Improving dynamic cluster assignment for clustered trace cache processors

This work examines dynamic cluster assignment for a clustered trace cache processor (CTCP). Previously pro-posed cluster assignment techniques run into unique prob-lems as issue width and cluster count increase. Realistic design conditions, such as variable data forwarding laten-cies between clusters and a heavily partitioned instruction window, increase the degree of difficulty for effective cluster assignment. In this work, the trace cache and fill unit are used to per-form dynamic cluster assignment. The retire-time fill unit analysis is aided by a dynamic profiling mechanism em-bedded within the trace cache. This mechanism provides information about inter-trace data dependencies, an ele-ment absent in previous retire-time CTCP cluster assign-ment work. The strategy proposed in this work leads to more intra-cluster data forwarding and shorter data forwarding distances. In addition, performing cluster assignment at re-tire time reduces issue-time complexity and eliminates early pipeline stages. This increases overall performance for in-teger programs by 11.5 % over our base CTCP architecture. This speedup is significantly higher than a previously pro-posed retire-time CTCP assignment strategy. Dynamic clus-ter assignment is also evaluated for several alternate cluster designs as well as for media benchmarks. 1