|
|
Richard Henderson wrote:
> On Thu, Jan 27, 2005 at 12:18:48AM +0100, Ulrich Weigand wrote:
> > Anyway, this is a pretty normal situation on s390; it arises
> > whenever the address of a string constant is loaded. The
> > string constant itself gets a constant pool label, but since
> > I cannot directly load addresses outside the per-function
> > literal pool, that address is forced into the constant pool again.
>
> Huh. I wouldn't have expected the string to get a constant pool
> label, since it's not in the constant pool.
Hmm. I may have been confused by both labels using the LC prefix.
You're right, the string label is only TREE_CONSTANT_POOL_ADDRESS_P,
not CONSTANT_POOL_ADDRESS_P.
So I don't know either why there would be a CONSTANT_POOL_ADDRESS_P
symbol in the constant pool. I've copied the check from the dwarf2
code where it was introduced by:
http://gcc.gnu.org/ml/gcc-patches/2002-02/msg01396.html
I guess it cannot harm either way ...
> Oh well, ok then.
Still OK, or should I remove the inner get_pool_constant_mark
test?
Bye,
Ulrich
--
Dr. Ulrich Weigand
Linux on zSeries Development
Ulrich.Weigand@xxxxxxxxxx
|
|