public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: Richard Biener <richard.guenther@gmail.com>
To: Oluwatamilore Adebayo <Oluwatamilore.Adebayo@arm.com>,
	 "gcc-patches@gcc.gnu.org" <gcc-patches@gcc.gnu.org>,
	 "richard.guenther@gmail.com" <richard.guenther@gmail.com>,
	richard.sandiford@arm.com
Subject: Re: [PATCH] vect: Missed opportunity to use [SU]ABD
Date: Wed, 10 May 2023 11:49:16 +0200	[thread overview]
Message-ID: <CAFiYyc0L1pmHebsppQM9_c2yH7s48WLhH-dP+_rVXv2hyNRb5A@mail.gmail.com> (raw)
In-Reply-To: <mpt3544o90f.fsf@arm.com>

On Wed, May 10, 2023 at 11:01 AM Richard Sandiford
<richard.sandiford@arm.com> wrote:
>
> Oluwatamilore Adebayo <Oluwatamilore.Adebayo@arm.com> writes:
> > From 0b5f469171c340ef61a48a31877d495bb77bd35f Mon Sep 17 00:00:00 2001
> > From: oluade01 <oluwatamilore.adebayo@arm.com>
> > Date: Fri, 14 Apr 2023 10:24:43 +0100
> > Subject: [PATCH 1/4] Missed opportunity to use [SU]ABD
> >
> > This adds a recognition pattern for the non-widening
> > absolute difference (ABD).
> >
> > gcc/ChangeLog:
> >
> >         * doc/md.texi (sabd, uabd): Document them.
> >         * internal-fn.def (ABD): Use new optab.
> >         * optabs.def (sabd_optab, uabd_optab): New optabs,
> >         * tree-vect-patterns.cc (vect_recog_absolute_difference):
> >         Recognize the following idiom abs (a - b).
> >         (vect_recog_sad_pattern): Refactor to use
> >         vect_recog_absolute_difference.
> >         (vect_recog_abd_pattern): Use patterns found by
> >         vect_recog_absolute_difference to build a new ABD
> >         internal call.
> > ---
> >  gcc/doc/md.texi           |  10 ++
> >  gcc/internal-fn.def       |   3 +
> >  gcc/optabs.def            |   2 +
> >  gcc/tree-vect-patterns.cc | 250 +++++++++++++++++++++++++++++++++-----
> >  4 files changed, 234 insertions(+), 31 deletions(-)
> >
> > diff --git a/gcc/doc/md.texi b/gcc/doc/md.texi
> > index 07bf8bdebffb2e523f25a41f2b57e43c0276b745..0ad546c63a8deebb4b6db894f437d1e21f0245a8 100644
> > --- a/gcc/doc/md.texi
> > +++ b/gcc/doc/md.texi
> > @@ -5778,6 +5778,16 @@ Other shift and rotate instructions, analogous to the
> >  Vector shift and rotate instructions that take vectors as operand 2
> >  instead of a scalar type.
> >
> > +@cindex @code{uabd@var{m}} instruction pattern
> > +@cindex @code{sabd@var{m}} instruction pattern
> > +@item @samp{uabd@var{m}}, @samp{sabd@var{m}}
> > +Signed and unsigned absolute difference instructions.  These
> > +instructions find the difference between operands 1 and 2
> > +then return the absolute value.  A C code equivalent would be:
> > +@smallexample
> > +op0 = abs (op0 - op1)
>
> op0 = abs (op1 - op2)
>
> But that isn't the correct calculation for unsigned (where abs doesn't
> really work).  It also doesn't handle some cases correctly for signed.
>
> I think it's more:
>
>   op0 = op1 > op2 ? (unsigned type) op1 - op2 : (unsigned type) op2 - op1
>
> or (conceptually) max minus min.
