[llvm] r216523 - InstCombine: Optimize GEP's involving ptrtointbetter

[Renato Golin]

On 28 August 2014 23:52, David Majnemer <david.majnemer at gmail.com> wrote:
> I'd be all for reverting it if we had LLVM IR that we can commit with the
> revert which shows the problem.
> Better yet, I could fix the bug if I had the problematic IR.

I'm not sure that's how it works...

I have limited time to investigate the problem and bisecting and
debugging on ARM is not particularly easy, as I'm sure is the same on
MIPS. Lots of bots are broken because of this patch for several days
already. Normally, when that happens, whomever committed the patch
reverts it and works with the affected people (me, Doug, Nuno) to
solve the issue, while the bots are back to green.

Since it might be a while to investigate this problem on the diverse
architectures, keeping the bots red, for what could be several more
days, for a not that critical patch is poor judgement, indeed. This
simplification can give you some edge on some corner case but it's not
at all worth having multiple bots red because of it.

cheers,
--renato