Ensure Queue is empty before finishing simulation

[h0lyalg0rithm]

No description provided.

[h0lyalg0rithm]

@knute-sifive I added this assert to make sure there are no pending stores and the current model fails.Do you some ideas on this

[ghost]

I ran the regression with your branch and the failure makes sense -- and doesn't. 😄

One of the tests terminates simulation after 1M instructions have retired:

olympia -i 1M --workload traces/core_riscv.zstf

which will leave "unretired" stores in the ROB. In other words -- that's ok!

This test works fine:

olympia --workload traces/core_riscv.zstf

[h0lyalg0rithm]

Oh I missed that.
Should I change the test cases to remove the instruction limit.

[ghost]

Well, you can remove the instruction limit if you want, but the simulator should not die if a user wants to only generate a pipeout or report or whatever for just the first few million instructions.

In other words, I do not think this assert is required.

[h0lyalg0rithm]

Got it, Is there another way to get around this issue?
Its pretty useful when developing to know that all the instructions were correctly handled in the LSU

[ghost]

You can make it a parameter to the LS to help with debugging. If you really want to get fancy, you can add a NotificationSource to Retire that is observed by LS. When Retire is completely drained AND is stopping simulation, that notification can be broadcasted, which LS can "hear" and set a boolean retire_done_and_is_drained_ and then you can check the ensure the LS issue queue is actually drained.