We characterize high-performance streaming applications as a new and distinct domain of programs that is becoming increasingly important.

With the increasing miniaturization of transistors, wire delays are becoming a dominant factor in microprocessor performance. To address this issue, a number of emerging architectures contain replicated processing units with software-exposed communication between one unit and another (e.g., Raw, iWarp, SmartMemories). However, for their use to be widespread, it will be necessary to develop compiler technology that enables a portable, high-level language to execute efficiently across a range of wireexposed architectures.

Streaming programs represent an increasingly important and widespread class of applications that holds unprecedented opportunities for high-impact compiler technology. Unlike sequential programs with obscured dependence information and complex communication patterns, a stream program is naturally written as a set of concurrent filters with regular steady-state communication. The StreamIt language aims to provide a natural, high-level syntax that improves programmer productivity in the streaming domain. At the same time, the language imposes a hierarchical structure on the stream graph that enables novel representations and optimizations within the StreamIt compiler. We define the "stream dependence function", a fundamental relationship between the input channels of two filters in a stream graph. We also describe a suite of stream optimizations, a denotational semantics for validating these optimizations, and a novel phased scheduling algorithm for stream graphs. In addition, we have implemented a prototype of the StreamIt optimizing compiler that is showing promising results.

This paper addresses the problem of trading-o between the minimization of program and data memory requirements of single-processor implementations of dataow programs. Based on the formal model of synchronous data ow (SDF) graphs [20], so called single appearance schedules are known to be program-memory optimal. Among these schedules, buer memory schedules are investigated and explored based on a two-step approach: (1) An Evolutionary Algorithm (EA) is applied to eÆciently explore the (in general) exponential search space of actor ring orders. (2) For each order, the buer costs are evaluated by applying a dynamic programming post-optimization step (GDPPO). This iterative approach is compared to existing heuristics for buer memory optimization. 1 Introduction In recent years, software synthesis has become an important component of the implementation process for embedded VLSI systems. This is due to important advantages of software such as exibility, and shorter design time (fa...