[Mlir-commits] [mlir] [mlir][linalg] Enable masked vectorisation for depthwise convolutions (PR #81625)

[Benjamin Maxwell]

================
@@ -2697,6 +2714,7 @@ struct Conv1DGenerator
         return;
       break;
     }
+    hasTensorSemantics = linalgOp.hasPureTensorSemantics();
----------------
MacDue wrote:

You have `op` in that function, is that the linalg op? If so you could do `cast<LinalgOp>(op).hasPureTensorSemantics()` within `maybeMaskXferOp`, rather than add a new member variable. 

https://github.com/llvm/llvm-project/pull/81625