AMD Neural Network Extension API

AMD Neural Network Extension API#

MIVisionX: Extension: AMD Neural Network Extension API

AMD OpenVX Neural Network Nodes Extension to enhance Khronos Convolutional Network Nodes Extension. More...

Data Structures
struct	_vx_amd_custom_params_t
	Input parameters for a convolution operation. More...

Macros
#define	VX_NN_ACTIVATION_LEAKY_RELU (VX_ENUM_BASE(VX_ID_AMD, VX_ENUM_NN_ACTIVATION_FUNCTION_TYPE) + 0x9)
	The Neural Network activation functions vx_nn_activation_function_e extension.

Functions
VX_API_ENTRY vx_node VX_API_CALL	vxBatchNormalizationLayer (vx_graph graph, vx_tensor inputs, vx_tensor mean, vx_tensor variance, vx_tensor scale, vx_tensor bias, vx_float32 eps, vx_tensor output)
	[Graph] Creates a Batch Normalization Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxScaleLayer (vx_graph graph, vx_tensor inputs, vx_tensor scale, vx_tensor bias, vx_tensor output)
	[Graph] Creates a Scale Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxArgmaxLayer (vx_graph graph, vx_tensor input, vx_reference output)
	[Graph] Creates a Argmax Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxConvertImageToTensorNode (vx_graph graph, vx_image input, vx_tensor output, vx_float32 a, vx_float32 b, vx_bool reverse_channel_order)
	[Graph] Creates a Image to Tensor Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxConvertTensorToImageNode (vx_graph graph, vx_tensor input, vx_image output, vx_float32 a, vx_float32 b, vx_bool reverse_channel_order)
	[Graph] Creates a Tensor to Image Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxConcatLayer (vx_graph graph, vx_tensor output, vx_tensor input1, vx_tensor input2, vx_tensor input3, vx_tensor input4, vx_tensor input5, vx_tensor input6, vx_tensor input7, vx_tensor input8, vx_int32 axis)
	[Graph] Creates a Concat Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxSliceLayer (vx_graph graph, vx_tensor input, vx_tensor output1, vx_tensor output2, vx_tensor output3, vx_tensor output4, vx_tensor output5, vx_tensor output6, vx_tensor output7, vx_tensor output8)
	[Graph] Creates a Slice Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxUpsampleNearestLayer (vx_graph graph, vx_tensor input, vx_tensor output)
	[Graph] Creates a Convolutional Network Upsampling Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxReshapeLayer (vx_graph graph, vx_tensor input, vx_tensor output)
	[Graph] Creates a Convolutional Network Reshape Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxPermuteLayer (vx_graph graph, vx_tensor input, vx_array order, vx_tensor output)
	[Graph] Creates a Permute Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxPriorBoxLayer (vx_graph graph, vx_tensor input_1, vx_tensor input_2, vx_float32 minSize, vx_array aspect_ratio, vx_int32 flip, vx_int32 clip, vx_float32 offset, vx_tensor output, vx_array variance, vx_float32 maxSize)
	[Graph] Creates a Prior Box Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxCropLayer (vx_graph graph, vx_tensor input, vx_tensor ref, vx_tensor output, vx_scalar axis, vx_scalar offset1, vx_scalar offset2, vx_scalar offset3, vx_scalar offset4)
	[Graph] Creates a Crop Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxCropAndResizeLayer (vx_graph graph, vx_tensor input, vx_tensor output, vx_scalar x_coord, vx_scalar y_coord, vx_scalar width, vx_scalar height, vx_scalar scaleFactor, vx_scalar mode)
	[Graph] Creates a Crop_And_Resize Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxTensorMinNode (vx_graph graph, vx_tensor input, vx_tensor input2, vx_tensor output)
	[Graph] Creates a Tensor_Min Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxTensorMaxNode (vx_graph graph, vx_tensor input, vx_tensor input2, vx_tensor output)
	[Graph] Creates a Tensor_Max Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxDetectionOutputLayer (vx_graph graph, vx_tensor input1, vx_tensor input2, vx_tensor input3, vx_int32 num_classes, vx_int32 share_location, vx_int32 background_label_id, vx_float32 nms_threshold, vx_int32 code_type, vx_int32 keep_top_k, vx_int32 variance_encoded_in_target, vx_tensor output)
	[Graph] Creates a Detection Output Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxCastLayer (vx_graph graph, vx_tensor input, vx_int32 output_data_type, vx_tensor output)
	[Graph] Creates a Cast Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxTensorExpNode (vx_graph graph, vx_tensor input, vx_tensor output)
	[Graph] Creates a Tensor_Exp Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxTensorLogNode (vx_graph graph, vx_tensor input, vx_tensor output)
	[Graph] Creates a Tensor_Log Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxNMSLayer (vx_graph graph, vx_tensor boxes, vx_tensor scores, vx_int32 center_point_box, vx_tensor output, vx_tensor max_output_boxes_per_class, vx_tensor iou_threshold, vx_tensor score_threshold)
	[Graph] Creates a Non Max Suppression Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxGatherLayer (vx_graph graph, vx_tensor input, vx_tensor indices, vx_tensor output, vx_int32 axis)
	[Graph] Creates a Gather Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxTopKLayer (vx_graph graph, vx_tensor x_tensor, vx_tensor k_tensor, vx_int32 axis, vx_int32 largest, vx_int32 sorted, vx_tensor values, vx_tensor indices)
	[Graph] Creates a TopK Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxReduceMinLayer (vx_graph graph, vx_tensor data, vx_array axes, vx_int32 keepdims, vx_tensor reduced)
	[Graph] Creates a Reduce Min Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxTileLayer (vx_graph graph, vx_tensor input, vx_tensor repeats, vx_tensor output)
	[Graph] Creates a Tile Layer Node. More...

