[Mlir-commits] [mlir] [mlir][spirv] Add definitions and (de)serialization for FPRoundingMode (PR #101546)

[Andrea Faulds]

andfau-amd wrote:

> Should we add some negative tests for when `fp_rounding_mode` is not a `#spirv.fp_rounding_mode` attribute?

Hmm, I'm not sure if it's an important enough case to test. The code that rejects it is very straightforward, I couldn't find any negative tests for similar attributes, and it doesn't seem that likely to me that someone would generate invalid MLIR like that anyway.

One observation I have there is that this is, effectively, validated by the serializer rather than a proper verifier method. I'm not sure if that's good, but it is what the other attributes do too, so...

https://github.com/llvm/llvm-project/pull/101546