AN 886: Intel Agilex® 7 Device Design Guidelines

ID 683634
Date 10/09/2023
Public
Document Table of Contents

5.1.8.2.3. MPU Sharing Data with FPGA

You can optimize data throughput by selecting the correct method of sharing data between the HPS and the FPGA. This section assumes that the HPS SDRAM is the data source and the FPGA require access to it. There are two main ways for the FPGA to access data that originates in HPS SDRAM:

  • FPGA accesses non-cached data directly through the FPGA-to-HPS bridge targeting the SDRAM.
  • FPGA accesses cached data through the FPGA-to-HPS bridge targeting the CCU.

If the data in the SDRAM is the most recent copy of the data (software managed coherency) then the highest throughput method of accessing the data is to have masters in the FPGA access the data directly through the FPGA-to-HPS bridge.

If the data in the SDRAM potentially is not the most recent copy of the data and software does not flush the MPU caches to ensure system wide coherency is maintained, then the FPGA master performs cacheable transactions to the FPGA-to-HPS bridge to ensure the most recent data is accessed.

GUIDELINE: Avoid using the HPS DMA controller to move data between the FPGA and HPS. Use a soft DMA controller in the FPGA fabric instead. Use the HPS DMA controller only for memory copies or peripheral data movements that remain inside the HPS.

It is not recommended to use the HPS DMA to move the data to the FPGA because the DMA bandwidth into the HPS SDRAM is limited. The HPS DMA is intended to be used to move buffers on behalf of the MPU or used for transfers between peripherals and memory. As a result, any time the FPGA needs access to buffers in HPS memory, or if the HPS requires access to data stored in the FPGA, it is always recommended to have masters in the FPGA perform these transfers instead of the HPS initiating them.