21.11
|
Namespaces | |
pad_fold | |
Classes | |
class | AddBroadcastReshapeLayerImpl |
class | AddDebugImpl |
struct | BFloat16ToFloat32 |
class | ConvertConstants |
class | ConvertFp32NetworkToBf16Impl |
class | ConvertFp32NetworkToFp16Impl |
struct | Float16ToFloat32 |
struct | Float32ToBFloat16 |
struct | Float32ToFloat16 |
class | FuseBatchNorm |
struct | IsBFloat16Layer |
struct | IsFloat16Layer |
struct | IsFloat32Layer |
class | MovePermuteUpImpl |
class | MoveTransposeUpImpl |
class | OptimizeConsecutiveReshapesImpl |
class | OptimizeInverseConversionsImpl |
class | OptimizeInversePermutesImpl |
class | PermuteAndBatchToSpaceAsDepthToSpaceImpl |
Replaces Permute leading into BatchToSpace with a DepthToSpace in the case where the Permute swaps the batch and channels dimensions such that the replacement is valid. More... | |
class | PermuteAsReshapeImpl |
class | RedirectMembersToConstantInputsImpl |
class | SquashEqualSiblingsImpl |
class | TransposeAsReshapeImpl |
Functions | |
template<typename LayerT > | |
LayerT * | ConvertWeight (Layer *l) |
Definition at line 94 of file AddBroadcastReshapeLayer.hpp.
Definition at line 151 of file ConvertConstants.hpp.
Definition at line 152 of file ConvertConstants.hpp.
Definition at line 155 of file ConvertConstants.hpp.
Definition at line 154 of file ConvertConstants.hpp.
using FoldPadIntoConvolution2d = OptimizeForExclusiveConnection<PadLayer, Convolution2dLayer, pad_fold::FoldPadIntoConvolution2dImpl> |
Definition at line 233 of file FoldPadIntoLayer2d.hpp.
using FoldPadIntoDepthwiseConvolution2d = OptimizeForExclusiveConnection <PadLayer, DepthwiseConvolution2dLayer, pad_fold::FoldPadIntoDepthwiseConvolution2dImpl> |
Definition at line 237 of file FoldPadIntoLayer2d.hpp.
using FoldPadIntoPooling2d = OptimizeForExclusiveConnection<PadLayer, Pooling2dLayer, pad_fold::FoldPadIntoPooling2dImpl> |
Definition at line 239 of file FoldPadIntoLayer2d.hpp.
Definition at line 76 of file ConvertFp32NetworkToBf16.hpp.
Definition at line 87 of file ConvertFp32NetworkToFp16.hpp.
using FuseBatchNormIntoConvolution2DFloat16 = OptimizeForExclusiveConnection<Convolution2dLayer, BatchNormalizationLayer, FuseBatchNorm<Convolution2dLayer, armnn::DataType::Float16> > |
Definition at line 183 of file FuseBatchNorm.hpp.
using FuseBatchNormIntoConvolution2DFloat32 = OptimizeForExclusiveConnection<Convolution2dLayer, BatchNormalizationLayer, FuseBatchNorm<Convolution2dLayer, armnn::DataType::Float32> > |
Definition at line 178 of file FuseBatchNorm.hpp.
Definition at line 193 of file FuseBatchNorm.hpp.
Definition at line 188 of file FuseBatchNorm.hpp.
using InsertDebugLayer = OptimizeForType<Layer, AddDebugImpl> |
Definition at line 34 of file AddDebug.hpp.
Definition at line 77 of file MovePermuteUp.hpp.
Definition at line 77 of file MoveTransposeUp.hpp.
using OptimizeConsecutiveReshapes = OptimizeForConnection<ReshapeLayer, ReshapeLayer, OptimizeConsecutiveReshapesImpl> |
Definition at line 61 of file OptimizeConsecutiveReshapes.hpp.
using OptimizeInverseConversionsFp16 = OptimizeForConnection<ConvertFp16ToFp32Layer, ConvertFp32ToFp16Layer, OptimizeInverseConversionsImpl> |
Definition at line 42 of file OptimizeInverseConversions.hpp.
using OptimizeInverseConversionsFp32 = OptimizeForConnection<ConvertFp32ToFp16Layer, ConvertFp16ToFp32Layer, OptimizeInverseConversionsImpl> |
Definition at line 44 of file OptimizeInverseConversions.hpp.
using OptimizeInversePermutes = OptimizeForConnection<PermuteLayer, PermuteLayer, OptimizeInversePermutesImpl<PermuteLayer> > |
Definition at line 43 of file OptimizeInversePermutes.hpp.
using OptimizeInverseTransposes = OptimizeForConnection<TransposeLayer, TransposeLayer, OptimizeInversePermutesImpl<TransposeLayer> > |
Definition at line 45 of file OptimizeInversePermutes.hpp.
using PermuteAndBatchToSpaceAsDepthToSpace = OptimizeForConnection<PermuteLayer, BatchToSpaceNdLayer, PermuteAndBatchToSpaceAsDepthToSpaceImpl<PermuteLayer> > |
Definition at line 102 of file PermuteAndBatchToSpaceAsDepthToSpace.hpp.
Definition at line 66 of file PermuteAsReshape.hpp.
using RedirectMembersToConstantInputs = OptimizeForType<FullyConnectedLayer, RedirectMembersToConstantInputsImpl> |
Definition at line 83 of file RedirectMembersToConstantInputs.hpp.
using SquashEqualPermuteSiblings = OptimizeForConnection<Layer, PermuteLayer, SquashEqualSiblingsImpl<PermuteLayer> > |
Definition at line 67 of file SquashEqualSiblings.hpp.
using SquashEqualReshapeSiblings = OptimizeForConnection<Layer, ReshapeLayer, SquashEqualSiblingsImpl<ReshapeLayer> > |
Definition at line 70 of file SquashEqualSiblings.hpp.
using SquashEqualTransposeSiblings = OptimizeForConnection<Layer, TransposeLayer, SquashEqualSiblingsImpl<TransposeLayer> > |
Definition at line 69 of file SquashEqualSiblings.hpp.
using TransposeAndBatchToSpaceAsDepthToSpace = OptimizeForConnection<TransposeLayer, BatchToSpaceNdLayer, PermuteAndBatchToSpaceAsDepthToSpaceImpl<TransposeLayer> > |
Definition at line 104 of file PermuteAndBatchToSpaceAsDepthToSpace.hpp.
Definition at line 77 of file TransposeAsReshape.hpp.
|
inline |
Definition at line 18 of file ConvertFp32NetworkToBf16.hpp.
References armnn::BFloat16, FloatingPointConverter::ConvertFloat32ToBFloat16(), armnn::Convolution2d, armnn::Float32, armnn::FullyConnected, TensorInfo::GetDataType(), TensorInfo::GetNumElements(), and armnn::info.