The Impact of Fetch Rate and Reorder Buffer Size on Speculative Pre-Execution

Pre-execution systems reduce the impact of cache misses and branch mispredictions by forking a slice, a code fragment derived from the program, in advance of frequently mispredicted branches and frequently missing loads in order to either resolve the branch or prefetch the load. Because unnecessary instructions are omitted the slice reaches the branch or load before the main thread does, for loads this time margin can reduce or even eliminate cache miss delay