Main Content

validateReferenceDesignForDeepLearning

Class: hdlcoder.ReferenceDesign
Namespace: hdlcoder

Checks property values in reference design object

Since R2021b

Syntax

validateReferenceDesignForDeepLearning

Description

validateReferenceDesignForDeepLearning checks that the hdlcoder.ReferenceDesign object is compatible with the deep learning processor IP core generation workflow.

Reference Design Requirements

The validateReferenceDesign method checks your hdlcoder.ReferenceDesign object for these requirements:

  • Your reference design must have an AXI4 Slave interface. You use this interface to access registers in the deep learning processor IP core.

  • Your reference design must have three AXI4 Master interfaces with InterfaceID values of "AXI4 Master Activation Data", "AXI4 Master Weight Data", and "AXI4 Master Debug". The deep learning processor IP core uses these AXI4 Master interfaces to access memory for network storage and intermediate calculations.

  • Your reference design must have a registered memory address space. Use the registerDeepLearningMemoryAddressSpace to register a memory address space. To run a variety of deep learning networks on your deep learning processor IP core, it is recommended to register a minimum of 512 MB of memory address space.

Tips

Register a target interface for your reference design by using the registerDeepLearningTargetInterface method. You can use the registered target interface to run your deep learning network on the generated deep learning processor IP core by using MATLAB® and a dlhdl.Workflow object.

Version History

Introduced in R2021b