Polaris: Improving the effectiveness of parallelizing compilers

Abstract

Abstract. It is the goal of the Polaris project to develop a new parallelizing compiler that will overcome limitations of current compilers. While current parallelizing compilers may succeed on small kernels, they often fail to extract any meaningful parallelism from large applications. After a study of application codes, it was concluded that by adding a few new techniques to current compilers, automatic parallelization becomes possible. The techniques needed are interprocedural analysis, scalar and array privatization, symbolic dependence analysis, and advanced induction and reduction recognition and elimination, along with run-time techniques to allow data dependent behavior.

Symbolic range propagation

“…This is done by calculating the forward di erence (x + 1 ) 2 ; (x + 1 ) ; (x 2 ; x) = 2 x, then determining whether 2 x is greater-equal or less-equal to zero by a recursive call to the expression comparison algorithm. Now, 2 x is greater than zero, since 2 x ) 2 …”

“…The 1 A strongly-connected component of a graph is a maximal subgraph of that graph where each v ertex in the subgraph can reach all other vertices in the subgraph. 2 A topological ordering of a directed acyclic graph is an ordering such t h a t i f t h e r e i s a p a t h f r o m v ertex u to vertex v then u occurs before v in this ordering. 3 By de nition of SCCs, one can always topologically sort the SCCs of a graph in respect to each other.…”

Gossamer: A Lightweight Approach to Using Multicore Machines

“…This approach is general and efficient, but it is not simple because the programmer has to control every aspect of concurrency and synchronization using low-level APIs. At the other extreme, one could use a compiler that automatically parallelizes code or uses optimistic concurrency techniques [8,12,10]. This is simple and it works well for some kinds of parallelism (e.g., independent loop iterations), but it is not general.…”

“…This frees the programmer from having to learn how to write a parallel program. Such compilers include Intel C compiler [10], Paradigm [29], Polaris [11] Rice Fortran D [1], SUIF [59], and Vienna Fortran [8];…”

Loop Selection for Thread-Level Speculation

“…One attractive method of fully utilizing such resources is to automatically extract parallel threads from existing programs. However, automatic parallelization [4,10] for general-purpose applications (e.g., compilers, spreadsheets, games, etc.) is difficult because of pointer aliasing, irregular array accesses, and complex control flow.…”