>
> E.g. for 16-bit values, the absolute difference between signed 0x7fff
> and signed -0x8000 is 0xffff (reinterpreted as -1 if you cast back
> to signed).  But, ignoring undefined behaviour:
>
>   0x7fff - 0x8000 = -1
>   abs(-1) = 1
>
> which gives the wrong answer.
>
> We might still be able to fold C abs(a - b) to abd for signed a and b
> by relying on undefined behaviour (TYPE_OVERFLOW_UNDEFINED).  But we
> can't do it for -fwrapv.
>
> Richi knows better than me what would be appropriate here.

The question is what does the hardware do?  For the widening [us]sad it's
obvious since the difference is computed in a wider signed mode and the
absolute value always fits.

So what does it actually do, esp. when the difference yields 0x8000?

Richard.

>
> Thanks,
> Richard
>
> > +@end smallexample
> > +
> >  @cindex @code{avg@var{m}3_floor} instruction pattern
> >  @cindex @code{uavg@var{m}3_floor} instruction pattern
> >  @item @samp{avg@var{m}3_floor}
> > diff --git a/gcc/internal-fn.def b/gcc/internal-fn.def
> > index 7fe742c2ae713e7152ab05cfdfba86e4e0aa3456..0f1724ecf37a31c231572edf90b5577e2d82f468 100644
> > --- a/gcc/internal-fn.def
> > +++ b/gcc/internal-fn.def
> > @@ -167,6 +167,9 @@ DEF_INTERNAL_OPTAB_FN (FMS, ECF_CONST, fms, ternary)
> >  DEF_INTERNAL_OPTAB_FN (FNMA, ECF_CONST, fnma, ternary)
> >  DEF_INTERNAL_OPTAB_FN (FNMS, ECF_CONST, fnms, ternary)
> >
> > +DEF_INTERNAL_SIGNED_OPTAB_FN (ABD, ECF_CONST | ECF_NOTHROW, first,
> > +                             sabd, uabd, binary)
> > +
> >  DEF_INTERNAL_SIGNED_OPTAB_FN (AVG_FLOOR, ECF_CONST | ECF_NOTHROW, first,
> >                               savg_floor, uavg_floor, binary)
> >  DEF_INTERNAL_SIGNED_OPTAB_FN (AVG_CEIL, ECF_CONST | ECF_NOTHROW, first,
> > diff --git a/gcc/optabs.def b/gcc/optabs.def
> > index 695f5911b300c9ca5737de9be809fa01aabe5e01..29bc92281a2175f898634cbe6af63c18021e5268 100644
> > --- a/gcc/optabs.def
> > +++ b/gcc/optabs.def
> > @@ -359,6 +359,8 @@ OPTAB_D (mask_fold_left_plus_optab, "mask_fold_left_plus_$a")
> >  OPTAB_D (extract_last_optab, "extract_last_$a")
> >  OPTAB_D (fold_extract_last_optab, "fold_extract_last_$a")
> >
> > +OPTAB_D (uabd_optab, "uabd$a3")
> > +OPTAB_D (sabd_optab, "sabd$a3")
> >  OPTAB_D (savg_floor_optab, "avg$a3_floor")
> >  OPTAB_D (uavg_floor_optab, "uavg$a3_floor")
> >  OPTAB_D (savg_ceil_optab, "avg$a3_ceil")
> > diff --git a/gcc/tree-vect-patterns.cc b/gcc/tree-vect-patterns.cc
> > index a49b09539776c0056e77f99b10365d0a8747fbc5..91e1f9d4b610275dd833ec56dc77f76367ee7886 100644
> > --- a/gcc/tree-vect-patterns.cc
> > +++ b/gcc/tree-vect-patterns.cc
> > @@ -770,6 +770,89 @@ vect_split_statement (vec_info *vinfo, stmt_vec_info stmt2_info, tree new_rhs,
> >      }
> >  }
> >
> > +/* Look for the following pattern
> > +       X = x[i]
> > +       Y = y[i]
> > +       DIFF = X - Y
> > +       DAD = ABS_EXPR<DIFF>
> > + */
> > +static bool
> > +vect_recog_absolute_difference (vec_info *vinfo, gassign *abs_stmt,
> > +                               tree *half_type, bool reject_unsigned,
