Xref: utzoo comp.arch:8459 comp.lang.misc:2696 comp.lang.c:16566
Path: utzoo!utgpu!utstat!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!cwjcc!ukma!gatech!bloom-beacon!mit-eddie!fenchurch!jbs
From: jbs@fenchurch.mit.edu (Jeff Siegal)
Newsgroups: comp.arch,comp.lang.misc,comp.lang.c
Subject: Re: Peephole optimisation
Message-ID: <11201@eddie.MIT.EDU>
Date: 27 Feb 89 19:16:20 GMT
References: <740@tetons.UUCP> <76700068@p.cs.uiuc.edu> <671@oracle.oracle.com> <730@microsoft.UUCP> <1153@l.cc.purdue.edu> <8650@aw.sei.cmu.edu>
Sender: uucp@eddie.MIT.EDU
Reply-To: jbs@fenchurch.UUCP (Jeff Siegal)
Organization: MIT, EE/CS Computer Facilities, Cambridge, MA
Lines: 19

>cik@l.cc.purdue.edu (Herman Rubin) writes:
>
>   [...]And how big is the peephole?

Every study I've seen claims that three instructions is very nearly
always sufficient.  The exception being very complex instructions
which might be recognized as a substitution for a long sequence of
simple instructions.  In reality, the long sequence of simple
instructions is very unlikely to *exactly* match the complex
instruction anyway, so the very long peephole would almost never be of
value.

For RISC machines, this is not likely to be an issue, and three
instructions would (probably) suffice.

Multiple passes can often be used to get the same effect as a longer
peephole (by using intermediate transformations).

Jeff Siegal