Queues: PCAP-based benchmarking harness

[polybeandip]

Closing #2304, we can generate data files for queues directly from packet captures using parse_pcap.py!

Changes:

• add parse_pcap.py, a PCAP parser for generating data files
• move synthesis, parsing, and plotting scripts into evaluation/
• add pcap_sim.py, a harness similar to queue_call for iterating through commands in our data file
• add insert_tuple_flow_inference to flow_inference.py, flow inference to be used with pcap_sim.py
• remove generate functions from binheap/round_robin.py, binheap/strict.py, and strict_or_rr.py
  • cleaner seperation of queue components and tests
• tweak round robin and strict tests to take command line argument sim-pcap
  • if sim-pcap is True, use pcap_sim.py instead of queue_call.py
• add calyx_py-to-calyx.rhai to fud2/scripts/
  • creates a state for CalyxPy and an operation from CalyxPy to Calyx
  • add associated fud2 tests to pass CI
• add plot_pcap_sim.py, a script for generating Gantt charts from output memories

Example

Say you want a data file of the format used by our Testing Harness, with timing based on packets 10 through 20 of kachow.pcap.

Simply run

python3 parse_pcap.py kachow.pcap kachow.data --start 10 --end 20

This produces two outputs

• kachow.data, the data file
• various statistics about the data file and PCAP: namely, num_cmds, the size of the commands memory in kachow.data

We can run kachow.data using sim_pcap.py on any of our round robin or strict queues:

1. pick your favorite round robin or strict test in tests/
2. generate Calyx with python3 <num_cmds> --sim-pcap <test_name>.py
3. run the produced Calyx with data file kachow.data and your simulator of choice

For example, we can combining steps 2 and 3 into a single fud2 command

fud2 rr_2flow_test.py -o rr_2flow_test.json \
    --through verilator \
    -s sim.data=kachow.data \
    -s py.args="<num_cmds> --sim-pcap" \

which produces the output memories rr_2flow_test.json.

We can visualize these memories with graphs, in the style of Formal Abstractions, by feeding them to plot_pcap_sim.py:

python3 plot_pcap_sim.py rr_2flow_test.json

This generates the Gantt chart rr_2flow_test.png.

[polybeandip]

Some Gantt charts from packets 1 through 20 of a PCAP (it's too big for me to upload and I forgot where I found it, sorry!)

2-flow Specialized Round Robin

2-flow Binary Heap Round Robin

2-flow Specialized Strict

2-flow Binary Heap Strict

The order packets are popped is the same (phew!), but it seems they stay in the queue much longer in the heap-based versions.

[anshumanmohan]

Thanks for all the great work, and for walking me through this IRL! Comments on Github are minimal because we spoke offline.

[anshumanmohan]

@jku20 Akash has added a fud2 stage that runs python3 on a file. This spares us from the usual pattern of:

python3 myfile.py | 
fud ...

Would you please take a look for style etc? Also we know this should have been its own PR; sorry about that!