Where to find paramteers defined in autotuner.json?

[dalyles]

Hello,

I am trying to do an example run of autotuner.json for the gcd circuit nangate45. I copied/pasted the autotuner.json file from the sky130hd gcd. It seems to have a number of parameters listed which I will list here:

_SDC_CLK_PERIOD
CORE_UTILIZATION
CORE_ASPECT_RATIO
CORE_MARGIN
CELL_PAD_IN_SITES_GLOBAL_PLACEMENT
CELL_PAD_IN_SITES_DETAIL_PLACEMENT
_FR_LAYER_ADJUST
PLACE_DENSITY_LB_ADDON
_PINS_DISTANCE
CTS_CLUSTER_SIZE
CTS_CLUSTER_DIAMETER
_FR_GR_OVERFLOW

So my question is, where do you find these parameters? I assume you have to set them to the values determined by the Autotuner. And also why do some parameters start with an underscore?

Thanks

[vijayank88]

@vvbandeira Can you please respond to this?

[vvbandeira]

@dalyles we do not have a docs to point to you at this moment. But in short:

• Variables without underscore are any variable you can give the make command. For these, we have a non-exhaustive list here.
• Variables with an underscore are special cases, mainly values that need to be written in the sdc and fastroute.tcl files. You can look at the code that handles them to better understand their behaviour here and here.

[dalyles]

Thanks,

I also have a followup question.
I successfully ran the autotuner for gcd in nangate45.
However, the run ID with the optimal parameters, 84eb1d92,has an incomplete list of logs that you would find in /OpenROAD-flow-scripts/flow/logs/nangate45/gcd/test-tune-2023-06-26-17-27-48/variant-AutoTunerBase-84eb1d92-or for example.

Right here shows the .json file with the corresponding run ID

Is this behavior expected or is the tool falsely giving optimal parameters without regards to where a run is completed?

Here is a screenshot of 2_2_floorplan_io.log for reference

Thanks

[vvbandeira]

This is a bug, please file an issue with the required steps and files to reproduce.

[vijayank88]

@dalyles
can you share the command and autotuner.json file used?

[dalyles]

Sure, the command I used is:

python3 distributed.py --design gcd --platform nangate45 --config ../designs/nangate45/gcd/autotuner.json tune

And here's the autotuner.json file: (for some reason github doesn't allow me to attach json files)

{
    "_SDC_FILE_PATH": "constraint.sdc",
    "_SDC_CLK_PERIOD": {
        "type": "float",
        "minmax": [
            1.0,
            4.3647
        ],
        "step": 0
    },
    "CORE_UTILIZATION": {
        "type": "int",
        "minmax": [
            5,
            100
        ],
        "step": 1
    },
    "CORE_ASPECT_RATIO": {
        "type": "float",
        "minmax": [
            0.5,
            2.0
        ],
        "step": 0
    },
    "CORE_MARGIN": {
        "type": "int",
        "minmax": [
            2,
            2
        ],
        "step": 0
    },
    "CELL_PAD_IN_SITES_GLOBAL_PLACEMENT": {
        "type": "int",
        "minmax": [
            0,
            5
        ],
        "step": 1
    },
    "CELL_PAD_IN_SITES_DETAIL_PLACEMENT": {
        "type": "int",
        "minmax": [
            0,
            5
        ],
        "step": 1
    },
    "_FR_LAYER_ADJUST": {
        "type": "float",
        "minmax": [
            0.1,
            0.7
        ],
        "step": 0
    },
    "PLACE_DENSITY_LB_ADDON": {
        "type": "float",
        "minmax": [
            0.0,
            0.99
        ],
        "step": 0
    },
    "_PINS_DISTANCE": {
        "type": "int",
        "minmax": [
            1,
            4
        ],
        "step": 1
    },
    "CTS_CLUSTER_SIZE": {
        "type": "int",
        "minmax": [
            10,
            200
        ],
        "step": 1
    },
    "CTS_CLUSTER_DIAMETER": {
        "type": "int",
        "minmax": [
            20,
            400
        ],
        "step": 1
    },
    "_FR_FILE_PATH": "../../../platforms/nangate45/fastroute.tcl",
    "_FR_GR_OVERFLOW": {
        "type": "int",
        "minmax": [
            1,
            1
        ],
        "step": 0
    }
}

[vijayank88]

@dalyles
See my screenshot used with your autotuner.json. I am getting all logs relevant to optimal json parameters.

[dalyles]

Hmm strange...so it looks like it successfully generates the logs on your machine when you use the same exact commands to reproduce the results.

[dalyles]

I did a rerun and it mysteriously has all the logfiles now.