From 3dd7de023be43bca7d6d45140e4fde42e22bc336 Mon Sep 17 00:00:00 2001 From: "Brian M. Rzycki" Date: Fri, 16 Feb 2018 16:35:17 +0000 Subject: [PATCH 2/2] [JumpThreading] PR36133 enable/disable DominatorTree for LVI analysis Summary: The LazyValueInfo pass caches a copy of the DominatorTree when available. Whenever there are pending DominatorTree updates within JumpThreading's DeferredDominance object we cannot use the cached DT for LVI analysis. This commit adds the new methods enableDT() and disableDT() to LVI. JumpThreading also sets the appropriate usage model before calling LVI analysis methods. Fixes https://bugs.llvm.org/show_bug.cgi?id=36133 Reviewers: sebpop, dberlin, kuhar Reviewed by: sebpop, kuhar Subscribers: uabelho, llvm-commits, aprantl, hiraditya, a.elovikov Differential Revision: https://reviews.llvm.org/D42717 git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@325356 91177308-0d34-0410-b5e6-96231b3b80d8 --- include/llvm/Analysis/LazyValueInfo.h | 7 ++++ include/llvm/IR/Dominators.h | 3 ++ lib/Analysis/LazyValueInfo.cpp | 30 +++++++++++++++- lib/IR/Dominators.cpp | 3 ++ lib/Transforms/Scalar/JumpThreading.cpp | 37 ++++++++++++++++++++ test/Transforms/JumpThreading/pr36133.ll | 44 ++++++++++++++++++++++++ 6 files changed, 123 insertions(+), 1 deletion(-) create mode 100644 test/Transforms/JumpThreading/pr36133.ll diff --git a/llvm/include/llvm/Analysis/LazyValueInfo.h b/llvm/include/llvm/Analysis/LazyValueInfo.h index 787c88cc6ec..cea5bf0df80 100644 --- a/llvm/include/llvm/Analysis/LazyValueInfo.h +++ b/llvm/include/llvm/Analysis/LazyValueInfo.h @@ -113,6 +113,13 @@ public: /// in LVI, so we need to pass it here as an argument. void printLVI(Function &F, DominatorTree &DTree, raw_ostream &OS); + /// Disables use of the DominatorTree within LVI. + void disableDT(); + + /// Enables use of the DominatorTree within LVI. Does nothing if the class + /// instance was initialized without a DT pointer. + void enableDT(); + // For old PM pass. Delete once LazyValueInfoWrapperPass is gone. void releaseMemory(); diff --git a/llvm/include/llvm/IR/Dominators.h b/llvm/include/llvm/IR/Dominators.h index c5373376ade..ccb18527abc 100644 --- a/llvm/include/llvm/IR/Dominators.h +++ b/llvm/include/llvm/IR/Dominators.h @@ -342,6 +342,9 @@ public: /// \brief Returns true if DelBB is awaiting deletion at a flush() event. bool pendingDeletedBB(BasicBlock *DelBB); + /// \brief Returns true if pending DT updates are queued for a flush() event. + bool pending(); + /// \brief Flushes all pending updates and block deletions. Returns a /// correct DominatorTree reference to be used by the caller for analysis. DominatorTree &flush(); diff --git a/llvm/lib/Analysis/LazyValueInfo.cpp b/llvm/lib/Analysis/LazyValueInfo.cpp index d7da669f6e7..fcbfb08cf1d 100644 --- a/llvm/lib/Analysis/LazyValueInfo.cpp +++ b/llvm/lib/Analysis/LazyValueInfo.cpp @@ -401,6 +401,7 @@ namespace { AssumptionCache *AC; ///< A pointer to the cache of @llvm.assume calls. const DataLayout &DL; ///< A mandatory DataLayout DominatorTree *DT; ///< An optional DT pointer. + DominatorTree *DisabledDT; ///< Stores DT if it's disabled. ValueLatticeElement getBlockValue(Value *Val, BasicBlock *BB); bool getEdgeValue(Value *V, BasicBlock *F, BasicBlock *T, @@ -463,13 +464,30 @@ namespace { TheCache.eraseBlock(BB); } + /// Disables use of the DominatorTree within LVI. + void disableDT() { + if (DT) { + assert(!DisabledDT && "Both DT and DisabledDT are not