How to perform uvm_do_on without randomization?

Question

I have a virtual sequencer from which I execute three transactions in parallel, each one on its corresponding sequencer. So I have something like this:

class top_vseqr extends uvm_seqr extends uvm_sequencer;
  type_a_seqr seqr_a;
  type_b_seqr seqr_b;
  type_c_seqr seqr_c;

...

endclass: top_vseqr

class simple_vseq extends uvm_sequence;
  `uvm_declare_p_sequencer(top_vseqr)

  type_a_seq seq_a;
  type_b_seq seq_b;
  type_c_seq seq_c;

  ...

  virtual task body();
    fork
      `uvm_do_on(seq_a, p_sequencer.seqr_a)
      `uvm_do_on(seq_b, p_sequencer.seqr_b)
      `uvm_do_on(seq_c, p_sequencer.seqr_c)
    join
  endtask: body

endclass: simple_vseq

But now I want to be able to drive specific transactions into the virtual sequencer, depending on the test I am running. To do so, I have a class with an analysis import that is updated every time the monitor sees a transaction in the interface, and a function that returns the next transaction to be driven. So now I want to do something like the following:

class test extends uvm_test;

  model model_a;
  simple_vseq seq;
  top_vseqr virt_seqr;

  ...

  task run_phase(uvm_phase phase);

    ...

    seq = simple_vseq::type_id::create("seq", this);
    seq.seq_a = model_a.get_sequence();
    seq.start(virt_seqr);

    ...

  endtask: run_phase

Digging through the UVM documentation I have seen that there is a 'uvm_send macro, but it doesn't allow you to select the sequencer to run the sequence on (i.e. I haven't seen a 'uvm_send_on or something like that). What can I do?

Thanks!

Answer 1

`uvm_do_on_with may statisfis your requirement, and you can also delete rand in your packet to disable randomization or add constraint

Answer 2

You can implement the contents of the uvm_do_on macro without the call to randomize() (like you showed in the second snippet) without any worries. This is anyway the suggested practice by some experts, because the sequencer/driver handshake mechanism is pretty simple. The `uvm_do* macros are not the norm, they're just there to help you out in the beginning.

Answer 3

I don't think there is a `uvm_send_on macro but there is a `uvm_create_on(SEQ_OR_ITEM, SEQR) macro which you can use. From the UVM documentation, this is the same as `uvm_create except that it also sets the parent sequence to the sequence in which the macro is invoked, and it sets the sequencer to the specified ~SEQR~ argument. In fact, the `uvm_create macro calls `uvm_create_on macro internally by passing m_sequencer by default. You can override it using the `uvm_create_on call.

Alternatively, you could also do a set_sequencer on your sequence_item object so that it sets the m_sequencer variable.

Hope this helps.