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: SSE fix 12 - argument alignment PR pending/8212


> On Sun, Oct 20, 2002 at 12:23:54AM +0200, Jan Hubicka wrote:
> > OK. Do you see different choice than walking recursivly the types and
> > looking for vector modes?
> 
> Worst case, no.
> 
> In the normal case, if TYPE_USER_ALIGN is _not_ set, then we
> can just use the alignment given without searching.
> 
> But if TYPE_USER_ALIGN _is_ set, then we have to search to 
> see if there are vector mode types that _havn't_ been set to
> have user alignment less than normal.

Hmm, even better...
OK, I can see it as an excersise in trees :)
So when vector mode is present, I will have alignment 128, ptherwise
I will have alignment 32 even when -malign-double double is present?
(that seems to be the case, but I don't quite understand why).

If I have type user align, I search for vector modes with no overwriten
alignment and align to 128 ifif I find any?

Honza
> 
> 
> 
> r~


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