20 #if defined(ARMCOMPUTENEON_ENABLED) 87 template<
typename ... Args>
88 bool IsNeonBackendSupported(Optional<std::string&> reasonIfUnsupported, Args... args)
91 #if defined(ARMCOMPUTENEON_ENABLED) 94 SetValueChecked(reasonIfUnsupported,
"The armnn library has been built without NEON support");
99 template<
typename FloatFunc,
typename Uint8Func,
typename ... Params>
100 bool IsSupportedForDataTypeNeon(Optional<std::string&> reasonIfUnsupported,
102 FloatFunc floatFuncPtr,
103 Uint8Func uint8FuncPtr,
106 return IsNeonBackendSupported(reasonIfUnsupported) &&
114 std::forward<Params>(params)...);
117 #if defined(ARMCOMPUTENEON_ENABLED) 118 template<
class FuncType,
class... Args>
119 inline bool IsWorkloadSupported(FuncType& func, Optional<std::string&> reasonIfUnsupported, Args&&... args)
122 const bool supported = (aclStatus.error_code() == arm_compute::ErrorCode::OK);
123 if (!supported && reasonIfUnsupported)
125 reasonIfUnsupported.value() = aclStatus.error_description();
130 #define FORWARD_WORKLOAD_VALIDATE_FUNC(func, reasonIfUnsupported, ...) \ 131 return IsWorkloadSupported(func, reasonIfUnsupported, __VA_ARGS__); 133 #define FORWARD_WORKLOAD_VALIDATE_FUNC(func, reasonIfUnsupported, ...) \ 134 return IsNeonBackendSupported(reasonIfUnsupported, __VA_ARGS__); 139 : m_ModelContextPtr(modelContextPtr)
144 : m_ModelContextPtr(nullptr)
149 const std::vector<TensorInfo>& infos,
160 *(PolymorphicDowncast<const ActivationDescriptor*>(&descriptor)),
161 reasonIfUnsupported);
167 *(PolymorphicDowncast<const ArgMinMaxDescriptor*>(&descriptor)),
168 reasonIfUnsupported);
176 *(PolymorphicDowncast<const BatchNormalizationDescriptor*>
178 reasonIfUnsupported);
182 *(PolymorphicDowncast<const BatchToSpaceNdDescriptor*>(&descriptor)),
183 reasonIfUnsupported);
188 *(PolymorphicDowncast<const ComparisonDescriptor*>(&descriptor)),
189 reasonIfUnsupported);
192 std::vector<const TensorInfo*> inputInfos;
193 for (uint32_t i = 0; i < (infos.size() - 1); i++)
195 inputInfos.push_back(&infos[i]);
198 infos[infos.size() - 1],
199 *(PolymorphicDowncast<const OriginsDescriptor*>(&
descriptor)),
214 if (infos.size() != 4)
217 "TensorInfos should be of format: {input, output, weights, biases}.");
220 auto desc = *(PolymorphicDowncast<const Convolution2dDescriptor*>(&
descriptor));
228 reasonIfUnsupported);
237 reasonIfUnsupported);
243 *(PolymorphicDowncast<const DepthToSpaceDescriptor*>(&descriptor)),
244 reasonIfUnsupported);
247 if (infos.size() != 4)
250 "TensorInfos should be of format: {input, output, weights, biases}.");
253 auto desc = *(PolymorphicDowncast<const DepthwiseConvolution2dDescriptor*>(&
descriptor));
261 reasonIfUnsupported);
270 reasonIfUnsupported);
280 *(PolymorphicDowncast<const ElementwiseUnaryDescriptor*>(&descriptor)),
281 reasonIfUnsupported);
285 *(PolymorphicDowncast<const FillDescriptor*>(&descriptor)),
286 reasonIfUnsupported);
294 *(PolymorphicDowncast<const FullyConnectedDescriptor*>(&descriptor)),
295 reasonIfUnsupported);
300 *(PolymorphicDowncast<const GatherDescriptor*>(&descriptor)),
301 reasonIfUnsupported);
307 *(PolymorphicDowncast<const InstanceNormalizationDescriptor*>
309 reasonIfUnsupported);
313 *(PolymorphicDowncast<const L2NormalizationDescriptor*>(&descriptor)),
314 reasonIfUnsupported);
319 *(PolymorphicDowncast<const LogicalBinaryDescriptor*>(&descriptor)),
320 reasonIfUnsupported);
324 *(PolymorphicDowncast<const LogSoftmaxDescriptor*>(&descriptor)),
325 reasonIfUnsupported);
334 *(PolymorphicDowncast<const LstmDescriptor*>(&descriptor)),
335 lstmParamsInfo.
