[Mlir-commits] [mlir] [mlir][ptr] Add nontemporal field to ptr.masked_load (PR #204987)

[Fabian Mora]

fabianmcg wrote:

> This is indeed a bit confusing. The nontemporal flag is just metadata so nothing prevents us from adding it to masked loads / stores. However, it is indeed a good question if this is something that the LLVM backends take into account?
> 
> I see some SVE tests that use nontemporal metadata but it is unclear to me if that is intentional or just a random artifact of these tests.

I would argue the even if some backends support the metadata for optimizations, it's not a first class member and as such it shouldn't live as intrinsic attributes in the op. However, I'm not opposed to amend translation such that if it's found in the discardable attributes it gets passthrough LLVM. Having said that, IMO that's also fragile. I would try to rally LLVM to add the nontemporal flag as first class or get more information from LLVM before making a change.


https://github.com/llvm/llvm-project/pull/204987