https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110305
Bug ID: 110305
Summary: Incorrect optimization with -O3 -fsignaling-nans
-fno-signed-zeros
Product: gcc
Version: og12 (devel/omp/gcc-12)
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: tkisuki at tachyum dot com
Target Milestone: ---
float f(float a)
{
return a + 0.0f;
}
Compiling the above code with -O3 -fsignaling-nans -fno-signed-zeros simplifies
'a + 0' to 'a'.
It seems '-fsignaling-nans' is not considered for this simplification.
Simplification is done by simplify_context::simplify_binary_operation_1()
function in gcc/simplify-rtx.cc.
'-fsignaling-nans' is experimental, but can the condition also check
!HONOR_SNANS(mode)?
--- a/gcc/simplify-rtx.cc
+++ b/gcc/simplify-rtx.cc
@@ -2698,7 +2698,8 @@ simplify_context::simplify_binary_operation_1 (rtx_code
code,
when x is NaN, infinite, or finite and nonzero. They aren't
when x is -0 and the rounding mode is not towards -infinity,
since (-0) + 0 is then 0. */
- if (!HONOR_SIGNED_ZEROS (mode) && trueop1 == CONST0_RTX (mode))
+ if (!HONOR_SIGNED_ZEROS (mode) && !HONOR_SNANS (mode)
+ && trueop1 == CONST0_RTX (mode))
return op0;
