Architecture-dependent register allocation and instruction scheduling on VLIW

Architecture-dependent optimization is very important for VLIW architecture. In this paper we have described a new architecture of VLIW-EPRC. And a new architecture-dependent optimization method based on micro-operation for exploiting the characteristics of target machine is described. The software bypassing based on static single-assignment is used to allocate pseudo registers, and the scheduling and register assignment algorithm with heuristics is adopted. Finally, the experimental results of Livermore and Whetstone are given.