News for hackers. Stuff that matters.

A comment on What's New in CPUs Since the 80s and How Does It Affect Programmers?.

Re:What's New in CPUs Since the 80s and How Does It Affect P... by acallan on Monday January 12, 2015 @08:30AM (#8873751)
The author is incorrect in the section about memory fences. x86 has strong memory ordering [1], which means that writes always appear in program order with respect to other cores. Use a memory fence to guarantee that reads and writes are memory bus visible.

The example that the author gives does not apply to x86.

[1]There are memory types that do not have strong memory ordering, and if you use non-temporal instructions for streaming SIMD, SFENCE/LFENCE/MFENCE are useful.

[ Parent | on HN ]
Re:What's New in CPUs Since the 80s and How Does It Affect P... by Animats on Monday January 12, 2015 @10:18PM (#8877929)
Modern X86 machines do not quite have strong memory ordering. Loads may be reordered with older stores to different locations. This breaks some of the overly clever "lock free" algorithms:

http://bartoszmilewski.com/2008/11/05/who-ordered-memory-fen...

http://preshing.com/20120515/memory-reordering-caught-in-the...

This whole area is very touchy and easy to get wrong. Further reading:

https://software.intel.com/en-us/articles/tsx-anti-patterns-...

[ Parent | on HN ]
Re:What's New in CPUs Since the 80s and How Does It Affect P... by martincmartin on Monday January 12, 2015 @09:36AM (#8874037)
The authors point still holds if writes might not appear in program order on other CPU sockets, not just other cores within the same socket.

Do you have a reference for the strong memory ordering on x86? I'd like to read more about it.

[ Parent | on HN ]
Re:What's New in CPUs Since the 80s and How Does It Affect P... by rayiner on Monday January 12, 2015 @10:01AM (#8874168)
x86 enforces (essentially) total store order across all sockets: http://www.cl.cam.ac.uk/~pes20/weakmemory/index3.html. The barriers are still useful for kernel code because other processors on the machine usually don't participate in the cache-coherency protocol.
[ Parent | on HN ]