In most implementations of IP, there is code in the module that handles outgoing datagrams to decide if a datagram can be sent directly to the destination on the local network or if it must be sent to a gateway.
Generally the code is something like this:
IF ip_net_number(dg.ip_dest) = ip_net_number(my_ip_addr)
THEN
send_dg_locally(dg, dg.ip_dest)
ELSE
send_dg_locally(dg,
gateway_to(ip_net_number(dg.ip_dest)))
(If the code supports multiply-connected networks, it will be more
complicated, but this is irrelevant to the current discussion.)
To support subnets, it is necessary to store one more 32-bit quantity, called my_ip_mask. This is a bit-mask with bits set in the fields corresponding to the IP network number, and additional bits set corresponding to the subnet number field.
The code then becomes:
IF bitwise_and(dg.ip_dest, my_ip_mask)
= bitwise_and(my_ip_addr, my_ip_mask)
THEN
send_dg_locally(dg, dg.ip_dest)
ELSE
send_dg_locally(dg,
gateway_to(bitwise_and(dg.ip_dest, my_ip_mask)))
Of course, part of the expression in the conditional can be
pre-computed.
It may or may not be necessary to modify the "gateway_to" function, so that it too takes the subnet field bits into account when performing comparisons.
To support multiply-connected hosts, the code can be changed to keep the "my_ip_addr" and "my_ip_mask" quantities on a per-interface basis; the expression in the conditional must then be evaluated for each interface.