TensorShape Class Reference

Shape of a tensor. More...

#include <TensorShape.h>

Collaboration diagram for TensorShape:

Public Member Functions
template<typename... Ts>
	TensorShape (Ts... dims)
	Constructor to initialize the tensor shape. More...
	TensorShape (const TensorShape &)=default
	Allow instances of this class to be copy constructed. More...
TensorShape &	operator= (const TensorShape &)=default
	Allow instances of this class to be copied. More...
	TensorShape (TensorShape &&)=default
	Allow instances of this class to be move constructed. More...
TensorShape &	operator= (TensorShape &&)=default
	Allow instances of this class to be moved. More...
	~TensorShape ()=default
	Default destructor. More...
TensorShape &	set (size_t dimension, size_t value, bool apply_dim_correction=true, bool increase_dim_unit=true)
	Accessor to set the value of one of the dimensions. More...
void	remove_dimension (size_t n)
	Accessor to remove the dimension n from the tensor shape. More...
void	collapse (size_t n, size_t first=0)
	Collapse the first n dimensions. More...
void	shift_right (size_t step)
	Shifts right the tensor shape increasing its dimensions. More...
TensorShape	collapsed_from (size_t start) const
	Return a copy with collapsed dimensions starting from a given point. More...
size_t	total_size () const
	Collapses all dimensions to a single linear total size. More...
size_t	total_size_upper (size_t dimension) const
	Collapses given dimension and above. More...
size_t	total_size_lower (size_t dimension) const
	Compute size of dimensions lower than the given one. More...
Public Member Functions inherited from Dimensions< size_t >
	Dimensions (Ts... dims)
	Constructor to initialize the tensor shape. More...
	Dimensions (const Dimensions &)=default
	Allow instances of this class to be copy constructed. More...
	Dimensions (Dimensions &&)=default
	Allow instances of this class to be move constructed. More...
Dimensions &	operator= (const Dimensions &)=default
	Allow instances of this class to be copied. More...
Dimensions &	operator= (Dimensions &&)=default
	Allow instances of this class to be moved. More...
void	set (size_t dimension, size_t value, bool increase_dim_unit=true)
	Accessor to set the value of one of the dimensions. More...
size_t	x () const
	Alias to access the size of the first dimension. More...
size_t	y () const
	Alias to access the size of the second dimension. More...
size_t	z () const
	Alias to access the size of the third dimension. More...
void	increment (size_t dim, size_t step=1)
	Increments the given dimension by a step size, avoiding overflows. More...
const size_t &	operator[] (size_t dimension) const
	Generic accessor to get the size of any dimension. More...
size_t &	operator[] (size_t dimension)
	Generic accessor to get the size of any dimension. More...
unsigned int	num_dimensions () const
	Returns the effective dimensionality of the tensor. More...
void	set_num_dimensions (size_t num_dimensions)
	Set number of dimensions. More...
void	collapse (const size_t n, const size_t first=0)
	Collapse dimensions. More...
void	collapse_from (size_t start)
	Collapse dimensions starting from a given point. More...
void	remove (size_t idx)
	Remove dimension of a given index. More...
std::array< size_t, num_max_dimensions >::iterator	begin ()
	Returns a read/write iterator that points to the first element in the dimension array. More...
std::array< size_t, num_max_dimensions >::const_iterator	begin () const
	Returns a read-only (constant) iterator that points to the first element in the dimension array. More...
std::array< size_t, num_max_dimensions >::const_iterator	cbegin () const
	Returns a read-only (constant) iterator that points to the first element in the dimension array. More...
std::array< size_t, num_max_dimensions >::iterator	end ()
	Returns a read/write iterator that points one past the last element in the dimension array. More...
std::array< size_t, num_max_dimensions >::const_iterator	end () const
	Returns a read-only (constant) iterator that points one past the last element in the dimension array. More...
std::array< size_t, num_max_dimensions >::const_iterator	cend () const
	Returns a read-only (constant) iterator that points one past the last element in the dimension array. More...

Static Public Member Functions
template<typename... Shapes>
static TensorShape	broadcast_shape (const Shapes &... shapes)
	If shapes are broadcast compatible, return the broadcasted shape. More...

Additional Inherited Members
Static Public Attributes inherited from Dimensions< size_t >
static constexpr size_t	num_max_dimensions
	Number of dimensions the tensor has. More...

Detailed Description

Shape of a tensor.

Definition at line 39 of file TensorShape.h.

