Compilation for Domain-Specific Computing

Monday, September 10, 2007
7:00 PM
Free and open to the public

As alternatives to multi-core CPUs and GPUs, field-programmable gate-arrays (FPGAs) can be used as customized computing engines for accelerating a wide range of applications. The recent developments by AMD to open up its HyperTransport bus and Intel to open up its Front-side Bus lead to new high-performance computing platforms with high-bandwidth communication between CPUs and FPGAs. The effective use of such hybrid computing platforms relies on efficient compilation tools and technologies for automatic synthesis of applications specified in typical software languages, such as C or C++, onto FPGAs. In this talk, I shall present the platform-based compilation and synthesis system, named xPilot, developed at UCLA. The xPilot provides advanced behavioral synthesis capability for compiling C or C++ applications to FPGA platforms for logic, interconnects, and memory optimization with flexibile performance and area trade-off. It includes a number of algorithmic innovations, such as scheduling based on the system of difference constraints, resource binding for distributed memory architectures, etc. The xPilot system has been licensed by AutoESL Design Technologies, Inc. for commercialization. The AutoPilot tool from AutoESL, derived from xPilot, has been successfully used to compile a number of computation-intensive applications in many domains, such as multimedia applications, financial engineering, and computer-aided designs. I shall report some of the application results.

x x


Jason Cong

UCLA Computer Science Department