Intel® Hyperflex™ Architecture High-Performance Design Handbook

ID 683353
Date 10/04/2021
Public
Document Table of Contents

8.2.1. Synchronous Resets and Limitations

Converting asynchronous resets to synchronous eases retiming restrictions, but does not remove all performance restrictions. The ALM register's dedicated LAB-wide signal often performs synchronous clears. The signal’s fan-out determines use of this signal during synthesis. The Compiler typically implements a synchronous clear with a small fan-out in logic. Larger fan-outs use this dedicated signal. Even if synthesis uses the synchronous clear, the Compiler still retimes the register into Hyper-Registers. The bypass mode of the ALM register enables this functionality. When the Compiler bypasses the register, the sclr signal and other control signals remain accessible.

In the following example, the LAB-wide synchronous clear feeds multiple ALM registers. A Hyper-Register is available along the synchronous clear path for every register.

Figure 134. Retiming Example for Synchronous ResetsCircles represent Hyper-Registers and rectangles represent ALM registers. An unfilled object represents an unoccupied location and a blue-filled object is occupied.

During retiming, the Compiler pushes top register in row (a) into a Hyper-Register. The Compiler implements this by bypassing the ALM register, but still using the SCLR logic that feeds that register. When you use the LAB-wide SCLR signal, an ALM register must exist on the data path, but you need not use the register.

Register retiming pushes the register in row (b) left into its data path. The register pushes through a signal split of the data path and synchronous clear. The Compiler must push this register onto both nets: one register in the data path, and one register in the synchronous clear path. This implementation is possible because each path has a Hyper-Register.

Retiming is more complex when another register pushes forward into the ALM. As shown in the following figure, a register from the synchronous clear port, and a register from the data path, merge together.

Figure 135. Retiming Example – Second Register Pushes out of ALM

Because other registers share the synchronous clear path, the register splits on the path to other synchronous clear ports.

Figure 136. Retiming Example – Register Splits on the Path to other Synchronous Clear Ports


In the following figure, the Hyper-Register at a synchronous clear is in use and cannot accept another register. The Compiler cannot retime this register for the second time through the ALM.

Figure 137. Retiming Example – Conflict at Synchronous Clear


Two key architectural components enable movement of ALM registers with a synchronous clear forward or backward:

  • The ability to bypass the ALM register
  • A Hyper-Register on the synchronous clear path

To push more registers through, retiming becomes difficult. Performance improvement is better with asynchronous reset removal than conversion to synchronous resets. Synchronous clears are often difficult to retime because of their wide broadcast nature.

Did you find the information on this page useful?

Characters remaining:

Feedback Message