practice

Doi: 10.1145/1785414.1785433

Article development led by queue.acm.org

Maybe it’s Fortran.
Or maybe it just doesn’t matter.
BY euGene Loh
the ideal hPc
Programming
Language

The darpa hiGh Productivity Computing Systems (HPCS) program sought a tenfold productivity improvement in trans-petaflop systems for high-performance computing (HPC). This article describes programmability studies undertaken by Sun Microsystems in its HPCS participation. These studies were distinct from Sun’s ongoing development of a new HPC programming language (Fortress) and the company’s broader HPCS productivity studies, though there was certainly overlap with both activities.

These programmability studies began with a focus on programming languages, but the focus quickly shifted to other topics. Existing languages—notably Fortran, which is arguably still the primary language in HPC—proved remarkably adequate. Programming challenges stem mostly from other factors.

What if programming did not mean having to learn a language someone else devised and then wrestling with the limitations of that language, its compilers,

and computers to implement your task? What if it meant, in a sense, the opposite? You could write your program in whatever way was most expressive for you, without regard for language rules imposed by someone else. Then it would be someone else’s job to define the programming language that would make sense of what you wrote, write the compilers to digest the program, and build the computers that would efficiently run the task you specified.

We undertook such an exercise to get a feel for what an “ideal” programming language for HPC applications might look like. Our approach was to take existing HPC programs and have someone rewrite them in whatever way suited that individual, not bound by the constraints of any existing computer, compiler, or language. Rather, he was invited to write whatever seemed most expressive. We might not be able to compile or run these programs, but we could at least see what the writer wanted.

Almost immediately, we were struck by what we were seeing. Of course, the rewritten code was much more compact and readable than the original, but, surprisingly, the “ideal” programming language was basically Fortran.

My first job here is to convince you that this finding is not ridiculous. I admit, the experiment was biased in that we were starting with existing code, mostly written in Fortran, and used a human subject who was not only familiar with Fortran but indeed embraced it. The main point, however, is less that every programmer would have ended up preferring Fortran and more that the problems with the original source code have more to do with reasons other than the limitations of existing programming languages. We look at some of these reasons here.

The DARPA HPCS program also sponsored the development of new programming languages: Chapel from Cray, Fortress from Sun, and X10 from IBM. Proponents of those languages would show early on how rewriting