Path: utzoo!mnetor!uunet!seismo!sundc!pitstop!sun!amdcad!ames!hao!gatech!bloom-beacon!mit-eddie!uw-beaver!cornell!rochester!PT.CS.CMU.EDU!sei!sei.cmu.edu!firth
From: firth@sei.cmu.edu (Robert Firth)
Newsgroups: comp.arch
Subject: Re: conditional branches
Message-ID: <4252@aw.sei.cmu.edu>
Date: 18 Feb 88 14:36:21 GMT
References: <191@telesoft.UUCP> <1556@gumby.mips.COM> <375@imagine.PAWL.RPI.EDU> <1610@gumby.mips.COM>
Sender: netnews@sei.cmu.edu
Reply-To: firth@bd.sei.cmu.edu.UUCP (Robert Firth)
Organization: Carnegie-Mellon University, SEI, Pgh, Pa
Lines: 28

Some of our work here confirms the view that condition codes are
not a very good idea.  Compiling the same code for MIPS M/500
and Vax-11, using the same compiler technology, gave the result

MIPS: 6.1% of instructions were conditional branches
      1.2% were 'set condition' instructions

VAX: 10.0% conditional branches
      8.9% tests or comparisons

MIPS provides only comparison against zero in the conditional
branch  This took care of about 80% of branches; the other 20%
required a prior set condition instruction.

Vax branches against the condition codes, but loses because
nearly all the branches then required a prior instruction to
set them.  Here is an extract from our report, discussing the
Vax statistics:

	"The code generator slaves the condition codes
	 religiously, both through linear code and across
	 control transfers.  Nevertheless, of 415 conditional
	 branches, 371 (almost 90%) required a prior test
	 or compare to set the condition codes; of 2169
	 normal instructions that set the condition codes,
	 only 44 (about 2%) did so to any purpose.  It is
	 hard to avoid the conclusion that condition codes
	 are a waste of time, effort, and silicon."