[DPAS]: Use 2d-loads instruction to load the operand of tt.dot

[etiotto]

The operands of the Triton's tt.dot operation should be loaded by using specialized instruction to load 2D blocks of the matrices.
Loading the operands in blocks is more efficient than loading them by using regular loads @llvm.genx.GenISA.LSCPrefetch.

We might need to leverage the semantic information associated with Tritons blocked pointers (https://triton-lang.org/main/getting-started/tutorials/08-experimental-block-pointer.html) in order to generate 2d-Blocked loads.

[tdeng5]

Liyang, please check how cuda handle stride_xx.

[LiyangLingIntel]

Liyang, please check how cuda handle stride_xx.

Triton CUDA pipeline lower 2d-load to TMALoadTiledOp which does not have limitation on the last dim strides.
For Triton XPU pipeline, if we want to leverage GenISA_LSC2DBlockRead, it requires last dim (stride[-1]=1) continuous for each block pointer.

For the first stage, in the pass of 2d load conversion lowering, we will check the stride attr type. If the type is a constant, and meets the 2d block load case, we can leverage genx.matrix.2Dblockload. Otherwise, it will fallback to regular loads. This is the fastest way to enable 2DblockLoad in our pipeline with limited functionalities.

For the second stage, we will consider the dynamic stride case. If the stride attr type is a variable, the lowering strategy is to using a conditional branching to decide the if the last dim stride is 1 at kernel runtime. Then pick the block loads or regular loads.

[vlad-penkin]

@LiyangLingIntel as per our discussion could you please split this ticket by stage

[LiyangLingIntel]

@LiyangLingIntel as per our discussion could you please split this ticket by stage

Sure, I have add 2 issues(#413 and #415) to split this ticket as 2 stages.

[etiotto]

Helping with refactoring and code review.