Constructor & Destructor Documentation

TensorShape() [1/3]

TensorShape ( Ts...  dims )

inline

Constructor to initialize the tensor shape.

Parameters

[in]

dims

Values to initialize the dimensions.

Definition at line 47 of file TensorShape.h.

References arm_compute::test::validation::fill(), TensorShape::operator=(), and TensorShape::~TensorShape().

        : Dimensions{ dims... }
    {
        // Initialize unspecified dimensions to 1
        if(_num_dimensions > 0)
        {
            std::fill(_id.begin() + _num_dimensions, _id.end(), 1);
        }

        // Correct number dimensions to ignore trailing dimensions of size 1
        apply_dimension_correction();
    }

TensorShape() [2/3]

TensorShape ( const TensorShape &  )

default

Allow instances of this class to be copy constructed.

TensorShape() [3/3]

TensorShape ( TensorShape &&  )

default

Allow instances of this class to be move constructed.

~TensorShape()

~TensorShape ( )

default

Default destructor.

Referenced by TensorShape::TensorShape().

Member Function Documentation

broadcast_shape()

static TensorShape broadcast_shape ( const Shapes &...  shapes )

inlinestatic

If shapes are broadcast compatible, return the broadcasted shape.

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]

shapes

Tensor shapes.

Returns

The broadcasted shape or an empty shape if the shapes are not broadcast compatible.

Definition at line 211 of file TensorShape.h.

References arm_compute::utility::for_each(), Dimensions< T >::num_dimensions(), Dimensions< size_t >::num_max_dimensions, TensorShape::set(), and arm_compute::U.

Referenced by PostOpCLKernelUtils::are_post_op_shapes_compliant(), arm_compute::test::validation::reference::arithmetic_division(), arm_compute::test::validation::reference::arithmetic_operation(), ITensorInfo::broadcast_shape_and_valid_region(), arm_compute::test::validation::reference::compare(), NELogicalKernel::configure(), CpuSubKernel::configure(), CpuMulKernel::configure(), CpuAddKernel::configure(), ClMulKernel::configure(), ClComplexMulKernel::configure(), CpuComplexMulKernel::configure(), EltwiseLayerNode::configure_output(), CpuComparisonKernel::get_available_kernels(), arm_compute::intersect_valid_regions(), arm_compute::test::validation::reference::logical_and(), arm_compute::test::validation::reference::logical_or(), CpuMulKernel::name(), arm_compute::test::validation::reference::pixel_wise_multiplication(), ClMulKernel::run_op(), NELogicalKernel::validate(), and ClComponentElementwiseBinary::validate().

    {
        TensorShape bc_shape;

        auto broadcast = [&bc_shape](const TensorShape & other)
        {
            if(bc_shape.num_dimensions() == 0)
            {
                bc_shape = other;
            }
            else if(other.num_dimensions() != 0)
            {
                for(size_t d = 0; d < TensorShape::num_max_dimensions; ++d)
                {
                    const size_t dim_min = std::min(bc_shape[d], other[d]);
                    const size_t dim_max = std::max(bc_shape[d], other[d]);

                    if((dim_min != 1) && (dim_min != dim_max))
                    {
                        bc_shape = TensorShape{ 0U };
                        break;
                    }

                    bc_shape.set(d, dim_max);
                }
            }
        };

        utility::for_each(broadcast, shapes...);

        return bc_shape;
    }

collapse()

void collapse ( size_t  n, size_t  first = 0  )

inline

Collapse the first n dimensions.

Parameters

[in]	n	Number of dimensions to collapse into first
[in]	first	Dimensions into which the following n are collapsed.

Definition at line 133 of file TensorShape.h.

References Dimensions< T >::collapse(), and arm_compute::test::validation::fill().

Referenced by TensorShape::collapsed_from(), arm_compute::misc::shape_calculator::compute_softmax_shape(), arm_compute::misc::shape_calculator::compute_weights_reshaped_shape(), ClTransposedConvolutionKernel::configure(), CpuGemmLowpMatrixMultiplyKernel::configure(), ClIndirectConv2dKernel::configure(), ClDirectConv2dKernel::configure(), FlattenLayerNode::configure_output(), and ClTemplateElementwiseBinary::get_window().

    {
        Dimensions::collapse(n, first);

        // Make sure all empty dimensions are filled with 1
        std::fill(_id.begin() + _num_dimensions, _id.end(), 1);
    }

collapsed_from()

TensorShape collapsed_from ( size_t  start ) const

inline

Return a copy with collapsed dimensions starting from a given point.

