Menu
▾
▴
#61 multi-threaded digit
Hello,
I am running into some issues when running digifil with multiple threads (both coherent filterbank and the normal on). The output filterbank file does not show the pulsar when folded again with dspsr. The input to digifil is complex baeband data. dspsr works on the same set of files and 1937+21 shows up pretty well.
Are there known issues when running digifil in multithreaded form?
Thanks!
--Ramesh
hi Ramesh,
Yes, there are some differences between multi- and single-threaded digifil in terms of which options work/don't. And I know of a couple bugs that I have been testing workarounds for locally but are not yet checked in. Could you send alond the full command lines you are using?
Cheers,
Paul
Hi Paul,
Happy new year! Thanks for the quick response. Here is the command line that I used:
digifil -threads 8 -F 16:D -D 71.04 -b 8 -t 20 -o out3.fil
/bin/ls -1 *.dada | head -2The output file appears to have some data in just two channels (see screen shot). Adding "-overlap" did not help, neither does processing off a data block (I used dada_diskdb to dump data on to a data block in RAM).
EDIT: Paul, I also confirm that disabling coherent filterbank does not help - again only in multi-threaded mode.
Cheers,
Ramesh
Last edit: Ramesh 2016-01-20
So, I think the main thing that is not quite working yet in the multi-threaded version is the data rescaling. Code still needs to be developed to allow the different threads to share scale/offset values appropriately. You can check that this is the issue by using floating-point output ("-b-32") and see if the data look better. You might also try disabling auto-rescaling by adding "-I0" but I think this should automatically happen in multithreaded mode.
For now, to get the data to fit into 8 or 16 bits you can play around with the "-s" option to manually apply a scale factor. If this does not help, let me know and we can investigate more..
Hi Paul,
Ok, scaling appears to be the problem. When I write out as float, things do work fine. Thanks, I shall use this work around for the moment.
Cheers,
Ramesh
Ok, great! I have noticed a few other bugs that come up when using coherent dedispersion digifil. I will check in my workarounds for these. I think they only happen when using 4-pol output though.