[PATCH] math: Fix float conversion regressions with gcc-12 [BZ #28713]

[Paul Zimmermann]

       Dear Szabolcs,

> > +# define M_PI_2f	1.57079632679489661923f	/* pi/2 */
> > 
> > -      __real__ res = (FLOAT) M_MLIT (M_PI_2) - __real__ y;
> > +      __real__ res = M_MLIT (M_PI_2) - __real__ y;
> > 
> > I wonder why your patch never uses M_PI_2f. Should it be M_PI_2f above?
> 
> this is template code under math/ and M_MLIT (M_PI_2) should
> expand to M_PI_2f when it is used for the float variant.
> 
> see the change in
> sysdeps/generic/math-type-macros-float.h

ok.

While looking at that patch, in math/math.h, I thought new constants should
be declared as hexadecimal values?

+# define M_Ef		2.7182818284590452354f	/* e */

Otherwise this patch seems good to me.

Paul