Thread Rating:
  • 0 Vote(s) - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5
[Simple profiling results] ~30% of time Waiting?
#1
Hi guys,
i was interested on how pcsx2 would look like on a profiler.
So here you go:
[Image: pcsxz.jpg]

Quite curious, so where does pcsx2 spend so much time "slowing down"
the overall process?

Roughly speaking, with all wait stuff ( SleepCondition, raising the exception)
its almost 58% CPU time all the way


Regards,
nato

EDIT: whoops, messed up the res, couldnt see a thing
Reply

Sponsored links

#2
One word answer: synchronization.
I guess I'm just a noob, but that's probably fairy accurate.
Reply
#3
But a program that spends ~50% with synchronization is improveable.
Besides the sync stuff there are no real bottlenecks (according to the profiling results) other than some readfile junk which is not important anyways.

Would be great to know where most of that sync time is spent Smile

Regards,
nato
Reply
#4
with threading you have to put waits in loops or the process will consume 100% cpu probably pcsx2 didn't have enough to do so it waits, or its synchronizing with sound or other threads.
Intel Core I7 930@3.5GHz(TB, HT), 6GB DDR3@1600MHz, Geforce 970 G1
Reply
#5
im aware of that Smile
Still 30% for avoiding 100% is never-ever real.

So the question is, where is that critical sync part that consumes so much time waiting on something.

Btw, this profiling was taken while playing GoW in 1080p dx 10 render size @99% EE, so 30% (50) is not from "idling" around Wink
^- Same results when i turn down settings, 30 - 50% waittime, just in case you think its because of the cpu cant handling all that at once and therefor forces threads to wait

Regards,
nato
Reply
#6
Is this wait time from the extra inactive cores of your processor...?
Reply
#7
was this performed on the most recent svn?
Reply
#8
Is this wait time from the extra inactive cores of your processor...?


^-- might be interested in explaining that one ? i'm not that expirienced and would appreciate if you would do so Smile
Thanks anyways


Yes this was tested with r5160 from the svn.
Got an i7 quad core, thus i enabled the VU which really helps out.
^-- as a side note
Thanks to you too.

Regards,
Nato
Reply
#9
there are intentional idle spin waits to emulate timing.
Reply
#10
Thanks for answering!
I'm impressed by your work guys, most of the time profiling reveals some nifty performance bottlenecks, but all of it seems intentional or not that big of a deal.

Regards,
natoyo
Reply




Users browsing this thread: 1 Guest(s)