Hi,

Just a quick progress update. I have completed NEON support for 34 out
of 74 libvolk kernels that were missing NEON implementations.

Average speedup is around 4x depending on kernel, not very surprising
since NEON SIMD vector size for float32 is 4...

Biggest surprise was volk_32fc_s32fc_x2_rotator_32fc that now is 14x
faster on Raspberry Pi 3b. This is nice because this kernel is used in
the frequently used (pun intended) Frequency Xlating FIR Filter.

https://github.com/gnuradio/volk/issues/243

So far kernels are only available in my (messy) branch but I will
gradually create pull requests into libvolk.


--Albin

_______________________________________________
Discuss-gnuradio mailing list
Discuss-gnuradio@gnu.org
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio

Reply via email to