[Mlir-commits] [mlir] Extending UniformQuantizedType with interface-based support for new storage types in Quant dialect (PR #152966)

[llvmlistbot at llvm.org]

Roman-Pevnyi wrote:

Hi all — thanks again for the helpful feedback so far. I’ve been thinking through the open questions, and here’s where things stand:

- Contiguous / sub-byte storage types: I believe the interface should support both contiguous and packed storage models. To move things forward, I’ve implemented a few basic packing methods that should cover common use cases.
- MX types: One motivation here is to better support mixed-precision types. The current interface extension makes it possible to plug in these new storage types more systematically, without duplicating logic.
- Examples: I’ve added examples of float types implementing this interface, which demonstrate the relative ease of adoption. If there are any other blockers or concerns, please let me know — I’d really like to move this toward merge.

Thanks!
@rengolin @javedabsar1 @anuragsingh-tt 

https://github.com/llvm/llvm-project/pull/152966