Store the tensor's metadata. More...

#include <ITensorInfo.h>

Collaboration diagram for ITensorInfo:

Public Types
using	TensorDimsState = std::vector< int >

using	Id = int32_t
	An id that uniquely identifies an ITensorInfo within some domain (e.g. More...

Public Member Functions
virtual	~ITensorInfo ()=default
	Default virtual destructor. More...

virtual ITensorInfo &	set_data_type (DataType data_type)=0
	Set the data type to the specified value. More...

virtual ITensorInfo &	set_num_channels (int num_channels)=0
	Set the number of channels to the specified value. More...

virtual ITensorInfo &	set_format (Format format)=0
	Set the format of an already initialized tensor. More...

virtual ITensorInfo &	set_tensor_shape (const TensorShape &shape)=0
	Set the shape of an already initialized tensor. More...

virtual ITensorInfo &	set_tensor_dims_state (const TensorDimsState &state)=0
	Set the state for each dimension of the tensor. More...

virtual ITensorInfo &	set_quantization_info (const QuantizationInfo &quantization_info)=0
	Set the quantization settings (scale and offset) of the tensor. More...

virtual ITensorInfo &	set_data_layout (const DataLayout &data_layout)=0
	Set the data layout of the tensor. More...

virtual ITensorInfo &	reset_padding ()=0
	Resets the padding settings of the tensor. More...

virtual bool	auto_padding ()=0
	Update the offset to the first element and the strides to automatically computed values. More...

virtual ITensorInfo &	set_lock_paddings (bool flag)=0
	Set the lock paddings flag of the tensor. More...

virtual bool	lock_paddings () const =0
	Get the lock paddings flag value. More...

virtual bool	extend_padding (const PaddingSize &padding)=0
	Update the offset to the first element, the strides and the total size. More...

virtual size_t	dimension (size_t index) const =0
	Return the size of the requested dimension. More...

virtual size_t	dimension (DataLayoutDimension dimension) const =0
	Return the size of the requested data layout dimension. More...

virtual const Strides &	strides_in_bytes () const =0
	The strides in bytes for accessing each dimension of the tensor. More...

virtual size_t	offset_first_element_in_bytes () const =0
	The offset from the beginning of the memory allocation to the first element of the tensor. More...

virtual int32_t	offset_element_in_bytes (const Coordinates &pos) const =0
	The offset in bytes from the beginning of the memory allocation to access the element at position (x, y, z ...) More...

virtual size_t	element_size () const =0
	Element size in bytes calculated as data_size() * num_channels() More...

virtual size_t	num_dimensions () const =0
	The number of dimensions of the tensor (rank) More...

virtual size_t	num_channels () const =0
	The number of channels for each tensor element. More...

virtual const TensorShape &	tensor_shape () const =0
	Size for each dimension of the tensor. More...

virtual const TensorDimsState &	tensor_dims_state () const =0
	State of each dimension of the tensor shape. More...

virtual DataType	data_type () const =0
	Data type used for each element of the tensor. More...

virtual Format	format () const =0
	Colour format of the image. More...

virtual size_t	total_size () const =0
	Returns the total size of the tensor in bytes. More...

virtual PaddingSize	padding () const =0
	Padding of tensor. More...

virtual bool	has_padding () const =0
	Checks if the tensor has been allocated with padding or not. More...

virtual bool	is_resizable () const =0
	Flag indicating whether the size of the tensor can be changed. More...

virtual bool	is_dynamic () const =0
	Flag indicating whether the shape of the tensor is dynamic, meaning that it can change on kernel/function execution. More...

virtual bool	are_values_constant () const =0
	Flag indicating whether the values of the tensor are constant, meaning that they can change on kernel/function execution. More...

virtual ITensorInfo &	set_is_resizable (bool is_resizable)=0
	Set the flag whether the tensor size can be changed. More...

virtual ITensorInfo &	set_are_values_constant (bool are_values_constant)=0
	Set the flag whether the tensor values can change during kernel/function execution. More...

virtual ValidRegion	valid_region () const =0
	Valid region of the tensor. More...

virtual void	set_valid_region (const ValidRegion &valid_region)=0
	Set the valid region of the tensor. More...

virtual QuantizationInfo	quantization_info () const =0
	Get the quantization settings (scale and offset) of the tensor. More...

virtual DataLayout	data_layout () const =0
	Get the data layout of the tensor. More...

virtual Id	id () const =0
	Get the workload tensor id of the tensor. More...

virtual ITensorInfo &	set_id (ITensorInfo::Id id)=0
	Set the tensor id. More...

bool	has_valid_id () const
	Check if the tensor id is valid. More...

