TensorRT 10.14.1
NvInferImpl.h
Go to the documentation of this file.
1/*
2 * SPDX-FileCopyrightText: Copyright (c) 1993-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
3 * SPDX-License-Identifier: Apache-2.0
4 *
5 * Licensed under the Apache License, Version 2.0 (the "License");
6 * you may not use this file except in compliance with the License.
7 * You may obtain a copy of the License at
8 *
9 * http://www.apache.org/licenses/LICENSE-2.0
10 *
11 * Unless required by applicable law or agreed to in writing, software
12 * distributed under the License is distributed on an "AS IS" BASIS,
13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14 * See the License for the specific language governing permissions and
15 * limitations under the License.
16 */
17
18#ifndef NV_INFER_IMPL_H
19#define NV_INFER_IMPL_H
20
21#include "NvInferLegacyDims.h"
22#include "NvInferRuntimeCommon.h"
23
24// @cond SuppressDoxyWarnings
25
26namespace nvinfer1
27{
28
29class ILogger;
30
31namespace v_1_0
32{
33class IProgressMonitor;
34} // namespace v_1_0
35using IProgressMonitor = v_1_0::IProgressMonitor;
36
37namespace v_1_0
38{
39class IAlgorithmSelector;
40} // namespace v_1_0
41using IAlgorithmSelector = v_1_0::IAlgorithmSelector;
42
43namespace v_1_0
44{
45class IProfiler;
46} // namespace v_1_0
47using IProfiler = v_1_0::IProfiler;
48
49namespace v_1_0
50{
51class IOutputAllocator;
52} // namespace v_1_0
53using IOutputAllocator = v_1_0::IOutputAllocator;
54
55namespace v_1_0
56{
57class IDebugListener;
58} // namespace v_1_0
59using IDebugListener = v_1_0::IDebugListener;
60
61class IActivationLayer;
62class IAlgorithm;
63class IAlgorithmContext;
64class IAlgorithmIOInfo;
65class IAlgorithmVariant;
66class IAssertionLayer;
67class IAttention;
68class IBuilder;
69class IBuilderConfig;
70class IConcatenationLayer;
71class IConditionLayer;
72class IConstantLayer;
73class IConvolutionLayer;
74class ICudaEngine;
75class ICumulativeLayer;
76class IDeconvolutionLayer;
77class IDequantizeLayer;
78class IDimensionExpr;
79class IDynamicQuantizeLayer;
80class IEinsumLayer;
81class IElementWiseLayer;
82class IEngineInspector;
83class IExecutionContext;
84class IFillLayer;
85class IGatherLayer;
86class IGridSampleLayer;
87class IHostMemory;
88class IIdentityLayer;
89class ICastLayer;
90class IIfConditional;
91class IIfConditionalInputLayer;
92class IIfConditionalOutputLayer;
93class IInt8Calibrator;
94class IIteratorLayer;
95class ILayer;
96class ILoop;
97class ILoopOutputLayer;
98class ILRNLayer;
99class IMatrixMultiplyLayer;
100class INetworkDefinition;
101class INormalizationLayer;
102class INMSLayer;
103class INonZeroLayer;
104class IOneHotLayer;
105class IOptimizationProfile;
106class IPaddingLayer;
107class IParametricReLULayer;
108class IPlugin;
109class IPluginExt;
110class IPluginFactory;
111class IPluginLayer;
112class IPluginRegistry;
113class IPluginV2Layer;
114class IRuntimeConfig;
115
116namespace v_1_0
117{
118class IPluginV3;
119} // namespace v_1_0
120using IPluginV3 = v_1_0::IPluginV3;
121
122namespace v_1_0
123{
124class IStreamReader;
125class IStreamWriter;
126} // namespace v_1_0
127using IStreamReader = v_1_0::IStreamReader;
128using IStreamWriter = v_1_0::IStreamWriter;
129namespace v_1_0
130{
131class IStreamReaderV2;
132} // namespace v_1_0
133using IStreamReaderV2 = v_1_0::IStreamReaderV2;
134
135class IPluginV3Layer;
136class IPoolingLayer;
137class IQuantizeLayer;
138class IRaggedSoftMaxLayer;
139class IRecurrenceLayer;
140class IReduceLayer;
141class IRefitter;
142class IResizeLayer;
143class IReverseSequenceLayer;
144class IRuntime;
145class IScaleLayer;
146class IScatterLayer;
147class ISelectLayer;
148class ISerializationConfig;
149class IShapeLayer;
150class IShuffleLayer;
151class ISliceLayer;
152class ISoftMaxLayer;
153class ISqueezeLayer;
154class ITensor;
155
156namespace v_1_0
157{
158struct TimingCacheKey;
159struct TimingCacheValue;
160} // namespace v_1_0
161using TimingCacheKey = v_1_0::TimingCacheKey;
162using TimingCacheValue = v_1_0::TimingCacheValue;
163
164class ITimingCache;
165class ITopKLayer;
166class ITripLimitLayer;
167class IUnaryLayer;
168class IUnsqueezeLayer;
169struct Permutation;
170class Weights;
171
172enum class ActivationType : int32_t;
173enum class AttentionNormalizationOp : int32_t;
174enum class BoundingBoxFormat : int32_t;
175enum class BuilderFlag : int32_t;
176enum class CalibrationAlgoType : int32_t;
177enum class CumulativeOperation : int32_t;
178enum class DeviceType : int32_t;
179enum class DimensionOperation : int32_t;
180enum class ElementWiseOperation : int32_t;
181enum class EngineCapability : int32_t;
182enum class FillOperation : int32_t;
183enum class GatherMode : int32_t;
184enum class LayerInformationFormat : int32_t;
185enum class LayerType : int32_t;
186enum class LoopOutput : int32_t;
187enum class MatrixOperation : int32_t;
188enum class MemoryPoolType : int32_t;
189enum class NetworkDefinitionCreationFlag : int32_t;
190enum class OptProfileSelector : int32_t;
191enum class PaddingMode : int32_t;
192enum class PoolingType : int32_t;
193enum class ProfilingVerbosity : int32_t;
194enum class QuantizationFlag : int32_t;
195enum class ReduceOperation : int32_t;
196enum class ResizeCoordinateTransformation : int32_t;
197enum class InterpolationMode : int32_t;
198enum class ResizeRoundMode : int32_t;
199enum class ResizeSelector : int32_t;
200enum class ScaleMode : int32_t;
201enum class ScatterMode : int32_t;
202enum class SampleMode : int32_t;
203enum class SerializationFlag : int32_t;
204enum class TensorIOMode : int32_t;
205enum class TensorLocation : int32_t;
206enum class TopKOperation : int32_t;
207enum class TripLimit : int32_t;
208enum class UnaryOperation : int32_t;
209enum class WeightsRole : int32_t;
210enum class PreviewFeature : int32_t;
211enum class HardwareCompatibilityLevel : int32_t;
212enum class ExecutionContextAllocationStrategy : int32_t;
213enum class RuntimePlatform : int32_t;
214enum class TilingOptimizationLevel : int32_t;
215enum class EngineStat : int32_t;
216
217
218using TacticSources = uint32_t;
219using TensorFormats = uint32_t;
220using BuilderFlags = uint32_t;
221using NetworkDefinitionCreationFlags = uint32_t;
222using QuantizationFlags = uint32_t;
223using TempfileControlFlags = uint32_t;
224using SerializationFlags = uint32_t;
225
233
234namespace apiv
235{
236
237class VRoot
238{
239public:
240 virtual ~VRoot() noexcept = default;
241};
242
243class VHostMemory : public VRoot
244{
245public:
246 virtual void* data() const noexcept = 0;
247 virtual std::size_t size() const noexcept = 0;
248 virtual DataType type() const noexcept = 0;
249};
250
251class VDimensionExpr : public VRoot
252{
253public:
254 virtual bool isConstant() const = 0;
255 virtual int64_t getConstantValue() const = 0;
256 virtual bool isSizeTensor() const = 0;
257};
258
259class VExprBuilder : public VRoot
260{
261public:
262 virtual IDimensionExpr const* constant(int64_t value) = 0;
263 virtual IDimensionExpr const* operation(
264 DimensionOperation op, IDimensionExpr const& first, IDimensionExpr const& second)
265 = 0;
266 virtual IDimensionExpr const* declareSizeTensor(
267 int32_t outputIndex, IDimensionExpr const& opt, IDimensionExpr const& upper)
268 = 0;
269};
270
271class VRuntime : public VRoot
272{
273public:
274 virtual IRuntime* getPImpl() noexcept = 0;
275 virtual nvinfer1::ICudaEngine* deserializeCudaEngine(void const* blob, std::size_t size) noexcept = 0;
276 virtual nvinfer1::ICudaEngine* deserializeCudaEngine(IStreamReader& streamReader) noexcept = 0;
277 virtual void setDLACore(int32_t dlaCore) noexcept = 0;
278 virtual int32_t getDLACore() const noexcept = 0;
279 virtual int32_t getNbDLACores() const noexcept = 0;
280 virtual void setGpuAllocator(IGpuAllocator* allocator) noexcept = 0;
281 virtual void setErrorRecorder(IErrorRecorder* recorder) noexcept = 0;
282 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
283 virtual ILogger* getLogger() const noexcept = 0;
284 virtual bool setMaxThreads(int32_t maxThreads) noexcept = 0;
285 virtual int32_t getMaxThreads() const noexcept = 0;
286 virtual void setTemporaryDirectory(char const*) noexcept = 0;
287 virtual char const* getTemporaryDirectory() const noexcept = 0;
288 virtual void setTempfileControlFlags(TempfileControlFlags) noexcept = 0;
289 virtual TempfileControlFlags getTempfileControlFlags() const noexcept = 0;
290 virtual IPluginRegistry& getPluginRegistry() noexcept = 0;
291 virtual void setPluginRegistryParent(IPluginRegistry* parent) noexcept = 0;
292 virtual IRuntime* loadRuntime(char const* path) noexcept = 0;
293 virtual void setEngineHostCodeAllowed(bool allowed) noexcept = 0;
294 virtual bool getEngineHostCodeAllowed() const noexcept = 0;
295 // Added in TensorRT version 10.7
296 virtual nvinfer1::ICudaEngine* deserializeCudaEngineV2(IStreamReaderV2& streamReader) noexcept = 0;
297};
298
299class VRefitter : public VRoot
300{
301public:
302 virtual IRefitter* getPImpl() noexcept = 0;
303 virtual bool setWeights(char const* layerName, WeightsRole role, const Weights weights) noexcept = 0;
