[Mlir-commits] [mlir] [mlir][Arith] Add FTZ (Flush-to-Zero) fast-math flag (PR #112700)

[Mehdi Amini]

joker-eph wrote:

LLVM often uses target intrinsics instead for this kind of “exotic” attributes, especially when it can’t be supported consistently across target.
Another aspect is how much the attribute would affect other transformations on this op than just lowerings.

I am not sure what’s the right thing to do here, but I wanted to mention about tradeoffs.

https://github.com/llvm/llvm-project/pull/112700