Load Scheduling with Profile Information

Within the past five years, many manufactures have added hardware performance counters to their microprocessors to generate profile data cheaply. We show how to use Compaq’s DCPI tool to determine load latencies which are at a fine, instruction granularity and use them as fodder for improving instruction scheduling. We validate our heuristic for using DCPI latency data to classify loads as hits and misses against simulation numbers. We map our classification into the Multiflow compiler’s intermediate representation, and use a locality sensitive Balanced scheduling algorithm. Our experiments illustrate that our algorithm improves run times by 1% on average, but up to 10% on a Compaq Alpha.