VX_API_ENTRY vx_node VX_API_CALL	vxTensorCompareNode (vx_graph graph, vx_tensor input, vx_tensor input2, vx_tensor output)
	[Graph] Creates a Tensor Compare Node. More...

Detailed Description

AMD OpenVX Neural Network Nodes Extension to enhance Khronos Convolutional Network Nodes Extension.

Function Documentation

◆ vxArgmaxLayer()

VX_API_ENTRY vx_node VX_API_CALL vxArgmaxLayer	(	vx_graph	graph,
		vx_tensor	input,
		vx_reference	output
	)

[Graph] Creates a Argmax Layer Node.

Parameters

[in]	graph	The handle to the graph.
[in]	inputs	The input tensor data.
[out]	outputs	The output tensor data.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxBatchNormalizationLayer()

VX_API_ENTRY vx_node VX_API_CALL vxBatchNormalizationLayer	(	vx_graph	graph,
		vx_tensor	inputs,
		vx_tensor	mean,
		vx_tensor	variance,
		vx_tensor	scale,
		vx_tensor	bias,
		vx_float32	eps,
		vx_tensor	output
	)

[Graph] Creates a Batch Normalization Layer Node.

Parameters

[in]	graph	The handle to the graph.
[in]	inputs	The input tensor data.
[in]	inputs	The mean tensor data.
[in]	inputs	The variance tensor data.
[in]	inputs	The scale tensor data.
[in]	inputs	The bias tensor data.
[in]	inputs	The eps vx_float32 data.
[out]	outputs	The output tensor data.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxCastLayer()

VX_API_ENTRY vx_node VX_API_CALL vxCastLayer	(	vx_graph	graph,
		vx_tensor	input,
		vx_int32	output_data_type,
		vx_tensor	output
	)

[Graph] Creates a Cast Layer Node.

Converts all the elements of the input tensor to the data type specified by input_2 of the node.
This function supports 2D or 4D tensors as input and output.

Parameters

[in]	graph	The handle to the graph.
[in]	input	The input tensor data. Can be VX_TYPE_FLOAT32, VX_TYPE_INT32, VX_TYPE_INT64.
[in]	output_data_type	The required output tensor data type. Integer value between 0-13.
[out]	output	The output tensor data. Output will have the same number of dimensions as input. Output tensor data type will be that specified by 'to'.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxConcatLayer()

VX_API_ENTRY vx_node VX_API_CALL vxConcatLayer	(	vx_graph	graph,
		vx_tensor	output,
		vx_tensor	input1,
		vx_tensor	input2,
		vx_tensor	input3,
		vx_tensor	input4,
		vx_tensor	input5,
		vx_tensor	input6,
		vx_tensor	input7,
		vx_tensor	input8,
		vx_int32	axis
	)

[Graph] Creates a Concat Layer Node.

