https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118243
--- Comment #11 from Martin Jambor <jamborm at gcc dot gnu.org> --- (In reply to Richard Biener from comment #9) > Indeed we don't seem to split a vector in the same way: > We do but the constant IPA-CP is different, is it is the entire vector in one constant, as opposed to separate constants for the real and the imaginary part which we have in the complex case. That means that the types of the accesses for IPA-SRA access matches nicely and there are no issues. That, together with the fact that complex numbers are scalar types, is the problem here. (It seems that creating a vector testcase is not possible however, as those are represented by BIT_FIELD_EXPR and IPA-CP does not gather constants from those.)
