[Mlir-commits] [mlir] [MLIR][Arith] Add denormal attribute to binary/unary operations (PR #112700)

[Diego Caballero]

dcaballe wrote:

> VFP is IEEE compliant while NEON is not (has denormals)

That's interesting. Does this mean that if the scalar code is IEEE compliant and we are under a precise fp mode, the code is not vectorize?

I'm also wondering what the interaction between FMFs and denormal models is. Which flags would allow changing an ieee denormal mode to ftz, for example? I think it's important that we document that.

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