[PATCH] Optimize memcmp(x, y, n)==0 for small n and suitably aligned x/y.

[Eelis]

Thanks Joerg, that's a great point!

I don't really know how to do that though. I see that the memcmp code in SelectionDAGBuilder.cpp uses allowsMisalignedMemoryAccesses(), but that function takes an EVT, a type which is not known in Transforms/Utils/SimplifyLibCalls.cpp.


http://reviews.llvm.org/D6952

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/