NetNews Usenet Archive 1993 #1/NN_1993_1.iso/spool/comp/unix/aix/13214

Internet Message Format | 1993-01-12 | 1.6 KB

Path: sparky!uunet!mcsun!sun4nl!orcenl!dk.oracle.com!bengsig
From: [e-mail] (Bjorn Engsig)
Newsgroups: comp.unix.aix
Subject: Re: optimisation and volatile
Message-ID: <[e-mail]>
Date: 12 Jan 93 10:52:41 GMT
References: <[e-mail]>
Sender: [e-mail] (News Administrator)
Reply-To: [e-mail] (Bjorn Engsig)
Organization: Oracle Denmark
Lines: 22
Nntp-Posting-Host: dkaix1.dk.oracle.com

Article <[e-mail]> by [e-mail] (phil) says:
|would like to compile the code with -O optimization. The last time we tried
|this the compiler would remove repeated shared memory reads with no intervening
|writes,
How should the compiler know that it needs to reload the data?  We've had
examples of the opposite: data are loaded even though they weren't needed,
and if the data was in unallocated shared memory, you'd get a segmentation
violation.  E.g.:  if(a) b++; may lay load b even if a is zero.  Declaring
b volatile avoids this [the actual code was more complicated than this].

As you suggest yourself:
|gcc has a switch `-fvolatile' : Consider all memory references through pointers
|to be volatile.
But why would you force all references to be considered volatile? The only
resonable solution is to declare your volatile variables as volatile.
|Is there an equivalent for xlc?
Not that I know of.
--
Bjorn Engsig,           Internet: [e-mail]   uucp: uunet!oracle!bengsig
ORACLE Corporation.    Private : [e-mail].

Every program uses a constant amount of memory, if memory is counted in dollars.