An array operation synthesis scheme to optimize Fortran 90 programs

Abstract

An increasing number of programming languages, such as Fortran 90 and APL, are providing a rich set of intrinsic array functions and array expressions. These constructs which constitute an important part of data parallel languages provide excellent opportunities for com-SHAPE, EOSHIFT, MERGE, and WHERE constructs together. Experimental results show speedups from 1.21 to 2.95 for code fragments from real applications on a Sequent multiprocessor machine by incorporating the proposed optimizations.

I/O-efficient statistical computing with RIOT

“…This approach, however, as Lewis et al [9] pointed out elegantly, "solves the problem at a greater conceptual distance from the source of the problem and at a greater cost." Instead, RIOT optimizes at the higher level of array operations, an approach also used by [4,6,9,8,15].…”

“…During evaluation, intermediate results are "streamed" in time, instead of being materialized in temporary arrays. Hwang et al [6] generalized the idea and applied it to Fortran 90. They also support statement merges, whereby certain adjacent statements can be merged into one and processed as a single loop without materializing temporary arrays.…”

CRAFT: a framework for F90/HPF compiler optimizations

“…Chatterjee et al [58,59] extended that work to allow loop nests. Ju, Wu and Carini [60] (and later Hwang, Lee and Ju [61]) proposed a synthesis scheme for combining consecutive data reference patterns to reduce communication. Another line of work optimizes communication using data-flow analysis, e.g.…”