value(),
344 *(PolymorphicDowncast<const QLstmDescriptor*>(&descriptor)),
345 lstmParamsInfo.
value(),
352 *(PolymorphicDowncast<const MeanDescriptor*>(&descriptor)),
353 reasonIfUnsupported);
361 *(PolymorphicDowncast<const NormalizationDescriptor*>(&descriptor)),
362 reasonIfUnsupported);
368 *(PolymorphicDowncast<const PadDescriptor*>(&descriptor)),
369 reasonIfUnsupported);
373 *(PolymorphicDowncast<const PermuteDescriptor*>(&descriptor)),
374 reasonIfUnsupported);
378 *(PolymorphicDowncast<const Pooling2dDescriptor*>(&descriptor)),
379 reasonIfUnsupported);
390 quantizedLstmParamsInfo.
value(),
395 *(PolymorphicDowncast<const ReshapeDescriptor*>(&descriptor)),
396 reasonIfUnsupported);
400 *(PolymorphicDowncast<const ResizeDescriptor*>(&descriptor)),
401 reasonIfUnsupported);
405 *(PolymorphicDowncast<const ReduceDescriptor*>(&descriptor)),
406 reasonIfUnsupported);
410 *(PolymorphicDowncast<const SliceDescriptor*>(&descriptor)),
411 reasonIfUnsupported);
415 *(PolymorphicDowncast<const SoftmaxDescriptor*>(&descriptor)),
416 reasonIfUnsupported);
420 *(PolymorphicDowncast<const SpaceToBatchNdDescriptor*>(&descriptor)),
421 reasonIfUnsupported);
425 *(PolymorphicDowncast<const SpaceToDepthDescriptor*>(&descriptor)),
426 reasonIfUnsupported);
429 std::vector<TensorInfo> outputInfos;
430 for (uint32_t i = 1; i < infos.size(); i++)
432 outputInfos.push_back(infos[i]);
435 {outputInfos.begin(), outputInfos.end()},
436 *(PolymorphicDowncast<const ViewsDescriptor*>(&
descriptor)),
441 std::vector<const TensorInfo*> inputInfos;
442 for (uint32_t i = 0; i < infos.size() - 1; i++)
444 inputInfos.push_back(&infos[i]);
447 infos[infos.size() - 1],
448 *(PolymorphicDowncast<const StackDescriptor*>(&
descriptor)),
454 *(PolymorphicDowncast<const StridedSliceDescriptor*>(&descriptor)),
455 reasonIfUnsupported);
461 *(PolymorphicDowncast<const TransposeDescriptor*>(&descriptor)),
462 reasonIfUnsupported);
465 if (infos.size() != 4)
468 "TensorInfos should be of format: {input, output, weights, biases}.");
471 auto desc = *(PolymorphicDowncast<const TransposeConvolution2dDescriptor*>(&
descriptor));
479 reasonIfUnsupported);
488 reasonIfUnsupported);
496 *(PolymorphicDowncast<const ChannelShuffleDescriptor*>(&descriptor)),
497 reasonIfUnsupported);
500 if (infos.size() != 4)
503 "TensorInfos should be of format: {input, output, weights, biases}.");
506 auto desc = *(PolymorphicDowncast<const Convolution3dDescriptor*>(&
descriptor));
514 reasonIfUnsupported);
523 reasonIfUnsupported);
532 auto desc = *(PolymorphicDowncast<const DetectionPostProcessDescriptor*>(&
descriptor));
541 reasonIfUnsupported);
551 reasonIfUnsupported);
557 reasonIfUnsupported);
681 SetValueChecked(reasonIfUnsupported,
"Neon Concat: Concat axis > Number of dimensions.");
686 if(concatInnerAxis < 3)
694 else if (concatInnerAxis == 3)
696 for (
auto& input : inputs)
700 SetValueChecked(reasonIfUnsupported,
"Neon Concat: Types and quantization parameters must match.");
708 SetValueChecked(reasonIfUnsupported,
"Neon Concat: Maximum of 4 dimensions supported.");
768 bool isFastMathEnabled =
false;
769 #if defined(ARMCOMPUTENEON_ENABLED) 770 if (m_ModelContextPtr)
772 if (m_ModelContextPtr.get() !=
nullptr)
801 bool isFastMathEnabled =
false;