> > +                               vect_unpromoted_value unprom[2],
> > +                               tree diff_oprnds[2])
> > +{
> > +  if (!abs_stmt)
> > +    return false;
> > +
> > +  /* FORNOW.  Can continue analyzing the def-use chain when this stmt in a phi
> > +     inside the loop (in case we are analyzing an outer-loop).  */
> > +  enum tree_code code = gimple_assign_rhs_code (abs_stmt);
> > +  if (code != ABS_EXPR && code != ABSU_EXPR)
> > +    return false;
> > +
> > +  tree abs_oprnd = gimple_assign_rhs1 (abs_stmt);
> > +  tree abs_type = TREE_TYPE (abs_oprnd);
> > +  if (!abs_oprnd)
> > +    return false;
> > +  if (reject_unsigned && TYPE_UNSIGNED (abs_type))
> > +    return false;
> > +  if (!ANY_INTEGRAL_TYPE_P (abs_type) || TYPE_OVERFLOW_WRAPS (abs_type))
> > +    return false;
> > +
> > +  /* Peel off conversions from the ABS input.  This can involve sign
> > +     changes (e.g.  from an unsigned subtraction to a signed ABS input)
> > +     or signed promotion, but it can't include unsigned promotion.
> > +     (Note that ABS of an unsigned promotion should have been folded
> > +     away before now anyway.)  */
> > +  vect_unpromoted_value unprom_diff;
> > +  abs_oprnd = vect_look_through_possible_promotion (vinfo, abs_oprnd,
> > +                                                   &unprom_diff);
> > +  if (!abs_oprnd)
> > +    return false;
> > +  if (TYPE_PRECISION (unprom_diff.type) != TYPE_PRECISION (abs_type)
> > +      && TYPE_UNSIGNED (unprom_diff.type))
> > +    if (!reject_unsigned)
> > +      return false;
> > +
> > +  /* We then detect if the operand of abs_expr is defined by a minus_expr.  */
> > +  stmt_vec_info diff_stmt_vinfo = vect_get_internal_def (vinfo, abs_oprnd);
> > +  if (!diff_stmt_vinfo)
> > +    return false;
> > +
> > +  bool assigned_oprnds = false;
> > +  gassign *diff = dyn_cast <gassign *> (STMT_VINFO_STMT (diff_stmt_vinfo));
> > +  if (diff_oprnds && diff && gimple_assign_rhs_code (diff) == MINUS_EXPR)
> > +  {
> > +    assigned_oprnds = true;
> > +    diff_oprnds[0] = gimple_assign_rhs1 (diff);
> > +    diff_oprnds[1] = gimple_assign_rhs2 (diff);
> > +  }
> > +
> > +  /* FORNOW.  Can continue analyzing the def-use chain when this stmt in a phi
> > +     inside the loop (in case we are analyzing an outer-loop).  */
> > +  if (vect_widened_op_tree (vinfo, diff_stmt_vinfo, MINUS_EXPR,
> > +                            WIDEN_MINUS_EXPR,
> > +                            false, 2, unprom, half_type))
> > +  {
> > +    if (diff_oprnds && !assigned_oprnds)
> > +    {
> > +      diff_oprnds[0] = unprom[0].op;
> > +      diff_oprnds[1] = unprom[1].op;
> > +    }
> > +  }
> > +  else if (!assigned_oprnds)
> > +  {
> > +    return false;
> > +  }
> > +  else
> > +  {
> > +    *half_type = NULL_TREE;
> > +  }
> > +
> > +  return true;
> > +}
> > +
> >  /* Convert UNPROM to TYPE and return the result, adding new statements
> >     to STMT_INFO's pattern definition statements if no better way is
> >     available.  VECTYPE is the vector form of TYPE.
