[Mlir-commits] [mlir] [mlir][tosa] Add add/sub/mul/div_floor/div_ceil_shape ops (PR #169321)

[Luke Hutton]

================
@@ -53,6 +46,29 @@ class Tosa_ElementwiseShapeOp<string mnemonic, list<Trait> traits = []>
 }
 
 
+//===----------------------------------------------------------------------===//
+// Operator: AddShape
----------------
lhutton1 wrote:

You're correct, this is a valid alternative approach. However, I believe there was concern about the separation of operator semantics with this (using tosa.add) approach. Broadcast was likely one of them, as you mentioned, another might be ensuring CTC status of the inputs, etc. @eric-k256 can likely go into more details here

https://github.com/llvm/llvm-project/pull/169321