[Mlir-commits] [mlir] [mlir][ArmSME] Add arith-to-arm-sme conversion pass (PR #78197)

[Benjamin Maxwell]

================
@@ -42,6 +49,16 @@ std::optional<ArmSMETileType> getSMETileType(VectorType);
 /// Verifies the tile ID (if set) on this tile operation is valid.
 LogicalResult verifyOperationHasValidTileId(Operation *);
 
+using LoopBodyBuilder =
+    std::function<void(OpBuilder &, Location, Value, Value)>;
+
+/// Generates a for loop over ZA tile slices where the induction variable is
+/// the tile slice index and each iteration yields a new tile. Loop body is
+/// built via LoopBodyBuilder, which returns the next tile value.
----------------
MacDue wrote:

This does not automatically insert the yields anymore. I'm not sure why though as all callers now manually add them?

https://github.com/llvm/llvm-project/pull/78197