https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68776
--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> --- (In reply to rguent...@suse.de from comment #6) > On December 17, 2015 4:19:00 PM GMT+01:00, "wschmidt at gcc dot gnu.org" > <gcc-bugzi...@gcc.gnu.org> wrote: > >https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68776 > > > >--- Comment #5 from Bill Schmidt <wschmidt at gcc dot gnu.org> --- > >(In reply to Bill Schmidt from comment #4) > >> Yep. I'll verify the fix and commit today if all goes well. Thanks > >for the > >> investigation! > > > >Actually, looking at check_effective_target_vect_int_mult, this won't > >work (and > >I've confirmed it doesn't): > > > > if { ([istarget powerpc*-*-*] && ![istarget powerpc-*-linux*paired*]) > > || [istarget spu-*-*] > > || [istarget i?86-*-*] || [istarget x86_64-*-*] > > || [istarget ia64-*-*] > > || [istarget aarch64*-*-*] > > || [check_effective_target_arm32] } { > > set et_vect_int_mult_saved 1 > > > >So the proposal will still expect powerpc64le to find 2 instances > >instead of 4. > > > >I've verified that we are generating the expected widening > >multiplication > >sequences for foo and bar, and generating 32-bit multiplies for baz. > > But if ppc can do int mult the mult pattern should not apply... Specifically: static gimple * vect_recog_mult_pattern (vec<gimple *> *stmts, tree *type_in, tree *type_out) { ... vectype = get_vectype_for_scalar_type (itype); if (vectype == NULL_TREE) return NULL; /* If the target can handle vectorized multiplication natively, don't attempt to optimize this. */ optab = optab_for_tree_code (MULT_EXPR, vectype, optab_default); if (optab != unknown_optab) { machine_mode vec_mode = TYPE_MODE (vectype); int icode = (int) optab_handler (optab, vec_mode); if (icode != CODE_FOR_nothing) return NULL; so powerpc can _not_ do integer multiplication (V2DI in this case). Ok, so if you take "int" literally it can do vect_int_mult and we don't have a vect_long_mult target. I would say it would help to make the "pattern recognized" printing more specific but that requires some extra work. Let me try sth simple...
