[Mlir-commits] [mlir] [mlir][ArmSME] Fold MoveTileSliceToVector + TransferWrite to StoreTileSlice (PR #95907)

[Cullen Rhodes]

================
@@ -666,14 +666,64 @@ struct VectorPrintToArmSMELowering : public OpRewritePattern<vector::PrintOp> {
   }
 };
 
+/// Folds a MoveTileSliceToVectorOp + TransferWriteOp to a StoreTileSliceOp.
+///
+///  BEFORE:
+///  ```mlir
+///  %slice = arm_sme.move_tile_slice_to_vector %tile[%index]
+///             : vector<[4]xf32> from vector<[4]x[4]xf32>
+///  vector.transfer_write %slice, %memref[%i, %j], %mask {in_bounds = [true]}
----------------
c-rhodes wrote:

the pattern doesn't check `in_bounds`

https://github.com/llvm/llvm-project/pull/95907