Public Member Functions inherited from ICloneable< ITensorInfo >
virtual	~ICloneable ()=default
	Default virtual desctructor. More...

virtual std::unique_ptr< ITensorInfo >	clone () const=0
	Provide a clone of the current object of class T. More...

Static Public Member Functions
static constexpr int32_t	get_dynamic_state_value ()
	Get the value representing dynamic dimension state. More...

static constexpr int32_t	get_static_state_value ()
	Get the value representing static dimension state. More...

template<typename... Infos>
static std::pair< TensorShape, ValidRegion >	broadcast_shape_and_valid_region (const Infos &...infos)
	If infos are broadcast compatible tensor info's, return the broadcasted shape and the intersection of the broadcasted valid regions of the tensors. More...

Static Public Attributes
static constexpr Id	invalid_tensor_id = 0
	An invalid tensor id within a domain. More...

Detailed Description

Store the tensor's metadata.

Definition at line 44 of file ITensorInfo.h.

Member Typedef Documentation

◆ Id

using Id = int32_t

An id that uniquely identifies an ITensorInfo within some domain (e.g.

a workload)

Definition at line 50 of file ITensorInfo.h.

◆ TensorDimsState

using TensorDimsState = std::vector<int>

Definition at line 47 of file ITensorInfo.h.

Constructor & Destructor Documentation

◆ ~ITensorInfo()

virtual ~ITensorInfo ( )

virtualdefault

Default virtual destructor.

Member Function Documentation

◆ are_values_constant()

virtual bool are_values_constant ( ) const

pure virtual

Flag indicating whether the values of the tensor are constant, meaning that they can change on kernel/function execution.

Returns: True if values are constant else false

Implemented in TensorInfo, and SubTensorInfo.

Referenced by SubTensorInfo::are_values_constant(), arm_compute::auto_init_if_empty(), CpuDepthwiseConv2dAssemblyDispatch::configure(), CLFullyConnectedLayer::configure(), CpuFullyConnected::configure(), ClFullyConnected::configure(), NEFullyConnectedLayer::configure(), CpuMatMul::validate(), ClFullyConnected::validate(), CLDepthwiseConvolutionLayer::validate(), NEConvolutionLayer::validate(), CLConvolutionLayer::validate(), GpuConv2d::validate_op(), and GpuDepthwiseConv2d::validate_op().

◆ auto_padding()

virtual bool auto_padding ( )

pure virtual

Update the offset to the first element and the strides to automatically computed values.

Note: The padding used by this method is really conservative so that the tensor can be used for most functions.

Returns: True if the strides or the offset to the first element have changed.

Implemented in TensorInfo, and SubTensorInfo.

Referenced by SubTensorInfo::auto_padding(), and AccessWindowAutoPadding::update_padding_if_needed().

◆ broadcast_shape_and_valid_region()

static std::pair<TensorShape, ValidRegion> broadcast_shape_and_valid_region ( const Infos &... infos )

inlinestatic

If infos are broadcast compatible tensor info's, return the broadcasted shape and the intersection of the broadcasted valid regions of the tensors.

Two tensor info's are broadcast compatible if their shapes are broadcast compatible.

Two tensor shapes are broadcast compatible if for each dimension, they're equal or one of them is 1.

If two shapes are compatible, each dimension in the broadcasted shape is the max of the original dimensions.

Parameters

[in] infos Tensor info's.

Returns: The broadcasted shape and valid region, or an empty shape and valid region if the info's are not broadcast compatible.

Definition at line 332 of file ITensorInfo.h.

     {
         TensorShape bc_shape = TensorShape::broadcast_shape(infos.tensor_shape()...);
         ValidRegion bc_valid_region{Coordinates(), bc_shape};
  
         auto broadcast_valid_region = [&bc_valid_region](const ITensorInfo &info)
         {
             if (info.num_dimensions() != 0)
             {
                 for (size_t d = 0; d < bc_valid_region.shape.num_dimensions(); ++d)
                 {
                     const bool is_broadcast = (info.tensor_shape()[d] == 1);
  
                     const int    anchor_max = std::max(bc_valid_region.anchor[d], info.valid_region().anchor[d]);
                     const size_t valid_min  = std::min(bc_valid_region.shape[d], info.valid_region().shape[d]);
  
                     if (!is_broadcast || (valid_min == 0))
                     {
                         bc_valid_region.anchor.set(d, anchor_max);
                         bc_valid_region.shape.set(d, valid_min);
                     }
                 }
             }
         };
  
         utility::for_each(broadcast_valid_region, infos...);
  
         return std::pair<TensorShape, ValidRegion>(bc_shape, bc_valid_region);
     }

References TensorShape::broadcast_shape(), arm_compute::utility::for_each(), and arm_compute::test::validation::info.

