unbreak optimized gelu test (#7987)

We don't run optimized gelu in OSS currently (soon, I hope!), so I missed the need to fix it.

Test plan: ran optimized gelu test internally.

Author: swolchok

kernels/optimized/cpu/op_gelu.cpp

@@ -13,6 +13,7 @@
 
 #include <cmath>
 
+#include <executorch/kernels/portable/cpu/util/activation_ops_util.h>
 #include <executorch/runtime/kernel/kernel_includes.h>
 #include <executorch/runtime/platform/assert.h>
 
@@ -116,30 +117,12 @@ Tensor& opt_gelu_out(
     Tensor& out) {
   (void)context;
   ET_KERNEL_CHECK(
-      context,
-      tensors_have_same_shape_and_dtype(input, out),
-      InvalidArgument,
-      out);
+      context, check_gelu_args(input, approximate, out), InvalidArgument, out);
 
-// helper for generating the cases for different data types
-#define GELU(ctype, dtype)                         \
-  case ScalarType::dtype:                          \
-    gelu<ctype>(context, input, approximate, out); \
-    break;
-
-  switch (input.scalar_type()) {
-    // TODO support Double as well
-    GELU(float, Float)
-    default:
-      ET_KERNEL_CHECK_MSG(
-          context,
-          false,
-          InvalidArgument,
-          out,
-          "Unhandled dtype %" PRId8,
-          static_cast<int8_t>(input.scalar_type()));
-  }
-#undef GELU
+  ET_SWITCH_FLOATHBF16_TYPES(
+      input.scalar_type(), context, "gelu.out", CTYPE, [&]() {
+        gelu<CTYPE>(context, input, approximate, out);
+      });
 
   return out;
 }

kernels/optimized/cpu/targets.bzl

@@ -33,7 +33,9 @@ _OPTIMIZED_ATEN_OPS = (
             "ovr_config//cpu:arm64": [
                 "fbsource//third-party/sleef:sleef_arm",
             ],
-        }),
+        }) + [
+            "//executorch/kernels/portable/cpu/util:activation_ops_util",
+        ],
     ),
     op_target(
         name = "op_le",