Re: Cortex M0 Floating Point Library

On Tue, Nov 13, 2018, at 1:19 AM, Emmanuel Blot wrote:
> Hi Daniel,
>
> Daniel Engel wrote on 12/11/2018 19:03:
> > Hi,
> >
> > Over the past couple of years, I have hand-assembled a new floating
> > point library for the ARM Cortex M0 architecture.  Last week I
> > started a thread on the GCC mailing list about contributing this
> > library as open source.
>
> > If you have any comments on this direction, please advise.
>
> I think it would be nice if it could integrate smoothly with
> LLVM/clang as well. I'm thinking about the compiler-rt library, which
> I think partially covers this topic. It's support for ARMv6m is quite
> shallow AFAICT.
>
> Thanks, Emmanuel.

I would be open to that.  I'm much more familiar with the "Arm Embedded Toolchain", but it looks as if compiler-rt tries to follow the libgcc specification.  Internally there are some significant style differences in the function preambles and local label definitions.  I'm not sure how particular each project is to this kind of style.  

At a glance, the generic "arm" assembly functions in compiler-rt should apply to cortex M0.  However, taking addsf3 as an example, it appears to occupy ~300 bytes as opposed to the ~120 bytes of my library.  I did not see assembly versions of mulsf3/aeabi_fmul or divsf3/aeabi_fdiv.  (My library also has a fixed overhead of ~150 bytes, shared by all floating point functions.)

I have not posted code yet because I have not discussed copyright assignment options with the FSF.  However, it seems that dual-licensing is fairly common with this sort of thing.  

Regards,
Daniel Engel