https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116142
--- Comment #8 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Xi Ruoyao from comment #7)
> (In reply to Richard Biener from comment #6)
>
> > Can you check if the following makes things work for you?
> >
> > diff --git a/gcc/tree-vect-stmts.cc b/gcc/tree-vect-stmts.cc
> > index 67f6e5df255..7496e31164c 100644
> > --- a/gcc/tree-vect-stmts.cc
> > +++ b/gcc/tree-vect-stmts.cc
> > @@ -14200,7 +14200,6 @@ supportable_widening_operation (vec_info *vinfo,
> > are properly set up for the caller. If we fail, we'll continue
> > with
> > a VEC_WIDEN_MULT_LO/HI_EXPR check. */
> > if (vect_loop
> > - && STMT_VINFO_RELEVANT (stmt_info) == vect_used_by_reduction
> > && !nested_in_vect_loop_p (vect_loop, stmt_info)
> > && supportable_widening_operation (vinfo,
> > VEC_WIDEN_MULT_EVEN_EXPR,
> > stmt_info, vectype_out,
>
> Still not. I get something like
>
> t1.c:5:20: note: ==> examining pattern statement: patt_24 = _1 w* _3;
> t1.c:5:20: note: vect_is_simple_use: operand x[i_15], type of def: internal
> t1.c:5:20: note: vect_is_simple_use: vectype vector(8) short int
> t1.c:5:20: note: vect_is_simple_use: operand y[i_15], type of def: internal
> t1.c:5:20: note: vect_is_simple_use: vectype vector(8) short int
> t1.c:5:20: missed: conversion not supported by target.
> t1.c:5:20: note: vect_is_simple_use: operand x[i_15], type of def: internal
> t1.c:5:20: note: vect_is_simple_use: vectype vector(8) short int
> t1.c:3:5: missed: not vectorized: relevant stmt not supported: patt_24 =
> _1 w* _3;
> t1.c:5:20: missed: bad operation or unsupported loop bound.
> t1.c:5:20: note: ***** Analysis failed with vector mode V8HI
>
> As an experiment I renamed the {even,odd} expanders to {lo,hi} (doing so is
> definitely wrong in reality) and I got the "expected" result.
You have to debug why it's still not happy with this.