https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116685
--- Comment #3 from Edwin Lu <ewlu at rivosinc dot com> ---
(In reply to Andrew Pinski from comment #1)
> -fno-vect-cost-model fixes some of these.
I hadn't taken a look with -fno-vect-cost-model until now but it seems like
there's some weird codegen with the 3 element dot products
dotDbl_3d(GenericVector3d<double> const&, GenericVector3d<double> const&):
fld fa5,16(a0)
fld fa4,16(a1)
vsetivli zero,2,e64,m1,ta,ma
vfmv.v.f v1,fa5
vfmv.v.f v2,fa4 <-- why not vector load?
fld fa5,0(a0)
fld fa4,0(a1)
vmv.s.x v3,zero
vfslide1down.vf v1,v1,fa5
vfslide1down.vf v2,v2,fa4
fld fa0,8(a0)
fld fa4,8(a1)
vfmul.vv v1,v1,v2
vfredusum.vs v1,v1,v3
vfmv.f.s fa5,v1
fmadd.d fa0,fa0,fa4,fa5
ret
