public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: Richard Biener <richard.guenther@gmail.com>
To: Robin Dapp <rdapp.gcc@gmail.com>
Cc: Richard Sandiford <richard.sandiford@arm.com>,
	gcc-patches <gcc-patches@gcc.gnu.org>,
	 Tamar Christina <Tamar.Christina@arm.com>
Subject: Re: [PATCH] vect: Use statement vectype for conditional mask.
Date: Mon, 20 Nov 2023 09:03:09 +0100	[thread overview]
Message-ID: <CAFiYyc0G4aOEdrEh7i18VO5ktWW-fNF3_G0CpjWS+KXZ9P6ioQ@mail.gmail.com> (raw)
In-Reply-To: <d8c540aa-1d15-454f-8291-9010d78ffa5a@gmail.com>

On Fri, Nov 17, 2023 at 8:20 PM Robin Dapp <rdapp.gcc@gmail.com> wrote:
>
> > No, you shouldn't place _7 != 0 inside the .COND_ADD but instead
> > have an extra pattern stmt producing that so
> >
> > patt_8 = _7 != 0;
> > patt_9 = .COND_ADD (patt_8, ...);
> >
> > that's probably still not enough, but I always quickly forget how
> > bool patterns work ... basically a comparison like patt_8 = _7 != 0
> > vectorizes to a mask (aka vector boolean) while any "data" uses
> > of bools are replaced by mask ? 1 : 0; - there's a complication for
> > bool data producing loads which is why we need to insert the
> > "fake" compares to produce a mask.  IIRC.
>
> I already had call handling to vect_recog_bool_pattern in working
> shape when I realized that vect_recog_mask_conversion_pattern already
> handles most of what I need.  The difference is that it doesn't do
>  patt_8 = _7 != 0
> but rather
>  patt_8 =  (<signed-boolean:1>) _7;
>
> It works equally well and most of the code can be reused.
>
> The attached was bootstrapped and regtested on x86 and aarch64
> and regtested on riscv.

OK.

Thanks,
Richard.

