[llvm-commits] [llvm] r41489 - in /llvm/trunk: include/llvm/CodeGen/SelectionDAGISel.h lib/CodeGen/SelectionDAG/SelectionDAGISel.cpp test/CodeGen/X86/memmove-0.ll test/CodeGen/X86/memmove-1.ll test/CodeGen/X86/memmove-2.ll test/CodeGen/X86/memmove-3.ll

[Chris Lattner]

On Aug 27, 2007, at 10:46 AM, Dan Gohman wrote:

>> This is nifty, but shouldn't this be done at the LLVM IR level?  I
>> can't think of cases where lowering would create new memmove calls.
>
> I'm not aware of any cases where memmove vs. memcpy matters before
> codegen currently. If it were done at the LLVM IR level, would it be
> too trivial to be in a pass by itself, or would you want to add it,
> along with an addRequired<AliasAnalysis>, to some existing pass?

I'm not really sure that there is any place that is truly natural.  I  
suspect that the best place would be the new GVN pass that Owen  
should be turning on "real soon now", as it already uses alias info.   
Owen, how's that going? :)

-Chris