TensorRT for RTX 1.2.0
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
37
38namespace v_1_0
39{
40class IProfiler;
41} // namespace v_1_0
42using IProfiler = v_1_0::IProfiler;
43
44namespace v_1_0
45{
46class IOutputAllocator;
47} // namespace v_1_0
48using IOutputAllocator = v_1_0::IOutputAllocator;
49
50namespace v_1_0
51{
52class IDebugListener;
53} // namespace v_1_0
54using IDebugListener = v_1_0::IDebugListener;
55
56class IActivationLayer;
57class IAssertionLayer;
58class IAttention;
59class IBuilder;
60class IBuilderConfig;
61class IConcatenationLayer;
62class IConditionLayer;
63class IConstantLayer;
64class IConvolutionLayer;
65class ICudaEngine;
66class ICumulativeLayer;
67class IDeconvolutionLayer;
68class IDequantizeLayer;
69class IDimensionExpr;
70class IDynamicQuantizeLayer;
71class IEinsumLayer;
72class IElementWiseLayer;
73class IEngineInspector;
74class IExecutionContext;
75class IFillLayer;
76class IGatherLayer;
77class IGridSampleLayer;
78class IHostMemory;
79class IIdentityLayer;
80class ICastLayer;
81class IIfConditional;
82class IIfConditionalInputLayer;
83class IIfConditionalOutputLayer;
84class IIteratorLayer;
85class ILayer;
86class ILoop;
87class ILoopOutputLayer;
88class ILRNLayer;
89class IMatrixMultiplyLayer;
90class INetworkDefinition;
91class INormalizationLayer;
92class INMSLayer;
93class INonZeroLayer;
94class IOneHotLayer;
95class IOptimizationProfile;
96class IPaddingLayer;
97class IParametricReLULayer;
98class IPlugin;
99class IPluginExt;
100class IPluginFactory;
101class IPluginLayer;
102class IPluginRegistry;
103class IPluginV2Layer;
104class IRuntimeCache;
105class IRuntimeConfig;
106
107namespace v_1_0
108{
109class IPluginV3;
110} // namespace v_1_0
111using IPluginV3 = v_1_0::IPluginV3;
112
113namespace v_1_0
114{
115class IStreamReader;
116class IStreamWriter;
117} // namespace v_1_0
118using IStreamReader = v_1_0::IStreamReader;
119using IStreamWriter = v_1_0::IStreamWriter;
120namespace v_1_0
121{
122class IStreamReaderV2;
123} // namespace v_1_0
124using IStreamReaderV2 = v_1_0::IStreamReaderV2;
125
126class IPluginV3Layer;
127class IPoolingLayer;
128class IQuantizeLayer;
129class IRaggedSoftMaxLayer;
130class IRecurrenceLayer;
131class IReduceLayer;
132class IRefitter;
133class IResizeLayer;
134class IReverseSequenceLayer;
135class IRuntime;
136class IScaleLayer;
137class IScatterLayer;
138class ISelectLayer;
139class ISerializationConfig;
140class IShapeLayer;
141class IShuffleLayer;
142class ISliceLayer;
143class ISoftMaxLayer;
144class ISqueezeLayer;
145class ITensor;
146
147namespace v_1_0
148{
149struct TimingCacheKey;
150struct TimingCacheValue;
151} // namespace v_1_0
152using TimingCacheKey = v_1_0::TimingCacheKey;
153using TimingCacheValue = v_1_0::TimingCacheValue;
154
155class ITimingCache;
156class ITopKLayer;
157class ITripLimitLayer;
158class IUnaryLayer;
159class IUnsqueezeLayer;
160struct Permutation;
161class Weights;
162
163enum class ActivationType : int32_t;
164enum class AttentionNormalizationOp : int32_t;
165enum class BoundingBoxFormat : int32_t;
166enum class BuilderFlag : int32_t;
167enum class CumulativeOperation : int32_t;
168enum class DeviceType : int32_t;
169enum class DimensionOperation : int32_t;
170enum class ElementWiseOperation : int32_t;
171enum class EngineCapability : int32_t;
172enum class FillOperation : int32_t;
173enum class GatherMode : int32_t;
174enum class LayerInformationFormat : int32_t;
175enum class LayerType : int32_t;
176enum class LoopOutput : int32_t;
177enum class MatrixOperation : int32_t;
178enum class MemoryPoolType : int32_t;
179enum class NetworkDefinitionCreationFlag : int32_t;
180enum class OptProfileSelector : int32_t;
181enum class PaddingMode : int32_t;
182enum class PoolingType : int32_t;
183enum class ProfilingVerbosity : int32_t;
184enum class QuantizationFlag : int32_t;
185enum class ReduceOperation : int32_t;
186enum class ResizeCoordinateTransformation : int32_t;
187enum class InterpolationMode : int32_t;
188enum class ResizeRoundMode : int32_t;
189enum class ResizeSelector : int32_t;
190enum class ScaleMode : int32_t;
191enum class ScatterMode : int32_t;
192enum class SampleMode : int32_t;
193enum class SerializationFlag : int32_t;
194enum class TensorIOMode : int32_t;
195enum class TensorLocation : int32_t;
196enum class TopKOperation : int32_t;
197enum class TripLimit : int32_t;
198enum class UnaryOperation : int32_t;
199enum class WeightsRole : int32_t;
200enum class PreviewFeature : int32_t;
201enum class HardwareCompatibilityLevel : int32_t;
202enum class ExecutionContextAllocationStrategy : int32_t;
203enum class RuntimePlatform : int32_t;
204enum class TilingOptimizationLevel : int32_t;
205enum class ComputeCapability : int32_t;
206enum class EngineValidity : int32_t;
207enum class EngineStat : int32_t;
208
209enum class DynamicShapesKernelSpecializationStrategy : int32_t;
210enum class CudaGraphStrategy : int32_t;
211
212using TacticSources = uint32_t;
213using TensorFormats = uint32_t;
214using BuilderFlags = uint32_t;
215using NetworkDefinitionCreationFlags = uint32_t;
216using QuantizationFlags = uint32_t;
217using TempfileControlFlags = uint32_t;
218using SerializationFlags = uint32_t;
219
227
228namespace apiv
229{
230
231class VRoot
232{
233public:
234 virtual ~VRoot() noexcept = default;
235};
236
237class VHostMemory : public VRoot
238{
239public:
240 virtual void* data() const noexcept = 0;
241 virtual std::size_t size() const noexcept = 0;
242 virtual DataType type() const noexcept = 0;
243};
244
245class VDimensionExpr : public VRoot
246{
247public:
248 virtual bool isConstant() const = 0;
249 virtual int64_t getConstantValue() const = 0;
250 virtual bool isSizeTensor() const = 0;
251};
252
253class VExprBuilder : public VRoot
254{
255public:
256 virtual IDimensionExpr const* constant(int64_t value) = 0;
257 virtual IDimensionExpr const* operation(
258 DimensionOperation op, IDimensionExpr const& first, IDimensionExpr const& second)
259 = 0;
260 virtual IDimensionExpr const* declareSizeTensor(
261 int32_t outputIndex, IDimensionExpr const& opt, IDimensionExpr const& upper)
262 = 0;
263};
264
265class VRuntime : public VRoot
266{
267public:
268 virtual IRuntime* getPImpl() noexcept = 0;
269 virtual nvinfer1::ICudaEngine* deserializeCudaEngine(void const* blob, std::size_t size) noexcept = 0;
270 virtual nvinfer1::ICudaEngine* deserializeCudaEngine(IStreamReader& streamReader) noexcept = 0;
271 virtual void setDLACore(int32_t dlaCore) noexcept = 0;
272 virtual int32_t getDLACore() const noexcept = 0;
273 virtual int32_t getNbDLACores() const noexcept = 0;
274 virtual void setGpuAllocator(IGpuAllocator* allocator) noexcept = 0;
275 virtual void setErrorRecorder(IErrorRecorder* recorder) noexcept = 0;
276 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
277 virtual ILogger* getLogger() const noexcept = 0;
278 virtual bool setMaxThreads(int32_t maxThreads) noexcept = 0;
279 virtual int32_t getMaxThreads() const noexcept = 0;
280 virtual void setTemporaryDirectory(char const*) noexcept = 0;
281 virtual char const* getTemporaryDirectory() const noexcept = 0;
282 virtual void setTempfileControlFlags(TempfileControlFlags) noexcept = 0;
283 virtual TempfileControlFlags getTempfileControlFlags() const noexcept = 0;
284 virtual IPluginRegistry& getPluginRegistry() noexcept = 0;
285 virtual void setPluginRegistryParent(IPluginRegistry* parent) noexcept = 0;
286 virtual IRuntime* loadRuntime(char const* path) noexcept = 0;
287 virtual void setEngineHostCodeAllowed(bool allowed) noexcept = 0;
288 virtual bool getEngineHostCodeAllowed() const noexcept = 0;
289 // Added in TensorRT version 10.7
290 virtual nvinfer1::ICudaEngine* deserializeCudaEngineV2(IStreamReaderV2& streamReader) noexcept = 0;
291 virtual int64_t getEngineHeaderSize() const noexcept = 0;
292 virtual EngineValidity getEngineValidity(void const* blob, int64_t blobSize, uint64_t* diagnostics) const noexcept = 0;
293};
294
295class VRefitter : public VRoot
296{
297public:
298 virtual IRefitter* getPImpl() noexcept = 0;
299 virtual bool setWeights(char const* layerName, WeightsRole role, const Weights weights) noexcept = 0;
300 virtual bool refitCudaEngine() noexcept = 0;
