diff options
Diffstat (limited to 'src/backends/neon/NeonLayerSupport.cpp')
| -rw-r--r-- | src/backends/neon/NeonLayerSupport.cpp | 468 |
1 files changed, 468 insertions, 0 deletions
diff --git a/src/backends/neon/NeonLayerSupport.cpp b/src/backends/neon/NeonLayerSupport.cpp new file mode 100644 index 0000000000..dfaea5c81c --- /dev/null +++ b/src/backends/neon/NeonLayerSupport.cpp @@ -0,0 +1,468 @@ +// +// Copyright © 2017 Arm Ltd. All rights reserved. +// SPDX-License-Identifier: MIT +// + +#include "NeonLayerSupport.hpp" + +#include <LayerSupportCommon.hpp> +#include <InternalTypes.hpp> + +#include <armnn/Descriptors.hpp> +#include <armnn/Types.hpp> +#include <armnn/Tensor.hpp> + +#include <boost/core/ignore_unused.hpp> + +#ifdef ARMCOMPUTENEON_ENABLED +#include "workloads/NeonAdditionFloatWorkload.hpp" +#include "workloads/NeonActivationFloatWorkload.hpp" +#include "workloads/NeonBatchNormalizationFloatWorkload.hpp" +#include "workloads/NeonConvolution2dBaseWorkload.hpp" +#include "workloads/NeonDepthwiseConvolutionBaseWorkload.hpp" +#include "workloads/NeonL2NormalizationFloatWorkload.hpp" +#include "workloads/NeonMultiplicationFloatWorkload.hpp" +#include "workloads/NeonNormalizationFloatWorkload.hpp" +#include "workloads/NeonFullyConnectedWorkload.hpp" +#include "workloads/NeonPermuteWorkload.hpp" +#include "workloads/NeonPooling2dBaseWorkload.hpp" +#include "workloads/NeonSoftmaxBaseWorkload.hpp" +#include "workloads/NeonSubtractionFloatWorkload.hpp" +#endif + +using namespace boost; + +namespace armnn +{ + +bool IsNeonDirectConvolutionPreferred(const TensorInfo& weightInfo, const Convolution2dDescriptor& desc) +{ + // See arm_compute::NEDirectConvolutionLayer documentation for the supported cases, + // and complement with NEDirectConvolutionLayerKernel::configure() implementation. + + // Only 1x1 is using direct convolution. Performance results and details are in: + // https://jira.arm.com/browse/IVGCVSW-1003 + // Measurements were taken as of clframework: f105ab972135bcd21304883eff040d7e587099bc + + const bool dataTypeSupported = (weightInfo.GetDataType() == armnn::DataType::Float32); + + // Strides: 1|2|3 + const bool strideSupported = (desc.m_StrideX == 1 || desc.m_StrideX == 2 || desc.m_StrideX == 3) && + (desc.m_StrideY == 1 || desc.m_StrideY == 2 || desc.m_StrideY == 3); + + auto paddingLargerThan = [](const Convolution2dDescriptor& conv2ddesc, unsigned int value) + { + return conv2ddesc.m_PadLeft > value || conv2ddesc.m_PadRight > value || + conv2ddesc.m_PadTop > value || conv2ddesc.m_PadBottom > value; + }; + + // Supported sizes and padding. + const bool sizeAndPaddingSupported = + // Pad > 0 not supported for 1x1 weights. + (weightInfo.GetShape()[2] == 1 && weightInfo.GetShape()[3] == 1 && !paddingLargerThan(desc, 0u)); + + const bool preferDirectConvolution = dataTypeSupported && + strideSupported && + sizeAndPaddingSupported && + // NEDirectConvolutionLayerKernel doesn't support NULL bias. + desc.m_BiasEnabled; + return preferDirectConvolution; +} + +bool IsNeonNormalizationDescParamsSupported(std::string* reasonIfUnsupported, const NormalizationDescriptor& parameters) +{ + if (parameters.m_NormMethodType != NormalizationAlgorithmMethod::LocalBrightness) + { + if (reasonIfUnsupported) + { + *reasonIfUnsupported = "Unsupported normalisation method type, only LocalBrightness is supported"; + } + return false; + } + if (parameters.m_NormSize % 2 == 0) + { + if (reasonIfUnsupported) + { + *reasonIfUnsupported = "Normalization size must be an odd number."; + } + return false; + } + + return true; +} + +bool IsNeonBackendSupported(std::string* reasonIfUnsupported) +{ +#if ARMCOMPUTENEON_ENABLED + return true; +#else + if (reasonIfUnsupported != nullptr) + { + *reasonIfUnsupported = "The armnn library has been built without NEON support"; + } + return false; +#endif +} + +template<typename FloatFunc, typename Uint8Func, typename ... Params> +bool IsSupportedForDataTypeNeon(std::string* reasonIfUnsupported, + DataType dataType, + FloatFunc floatFuncPtr, + Uint8Func uint8FuncPtr, + Params&&... params) +{ + return IsNeonBackendSupported(reasonIfUnsupported) && + IsSupportedForDataTypeGeneric(reasonIfUnsupported, + dataType, + floatFuncPtr, + floatFuncPtr, + uint8FuncPtr, + std::forward<Params>(params)...); +} + +#if ARMCOMPUTENEON_ENABLED +template<class FuncType, class... Args> +inline bool IsWorkloadSupported(FuncType& func, std::string* reasonIfUnsupported, Args&&... args) +{ + arm_compute::Status aclStatus = func(std::forward<Args>(args)...); + const bool supported = (aclStatus.error_code() == arm_compute::ErrorCode::OK); + if (!supported && reasonIfUnsupported) + { + *reasonIfUnsupported = aclStatus.error_description(); + } + return supported; +} + +#define FORWARD_WORKLOAD_VALIDATE_FUNC(func, reasonIfUnsupported, ...) \ + return IsWorkloadSupported(func, reasonIfUnsupported, __VA_ARGS__); +#else +#define FORWARD_WORKLOAD_VALIDATE_FUNC(func, reasonIfUnsupported, ...) \ + return IsNeonBackendSupported(reasonIfUnsupported); +#endif + +bool IsActivationSupportedNeon(const TensorInfo& input, + const TensorInfo& output, + const ActivationDescriptor& descriptor, + std::string* reasonIfUnsupported) +{ + ignore_unused(descriptor); + FORWARD_WORKLOAD_VALIDATE_FUNC(NeonActivationWorkloadValidate, + reasonIfUnsupported, + input, + output, + descriptor); +} + +bool IsAdditionSupportedNeon(const TensorInfo& input0, + const TensorInfo& input1, + const TensorInfo& output, + std::string* reasonIfUnsupported) +{ + FORWARD_WORKLOAD_VALIDATE_FUNC(NeonAdditionWorkloadValidate, + reasonIfUnsupported, + input0, + input1, + output); +} + +bool IsBatchNormalizationSupportedNeon(const TensorInfo& input, + const TensorInfo& output, + const TensorInfo& mean, + const TensorInfo& var, + const TensorInfo& beta, + const TensorInfo& gamma, + const BatchNormalizationDescriptor& descriptor, + std::string* reasonIfUnsupported) +{ + FORWARD_WORKLOAD_VALIDATE_FUNC(NeonBatchNormalizationValidate, + reasonIfUnsupported, + input, + output, + mean, + var, + beta, + gamma, + descriptor); +} + +bool IsConstantSupportedNeon(const TensorInfo& output, + std::string* reasonIfUnsupported) +{ + return IsSupportedForDataTypeNeon(reasonIfUnsupported, + output.GetDataType(), + &TrueFunc<>, + &TrueFunc<>); +} + +bool IsConvolution2dSupportedNeon(const TensorInfo& input, + const TensorInfo& output, + const Convolution2dDescriptor& descriptor, + const TensorInfo& weights, + const boost::optional<TensorInfo>& biases, + std::string* reasonIfUnsupported) +{ + FORWARD_WORKLOAD_VALIDATE_FUNC(NeonConvolution2dWorkloadValidate, + reasonIfUnsupported, + input, + output, + descriptor, + weights, + biases); +} + +bool IsDepthwiseConvolutionSupportedNeon(const TensorInfo& input, + const TensorInfo& output, + const DepthwiseConvolution2dDescriptor& descriptor, + const TensorInfo& weights, + const boost::optional<TensorInfo>& biases, + std::string* reasonIfUnsupported) +{ + FORWARD_WORKLOAD_VALIDATE_FUNC(NeonDepthwiseConvolutionWorkloadValidate, + reasonIfUnsupported, + input, + output, + descriptor, + weights, + biases); +} + +bool IsDivisionSupportedNeon(const TensorInfo& input0, + const TensorInfo& input1, + const TensorInfo& output, + std::string* reasonIfUnsupported) +{ + // At the moment division is not supported + return false; +} + +bool IsSubtractionSupportedNeon(const TensorInfo& input0, + const TensorInfo& input1, + const TensorInfo& output, + std::string* reasonIfUnsupported) +{ + FORWARD_WORKLOAD_VALIDATE_FUNC(NeonSubtractionWorkloadValidate, + reasonIfUnsupported, + input0, + input1, + output); +} + +bool IsFullyConnectedSupportedNeon(const TensorInfo& input, + const TensorInfo& output, + const TensorInfo& weights, + const TensorInfo& biases, + const FullyConnectedDescriptor& descriptor, + std::string* reasonIfUnsupported) +{ + // At the moment U8 is unsupported + if (input.GetDataType() == DataType::QuantisedAsymm8) + { + return false; + } + FORWARD_WORKLOAD_VALIDATE_FUNC(NeonFullyConnectedWorkloadValidate, + reasonIfUnsupported, + input, + output, + weights, + biases, + descriptor); +} + +bool IsInputSupportedNeon(const TensorInfo& input, + std::string* reasonIfUnsupported) +{ + return IsSupportedForDataTypeNeon(reasonIfUnsupported, + input.GetDataType(), + &TrueFunc<>, + &TrueFunc<>); +} + +bool IsL2NormalizationSupportedNeon(const TensorInfo& input, + const TensorInfo& output, + std::string* reasonIfUnsupported) +{ + FORWARD_WORKLOAD_VALIDATE_FUNC(NeonL2NormalizationWorkloadValidate, reasonIfUnsupported, input, output); +} + +bool IsMergerSupportedNeon(const std::vector<const TensorInfo*> inputs, + const OriginsDescriptor& descriptor, + std::string* reasonIfUnsupported) +{ + ignore_unused(descriptor); + return IsSupportedForDataTypeNeon(reasonIfUnsupported, + inputs[0]->GetDataType(), + &TrueFunc<>, + &TrueFunc<>); +} + +bool IsMultiplicationSupportedNeon(const TensorInfo& input0, + const TensorInfo& input1, + const TensorInfo& output, + std::string* reasonIfUnsupported) +{ + FORWARD_WORKLOAD_VALIDATE_FUNC(NeonMultiplicationWorkloadValidate, + reasonIfUnsupported, + input0, + input1, + output); +} + +bool IsNormalizationSupportedNeon(const TensorInfo& input, + const TensorInfo& output, + const NormalizationDescriptor& descriptor, + std::string* reasonIfUnsupported) +{ + FORWARD_WORKLOAD_VALIDATE_FUNC(NeonNormalizationWorkloadValidate, reasonIfUnsupported, input, output, descriptor); +} + +bool IsOutputSupportedNeon(const TensorInfo& output, + std::string* reasonIfUnsupported) +{ + return IsSupportedForDataTypeNeon(reasonIfUnsupported, + output.GetDataType(), + &TrueFunc<>, + &TrueFunc<>); +} + +bool IsPermuteSupportedNeon(const TensorInfo& input, + const TensorInfo& output, + const PermuteDescriptor& descriptor, + std::string* reasonIfUnsupported) +{ + FORWARD_WORKLOAD_VALIDATE_FUNC(NeonPermuteWorkloadValidate, reasonIfUnsupported, input, output, descriptor); +} + +bool IsPooling2dSupportedNeon(const TensorInfo& input, + const TensorInfo& output, + const Pooling2dDescriptor& descriptor, + std::string* reasonIfUnsupported) +{ + FORWARD_WORKLOAD_VALIDATE_FUNC(NeonPooling2dWorkloadValidate, reasonIfUnsupported, input, output, descriptor); +} + +bool IsResizeBilinearSupportedNeon(const TensorInfo& input, + std::string* reasonIfUnsupported) +{ + ignore_unused(input); + return false; +} + +bool IsSoftmaxSupportedNeon(const TensorInfo& input, + const TensorInfo& output, + const