George Nychis wrote:
So it looks like there is a lot of time spent scheduling the threads in
the kernel. We're not concerned about optimizing the application itself.
Not really.
Look at just the functions that take up more than 1% of CPU time (the
bottom 21 lines.) These total about 70% of the overall time spent. Of
these (approximate numbers)
26% libm, libc or libstdc++
19% libpmt
12% test_usrp_inband_tx
5% libusrp-inband
5% vmlinux or usbcore
3% libmblock
So it doesn't appear whatever performance issue you are seeing is
directly related to CPU time spent in the kernel. I have some other
ideas for profiling but don't have the time until later today perhaps to
email to you.