802 #if defined(ARMCOMPUTENEON_ENABLED) 803 if (m_ModelContextPtr)
805 if (m_ModelContextPtr.get() !=
nullptr)
939 return IsNeonBackendSupported(reasonIfUnsupported);
947 return IsNeonBackendSupported(reasonIfUnsupported) &&
991 return IsNeonBackendSupported(reasonIfUnsupported, input);
1000 reasonIfUnsupported,
1024 reasonIfUnsupported,
1030 reasonIfUnsupported,
1059 reasonIfUnsupported,
1077 reasonIfUnsupported,
1089 reasonIfUnsupported,
1101 reasonIfUnsupported,
1113 reasonIfUnsupported,
1126 reasonIfUnsupported,
1139 reasonIfUnsupported,
1148 return IsNeonBackendSupported(reasonIfUnsupported, output);
1157 reasonIfUnsupported,
1206 reasonIfUnsupported,
1208 previousCellStateIn,
1227 reasonIfUnsupported,
1241 reasonIfUnsupported,
1256 reasonIfUnsupported,
1269 reasonIfUnsupported,
1280 reasonIfUnsupported,
1292 reasonIfUnsupported,
1312 reasonIfUnsupported,
1324 reasonIfUnsupported,
1331 const std::vector<std::reference_wrapper<TensorInfo>>&
outputs,
1335 #if defined(ARMCOMPUTENEON_ENABLED) 1345 reasonIfUnsupported,
1348 *splitAxis.begin());
1356 SetValueChecked(reasonIfUnsupported,
"Neon Splitter: Types and quantization parameters must match.");
1369 reasonIfUnsupported,
1381 reasonIfUnsupported,
1393 reasonIfUnsupported,
1408 reasonIfUnsupported,
arm_compute::Status NeonGatherWorkloadValidate(const TensorInfo &input, const TensorInfo &indices, const TensorInfo &output, const GatherDescriptor &descriptor)
arm_compute::Status NeonNegWorkloadValidate(const TensorInfo &input, const TensorInfo &output)
bool IsConcatSupported(const std::vector< const TensorInfo *> inputs, const TensorInfo &output, const OriginsDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
UnaryOperation m_Operation
Specifies the elementwiseUnary operation to execute.
bool IsSliceSupported(const TensorInfo &input, const TensorInfo &output, const SliceDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsSoftmaxSupported(const TensorInfo &input, const TensorInfo &output, const SoftmaxDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsConvolution2dSupported(const TensorInfo &input, const TensorInfo &output, const Convolution2dDescriptor &descriptor, const TensorInfo &weights, const Optional< TensorInfo > &biases, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
A ViewsDescriptor for the SplitterLayer.
bool IsTypeSpaceMatch(const TensorInfo &other) const
Check that the types are the same and, if quantize, that the quantization parameters are the same...
bool IsConvertFp32ToFp16Supported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonSpaceToDepthWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const SpaceToDepthDescriptor &descriptor)
arm_compute::Status NeonSplitterWorkloadValidate(const TensorInfo &input, const std::vector< std::reference_wrapper< TensorInfo >> &outputs, unsigned int splitAxis)
A TransposeConvolution2dDescriptor for the TransposeConvolution2dLayer.
const TensorInfo const TensorInfo const TensorInfo const TensorInfo const TensorInfo const TensorInfo const LstmDescriptor const LstmInputParamsInfo & paramsInfo
const TensorShape & GetShape() const
arm_compute::Status NeonLogSoftmaxWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const LogSoftmaxDescriptor &descriptor)
A ReshapeDescriptor for the ReshapeLayer.
