This is a simple one line fix to the regression PR middle-end/103406,
where x - x is being folded to 0.0 even when x is +Inf or -Inf.
In GCC 11 and previously, we'd check whether the type honored NaNs
(which implicitly covered the case where the type honors infinities),
but my patch to test whether the operand could potentially be NaN
failed to also check whether the operand could potentially be Inf.
This patch doesn't address the issue of NaN signedness from binary
arithmetic operations, just the regression.
This patch has been tested on x86_64-pc-linux-gnu with make bootstrap
and make -k check with no new failures. Ok for mainline?
2021-11-25 Roger Sayle <ro...@nextmovesoftware.com>
gcc/ChangeLog
PR middle-end/103406
* match.pd (minus @0 @0): Check tree_expr_maybe_infinite_p.
gcc/testsuite/ChangeLog
PR middle-end/103406
* gcc.dg/pr103406.c: New test case.
Thanks in advance (and sorry for the inconvenience),
Roger
--
diff --git a/gcc/match.pd b/gcc/match.pd
index f059b47..d28dfe2 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -232,7 +232,9 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
is volatile. */
(simplify
(minus @0 @0)
- (if (!FLOAT_TYPE_P (type) || !tree_expr_maybe_nan_p (@0))
+ (if (!FLOAT_TYPE_P (type)
+ || (!tree_expr_maybe_nan_p (@0)
+ && !tree_expr_maybe_infinite_p (@0)))
{ build_zero_cst (type); }))
(simplify
(pointer_diff @@0 @0)
diff --git a/gcc/testsuite/gcc.dg/pr103406.c b/gcc/testsuite/gcc.dg/pr103406.c
new file mode 100644
index 0000000..9c7b83b
--- /dev/null
+++ b/gcc/testsuite/gcc.dg/pr103406.c
@@ -0,0 +1,14 @@
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-tree-optimized" } */
+
+#define HUGE __DBL_MAX__
+#define INF (HUGE + HUGE)
+#define NAN (INF - INF)
+
+double foo() {
+ double x = -NAN;
+ double y = NAN;
+ return x + y;
+}
+
+/* { dg-final { scan-tree-dump-not "return 0\.0" "optimized" } } */
