This is the mail archive of the
mailing list for the GCC project.
Re: [PATCH] Improve Gimplification of return for vectors and complex variables
On 12/26/06, Richard Guenther <firstname.lastname@example.org> wrote:
This is ok. Can you produce a followup to make the test for COMPLEX
or VECTOR type
a predicate GIMPLE_REG_TYPE_P? (or rather GIMPLE_REG_TYPE_POSSIBLE_P -
but that's so long...)
I was going to change GIMPLE_REG_TYPE_P to be not only on complex and
vector types in the near future as it is needed for some statements
which have VIEW_CONVERT_EXPR on the left hand side like:
(VIEW_CONVERT_EXPR<struct b>(a)).c =1;
Which is a partial kill (subreg type of a kill) of a.
After I fix that issue, I also need to also go through the IR and try
to set GIMPLE_REG_TYPE_P on some decls, otherwise we will cause
TREE_ADDRESSABLE not to be set but still get non gimple register.