301 virtual int32_t getMissing(int32_t size, char const** layerNames, WeightsRole* roles) noexcept = 0;
302 virtual int32_t getAll(int32_t size, char const** layerNames, WeightsRole* roles) noexcept = 0;
303 virtual bool setDynamicRange(char const* tensorName, float min, float max) noexcept = 0;
304 virtual float getDynamicRangeMin(char const* tensorName) const noexcept = 0;
305 virtual float getDynamicRangeMax(char const* tensorName) const noexcept = 0;
306 virtual int32_t getTensorsWithDynamicRange(int32_t size, char const** tensorNames) const noexcept = 0;
307 virtual void setErrorRecorder(IErrorRecorder* recorder) noexcept = 0;
308 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
309 virtual bool setNamedWeights(char const* name, Weights weights) noexcept = 0;
310 virtual int32_t getMissingWeights(int32_t size, char const** weightsNames) noexcept = 0;
311 virtual int32_t getAllWeights(int32_t size, char const** weightsNames) noexcept = 0;
312 virtual ILogger* getLogger() const noexcept = 0;
313 virtual bool setMaxThreads(int32_t maxThreads) noexcept = 0;
314 virtual int32_t getMaxThreads() const noexcept = 0;
315 virtual bool setNamedWeightsWithLocation(char const* name, Weights weights, TensorLocation location) noexcept = 0;
316 virtual Weights getNamedWeights(char const* weightsName) const noexcept = 0;
317 virtual TensorLocation getWeightsLocation(char const* weightsName) const noexcept = 0;
318 virtual bool unsetNamedWeights(char const* weightsName) noexcept = 0;
319 virtual void setWeightsValidation(bool weightsValidation) noexcept = 0;
320 virtual bool getWeightsValidation() const noexcept = 0;
321 virtual bool refitCudaEngineAsync(cudaStream_t stream) noexcept = 0;
322 virtual Weights getWeightsPrototype(char const* weightsName) const noexcept = 0;
323};
324
325class VOptimizationProfile : public VRoot
326{
327public:
328 virtual bool setDimensions(char const* inputName, OptProfileSelector select, Dims const& dims) noexcept = 0;
329 virtual Dims getDimensions(char const* inputName, OptProfileSelector select) const noexcept = 0;
330 virtual bool setShapeValues(
331 char const* inputName, OptProfileSelector select, int32_t const* values, int32_t nbValues) noexcept = 0;
332 virtual int32_t getNbShapeValues(char const* inputName) const noexcept = 0;
333 virtual int32_t const* getShapeValues(char const* inputName, OptProfileSelector select) const noexcept = 0;
334 virtual bool setExtraMemoryTarget(float target) noexcept = 0;
335 virtual float getExtraMemoryTarget() const noexcept = 0;
336 virtual bool isValid() const noexcept = 0;
337 // Added in TensorRT 10.11
338 TRT_NODISCARD virtual bool setShapeValuesV2(
339 char const* inputName, OptProfileSelector select, int64_t const* values, int32_t nbValues) noexcept = 0;
340 TRT_NODISCARD virtual int64_t const* getShapeValuesV2(
341 char const* inputName, OptProfileSelector select) const noexcept = 0;
342};
343
344class VCudaEngine : public VRoot
345{
346public:
347 virtual ICudaEngine* getPImpl() noexcept = 0;
348 virtual int32_t getNbLayers() const noexcept = 0;
349 virtual IHostMemory* serialize() const noexcept = 0;
350 virtual IExecutionContext* createExecutionContext(ExecutionContextAllocationStrategy strategy) noexcept = 0;
351 virtual bool isRefittable() const noexcept = 0;
352 virtual char const* getName() const noexcept = 0;
353 virtual int32_t getNbOptimizationProfiles() const noexcept = 0;
354 virtual int32_t const* getProfileTensorValues(
355 char const* tensorName, int32_t profileIndex, OptProfileSelector select) const noexcept = 0;
356 virtual EngineCapability getEngineCapability() const noexcept = 0;
357 virtual void setErrorRecorder(IErrorRecorder* recorder) noexcept = 0;
358 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
359 virtual bool hasImplicitBatchDimension() const noexcept = 0;
360 virtual TacticSources getTacticSources() const noexcept = 0;
361 virtual ProfilingVerbosity getProfilingVerbosity() const noexcept = 0;
362 virtual IEngineInspector* createEngineInspector() const noexcept = 0;
363 virtual Dims getTensorShape(char const* tensorName) const noexcept = 0;
364 virtual DataType getTensorDataType(char const* tensorName) const noexcept = 0;
365 virtual TensorLocation getTensorLocation(char const* tensorName) const noexcept = 0;
366 virtual bool isShapeInferenceIO(char const* tensorName) const noexcept = 0;
367 virtual TensorIOMode getTensorIOMode(char const* tensorName) const noexcept = 0;
368 virtual int32_t getTensorBytesPerComponent(char const* tensorName) const noexcept = 0;
369 virtual int32_t getTensorComponentsPerElement(char const* tensorName) const noexcept = 0;
370 virtual TensorFormat getTensorFormat(char const* tensorName) const noexcept = 0;
371 virtual char const* getTensorFormatDesc(char const* tensorName) const noexcept = 0;
372 virtual int32_t getTensorVectorizedDim(char const* tensorName) const noexcept = 0;
373 virtual Dims getProfileShape(
374 char const* tensorName, int32_t profileIndex, OptProfileSelector select) const noexcept = 0;
375 virtual int32_t getNbIOTensors() const noexcept = 0;
376 virtual char const* getIOTensorName(int32_t index) const noexcept = 0;
377 virtual HardwareCompatibilityLevel getHardwareCompatibilityLevel() const noexcept = 0;
378 virtual int32_t getNbAuxStreams() const noexcept = 0;
379
380 virtual int32_t getTensorBytesPerComponentV2(char const* tensorName, int32_t profileIndex) const noexcept = 0;
381 virtual int32_t getTensorComponentsPerElementV2(char const* tensorName, int32_t profileIndex) const noexcept = 0;
382 virtual TensorFormat getTensorFormatV2(char const* tensorName, int32_t profileIndex) const noexcept = 0;
383 virtual char const* getTensorFormatDescV2(char const* tensorName, int32_t profileIndex) const noexcept = 0;
384 virtual int32_t getTensorVectorizedDimV2(char const* tensorName, int32_t profileIndex) const noexcept = 0;
385
386 virtual ISerializationConfig* createSerializationConfig() noexcept = 0;
387 virtual IHostMemory* serializeWithConfig(ISerializationConfig& config) const noexcept = 0;
388
389 virtual size_t getDeviceMemorySizeForProfile(int32_t profileIndex) const noexcept = 0;
390 virtual IRefitter* createRefitter(ILogger& logger) noexcept = 0;
391
392 virtual bool setWeightStreamingBudget(int64_t gpuMemoryBudget) noexcept = 0;
393 virtual int64_t getWeightStreamingBudget() const noexcept = 0;
394 virtual int64_t getMinimumWeightStreamingBudget() const noexcept = 0;
395 virtual int64_t getStreamableWeightsSize() const noexcept = 0;
396
397 virtual bool isDebugTensor(char const* name) const noexcept = 0;
398
399 // Added in TensorRT 10.1
400 virtual bool setWeightStreamingBudgetV2(int64_t gpuMemoryBudget) noexcept = 0;
401 virtual int64_t getWeightStreamingBudgetV2() const noexcept = 0;
402 virtual int64_t getWeightStreamingAutomaticBudget() const noexcept = 0;
403 virtual int64_t getWeightStreamingScratchMemorySize() const noexcept = 0;
404 virtual int64_t getDeviceMemorySizeV2() const noexcept = 0;
405 virtual int64_t getDeviceMemorySizeForProfileV2(int32_t profileIndex) const noexcept = 0;
406 // Added in TensorRT 10.11
407 TRT_NODISCARD virtual int64_t const* getProfileTensorValuesV2(
408 char const* tensorName, int32_t profileIndex, OptProfileSelector select) const noexcept = 0;
409 TRT_NODISCARD virtual IExecutionContext* createExecutionContextWithRuntimeConfig(
410 IRuntimeConfig* runtimeConfig) noexcept = 0;
411 TRT_NODISCARD virtual IRuntimeConfig* createRuntimeConfig() noexcept = 0;
412 TRT_NODISCARD virtual int64_t getEngineStat(EngineStat stat) const noexcept = 0;
413};
414
415class VExecutionContext : public VRoot
416{
417public:
418 virtual IExecutionContext* getPImpl() noexcept = 0;
419 virtual void setDebugSync(bool sync) noexcept = 0;
420 virtual bool getDebugSync() const noexcept = 0;
421 virtual void setProfiler(IProfiler*) noexcept = 0;
422 virtual IProfiler* getProfiler() const noexcept = 0;
423 virtual ICudaEngine const& getEngine() const noexcept = 0;
424 virtual void setName(char const* name) noexcept = 0;
425 virtual char const* getName() const noexcept = 0;
426 virtual void setDeviceMemory(void* memory) noexcept = 0;
427 virtual int32_t getOptimizationProfile() const noexcept = 0;
428 virtual bool allInputDimensionsSpecified() const noexcept = 0;
