netdev
[Top] [All Lists]

Re: MSEC_TO_JIFFIES is messed up...

To: Davide Libenzi <davidel@xxxxxxxxxxxxxxx>
Subject: Re: MSEC_TO_JIFFIES is messed up...
From: "J. Bruce Fields" <bfields@xxxxxxxxxxxx>
Date: Wed, 12 May 2004 17:39:13 -0400
Cc: Ingo Molnar <mingo@xxxxxxx>, Jeff Garzik <jgarzik@xxxxxxxxx>, Greg KH <greg@xxxxxxxxx>, Andrew Morton <akpm@xxxxxxxx>, Linux Kernel Mailing List <linux-kernel@xxxxxxxxxxxxxxx>, Netdev <netdev@xxxxxxxxxxx>
In-reply-to: <Pine.LNX.4.58.0405121400360.11950@bigblue.dev.mdolabs.com>
References: <20040512020700.6f6aa61f.akpm@osdl.org> <20040512181903.GG13421@kroah.com> <40A26FFA.4030701@pobox.com> <20040512193349.GA14936@elte.hu> <Pine.LNX.4.58.0405121247011.11950@bigblue.dev.mdolabs.com> <20040512200305.GA16078@elte.hu> <Pine.LNX.4.58.0405121400360.11950@bigblue.dev.mdolabs.com>
Sender: netdev-bounce@xxxxxxxxxxx
User-agent: Mutt/1.5.6i
On Wed, May 12, 2004 at 02:01:58PM -0700, Davide Libenzi wrote:
> On Wed, 12 May 2004, Ingo Molnar wrote:
> > the compiler cannot discard the multiplication and the division from the
> > following:
> > 
> >     x * 1000 / 1000
> > 
> > due to overflows.
> 
> $ cat foo.c
> 
> int foo(int i) {
> 
> 
>         return i * 1000 / 1000;
> }

If gcc really optimizes that to just the identity function, then surely
that's a gcc bug?  Multiplication is left-associative, so i * 1000 /
1000 = (i * 1000) / 1000, but (i * 1000) should be zero for any i
divisible by i^(sizeof(int) - 12).

It shouldn't be able to optimize out the 1000 here for exactly the same
reason it shouldn't be able to optimize out the shifts in, e.g.,

        i << 12 >> 12

--Bruce Fields

<Prev in Thread] Current Thread [Next in Thread>