This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: real.c on unicosmk
- From: Zack Weinberg <zack at codesourcery dot com>
- To: Stephen L Moshier <steve at moshier dot net>
- Cc: Roman Lechtchinsky <rl at cs dot tu-berlin dot de>, gcc at gcc dot gnu dot org
- Date: Wed, 22 May 2002 10:43:25 -0700
- Subject: Re: real.c on unicosmk
- References: <Pine.LNX.4.33.0205221311470.25146-100000@moshier.net>
On Wed, May 22, 2002 at 01:25:32PM -0400, Stephen L Moshier wrote:
>
> > Bonus points if you can persuade EMUSHORT to be 16 bits on 32-bit
> > architectures and 32 bits on 64-bit architectures (this would be most
> > efficient on both).
>
> You can get a 32-bit version of the arithmetic from
> www.netlib.org/cephes/qfloat.tgz.
> It runs about twice as fast as the 16-bit version.
Thanks, I'll check that out.
> That would not solve the problem that EMUSHORT is used for
> packing and unpacking data formats.
This is the requirement that Roman and I are trying to get rid of.
> There doesn't seem to be any reason why REAL_ARITHMETIC is required
> in alpha.{c,h,md} (I'm looking in gcc-3.1).
In 3.2, REAL_ARITHMETIC is unconditionally on all the time. It's
amazing how much crap vanishes when there is only one supported method
of doing floating-point calculations.
I don't think any of that made it into 3.1.
zw