Lamp: tools for creating application-specific fpga coprocessors

摘要: Field Programmable Gate Arrays (FPGAs) have begun to appear as accelerators for general computation. Their potential massive parallelism, high on-chip memory bandwidth, and customizable interconnection networks all contribute demonstrated 100-1000× increases in application performance relative current PCs. FPGA coprocessors been available niche markets years, are now appearing mainstream supercomputers from vendors including Cray Silicon Graphics. Available development tools do not address developers of computing applications, however. Traditional design meet the gate-level needs logic designers, but present a model that vanishingly few software can use. Likewise, designers understand structures performance, rarely know biology, biochemistry, or other applications need acceleration. Logic must both participate creating efficient, useful accelerators, their different kinds participation supported by tools. This work presents two major sets contributions. The first is proof example FPGAs give speedups large families bioinformatics computational biology (BCB), sequence alignment, molecule docking, string analysis. These demonstrations also provide beginnings library reusable structures. The second set contributions novel features accelerator based on Architecture Model Parameterization (LAMP). LAMP broad, point solutions narrow problem statements. separates who create efficient hardware structures, specialists tailor specific members family. This separation enables customization without access skills. Finally, provides mechanisms automating tradeoff between complexity quantity parallel processing elements (PEs), allowing fewer PEs larger numbers small ones, subject FPGA's resource constraints. creates unique ability allocate resources differently each member an family, according datatypes functions family member. Performance results prototype presented, using sample BCB applications.