Matching SVF’s alias pair enumeration with LLVM’s aa-eval

[parjavadian]

Hi, I am trying to compare SVF’s alias analysis results against LLVM’s aa-eval in a fair way. Is there a flag or recommended approach to make SVF enumerate the same pointer pairs that aa-eval checks (i.e., only pairs arising from real use sites such as loads, stores, calls, and returns, normalized to top-level pointers)?

On the same bitcode, LLVM’s aa-eval reports 1165 NoAlias and 702 MayAlias, while SVF (using wpa -ander -print-aliases) reports 104,925 NoAlias and 272 MayAlias. This suggests that the sets of pointer pairs differ significantly, making it hard to compare precision directly.

[yuleisui]

I don't think there is a flag that can enumerate the same pointer pairs, as SVF performs interprocedural analysis, the nodes on PAG are the pointers/objects for the entire program.