[PATCH] Add peephole optimization to use LEA instructions of Intel Atom

[Nadav Rotem]

Hi Preston, 

I understand that your in-order processor is affected by the scheduling decision and you need some peepholes after RA/scheduling.  But please don't change X86InstrInfo::convertToThreeAddress because this function is meant to be called before register allocation. Your change adds complexity to this function and does not help the non FixupLeaPass users.  Also, you really don't reuse much code from convertToThreeAddress. You don't need all of the code that handles the INCs and the 64bit instructions and the Shifts, etc. What you really need is a few peepholes inside your own pass. 

Also, maybe you should merge it with your other in-order ATOM pass that pads small functions that don't get inlined. It looks like all of these optimizations fall into the same category and I am not sure that we need lots of different passes for every micro architectural issue. 

Thanks,
Nadav 
  
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20130416/a639655b/attachment.html>