https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88576
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Attachment #45315|0 |1
is obsolete| |
--- Comment #23 from Richard Biener <rguenth at gcc dot gnu.org> ---
Created attachment 63981
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=63981&action=edit
patch
ISTR this was important for Fortran code (sic!), there is currently no way to
tell that 'errno' isn't a thing. And with LTO this may or may not be true -
fortran code dispatching to the C library changing errno can at least be
observed by C code calling into such fortran code.
So the Fortran frontend resorts to defaulting to -fno-errno-math.
For testing I'm benchmarking the attached patch on SPEC CPU 2017 though
possibly polyhedron might be more interesting as a pure fortran benchmark.
As I said, we need at least some kind of plan how to deal with such language
differences in treating core C runtime parts (and also document
inter-operability).
