Deep Learning HDL Output Format

R2026b

Convert deep learning processor outputs in external memory into per-element data and indices

Since R2026b

• 

Deep Learning HDL Output Format block

Description

The Deep Learning HDL Output Format block converts a stream of deep learning processor outputs in external memory into per element data and indices. Use the converted output data to implement post-processing algorithms in Simulink^® or to generate HDL code.

To access the block open the dlhdlutils library. At the MATLAB^® command line, enter:

open_system("dlhdlutils")

Ports

Input

DataIn — Streaming input data from external memory

scalar

Streaming input data from external memory, specified as a single or int8 data type scalar.

Data Types: single | int8

ValidIn — Input valid data flag

scalar

Input valid data flag, specified as a Boolean scalar.

Data Types: Boolean

DataStart — Trigger to start streaming input data

scalar

Trigger to start streaming input data, specified as a Boolean scalar.

Data Types: Boolean

Output

DataOut — Formatted output data

scalar

Formatted output data, returned as a single or int8 data type scalar.

Data Types: single | int8

ValidOut — Trigger to read processed output data

scalar

Trigger to read processed output data, returned as a Boolean scalar.

Data Types: Boolean

RowIdx — Row index of formatted output data

scalar

Row index of the formatted output data, returned as an uint16 data type scalar. To see which element the row index corresponds to, see Read Convolution Memory Format Data or Read Fully Connected Memory Format Data.

Data Types: uint16

ColIdx — Column index of formatted output data

scalar

Column index of the formatted output data, returned as an uint16 data type scalar. To see which element the column index corresponds to, see Read Convolution Memory Format Data or Read Fully Connected Memory Format Data.

Data Types: uint16

DepthIdx — Depth index of formatted output data

scalar

Depth index of the formatted output data, returned as an uint16 data type scalar. To see which element the depth index corresponds to, see Read Convolution Memory Format Data or Read Fully Connected Memory Format Data.

Data Types: uint16

OutputIdx — Output selection index for multiple-output networks

scalar

Output selection index for multiple-output networks, returned as an uint16 data type scalar.

Data Types: uint16

FrameCount — Number of output data frames

scalar

Number of output data frames, returned as an uint16 data type scalar. To see which element the frame count corresponds to, see Read Convolution Memory Format Data or Read Fully Connected Memory Format Data.

Dependencies

To enable this port, select Enable FrameCount.

Data Types: uint16

Parameters

To edit block parameters interactively, use the Property Inspector. From the Simulink Toolstrip, on the Simulation tab, in the Prepare gallery, select Property Inspector.

Network — Source of network object

Network from MATLAB function (default) | Network from MAT-file

Location of the network object to use in the block.

• Network from MATLAB function — Provide the name of the MATLAB function that returns a network object.

• Network from MAT-file — Provide the path to the MAT file that stores the network object.

Programmatic Use

To set the block parameter value programmatically, use the set_param (Simulink) function.

To get the block parameter value programmatically, use the get_param (Simulink) function.

Parameter:	networkOption
Values:	Network from MATLAB function (default) | Network from MAT-FILE

Example: get_param(gcb,"networkOption")

MATLAB Function — Name of network MATLAB function

getDigitsNetwork (default)

Name of the MATLAB function that returns a network object.

Dependencies

To enable this parameter, set Network to Network from MATLAB function.

Programmatic Use

To set the block parameter value programmatically, use the set_param (Simulink) function.

To get the block parameter value programmatically, use the get_param (Simulink) function.

Parameter:	netFunction
Values:	getDigitsNetwork (default)

Example: get_param(gcb,"netFunction")

File Path — Path to MAT-file

tinyyolov4.mat (default)

Path to the MAT-file that contains the network object.

Dependencies

To enable this parameter, set Network to Network from MAT-file.

Programmatic Use

To set the block parameter value programmatically, use the set_param (Simulink) function.

To get the block parameter value programmatically, use the get_param (Simulink) function.

