[Mlir-commits] [mlir] [MLIR][XeGPU] Disable block count usage in layout propagation (PR #168504)

[Jianhui Li]

================
@@ -37,6 +37,18 @@ def XeGPUPropagateLayout : Pass<"xegpu-propagate-layout"> {
     propagate the layouts required for their operands to the producers. With
     this propagated layout information, pass will then update op result type
     with the layout information.
+
----------------
Jianhui-Li wrote:

Can we put a simple description for both `inst` and `lane`? 
Consider the following:  

`inst`
Propagate the `inst_data` field of the layout attribute. The default is chosen to maximize instruction-level granularity so that the user shape can be processed with the fewest instructions.
For N-D operations, this granularity depends on W (width) and H (height) of the instruction shape. The B (block) dimension—i.e., the array length—is not included in the default configuration and must be enabled via a separate optimization pass.

`lane`
Propagate the `lane_layout` and `lane_data` fields of the layout attribute. Default values are selected to align with hardware expectations.

https://github.com/llvm/llvm-project/pull/168504