[Mlir-commits] [mlir] [mlir][ArmSME] Lower transfer_write + transpose to vertical store (PR #71181)

[Cullen Rhodes]

================
@@ -136,13 +136,31 @@ struct TransferReadToArmSMELowering
 
 /// Conversion pattern for vector.transfer_write.
 ///
-///   vector.transfer_write %vector, %source[%c0, %c0] : vector<[16]x[16]xi8>,
-///                                                      memref<?x?xi8>
+/// ---
+///
+/// Example 1: op with identity permutation map to horizontal
+///            arm_sme.tile_store:
+///
+///   vector.transfer_write %vector, %source[%c0, %c0]
+///     {in_bounds = [true, true]} : vector<[16]x[16]xi8>, memref<?x?xi8>
 ///
 /// is converted to:
 ///
 ///   arm_sme.tile_store %vector, %source[%c0, %c0] : memref<?x?xi8>,
----------------
c-rhodes wrote:

horizontal layout is the default and isn't printed, but I could add it to be explicit if you think that's clearer?

https://github.com/llvm/llvm-project/pull/71181