> > @@ -1308,40 +1391,13 @@ vect_recog_sad_pattern (vec_info *vinfo,
> >    /* FORNOW.  Can continue analyzing the def-use chain when this stmt in a phi
> >       inside the loop (in case we are analyzing an outer-loop).  */
> >    gassign *abs_stmt = dyn_cast <gassign *> (abs_stmt_vinfo->stmt);
> > -  if (!abs_stmt
> > -      || (gimple_assign_rhs_code (abs_stmt) != ABS_EXPR
> > -         && gimple_assign_rhs_code (abs_stmt) != ABSU_EXPR))
> > -    return NULL;
> > -
> > -  tree abs_oprnd = gimple_assign_rhs1 (abs_stmt);
> > -  tree abs_type = TREE_TYPE (abs_oprnd);
> > -  if (TYPE_UNSIGNED (abs_type))
> > -    return NULL;
> > -
> > -  /* Peel off conversions from the ABS input.  This can involve sign
> > -     changes (e.g. from an unsigned subtraction to a signed ABS input)
> > -     or signed promotion, but it can't include unsigned promotion.
> > -     (Note that ABS of an unsigned promotion should have been folded
> > -     away before now anyway.)  */
> > -  vect_unpromoted_value unprom_diff;
> > -  abs_oprnd = vect_look_through_possible_promotion (vinfo, abs_oprnd,
> > -                                                   &unprom_diff);
> > -  if (!abs_oprnd)
> > -    return NULL;
> > -  if (TYPE_PRECISION (unprom_diff.type) != TYPE_PRECISION (abs_type)
> > -      && TYPE_UNSIGNED (unprom_diff.type))
> > -    return NULL;
> >
> > -  /* We then detect if the operand of abs_expr is defined by a minus_expr.  */
> > -  stmt_vec_info diff_stmt_vinfo = vect_get_internal_def (vinfo, abs_oprnd);
> > -  if (!diff_stmt_vinfo)
> > +  vect_unpromoted_value unprom[2];
> > +  if (!vect_recog_absolute_difference (vinfo, abs_stmt, &half_type,
> > +                                      true, unprom, NULL))
> >      return NULL;
> >
> > -  /* FORNOW.  Can continue analyzing the def-use chain when this stmt in a phi
> > -     inside the loop (in case we are analyzing an outer-loop).  */
> > -  vect_unpromoted_value unprom[2];
> > -  if (!vect_widened_op_tree (vinfo, diff_stmt_vinfo, MINUS_EXPR, WIDEN_MINUS_EXPR,
> > -                            false, 2, unprom, &half_type))
> > +  if (!half_type)
> >      return NULL;
> >
> >    vect_pattern_detected ("vect_recog_sad_pattern", last_stmt);
> > @@ -1363,6 +1419,137 @@ vect_recog_sad_pattern (vec_info *vinfo,
> >    return pattern_stmt;
> >  }
> >
> > +/* Function vect_recog_abd_pattern
> > +
> > +   Try to find the following ABsolute Difference (ABD) pattern:
> > +
> > +     VTYPE x, y, out;
> > +     type diff;
> > +   loop i in range:
> > +     S1 diff = x[i] - y[i]
> > +     S2 out[i] = ABS_EXPR <diff>;
> > +
> > +   where 'type' is a integer and 'VTYPE' is a vector of integers
> > +   the same size as 'type'
> > +
> > +   Input:
> > +
> > +   * STMT_VINFO: The stmt from which the pattern search begins
> > +
> > +   Output:
> > +
> > +   * TYPE_out: The type of the output of this pattern
> > +
> > +   * Return value: A new stmt that will be used to replace the sequence of
> > +     stmts that constitute the pattern; either SABD or UABD:
> > +       SABD_EXPR<x, y, out>
> > +       UABD_EXPR<x, y, out>
> > +
> > +      UABD expressions are used when the input types are
