Kunwar Grover
Kunwar Grover
Not sure where to add tests for checking if serialization works
> Isn't thread id Added here same as GPU thread id? It's fine to have this, but it seems redundant. You could also add subgroup id to the GPU dialect...
> This is just deferring the lowering to ids. Is there anything we want to do with these operations? Yes, this just defers the lowering. No plan to do anything...
I decided this patch is not worth doing. It's simpler to just do an affine.apply
I have proofs that this is equivalent to the basis form and how the bidrectional `tid` -> `virtual tid` mapping works. I will include them in docs once I have...
> batches_per_subgroup = [1, 4, 0] Looks like the config setting the layout for contraction messed up there. I can have a look tomorrow and send a fix.
Can you give me the original mlir file? I can at least add a failure when there is no intrinsic available for this shape.
If this bot is in > 100 servers we have other things to worry about: https://support.discord.com/hc/en-us/articles/360040720412-Bot-Verification-and-Data-Whitelisting#what-if-im-already-in-100-guilds
Not planned anymore
Stale PR, already done