[Mlir-commits] [mlir] [mlir][linalg] Type conversion of operands in new elementwise-op. (PR #131542)

[Andrzej Warzyński]

================
@@ -563,13 +563,16 @@ def ElementwiseOp : LinalgStructuredBase_Op<"elementwise", [
     The number of dims of the iterator-types are inferred from the rank of
     the result type.
 
+    Numeric casting is performed on the input operand, promoting it to the same
+    data type as the result.
+
     Example:
 
     Defining a unary linalg.elemwise with default indexing-map:
     ```mlir
     %exp = linalg.elemwise
         kind=#linalg.elemwise_kind<exp>
-        ins(%x : tensor<4x16x8xf32>)
+        ins(%x : tensor<4x16x8xf16>)
----------------
banach-space wrote:

So you have changed this example so that there's casting. But what kind of casting? And why is it crucial? It would be good to expand docs.

https://github.com/llvm/llvm-project/pull/131542