Re: [PATCH] [PR24021] Implement PLUS_EXPR range-op entry for floats.

[Aldy Hernandez <aldyh@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1187 bytes --]

This patch fixes the oversight.

Tested on x86-64 Linux.

Pushed.

On Wed, Nov 9, 2022 at 12:05 AM Aldy Hernandez <aldyh@redhat.com> wrote:
>
> Sigh, one more thing.
>
> There are further possibilities for a NAN result, even if the operands
> are !NAN and the result from frange_arithmetic is free of NANs.
> Adding different signed infinities.
>
> For example, [-INF,+INF] + [-INF,+INF] has the possibility of adding
> -INF and +INF, which is a NAN.  Since we end up calling frange
> arithmetic on the lower bounds and then on the upper bounds, we miss
> this, and mistakenly think we're free of NANs.
>
> I have a patch in testing, but FYI, in case anyone notices this before
> I get around to it tomorrow.
>
> Aldy
>
> On Tue, Nov 8, 2022 at 3:11 PM Jakub Jelinek <jakub@redhat.com> wrote:
> >
> > On Tue, Nov 08, 2022 at 03:06:53PM +0100, Aldy Hernandez wrote:
> > > +// If either operand is a NAN, set R to the combination of both NANs
> > > +// signwise and return TRUE.
> >
> > This comment doesn't describe what it does now.
> > If either operand is a NAN, set R to NAN with unspecified sign bit and return
> > TRUE.
> > ?
> >
> > Other than this LGTM.
> >
> >         Jakub
> >

[-- Attachment #2: 0001-range-op-float-Set-NAN-possibility-for-INF-INF-and-v.patch --]
[-- Type: text/x-patch, Size: 2276 bytes --]

From 68b0615be2aaff3a8ce91ba7cd0f69ebbd93702c Mon Sep 17 00:00:00 2001
From: Aldy Hernandez <aldyh@redhat.com>
Date: Tue, 8 Nov 2022 23:42:04 +0100
Subject: [PATCH] [range-op-float] Set NAN possibility for INF + (-INF) and
 vice versa.

Some combinations of operations can yield a NAN even if no operands
have the possiblity of a NAN.  For example, [-INF] + [+INF] = NAN and
vice versa.

For [-INF,+INF] + [-INF,+INF], frange_arithmetic will not return a
NAN, and since the operands have no possibility of a NAN, we will
mistakenly assume the result cannot have a NAN.  This fixes the
oversight.

gcc/ChangeLog:

	* range-op-float.cc (foperator_plus::fold_range): Set NAN for
	addition of different signed infinities.
	(range_op_float_tests): New test.
---
 gcc/range-op-float.cc | 26 +++++++++++++++++++++++++-
 1 file changed, 25 insertions(+), 1 deletion(-)

diff --git a/gcc/range-op-float.cc b/gcc/range-op-float.cc
index 3bc6cc8849d..8282c912fc4 100644
--- a/gcc/range-op-float.cc
+++ b/gcc/range-op-float.cc
@@ -1863,7 +1863,21 @@ foperator_plus::fold_range (frange &r, tree type,
 
   r.set (type, lb, ub);
 
-  if (lb_nan || ub_nan)
+  // Some combinations can yield a NAN even if no operands have the
+  // possibility of a NAN.
+  bool maybe_nan;
+  // [-INF] + [+INF] = NAN
+  if (real_isinf (&op1.lower_bound (), true)
+      && real_isinf (&op2.upper_bound (), false))
+    maybe_nan = true;
+  // [+INF] + [-INF] = NAN
+  else if (real_isinf (&op1.upper_bound (), false)
+	   && real_isinf (&op2.lower_bound (), true))
+    maybe_nan = true;
+  else
+    maybe_nan = false;
+
+  if (lb_nan || ub_nan || maybe_nan)
     // Keep the default NAN (with a varying sign) set by the setter.
     ;
   else if (!op1.maybe_isnan () && !op2.maybe_isnan ())
@@ -1960,6 +1974,16 @@ range_op_float_tests ()
   r1 = frange_float ("-1", "-0");
   r1.update_nan (false);
   ASSERT_EQ (r, r1);
+
+  // [-INF,+INF] + [-INF,+INF] could be a NAN.
+  range_op_handler plus (PLUS_EXPR, float_type_node);
+  r0.set_varying (float_type_node);
+  r1.set_varying (float_type_node);
+  r0.clear_nan ();
+  r1.clear_nan ();
+  plus.fold_range (r, float_type_node, r0, r1);
+  if (HONOR_NANS (float_type_node))
+    ASSERT_TRUE (r.maybe_isnan ());
 }
 
 } // namespace selftest
-- 
2.38.1