Adding a backend to LLVM is not an easy thing to do, there was a discussion
in the llvm mailing list some time ago about merging experimental backends,
and even the devs didnt seem to agree too much. They don't like mantaining
backends each time a modification has to be done. I think they proposed to
have new backends in some experimental branch and if after some time they
got positive feedback they would be finally merged to trunk.
The biggest problem I see now is that we would have to merge our out of
tree patches, some of them are very hacky and i know they're not going to
like them.
2013/6/10 Weddington, Eric <Eri...@at...>
>
> > -----Original Message-----
> > From: Stepan Dyatkovskiy [mailto:stp...@na...]
> > Sent: Monday, June 10, 2013 12:03 PM
> > To: John Myers; Borja Ferrer; avr-llvm
> > Subject: [avr-llvm-devel] May be join to LLVM?
> >
> > Guys,
> >
> > I propose to make your back-end stable and join to LLVM project then.
> > Yes, just add your back-end to LLVM. There are few advantages:
> > 1. You won't synchronize it anymore. If somebody will brake AVR
> > backend,
> > then, according to LLVM policy: it *his* blame, and he will fix it, not
> > you and me. You will get some kind of protection.
> > 2. Being in main llvm repo, AVR backend will attract more developers
> > and
> > testers, more people that may be interested in your code.
> > 3. You lose nothing in that case. Benefits only.
> > 4. You will still code owners, of course.
> >
> > What do you think?
> >
>
> Well the thought was crossing my mind this morning...
>
> The only reason to have a separate project was to develop the AVR backend
> sufficiently enough that it worked. The intent, if I recall, was always to
> merge with LLVM at some point (to be determined). Are we at that point now
> that it will benefit us to merge with LLVM?
>
> Eric
>
|
