Re: math, math, math...

[Andreas Jaeger <aj at suse dot de>]

Ulrich Drepper <drepper@redhat.com> writes:

> You might have seen the batch of changes which I added yesterday.
> These sources are mainly Intel's hand-written, optimized, and fairly
> accurate libm for IA-64.  I've checked in all the code which already
> works, some files have errors.  I'll do accuracy measurements using
> the programs I wrote several years ago once I have a bit of time.
> 
> THe other change which went in was a 96bit long double implementation
> of j1.  Stephen Moshier is hopefully providing appropriate versions of
> the other missing functions are well (and perhaps 128bit versions).

Excellent!

> But that's not all.  I have another giant patch waiting here.  It's a
> complete rewrite of quite a lot of math functions contributed by IBM.
> Their guys in Haifa are really today's elite in math library
> implementations.  The functions have last bit accuracy, no bit wrong.
> As for the Intel libm integrating this one will be tricky.  The
> structure is different from what we expect.  I cannot do it right now.

Do you have any documentation about the functions or can you provide
me with the current version of the patches?

> After yesterday's partly incomplete patch things are back to normal
> now.  I've successfully compiled the current veriso on x86, IA64,
> Alpha, PPC, and Arm.

Thanks,
Andreas
-- 
 Andreas Jaeger
  SuSE Labs aj@suse.de
   private aj@arthur.inka.de
    http://www.suse.de/~aj