Re:Who ordered memory fences on an x86? (2008)
by pkhuong on Friday November 07, 2014 @12:01PM
(#8573461)
The fun thing about membars on x86 is that, unless you're playing with nontemporal stores or non-standard memory types, LOCKed ops are more efficient fences than mfence.
[ Parent
| on HN ]
Re:Who ordered memory fences on an x86? (2008)
by mattnewport on Friday November 07, 2014 @12:37PM
(#8573624)
Missed this somehow before making my comment saying essentially the same thing. Unfortunately the code base I've been maintaining heavily overuses mfences at a measurable performance penalty on x64.