Support nested loop

[tancheng]

Two issues here

• Outer loop's control (i.e., branch) no need to reach inner loop's basic block
• Overall II can be decreased with HW supported fusion

[Aegithalos-caudatus]

Also for multi-dimensional address generation, the getelementptr cannot be mapped to a simple ADD anymore.

[tancheng]

Also for multi-dimensional address generation, the getelementptr cannot be mapped to a simple ADD anymore.

Thanks for the input!