304 virtual bool refitCudaEngine() noexcept = 0;
305 virtual int32_t getMissing(int32_t size, char const** layerNames, WeightsRole* roles) noexcept = 0;
306 virtual int32_t getAll(int32_t size, char const** layerNames, WeightsRole* roles) noexcept = 0;
307 virtual bool setDynamicRange(char const* tensorName, float min, float max) noexcept = 0;
308 virtual float getDynamicRangeMin(char const* tensorName) const noexcept = 0;
309 virtual float getDynamicRangeMax(char const* tensorName) const noexcept = 0;
310 virtual int32_t getTensorsWithDynamicRange(int32_t size, char const** tensorNames) const noexcept = 0;
311 virtual void setErrorRecorder(IErrorRecorder* recorder) noexcept = 0;
312 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
313 virtual bool setNamedWeights(char const* name, Weights weights) noexcept = 0;
314 virtual int32_t getMissingWeights(int32_t size, char const** weightsNames) noexcept = 0;
315 virtual int32_t getAllWeights(int32_t size, char const** weightsNames) noexcept = 0;
316 virtual ILogger* getLogger() const noexcept = 0;
317 virtual bool setMaxThreads(int32_t maxThreads) noexcept = 0;
318 virtual int32_t getMaxThreads() const noexcept = 0;
319 virtual bool setNamedWeightsWithLocation(char const* name, Weights weights, TensorLocation location) noexcept = 0;
320 virtual Weights getNamedWeights(char const* weightsName) const noexcept = 0;
321 virtual TensorLocation getWeightsLocation(char const* weightsName) const noexcept = 0;
322 virtual bool unsetNamedWeights(char const* weightsName) noexcept = 0;
323 virtual void setWeightsValidation(bool weightsValidation) noexcept = 0;
324 virtual bool getWeightsValidation() const noexcept = 0;
325 virtual bool refitCudaEngineAsync(cudaStream_t stream) noexcept = 0;
326 virtual Weights getWeightsPrototype(char const* weightsName) const noexcept = 0;
327};
328
329class VOptimizationProfile : public VRoot
330{
331public:
332 virtual bool setDimensions(char const* inputName, OptProfileSelector select, Dims const& dims) noexcept = 0;
333 virtual Dims getDimensions(char const* inputName, OptProfileSelector select) const noexcept = 0;
334 virtual bool setShapeValues(
335 char const* inputName, OptProfileSelector select, int32_t const* values, int32_t nbValues) noexcept = 0;
336 virtual int32_t getNbShapeValues(char const* inputName) const noexcept = 0;
337 virtual int32_t const* getShapeValues(char const* inputName, OptProfileSelector select) const noexcept = 0;
338 virtual bool setExtraMemoryTarget(float target) noexcept = 0;
339 virtual float getExtraMemoryTarget() const noexcept = 0;
340 virtual bool isValid() const noexcept = 0;
341 // Added in TensorRT 10.11
342 TRT_NODISCARD virtual bool setShapeValuesV2(
343 char const* inputName, OptProfileSelector select, int64_t const* values, int32_t nbValues) noexcept = 0;
344 TRT_NODISCARD virtual int64_t const* getShapeValuesV2(
345 char const* inputName, OptProfileSelector select) const noexcept = 0;
346};
347
348class VCudaEngine : public VRoot
349{
350public:
351 virtual ICudaEngine* getPImpl() noexcept = 0;
352 virtual int32_t getNbLayers() const noexcept = 0;
353 virtual IHostMemory* serialize() const noexcept = 0;
354 virtual IExecutionContext* createExecutionContext(ExecutionContextAllocationStrategy strategy) noexcept = 0;
355 virtual IExecutionContext* createExecutionContextWithoutDeviceMemory() noexcept = 0;
356 virtual size_t getDeviceMemorySize() const noexcept = 0;
357 virtual bool isRefittable() const noexcept = 0;
358 virtual char const* getName() const noexcept = 0;
359 virtual int32_t getNbOptimizationProfiles() const noexcept = 0;
360 virtual int32_t const* getProfileTensorValues(
361 char const* tensorName, int32_t profileIndex, OptProfileSelector select) const noexcept = 0;
362 virtual EngineCapability getEngineCapability() const noexcept = 0;
363 virtual void setErrorRecorder(IErrorRecorder* recorder) noexcept = 0;
364 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
365 virtual bool hasImplicitBatchDimension() const noexcept = 0;
366 virtual TacticSources getTacticSources() const noexcept = 0;
367 virtual ProfilingVerbosity getProfilingVerbosity() const noexcept = 0;
368 virtual IEngineInspector* createEngineInspector() const noexcept = 0;
369 virtual Dims getTensorShape(char const* tensorName) const noexcept = 0;
370 virtual DataType getTensorDataType(char const* tensorName) const noexcept = 0;
371 virtual TensorLocation getTensorLocation(char const* tensorName) const noexcept = 0;
372 virtual bool isShapeInferenceIO(char const* tensorName) const noexcept = 0;
373 virtual TensorIOMode getTensorIOMode(char const* tensorName) const noexcept = 0;
374 virtual int32_t getTensorBytesPerComponent(char const* tensorName) const noexcept = 0;
375 virtual int32_t getTensorComponentsPerElement(char const* tensorName) const noexcept = 0;
376 virtual TensorFormat getTensorFormat(char const* tensorName) const noexcept = 0;
377 virtual char const* getTensorFormatDesc(char const* tensorName) const noexcept = 0;
378 virtual int32_t getTensorVectorizedDim(char const* tensorName) const noexcept = 0;
379 virtual Dims getProfileShape(
380 char const* tensorName, int32_t profileIndex, OptProfileSelector select) const noexcept = 0;
381 virtual int32_t getNbIOTensors() const noexcept = 0;
382 virtual char const* getIOTensorName(int32_t index) const noexcept = 0;
383 virtual HardwareCompatibilityLevel getHardwareCompatibilityLevel() const noexcept = 0;
384 virtual int32_t getNbAuxStreams() const noexcept = 0;
385
386 virtual int32_t getTensorBytesPerComponentV2(char const* tensorName, int32_t profileIndex) const noexcept = 0;
387 virtual int32_t getTensorComponentsPerElementV2(char const* tensorName, int32_t profileIndex) const noexcept = 0;
388 virtual TensorFormat getTensorFormatV2(char const* tensorName, int32_t profileIndex) const noexcept = 0;
389 virtual char const* getTensorFormatDescV2(char const* tensorName, int32_t profileIndex) const noexcept = 0;
390 virtual int32_t getTensorVectorizedDimV2(char const* tensorName, int32_t profileIndex) const noexcept = 0;
391
392 virtual ISerializationConfig* createSerializationConfig() noexcept = 0;
393 virtual IHostMemory* serializeWithConfig(ISerializationConfig& config) const noexcept = 0;
394
395 virtual size_t getDeviceMemorySizeForProfile(int32_t profileIndex) const noexcept = 0;
396 virtual IRefitter* createRefitter(ILogger& logger) noexcept = 0;
397
398 virtual bool setWeightStreamingBudget(int64_t gpuMemoryBudget) noexcept = 0;
399 virtual int64_t getWeightStreamingBudget() const noexcept = 0;
400 virtual int64_t getMinimumWeightStreamingBudget() const noexcept = 0;
401 virtual int64_t getStreamableWeightsSize() const noexcept = 0;
402
403 virtual bool isDebugTensor(char const* name) const noexcept = 0;
404
405 // Added in TensorRT 10.1
406 virtual bool setWeightStreamingBudgetV2(int64_t gpuMemoryBudget) noexcept = 0;
407 virtual int64_t getWeightStreamingBudgetV2() const noexcept = 0;
408 virtual int64_t getWeightStreamingAutomaticBudget() const noexcept = 0;
409 virtual int64_t getWeightStreamingScratchMemorySize() const noexcept = 0;
410 virtual int64_t getDeviceMemorySizeV2() const noexcept = 0;
411 virtual int64_t getDeviceMemorySizeForProfileV2(int32_t profileIndex) const noexcept = 0;
412 // Added in TensorRT 10.11
413 TRT_NODISCARD virtual int64_t const* getProfileTensorValuesV2(
414 char const* tensorName, int32_t profileIndex, OptProfileSelector select) const noexcept = 0;
415 TRT_NODISCARD virtual IExecutionContext* createExecutionContextWithRuntimeConfig(
416 IRuntimeConfig* runtimeConfig) noexcept = 0;
417 TRT_NODISCARD virtual IRuntimeConfig* createRuntimeConfig() noexcept = 0;
418 TRT_NODISCARD virtual int64_t getEngineStat(EngineStat stat) const noexcept = 0;
419};
420
421class VExecutionContext : public VRoot
422{
423public:
424 virtual IExecutionContext* getPImpl() noexcept = 0;
425 virtual void setDebugSync(bool sync) noexcept = 0;
426 virtual bool getDebugSync() const noexcept = 0;
427 virtual void setProfiler(IProfiler*) noexcept = 0;
428 virtual IProfiler* getProfiler() const noexcept = 0;
429 virtual ICudaEngine const& getEngine() const noexcept = 0;
430 virtual void setName(char const* name) noexcept = 0;
431 virtual char const* getName() const noexcept = 0;
432 virtual void setDeviceMemory(void* memory) noexcept = 0;
433 virtual int32_t getOptimizationProfile() const noexcept = 0;
434 virtual bool allInputDimensionsSpecified() const noexcept = 0;
435 virtual bool allInputShapesSpecified() const noexcept = 0;
436 virtual void setErrorRecorder(IErrorRecorder* recorder) noexcept = 0;
437 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
438 virtual bool executeV2(void* const* bindings) noexcept = 0;
439 virtual bool setOptimizationProfileAsync(int32_t profileIndex, cudaStream_t stream) noexcept = 0;
440 virtual void setEnqueueEmitsProfile(bool enqueueEmitsProfile) noexcept = 0;
441 virtual bool getEnqueueEmitsProfile() const noexcept = 0;
442 virtual bool reportToProfiler() const noexcept = 0;
443 virtual bool setInputShape(char const* tensorName, Dims const& dims) noexcept = 0;
