You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In bea8510 we had to revert #6719. This change caused some memories that were previously targeted from OMIR to fail in the EmitOMIR pass. Opening this issue for tracking the effort to re-land the change, and will attempt to reduce the problem with a FIRRTL example here.
The text was updated successfully, but these errors were encountered:
mikeurbach
changed the title
[FIRRTL] LowerMemory change in #6719 blocks some memories from deduping
[FIRRTL] LowerMemory change in #6719 leads to ambiguous targets in EmitOMIR
Mar 14, 2024
I've updated the title and description. I still haven't been able to come up with a simple reproducer, but the gist of the problem is we have OMIR targeting the memories, and by the time we get to EmitOMIR, we expect a single path through the instance hierarchy to each memory:
I don't think anyone has been looking at addressing this. I could reproduce the issue on an internal design, but I never got around to writing a small FIRRTL reproducer based on the observations above. It should be pretty easy to write a test case that dedupes without the change in #6719 , but then fails to dedupe after it. We should include such a test case before we re-land this change.
In bea8510 we had to revert #6719. This change caused some memories that were previously targeted from OMIR to fail in the EmitOMIR pass. Opening this issue for tracking the effort to re-land the change, and will attempt to reduce the problem with a FIRRTL example here.
The text was updated successfully, but these errors were encountered: