On Fri, Sep 01, 2000 at 04:01:43PM +0300, Matti Aarnio wrote:
> On Fri, Sep 01, 2000 at 02:44:04PM +0200, Andi Kleen wrote:
> > > To my knowlege it's only been speed related issues, not
> > > correctness issues, that have been the cause for the
> > > fear and loathing of long long.
> >
> > There are several parts of XFS which do not compile correctly with gcc
> > 2.95.2, but do with egcs 1.1
> >
> > For example consider the appended test case. It breaks with 2.95-stable
> > (from CVS March or so) because the bh->b_blocknr >>= block_bits;
> > shift is miscompiled. The problem seems to be that it forgets to reload
> > the %cl register used for the variable shift after the long long shift
> > before, leading to a bogus shift.
>
> And mind you, that shift operates on 'unsigned long' !
> ( Not 'unsigned long long' )
The shift on pbm_offset operates on long long.
The previous analysis was not quite right though (%cl is actually loaded,
just %eax gets bogus input from the long long shift)
>
> So far there has been only talk of doing the change. Brr...
>
> > Another problem is that linus' do_div in asm-i386/div64.h seems to cause
> > miscompiled code when used in anything more complicated than printk (no
> > extracted test case yet sorry)
>
> It really isn't intended for anything else.
> It is rather terrible side-effect monster..
So what do you propose to use when a long long division is needed (after
much thought and considering all alternatives etc.etc.) ?
-Andi
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Thu Sep 07 2000 - 21:00:11 EST