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]
Other format: [Raw text]

Re: not too big an alignment


On Tue, Nov 12, 2013 at 01:42:00PM -0800, Mike Stump wrote:
> On Nov 12, 2013, at 1:16 PM, Jakub Jelinek <jakub@redhat.com> wrote:
> > On Tue, Nov 12, 2013 at 01:11:04PM -0800, Mike Stump wrote:
> >> Alignments are stored in a byte, large alignments don't actually work nicely.  This caps the alignment to 128, as most ports would define BIGGEST_ALIGNMENT to be smaller than this.  The competing change would to be to make it a short, but, I'd be happy to punt that until such time as someone actually needs that.
> >> 
> >> Ports break down this way currently:
> >> 
> >>  12 #define BIGGEST_ALIGNMENT 64
> >>  10 #define BIGGEST_ALIGNMENT 32
> >>   6 #define BIGGEST_ALIGNMENT 128
> >>   3 #define BIGGEST_ALIGNMENT 8
> >>   8 #define BIGGEST_ALIGNMENT 16
> > 
> > You are missing i386 that has BIGGEST_ALIGNMENT 512 (or less, depending on
> > compiler options).  So this doesn't look right.
> 
> And yet alignments for modes with sizes like 256 won't work and i386 has no mode with alignment bigger than 128 in this table.

Well, BIGGEST_ALIGNMENT is in bits, while mode_base_align seems to be in bytes it
seems:

unsigned int
get_mode_alignment (enum machine_mode mode)
{
  return MIN (BIGGEST_ALIGNMENT, MAX (1, mode_base_align[mode]*BITS_PER_UNIT));
}

So, supposedly it works up to 1024 bit alignment right now.

	Jakub


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