[Mlir-commits] [mlir] [MLIR][XeGPU] Introduce `xegpu::uArch` usage in target-sensitive passes (PR #163801)

[Charitha Saumya]

================
@@ -456,7 +482,22 @@ void LayoutInfoPropagation::visitPrefetchNdOp(
   // Here we assign the default layout to the tensor descriptor operand of
   // prefetch.
   auto tdescTy = prefetch.getTensorDescType();
-  auto prefetchLayout = getDefaultSIMTLayoutInfo(tdescTy);
+
+  auto uArch = getUArch(getChipStr(prefetch).value_or(""));
----------------
charithaintc wrote:

I think rather than defaulting to a certain arch, its better to fail the propagation early on if module lacks the arch info. 

Assuming the arch can have some unintended effect for the user. We should let them know as early as possible that they are doing something wrong by not providing us an arch. 

https://github.com/llvm/llvm-project/pull/163801