444 virtual Dims getTensorShape(char const* tensorName) const noexcept = 0;
445 virtual Dims getTensorStrides(char const* tensorName) const noexcept = 0;
446 virtual bool setTensorAddress(char const* tensorName, void* data) noexcept = 0;
447 virtual void const* getTensorAddress(char const* tensorName) const noexcept = 0;
448 virtual bool setInputTensorAddress(char const* tensorName, void const* data) noexcept = 0;
449 virtual bool setOutputTensorAddress(char const* tensorName, void* data) noexcept = 0;
450 virtual int32_t inferShapes(int32_t nbMaxNames, char const** tensorNames) noexcept = 0;
451 virtual bool setInputConsumedEvent(cudaEvent_t event) noexcept = 0;
452 virtual cudaEvent_t getInputConsumedEvent() const noexcept = 0;
453 virtual void* getOutputTensorAddress(char const* tensorName) const noexcept = 0;
454 virtual bool setOutputAllocator(char const* tensorName, IOutputAllocator* outputAllocator) noexcept = 0;
455 virtual IOutputAllocator* getOutputAllocator(char const* name) noexcept = 0;
456 virtual int64_t getMaxOutputSize(char const* tensorName) const noexcept = 0;
457 virtual bool setTemporaryStorageAllocator(IGpuAllocator* allocator) noexcept = 0;
458 virtual IGpuAllocator* getTemporaryStorageAllocator() const noexcept = 0;
459 virtual bool enqueueV3(cudaStream_t stream) noexcept = 0;
460 virtual void setPersistentCacheLimit(size_t size) noexcept = 0;
461 virtual size_t getPersistentCacheLimit() const noexcept = 0;
462 virtual bool setNvtxVerbosity(ProfilingVerbosity verbosity) noexcept = 0;
463 virtual ProfilingVerbosity getNvtxVerbosity() const noexcept = 0;
464 virtual void setAuxStreams(cudaStream_t* auxStreams, int32_t nbStreams) noexcept = 0;
465 virtual bool setDebugListener(IDebugListener* listener) noexcept = 0;
466 virtual IDebugListener* getDebugListener() noexcept = 0;
467 virtual bool setTensorDebugState(char const* name, bool flag) noexcept = 0;
468 virtual bool getDebugState(char const* name) const noexcept = 0;
469 virtual bool setAllTensorsDebugState(bool flag) noexcept = 0;
470 virtual size_t updateDeviceMemorySizeForShapes() noexcept = 0;
471 virtual void setDeviceMemoryV2(void* memory, int64_t size) noexcept = 0;
472 TRT_NODISCARD virtual IRuntimeConfig* getRuntimeConfig() const noexcept = 0;
473 virtual bool setUnfusedTensorsDebugState(bool flag) noexcept = 0;
474 virtual bool getUnfusedTensorsDebugState() const noexcept = 0;
475#if ENABLE_FEATURE_DISABLE_RUNTIME_ALLOCATION
476 virtual bool isStreamCapturable(cudaStream_t stream) const noexcept = 0;
477#endif // ENABLE_FEATURE_DISABLE_RUNTIME_ALLOCATION
478};
479
480class VEngineInspector : public VRoot
481{
482public:
483 virtual IEngineInspector* getPImpl() noexcept = 0;
484 virtual bool setExecutionContext(IExecutionContext const* context) noexcept = 0;
485 virtual IExecutionContext const* getExecutionContext() const noexcept = 0;
486 virtual char const* getLayerInformation(int32_t layerIndex, LayerInformationFormat format) const noexcept = 0;
487 virtual char const* getEngineInformation(LayerInformationFormat format) const noexcept = 0;
488 virtual void setErrorRecorder(IErrorRecorder* recorder) noexcept = 0;
489 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
490};
491
492class VTensor : public VRoot
493{
494public:
495 virtual void setName(char const* name) noexcept = 0;
496 virtual char const* getName() const noexcept = 0;
497 virtual void setDimensions(Dims const& dimensions) noexcept = 0;
498 virtual Dims getDimensions() const noexcept = 0;
499 virtual void setType(DataType type) noexcept = 0;
500 virtual DataType getType() const noexcept = 0;
501 virtual bool setDynamicRange(float min, float max) noexcept = 0;
502 virtual bool isNetworkInput() const noexcept = 0;
503 virtual bool isNetworkOutput() const noexcept = 0;
504 virtual void setBroadcastAcrossBatch(bool broadcastAcrossBatch) noexcept = 0;
505 virtual bool getBroadcastAcrossBatch() const noexcept = 0;
506 virtual TensorLocation getLocation() const noexcept = 0;
507 virtual void setLocation(TensorLocation location) noexcept = 0;
508 virtual bool dynamicRangeIsSet() const noexcept = 0;
509 virtual void resetDynamicRange() noexcept = 0;
510 virtual float getDynamicRangeMin() const noexcept = 0;
511 virtual float getDynamicRangeMax() const noexcept = 0;
512 virtual void setAllowedFormats(TensorFormats formats) noexcept = 0;
513 virtual TensorFormats getAllowedFormats() const noexcept = 0;
514 virtual bool isShapeTensor() const noexcept = 0;
515 virtual bool isExecutionTensor() const noexcept = 0;
516 virtual void setDimensionName(int32_t index, char const* name) noexcept = 0;
517 virtual char const* getDimensionName(int32_t index) const noexcept = 0;
518};
519
520class VLayer : public VRoot
521{
522public:
523 virtual LayerType getType() const noexcept = 0;
524 virtual void setName(char const* name) noexcept = 0;
525 virtual char const* getName() const noexcept = 0;
526 virtual int32_t getNbInputs() const noexcept = 0;
527 virtual ITensor* getInput(int32_t index) const noexcept = 0;
528 virtual int32_t getNbOutputs() const noexcept = 0;
529 virtual ITensor* getOutput(int32_t index) const noexcept = 0;
530 virtual void setInput(int32_t index, ITensor& tensor) noexcept = 0;
531 virtual void setPrecision(DataType dataType) noexcept = 0;
532 virtual DataType getPrecision() const noexcept = 0;
533 virtual bool precisionIsSet() const noexcept = 0;
534 virtual void resetPrecision() noexcept = 0;
535 virtual void setOutputType(int32_t index, DataType dataType) noexcept = 0;
536 virtual DataType getOutputType(int32_t index) const noexcept = 0;
537 virtual bool outputTypeIsSet(int32_t index) const noexcept = 0;
538 virtual void resetOutputType(int32_t index) noexcept = 0;
539 virtual void setMetadata(char const* docString) noexcept = 0;
540 virtual char const* getMetadata() const noexcept = 0;
541};
542
543class VConvolutionLayer : public VRoot
544{
545public:
546 virtual void setNbOutputMaps(int64_t nbOutputMaps) noexcept = 0;
547 virtual int64_t getNbOutputMaps() const noexcept = 0;
548 virtual void setNbGroups(int64_t nbGroups) noexcept = 0;
549 virtual int64_t getNbGroups() const noexcept = 0;
550 virtual void setKernelWeights(Weights weights) noexcept = 0;
551 virtual Weights getKernelWeights() const noexcept = 0;
552 virtual void setBiasWeights(Weights weights) noexcept = 0;
553 virtual Weights getBiasWeights() const noexcept = 0;
554 virtual void setPrePadding(Dims const& padding) noexcept = 0;
555 virtual Dims getPrePadding() const noexcept = 0;
556 virtual void setPostPadding(Dims const& padding) noexcept = 0;
557 virtual Dims getPostPadding() const noexcept = 0;
558 virtual void setPaddingMode(PaddingMode paddingMode) noexcept = 0;
559 virtual PaddingMode getPaddingMode() const noexcept = 0;
560 virtual void setKernelSizeNd(Dims const& kernelSize) noexcept = 0;
561 virtual Dims getKernelSizeNd() const noexcept = 0;
562 virtual void setStrideNd(Dims const& stride) noexcept = 0;
563 virtual Dims getStrideNd() const noexcept = 0;
564 virtual void setPaddingNd(Dims const& padding) noexcept = 0;
565 virtual Dims getPaddingNd() const noexcept = 0;
566 virtual void setDilationNd(Dims const& dilation) noexcept = 0;
567 virtual Dims getDilationNd() const noexcept = 0;
568};
569
570class VActivationLayer : public VRoot
571{
572public:
573 virtual void setActivationType(ActivationType type) noexcept = 0;
574 virtual ActivationType getActivationType() const noexcept = 0;
575 virtual void setAlpha(float alpha) noexcept = 0;
576 virtual void setBeta(float beta) noexcept = 0;
577 virtual float getAlpha() const noexcept = 0;
578 virtual float getBeta() const noexcept = 0;
579};
580
581class VPoolingLayer : public VRoot
582{
583public:
584 virtual void setPoolingType(PoolingType type) noexcept = 0;
585 virtual PoolingType getPoolingType() const noexcept = 0;
586 virtual void setBlendFactor(float blendFactor) noexcept = 0;
587 virtual float getBlendFactor() const noexcept = 0;
588 virtual void setAverageCountExcludesPadding(bool exclusive) noexcept = 0;
589 virtual bool getAverageCountExcludesPadding() const noexcept = 0;
590 virtual void setPrePadding(Dims const& padding) noexcept = 0;
591 virtual Dims getPrePadding() const noexcept = 0;
592 virtual void setPostPadding(Dims const& padding) noexcept = 0;
593 virtual Dims getPostPadding() const noexcept = 0;
594 virtual void setPaddingMode(PaddingMode paddingMode) noexcept = 0;
595 virtual PaddingMode getPaddingMode() const noexcept = 0;
596 virtual void setWindowSizeNd(Dims const& windowSize) noexcept = 0;
597 virtual Dims getWindowSizeNd() const noexcept = 0;
598 virtual void setStrideNd(Dims const& stride) noexcept = 0;
599 virtual Dims getStrideNd() const noexcept = 0;
600 virtual void setPaddingNd(Dims const& padding) noexcept = 0;
601 virtual Dims getPaddingNd() const noexcept = 0;
602};
603
604class VLRNLayer : public VRoot
605{
606public:
607 virtual void setWindowSize(int64_t windowSize) noexcept = 0;
608 virtual int64_t getWindowSize() const noexcept = 0;
609 virtual void setAlpha(float alpha) noexcept = 0;
