Mitch Hayenga got review request #2002!
mem: Add additional tolerance to stride prefetcher (prefetcher patch #3)
Review Request #2002 - Created Sept. 6, 2013 and submitted
| Information | |
|---|---|
| Mitch Hayenga | |
| gem5 | |
| Reviewers | |
| Default | |
Adds tolerance to the existing stride prefetcher. Forces the prefetcher to mispredict twice in a row before resetting the confidence of prefetching. This helps cases where a load PC strides by a constant factor, however it may operate on different arrays at times. Avoids the cost of retraining. Primarily helps with small iteration loops. Note: Relies on the patch from review request #2000 Improves performance by 2.1% on SPECINT, 0.3% on SPECFP (geomean).
Posted (Sept. 12, 2013, 12:45 p.m.)
Is this a 1 bit counter? If so, would it be worth making it an int and allow values bigger than 1? Some comments would be great as well :-)
seems fine to me.. only feedback in tolerace seems like it should be configurable and an int rather than a bool