Re: [r17-1485 Regression] FAIL: g++.dg/vect/slp-pr98855.cc -std=c++26 scan-tree-dump slp1 "not vectorized: vectorization is not profitable" on Linux/x86_64

Richard Biener via Gcc-regression <[email protected]> Mon, 15 Jun 2026 09:17:54 +0200
Newsgroups gmane.comp.gcc.regression,gmane.comp.gcc.patches
Message-ID <CAFiYyc2gq7Toj2qU7Rg7nmAGu6puc_vqfSOhOEBc=Ec3Y85dig@mail.gmail.com>
On Sat, Jun 13, 2026 at 5:00 AM Zhongyao Chen
<[email protected]> wrote:
>
> On Fri, Jun 12, 2026 at 2:49 PM Haochen Jiang <[email protected]> wrote:
> >
> > On Linux/x86_64,
> >
> > 9f8409f2e2c28f64bf6a584bc81afcae0f084785 is the first bad commit
> > commit 9f8409f2e2c28f64bf6a584bc81afcae0f084785
> > Author: Zhongyao Chen <[email protected]>
> > Date:   Wed Jun 3 20:44:59 2026 +0800
> >
> >     vect: Avoid external fallback while operand swap retry is possible
> >
> > caused
> >
> > FAIL: g++.dg/vect/slp-pr98855.cc  -std=c++26  scan-tree-dump-not slp1 "vectorizing stmts using SLP"
> > FAIL: g++.dg/vect/slp-pr98855.cc  -std=c++26  scan-tree-dump slp1 "not vectorized: vectorization is not profitable"
> >
> > with GCC configured with
> >
> > ../../gcc/configure --prefix=/export/users3/haochenj/src/gcc-bisect/master/master/r17-1485/usr --enable-clocale=gnu --with-system-zlib --with-demangler-in-ld --with-fpmath=sse --enable-languages=c,c++,fortran --enable-cet --without-isl --enable-libmpx x86_64-linux --disable-bootstrap
> >
> > To reproduce:
> >
> > $ cd {build_dir}/gcc && make check RUNTESTFLAGS="vect.exp=g++.dg/vect/slp-pr98855.cc --target_board='unix{-m64\ -march=cascadelake}'"
> >
> > (Please directly reply to this email for question about this report.)
> > (If you met problems with cascadelake related, disabling AVX512F in command line might save that.)
> > (However, please make sure that there is no potential problems with AVX512.)
>
> I took a quick look at PR98855. The current FAIL does not look like
> the original issue there.
> The inner loop is still scalar, and it seems we only vectorize the
> final store now.
>
> So I suspect this is more about cost model, the estimated costs are close.
> And I'm not sure which is really better, but from the asm, the scalar
> version looks better, since it uses fewer instructions.

One option is to add -mtune=generic to the testcase, since that
reflects what it was built
for.

>
> --
> Regards,
> Zhongyao