Re: [PATCH v3] Add sinh(atanh(x)) and cosh(atanh(x)) optimizations

[Giuliano Augusto Faulin Belinassi <giuliano dot belinassi at usp dot br>]

On Wed, Nov 14, 2018 at 7:41 AM Richard Biener
<richard.guenther@gmail.com> wrote:
>
> On Tue, Nov 13, 2018 at 10:25 PM Giuliano Augusto Faulin Belinassi
> <giuliano.belinassi@usp.br> wrote:
> >
> > Only do the optimization if flag_signed_zeros &&
> > !flag_finite_math_only is set, as suggested in the previous iteration.
> >
> > Before, the patch did the optimization even when -fno-signed-zeros and
> > -ffinite-math-only was set. This could generate badly incorrect
> > results for targets that do not support infinite or signed zeros.
>
> How's the result wrong if there are no signed zeros?  Note that -ffast-math
> enables -fno-signed-zeros for example.  So both of your check look
> backwards.

Indeed. After plotting the graph of both functions, it is very clear
that this check isn't required. Sorry about that.

> Also the support for signed zeros and infs/nans should be guarded with
>
> !HONOR_SIGNED_ZEROS (type) && !HONOR_NANS (type) && !HONOR_INFINITIES (type)
>
> which then means there's no difference between -0. and 0. and there are no
> NaNs or Infs in the inputs and ouptut NaNs or Infs need not be produced.

There can be NaNs and Infinities. For NaNs, take any input that is
outside the [-1, 1] line.
For Infinities, take x = -1, or x = 1. I think these must be 'honored'
as to ensure compatibility with the original expression.

so I must check for
!HONOR_SIGNED_ZEROS (type) && HONOR_NANS (type) && HONOR_INFINITIES (type)
that is correct? Also, is it safe to remove the !finite_math_only with
this, as now it is stated that the type supports infinity and NaNs?

However, I am not sure if it is OK to remove unsafe-math-optimizations
even if it enables
finite_math_only because of the 2 ULP error. As stated in the first
iteration, the user can be
using a very precise math library that yields 0 ULP.

> Richard.
>
> > I also updated the tests with the proper flags.
> >
> > gcc/ChangeLog
> > 2018-11-13  Giuliano Belinassi  <giuliano.belinassi@usp.br>
> >
> >     * match.pd (sinh (atanh (x))): New simplification rules.
> >     (cosh (atanh (x))): Likewise.
> >
> > gcc/testsuite/ChangeLog
> > 2018-11-13  Giuliano Belinassi  <giuliano.belinassi@usp.br>
> >
> >     * gcc.dg/sinhatanh-1.c: New test.
> >     * gcc.dg/sinhatanh-2.c: New test.
> >
> > There are no tests in trunk that seems to be breaking because of this patch.