429 virtual bool allInputShapesSpecified() const noexcept = 0;
430 virtual void setErrorRecorder(IErrorRecorder* recorder) noexcept = 0;
431 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
432 virtual bool executeV2(void* const* bindings) noexcept = 0;
433 virtual bool setOptimizationProfileAsync(int32_t profileIndex, cudaStream_t stream) noexcept = 0;
434 virtual void setEnqueueEmitsProfile(bool enqueueEmitsProfile) noexcept = 0;
435 virtual bool getEnqueueEmitsProfile() const noexcept = 0;
436 virtual bool reportToProfiler() const noexcept = 0;
437 virtual bool setInputShape(char const* tensorName, Dims const& dims) noexcept = 0;
438 virtual Dims getTensorShape(char const* tensorName) const noexcept = 0;
439 virtual Dims getTensorStrides(char const* tensorName) const noexcept = 0;
440 virtual bool setTensorAddress(char const* tensorName, void* data) noexcept = 0;
441 virtual void const* getTensorAddress(char const* tensorName) const noexcept = 0;
442 virtual bool setInputTensorAddress(char const* tensorName, void const* data) noexcept = 0;
443 virtual bool setOutputTensorAddress(char const* tensorName, void* data) noexcept = 0;
444 virtual int32_t inferShapes(int32_t nbMaxNames, char const** tensorNames) noexcept = 0;
445 virtual bool setInputConsumedEvent(cudaEvent_t event) noexcept = 0;
446 virtual cudaEvent_t getInputConsumedEvent() const noexcept = 0;
447 virtual void* getOutputTensorAddress(char const* tensorName) const noexcept = 0;
448 virtual bool setOutputAllocator(char const* tensorName, IOutputAllocator* outputAllocator) noexcept = 0;
449 virtual IOutputAllocator* getOutputAllocator(char const* name) noexcept = 0;
450 virtual int64_t getMaxOutputSize(char const* tensorName) const noexcept = 0;
451 virtual bool setTemporaryStorageAllocator(IGpuAllocator* allocator) noexcept = 0;
452 virtual IGpuAllocator* getTemporaryStorageAllocator() const noexcept = 0;
453 virtual bool enqueueV3(cudaStream_t stream) noexcept = 0;
454 virtual void setPersistentCacheLimit(size_t size) noexcept = 0;
455 virtual size_t getPersistentCacheLimit() const noexcept = 0;
456 virtual bool setNvtxVerbosity(ProfilingVerbosity verbosity) noexcept = 0;
457 virtual ProfilingVerbosity getNvtxVerbosity() const noexcept = 0;
458 virtual void setAuxStreams(cudaStream_t* auxStreams, int32_t nbStreams) noexcept = 0;
459 virtual bool setDebugListener(IDebugListener* listener) noexcept = 0;
460 virtual IDebugListener* getDebugListener() noexcept = 0;
461 virtual bool setTensorDebugState(char const* name, bool flag) noexcept = 0;
462 virtual bool getDebugState(char const* name) const noexcept = 0;
463 virtual bool setAllTensorsDebugState(bool flag) noexcept = 0;
464 virtual size_t updateDeviceMemorySizeForShapes() noexcept = 0;
465 virtual void setDeviceMemoryV2(void* memory, int64_t size) noexcept = 0;
466 TRT_NODISCARD virtual IRuntimeConfig* getRuntimeConfig() const noexcept = 0;
467 virtual bool setUnfusedTensorsDebugState(bool flag) noexcept = 0;
468 virtual bool getUnfusedTensorsDebugState() const noexcept = 0;
469#if ENABLE_FEATURE_DISABLE_RUNTIME_ALLOCATION
470 virtual bool isStreamCapturable(cudaStream_t stream) const noexcept = 0;
471#endif // ENABLE_FEATURE_DISABLE_RUNTIME_ALLOCATION
472};
473
474class VEngineInspector : public VRoot
475{
476public:
477 virtual IEngineInspector* getPImpl() noexcept = 0;
478 virtual bool setExecutionContext(IExecutionContext const* context) noexcept = 0;
479 virtual IExecutionContext const* getExecutionContext() const noexcept = 0;
480 virtual char const* getLayerInformation(int32_t layerIndex, LayerInformationFormat format) const noexcept = 0;
481 virtual char const* getEngineInformation(LayerInformationFormat format) const noexcept = 0;
482 virtual void setErrorRecorder(IErrorRecorder* recorder) noexcept = 0;
483 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
484};
485
486class VTensor : public VRoot
487{
488public:
489 virtual void setName(char const* name) noexcept = 0;
490 virtual char const* getName() const noexcept = 0;
491 virtual void setDimensions(Dims const& dimensions) noexcept = 0;
492 virtual Dims getDimensions() const noexcept = 0;
493 virtual void setType(DataType type) noexcept = 0;
494 virtual DataType getType() const noexcept = 0;
495 virtual bool isNetworkInput() const noexcept = 0;
496 virtual bool isNetworkOutput() const noexcept = 0;
497 virtual void setAllowedFormats(TensorFormats formats) noexcept = 0;
498 virtual TensorFormats getAllowedFormats() const noexcept = 0;
499 virtual bool isShapeTensor() const noexcept = 0;
500 virtual bool isExecutionTensor() const noexcept = 0;
501 virtual void setDimensionName(int32_t index, char const* name) noexcept = 0;
502 virtual char const* getDimensionName(int32_t index) const noexcept = 0;
503};
504
505class VLayer : public VRoot
506{
507public:
508 virtual LayerType getType() const noexcept = 0;
509 virtual void setName(char const* name) noexcept = 0;
510 virtual char const* getName() const noexcept = 0;
511 virtual int32_t getNbInputs() const noexcept = 0;
512 virtual ITensor* getInput(int32_t index) const noexcept = 0;
513 virtual int32_t getNbOutputs() const noexcept = 0;
514 virtual ITensor* getOutput(int32_t index) const noexcept = 0;
515 virtual void setInput(int32_t index, ITensor& tensor) noexcept = 0;
516 virtual void setPrecision(DataType dataType) noexcept = 0;
517 virtual DataType getPrecision() const noexcept = 0;
518 virtual bool precisionIsSet() const noexcept = 0;
519 virtual void resetPrecision() noexcept = 0;
520 virtual void setOutputType(int32_t index, DataType dataType) noexcept = 0;
521 virtual DataType getOutputType(int32_t index) const noexcept = 0;
522 virtual bool outputTypeIsSet(int32_t index) const noexcept = 0;
523 virtual void resetOutputType(int32_t index) noexcept = 0;
524 virtual void setMetadata(char const* docString) noexcept = 0;
525 virtual char const* getMetadata() const noexcept = 0;
526};
527
528class VConvolutionLayer : public VRoot
529{
530public:
531 virtual void setNbOutputMaps(int64_t nbOutputMaps) noexcept = 0;
532 virtual int64_t getNbOutputMaps() const noexcept = 0;
533 virtual void setNbGroups(int64_t nbGroups) noexcept = 0;
534 virtual int64_t getNbGroups() const noexcept = 0;
535 virtual void setKernelWeights(Weights weights) noexcept = 0;
536 virtual Weights getKernelWeights() const noexcept = 0;
537 virtual void setBiasWeights(Weights weights) noexcept = 0;
538 virtual Weights getBiasWeights() const noexcept = 0;
539 virtual void setPrePadding(Dims const& padding) noexcept = 0;
540 virtual Dims getPrePadding() const noexcept = 0;
541 virtual void setPostPadding(Dims const& padding) noexcept = 0;
542 virtual Dims getPostPadding() const noexcept = 0;
543 virtual void setPaddingMode(PaddingMode paddingMode) noexcept = 0;
544 virtual PaddingMode getPaddingMode() const noexcept = 0;
545 virtual void setKernelSizeNd(Dims const& kernelSize) noexcept = 0;
546 virtual Dims getKernelSizeNd() const noexcept = 0;
547 virtual void setStrideNd(Dims const& stride) noexcept = 0;
548 virtual Dims getStrideNd() const noexcept = 0;
549 virtual void setPaddingNd(Dims const& padding) noexcept = 0;
550 virtual Dims getPaddingNd() const noexcept = 0;
551 virtual void setDilationNd(Dims const& dilation) noexcept = 0;
552 virtual Dims getDilationNd() const noexcept = 0;
553};
554
555class VActivationLayer : public VRoot
556{
557public:
558 virtual void setActivationType(ActivationType type) noexcept = 0;
559 virtual ActivationType getActivationType() const noexcept = 0;
560 virtual void setAlpha(float alpha) noexcept = 0;
561 virtual void setBeta(float beta) noexcept = 0;
562 virtual float getAlpha() const noexcept = 0;
563 virtual float getBeta() const noexcept = 0;
564};
565
566class VPoolingLayer : public VRoot
567{
568public:
569 virtual void setPoolingType(PoolingType type) noexcept = 0;
570 virtual PoolingType getPoolingType() const noexcept = 0;
571 virtual void setBlendFactor(float blendFactor) noexcept = 0;
572 virtual float getBlendFactor() const noexcept = 0;
573 virtual void setAverageCountExcludesPadding(bool exclusive) noexcept = 0;
574 virtual bool getAverageCountExcludesPadding() const noexcept = 0;
575 virtual void setPrePadding(Dims const& padding) noexcept = 0;
