Instruction bundling on IA64 and the Intel Assembler

[Richard Kenner]

    NOPs have to be considered when determining bundle boundaries, so I think
    that the naive three-insns-farther approach will be correct.

Sure, but that's not what I meant.  What I meant is that you can't simply
look three insns ahead when the bundle start insn is emitted because the
nops haven't yet been emitted.  So you have to know where to look forward
and that requires *some* knowlege of the function in question.