The EPIDs are only used by the stream endpoints. I think it will always be
set to 0 when it is sent to your block, and anything you put in there will
be ignored by downstream blocks. In theory, I suppose it could be used the
way you described, but I think that would require pretty significant
changes in both UHD and in the FPGA.

Could you use two stream endpoints instead to double the number of virtual
channels?

Thanks,

Wade

On Thu, Apr 17, 2025 at 11:38 PM Brian Padalino bpadalino@gmail.com wrote:

On Sun, Apr 27, 2025 at 8:44 PM Wade Fife wade.fife@ettus.com wrote:

Maybe I could do that. So have 4 endpoints be the lower 64 channels, and
then have 4 endpoints be the upper 64 channels and utilize only the VC
field?

When it comes to SeqNum, do the VC's count as being separate or would the
SeqNum also increase since the DestEPID is still the same?

Thanks,
Brian

Right, up to 64 VCs per stream endpoint.

Each stream endpoint only supports a single sequence number counter, so
that single counter would be shared by all virtual channels.

Wade

On Sun, Apr 27, 2025 at 8:27 PM Brian Padalino bpadalino@gmail.com wrote:

On Mon, Apr 28, 2025 at 8:47 AM Wade Fife wade.fife@ettus.com wrote:

Thanks for the clarification on this.

From your perspective, what is the feasibility of having 1024 individual
endpoints in a system?

Thanks,
Brian

In practice, you can't have a large number of stream endpoints in a single
USRP, because the crossbar and associated logic adds up. Something on the
order of 16 or so might be a practical limit, depending a lot on what's on
those endpoints. If you need to distinguish between more data streams, then
you'd want to use something like virtual channels or prepending your data
with some kind of identifier.

With many USRPs interconnected, you could have a lot more endpoints in
theory.

Wade

On Mon, Apr 28, 2025 at 7:56 AM Brian Padalino bpadalino@gmail.com wrote:

On Mon, Apr 28, 2025 at 10:33 AM Wade Fife wade.fife@ettus.com wrote:

Thanks for this answer.

Does static routing help with this or not particularly?

I have only a single configuration I ever want to run, and it's endpoints
directly into a modified radio block - no other blocks.

Thanks,
Brian

I don't think static routing will help in your case. Stream endpoints are
only for communicating over dynamic routes and the EPID is used for that
routing.

Wade

On Mon, Apr 28, 2025 at 9:37 AM Brian Padalino bpadalino@gmail.com wrote:

One thing to keep in mind: If you have many many virtual channels, then you
probably still have a lot of blocks. The control crossbar also increases
with N^2 in size, where N is the number of blocks (not the number of stream
endpoints, which affects the size of the CHDR crossbar).

Maybe what you're trying to do is better handled at the application level.
Meaning that you have bigger blocks that do some kind of channelization
internally, and then pull apart the stream in your application. After all,
there are very few rules on what kind of data you can stream to the host
(or vice versa). For example, when we stream FFT data to the host, we don't
separate the bins into virtual channels, but we put the onus on the
application to understand that every vector of FFT-size needs to be
understood as such.

Not sure if this is helpful to you -- of course, for most cases, RFNoC
should be doing all the hard work for you. But in some cases, it might be
easier to do some of the work in your app.

--M

On Thu, May 1, 2025 at 4:42 AM Wade Fife wade.fife@ettus.com wrote: