Open tobiasgrosser opened 5 days ago
When doing the keyword implementations, I skipped over exact (and disjoint) because they are not part of MLIR's LLVM dialect https://mlir.llvm.org/docs/Dialects/LLVM/#llvmsdiv-llvmsdivop They could be added though
Right, adding them could be useful. Meanwhile, maybe filter out test cases that contain the exact keyword.
Now, maybe adding exact support to MLIR's llvm dialect is easier than skipping over these benchmarks?
It seems the exact keyword in the generated proof is not preserved: https://github.com/opencompl/lean-mlir/blob/6e76106791a32338882c53685cd860527adda64d/SSA/Projects/InstCombine/tests/LLVM/gexact.lean#L20
The original test case for this is:
Our incomplete semantics currently lead to verification failures in the test data @luisacicolini generates. @lfrenot, can you look into this on Monday?