[Mlir-commits] [mlir] [MLIR][LLVM] Add `llvm.experimental.constrained.fptrunc` operation (PR #86260)

[Victor Perez]

victor-eds wrote:

> > I guess we could have some kind of argument for the base intrinsic class in a similar way to immediate values, but that'd require further modifications. I don't think it'd be 100 % automatic anyway...
> 
> A solution where the arguments are defined in a derived class being the actual intrinsic or some unary / binary etc base class could be nice. Especially since that may allow us to properly type the intrinsic operands etc. However, with tablegen it is really hard to guess how complex things are in the end. If you have a good idea you are very welcome to give it a shot.
> 
> Otherwise it is also ok to land and follow up if needed (e.g. after adding more intrinsics).

Yes, that's more or less what I was thinking of. I'm depending on this PR for further work in other dialects, so I'd rather get this merged and give this a shot in the future. I'll wait till CI passes. Thanks a lot!

https://github.com/llvm/llvm-project/pull/86260