Deep Learning HDL Processing System

Simulate deep learning processor IP core interface

Since R2023b

Libraries:

Description

The Deep Learning HDL Processing System block simulates the deep learning processor IP core and models, simulates, and validates the hand-shaking logic between the pre- and post-processing systems and the deep learning processor IP core. The block icon changes depending on the RunTimeControl property setting of the dlhdl.ProcessorConfig object. To access the block at the MATLAB^® command line, enter:

open_system('dlhdllib')

RunTimeControl Setting	Block Icon
Register
Port

Examples

Deploy and Verify YOLO v2 Vehicle Detector on FPGA

Deploy a you only look once (YOLO) v2 vehicle detector on FPGA and verify the end-to-end application using MATLAB.

(Vision HDL Toolbox)

Limitations

• Batch processing mode for the deep learning processor IP core is not supported for simulation.

• The data type of the dlhdl.ProcessorConfig data type must be set to single. The int8 data type is not supported.

• The top-level properties of the dlhdl.ProcessorConfig object must all be set to "Port" or "Register". A mix of port and register parameter settings is not supported.

• Multiple input networks are not supported.

Ports

Input

RegisterWriteCtrlIn — Input write data control bus

bus

Input write data control bus, specified as a bus. The bus control signals are of type BusAXIWriteCtrlM2S and includes these signals:

Signal Name	Data Type
wr_addr	uint32 scalar
wr_len	uint32 scalar
wr_valid	Boolean scalar

Data Types: BusAXIWriteCtrlM2S

RegisterWriteData — Input data

scalar | vector

Input data, specified as a scalar or vector.

Data Types: uint32

RegisterReadCtrlIn — Input read data control bus

bus

Input data read control bus, specified as a bus. The bus control signals are of type BusAXIReadCtrlM2S and includes these signals:

Signal Name	Data Type
rd_addr	uint32 scalar
rd_len	uint32 scalar
rd_avalid	boolean scalar
rd_dready	boolean scalar

Data Types: BusAXIReadCtrlM2S

MemoryWriteCtrlIn — Input DDR memory write data control bus

bus

Input DDR memory write data control bus, specified as a bus. The bus control signals are of type BusAXIWriteCtrlM2S and includes these signals:

Signal Name	Data Type
wr_addr	uint32 scalar
wr_len	uint32 scalar
wr_valid	Boolean scalar

Data Types: BusAXIWriteCtrlm2S

MemoryWriteData — Input DDR memory data

scalar | vector

Input DDR memory data specified as a scalar or vector.

Data Types: ufix128

MemoryReadCtrlIn — Input DDR memory read data control bus

bus

Input DDR memory read data control bus, specified as a bus. The bus data type is BusAXIReadCtrlM2S and includes these signals:

Signal Name	Data Type
rd_addr	uint32 scalar
rd_len	uint32 scalar
rd_avalid	boolean scalar
rd_dready	boolean scalar

Data Types: BusAXIReadCtrlM2S

InputStart — Data processing start signal

scalar

Data processing start signal, specified as a boolean.

Dependencies

To enable this port, set the RunTimeControl property of the dlhdl.ProcessorConfig object to "port".

Data Types: Boolean

FrameCount — Number of input data frames

scalar

Number of input data frames, specified as a int32 scalar.

Dependencies

To enable this port, set the RunTimeControl property of the dlhdl.ProcessorConfig object to "port".

Data Types: int32

InputStop — Data streaming stop signal

scalar

Data streaming stop signal, specified as a boolean.

Dependencies

To enable this port, set the RunTimeControl property of the dlhdl.ProcessorConfig object to "port".

Data Types: Boolean

InputNext — Next input frame ready signal

scalar

Next input frame ready signal, specified as a boolean.

Dependencies

To enable this port, set the RunTimeControl property of the dlhdl.ProcessorConfig object to "port".

Data Types: Boolean

OutputNext — Current output data read complete signal

scalar

Current output data read complete signal, specified as a boolean.

