18#ifndef NV_INFER_IMPL_H
19#define NV_INFER_IMPL_H
61class IActivationLayer;
63class IAlgorithmContext;
64class IAlgorithmIOInfo;
65class IAlgorithmVariant;
70class IConcatenationLayer;
73class IConvolutionLayer;
75class ICumulativeLayer;
76class IDeconvolutionLayer;
77class IDequantizeLayer;
79class IDynamicQuantizeLayer;
81class IElementWiseLayer;
82class IEngineInspector;
83class IExecutionContext;
86class IGridSampleLayer;
91class IIfConditionalInputLayer;
92class IIfConditionalOutputLayer;
95class IKVCacheUpdateLayer;
98class ILoopOutputLayer;
100class IMatrixMultiplyLayer;
102class IDistCollectiveLayer;
103class INetworkDefinition;
104class INormalizationLayer;
108class IOptimizationProfile;
110class IParametricReLULayer;
115class IPluginRegistry;
117class IRotaryEmbeddingLayer;
142class IRaggedSoftMaxLayer;
143class IRecurrenceLayer;
147class IReverseSequenceLayer;
152class ISerializationConfig;
162struct TimingCacheKey;
163struct TimingCacheValue;
165using TimingCacheKey = v_1_0::TimingCacheKey;
166using TimingCacheValue = v_1_0::TimingCacheValue;
170class ITripLimitLayer;
172class IUnsqueezeLayer;
247 virtual ~VRoot() noexcept = default;
250class VHostMemory : public VRoot
253 virtual void* data() const noexcept = 0;
254 virtual std::
size_t size() const noexcept = 0;
255 virtual
DataType type() const noexcept = 0;
258class VDimensionExpr : public VRoot
261 virtual bool isConstant()
const = 0;
262 virtual int64_t getConstantValue()
const = 0;
263 virtual bool isSizeTensor()
const = 0;
266class VExprBuilder :
public VRoot
269 virtual IDimensionExpr
const* constant(int64_t value) = 0;
270 virtual IDimensionExpr
const* operation(
273 virtual IDimensionExpr
const* declareSizeTensor(
274 int32_t outputIndex, IDimensionExpr
const& opt, IDimensionExpr
const& upper)
278class VRuntime :
public VRoot
281 virtual IRuntime* getPImpl() noexcept = 0;
282 virtual
nvinfer1::ICudaEngine* deserializeCudaEngine(
void const* blob, std::
size_t size) noexcept = 0;
284 virtual
void setDLACore(int32_t dlaCore) noexcept = 0;
285 virtual int32_t getDLACore() const noexcept = 0;
286 virtual int32_t getNbDLACores() const noexcept = 0;
287 virtual
void setGpuAllocator(
IGpuAllocator* allocator) noexcept = 0;
288 virtual
void setErrorRecorder(
IErrorRecorder* recorder) noexcept = 0;
290 virtual ILogger*
getLogger() const noexcept = 0;
291 virtual
bool setMaxThreads(int32_t maxThreads) noexcept = 0;
292 virtual int32_t getMaxThreads() const noexcept = 0;
293 virtual
void setTemporaryDirectory(
char const*) noexcept = 0;
294 virtual
char const* getTemporaryDirectory() const noexcept = 0;
298 virtual
void setPluginRegistryParent(IPluginRegistry* parent) noexcept = 0;
299 virtual IRuntime* loadRuntime(
char const* path) noexcept = 0;
300 virtual
void setEngineHostCodeAllowed(
bool allowed) noexcept = 0;
301 virtual
bool getEngineHostCodeAllowed() const noexcept = 0;
306class VRefitter : public VRoot
309 virtual IRefitter* getPImpl() noexcept = 0;
310 virtual
bool setWeights(
char const* layerName,
WeightsRole role, const Weights weights) noexcept = 0;
311 virtual
bool refitCudaEngine() noexcept = 0;
312 virtual int32_t getMissing(int32_t size,
char const** layerNames,
WeightsRole* roles) noexcept = 0;
313 virtual int32_t getAll(int32_t size,
char const** layerNames,
WeightsRole* roles) noexcept = 0;
314 virtual
bool setDynamicRange(
char const* tensorName,
float min,
float max) noexcept = 0;
315 virtual
float getDynamicRangeMin(
char const* tensorName) const noexcept = 0;
316 virtual
float getDynamicRangeMax(
char const* tensorName) const noexcept = 0;
317 virtual int32_t getTensorsWithDynamicRange(int32_t size,
char const** tensorNames) const noexcept = 0;
318 virtual
void setErrorRecorder(
IErrorRecorder* recorder) noexcept = 0;
320 virtual
bool setNamedWeights(
char const* name, Weights weights) noexcept = 0;
321 virtual int32_t getMissingWeights(int32_t size,
char const** weightsNames) noexcept = 0;
322 virtual int32_t getAllWeights(int32_t size,
char const** weightsNames) noexcept = 0;
323 virtual ILogger*
getLogger() const noexcept = 0;
324 virtual
bool setMaxThreads(int32_t maxThreads) noexcept = 0;
325 virtual int32_t getMaxThreads() const noexcept = 0;
326 virtual
bool setNamedWeightsWithLocation(
char const* name, Weights weights,
TensorLocation location) noexcept = 0;
327 virtual Weights getNamedWeights(
char const* weightsName) const noexcept = 0;
328 virtual
TensorLocation getWeightsLocation(
char const* weightsName) const noexcept = 0;
329 virtual
bool unsetNamedWeights(
char const* weightsName) noexcept = 0;
330 virtual
void setWeightsValidation(
bool weightsValidation) noexcept = 0;
331 virtual
bool getWeightsValidation() const noexcept = 0;
332 virtual
bool refitCudaEngineAsync(cudaStream_t stream) noexcept = 0;
333 virtual Weights getWeightsPrototype(
char const* weightsName) const noexcept = 0;
336class VOptimizationProfile : public VRoot
339 virtual bool setDimensions(
char const* inputName,
OptProfileSelector select,
Dims const& dims)
noexcept = 0;
341 virtual bool setShapeValues(
342 char const* inputName,
OptProfileSelector select, int32_t
const* values, int32_t nbValues)
noexcept = 0;
343 virtual int32_t getNbShapeValues(
char const* inputName)
const noexcept = 0;
344 virtual int32_t
const* getShapeValues(
char const* inputName,
OptProfileSelector select)
const noexcept = 0;
345 virtual bool setExtraMemoryTarget(
float target)
noexcept = 0;
346 virtual float getExtraMemoryTarget() const noexcept = 0;
347 virtual
bool isValid() const noexcept = 0;
350 char const* inputName,
OptProfileSelector select, int64_t const* values, int32_t nbValues) noexcept = 0;
355class VCudaEngine : public VRoot
358 virtual ICudaEngine* getPImpl() noexcept = 0;
359 virtual int32_t getNbLayers() const noexcept = 0;
360 virtual IHostMemory* serialize() const noexcept = 0;
362 virtual IExecutionContext* createExecutionContextWithoutDeviceMemory() noexcept = 0;
363 virtual
size_t getDeviceMemorySize() const noexcept = 0;
364 virtual
bool isRefittable() const noexcept = 0;
365 virtual
char const* getName() const noexcept = 0;
366 virtual int32_t getNbOptimizationProfiles() const noexcept = 0;
367 virtual int32_t const* getProfileTensorValues(
368 char const* tensorName, int32_t profileIndex,
OptProfileSelector select) const noexcept = 0;
370 virtual
void setErrorRecorder(
IErrorRecorder* recorder) noexcept = 0;
372 virtual
bool hasImplicitBatchDimension() const noexcept = 0;
375 virtual IEngineInspector* createEngineInspector() const noexcept = 0;
376 virtual
Dims getTensorShape(
char const* tensorName) const noexcept = 0;
377 virtual
DataType getTensorDataType(
char const* tensorName) const noexcept = 0;
378 virtual
TensorLocation getTensorLocation(
char const* tensorName) const noexcept = 0;
379 virtual
bool isShapeInferenceIO(
char const* tensorName) const noexcept = 0;
380 virtual
TensorIOMode getTensorIOMode(
char const* tensorName) const noexcept = 0;
381 virtual int32_t getTensorBytesPerComponent(
char const* tensorName) const noexcept = 0;
382 virtual int32_t getTensorComponentsPerElement(
char const* tensorName) const noexcept = 0;
383 virtual
TensorFormat getTensorFormat(
char const* tensorName) const noexcept = 0;
384 virtual
char const* getTensorFormatDesc(
char const* tensorName) const noexcept = 0;
385 virtual int32_t getTensorVectorizedDim(
char const* tensorName) const noexcept = 0;
386 virtual
