Only run __syncthreads when there is data to preload

src/targets/gpu/kernels/include/migraphx/kernels/preload.hpp

@@ -186,6 +186,7 @@ __device__ auto auto_preload(index idx)
 {
     return make_transform([=](auto f, auto... xs) {
         auto invoke = [=](auto... ys) {
+            if constexpr((Bs or ...))
                 __syncthreads();
             f(ys...);
         };