◆ data_layout()

virtual DataLayout data_layout ( ) const

pure virtual

Get the data layout of the tensor.

Returns: A DataLayout containing the layout data information.

Implemented in TensorInfo, and SubTensorInfo.

Referenced by arm_compute::auto_init_if_empty(), arm_compute::misc::shape_calculator::compute_depthwise_convolution_shape(), arm_compute::misc::shape_calculator::compute_weights_reshaped_shape(), CLComparisonKernel::configure(), NEFuseBatchNormalizationKernel::configure(), NEBatchNormalizationLayerKernel::configure(), CLPriorBoxLayerKernel::configure(), CpuDirectConv3d::configure(), CLDepthwiseConvolutionLayer::configure(), NEGenerateProposalsLayer::configure(), CLGenerateProposalsLayer::configure(), CLDirectDeconvolutionLayer::configure(), Accessor::data_layout(), SubTensorInfo::data_layout(), SubTensorInfo::dimension(), ClTemplateDirectConv2d::get_build_options(), ClTemplateDirectConv2d::get_component_code(), ClTemplateDirectConv2d::get_config_id(), ClTemplatePool2d::get_config_id(), CPPUpsampleKernel::run(), NEChannelShuffleLayerKernel::run(), NEReorgLayerKernel::run(), CLInstanceNormalizationLayerKernel::run(), NEROIAlignLayerKernel::run(), NEBatchNormalizationLayerKernel::run(), CLROIAlignLayerKernel::run(), CLComputeMeanVariance::run(), arm_compute::cl_dwc::use_cl_image_for_weights(), CLDepthwiseConvolutionLayer::validate(), NEGenerateProposalsLayer::validate(), NEDeconvolutionLayer::validate(), CLGenerateProposalsLayer::validate(), GpuCkwMatMul::write_component_code(), and GpuCkwDirectConv2d::write_component_code().

◆ data_type()

virtual DataType data_type ( ) const

pure virtual

Data type used for each element of the tensor.

Returns: Tensor data type

Implemented in TensorInfo, and SubTensorInfo.

◆ dimension() [1/2]

virtual size_t dimension ( DataLayoutDimension dimension ) const

pure virtual

Return the size of the requested data layout dimension.

Parameters

[in] dimension DataLayoutDimension of the dimension

Returns: Dimension of the requested dimension

Implemented in TensorInfo, and SubTensorInfo.

◆ dimension() [2/2]

virtual size_t dimension ( size_t index ) const

pure virtual

Return the size of the requested dimension.

Parameters

[in] index Index of the dimension

Returns: Dimension of the requested dimension

Implemented in TensorInfo, and SubTensorInfo.

◆ element_size()

virtual size_t element_size ( ) const

pure virtual

Element size in bytes calculated as data_size() * num_channels()

Returns: The size of one element in bytes

Implemented in TensorInfo, and SubTensorInfo.

Referenced by arm_compute::calculate_squashed_or_max_window(), ClTransposedConvolutionKernel::configure(), ClWidthConcatenate2TensorsKernel::configure(), ClWidthConcatenate4TensorsKernel::configure(), CLBitwiseKernel::configure(), CLSelectKernel::configure(), NEPadLayerKernel::configure(), CLComparisonKernel::configure(), CLRangeKernel::configure(), ClMulKernel::configure(), ITensor::copy_from(), Accessor::element_size(), SubTensorInfo::element_size(), arm_compute::cpu::fused_batch_normalization_conv(), arm_compute::cpu::fused_batch_normalization_dwc_nchw(), arm_compute::cpu::fused_batch_normalization_dwc_nhwc(), GpuCkwActivation::get_window(), GpuCkwCast::get_window(), GpuCkwElementwiseBinary::get_window(), GpuCkwResize::get_window(), ClTemplateCast::get_window(), ClTemplateReshape::get_window(), ClTemplateResize::get_window(), ClTemplateActivation::get_window(), arm_compute::cpu::in_bounds_crop_window(), CpuWinogradConv2d::prepare(), NEQLSTMLayer::prepare(), CPPUpsampleKernel::run(), NETileKernel::run(), NEDepthToSpaceLayerKernel::run(), NESpaceToDepthLayerKernel::run(), NEReverseKernel::run(), NEReorgLayerKernel::run(), NEBatchToSpaceLayerKernel::run(), NESpaceToBatchLayerKernel::run(), NELSTMLayer::run(), CpuWinogradConv2dTransformInputKernel::run_op(), CpuWinogradConv2dTransformOutputKernel::run_op(), arm_compute::cpu::select_op_not_same_rank(), and CpuGemmAssemblyDispatch::validate().

◆ extend_padding()

virtual bool extend_padding ( const PaddingSize & padding )

pure virtual

Update the offset to the first element, the strides and the total size.

Note: This function can only increase the offset, strides and total size.

Parameters

[in] padding Padding around the XY plane in number of elements.

Returns: True if the strides, offset and total size have changed.

Implemented in TensorInfo, and SubTensorInfo.

Referenced by ClIndirectConv2dKernel::configure(), SubTensorInfo::extend_padding(), AccessWindowStatic::update_padding_if_needed(), and AccessWindowRectangle::update_padding_if_needed().

◆ format()

virtual Format format ( ) const

pure virtual

Colour format of the image.

Returns: Colour format of the image

Implemented in TensorInfo, and SubTensorInfo.

Referenced by Accessor::format(), SubTensorInfo::format(), and SubTensorInfo::set_tensor_shape().

◆ get_dynamic_state_value()

static constexpr int32_t get_dynamic_state_value ( )

inlinestaticconstexpr

Get the value representing dynamic dimension state.

Returns: Value representing dynamic dimension state

Definition at line 58 of file ITensorInfo.h.

     {
         return _dynamic_dimension;
     }

Referenced by arm_compute::test::construct_dynamic_dims_state(), and TensorInfo::is_dynamic().

◆ get_static_state_value()

static constexpr int32_t get_static_state_value ( )

inlinestaticconstexpr

Get the value representing static dimension state.

Returns: Value representing static dimension state

Definition at line 67 of file ITensorInfo.h.

     {
         return _static_dimension;
     }

Referenced by arm_compute::test::construct_static_dims_state().

◆ has_padding()

virtual bool has_padding ( ) const

pure virtual

Checks if the tensor has been allocated with padding or not.

Returns: True if padding is allocated in the tensor, otherwise false.

Implemented in TensorInfo, and SubTensorInfo.

Referenced by NEPadLayerKernel::configure(), SubTensorInfo::has_padding(), NECropKernel::run(), and NECropKernel::validate().

◆ has_valid_id()

bool has_valid_id ( ) const

inline

Check if the tensor id is valid.

Definition at line 313 of file ITensorInfo.h.

     {
         return id() != invalid_tensor_id;
     }

References ITensorInfo::id(), and ITensorInfo::invalid_tensor_id.

Referenced by ClTemplateDepthwiseConv2d::declare_variables(), ClTemplateDirectConv2d::declare_variables(), ClTemplateDepthwiseConv2d::get_component_code(), ClTemplateDirectConv2d::get_component_code(), ClTemplateDepthwiseConv2d::get_tag_lut(), ClTemplateDirectConv2d::get_tag_lut(), GpuKernelArgument::has_valid_id(), GpuWorkloadArgument::has_valid_id(), GpuConv2d::validate_op(), GpuDepthwiseConv2d::validate_op(), GpuElementwiseBinaryCommon::validate_op(), and GpuCkwDepthwiseConv2d::write_component_code().

◆ id()

virtual Id id ( ) const

pure virtual

Get the workload tensor id of the tensor.

Returns: Workload tensor id of the tensor

Implemented in TensorInfo, and SubTensorInfo.

Referenced by ITensorInfo::has_valid_id(), and SubTensorInfo::id().

◆ is_dynamic()

virtual bool is_dynamic ( ) const

pure virtual

Flag indicating whether the shape of the tensor is dynamic, meaning that it can change on kernel/function execution.

Returns: True if its dynamic else false

Implemented in TensorInfo, and SubTensorInfo.

Referenced by SubTensorInfo::is_dynamic().

◆ is_resizable()

virtual bool is_resizable ( ) const

pure virtual

Flag indicating whether the size of the tensor can be changed.

Returns: True if the tensor size can be changed.

Implemented in TensorInfo, and SubTensorInfo.

