search for: 9978022

...houldMergeGEPs. I just tried with ppc64 and x86-64 and I am also getting better code without the GEP merging in InstructionCombining. I am not sure what the solution is yet but I think we are too aggressive when merging GEPs in InstructionCombining. Here is the details for ppc64: http://pastie.org/9978022 -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20150224/606b7b39/attachment.html>

Hello I am not sure I understand the logic for merging GEPs in InstructionCombining.cpp: static bool shouldMergeGEPs(GEPOperator &GEP, GEPOperator &Src) { // If this GEP has only 0 indices, it is the same pointer as // Src. If Src is not a trivial GEP too, don't combine // the indices. if (GEP.hasAllZeroIndices() && !Src.hasAllZeroIndices() &&

...g better > > code > > without the GEP merging in InstructionCombining. I am not sure what > > the solution is yet but I think we are too aggressive when merging > > GEPs in InstructionCombining. > > > > > > Here is the details for ppc64: http://pastie.org/9978022 > > > > > > First, thanks for posting this, it is quite useful. LLVM trunk's > PowerPC backend does a slightly better job now, but for an > irrelevant reason, so to stick with the code you generated, we have: > > .LBB0_1: # %for.b...