https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86420
Bug ID: 86420
Summary: [9 regression] nextafter(0x1p-1022,0) is constant
folded
Product: gcc
Version: 9.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c
Assignee: unassigned at gcc dot gnu.org
Reporter: nsz at gcc dot gnu.org
Target Milestone: ---
nextafterf(FLT_MIN,0), nextafter(DBL_MIN,0), nextafterl(LDBL_MIN,0) should
signal the underflow exception (c99 annex F requirement), but now gcc constant
folds them, this causes the following glibc test failures with gcc trunk:
FAIL: math/bug-nextafter
FAIL: math/bug-nexttoward
for
double f(void)
{
return __builtin_nextafter (0x1p-1022, 0);
}
on aarch64 with gcc -O2 -S -std=c11 -frounding-math
i get
f:
mov x0, 4503599627370495
fmov d0, x0
ret
with gcc-8 it is
f:
movi d1, #0
mov x0, 4503599627370496
fmov d0, x0
b nextafter
gcc has no flag to say 'floating-point exceptions matter' (like -frounding-math
for non-default rounding mode) so i think gcc should not do such optimizations
in standard conform mode where users might test for exceptions.