bool IsGatherSupported(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, const GatherDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported) const override
arm_compute::Status NeonBatchNormalizationValidate(const TensorInfo &input, const TensorInfo &output, const TensorInfo &mean, const TensorInfo &var, const TensorInfo &beta, const TensorInfo &gamma, const BatchNormalizationDescriptor &descriptor, const ActivationDescriptor *activationDescriptor)
bool IsDetectionPostProcessSupported(const TensorInfo &boxEncodings, const TensorInfo &scores, const TensorInfo &anchors, const TensorInfo &detectionBoxes, const TensorInfo &detectionClasses, const TensorInfo &detectionScores, const TensorInfo &numDetections, const DetectionPostProcessDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsMemImportSupported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsDilatedDepthwiseConvolutionSupported(const TensorInfo &input, const TensorInfo &output, const DepthwiseConvolution2dDescriptor &descriptor, const TensorInfo &weights, const Optional< TensorInfo > &biases, Optional< std::string &> reason=EmptyOptional()) const override
bool IsQuantizedLstmSupported(const TensorInfo &input, const TensorInfo &cellStateIn, const TensorInfo &outputStateIn, const TensorInfo &cellStateOut, const TensorInfo &outputStateOut, const QuantizedLstmInputParamsInfo ¶msInfo, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
uint32_t GetNumDimensions() const
Get the number of dimensions.
A ComparisonDescriptor for the ComparisonLayer.
bool IsDepthToSpaceSupported(const TensorInfo &input, const TensorInfo &output, const DepthToSpaceDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
const TensorInfo const TensorInfo const TensorInfo const TensorInfo const TensorInfo & gamma
bool IsL2NormalizationSupported(const TensorInfo &input, const TensorInfo &output, const L2NormalizationDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsCastSupported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsStridedSliceSupported(const TensorInfo &input, const TensorInfo &output, const StridedSliceDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
const std::vector< std::reference_wrapper< TensorInfo > > & outputs
bool IsLayerSupported(const LayerType &type, const std::vector< TensorInfo > &infos, const BaseDescriptor &descriptor, const Optional< LstmInputParamsInfo > &lstmParamsInfo, const Optional< QuantizedLstmInputParamsInfo > &quantizedLstmParamsInfo, Optional< std::string &> reasonIfUnsupported) const override
bool IsConvertFp32ToBf16Supported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsAdditionSupported(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
A Convolution2dDescriptor for the Convolution2dLayer.
arm_compute::Status NeonDepthwiseConvolutionWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const DepthwiseConvolution2dDescriptor &descriptor, const TensorInfo &weights, const Optional< TensorInfo > &biases, const ActivationDescriptor *activationDescriptor)
arm_compute::Status NeonActivationWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const ActivationDescriptor &descriptor)
arm_compute::Status NeonMinimumWorkloadValidate(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output)
Validate function for validating the inputs and output.
arm_compute::Status NeonStridedSliceWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const StridedSliceDescriptor &descriptor)
const TensorInfo const ActivationDescriptor Optional< std::string & > reasonIfUnsupported
arm_compute::Status NeonTransposeConvolution2dWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const TransposeConvolution2dDescriptor &descriptor, const TensorInfo &weights, const Optional< TensorInfo > &biases)
arm_compute::Status NeonLstmFloatWorkloadValidate(const TensorInfo &input, const TensorInfo &outputStateIn, const TensorInfo &cellStateIn, const TensorInfo &scratchBuffer, const TensorInfo &outputStateOut, const TensorInfo &cellStateOut, const TensorInfo &output, const LstmDescriptor &descriptor, const LstmInputParamsInfo ¶msInfo)
bool IsShapeSupported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonQLstmWorkloadValidate(const TensorInfo &input, const TensorInfo &cellStateIn, const TensorInfo &outputStateIn, const TensorInfo &cellStateOut, const TensorInfo &outputStateOut, const TensorInfo &output, const QLstmDescriptor &descriptor, const LstmInputParamsInfo ¶msInfo)
arm_compute::Status NeonSliceWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const SliceDescriptor &descriptor)
arm_compute::Status NeonFullyConnectedWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const TensorInfo &weights, const Optional< TensorInfo > &biases, const FullyConnectedDescriptor &descriptor, const ActivationDescriptor *activationDescriptor)
bool IsFastMathEnabled() const
const TensorInfo const TensorInfo const TensorInfo const TensorInfo & outputStateOut
A LogicalBinaryDescriptor for the LogicalBinaryLayer.
bool IsConstantSupported(const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
Copyright (c) 2021 ARM Limited and Contributors.
arm_compute::Status NeonQuantizeWorkloadValidate(const TensorInfo &input, const TensorInfo &output)
bool IsPadSupported(const TensorInfo &input, const TensorInfo &output, const PadDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
void IgnoreUnused(Ts &&...)
