https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116571
--- Comment #6 from Andrew Stubbs <ams at gcc dot gnu.org> ---
(In reply to Richard Biener from comment #5)
> (In reply to Thomas Schwinge from comment #4)
> > The GCN target FAILs that I originally had reported here:
> >
> > > [-PASS:-]{+FAIL:+} gcc.dg/vect/slp-11a.c scan-tree-dump-times vect
> > > "vectorizing stmts using SLP" [-0-]{+1+}
> >
> > > [-PASS:-]{+FAIL:+} gcc.dg/vect/slp-12a.c scan-tree-dump-times vect
> > > "vectorizing stmts using SLP" [-0-]{+1+}
> >
> > ... are back to PASS as of recently; should we close this PR?
>
> I'd say so.
>
> >
> > Andrew, anything to be done anyway, regarding the following?
> >
> > (In reply to Richard Biener from comment #3)
> > > Possibly for GCN the issue is the vect_strided8 which is implemented as
> > >
> > > foreach N {2 3 4 5 6 7 8} {
> > > eval [string map [list N $N] {
> > > # Return 1 if the target supports 2-vector interleaving
> > > proc check_effective_target_vect_stridedN { } {
> > > return [check_cached_effective_target_indexed vect_stridedN {
> > > if { (N & -N) == N
> > > && [check_effective_target_vect_interleave]
> > > && [check_effective_target_vect_extract_even_odd] } {
> > > return 1
> > > }
> > > if { ([istarget arm*-*-*]
> > > || [istarget aarch64*-*-*]) && N >= 2 && N <= 4 } {
> > > return 1
> > > }
> > > if { ([istarget riscv*-*-*]) && N >= 2 && N <= 8 } {
> > > return 1
> > > }
> > > if [check_effective_target_vect_fully_masked] {
> > > return 1
> > > }
> > >
> > > not sure if gcn really supports a load/store-lane with 8 elements.
>
> Note I might have misunderstood the vect_stridedN effective target, but
> the last line looks odd (x86 also can do fully masked loops with AVX512
> but definitely cannot do arbitrary interleaving schemes just because of
> that).
OK, but unless I'm missing something x86 is not one of the targets where
check_effective_target_vect_fully_masked is true.
> I'd remove that last line, it was added by you, Andrew, in
> r9-5484-g674931d2b7bd88 ...
I will switch it to test for GCN specifically.
As far as I know, GCN is fine with 8 element vectors and can load store them
from any arbitrary pattern you can generate.