Dims getProfileShape(
387 char const* tensorName, int32_t profileIndex,
OptProfileSelector select) const noexcept = 0;
388 virtual int32_t getNbIOTensors() const noexcept = 0;
389 virtual
char const* getIOTensorName(int32_t index) const noexcept = 0;
391 virtual int32_t getNbAuxStreams() const noexcept = 0;
393 virtual int32_t getTensorBytesPerComponentV2(
char const* tensorName, int32_t profileIndex) const noexcept = 0;
394 virtual int32_t getTensorComponentsPerElementV2(
char const* tensorName, int32_t profileIndex) const noexcept = 0;
395 virtual
TensorFormat getTensorFormatV2(
char const* tensorName, int32_t profileIndex) const noexcept = 0;
396 virtual
char const* getTensorFormatDescV2(
char const* tensorName, int32_t profileIndex) const noexcept = 0;
397 virtual int32_t getTensorVectorizedDimV2(
char const* tensorName, int32_t profileIndex) const noexcept = 0;
399 virtual ISerializationConfig* createSerializationConfig() noexcept = 0;
400 virtual IHostMemory* serializeWithConfig(ISerializationConfig& config) const noexcept = 0;
402 virtual
size_t getDeviceMemorySizeForProfile(int32_t profileIndex) const noexcept = 0;
403 virtual IRefitter* createRefitter(ILogger& logger) noexcept = 0;
405 virtual
bool setWeightStreamingBudget(int64_t gpuMemoryBudget) noexcept = 0;
406 virtual int64_t getWeightStreamingBudget() const noexcept = 0;
407 virtual int64_t getMinimumWeightStreamingBudget() const noexcept = 0;
408 virtual int64_t getStreamableWeightsSize() const noexcept = 0;
410 virtual
bool isDebugTensor(
char const* name) const noexcept = 0;
413 virtual
bool setWeightStreamingBudgetV2(int64_t gpuMemoryBudget) noexcept = 0;
414 virtual int64_t getWeightStreamingBudgetV2() const noexcept = 0;
415 virtual int64_t getWeightStreamingAutomaticBudget() const noexcept = 0;
416 virtual int64_t getWeightStreamingScratchMemorySize() const noexcept = 0;
417 virtual int64_t getDeviceMemorySizeV2() const noexcept = 0;
418 virtual int64_t getDeviceMemorySizeForProfileV2(int32_t profileIndex) const noexcept = 0;
420 TRT_NODISCARD virtual int64_t const* getProfileTensorValuesV2(
421 char const* tensorName, int32_t profileIndex,
OptProfileSelector select) const noexcept = 0;
422 TRT_NODISCARD virtual IExecutionContext* createExecutionContextWithRuntimeConfig(
423 IRuntimeConfig* runtimeConfig) noexcept = 0;
424 TRT_NODISCARD virtual IRuntimeConfig* createRuntimeConfig() noexcept = 0;
427 TRT_NODISCARD virtual
char const* getAliasedInputTensor(
char const* tensorName) const noexcept = 0;
430class VExecutionContext : public VRoot
433 virtual IExecutionContext* getPImpl() noexcept = 0;
434 virtual
void setDebugSync(
bool sync) noexcept = 0;
435 virtual
bool getDebugSync() const noexcept = 0;
436 virtual
void setProfiler(
IProfiler*) noexcept = 0;
437 virtual
IProfiler* getProfiler() const noexcept = 0;
438 virtual ICudaEngine const& getEngine() const noexcept = 0;
439 virtual
void setName(
char const* name) noexcept = 0;
440 virtual
char const* getName() const noexcept = 0;
441 virtual
void setDeviceMemory(
void* memory) noexcept = 0;
442 virtual int32_t getOptimizationProfile() const noexcept = 0;
443 virtual
bool allInputDimensionsSpecified() const noexcept = 0;
444 virtual
bool allInputShapesSpecified() const noexcept = 0;
445 virtual
void setErrorRecorder(
IErrorRecorder* recorder) noexcept = 0;
447 virtual
bool executeV2(
void* const* bindings) noexcept = 0;
448 virtual
bool setOptimizationProfileAsync(int32_t profileIndex, cudaStream_t stream) noexcept = 0;
449 virtual
void setEnqueueEmitsProfile(
bool enqueueEmitsProfile) noexcept = 0;
450 virtual
bool getEnqueueEmitsProfile() const noexcept = 0;
451 virtual
bool reportToProfiler() const noexcept = 0;
452 virtual
bool setInputShape(
char const* tensorName,
Dims const& dims) noexcept = 0;
453 virtual
Dims getTensorShape(
char const* tensorName) const noexcept = 0;
454 virtual
Dims getTensorStrides(
char const* tensorName) const noexcept = 0;
455 virtual
bool setTensorAddress(
char const* tensorName,
void* data) noexcept = 0;
456 virtual
void const* getTensorAddress(
char const* tensorName) const noexcept = 0;
457 virtual
bool setInputTensorAddress(
char const* tensorName,
void const* data) noexcept = 0;
458 virtual
bool setOutputTensorAddress(
char const* tensorName,
void* data) noexcept = 0;
459 virtual int32_t inferShapes(int32_t nbMaxNames,
char const** tensorNames) noexcept = 0;
460 virtual
bool setInputConsumedEvent(cudaEvent_t event) noexcept = 0;
461 virtual cudaEvent_t getInputConsumedEvent() const noexcept = 0;
462 virtual
void* getOutputTensorAddress(
char const* tensorName) const noexcept = 0;
463 virtual
bool setOutputAllocator(
char const* tensorName,
IOutputAllocator* outputAllocator) noexcept = 0;
464 virtual
IOutputAllocator* getOutputAllocator(
char const* name) noexcept = 0;
465 virtual int64_t getMaxOutputSize(
char const* tensorName) const noexcept = 0;
466 virtual
bool setTemporaryStorageAllocator(
IGpuAllocator* allocator) noexcept = 0;
467 virtual
IGpuAllocator* getTemporaryStorageAllocator() const noexcept = 0;
468 virtual
bool enqueueV3(cudaStream_t stream) noexcept = 0;
469 virtual
void setPersistentCacheLimit(
size_t size) noexcept = 0;
470 virtual
size_t getPersistentCacheLimit() const noexcept = 0;
473 virtual
void setAuxStreams(cudaStream_t* auxStreams, int32_t nbStreams) noexcept = 0;
474 virtual
bool setDebugListener(
IDebugListener* listener) noexcept = 0;
476 virtual
bool setTensorDebugState(
char const* name,
bool flag) noexcept = 0;
477 virtual
bool getDebugState(
char const* name) const noexcept = 0;
478 virtual
bool setAllTensorsDebugState(
bool flag) noexcept = 0;
479 virtual
size_t updateDeviceMemorySizeForShapes() noexcept = 0;
480 virtual
void setDeviceMemoryV2(
void* memory, int64_t size) noexcept = 0;
481 TRT_NODISCARD virtual IRuntimeConfig* getRuntimeConfig() const noexcept = 0;
482 virtual
bool setUnfusedTensorsDebugState(
bool flag) noexcept = 0;
483 virtual
bool getUnfusedTensorsDebugState() const noexcept = 0;
484#if ENABLE_FEATURE_DISABLE_RUNTIME_ALLOCATION
485 virtual bool isStreamCapturable(cudaStream_t stream)
const noexcept = 0;
487 virtual bool setCommunicator(
void* communicator)
noexcept = 0;
490class VEngineInspector :
public VRoot
493 virtual IEngineInspector* getPImpl() noexcept = 0;
494 virtual
bool setExecutionContext(IExecutionContext const* context) noexcept = 0;
495 virtual IExecutionContext const* getExecutionContext() const noexcept = 0;
496 virtual
char const* getLayerInformation(int32_t layerIndex,
LayerInformationFormat format) const noexcept = 0;
498 virtual
void setErrorRecorder(
IErrorRecorder* recorder) noexcept = 0;
502class VTensor : public VRoot
505 virtual void setName(
char const* name)
noexcept = 0;
506 virtual char const* getName() const noexcept = 0;
507 virtual
void setDimensions(
Dims const& dimensions) noexcept = 0;
508 virtual
Dims getDimensions() const noexcept = 0;
509 virtual
void setType(
DataType type) noexcept = 0;
510 virtual
DataType getType() const noexcept = 0;
511 virtual
bool setDynamicRange(
float min,
float max) noexcept = 0;
512 virtual
bool isNetworkInput() const noexcept = 0;
513 virtual
bool isNetworkOutput() const noexcept = 0;
514 virtual
void setBroadcastAcrossBatch(
bool broadcastAcrossBatch) noexcept = 0;
515 virtual
bool getBroadcastAcrossBatch() const noexcept = 0;
518 virtual
bool dynamicRangeIsSet() const noexcept = 0;
519 virtual