const TensorInfo const ActivationDescriptor & descriptor
std::set< unsigned int > ComputeSplitAxis(const armnn::SplitterDescriptor &desc, const TensorShape &input)
arm_compute::Status NeonAdditionWorkloadValidate(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, const ActivationDescriptor *activationDescriptor)
A SpaceToDepthDescriptor for the SpaceToDepthLayer.
arm_compute::Status NeonLogWorkloadValidate(const TensorInfo &input, const TensorInfo &output)
arm_compute::Status NeonLogicalAndWorkloadValidate(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output)
arm_compute::Status NeonInstanceNormalizationWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const InstanceNormalizationDescriptor &descriptor)
bool IsPooling2dSupported(const TensorInfo &input, const TensorInfo &output, const Pooling2dDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonLogicalOrWorkloadValidate(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output)
LogicalBinaryOperation m_Operation
Specifies the logical operation to execute.
A BatchToSpaceNdDescriptor for the BatchToSpaceNdLayer.
bool IsActivationSupported(const TensorInfo &input, const TensorInfo &output, const ActivationDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
The NeonBackendModelContext is used to pass in Neon specific backend ModelOptions.
bool IsMultiplicationSupported(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
const TensorInfo & outputStateIn
const TensorInfo const TensorInfo & previousCellStateIn
bool IsComparisonSupported(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, const ComparisonDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
A ResizeBilinearDescriptor for the ResizeBilinearLayer.
arm_compute::Status NeonL2NormalizationWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const L2NormalizationDescriptor &descriptor)
arm_compute::Status NeonAbsWorkloadValidate(const TensorInfo &input, const TensorInfo &output)
Base class for all descriptors.
A StackDescriptor for the StackLayer.
arm_compute::Status NeonQuantizedLstmWorkloadValidate(const TensorInfo &input, const TensorInfo &cellStateIn, const TensorInfo &outputStateIn, const TensorInfo &cellStateOut, const TensorInfo &outputStateOut, const QuantizedLstmInputParamsInfo ¶msInfo)
arm_compute::Status NeonStackWorkloadValidate(const std::vector< const TensorInfo *> &inputs, const TensorInfo &output, const StackDescriptor &descriptor)
arm_compute::Status NeonMeanWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const MeanDescriptor &descriptor)
arm_compute::Status NeonReduceWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const ReduceDescriptor &descriptor)
arm_compute::Status NeonSpaceToBatchNdWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const SpaceToBatchNdDescriptor &descriptor)
arm_compute::Status NeonSubtractionWorkloadValidate(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, const ActivationDescriptor *activationDescriptor)
A PadDescriptor for the PadLayer.
bool IsLogicalBinarySupported(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, const LogicalBinaryDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported) const override
bool IsConvolution3dSupported(const TensorInfo &input, const TensorInfo &output, const Convolution3dDescriptor &descriptor, const TensorInfo &weights, const Optional< TensorInfo > &biases, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
const TensorInfo const TensorInfo & cellStateIn
bool IsFillSupported(const TensorInfo &input, const TensorInfo &output, const FillDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
An LstmDescriptor for the LstmLayer.
arm_compute::Status NeonExpWorkloadValidate(const TensorInfo &input, const TensorInfo &output)
arm_compute::Status NeonReshapeWorkloadValidate(const TensorInfo &input, const TensorInfo &output)
bool IsInputSupported(const TensorInfo &input, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonBatchToSpaceNdWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const BatchToSpaceNdDescriptor &descriptor)
bool IsStackSupported(const std::vector< const TensorInfo *> &inputs, const TensorInfo &output, const StackDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
std::shared_ptr< IBackendModelContext > IBackendSpecificModelContextPtr
A L2NormalizationDescriptor for the L2NormalizationLayer.
const TensorInfo const TensorInfo const TensorInfo & var
bool IsMinimumSupported(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsPreluSupported(const TensorInfo &input, const TensorInfo &alpha, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsOutputSupported(const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
An ArgMinMaxDescriptor for ArgMinMaxLayer.
bool IsInstanceNormalizationSupported(const TensorInfo &input, const TensorInfo &output, const InstanceNormalizationDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
DataType GetDataType() const
An OriginsDescriptor for the ConcatLayer.
A ReduceDescriptor for the REDUCE operators.
