Accelerating information experts through compiler design

Abstract

Dyalog APL is a tool of thought for information experts, enabling rapid development of domain-centric software without the costly software engineering feedback loop often required. The Dyalog APL interpreter introduces performance constraints that hinder the analysis of large data sets, especially on highly-parallel computing architectures. The Co-dfns compiler project aims to reduce the overheads involved in creating high-performance code in APL. It focuses on integrating with the APL environment and compiles a familiar subset of the language, delivering significant performance and platform independence to information experts without requiring code rewrites and conversion into other languages.The design of the Co-dfns compiler, itself an APL program, possesses a unique architecture that permits implementation without branching, recursion, or other complex forms of control flow. By integrating specific optimizations, the generated code competes with hand-written C code in the domain of financial simulations, exceeding it when integrated into the environment. Preliminary results demonstrate platform independent performance across CPUs and GPUs without modification of the source. Work continues to improve performance both of the architecture and the generated code. Eventually, the project hopes to convincingly demonstrate a wider range of techniques that extend the suitable domain for effective array programming.

The key to a data parallel compiler

“…The Co-dfns compiler is an experimental commercial compiler funded by Dyalog, Ltd. [13,14] It is intended to provide increased scaling and performance for APL programmers using the dfns syntax (lexically scoped, functionally oriented APL). These techniques form one of the foundational elements to the architecture of the compiler, which is built in the style of a Nanopass [20] compiler, with very small passes chained together through composition, with the added requirements that each compiler pass must be fully data parallel.…”

“…This enables us to perform arbitrary computation over arbitrary sub-trees of an AST selected by parentchild relationships in the manner often seen in compiler transformations. These ideas have been further implemented and tested through the implementation of a complete data-parallel by construction compiler, called Co-dfns, that compiles a lexically scoped, functionally oriented dialect of APL with nested functions [13,14]. The compiler targets both the "PU and the CPU, and its core is implemented in this pure, restricted language that uses only function composition over data-parallel primitives.…”