Parameters

[in]	graph	The handle to the graph.
[out]	outputs	The output tensor data.
[in]	inputs	The input 1 tensor data.
[in]	inputs	The input 2 tensor data.
[in]	inputs	The input 3 tensor data.
[in]	inputs	The input 4 tensor data.
[in]	inputs	The input 5 tensor data.
[in]	inputs	The input 6 tensor data.
[in]	inputs	The input 7 tensor data.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxConvertImageToTensorNode()

VX_API_ENTRY vx_node VX_API_CALL vxConvertImageToTensorNode	(	vx_graph	graph,
		vx_image	input,
		vx_tensor	output,
		vx_float32	a,
		vx_float32	b,
		vx_bool	reverse_channel_order
	)

[Graph] Creates a Image to Tensor Node.

Parameters

[in]	graph	The handle to the graph.
[in]	inputs	The input tensor data.
[out]	outputs	The output tensor data.
[in]	inputs	The a vx_float32 data.
[in]	inputs	The b vx_float32 data.
[in]	inputs	The reverse channel order vx_bool data.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxConvertTensorToImageNode()

VX_API_ENTRY vx_node VX_API_CALL vxConvertTensorToImageNode	(	vx_graph	graph,
		vx_tensor	input,
		vx_image	output,
		vx_float32	a,
		vx_float32	b,
		vx_bool	reverse_channel_order
	)

[Graph] Creates a Tensor to Image Node.

Parameters

[in]	graph	The handle to the graph.
[in]	inputs	The input tensor data.
[out]	outputs	The output tensor data.
[in]	inputs	The a vx_float32 data.
[in]	inputs	The b vx_float32 data.
[in]	inputs	The reverse channel order vx_bool data.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxCropAndResizeLayer()

VX_API_ENTRY vx_node VX_API_CALL vxCropAndResizeLayer	(	vx_graph	graph,
		vx_tensor	input,
		vx_tensor	output,
		vx_scalar	x_coord,
		vx_scalar	y_coord,
		vx_scalar	width,
		vx_scalar	height,
		vx_scalar	scaleFactor,
		vx_scalar	mode
	)

[Graph] Creates a Crop_And_Resize Layer Node.

Cropping and Resizing is done on the width and height dimensions of the vx_tensor. Like Upsampling Layer Node, we use the term x for the width dimension and y for the height dimension.
This function supports 4D tensors as input and ouput. The type of the tensor can be either float32 or float16. There are two modes for the resize: NEAREST_NEIGHBOR(mode = 0, default) and BILINEAR_INTERPOLATION(mode = 1).

Parameters

[in]	graph	The handle to the graph.
[in]	inputs	The input tensor data.
[out]	outputs	The output tensor data. Output will have the same number of dimensions as input. Output tensor data type must be same as the inputs.
	[x_coord]	x_coord The x coordinate of the upper left point that will be cropped.
	[y_coord]	y_coord The y coordinate of the upper left point that will be cropped.
	[width]	width The width of the area that will be cropped.
	[height]	height The height of the are that will be cropped.
	[scaleFactor]	scaleFactor The scale factor that will be used to resize the cropped tensor.
	[mode]	mode The mode to decide which method will be used for the resize.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxCropLayer()

VX_API_ENTRY vx_node VX_API_CALL vxCropLayer	(	vx_graph	graph,
		vx_tensor	input,
		vx_tensor	ref,
		vx_tensor	output,
		vx_scalar	axis,
		vx_scalar	offset1,
		vx_scalar	offset2,
		vx_scalar	offset3,
		vx_scalar	offset4
	)

[Graph] Creates a Crop Layer Node.

Cropping is done on the dimensions of the input vx_tensor to fit the dimensions of the reference tensor. This function supports 4D tensors as input and ouput. The type of the tensor can be either float32 or float16.

