Re: freeze for sbcl-2.6.0

Yes. I also decided to disable the other float expt transforms instead
of also reverting the change for (expt x 4), which is good for
rationals.
NEWS doesn't yet reflect that (expt x 3) or (expt x 1/2) no longer get
transformed, unless float-accuracy 0.

sb-c::float-accuracy being in sb-c:: is also not a great interface,
ought to move it, and things like INSERT-ARRAY-BOUNDS-CHECKS,
somewhere more official.

On Sun, Dec 28, 2025 at 12:52 AM Christophe Rhodes <[email protected]> wrote:
>
> Stas Boukarev <[email protected]> writes:
>
> > I'm not sure anymore about the (expt x constant) transforms to floats.
> > They produce different values from pow(). GCC and clang only convert
> > pow(d, 2) to d*d. Unless -ffast-math. The same goes for converting to
> > sqrt.
> > So I think all these should be hidden behind float-accuracy = 0.
> > But I made a change for the out of line expt to call sqrt for 1/2,
> > which is both faster and consistent with the inline version.
> > Maybe it should be reverted so that things don't swing back and forth
> > across multiple releases?
>
> Thanks.  I saw the commits this evening -- are you finished?  Does NEWS
> reflect what's currently in the tree now?  (I expect to release some
> time tomorrow, in 12-18 hours from sending this message).
>
> Thanks,
>
> Christophe


_______________________________________________
Sbcl-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/sbcl-devel