Dependencies

To enable this port, set the RunTimeControl property of the dlhdl.ProcessorConfig object to "port".

Data Types: Boolean

Output

RegisterWriteCtrlOut — Output data write control bus

bus

Output data write control bus, returned as a bus. The bus control signals are of type BusAXIWriteCtrlS2M and includes these signals:

Signal Name	Data Type
wr_ready	boolean scalar
wr_complete	boolean scalar

Data Types: BusAXIWriteCtrlS2M

RegisterReadData — Output data

scalar | vector

Output data returned as a scalar or vector.

Data Types: uint32

RegisterReadCtrlOut — Output data read control bus

bus

Output data read control bus, returned as a bus. The bus control signals are of type BusAXIReadCtrlS2M and includes these signals:

Signal Name	Data Type
rd_aready	boolean scalar
rd_dvalid	boolean scalar

Data Types: BusAXIReadCtrlS2M

MemoryWriteCtrlOut — Output DDR memory data write control bus

bus

Output DDR memory data write control bus, returned as a bus. The bus control signals are of type BusAXIWriteCtrlS2M and includes these signals:

Signal Name	Data Type
wr_ready	boolean scalar
wr_complete	boolean scalar

Data Types: BusAXIWriteCtrlS2M

MemoryReadData — Output DDR memory data

scalar | vector

Output DDR memory data returned as a scalar or vector.

Data Types: ufix128

MemoryReadCtrlOut — Output DDR memory data read control bus

bus

Output DDR memory data read control bus, returned as a bus. The bus control signals are of type BusAXIReadCtrlS2M and includes these signals:

Signal Name	Data Type
rd_aready	boolean scalar
rd_dvalid	boolean scalar

Data Types: BusAXIReadCtrlS2M

InputAddr — Input data memory address signal

scalar

Input data memory address signal returned, as a uint32 scalar.

Dependencies

To enable this port, set the RunTimeControl property of the dlhdl.ProcessorConfig object to "port".

Data Types: uint32

InputSize — Size of next input data frame

scalar

Size of next input data frame, returned as a uint32 scalar. The input data frame size is measured in bytes.

Dependencies

To enable this port, set the RunTimeControl property of the dlhdl.ProcessorConfig object to "port".

Data Types: uint32

InputValid — Input data valid signal

scalar

Input data valid signal, returned as a boolean.

Dependencies

To enable this port, set the RunTimeControl property of the dlhdl.ProcessorConfig object to "port".

Data Types: Boolean

OutputAddr — Processed output data memory address signal

scalar

Processed output data memory address signal, returned as a uint32 scalar.

Dependencies

To enable this port, set the RunTimeControl property of the dlhdl.ProcessorConfig object to "port".

Data Types: uint32

OutputSize — Size of next output data frame signal

scalar

Size of next output data frame signal, returned as a uint32 scalar. Use this signal when the OutputValid signal is on.

Dependencies

To enable this port, set the RunTimeControl property of the dlhdl.ProcessorConfig object to "port".

Data Types: uint32

OutputValid — Output data valid signal

scalar

Output data valid signal, returned as a boolean scalar.

Dependencies

To enable this port, set the RunTimeControl property of the dlhdl.ProcessorConfig object to "port".

Data Types: Boolean

Done — Data processing complete signal

scalar

Data processing complete signal, returned as a boolean scalar.

Dependencies

To enable this port, set the RunTimeControl property of the dlhdl.ProcessorConfig object to "port".

Data Types: Boolean

Parameters

Network — Name of network object

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

Name of network object, specified as one of these options:

• 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.

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.

File Path — Path to MAT-file

untitled.mat (default)

Path to MAT-file that contains the network object.

Dependencies

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

Activation Layer — Name of network activation layer

pool10 (default)

Name of network activation layer to simulate. The default activation layer is the last layer of the network object specified by the Network parameter.

Maximum input frame number limit for memory access allocation — Maximum input frame number limit

30 (default) | positive scalar

Maximum input frame number limit to calculate DDR memory access allocation.