Parameters

[in]	graph	The handle to the graph.
[in]	input	The input tensor data.
[in]	ref	The reference tensor data.
[out]	output	The cropped tensor data.
	[axis]	The dimensions including and trailing 'axis' are cropped. [n x c x h x w]
	[offset1]	The offset to set the shift for dimension n.
	[offset2]	The offset to set the shift for dimension c.
	[offset3]	The offset to set the shift for dimension h.
	[offset4]	The offset to set the shift for dimension w.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxDetectionOutputLayer()

VX_API_ENTRY vx_node VX_API_CALL vxDetectionOutputLayer	(	vx_graph	graph,
		vx_tensor	input1,
		vx_tensor	input2,
		vx_tensor	input3,
		vx_int32	num_classes,
		vx_int32	share_location,
		vx_int32	background_label_id,
		vx_float32	nms_threshold,
		vx_int32	code_type,
		vx_int32	keep_top_k,
		vx_int32	variance_encoded_in_target,
		vx_tensor	output
	)

[Graph] Creates a Detection Output Layer Node.

Gives the details of the detected ouutputs in an image with their label, confidence and the bounding box coordinates. This function supports three 4D tensors as input and one 4D tensor as ouput. The type of the tensor can be either float32 or float16.

Parameters

[in]	graph	The handle to the graph.
[in]	input1	The first input tensor data (location values).
[in]	input2	The second input tensor data (confidence values).
[in]	input3	The third input tensor data (prior box values).
[in]	num_classes	Integer value: Number of output classes. (example: tiny yolo = 20 classes)
[in]	share_location	Integer value: Label values change based on this
[in]	background_label_id	Integer value: Ignores the background classes
[in]	nms_threshold	Float value: NMS-Threshold for output boxes
[in]	code_type	Integer value: Decides if the bounding boxes are Center-type or Corner-type
[in]	keep_top_k	Integer value: Tells the number of output boxes to keep
[in]	variance_encoded_in_target	Integer value: Helps calculate the bounding box coordinates
[out]	output	The output tensor data with the same dimensions as the input tensor data.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxGatherLayer()

VX_API_ENTRY vx_node VX_API_CALL vxGatherLayer	(	vx_graph	graph,
		vx_tensor	input,
		vx_tensor	indices,
		vx_tensor	output,
		vx_int32	axis
	)

[Graph] Creates a Gather Layer Node.

Given data tensor of rank r >= 1, and indices tensor of rank q, gather entries of the axis dimension of data (by default outer-most one as axis=0) indexed by indices, and concatenates them in an output tensor of rank q + (r - 1).

Parameters

[in]	graph	The handle to the graph.
[in]	input	The input tensor data to gather elements on. The type of the tensor can be either float32 or float16.
[in]	indices	The indices tensor containing the index data. The type of the tensor can be either int32 or int64.
[out]	output	The output tensor data with the same type as the input tensor data.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxNMSLayer()

VX_API_ENTRY vx_node VX_API_CALL vxNMSLayer	(	vx_graph	graph,
		vx_tensor	boxes,
		vx_tensor	scores,
		vx_int32	center_point_box,
		vx_tensor	output,
		vx_tensor	max_output_boxes_per_class,
		vx_tensor	iou_threshold,
		vx_tensor	score_threshold
	)

[Graph] Creates a Non Max Suppression Layer Node.

Filter out boxes that have high intersection-over-union (IOU) overlap with previously selected boxes. This function supports 4D tensors as input and ouput. The type of the tensor is int64.

Parameters

[in]	graph	The handle to the graph.
[in]	boxes	The input tensor data which has coordinates for all the bounding boxes.
[in]	scores	The input tensor data which has scores for all the bounding boxes.
[in]	center_point_box	The input scalar data which tells the format of the coordinates of bounding boxes.
[in]	max_output_boxes_per_class	The input tensor int64 data representing the maximum number of boxes to be selected per batch per class.
[in]	iou_threshold	The input tensor float data representing the maximum number of boxes to be selected per batch per class.
[in]	score_threshold	The input tensor float data representing the threshold for deciding when to remove boxes based on score.
[out]	output	The output tensor data with the dimensions based on the number of selected boxes.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxPermuteLayer()

VX_API_ENTRY vx_node VX_API_CALL vxPermuteLayer	(	vx_graph	graph,
		vx_tensor	input,
		vx_array	order,
		vx_tensor	output
	)

[Graph] Creates a Permute Layer Node.

