Intel® Hyperflex™ Architecture High-Performance Design Handbook

ID 683353
Date 10/04/2021
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents

5.2.3.1. Optimizing Path Limit

Evaluate the Fast Forward recommendations. If your critical chain has a limiting reason of Path Limit, and the chain is entirely in the core logic and in the routing elements of the Intel FPGA fabric, the design can run at the maximum performance of the core fabric. When the critical chain has a limiting reason of Path limit, and chain is through a DSP block or hard memory block, you can improve performance by optimizing the path limit.

To optimize path limit, enable the optional input and output registers for DSP blocks and hard memory blocks. If you do not use the optional input and output registers for DSP blocks and memory blocks, the locations for the optional registers are not available for Hyper-Retiming, and do not appear as bypassed Hyper-Registers in the critical chain. The path limit is the silicon limit of the path, without the optional input or output registers. You can improve the performance by enabling optional input and output registers.

Turn on optional registers using the IP parameter editor to parameterize hard DSP or memory blocks. If you infer DSP or memory functions from your RTL, ensure that you follow the Recommended HDL Coding Styles to ensure that you use the optional input and output registers of the hard blocks. The Compiler does not retime into or out of DSP and hard memory block registers. Instantiate the optional registers to achieve maximum performance.

If your critical chain includes true dual port memory, refer to True Dual-Port Memory for optimizing techniques.