[clang-tools-extra] [mlir] [MLIR][XeGPU] Refactor Layout access interface (PR #172125)

[Tuomas Kärnä via cfe-commits]

================
@@ -20,18 +20,20 @@ gpu.module @xevm_module{
     %0 = xegpu.create_nd_tdesc %arg0 : memref<8x16xf16>
       -> !xegpu.tensor_desc<8x16xf16, #xegpu.layout<lane_layout = [1, 16], lane_data = [1, 1]>>
     %1 = xegpu.load_nd %0[%c0, %c0]
-      {layout_result_0 = #xegpu.layout<lane_layout = [1, 16], lane_data = [1, 1]>} :
+      {layout = #xegpu.layout<lane_layout = [1, 16], lane_data = [1, 1]>} :
----------------
tkarna wrote:

The tests seem to be using a mixture of `{layout = ...}` and `<{layout = ...}>` format which I find a little confusing. The former format suggests just an arbitrary attribute applicable to any op, whereas here we mean the anchor layout that is only supported by certain ops.

https://github.com/llvm/llvm-project/pull/172125