Re: [PATCH v14 04/10] accel/tcg: add jit stats and time to TBStatistics

[Richard Henderson]

On 5/31/23 23:48, Wu, Fei wrote:
> > > -int64_t tcg_cpu_exec_time(void);
> > > +uint64_t tcg_cpu_exec_time(void);
> >
> > Why?  (Also, probably wants removing, per above.)
> The original patch mentioned 'make it an uint64_t as we won't be dealing
> in negative numbers.'

The signed vs unsigned thing is something that should be handled throughout everything 
that handles times, not adjusted here and there by only profiling.

> > > --- a/softmmu/runstate.c
> > > +++ b/softmmu/runstate.c
> > > @@ -728,9 +728,18 @@ static bool main_loop_should_exit(int *status)
> > >    int qemu_main_loop(void)
> > >    {
> > >        int status = EXIT_SUCCESS;
> > > +#ifdef CONFIG_TCG
> > > +    uint64_t ti;
> > > +#endif
> > >          while (!main_loop_should_exit(&status)) {
> > > +#ifdef CONFIG_TCG
> > > +        ti = profile_getclock();
> > > +#endif
> > >            main_loop_wait(false);
> > > +#ifdef CONFIG_TCG
> > > +        dev_time += profile_getclock() - ti;
> > > +#endif
> > >        }
> >
> > What is this intending to collect?  Because I don't think it measures
> > anything.  Certainly nothing related to TCG, CPUs or even devices.
> It's exported to hmp cmd in dump_jit_exec_time_info() together with
> cpu_exec_time.

That doesn't answer my question: What do you think it measures?

r~