================ ---------------- osa1 wrote: If the predecessor is unreachable, could we wrap the incoming value with the original trunc we're elminating and still reduce the rest of the graph? https://github.com/llvm/llvm-project/pull/226383