[m-rev.] for review: Add news about data representation and float boxing changes

[Peter Wang]

On Mon, 29 Apr 2013 13:31:48 +1000, Paul Bone <paul at bone.id.au> wrote:
> On Mon, Apr 29, 2013 at 12:46:24PM +1000, Peter Wang wrote:
> > Plus one more, which was disabled on 11.07:
> 
> All the changes seem fine to me.
> 
> > On Mon, 29 Apr 2013 12:36:41 +1000, Peter Wang <novalazy at gmail.com> wrote:
> > > +
> > > +* Double-precision `float' constructor arguments are now stored in
> > > +  unboxed form on 32-bit architectures in the low-level C grade.
> > > +  This was previously so for 64-bit architectures.
> > > +
> 
> Maybe I didn't realize when this was implemented, but does this mean that a
> double-precision 'float' takes two 'slots' in the structure?

Yes.

> Do we control
> packing at all or is this done by the C compiler?

We do it.  IIRC for high-level C we generate structure definitions
containing MR_Float.

Peter