[Mlir-commits] [mlir] [mlir][SME] Add vector.splat -> SME conversion (PR #67659)

[Cullen Rhodes]

================
@@ -240,6 +240,63 @@ struct BroadcastOpToArmSMELowering
   }
 };
 
+/// Conversion pattern for vector.splat.
+///
+/// Example:
+///
+///   %splat_to_tile = vector.splat %src : i32 to vector<[4]x[4]xi32>
+///
+/// is converted to:
+///
+///   %broadcast_to_1d = vector.broadcast %src : i32 to vector<[4]xi32>
+///   scf.for %tile_slice_index = %c0 to %num_tile_slices step %c1 {
+///     arm_sme.move_vector_to_tile_slice %broadcast_to_1d, %tile,
+///       %tile_slice_index : vector<[4]xi32> into vector<[4]x[4]xi32>
+///   }
+///
+/// This should, in practice, be identical to vector.broadcast when
+/// broadcasting a scalar.
----------------
c-rhodes wrote:

It's identical, I think the ambiguity can be removed
```suggestion
/// This is identical to vector.broadcast of a scalar.
```

https://github.com/llvm/llvm-project/pull/67659