Referenced by arm_compute::test::validation::DATA_TEST_CASE(), SubTensorInfo::extend_padding(), SubTensorInfo::is_resizable(), arm_compute::test::validation::TEST_CASE(), AccessWindowAutoPadding::update_padding_if_needed(), AccessWindowStatic::update_padding_if_needed(), AccessWindowRectangle::update_padding_if_needed(), AccessWindowStatic::update_window_if_needed(), and AccessWindowRectangle::update_window_if_needed().

◆ lock_paddings()

virtual bool lock_paddings ( ) const

pure virtual

Get the lock paddings flag value.

Returns: lock paddings flag value

Implemented in TensorInfo, and SubTensorInfo.

Referenced by ClMatMulNativeDefaultConfigValhall::configure(), and ClMatMulNativeKernel::configure().

◆ num_channels()

virtual size_t num_channels ( ) const

pure virtual

The number of channels for each tensor element.

Returns: The number of channels for each tensor element

Implemented in TensorInfo, and SubTensorInfo.

Referenced by arm_compute::auto_init_if_empty(), CLFFTScaleKernel::configure(), NEReduceMean::configure(), CLReduceMean::configure(), NEFFT1D::configure(), CLFFT1D::configure(), CpuComplexMulKernel::configure(), ITensor::copy_from(), Accessor::num_channels(), SubTensorInfo::num_channels(), NEFFT1D::validate(), and CLFFT1D::validate().

◆ num_dimensions()

virtual size_t num_dimensions ( ) const

pure virtual

The number of dimensions of the tensor (rank)

Returns: The number of dimensions of the tensor (rank)

Implemented in TensorInfo, and SubTensorInfo.

◆ offset_element_in_bytes()

virtual int32_t offset_element_in_bytes ( const Coordinates & pos ) const

pure virtual

The offset in bytes from the beginning of the memory allocation to access the element at position (x, y, z ...)

Parameters

[in] pos Vector with the coordinates of the element to access. The size of this vector must be equal to the number of dimensions of the tensor

Returns: Offset in bytes from the beginning of the memory allocation to access the element (x, y, z, ...)

Implemented in TensorInfo, and SubTensorInfo.

Referenced by SubTensorInfo::offset_first_element_in_bytes(), ITensor::ptr_to_element(), and NEReorgLayerKernel::run().

◆ offset_first_element_in_bytes()

virtual size_t offset_first_element_in_bytes ( ) const

pure virtual

The offset from the beginning of the memory allocation to the first element of the tensor.

This can be used to access efficiently elements in a 2D tensor

Returns: The offset in bytes to access the first element of the tensor.

Implemented in TensorInfo, and SubTensorInfo.

Referenced by arm_compute::cpu::bounding_box_transform(), arm_compute::cpu::bounding_box_transform_qsymm16(), arm_compute::cpu::directconv3d_float_neon_ndhwc(), arm_compute::cpu::directconv3d_quantized_neon_ndhwc(), CpuDepthwiseConv2dAssemblyDispatch::prepare(), CpuWinogradConv2d::prepare(), CpuWinogradConv2dTransformInputKernel::run_op(), CpuPool2dAssemblyWrapperKernel::run_op(), CpuDepthwiseConv2dAssemblyWrapperKernel::run_op(), CpuWinogradConv2dTransformOutputKernel::run_op(), AccessWindowStatic::update_window_if_needed(), and AccessWindowRectangle::update_window_if_needed().

◆ padding()

virtual PaddingSize padding ( ) const

pure virtual

Padding of tensor.

Returns: Padding.

Implemented in TensorInfo, and SubTensorInfo.

Referenced by ClIndirectConv2dKernel::configure(), arm_compute::test::validation::DATA_TEST_CASE(), Accessor::padding(), SubTensorInfo::padding(), CpuDepthwiseConv2dAssemblyDispatch::prepare(), NEFFTRadixStageKernel::run(), ClGemm::run(), CpuGemmConv2d::run(), arm_compute::test::validation::TEST_CASE(), and AccessWindowRectangle::update_window_if_needed().

◆ quantization_info()

virtual QuantizationInfo quantization_info ( ) const

pure virtual

Get the quantization settings (scale and offset) of the tensor.

Returns: A QuantizationInfo containing the scale and offset.

Implemented in TensorInfo, and SubTensorInfo.

◆ reset_padding()

virtual ITensorInfo& reset_padding ( )

pure virtual

Resets the padding settings of the tensor.

