[PATCH] [AArch64] Add pass to enable additional comparison optimizations by CSE

Sergey Dmitrouk sdmitrouk at accesssoftek.com
Wed Aug 27 09:46:39 PDT 2014

> Rather, is the implementation necessarily target-specific? I imagine other targets would benefit, but I'm asking if we could have a more generic solution.

I doubt whether more generic solution will work. Here is some [[ http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20140721/227116.html | context ]].
Final code heavily depends on how instructions are lowered (details of SelectionDAG construction is just one of things that affect it).
So I guess this optimization must be implementation-specific to be meaningful.

> I'll leave it to your good judgment to decide if this discussion is really necessary.

Then I won't submit a bug as chances that it'll get us somewhere are quite low.
Anyway, I'll keep generalization of this in mind. Thanks for bringing this up.



More information about the llvm-commits mailing list