> Thanks David. I have BigLittle board with 6 cores, I'll try it on then.
> Do you intend to apply the same NN treatment to mode 700C in the future?
I'd like too, but it's very much early days.
Multiple cores won't help with LPCNet as it works sample by sample (like
an IIR filter) so isn't
Thanks David. I have BigLittle board with 6 cores, I'll try it on then.
Do you intend to apply the same NN treatment to mode 700C in the future?
Regards,
Adrian
On December 15, 2019 7:12:54 PM UTC, David Rowe wrote:
>Hi Adrian,
>
>The LPCNet codec does have some of the DSP functions ported to
Hi Adrian,
The LPCNet codec does have some of the DSP functions ported to NEON as
well as AVX, it's a build time option.
About a year ago I did some work on getting the codec running in real
time on a Pi 3, but it wasn't fast enough. I have been told it runs in
real time on some faster ARM
Hi Adrian,
On reading the Intel/AMD docs, the AVX/AVX2 additions in the i3 and up do
256bit.
ie. 8 simultaneous 16bit number crunches.
But the aarch64 NEON, as far as I know only do 128bit.
ie. Only 4.
I'd look at utilising the other CPU cores, perhaps for alternate frames.
Or the GUI.
I'm
Hi David,
Do you have any idea if we could use any ARM optimizations for 2020 mode?
I'm thinking of libvolk which can use SSE3, AVX and NEON and is the backend for
GNU radio on ARM platforms. I have an ARM64 board and thinking of trying 2020
on it.
Regards,
