Re: #pragma STDC FENV_ACCESS ON

On Wed, Jul 17, 2024 at 4:27 PM Steve Kargl
<sgk@troutmask.apl.washington.edu> wrote:
>
> On Wed, Jul 17, 2024 at 04:16:33PM -0400, enh wrote:
> > On Wed, Jul 17, 2024 at 4:01 PM Steve Kargl
> > <sgk@troutmask.apl.washington.edu> wrote:
> > >
> > > On Wed, Jul 17, 2024 at 01:30:39PM -0400, enh wrote:
> > > > i see lib/msun/src/e_sqrtl.c has `#pragma STDC FENV_ACCESS ON`, but
> > > > s_fma.c, s_fmaf.c, s_fmal.c, and s_nearbyint.c all use fesetround() or
> > > > fesetenv() too --- shouldn't they also have the pragma?
> > > >
> > > > (Android currently builds all of libm with that pragma, but (a) that
> > > > seems like a big hammer and (b) afaict upstream [you] don't, so this
> > > > is potentially causing problems^W^Wallowing the compiler to cause
> > > > problems for you?)
> > >
> > > Some of the compilers, which that are used to compile FreeBSD,
> > > do not support "#pragma STDC FENV_ACCESS ON".
> >
> > that's why you have a line in the Makefile to quiet the warning :-)
> >
> > (remember you _do_ have this #pragma in e_sqrtl.c already.)
> >
>
> Then your original post is simply a rhetorical question?

no. i said:

1. e_sqrtl.c has `#pragma STDC FENV_ACCESS ON`

2. s_fma.c, s_fmaf.c, s_fmal.c, and s_nearbyint.c all use fesetround()
or fesetenv() but _don't_ have the pragma

that is: #1 implies you think this is needed, but #2 shows that it's
inconsistently applied in places where it's needed.

> Consider it future-proofing for when all compilers supprt
> the pragma.

"future proofing" is exactly why i'm suggesting you should probably
have the pragma in the other files that use the relevant functions :-)

> --
> Steve
>