In generated code we sometimes see duplicated DMB instructions. We can make this more efficient by merging them.

For example, in (C2-compiled) java.util.concurrent.ConcurrentHashMap$Node::<init> we see:

  0x000003ffa85d73d0: dmb ish
  0x000003ffa85d73d4: mov x11, x4
  0x000003ffa85d73d8: lsr xmethod, x1, #9
  0x000003ffa85d73dc: str w11, [x1,#20]
  0x000003ffa85d73e0: dmb ishst
  0x000003ffa85d73e4: strb wzr, [x10,x12,lsl #0]
 ;; membar_volatile
  0x000003ffa85d73e8: dmb ish ;*putfield val
                                                ; - java.util.concurrent.ConcurrentHashMap$Node::<init>@16 (line 629)

 ;; membar_release
  0x000003ffa85d73ec: dmb ish
  0x000003ffa85d73f0: mov x11, x5
  0x000003ffa85d73f4: lsr xmethod, x1, #9
  0x000003ffa85d73f8: str w11, [x1,#24]
  0x000003ffa85d73fc: dmb ishst
  0x000003ffa85d7400: strb wzr, [x10,x12,lsl #0]
 ;; membar_volatile
  0x000003ffa85d7404: dmb ish ;*putfield next
                                                ; - java.util.concurrent.ConcurrentHashMap$Node::<init>@22 (line 630)

We see essentially the same effect in C1-compiled code.