[Mlir-commits] [mlir] [mlir] Add FP software implementation lowering pass: `arith-to-apfloat` (PR #166618)

[Maksim Levental]

makslevental wrote:

> but perhaps that already works (although messing up rounding is easy) 

extf doesn't work for sure (dunno about truncf) - I know because last week before @matthias-springer proposed those I ran into the obviously wrong `llvm.extf(%int8)` or something like that when trying to naively lower.

> or perhaps as you are focusing not on heterogeneous floating point computations (yet).

Need to check if my use case calls for this (I actually think it does). Anyway I plan to send more PRs after this one for more of the arithmetic operations.

https://github.com/llvm/llvm-project/pull/166618