Enable logging after trigger

[danbone]

Is there a way I could enable logging only once a certain condition happens i.e. enable logging when the retired instruction counter passes 100k.

[ghost]

Yes! In the Olympia Performance Simulator (and in the Sparta CoreModel), you can use the command line option --debug-on-icount 100k to start whatever collection mechanisms you've defined on the command line. For example:

olympia -l top info my_info_log.out --debug-on-icount 100k traces/dhry_riscv.zstf 

This will start collecting a log on the top node after 100k instructions have been retired.

How this works...

Since the command line option is a Sparta command line option, Sparta needs to know which counter in the implemented simulator that represents the icount. This is communicated to Sparta via the derived sparta::app::Simulation instance overriding the method findSemanticCounter_

For example:

    class MySim : public sparta::app::Simulation
    {
        // lots of stuff
    private:
        const sparta::CounterBase* findSemanticCounter_(CounterSemantic sem) const override {
             switch(sem){
             case CSEM_INSTRUCTIONS:
                return getRoot()->getChildAs<const sparta::CounterBase>("cpu.core0.rob.stats.total_number_retired");
                break;
             default:
                return nullptr;
        }
    };

[ghost]

Keep in mind that the command line options --debug-on <cycle> and --debug-on-icount <inst count> work on logging, pipeout collection, and pevent collection.

[danbone]

Thank you Knute, exactly what I was looking for. Can't believe I missed it in the help message.

[klingaard]

No prob. This needed to be documented better.