Add cmul and cdiv for float16 type

zhongkaifu · Oct 8, 2024 · b3b8fed · b3b8fed
1 parent 7c93f7a
commit b3b8fed
Showing 1 changed file with 3 additions and 0 deletions.
diff --git a/TensorSharp.CUDA/DeviceCode/ElementwiseOpKernels.cs b/TensorSharp.CUDA/DeviceCode/ElementwiseOpKernels.cs
@@ -64,6 +64,9 @@ private static string GetFullCode()
                 AppendTTSHalfFunc(result, "mul", "__hmul");
                 AppendTTSHalfFunc(result, "div", "__hdiv");
                 AppendTTTFunc(result, "cadd", "__hadd", DType.Float16);
+                AppendTTTFunc(result, "cmul", "__hmul", DType.Float16);
+                AppendTTTFunc(result, "cdiv", "__hdiv", DType.Float16);
+
                 AppendAtomicAdd(result, "atomicAdd", DType.Float16);
             }