I’m always surprised to learn LLVM is so slow given that was one of the original motivations for developing it. I don’t know if that’s down to feature creep or intrinsic complexity being higher than people presumed was the case for GCC.