Fast membership performance

perf stat says:

#    3.483 GHz
#    3.81  insn per cycle

⇔ cycles per nanosecond

The longest dependency is 2 cycles
There are 8 instructions in a loop, giving 8÷3.81 = 2.10 cycles

      y←?200⍴1e9                 ⍝ The algorithm is branchless,
      x←?1e6⍴1e9                 ⍝ so the data doesn't matter
      ⊢t←1e6×1 0 0 5 cmpx 'x∊y'  ⍝ Time per element in ps
759.8876953
      t×3.483×1e¯3               ⍝ in cycles
2.646688843