[PATCH] Implement ADRP CSE for global symbols

[Quentin Colombet]

Hi Jiangning,

A target hook seems like a good idea.
I would be more agressive on its scope though.

> What about we create a target specific hook and allow target/back-end to choose the alignment between RoundupToPowerOfTwo(sizeof(MergedGlobal)) and NaturalAlignment(MergedGlobal)?

Instead of choosing the alignment between RoundupToPowerOfTwo(sizeof(MergedGlobal)) and NaturalAlignment(MergedGlobal), what about choosing an alignment period :).
We could then assert that the returned alignment is at least equal to the natural alignment. The default implementation of the target hook could just return the natural alignment.

What do you think?

Thanks,
-Quentin

http://reviews.llvm.org/D3432