Polaris: A New-Generation Parallelizing Compiler for MPPs

ion for Inner Loops When the algorithm nds a loop nested inside a loop body, it will recursively call itself on the inner loop. To hide the control ow of an inner loop, we introduce some abstraction and extend the previous de nition from a basic block to a complete loop. We start by de ning the information for one iteration of the loop. De nition 5 Let L be a loop and V AR be the variables in the program. We de ne the following set as summary set for body(L). 1. DEFb(L) := fv 2 VAR : v has a MRD reaching all exits node of body(L) g 2. USEb(L) := fv 2 V AR : v has an outward exposed use in body(L) g 3. KILLb(L) := DEFb(L) 4. PRIb(L) := fv 2 VAR : every use of v has a reaching MRD in body(L) g 2 The summary set is an abstraction of the e ect of a loop iteration on the data ow values. Using the summary set, we can ignore the structure of the inner loops in the analysis of the outer loop. The trade-o is that we have to make a conservative approximation and may lose information in the process. DEFb(L) is themust de ne variables for one iteration of L; i.e. the must de ne variables upon exiting the iteration: DEFb(L) = \(MRDout(t) : t 2 exits(L))