LLVM 4 #3
LLVM 4 #3
Conversation
…t should be very close
… files is the way to go here...
…source copy is compiling so...
… what's in the PassBuilder.cpp of llvm
…nying header); also changed the CMake file. This still has a linkage error but the missing symbols for the pass registration are not an issue any longer. This is more verbose and perhaps contains items not necessary than before, but with the new pass manager this is clearly the way to move (I think) and so... commit
…LLVM 4 build being built without RTTI, recompiling fixed that as that's what seems what is wanted.
…ping it as I did will make this a bit more teidious than it should be...
… the Internal header
|
Sorry my fault this is taking too long. |
|
@caballa no worries... honestly, I've got a bunch to do before the 4th, so I wasn't even thinking of checking in until after the holiday, and I'm not in a rush. |
| @@ -488,16 +494,16 @@ void PassManagerBuilder::addLTOOptimizationPasses(PassManagerBase &PM) { | |||
| if (UseNewSROA) | |||
There was a problem hiding this comment.
No need of if-then-else here
| @@ -1760,7 +1683,7 @@ Value *InstCombiner::FoldOrOfICmps(ICmpInst *LHS, ICmpInst *RHS, | |||
| // 4) LowRange1 ^ LowRange2 and HighRange1 ^ HighRange2 are one-bit mask. | |||
| // This implies all values in the two ranges differ by exactly one bit. | |||
|
|
|||
| if (!AvoidBv && (LHSCC == ICmpInst::ICMP_ULT || LHSCC == ICmpInst::ICMP_ULE) && | |||
| if ((LHSCC == ICmpInst::ICMP_ULT || LHSCC == ICmpInst::ICMP_ULE) && | |||
There was a problem hiding this comment.
Forgot here to add (!AvoidBv)
| bool MinimizeSize, bool ExpensiveCombines, AliasAnalysis *AA, | ||
| AssumptionCache &AC, TargetLibraryInfo &TLI, | ||
| DominatorTree &DT, const DataLayout &DL, LoopInfo *LI) | ||
| : Worklist(Worklist), Builder(Builder), MinimizeSize(MinimizeSize), |
There was a problem hiding this comment.
AvoidBv is not initialized. It should be initialized to true in the constructor.
| class DbgDeclareInst; | ||
| class MemIntrinsic; | ||
| class MemSetInst; | ||
|
|
There was a problem hiding this comment.
After the above forward declarations it should be part of the llvm_seahorn namespace
|
|
||
| FunctionPass *llvm::createInstructionCombiningPass(bool ExpensiveCombines) { |
There was a problem hiding this comment.
Why are we keeping llvm::InstructionCombiningPass? We just care about llvm_seahorn::SeaInstructionCombiningPass
| if (L->contains(BI->getSuccessor(0))) | ||
| P = ICmpInst::ICMP_SLT; | ||
| P = ICmpInst::ICMP_NE; |
There was a problem hiding this comment.
It should be ICMP_SLT to avoid disequalities. This change is important, it was actually the main reason for duplicating IndVarSimplify.cc
| else | ||
| P = ICmpInst::ICMP_EQ; | ||
|
|
||
| DEBUG(dbgs() << "INDVARS: Rewriting loop exit condition to:\n" | ||
| << " LHS:" << *CmpIndVar << '\n' | ||
| << " op:\t" | ||
| << (P == ICmpInst::ICMP_EQ ? "==" : "<") << "\n" | ||
| << (P == ICmpInst::ICMP_NE ? "!=" : "==") << "\n" |
| static void AddStandardLinkPasses(PassManagerBase &PM) { | ||
| llvm_seahorn::PassManagerBuilder Builder; | ||
| static void AddStandardLinkPasses(legacy::PassManagerBase &PM) { | ||
| PassManagerBuilder Builder; |
There was a problem hiding this comment.
It should be llvm_seahorn::PassManagerBuilder. Otherwise, we will use the llvm passes.
| @@ -0,0 +1,1172 @@ | |||
| //===- Parsing, selection, and construction of pass pipelines -------------===// | |||
There was a problem hiding this comment.
There is already a comment in tools/opt/opt.cpp. We need to use the PassBuilder defined in llvm_seahorn, otherwise seaopt will call the llvm pass rather than the seahorn passes. So I don't think we need this file.
| @@ -0,0 +1,314 @@ | |||
| //===- Parsing, selection, and construction of pass pipelines --*- C++ -*--===// | |||
There was a problem hiding this comment.
We don't need this file for same reason as described in tools/opt/PassBuilder.cpp.
Alright, here goes again. I think this ought to be much better. That said, two areas I would look at (beyond the AvoidBv flag, for instance) are the IndVarSimplify.cc file, where in 4.0 there are a fair number of changes making the 3.8 into a legacy pass (and I did try to reference the 3.8 branch - that was helpful) and so I didn't move this into the llvm_seahorn namespace; and then similarly in the opt folder I included some of the current 4.0 code because it didn't seem as if it would hurt and figured it might make it easier to integrate then with 4.x...
But please, let me know if I've got something wrong. Thanks again for all the help and pointers, I definitely made this into more of a project than I needed to, but.. there's a first time to everything.