[PATCH 1/1] Keep the denormal-operand exception masked; modify FE_ALL_EXCEPT accordingly.
Wed Aug 15 16:05:00 GMT 2018
On Aug 15 17:43, Houder wrote:
> On 2018-08-15 17:26, Corinna Vinschen wrote:
> > On Aug 15 17:01, Houder wrote:
> > > On 2018-08-15 16:54, Corinna Vinschen wrote:
> > > > Shouldn't FE_ALL_EXCEPT_X86 be defined locally in fenv.cc only?
> > > > I don't see that Linux exports that definition.
> > >
> > > Ah, Sorry. Do I have to resubmit my patch? Or is it easy enough for
> > > you to
> > > make this modification?
> > It's easy enough but I'm still mulling over __FE_DENORM. The glibc
> > fenv.h header defines it, so I guess we should stick to it. In that
> > case it might make sense to revert the original comment and just move
> > __FE_ALL_EXCEPT_X86.
> ... uhm, my intention was to remove FE_DENORMAL from fenv.h, because it
> is no longer part of the interface.
> I should have defined "a mask" in fenv.cc, that would have enabled me to
> initialize the MXCSR register (i.e. mask ALL exceptions).
> I defined __FE_ALL_EXCEPT_X86 (and should have defined it as 0x3f).
> Basically, there was no reason to define __FE_DENORM (or __FE_DENORMAL),
> now that I think it over.
> (I needed __FE_DENORM(AL) for testing)
I think it's more clear to have the definition, even if only used
locally. You're using __FE_DENORM in a comment, too. So I just
moved it to fenv.cc.
Corinna Vinschen Please, send mails regarding Cygwin to
Cygwin Maintainer cygwin AT cygwin DOT com
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 833 bytes
Desc: not available
More information about the Cygwin-patches