From 4eed756153b84c211114a3e9186bf0cb55d4b394 Mon Sep 17 00:00:00 2001 From: Benjamin Kramer Date: Mon, 17 Jun 2013 19:00:36 +0000 Subject: [PATCH] Switch spill weights from a basic loop depth estimation to BlockFrequencyInfo. The main advantages here are way better heuristics, taking into account not just loop depth but also __builtin_expect and other static heuristics and will eventually learn how to use profile info. Most of the work in this patch is pushing the MachineBlockFrequencyInfo analysis into the right places. This is good for a 5% speedup on zlib's deflate (x86_64), there were some very unfortunate spilling decisions in its hottest loop in longest_match(). Other benchmarks I tried were mostly neutral. This changes register allocation in subtle ways, update the tests for it. 2012-02-20-MachineCPBug.ll was deleted as it's very fragile and the instruction it looked for was gone already (but the FileCheck pattern picked up unrelated stuff). git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@184105 91177308-0d34-0410-b5e6-96231b3b80d8 --- include/llvm/CodeGen/CalcSpillWeights.h | 7 +- include/llvm/CodeGen/LiveIntervalAnalysis.h | 3 +- include/llvm/CodeGen/LiveRangeEdit.h | 4 +- include/llvm/CodeGen/RegAllocPBQP.h | 6 +- lib/CodeGen/CalcSpillWeights.cpp | 13 ++-- lib/CodeGen/InlineSpiller.cpp | 7 +- lib/CodeGen/LiveIntervalAnalysis.cpp | 19 +---- lib/CodeGen/LiveRangeEdit.cpp | 8 +- lib/CodeGen/RegAllocBasic.cpp | 3 + lib/CodeGen/RegAllocGreedy.cpp | 7 +- lib/CodeGen/RegAllocPBQP.cpp | 21 +++-- lib/CodeGen/SpillPlacement.cpp | 8 +- lib/CodeGen/SplitKit.cpp | 6 +- lib/CodeGen/SplitKit.h | 4 +- lib/CodeGen/StackSlotColoring.cpp | 14 ++-- test/CodeGen/ARM/lsr-unfolded-offset.ll | 3 +- test/CodeGen/SPARC/2011-01-19-DelaySlot.ll | 2 +- test/CodeGen/X86/2012-02-20-MachineCPBug.ll | 78 ------------------- .../X86/atom-bypass-slow-division-64.ll | 15 ++-- 19 files changed, 84 insertions(+), 144 deletions(-) delete mode 100644 test/CodeGen/X86/2012-02-20-MachineCPBug.ll diff --git a/include/llvm/CodeGen/CalcSpillWeights.h b/include/llvm/CodeGen/CalcSpillWeights.h index 9cd2decfacf..c8ec76410ce 100644 --- a/include/llvm/CodeGen/CalcSpillWeights.h +++ b/include/llvm/CodeGen/CalcSpillWeights.h @@ -18,6 +18,7 @@ namespace llvm { class LiveInterval; class LiveIntervals; + class MachineBlockFrequencyInfo; class MachineLoopInfo; /// normalizeSpillWeight - The spill weight of a live interval is computed as: @@ -43,11 +44,13 @@ namespace llvm { MachineFunction &MF; LiveIntervals &LIS; const MachineLoopInfo &Loops; + const MachineBlockFrequencyInfo &MBFI; DenseMap Hint; public: VirtRegAuxInfo(MachineFunction &mf, LiveIntervals &lis, - const MachineLoopInfo &loops) : - MF(mf), LIS(lis), Loops(loops) {} + const MachineLoopInfo &loops, + const MachineBlockFrequencyInfo &mbfi) + : MF(mf), LIS(lis), Loops(loops), MBFI(mbfi) {} /// CalculateWeightAndHint - (re)compute li's spill weight and allocation /// hint. diff --git a/include/llvm/CodeGen/LiveIntervalAnalysis.h b/include/llvm/CodeGen/LiveIntervalAnalysis.h index 7d72f37255b..ffb07a5487d 100644 --- a/include/llvm/CodeGen/LiveIntervalAnalysis.h +++ b/include/llvm/CodeGen/LiveIntervalAnalysis.h @@ -35,6 +35,7 @@ namespace llvm { class AliasAnalysis; class BitVector; + class BlockFrequency; class LiveRangeCalc; class LiveVariables; class