vfdff wrote: Thank you for your advice, so I need to adjust to **i8 GEP** directly here too? I've tried to adjust directly to i8 GEP before, but I see some test cases regression, then I'll see why. https://github.com/llvm/llvm-project/pull/107848