A DYNAMIC MULTITHREADING PROCESSOR

An architecture that features dynamic multithreading execution of a single program is studied in this dissertation. Threads are created automatically by hardware at procedure and loop boundaries and executed speculatively on a simultaneous multithreading pipeline. Data prediction is used to alleviate dependency constraints and enable lookahead execution of the threads. A two-level hierarchy significantly enlarges the instruction window. Selective recovery from the second level instruction window takes place after a midspredicted input to a thread is corrected. The second level is slower to access but has the advantage of large storage capacity. We show several advantages of this architecture: (1) it increases a superscalar’s execution throughput and functional units ' utilization, (2) it adds minimal complexity to the critical execution path, therefore, throughput increase is realizable at little or no cycle time cost, (3) it minimizes the impact of ICache misses and branch mispredictions by fetching and dispatching instructions out-of-order, (4) it uses a novel value prediction and recovery mechanism to reduce artificial data dependencies created by the use of a stack to manage run-time storage. (5) it achieves a speedup of almost 30 % on the integer SPEC95 benchmarks, with equal execution units. This speedup was measured on single thread binaries, without any compiler support, and using a detailed performance simulator