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
It is unclear whether memory and/or register loads should be considered side-effecting. This affects downstream uses and transformations in significant ways. Most notably, side-effecting operations should be excluded from unused code elimination.
Currently, these are treated as not side-effecting, but this causes inaccuracies in the later Alive2 evaluation. Since the loads may be—but are not always—removed, this can create mismatches in the translation validation where it sees an unmatched load() intrinsic on only one side.
To make this more consistent, we would need to define the semantics and implement them across the toolchain. Additionally, it would be nice to verify that other lifters under test are similar in this way.
It is unclear whether memory and/or register loads should be considered side-effecting. This affects downstream uses and transformations in significant ways. Most notably, side-effecting operations should be excluded from unused code elimination.
Currently, these are treated as not side-effecting, but this causes inaccuracies in the later Alive2 evaluation. Since the loads may be—but are not always—removed, this can create mismatches in the translation validation where it sees an unmatched load() intrinsic on only one side.
To make this more consistent, we would need to define the semantics and implement them across the toolchain. Additionally, it would be nice to verify that other lifters under test are similar in this way.
This issue is to record some thoughts from #21.
The text was updated successfully, but these errors were encountered: