Observation for <<The Superblock: An Effective Technique>>
     The paper shows that superblock is an effective way to increase ILP
in programs. To show its effectiveness, the authors compared it with
the base level compilation. To form superblock, we need to know which
basic block is important and which is not. This can only be approximately
decided by profiling. If the profiling is different from actural result, what
is the effect on superblock scheduling?