MachineDominatorTree; @@ -99,7 +100,7 @@ namespace llvm { virtual ~LiveIntervals(); // Calculate the spill weight to assign to a single instruction. - static float getSpillWeight(bool isDef, bool isUse, unsigned loopDepth); + static float getSpillWeight(bool isDef, bool isUse, BlockFrequency freq); LiveInterval &getInterval(unsigned Reg) { LiveInterval *LI = VirtRegIntervals[Reg]; diff --git a/include/llvm/CodeGen/LiveRangeEdit.h b/include/llvm/CodeGen/LiveRangeEdit.h index e59276fd6d9..545bd0c951a 100644 --- a/include/llvm/CodeGen/LiveRangeEdit.h +++ b/include/llvm/CodeGen/LiveRangeEdit.h @@ -27,6 +27,7 @@ namespace llvm { class AliasAnalysis; class LiveIntervals; +class MachineBlockFrequencyInfo; class MachineLoopInfo; class MachineRegisterInfo; class VirtRegMap; @@ -201,7 +202,8 @@ public: /// calculateRegClassAndHint - Recompute register class and hint for each new /// register. void calculateRegClassAndHint(MachineFunction&, - const MachineLoopInfo&); + const MachineLoopInfo&, + const MachineBlockFrequencyInfo&); }; } diff --git a/include/llvm/CodeGen/RegAllocPBQP.h b/include/llvm/CodeGen/RegAllocPBQP.h index 8b8e3d90f73..6f2d1394533 100644 --- a/include/llvm/CodeGen/RegAllocPBQP.h +++ b/include/llvm/CodeGen/RegAllocPBQP.h @@ -26,8 +26,8 @@ namespace llvm { class LiveIntervals; + class MachineBlockFrequencyInfo; class MachineFunction; - class MachineLoopInfo; class TargetRegisterInfo; template class OwningPtr; @@ -125,7 +125,7 @@ namespace llvm { /// Build a PBQP instance to represent the register allocation problem for /// the given MachineFunction. virtual PBQPRAProblem *build(MachineFunction *mf, const LiveIntervals *lis, - const MachineLoopInfo *loopInfo, + const MachineBlockFrequencyInfo *mbfi, const RegSet &vregs); private: @@ -144,7 +144,7 @@ namespace llvm { /// Build a PBQP instance to represent the register allocation problem for /// the given MachineFunction. virtual PBQPRAProblem *build(MachineFunction *mf, const LiveIntervals *lis, - const MachineLoopInfo *loopInfo, + const MachineBlockFrequencyInfo *mbfi, const RegSet &vregs); private: diff --git a/lib/CodeGen/CalcSpillWeights.cpp b/lib/CodeGen/CalcSpillWeights.cpp index 38ae17d2315..b03c325c895 100644 --- a/lib/CodeGen/CalcSpillWeights.cpp +++ b/lib/CodeGen/CalcSpillWeights.cpp @@ -12,6 +12,7 @@ #include "llvm/ADT/SmallSet.h" #include "llvm/CodeGen/CalcSpillWeights.h" #include "llvm/CodeGen/LiveIntervalAnalysis.h" +#include "llvm/CodeGen/MachineBlockFrequencyInfo.h" #include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineLoopInfo.h" #include "llvm/CodeGen/MachineRegisterInfo.h" @@ -33,6 +34,7 @@ INITIALIZE_PASS_END(CalculateSpillWeights, "calcspillweights", void CalculateSpillWeights::getAnalysisUsage(AnalysisUsage &au) const { au.addRequired(); + au.addRequired(); au.addRequired(); au.setPreservesAll(); MachineFunctionPass::getAnalysisUsage(au); @@ -45,7 +47,8 @@ bool CalculateSpillWeights::runOnMachineFunction(MachineFunction &MF) { LiveIntervals &LIS = getAnalysis(); MachineRegisterInfo &MRI = MF.getRegInfo(); - VirtRegAuxInfo VRAI(MF, LIS, getAnalysis()); + VirtRegAuxInfo VRAI(MF, LIS, getAnalysis(), + getAnalysis()); for (unsigned i = 0, e = MRI.getNumVirtRegs(); i != e; ++i) { unsigned Reg = TargetRegisterInfo::index2VirtReg(i); if (MRI.reg_nodbg_empty(Reg)) @@ -107,12 +110,12 @@ static bool isRematerializable(const