Fastcode MM challenge: New Address Space Creep (Larger Blocks) Benchmark


2005-06-08 09:19:28 PM
delphi206
Hi All,
Posted to attachments. The previous version didn't touch all allocated
pages.
Best would probably be to touch every cache line, but then the slowdown
would overshadow the speed of the MM itself. I suppose touching every page
is a reasonable compromise.
I think we should do the same in the replay benchmarks, and all other
benchmarks that do not currently at least access every page it allocates.
Regards,
Pierre