[PATCH] Do not specify NAN sign in frange::set_nonnegative.

[PATCH] Do not specify NAN sign in frange::set_nonnegative.

[Aldy Hernandez]

[Jakub, how's this?  Do you agree?]

After further reading of the IEEE 754 standard, it has become clear
that there are no guarantees with regards to the sign of a NAN when it
comes to any operation other than copy, copysign, abs, and negate.

Currently, set_nonnegative() is only used in one place in ranger
applicable to floating point values, when expanding unknown calls.
Since we already specially handle copy, copysign, abs, and negate, all
the calls to set_nonnegative() must be NAN-sign agnostic.

The cleanest solution is to leave the sign unspecificied in
frange::set_nonnegative().  Any special case, must be handled by the
caller.

gcc/ChangeLog:

	* value-range.cc (frange::set_nonnegative): Remove NAN sign handling.
	(range_tests_signed_zeros): Adjust test.
---
 gcc/value-range.cc | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/gcc/value-range.cc b/gcc/value-range.cc
index 852ac09f2c4..d55d85846c1 100644
--- a/gcc/value-range.cc
+++ b/gcc/value-range.cc
@@ -797,14 +797,17 @@ frange::zero_p () const
 	  && real_iszero (&m_max));
 }
 
+// Set the range to non-negative numbers, that is [+0.0, +INF].
+//
+// The NAN in the resulting range (if HONOR_NANS) has a varying sign
+// as there are no guarantees in IEEE 754 wrt to the sign of a NAN,
+// except for copy, abs, and copysign.  It is the responsibility of
+// the caller to set the NAN's sign if desired.
+
 void
 frange::set_nonnegative (tree type)
 {
   set (type, dconst0, frange_val_max (type));
-
-  // Set +NAN as the only possibility.
-  if (HONOR_NANS (type))
-    update_nan (/*sign=*/false);
 }
 
 // Here we copy between any two irange's.  The ranges can be legacy or
@@ -3923,7 +3926,6 @@ range_tests_signed_zeros ()
     ASSERT_TRUE (r0.undefined_p ());
 
   r0.set_nonnegative (float_type_node);
-  ASSERT_TRUE (r0.signbit_p (signbit) && !signbit);
   if (HONOR_NANS (float_type_node))
     ASSERT_TRUE (r0.maybe_isnan ());
 }
-- 
2.38.1

Re: [PATCH] Do not specify NAN sign in frange::set_nonnegative.

[Jakub Jelinek]

On Thu, Nov 10, 2022 at 04:03:46PM +0100, Aldy Hernandez wrote:
> [Jakub, how's this?  Do you agree?]
> 
> After further reading of the IEEE 754 standard, it has become clear
> that there are no guarantees with regards to the sign of a NAN when it
> comes to any operation other than copy, copysign, abs, and negate.
> 
> Currently, set_nonnegative() is only used in one place in ranger
> applicable to floating point values, when expanding unknown calls.
> Since we already specially handle copy, copysign, abs, and negate, all
> the calls to set_nonnegative() must be NAN-sign agnostic.
> 
> The cleanest solution is to leave the sign unspecificied in
> frange::set_nonnegative().  Any special case, must be handled by the
> caller.
> 
> gcc/ChangeLog:
> 
> 	* value-range.cc (frange::set_nonnegative): Remove NAN sign handling.
> 	(range_tests_signed_zeros): Adjust test.

LGTM, thanks.

	Jakub