[PATCH] D20325: Add ARM cdp intrinsics

[Tim Northover via cfe-commits]

On 17 May 2016 at 10:32, Renato Golin <renato.golin at linaro.org> wrote:
> I don't think a small future ACLE builtin is on the same league as a whole new back-end. :)

Well yes, it's probably got orders of magnitude less bugs than the
backend for a start.

> Nor I think that having this when the docs are public, rather than now, will affect users in any way.

True. It's an extremely niche feature.

> It's been our stance for a long time to require docs to approve changes, however small. I don't want to relax that which I think is a good constraint, not for such a seemly irrelevant issue.

The only place I've seen that policy in play is with asm aliases
(which have a habit of getting documented nowhere) and even there we
generally grudgingly accept promises to document.

Generally we're far more relaxed as long as a specification is on the
way. We're up to to v8.2 in LLVM proper already, and I don't think
even the v8.1 specification is public (all I can find is a blog
announcing it).

> Or, maybe I am mistaken, and this is really that important... is it?

Goodness no! The entire coprocessor instruction space was
deprecated/reclaimed in v8 as far as I'm aware. But since I don't
think timing matters particularly either way and ARM themselves are
the ones who will be supporting the majority of these users, it's
pretty harmless for us to fall in with their desired schedules.

</waffle mode>

Well, that's how I view these kinds of things anyway.

Tim.