From: Richard Biener <rguenther@suse.de>
Date: Mon, 15 Nov 2021 11:13:40 +0000 (+0100)
Subject: middle-end/103193 - avoid canonicalizing <= and >= to == for floats
X-Git-Tag: basepoints/gcc-13~2818
X-Git-Url: http://git.ipfire.org/?a=commitdiff_plain;h=d9ca2ca381e44a332703155d07b50b84aa21f80d;p=thirdparty%2Fgcc.git

middle-end/103193 - avoid canonicalizing <= and >= to == for floats

This avoids doing aforementioned canoncalization when -ftrapping-math
is in effect and we honor NaNs.

2021-11-15  Richard Biener  <rguenther@suse.de>

	PR middle-end/103193
	* match.pd: Avoid canonicalizing (le/ge @0 @0) to (eq @0 @0)
	with NaNs and -ftrapping-math.
---

diff --git a/gcc/match.pd b/gcc/match.pd
index 3e54e2cf5a6d..5adcd6bd02c5 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -4671,7 +4671,10 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
   (if (! FLOAT_TYPE_P (TREE_TYPE (@0))
        || ! HONOR_NANS (@0))
    { constant_boolean_node (true, type); }
-   (if (cmp != EQ_EXPR)
+   (if (cmp != EQ_EXPR
+	/* With -ftrapping-math conversion to EQ loses an exception.  */
+	&& (! FLOAT_TYPE_P (TREE_TYPE (@0))
+	    || ! flag_trapping_math))
     (eq @0 @0)))))
 (for cmp (ne gt lt)
  (simplify