Processor Configuration Properties

Import Processor Configuration from MATLAB workspace — Enable import of existing dlhdl.ProcessorConfig object

dlhdl.ProcessorConfig (default) | dlhdl.ProcessorConfig('Bitstream','zcu102_single') | dlhdl.ProcessorConfig('Bitstream','arria10soc_single') | dlhdl.ProcessorConfig('Bitstream','zc706_single')

Import an existing dlhdl.ProcessorConfig object. if you specify a function name, it must be a MATLAB function. For example, this MATLAB function returns a dlhdl.ProcessorConfig object called hPCinMATLAB.

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

After you add the MATLAB function name or dlhdl.ProcessorConfig object, click Apply to import the processor configuration or MATLAB configuration into the block.

Export Processor Configuration to MATLAB workspace — Enable export of dlhdl.ProcessorConfig object

hPCInSimulink (default)

Export a dlhdl.ProcessorConfig object. The default object name is hPCInSimulink.

Convolution Module

Module Generation — Enable or disable convolution module generation

on (default) | off

Enable or disable convolution module generation. Use this parameter to control generation of the convolution module as a part of the deep learning processor configuration.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

LRN Block Generation — Enable or disable local response normalization (LRN) block generation

off (default) | on

Enable or disable LRN block generation. Use this parameter to control generation of the LRN block as a part of the convolution module of the deep learning processor configuration.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Segmentation Block Generation — Enable or disable segmentation block generation

on (default) | off

Enable or disable segmentation block generation. Use this parameter to control generation of the segmentation block as a part of the convolution module of the deep learning processor configuration.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Convolution Thread Number — Number of parallel convolution processor kernel threads

16 (default) | 4 | 9 | 16 | 25 | 36 | 49 | 64

Number of parallel convolution processor kernel threads.

This parameter is the number of parallel 3-by-3 convolution kernel threads that are a part of the conv module in the dlhdl.ProcessorConfig object.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Input Memory Size — Input memory cache block RAM (BRAM) size

Read-only: [227 227 3] (default) | 3-D positive integer array

This parameter is read-only.

Input memory cache BRAM size. This parameter is a 3-D matrix that represents the maximum input image size permitted by the BRAM of the conv module in the dlhdl.ProcessorConfig object.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Output Memory Size — Output memory cache block RAM (BRAM) sizes

[227 227 3] (default) | 3-D positive integer array

Output memory cache BRAM size.

This parameter is a 3-D matrix that represents the maximum output image size permitted by the BRAM of the conv module in the dlhdl.ProcessorConfig object.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Feature Size Limit — Maximum input and output feature size

2048 (default) | positive integer

Maximum input and output feature size.

This parameter is a positive integer that represents the maximum input and output feature size as a part of the conv module in the dlhdl.ProcessorConfig object.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

FC Module

Module Generation — Enable or disable fully connected module generation

on (default) | off

Enable or disable fully connected module generation. Use this parameter to control generation of the fully connected module as a part of the deep learning processor configuration.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Softmax Block Generation — Enable or disable Softmax block generation

off (default) | on

Enable or disable Softmax block generation. Use this parameter to control generation of the Softmax block as a part of the fully connected module of the deep learning processor configuration. When you clear this parameter, the Softmax layer is still implemented in software.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

FC Thread Number — Number of parallel fully connected (fc) MAC threads

4 (default) | 4 | 8 | 16

Number of parallel fully connected (FC) MAC threads. This parameter is the number of parallel FC MAC threads that are a part of the fc module in the dlhdl.ProcessorConfig object.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Input Memory Size — Input memory cache block RAM (BRAM) sizes

25088 (default) | unsigned integer

Input memory cache BRAM size. This parameter is an unsigned integer that represents the input cache memory size permitted by the BRAM of the fc module in the dlhdl.ProcessorConfig object.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Output Memory Size — Cache block RAM (BRAM) sizes

4096 (default) | unsigned integer

Output memory cache BRAM size. This parameter is an unsigned integer that represents the output cache memory size permitted by the BRAM of the fc module in the dlhdl.ProcessorConfig object.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Custom Module

