Re: accept 64-bit constant addresses in mips n32

[Eric Christopher <echristo at redhat dot com>]

On Tue, 2003-04-08 at 03:59, Alexandre Oliva wrote:
> On Apr  8, 2003, Thiemo Seufer <ica2_ts at csv dot ica dot uni-stuttgart dot de> wrote:
> 
> > Trditionally it does more than that. The purpose of .set mipsX is to
> > allow full exploitation of the architecture's capabilities. ABI-imposed
> > register width restrictions are the obvious example.
> 
> Fine.  One thing is to say `it's now ok to use the full 64 bits of
> this register for an integer'.  Saying `assume pointers can now be
> 64-bits wide' is pushing it a bit too far.  Consider non-literal
> values, that have to be loaded from a GOT.  Would we have to somehow
> generate 64-bit GOT entries if we find .mips3 in an n32 program?  I
> hope not :-)

Good lord no. :)

While I don't particularly like the idea of 64-bit constant addresses in
n32 (especially on the number of bogus things in the port) the savings
are fairly impressive which is why I allowed it. Alex: can you document
what's going on a bit more as well and then throw it in?

Thanks.

-eric

-- 
Eric Christopher <echristo at redhat dot com>