void resetDynamicRange() noexcept = 0;
520 virtual
float getDynamicRangeMin() const noexcept = 0;
521 virtual
float getDynamicRangeMax() const noexcept = 0;
522 virtual
void setAllowedFormats(
TensorFormats formats) noexcept = 0;
523 virtual
TensorFormats getAllowedFormats() const noexcept = 0;
524 virtual
bool isShapeTensor() const noexcept = 0;
525 virtual
bool isExecutionTensor() const noexcept = 0;
526 virtual
void setDimensionName(int32_t index,
char const* name) noexcept = 0;
527 virtual
char const* getDimensionName(int32_t index) const noexcept = 0;
530class VLayer : public VRoot
533 virtual LayerType getType() const noexcept = 0;
534 virtual
void setName(
char const* name) noexcept = 0;
535 virtual
char const* getName() const noexcept = 0;
536 virtual int32_t getNbInputs() const noexcept = 0;
537 virtual ITensor* getInput(int32_t index) const noexcept = 0;
538 virtual int32_t getNbOutputs() const noexcept = 0;
539 virtual ITensor* getOutput(int32_t index) const noexcept = 0;
540 virtual
void setInput(int32_t index, ITensor& tensor) noexcept = 0;
541 virtual
void setPrecision(
DataType dataType) noexcept = 0;
542 virtual
DataType getPrecision() const noexcept = 0;
543 virtual
bool precisionIsSet() const noexcept = 0;
544 virtual
void resetPrecision() noexcept = 0;
545 virtual
void setOutputType(int32_t index,
DataType dataType) noexcept = 0;
546 virtual
DataType getOutputType(int32_t index) const noexcept = 0;
547 virtual
bool outputTypeIsSet(int32_t index) const noexcept = 0;
548 virtual
void resetOutputType(int32_t index) noexcept = 0;
549 virtual
void setMetadata(
char const* docString) noexcept = 0;
550 virtual
char const* getMetadata() const noexcept = 0;
551 virtual
bool setNbRanks(int32_t nbRanks) noexcept = 0;
552 virtual int32_t getNbRanks() const noexcept = 0;
555class VConvolutionLayer : public VRoot
558 virtual void setNbOutputMaps(int64_t nbOutputMaps)
noexcept = 0;
559 virtual int64_t getNbOutputMaps() const noexcept = 0;
560 virtual
void setNbGroups(int64_t nbGroups) noexcept = 0;
561 virtual int64_t getNbGroups() const noexcept = 0;
562 virtual
void setKernelWeights(Weights weights) noexcept = 0;
563 virtual Weights getKernelWeights() const noexcept = 0;
564 virtual
void setBiasWeights(Weights weights) noexcept = 0;
565 virtual Weights getBiasWeights() const noexcept = 0;
566 virtual
void setPrePadding(
Dims const& padding) noexcept = 0;
567 virtual
Dims getPrePadding() const noexcept = 0;
568 virtual
void setPostPadding(
Dims const& padding) noexcept = 0;
569 virtual
Dims getPostPadding() const noexcept = 0;
570 virtual
void setPaddingMode(
PaddingMode paddingMode) noexcept = 0;
571 virtual
PaddingMode getPaddingMode() const noexcept = 0;
572 virtual
void setKernelSizeNd(
Dims const& kernelSize) noexcept = 0;
573 virtual
Dims getKernelSizeNd() const noexcept = 0;
574 virtual
void setStrideNd(
Dims const& stride) noexcept = 0;
575 virtual
Dims getStrideNd() const noexcept = 0;
576 virtual
void setPaddingNd(
Dims const& padding) noexcept = 0;
577 virtual
Dims getPaddingNd() const noexcept = 0;
578 virtual
void setDilationNd(
Dims const& dilation) noexcept = 0;
579 virtual
Dims getDilationNd() const noexcept = 0;
582class VActivationLayer : public VRoot
587 virtual
void setAlpha(
float alpha) noexcept = 0;
588 virtual
void setBeta(
float beta) noexcept = 0;
589 virtual
float getAlpha() const noexcept = 0;
590 virtual
float getBeta() const noexcept = 0;
593class VPoolingLayer : public VRoot
596 virtual void setPoolingType(
PoolingType type)
noexcept = 0;
597 virtual PoolingType getPoolingType() const noexcept = 0;
598 virtual
void setBlendFactor(
float blendFactor) noexcept = 0;
599 virtual
float getBlendFactor() const noexcept = 0;
600 virtual
void setAverageCountExcludesPadding(
bool exclusive) noexcept = 0;
601 virtual
bool getAverageCountExcludesPadding() const noexcept = 0;
602 virtual
void setPrePadding(
Dims const& padding) noexcept = 0;
603 virtual
Dims getPrePadding() const noexcept = 0;
604 virtual
void setPostPadding(
Dims const& padding) noexcept = 0;
605 virtual
Dims getPostPadding() const noexcept = 0;
606 virtual
void setPaddingMode(
PaddingMode paddingMode) noexcept = 0;
607 virtual
PaddingMode getPaddingMode() const noexcept = 0;
608 virtual
void setWindowSizeNd(
Dims const& windowSize) noexcept = 0;
609 virtual
Dims getWindowSizeNd() const noexcept = 0;
610 virtual
void setStrideNd(
Dims const& stride) noexcept = 0;
611 virtual
Dims getStrideNd() const noexcept = 0;
612 virtual
void setPaddingNd(
Dims const& padding) noexcept = 0;
613 virtual
Dims getPaddingNd() const noexcept = 0;
616class VLRNLayer : public VRoot
619 virtual void setWindowSize(int64_t windowSize)
noexcept = 0;
620 virtual int64_t getWindowSize() const noexcept = 0;
621 virtual
void setAlpha(
float alpha) noexcept = 0;
622 virtual
float getAlpha() const noexcept = 0;
623 virtual
void setBeta(
float beta) noexcept = 0;
624 virtual
float getBeta() const noexcept = 0;
625 virtual
void setK(
float k) noexcept = 0;
626 virtual
float getK() const noexcept = 0;
629class VScaleLayer : public VRoot
632 virtual void setMode(
ScaleMode mode)
noexcept = 0;
633 virtual ScaleMode getMode() const noexcept = 0;
634 virtual
void setShift(Weights shift) noexcept = 0;
635 virtual Weights getShift() const noexcept = 0;
636 virtual
void setScale(Weights scale) noexcept = 0;
637 virtual Weights getScale() const noexcept = 0;
638 virtual
void setPower(Weights power) noexcept = 0;
639 virtual Weights getPower() const noexcept = 0;
640 virtual int32_t getChannelAxis() const noexcept = 0;
641 virtual
void setChannelAxis(int32_t channelAxis) noexcept = 0;
644class VSoftMaxLayer : public VRoot
647 virtual void setAxes(uint32_t axes)
noexcept = 0;
648 virtual uint32_t getAxes() const noexcept = 0;
651class VConcatenationLayer : public VRoot
654 virtual void setAxis(int32_t axis)
noexcept = 0;
655 virtual int32_t getAxis() const noexcept = 0;
658class VDeconvolutionLayer : public VRoot
661 virtual void setNbOutputMaps(int64_t nbOutputMaps)
noexcept = 0;
662 virtual int64_t getNbOutputMaps() const noexcept = 0;
663 virtual
void setNbGroups(int64_t nbGroups) noexcept = 0;
664 virtual int64_t getNbGroups() const noexcept = 0;
665 virtual
void setKernelWeights(Weights weights) noexcept = 0;
666 virtual Weights getKernelWeights() const noexcept = 0;
667 virtual
void setBiasWeights(Weights weights) noexcept = 0;
668 virtual Weights getBiasWeights() const noexcept = 0;
669 virtual
void setPrePadding(
Dims const& padding) noexcept = 0;
670 virtual
Dims getPrePadding() const noexcept = 0;
671 virtual
void setPostPadding(
Dims const& padding) noexcept = 0;
672 virtual
Dims getPostPadding() const noexcept = 0;
673 virtual
void setPaddingMode(
PaddingMode paddingMode) noexcept = 0;
674 virtual
PaddingMode getPaddingMode() const noexcept = 0;
675 virtual
void setKernelSizeNd(
Dims const& kernelSize) noexcept = 0;
676 virtual
Dims getKernelSizeNd() const noexcept = 0;
677 virtual
void setStrideNd(
Dims const& stride) noexcept = 0;
678 virtual
Dims getStrideNd() const noexcept = 0;
679 virtual
void setPaddingNd(
Dims const& padding) noexcept = 0;
680 virtual
Dims getPaddingNd() const noexcept = 0;
681 virtual
void setDilationNd(
Dims const& dilation) noexcept = 0;
682 virtual
Dims getDilationNd() const noexcept = 0;
685class VElementWiseLayer : public VRoot
692class VGatherLayer : public VRoot
695 virtual void setGatherAxis(int32_t axis)
