[Mlir-commits] [mlir] [mlir][OpenACC][OpenMP] Modify atomic capture to allow update/write (PR #167963)

[Erich Keane]

erichkeane wrote:

> > I think it is supposed to be an update + read in the cases I was concerned about in OpenACC as well
> 
> OK - I agree - the `v = x` is a read not a write. I also misunderstood this when I reviewed your initial proposal. I was hyper-focused on the fact that in Fortran we did not have `x++` so I concluded that there was a chance we did not have complete support in the representation. Thank you for taking a deeper look.

Ok, thanks for confirming! I was messing around with it a bunch today and convinced myself I was wrong when I asked about it, and figured I must have talked you into it :)   

I was so focused on the other components/how weird it looked, I got distracted.  Fortunately while writing my tests to use it today I realized it was strange that there wasn't a 2-statement form of it, until it hit me :D 

https://github.com/llvm/llvm-project/pull/167963