[PATCH] D108443: [MCA] Fixing bug causing LSUnit to not be notified when a 0 latency instruction finishes executing.

[Andrea Di Biagio via llvm-commits]

On Fri, 20 Aug 2021 at 18:58, Patrick Holland via Phabricator <
reviews at reviews.llvm.org> wrote:

> holland11 added a comment.
>
> Out of curiosity, how would creating a lit test work for this bug? The bug
> itself (under the input file and scheduling model that I have with my
> downstream target) causes mca to never terminate. Does lit have a default
> timeout so that we could just give the test an input that would cause this
> infinite run-time and then make sure that mca actually terminates without
> hitting the timeout?
>

Unfortunately there isn’t a good way to test bugs like this.
What you described is probably the only way. That being said, it isn’t nice
for a test to be dependent on a timeout only.

It could be useful to add a debug option to “force fail” the entire
simulation after N cycles (with N customisable from command line).

That option would simplify writing tests for cases where stages are stuck
in a endless loop of inactivity.


>
> Repository:
>   rG LLVM Github Monorepo
>
> CHANGES SINCE LAST ACTION
>   https://reviews.llvm.org/D108443/new/
>
> https://reviews.llvm.org/D108443
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210821/94651178/attachment.html>