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

[Joerg Wunsch]

"Ben L. Titzer" <address@hidden> wrote:

> Double precision floating point numbers should be 64 bits--8 bytes,
> on all desktop platforms.

The AVR is not a desktop platform.

Indeed, currently AVR-GCC and avr-libc are using 32-bit "double"
values.

As this violates the C standard, *maybe* this could change in the far
future, but not until someone has donated an efficient 64-bit double
library.  Preferrably, I'd rather provide a compiler option a la
-mint8 even then, to allow for some backwards compatibility.  This
would then offer two choices, like

choice   sizeof(float)  sizeof(double)  sizeof(long double)
#1       4              4               8
#2       4              8               8

Choice #2 complies with the C standard, but breaks compatibility with
all existing code, and is likely to cause even more bloat (due to the
internal default promotion of floating point expressions to type
double).

But again, it requires a volunteer to handle all this...

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