Why the presence of dynamic parallelism in unused kernel breaks cooperative groups?

Question

I have two kernels in my program. kernel1 uses cooperative groups and is supposed to be launched with cudaLaunchCooperativeKernel. kernel2