Re: [PATCH] s390x: Add single instruction math functions

[Rich Felker <dalias@libc.org>]

On Tue, Jun 13, 2017 at 05:55:13PM +0200, Szabolcs Nagy wrote:
> > > diff --git a/configure b/configure
> > > index c2db298c..bcaf3a7d 100755
> > > --- a/configure
> > > +++ b/configure
> > > @@ -656,6 +656,12 @@ trycppif __LITTLE_ENDIAN__ "$t" && SUBARCH=${SUBARCH}le
> > >  trycppif _SOFT_FLOAT "$t" && fail "$0: error: soft-float not supported on powerpc64"
> > >  fi
> > >
> > > +if test "$ARCH" = "s390x" ; then
> > > +echo 'float x; void f(){__asm__("fiebra %0,6,%1,4":"=f"(x):"f"(x));}' > "$tmpc"
> > > +$CC $CFLAGS_C99FSE $CPPFLAGS $CFLAGS -c -o /dev/null "$tmpc" >/dev/null 2>&1 ||
> > > +  fail "$0: error: s390x isa level is too low, use at least -march=z196"
> > > +fi
> > > +
> > >  if test "$ARCH" = "sh" ; then
> > >  tryflag CFLAGS_AUTO -Wa,--isa=any
> > >  trycppif __BIG_ENDIAN__ "$t" && SUBARCH=${SUBARCH}eb
> > 
> > Why is the x86 configure logic that I used as a template correct?
> 
> hm i didnt know about that
> then the patch is probably ok.

The x86 logic is actually implementing policy to the opposite effect,
ignoring the toolchain's default -march and forcing the minimum
baseline isa (i486) unless the user manually puts a specific -march in
CFLAGS. I'm doubtful that it's actually a good idea to be doing that
any more; at least it's inconsistent with how other archs are treated
(using the toolchain's default).

Rich