LiveInterval &LI, return true; } -void VirtRegAuxInfo::CalculateWeightAndHint(LiveInterval &li) { +void +VirtRegAuxInfo::CalculateWeightAndHint(LiveInterval &li) { MachineRegisterInfo &mri = MF.getRegInfo(); const TargetRegisterInfo &tri = *MF.getTarget().getRegisterInfo(); MachineBasicBlock *mbb = 0; MachineLoop *loop = 0; - unsigned loopDepth = 0; bool isExiting = false; float totalWeight = 0; SmallPtrSet visited; @@ -140,14 +143,14 @@ void VirtRegAuxInfo::CalculateWeightAndHint(LiveInterval &li) { if (mi->getParent() != mbb) { mbb = mi->getParent(); loop = Loops.getLoopFor(mbb); - loopDepth = loop ? loop->getLoopDepth() : 0; isExiting = loop ? loop->isLoopExiting(mbb) : false; } // Calculate instr weight. bool reads, writes; tie(reads, writes) = mi->readsWritesVirtualRegister(li.reg); - weight = LiveIntervals::getSpillWeight(writes, reads, loopDepth); + weight = LiveIntervals::getSpillWeight( + writes, reads, MBFI.getBlockFreq(mi->getParent())); // Give extra weight to what looks like a loop induction variable update. if (writes && isExiting && LIS.isLiveOutOfMBB(li, mbb)) diff --git a/lib/CodeGen/InlineSpiller.cpp b/lib/CodeGen/InlineSpiller.cpp index dd00d77c54c..db63b526383 100644 --- a/lib/CodeGen/InlineSpiller.cpp +++ b/lib/CodeGen/InlineSpiller.cpp @@ -22,6 +22,7 @@ #include "llvm/CodeGen/LiveRangeEdit.h" #include "llvm/CodeGen/LiveStackAnalysis.h" #include "llvm/CodeGen/MachineDominators.h" +#include "llvm/CodeGen/MachineBlockFrequencyInfo.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineInstrBuilder.h" @@ -65,6 +66,7 @@ class InlineSpiller : public Spiller { MachineRegisterInfo &MRI; const TargetInstrInfo &TII; const TargetRegisterInfo &TRI; + const MachineBlockFrequencyInfo &MBFI; // Variables that are valid during spill(), but used by multiple methods. LiveRangeEdit *Edit; @@ -148,7 +150,8 @@ public: MFI(*mf.getFrameInfo()), MRI(mf.getRegInfo()), TII(*mf.getTarget().getInstrInfo()), - TRI(*mf.getTarget().getRegisterInfo()) {} + TRI(*mf.getTarget().getRegisterInfo()), + MBFI(pass.getAnalysis()) {} void spill(LiveRangeEdit &); @@ -1290,5 +1293,5 @@ void InlineSpiller::spill(LiveRangeEdit &edit) { if (!RegsToSpill.empty()) spillAll(); - Edit->calculateRegClassAndHint(MF, Loops); + Edit->calculateRegClassAndHint(MF, Loops, MBFI); } diff --git a/lib/CodeGen/LiveIntervalAnalysis.cpp b/lib/CodeGen/LiveIntervalAnalysis.cpp index 1ca2d46cc29..18eac4c7f96 100644 --- a/lib/CodeGen/LiveIntervalAnalysis.cpp +++ b/lib/CodeGen/LiveIntervalAnalysis.cpp @@ -28,6 +28,7 @@ #include "llvm/CodeGen/Passes.h" #include "llvm/CodeGen/VirtRegMap.h" #include "llvm/IR/Value.h" +#include "llvm/Support/BlockFrequency.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" @@ -605,21 +606,9 @@ LiveIntervals::hasPHIKill(const LiveInterval &LI, const VNInfo *VNI) const { } float -LiveIntervals::getSpillWeight(bool isDef, bool isUse, unsigned loopDepth) { - // Limit the loop depth ridiculousness. - if (loopDepth > 200) - loopDepth = 200; - - // The loop depth is used to roughly estimate the number of times the - // instruction is executed. Something like 10^d is simple, but will quickly - // overflow a float. This expression behaves like 10^d for small d, but is - // more tempered for large d. At d=200 we get 6.7e33 which leaves a bit of - // headroom before overflow. - // By the way, powf() might