Path: utzoo!utgpu!water!watmath!clyde!rutgers!mcnc!decvax!dartvax!eleazar!davidg
From: davidg@eleazar.Dartmouth.EDU (David Gelhar)
Newsgroups: comp.arch
Subject: Re: taken -vs- untaken branches
Message-ID: <7891@eleazar.Dartmouth.EDU>
Date: 9 Jan 88 04:36:06 GMT
References: <496@cresswell.quintus.UUCP> <638@l.cc.purdue.edu> <20339@amdahl.amdahl.com>
Reply-To: davidg@eleazar.Dartmouth.EDU (David Gelhar)
Organization: Dartmouth College, Hanover, NH
Lines: 27

In article <20339@amdahl.amdahl.com> chuck@amdahl.amdahl.com (Charles Simmons) writes:
>[...]to estimate the number of legal values that one side of the expression
>can take on, and the number of legal values that the other side of
>the expression can take on.  These two estimates can then be used
>to estimate how often the expression will be TRUE.
>For example, consider the fragment of code:
>    if (p == NULL) {
>	...
>    }
>Here, the compiler might think "'p' can take on 2**32 values,
>NULL can take on one value, so the estimated number of times
>these two expressions are equal is 1/2**32.  Since this number
>is less than 1/2, the condition will usually be false."


This code fragment strikes me as an excellent example of where this
heuristic would NOT produce useful results; I think it's pretty clear
that in practice the values of pointer variables are not evenly
distributed over the 2**32 possibilities.  In fact, my estimate of
the probability of an arbitrary pointer being NULL is more like 1/2
than 1/2**32.  (Hmmmm... I guess that doesn't help much in predicting
which way the branch will go, eh? :-)).




(Hi chuck.)