Path: utzoo!mnetor!uunet!husc6!hao!ames!ll-xn!mit-eddie!uw-beaver!uw-june!uw-entropy!dataio!bright
From: bright@Data-IO.COM (Walter Bright)
Newsgroups: comp.lang.c
Subject: noalias stuff
Message-ID: <1457@dataio.Data-IO.COM>
Date: 11 Jan 88 19:00:34 GMT
References: <11189@brl-adm.ARPA> <517@cresswell.quintus.UUCP>
Reply-To: bright@dataio.UUCP (Walter Bright)
Organization: Data I/O Corporation; Redmond, WA
Lines: 46

In article <517@cresswell.quintus.UUCP> ok@quintus.UUCP (Richard A. O'Keefe) writes:
<In article <11189@brl-adm.ARPA>, TLIMONCE%DREW.BITNET@CUNYVM.CUNY.EDU writes:
<< noalias int i;
<< for (i=0; i<100; i++) {...anything...}
<< the default is the "safe" method.
<< If the compiler notices that I don't modify i in the loop
<	  ***********************
<If the compiler is smart enough to notice that the loop body does not
<change 'i', why do you need to tell it 'noalias'?  A *local* variable
<whose address is never taken cannot possibly be aliased, because there
<is no other *legal* way to get at it.
<Just how much of a problem do we really have here, anyway?
<The problem of aliasing is a problem of having multiple legal access
<paths to the same variable.  How often is the controlled variable of
<a loop global?  How often does it have its address taken?

I have only run across two general cases where people have used globals for
loop indices:

	1. The programmer was new to C.
	2. So the variable could be accessed by a debugger.

Neither of those cases warrants adding a new keyword.

The primary use of noalias is to soften the following worst cases:
	if (an assignment to a global or static occurs ||
	    an assignment through a pointer occurs ||
	    a function is called for which we have no information)
	{	Mark all global and static variables as possibly modified;
		Mark all indirect references as possibly modified;
	}

Marking an expression as being possibly modified means that if its value is
in a register, it must be reloaded. Also, it prevents common subexpressions
from being formed across the modification point.

Since C is a 'pointy' language, in order to improve code generation, a
real wad of 'noalias' casts would have to be used. I suspect people won't
use it because its ugly.

Does anyone have any statistics on the code generation improvement if
'noalias' is used to the max? I suspect it may not be worth the effort.
My vote is to dump the noalias keyword.

P.S. I have written a globally optimizing compiler (Datalight C) that
does full flow analysis, and uses the algorithm above.