March 10, 1999
We are pleased to announce that Cygnus has donated a lazy code motion optimizer framework (LCM).
LCM is a general technique for performing global optimizations such as global cse and partial redundancy elimination.
LCM is an improvement over other global redundancy elimination algorithms because it can typically find more redundancies and minimize register lifetimes. In fact, in a 100% complete implementation LCM can be shown to be computationally optimal.
The LCM algorithm Cygnus has donated is based on the lazy code motion algorithms found in Steven Muchnick's book, Advanced Compiler Design and Implementation (pages 407-415). LCM optimizers are provided which work on both the forward and the reverse flow graphs.
The basic concept behind LCM is to initially move all expressions as far up in the flowgraph as possible. This exposes the maximum number of redundancies. Then expressions are then pushed as far down in the flow graph as possible, which minimizes register lifetimes.
This implementation is missing one key feature which prevents it from being computationally optimal, specifically critical edge splitting. While the optimizer works and will perform better than most global optimizers, the lack of critical edge splitting prevents the resulting code from being computationally optimal.
We will be shortly reworking the global cse optimizer to use the lazy code motion framework. We expect other optimizations using the lazy code motion framework to be added to the compiler in the future. For example, the lazy code motion framework can be used to implement spill code motion.
Copyright (C) Free Software Foundation, Inc. Verbatim copying and distribution of this entire article is permitted in any medium, provided this notice is preserved.
These pages are maintained by the GCC team. Last modified 2022-10-26.