Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!wuarchive!udel!rochester!pt.cs.cmu.edu!gandalf.cs.cmu.edu!lindsay
From: lindsay@gandalf.cs.cmu.edu (Donald Lindsay)
Newsgroups: comp.arch
Subject: Re: Compilers and efficiency
Message-ID: <12784@pt.cs.cmu.edu>
Date: 25 Apr 91 03:15:20 GMT
References: <1991Apr23.162814.20093@Think.COM>
Organization: Carnegie Mellon
Lines: 24

In article <1991Apr23.162814.20093@Think.COM> barmar@think.com 
	(Barry Margolin) writes:
| ... The reason it was easy was because the machine code for a
| procedure prologue, contained (get this) a *call* to the runtime
| package...

>...alternate versions of the operators could be
>dynamically linked into the process (similar to the way personal computer
>customizations patch trap vectors).  The most notable uses of this are the
>procedure call tracing mechanism and the metering facility.

Another debugging feature available via this trick is break-when-
expression-true.  The expression is of course only evaluated on every
procedure entry, not on every cycle. However, my personal experience
(on a DEC 20) was very postive. I was usually trying to find out
which routine was using a wild pointer, so per-call was the perfect
granularity.

Of course, there are other ways to achieve the same effect on vanilla
hardware. Aside from page protection cleverness, one can run a snoop
process on one of the other CPUs. (Multiprocessors *are* vanilla,
right?)
-- 
Don		D.C.Lindsay 	Carnegie Mellon Robotics Institute