Parameter:	netFile
Values:	tinyyolov4.mat (default)

Example: get_param(gcb,"netFile")

Deep Learning Processor Configuration — Name of configuration object or MATLAB function

dlhdl.ProcessorConfig (default)

Name of a dlhdl.ProcessorConfig object or a MATLAB function that returns a dlhdl.ProcessorConfig object. If you set Network to MATLAB function, you must enter the name of a MATLAB function that returns a dlhdl.ProcessorConfig object. For example, this MATLAB function returns a dlhdl.ProcessorConfig object called hPCinMATLAB.

function hPCinMATLAB = hPCinMATLAB()
hPC = dlhdl.ProcessorConfig;
hPC.RunTimeControl = "port";
hPCinMATLAB = hPC;
end

Programmatic Use

To set the block parameter value programmatically, use the set_param (Simulink) function.

To get the block parameter value programmatically, use the get_param (Simulink) function.

Parameter:	processorConfig
Values:	on (default) | off

Example: get_param(gcb,"processorConfig")

Enable FrameCount — Option to output frame count data

on (default) | off

Set this parameter to enable the FrameCount output port.

Programmatic Use

To set the block parameter value programmatically, use the set_param (Simulink) function.

To get the block parameter value programmatically, use the get_param (Simulink) function.

Parameter:	enableFrameCount
Values:	on (default) | off

Example: get_param(gcb,"enableFrameCount")

Algorithms

Read Convolution Memory Format Data

For layers that use the deep learning processor convolution module data format SSCB (spatial, spatial, channel, batch), the block outputs row, column, depth, and batch indices. This table shows how those indices correspond to the spatial, channel, and batch elements of the SSCB data format.

Port Name	Corresponding Element
RowIdx	S (first spatial field)
ColIdx	S (second spatial field)
DepthIdx	C (channel field)
FrameCount	B (batch field)

This timing diagram shows the block operation when reading convolution module format data from external memory. The DataIn port is the data read from external memory. When the input data is available, you must set the signal to the ValidIn port to true and pulse the signal input to the DataStart port to true to read the first valid data sample from memory. The block then indicates that the data is ready to read by outputting a true signal from the ValidOut port, sets the row, column, and depth indices and outputs a value to the DataOut port. The input data to the block is stored in memory as a one-dimensional vector. To learn how the data is converted, see Convolution Module External Memory Data Format.

In this timing diagram, when you set the input signal to the ValidIn port to true and you pulse the input signal to the DataStart port, the block sets the ValidOut port signal to true to indicate that the data is ready to read, sets the row index to 1, column index to 1, depth index to 1, and outputs the first data in the DataOut port. The block then keeps the row and column indices at 1, sets the depth index to 2, and outputs the next data value into the DataOut port. After the block completes reading all the channel elements for the first row and column, it continues to the next column and then continues to the next row after completing the first row.

Read Fully Connected Memory Format Data

For layers that use the deep learning processor fully connected module data format TCB (time, channel, batch), or CB (channel, batch) the block outputs row, column, depth, and batch indices. This table shows how those indices correspond to the time, channel, batch of the TCB data format, or channel, batch of the CB data format.

Port Name	Corresponding Element
RowIdx	1
ColIdx	1
DepthIdx	C (channel field)
FrameCount	T (time field in TCB) or B (batch field in CB)

When the input data is available, you must set the signal to the ValidIn port to true and pulse the signal input to the DataStart port to true to read the first valid data sample from memory. The block then indicates that the data is ready to read by outputting a true signal from the ValidOut port, sets the row, column, and depth indices and outputs a value to the DataOut port. The input data to the block is stored in memory as a one-dimensional vector. For data in the fully connected memory data format, the row and column indices are always 1. To learn how the data is converted, see Fully Connected Module External Memory Data Format.

In this timing diagram, when you set the input signal to the ValidIn port to true and you pulse the input signal to the DataStart port, the block sets the ValidOut port signal to true to indicate that the data is ready to read, sets the row index to 1, column index to 1, depth index to 1, and outputs the first data in the DataOut port. The block then keeps the row and column indices at 1, sets the depth index to 2, and outputs the next data value into the DataOut port. After the block completes reading all the channel elements, it resets all the indices to 0.

Version History

Introduced in R2026b