Looking for OpenNIC simulation model

[adki]

Hi,

I'm trying to figure out how OpenNIC works using Alveo U280 and I could not get it work yet.
So, I'm looking for simulation environment for 'p2p_250mhz'.
The reference guide says there was 'simulating OpenNIC_shell' for prior to 1.0.
I believe that hardware developer needs an environment to check his/her block in the context of OpenNIC as simple as possible,
where his/her block will be 'p2p_250mhz' or 'p2p_322mhz'.
So, is there behavioral model for QDMA subsystem / CMAC subsystem?
If there is not, is it possible to get more details regarding packet format of AXI stream at the boundary of 'p2p_250mhz'.

Thanks to OpenNIC team for your contributions.

[cneely-amd]

The early simulation support wasn't maintained, when OpenNIC shell underwent substantial refactoring for version 1.0.

@108anup, can you maybe please say a bit about the testbench that you've created for simulating OpenNIC shell?

[108anup]

As a quick answer, I used modelsim and cocotb.

cocotb (https://www.cocotb.org/) enables cosimulation of RTL and a test bench written in python (e.g., https://github.com/alexforencich/verilog-axis/blob/master/tb/axis_switch/test_axis_switch.py).
The open nic shell design uses some Xilinx IPs so their simulation models need to be exported. Vivado has an option to do this for a lot of common simulators. Likewise cocotb supports a number of simulators as its backend. Using cocotb is just a convenience, one can very well write a test bench in verilog to activate signals at the boundary of the user plugin.

For the packet format, all the packet contents are in the axis_tdata signal in the AXI stream interface. So one can write a test bench similar to the verilog-axis test bench shown above.

I can open a pull request with a simple example for reference.

[adki]

Hi Chris and Anup,
Thanks for the comments,

I am trying to prepare some sort of packet parser in p2p_250mhz as shown below,

packets form host ====>|p2p_250mhz|=====> pakets to adap
packets to host <======|p2p_250mha|<===== packets from adap

I could not find how Ethernet(UDP/TCP over IP) is carried on the s_axis_tdata/tkeep/tlast/size/src/dst stream.
I presume the packet will be placed on the stream as big-endian format, right?

I believe that the host and adap side model can utilize cocotb for further applications.

Thanks.

[108anup]

See #30 for cocotb+modelsim setup. It shows how a packet might be sent.

All packet contents are carried in axis_tdata. All other signals are control signals.

Regarding byte order:
In simulation if you send packet with content MSB \x00 \x11 \x22 LSB from cocotb, it appears as as \x22 \x11 \x00, i.e, axis_data[7:0]='h00, axis_tdata[15:8]='h11, axis_tdata[23:16]='h22. Please correct me if I am wrong here @cneely-amd.