[Mlir-commits] [flang] [mlir] [RFC][mlir] Conditional support for fast-math attributes. (PR #125620)

[Slava Zakharin]

vzakhari wrote:

> Can you link to some example that shows why this is necessary? I'd think that `arith.select` preserves bitpatterns of its operands, so I struggle to see why it needs fast math flags; if llvm needs them, couldn't it calculate them as the intersection of the fast math flags of the operands?

I am sorry, I do not have examples readily available for you.

There is LLVM Floating Point Working Group (https://discourse.llvm.org/t/floating-point-working-group/76907/10) that discussed the need for fast-math flags on `select` instructions here: https://github.com/llvm/llvm-project/issues/51601 (see also notes from February 21, 2024 in https://docs.google.com/document/d/1QcmUlWftPlBi-Wz6b6PipqJfvjpJ-OuRMRnN9Dm2t0c/edit?tab=t.0#heading=h.k3uvggph248w

The initial addition of fast-math support for `select` was done in 2019: https://github.com/llvm/llvm-project/commit/5a4f7cf2ff3fc15d82ca062ee64dd8bd01a68883

https://github.com/llvm/llvm-project/pull/125620