noexcept = 0;
696 virtual int32_t getGatherAxis() const noexcept = 0;
697 virtual
void setNbElementWiseDims(int32_t k) noexcept = 0;
698 virtual int32_t getNbElementWiseDims() const noexcept = 0;
699 virtual
void setMode(
GatherMode mode) noexcept = 0;
700 virtual
GatherMode getMode() const noexcept = 0;
703class VPluginLayer : public VRoot
706 virtual IPlugin& getPlugin() noexcept = 0;
709class VPluginV2Layer : public VRoot
712 virtual IPluginV2& getPlugin() noexcept = 0;
715class VPluginV3Layer : public VRoot
718 virtual IPluginV3& getPlugin() noexcept = 0;
721class VUnaryLayer : public VRoot
728class VReduceLayer : public VRoot
733 virtual
void setReduceAxes(uint32_t reduceAxes) noexcept = 0;
734 virtual uint32_t getReduceAxes() const noexcept = 0;
735 virtual
void setKeepDimensions(
bool keepDimensions) noexcept = 0;
736 virtual
bool getKeepDimensions() const noexcept = 0;
739class VPaddingLayer : public VRoot
742 virtual void setPrePaddingNd(
Dims const& padding)
noexcept = 0;
743 virtual Dims getPrePaddingNd() const noexcept = 0;
744 virtual
void setPostPaddingNd(
Dims const& padding) noexcept = 0;
745 virtual
Dims getPostPaddingNd() const noexcept = 0;
748class VShuffleLayer : public VRoot
751 virtual void setFirstTranspose(Permutation
const& permutation)
noexcept = 0;
752 virtual Permutation
const& getFirstTranspose() const noexcept = 0;
753 virtual
void setReshapeDimensions(
Dims const& dimensions) noexcept = 0;
754 virtual
Dims getReshapeDimensions() const noexcept = 0;
755 virtual
void setSecondTranspose(Permutation const& permutation) noexcept = 0;
756 virtual Permutation const& getSecondTranspose() const noexcept = 0;
757 virtual
void setZeroIsPlaceholder(
bool zeroIsPlaceholder) noexcept = 0;
758 virtual
bool getZeroIsPlaceholder() const noexcept = 0;
761class VSliceLayer : public VRoot
764 virtual void setStart(
Dims const& start)
noexcept = 0;
765 virtual Dims getStart() const noexcept = 0;
766 virtual
void setSize(
Dims const& size) noexcept = 0;
767 virtual
Dims getSize() const noexcept = 0;
768 virtual
void setStride(
Dims const& stride) noexcept = 0;
769 virtual
Dims getStride() const noexcept = 0;
770 virtual
void setMode(
SampleMode mode) noexcept = 0;
771 virtual
SampleMode getMode() const noexcept = 0;
772 virtual
void setAxes(
Dims const& axes) noexcept = 0;
773 virtual
Dims getAxes() const noexcept = 0;
776class VShapeLayer : public VRoot
781class VTopKLayer :
public VRoot
786 virtual
void setK(int32_t k) noexcept = 0;
787 virtual int32_t getK() const noexcept = 0;
788 virtual
void setReduceAxes(uint32_t reduceAxes) noexcept = 0;
789 virtual uint32_t getReduceAxes() const noexcept = 0;
790 virtual
bool setIndicesType(
DataType type) noexcept = 0;
791 virtual
DataType getIndicesType() const noexcept = 0;
794class VMatrixMultiplyLayer : public VRoot
797 virtual void setOperation(int32_t index,
MatrixOperation op)
noexcept = 0;
798 virtual MatrixOperation getOperation(int32_t index)
const noexcept = 0;
801class VNonZeroLayer :
public VRoot
804 virtual bool setIndicesType(
DataType type)
noexcept = 0;
805 virtual DataType getIndicesType() const noexcept = 0;
808class VRaggedSoftMaxLayer : public VRoot
813class VIdentityLayer :
public VRoot
818class VCastLayer :
public VRoot
821 virtual void setToType(
DataType toType)
noexcept = 0;
822 virtual DataType getToType() const noexcept = 0;
825class VConstantLayer : public VRoot
828 virtual void setWeights(Weights weights)
noexcept = 0;
829 virtual Weights getWeights() const noexcept = 0;
830 virtual
void setDimensions(
Dims const& dimensions) noexcept = 0;
831 virtual
Dims getDimensions() const noexcept = 0;
834class VParametricReLULayer : public VRoot
839class VResizeLayer :
public VRoot
842 virtual void setOutputDimensions(
Dims const& dimensions)
noexcept = 0;
843 virtual Dims getOutputDimensions() const noexcept = 0;
844 virtual
void setScales(
float const* scales, int32_t nbScales) noexcept = 0;
845 virtual int32_t getScales(int32_t size,
float* scales) const noexcept = 0;
850 virtual
void setSelectorForSinglePixel(
ResizeSelector selector) noexcept = 0;
851 virtual
ResizeSelector getSelectorForSinglePixel() const noexcept = 0;
854 virtual
void setCubicCoeff(
float value) noexcept = 0;
855 virtual
float getCubicCoeff() const noexcept = 0;
856 virtual
void setExcludeOutside(
bool value) noexcept = 0;
857 virtual
bool getExcludeOutside() const noexcept = 0;
860class VLoopBoundaryLayer : public VRoot
863 virtual ILoop* getLoop() const noexcept = 0;
866class VRecurrenceLayer : public VRoot
871class VLoopOutputLayer :
public VRoot
874 virtual LoopOutput getLoopOutput() const noexcept = 0;
875 virtual
void setAxis(int32_t axis) noexcept = 0;
876 virtual int32_t getAxis() const noexcept = 0;
879class VTripLimitLayer : public VRoot
882 virtual TripLimit getTripLimit() const noexcept = 0;
885class VIteratorLayer : public VRoot
888 virtual void setAxis(int32_t axis)
noexcept = 0;
889 virtual int32_t getAxis() const noexcept = 0;
890 virtual
void setReverse(
bool reverse) noexcept = 0;
891 virtual
bool getReverse() const noexcept = 0;
893class VLoop : public VRoot
896 virtual IRecurrenceLayer* addRecurrence(ITensor& initialValue)
noexcept = 0;
897 virtual ITripLimitLayer* addTripLimit(ITensor& tensor,
TripLimit limit)
noexcept = 0;
898 virtual IIteratorLayer* addIterator(ITensor& tensor, int32_t axis = 0,
bool reverse =
false) noexcept = 0;
899 virtual ILoopOutputLayer* addLoopOutput(ITensor& tensor,
LoopOutput outputKind, int32_t axis = 0) noexcept = 0;
900 virtual
void setName(
char const* name) noexcept = 0;
901 virtual
char const* getName() const noexcept = 0;
904class VConditionalBoundaryLayer : public VRoot
907 virtual IIfConditional* getConditional() const noexcept = 0;
910class VConditionLayer : public VRoot
915class VConditionalInputLayer :
public VRoot
920class VConditionalOutputLayer :
public VRoot
925class VIfConditional :
public VRoot
928 virtual IConditionLayer* setCondition(ITensor& tensor)
noexcept = 0;
929 virtual IIfConditionalInputLayer* addInput(ITensor& tensor)
noexcept = 0;
930 virtual IIfConditionalOutputLayer* addOutput(ITensor& trueTensor, ITensor& falseTensor)
noexcept = 0;
931 virtual void setName(
char const* name)
noexcept = 0;
932 virtual char const* getName() const noexcept = 0;
935class VAttentionBoundaryLayer : public VRoot
938 virtual IAttention* getAttention() const noexcept = 0;
941class VAttentionInputLayer : public VRoot
946class VAttentionOutputLayer :
public VRoot
951class VAttention :
public VRoot
954 TRT_NODISCARD virtual bool setInput(int32_t index, ITensor& input)
noexcept = 0;
955 TRT_NODISCARD virtual int32_t getNbInputs() const noexcept = 0;
956 TRT_NODISCARD virtual ITensor* getInput(int32_t index) const noexcept = 0;
957 TRT_NODISCARD virtual int32_t getNbOutputs() const noexcept = 0;
958 TRT_NODISCARD virtual ITensor* getOutput(int32_t index) const noexcept = 0;
959 TRT_NODISCARD virtual
bool setName(
char const* name) noexcept = 0;
960 TRT_NODISCARD virtual
char const* getName() const noexcept = 0;
963 TRT_NODISCARD virtual
bool setCausal(
bool isCausal) noexcept = 0;
965 TRT_NODISCARD virtual
bool setMask(ITensor& mask) noexcept = 0;
967 TRT_NODISCARD virtual
bool setDecomposable(
bool decomposable) noexcept = 0;
968 TRT_NODISCARD virtual
bool getDecomposable() const noexcept = 0;
969 TRT_NODISCARD virtual
bool setNormalizationQuantizeScale(ITensor& tensor) noexcept = 0;
