On Wed, Sep 6, 2017 at 5:04 PM, Jakob Kummerow <[email protected]> wrote:
> Ideally, I want to look at per-function tick counting when running >> speedometer2.0, but I have not gotten to a point running speedometer 2.0 on >> a command line. > > > True, Speedometer doesn't run in d8, it requires a browser (at least > "content_shell", or full Chromium), but you can start that from the > command-line, and pass --prof or --runtime-call-stats to V8 via > --js-flags="...". > I guess to build content_shell, I would have to download chromium source. So far, I have been only playing around with V8. Let me try to run speedometer 2.0 on content_shell. Thanks for pointing this out. > > It looks like its taking measurement is based on the calls/counters >> (RuntimeCallStats) that is placed at the entry of the functions that it >> wants to instrument eg. ParseFunctionLiteral. > > > Correct, --runtime-call-stats measures manually defined counters. That's > intentional -- for a number of investigations that the team is interested > in, it provides more actionable data than plain old statistical profiling. > It doesn't try or claim to generate per-function profiling results, so it's > not directly comparable to the other tools you mentioned. > > You can (obviously) use whichever tool you find the most useful. You can > use a profiler you know and like. You can extend RuntimeCallStats for your > purposes (that's how it was built in the first place). If you dig deep > enough into an area nobody has explored before, you will probably soon want > to build your own instrumentation -- but nobody can tell you what that > might be, it depends on what you want to find out. > > I understand you have not used valgrind to measure performance, and you >> have not seen this data. > > > I have seen callgrind output, and it looked interesting. I did not find it > very actionable, but maybe that's just due to what I was working on at the > time. > > Hence, I wanted to know if there are specific options that measures >> parsing in a little more detail. > > > I'm not aware of any parsing specific profiling options. And I still don't > understand what you mean by "more detail". It doesn't get more detailed > than instruction-level profiling with linux perf (which I have mentioned > before). > All I am refering to is with valgrind/callgrind I see parsing in V8 is one of the performance hogs, and I wanted to see if there are any specific options that instruments parsing. I guess there are none, and if I want to investigate further, I would have add instrumentation myself. Thanks. > > -- > -- > v8-dev mailing list > [email protected] > http://groups.google.com/group/v8-dev > --- > You received this message because you are subscribed to a topic in the > Google Groups "v8-dev" group. > To unsubscribe from this topic, visit https://groups.google.com/d/ > topic/v8-dev/S4p0xaSZrr0/unsubscribe. > To unsubscribe from this group and all its topics, send an email to > [email protected]. > For more options, visit https://groups.google.com/d/optout. > -- -- v8-dev mailing list [email protected] http://groups.google.com/group/v8-dev --- You received this message because you are subscribed to the Google Groups "v8-dev" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
