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A newer version of this document is available.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

5.2.9. Domain Boundary Entry and Domain Boundary Exit

The Path Info column lists the Domain Boundary Entry or Domain Boundary Exit for a critical chain. Domain boundary entry and domain boundary exit refer to paths that are unconstrained, paths between asynchronous clock domains, or between a clock domain and top-level device input-outputs. Domain boundary entry and exit can also be indicated for some false paths as well.

A domain boundary entry refers to a point in the design topology, at a clock domain boundary, where Hyper-Retiming can insert register stages (where latency can enter the clock domain) if Hyper-Pipelining is enabled. The concept of a domain boundary entry is independent of the dataflow direction. Hyper-Retiming can insert register stages at the input of a module, and perform forward retiming pushes. Hyper-Retiming can also insert register stages at the output of a module, and perform backward retiming pushes. These insertions occur at domain boundary entry points.

A domain boundary exit refers to a point in the design topology, at a clock domain boundary, where Hyper-Retiming can remove register stages and the latency can exit the clock domain, if Hyper-Pipelining is enabled. Removing a register seems counter intuitive. However, this method is often necessary to retain functional correctness, depending on other optimizations that Hyper-Retiming performs.

Sometimes a critical chain indicates a domain boundary entry or exit when there is an unregistered I/O feeding combinational logic on a register-to-register path as shown in the following figure.

Figure 121. Domain Boundary with Unregistered Input/Output

The register-to-register path might be shown as a critical chain segment with a domain boundary entry or a domain boundary exit, depending on how the path restricts Hyper-Retiming. The unregistered input prevents the Hyper-Retiming from inserting register stages at the domain boundary, because the input is unregistered. Likewise, the unregistered input can also prevent Hyper-Retiming from removing register stages at the domain boundary.

Critical chains with a domain boundary exit do not provide complete information for you to determine what prevents retiming a register out of the clock domain. To determine why a register cannot retime, review the design to identify the signals that connect to the other side of a register associated with a domain boundary exit.

Domain boundary entry and domain boundary exit can appear independently in critical chains. They can also appear in combination such as, a domain boundary exit without a domain boundary entry, or a domain boundary entry at the beginning and end of a critical chain.

The following critical chain begins and ends with domain boundary entry. The domain boundary entries are the input and output registers connecting to top-level device I/Os. The input register is round_robin_requests_r and the output register is round_robin_next.

Figure 122. Critical Chain Schematic with Domain Boundary

The limiting reason for the base compile is Insufficient Registers.

Figure 123. Retiming Limit Summary with Insufficient Registers

The following parts of the critical chain report show that the endpoints are labeled with Domain Boundary Entry.

Figure 124. Critical Chain with Domain Boundary Entry

Both the input and output registers are indicated as Domain Boundary Entry because the Fast Forward Compile often inserts register stages at these boundaries if Hyper-Pipelining were enabled.