Re: AW: [avr-gcc-list] Double desirability

[Joerg Wunsch]

"Haase Bjoern (PT-BEU/EMT) *" <address@hidden> wrote:

> I have doubts, whether one could succeed in realizing similar
> efficiency with plain C functions. However, one might stand chances
> to succeed when using inline assembly.

> In my opinion, one might even consider to integrate the floating
> point support into gcc itself.

The FP basics are already implemented inside GCC right now.  I didn't
look how it's done, I assume it's just compiled into libgcc.a, using
GCC's generic code (which is IMHO semi-automatically generated C
code).

The fplib part of avr-libc only affects standard functions (like the
trigonometrical functions etc).

> This would also resolve the bug, that
> fp support is broken when forgetting to link with -lm.

We (the avr-libc developers) already basically agreed that libm.a is
scheduled to disappear in the long run.  The division into libc.a and
libm.a probably made some sense on the PDP-11, but it makes no sense
at all in avr-libc, in particular given that some functions declared
in <stdlib.h> are actually being implemented in libm.a.

In order to not break old Makefiles, we'll probably ship a dummy
libm.a then.

My intention is to make this move when we finally adopted a scheme to
go for per-device libraries.

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