970 TRT_NODISCARD virtual ITensor* getNormalizationQuantizeScale() const noexcept = 0;
973 TRT_NODISCARD virtual
bool setMetadata(
char const* docString) noexcept = 0;
974 TRT_NODISCARD virtual
char const* getMetadata() const noexcept = 0;
975 TRT_NODISCARD virtual
bool setNbRanks(int32_t nbRanks) noexcept = 0;
976 TRT_NODISCARD virtual int32_t getNbRanks() const noexcept = 0;
979class VSelectLayer : public VRoot
983class VAssertionLayer :
public VRoot
986 virtual void setMessage(
char const* message)
noexcept = 0;
987 virtual char const* getMessage() const noexcept = 0;
990class VFillLayer : public VRoot
993 virtual void setDimensions(
Dims const& dimensions)
noexcept = 0;
994 virtual Dims getDimensions() const noexcept = 0;
997 virtual
void setAlpha(
double alpha) noexcept = 0;
998 virtual
double getAlpha() const noexcept = 0;
999 virtual
void setBeta(
double beta) noexcept = 0;
1000 virtual
double getBeta() const noexcept = 0;
1001 virtual
void setAlphaInt64(int64_t alpha) noexcept = 0;
1002 virtual int64_t getAlphaInt64() const noexcept = 0;
1003 virtual
void setBetaInt64(int64_t beta) noexcept = 0;
1004 virtual int64_t getBetaInt64() const noexcept = 0;
1005 virtual
bool isAlphaBetaInt64() const noexcept = 0;
1006 virtual
DataType getToType() const noexcept = 0;
1007 virtual
void setToType(
DataType toType) noexcept = 0;
1010class VQuantizeLayer : public VRoot
1013 virtual int32_t getAxis() const noexcept = 0;
1014 virtual
void setAxis(int32_t axis) noexcept = 0;
1015 virtual
DataType getToType() const noexcept = 0;
1016 virtual
void setToType(
DataType toType) noexcept = 0;
1017 virtual
Dims getBlockShape() const noexcept = 0;
1018 virtual
bool setBlockShape(
Dims const& blockShape) noexcept = 0;
1021class VDequantizeLayer : public VRoot
1024 virtual int32_t getAxis() const noexcept = 0;
1025 virtual
void setAxis(int32_t axis) noexcept = 0;
1026 virtual
DataType getToType() const noexcept = 0;
1027 virtual
void setToType(
DataType toType) noexcept = 0;
1028 virtual
Dims getBlockShape() const noexcept = 0;
1029 virtual
bool setBlockShape(
Dims const& blockShape) noexcept = 0;
1032class VDynamicQuantizeLayer : public VRoot
1037 TRT_DEPRECATED virtual int32_t getBlockSize() const noexcept = 0;
1038 TRT_DEPRECATED virtual
void setBlockSize(int32_t axis) noexcept = 0;
1039 virtual
DataType getScaleType() const noexcept = 0;
1040 virtual
void setScaleType(
DataType axis) noexcept = 0;
1041 virtual
DataType getToType() const noexcept = 0;
1042 virtual
void setToType(
DataType toType) noexcept = 0;
1043 virtual
Dims getBlockShape() const noexcept = 0;
1044 virtual
void setBlockShape(
Dims const& blockShape) noexcept = 0;
1047class VScatterLayer : public VRoot
1050 virtual void setMode(
ScatterMode mode)
noexcept = 0;
1052 virtual
void setAxis(int32_t axis) noexcept = 0;
1053 virtual int32_t getAxis() const noexcept = 0;
1056class VEinsumLayer : public VRoot
1059 virtual bool setEquation(
char const* equation)
noexcept = 0;
1060 virtual char const* getEquation() const noexcept = 0;
1063class VOneHotLayer : public VRoot
1066 virtual int32_t getAxis() const noexcept = 0;
1067 virtual
void setAxis(int32_t axis) noexcept = 0;
1070class VGridSampleLayer : public VRoot
1075 virtual
void setAlignCorners(
bool alignCorners) noexcept = 0;
1076 virtual
bool getAlignCorners() const noexcept = 0;
1077 virtual
bool setSampleMode(
SampleMode mode) noexcept = 0;
1078 virtual
SampleMode getSampleMode() const noexcept = 0;
1081class VNMSLayer : public VRoot
1086 virtual
void setTopKBoxLimit(int32_t limit) noexcept = 0;
1087 virtual int32_t getTopKBoxLimit() const noexcept = 0;
1088 virtual
bool setIndicesType(
DataType type) noexcept = 0;
1089 virtual
DataType getIndicesType() const noexcept = 0;
1092class VReverseSequenceLayer : public VRoot
1095 virtual void setBatchAxis(int32_t batchAxis)
noexcept = 0;
1096 virtual int32_t getBatchAxis() const noexcept = 0;
1098 virtual
void setSequenceAxis(int32_t sequenceAxis) noexcept = 0;
1099 virtual int32_t getSequenceAxis() const noexcept = 0;
1102class VNormalizationLayer : public VRoot
1105 virtual void setEpsilon(
float eps)
noexcept = 0;
1106 virtual float getEpsilon() const noexcept = 0;
1107 virtual
void setAxes(uint32_t axesMask) noexcept = 0;
1108 virtual uint32_t getAxes() const noexcept = 0;
1109 virtual
void setNbGroups(int64_t nbGroups) noexcept = 0;
1110 virtual int64_t getNbGroups() const noexcept = 0;
1111 virtual
void setComputePrecision(
DataType type) noexcept = 0;
1112 virtual
DataType getComputePrecision() const noexcept = 0;
1113 virtual
bool isV2() const noexcept = 0;
1116class VSqueezeLayer : public VRoot
1120class VUnsqueezeLayer :
public VRoot
1124class VCumulativeLayer :
public VRoot
1129 virtual
void setExclusive(
bool exclusive) noexcept = 0;
1130 virtual
bool getExclusive() const noexcept = 0;
1131 virtual
void setReverse(
bool reverse) noexcept = 0;
1132 virtual
bool getReverse() const noexcept = 0;
1135class VRotaryEmbeddingLayer : public VRoot
1138 virtual void setInterleaved(
bool interleaved)
noexcept = 0;
1139 virtual bool getInterleaved() const noexcept = 0;
1140 virtual
bool setRotaryEmbeddingDim(int32_t rotaryEmbeddingDim) noexcept = 0;
1141 virtual int32_t getRotaryEmbeddingDim() const noexcept = 0;
1142 virtual
void setInput(int32_t index, ITensor& input) noexcept = 0;
1145class VKVCacheUpdateLayer : public VRoot
1152class VMoELayer : public VRoot
1155 virtual void setGatedWeights(ITensor& fcGateWeights, ITensor& fcUpWeights, ITensor& fcDownWeights,
MoEActType activationType)
noexcept = 0;
1156 virtual void setGatedBiases(ITensor& fcGateBiases, ITensor& fcUpBiases, ITensor& fcDownBiases)
noexcept = 0;
1157 virtual void setActivationType(
MoEActType activationType)
noexcept = 0;
1158 virtual MoEActType getActivationType() const noexcept = 0;
1159 virtual
void setQuantizationStatic(ITensor& fcDownActivationScale,
DataType dataType) noexcept = 0;
1160 virtual
void setQuantizationDynamicDblQ(ITensor& fcDownActivationDblQScale,
DataType dataType,
Dims const& blockShape,
DataType dynQOutputScaleType) noexcept = 0;
1161 virtual
void setQuantizationToType(
DataType type) noexcept = 0;
1162 virtual
DataType getQuantizationToType() const noexcept = 0;
1163 virtual
void setQuantizationBlockShape(
Dims const& blockShape) noexcept = 0;
1164 virtual
Dims getQuantizationBlockShape() const noexcept = 0;
1165 virtual
void setDynQOutputScaleType(
DataType type) noexcept = 0;
1166 virtual
DataType getDynQOutputScaleType() const noexcept = 0;
1167 virtual
void setSwigluParams(
float limit,
float alpha,
float beta) noexcept = 0;
1168 virtual
void setSwigluParamLimit(
float limit) noexcept = 0;
1169 virtual
float getSwigluParamLimit() const noexcept = 0;
1170 virtual
void setSwigluParamAlpha(
float alpha) noexcept = 0;
1171 virtual
float getSwigluParamAlpha() const noexcept = 0;
1172 virtual
void setSwigluParamBeta(
float beta) noexcept = 0;
1173 virtual
float getSwigluParamBeta() const noexcept = 0;
1174 virtual
void setInput(int32_t index, ITensor& tensor) noexcept = 0;
1178class VNetworkDefinition : public VRoot
1181 virtual ITensor* addInput(
char const* name,
DataType type,
Dims const& dimensions)
noexcept = 0;
1182 virtual void markOutput(ITensor& tensor)
noexcept = 0;
1183 virtual IActivationLayer* addActivation(ITensor& input,
ActivationType type)
noexcept = 0;
1184 virtual ILRNLayer* addLRN(ITensor& input, int64_t window,
float alpha,
