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

[Charitha Saumya]

================
@@ -495,8 +495,7 @@ void LayoutInfoPropagation::visitPrefetchNdOp(
   auto [bWidth, bHeight, bCount] = blockWHC.value();
   SmallVector<int> instData;
   int instWidth = xegpu::getLargestDivisor(
-      static_cast<int>(tdescTy.getDimSize(tdescTy.getRank() - 1)), bWidth,
-      bCount);
+      static_cast<int>(tdescTy.getDimSize(tdescTy.getRank() - 1)), bWidth);
----------------
charithaintc wrote:

just to clarify, for dtype f16 the largest width allowed is 16 not 32 right? to get 32 we have to use bCount = 2. 

I wonder in that case we can not generate a 32 width load with bCount = 1? (we already did in VC). 

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