Permute is done if the output tensor dimensions change. Otherwise it will do a copy.
This function supports 4D tensors as input and output.

Parameters

[in]	graph	The handle to the graph.
[in]	input	The input tensor data.
[in]	order	The required output tensor dimensions.

Note

Order takes values:

'0' : 0,1,2,3 (eg:nchw->nchw)
'1' : 0,2,3,1 (eg:nchw->nhwc)

Parameters

[out] output The output tensor data. Output will have the same number of dimensions as input. Output tensor data type must be same as the inputs. The width, height, batch and channel dimensions of output can be rearranged, but value must be the same as input.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxPriorBoxLayer()

VX_API_ENTRY vx_node VX_API_CALL vxPriorBoxLayer	(	vx_graph	graph,
		vx_tensor	input_1,
		vx_tensor	input_2,
		vx_float32	minSize,
		vx_array	aspect_ratio,
		vx_int32	flip,
		vx_int32	clip,
		vx_float32	offset,
		vx_tensor	output,
		vx_array	variance,
		vx_float32	maxSize
	)

[Graph] Creates a Prior Box Layer Node.

Prior box gives the coordinates of the bounding boxes for an image. This function supports 4D tensors as input and 3D tensor as output.

Parameters

[in]	graph	The handle to the graph.
[in]	input_1	The input tensor data (output of the previous layer)
[in]	input_2	The input tensor data (image data)
[in]	minSize	The size for the first prior
[in]	aspect_ratio	Array of floats for different bounding boxes, with varying aspect ratio
[in]	flip	Input indicating whether aspect ratio can be flipped. Can take values 1(true)/0(false)
[in]	clip	Input indicating whether bounding box coordinates should be within [0,1]. Can take values 1(true)/0(false)
[in]	offset	Float value to give an offset for each bounding box
[out]	outputs	The output tensor data. Output tensor data type must be float. The batch dimensions of output and input must be same. The channel dimensions of the output will be 2. The 3rd dimension will depend on number of boxes.
[in]	maxSize	The size for the first prior (optional)
[in]	variance	The variance of each prior (optional)

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxReduceMinLayer()

VX_API_ENTRY vx_node VX_API_CALL vxReduceMinLayer	(	vx_graph	graph,
		vx_tensor	data,
		vx_array	axes,
		vx_int32	keepdims,
		vx_tensor	reduced
	)

[Graph] Creates a Reduce Min Layer Node.

Computes the min of the input tensor's element along the provided axes.

Parameters

[in]	graph	The handle to the graph.
[in]	data	The input tensor data.
[in]	axes	A list of integers, along which to reduce. The default is to reduce over all the dimensions of the input tensor. Accepted range is [-r, r-1] where r = rank(data).
[in]	keepdims	Keep the reduced dimension or not, default 1 mean keep reduced dimension.
[out]	reduced	The output tensor data with the dimensions based axes and keepdims.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxReshapeLayer()

VX_API_ENTRY vx_node VX_API_CALL vxReshapeLayer	(	vx_graph	graph,
		vx_tensor	input,
		vx_tensor	output
	)

[Graph] Creates a Convolutional Network Reshape Layer Node.

Reshaping is done with alias if available (output tensor will point to input tensor memory). Otherwise it will do a copy.
This function supports 4D tensors as input and output.

Parameters

[in]	graph	The handle to the graph.
[in]	inputs	The input tensor data.
[out]	outputs	The output tensor data. Output will have the same number of dimensions as input. Output tensor data type must be same as the inputs. The width and height dimensions of output must be integer multiple of input. The batch and channel dimensions of output and input must be same.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxScaleLayer()

VX_API_ENTRY vx_node VX_API_CALL vxScaleLayer	(	vx_graph	graph,
		vx_tensor	inputs,
		vx_tensor	scale,
		vx_tensor	bias,
		vx_tensor	output
	)

[Graph] Creates a Scale Layer Node.

