E ectively Exploiting Indirect Jumps

This paper describes a general code-improving transformation that can coalesce conditional branches into an indirect jump from a table. Applying this transformation allows an optimizer to exploit indirect jumps for many other coalescing opportunities besides the translation of multiway branch statements. First, dataaow analysis is performed to detect a set of coalescent conditional branches, which are often separated by blocks of intervening instructions. Second, several techniques are applied to reduce the cost of performing an indirect jump operation, often requiring the execution of only two instructions on a SPARC. Finally, the control ow is restructured using code duplication to replace the set of branches with an indirect jump. Thus, the transformation essentially provides early resolution of conditional branches that may originally have been some distance from the point where the indirect jump is inserted. The transformation can be frequently applied with often signiicant reductions in the number of instructions executed, total cache work, and execution time. In addition, we show that with branch target buuer support, indirect jumps improve branch prediction since they cause fewer mispredictions than the set of branches they replaced. 1. INTRODUCTION Most high-level languages provide multiway branch statements to allow programmers to write more readable code. The characteristic feature of a multiway statement is the ability to select an action based on the value of a control expression. Without performing any optimization, a compiler would translate each case label of the multiway statement into a conditional branch. Because of the widespread usage of multiway statements, instruction sets commonly support an indirect jump from a table in order to reduce the cost of such sequences of conditional branches. As a result, compiler front-ends typically generate an indirect jump from a table as one translation alternative 1 for multiway statements Sale 1981; Spuler 1994]. This traditional approach for using indirect jumps poses two problems. First, it is diicult to determine when the indirect jump can be eeectively used in a machine-independent fashion since an accurate cost-beneet estimate can only be made after generating machine code. Second, many code-improving opportunities