Parameters

[in]

start

Starting point of collapsing dimensions.

Returns

A copy with collapse dimensions starting from start.

Definition at line 161 of file TensorShape.h.

References TensorShape::collapse(), arm_compute::test::validation::reference::copy(), and Dimensions< size_t >::num_dimensions().

Referenced by CpuMatMul::configure(), CpuMatMul::run(), CLComparisonKernel::run(), ClElementwiseKernel::run_op(), ClMulKernel::run_op(), and ClComplexMulKernel::run_op().

    {
        TensorShape copy(*this);
        copy.collapse(num_dimensions() - start, start);
        return copy;
    }

operator=() [1/2]

TensorShape& operator= ( const TensorShape &  )

default

Allow instances of this class to be copied.

Referenced by TensorShape::TensorShape().

operator=() [2/2]

TensorShape& operator= ( TensorShape &&  )

default

Allow instances of this class to be moved.

remove_dimension()

void remove_dimension ( size_t  n )

inline

Accessor to remove the dimension n from the tensor shape.

Note

The upper dimensions of the tensor shape will be shifted down by 1

Parameters

[in]

n

Dimension to remove

Definition at line 111 of file TensorShape.h.

References ARM_COMPUTE_ERROR_ON, arm_compute::test::validation::reference::copy(), arm_compute::test::validation::fill(), and arm_compute::test::validation::n.

Referenced by arm_compute::misc::shape_calculator::calculate_reduce_mean_shape(), arm_compute::misc::shape_calculator::calculate_unstack_shape(), arm_compute::misc::shape_calculator::compute_lhs_reshaped_shape(), arm_compute::misc::shape_calculator::compute_reductionA_shape(), arm_compute::misc::shape_calculator::compute_reductionB_shape(), NEReduceMean::configure(), CLReduceMean::configure(), NEFFTConvolutionLayer::configure(), and CLFFTConvolutionLayer::configure().

    {
        ARM_COMPUTE_ERROR_ON(_num_dimensions < 1);
        ARM_COMPUTE_ERROR_ON(n >= _num_dimensions);

        std::copy(_id.begin() + n + 1, _id.end(), _id.begin() + n);

        // Reduce number of dimensions
        _num_dimensions--;

        // Make sure all empty dimensions are filled with 1
        std::fill(_id.begin() + _num_dimensions, _id.end(), 1);

        // Correct number dimensions to ignore trailing dimensions of size 1
        apply_dimension_correction();
    }

set()

TensorShape& set ( size_t  dimension, size_t  value, bool  apply_dim_correction = true, bool  increase_dim_unit = true  )

inline

Accessor to set the value of one of the dimensions.

Parameters

[in]	dimension	Dimension for which the value is set.
[in]	value	Value to be set for the dimension.
[in]	apply_dim_correction	(Optional) Flag to state whether apply dimension correction after setting one dimension. E.g. when permuting NCHW -> NHWC, 1x1x2 would become 2x1x1, but _num_dimensions should be 3 rather than 1.
[in]	increase_dim_unit	(Optional) Set to true if new unit dimensions increase the number of dimensions of the shape.

Returns

*this.

Definition at line 79 of file TensorShape.h.

References arm_compute::test::validation::fill(), and Dimensions< T >::set().

