[Mlir-commits] [mlir] [MLIR][XeGPU] Add wg-to-sg distirbution for dpasmx, bitcast, interleave, and deinterleave (PR #194985)

[Charitha Saumya]

================
@@ -1403,19 +1462,116 @@ struct WgToSgVectorMaskOp : public OpConversionPattern<MaskOpType> {
 
 using WgToSgVectorConstantMaskOp = WgToSgVectorMaskOp<vector::ConstantMaskOp>;
 using WgToSgVectorCreateMaskOp = WgToSgVectorMaskOp<vector::CreateMaskOp>;
+
+// This pattern transforms vector.bitcast ops to work at subgroup level.
+struct WgToSgVectorBitCastOp : public OpConversionPattern<vector::BitCastOp> {
+  using OpConversionPattern<vector::BitCastOp>::OpConversionPattern;
+
+  LogicalResult
+  matchAndRewrite(vector::BitCastOp op, OneToNOpAdaptor adaptor,
+                  ConversionPatternRewriter &rewriter) const override {
----------------
charithaintc wrote:

shouldn't we check if the bitcast require shuffling data across subgroups here? (and maybe avoid handling that)

https://github.com/llvm/llvm-project/pull/194985