> > +      narrower than the output types or the output type is narrower
> > +      than 32 bits
> > + */
> > +
> > +static gimple *
> > +vect_recog_abd_pattern (vec_info *vinfo,
> > +               stmt_vec_info stmt_vinfo, tree *type_out)
> > +{
> > +  /* Look for the following patterns
> > +       X = x[i]
> > +       Y = y[i]
> > +       DIFF = X - Y
> > +       DAD = ABS_EXPR<DIFF>
> > +       out[i] = DAD
> > +
> > +     In which
> > +      - X, Y, DIFF, DAD all have the same type
> > +      - x, y, out are all vectors of the same type
> > +  */
> > +  gassign *last_stmt = dyn_cast <gassign *> (STMT_VINFO_STMT (stmt_vinfo));
> > +  if (!last_stmt)
> > +    return NULL;
> > +
> > +  tree out_type = TREE_TYPE (gimple_assign_lhs (last_stmt));
> > +
> > +  gassign *abs_stmt = last_stmt;
> > +  if (gimple_assign_cast_p (last_stmt))
> > +  {
> > +    tree last_rhs = gimple_assign_rhs1 (last_stmt);
> > +    if (!SSA_VAR_P (last_rhs))
> > +      return NULL;
> > +
> > +    abs_stmt = dyn_cast <gassign *> (SSA_NAME_DEF_STMT (last_rhs));
> > +    if (!abs_stmt)
> > +      return NULL;
> > +  }
> > +
> > +  vect_unpromoted_value unprom[2];
> > +  tree diff_oprnds[2];
> > +  tree half_type;
> > +  if (!vect_recog_absolute_difference (vinfo, abs_stmt, &half_type,
> > +                                      false, unprom, diff_oprnds))
> > +    return NULL;
> > +
> > +#define SAME_TYPE(A, B) (TYPE_PRECISION (A) == TYPE_PRECISION (B))
> > +
> > +  tree abd_oprnds[2];
> > +  if (half_type)
> > +  {
> > +    if (!SAME_TYPE (unprom[0].type, unprom[1].type))
> > +      return NULL;
> > +
> > +    tree diff_type = TREE_TYPE (diff_oprnds[0]);
> > +    if (TYPE_PRECISION (out_type) != TYPE_PRECISION (diff_type))
> > +    {
> > +      vect_convert_inputs (vinfo, stmt_vinfo, 2, abd_oprnds, half_type, unprom,
> > +                          get_vectype_for_scalar_type (vinfo, half_type));
> > +    }
> > +    else
> > +    {
> > +      abd_oprnds[0] = diff_oprnds[0];
> > +      abd_oprnds[1] = diff_oprnds[1];
> > +    }
> > +  }
> > +  else
> > +  {
> > +    if (unprom[0].op && unprom[1].op
> > +       && (!SAME_TYPE (unprom[0].type, unprom[1].type)
> > +       || !SAME_TYPE (unprom[0].type, out_type)))
> > +      return NULL;
> > +
> > +    unprom[0].op = diff_oprnds[0];
> > +    unprom[1].op = diff_oprnds[1];
> > +    tree signed_out = signed_type_for (out_type);
> > +    tree signed_out_vectype = get_vectype_for_scalar_type (vinfo, signed_out);
> > +    vect_convert_inputs (vinfo, stmt_vinfo, 2, abd_oprnds,
> > +                        signed_out, unprom, signed_out_vectype);
> > +
> > +    if (!SAME_TYPE (TREE_TYPE (diff_oprnds[0]), TREE_TYPE (abd_oprnds[0])))
> > +      return NULL;
> > +  }
> > +
> > +  if (!SAME_TYPE (TREE_TYPE (abd_oprnds[0]), TREE_TYPE (abd_oprnds[1]))
> > +      || !SAME_TYPE (TREE_TYPE (abd_oprnds[0]), out_type))
> > +    return NULL;
> > +
> > +  vect_pattern_detected ("vect_recog_abd_pattern", last_stmt);
> > +
> > +  tree vectype = get_vectype_for_scalar_type (vinfo, out_type);
> > +  if (!vectype
> > +      || !direct_internal_fn_supported_p (IFN_ABD, vectype,