Module Generation — Enable or disable custom module generation

on (default) | off

Enable or disable custom module generation. Use this parameter to control generation of the adder module as a part of the deep learning processor configuration.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Addition — Enable or disable addition layer generation

on (default) | off

Enable or disable addition layer generation. Use this parameter to control generation of the addition layer as a part of the custom module of the deep learning processor configuration.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Multiplication — Enable or disable multiplication layer generation

on (default) | off

Enable or disable multiplication layer generation. Use this parameter to control generation of the multiplication layer as a part of the custom module of the deep learning processor configuration.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Resize2D — Enable or disable 2-D resize layer generation

off (default) | on

Enable or disable 2-D resize layer generation. Use this parameter to control generation of the 2-D resize layer as a part of the custom module of the deep learning processor configuration.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Sigmoid — Enable or disable sigmoid layer generation

off (default) | on

Enable or disable sigmoid layer generation. Use this parameter to control generation of the sigmoid layer as a part of the custom module of the deep learning processor configuration.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

TanhLayer — Enable or disable tanh layer generation

off (default) | on

Enable or disable tanh layer generation. Use this parameter to control generation of the tanh layer as a part of the custom module of the deep learning processor configuration.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

MishLayer — Enable or disable mish layer generation

off (default) | on

Enable or disable mish layer generation. Use this parameter to control generation of the mish layer as a part of the custom module of the deep learning processor configuration.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

SwishLayer — Enable or disable swish layer generation

off (default) | on

Enable or disable swish layer generation. Use this parameter to control generation of the swish layer as a part of the custom module of the deep learning processor configuration.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Input Memory Size — Input memory cache block RAM (BRAM) sizes

40 (default) | unsigned integer

Input memory cache BRAM size. This parameter is an unsigned integer that represents the input cache memory size permitted by the BRAM of the custom module in the dlhdl.ProcessorConfig object.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Output Memory Size — Output memory cache block RAM (BRAM) sizes

120 (default) | unsigned integer

Output memory cache BRAM size. This parameter is an unsigned integer that represents the input cache memory size permitted by the BRAM of the custom module in the dlhdl.ProcessorConfig object.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Processor Top Level

Choose Processinf Mode of All Interface Signals — Select the interface type for all deep learning processor IP core interface signals

Read-only: register (default) | port

This parameter is read-only.

Select the interface type for all deep learning processor IP core interface signals. Specify whether the run-time input signals, run-time feedback signals, input interface control input signals, and input interface control feedback signals are implemented as registers or ports.

Dependencies

To enable and edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

RunTimeControl — Deep learning processor IP core mode setting

Read-only: register (default) | port

This parameter is read-only.

Deep learning processor IP core mode setting. Specify whether the run-time input signals to the deep learning processor IP core are implemented as registers or ports.

RunTimeStatus — Feedback signals from deep learning processor IP core control

Read-only: register (default) | port

This parameter is read-only.

Feedback signals from deep learning processor IP core control. Specify whether the run-time output signals from the deep learning processor IP core are implemented as registers or ports.

InputStreamControl — Deep learning processor IP core input interface control

Read-only: register (default) | port

This parameter is read-only.

Deep learning processor IP core input interface control.

OutputStreamControl — Deep learning processor IP core output interface control

Read-only: register (default) | port

This parameter is read-only.

Deep learning processor IP core output interface control.

SetupControl — Deep learning processor IP core user interface control

Read-only: register (default) | port

This parameter is read-only.

Deep learning processor IP core user interface control.

ProcessorDataType — Deep learning processor IP core module data type

single (default)

Deep learning processor IP core module data type.

Dependencies

To edit this parameter, enable Export Processor Configuration to MATLAB Workspace.

Export Deep Learning Processor IP Core Register addresses to MATLAB workspace — Export register addresses

Read-only: on (default) | off

This parameter is read-only.

Enable this option to export a structure containing address locations to the MATLAB workspace.

Version History

Introduced in R2023b