================
@@ -168,6 +173,12 @@ class OpLowerer {
         }
       } else if (IsVectorArgExpansion) {
         Args = argVectorFlatten(CI, OpBuilder.getIRB());
+      } else if (F.getIntrinsicID() == Intrinsic::dx_dot2add) {
----------------
farzonl wrote:

You are correct this is a one off. That said I kind of like the placement 
because it keeps things within the same spatial locality to the 
`IsVectorArgExpansion` cases of  `argVectorFlatten`.

 The only difference with this version is that the implementer of the `dot2add` 
spec decided to change the argument ordering o so now we can't reuse the same 
code we had for dot product and instead need to fetch the last argument so we 
can process it first.

Moving it out of here feel like we would lose the similarity context of these 
two implementations.

https://github.com/llvm/llvm-project/pull/131237
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to