> -----Original Message-----
> From: Richard Biener <richard.guent...@gmail.com>
> Sent: 09 June 2020 14:24
> To: Alex Coplan <alex.cop...@arm.com>
> Cc: Richard Biener <rguent...@suse.de>; gcc-patches@gcc.gnu.org
> Subject: Re: [PATCH] move permutation validity check
> 
> On Tue, Jun 9, 2020 at 3:22 PM Alex Coplan <alex.cop...@arm.com> wrote:
> >
> > > -----Original Message-----
> > > From: Gcc-patches <gcc-patches-boun...@gcc.gnu.org> On Behalf Of Richard
> > > Biener
> > > Sent: 05 May 2020 14:49
> > > To: gcc-patches@gcc.gnu.org
> > > Subject: [PATCH] move permutation validity check
> > >
> > >
> > > This delays the SLP permutation check to vectorizable_load and optimizes
> > > permutations only after all SLP instances have been generated and the
> > > vectorization factor is determined.
> > >
> > > Bootstrap and regtest running on x86_64-unknown-linux-gnu.
> > >
> > > Richard.
> >
> > Hi Richard,
> >
> > This patch (bc484e25) introduced a test failure on arm and aarch64. The
> > test gcc.dg/vect/slp-perm-9.c is now failing with:
> >
> > FAIL: gcc.dg/vect/slp-perm-9.c scan-tree-dump-times vect "permutation
> requires at least three vectors" 1
> >
> > It seems this assertion isn't run on x86-64 which explains why this was
> > missed.
> 
> Please open a bugreport if there's none already.
> 
> Thanks,
> Richard.

Ah, it appears there already is one: 
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=95056

Apologies, I should have checked Bugzilla first.

Thanks,
Alex

> 
> >
> > Thanks,
> > Alex
> >
> > >
> > > 2020-05-05  Richard Biener  <rguent...@suse.de>
> > >
> > >       * tree-vectorizer.h (vec_info::slp_loads): New.
> > >       (vect_optimize_slp): Declare.
> > >       * tree-vect-slp.c (vect_attempt_slp_rearrange_stmts):  Do
> > >       nothing when there are no loads.
> > >       (vect_gather_slp_loads): Gather loads into a vector.
> > >       (vect_supported_load_permutation_p): Remove.
> > >       (vect_analyze_slp_instance): Do not verify permutation
> > >       validity here.
> > >       (vect_analyze_slp): Optimize permutations of reductions
> > >       after all SLP instances have been gathered and gather
> > >       all loads.
> > >       (vect_optimize_slp): New function split out from
> > >       vect_supported_load_permutation_p.  Elide some permutations.
> > >       (vect_slp_analyze_bb_1): Call vect_optimize_slp.
> > >       * tree-vect-loop.c (vect_analyze_loop_2): Likewise.
> > >       * tree-vect-stmts.c (vectorizable_load): Check whether
> > >       the load can be permuted.  When generating code assert we can.
> > >
> > >       * gcc.dg/vect/bb-slp-pr68892.c: Adjust for not supported
> > >       SLP permutations becoming builds from scalars.
> > >       * gcc.dg/vect/bb-slp-pr78205.c: Likewise.
> > >       * gcc.dg/vect/bb-slp-34.c: Likewise.
> > > ---
> > >  gcc/testsuite/gcc.dg/vect/bb-slp-34.c      |   3 +-
> > >  gcc/testsuite/gcc.dg/vect/bb-slp-pr68892.c |   7 +-
> > >  gcc/testsuite/gcc.dg/vect/bb-slp-pr78205.c |   6 +-
> > >  gcc/tree-vect-loop.c                       |   3 +
> > >  gcc/tree-vect-slp.c                        | 262 +++++++++++------------
> > > ------
> > >  gcc/tree-vect-stmts.c                      |  50 +++++-
> > >  gcc/tree-vectorizer.h                      |   4 +-
> > >  7 files changed, 154 insertions(+), 181 deletions(-)

Reply via email to