float beta,
float k)
noexcept = 0;
1185 virtual IScaleLayer* addScale(
1186 ITensor& input,
ScaleMode mode, Weights shift, Weights scale, Weights power)
noexcept = 0;
1187 virtual ISoftMaxLayer* addSoftMax(ITensor& input)
noexcept = 0;
1188 virtual IConcatenationLayer* addConcatenation(ITensor*
const* inputs, int32_t nbInputs)
noexcept = 0;
1189 virtual IElementWiseLayer* addElementWise(ITensor& input1, ITensor& input2,
ElementWiseOperation op)
noexcept = 0;
1190 virtual IUnaryLayer* addUnary(ITensor& input,
UnaryOperation operation)
noexcept = 0;
1191 virtual IShuffleLayer* addShuffle(ITensor& input)
noexcept = 0;
1192 virtual int32_t getNbLayers() const noexcept = 0;
1193 virtual ILayer* getLayer(int32_t index) const noexcept = 0;
1194 virtual int32_t getNbInputs() const noexcept = 0;
1195 virtual ITensor* getInput(int32_t index) const noexcept = 0;
1196 virtual int32_t getNbOutputs() const noexcept = 0;
1197 virtual ITensor* getOutput(int32_t index) const noexcept = 0;
1198 virtual IReduceLayer* addReduce(
1199 ITensor& input,
ReduceOperation operation, uint32_t reduceAxes,
bool keepDimensions) noexcept
1201 virtual ITopKLayer* addTopK(ITensor& input,
TopKOperation op, int32_t k, uint32_t reduceAxes) noexcept = 0;
1202 virtual IGatherLayer* addGather(ITensor& data, ITensor& indices, int32_t axis) noexcept = 0;
1203 virtual IRaggedSoftMaxLayer* addRaggedSoftMax(ITensor& input, ITensor& bounds) noexcept = 0;
1204 virtual IMatrixMultiplyLayer* addMatrixMultiply(
1206 virtual IConstantLayer* addConstant(
Dims const& dimensions, Weights weights) noexcept = 0;
1207 virtual IIdentityLayer* addIdentity(ITensor& input) noexcept = 0;
1208 virtual
void removeTensor(ITensor& tensor) noexcept = 0;
1209 virtual
void unmarkOutput(ITensor& tensor) noexcept = 0;
1210 virtual IPluginV2Layer* addPluginV2(ITensor* const* inputs, int32_t nbInputs, IPluginV2& plugin) noexcept = 0;
1211 virtual IPluginV3Layer* addPluginV3(ITensor* const* inputs, int32_t nbInputs, ITensor* const* shapeInputs,
1212 int32_t nbShapeInputs,
IPluginV3& plugin) noexcept = 0;
1213 virtual ISliceLayer* addSlice(ITensor& input,
Dims const& start,
Dims const& size,
Dims const& stride) noexcept = 0;
1214 virtual
void setName(
char const* name) noexcept = 0;
1215 virtual
char const* getName() const noexcept = 0;
1216 virtual IShapeLayer* addShape(ITensor& input) noexcept = 0;
1217 virtual
bool hasImplicitBatchDimension() const noexcept = 0;
1218 virtual
bool markOutputForShapes(ITensor& tensor) noexcept = 0;
1219 virtual
bool unmarkOutputForShapes(ITensor& tensor) noexcept = 0;
1220 virtual IParametricReLULayer* addParametricReLU(ITensor& input, ITensor& slope) noexcept = 0;
1221 virtual IConvolutionLayer* addConvolutionNd(
1222 ITensor& input, int64_t nbOutputMaps,
Dims const& kernelSize, Weights kernelWeights, Weights biasWeights) noexcept
1224 virtual IPoolingLayer* addPoolingNd(ITensor& input,
PoolingType type,
Dims const& windowSize) noexcept = 0;
1225 virtual IDeconvolutionLayer* addDeconvolutionNd(
1226 ITensor& input, int64_t nbOutputMaps,
Dims const& kernelSize, Weights kernelWeights, Weights biasWeights) noexcept
1228 virtual IScaleLayer* addScaleNd(
1229 ITensor& input,
ScaleMode mode, Weights shift, Weights scale, Weights power, int32_t channelAxis) noexcept = 0;
1230 virtual IResizeLayer* addResize(ITensor& input) noexcept = 0;
1231 virtual ILoop* addLoop() noexcept = 0;
1232 virtual ISelectLayer* addSelect(ITensor& condition, ITensor& thenInput, ITensor& elseInput) noexcept = 0;
1233 virtual IFillLayer* addFill(
Dims const& dimensions,
FillOperation op) noexcept = 0;
1234 virtual IPaddingLayer* addPaddingNd(ITensor& input,
Dims const& prePadding,
Dims const& postPadding) noexcept = 0;
1235 virtual
bool setWeightsName(Weights weights,
char const* name) noexcept = 0;
1236 virtual
void setErrorRecorder(
IErrorRecorder* recorder) noexcept = 0;
1238 virtual IDequantizeLayer* addDequantize(ITensor& input, ITensor& scale) noexcept = 0;
1239 virtual IQuantizeLayer* addQuantize(ITensor& input, ITensor& scale) noexcept = 0;
1240 virtual IGatherLayer* addGatherV2(ITensor& data, ITensor& indices,
GatherMode mode) noexcept = 0;
1241 virtual IIfConditional* addIfConditional() noexcept = 0;
1242 virtual IScatterLayer* addScatter(ITensor& data, ITensor& indices, ITensor& updates,
ScatterMode mode) noexcept = 0;
1243 virtual IEinsumLayer* addEinsum(ITensor* const* inputs, int32_t nbInputs,
char const* equation) noexcept = 0;
1244 virtual IAssertionLayer* addAssertion(ITensor& condition,
char const* message) noexcept = 0;
1245 virtual IOneHotLayer* addOneHot(ITensor& indices, ITensor& values, ITensor& depth, int32_t axis) noexcept = 0;
1246 virtual INonZeroLayer* addNonZero(ITensor& input) noexcept = 0;
1247 virtual IGridSampleLayer* addGridSample(ITensor& input, ITensor& grid) noexcept = 0;
1248 virtual INMSLayer* addNMS(ITensor& boxes, ITensor& scores, ITensor& maxOutputBoxesPerClass) noexcept = 0;
1249 virtual IReverseSequenceLayer* addReverseSequence(ITensor& input, ITensor& sequenceLens) noexcept = 0;
1250 virtual INormalizationLayer* addNormalization(
1251 ITensor& input, ITensor& scale, ITensor& bias, uint32_t axesMask) noexcept = 0;
1252 virtual ICastLayer* addCast(ITensor& input,
DataType toType) noexcept = 0;
1253 virtual IBuilder& getBuilder() const noexcept = 0;
1256 virtual IQuantizeLayer* addQuantizeV2(ITensor& input, ITensor& scale,
DataType outputType) noexcept = 0;
1257 virtual IDequantizeLayer* addDequantizeV2(ITensor& input, ITensor& scale,
DataType outputType) noexcept = 0;
1259 virtual
bool markDebug(ITensor& tensor) noexcept = 0;
1260 virtual
bool unmarkDebug(ITensor& tensor) noexcept = 0;
1261 virtual
bool isDebugTensor(ITensor const& tensor) const noexcept = 0;
1262 virtual
bool markWeightsRefittable(
char const* name) noexcept = 0;
1263 virtual
bool unmarkWeightsRefittable(
char const* name) noexcept = 0;
1264 virtual
bool areWeightsMarkedRefittable(
char const* name) const noexcept = 0;
1265 virtual ISqueezeLayer* addSqueeze(ITensor& input, ITensor& axes) noexcept = 0;
1266 virtual IUnsqueezeLayer* addUnsqueeze(ITensor& input, ITensor& axes) noexcept = 0;
1267 virtual IDynamicQuantizeLayer* addDynamicQuantize(
1268 ITensor& input, int32_t axis, int32_t blockSize,
DataType toType,
DataType scaleType) noexcept = 0;
1269 virtual ICumulativeLayer* addCumulative(
1270 ITensor& input, ITensor& axis,
CumulativeOperation operation,
bool exclusive,
bool reverse) noexcept = 0;
1271 virtual
bool markUnfusedTensorsAsDebugTensors() noexcept = 0;
1272 virtual
bool unmarkUnfusedTensorsAsDebugTensors() noexcept = 0;
1273 virtual ITopKLayer* addTopKV2(
1274 ITensor& input,
TopKOperation op, int32_t k, uint32_t reduceAxes,
DataType indicesType) noexcept = 0;
1275 virtual INonZeroLayer* addNonZeroV2(ITensor& input,
DataType indicesType) noexcept = 0;
1276 virtual INMSLayer* addNMSV2(
1277 ITensor& boxes, ITensor& scores, ITensor& maxOutputBoxesPerClass,
DataType indicesType) noexcept = 0;
1278 virtual IAttention* addAttention(
1280 virtual IRotaryEmbeddingLayer* addRotaryEmbedding(ITensor& input, ITensor& cosCache, ITensor& sinCache,
1281 bool interleaved, int32_t rotaryEmbeddingDim) noexcept = 0;
1282 virtual IDynamicQuantizeLayer* addDynamicQuantizeV2(
1284 virtual IKVCacheUpdateLayer* addKVCacheUpdate(
1285 ITensor& cache, ITensor& update, ITensor& writeIndices,
KVCacheMode cacheMode) noexcept = 0;
1286 virtual INormalizationLayer* addNormalizationV2(
