`-mt > 1` Ignores Violations about freerouting HOT 11 CLOSED

Possibly related tickets:

• #191
• #155

from freerouting.

Trialling on a smaller version of the problem that runs faster (I will upload shortly), it could be that the violation is not kept as a viable solution, but is still displayed. It is unfortunate if this is the case, as for autorouting that requires more computation it can be a very long time before another potential solution is rendered - so it looks like it settles on a bad solution.

from freerouting.

Obviously a problems in the above image: blue traces overlap behind the large red square.

java -jar freerouting-executable.jar -l en -de board.dsn -do board.ses -mp 1000 -mt $(np) -oit 0 -us hybrid -hr 1:1 -is random

In the terminal I see:

2023-08-31 10:47:33.678 [Thread-503] WARN  OptViaAlgo.opt_via_location: move via failed
2023-08-31 10:47:34.580 [Thread-554] WARN  OptViaAlgo.opt_via_location: move via failed
2023-08-31 10:47:51.161 [Thread-0] INFO  Saving '/tmp/freerouting-f874feab.frb'...
2023-08-31 10:47:57.338 [Thread-1863] WARN  OptViaAlgo.opt_via_location: move via failed

And it now appears to be 'stuck' on a bad solution.

from freerouting.

Also to prove it is possible to pass DRC (for the traces):

from freerouting.

It has come to a pretty similar solution, but still has this issue with crossing traces at the bottom and the warning message being spit out:

2023-08-31 11:34:42.896 [Thread-210361] WARN  OptViaAlgo.opt_via_location: move via failed

I can't manually drag it either.

from freerouting.

Meanwhile, -mt 1 solves this problem in 0.27ms:

2023-08-31 12:49:04.571 [main] INFO  Freerouting v1.8.0 (build-date: 2023-08-31)
2023-08-31 12:49:04.580 [main] INFO  Opening 'board.dsn'...
2023-08-31 12:49:05.099 [main] INFO  Opening 'board.rules'...
2023-08-31 12:49:05.100 [main] ERROR File 'board.rules' was not found.
2023-08-31 12:49:05.100 [main] ERROR File 'board.rules' was not found.
2023-08-31 12:49:07.545 [Thread-0] INFO  Starting auto-routing...
2023-08-31 12:49:07.683 [Thread-0] INFO  Saving '/tmp/freerouting-f874feab.frb'...
2023-08-31 12:49:07.818 [Thread-0] INFO  Auto-routing was completed in 0.27 seconds.
2023-08-31 12:49:07.825 [Thread-0] INFO  Saving 'board.ses'...

Definitely a big problem here in the threading.

from freerouting.

Thank you @daniel-theia for the detailed report!

Yes, multithreading is broken and it's likely to be permanently disabled in the future.

I'll add an extra warning message for now to discourage the usage of the -mt command line argument.

from freerouting.

Thank you @daniel-theia for the detailed report!

Yes, multithreading is broken and it's likely to be permanently disabled in the future.

I'll add an extra warning message for now to discourage the usage of the -mt command line argument.

Any idea how it might be fixed? I have some complex boards that could do with any acceleration they can get. I did have a look through for a few hours, but never really got to the bottom of anything.

I'm still yet to fully understand what the approach in the routing software, in places it reads a lot like C code? If you have some high-level idea how it is searching this enormous space it might add some context when reading the code.

from freerouting.

I know, it's not easy at all, it challenges me as well. One of the problems is that I inherited this code a few years ago from the author who seems to be unreachable, so there are still blurry parts for me too.

Answering to your questions though: it doesn't use C code, it's Java only.
It reads the data from the DSN files using the SpecctraDsnFileReader and we will get a object-representation of the board in the end. That representation consists of classes from the designforms.specctra folder.
The auto-router does its thing through the AutorouteEngine class and then we either run BatchOptRoute (single-threaded) or BatchOptRouteMT (multi-threaded) route optimizer.

The important takeaway here is that the auto-routing is always single-threaded, only the second stage, the route-optimization gets multi-threaded (if enabled). So your potential gain is limited.

I hope it helps somewhat!

from freerouting.

from freerouting.

@C64HexCoder That's not something we control. You can check your GitHub settings again or set an email filter.

I know, it's not easy at all, it challenges me as well. One of the problems is that I inherited this code a few years ago from the author who seems to be unreachable, so there are still blurry parts for me too.

Ah okay, you've done a great job so far it seems, so well done on that. It's a super valuable tool, it just needs some instability ironing out.

Answering to your questions though: it doesn't use C code, it's Java only.

I understand that, I meant that it reads like it was original written in C, or that the original developer liked C.

The important takeaway here is that the auto-routing is always single-threaded, only the second stage, the route-optimization gets multi-threaded (if enabled). So your potential gain is limited.

Interesting, thank you. Maybe I could take a look at BatchOptRoute and write a new multi-threading system around it. It would probably also teach me a lot about how this all actually works too.

from freerouting.