Add global history predictor (Fix issue #126)

Created by: zachzzc

Add a gshare global history predictor to enhance the performance and fix #126
By compare the mispredictions on benchmarks, a global history register with length = 4 performed best. Here is the comparison between the original and the new branch predictor result.

It doesn't perform well on qsort benchmark. My understanding is that in the insertion sort, the pattern between taken and untaken is not regular. It makes global history varies and make gshare harder to learn the prediction.

Synthesis test was run and passed.