The PI will work towards rapid compiler prototyping for the Explicit Multi-Threading (XMT) project at the University of Maryland. This target "approximate" compiler will correctly translate a high-level code into the assembly code for the XMT architecture; the performance of the assembly code is expected be within some small percentage from optimal. One graduate student will devote a full year to this effort. In a second project, the PI plans to design an on-chip interconnect for the memory system. This will explore several alternative ways for connecting the local registers in eash thread control unit (TCU) with the first level of the memory, which is to be interleaved; this is a demanding challenge as the interconnect needs to provide high bandwidth and low latency. A second graduate student will devode a full year to this effort.
