https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123108

--- Comment #10 from Jeffrey A. Law <law at gcc dot gnu.org> ---
I think we've got a post RA split pass, but it's actions are somewhat limited
as it can't create new pseudos.

What I was thinking about as a fallback was to have curr_insn_transform treat
the case where there are no constraints and the predicates fail as having a
single "r" constraint.  I think that would work around the current problem, but
would likely cause more in the long term the more I think about it.

I tend to think we just need to either bite the bullet and add the constraints
(likely in the next stage1 cycle) or really introduce a way to do a more
gradual lowering, primarily for vector, but we'd likely use the same mechanism
for things like mvconst_internal.

Reply via email to