Re: [Discuss-gnuradio] oprofile inband code results

[Eric Blossom]

On Tue, Oct 09, 2007 at 01:33:12PM -0400, George Nychis wrote:
> 
> 
> George Nychis wrote:
> >I've been digging around the oprofile docs and google for this, but 
> >haven't turned up with anything yet as many of the docs show libc 
> >symbols in oprofile... but any ideas why my results are not achieving 
> >per symbol analysis of libc?  Instead, it just says "(no symbols)"
> >
> 
> Finally found it:
> sudo apt-get install libc-dbg
> 
> So in that case:
> http://www.andrew.cmu.edu/user/gnychis/inband_tx_10
> 
> And I've stopped reverse sorting :P heh

Glad to see it ;)

> So libm's bulk comes from sincosf, which I'm not worrying about, but 
> that solves the where question for that lib.
> 
> When it comes to libc, its free and mallocs that are killing us.  If you 
> look down libc to even the smaller percentages, its free/malloc related 
> (malloc_consolidate, _int_malloc, _int_free ... etc).
> 
> This happens in two major areas that I can think of: once in the 
> application which stores the samples (our sine wave) and passes them 
> down.  Another is in the usrp_server which allocates new memory for the 
> USB packets for which the samples are then treated transparently and 
> copied in to.  Then of course, this memory gets free.
> 
> Thoughts/comments? :)

The malloc/free's could be coming from the C++ allocations for the pmt
objects.  Thanks for investigating!

Eric