Hey everyone,
If you don’t track our check-ins closely, you might have missed a
fairly large change to the code yesterday. We pushed a new component
called gr-filter that consolidates all of the filter blocks and
related code into one top-level component. If you are working off
master, all of the current filter functions still exist in the gr
namespace, but we’re working to move to the new filter namespace in
preparation for version 3.7.
Part of this move was to convert all of the hand-coded assembly over
to using VOLK. This is designed to make adding SIMD code to filters
much easier in the future, but we could still see a few bumps. I’ve
already seen some machines machines fail QA, so we’re working to fix
those, but I’d appreciate more reports from various architectures.
So if you run ‘make test’ after getting the most recent master branch
and you see any test failures, let me know. If you get failures, the
first thing to do is run “volk_profile”. This is a program that is
installed into $prefix/bin that will test all Volk kernels and select
the fastest version for your machine. Rerun ‘make test’ and let me
know if this fixes any problems. If you still have problems, please
send me the unit test output files found at
$builddir/gr-filter/lib/.unittests/gr_filter.xml and
$builddir/gr-filter/python/unittests/python/tests_*.xml. That should
help me get an understanding for what’s going wrong.
Thanks!
Tom