The classification, fusion, and parallelization of array language primitives

Abstract

Absstncct-We present a classification scheme for array language primitives that quantifies the variation in parallelism and data locality that results from the fusion of any two primitives. We also present an algorithm based on thii scheme that efficiently determines when it is beneficial to fuse any two primitives. Experhentnl results show that five LINPACK routines report 50% performance improvement from the fusion of array operators.

A Function-Composition Approach to Synthesize Fortran 90 Array Operations

“…In addition, statement merge permits more opportunities for synthesizing consecutive array functions by concatenating statements. Statement merge needs an accurate data flow analysis of the reaching definition and use information, and this method is well treated in [17]. The main issue is to decide whether an array is a temporary array or not.…”

“…Based on the statement merge optimization in [17], this synthesis framework can be applied to synthesize the array operations in different array assignment statements. For example, if T is a program temporary array, the following two array expressions: T=TRANSPOSE(CSHIFT(A, 1, 1))…”

“…Based on the statement merge optimization in [17], this synthesis framework can be applied to synthesize the array operations in different array assignment statements. In this paper, we enable the statement merge techniques to include the Fortran 90 WHERE and ELSE-WHERE constructs.…”

“…In addition, we employ the solution to address the performance anomaly problem in this test case. The code fragment in the sixth suite is modified from the Sandia Wave application [17]. The Sandia Wave program solves a 2-D wave proram in a periodic domain with an obstacle.…”

“…To obtain this information, we need to perform liveness analysis [2]. For example, see the following code fragment which is in the inner-most loop of Sandia Wave application [17].…”