Benchmarks in the CuPBoP/examples folder are based on the Rodinia benchmarks. Refer to the files in the Rodinia repo for each benchmarks when creating new benchmarks. Also refer to the following script to serve as the foundation for your script: General kjrun_llvm18.sh Script
If you look at all benchmarks in the CuPBoP/examples folders have similar kjrun_llvm18.sh scripts.
Seeing a divergent branch isn't a main issue of concern. If you see some other issues like not defined errors within a file or executing a script leads to a non-stop loop, it suggests that CuPBoP doesn't support that benchmark.
Some issues in regards to using the Makefile with the benchmarks are reported here