Re: [avr-gcc-list] linear regression on 128L

[Joerg Wunsch]

"Daniel O'Connor" <address@hidden> wrote:

> Surely if double =3D=3D 4 bytes and float =3D=3D 4 bytes then double
> is wor= thless...

Many people obviously don't find it worthless. ;-)

The point is, C basically depends on ``floating point support'' being
``type double supported''.  If all you have is a 32-bit FP library,
you can then decide whether you don't want to offer *any* FP support,
or whether you ship with 32-bit FP being declared as `double'.

> Is there some hidden cost for double =3D=3D 8 that doesn't affect
> people us= ing=20 doubles?

Yes, C assumes on all ends that the default FP type is double.  FP
expressions promote to it, much the same way integer expressions
promote to `int' (which has already caused the infamous invention of
the -mint8 compiler option).  *Many* people wouldn't want to pay for
the bloat.

Also, remember that a 64-bit FP library for the AVR is still
vapourware (at least one in a form that could be integrated smoothely
with the compiler).  *Once* we *might* have it, but then, the
installed base which assumes 32-bit doubles has to be considered as
historical baggage to care for.

-- 
J"org Wunsch                                           Unix support engineer
address@hidden        http://www.interface-systems.de/~j/