576 virtual Dims getPrePadding() const noexcept = 0;
577 virtual void setPostPadding(Dims const& padding) noexcept = 0;
578 virtual Dims getPostPadding() const noexcept = 0;
579 virtual void setPaddingMode(PaddingMode paddingMode) noexcept = 0;
580 virtual PaddingMode getPaddingMode() const noexcept = 0;
581 virtual void setWindowSizeNd(Dims const& windowSize) noexcept = 0;
582 virtual Dims getWindowSizeNd() const noexcept = 0;
583 virtual void setStrideNd(Dims const& stride) noexcept = 0;
584 virtual Dims getStrideNd() const noexcept = 0;
585 virtual void setPaddingNd(Dims const& padding) noexcept = 0;
586 virtual Dims getPaddingNd() const noexcept = 0;
587};
588
589class VLRNLayer : public VRoot
590{
591public:
592 virtual void setWindowSize(int64_t windowSize) noexcept = 0;
593 virtual int64_t getWindowSize() const noexcept = 0;
594 virtual void setAlpha(float alpha) noexcept = 0;
595 virtual float getAlpha() const noexcept = 0;
596 virtual void setBeta(float beta) noexcept = 0;
597 virtual float getBeta() const noexcept = 0;
598 virtual void setK(float k) noexcept = 0;
599 virtual float getK() const noexcept = 0;
600};
601
602class VScaleLayer : public VRoot
603{
604public:
605 virtual void setMode(ScaleMode mode) noexcept = 0;
606 virtual ScaleMode getMode() const noexcept = 0;
607 virtual void setShift(Weights shift) noexcept = 0;
608 virtual Weights getShift() const noexcept = 0;
609 virtual void setScale(Weights scale) noexcept = 0;
610 virtual Weights getScale() const noexcept = 0;
611 virtual void setPower(Weights power) noexcept = 0;
612 virtual Weights getPower() const noexcept = 0;
613 virtual int32_t getChannelAxis() const noexcept = 0;
614 virtual void setChannelAxis(int32_t channelAxis) noexcept = 0;
615};
616
617class VSoftMaxLayer : public VRoot
618{
619public:
620 virtual void setAxes(uint32_t axes) noexcept = 0;
621 virtual uint32_t getAxes() const noexcept = 0;
622};
623
624class VConcatenationLayer : public VRoot
625{
626public:
627 virtual void setAxis(int32_t axis) noexcept = 0;
628 virtual int32_t getAxis() const noexcept = 0;
629};
630
631class VDeconvolutionLayer : public VRoot
632{
633public:
634 virtual void setNbOutputMaps(int64_t nbOutputMaps) noexcept = 0;
635 virtual int64_t getNbOutputMaps() const noexcept = 0;
636 virtual void setNbGroups(int64_t nbGroups) noexcept = 0;
637 virtual int64_t getNbGroups() const noexcept = 0;
638 virtual void setKernelWeights(Weights weights) noexcept = 0;
639 virtual Weights getKernelWeights() const noexcept = 0;
640 virtual void setBiasWeights(Weights weights) noexcept = 0;
641 virtual Weights getBiasWeights() const noexcept = 0;
642 virtual void setPrePadding(Dims const& padding) noexcept = 0;
643 virtual Dims getPrePadding() const noexcept = 0;
644 virtual void setPostPadding(Dims const& padding) noexcept = 0;
645 virtual Dims getPostPadding() const noexcept = 0;
646 virtual void setPaddingMode(PaddingMode paddingMode) noexcept = 0;
647 virtual PaddingMode getPaddingMode() const noexcept = 0;
648 virtual void setKernelSizeNd(Dims const& kernelSize) noexcept = 0;
649 virtual Dims getKernelSizeNd() const noexcept = 0;
650 virtual void setStrideNd(Dims const& stride) noexcept = 0;
651 virtual Dims getStrideNd() const noexcept = 0;
652 virtual void setPaddingNd(Dims const& padding) noexcept = 0;
653 virtual Dims getPaddingNd() const noexcept = 0;
654 virtual void setDilationNd(Dims const& dilation) noexcept = 0;
655 virtual Dims getDilationNd() const noexcept = 0;
656};
657
658class VElementWiseLayer : public VRoot
659{
660public:
661 virtual void setOperation(ElementWiseOperation op) noexcept = 0;
662 virtual ElementWiseOperation getOperation() const noexcept = 0;
663};
664
665class VGatherLayer : public VRoot
666{
667public:
668 virtual void setGatherAxis(int32_t axis) noexcept = 0;
669 virtual int32_t getGatherAxis() const noexcept = 0;
670 virtual void setNbElementWiseDims(int32_t k) noexcept = 0;
671 virtual int32_t getNbElementWiseDims() const noexcept = 0;
672 virtual void setMode(GatherMode mode) noexcept = 0;
673 virtual GatherMode getMode() const noexcept = 0;
674};
675
676class VPluginLayer : public VRoot
677{
678public:
679 virtual IPlugin& getPlugin() noexcept = 0;
680};
681
682class VPluginV2Layer : public VRoot
683{
684public:
685 virtual IPluginV2& getPlugin() noexcept = 0;
686};
687
688class VPluginV3Layer : public VRoot
689{
690public:
691 virtual IPluginV3& getPlugin() noexcept = 0;
692};
693
694class VUnaryLayer : public VRoot
695{
696public:
697 virtual void setOperation(UnaryOperation op) noexcept = 0;
698 virtual UnaryOperation getOperation() const noexcept = 0;
699};
700
701class VReduceLayer : public VRoot
702{
703public:
704 virtual void setOperation(ReduceOperation op) noexcept = 0;
705 virtual ReduceOperation getOperation() const noexcept = 0;
706 virtual void setReduceAxes(uint32_t reduceAxes) noexcept = 0;
707 virtual uint32_t getReduceAxes() const noexcept = 0;
708 virtual void setKeepDimensions(bool keepDimensions) noexcept = 0;
709 virtual bool getKeepDimensions() const noexcept = 0;
710};
711
712class VPaddingLayer : public VRoot
713{
714public:
715 virtual void setPrePaddingNd(Dims const& padding) noexcept = 0;
716 virtual Dims getPrePaddingNd() const noexcept = 0;
717 virtual void setPostPaddingNd(Dims const& padding) noexcept = 0;
718 virtual Dims getPostPaddingNd() const noexcept = 0;
719};
720
721class VShuffleLayer : public VRoot
722{
723public:
724 virtual void setFirstTranspose(Permutation const& permutation) noexcept = 0;
725 virtual Permutation const& getFirstTranspose() const noexcept = 0;
726 virtual void setReshapeDimensions(Dims const& dimensions) noexcept = 0;
727 virtual Dims getReshapeDimensions() const noexcept = 0;
728 virtual void setSecondTranspose(Permutation const& permutation) noexcept = 0;
729 virtual Permutation const& getSecondTranspose() const noexcept = 0;
730 virtual void setZeroIsPlaceholder(bool zeroIsPlaceholder) noexcept = 0;
731 virtual bool getZeroIsPlaceholder() const noexcept = 0;
732};
733
734class VSliceLayer : public VRoot
735{
736public:
737 virtual void setStart(Dims const& start) noexcept = 0;
738 virtual Dims getStart() const noexcept = 0;
739 virtual void setSize(Dims const& size) noexcept = 0;
740 virtual Dims getSize() const noexcept = 0;
741 virtual void setStride(Dims const& stride) noexcept = 0;
742 virtual Dims getStride() const noexcept = 0;
743 virtual void setMode(SampleMode mode) noexcept = 0;
744 virtual SampleMode getMode() const noexcept = 0;
745 virtual void setAxes(Dims const& axes) noexcept = 0;
746 virtual Dims getAxes() const noexcept = 0;
747};
748
749class VShapeLayer : public VRoot
750{
751public:
752};
753
754class VTopKLayer : public VRoot
755{
756public:
757 virtual void setOperation(TopKOperation op) noexcept = 0;
758 virtual TopKOperation getOperation() const noexcept = 0;
759 virtual void setK(int32_t k) noexcept = 0;
760 virtual int32_t getK() const noexcept = 0;
761 virtual void setReduceAxes(uint32_t reduceAxes) noexcept = 0;
762 virtual uint32_t getReduceAxes() const noexcept = 0;
763 virtual bool setIndicesType(DataType type) noexcept = 0;
764 virtual DataType getIndicesType() const noexcept = 0;
765};
766
767class VMatrixMultiplyLayer : public VRoot
768{
769public:
770 virtual void setOperation(int32_t index, MatrixOperation op) noexcept = 0;
771 virtual MatrixOperation getOperation(int32_t index) const noexcept = 0;
772};
773
774class VNonZeroLayer : public VRoot
775{
776public:
777 virtual bool setIndicesType(DataType type) noexcept = 0;
778 virtual DataType getIndicesType() const noexcept = 0;
779};
780
781class VRaggedSoftMaxLayer : public VRoot
782{
783public:
784};
785
786class VIdentityLayer : public VRoot
787{
788public:
789};
790
791class VCastLayer : public VRoot
792{
793public:
794 virtual void setToType(DataType toType) noexcept = 0;
795 virtual DataType getToType() const noexcept = 0;
796};
797
798class VConstantLayer : public VRoot
799{
800public:
801 virtual void setWeights(Weights weights) noexcept = 0;
802 virtual Weights getWeights() const noexcept = 0;
803 virtual void setDimensions(Dims const& dimensions) noexcept = 0;
804 virtual Dims getDimensions() const noexcept = 0;
805};
806