610 virtual float getAlpha() const noexcept = 0;
611 virtual void setBeta(float beta) noexcept = 0;
612 virtual float getBeta() const noexcept = 0;
613 virtual void setK(float k) noexcept = 0;
614 virtual float getK() const noexcept = 0;
615};
616
617class VScaleLayer : public VRoot
618{
619public:
620 virtual void setMode(ScaleMode mode) noexcept = 0;
621 virtual ScaleMode getMode() const noexcept = 0;
622 virtual void setShift(Weights shift) noexcept = 0;
623 virtual Weights getShift() const noexcept = 0;
624 virtual void setScale(Weights scale) noexcept = 0;
625 virtual Weights getScale() const noexcept = 0;
626 virtual void setPower(Weights power) noexcept = 0;
627 virtual Weights getPower() const noexcept = 0;
628 virtual int32_t getChannelAxis() const noexcept = 0;
629 virtual void setChannelAxis(int32_t channelAxis) noexcept = 0;
630};
631
632class VSoftMaxLayer : public VRoot
633{
634public:
635 virtual void setAxes(uint32_t axes) noexcept = 0;
636 virtual uint32_t getAxes() const noexcept = 0;
637};
638
639class VConcatenationLayer : public VRoot
640{
641public:
642 virtual void setAxis(int32_t axis) noexcept = 0;
643 virtual int32_t getAxis() const noexcept = 0;
644};
645
646class VDeconvolutionLayer : public VRoot
647{
648public:
649 virtual void setNbOutputMaps(int64_t nbOutputMaps) noexcept = 0;
650 virtual int64_t getNbOutputMaps() const noexcept = 0;
651 virtual void setNbGroups(int64_t nbGroups) noexcept = 0;
652 virtual int64_t getNbGroups() const noexcept = 0;
653 virtual void setKernelWeights(Weights weights) noexcept = 0;
654 virtual Weights getKernelWeights() const noexcept = 0;
655 virtual void setBiasWeights(Weights weights) noexcept = 0;
656 virtual Weights getBiasWeights() const noexcept = 0;
657 virtual void setPrePadding(Dims const& padding) noexcept = 0;
658 virtual Dims getPrePadding() const noexcept = 0;
659 virtual void setPostPadding(Dims const& padding) noexcept = 0;
660 virtual Dims getPostPadding() const noexcept = 0;
661 virtual void setPaddingMode(PaddingMode paddingMode) noexcept = 0;
662 virtual PaddingMode getPaddingMode() const noexcept = 0;
663 virtual void setKernelSizeNd(Dims const& kernelSize) noexcept = 0;
664 virtual Dims getKernelSizeNd() const noexcept = 0;
665 virtual void setStrideNd(Dims const& stride) noexcept = 0;
666 virtual Dims getStrideNd() const noexcept = 0;
667 virtual void setPaddingNd(Dims const& padding) noexcept = 0;
668 virtual Dims getPaddingNd() const noexcept = 0;
669 virtual void setDilationNd(Dims const& dilation) noexcept = 0;
670 virtual Dims getDilationNd() const noexcept = 0;
671};
672
673class VElementWiseLayer : public VRoot
674{
675public:
676 virtual void setOperation(ElementWiseOperation op) noexcept = 0;
677 virtual ElementWiseOperation getOperation() const noexcept = 0;
678};
679
680class VGatherLayer : public VRoot
681{
682public:
683 virtual void setGatherAxis(int32_t axis) noexcept = 0;
684 virtual int32_t getGatherAxis() const noexcept = 0;
685 virtual void setNbElementWiseDims(int32_t k) noexcept = 0;
686 virtual int32_t getNbElementWiseDims() const noexcept = 0;
687 virtual void setMode(GatherMode mode) noexcept = 0;
688 virtual GatherMode getMode() const noexcept = 0;
689};
690
691class VPluginLayer : public VRoot
692{
693public:
694 virtual IPlugin& getPlugin() noexcept = 0;
695};
696
697class VPluginV2Layer : public VRoot
698{
699public:
700 virtual IPluginV2& getPlugin() noexcept = 0;
701};
702
703class VPluginV3Layer : public VRoot
704{
705public:
706 virtual IPluginV3& getPlugin() noexcept = 0;
707};
708
709class VUnaryLayer : public VRoot
710{
711public:
712 virtual void setOperation(UnaryOperation op) noexcept = 0;
713 virtual UnaryOperation getOperation() const noexcept = 0;
714};
715
716class VReduceLayer : public VRoot
717{
718public:
719 virtual void setOperation(ReduceOperation op) noexcept = 0;
720 virtual ReduceOperation getOperation() const noexcept = 0;
721 virtual void setReduceAxes(uint32_t reduceAxes) noexcept = 0;
722 virtual uint32_t getReduceAxes() const noexcept = 0;
723 virtual void setKeepDimensions(bool keepDimensions) noexcept = 0;
724 virtual bool getKeepDimensions() const noexcept = 0;
725};
726
727class VPaddingLayer : public VRoot
728{
729public:
730 virtual void setPrePaddingNd(Dims const& padding) noexcept = 0;
731 virtual Dims getPrePaddingNd() const noexcept = 0;
732 virtual void setPostPaddingNd(Dims const& padding) noexcept = 0;
733 virtual Dims getPostPaddingNd() const noexcept = 0;
734};
735
736class VShuffleLayer : public VRoot
737{
738public:
739 virtual void setFirstTranspose(Permutation const& permutation) noexcept = 0;
740 virtual Permutation const& getFirstTranspose() const noexcept = 0;
741 virtual void setReshapeDimensions(Dims const& dimensions) noexcept = 0;
742 virtual Dims getReshapeDimensions() const noexcept = 0;
743 virtual void setSecondTranspose(Permutation const& permutation) noexcept = 0;
744 virtual Permutation const& getSecondTranspose() const noexcept = 0;
745 virtual void setZeroIsPlaceholder(bool zeroIsPlaceholder) noexcept = 0;
746 virtual bool getZeroIsPlaceholder() const noexcept = 0;
747};
748
749class VSliceLayer : public VRoot
750{
751public:
752 virtual void setStart(Dims const& start) noexcept = 0;
753 virtual Dims getStart() const noexcept = 0;
754 virtual void setSize(Dims const& size) noexcept = 0;
755 virtual Dims getSize() const noexcept = 0;
756 virtual void setStride(Dims const& stride) noexcept = 0;
757 virtual Dims getStride() const noexcept = 0;
758 virtual void setMode(SampleMode mode) noexcept = 0;
759 virtual SampleMode getMode() const noexcept = 0;
760 virtual void setAxes(Dims const& axes) noexcept = 0;
761 virtual Dims getAxes() const noexcept = 0;
762};
763
764class VShapeLayer : public VRoot
765{
766public:
767};
768
769class VTopKLayer : public VRoot
770{
771public:
772 virtual void setOperation(TopKOperation op) noexcept = 0;
773 virtual TopKOperation getOperation() const noexcept = 0;
774 virtual void setK(int32_t k) noexcept = 0;
775 virtual int32_t getK() const noexcept = 0;
776 virtual void setReduceAxes(uint32_t reduceAxes) noexcept = 0;
777 virtual uint32_t getReduceAxes() const noexcept = 0;
778 virtual bool setIndicesType(DataType type) noexcept = 0;
779 virtual DataType getIndicesType() const noexcept = 0;
780};
781
782class VMatrixMultiplyLayer : public VRoot
783{
784public:
785 virtual void setOperation(int32_t index, MatrixOperation op) noexcept = 0;
786 virtual MatrixOperation getOperation(int32_t index) const noexcept = 0;
787};
788
789class VNonZeroLayer : public VRoot
790{
791public:
792 virtual bool setIndicesType(DataType type) noexcept = 0;
793 virtual DataType getIndicesType() const noexcept = 0;
794};
795
796class VRaggedSoftMaxLayer : public VRoot
797{
798public:
799};
800
801class VIdentityLayer : public VRoot
802{
803public:
804};
805
806class VCastLayer : public VRoot
807{
808public:
809 virtual void setToType(DataType toType) noexcept = 0;
810 virtual DataType getToType() const noexcept = 0;
811};
812
813class VConstantLayer : public VRoot
814{
815public:
816 virtual void setWeights(Weights weights) noexcept = 0;
817 virtual Weights getWeights() const noexcept = 0;
818 virtual void setDimensions(Dims const& dimensions) noexcept = 0;
819 virtual Dims getDimensions() const noexcept = 0;
820};
821
822class VParametricReLULayer : public VRoot
823{
824public:
825};
826
827class VResizeLayer : public VRoot
828{
829public:
830 virtual void setOutputDimensions(Dims const& dimensions) noexcept = 0;
831 virtual Dims getOutputDimensions() const noexcept = 0;
832 virtual void setScales(float const* scales, int32_t nbScales) noexcept = 0;
833 virtual int32_t getScales(int32_t size, float* scales) const noexcept = 0;
834 virtual void setResizeMode(InterpolationMode interpolationMode) noexcept = 0;
835 virtual InterpolationMode getResizeMode() const noexcept = 0;
836 virtual void setCoordinateTransformation(ResizeCoordinateTransformation coordTransform) noexcept = 0;
837 virtual ResizeCoordinateTransformation getCoordinateTransformation() const noexcept = 0;
838 virtual void setSelectorForSinglePixel(ResizeSelector selector) noexcept = 0;
839 virtual ResizeSelector getSelectorForSinglePixel() const noexcept = 0;
840 virtual void setNearestRounding(ResizeRoundMode value) noexcept = 0;
841 virtual ResizeRoundMode getNearestRounding() const noexcept = 0;
842 virtual void setCubicCoeff(float value) noexcept = 0;
843 virtual float getCubicCoeff() const noexcept = 0;
844 virtual void setExcludeOutside(bool value) noexcept = 0;
845 virtual bool getExcludeOutside() const noexcept = 0;
846};
847
848class VLoopBoundaryLayer : public VRoot
849{
850public:
851 virtual ILoop* getLoop() const noexcept = 0;
852};
853
854class VRecurrenceLayer : public VRoot
855{
856public:
857};
858
859class VLoopOutputLayer : public VRoot
860{
861public:
862 virtual LoopOutput getLoopOutput() const noexcept = 0;
863 virtual void setAxis(int32_t axis) noexcept = 0;
864 virtual int32_t getAxis() const noexcept = 0;
865};
866
867class VTripLimitLayer : public VRoot