Returns: Reference to this ITensorInfo object

Implemented in TensorInfo, and SubTensorInfo.

Referenced by CpuLogits1DSoftmaxKernel< IS_LOG >::configure(), and SubTensorInfo::reset_padding().

◆ set_are_values_constant()

virtual ITensorInfo& set_are_values_constant ( bool are_values_constant )

pure virtual

Set the flag whether the tensor values can change during kernel/function execution.

Parameters

[in] are_values_constant Flag that marks the tensor values if they can be changed or not.

Returns: Reference to this ITensorInfo object

Implemented in TensorInfo, and SubTensorInfo.

Referenced by arm_compute::auto_init_if_empty(), and SubTensorInfo::set_are_values_constant().

◆ set_data_layout()

virtual ITensorInfo& set_data_layout ( const DataLayout & data_layout )

pure virtual

Set the data layout of the tensor.

Parameters

[in] data_layout DataLayout containing the layout data information.

Returns: Reference to this ITensorInfo object

Implemented in TensorInfo, and SubTensorInfo.

Referenced by arm_compute::auto_init_if_empty(), NEROIAlignLayerKernel::configure(), NEInstanceNormalizationLayer::configure(), CLROIAlignLayerKernel::configure(), NEFFTConvolutionLayer::configure(), CpuGemmConv2d::configure(), ClGemmConv2d::configure(), and SubTensorInfo::set_data_layout().

◆ set_data_type()

virtual ITensorInfo& set_data_type ( DataType data_type )

pure virtual

Set the data type to the specified value.

Warning: This resets the format to UNKNOWN.

Parameters

[in] data_type The new data type.

Returns: Reference to this ITensorInfo object

Implemented in TensorInfo, and SubTensorInfo.

Referenced by arm_compute::auto_init_if_empty(), main(), SubTensorInfo::set_data_type(), and arm_compute::graph::backends::detail::validate_convolution_layer().

◆ set_format()

virtual ITensorInfo& set_format ( Format format )

pure virtual

Set the format of an already initialized tensor.

Note: If the data type has already been configured (i.e. not UNKNOWN) it must match the new format. If data type hasn't been configured it will be based on the format.

Parameters

[in] format Single-plane format of the tensor.

Returns: Reference to this ITensorInfo object

Implemented in TensorInfo, and SubTensorInfo.

Referenced by SubTensorInfo::set_format().

◆ set_id()

virtual ITensorInfo& set_id ( ITensorInfo::Id id )

pure virtual

Set the tensor id.

Implemented in TensorInfo, and SubTensorInfo.

Referenced by SubTensorInfo::set_id().

◆ set_is_resizable()

virtual ITensorInfo& set_is_resizable ( bool is_resizable )

pure virtual

Set the flag whether the tensor size can be changed.

Parameters

[in] is_resizable Flag that marks the tensor if it can be changed or not.

Returns: Reference to this ITensorInfo object

Implemented in TensorInfo, and SubTensorInfo.

Referenced by CpuGemmConv2d::configure(), and SubTensorInfo::set_is_resizable().

◆ set_lock_paddings()

virtual ITensorInfo& set_lock_paddings ( bool flag )

pure virtual

Set the lock paddings flag of the tensor.

It should be set to True, when the tensor could be mapped to camera or frame buffer.

Returns: Reference to this ITensorInfo object

Implemented in TensorInfo, and SubTensorInfo.

◆ set_num_channels()

virtual ITensorInfo& set_num_channels ( int num_channels )

pure virtual

Set the number of channels to the specified value.

Warning: This resets the format to UNKNOWN.

Parameters

[in] num_channels New number of channels.

Returns: Reference to this ITensorInfo object

Implemented in TensorInfo, and SubTensorInfo.

Referenced by arm_compute::auto_init_if_empty(), main(), SubTensorInfo::set_num_channels(), NEReductionOperation::validate(), CLArgMinMaxLayer::validate(), and CLReductionOperation::validate().

◆ set_quantization_info()

virtual ITensorInfo& set_quantization_info ( const QuantizationInfo & quantization_info )

pure virtual

Set the quantization settings (scale and offset) of the tensor.

Parameters

[in] quantization_info QuantizationInfo containing the scale and offset

Returns: Reference to this ITensorInfo object

Implemented in TensorInfo, and SubTensorInfo.

