The old switch-case instruction decode implementation results in a
sparse table, which is not resolved by a fast jump table routing.
Therefore, it is really slow in instruction decoding phase.

R=joransiu@ca.ibm.com, michael_dawson@ca.ibm.com, mbrandy@us.ibm.com
BUG=

Review URL: https://codereview.chromium.org/1873963002

Cr-Commit-Position: refs/heads/master@{#35589}