[PATCH] D113578: [SCEV] Apply loop guards when computing max BTC for arbitrary steps.

[Alexander Kornienko via Phabricator via llvm-commits]

alexfh added a comment.

Hi Florian,

this revision is causing compiler crashes on a number of translation units in our code. This _may_ be an increase in stack depth, but I'm not sure yet. Is significantly increased stack depth expected after this patch? Is it justifiable?

The failures look more or less like this:

  1.	<eof> parser at end of file
  2.	Code generation
  3.	Running pass 'Function Pass Manager' on module '...'.
  4.	Running pass 'Loop Pass Manager' on function '@...'
  5.	Running pass 'Induction Variable Users' on basic block '%107409'
    #0 0x0000562126256c38 llvm::sys::RunSignalHandlers() (clang+0x6c56c38)
    #1 0x000056212625926c SignalHandler(int) (clang+0x6c5926c)
    #2 0x00007faea56f9750 __restore_rt (libpthread.so.0+0x15750)
    #3 0x0000562125ec3b1b computeKnownBitsFromAssume(llvm::Value const*, llvm::KnownBits&, unsigned int, (anonymous namespace)::Query const&) (clang+0x68c3b1b)
    #4 0x0000562125eae55d computeKnownBits(llvm::Value const*, llvm::KnownBits&, unsigned int, (anonymous namespace)::Query const&) (clang+0x68ae55d)
    #5 0x0000562125ec2de5 computeKnownBitsFromOperator(llvm::Operator const*, llvm::APInt const&, llvm::KnownBits&, unsigned int, (anonymous namespace)::Query const&) (clang+0x68c2de5)
    #6 0x0000562125eaeb6e computeKnownBits(llvm::Value const*, llvm::APInt const&, llvm::KnownBits&, unsigned int, (anonymous namespace)::Query const&) (clang+0x68aeb6e)
    #7 0x0000562125eae55d computeKnownBits(llvm::Value const*, llvm::KnownBits&, unsigned int, (anonymous namespace)::Query const&) (clang+0x68ae55d)
    #8 0x0000562125eaecd9 llvm::computeKnownBits(llvm::Value const*, llvm::DataLayout const&, unsigned int, llvm::AssumptionCache*, llvm::Instruction const*, llvm::DominatorTree const*, llvm::OptimizationRemark
  Emitter*, bool) (clang+0x68aecd9)
    #9 0x0000562125e47339 llvm::ScalarEvolution::GetMinTrailingZerosImpl(llvm::SCEV const*) (clang+0x6847339)
   #10 0x0000562125e3143a llvm::ScalarEvolution::GetMinTrailingZeros(llvm::SCEV const*) (clang+0x683143a)
   #11 0x0000562125e482ea llvm::ScalarEvolution::getRangeRef(llvm::SCEV const*, llvm::ScalarEvolution::RangeSignHint) (clang+0x68482ea)
   #12 0x0000562125e3ba27 StrengthenNoWrapFlags(llvm::ScalarEvolution*, llvm::SCEVTypes, llvm::ArrayRef<llvm::SCEV const*>, llvm::SCEV::NoWrapFlags) (clang+0x683ba27)
   #13 0x0000562125e2facf llvm::ScalarEvolution::getAddExpr(llvm::SmallVectorImpl<llvm::SCEV const*>&, llvm::SCEV::NoWrapFlags, unsigned int) (clang+0x682facf)
   #14 0x0000562125e3c19d llvm::ScalarEvolution::getGEPExpr(llvm::GEPOperator*, llvm::SmallVectorImpl<llvm::SCEV const*> const&) (clang+0x683c19d)
   #15 0x0000562125e46e7e llvm::ScalarEvolution::createNodeForGEP(llvm::GEPOperator*) (clang+0x6846e7e)
   #16 0x0000562125e3fe3d llvm::ScalarEvolution::createSCEV(llvm::Value*) (clang+0x683fe3d)
   #17 0x0000562125e38c60 llvm::ScalarEvolution::getSCEV(llvm::Value*) (clang+0x6838c60)
   #18 0x0000562125e452f4 llvm::ScalarEvolution::createAddRecFromPHI(llvm::PHINode*) (clang+0x68452f4)
   #19 0x0000562125e46938 llvm::ScalarEvolution::createNodeForPHI(llvm::PHINode*) (clang+0x6846938)
   #20 0x0000562125e3ff14 llvm::ScalarEvolution::createSCEV(llvm::Value*) (clang+0x683ff14)
   #21 0x0000562125e38c60 llvm::ScalarEvolution::getSCEV(llvm::Value*) (clang+0x6838c60)
   #22 0x0000562125e3bd47 llvm::ScalarEvolution::getGEPExpr(llvm::GEPOperator*, llvm::SmallVectorImpl<llvm::SCEV const*> const&) (clang+0x683bd47)
   #23 0x0000562125e46e7e llvm::ScalarEvolution::createNodeForGEP(llvm::GEPOperator*) (clang+0x6846e7e)
   #24 0x0000562125e3fe3d llvm::ScalarEvolution::createSCEV(llvm::Value*) (clang+0x683fe3d)
   #25 0x0000562125e38c60 llvm::ScalarEvolution::getSCEV(llvm::Value*) (clang+0x6838c60)
   #26 0x0000562125e4dd1d llvm::ScalarEvolution::applyLoopGuards(llvm::SCEV const*, llvm::Loop const*) (clang+0x684dd1d)
  ...

I'm now working on a self-contained test case, but if you have any ideas of fixes, I can verify them on our real code.

Regards,
Alex


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D113578/new/

https://reviews.llvm.org/D113578