> Regards
>  Robin
>
> Subject: [PATCH] vect: Add bool pattern handling for COND_OPs.
>
> In order to handle masks properly for conditional operations this patch
> teaches vect_recog_mask_conversion_pattern to also handle conditional
> operations.  Now we convert e.g.
>
>  _mask = *_6;
>  _ifc123 = COND_OP (_mask, ...);
>
> into
>  _mask = *_6;
>  patt200 = (<signed-boolean:1>) _mask;
>  patt201 = COND_OP (patt200, ...);
>
> This way the mask will be properly recognized as boolean mask and the
> correct vector mask will be generated.
>
> gcc/ChangeLog:
>
>         PR middle-end/112406
>
>         * tree-vect-patterns.cc (build_mask_conversion):
>         (vect_convert_mask_for_vectype):
>
> gcc/testsuite/ChangeLog:
>
>         * gfortran.dg/pr112406.f90: New test.
> ---
>  gcc/testsuite/gfortran.dg/pr112406.f90 | 21 +++++++++++++++++++++
>  gcc/tree-vect-patterns.cc              | 26 ++++++++++++++++++--------
>  2 files changed, 39 insertions(+), 8 deletions(-)
>  create mode 100644 gcc/testsuite/gfortran.dg/pr112406.f90
>
> diff --git a/gcc/testsuite/gfortran.dg/pr112406.f90 b/gcc/testsuite/gfortran.dg/pr112406.f90
> new file mode 100644
> index 00000000000..27e96df7e26
> --- /dev/null
> +++ b/gcc/testsuite/gfortran.dg/pr112406.f90
> @@ -0,0 +1,21 @@
> +! { dg-do compile { target { aarch64-*-* || riscv*-*-* } } }
> +! { dg-options "-Ofast -w -fprofile-generate" }
> +! { dg-additional-options "-march=rv64gcv -mabi=lp64d" { target riscv*-*-* } }
> +! { dg-additional-options "-march=armv8-a+sve" { target aarch64-*-* } }
> +
> +module brute_force
> +  integer, parameter :: r=9
> +   integer sudoku1(1, r)
> +  contains
> +subroutine brute
> +integer l(r), u(r)
> +   where(sudoku1(1, :) /= 1)
> +        l = 1
> +      u = 1
> +   end where
> +do i1 = 1, u(1)
> +   do
> +      end do
> +   end do
> +end
> +end
> diff --git a/gcc/tree-vect-patterns.cc b/gcc/tree-vect-patterns.cc
> index 7debe7f0731..696b70b76a8 100644
> --- a/gcc/tree-vect-patterns.cc
> +++ b/gcc/tree-vect-patterns.cc
> @@ -5830,7 +5830,8 @@ vect_recog_mask_conversion_pattern (vec_info *vinfo,
>    tree rhs1_op0 = NULL_TREE, rhs1_op1 = NULL_TREE;
>    tree rhs1_op0_type = NULL_TREE, rhs1_op1_type = NULL_TREE;
>
> -  /* Check for MASK_LOAD ans MASK_STORE calls requiring mask conversion.  */
> +  /* Check for MASK_LOAD and MASK_STORE as well as COND_OP calls requiring mask
> +     conversion.  */
>    if (is_gimple_call (last_stmt)
>        && gimple_call_internal_p (last_stmt))
>      {
> @@ -5842,6 +5843,7 @@ vect_recog_mask_conversion_pattern (vec_info *vinfo,
>         return NULL;
>
>        bool store_p = internal_store_fn_p (ifn);
> +      bool load_p = internal_store_fn_p (ifn);
>        if (store_p)
>         {
>           int rhs_index = internal_fn_stored_value_index (ifn);
> @@ -5856,15 +5858,21 @@ vect_recog_mask_conversion_pattern (vec_info *vinfo,
>           vectype1 = get_vectype_for_scalar_type (vinfo, TREE_TYPE (lhs));
>         }
>
> +      if (!vectype1)
> +       return NULL;
> +
>        tree mask_arg = gimple_call_arg (last_stmt, mask_argno);
>        tree mask_arg_type = integer_type_for_mask (mask_arg, vinfo);
> -      if (!mask_arg_type)
> -       return NULL;
> -      vectype2 = get_mask_type_for_scalar_type (vinfo, mask_arg_type);
> +      if (mask_arg_type)
> +       {
> +         vectype2 = get_mask_type_for_scalar_type (vinfo, mask_arg_type);
>
> -      if (!vectype1 || !vectype2
> -         || known_eq (TYPE_VECTOR_SUBPARTS (vectype1),
> -                      TYPE_VECTOR_SUBPARTS (vectype2)))
> +         if (!vectype2
> +             || known_eq (TYPE_VECTOR_SUBPARTS (vectype1),
> +                          TYPE_VECTOR_SUBPARTS (vectype2)))
> +           return NULL;
> +       }
> +      else if (store_p || load_p)
>         return NULL;
>
>        tmp = build_mask_conversion (vinfo, mask_arg, vectype1, stmt_vinfo);
> @@ -5883,7 +5891,9 @@ vect_recog_mask_conversion_pattern (vec_info *vinfo,
>           lhs = vect_recog_temp_ssa_var (TREE_TYPE (lhs), NULL);
>           gimple_call_set_lhs (pattern_stmt, lhs);
>         }
> -      gimple_call_set_nothrow (pattern_stmt, true);
> +
> +      if (load_p || store_p)
> +       gimple_call_set_nothrow (pattern_stmt, true);
>
>        pattern_stmt_info = vinfo->add_stmt (pattern_stmt);
>        if (STMT_VINFO_DATA_REF (stmt_vinfo))
> --
> 2.41.0
>

      reply	other threads:[~2023-11-20  8:06 UTC|newest]

Thread overview: 12+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-11-08 16:18 Robin Dapp
2023-11-10  9:33 ` Richard Biener
2023-11-16 22:30   ` Robin Dapp
2023-11-17  8:24     ` Richard Biener
2023-11-17  8:45       ` Robin Dapp
2023-11-17  8:47         ` Richard Biener
2023-11-17 13:04           ` Robin Dapp
2023-12-03 18:32             ` [PATCH] testsuite: Fix up gcc.target/aarch64/pr112406.c for modern C [PR112406] Jakub Jelinek
2023-12-03 18:32               ` Jakub Jelinek
2023-12-03 18:59               ` Richard Biener
2023-11-17 19:20       ` [PATCH] vect: Use statement vectype for conditional mask Robin Dapp
2023-11-20  8:03         ` Richard Biener [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=CAFiYyc0G4aOEdrEh7i18VO5ktWW-fNF3_G0CpjWS+KXZ9P6ioQ@mail.gmail.com \
    --to=richard.guenther@gmail.com \
    --cc=Tamar.Christina@arm.com \
    --cc=gcc-patches@gcc.gnu.org \
    --cc=rdapp.gcc@gmail.com \
    --cc=richard.sandiford@arm.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).