   This paper proposed a hardware structure and corresponding compiler
support to handle dynamic memory disambiguation. From Table 2, we can
see that the code size increased by an average of 15.7% for the benchmarks.
This will probably affect cash miss rate and page fault rate. Since the simulated
architecture in Table 1 didn't consider memory size and TLB, we can not see
to what extent this will affect page fault rate.
    The check instruction is actually a branch instruction. This means the MCB
method will increase the total branch instructions in a program. Though the
frequency that this checks are taken is small, it may hurt the overall performance.
    The effectiveness of MCB depends on other compiler techniques. For example,
if the pipeline is pretty full before MCB, then applying MCB may actually bring
negative performance gain.

Zhigang Hu