{
 return static_cast<IRuntime*>(createInferRuntime_INTERNAL(&logger, NV_TENSORRT_VERSION));
}
 
inline IRefitter* createInferRefitter(ICudaEngine& engine, ILogger& logger) noexcept
{
 return static_cast<IRefitter*>(createInferRefitter_INTERNAL(&engine, &logger, NV_TENSORRT_VERSION));
}
 
} // namespace 命名空间
 
template <typename T>
class PluginRegistrar
{
public
 PluginRegistrar()
    {
 getPluginRegistry()->registerCreator(instance, "");
    }
 
private
 T instance{};
};
 
} // namespace nvinfer1
 
#define REGISTER_TENSORRT_PLUGIN(name) \
 static nvinfer1::PluginRegistrar<name> pluginRegistrar##name {}
 
namespace nvinfer1
{
class ILoggerFinder
{
public
 virtual ILogger* findLogger() = 0;
 
protected
 virtual ~ILoggerFinder() = default;
};
 
namespace v_1_0
{
 
class IGpuAsyncAllocator : public IGpuAllocator
{
public
 IGpuAsyncAllocator() = default;
 ~IGpuAsyncAllocator() override = default;
 
 void* allocateAsync(uint64_t const size, uint64_t const alignment, AllocatorFlags const flags,
 cudaStream_t /*stream*/) noexcept override = 0;
 
 bool deallocateAsync(void* const memory, cudaStream_t /*stream*/) noexcept override = 0;
 
 TRT_DEPRECATED void* allocate(
 uint64_t const size, uint64_t const alignment, AllocatorFlags const flags) noexcept override
    {
 return allocateAsync(size, alignment, flags, nullptr);
    }
 
 TRT_DEPRECATED bool deallocate(void* const memory) noexcept override
    {
 return deallocateAsync(memory, nullptr);
    }
 
 InterfaceInfo getInterfaceInfo() const noexcept override
    {
 return {"IGpuAllocator", 1, 0};
    }
};
 
class IPluginCreatorV3One : public IPluginCreatorInterface
{
public
 InterfaceInfo getInterfaceInfo() const noexcept override
    {
 return InterfaceInfo{"PLUGIN CREATOR_V3ONE", 1, 0};
    }
 
 virtual IPluginV3* createPlugin(
 AsciiChar const* name, PluginFieldCollection const* fc, TensorRTPhase phase) noexcept = 0;
 
 virtual PluginFieldCollection const* getFieldNames() noexcept = 0;
 
 virtual AsciiChar const* getPluginName() const noexcept = 0;
 
 virtual AsciiChar const* getPluginVersion() const noexcept = 0;
 
 virtual AsciiChar const* getPluginNamespace() const noexcept = 0;
 
 IPluginCreatorV3One() = default;
 virtual ~IPluginCreatorV3One() = default;
 
protected
 IPluginCreatorV3One(IPluginCreatorV3One const&) = default;
 IPluginCreatorV3One(IPluginCreatorV3One&&) = default;
 IPluginCreatorV3One& operator=(IPluginCreatorV3One const&) & = default;
 IPluginCreatorV3One& operator=(IPluginCreatorV3One&&) & = default;
};
 
} // namespace v_1_0
 
using IGpuAsyncAllocator = v_1_0::IGpuAsyncAllocator;
 
using IPluginCreatorV3One = v_1_0::IPluginCreatorV3One;
 
} // namespace nvinfer1
 
extern "C" TENSORRTAPI int32_t getInferLibMajorVersion() noexcept;
extern "C" TENSORRTAPI int32_t getInferLibMinorVersion() noexcept;
extern "C" TENSORRTAPI int32_t getInferLibPatchVersion() noexcept;
extern "C" TENSORRTAPI int32_t getInferLibBuildVersion() noexcept;
 
#endif // NV_INFER_RUNTIME_H