1287 ITensor& input, ITensor& scale, ITensor& bias, uint32_t axesMask) noexcept = 0;
1288 virtual IMoELayer* addMoE(
1289 ITensor& hiddenStates, ITensor& selectedExpertsForTokens, ITensor& scoresForSelectedExperts) noexcept = 0;
1290 virtual IDistCollectiveLayer* addDistCollective(ITensor& input,
CollectiveOperation distCollectiveOp,
1291 ReduceOperation reduceOp, int64_t root, int64_t* groups, int64_t groupSize) noexcept = 0;
1294class VAlgorithmIOInfo : public VRoot
1297 virtual DataType getDataType() const noexcept = 0;
1298 virtual
Dims getStrides() const noexcept = 0;
1299 virtual int64_t getVectorizedDim() const noexcept = 0;
1300 virtual int64_t getComponentsPerElement() const noexcept = 0;
1303class VAlgorithmVariant : public VRoot
1306 virtual int64_t getImplementation() const noexcept = 0;
1307 virtual int64_t getTactic() const noexcept = 0;
1310class VAlgorithmContext : public VRoot
1313 virtual char const* getName() const noexcept = 0;
1315 virtual int32_t getNbInputs() const noexcept = 0;
1316 virtual int32_t getNbOutputs() const noexcept = 0;
1319class VAlgorithm : public VRoot
1322 virtual IAlgorithmVariant
const& getAlgorithmVariant() const noexcept = 0;
1323 virtual
float getTimingMSec() const noexcept = 0;
1324 virtual std::
size_t getWorkspaceSize() const noexcept = 0;
1325 virtual IAlgorithmIOInfo const* getAlgorithmIOInfoByIndex(int32_t index) const noexcept = 0;
1329class VTimingCache : public VRoot
1333 virtual
bool combine(ITimingCache const& inputCache,
bool ignoreMismatch) noexcept = 0;
1334 virtual
bool reset() noexcept = 0;
1335 virtual int64_t queryKeys(TimingCacheKey* keyBuffer, int64_t capacity) const noexcept = 0;
1336 virtual TimingCacheValue query(TimingCacheKey const& key) const noexcept = 0;
1337 virtual
bool update(TimingCacheKey const& key, TimingCacheValue const& value) noexcept = 0;
1340class VBuilderConfig : public VRoot
1343 virtual void setAvgTimingIterations(int32_t avgTiming)
noexcept = 0;
1344 virtual int32_t getAvgTimingIterations() const noexcept = 0;
1345 virtual
void setEngineCapability(
EngineCapability capability) noexcept = 0;
1347 virtual
void setInt8Calibrator(IInt8Calibrator* calibrator) noexcept = 0;
1348 virtual IInt8Calibrator* getInt8Calibrator() const noexcept = 0;
1349 virtual
void setFlags(
BuilderFlags builderFlags) noexcept = 0;
1351 virtual
void clearFlag(
BuilderFlag builderFlag) noexcept = 0;
1352 virtual
void setFlag(
BuilderFlag builderFlag) noexcept = 0;
1353 virtual
bool getFlag(
BuilderFlag builderFlag) const noexcept = 0;
1354 virtual
void setDeviceType(ILayer const* layer,
DeviceType deviceType) noexcept = 0;
1355 virtual
DeviceType getDeviceType(ILayer const* layer) const noexcept = 0;
1356 virtual
bool isDeviceTypeSet(ILayer const* layer) const noexcept = 0;
1357 virtual
void resetDeviceType(ILayer const* layer) noexcept = 0;
1358 virtual
bool canRunOnDLA(ILayer const* layer) const noexcept = 0;
1359 virtual
void setDLACore(int32_t dlaCore) noexcept = 0;
1360 virtual int32_t getDLACore() const noexcept = 0;
1361 virtual
void setDefaultDeviceType(
DeviceType deviceType) noexcept = 0;
1362 virtual
DeviceType getDefaultDeviceType() const noexcept = 0;
1363 virtual
void reset() noexcept = 0;
1364 virtual
void setProfileStream(const cudaStream_t stream) noexcept = 0;
1365 virtual cudaStream_t getProfileStream() const noexcept = 0;
1366 virtual int32_t addOptimizationProfile(IOptimizationProfile const* profile) noexcept = 0;
1367 virtual int32_t getNbOptimizationProfiles() const noexcept = 0;
1372 virtual
bool setCalibrationProfile(IOptimizationProfile const* profile) noexcept = 0;
1373 virtual IOptimizationProfile const* getCalibrationProfile() noexcept = 0;
1378 virtual
bool getQuantizationFlag(
QuantizationFlag flag) const noexcept = 0;
1379 virtual
bool setTacticSources(
TacticSources tacticSources) noexcept = 0;
1380 virtual
TacticSources getTacticSources() const noexcept = 0;
1381 virtual
nvinfer1::ITimingCache* createTimingCache(
void const* blob, std::
size_t size) const noexcept = 0;
1382 virtual
bool setTimingCache(ITimingCache const& cache,
bool ignoreMismatch) noexcept = 0;
1383 virtual
nvinfer1::ITimingCache const* getTimingCache() const noexcept = 0;
1384 virtual
void setMemoryPoolLimit(
MemoryPoolType pool, std::
size_t poolSize) noexcept = 0;
1385 virtual std::
size_t getMemoryPoolLimit(
MemoryPoolType pool) const noexcept = 0;
1386 virtual
void setPreviewFeature(
PreviewFeature feature,
bool enable) noexcept = 0;
1387 virtual
bool getPreviewFeature(
PreviewFeature feature) const noexcept = 0;
1388 virtual
void setBuilderOptimizationLevel(int32_t level) noexcept = 0;
1389 virtual int32_t getBuilderOptimizationLevel() const noexcept = 0;
1392 virtual
void setPluginsToSerialize(
char const* const* paths, int32_t nbPaths) noexcept = 0;
1393 virtual
char const* getPluginToSerialize(int32_t index) const noexcept = 0;
1394 virtual int32_t getNbPluginsToSerialize() const noexcept = 0;
1395 virtual
void setMaxAuxStreams(int32_t nbStreams) noexcept = 0;
1396 virtual int32_t getMaxAuxStreams() const noexcept = 0;
1399 virtual
void setRuntimePlatform(
RuntimePlatform runtimePlatform) noexcept = 0;
1401 virtual
void setMaxNbTactics(int32_t maxTactics) noexcept = 0;
1402 virtual int32_t getMaxNbTactics() const noexcept = 0;
1405 virtual
bool setL2LimitForTiling(int64_t size) noexcept = 0;
1406 virtual int64_t getL2LimitForTiling() const noexcept = 0;
1407 virtual
bool setRemoteAutoTuningConfig(
char const* config) noexcept = 0;
1408 virtual
char const* getRemoteAutoTuningConfig() const noexcept = 0;
1411class VSerializationConfig : public VRoot
1421class VBuilder : public VRoot
1424 virtual bool platformHasFastFp16() const noexcept = 0;
1425 virtual
bool platformHasFastInt8() const noexcept = 0;
1426 virtual int32_t getMaxDLABatchSize() const noexcept = 0;
1427 virtual int32_t getNbDLACores() const noexcept = 0;
1428 virtual
void setGpuAllocator(
IGpuAllocator* allocator) noexcept = 0;
1429 virtual
nvinfer1::IBuilderConfig* createBuilderConfig() noexcept = 0;
1431 virtual
nvinfer1::IOptimizationProfile* createOptimizationProfile() noexcept = 0;
1432 virtual
void setErrorRecorder(
IErrorRecorder* recorder) noexcept = 0;
1434 virtual
void reset() noexcept = 0;
1435 virtual
bool platformHasTf32() const noexcept = 0;
1436 virtual
nvinfer1::IHostMemory* buildSerializedNetwork(
1437 INetworkDefinition& network, IBuilderConfig& config) noexcept = 0;
1438 virtual
bool isNetworkSupported(INetworkDefinition const& network, IBuilderConfig const& config) const noexcept = 0;
1439 virtual ILogger*
getLogger() const noexcept = 0;
1440 virtual
bool setMaxThreads(int32_t maxThreads) noexcept = 0;
1441 virtual int32_t getMaxThreads() const noexcept = 0;
1443 virtual ICudaEngine* buildEngineWithConfig(INetworkDefinition& network, IBuilderConfig& config) noexcept = 0;
1444 virtual
bool buildSerializedNetworkToStream(
1445 INetworkDefinition& network, IBuilderConfig& config,
IStreamWriter& writer) noexcept = 0;
1446 virtual
nvinfer1::IHostMemory* buildSerializedNetworkWithKernelText(
1447 INetworkDefinition& network, IBuilderConfig& config, IHostMemory*& kernelText) noexcept
1451class VRuntimeConfig : public VRoot
1454 virtual IRuntimeConfig* getPImpl() noexcept = 0;
1460class VDistCollectiveLayer : public VRoot
TENSORRTAPI nvinfer1::IPluginRegistry * getPluginRegistry() noexcept
Return the plugin registry.