A FullyConnectedDescriptor for the FullyConnectedLayer.
arm_compute::Status NeonLogicalNotWorkloadValidate(const TensorInfo &input, const TensorInfo &output)
bool IsNormalizationSupported(const TensorInfo &input, const TensorInfo &output, const NormalizationDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsElementwiseUnarySupported(const TensorInfo &input, const TensorInfo &output, const ElementwiseUnaryDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonCastValidate(const TensorInfo &input, const TensorInfo &output)
A GatherDescriptor for the GatherLayer.
arm_compute::Status NeonChannelShuffleValidate(const TensorInfo &input, const TensorInfo &output, const ChannelShuffleDescriptor &descriptor)
arm_compute::Status NeonComparisonWorkloadValidate(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, const ComparisonDescriptor &descriptor)
bool IsDivisionSupported(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
const TensorInfo const TensorInfo const TensorInfo const TensorInfo & beta
arm_compute::Status NeonConcatWorkloadValidate(const std::vector< const TensorInfo *> &inputs, const TensorInfo &output, const OriginsDescriptor &descriptor)
arm_compute::Status NeonPermuteWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const PermuteDescriptor &descriptor)
bool IsFloorSupported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
A QLstmDescriptor for the QLstmLayer.
bool IsMergeSupported(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonConvolution2dWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const Convolution2dDescriptor &descriptor, const TensorInfo &weights, const Optional< TensorInfo > &biases, bool isFastMathEnabled, const ActivationDescriptor *activationDescriptor)
bool IsLogSoftmaxSupported(const TensorInfo &input, const TensorInfo &output, const LogSoftmaxDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
An ActivationDescriptor for the ActivationLayer.
bool IsResizeSupported(const TensorInfo &input, const TensorInfo &output, const ResizeDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsMemCopySupported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsTransposeConvolution2dSupported(const TensorInfo &input, const TensorInfo &output, const TransposeConvolution2dDescriptor &descriptor, const TensorInfo &weights, const Optional< TensorInfo > &biases, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
A SliceDescriptor for the SliceLayer.
A Convolution3dDescriptor for the Convolution3dLayer.
arm_compute::Status NeonDequantizeWorkloadValidate(const TensorInfo &input, const TensorInfo &output)
bool IsSplitterSupported(const TensorInfo &input, const std::vector< std::reference_wrapper< TensorInfo >> &outputs, const ViewsDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
const TensorInfo & previousOutputIn
bool IsReshapeSupported(const TensorInfo &input, const TensorInfo &output, const ReshapeDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonDivisionWorkloadValidate(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, const ActivationDescriptor *activationDescriptor)
bool IsBatchNormalizationSupported(const TensorInfo &input, const TensorInfo &output, const TensorInfo &mean, const TensorInfo &var, const TensorInfo &beta, const TensorInfo &gamma, const BatchNormalizationDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonConvolution3dWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const Convolution3dDescriptor &descriptor, const TensorInfo &weights, const Optional< TensorInfo > &biases, bool isFastMathEnabled, const ActivationDescriptor *activationDescriptor)
void SetValueChecked(Optional< T &> optionalRef, V &&val)
const TensorInfo & output
A SpaceToBatchNdDescriptor for the SpaceToBatchNdLayer.
arm_compute::Status NeonNormalizationWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const NormalizationDescriptor &descriptor)
bool IsQuantizeSupported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
EmptyOptional is used to initialize the Optional class in case we want to have default value for an O...
arm_compute::Status NeonRsqrtWorkloadValidate(const TensorInfo &input, const TensorInfo &output)
bool IsConvertFp16ToFp32Supported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonPadWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const PadDescriptor &descriptor)
A ElementwiseUnaryDescriptor for the ElementwiseUnaryLayer.
bool IsSubtractionSupported(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsLstmSupported(const TensorInfo &input, const TensorInfo &outputStateIn, const TensorInfo &cellStateIn, const TensorInfo &scratchBuffer, const TensorInfo &outputStateOut, const TensorInfo &cellStateOut, const TensorInfo &output, const LstmDescriptor &descriptor, const LstmInputParamsInfo ¶msInfo, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonArgMinMaxWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const ArgMinMaxDescriptor &descriptor)
arm_compute::Status NeonSoftmaxWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const SoftmaxDescriptor &descriptor)
bool IsSpaceToBatchNdSupported(const TensorInfo &input, const TensorInfo &output, const SpaceToBatchNdDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonTransposeWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const TransposeDescriptor &descriptor)
const TensorInfo const Convolution2dDescriptor const TensorInfo const Optional< TensorInfo > & biases
uint32_t GetNumDimensions() const
Get the number of dimensions.
