Scalable cross-module optimization

Abstract

Large applications are typically partitioned into separately compiled modules. Large performance gains in these applications are available by optimizing across module boundaries. One barrier to applying crossmodule optimization (CMO) to large applications is the potentially enormous amount of time and space consumed by the optimization process.We describe a framework for scalable CMO that provides large gains in performance on applications that contain millions of lines of code. Two major techniques are described. First, careful management of in-memory data structures results in sub-linear memory occupancy when compared to the number of lines of code being optimized. Second, profile data is used to focus optimization effort on the performance-critical portions of applications. We also present practical issues that arise in deploying this framework in a production environment. These issues include debuggability and compatibility with existing development tools, such as make . Our framework is deployed in Hewlett-Packard's (HP) UNIX compiler products and speeds up shipped independent software vendors' applications by as much as 71%.

alto: a link‐time optimizer for the Compaq Alpha

“…Link-time register allocation, aimed at allowing global variables to be kept in registers and reducing register saves and restores at inter-module calls, is discussed by Santhanam and Odnert [27] and Wall [15]. Ayers et al [30] describe a production-quality link-time optimizer for Hewlett-Packard systems running HP-UX, which is distinguished by its ability to perform whole-program optimizations on very large programs, by virtue of the careful attention paid to memory management issues. Ayers et al [30] describe a production-quality link-time optimizer for Hewlett-Packard systems running HP-UX, which is distinguished by its ability to perform whole-program optimizations on very large programs, by virtue of the careful attention paid to memory management issues.…”

“…The Zuse Translation System [28] and the mld link-time optimizer [29] are aimed at reducing the cost of abstraction in object-oriented languages. These works rely on specially engineered compilers that produce either object files containing special annotations to assist the link-time optimizer [15], or an intermediate representation of the program (together with semantic information about it) that is subsequently optimized and translated to executable code by the linker [27][28][29][30]. These works rely on specially engineered compilers that produce either object files containing special annotations to assist the link-time optimizer [15], or an intermediate representation of the program (together with semantic information about it) that is subsequently optimized and translated to executable code by the linker [27][28][29][30].…”

Inlining java native calls at runtime

“…To do this efficiently, we exploit the ability to store IL alongside native executable code in the same binary object file or library. This strategy of retaining the IL generated by a traditional static compiler to support future optimizations is similar to strategies used to support linktime cross-module inlining optimizations in several commercial compilers, include those from HP [2] and IBM [20]. It is also reminiscent of the strategy for supporting "life-long" program optimization used in LLVM [21].…”

A sparse algorithm for predicated global value numbering

“…When the algorithm has finished, blocks and edges not in REACHABLE are unreachable. 2 Blocks are also kept in TOUCHED (see section 2.8). Initialize the reachable dominator tree…”

“…We have implemented the practical algorithm in the high level optimizer (HLO) [2] component of an internal version of the PA-RISC C compiler on HP-UX. HLO performs whole program, inter-procedural, parallelizing and scalar optimizations on a high level SSA form intermediate representation.…”