[llvm-commits] PATCH: Add a few intrinsic definitions for r600

[Tom Stellard]

On Tue, Mar 20, 2012 at 03:33:04PM -0700, Chandler Carruth wrote:
> On Tue, Mar 20, 2012 at 3:16 PM, Jim Grosbach <grosbach at apple.com> wrote:
> 
> > What's the benefit of adding the intrinsics to trunk llvm if the backend
> > itself is going to remain separate?
> 
> 
> I think this is the key point -- there is no particular benefit, and in
> fact there is a cost because we won't be able to properly test them.
> 
> That said, I think Tom's concern is both valid, and addressable. We should
> just work to get this backend into the tree.
> 
> 
> You essentially *always* have the ability to make *your own* patch release
> of LLVM. If you contribute your backend, and it lives in-tree, and you, as
> a primary consumer, need to patch a release, you can create a
> 'mesa_3.1.1_release_branch' or whatever, and patch it.
> 
> The question is whether you need the typical round of release testing,
> qualification, and pushing to various distros etc that happen for LLVM's
> normal release. If not (that is, if you can directly use the patch LLVM
> code from that branch), then you're done. It sounds like this is the state
> you're in, as otherwise having your backend out-of-tree wouldn't help?
> 
> 
> If, for some reason, you need all the distros to pick up the patched
> version, then it's a bit harder. That would mean you essentially need to
> take on the effort of doing the qualification and pushing of the point
> release. I'm not sure you have the resources, but if you're the only one
> who needs the point release I think it would fall on you to do whatever was
> needed to get it out the door.
> 
> 
> I don't think there is any objection to someone doing either of these if
> they both need it and are willing to do the leg work. It sounds like the
> answer to both for you is "yes". If so, I vote we start working on getting
> the backend into the tree, and that will make patches such as this entirely
> reasonable.

Thanks for the response, I think we have enough options available to
solve the 'stable release' problem.  I'm not sure exactly what we will
end up doing, but I think we can make it work.  I'll start trying to
clean up the backend and get it ready for inclusion into the LLVM tree.

Thanks,
Tom