AHA PNR Result

[MinjaeLee94]

Hello, I have a question regarding Amber CGRA mapping.

Using the provided Gaussian application example, we obtained the final PNR result using the aha map and pnr commands. After checking, the placement of PE tiles used was relatively widespread. Can you tell me why?

I understand that in order to solve the placement problem in the thunder library used in the AHA PNR process, HPWL (sum of squared distances from the center of the net to each block (cluster)), spreading potential, legalization energy, etc. are calculated. As a result, isn't the placement of the PE tiles used as a PNR result close to each other?

The picture below shows the results of gaussian application pnr in the order described in the AHA wiki (https://github.com/StanfordAHA/aha/wiki/CGRA-Design-Flow). The layout seems to be different from what I expected, so I would appreciate it if you could let me know if I misunderstood anything.

And the provided application does not perform global placement because "DISABLE_GP" is all set to 1 in the aha/util/application_parameters.json file. Is there a reason? I arbitrarily tried to perform PNR with the DISABLE_GP setting removed, but an error occurred during the PNR process and it stopped midway. Is “DISABLE_GP” all set to 1 because of this phenomenon?

Thank you for reading the question.