Parameters

[in]	graph	The handle to the graph.
[in]	inputs	The input tensor data.
[in]	inputs	The scale tensor data.
[in]	inputs	The bias tensor data.
[out]	outputs	The output tensor data.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxSliceLayer()

VX_API_ENTRY vx_node VX_API_CALL vxSliceLayer	(	vx_graph	graph,
		vx_tensor	input,
		vx_tensor	output1,
		vx_tensor	output2,
		vx_tensor	output3,
		vx_tensor	output4,
		vx_tensor	output5,
		vx_tensor	output6,
		vx_tensor	output7,
		vx_tensor	output8
	)

[Graph] Creates a Slice Layer Node.

Parameters

[in]	graph	The handle to the graph.
[in]	inputs	The input tensor data.
[out]	inputs	The output 1 tensor data.
[out]	inputs	The output 2 tensor data.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxTensorCompareNode()

VX_API_ENTRY vx_node VX_API_CALL vxTensorCompareNode	(	vx_graph	graph,
		vx_tensor	input,
		vx_tensor	input2,
		vx_tensor	output
	)

[Graph] Creates a Tensor Compare Node.

Returns the tensor resulted from performing the comparison operation elementwise on the input tensors A and B This function supports 4D tensors as input and ouput.

Parameters

[in]	graph	The handle to the graph.
[in]	input	The first input tensor data.
[in]	input2	The second input tensor data.
[out]	output	The output tensor data with the same dimensions as the input tensor data. The type of the output is constraint to boolean.
[in]	mode	The mode value for the comparison.

Note

Supports the following mode values:

0 - Less than (<)
1 - Greater than (>)
2 - Less than or equal to (<=)
3 - Greater than or equal to (>=)
4 - Equal to (==)
5 - Not equal to (!=)

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxTensorExpNode()

VX_API_ENTRY vx_node VX_API_CALL vxTensorExpNode	(	vx_graph	graph,
		vx_tensor	input,
		vx_tensor	output
	)

[Graph] Creates a Tensor_Exp Layer Node.

Calculates the element-wise exponential of the element values in the input vx_tensor. This function supports 4D tensors as input and ouput. The type of the tensor can be either float32 or float16.

Parameters

[in]	graph	The handle to the graph.
[in]	input	The input tensor data.
[out]	output	The output tensor data with the same dimensions as the input tensor data.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxTensorLogNode()

VX_API_ENTRY vx_node VX_API_CALL vxTensorLogNode	(	vx_graph	graph,
		vx_tensor	input,
		vx_tensor	output
	)

[Graph] Creates a Tensor_Log Layer Node.

Calculates the element-wise natural log of the element values in the input vx_tensor. This function supports 4D tensors as input and ouput. The type of the tensor can be either float32 or float16.

Parameters

[in]	graph	The handle to the graph.
[in]	input	The input tensor data.
[out]	output	The output tensor data with the same dimensions as the input tensor data.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxTensorMaxNode()

VX_API_ENTRY vx_node VX_API_CALL vxTensorMaxNode	(	vx_graph	graph,
		vx_tensor	input,
		vx_tensor	input2,
		vx_tensor	output
	)

[Graph] Creates a Tensor_Max Layer Node.

Performs element-wise max on element values in the input vx_tensor. This function supports 4D tensors as input and ouput. The type of the tensor can be either float32 or float16.

Parameters

[in]	graph	The handle to the graph.
[in]	input	The first input tensor data.
[in]	input2	The second input tensor data. The dimensions and sizes of input2 match those of input1, unless the vx_tensor of one or more dimensions in input2 is 1. In this case, those dimensions are treated as if this tensor was expanded to match the size of the corresponding dimension of input1, and data was duplicated on all terms in that dimension. After this expansion, the dimensions will be equal. The data type must match the data type of input1.
[out]	output	The output tensor data with the same dimensions as the input tensor data.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxTensorMinNode()

VX_API_ENTRY vx_node VX_API_CALL vxTensorMinNode	(	vx_graph	graph,
		vx_tensor	input,
		vx_tensor	input2,
		vx_tensor	output
	)

[Graph] Creates a Tensor_Min Layer Node.

Performs element-wise min on element values in the input vx_tensor. This function supports 4D tensors as input and ouput. The type of the tensor can be either float32 or float16.

