[Mlir-commits] [mlir] [MLIR][XeGPU] Scattered ops sg-to-wi distribution (PR #154949)

[Charitha Saumya]

charithaintc wrote:

> That is the point, the offsets are not
> 
> > %offsets = arith.constant dense<0> : vector<16xindex>
> 
> They can be arbitrary, how does the proposed distribution work with a vector of `[0, 5, 2, 11, 1]` instead of sg-uniform value?
> 
> UPD: I see, let me think about it

I feel such random offsets are more likely to come as a memory buffer and not likely to be static at SG level. 

https://github.com/llvm/llvm-project/pull/154949