This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: real.c on unicosmk


On Thu, May 23, 2002 at 11:00:26AM +0200, Roman Lechtchinsky wrote:
> On Wed, 22 May 2002, Zack Weinberg wrote:
> 
> > I don't see any good reason.  The ereal_from_* functions still have to
> > care about endianness, but I don't think that's your problem -- you're
> > talking about etoeXX and eXXtoe, right?
> 
> Right. ereal_from_*, ereal_unto_* and etar* depend on endianness already,
> so this won't add a lot of code to these functions. But it will simplify
> etoe* and e*toe. I think I will prepare a patch for this first.

That's a good idea.  Small patches are always easier to get applied.

There's a *lot* of duplicated code in the e*toe, etoe*, and to*
functions.  If you feel like refactoring it, that would be worthwhile
too.

> Hmm, actually, the external representation is little endian at the moment
> while the internal is big endian. We might as well be consistent and use
> big-endian throughout. In fact, using big endian for the significand and
> storing the exponent in the *last* 16-bit word (or the lower 16 bits of
> the last 32-bit word) will also simplify emovo and emovi in the 32/64 case
> - they have to shift the significand by 16 bits now. If you don't have any
> objections, I'll include this change in the patch mentioned above.

Yes, please.  Anything you can do to make real.c simpler and easier to
understand is worth doing.

zw


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]