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

[Zoran Zomborat]

ZoranZomborat wrote:


> Misc. point: If someone brings up 6‑bit (`fp6`) formats those generally require a composite packing scheme (ex: 4+2 bit streams) and don’t divide 8 cleanly. That case argues for either (a) a richer packing descriptor than just `elementsPerByte`, or (b) modeling fp6 as a composite/“MX” type (as discussed above) rather than a single simple carrier. So these helpers might be useful for the common divisors {1,2,4,8} but outliers would need a separate path.

Sounds good, and yeah FP6 seems hard for sure; not only do you need a combination of 4+2 bit packing; you'd also need to know at which point in the fp6 sequence you are so to correctly decode:
[[6+2],[4,4],[2,6] ....] definitely a head scratchier;
As you said also int2 is getting a lot of traction and there are more and more 4 bit float variants;
Even if HW has some native int6, fp6 processing, I see a lot of challenges into getting this functional and optimal.

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