807class VParametricReLULayer : public VRoot
808{
809public:
810};
811
812class VResizeLayer : public VRoot
813{
814public:
815 virtual void setOutputDimensions(Dims const& dimensions) noexcept = 0;
816 virtual Dims getOutputDimensions() const noexcept = 0;
817 virtual void setScales(float const* scales, int32_t nbScales) noexcept = 0;
818 virtual int32_t getScales(int32_t size, float* scales) const noexcept = 0;
819 virtual void setResizeMode(InterpolationMode interpolationMode) noexcept = 0;
820 virtual InterpolationMode getResizeMode() const noexcept = 0;
821 virtual void setCoordinateTransformation(ResizeCoordinateTransformation coordTransform) noexcept = 0;
822 virtual ResizeCoordinateTransformation getCoordinateTransformation() const noexcept = 0;
823 virtual void setSelectorForSinglePixel(ResizeSelector selector) noexcept = 0;
824 virtual ResizeSelector getSelectorForSinglePixel() const noexcept = 0;
825 virtual void setNearestRounding(ResizeRoundMode value) noexcept = 0;
826 virtual ResizeRoundMode getNearestRounding() const noexcept = 0;
827 virtual void setCubicCoeff(float value) noexcept = 0;
828 virtual float getCubicCoeff() const noexcept = 0;
829 virtual void setExcludeOutside(bool value) noexcept = 0;
830 virtual bool getExcludeOutside() const noexcept = 0;
831};
832
833class VLoopBoundaryLayer : public VRoot
834{
835public:
836 virtual ILoop* getLoop() const noexcept = 0;
837};
838
839class VRecurrenceLayer : public VRoot
840{
841public:
842};
843
844class VLoopOutputLayer : public VRoot
845{
846public:
847 virtual LoopOutput getLoopOutput() const noexcept = 0;
848 virtual void setAxis(int32_t axis) noexcept = 0;
849 virtual int32_t getAxis() const noexcept = 0;
850};
851
852class VTripLimitLayer : public VRoot
853{
854public:
855 virtual TripLimit getTripLimit() const noexcept = 0;
856};
857
858class VIteratorLayer : public VRoot
859{
860public:
861 virtual void setAxis(int32_t axis) noexcept = 0;
862 virtual int32_t getAxis() const noexcept = 0;
863 virtual void setReverse(bool reverse) noexcept = 0;
864 virtual bool getReverse() const noexcept = 0;
865};
866class VLoop : public VRoot
867{
868public:
869 virtual IRecurrenceLayer* addRecurrence(ITensor& initialValue) noexcept = 0;
870 virtual ITripLimitLayer* addTripLimit(ITensor& tensor, TripLimit limit) noexcept = 0;
871 virtual IIteratorLayer* addIterator(ITensor& tensor, int32_t axis = 0, bool reverse = false) noexcept = 0;
872 virtual ILoopOutputLayer* addLoopOutput(ITensor& tensor, LoopOutput outputKind, int32_t axis = 0) noexcept = 0;
873 virtual void setName(char const* name) noexcept = 0;
874 virtual char const* getName() const noexcept = 0;
875};
876
877class VConditionalBoundaryLayer : public VRoot
878{
879public:
880 virtual IIfConditional* getConditional() const noexcept = 0;
881};
882
883class VConditionLayer : public VRoot
884{
885public:
886};
887
888class VConditionalInputLayer : public VRoot
889{
890public:
891};
892
893class VConditionalOutputLayer : public VRoot
894{
895public:
896};
897
898class VIfConditional : public VRoot
899{
900public:
901 virtual IConditionLayer* setCondition(ITensor& tensor) noexcept = 0;
902 virtual IIfConditionalInputLayer* addInput(ITensor& tensor) noexcept = 0;
903 virtual IIfConditionalOutputLayer* addOutput(ITensor& trueTensor, ITensor& falseTensor) noexcept = 0;
904 virtual void setName(char const* name) noexcept = 0;
905 virtual char const* getName() const noexcept = 0;
906};
907
908class VAttentionBoundaryLayer : public VRoot
909{
910public:
911 virtual IAttention* getAttention() const noexcept = 0;
912};
913
914class VAttentionInputLayer : public VRoot
915{
916public:
917};
918
919class VAttentionOutputLayer : public VRoot
920{
921public:
922};
923
924class VAttention : public VRoot
925{
926public:
927 TRT_NODISCARD virtual bool setInput(int32_t index, ITensor& input) noexcept = 0;
928 TRT_NODISCARD virtual int32_t getNbInputs() const noexcept = 0;
929 TRT_NODISCARD virtual ITensor* getInput(int32_t index) const noexcept = 0;
930 TRT_NODISCARD virtual int32_t getNbOutputs() const noexcept = 0;
931 TRT_NODISCARD virtual ITensor* getOutput(int32_t index) const noexcept = 0;
932 TRT_NODISCARD virtual bool setName(char const* name) noexcept = 0;
933 TRT_NODISCARD virtual char const* getName() const noexcept = 0;
934 TRT_NODISCARD virtual bool setNormalizationOperation(AttentionNormalizationOp op) noexcept = 0;
935 TRT_NODISCARD virtual AttentionNormalizationOp getNormalizationOperation() const noexcept = 0;
936 TRT_NODISCARD virtual bool setCausal(bool isCausal) noexcept = 0;
937 TRT_NODISCARD virtual bool getCausal() const noexcept = 0;
938 TRT_NODISCARD virtual bool setMask(ITensor& mask) noexcept = 0;
939 TRT_NODISCARD virtual ITensor* getMask() const noexcept = 0;
940 TRT_NODISCARD virtual bool setDecomposable(bool decomposable) noexcept = 0;
941 TRT_NODISCARD virtual bool getDecomposable() const noexcept = 0;
942 TRT_NODISCARD virtual bool setNormalizationQuantizeScale(ITensor& tensor) noexcept = 0;
943 TRT_NODISCARD virtual ITensor* getNormalizationQuantizeScale() const noexcept = 0;
944 TRT_NODISCARD virtual bool setNormalizationQuantizeToType(DataType type) noexcept = 0;
945 TRT_NODISCARD virtual DataType getNormalizationQuantizeToType() const noexcept = 0;
946}; // class VAttention
947
948class VSelectLayer : public VRoot
949{
950};
951
952class VAssertionLayer : public VRoot
953{
954public:
955 virtual void setMessage(char const* message) noexcept = 0;
956 virtual char const* getMessage() const noexcept = 0;
957};
958
959class VFillLayer : public VRoot
960{
961public:
962 virtual void setDimensions(Dims const& dimensions) noexcept = 0;
963 virtual Dims getDimensions() const noexcept = 0;
964 virtual void setOperation(FillOperation op) noexcept = 0;
965 virtual FillOperation getOperation() const noexcept = 0;
966 virtual void setAlpha(double alpha) noexcept = 0;
967 virtual double getAlpha() const noexcept = 0;
968 virtual void setBeta(double beta) noexcept = 0;
969 virtual double getBeta() const noexcept = 0;
970 virtual void setAlphaInt64(int64_t alpha) noexcept = 0;
971 virtual int64_t getAlphaInt64() const noexcept = 0;
972 virtual void setBetaInt64(int64_t beta) noexcept = 0;
973 virtual int64_t getBetaInt64() const noexcept = 0;
974 virtual bool isAlphaBetaInt64() const noexcept = 0;
975 virtual DataType getToType() const noexcept = 0;
976 virtual void setToType(DataType toType) noexcept = 0;
977};
978
979class VQuantizeLayer : public VRoot
980{
981public:
982 virtual int32_t getAxis() const noexcept = 0;
983 virtual void setAxis(int32_t axis) noexcept = 0;
984 virtual DataType getToType() const noexcept = 0;
985 virtual void setToType(DataType toType) noexcept = 0;
986};
987
988class VDequantizeLayer : public VRoot
989{
990public:
991 virtual int32_t getAxis() const noexcept = 0;
992 virtual void setAxis(int32_t axis) noexcept = 0;
993 virtual DataType getToType() const noexcept = 0;
994 virtual void setToType(DataType toType) noexcept = 0;
995};
996
997class VDynamicQuantizeLayer : public VRoot
998{
999public:
1000 virtual int32_t getAxis() const noexcept = 0;
1001 virtual void setAxis(int32_t axis) noexcept = 0;
1002 virtual int32_t getBlockSize() const noexcept = 0;
1003 virtual void setBlockSize(int32_t axis) noexcept = 0;
1004 virtual DataType getScaleType() const noexcept = 0;
1005 virtual void setScaleType(DataType axis) noexcept = 0;
1006 virtual DataType getToType() const noexcept = 0;
1007 virtual void setToType(DataType toType) noexcept = 0;
1008};
1009
1010class VScatterLayer : public VRoot
1011{
1012public:
1013 virtual void setMode(ScatterMode mode) noexcept = 0;
1014 virtual ScatterMode getMode() const noexcept = 0;
1015 virtual void setAxis(int32_t axis) noexcept = 0;
1016 virtual int32_t getAxis() const noexcept = 0;
1017}; // class VScatterLayer
1018
1019class VEinsumLayer : public VRoot
1020{
1021public:
1022 virtual bool setEquation(char const* equation) noexcept = 0;
1023 virtual char const* getEquation() const noexcept = 0;
1024};
1025
1026class VOneHotLayer : public VRoot
1027{
1028public:
1029 virtual int32_t getAxis() const noexcept = 0;
1030 virtual void setAxis(int32_t axis) noexcept = 0;
1031}; // class VOneHotLayer
1032
1033class VGridSampleLayer : public VRoot
