Automatic Design of Custom Wide-Issue Counterflow Pipelines

Abstract

Application-specific processor design is a promising approach for meeting the performance and cost goals of a system. Application-specific integrated processors (ASIP's) are especially promising for embedded systems (e.g., automobile control systems, avionics, cellular phones, etc.) where a small increase in performance and decrease in cost can have a large impact on a product's viability. Sutherland, Sproull, and Molnar have proposed a new pipeline organization called the Counterflow Pipeline (CFP) that may be appropriate for ASIP design. This paper extends the original CFP microarchitecture to a very long instruction word (VLIW) organization for the custom design of instruction-level parallel processors tailored to the requirements of computation-intensive inner loops. First, we describe our extensions to the CFP to support automatic customization of application-specific VLIW processors. Second, we present an automatic design system that customizes a wide-issue counterflow pipeline (WCFP) to the resource and data flow requirements of a software pipeline loop. Third, we show that custom asynchronous WCFP's achieve cycles per operation measurements that are competitive with custom VLIW organizations at a potentially low design complexity. Finally, the paper describes several enhancements that can be made to WCFP's to further improve performance of kernel loops.