[llvm] [LLVM][LangRef] Remove "n > 0" restriction from get.active.lanes.mask. (PR #152140)

[Sjoerd Meijer via llvm-commits]

sjoerdmeijer wrote:

> See https://reviews.llvm.org/D86147 and https://reviews.llvm.org/D79175 for the original discussion of the restriction. CC @sjoerdmeijer @sparker-arm .
> 
> From what I recall, there's some interaction with MVE tail predication that breaks with n=0. But I haven't thought about this for a long time, so I'm not sure if that's still true.

Yep, that's right, we rely on this behaviour for MVE and tail-predicated hardware-loops. 
With n=0, the loop wouldn't be executed, but incorrectly INT_MAX times, or something like that. But it's been a long time since I looked into MVE. I am hoping @davemgreen or @sparker-arm know more details from memory, otherwise we need to look further into this. 

https://github.com/llvm/llvm-project/pull/152140