nullptr!"); + std::swap(DT, DisabledDT); + } + } + + /// Enables use of the DominatorTree within LVI. Does nothing if the class + /// instance was initialized without a DT pointer. + void enableDT() { + if (DisabledDT) { + assert(!DT && "Both DT and DisabledDT are not nullptr!"); + std::swap(DT, DisabledDT); + } + } + /// This is the update interface to inform the cache that an edge from /// PredBB to OldSucc has been threaded to be from PredBB to NewSucc. void threadEdge(BasicBlock *PredBB,BasicBlock *OldSucc,BasicBlock *NewSucc); LazyValueInfoImpl(AssumptionCache *AC, const DataLayout &DL, DominatorTree *DT = nullptr) - : AC(AC), DL(DL), DT(DT) {} + : AC(AC), DL(DL), DT(DT), DisabledDT(nullptr) {} }; } // end anonymous namespace @@ -1791,6 +1809,16 @@ void LazyValueInfo::printLVI(Function &F, DominatorTree &DTree, raw_ostream &OS) } } +void LazyValueInfo::disableDT() { + if (PImpl) + getImpl(PImpl, AC, DL, DT).disableDT(); +} + +void LazyValueInfo::enableDT() { + if (PImpl) + getImpl(PImpl, AC, DL, DT).enableDT(); +} + // Print the LVI for the function arguments at the start of each basic block. void LazyValueInfoAnnotatedWriter::emitBasicBlockStartAnnot( const BasicBlock *BB, formatted_raw_ostream &OS) { diff --git a/llvm/lib/IR/Dominators.cpp b/llvm/lib/IR/Dominators.cpp index e44e845b324..5efd0df5db9 100644 --- a/llvm/lib/IR/Dominators.cpp +++ b/llvm/lib/IR/Dominators.cpp @@ -453,6 +453,9 @@ bool DeferredDominance::pendingDeletedBB(BasicBlock *DelBB) { return DeletedBBs.count(DelBB) != 0; } +/// \brief Returns true if pending DT updates are queued for a flush() event. +bool DeferredDominance::pending() { return !PendUpdates.empty(); } + /// \brief Flushes all pending updates and block deletions. Returns a /// correct DominatorTree reference to be used by the caller for analysis. DominatorTree &DeferredDominance::flush() { diff --git a/llvm/lib/Transforms/Scalar/JumpThreading.cpp b/llvm/lib/Transforms/Scalar/JumpThreading.cpp index 4d366e8e392..9ff91cf4924 100644 --- a/llvm/lib/Transforms/Scalar/JumpThreading.cpp +++ b/llvm/lib/Transforms/Scalar/JumpThreading.cpp @@ -425,6 +425,7 @@ bool JumpThreadingPass::runImpl(Function &F, TargetLibraryInfo *TLI_, LoopHeaders.clear(); DDT->flush(); + LVI->enableDT(); return EverChanged; } @@ -617,6 +618,10 @@ bool JumpThreadingPass::ComputeValueKnownInPredecessors( // "X < 4" and "X < 3" is known true but "X < 4" itself is not available. // Perhaps getConstantOnEdge should be smart enough to do this? + if (DDT->pending()) + LVI->disableDT(); + else + LVI->enableDT(); for (BasicBlock *P : predecessors(BB)) { // If the value is known by LazyValueInfo to be a constant in a // predecessor, use that information to try to thread this block. @@ -630,6 +635,10 @@ bool JumpThreadingPass::ComputeValueKnownInPredecessors( /// If I is a PHI node, then we know the incoming values for any constants. if (PHINode *PN = dyn_cast(I)) { + if (DDT->pending()) + LVI->disableDT(); + else + LVI->enableDT(); for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) { Value *InVal = PN->getIncomingValue(i); if (Constant *KC = getKnownConstant(InVal, Preference)) { @@ -759,6 +768,10 @@ bool JumpThreadingPass::ComputeValueKnownInPredecessors( const DataLayout &DL = PN->getModule()->getDataLayout(); // We can do this simplification if any comparisons fold to true or false. // See if any do. + if (DDT->pending()) + LVI->disableDT(); + else + LVI->enableDT(); for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) { BasicBlock *PredBB = PN->getIncomingBlock(i); Value *LHS = PN->getIncomingValue(i); @@ -792,6 +805,10 @@ bool JumpThreadingPass::ComputeValueKnownInPredecessors( if (!isa(CmpLHS) || cast(CmpLHS)->getParent() != BB) { + if (DDT->pending()) + LVI->disableDT(); + else + LVI->enableDT(); for (BasicBlock *P : predecessors(BB)) { // If the value is known by LazyValueInfo to be a constant in a // predecessor, use that information to try to thread this block. @@ -820,6 +837,10 @@ bool JumpThreadingPass::ComputeValueKnownInPredecessors( match(CmpLHS, m_Add(m_Value(AddLHS), m_ConstantInt(AddConst)))) { if (!isa(AddLHS) || cast(AddLHS)->getParent() != BB) { + if (DDT->pending()) + LVI->disableDT(); + else + LVI->enableDT(); for (BasicBlock *P : predecessors(BB)) { // If the value is known by LazyValueInfo to be a ConstantRange in // a predecessor, use that information to try to thread this @@ -901,6 +922,10 @@ bool JumpThreadingPass::ComputeValueKnownInPredecessors( } // If all else fails, see if LVI can figure out a constant value for us. + if (DDT->pending()) + LVI->disableDT(); + else + LVI->enableDT(); Constant *CI = LVI->getConstant(V, BB, CxtI); if (Constant *KC = getKnownConstant(CI, Preference)) { for (BasicBlock *Pred : predecessors(BB)) @@ -1102,6 +1127,10 @@ bool JumpThreadingPass::ProcessBlock(BasicBlock *BB) { // threading is concerned. assert(CondBr->isConditional() && "Threading on unconditional terminator"); + if (DDT->pending()) + LVI->disableDT(); + else + LVI->enableDT(); LazyValueInfo::Tristate Ret = LVI->getPredicateAt(CondCmp->getPredicate(), CondCmp->getOperand(0), CondConst, CondBr); @@ -1899,6 +1928,10 @@ bool JumpThreadingPass::ThreadEdge(BasicBlock *BB, << ", across block:\n " << *BB << "\n"); + if (DDT->pending()) + LVI->disableDT(); + else + LVI->enableDT(); LVI->threadEdge(PredBB, BB, SuccBB); // We are going to have to map operands from the original BB block to the new @@ -2368,6 +2401,10 @@ bool JumpThreadingPass::TryToUnfoldSelect(CmpInst *CondCmp, BasicBlock *BB) { // Now check if one of the select values would allow us to constant fold the // terminator in BB. We don't do the transform if both sides fold, those // cases will be threaded in any case. + if (DDT->pending()) + LVI->disableDT(); + else + LVI->enableDT(); LazyValueInfo::Tristate LHSFolds = LVI->getPredicateOnEdge(CondCmp->getPredicate(), SI->getOperand(1), CondRHS, Pred, BB, CondCmp); diff --git a/llvm/test/Transforms/JumpThreading/pr36133.ll b/llvm/test/Transforms/JumpThreading/pr36133.ll new file mode 100644 index 00000000000..b8d8c5fac46 --- /dev/null +++ b/llvm/test/Transforms/JumpThreading/pr36133.ll @@ -0,0 +1,44 @@ +; RUN: opt -jump-threading -S < %s | FileCheck %s +@global = external global i8*, align 8 + +define i32 @foo(i32 %arg) { +; CHECK-LABEL: @foo +; CHECK-LABEL: bb: +; CHECK: icmp eq +; CHECK-NEXT: br i1 %tmp1, label %bb7, label %bb7 +bb: + %tmp = load i8*, i8** @global, align 8 + %tmp1 = icmp eq i8* %tmp, null + br i1 %tmp1, label %bb3, label %bb2 + +; CHECK-NOT: bb2: +bb2: + br label %bb3 + +; CHECK-NOT: bb3: +bb3: + %tmp4 = phi i8 [ 1, %bb2 ], [ 0, %bb ] + %tmp5 = icmp eq i8 %tmp4, 0 + br i1 %tmp5, label %bb7, label %bb6 + +; CHECK-NOT: bb6: +bb6: + br label %bb7 + +; CHECK-LABEL: bb7: +bb7: + %tmp8 = icmp eq i32 %arg, -1 + br i1 %tmp8, label %bb9, label %bb10 + +; CHECK-LABEL: bb9: +bb9: + ret i32 0 + +; CHECK-LABEL: bb10: +bb10: + %tmp11 = icmp sgt i32 %arg, -1 + call void @llvm.assume(i1 %tmp11) + ret i32 1 +} + +declare void @llvm.assume(i1) -- 2.18.0