minimal example that reproduce hang on ara-2-lanes.

[yanghao]

Note: this minimal example is not as strong as the full blown program, which not only hangs on ara-2-lanes but also ara-4-lanes. And it seems there is a pattern that VL==128 hangs ara-2-lanes and VL==256 hangs ara-4-lanes (those are reproduced), and maybe VL==512 would hang ara-8-lanes, VL==1024 would hang ara-16-lanes (those are not tried out yet).

[mp-17]

Thanks a lot, @yanghao! I answered here -> https://github.com/pulp-platform/ara/issues/87

[mp-17]

Hello @yanghao, sorry for the very long delay.

I should have fixed the hang issue in the PR https://github.com/pulp-platform/ara/pull/102. Check that out and see if your full program still hangs!

Best, Matteo

[yanghao]

The minimal program that reproduce the hang indeed can pass now.

However, the full-blown program, even with reduced vector length, which was previously passing, is now hanging. Same for bigger values of vector length.

[mp-17]

What a shame! The number of fetched mask elements should be okay now; can you share the full program? If yes, I will work on it.

[yanghao]

The full program unfortunately cannot be shared ... let me try to see if I can create another smaller program to reproduce it. (could take a while)

[mp-17]

We have implemented new fixes for Ara. Two PRs should still be approved but are ready:

https://github.com/pulp-platform/ara/pull/114 https://github.com/pulp-platform/ara/pull/123

Do you want to check them out with your full program?

[mp-17]

The fixes are in the main branch now; feel free to try your program again and let us know! :-)

[yanghao]

@mp-17 Sorry for getting back late ... I finally got my workspace again and tested it out, all working fine now! Thanks for the effort guys!