Pillar: A Parallel Implementation Language

Abstract

As parallelism in microprocessors becomes mainstream, new programming languages and environments are emerging to meet the challenges of parallel programming. To support research on these languages, we are developing a lowlevel language infrastructure called Pillar (derived from Parallel Implementation Language). Although Pillar programs are intended to be automatically generated from source programs in each parallel language, Pillar programs can also be written by expert programmers. The language is defined as a small set of extensions to C. As a result, Pillar is familiar to C programmers, but more importantly, it is practical to reuse an existing optimizing compiler like gcc [1] or Open64 [2] to implement a Pillar compiler. Pillar's concurrency features include constructs for threading, synchronization, and explicit data-parallel operations. The threading constructs focus on creating new threads only when hardware resources are idle, and otherwise executing parallel work within existing threads, thus minimizing thread creation overhead. In addition to the usual synchronization constructs, Pillar includes transactional memory. Its sequential features include stack walking, second-class continuations, support for precise garbage collection, tail calls, and seamless integration of Pillar and legacy code. This paper describes the design and implementation of the Pillar software stack, including the language, compiler, runtime, and high-level converters (that translate high-level language programs into Pillar programs). It also reports on early experience with three high-level languages that target Pillar. for these languages will have strong similarities. Pillar factors out these similarities and provides a single set of components to ease the implementation and optimization of a compiler and its runtime for any parallel language. The core idea of Pillar is to define a low-level language and runtime that can be used to express the sequential and concurrency features of higher-level parallel languages. The Pillar infrastructure consists of three main components: the Pillar language, a Pillar compiler, and the Pillar runtime.

The Intel labs Haskell research compiler

“…The boxes represent transformation passes of the compiler, and lines represent data flowing through the compiler, annotated with the specific IR going into and coming out of each pass. When we compile a set of input Haskell files, we first invoke GHC to compile them to external Core files, read them back in, then go through various internal transformations and multiple IRs, before outputting a program in a language called Pillar [2] (an extension to the C language inspired by C-- [23]). A tool called Pillar2c is used to translate the Pillar code to C code, and the Intel C compiler is used to produce the final machine executable.…”

“…FLRC and Pillar [2] were concurrently developed in the same lab and one goal of the FLRC project was to act as a test case for language development on top of Pillar. Pillar is a language, compiler, and runtime that provides programming language infrastructure.…”

“…We use GHC to perform de-sugaring and high-level optimization; intercept the Core IR from GHC and translate the lazy Core language into a strict, lower-level, general-purpose IR; perform aggressive wholeprogram compilation on this IR; and compile the result eventually to Pillar [2], also inspired by C--. Our choice of building a Haskell compiler by marrying two compiler platforms together is deliberate.…”

Automatic SIMD vectorization for Haskell

“…In order to get immutable code written in this style from GHC and the libraries discussed above, we modified GHC with new primitive immutable array types 1 and primitive operations on them. The additional operations include creation of a new uninitialized array, initializing writes, and a subscript operation.…”

“…In this case, the variable b s is loop invariant, and consequently b v will consist simply of eight copies of the same value. Semantically then, we can replace our use of the general vector load instruction with the simpler gather 1), is the induction variable computation, and as such requires special treatment. The result of this instruction is used for three purposes: in the computation of the test to decide whether to proceed with another iteration, to provide a new value for the induction variable in the next iteration, and to provide the value passed out on the exit edge.…”