Intel® Hyperflex™ Architecture High-Performance Design Handbook

ID 683353
Date 12/08/2023
Document Table of Contents

2.3.1. Conventional Versus Hyper-Pipelining

Hyper-Pipelining simplifies this process of conventional pipelining. Conventional pipelining includes the following design modifications:
  • Add two registers between logic clouds.
  • Modify HDL to insert a third register (or pipeline stage) into the design’s logic cloud, which is Logic Cloud 2. This register insertion effectively creates Logic Cloud 2a and Logic Cloud 2b in the HDL
Figure 28. Conventional Pipelining User Modifications

Figure 29. Hyper-Pipelining User ModificationsHyper-Pipelining simplifies the process of adding registers. Add the registers—Pipe 1, Pipe 2, and Pipe 3—in aggregate at one location in the design RTL. The Compiler retimes the registers throughout the circuit to find the optimal placement along the path. This optimization reduces path delay and maximizes the design's operating frequency.

Figure 30. Hyper-Pipelining and Hyper-Retiming ImplementationThe following figure shows implementation of additional registers after the retiming stage completes optimization.

The resulting implementation in the Hyper-Pipelining flow differs from the conventional pipelining flow by the location of the Pipe 3 register. Because the Compiler is aware of the current circuit implementation, including routing, the Compiler can more effectively locate the aggregate registers to meet the design’s maximum operating frequency. Hyper-Pipelining requires significantly less effort than conventional pipelining techniques because you can place registers at a convenient location in a data path. The Compiler optimizes the register placements automatically.