[P2] Simulation traces are not understandable for human being

[tancheng]

5: [tile0]: 00000000.0.0.0|00000000.0.0.0|00000000.0.0.0|00000000.0.0.0 => [(opt)1c|(in)1-2-3-4|(p)0|(out)5-0-0-0-0-0-0-0|(p_in)0-0-0-0-0-0] ([recv: 00000000.0.0.0,00000000.0.0.0,00000000.0.0.0,00000000.0.0.0] (ldcst)(P0) (const: 00000000.0.0.0, en: 1) ] = [out: 00000000.1.0.0,00000000.0.0.0] (recv_opt.rdy: 1, recv_in[0].rdy: 1, recv_in[1].rdy: 1, recv_predicate.msg: 0.0, (ldcst), recv_opt.en: 1, send[0].en: 1) ) => 00000000.1.1.0|00000000.0.0.0|00000000.0.0.0|00000000.0.0.0|00000000.0.0.0|00000000.0.0.0|00000000.0.0.0|00000000.0.0.0 => 00000000.1.0.0|00000000.0.0.0|00000000.0.0.0|00000000.0.0.0|00000000.0.0.0|00000000.0.0.0|00000000.0.0.0|00000000.0.0.0 => (00000000,1)|(00000000,0)|(00000000,0)|(00000000,0)(opt)1c|(in)1-2-3-4|(p)0|(out)5-0-0-0-0-0-0-0|(p_in)0-0-0-0-0-0 : [(opt)1c|(in)1-2-3-4|(p)0|(out)5-0-0-0-0-0-0-0|(p_in)0-0-0-0-0-0||(opt)1c|(in)1-2-3-4|(p)0|(out)5-0-0-0-0-0-0-0|(p_in)0-0-0-0-0-0||(opt)00|(in)0-0-0-0|(p)0|(out)0-0-0-0-0-0-0-0|(p_in)0-0-0-0-0-0||(opt)00|(in)0-0-0-0|(p)0|(out)0-0-0-0-0-0-0-0|(p_in)0-0-0-0-0-0||(opt)00|(in)0-0-0-0|(p)0|(out)0-0-0-0-0-0-0-0|(p_in)0-0-0-0-0-0||(opt)00|(in)0-0-0-0|(p)0|(out)0-0-0-0-0-0-0-0|(p_in)0-0-0-0-0-0||(opt)00|(in)0-0-0-0|(p)0|(out)0-0-0-0-0-0-0-0|(p_in)0-0-0-0-0-0||(opt)00|(in)0-0-0-0|(p)0|(out)0-0-0-0-0-0-0-0|(p_in)0-0-0-0-0-0] : (opt)1c|(in)1-2-3-4|(p)0|(out)5-0-0-0-0-0-0-0|(p_in)0-0-0-0-0-0||

I don't think existing simulation trace is human-friendly. HW-oriented development is not the way to go towards HW/SW co-design in my personal opinion. At least we need to add some prefix to each field to indicate what the waveform/data is for.

[yuqisun]

hi @tancheng , want to confirm the port order, is it same as

1. xbarTypeList = ["W", "E", "N", "S", "NE", "NW", "SE", "SW"] or
2. [NW(0, 0), N(0, 1) ... , SE(2, 2)] or
3. {0: "NORTH", 1: "SOUTH", 2: "WEST", 3: "EAST", 4: "NORTHWEST", 5: "NORTHEAST", 6: "SOUTHEAST", 7: "SOUTHWEST"}

So I can add direction in the beginning of trace log:
north_inport: {'payload': Bits64(0x0000000000000000) ...
...

[tancheng]

I didn't get the second one, I thought that is just for GUI?
The first one can be found here? https://github.com/tancheng/CGRA-Flow/blob/29a942e4fdde2847f4bba5f1be1518e134bdfc84/mode_dark_light.py#L43, which is same as 3?

[yuqisun]

Ok, working to format in a markdown table, looks better?

[tancheng]

Looks awesome. I think this work is super ChatGPT-friendly...
Your markdown table inspires me that can we have sth similar to ↓?

[tancheng]

It could also be based on verbosity, i.e., verbosity == 1 shows above figure, verbosity == 2 shows your markdown; It could also be a GUI that can be clickable, i.e., click one tile in above figure, shows how the dataflow is happening on the selected cycle.

LOL, might be to much, you can make the call~! Markdown already looks fancy!