I'm reading the source code for the MLIR side of TritonGPU. I have some more questions I would like to ask.

[linuxlonelyeagle]

Hello everyone, I'm very interested in the triton mlir part but I'm having some difficulties and I hope someone can help me.Thanks!
The following is the code from 2.0.x, but I can't understand it.

class TritonGPU_Attr<string name, list<Trait> traits = [],
                     string baseCppClass = "::mlir::Attribute">
  : AttrDef<TritonGPU_Dialect, name, traits, baseCppClass> {

  let description = [{
TritonGPU Tensors differ from usual tensors in that they contain a _layout_ attribute which determines
how the data should be partitioned across CUDA threads. Formally speaking, we define a layout as a function
\mathcal{L} that maps a multi-dimensional tensor index $i \in \mathbb{Z}^d$ to a set of integers T corresponding
to the indices of the CUDA threads allowed to access some data at index $i$.

For example, let us consider the layout function:
\mathcal{L}(0, 0) = {0, 4}
\mathcal{L}(0, 1) = {1, 5}
\mathcal{L}(1, 0) = {2, 6}
\mathcal{L}(1, 1) = {3, 7}

Then, attaching $\mathcal{L} to a tensor $T$ would mean that:
- T[0,0] is owned by both cuda thread 0 and 4
- T[0,1] is owned by both cuda thread 1 and 5
- T[1,0] is owned by both cuda thread 2 and 6
- T[1,1] is owned by both cuda thread 3 and 7

Right now, Triton implements two classes of layouts: shared, and distributed.
  }];

class DistributedEncoding<string name> : TritonGPU_Attr<name> {
  let description = [{
Distributed encodings have a layout function that is entirely characterized
by a d-dimensional tensor L. Note that L doesn't need to have the same shape
(or even the same rank) as the tensor it is encoding.

The layout function \mathcal{L} of this layout is then defined, for an
index `i` \in R^D, as follows:

\mathcal{L}(A)[i_d] = L[(i_d + k_d*A.shape[d]) % L.shape[d]] \forall k_d such as i_d + k_d*A.shape[d] < L.shape[d]

For example, for a tensor/layout pair
A = [x  x  x  x  x  x  x  x]
    [x  x  x  x  x  x  x  x]
L = [0  1  2  3 ]
    [4  5  6  7 ]
    [8  9  10 11]
    [12 13 14 15]

Then the data of A would be distributed as follow between the 16 CUDA threads:
L(A) = [ {0,8} , {1,9} , {2,10}, {3,11}, {0,8} , {1, 9} , {2, 10}, {3, 11},
         {4,12}, {5,13}, {6,14}, {7,15}, {4,12}, {5, 13}, {6, 14}, {7, 15} ]
  }];

  let extraClassDeclaration = extraBaseClassDeclaration;
}

It's mostly the code in the description that I can't read the meaning of, but of course there are still other classes, but I'm looking at these two at the moment.
The point of the question has to be how to understand what's going on here and how should I think about it?I hope someone can help me, thanks！

[linuxlonelyeagle]

Sorry, I found the answer here #2856 I'd like to know something further, why did you do it?