SoftmaxDescriptor& descriptor, + std::string* reasonIfUnsupported) +{ + FORWARD_WORKLOAD_VALIDATE_FUNC(NeonSoftmaxWorkloadValidate, reasonIfUnsupported, input, output, descriptor); +} + +bool IsSplitterSupportedNeon(const TensorInfo& input, + const ViewsDescriptor& descriptor, + std::string* reasonIfUnsupported) +{ + ignore_unused(descriptor); + return IsSupportedForDataTypeNeon(reasonIfUnsupported, + input.GetDataType(), + &TrueFunc<>, + &TrueFunc<>); +} + +bool IsFakeQuantizationSupportedNeon(const TensorInfo& input, + const FakeQuantizationDescriptor& descriptor, + std::string* reasonIfUnsupported) +{ + ignore_unused(input); + ignore_unused(descriptor); + return false; +} + +bool IsReshapeSupportedNeon(const TensorInfo& input, + std::string* reasonIfUnsupported) +{ + return IsSupportedForDataTypeNeon(reasonIfUnsupported, + input.GetDataType(), + &TrueFunc<>, + &TrueFunc<>); +} + +bool IsFloorSupportedNeon(const TensorInfo& input, + const TensorInfo& output, + std::string* reasonIfUnsupported) +{ + ignore_unused(output); + return IsNeonBackendSupported(reasonIfUnsupported) && + IsSupportedForDataTypeGeneric(reasonIfUnsupported, + input.GetDataType(), + &FalseFuncF16<>, + &TrueFunc<>, + &FalseFuncU8<>); +} + +bool IsLstmSupportedNeon(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 TensorInfo& inputToForgetWeights, const TensorInfo& inputToCellWeights, + const TensorInfo& inputToOutputWeights, const TensorInfo& recurrentToForgetWeights, + const TensorInfo& recurrentToCellWeights, const TensorInfo& recurrentToOutputWeights, + const TensorInfo& forgetGateBias, const TensorInfo& cellBias, + const TensorInfo& outputGateBias, const TensorInfo* inputToInputWeights, + const TensorInfo* recurrentToInputWeights, const TensorInfo* cellToInputWeights, + const TensorInfo* inputGateBias, const TensorInfo* projectionWeights, + const TensorInfo* projectionBias, const TensorInfo* cellToForgetWeights, + const TensorInfo* cellToOutputWeights, std::string* reasonIfUnsupported) +{ + ignore_unused(input); + ignore_unused(outputStateIn); + ignore_unused(cellStateIn); + ignore_unused(scratchBuffer); + ignore_unused(outputStateOut); + ignore_unused(cellStateOut); + ignore_unused(output); + ignore_unused(descriptor); + ignore_unused(inputToForgetWeights); + ignore_unused(inputToCellWeights); + ignore_unused(inputToOutputWeights); + ignore_unused(recurrentToForgetWeights); + ignore_unused(recurrentToCellWeights); + ignore_unused(recurrentToOutputWeights); + ignore_unused(forgetGateBias); + ignore_unused(cellBias); + ignore_unused(outputGateBias); + ignore_unused(inputToInputWeights); + ignore_unused(recurrentToInputWeights); + ignore_unused(cellToInputWeights); + ignore_unused(inputGateBias); + ignore_unused(projectionWeights); + ignore_unused(projectionBias); + ignore_unused(cellToForgetWeights); + ignore_unused(cellToOutputWeights); + return false; +} + +bool IsConvertFp16ToFp32SupportedNeon(const TensorInfo& input, + const TensorInfo& output, + std::string* reasonIfUnsupported) +{ + ignore_unused(input); + ignore_unused(output); + return true; +} + +bool IsConvertFp32ToFp16SupportedNeon(const TensorInfo& input, + const TensorInfo& output, + std::string* reasonIfUnsupported) +{ + ignore_unused(input); + ignore_unused(output); + return true; +} + +bool IsMeanSupportedNeon(const TensorInfo& input, + const TensorInfo& output, + const MeanDescriptor& descriptor, + std::string* reasonIfUnsupported) +{ + return false; +} + +} |