Re: [Valgrind-users] Expected slowdown

[Christoph B. <bar...@or...>]

Am Dienstag, 24. Oktober 2006 17:51 schrieb Nicholas Nethercote:

>
> >               :   Int         i;
> >               :   Addr        ips[VG_DEEPEST_BACKTRACE];
> >               :   Bool        same;
> >               :   UWord       hash;
> >               :   ExeContext* new_ec;
> >               :   ExeContext* list;
> >               :   UInt        n_ips;
> >
> >     7 2.2e-05 :   init_ExeContext_storage();
> >   276 8.8e-04 :   vg_assert(VG_(clo_backtrace_size) >= 1 &&
> >
> >               :             VG_(clo_backtrace_size) <=
> >               : VG_DEEPEST_BACKTRACE);
>
> Do you know what these two columns mean?
>

OProfile is a stochastic profiler. Each 100000 events of type CPU_CLK_UNHALTED 
it checks where the execution currently is. Then the counter for this 
position is incremented. Here we see the counter.

For example: During the whole profiling run oprofile stopped 7 times in the 
line with 

init_ExeContext_storage();

It did not stop inside of the called function but in the code of the current 
function preparing or evaluating the call to init_ExeContext_storage();

The second number is the percantage that this line has compared to all events. 
Here the number is neglible. Another example is:

11124624 35.3958 :         if (list->ips[i] != ips[i]) {

Here we see that 35.4% of all taken samples have been in this line of code. If 
the profiling session is long enough (and I think mine was long enough) one 
can conclude that 35.4% of the whole runtime during the profiling session has 
been spent in this line of code.

Christoph Bartoschek