Re: optimization/8049: SSE unaligned vector stores crash with -O0

[Jan Hubicka <jh at suse dot cz>]

The following reply was made to PR optimization/8049; it has been noted by GNATS.

From: Jan Hubicka <jh@suse.cz>
To: Ian Ollmann <iano@cco.caltech.edu>
Cc: Jan Hubicka <jh@suse.cz>, hubicka@gcc.gnu.org,
	gcc-bugs@gcc.gnu.org, gcc-prs@gcc.gnu.org, gcc-gnats@gcc.gnu.org
Subject: Re: optimization/8049: SSE unaligned vector stores crash with -O0
Date: Thu, 10 Oct 2002 20:03:17 +0200

 > On Thu, 10 Oct 2002, Jan Hubicka wrote:
 > 
 > > > On 10 Oct 2002 hubicka@gcc.gnu.org wrote:
 > > >
 > > > > Synopsis: SSE unaligned vector stores crash with -O0
 > > > >
 > > > > State-Changed-From-To: open->closed
 > > > > State-Changed-By: hubicka
 > > > > State-Changed-When: Thu Oct 10 09:45:59 2002
 > > > > State-Changed-Why:
 > > > >     It is runtime bug to not align stack properly for main.
 > > > >     It will go away with runtime compiled using gcc 3.2 or can be workarounded by avoiding vector stuff in main.
 > > > >
 > > > > http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=8049
 > > >
 > > > I see this in functions that are not main(). Didn't I provide an example
 > > > of one?
 > >
 > > I didn't see it.  Can you send me some?
 > 
 > The original example is one.  The three buffers passed into MatrixMultiply
 > happen to be aligned on my system. The actual crash happens when the stack
 > copy of C1 is loaded, before the _mm_store_ps (uninlined) function is
 > called.
 
 This looks strange.  All stores to C1 works properly and then you get
 movaps crash when you load it to store into output?
 I believe it is the output that is missaligned (destination of store)
 because the destination array is missaligned from the caller.
 
 Honza
 > 
 > Ian
 > 
 > ---------------------------------------------------
 >    Ian Ollmann, Ph.D.       iano@cco.caltech.edu
 > ---------------------------------------------------