Path: utzoo!mnetor!uunet!lll-winken!lll-lcc!ames!nrl-cmf!cmcl2!brl-adm!umd5!purdue!i.cc.purdue.edu!j.cc.purdue.edu!pur-ee!uiucdcs!uxc.cso.uiuc.edu!ccvaxa!aglew
From: aglew@ccvaxa.UUCP
Newsgroups: comp.arch
Subject: Re: The advantages of FREQUENCY
Message-ID: <28200095@ccvaxa>
Date: 9 Feb 88 00:58:00 GMT
References: <3783@hall.cray.com>
Lines: 33
Nf-ID: #R:hall.cray.com:3783:ccvaxa:28200095:000:1532
Nf-From: ccvaxa.UUCP!aglew    Feb  8 18:58:00 1988


>The performance benefits of FREQUENCY in large scale architectures would
>likely be very slight. In more complex architectures with significant
>instruction buffers (such as supercomputers), jumps are very expensive
>and, unless you moved the less frequent block totally out of the flow of
>the program, you would be jumping around one of the blocks either way.

Well, I would like to remove the really infrequent cases completely
out of the flow of the program - at the very least to another page,
so that I'm only getting pagefaults and cache misses due to instruction
prefetch on code I really need to execute.

I've traced several of the critical paths in UNIX, and well over half of 
the machine code produced is for paranoid cases that hardly ever happen:

	if( cond )
		panic("cond");

Part of me wants to ifdef such code out, but naw, I wouldn't do that now...

A good trace scheduling compiler, with manual and automatic profiling feedback,
would be useful on a lot more than a VLIW machine.



Andy "Krazy" Glew. Gould CSD-Urbana.    1101 E. University, Urbana, IL 61801   
    aglew@gould.com     	- preferred, if you have nameserver
    aglew@gswd-vms.gould.com    - if you don't
    aglew@gswd-vms.arpa 	- if you use DoD hosttable
    aglew%mycroft@gswd-vms.arpa - domains are supposed to make things easier?
   
My opinions are my own, and are not the opinions of my employer, or any
other organisation. I indicate my company only so that the reader may
account for any possible bias I may have towards our products.