On Mon, Jun 18, 2018 at 07:40:46PM +0000, Joseph Myers wrote: > On Mon, 18 Jun 2018, Michael Meissner wrote: > > > gcc.target/powerpc/pr70117.c (bug in isnormal on > > IEEE 128) > > Doesn't look like a bug in isnormal to me. That test is explicitly using > an IBM long double representation and can't possibly be expected to work > with IEEE long double - either it needs to be disabled for IEEE long > double, or to use -mabi=ibmlongdouble explicitly, or to use __ibm128 (but > if you use __ibm128 you can't then test the type-specific __builtin_isinfl > like it does at present).
I've fixed the test to use __ibm128 already in my branch. The issue is the compiler generates different code for __ibm128 (when long double is IEEE 128-bit) than it does for long double using IBM extended double. The code fails for __ibm128, but does not fail normally. I haven't fully investigated why the code is different. -- Michael Meissner, IBM IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA email: meiss...@linux.ibm.com, phone: +1 (978) 899-4797
