From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-bugzilla@gcc.gnu.org>
Received: by sourceware.org (Postfix, from userid 48)
	id 13AC138555A0; Wed,  8 Mar 2023 18:56:52 +0000 (GMT)
DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 13AC138555A0
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org;
	s=default; t=1678301813;
	bh=kfW133QSRXfNRGGXVo+ilS2YCLFkcfmgLxH2yXwWZUA=;
	h=From:To:Subject:Date:In-Reply-To:References:From;
	b=PAtvaNdjaTNZqucoXcLrUbEFMZAkM2NeRQOMYMXxRAcgQl5J6K9WM7avhQ6Qd3IgS
	 BirrwQJaTAUbhJq6wcrf24uCqu3AQJu2O+jL2LRbQYbglu2A+/YXobSyhiLo9wxUzU
	 8vXfbB/7o40OoBqdYvgSpNjLyrxV5pcZ32teAVLs=
From: "jakub at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug tree-optimization/109008] [13 Regression] Wrong code in scipy
 package since r13-3926-gd4c2f1d376da6f
Date: Wed, 08 Mar 2023 18:56:51 +0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: tree-optimization
X-Bugzilla-Version: 13.0
X-Bugzilla-Keywords: wrong-code
X-Bugzilla-Severity: normal
X-Bugzilla-Who: jakub at gcc dot gnu.org
X-Bugzilla-Status: NEW
X-Bugzilla-Resolution: 
X-Bugzilla-Priority: P1
X-Bugzilla-Assigned-To: unassigned at gcc dot gnu.org
X-Bugzilla-Target-Milestone: 13.0
X-Bugzilla-Flags: 
X-Bugzilla-Changed-Fields: 
Message-ID: <bug-109008-4-0AwfZQNS6i@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-109008-4@http.gcc.gnu.org/bugzilla/>
References: <bug-109008-4@http.gcc.gnu.org/bugzilla/>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/
Auto-Submitted: auto-generated
MIME-Version: 1.0
List-Id: <gcc-bugs.sourceware.org>

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=3D109008

--- Comment #37 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
I've additionally ran
#include <math.h>

#include "pr109008-4.c"

struct S { float (*fn) (float); float lb, ub; };
struct S arr[] =3D {
#include "pr109008-4.h"
};

int
main ()
{
  float plus_inf =3D __builtin_inf ();
  float minus_inf =3D -plus_inf;
  for (int i =3D 0; i < sizeof (arr) / sizeof (arr[0]); ++i)
    {
      int n1, n2;
      float lb =3D nextafterf (arr[i].lb, minus_inf);
      float ub =3D nextafterf (arr[i].ub, plus_inf);
      if (!__builtin_isnan (arr[i].fn (lb)) || !__builtin_isnan (arr[i].fn
(ub)))
        __builtin_printf ("%p err\n", arr[i].fn);
      lb =3D arr[i].lb;
      ub =3D arr[i].ub;
      for (n1 =3D 0; n1 < 50; n1++)
        if (!__builtin_isnan (arr[i].fn (lb)))
          break;
        else if (lb > ub)
          {
            n1 =3D -1;
            n2 =3D -1;
            break;
          }
        else
          lb =3D nextafterf (lb, plus_inf);
      if (n1 !=3D -1)
        for (n2 =3D 0; n2 < 50; n2++)
          if (!__builtin_isnan (arr[i].fn (ub)))
            break;
          else
            ub =3D nextafterf (ub, minus_inf);
      __builtin_printf ("%d %d\n", n1, n2);
    }
}
on the header from the #c32+#c33 patches, and
./pr109008-4-main 2>&1 | sort | uniq -c | sort -n
      1 1 11
      1 11 1
      1 1 17
      1 1 25
      1 1 28
      1 1 3
      1 1 4
      1 19 1
      1 2 13
      1 2 16
      1 2 17
      1 2 36
      1 2 40
      1 3 2
      1 39 1
      1 5 1
      1 5 2
      2 1 32
      2 1 7
      2 3 1
      2 3 50
      3 4 1
     12 50 2
     16 2 50
     27 50 1
     39 1 50
     48 32 32
     53 33 33
     57 17 17
     73 16 16
     77 8 8
     78 9 9
    114 4 4
    124 5 5
    170 3 3
   8533 2 2
  15203 1 1
  24287 50 50
  59288 2 1
  59613 1 2
 130096 -1 -1

The -1 -1 case means that in reality the range is empty (UNDEFINED), there =
are
no values of eps for which the function returns it, as the difference in
exponents is too large for such a number to exist.  50 means actually 50 or
more ulps.=