Re: [Linux-ia64] Unaligned access in ia64

From: <>
Date: 2001-11-13 04:16:02

Looks like you have a bigger problem than just unaligned access :-)

I think your basic problem is that for IA32 int's, long's and
pointers are all 4-bytes while for IA64 only int's are 4-bytes,
long's and pointer's are 8-bytes.

This means that the compiler warning about the cast to different
sizes is VERY significant and probably means your program is incorrect.
When you cast the pointer to the int you lost the upper half of the
pointer and, when you cast the int back to a pointer, you wound up with
a bad pointer.

The simple solution is to take all of those int variables that you
are putting pointers into and change them into longs.  Of course, if
you move any of those long variables into a different interrmediate
int variable you'll have to change the intermediate variable into a
long also.  And so on, and so on...

(As an aside I'd ask the more basic question of why are you casting
pointers to an integral type?  This is inherently unsafe and you should
avoid doing this if at all possible.  Why don't you just always treat
pointers as pointers and then avoid this whole problem?)

On Mon, Nov 12, 2001 at 08:56:10AM -0800, Pereira, D LailaX E wrote:
> Hi 
> My program still does not port properly gives the "unaligned access
> warnings" while running. Also, When I make the files, I also get the
> warnings of the following type::
> warning: cast from pointer to integer of different size
> This happens when I convert (void*) pointer to int i. 
> The following warning is also displayed: 
> cast to pointer from integer of different size 
> This happens when I convert int i to (void*) pointer . 
> for example::  function((void*)i);
> How can these warnings be avoided? 
> d'laila 
> -----Original Message-----
> From: []
> Sent: Monday, November 12, 2001 10:38 AM
> To: Pereira, D LailaX E
> Cc: ''
> Subject: Re: [Linux-ia64] Unaligned access in ia64
> D'Laila-
> I should first point out that those were warnings, not errors.  Your
> program should have executed properly, all be it a lot slower that if
> it didn't do the unaligned accesses.
> The simple answer is that all accesses should be to natively aligned
> boundaries.  That means 2-byte shorts should be on 2-byte boundaries,
> 4-byte ints should be on 4-byte boundaries and so on.
> Also, you should note that you are probably paying a performance
> penality on your Intelx86 machine also, you just don't know about
> it because the CPU doesn't even tell you when you make an unaligned
> access.
> On Mon, Nov 12, 2001 at 08:24:38AM -0800, Pereira, D LailaX E wrote:
> > Hi 
> > I have an application running fine on the Intelx86 machines. When I run
> the
> > same on itanium machine ( linux OS) , I get the following  errors:
> > stress.udp(16868):unaligned access  to 0x4000000000003541,
> > ip=0x400000000001cfe0
> > stress.udp(16868):unaligned access  to 0x4000000000003541,
> > ip=0x400000000001c551
> > 
> > What could be the possible causes for this? Can anyone please point out?
> > what are the general cautions to be taken while porting applications to
> > itanium? for eg any pointer conversions , int conversions etc 
> > 
> > thanks
> > d'laila 
> > 
> > 
> > _______________________________________________
> > Linux-IA64 mailing list
> >
> >
> -- 
> Don Dugger
> "Censeo Toto nos in Kansa esse decisse." - D. Gale
> Ph: 303/652-0870x117
> _______________________________________________
> Linux-IA64 mailing list

Don Dugger
"Censeo Toto nos in Kansa esse decisse." - D. Gale
Ph: 303/652-0870x117
Received on Mon Nov 12 09:16:09 2001

This archive was generated by hypermail 2.1.8 : 2005-08-02 09:20:05 EST