Effectiveness of a machine-level, global optimizer

Abstract

We present an overview of the design of a machine-code-level, global (intraprocedural) optimizer that supports several front-ends producing code for the Hewlett-Packard Precision Architecture family of machines. The basic optimization strategy is described, including information about the division of responsibilities between various components of the compiler.Optimization algorithms are described, including a discussion of the dataflow information they require. Measurements showing the collective and individual effects of various optimizer components are presented.The performance data presented here was collected using a preliminary version of the optimizer. Development is continuing and further improvements are expected. The SettingThe HP Precision Architecture defines a machine with a simple register-based instruction set using 32 general-purpose registers.2 3 Only load and store instructions reference memory. The architecture acknowledges the use of pipelined hardware by defining branches to not take effect until the second instruction cycle following the branch. Referencing the target of a load instruction in the instruction which follows the load will result in a one-instruction hardware interlock.

alto: a link‐time optimizer for the Compaq Alpha

“…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. Machine-level global optimization is discussed also by Johnson and Miller [31], but, unlike alto, this system does not carry out interprocedural analysis and optimizations. One implication of this is that performance-critical modules written in hand-coded assembly language, third-party software such as libraries for which source code is not available, or code that is not in the source language supported by the compiler, is not amenable to optimization by these tools.…”

Compiler optimizations for the PA-8000

“…The user selects desired optimizing components with command-line options; for example, the low level optimizer may be invoked via -0 and the high level optimizer is typically invoked via + 0 3 or +04. The [29,21]. It consists of the following phases: basic block formation and optimization, interval construction and calculation of reaching dataflow information [27], global common subexpression elimination, calculation of need dataflow information, promotion of' memory variables, loop optimizations, formation and optimization of definitionhe webs, instruction scheduling, coloring register allocation, peephole optimization, branch optimization, branch reachability fixing, and (again) instruction scheduling.…”

Alias analysis of executable code

“…Recent years have seen increasing interest in reasoning about and manipulating executable les 5,15,20,25,27,30,31,33]. When working with an executable le, we typically have information about the entire program|including, potentially, library functions| that is usually not available at compile time.…”