Referenced by GraphBuilder::add_convolution_node(), GraphBuilder::add_deconvolution_node(), GraphBuilder::add_depthwise_convolution_node(), GraphBuilder::add_scale_layer(), arm_compute::adjust_odd_shape(), TensorShape::broadcast_shape(), ITensorInfo::broadcast_shape_and_valid_region(), arm_compute::misc::shape_calculator::calculate_concatenate_shape(), arm_compute::misc::shape_calculator::calculate_reduce_mean_shape(), arm_compute::calculate_subsampled_shape(), arm_compute::calculate_valid_region_scale(), arm_compute::misc::shape_calculator::compute_deconvolution_upsampled_shape(), arm_compute::misc::shape_calculator::compute_gather_shape(), arm_compute::misc::shape_calculator::compute_interleaved_shape(), arm_compute::misc::shape_calculator::compute_lhs_reshaped_shape(), ReorgLayerNode::compute_output_descriptor(), DetectionOutputLayerNode::compute_output_descriptor(), PoolingLayerNode::compute_output_descriptor(), PriorBoxLayerNode::compute_output_descriptor(), DeconvolutionLayerNode::compute_output_descriptor(), SplitLayerNode::compute_output_descriptor(), FusedConvolutionBatchNormalizationWithPostOpsNode::compute_output_descriptor(), FusedConvolutionBatchNormalizationNode::compute_output_descriptor(), FusedDepthwiseConvolutionBatchNormalizationNode::compute_output_descriptor(), DepthwiseConvolutionLayerNode::compute_output_descriptor(), ConvolutionLayerNode::compute_output_descriptor(), FusedConvolutionWithPostOpNode::compute_output_descriptor(), arm_compute::misc::shape_calculator::compute_padded_shape(), arm_compute::misc::shape_calculator::compute_reductionB_shape(), arm_compute::misc::shape_calculator::compute_rhs_reshaped_shape(), arm_compute::misc::shape_calculator::compute_split_shape(), arm_compute::misc::shape_calculator::compute_stack_shape(), arm_compute::helpers::tensor_transform::compute_strided_slice_output_shape(), arm_compute::misc::shape_calculator::compute_tiled_shape(), arm_compute::misc::shape_calculator::compute_transpose1xW_shape(), arm_compute::misc::shape_calculator::compute_transpose1xW_with_element_size_shape(), arm_compute::misc::shape_calculator::compute_transposed_shape(), arm_compute::misc::shape_calculator::compute_unpool_shape(), AccessWindowTranspose::compute_valid_region(), AccessWindowStatic::compute_valid_region(), AccessWindowRectangle::compute_valid_region(), arm_compute::misc::shape_calculator::compute_vector_to_tensor_output_shape(), arm_compute::misc::shape_calculator::compute_weights_reshaped_shape(), CpuGemmMatrixMultiplyKernel::configure(), CpuSoftmaxGeneric< IS_LOG >::configure(), NEReduceMean::configure(), CLReduceMean::configure(), ClGemmConv2d::configure(), CpuGemmConv2d::configure(), ResizeLayerNode::configure_output(), PadLayerNode::configure_output(), ROIAlignLayerNode::configure_output(), SplitLayerNode::configure_output(), GpuSoftmax::create_op(), arm_compute::test::validation::DATA_TEST_CASE(), arm_compute::test::validation::reference::deconvolution_layer(), NPYLoader::fill_tensor(), arm_compute::test::index2coord(), arm_compute::index2coords(), NPYLoader::init_tensor(), arm_compute::intersect_valid_regions(), GpuSoftmax::is_supported_op(), arm_compute::permute(), arm_compute::test::validation::reference::scale_core(), GraphEdsr::setup(), Window::shape(), arm_compute::test::shape_to_valid_region(), arm_compute::test::validation::transpose(), ClSoftmax::validate(), CpuSoftmaxGeneric< IS_LOG >::validate(), NEReductionOperation::validate(), CLArgMinMaxLayer::validate(), ClGemmConv2d::validate(), CpuGemmConv2d::validate(), and CpuGemmLowpMatrixMultiplyCore::validate().

    {
        // Clear entire shape if one dimension is zero
        if(value == 0)
        {
            _num_dimensions = 0;
            std::fill(_id.begin(), _id.end(), 0);
        }
        else
        {
            // Make sure all empty dimensions are filled with 1
            std::fill(_id.begin() + _num_dimensions, _id.end(), 1);

            // Set the specified dimension and increase the number of dimensions if
            // necessary
            Dimensions::set(dimension, value, increase_dim_unit);

            // Correct number dimensions to ignore trailing dimensions of size 1
            if(apply_dim_correction)
            {
                apply_dimension_correction();
            }
        }
        return *this;
    }

shift_right()

void shift_right ( size_t  step )

inline

Shifts right the tensor shape increasing its dimensions.

Parameters

[in]

step

Rotation step

Definition at line 144 of file TensorShape.h.

References ARM_COMPUTE_ERROR_ON, Dimensions< size_t >::begin(), Dimensions< size_t >::end(), Dimensions< size_t >::num_dimensions(), Dimensions< size_t >::num_max_dimensions, and arm_compute::cpu::step.

Referenced by arm_compute::misc::shape_calculator::compute_col2im_shape(), arm_compute::misc::shape_calculator::compute_output_stage_shape(), and arm_compute::misc::shape_calculator::compute_softmax_shape().