> > +                                         OPTIMIZE_FOR_SPEED))
> > +    return NULL;
> > +
> > +  *type_out = STMT_VINFO_VECTYPE (stmt_vinfo);
> > +
> > +  tree var = vect_recog_temp_ssa_var (out_type, NULL);
> > +  gcall *abd_stmt = gimple_build_call_internal (IFN_ABD, 2,
> > +                                               abd_oprnds[0], abd_oprnds[1]);
> > +  gimple_call_set_lhs (abd_stmt, var);
> > +  gimple_set_location (abd_stmt, gimple_location (last_stmt));
> > +  return abd_stmt;
> > +}
> > +
> >  /* Recognize an operation that performs ORIG_CODE on widened inputs,
> >     so that it can be treated as though it had the form:
> >
> > @@ -6439,6 +6626,7 @@ struct vect_recog_func
> >  static vect_recog_func vect_vect_recog_func_ptrs[] = {
> >    { vect_recog_bitfield_ref_pattern, "bitfield_ref" },
> >    { vect_recog_bit_insert_pattern, "bit_insert" },
> > +  { vect_recog_abd_pattern, "abd" },
> >    { vect_recog_over_widening_pattern, "over_widening" },
> >    /* Must come after over_widening, which narrows the shift as much as
> >       possible beforehand.  */
> > --
> > 2.25.1

  reply	other threads:[~2023-05-10  9:51 UTC|newest]

Thread overview: 25+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2023-05-09 16:07 Oluwatamilore Adebayo
2023-05-10  9:01 ` Richard Sandiford
2023-05-10  9:49   ` Richard Biener [this message]
2023-05-10  9:51     ` Richard Biener
2023-05-10 15:27       ` Richard Sandiford
2023-05-17 12:21         ` oluwatamilore.adebayo
2023-05-18  8:39           ` [PATCH 1/4] " Oluwatamilore.Adebayo
2023-05-18 17:59             ` Richard Sandiford
2023-05-22 13:32               ` Richard Biener
2023-05-23 14:27               ` [PATCH] vect: " Oluwatamilore Adebayo
2023-05-23 14:34                 ` [PATCH 1/2] " Oluwatamilore Adebayo
2023-05-24  9:48                   ` Richard Sandiford
2023-06-06  9:50                     ` [PATCH] vect: " Oluwatamilore Adebayo
2023-06-06  9:53                       ` [PATCH 1/2] " Oluwatamilore Adebayo
2023-06-06 12:56                         ` Richard Sandiford
2023-06-06 14:34                           ` Oluwatamilore Adebayo
2023-06-08 10:28                             ` [PATCH] vect: " Oluwatamilore Adebayo
2023-06-08 10:31                               ` [PATCH 1/2] " Oluwatamilore Adebayo
2023-06-13  8:26                                 ` Oluwatamilore Adebayo
2023-06-14 11:15                                   ` Richard Sandiford
2023-06-14 15:26                                     ` Oluwatamilore Adebayo
2023-06-15  6:38                                       ` Richard Sandiford
2023-05-10 13:29     ` [PATCH] vect: " Oluwatamilore Adebayo
2023-05-15 12:35       ` Oluwatamilore Adebayo
  -- strict thread matches above, loose matches on Subject: below --
2023-05-09 16:00 Oluwatamilore Adebayo

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=CAFiYyc0L1pmHebsppQM9_c2yH7s48WLhH-dP+_rVXv2hyNRb5A@mail.gmail.com \
    --to=richard.guenther@gmail.com \
    --cc=Oluwatamilore.Adebayo@arm.com \
    --cc=gcc-patches@gcc.gnu.org \
    --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).