Referenced by arm_compute::auto_init_if_empty(), NEQLSTMLayerNormalizationKernel::configure(), NELSTMLayerQuantized::configure(), CLGEMMDeconvolutionLayer::configure(), NEQLSTMLayer::configure(), arm_compute::test::validation::DATA_TEST_CASE(), SubTensorInfo::set_quantization_info(), NEReductionOperation::validate(), CLArgMinMaxLayer::validate(), CLReductionOperation::validate(), NEGenerateProposalsLayer::validate(), and CLGenerateProposalsLayer::validate().

◆ set_tensor_dims_state()

virtual ITensorInfo& set_tensor_dims_state ( const TensorDimsState & state )

pure virtual

Set the state for each dimension of the tensor.

This sets the state of each dimension of the shape in terms of dynamic behavior using -1 where appropriate. The index in the state is a 1 to 1 mapping with the shape dimension index. For example if you want to express [?, 3, 3] as a dynamic input then [-1, 3, 3] has to be set as a state

Parameters

[in] state Tensor dimensions state

Returns: Reference to this ITensorInfo object

Implemented in TensorInfo, and SubTensorInfo.

◆ set_tensor_shape()

virtual ITensorInfo& set_tensor_shape ( const TensorShape & shape )

pure virtual

Set the shape of an already initialized tensor.

Warning: Changing the shape requires to recompute the strides and is therefore only possible if the tensor hasn't been allocated yet.

Parameters

[in] shape New tensor shape.

Returns: Reference to this ITensorInfo object

Implemented in TensorInfo, and SubTensorInfo.

Referenced by arm_compute::auto_init_if_empty(), NEQLSTMLayer::configure(), NECropKernel::configure_output_shape(), CpuMatMul::run(), SubTensorInfo::set_tensor_shape(), NEReductionOperation::validate(), CLArgMinMaxLayer::validate(), and CLReductionOperation::validate().

◆ set_valid_region()

virtual void set_valid_region ( const ValidRegion & valid_region )

pure virtual

Set the valid region of the tensor.

Parameters

[in] valid_region Valid region to set.

Implemented in TensorInfo, and SubTensorInfo.

Referenced by CPPDetectionOutputLayer::configure(), CPPUpsampleKernel::configure(), CPPPermuteKernel::configure(), ITensor::copy_from(), CPPDetectionOutputLayer::run(), SubTensorInfo::set_tensor_shape(), AccessWindowAutoPadding::set_valid_region(), AccessWindowStatic::set_valid_region(), and AccessWindowRectangle::set_valid_region().

◆ strides_in_bytes()

virtual const Strides& strides_in_bytes ( ) const

pure virtual

The strides in bytes for accessing each dimension of the tensor.

Returns: Strides in bytes for each tensor dimension

Implemented in TensorInfo, and SubTensorInfo.

Referenced by arm_compute::calculate_squashed_or_max_window(), arm_compute::colorconvert_iyuv_to_rgb(), arm_compute::colorconvert_nv12_to_rgb(), NEGatherKernel::configure(), arm_compute::cpu::kernels::convolve_nchw(), arm_compute::cpu::kernels::convolve_nhwc(), arm_compute::cpu::directconv3d_float_neon_ndhwc(), arm_compute::cpu::directconv3d_quantized_neon_ndhwc(), arm_compute::cpu::matrix_matrix_multiply_f32(), arm_compute::cpu::max_unpooling(), CpuWinogradConv2d::prepare(), CLROIPoolingLayerKernel::run(), CLROIAlignLayerKernel::run(), CLDepthwiseConvolutionLayerNativeKernel::run(), CpuWinogradConv2dTransformInputKernel::run_op(), ClSoftmaxKernel::run_op(), ClMatMulNativeKernel::run_op(), CpuWinogradConv2dTransformOutputKernel::run_op(), SubTensorInfo::strides_in_bytes(), AccessWindowStatic::update_window_if_needed(), AccessWindowRectangle::update_window_if_needed(), and arm_compute::cpu::vector_matrix_multiply_f32().

◆ tensor_dims_state()

virtual const TensorDimsState& tensor_dims_state ( ) const

pure virtual

State of each dimension of the tensor shape.

Returns: A vector with the state for each dimension of the tensor, where -1 specifies dynamic dimension

Implemented in TensorInfo, and SubTensorInfo.

◆ tensor_shape()

virtual const TensorShape& tensor_shape ( ) const

pure virtual

Size for each dimension of the tensor.

Returns: A vector with the size for each dimension of the tensor

Implemented in TensorInfo, and SubTensorInfo.

◆ total_size()

virtual size_t total_size ( ) const

pure virtual

Returns the total size of the tensor in bytes.

