release_35 patches for unroll pragma

[Mark Heffernan]

On Mon, Aug 4, 2014 at 10:35 AM, Aaron Ballman <aaron.ballman at gmail.com>
wrote:

> On Mon, Aug 4, 2014 at 1:33 PM, Mark Heffernan <meheff at google.com> wrote:
> > On Mon, Aug 4, 2014 at 10:24 AM, Aaron Ballman <aaron.ballman at gmail.com>
> > wrote:
> >>
> >> On Mon, Aug 4, 2014 at 1:22 PM, Eric Christopher <echristo at gmail.com>
> >> wrote:
> >> > Any reason that we need them in 3.5? Correctness?
> >>
> >> My only concern is that the feature is partially in 3.5, but a
> >> user-facing part of that feature was changed once the freeze happened.
> >> Eg)  #pragma clang loop unroll(enable) became  #pragma clang loop
> >> unroll(full)
> >
> >
> > That's my primary concern as well.  Having one release with one
> particular
> > syntax, then switch it to something else for the next release is not
> great.
> > All-in-all I'd probably prefer not supporting the unroll pragma at all in
> > 3.5 than have a (slightly) buggy one whose syntax will change.  However,
> > rolling back support completely would be a bigger change than these
> patches.
>
> An alternate option would be to update the documentation to remove
> mention of the feature. That's a much smaller change. ;-)
>
> ~Aaron
>

If having a stealth feature like that is reasonable, I'm happy to remove
mention of it from the docs.  More specifically any mention of the
following would be removed: '#pragma unroll', '#pragma clang loop unroll',
'#pragma clang loop unroll_count', and 'llvm.loop.unroll.*' metadata.

Mark
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20140804/a931c61d/attachment.html>