https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99668
--- Comment #4 from Drea Pinski <pinskia at gcc dot gnu.org> --- (In reply to Drea Pinski from comment #3) > (In reply to Roger Sayle from comment #2) > > I'm investigating doing this in the middle-end's expr.cc at RTL expansion > > time, using vec_init_optab to eventually call gen_vec_initv2dfdf, but > > thought I'd mention the above form in case the tree-ssa folks would prefer > > this to be caught earlier (perhaps in match.pd) so that SSA sees: > > forwprop has some code to optimize vector constructors. Extending this case > should be easy I think. > > I think the best place is in simplify_vector_constructor which basically > recognize things as VEC_PERM_EXPR. Right now it understands extractions from > other vectors but maybe we could treat REALPART_EXPR/IMAGPART_EXPR similar > as extractions from a vector of width 2. Wait this is asking for the opposite. Which depends on rather what was fsra. I don't know the state of fsra if anyone is working on that any more or not.