1034{
1035public:
1036 virtual void setInterpolationMode(InterpolationMode mode) noexcept = 0;
1037 virtual InterpolationMode getInterpolationMode() const noexcept = 0;
1038 virtual void setAlignCorners(bool alignCorners) noexcept = 0;
1039 virtual bool getAlignCorners() const noexcept = 0;
1040 virtual bool setSampleMode(SampleMode mode) noexcept = 0;
1041 virtual SampleMode getSampleMode() const noexcept = 0;
1042}; // class VGridSampleLayer
1043
1044class VNMSLayer : public VRoot
1045{
1046public:
1047 virtual void setBoundingBoxFormat(BoundingBoxFormat fmt) noexcept = 0;
1048 virtual BoundingBoxFormat getBoundingBoxFormat() const noexcept = 0;
1049 virtual void setTopKBoxLimit(int32_t limit) noexcept = 0;
1050 virtual int32_t getTopKBoxLimit() const noexcept = 0;
1051 virtual bool setIndicesType(DataType type) noexcept = 0;
1052 virtual DataType getIndicesType() const noexcept = 0;
1053}; // class VNMSLayer
1054
1055class VReverseSequenceLayer : public VRoot
1056{
1057public:
1058 virtual void setBatchAxis(int32_t batchAxis) noexcept = 0;
1059 virtual int32_t getBatchAxis() const noexcept = 0;
1060
1061 virtual void setSequenceAxis(int32_t sequenceAxis) noexcept = 0;
1062 virtual int32_t getSequenceAxis() const noexcept = 0;
1063}; // class VReverseSequenceLayer
1064
1065class VNormalizationLayer : public VRoot
1066{
1067public:
1068 virtual void setEpsilon(float eps) noexcept = 0;
1069 virtual float getEpsilon() const noexcept = 0;
1070 virtual void setAxes(uint32_t axesMask) noexcept = 0;
1071 virtual uint32_t getAxes() const noexcept = 0;
1072 virtual void setNbGroups(int64_t nbGroups) noexcept = 0;
1073 virtual int64_t getNbGroups() const noexcept = 0;
1074 virtual void setComputePrecision(DataType type) noexcept = 0;
1075 virtual DataType getComputePrecision() const noexcept = 0;
1076}; // class VNormalizationLayer
1077
1078class VSqueezeLayer : public VRoot
1079{
1080};
1081
1082class VUnsqueezeLayer : public VRoot
1083{
1084};
1085
1086class VCumulativeLayer : public VRoot
1087{
1088public:
1089 virtual bool setOperation(CumulativeOperation op) noexcept = 0;
1090 virtual CumulativeOperation getOperation() const noexcept = 0;
1091 virtual void setExclusive(bool exclusive) noexcept = 0;
1092 virtual bool getExclusive() const noexcept = 0;
1093 virtual void setReverse(bool reverse) noexcept = 0;
1094 virtual bool getReverse() const noexcept = 0;
1095}; // class VCumulativeLayer
1096
1097
1098class VNetworkDefinition : public VRoot
1099{
1100public:
1101 virtual ITensor* addInput(char const* name, DataType type, Dims const& dimensions) noexcept = 0;
1102 virtual void markOutput(ITensor& tensor) noexcept = 0;
1103 virtual IActivationLayer* addActivation(ITensor& input, ActivationType type) noexcept = 0;
1104 virtual ILRNLayer* addLRN(ITensor& input, int64_t window, float alpha, float beta, float k) noexcept = 0;
1105 virtual IScaleLayer* addScale(
1106 ITensor& input, ScaleMode mode, Weights shift, Weights scale, Weights power) noexcept = 0;
1107 virtual ISoftMaxLayer* addSoftMax(ITensor& input) noexcept = 0;
1108 virtual IConcatenationLayer* addConcatenation(ITensor* const* inputs, int32_t nbInputs) noexcept = 0;
1109 virtual IElementWiseLayer* addElementWise(ITensor& input1, ITensor& input2, ElementWiseOperation op) noexcept = 0;
1110 virtual IUnaryLayer* addUnary(ITensor& input, UnaryOperation operation) noexcept = 0;
1111 virtual IShuffleLayer* addShuffle(ITensor& input) noexcept = 0;
1112 virtual int32_t getNbLayers() const noexcept = 0;
1113 virtual ILayer* getLayer(int32_t index) const noexcept = 0;
1114 virtual int32_t getNbInputs() const noexcept = 0;
1115 virtual ITensor* getInput(int32_t index) const noexcept = 0;
1116 virtual int32_t getNbOutputs() const noexcept = 0;
1117 virtual ITensor* getOutput(int32_t index) const noexcept = 0;
1118 virtual IReduceLayer* addReduce(
1119 ITensor& input, ReduceOperation operation, uint32_t reduceAxes, bool keepDimensions) noexcept
1120 = 0;
1121 virtual ITopKLayer* addTopK(ITensor& input, TopKOperation op, int32_t k, uint32_t reduceAxes) noexcept = 0;
1122 virtual IGatherLayer* addGather(ITensor& data, ITensor& indices, int32_t axis) noexcept = 0;
1123 virtual IRaggedSoftMaxLayer* addRaggedSoftMax(ITensor& input, ITensor& bounds) noexcept = 0;
1124 virtual IMatrixMultiplyLayer* addMatrixMultiply(
1125 ITensor& input0, MatrixOperation op0, ITensor& input1, MatrixOperation op1) noexcept = 0;
1126 virtual IConstantLayer* addConstant(Dims const& dimensions, Weights weights) noexcept = 0;
1127 virtual IIdentityLayer* addIdentity(ITensor& input) noexcept = 0;
1128 virtual void removeTensor(ITensor& tensor) noexcept = 0;
1129 virtual void unmarkOutput(ITensor& tensor) noexcept = 0;
1130 virtual ISliceLayer* addSlice(ITensor& input, Dims const& start, Dims const& size, Dims const& stride) noexcept = 0;
1131 virtual void setName(char const* name) noexcept = 0;
1132 virtual char const* getName() const noexcept = 0;
1133 virtual IShapeLayer* addShape(ITensor& input) noexcept = 0;
1134 virtual bool hasImplicitBatchDimension() const noexcept = 0;
1135 virtual bool markOutputForShapes(ITensor& tensor) noexcept = 0;
1136 virtual bool unmarkOutputForShapes(ITensor& tensor) noexcept = 0;
1137 virtual IParametricReLULayer* addParametricReLU(ITensor& input, ITensor& slope) noexcept = 0;
1138 virtual IConvolutionLayer* addConvolutionNd(
1139 ITensor& input, int64_t nbOutputMaps, Dims const& kernelSize, Weights kernelWeights, Weights biasWeights) noexcept
1140 = 0;
1141 virtual IPoolingLayer* addPoolingNd(ITensor& input, PoolingType type, Dims const& windowSize) noexcept = 0;
1142 virtual IDeconvolutionLayer* addDeconvolutionNd(
1143 ITensor& input, int64_t nbOutputMaps, Dims const& kernelSize, Weights kernelWeights, Weights biasWeights) noexcept
1144 = 0;
1145 virtual IScaleLayer* addScaleNd(
1146 ITensor& input, ScaleMode mode, Weights shift, Weights scale, Weights power, int32_t channelAxis) noexcept = 0;
1147 virtual IResizeLayer* addResize(ITensor& input) noexcept = 0;
1148 virtual ILoop* addLoop() noexcept = 0;
1149 virtual ISelectLayer* addSelect(ITensor& condition, ITensor& thenInput, ITensor& elseInput) noexcept = 0;
1150 virtual IFillLayer* addFill(Dims const& dimensions, FillOperation op) noexcept = 0;
1151 virtual IPaddingLayer* addPaddingNd(ITensor& input, Dims const& prePadding, Dims const& postPadding) noexcept = 0;
1152 virtual bool setWeightsName(Weights weights, char const* name) noexcept = 0;
1153 virtual void setErrorRecorder(IErrorRecorder* recorder) noexcept = 0;
1154 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
1155 virtual IGatherLayer* addGatherV2(ITensor& data, ITensor& indices, GatherMode mode) noexcept = 0;
1156 virtual IIfConditional* addIfConditional() noexcept = 0;
1157 virtual IScatterLayer* addScatter(ITensor& data, ITensor& indices, ITensor& updates, ScatterMode mode) noexcept = 0;
1158 virtual IEinsumLayer* addEinsum(ITensor* const* inputs, int32_t nbInputs, char const* equation) noexcept = 0;
1159 virtual IAssertionLayer* addAssertion(ITensor& condition, char const* message) noexcept = 0;
1160 virtual IOneHotLayer* addOneHot(ITensor& indices, ITensor& values, ITensor& depth, int32_t axis) noexcept = 0;
1161 virtual INonZeroLayer* addNonZero(ITensor& input) noexcept = 0;
1162 virtual IGridSampleLayer* addGridSample(ITensor& input, ITensor& grid) noexcept = 0;
1163 virtual INMSLayer* addNMS(ITensor& boxes, ITensor& scores, ITensor& maxOutputBoxesPerClass) noexcept = 0;
1164 virtual IReverseSequenceLayer* addReverseSequence(ITensor& input, ITensor& sequenceLens) noexcept = 0;
1165 virtual INormalizationLayer* addNormalization(
1166 ITensor& input, ITensor& scale, ITensor& bias, uint32_t axesMask) noexcept = 0;
1167 virtual ICastLayer* addCast(ITensor& input, DataType toType) noexcept = 0;
1168 virtual IBuilder& getBuilder() const noexcept = 0;
1169 virtual NetworkDefinitionCreationFlags getFlags() const noexcept = 0;
1170 virtual bool getFlag(NetworkDefinitionCreationFlag networkDefinitionCreationFlag) const noexcept = 0;
1171 virtual IQuantizeLayer* addQuantizeV2(ITensor& input, ITensor& scale, DataType outputType) noexcept = 0;
1172 virtual IDequantizeLayer* addDequantizeV2(ITensor& input, ITensor& scale, DataType outputType) noexcept = 0;
