grt: why do guide segments for local nets always start at layer index 1?

[donn]

I mentioned this before but I was working a proprietary PDK with a metal stack such that all signals were only allowed to use the two topmost metal layers, and the cells were designed with this in mind and only have pins on the second-to-last metal layer.

I recently had the chance to resume this work and ran into an issue where the global router, despite set_routing_layers, would continue to create guides on layers that it was not allowed to use (and which had no pins).

With some tinkering I found this line:

OpenROAD/src/grt/src/GlobalRouter.cpp

Line 3197 in a5b10a7

for (int l = 1; l <= last_layer; l++) {

Changing the l = 1 to be essentially l = min(pin_layers, min_routing_layer) solves the issue in my specific situation, but I have to wonder if I'm missing something…? I don't know much about how the global router works so this could be bad for other reasons.

[maliberty]

@eder-matheus

[eder-matheus]

@donn Sorry for the delay on answering it, it got lost on my notifications.
It seems to be a code that was necessary in the past, where GRT didn't had a good awareness of pin layer (it basically assumed that all pins were on layer 1). I'll try the fix you mentioned with our designs and if everything looks good, I'll create a PR for it.

[eder-matheus]

@donn The fix was merged with master. Could you confirm it works in your test case?

[donn]

Pulled #9975 and testing…

[donn]

Fix from #9975 works great. Thank you.