arm_compute Namespace Reference

Namespaces
namespace	misc
namespace	weights_transformations

Data Structures
class	CLCastBool
	Class to run CLCastBoolKernel. This converts the boolean input tensor to the output tensor's type. More...
class	CLCastBoolKernel
	Class for the kernel converting boolean type. More...
class	CLDirectTransposeConvLayer
class	CLEmbeddingLookupKernel
	Class to perform EmbeddingLookup operation with opencl kernel. More...
class	CLFullyConnectedHybridLayer
class	CLFullyConnectedLayerEx
class	CLFullyConnectedReshapingLayer
	Class to run FullyConnected Layer after reshaping input tensor. More...
class	CLGatherExKernel
	Class to define an interface for the gather kernel. More...
class	CLGEMMMatrixAccumulateBiasesKernel
class	CLHashtableLookup
	Class to perform HashtableLookup operation. More...
class	CLHashtableLookupKernel
	Class to perform HashtableLookup operation with opencl kernel. More...
class	CLInstanceNormalizationLayerKernelEx
class	CLKernelLibraryEx
	Class to build OpenCL kernels added from nnfw. More...
class	CLMemsetKernel
class	CLMultiplyScaleFactorKernel
class	CLOneHot
class	CLOneHotKernel
class	CLPadLayerEx
class	CLPadLayerKernelEx
class	CLQuantizationSymmetricKernel
class	CLReduceOperation
	Class to perform ReduceOperation. More...
class	CLReduceOperationKernel
	Class to define interface for the reduce operation kernel. More...
class	CLScaleFactorSymm8Kernel
class	CLTopKV2
	Class to execute TopKV2 operation. More...
class	CLTransposeConvLayer
class	NECastBool
	Class to run INESimpleFunctionNoBorder. More...
class	NECastBoolKernel
	Class for the kernel converting boolean type. More...
class	NEEmbeddingLookupKernel
class	NEFullyConnectedHybridLayer
class	NEFullyConnectedLayerEx
class	NEFullyConnectedReshapingLayer
	Class to run FullyConnected Layer after reshaping input tensor. More...
class	NEGEMMMatrixAccumulateBiasesKernel
class	NEHashtableLookup
	Class to perform HashtableLookup operation. More...
class	NEHashtableLookupKernel
class	NEInstanceNormalizationLayerKernelEx
class	NEMultiplyScaleFactorKernel
class	NEOneHot
class	NEOneHotKernel
class	NEQuantizationSymmetricKernel
class	NEReduceOperation
class	NEReduceSum

Functions
const std::pair< unsigned int, unsigned int >	transposeconv_output_dimensions (unsigned int in_width, unsigned int in_height, unsigned int kernel_width, unsigned int kernel_height, const PadStrideInfo &info, unsigned int invalid_right, unsigned int invalid_top)

Function Documentation

transposeconv_output_dimensions()

const std::pair< unsigned int, unsigned int > arm_compute::transposeconv_output_dimensions	(	unsigned int	in_width,
		unsigned int	in_height,
		unsigned int	kernel_width,
		unsigned int	kernel_height,
		const PadStrideInfo &	info,
		unsigned int	invalid_right,
		unsigned int	invalid_top
	)

Returns expected width and height of the transpose convolution's output tensor.

Note

This function was copied in order to fix a bug computing to wrong output dimensions.

Parameters

[in]	in_width	Width of input tensor (Number of columns)
[in]	in_height	Height of input tensor (Number of rows)
[in]	kernel_width	Kernel width.
[in]	kernel_height	Kernel height.
[in]	info	padding and stride info.
[in]	invalid_right	The number of zeros added to right edge of the output.
[in]	invalid_top	The number of zeros added to bottom edge of the output.

Returns

A pair with the new width in the first position and the new height in the second.

Definition at line 47 of file UtilsEx.cpp.

{
  const auto [stride_x, stride_y] = info.stride();
  const unsigned int padx = info.pad_left() + info.pad_right();
  const unsigned int pady = info.pad_top() + info.pad_bottom();
 
  ARM_COMPUTE_ERROR_ON(in_width < 1 || in_height < 1);
  ARM_COMPUTE_ERROR_ON(kernel_width <= padx);
  ARM_COMPUTE_ERROR_ON(kernel_height <= pady);
 
  // Find the transpose conv out dimensions
  // transpose conv out:
  //    tconv_out + pad = 1 + (in - 1) * stride + invalid
  //    tconv_out = 1 + (in - 1) * stride + invalid - pad
  const int w = stride_x * (in_width - 1) + kernel_width - padx + invalid_right;
  const int h = stride_y * (in_height - 1) + kernel_height - pady + invalid_bottom;
 
  return std::make_pair<unsigned int, unsigned int>(w, h);
}

References info.

Referenced by arm_compute::CLDirectTransposeConvLayer::configure(), and arm_compute::CLDirectTransposeConvLayer::validate().