https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89821
Bug ID: 89821
Summary: Get a SIGFPE on a simple test of a kind=real128
variable with -ffpe-trap=invalid switch
Product: gcc
Version: 7.3.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: fortran
Assignee: unassigned at gcc dot gnu.org
Reporter: urbanjost at comcast dot net
Target Milestone: ---
Created attachment 46019
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=46019&action=edit
test program that fails if -ffpe-trap=invalid is present on kind=real128 test
but ok on kind=real32 and kind=real64 tests even with switch on
Although there may be IEEE methods that are now preferred with some compilers
the simplified code in the attachment shows a method for testing for a NaN
value that I believe is standard and has worked with 32-bit and 64-bit REAL
variables in more than one programming environment including gfortran(1) with
and without the -ffpe-trap=invalid switch on the gfortran compile. The test
fails on kind=real128 variables (on Cygwin, at least) with the error
signal SIGFPE: Floating-point exception - erroneous arithmetic operation.
but the same kind of test works without an error on kind=real32 and kind=real64
variables, as it always has up to this point. I do not think the test is an
arithmetic operation and should be allowed even with the -ffpe-trap=invalid
switch present, as it does for kind=real32 and kind=real128 floating point
variables. Works without any VISIBLE problem without the -ffpe-trap=invalid
switch.
