https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117007
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Target| |powerpc
Keywords| |missed-optimization
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
It's likely those constants are now exposed to GIMPLE somehow so we expand
a constant vector CTOR for which the backend should implement a vec_init
optab, recognizing constants it can create efficiently (I think we try
to expand uniform vector constants with splats already).
So you need to analyze why RTL expansion ends up with a constant pool load
initially.
