Re: math library for ARM?

[Ian Lance Taylor <ian at zembu dot com>]

   Date: Thu, 16 Dec 1999 17:41:09 +0800
   From: Bill Su <ycsu@zyxel.com.tw>

     I have used arm-elf from the development version of binutils.
   It can be read if the output is stripped.  Thanks for help. Now
   I have a problem on floating library.  I use newlib from Cygnus.
   However, it's very slow on whetstone and my project is timing
   critical on floating point performance.  Is there any better
   floating library?  I saw the libm in NetBSD/OpenBSD comes from
   SunPro.  Is it better?  Where can I get the archive?  Thanks.

The newlib sources are also based on fdlibm, just like the
NetBSD/OpenBSD sources.

Are you using hardware floating point?  If you are not, then the basic
floating point operations--addition, multiplicaton, etc.--are being
done by code in gcc.  That code (in gcc/config/fp-bit.c) is generic,
and slow.

I note that the default for arm-elf appears to be software floating
point.  If your ARM chip has hardware support for floating point
operation, try compiling and linking with -mhard-float.  Make sure
that the version of newlib you link with has been compiled with
-mhard-float (this may happen automatically thanks to the magic of
multilib).

Ian

------
Want more information?  See the CrossGCC FAQ, http://www.objsw.com/CrossGCC/
Want to unsubscribe? Send a note to crossgcc-unsubscribe@sourceware.cygnus.com