discussion forum
message
| Name: |
Oliver |
| eMail: |
Follow.Me@gmx.net |
| Date: |
June 17, 2003 at 06:00:12 |
| Subject: |
Re: Hyperthreading and 'lock' prefix |
| In Reply To: |
Re: Hyperthreading and 'lock' prefix by Stephan Keil on June 16, 2003 at 06:41:05 |
| Text: |
> Have you tried your program on a HT system? I've posted the following to some newsgroups: || Subject: Can anyone run this on a HT-enabled system ? || From: "Oliver S." || Newsgroups: comp.programming.threads,comp.arch,comp.sys.intel || || I've developed a little Win32 test-app that verifies if on a HT-enabled || P4-processor (or more precisely - on current HT-enabled P4-processors) the || CMPXCHG-instruction is atomic without a lock-prefix. My work is related to || a question in the discussion-forum on sandpile (http://www.sandpile.org || /post/msgs/20003998.htm): || || | the x86 instruction set supports the 'lock' prefix for various || | instructions (e.g. inc, add) to guarantee atomicity in multiprocessor || | environments, i.e. multiple _physical_ processors. But is this || | prefix also necessary for hyperthreading processors, i.e. two || | logical processors within one physical processor? || || As I don't own a HT-capable P4-processor, I'm looking for someone who || will try this on his HT-enabled machine. So please download my test-app || (including the soure) from http://520032173347-0001.bei.t-online.de || /HtLock.zip and report your experience with this test here. || || PS: I don't know if the app will run correctly on HT-enabled systems with || multiple physical CPUs (I don't know if SetThreadAffinity will assign || the two threads to the same physical processor in this case) only re- || sults from systems with a single physical CPU are useful. And some minutes ago, I got the first reply in the NGs: > Just the fact: I tried it on HT system (P4 2.8Ghz, Family 15, Model > 2, Stepping 7) and it showed that unlocked instruction is not HT-safe > (adding lock prefix to cmpxchg fixed thread safety). |
post a followup message
(Be nice... or be blocked. Be technical... or be erased.)
currently posted followup messages
(You may have to press the RELOAD button of your browser.)
|