18 #if defined(ARMCOMPUTENEON_ENABLED) 70 template<
typename ... Args>
71 bool IsNeonBackendSupported(Optional<std::string&> reasonIfUnsupported, Args... args)
74 #if defined(ARMCOMPUTENEON_ENABLED) 77 SetValueChecked(reasonIfUnsupported,
"The armnn library has been built without NEON support");
82 template<
typename FloatFunc,
typename Uint8Func,
typename ... Params>
83 bool IsSupportedForDataTypeNeon(Optional<std::string&> reasonIfUnsupported,
85 FloatFunc floatFuncPtr,
86 Uint8Func uint8FuncPtr,
89 return IsNeonBackendSupported(reasonIfUnsupported) &&
97 std::forward<Params>(params)...);
100 #if defined(ARMCOMPUTENEON_ENABLED) 101 template<
class FuncType,
class... Args>
102 inline bool IsWorkloadSupported(FuncType& func, Optional<std::string&> reasonIfUnsupported, Args&&... args)
105 const bool supported = (aclStatus.error_code() == arm_compute::ErrorCode::OK);
106 if (!supported && reasonIfUnsupported)
108 reasonIfUnsupported.value() = aclStatus.error_description();
113 #define FORWARD_WORKLOAD_VALIDATE_FUNC(func, reasonIfUnsupported, ...) \ 114 return IsWorkloadSupported(func, reasonIfUnsupported, __VA_ARGS__); 116 #define FORWARD_WORKLOAD_VALIDATE_FUNC(func, reasonIfUnsupported, ...) \ 117 return IsNeonBackendSupported(reasonIfUnsupported, __VA_ARGS__); 223 SetValueChecked(reasonIfUnsupported,
"Neon Concat: Concat axis > Number of dimensions.");
228 if(concatInnerAxis < 3)
236 else if (concatInnerAxis == 3)
238 for (
auto& input : inputs)
242 SetValueChecked(reasonIfUnsupported,
"Neon Concat: Types and quantization parameters must match.");
250 SetValueChecked(reasonIfUnsupported,
"Neon Concat: Maximum of 4 dimensions supported.");
258 return IsSupportedForDataTypeNeon(reasonIfUnsupported,
382 return IsNeonBackendSupported(reasonIfUnsupported) &&
420 return IsNeonBackendSupported(reasonIfUnsupported, input);
550 return IsNeonBackendSupported(reasonIfUnsupported, output);
713 return IsSupportedForDataTypeNeon(reasonIfUnsupported,
720 const std::vector<std::reference_wrapper<TensorInfo>>& outputs,
724 #if defined(ARMCOMPUTENEON_ENABLED) 741 for (
auto output : outputs)
745 SetValueChecked(reasonIfUnsupported,
"Neon Splitter: Types and quantization parameters must match.");
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)
arm_compute::Status NeonMultiplicationWorkloadValidate(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output)
A TransposeConvolution2dDescriptor for the TransposeConvolution2dLayer.
const TensorShape & GetShape() const
A ReshapeDescriptor for the ReshapeLayer.
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.
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)
A ComparisonDescriptor for the ComparisonLayer.
bool IsDepthToSpaceSupported(const TensorInfo &input, const TensorInfo &output, const DepthToSpaceDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsL2NormalizationSupported(const TensorInfo &input, const TensorInfo &output, const L2NormalizationDescriptor &descriptor, 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
ISubgraphViewConverter supported
bool IsAdditionSupported(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
A Convolution2dDescriptor for the Convolution2dLayer.
bool IsConcatSupported(const std::vector< const TensorInfo *> inputs, const TensorInfo &output, const ConcatDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsSplitterSupported(const TensorInfo &input, const ViewsDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonMeanWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const MeanDescriptor &desc)
arm_compute::Status NeonAdditionWorkloadValidate(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output)
ResizeMethod m_Method
The Interpolation method to use (Bilinear, NearestNeighbor).
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)
arm_compute::Status NeonTransposeConvolution2dWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const TransposeConvolution2dDescriptor &descriptor, const TensorInfo &weights, const Optional< TensorInfo > &biases)
bool IsGreaterSupported(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
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 IsResizeBilinearSupported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonSliceWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const SliceDescriptor &descriptor)
bool IsConstantSupported(const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
Copyright (c) 2020 ARM Limited.
