A branch target instruction prefetchnig technique for improved performance

Modern processors are much faster than the main memory. Cache memories are introduced to reduce this speed gap. However, instruction cache misses can severely limit the performance of today's superscalar processors. Prefetch algorithms attempt to reduce the performance degradation by bringing cache lines into the instruction cache. Different algorithms like next line, target line and wrong-path prefetching are well studied. A new Branch Target Address (BTA) prefetching scheme is proposed. This technique substantially reduces the cycles loss due to branch instruction cache misses. It has achieved substantial performance improvement over other prefetching techniques; for example 10-15% improvement over wrong-path instruction prefetching. With the help of a small size additional buffer, cache access rate as well as cache pollution has been reduced drastically. The new scheme works better in processor designs where memory latencies are likely to be longer.