Sign In

Communications of the ACM

Research highlights

Automatically Accelerating Non-Numerical Programs By Architecture-Compiler Co-Design

Automatically Accelerating Non-Numerical Programs by Architecture-Compiler Co-Design, illustration

Credit: Getty Images

Because of the high cost of communication between processors, compilers that parallelize loops automatically have been forced to skip a large class of loops that are both critical to performance and rich in latent parallelism. HELIX-RC is a compiler/microprocessor co-design that opens those loops to parallelization by decoupling communication from thread execution in conventional multicore architecures. Simulations of HELIX-RC, applied to a processor with 16 Intel Atom-like cores, show an average of 6.85× performance speedup for six SPEC CINT2000 benchmarks.

Back to Top

1. Introduction

On a multicore processor, the performance of a program depends largely on how well it exploits parallel threads. Some computing problems are solved by numerical programs that are either inherently parallel or easy to parallelize. Historically, successful parallelization tools have been able to transform the sequential loops of such programs into parallel form, boosting performance significantly. Most software, however, is still sequentially designed and largely non-numerical, with irregular control and data flow. Because manual parallelization of such software is error-prone and time-consuming, automatic parallelization of non-numerical programs remains an important open problem.

The last decade has seen impressive steps toward a solution, but when targeting commodity processors, existing parallelizers still leave much of the latent parallelism in loops unrealized.5 The larger loops in a program can be so hard to analyze accurately that apparent dependences often flood communication channels between cores. Smaller loops are more amenable to accurate analysis, and our work shows that there is a lot of parallelism between the iterations of small loops in non-numerical programs represented by SPECint2000 benchmarks.4 But even after intense optimization, small loops typically include loop-carried dependences, so their iterations cannot be entirely independent—they must communicate. Because the iterations of a small loop are short (25 clock cycles on average for SPECint2000), their communications are frequent.


No entries found

Log in to Read the Full Article

Sign In

Sign in using your ACM Web Account username and password to access premium content if you are an ACM member, Communications subscriber or Digital Library subscriber.

Need Access?

Please select one of the options below for access to premium content and features.

Create a Web Account

If you are already an ACM member, Communications subscriber, or Digital Library subscriber, please set up a web account to access premium content on this site.

Join the ACM

Become a member to take full advantage of ACM's outstanding computing information resources, networking opportunities, and other benefits.

Subscribe to Communications of the ACM Magazine

Get full access to 50+ years of CACM content and receive the print version of the magazine monthly.

Purchase the Article

Non-members can purchase this article or a copy of the magazine in which it appears.
Sign In for Full Access
» Forgot Password? » Create an ACM Web Account