By Paul Biggar and David Gregg (Trinity College Dublin, Ireland)
Summary of review comments:
The authors sketch an algorithm for constructing an SSA intermediate representation while performing analysis that typically preceded SSA construction. The goal is to combine the benefits of SSA (sparse representation) with costly analysis so that the analysis is positively impacted (faster, more precise).
SSA has some nice properties that serve some purposes well, but it's not obviously clear that it's always practical. Is an SSA-based representation *the* representation to "rule them all". For example, is register spilling any simpler for code in SSA form compared to standard code? Similarly, flow-insensitive pointer analyses are already sparse and HSSA based on them may be precise enough. How much are we losing by ordering the analyses? Further, it's not clear why the authors want to build on SCCP when incremental SSA construction algorithms exist.
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment