[PATCH] [SeparateConstOffsetFromGEP] Fix bugs and improve the current SeparateConstOffsetFromGEP

[Hao Liu]

Hi Jingyue,

I agree with you.
The test case about "1 + 2" are corner cases, it won't affect too much about the performance.
For the "((a << 2) + 12) | 1"  issue, it is more common than above case. If instcombine can do optimization for such case, I think it can have some small improvement on performance.

I agree to not do such improvement. For other minor changes which you agree, I've added them to the patch in D5864, which is the key thing I want to do. I hope you can have a look and review it.

Thanks,
-Hao

http://reviews.llvm.org/D5863