The standard plumbing of TLM 2.0. How Initiators talk to Targets.
In TLM 2.0, the unidirectional Port/Export model is replaced by the Socket. A socket is a combined object that handles both the Initiator (Requestor) and Target (Responder) logic in a single connection.
The b_transport task is the workhorse of TLM 2.0. It is a
blocking call that passes a Generic Payload and a Time Delay object.
task b_transport(uvm_tlm_generic_payload trans, uvm_tlm_time delay);
// 1. Validate the command
if (trans.get_command() == UVM_TLM_READ_COMMAND) begin
// Perform read from memory array
// ...
trans.set_response_status(UVM_TLM_OK_RESPONSE);
end
// 2. Add local delay (Temporal Decoupling)
delay.add(10, UVM_NS);
endtask
Why pass a uvm_tlm_time object instead of calling #10ns?
This enables Loosely-Timed (LT) modeling. Components accumulate local time
offsets and only synchronize with the simulation kernel periodically.
By avoiding actual simulator #wait calls, TLM 2.0 models can run
100x to 1000x faster than RTL, making them ideal for software
development and architectural exploration.
UVM provides "Simple Sockets" (Initiator and Target) which handle most of the boilerplate code like registration and callback binding.
function void connect_phase(uvm_phase phase);
// Simple 1-to-1 socket binding
my_master.isock.connect(my_slave.tsock);
endfunction
Unlike TLM 1.0, sockets are bound to the Generic Payload type by default. If you need custom fields, use the Extension Mechanism rather than creating custom socket types. This maintains interoperability.