A Spill Code Placement Framework for Code Scheduling

Many of compiler optimization techniques either reply on or can benefit from the timing information associated with each instruction and each variable in a given program. Using such information, in this paper, we develop an analytical approach that helps the parallelizing compiler to determine whether to spill a variable or not, which variable to spill and the places where spill code needs to be added, when a register is needed for a computation but all available registers are in use. The preliminary experimental results show that this new approach produces better object code.