X86 branch prediction question

Hi all,

Does anyone know which of the following pieces of code is faster on Pentium & up processors? I'm not sure what the effect of static misprediction in the first piece is.

Code 1:

  CMP [reg1],reg2
  JA theend
  INT 5
theend:

Since "JA theend" is a conditional forward reference its static prediction is "not taken", but it will almost always be taken (in the case that most array indexes are valid)

Code 2:

  JMP test
fail:
  INT 5
test:
  CMP [reg1],reg2
  JNA fail