Re: [RFC][PATCH] New expf, exp2f, logf, log2f and powf implementations

[Joel Sherrill <joel dot sherrill at oarcorp dot com>]

On 8/30/2017 10:04 AM, Corinna Vinschen wrote:
> Hi Szabolcs,
>
> On Aug 30 15:18, Szabolcs Nagy wrote:
> > Based on code from https://github.com/ARM-software/optimized-routines/
> >
> > This patch adds a highly optimized generic implementation of expf,
> > exp2f, logf, log2f and powf.  The new functions are not only
> > significantly faster, but are also smaller, more accurate and fix
> > several correctness issues.  In order to achieve this, the algorithm
> > uses double precision arithmetic for accuracy, avoids divisions and
> > uses small table lookups to minimize the polynomials.  Special cases
> > are handled inline to avoid the unnecessary overhead of wrapper
> > functions and set errno to POSIX requirements.
> >
> > The new functions are added under newlib/libm/common and are
> > currently enabled for AArch64 and AArch32 with VFP format.  The
> > new functions are written in C99 and do not support non-IEEE
> > representations, mixed endian doubles or non-standard errno setting.
> > Targets can enable the new math code by undefining OBSOLETE_MATH.
> > Targets with a single precision FPU may still prefer the old
> > implementation. It was not clear what's the best way to add such
> > alternative math implementations to newlib so this is an RFC
> > patch and feedback is welcome.
>
> I'm not confident enough with math stuff to review this, so I'd
> rather like to defer to Jeff here.
>
> However, two points:
>
> - Please send the patch as generated by `git format-patch'.  Ideally
>    as a patchset of smaller patches.
>
> - Why the OBSOLETE_MATH stuff?  If these implementations are generic
>    and sufficiently tested, wouldn't it make sense to enable them by
>    default without forcing other targets to enable the OBSOLETE_MATH
>    setting?

Any comment on the difference in code and data size? Just curious
if we are considering it on all targets.

What capabilities does it require of the target architecture? You
mentioned VFP. Does it require vector operations. If something
is assumed, if the minimum capabilities are not met, how well
does this code perform?

> Thanks,
> Corinna

--joel