Parameters

[in]	graph	The handle to the graph.
[in]	input	The first input tensor data.
[in]	input2	The second input tensor data. The dimensions and sizes of input2 match those of input1, unless the vx_tensor of one or more dimensions in input2 is 1. In this case, those dimensions are treated as if this tensor was expanded to match the size of the corresponding dimension of input1, and data was duplicated on all terms in that dimension. After this expansion, the dimensions will be equal. The data type must match the data type of input1.
[out]	output	The output tensor data with the same dimensions as the input tensor data.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxTileLayer()

VX_API_ENTRY vx_node VX_API_CALL vxTileLayer	(	vx_graph	graph,
		vx_tensor	input,
		vx_tensor	repeats,
		vx_tensor	output
	)

[Graph] Creates a Tile Layer Node.

Constructs a tensor by tiling a given tensor.

Parameters

[in]	graph	The handle to the graph.
[in]	input	The input tensor data.
[in]	repeats	1D int64 tensor of the same length as input's dimension number, includes numbers of repeated copies along input's dimensions.
[out]	output	Output tensor of the same dimension and type as tensor input. output_dim[i] = input_dim[i] * repeats[i]

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxTopKLayer()

VX_API_ENTRY vx_node VX_API_CALL vxTopKLayer	(	vx_graph	graph,
		vx_tensor	x_tensor,
		vx_tensor	k_tensor,
		vx_int32	axis,
		vx_int32	largest,
		vx_int32	sorted,
		vx_tensor	values,
		vx_tensor	indices
	)

[Graph] Creates a TopK Layer Node.

Retrieve the top-K largest or smallest elements along a specified axis This function supports 4D tensors as input and ouput.

Parameters

[in]	graph	The handle to the graph.
[in]	x_tensor	The input tensor data.
[in]	k_tensor	The 1-D input tensor data containing a single positive value corresponding to the number of top elements to retrieve.
[out]	values	The output tensor data containing top K values from the input 'x_tensor'.
[out]	indices	The output tensor data containing the corresponding input tensor indices for the top K values.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

◆ vxUpsampleNearestLayer()

VX_API_ENTRY vx_node VX_API_CALL vxUpsampleNearestLayer	(	vx_graph	graph,
		vx_tensor	input,
		vx_tensor	output
	)

[Graph] Creates a Convolutional Network Upsampling Layer Node.

Upsampling is done on the width and height dimensions of the vx_tensor. Therefore, we use here the term x for the width dimension and y for the height dimension.
The Upsampling accept input images as tensors of several types. They always output resized images as float32 tensors. This function supports 4D and 3D tensors as input and output. 4D tensors are for batches of images, 3D tensors for individual images. Upsampling use resize method NEAREST_NEIGHBOR.

Parameters

[in]	graph	The handle to the graph.
[in]	inputs	The input tensor data.
[out]	outputs	The output tensor data. Output will have the same number of dimensions as input. Output tensor data type must be same as the inputs. The width and height dimensions of output must be integer multiple of input. The batch and channel dimensions of output and input must be same.

Returns: vx_node.; A node reference vx_node. Any possible errors preventing a successful creation should be checked using vxGetStatus.

AMD Neural Network Extension API

AMD Neural Network Extension API#

Data Structures

Macros

Functions

Detailed Description

Function Documentation

◆ vxArgmaxLayer()

◆ vxBatchNormalizationLayer()

◆ vxCastLayer()

◆ vxConcatLayer()

◆ vxConvertImageToTensorNode()

◆ vxConvertTensorToImageNode()

◆ vxCropAndResizeLayer()

◆ vxCropLayer()

◆ vxDetectionOutputLayer()

◆ vxGatherLayer()

◆ vxNMSLayer()

◆ vxPermuteLayer()

◆ vxPriorBoxLayer()

◆ vxReduceMinLayer()

◆ vxReshapeLayer()

◆ vxScaleLayer()

◆ vxSliceLayer()

◆ vxTensorCompareNode()

◆ vxTensorExpNode()

◆ vxTensorLogNode()

◆ vxTensorMaxNode()

◆ vxTensorMinNode()

◆ vxTileLayer()

◆ vxTopKLayer()

◆ vxUpsampleNearestLayer()