|Subject:||Re: [Discuss-gnuradio] segfault in volk_32f_x2_dot_prod_32f_a_generic|
|Date:||Sun, 19 Oct 2014 17:45:10 +0200|
|User-agent:||Mozilla/5.0 (X11; Linux x86_64; rv:24.0) Gecko/20100101 Thunderbird/24.7.0|
the segfault happens in a Volk kernel, which is an optimized numeric algorithm for a length of uniform data. The fact that the base pointer is reachable by gdb but the kernel segfaults somewhere in the middle of operation indicates things are fundamentally wrong with over how much data the kernel operates.
Now, the fft_filters are quite well-QA-tested, I think, so debugging this is interesting. Thus, I took a look at your GRC file: In you AM demod block, you use a decimation of samp_rate (2.4e6), which means the ration of input to output blocks is 2.4e6... I think you meant to use 1 there. With that, for me, the flowgraph doesn't crash.
Also, if you experience high CPU load (and anyway, in any case), run volk_profile. That will try out all the optimized VOLK kernel implementations, and will store the optimal algorithm in a file, so that the next time you start a flow graph, the best implementation is automatically used. I just mention that because your fft_filter internally uses volk_32f_x2_dop_prod_32f, and runs the "generic" implementation, which on all my 64bit machines is much slower than the SSE version.
On 19.10.2014 16:54, Sergey Alirzaev wrote:
Hello and thank you for the gnuradio project! Seems like this mailing list silently drops the messages from non-subscribers, that is not mentioned in <http://gnuradio.org/redmine/projects/gnuradio/wiki/Development>. I'm trying to use an AM Demod block but encounter a segfault: <https://bpaste.net/show/bbeefa9ff787>. It reproduces on 220.127.116.11, 3.7.5 and the git HEAD. My .grc is there: <https://bpaste.net/show/c274b44c57a6>. I've poked gdb a bit but couldn't see what's wrong as the pointers seem to be reachable from gdb. What can I do about this?
|[Prev in Thread]||Current Thread||[Next in Thread]|