I am trying to write a custom operator, but the document in https://pytorch.org/tutorials/advanced/cpp_custom_ops.html#creating-mutable-operators is confusing at:

TORCH_LIBRARY(extension_cpp, m) {
 m.def("mymuladd(Tensor a, Tensor b, float c) -> Tensor");
 m.def("mymul(Tensor a, Tensor b) -> Tensor");
 // New!
 m.def("myadd_out(Tensor a, Tensor b, Tensor(a!) out) -> ()");
}
TORCH_LIBRARY_IMPL(extension_cpp, CPU, m) {
 m.impl("mymuladd", &mymuladd_cpu);
 m.impl("mymul", &mymul_cpu);
 // New!
 m.impl("myadd_out", &myadd_out_cpu);
}

It says

You may wish to author a custom operator that mutates its inputs. Use Tensor(a!) to specify each mutable Tensor in the schema; otherwise, there will be undefined behavior. If there are multiple mutated Tensors, use different names (for example, Tensor(a!), Tensor(b!), Tensor(c!)) for each mutable Tensor.

But I do not know why we need a name a or b for mutable tensor, it seems myadd_out(Tensor a, Tensor b, Tensor(!) out) -> () is enough to represent out is mutable, what is the meaning of the name after ! ?