Three Wasted Cycles

This Y86-64 sequence runs on the CS:APP PIPE processor: five stages (F, D, E, M, W), full forwarding, and a branch predictor that always predicts taken.

1  mrmovq 8(%rbx), %rax
2  addq   %rax, %rdx
3  subq   %rdx, %rcx
4  jne    L1          # predicted taken; at run time it is NOT taken
5  irmovq $1, %rsi     # the instruction after the jne

How many bubble cycles does the pipeline inject while these instructions flow through?