[USRP-users] rfnoc blocks with multiple inputs/outputs
michael.west at ettus.com
Fri Oct 13 16:35:42 EDT 2017
OK. I dug in a bit more and I can now tell you are most likely using the 2
streams from a single TwinRX daughterboard by the nature of the issue. The
root cause is that Twin RX uses ports 0 and 1 on the radio block (all other
daughterboards only use port 0) and only port 0 is receiving the stream
command when there is a single output port on your block. The root of the
issue is in the code block found here:
Notice that the "chan" parameter is used to forward the command to the
upstream block. That parameter is the port on the block. Since your block
has only one output channel, chan is set to 0. When you add the second
output to your block, the function is called twice; once with chan=0 and
once with chan=1.
Your workaround adding a second output port will work for now until the
issue is resolved. Since this only affects TwinRX in this particular use
case and there are several other use cases we must consider and test for
any proposed fix, a proper permanent resolution may take some time.
On Fri, Oct 13, 2017 at 12:38 PM, Dario Pennisi <dario at iptronix.com> wrote:
> Hi Mike,
> Unfortunately what you write is not what I see. Split stream has 1 input
> and 2 outputs and that works. What I am saying is that the opposite
> doesn't. This seems to be due to the fact that for some reason blocks
> feeding second input are not started if block does not have a second port.
> With the same block, even without adding code for a second output but just
> adding the fake output to the block descriptors made it work...
> I'm sure there's a bug somewhere in uhd or gr-ettus as he is fine.
> Best regards,
> Dario Pennisi
> On Fri, Oct 13, 2017 at 9:32 PM +0200, "Michael West" <
> michael.west at ettus.com> wrote:
> Hi Dario,
>> A block can have any number of inputs and outputs, so 2 inputs and 1
>> output is fine. The Split Stream block is an example of an asymmetric
>> The UHD XML description of the Split Stream block can be found here:
>> The GnuRadio XML description of the Split Stream block can be found
>> here: https://github.com/EttusResearch/gr-ettus/blob/
>> Hopefully, those examples will help.
>> As far as the multiple ports at 200 Msps on a single block, it is a known
>> issue. The bus clock would need to be raised to the CE clock of 214.286
>> MHz to allow 2 ports to operate at 200 Msps in one block, but doing that
>> causes the FPGA build to fail timing. As far as a workaround or trick,
>> there is a way where you can hack up the core FPGA code and make a single
>> module with 2 noc_shells connected to separate crossbar ports in the
>> x300_core code, but that is a bit more involved. If you are daring and
>> curious, look at the way noc_block_axi_dma_fifo and noc_block_radio_core
>> blocks are instantiated and connected to the crossbar in x300_core.v
>> On Wed, Oct 4, 2017 at 11:57 PM, Dario Pennisi via USRP-users <
>> usrp-users at lists.ettus.com> wrote:
>>> I am still struggling with a block which should have 2 inputs and 1
>>> output. Apparently everything is fine with both FPGA and software but the
>>> second input is not fed with data.
>>> I could not find any example of an asymmetric block with different
>>> number of inputs/outputs so I am starting to suspect this is not supported.
>>> Can anyone tell me if this is the case and I need to add a dummy output?
>>> Also, when defining block I see that in the xml there is the possibility
>>> to specify nports parameter so that a single port definition provides more
>>> than one port. If i set nports=2 for the input I am not sure how to define
>>> the gnuradio xml descriptor and If I just describe two ports as usual block
>>> will not initialize properly.
>>> Finally, I understand that system clock rate on AXI bus is 166 MHz and
>>> since NOC ports are 64 bits, on X310 a single block should not be able to
>>> process two input streams at 200 MSPS.
>>> Even raising AXI bus clock to 200 MHz would fail as but would still not
>>> provide sufficient overhead for packet headers.
>>> Is there any suggested trick to handle this situation? Is it eventually
>>> possible to have a single rfnoc block connect to two AXI ports rather than
>>> just one?
>>> Dario Pennisi
>>> USRP-users mailing list
>>> USRP-users at lists.ettus.com
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the USRP-users