Returns: Total size of the tensor in bytes.

Implemented in TensorInfo, and SubTensorInfo.

Referenced by CLAuxTensorHandler::CLAuxTensorHandler(), CpuMatMul::configure(), CpuGemmDirectConv2d::configure(), CpuAuxTensorHandler::CpuAuxTensorHandler(), SubTensorInfo::extend_padding(), ITensorV2::get_size(), ClGemmLowpMatrixMultiplyCore::prepare(), NEQLSTMLayer::prepare(), CPPUpsampleKernel::run(), NELSTMLayer::run(), arm_compute::cpu::select_op_not_same_rank(), Accessor::size(), SubTensorInfo::total_size(), AccessWindowStatic::update_window_if_needed(), AccessWindowRectangle::update_window_if_needed(), NELogicalKernel::validate(), NEQLSTMLayerNormalizationKernel::validate(), NEFlattenLayer::validate(), NESelectKernel::validate(), NEFFT2D::validate(), CLFlattenLayer::validate(), NEFFT1D::validate(), NECropKernel::validate(), CLFFT2D::validate(), CLFFT1D::validate(), CLROIPoolingLayerKernel::validate(), NEPadLayer::validate(), NECropResize::validate(), ClGemmLowpMatrixMultiplyCore::validate(), CLArgMinMaxLayer::validate(), CpuGemm::validate(), CLReductionOperation::validate(), NEGenerateProposalsLayer::validate(), NEFFTConvolutionLayer::validate(), CLCropResize::validate(), NELSTMLayerQuantized::validate(), CLFFTConvolutionLayer::validate(), CLGenerateProposalsLayer::validate(), CLLSTMLayerQuantized::validate(), NEQLSTMLayer::validate(), and CLQLSTMLayer::validate().

◆ valid_region()

virtual ValidRegion valid_region ( ) const

pure virtual

Valid region of the tensor.

All elements in the valid region have defined values, i.e. are not undefined.

Returns: The valid region.

Implemented in TensorInfo, and SubTensorInfo.

Referenced by arm_compute::cpu::neon_softmax_logits_1d_float(), arm_compute::cpu::neon_softmax_logits_1d_quantized(), SubTensorInfo::set_valid_region(), arm_compute::cpu::sve2_softmax_logits_1d_quantized(), and arm_compute::cpu::sve_softmax_logits_1d_float().

Field Documentation

◆ invalid_tensor_id

constexpr Id invalid_tensor_id = 0

staticconstexpr

An invalid tensor id within a domain.

Definition at line 52 of file ITensorInfo.h.

Referenced by GpuSoftmax::create_op(), GpuKernelVariableTable::TensorVariable::has_valid_id(), ITensorInfo::has_valid_id(), arm_compute::experimental::dynamic_fusion::is_alloc_tensor(), and arm_compute::experimental::dynamic_fusion::is_noalloc_tensor().

The documentation for this class was generated from the following file:

arm_compute/core/ITensorInfo.h

Public Types

Public Member Functions

Static Public Member Functions

Static Public Attributes

Detailed Description

Member Typedef Documentation

◆ Id

◆ TensorDimsState

Constructor & Destructor Documentation

◆ ~ITensorInfo()

Member Function Documentation

◆ are_values_constant()

◆ auto_padding()

◆ broadcast_shape_and_valid_region()

◆ data_layout()

◆ data_type()

◆ dimension() [1/2]

◆ dimension() [2/2]

◆ element_size()

◆ extend_padding()

◆ format()

◆ get_dynamic_state_value()

◆ get_static_state_value()

◆ has_padding()

◆ has_valid_id()

◆ id()

◆ is_dynamic()

◆ is_resizable()

◆ lock_paddings()

◆ num_channels()

◆ num_dimensions()

◆ offset_element_in_bytes()

◆ offset_first_element_in_bytes()

◆ padding()

◆ quantization_info()

◆ reset_padding()

◆ set_are_values_constant()

◆ set_data_layout()

◆ set_data_type()

◆ set_format()

◆ set_id()

◆ set_is_resizable()

◆ set_lock_paddings()

◆ set_num_channels()

◆ set_quantization_info()

◆ set_tensor_dims_state()

◆ set_tensor_shape()

◆ set_valid_region()

◆ strides_in_bytes()

◆ tensor_dims_state()

◆ tensor_shape()

◆ total_size()

◆ valid_region()

Field Documentation

◆ invalid_tensor_id