[Tinyos-devel] external (nx*) types and floating point

[Vlado Handziski]

Yes, 2x word size, not 2 bytes.

Vlado

On Mon, May 5, 2008 at 8:21 PM, David Gay <dgay42 at gmail.com> wrote:

> On Mon, May 5, 2008 at 10:44 AM, Vlado Handziski
> <handzisk at tkn.tu-berlin.de> wrote:
> > Sure, that sounds reasonable. Do we have any use cases for nx_double?
> MSPGCC
> > supports only single precision floats which take 2 bytes.
>
> I assume you mean 4 for single precision floats ;-) I agree that
> double is unlikely to be useful in the short term. Btw, the work in
> doing all this is digging up exactly what floating-point rep avr-gcc,
> msp430-gcc and x86-gcc are currently using.
>
> David
>
> >
> > Vlado
> >
> >
> >
> > On Mon, May 5, 2008 at 7:30 PM, David Gay <dgay42 at gmail.com> wrote:
> >
> > >
> > > On Mon, May 5, 2008 at 9:28 AM, Vlado Handziski
> > > <handzisk at tkn.tu-berlin.de> wrote:
> > > > I think I lost you there. What do you understand under MCU specific
> > > > nx_float/nx_double definitions? How the serialization is implemented?
> > Cause
> > > > the whole point of nx_* was to provide platform independent types, so
> > the
> > > > "semantics" of the type should be the same all over, isn't it?
> > >
> > > The platform-independent rep of floating point types would be the
> > > standard IEEE 32 (or 64) bit floating point formats, in some standard
> > > bit order. The platform-specific code would convert to/from that rep
> > > (I can't think of any efficient, or even practical, way to write that
> > > code in a platform-independent way).
> > >
> > > David
> > >
> >
> >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://www.millennium.berkeley.edu/pipermail/tinyos-devel/attachments/20080505/ec85fafc/attachment.htm