https://gcc.gnu.org/bugzilla/show_bug.cgi?id=119091

--- Comment #5 from Jeffrey A. Law <law at gcc dot gnu.org> ---
I do think that removal of mvconst_internal is the right high level direction
and will help this space.  But blindly removing it will create a host of new
problems (poor code generation), particularly for combine due to its internal
limitations.

I've got a patch series that goes in that direction, but it's a rats nest of
inter-related problems.  I think that reducing our reliance on other
define_insn_and_split patterns will be necessary before we can start to really
make progress on removal of mvconst_internal.

Given the size and scope of that problem it may make sense to have a shared
branch for that effort if others are going to help.

Reply via email to