Hi,
I am using axi_rate_change with my custom noc block and seeing the
following error show up at the end-of-burst (i.e., everything seems to work
fine until I issue the stop streaming command).  As a result, subsequent
bursts do not behave since things are left in a bad state.

*[ERROR] [RX FLOW CTRL] *Error unpacking packet: ValueError: Bad CHDR
or packet fragment*[ERROR] [STREAMER] *The receive packet handler
caught a value exception.

My noc graph is the following:
  Radio => DDC => Myblock => host

One thing of note about my MyBlock is that it has long latency as it
contains 4 FFT cores (max 4096, but set to 1024) in series along with some
other stuff.  MyBlock also has capability to bypass the 2nd half of the
block (containing 2 of the 4 FFTs). When I exercise this bypass switch, the
frequency of this error drops drastically (although not entirely) from near
100% to perhaps 10%.

So, my question is: is there any issue with long latency in user IP when
using "axi_rate_change" (and if so, is there a way to fix it?)?

I reviewed "axi_rate_change" since the error indicates a bad CHDR (which my
IP does not touch since it is handled in axi_rate_change). A few remarks:

   - In the comments at the top, there is a note that says "User code with
   long pipelines may need to increase HEADER_FIFOSIZE", but this appears to
   me to be a "stale" comment that was applicable to an old version of this
   block.  Perhaps there previously was a FIFO for the header, but I think
   the block was re-architected to eliminate the need for this FIFO.
   - After reviewing the current code in the block, I can't see any problem
   with user IP having long latency.  For very long latency on the order of
   perhaps 80ms, it seems that the error/warning signals would be corrupted
   because of a throttle counter overflow, but I can't see how this would
   impact the streaming operation

Any help would be greatly appreciated.  Thanks.
Rob
_______________________________________________
USRP-users mailing list
[email protected]
http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com

Reply via email to