868{
869public:
870 virtual TripLimit getTripLimit() const noexcept = 0;
871};
872
873class VIteratorLayer : public VRoot
874{
875public:
876 virtual void setAxis(int32_t axis) noexcept = 0;
877 virtual int32_t getAxis() const noexcept = 0;
878 virtual void setReverse(bool reverse) noexcept = 0;
879 virtual bool getReverse() const noexcept = 0;
880};
881class VLoop : public VRoot
882{
883public:
884 virtual IRecurrenceLayer* addRecurrence(ITensor& initialValue) noexcept = 0;
885 virtual ITripLimitLayer* addTripLimit(ITensor& tensor, TripLimit limit) noexcept = 0;
886 virtual IIteratorLayer* addIterator(ITensor& tensor, int32_t axis = 0, bool reverse = false) noexcept = 0;
887 virtual ILoopOutputLayer* addLoopOutput(ITensor& tensor, LoopOutput outputKind, int32_t axis = 0) noexcept = 0;
888 virtual void setName(char const* name) noexcept = 0;
889 virtual char const* getName() const noexcept = 0;
890};
891
892class VConditionalBoundaryLayer : public VRoot
893{
894public:
895 virtual IIfConditional* getConditional() const noexcept = 0;
896};
897
898class VConditionLayer : public VRoot
899{
900public:
901};
902
903class VConditionalInputLayer : public VRoot
904{
905public:
906};
907
908class VConditionalOutputLayer : public VRoot
909{
910public:
911};
912
913class VIfConditional : public VRoot
914{
915public:
916 virtual IConditionLayer* setCondition(ITensor& tensor) noexcept = 0;
917 virtual IIfConditionalInputLayer* addInput(ITensor& tensor) noexcept = 0;
918 virtual IIfConditionalOutputLayer* addOutput(ITensor& trueTensor, ITensor& falseTensor) noexcept = 0;
919 virtual void setName(char const* name) noexcept = 0;
920 virtual char const* getName() const noexcept = 0;
921};
922
923class VAttentionBoundaryLayer : public VRoot
924{
925public:
926 virtual IAttention* getAttention() const noexcept = 0;
927};
928
929class VAttentionInputLayer : public VRoot
930{
931public:
932};
933
934class VAttentionOutputLayer : public VRoot
935{
936public:
937};
938
939class VAttention : public VRoot
940{
941public:
942 TRT_NODISCARD virtual bool setInput(int32_t index, ITensor& input) noexcept = 0;
943 TRT_NODISCARD virtual int32_t getNbInputs() const noexcept = 0;
944 TRT_NODISCARD virtual ITensor* getInput(int32_t index) const noexcept = 0;
945 TRT_NODISCARD virtual int32_t getNbOutputs() const noexcept = 0;
946 TRT_NODISCARD virtual ITensor* getOutput(int32_t index) const noexcept = 0;
947 TRT_NODISCARD virtual bool setName(char const* name) noexcept = 0;
948 TRT_NODISCARD virtual char const* getName() const noexcept = 0;
949 TRT_NODISCARD virtual bool setNormalizationOperation(AttentionNormalizationOp op) noexcept = 0;
950 TRT_NODISCARD virtual AttentionNormalizationOp getNormalizationOperation() const noexcept = 0;
951 TRT_NODISCARD virtual bool setCausal(bool isCausal) noexcept = 0;
952 TRT_NODISCARD virtual bool getCausal() const noexcept = 0;
953 TRT_NODISCARD virtual bool setMask(ITensor& mask) noexcept = 0;
954 TRT_NODISCARD virtual ITensor* getMask() const noexcept = 0;
955 TRT_NODISCARD virtual bool setDecomposable(bool decomposable) noexcept = 0;
956 TRT_NODISCARD virtual bool getDecomposable() const noexcept = 0;
957 TRT_NODISCARD virtual bool setNormalizationQuantizeScale(ITensor& tensor) noexcept = 0;
958 TRT_NODISCARD virtual ITensor* getNormalizationQuantizeScale() const noexcept = 0;
959 TRT_NODISCARD virtual bool setNormalizationQuantizeToType(DataType type) noexcept = 0;
960 TRT_NODISCARD virtual DataType getNormalizationQuantizeToType() const noexcept = 0;
961}; // class VAttention
962
963class VSelectLayer : public VRoot
964{
965};
966
967class VAssertionLayer : public VRoot
968{
969public:
970 virtual void setMessage(char const* message) noexcept = 0;
971 virtual char const* getMessage() const noexcept = 0;
972};
973
974class VFillLayer : public VRoot
975{
976public:
977 virtual void setDimensions(Dims const& dimensions) noexcept = 0;
978 virtual Dims getDimensions() const noexcept = 0;
979 virtual void setOperation(FillOperation op) noexcept = 0;
980 virtual FillOperation getOperation() const noexcept = 0;
981 virtual void setAlpha(double alpha) noexcept = 0;
982 virtual double getAlpha() const noexcept = 0;
983 virtual void setBeta(double beta) noexcept = 0;
984 virtual double getBeta() const noexcept = 0;
985 virtual void setAlphaInt64(int64_t alpha) noexcept = 0;
986 virtual int64_t getAlphaInt64() const noexcept = 0;
987 virtual void setBetaInt64(int64_t beta) noexcept = 0;
988 virtual int64_t getBetaInt64() const noexcept = 0;
989 virtual bool isAlphaBetaInt64() const noexcept = 0;
990 virtual DataType getToType() const noexcept = 0;
991 virtual void setToType(DataType toType) noexcept = 0;
992};
993
994class VQuantizeLayer : public VRoot
995{
996public:
997 virtual int32_t getAxis() const noexcept = 0;
998 virtual void setAxis(int32_t axis) noexcept = 0;
999 virtual DataType getToType() const noexcept = 0;
1000 virtual void setToType(DataType toType) noexcept = 0;
1001};
1002
1003class VDequantizeLayer : public VRoot
1004{
1005public:
1006 virtual int32_t getAxis() const noexcept = 0;
1007 virtual void setAxis(int32_t axis) noexcept = 0;
1008 virtual DataType getToType() const noexcept = 0;
1009 virtual void setToType(DataType toType) noexcept = 0;
1010};
1011
1012class VDynamicQuantizeLayer : public VRoot
1013{
1014public:
1015 virtual int32_t getAxis() const noexcept = 0;
1016 virtual void setAxis(int32_t axis) noexcept = 0;
1017 virtual int32_t getBlockSize() const noexcept = 0;
1018 virtual void setBlockSize(int32_t axis) noexcept = 0;
1019 virtual DataType getScaleType() const noexcept = 0;
1020 virtual void setScaleType(DataType axis) noexcept = 0;
1021 virtual DataType getToType() const noexcept = 0;
1022 virtual void setToType(DataType toType) noexcept = 0;
1023};
1024
1025class VScatterLayer : public VRoot
1026{
1027public:
1028 virtual void setMode(ScatterMode mode) noexcept = 0;
1029 virtual ScatterMode getMode() const noexcept = 0;
1030 virtual void setAxis(int32_t axis) noexcept = 0;
1031 virtual int32_t getAxis() const noexcept = 0;
1032}; // class VScatterLayer
1033
1034class VEinsumLayer : public VRoot
1035{
1036public:
1037 virtual bool setEquation(char const* equation) noexcept = 0;
1038 virtual char const* getEquation() const noexcept = 0;
1039};
1040
1041class VOneHotLayer : public VRoot
1042{
1043public:
1044 virtual int32_t getAxis() const noexcept = 0;
1045 virtual void setAxis(int32_t axis) noexcept = 0;
1046}; // class VOneHotLayer
1047
1048class VGridSampleLayer : public VRoot
1049{
1050public:
1051 virtual void setInterpolationMode(InterpolationMode mode) noexcept = 0;
1052 virtual InterpolationMode getInterpolationMode() const noexcept = 0;
1053 virtual void setAlignCorners(bool alignCorners) noexcept = 0;
1054 virtual bool getAlignCorners() const noexcept = 0;
1055 virtual bool setSampleMode(SampleMode mode) noexcept = 0;
1056 virtual SampleMode getSampleMode() const noexcept = 0;
1057}; // class VGridSampleLayer
1058
1059class VNMSLayer : public VRoot
1060{
1061public:
1062 virtual void setBoundingBoxFormat(BoundingBoxFormat fmt) noexcept = 0;
1063 virtual BoundingBoxFormat getBoundingBoxFormat() const noexcept = 0;
1064 virtual void setTopKBoxLimit(int32_t limit) noexcept = 0;
1065 virtual int32_t getTopKBoxLimit() const noexcept = 0;
1066 virtual bool setIndicesType(DataType type) noexcept = 0;
1067 virtual DataType getIndicesType() const noexcept = 0;
1068}; // class VNMSLayer
1069
1070class VReverseSequenceLayer : public VRoot
1071{
1072public:
1073 virtual void setBatchAxis(int32_t batchAxis) noexcept = 0;
1074 virtual int32_t getBatchAxis() const noexcept = 0;
1075
1076 virtual void setSequenceAxis(int32_t sequenceAxis) noexcept = 0;
1077 virtual int32_t getSequenceAxis() const noexcept = 0;
1078}; // class VReverseSequenceLayer
1079
1080class VNormalizationLayer : public VRoot
1081{
1082public:
1083 virtual void setEpsilon(float eps) noexcept = 0;
1084 virtual float getEpsilon() const noexcept = 0;
1085 virtual void setAxes(uint32_t axesMask) noexcept = 0;
1086 virtual uint32_t getAxes() const noexcept = 0;
1087 virtual void setNbGroups(int64_t nbGroups) noexcept = 0;
1088 virtual int64_t getNbGroups() const noexcept = 0;
1089 virtual void setComputePrecision(DataType type) noexcept = 0;
1090 virtual DataType getComputePrecision() const noexcept = 0;
1091}; // class VNormalizationLayer
1092
1093class VSqueezeLayer : public VRoot
1094{
1095};
1096
1097class VUnsqueezeLayer : public VRoot
1098{
1099};
1100
1101class VCumulativeLayer : public VRoot
1102{
1103public:
1104 virtual bool setOperation(CumulativeOperation op) noexcept = 0;
1105 virtual CumulativeOperation getOperation() const noexcept = 0;
1106 virtual void setExclusive(bool exclusive) noexcept = 0;
1107 virtual bool getExclusive() const noexcept = 0;
1108 virtual void setReverse(bool reverse) noexcept = 0;
1109 virtual bool getReverse() const noexcept = 0;
1110}; // class VCumulativeLayer
1111
1112
1113class VNetworkDefinition : public VRoot
1114{
1115public:
1116 virtual ITensor* addInput(char const* name, DataType type, Dims const& dimensions) noexcept = 0;
1117 virtual void markOutput(ITensor& tensor) noexcept = 0;
1118 virtual IActivationLayer* addActivation(ITensor& input, ActivationType type) noexcept = 0;
1119 virtual ILRNLayer* addLRN(ITensor& input, int64_t window, float alpha, float beta, float k) noexcept = 0;
1120 virtual IScaleLayer* addScale(
1121 ITensor& input, ScaleMode mode, Weights shift, Weights scale, Weights power) noexcept = 0;
1122 virtual ISoftMaxLayer* addSoftMax(ITensor& input) noexcept = 0;
1123 virtual IConcatenationLayer* addConcatenation(ITensor* const* inputs, int32_t nbInputs) noexcept = 0;
1124 virtual IElementWiseLayer* addElementWise(ITensor& input1, ITensor& input2, ElementWiseOperation op) noexcept = 0;
1125 virtual IUnaryLayer* addUnary(ITensor& input, UnaryOperation operation) noexcept = 0;
1126 virtual IShuffleLayer* addShuffle(ITensor& input) noexcept = 0;
1127 virtual int32_t getNbLayers() const noexcept = 0;
1128 virtual ILayer* getLayer(int32_t index) const noexcept = 0;
1129 virtual int32_t getNbInputs() const noexcept = 0;
1130 virtual ITensor* getInput(int32_t index) const noexcept = 0;
1131 virtual int32_t getNbOutputs() const noexcept = 0;
1132 virtual ITensor* getOutput(int32_t index) const noexcept = 0;
1133 virtual IReduceLayer* addReduce(
1134 ITensor& input, ReduceOperation operation, uint32_t reduceAxes, bool keepDimensions) noexcept
1135 = 0;
1136 virtual ITopKLayer* addTopK(ITensor& input, TopKOperation op, int32_t k, uint32_t reduceAxes) noexcept = 0;
1137 virtual IGatherLayer* addGather(ITensor& data, ITensor& indices, int32_t axis) noexcept = 0;
1138 virtual IRaggedSoftMaxLayer* addRaggedSoftMax(ITensor& input, ITensor& bounds) noexcept = 0;
1139 virtual IMatrixMultiplyLayer* addMatrixMultiply(
1140 ITensor& input0, MatrixOperation op0, ITensor& input1, MatrixOperation op1) noexcept = 0;
1141 virtual IConstantLayer* addConstant(Dims const& dimensions, Weights weights) noexcept = 0;
1142 virtual IIdentityLayer* addIdentity(ITensor& input) noexcept = 0;
1143 virtual void removeTensor(ITensor& tensor) noexcept = 0;
1144 virtual void unmarkOutput(ITensor& tensor) noexcept = 0;
1145 virtual IPluginV2Layer* addPluginV2(ITensor* const* inputs, int32_t nbInputs, IPluginV2& plugin) noexcept = 0;
1146 virtual IPluginV3Layer* addPluginV3(ITensor* const* inputs, int32_t nbInputs, ITensor* const* shapeInputs,
1147 int32_t nbShapeInputs, IPluginV3& plugin) noexcept = 0;
1148 virtual ISliceLayer* addSlice(ITensor& input, Dims const& start, Dims const& size, Dims const& stride) noexcept = 0;
1149 virtual void setName(char const* name) noexcept = 0;
1150 virtual char const* getName() const noexcept = 0;
1151 virtual IShapeLayer* addShape(ITensor& input) noexcept = 0;
1152 virtual bool hasImplicitBatchDimension() const noexcept = 0;
1153 virtual bool markOutputForShapes(ITensor& tensor) noexcept = 0;
1154 virtual bool unmarkOutputForShapes(ITensor& tensor) noexcept = 0;
1155 virtual IParametricReLULayer* addParametricReLU(ITensor& input, ITensor& slope) noexcept = 0;
1156 virtual IConvolutionLayer* addConvolutionNd(
1157 ITensor& input, int64_t nbOutputMaps, Dims const& kernelSize, Weights kernelWeights, Weights biasWeights) noexcept
1158 = 0;
1159 virtual IPoolingLayer* addPoolingNd(ITensor& input, PoolingType type, Dims const& windowSize) noexcept = 0;
1160 virtual IDeconvolutionLayer* addDeconvolutionNd(
1161 ITensor& input, int64_t nbOutputMaps, Dims const& kernelSize, Weights kernelWeights, Weights biasWeights) noexcept
1162 = 0;
1163 virtual IScaleLayer* addScaleNd(
1164 ITensor& input, ScaleMode mode, Weights shift, Weights scale, Weights power, int32_t channelAxis) noexcept = 0;
1165 virtual IResizeLayer* addResize(ITensor& input) noexcept = 0;
1166 virtual ILoop* addLoop() noexcept = 0;
1167 virtual ISelectLayer* addSelect(ITensor& condition, ITensor& thenInput, ITensor& elseInput) noexcept = 0;
1168 virtual IFillLayer* addFill(Dims const& dimensions, FillOperation op) noexcept = 0;
1169 virtual IPaddingLayer* addPaddingNd(ITensor& input, Dims const& prePadding, Dims const& postPadding) noexcept = 0;
1170 virtual bool setWeightsName(Weights weights, char const* name) noexcept = 0;
1171 virtual void setErrorRecorder(IErrorRecorder* recorder) noexcept = 0;
1172 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
1173 virtual IDequantizeLayer* addDequantize(ITensor& input, ITensor& scale) noexcept = 0;
1174 virtual IQuantizeLayer* addQuantize(ITensor& input, ITensor& scale) noexcept = 0;
1175 virtual IGatherLayer* addGatherV2(ITensor& data, ITensor& indices, GatherMode mode) noexcept = 0;
1176 virtual IIfConditional* addIfConditional() noexcept = 0;
1177 virtual IScatterLayer* addScatter(ITensor& data, ITensor& indices, ITensor& updates, ScatterMode mode) noexcept = 0;
1178 virtual IEinsumLayer* addEinsum(ITensor* const* inputs, int32_t nbInputs, char const* equation) noexcept = 0;
1179 virtual IAssertionLayer* addAssertion(ITensor& condition, char const* message) noexcept = 0;
1180 virtual IOneHotLayer* addOneHot(ITensor& indices, ITensor& values, ITensor& depth, int32_t axis) noexcept = 0;
1181 virtual INonZeroLayer* addNonZero(ITensor& input) noexcept = 0;
1182 virtual IGridSampleLayer* addGridSample(ITensor& input, ITensor& grid) noexcept = 0;
1183 virtual INMSLayer* addNMS(ITensor& boxes, ITensor& scores, ITensor& maxOutputBoxesPerClass) noexcept = 0;
1184 virtual IReverseSequenceLayer* addReverseSequence(ITensor& input, ITensor& sequenceLens) noexcept = 0;
1185 virtual INormalizationLayer* addNormalization(
1186 ITensor& input, ITensor& scale, ITensor& bias, uint32_t axesMask) noexcept = 0;
1187 virtual ICastLayer* addCast(ITensor& input, DataType toType) noexcept = 0;
1188 virtual IBuilder& getBuilder() const noexcept = 0;
1189 virtual NetworkDefinitionCreationFlags getFlags() const noexcept = 0;
1190 virtual bool getFlag(NetworkDefinitionCreationFlag networkDefinitionCreationFlag) const noexcept = 0;
1191 virtual IQuantizeLayer* addQuantizeV2(ITensor& input, ITensor& scale, DataType outputType) noexcept = 0;
1192 virtual IDequantizeLayer* addDequantizeV2(ITensor& input, ITensor& scale, DataType outputType) noexcept = 0;
1193 virtual IFillLayer* addFillV2(Dims const& dimensions, FillOperation op, DataType outputType) noexcept = 0;
1194 virtual bool markDebug(ITensor& tensor) noexcept = 0;
1195 virtual bool unmarkDebug(ITensor& tensor) noexcept = 0;
1196 virtual bool isDebugTensor(ITensor const& tensor) const noexcept = 0;
1197 virtual bool markWeightsRefittable(char const* name) noexcept = 0;
1198 virtual bool unmarkWeightsRefittable(char const* name) noexcept = 0;
1199 virtual bool areWeightsMarkedRefittable(char const* name) const noexcept = 0;
1200 virtual ISqueezeLayer* addSqueeze(ITensor& input, ITensor& axes) noexcept = 0;
1201 virtual IUnsqueezeLayer* addUnsqueeze(ITensor& input, ITensor& axes) noexcept = 0;
1202 virtual IDynamicQuantizeLayer* addDynamicQuantize(
1203 ITensor& input, int32_t axis, int32_t blockSize, DataType toType, DataType scaleType) noexcept
1204 = 0;
1205 virtual ICumulativeLayer* addCumulative(
1206 ITensor& input, ITensor& axis, CumulativeOperation operation, bool exclusive, bool reverse) noexcept
1207 = 0;
1208 virtual bool markUnfusedTensorsAsDebugTensors() noexcept = 0;
1209 virtual bool unmarkUnfusedTensorsAsDebugTensors() noexcept = 0;
1210 virtual ITopKLayer* addTopKV2(
1211 ITensor& input, TopKOperation op, int32_t k, uint32_t reduceAxes, DataType indicesType) noexcept
1212 = 0;
1213 virtual INonZeroLayer* addNonZeroV2(ITensor& input, DataType indicesType) noexcept = 0;
1214 virtual INMSLayer* addNMSV2(
1215 ITensor& boxes, ITensor& scores, ITensor& maxOutputBoxesPerClass, DataType indicesType) noexcept
1216 = 0;
1217 virtual IAttention* addAttention(
1218 ITensor& query, ITensor& key, ITensor& value, AttentionNormalizationOp normOp, bool isCausal) noexcept
1219 = 0;
1220};
1221
1222class VAlgorithmIOInfo : public VRoot
1223{
1224public:
1225 virtual DataType getDataType() const noexcept = 0;
1226 virtual Dims getStrides() const noexcept = 0;
1227 virtual int64_t getVectorizedDim() const noexcept = 0;
1228 virtual int64_t getComponentsPerElement() const noexcept = 0;
1229};
1230
1231class VAlgorithmVariant : public VRoot
1232{
1233public:
1234 virtual int64_t getImplementation() const noexcept = 0;
1235 virtual int64_t getTactic() const noexcept = 0;
1236};
1237
1238class VAlgorithmContext : public VRoot
1239{
1240public:
1241 virtual char const* getName() const noexcept = 0;
1242 virtual Dims getDimensions(int32_t index, OptProfileSelector select) const noexcept = 0;
1243 virtual int32_t getNbInputs() const noexcept = 0;
1244 virtual int32_t getNbOutputs() const noexcept = 0;
1245};
1246
1247class VAlgorithm : public VRoot
1248{
1249public:
1250 virtual IAlgorithmVariant const& getAlgorithmVariant() const noexcept = 0;
1251 virtual float getTimingMSec() const noexcept = 0;
1252 virtual std::size_t getWorkspaceSize() const noexcept = 0;
1253 virtual IAlgorithmIOInfo const* getAlgorithmIOInfoByIndex(int32_t index) const noexcept = 0;
1254};
1255
1256