1173 virtual IFillLayer* addFillV2(Dims const& dimensions, FillOperation op, DataType outputType) noexcept = 0;
1174 virtual bool markDebug(ITensor& tensor) noexcept = 0;
1175 virtual bool unmarkDebug(ITensor& tensor) noexcept = 0;
1176 virtual bool isDebugTensor(ITensor const& tensor) const noexcept = 0;
1177 virtual bool markWeightsRefittable(char const* name) noexcept = 0;
1178 virtual bool unmarkWeightsRefittable(char const* name) noexcept = 0;
1179 virtual bool areWeightsMarkedRefittable(char const* name) const noexcept = 0;
1180 virtual ISqueezeLayer* addSqueeze(ITensor& input, ITensor& axes) noexcept = 0;
1181 virtual IUnsqueezeLayer* addUnsqueeze(ITensor& input, ITensor& axes) noexcept = 0;
1182 virtual IDynamicQuantizeLayer* addDynamicQuantize(
1183 ITensor& input, int32_t axis, int32_t blockSize, DataType toType, DataType scaleType) noexcept
1184 = 0;
1185 virtual ICumulativeLayer* addCumulative(
1186 ITensor& input, ITensor& axis, CumulativeOperation operation, bool exclusive, bool reverse) noexcept
1187 = 0;
1188 virtual bool markUnfusedTensorsAsDebugTensors() noexcept = 0;
1189 virtual bool unmarkUnfusedTensorsAsDebugTensors() noexcept = 0;
1190 virtual ITopKLayer* addTopKV2(
1191 ITensor& input, TopKOperation op, int32_t k, uint32_t reduceAxes, DataType indicesType) noexcept
1192 = 0;
1193 virtual INonZeroLayer* addNonZeroV2(ITensor& input, DataType indicesType) noexcept = 0;
1194 virtual INMSLayer* addNMSV2(
1195 ITensor& boxes, ITensor& scores, ITensor& maxOutputBoxesPerClass, DataType indicesType) noexcept
1196 = 0;
1197 virtual IAttention* addAttention(
1198 ITensor& query, ITensor& key, ITensor& value, AttentionNormalizationOp normOp, bool isCausal) noexcept
1199 = 0;
1200};
1201
1202
1203class VTimingCache : public VRoot
1204{
1205public:
1206 virtual nvinfer1::IHostMemory* serialize() const noexcept = 0;
1207 virtual bool combine(ITimingCache const& inputCache, bool ignoreMismatch) noexcept = 0;
1208 virtual bool reset() noexcept = 0;
1209 virtual int64_t queryKeys(TimingCacheKey* keyBuffer, int64_t capacity) const noexcept = 0;
1210 virtual TimingCacheValue query(TimingCacheKey const& key) const noexcept = 0;
1211 virtual bool update(TimingCacheKey const& key, TimingCacheValue const& value) noexcept = 0;
1212};
1213
1214class VBuilderConfig : public VRoot
1215{
1216public:
1217 virtual void setAvgTimingIterations(int32_t avgTiming) noexcept = 0;
1218 virtual int32_t getAvgTimingIterations() const noexcept = 0;
1219 virtual void setEngineCapability(EngineCapability capability) noexcept = 0;
1220 virtual EngineCapability getEngineCapability() const noexcept = 0;
1221 virtual void setFlags(BuilderFlags builderFlags) noexcept = 0;
1222 virtual BuilderFlags getFlags() const noexcept = 0;
1223 virtual void clearFlag(BuilderFlag builderFlag) noexcept = 0;
1224 virtual void setFlag(BuilderFlag builderFlag) noexcept = 0;
1225 virtual bool getFlag(BuilderFlag builderFlag) const noexcept = 0;
1226 virtual void setDeviceType(ILayer const* layer, DeviceType deviceType) noexcept = 0;
1227 virtual DeviceType getDeviceType(ILayer const* layer) const noexcept = 0;
1228 virtual bool isDeviceTypeSet(ILayer const* layer) const noexcept = 0;
1229 virtual void resetDeviceType(ILayer const* layer) noexcept = 0;
1230 virtual bool canRunOnDLA(ILayer const* layer) const noexcept = 0;
1231 virtual void setDLACore(int32_t dlaCore) noexcept = 0;
1232 virtual int32_t getDLACore() const noexcept = 0;
1233 virtual void setDefaultDeviceType(DeviceType deviceType) noexcept = 0;
1234 virtual DeviceType getDefaultDeviceType() const noexcept = 0;
1235 virtual void reset() noexcept = 0;
1236 virtual void setProfileStream(const cudaStream_t stream) noexcept = 0;
1237 virtual cudaStream_t getProfileStream() const noexcept = 0;
1238 virtual int32_t addOptimizationProfile(IOptimizationProfile const* profile) noexcept = 0;
1239 virtual int32_t getNbOptimizationProfiles() const noexcept = 0;
1240 virtual void setProfilingVerbosity(ProfilingVerbosity verbosity) noexcept = 0;
1241 virtual ProfilingVerbosity getProfilingVerbosity() const noexcept = 0;
1242 virtual bool setTacticSources(TacticSources tacticSources) noexcept = 0;
1243 virtual TacticSources getTacticSources() const noexcept = 0;
1244 virtual nvinfer1::ITimingCache* createTimingCache(void const* blob, std::size_t size) const noexcept = 0;
1245 virtual bool setTimingCache(ITimingCache const& cache, bool ignoreMismatch) noexcept = 0;
1246 virtual nvinfer1::ITimingCache const* getTimingCache() const noexcept = 0;
1247 virtual void setMemoryPoolLimit(MemoryPoolType pool, std::size_t poolSize) noexcept = 0;
1248 virtual std::size_t getMemoryPoolLimit(MemoryPoolType pool) const noexcept = 0;
1249 virtual void setPreviewFeature(PreviewFeature feature, bool enable) noexcept = 0;
1250 virtual bool getPreviewFeature(PreviewFeature feature) const noexcept = 0;
1251 virtual void setBuilderOptimizationLevel(int32_t level) noexcept = 0;
1252 virtual int32_t getBuilderOptimizationLevel() const noexcept = 0;
1253 virtual void setHardwareCompatibilityLevel(HardwareCompatibilityLevel hardwareCompatibilityLevel) noexcept = 0;
1254 virtual HardwareCompatibilityLevel getHardwareCompatibilityLevel() const noexcept = 0;
1255 virtual void setPluginsToSerialize(char const* const* paths, int32_t nbPaths) noexcept = 0;
1256 virtual char const* getPluginToSerialize(int32_t index) const noexcept = 0;
1257 virtual int32_t getNbPluginsToSerialize() const noexcept = 0;
1258 virtual void setMaxAuxStreams(int32_t nbStreams) noexcept = 0;
1259 virtual int32_t getMaxAuxStreams() const noexcept = 0;
1260 virtual void setProgressMonitor(IProgressMonitor* monitor) noexcept = 0;
1261 virtual IProgressMonitor* getProgressMonitor() const noexcept = 0;
1262 virtual void setRuntimePlatform(RuntimePlatform runtimePlatform) noexcept = 0;
1263 virtual RuntimePlatform getRuntimePlatform() const noexcept = 0;
1264 virtual void setMaxNbTactics(int32_t maxTactics) noexcept = 0;
1265 virtual int32_t getMaxNbTactics() const noexcept = 0;
1266 virtual bool setTilingOptimizationLevel(TilingOptimizationLevel level) noexcept = 0;
1267 virtual TilingOptimizationLevel getTilingOptimizationLevel() const noexcept = 0;
1268 virtual bool setL2LimitForTiling(int64_t size) noexcept = 0;
1269 virtual int64_t getL2LimitForTiling() const noexcept = 0;
1270 TRT_NODISCARD virtual bool setNbComputeCapabilities(int32_t maxNbComputeCapabilities) noexcept = 0;
1271 TRT_NODISCARD virtual int32_t getNbComputeCapabilities() const noexcept = 0;
1272 TRT_NODISCARD virtual bool setComputeCapability(ComputeCapability computeCapability, int32_t index) noexcept = 0;
1273 TRT_NODISCARD virtual ComputeCapability getComputeCapability(int32_t index) const noexcept = 0;
1274};
1275
1276class VSerializationConfig : public VRoot
1277{
1278public:
1279 virtual bool setFlags(SerializationFlags serializationFlags) noexcept = 0;
1280 virtual SerializationFlags getFlags() const noexcept = 0;
1281 virtual bool clearFlag(SerializationFlag serializationFlag) noexcept = 0;
1282 virtual bool setFlag(SerializationFlag serializationFlag) noexcept = 0;
1283 virtual bool getFlag(SerializationFlag serializationFlag) const noexcept = 0;
1284};
1285
1286class VBuilder : public VRoot
1287{
1288public:
1289 virtual bool platformHasFastFp16() const noexcept = 0;
1290 virtual bool platformHasFastInt8() const noexcept = 0;
1291 virtual int32_t getMaxDLABatchSize() const noexcept = 0;
1292 virtual int32_t getNbDLACores() const noexcept = 0;
1293 virtual void setGpuAllocator(IGpuAllocator* allocator) noexcept = 0;
1294 virtual nvinfer1::IBuilderConfig* createBuilderConfig() noexcept = 0;
1295 virtual nvinfer1::INetworkDefinition* createNetworkV2(NetworkDefinitionCreationFlags flags) noexcept = 0;
1296 virtual nvinfer1::IOptimizationProfile* createOptimizationProfile() noexcept = 0;
1297 virtual void setErrorRecorder(IErrorRecorder* recorder) noexcept = 0;
1298 virtual IErrorRecorder* getErrorRecorder() const noexcept = 0;
1299 virtual void reset() noexcept = 0;
1300 virtual bool platformHasTf32() const noexcept = 0;
1301 virtual nvinfer1::IHostMemory* buildSerializedNetwork(
1302 INetworkDefinition& network, IBuilderConfig& config) noexcept = 0;