    {
        ARM_COMPUTE_ERROR_ON(step > TensorShape::num_max_dimensions - num_dimensions());

        std::rotate(begin(), begin() + TensorShape::num_max_dimensions - step, end());
        _num_dimensions += step;

        // Correct number dimensions to ignore trailing dimensions of size 1
        apply_dimension_correction();
    }

total_size()

size_t total_size ( ) const

inline

Collapses all dimensions to a single linear total size.

Returns

The total tensor size in terms of elements.

Definition at line 172 of file TensorShape.h.

References arm_compute::test::validation::reference::accumulate().

Referenced by arm_compute::auto_init_if_empty(), TensorInfo::auto_padding(), arm_compute::test::validation::reference::col2im(), arm_compute::misc::shape_calculator::compute_gather_shape(), ClMatMulLowpNativeKernel::configure(), CLSelectKernel::configure(), NESpaceToBatchLayer::configure(), CLSpaceToBatchLayer::configure(), GenerateProposalsLayerNode::configure_output(), EltwiseLayerNode::configure_output(), arm_compute::test::coord2index(), arm_compute::coords2index(), arm_compute::test::validation::DATA_TEST_CASE(), arm_compute::export_to_cl_image(), CpuComparisonKernel::get_available_kernels(), ClTemplateCast::get_window(), ClTemplateLogits1DMaxShiftExpSum::get_window(), ClTemplateLogits1DNorm::get_window(), ClTemplateReshape::get_window(), ClTemplateDepthwiseConv2d::get_window(), ClTemplateDirectConv2d::get_window(), ClTemplateElementwiseBinary::get_window(), ClTemplateResize::get_window(), ClTemplateActivation::get_window(), ClTemplatePool2d::get_window(), arm_compute::test::index2coord(), arm_compute::index2coords(), CpuMulKernel::name(), Accessor::num_elements(), CLAccessor::num_elements(), NumPyAccessor::NumPyAccessor(), arm_compute::test::validation::reference::reshape_layer(), CLComparisonKernel::run(), ClElementwiseKernel::run_op(), ClMulKernel::run_op(), ClComplexMulKernel::run_op(), arm_compute::set_shape_if_empty(), SubTensorInfo::set_tensor_shape(), SubTensorInfo::set_valid_region(), arm_compute::test::validation::TEST_CASE(), CPPSplit< CLSlice, ICLTensor >::validate(), NELogicalKernel::validate(), CpuConcatenate::validate(), ClConcatenate::validate(), ClComponentActivation::validate(), ClComponentResize::validate(), ClComponentElementwiseBinary::validate(), ClComponentLogits1DNorm::validate(), ClComponentLogits1DMaxShiftExpSum::validate(), and NEDeconvolutionLayer::validate().

    {
        return std::accumulate(_id.begin(), _id.end(), 1, std::multiplies<size_t>());
    }

total_size_lower()

size_t total_size_lower ( size_t  dimension ) const

inline

Compute size of dimensions lower than the given one.

Parameters

[in]

dimension

Upper boundary.

Returns

The linear size of the collapsed dimensions.

Definition at line 194 of file TensorShape.h.

References arm_compute::test::validation::reference::accumulate(), ARM_COMPUTE_ERROR_ON, and Dimensions< size_t >::num_max_dimensions.

Referenced by ClPool2dKernel::configure(), CpuConvertFullyConnectedWeightsKernel::validate(), and ClConvertFullyConnectedWeightsKernel::validate().

    {
        ARM_COMPUTE_ERROR_ON(dimension > TensorShape::num_max_dimensions);
        return std::accumulate(_id.begin(), _id.begin() + dimension, 1, std::multiplies<size_t>());
    }

total_size_upper()

size_t total_size_upper ( size_t  dimension ) const

inline

Collapses given dimension and above.

Parameters

[in]

dimension

Size of the wanted dimension

Returns

The linear size of the collapsed dimensions

Definition at line 182 of file TensorShape.h.

References arm_compute::test::validation::reference::accumulate(), ARM_COMPUTE_ERROR_ON, and Dimensions< size_t >::num_max_dimensions.

Referenced by ClKernelRuntime::configure(), ClWinogradInputTransformKernel::configure(), ClWinogradOutputTransformKernel::configure(), ITensor::copy_from(), arm_compute::test::validation::reference::fully_connected_layer(), and ClWinogradInputTransformKernel::run_op().

    {
        ARM_COMPUTE_ERROR_ON(dimension >= TensorShape::num_max_dimensions);
        return std::accumulate(_id.begin() + dimension, _id.end(), 1, std::multiplies<size_t>());
    }

---

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

• arm_compute/core/TensorShape.h