1257class VTimingCache : public VRoot
1258{
1259public:
1260 virtual nvinfer1::IHostMemory* serialize() const noexcept = 0;
1261 virtual bool combine(ITimingCache const& inputCache, bool ignoreMismatch) noexcept = 0;
1262 virtual bool reset() noexcept = 0;
1263 virtual int64_t queryKeys(TimingCacheKey* keyBuffer, int64_t capacity) const noexcept = 0;
1264 virtual TimingCacheValue query(TimingCacheKey const& key) const noexcept = 0;
1265 virtual bool update(TimingCacheKey const& key, TimingCacheValue const& value) noexcept = 0;
1266};
1267
1268class VBuilderConfig : public VRoot
1269{
1270public:
1271 virtual void setAvgTimingIterations(int32_t avgTiming) noexcept = 0;
1272 virtual int32_t getAvgTimingIterations() const noexcept = 0;
1273 virtual void setEngineCapability(EngineCapability capability) noexcept = 0;
1274 virtual EngineCapability getEngineCapability() const noexcept = 0;
1275 virtual void setInt8Calibrator(IInt8Calibrator* calibrator) noexcept = 0;
1276 virtual IInt8Calibrator* getInt8Calibrator() const noexcept = 0;
1277 virtual void setFlags(BuilderFlags builderFlags) noexcept = 0;
1278 virtual BuilderFlags getFlags() const noexcept = 0;
1279 virtual void clearFlag(BuilderFlag builderFlag) noexcept = 0;
1280 virtual void setFlag(BuilderFlag builderFlag) noexcept = 0;
1281 virtual bool getFlag(BuilderFlag builderFlag) const noexcept = 0;
1282 virtual void setDeviceType(ILayer const* layer, DeviceType deviceType) noexcept = 0;
1283 virtual DeviceType getDeviceType(ILayer const* layer) const noexcept = 0;
1284 virtual bool isDeviceTypeSet(ILayer const* layer) const noexcept = 0;
1285 virtual void resetDeviceType(ILayer const* layer) noexcept = 0;
1286 virtual bool canRunOnDLA(ILayer const* layer) const noexcept = 0;
1287 virtual void setDLACore(int32_t dlaCore) noexcept = 0;
1288 virtual int32_t getDLACore() const noexcept = 0;
1289 virtual void setDefaultDeviceType(DeviceType deviceType) noexcept = 0;
1290 virtual DeviceType getDefaultDeviceType() const noexcept = 0;
1291 virtual void reset() noexcept = 0;
1292 virtual void setProfileStream(const cudaStream_t stream) noexcept = 0;
1293 virtual cudaStream_t getProfileStream() const noexcept = 0;
1294 virtual int32_t addOptimizationProfile(IOptimizationProfile const* profile) noexcept = 0;
1295 virtual int32_t getNbOptimizationProfiles() const noexcept = 0;
1296 virtual void setProfilingVerbosity(ProfilingVerbosity verbosity) noexcept = 0;
1297 virtual ProfilingVerbosity getProfilingVerbosity() const noexcept = 0;
1298 virtual void setAlgorithmSelector(IAlgorithmSelector* selector) noexcept = 0;
1299 virtual IAlgorithmSelector* getAlgorithmSelector() const noexcept = 0;
1300 virtual bool setCalibrationProfile(IOptimizationProfile const* profile) noexcept = 0;
1301 virtual IOptimizationProfile const* getCalibrationProfile() noexcept = 0;
1302 virtual void setQuantizationFlags(QuantizationFlags flags) noexcept = 0;
1303 virtual QuantizationFlags getQuantizationFlags() const noexcept = 0;
1304 virtual void clearQuantizationFlag(QuantizationFlag flag) noexcept = 0;
1305 virtual void setQuantizationFlag(QuantizationFlag flag) noexcept = 0;
1306 virtual bool getQuantizationFlag(QuantizationFlag flag) const noexcept = 0;
1307 virtual bool setTacticSources(TacticSources tacticSources) noexcept = 0;
1308 virtual TacticSources getTacticSources() const noexcept = 0;
1309 virtual nvinfer1::ITimingCache* createTimingCache(void const* blob, std::size_t size) const noexcept = 0;
1310 virtual bool setTimingCache(ITimingCache const& cache, bool ignoreMismatch) noexcept = 0;
1311 virtual nvinfer1::ITimingCache const* getTimingCache() const noexcept = 0;
1312 virtual void setMemoryPoolLimit(MemoryPoolType pool, std::size_t poolSize) noexcept = 0;
1313 virtual std::size_t getMemoryPoolLimit(MemoryPoolType pool) const noexcept = 0;
1314 virtual void setPreviewFeature(PreviewFeature feature, bool enable) noexcept = 0;
1315 virtual bool getPreviewFeature(PreviewFeature feature) const noexcept = 0;
1316 virtual void setBuilderOptimizationLevel(int32_t level) noexcept = 0;
1317 virtual int32_t getBuilderOptimizationLevel() const noexcept = 0;
1318 virtual void setHardwareCompatibilityLevel(HardwareCompatibilityLevel hardwareCompatibilityLevel) noexcept = 0;
1319 virtual HardwareCompatibilityLevel getHardwareCompatibilityLevel() const noexcept = 0;
1320 virtual void setPluginsToSerialize(char const* const* paths, int32_t nbPaths) noexcept = 0;
1321 virtual char const* getPluginToSerialize(int32_t index) const noexcept = 0;
1322 virtual int32_t getNbPluginsToSerialize() const noexcept = 0;
1323 virtual void setMaxAuxStreams(int32_t nbStreams) noexcept = 0;
1324 virtual int32_t getMaxAuxStreams() const noexcept = 0;
1325 virtual void setProgressMonitor(IProgressMonitor* monitor) noexcept = 0;
1326 virtual IProgressMonitor* getProgressMonitor() const noexcept = 0;
1327 virtual void setRuntimePlatform(RuntimePlatform runtimePlatform) noexcept = 0;
1328 virtual RuntimePlatform getRuntimePlatform() const noexcept = 0;
1329 virtual void setMaxNbTactics(int32_t maxTactics) noexcept = 0;
1330 virtual int32_t getMaxNbTactics() const noexcept = 0;
1331 virtual bool setTilingOptimizationLevel(TilingOptimizationLevel level) noexcept = 0;
1332 virtual TilingOptimizationLevel getTilingOptimizationLevel() const noexcept = 0;
1333 virtual bool setL2LimitForTiling(int64_t size) noexcept = 0;
1334 virtual int64_t getL2LimitForTiling() const noexcept = 0;
1335 virtual bool setRemoteAutoTuningConfig(char const* config) noexcept = 0;
1336 virtual char const* getRemoteAutoTuningConfig() const noexcept = 0;
1337};
1338
1339class VSerializationConfig : public VRoot
1340{
1341public:
1342 virtual bool setFlags(SerializationFlags serializationFlags) noexcept = 0;
1343 virtual SerializationFlags getFlags() const noexcept = 0;
1344 virtual bool clearFlag(SerializationFlag serializationFlag) noexcept = 0;
1345 virtual bool setFlag(SerializationFlag serializationFlag) noexcept = 0;
1346 virtual bool getFlag(SerializationFlag serializationFlag) const noexcept = 0;
1347};
1348
1349class VBuilder : public VRoot
1350{
1351public:
1352 virtual bool platformHasFastFp16() const noexcept = 0;
1353 virtual bool platformHasFastInt8() const noexcept = 0;
1354 virtual int32_t getMaxDLABatchSize() const noexcept = 0;
1355 virtual int32_t getNbDLACores() const noexcept = 0;
1356 virtual void setGpuAllocator(IGpuAllocator* allocator) noexcept = 0;
1357 virtual nvinfer1::IBuilderConfig* createBuilderConfig() noexcept = 0;
1358 virtual nvinfer1::INetworkDefinition* createNetworkV2(NetworkDefinitionCreationFlags flags) noexcept = 0;
1359 virtual nvinfer1::IOptimizationProfile* createOptimizationProfile() noexcept = 0;
1360 virtual void setErrorRecorder(IErrorRecorder* recorder) noexcept = 0;
1361 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
1362 virtual void reset() noexcept = 0;
1363 virtual bool platformHasTf32() const noexcept = 0;
1364 virtual nvinfer1::IHostMemory* buildSerializedNetwork(
1365 INetworkDefinition& network, IBuilderConfig& config) noexcept = 0;
1366 virtual bool isNetworkSupported(INetworkDefinition const& network, IBuilderConfig const& config) const noexcept = 0;
1367 virtual ILogger* getLogger() const noexcept = 0;
1368 virtual bool setMaxThreads(int32_t maxThreads) noexcept = 0;
1369 virtual int32_t getMaxThreads() const noexcept = 0;
1370 virtual IPluginRegistry& getPluginRegistry() noexcept = 0;
1371 virtual ICudaEngine* buildEngineWithConfig(INetworkDefinition& network, IBuilderConfig& config) noexcept = 0;
1372 virtual bool buildSerializedNetworkToStream(
1373 INetworkDefinition& network, IBuilderConfig& config, IStreamWriter& writer) noexcept = 0;
1374 virtual nvinfer1::IHostMemory* buildSerializedNetworkWithKernelText(
1375 INetworkDefinition& network, IBuilderConfig& config, IHostMemory*& kernelText) noexcept = 0;
1376};
1377
1378class VRuntimeConfig : public VRoot
1379{
1380public:
1381 virtual IRuntimeConfig* getPImpl() noexcept = 0;
1382 virtual void setExecutionContextAllocationStrategy(ExecutionContextAllocationStrategy strategy) noexcept = 0;
1383 virtual ExecutionContextAllocationStrategy getExecutionContextAllocationStrategy() const noexcept = 0;
1384};
1385
1386
1387} // namespace apiv
1388} // namespace nvinfer1
1389
1390// @endcond
1391
1392#endif // NV_INFER_RUNTIME_IMPL_H
getPluginRegistry
TENSORRTAPI nvinfer1::IPluginRegistry * getPluginRegistry() noexcept
Return the plugin registry.
getLogger
TENSORRTAPI nvinfer1::ILogger * getLogger() noexcept
Return the logger object.
TRT_NODISCARD
#define TRT_NODISCARD
A stand-in for [[nodiscard]] and [[nodiscard(REASON)]] that works with older compilers.
Definition: NvInferRuntimeBase.h:57
Dims
Structure to define the dimensions of a tensor.
IAlgorithmSelector
Interface implemented by application for selecting and reporting algorithms of a layer provided by th...
IDebugListener
User-implemented callback for notification when value of a debug tensor is updated.
IErrorRecorder
Reference counted application-implemented error reporting interface for TensorRT objects.
