Re: [Discuss-gnuradio] oprofile callgraph question

[Mikyung Han]

Thanks eric,
Well, yes it will be helpful to have better symbol names,
but the core problem is doing callgraph.
Is there a way that I can figure out .loop and .loop2 is called?

How do I know that gr_fit_xxx.cc, gr_fir_filter_xxx.cc calls .loop and .loop2?

Thanks!

Mikie

On Fri, Jun 20, 2008 at 4:45 PM, Eric Blossom <address@hidden> wrote:

On Fri, Jun 20, 2008 at 04:32:35PM -0400, Mikyung Han wrote:
> Hi all,
>
> I am profiling gnuradio-3.1.1/gr-atsc/src/python/ example.
> I've already collected seinfeld.dat (the captured data from fox TV).
>
> opcontrol --setup --vmlinux=my_path_to_vmlinux/vmlinux
> opcontrol --start
> opcontrol --reset
>
> # run atsc example ( *.py files)
> xterm -e ./interp.py seinfeld.dat &
> xterm -e ./xlate.py
> xterm -e ./btl-fsd.py
> xterm -e ./viterbi-out.py seinfeld.mpg
>
> opcontrol --save=my_session_name
> opcontrol --dump
> opcontrol --stop
> opreport -l session:my_session_name
>
> So I do get the seinfeld.mpg correctly.
> And I can actually play with xine.
> Also I do get the profiled result.
>
> Here are top 10 application and symbols that consumed resources.
>
> atsci_equalizer_lms  26.466%
> .loop from  libgnuradio-core.so.0.0.0 12.52%
> atsci_single_viterbi 10.5%
> .cleanup from libgnuradio-core.so.0.0.0 6.7%
> __ieee754_atan2 from libm-2.5.so 5.53 %
> .loop2 from libgnuradio-core.so.0.0.0 5.5%
> memcpy from libc-2.5 so 3.03%
> gr_freq_xlating_fir_filter_ccf 2.68%
> atsc_fpll 1.8%

.loop and .loop2 are the inner loops of the SSE filter kernels.
They'll be called from gr_fir_XXX.cc, which is called from gr_fir_filter_XXX.cc

One of my developer branches has these unfortunate labels renamed to
something more enlightening.  Those changes will eventually make it to
the trunk.

BTW, I think that you'll find the ATSC code in the trunk is
considerably more up-to-date than that in 3.1.1.  Chuck Swiger has
been working on it.

Eric