[rfc] add real*12 and real*16

[Paul Brook]

On Tuesday 24 August 2004 19:33, Richard Henderson wrote:
> Well, basically revamp how a lot of builtin type stuff happens.
> Of note are:
>
>   (1) Real kinds come from formats found from machine modes and real.c.
>       We're now polling the actual format, rather than hard-coding IEEE.
>
>   (2) That means we find XFmode and TFmode, if the target mentions them.
>
>       Note that this is not quite 100% the same as *supporting* them.
>       A good example here is i386, which doesn't support TFmode, but the
>       compiler is built with amd64 support, which does use TFmode.  We
>       need a bit better information from the backend to fix this.
>
>       I arbitrarily chose real*12 for IEEE extended format.
>
>   (3) Lots of cleanup in trans-types.c.  IMO there's lots more to be
>       done as well.  We ought to do something similar for integer types.
>       For instance, c4x doesn't support 8-bit types, so we shouldn't
>       advertise that we have one.
>
> I'm running out of time this morning, so this is just a dump of the
> patch so far.  It does work, as far as it goes, just needs a bit of
> cleanup.  And the extra info from the backend that I mentioned.
>
> Comments?

Looks good to me.

We also want to look again at the implementation of intrinsic functions. Eg. 
the mapping of kind 4 and 8 to sinf and sin, and what to do with other types. 
That's a whole extra issue though.

Paul