A MeanDescriptor for the MeanLayer.
bool IsDepthwiseConvolutionSupported(const TensorInfo &input, const TensorInfo &output, const DepthwiseConvolution2dDescriptor &descriptor, const TensorInfo &weights, const Optional< TensorInfo > &biases, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonMaximumWorkloadValidate(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output)
bool IsArgMinMaxSupported(const TensorInfo &input, const TensorInfo &output, const ArgMinMaxDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsSpaceToDepthSupported(const TensorInfo &input, const TensorInfo &output, const SpaceToDepthDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
A TransposeDescriptor for the TransposeLayer.
A StridedSliceDescriptor for the StridedSliceLayer.
bool IsTransposeSupported(const TensorInfo &input, const TensorInfo &output, const TransposeDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsMaximumSupported(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonConstantWorkloadValidate(const TensorInfo &output)
const TensorInfo & input1
bool IsFullyConnectedSupported(const TensorInfo &input, const TensorInfo &output, const TensorInfo &weights, const TensorInfo &biases, const FullyConnectedDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsMeanSupported(const TensorInfo &input, const TensorInfo &output, const MeanDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsBatchToSpaceNdSupported(const TensorInfo &input, const TensorInfo &output, const BatchToSpaceNdDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonPreluWorkloadValidate(const TensorInfo &input, const TensorInfo &alpha, const TensorInfo &output)
#define FORWARD_WORKLOAD_VALIDATE_FUNC(func, reasonIfUnsupported,...)
arm_compute::Status NeonDepthToSpaceWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const DepthToSpaceDescriptor &descriptor)
A Pooling2dDescriptor for the Pooling2dLayer.
arm_compute::Status NeonResizeWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const ResizeDescriptor &descriptor)
A NormalizationDescriptor for the NormalizationLayer.
bool IsQLstmSupported(const TensorInfo &input, const TensorInfo &previousOutputIn, const TensorInfo &previousCellStateIn, const TensorInfo &outputStateOut, const TensorInfo &cellStateOut, const TensorInfo &output, const QLstmDescriptor &descriptor, const LstmInputParamsInfo ¶msInfo, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsConvertBf16ToFp32Supported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
const TensorInfo const TensorInfo const TensorInfo & scratchBuffer
An InstanceNormalizationDescriptor for InstanceNormalizationLayer.
unsigned int GetConcatAxis() const
Get the concatenation axis value.
A ChannelShuffleDescriptor for the ChannelShuffle operator.
const TensorInfo const TensorInfo const TensorInfo const TensorInfo const TensorInfo & cellStateOut
bool IsSupportedForDataTypeGeneric(Optional< std::string &> reasonIfUnsupported, DataType dataType, Float16Func float16FuncPtr, Float32Func float32FuncPtr, Uint8Func uint8FuncPtr, Int32Func int32FuncPtr, BooleanFunc booleanFuncPtr, Params &&... params)
A SoftmaxDescriptor for the SoftmaxLayer.
bool IsReduceSupported(const TensorInfo &input, const TensorInfo &output, const ReduceDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
const TensorInfo const Convolution2dDescriptor const TensorInfo & weights
arm_compute::Status NeonPooling2dWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const Pooling2dDescriptor &descriptor)
arm_compute::Status NeonSinWorkloadValidate(const TensorInfo &input, const TensorInfo &output)
bool IsDequantizeSupported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
A DepthwiseConvolution2dDescriptor for the DepthwiseConvolution2dLayer.
A FillDescriptor for the FillLayer.
A BatchNormalizationDescriptor for the BatchNormalizationLayer.
arm_compute::Status NeonMultiplicationWorkloadValidate(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, const ActivationDescriptor *activationDescriptor)
bool IsChannelShuffleSupported(const TensorInfo &input, const TensorInfo &output, const ChannelShuffleDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
const TensorInfo const TensorInfo & mean
A PermuteDescriptor for the PermuteLayer.
LayerType
When adding a new layer, adapt also the LastLayer enum value in the enum class LayerType below...
bool IsPermuteSupported(const TensorInfo &input, const TensorInfo &output, const PermuteDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override