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 &&...)
std::set< unsigned int > ComputeSplitAxis(const armnn::SplitterDescriptor &desc, const TensorShape &input)
A SpaceToDepthDescriptor for the SpaceToDepthLayer.
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
A BatchToSpaceNdDescriptor for the BatchToSpaceNdLayer.
bool IsActivationSupported(const TensorInfo &input, const TensorInfo &output, const ActivationDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsMultiplicationSupported(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsComparisonSupported(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, const ComparisonDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
A ResizeDescriptor for the ResizeLayer.
arm_compute::Status NeonL2NormalizationWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const L2NormalizationDescriptor &descriptor)
arm_compute::Status NeonAbsWorkloadValidate(const TensorInfo &input, const TensorInfo &output)
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 NeonSpaceToBatchNdWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const SpaceToBatchNdDescriptor &descriptor)
A PadDescriptor for the PadLayer.
bool IsAbsSupported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
An LstmDescriptor for the LstmLayer.
arm_compute::Status NeonReshapeWorkloadValidate(const TensorInfo &input, const TensorInfo &output)
bool IsInputSupported(const TensorInfo &input, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
bool IsStackSupported(const std::vector< const TensorInfo *> &inputs, const TensorInfo &output, const StackDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
A L2NormalizationDescriptor for the L2NormalizationLayer.
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 FullyConnectedDescriptor for the FullyConnectedLayer.
bool IsMergerSupported(const std::vector< const TensorInfo *> inputs, const TensorInfo &output, const MergerDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
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
uint32_t m_TargetWidth
Target width value.
bool IsDivisionSupported(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
arm_compute::Status NeonFullyConnectedWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const TensorInfo &weights, const TensorInfo &biases, const FullyConnectedDescriptor &descriptor)
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
An ActivationDescriptor for the ActivationLayer.
bool IsResizeSupported(const TensorInfo &input, const TensorInfo &output, const ResizeDescriptor &descriptor, 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 IsTransposeConvolution2dSupported(const TensorInfo &input, const TensorInfo &output, const TransposeConvolution2dDescriptor &descriptor, const TensorInfo &weights, const Optional< TensorInfo > &biases, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
uint32_t m_TargetHeight
Target height value.
A SliceDescriptor for the SliceLayer.
arm_compute::Status NeonDequantizeWorkloadValidate(const TensorInfo &input, const TensorInfo &output)
bool IsReshapeSupported(const TensorInfo &input, const TensorInfo &output, const ReshapeDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
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 NeonDepthwiseConvolutionWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const DepthwiseConvolution2dDescriptor &descriptor, const TensorInfo &weights, const Optional< TensorInfo > &biases)
void SetValueChecked(Optional< T &> optionalRef, V &&val)
ComparisonOperation m_Operation
Specifies the comparison operation to execute.
A SpaceToBatchNdDescriptor for the SpaceToBatchNdLayer.
arm_compute::Status NeonNormalizationWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const NormalizationDescriptor &descriptor)
arm_compute::Status NeonBatchToSpaceNdWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const BatchToSpaceNdDescriptor &desc)
bool IsQuantizeSupported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
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)
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
arm_compute::Status NeonSubtractionWorkloadValidate(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output)
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
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)
arm_compute::Status NeonGreaterWorkloadValidate(const TensorInfo &input0, const TensorInfo &input1, 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.
DataLayout m_DataLayout
The data layout to be used (NCHW, NHWC).
An InstanceNormalizationDescriptor for InstanceNormalizationLayer.
unsigned int GetConcatAxis() const
Get the concatenation axis value.
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.
arm_compute::Status NeonDivisionWorkloadValidate(const TensorInfo &input0, const TensorInfo &input1, const TensorInfo &output)
arm_compute::Status NeonPooling2dWorkloadValidate(const TensorInfo &input, const TensorInfo &output, const Pooling2dDescriptor &descriptor)
bool IsDequantizeSupported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
A DepthwiseConvolution2dDescriptor for the DepthwiseConvolution2dLayer.
bool IsRsqrtSupported(const TensorInfo &input, const TensorInfo &output, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override
A BatchNormalizationDescriptor for the BatchNormalizationLayer.
A PermuteDescriptor for the PermuteLayer.
bool IsPermuteSupported(const TensorInfo &input, const TensorInfo &output, const PermuteDescriptor &descriptor, Optional< std::string &> reasonIfUnsupported=EmptyOptional()) const override