Abstract: A system and method to use stream filters to defer deallocation of a stream based on the activity level of the stream, thereby preventing a stream thrashing situation from occurring. The least recently used ("LRU") stream is deallocated only after a number of potential new streams are detected.

Abstract: A method for selectively pre-fetching Line M+1 into an L1 instruction cache from an L2 cache or from main memory during the execution of Line M. If unresolved branches exist in pending Line M, Line M+1 is speculative and may be pre-fetched into L1 instruction cache only from L2 cache, not from main memory. Unresolved branches in pending Line M are resolved before Line M+1 is pre-fetched from main memory. If no unresolved branches exist, Line M is committed ("inevitable-speculative") and is pre-fetched from main memory. In this way, no potentially wasteful pre-fetches are performed and main memory bandwidth is preserved.

Abstract: Within a data processing system implementing L1 and L2 caches and stream filters and buffers, prefetching of cache lines is performed in a progressive manner. In one mode, data may not be prefetched. In a second mode, two cache lines are prefetched wherein one line is prefetched into the L1 cache and the next line is prefetched into a stream buffer. In a third mode, more than two cache lines are prefetched at a time. As a result, additional cache lines are progressively prefetched to a data cache as the sequentiality of the accessing of cache lines in memory is demonstrated through sequential addressing requests along a data stream. Furthermore, the stream is physically distributed. In other words, at least one line, but not all lines, of the stream are placed within the cache.