home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.arch
- Path: sparky!uunet!sun-barr!cs.utexas.edu!sdd.hp.com!caen!sol.ctr.columbia.edu!eff!iWarp.intel.com|ichips!ichips!glew
- From: glew@pdx007.intel.com (Andy Glew)
- Subject: Re: interrupt overhead
- In-Reply-To: henry@zoo.toronto.edu's message of 23 Jul 92 21:08:31 GMT
- Message-ID: <GLEW.92Aug18145622@pdx007.intel.com>
- Sender: news@ichips.intel.com (News Account)
- Organization: Intel Corp., Hillsboro, Oregon
- References: <13v85hINN2og@rodan.UU.NET> <GLEW.92Jul14234349@pdx007.intel.com>
- <Brsx7o.G69@zoo.toronto.edu> <1992Jul22.163956.57436@cc.usu.edu>
- <Brus8r.2K7@zoo.toronto.edu> <Brv1E9.76t@zoo.toronto.edu>
- Date: Tue, 18 Aug 1992 22:56:22 GMT
- Lines: 25
-
- >Pretty much the inescapable minimum is two pipeline breaks plus a cycle for
- >the "return from interrupt" instruction...
-
- John Carr has pointed out to me, in private mail, that a clever CPU
- doesn't actually need to flush its pipelines, if the things in the
- pipeline are guaranteed not to incur exceptions. So it's potentially
- not much worse than a couple of taken branches.
-
- This is a latency versus thruput tradeoff.
-
- If you insert interrupts at the end of the pipe, instead of clearing
- the pipe, you are potentially adding to the interrupt delivery
- latency.
-
-
- --
-
- Andy Glew, glew@ichips.intel.com
- Intel Corp., M/S JF1-19, 5200 NE Elam Young Pkwy,
- Hillsboro, Oregon 97124-6497
-
- This is a private posting; it does not indicate opinions or positions
- of Intel Corp.
-
- Intel Inside (tm)
-