This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: not so trivial powerpc regression fix


> On Fri, May 18, 2001 at 11:41:43PM +0200, Jan Hubicka wrote:
> > +       /* Avoid memories with nonzero SUBREG_BYTE, as offsetting the memory
> > +          may result in incorrect reference.  We should simplify all valid
> > +          subregs of MEM anyway.  */
> > +       if (SUBREG_BYTE (op) && GET_CODE (SUBREG_REG (op)) == MEM)
> > +         return 0;
> 
> This is incorrect.  The correct way to tell if we're offsettting
> the memory is to check subreg_lowpart_p.
Really? On little endian machine, lowpart lies on offsetted memory.

As I understand post SUBREG_BYTE situation, the SUBREG_BYTE means offset
as if value were stored in memory in the target format.

Honza
> 
> 
> r~


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]