An evaluation of a microprocessor with two independent hardware execution threads coupled through a shared cache
share
We investigate the utility of augmenting a microprocessor with a single execution pipeline by adding a second copy of the execution pipeline in parallel with the existing one. The resulting dual-hardware-threaded microprocessor has two identical, independent, single-issue in-order execution pipelines (hardware threads) which share a common memory sub-system (consisting of instruction and data caches together with a memory management unit). From a design perspective, the assembly and verification of the dual threaded processor is simplified by the use of existing verified implementations of the execution pipeline and a memory unit. Because the memory unit is shared by the two hardware threads, the relative area overhead of adding the second hardware thread is 25% of the area of the existing single threaded processor. Using an FPGA implementation we evaluate the performance of the dual threaded processor relative to the single threaded one. On applications which can be parallelized, we observe speedups of 1.6X to 1.88X. For applications that are not parallelizable, the speedup is more modest. We also observe that the dual threaded processor performance is degraded on applications which generate large numbers of cache misses.
READ FULL TEXT