IGpuAllocator
Application-implemented class for controlling allocation on the GPU.
IOutputAllocator
Callback from ExecutionContext::enqueueV3()
IPluginV3
Plugin class for the V3 generation of user-implemented layers.
IProfiler
Application-implemented interface for profiling.
IProgressMonitor
Application-implemented progress reporting interface for TensorRT.
IStreamReader
Application-implemented class for reading data in a stream-based manner.
IStreamReaderV2
Application-implemented class for reading data in a stream-based manner asynchronously....
IStreamWriter
Application-implemented class for writing data in a stream-based manner.
nvinfer1::IHostMemory
Class to handle library allocated memory that is accessible to the user.
Definition: NvInferRuntime.h:142
nvinfer1
The TensorRT API version 1 namespace.
nvinfer1::TacticSources
uint32_t TacticSources
Represents a collection of one or more TacticSource values combine using bitwise-OR operations.
Definition: NvInferRuntime.h:2961
nvinfer1::IOutputAllocator
v_1_0::IOutputAllocator IOutputAllocator
Definition: NvInferRuntime.h:4245
nvinfer1::ResizeSelector
ResizeSelector
The coordinate selector when resize to single pixel output.
Definition: NvInfer.h:4112
nvinfer1::EngineCapability
EngineCapability
List of supported engine capability flows.
Definition: NvInferRuntime.h:76
nvinfer1::DimensionOperation
DimensionOperation
An operation on two IDimensionExpr, which represent integer expressions used in dimension computation...
Definition: NvInferRuntime.h:179
nvinfer1::MemoryPoolType
MemoryPoolType
The type for memory pools used by TensorRT.
Definition: NvInfer.h:9824
nvinfer1::ScaleMode
ScaleMode
Controls how shift, scale and power are applied in a Scale layer.
Definition: NvInfer.h:1853
nvinfer1::RuntimePlatform
RuntimePlatform
Describes the intended runtime platform (operating system and CPU architecture) for the execution of ...
Definition: NvInfer.h:9407
nvinfer1::QuantizationFlags
uint32_t QuantizationFlags
Represents one or more QuantizationFlag values using binary OR operations.
Definition: NvInfer.h:9359
nvinfer1::TensorIOMode
TensorIOMode
Definition of tensor IO Mode.
Definition: NvInferRuntimeBase.h:658
nvinfer1::HardwareCompatibilityLevel
HardwareCompatibilityLevel
Describes requirements of compatibility with GPU architectures other than that of the GPU on which th...
Definition: NvInfer.h:9943
nvinfer1::CumulativeOperation
CumulativeOperation
Enumerates the cumulative operations that may be performed by a Cumulative layer.
Definition: NvInfer.h:6624
nvinfer1::SerializationFlag
SerializationFlag
List of valid flags that the engine can enable when serializing the bytes.
Definition: NvInferRuntime.h:3002
nvinfer1::BoundingBoxFormat
BoundingBoxFormat
Representation of bounding box data used for the Boxes input tensor in INMSLayer.
Definition: NvInfer.h:6178
nvinfer1::IPluginV3
v_1_0::IPluginV3 IPluginV3
Definition: NvInferPluginBase.h:273
nvinfer1::UnaryOperation
UnaryOperation
Enumerates the unary operations that may be performed by a Unary layer.
Definition: NvInfer.h:2798
nvinfer1::IStreamWriter
v_1_0::IStreamWriter IStreamWriter
Definition: NvInferRuntime.h:720
nvinfer1::IProfiler
v_1_0::IProfiler IProfiler
Definition: NvInferRuntime.h:1318
nvinfer1::IAlgorithmSelector
v_1_0::IAlgorithmSelector IAlgorithmSelector
Definition: NvInfer.h:9351
nvinfer1::IStreamReaderV2
v_1_0::IStreamReaderV2 IStreamReaderV2
Definition: NvInferRuntime.h:797
nvinfer1::ActivationType
ActivationType
Enumerates the types of activation to perform in an activation layer.
Definition: NvInfer.h:139
nvinfer1::TempfileControlFlags
uint32_t TempfileControlFlags
Represents a collection of one or more TempfileControlFlag values combined using bitwise-OR operation...
Definition: NvInferRuntime.h:1396
nvinfer1::FillOperation
FillOperation
Enumerates the tensor fill operations that may performed by a fill layer.
Definition: NvInfer.h:5093
nvinfer1::EngineStat
EngineStat
The kind of engine statistics that queried from the ICudaEngine.
Definition: NvInferRuntime.h:3170
nvinfer1::ResizeRoundMode
ResizeRoundMode
The rounding mode for nearest neighbor resize.
Definition: NvInfer.h:4142
nvinfer1::PaddingMode
PaddingMode
Enumerates the modes of padding to perform in convolution, deconvolution and pooling layer,...
Definition: NvInfer.h:1031
nvinfer1::TripLimit
TripLimit
Enum that describes kinds of trip limits.
Definition: NvInfer.h:4501
nvinfer1::NetworkDefinitionCreationFlags
uint32_t NetworkDefinitionCreationFlags
Represents one or more NetworkDefinitionCreationFlag flags using binary OR operations....
Definition: NvInfer.h:11077
nvinfer1::PreviewFeature
PreviewFeature
Define preview features.
Definition: NvInfer.h:9899
nvinfer1::TilingOptimizationLevel
TilingOptimizationLevel
Define the optimization levels for Tiling.
Definition: NvInfer.h:9996
nvinfer1::DataType
DataType
The type of weights and tensors. The datatypes other than kBOOL, kINT32, and kINT64 are "activation d...
Definition: NvInferRuntimeBase.h:145
nvinfer1::BuilderFlags
uint32_t BuilderFlags
Represents one or more BuilderFlag values using binary OR operations, e.g., 1U << BuilderFlag::kFP16 ...
Definition: NvInfer.h:9439
nvinfer1::DeviceType
DeviceType
The device that this layer/network will execute on.
Definition: NvInferRuntime.h:1350
nvinfer1::CalibrationAlgoType
CalibrationAlgoType
Version of calibration algorithm to use.
Definition: NvInfer.h:8785
nvinfer1::LayerType
LayerType
The type values of layer classes.
Definition: NvInfer.h:58
nvinfer1::SampleMode
SampleMode
Controls how ISliceLayer and IGridSample handle out-of-bounds coordinates.
Definition: NvInfer.h:3256
nvinfer1::GatherMode
GatherMode
Control form of IGatherLayer.
Definition: NvInfer.h:2545
nvinfer1::IDebugListener
v_1_0::IDebugListener IDebugListener
Definition: NvInferRuntime.h:4286
nvinfer1::WeightsRole
WeightsRole
How a layer uses particular Weights.
Definition: NvInferRuntime.h:1328
nvinfer1::TensorFormats
uint32_t TensorFormats
It is capable of representing one or more TensorFormat by binary OR operations, e....
Definition: NvInfer.h:131
nvinfer1::ProfilingVerbosity
ProfilingVerbosity
List of verbosity levels of layer information exposed in NVTX annotations and in IEngineInspector.
Definition: NvInferRuntime.h:2973
nvinfer1::NetworkDefinitionCreationFlag
NetworkDefinitionCreationFlag
List of immutable network properties expressed at network creation time. NetworkDefinitionCreationFla...
Definition: NvInfer.h:11088
nvinfer1::ElementWiseOperation
ElementWiseOperation
Enumerates the binary operations that may be performed by an ElementWise layer.
Definition: NvInfer.h:2455
nvinfer1::QuantizationFlag
QuantizationFlag
List of valid flags for quantizing the network to int8.
Definition: NvInfer.h:9371
nvinfer1::SerializationFlags
uint32_t SerializationFlags
Represents one or more SerializationFlag values using binary OR operations, e.g., 1U << Serialization...
Definition: NvInferRuntime.h:2992
nvinfer1::InterpolationMode
InterpolationMode
Enumerates various modes of interpolation.
Definition: NvInfer.h:4030
nvinfer1::BuilderFlag
BuilderFlag
List of valid modes that the builder can enable when creating an engine from a network definition.
Definition: NvInfer.h:9449
nvinfer1::TensorFormat
TensorFormat
Format of the input/output tensors.
Definition: NvInferRuntime.h:1430
nvinfer1::ExecutionContextAllocationStrategy
ExecutionContextAllocationStrategy
Different memory allocation behaviors for IExecutionContext.
Definition: NvInferRuntime.h:3107
nvinfer1::TopKOperation
TopKOperation
Enumerates the operations that may be performed by a TopK layer.
Definition: NvInfer.h:3537
nvinfer1::ReduceOperation
ReduceOperation
Enumerates the reduce operations that may be performed by a Reduce layer.
Definition: NvInfer.h:2893
nvinfer1::LayerInformationFormat
LayerInformationFormat
The format in which the IEngineInspector prints the layer information.
Definition: NvInferRuntime.h:5242
nvinfer1::ScatterMode
ScatterMode
Control form of IScatterLayer.
Definition: NvInfer.h:5912
nvinfer1::MatrixOperation
MatrixOperation
Enumerates the operations that may be performed on a tensor by IMatrixMultiplyLayer before multiplica...
Definition: NvInfer.h:3690
nvinfer1::ResizeCoordinateTransformation
ResizeCoordinateTransformation
The resize coordinate transformation function.
Definition: NvInfer.h:4058
nvinfer1::LoopOutput
LoopOutput
Enum that describes kinds of loop outputs.
Definition: NvInfer.h:4473
nvinfer1::IStreamReader
v_1_0::IStreamReader IStreamReader
Definition: NvInferRuntime.h:710
nvinfer1::PoolingType
PoolingType
The type of pooling to perform in a pooling layer.
Definition: NvInfer.h:1467
nvinfer1::IProgressMonitor
v_1_0::IProgressMonitor IProgressMonitor
Definition: NvInfer.h:10112
nvinfer1::TensorLocation
TensorLocation
The location for tensor data storage, device or host.
Definition: NvInferRuntime.h:204
nvinfer1::OptProfileSelector
OptProfileSelector
When setting or querying optimization profile parameters (such as shape tensor inputs or dynamic dime...
Definition: NvInferRuntime.h:2635
nvinfer1::AttentionNormalizationOp
AttentionNormalizationOp
Enumerates the operations that may be performed by the normalization in the attention subgraph.
Definition: NvInfer.h:6759

  Copyright © 2024 NVIDIA Corporation
  Privacy Policy | Manage My Privacy | Do Not Sell or Share My Data | Terms of Service | Accessibility | Corporate Policies | Product Security | Contact