FPGA AI Suite: SoC Design Example User Guide

ID 768979
Date 7/31/2024
Public
Document Table of Contents

6.3.3.2. Streaming System Inference Job Management

In a M2M system, the host CPU handles pushing jobs into the FPGA AI Suite IP job queue by writing to the IP registers. In the streaming configuration, this task is offloaded to the Nios® system and must be done in a coordinated way with the input buffer writing.

In streaming mode, the job-queue management is pushed to the mailbox instead of being managed by the host application. The job queue entry is then received by the Nios® processor. After an input buffer is written, the mSGDMA interrupts the Nios® processor, and the Nios® processor now pushes one job into the FPGA AI Suite IP.

For every buffer stored by the mSGDMA, the Nios® processor attempts to start another job.

For more details about the Nios® V Stream Controller and the mailbox communication protocol, refer to Streaming-to-Memory (S2M) Streaming Demonstration.