Register Pressure Sensitive Redundancy Elimination

Redundancy elimination optimizations avoid repeated computation of the same value by computing the value once, saving it in a temporary, and reusing the value from the temporary when it is needed again. Examples of redundancy elimination optimizations include common subexpression elimination, loop invariant code motion and partial redundancy elimination. We demonstrate that the introduction of tem-poraries to save computed values can result in a signiicant increase in register pressure. An increase in register pressure may in turn trigger generation of spill code which can more than ooset the gains derived from redundancy elimination. While current techniques minimize increases in register pressure, to avoid spill code generation it is instead necessary to ensure that register pressure does not exceed the number of available registers. In this paper we develop a redundancy elimination algorithm that is sensitive to register pressure: our novel technique rst sets upper limits on allowed register pressure and then performs redundancy elimination within these limits. By setting strict register pressure limits for frequently executed (hot) blocks and higher limits for infrequently executed (cold) blocks, our algorithm permits trade-oo between redundancy removal from hot blocks at the expense of introducing spill code in cold blocks. In addition, the program proole is also used to prioritize opti-mizable expressions; when not enough registers are available, the most prootable redundancies are removed rst. To increase redundancy elimination within the allowed register pressure, our algorithm lowers the pressure with two new program transformation techniques: (a) whenever possible, we avoid inserting a temporary and instead access the reused value from existing variables, which reduces the life time of the temporary beyond existing live-range optimal algorithms; and (b) the live ranges of variables referenced by the expressions are shortened by combining expression hoisting with assignment sinking.