TENSORRTAPI nvinfer1::ILogger * getLogger() noexcept
Return the logger object.
#define TRT_NODISCARD
A stand-in for [[nodiscard]] and [[nodiscard(REASON)]] that works with older compilers.
Definition: NvInferRuntimeBase.h:57
#define TRT_DEPRECATED
Definition: NvInferRuntimeBase.h:42
Structure to define the dimensions of a tensor.
Interface implemented by application for selecting and reporting algorithms of a layer provided by th...
User-implemented callback for notification when value of a debug tensor is updated.
Reference counted application-implemented error reporting interface for TensorRT objects.
Application-implemented class for controlling allocation on the GPU.
Callback from ExecutionContext::enqueueV3()
Plugin class for the V3 generation of user-implemented layers.
Application-implemented interface for profiling.
Application-implemented progress reporting interface for TensorRT.
Application-implemented class for reading data in a stream-based manner.
Application-implemented class for reading data in a stream-based manner asynchronously....
Application-implemented class for writing data in a stream-based manner.
Class to handle library allocated memory that is accessible to the user.
Definition: NvInferRuntime.h:142
The TensorRT API version 1 namespace.
Definition: NvInferSafePlugin.h:33
uint32_t TacticSources
Represents a collection of one or more TacticSource values combine using bitwise-OR operations.
Definition: NvInferRuntime.h:2961
v_1_0::IOutputAllocator IOutputAllocator
Definition: NvInferRuntime.h:4264
ResizeSelector
The coordinate selector when resize to single pixel output.
Definition: NvInfer.h:4181
EngineCapability
List of supported engine capability flows.
Definition: NvInferRuntime.h:76
DimensionOperation
An operation on two IDimensionExpr, which represent integer expressions used in dimension computation...
Definition: NvInferRuntime.h:179
MemoryPoolType
The type for memory pools used by TensorRT.
Definition: NvInfer.h:10841
ScaleMode
Controls how shift, scale and power are applied in a Scale layer.
Definition: NvInfer.h:1889
RuntimePlatform
Describes the intended runtime platform (operating system and CPU architecture) for the execution of ...
Definition: NvInfer.h:10421
uint32_t QuantizationFlags
Represents one or more QuantizationFlag values using binary OR operations.
Definition: NvInfer.h:10373
TensorIOMode
Definition of tensor IO Mode.
Definition: NvInferRuntimeBase.h:659
HardwareCompatibilityLevel
Describes requirements of compatibility with GPU architectures other than that of the GPU on which th...
Definition: NvInfer.h:10969
CumulativeOperation
Enumerates the cumulative operations that may be performed by a Cumulative layer.
Definition: NvInfer.h:6790
SerializationFlag
List of valid flags that the engine can enable when serializing the bytes.
Definition: NvInferRuntime.h:3002
BoundingBoxFormat
Representation of bounding box data used for the Boxes input tensor in INMSLayer.
Definition: NvInfer.h:6328
UnaryOperation
Enumerates the unary operations that may be performed by a Unary layer.
Definition: NvInfer.h:2834
v_1_0::IStreamWriter IStreamWriter
Definition: NvInferRuntime.h:720
v_1_0::IProfiler IProfiler
Definition: NvInferRuntime.h:1318
v_1_0::IAlgorithmSelector IAlgorithmSelector
Definition: NvInfer.h:10365
v_1_0::IStreamReaderV2 IStreamReaderV2
Definition: NvInferRuntime.h:797
ActivationType
Enumerates the types of activation to perform in an activation layer.
Definition: NvInfer.h:143
uint32_t TempfileControlFlags
Represents a collection of one or more TempfileControlFlag values combined using bitwise-OR operation...
Definition: NvInferRuntime.h:1396
FillOperation
Enumerates the tensor fill operations that may performed by a fill layer.
Definition: NvInfer.h:5162
EngineStat
The kind of engine statistics that queried from the ICudaEngine.
Definition: NvInferRuntime.h:3170
ResizeRoundMode
The rounding mode for nearest neighbor resize.
Definition: NvInfer.h:4211
PaddingMode
Enumerates the modes of padding to perform in convolution, deconvolution and pooling layer,...
Definition: NvInfer.h:1067
TripLimit
Enum that describes kinds of trip limits.
Definition: NvInfer.h:4570
uint32_t NetworkDefinitionCreationFlags
Represents one or more NetworkDefinitionCreationFlag flags using binary OR operations....
Definition: NvInfer.h:12103
PreviewFeature
Define preview features.
Definition: NvInfer.h:10916
TilingOptimizationLevel
Define the optimization levels for Tiling.
Definition: NvInfer.h:11022
DataType
The type of weights and tensors. The datatypes other than kBOOL, kINT32, and kINT64 are "activation d...
Definition: NvInferRuntimeBase.h:146
uint32_t BuilderFlags
Represents one or more BuilderFlag values using binary OR operations, e.g., 1U << BuilderFlag::kFP16 ...
Definition: NvInfer.h:10453
DeviceType
The device that this layer/network will execute on.
Definition: NvInferRuntime.h:1350
CalibrationAlgoType
Version of calibration algorithm to use.
Definition: NvInfer.h:9799
LayerType
The type values of layer classes.
Definition: NvInfer.h:58
SampleMode
Controls how ISliceLayer and IGridSample handle out-of-bounds coordinates.
Definition: NvInfer.h:3325
GatherMode
Control form of IGatherLayer.
Definition: NvInfer.h:2581
v_1_0::IDebugListener IDebugListener
Definition: NvInferRuntime.h:4305
MoEActType
Enumerates the activation type for the MoE layer.
Definition: NvInfer.h:7554
WeightsRole
How a layer uses particular Weights.
Definition: NvInferRuntime.h:1328
uint32_t TensorFormats
It is capable of representing one or more TensorFormat by binary OR operations, e....
Definition: NvInfer.h:135
ProfilingVerbosity
List of verbosity levels of layer information exposed in NVTX annotations and in IEngineInspector.
Definition: NvInferRuntime.h:2973
NetworkDefinitionCreationFlag
List of immutable network properties expressed at network creation time. NetworkDefinitionCreationFla...
Definition: NvInfer.h:12114
ElementWiseOperation
Enumerates the binary operations that may be performed by an ElementWise layer.
Definition: NvInfer.h:2491
QuantizationFlag
List of valid flags for quantizing the network to int8.
Definition: NvInfer.h:10385
CollectiveOperation
Enumerates the collective operations that may be performed by a DistCollective layer.
Definition: NvInfer.h:2962
uint32_t SerializationFlags
Represents one or more SerializationFlag values using binary OR operations, e.g., 1U << Serialization...
Definition: NvInferRuntime.h:2992
InterpolationMode
Enumerates various modes of interpolation.
Definition: NvInfer.h:4099
BuilderFlag
List of valid modes that the builder can enable when creating an engine from a network definition.
Definition: NvInfer.h:10463
TensorFormat
Format of the input/output tensors.
Definition: NvInferRuntime.h:1430
ExecutionContextAllocationStrategy
Different memory allocation behaviors for IExecutionContext.
Definition: NvInferRuntime.h:3107
TopKOperation
Enumerates the operations that may be performed by a TopK layer.
Definition: NvInfer.h:3606
ReduceOperation
Enumerates the reduce operations that may be performed by a Reduce layer.
Definition: NvInfer.h:2934
LayerInformationFormat
The format in which the IEngineInspector prints the layer information.
Definition: NvInferRuntime.h:5279
ScatterMode
Control form of IScatterLayer.
Definition: NvInfer.h:6062
MatrixOperation
Enumerates the operations that may be performed on a tensor by IMatrixMultiplyLayer before multiplica...
Definition: NvInfer.h:3759
ResizeCoordinateTransformation
The resize coordinate transformation function.
Definition: NvInfer.h:4127
LoopOutput
Enum that describes kinds of loop outputs.
Definition: NvInfer.h:4542
KVCacheMode
Enumerates the KVCache modes that may be performed by a KVCacheUpdate layer.
Definition: NvInfer.h:7466
v_1_0::IStreamReader IStreamReader
Definition: NvInferRuntime.h:710
PoolingType
The type of pooling to perform in a pooling layer.
Definition: NvInfer.h:1503
v_1_0::IProgressMonitor IProgressMonitor
Definition: NvInfer.h:11138
TensorLocation
The location for tensor data storage, device or host.
Definition: NvInferRuntime.h:204
OptProfileSelector
When setting or querying optimization profile parameters (such as shape tensor inputs or dynamic dime...
Definition: NvInferRuntime.h:2635
AttentionNormalizationOp
Enumerates the operations that may be performed by the normalization in the attention subgraph.
Definition: NvInfer.h:6925
nvinfer1::IPluginV3 IPluginV3
Definition: NvInferForwardDecl.h:47