What will be the access pattern of CUDA threads in shared memory during a Bank Conflict?

Question

I\'m a little bit confused about the access to the shared memory during a bank conflict. Let\'s assume that the word size is 4 Bytes and there is a total of 32 Banks in the shar