1303 virtual bool isNetworkSupported(INetworkDefinition const& network, IBuilderConfig const& config) const noexcept = 0;
1304 virtual ILogger* getLogger() const noexcept = 0;
1305 virtual bool setMaxThreads(int32_t maxThreads) noexcept = 0;
1306 virtual int32_t getMaxThreads() const noexcept = 0;
1307 virtual IPluginRegistry& getPluginRegistry() noexcept = 0;
1308 virtual bool buildSerializedNetworkToStream(
1309 INetworkDefinition& network, IBuilderConfig& config, IStreamWriter& writer) noexcept = 0;
1310};
1311
1312class VRuntimeConfig : public VRoot
1313{
1314public:
1315 virtual IRuntimeConfig* getPImpl() noexcept = 0;
1316 virtual void setExecutionContextAllocationStrategy(ExecutionContextAllocationStrategy strategy) noexcept = 0;
1317 virtual ExecutionContextAllocationStrategy getExecutionContextAllocationStrategy() const noexcept = 0;
1318 TRT_NODISCARD virtual nvinfer1::IRuntimeCache* createRuntimeCache() const noexcept = 0;
1319 virtual bool setRuntimeCache(nvinfer1::IRuntimeCache const& cache) noexcept = 0;
1320 TRT_NODISCARD virtual nvinfer1::IRuntimeCache* getRuntimeCache() const noexcept = 0;
1321 virtual void setDynamicShapesKernelSpecializationStrategy(
1322 DynamicShapesKernelSpecializationStrategy dynamicShapesKernelSpecializationStrategy) noexcept = 0;
1323 virtual DynamicShapesKernelSpecializationStrategy getDynamicShapesKernelSpecializationStrategy() const noexcept = 0;
1324 virtual bool setCudaGraphStrategy(CudaGraphStrategy strategy) noexcept = 0;
1325 virtual CudaGraphStrategy getCudaGraphStrategy() const noexcept = 0;
1326};
1327
1328class VRuntimeCache : public VRoot
1329{
1330public:
1331 TRT_NODISCARD virtual nvinfer1::IHostMemory* serialize() const noexcept = 0;
1332 virtual bool deserialize(void const* blob, size_t size) noexcept = 0;
1333 virtual bool reset() noexcept = 0;
1334};
1335
1336} // namespace apiv
1337} // namespace nvinfer1
1338
1339// @endcond
1340
1341#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.
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-RTX 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:2958
nvinfer1::IOutputAllocator
v_1_0::IOutputAllocator IOutputAllocator
Definition: NvInferRuntime.h:4274
nvinfer1::ResizeSelector
ResizeSelector
The coordinate selector when resize to single pixel output.
Definition: NvInfer.h:3992
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:8977
nvinfer1::ScaleMode
ScaleMode
Controls how shift, scale and power are applied in a Scale layer.
Definition: NvInfer.h:1733
nvinfer1::RuntimePlatform
RuntimePlatform
Describes the intended runtime platform (operating system and CPU architecture) for the execution of ...
Definition: NvInfer.h:8554
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:9096
nvinfer1::CumulativeOperation
CumulativeOperation
Enumerates the cumulative operations that may be performed by a Cumulative layer.
Definition: NvInfer.h:6504
nvinfer1::SerializationFlag
SerializationFlag
List of valid flags that the engine can enable when serializing the bytes.
Definition: NvInferRuntime.h:2999
nvinfer1::BoundingBoxFormat
BoundingBoxFormat
Representation of bounding box data used for the Boxes input tensor in INMSLayer.
Definition: NvInfer.h:6058
nvinfer1::ComputeCapability
ComputeCapability
Describes compute capability that an engine will be built for.
Definition: NvInfer.h:9145
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:2678
nvinfer1::IStreamWriter
v_1_0::IStreamWriter IStreamWriter
Definition: NvInferRuntime.h:720
nvinfer1::IProfiler
v_1_0::IProfiler IProfiler
Definition: NvInferRuntime.h:1318
nvinfer1::DynamicShapesKernelSpecializationStrategy
DynamicShapesKernelSpecializationStrategy
Different kernel specialization strategies for dynamic shapes.
Definition: NvInferRuntime.h:3178
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:138
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:4973
nvinfer1::EngineStat
EngineStat
The kind of engine statistics that queried from the ICudaEngine.
Definition: NvInferRuntime.h:3374
nvinfer1::CudaGraphStrategy
CudaGraphStrategy
Strategies available for CUDA graphs optimizations for JIT (Just-In-Time) inference.
Definition: NvInferRuntime.h:3217
nvinfer1::ResizeRoundMode
ResizeRoundMode
The rounding mode for nearest neighbor resize.
Definition: NvInfer.h:4022
nvinfer1::EngineValidity
EngineValidity
Whether a TensorRT-RTX engine is likely to be valid on the current system.
Definition: NvInferRuntime.h:1863
nvinfer1::PaddingMode
PaddingMode
Enumerates the modes of padding to perform in convolution, deconvolution and pooling layer,...
Definition: NvInfer.h:911
nvinfer1::TripLimit
TripLimit
Enum that describes kinds of trip limits.
Definition: NvInfer.h:4381
nvinfer1::NetworkDefinitionCreationFlags
uint32_t NetworkDefinitionCreationFlags
Represents one or more NetworkDefinitionCreationFlag flags using binary OR operations....
Definition: NvInfer.h:10148
nvinfer1::PreviewFeature
PreviewFeature
Define preview features.
Definition: NvInfer.h:9052
nvinfer1::TilingOptimizationLevel
TilingOptimizationLevel
Define the optimization levels for Tiling.
Definition: NvInfer.h:9171
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:8586
nvinfer1::DeviceType
DeviceType
The device that this layer/network will execute on.
Definition: NvInferRuntime.h:1350
nvinfer1::LayerType
LayerType
The type values of layer classes.
Definition: NvInfer.h:57
nvinfer1::SampleMode
SampleMode
Controls how ISliceLayer and IGridSample handle out-of-bounds coordinates.
Definition: NvInfer.h:3136
nvinfer1::GatherMode
GatherMode
Control form of IGatherLayer.
Definition: NvInfer.h:2425
nvinfer1::IDebugListener
v_1_0::IDebugListener IDebugListener
Definition: NvInferRuntime.h:4315
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:130
nvinfer1::ProfilingVerbosity
ProfilingVerbosity
List of verbosity levels of layer information exposed in NVTX annotations and in IEngineInspector.
Definition: NvInferRuntime.h:2970
nvinfer1::NetworkDefinitionCreationFlag
NetworkDefinitionCreationFlag
List of immutable network properties expressed at network creation time. NetworkDefinitionCreationFla...
Definition: NvInfer.h:10159
nvinfer1::ElementWiseOperation
ElementWiseOperation
Enumerates the binary operations that may be performed by an ElementWise layer.
Definition: NvInfer.h:2335
nvinfer1::SerializationFlags
uint32_t SerializationFlags
Represents one or more SerializationFlag values using binary OR operations, e.g., 1U << Serialization...
Definition: NvInferRuntime.h:2989
nvinfer1::InterpolationMode
InterpolationMode
Enumerates various modes of interpolation.
Definition: NvInfer.h:3910
nvinfer1::BuilderFlag
BuilderFlag
List of valid modes that the builder can enable when creating an engine from a network definition.
Definition: NvInfer.h:8596
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:3104
nvinfer1::TopKOperation
TopKOperation
Enumerates the operations that may be performed by a TopK layer.
Definition: NvInfer.h:3417
nvinfer1::ReduceOperation
ReduceOperation
Enumerates the reduce operations that may be performed by a Reduce layer.
Definition: NvInfer.h:2773
nvinfer1::LayerInformationFormat
LayerInformationFormat
The format in which the IEngineInspector prints the layer information.
Definition: NvInferRuntime.h:5254
nvinfer1::ScatterMode
ScatterMode
Control form of IScatterLayer.
Definition: NvInfer.h:5792
nvinfer1::MatrixOperation
MatrixOperation
Enumerates the operations that may be performed on a tensor by IMatrixMultiplyLayer before multiplica...
Definition: NvInfer.h:3570
nvinfer1::ResizeCoordinateTransformation
ResizeCoordinateTransformation
The resize coordinate transformation function.
Definition: NvInfer.h:3938
nvinfer1::LoopOutput
LoopOutput
Enum that describes kinds of loop outputs.
Definition: NvInfer.h:4353
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:1347
nvinfer1::IProgressMonitor
v_1_0::IProgressMonitor IProgressMonitor
Definition: NvInfer.h:9287
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:2632
nvinfer1::AttentionNormalizationOp
AttentionNormalizationOp
Enumerates the operations that may be performed by the normalization in the attention subgraph.
Definition: NvInfer.h:6639

  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