Reservation unit as FIFO

[Vladimir N. Makarov]

Joern Rennecke wrote:

> >   Yes, it is interesting.  In general, to model FIFO we need more
> > powerful model than DFA and regular expressions.  But regular expression
>
> You could just pretrend you are scheduling two instructions instead of one.
> One for the part that is done before and in the FIFO, and another one
> that does the work after the FIFO, and data-dependent on the former.
>

An interesting approach too.  But the insn scheduler has no such
infrastructure.  Even if the insn could be splitted in two insns, the
scheduler can create a big gap between the two insns.  That is wrong.  So we
need to differ such dependencies.  In general it is a big work.  Although this
infrastructure could be used for other insn scheduling problems too.

Vlad