be unavailable here. For consistency, - // We may take pow(double,double). - float lc = std::pow(1 + (100.0 / (loopDepth + 10)), (double)loopDepth); - - return (isDef + isUse) * lc; +LiveIntervals::getSpillWeight(bool isDef, bool isUse, BlockFrequency freq) { + const float Scale = 1.0f / BlockFrequency::getEntryFrequency(); + return (isDef + isUse) * (freq.getFrequency() * Scale); } LiveRange LiveIntervals::addLiveRangeToEndOfBlock(unsigned reg, diff --git a/lib/CodeGen/LiveRangeEdit.cpp b/lib/CodeGen/LiveRangeEdit.cpp index 7793e96c354..b1de6a09fe3 100644 --- a/lib/CodeGen/LiveRangeEdit.cpp +++ b/lib/CodeGen/LiveRangeEdit.cpp @@ -374,9 +374,11 @@ void LiveRangeEdit::eliminateDeadDefs(SmallVectorImpl &Dead, } } -void LiveRangeEdit::calculateRegClassAndHint(MachineFunction &MF, - const MachineLoopInfo &Loops) { - VirtRegAuxInfo VRAI(MF, LIS, Loops); +void +LiveRangeEdit::calculateRegClassAndHint(MachineFunction &MF, + const MachineLoopInfo &Loops, + const MachineBlockFrequencyInfo &MBFI) { + VirtRegAuxInfo VRAI(MF, LIS, Loops, MBFI); for (iterator I = begin(), E = end(); I != E; ++I) { LiveInterval &LI = **I; if (MRI.recomputeRegClass(LI.reg, MF.getTarget())) diff --git a/lib/CodeGen/RegAllocBasic.cpp b/lib/CodeGen/RegAllocBasic.cpp index 7fcfe9e88be..d6a7d6f4287 100644 --- a/lib/CodeGen/RegAllocBasic.cpp +++ b/lib/CodeGen/RegAllocBasic.cpp @@ -24,6 +24,7 @@ #include "llvm/CodeGen/LiveRangeEdit.h" #include "llvm/CodeGen/LiveRegMatrix.h" #include "llvm/CodeGen/LiveStackAnalysis.h" +#include "llvm/CodeGen/MachineBlockFrequencyInfo.h" #include "llvm/CodeGen/MachineFunctionPass.h" #include "llvm/CodeGen/MachineInstr.h" #include "llvm/CodeGen/MachineLoopInfo.h" @@ -145,6 +146,8 @@ void RABasic::getAnalysisUsage(AnalysisUsage &AU) const { AU.addRequired(); AU.addRequired(); AU.addPreserved(); + AU.addRequired(); + AU.addPreserved(); AU.addRequiredID(MachineDominatorsID); AU.addPreservedID(MachineDominatorsID); AU.addRequired(); diff --git a/lib/CodeGen/RegAllocGreedy.cpp b/lib/CodeGen/RegAllocGreedy.cpp index 49748289dac..bdcef6ff122 100644 --- a/lib/CodeGen/RegAllocGreedy.cpp +++ b/lib/CodeGen/RegAllocGreedy.cpp @@ -29,6 +29,7 @@ #include "llvm/CodeGen/LiveRangeEdit.h" #include "llvm/CodeGen/LiveRegMatrix.h" #include "llvm/CodeGen/LiveStackAnalysis.h" +#include "llvm/CodeGen/MachineBlockFrequencyInfo.h" #include "llvm/CodeGen/MachineDominators.h" #include "llvm/CodeGen/MachineFunctionPass.h" #include "llvm/CodeGen/MachineLoopInfo.h" @@ -71,6 +72,7 @@ class RAGreedy : public MachineFunctionPass, // analyses SlotIndexes *Indexes; + MachineBlockFrequencyInfo *MBFI; MachineDominatorTree *DomTree; MachineLoopInfo *Loops; EdgeBundles *Bundles; @@ -320,6 +322,8 @@ RAGreedy::RAGreedy(): MachineFunctionPass(ID) { void RAGreedy::getAnalysisUsage(AnalysisUsage &AU) const { AU.setPreservesCFG(); + AU.addRequired(); + AU.addPreserved(); AU.addRequired(); AU.addPreserved(); AU.addRequired(); @@ -1770,6 +1774,7 @@ bool RAGreedy::runOnMachineFunction(MachineFunction &mf) { getAnalysis(), getAnalysis()); Indexes = &getAnalysis(); + MBFI = &getAnalysis(); DomTree = &getAnalysis(); SpillerInstance.reset(createInlineSpiller(*this, *MF, *VRM)); Loops = &getAnalysis(); @@ -1778,7 +1783,7 @@ bool RAGreedy::runOnMachineFunction(MachineFunction &mf) { DebugVars = &getAnalysis(); SA.reset(new SplitAnalysis(*VRM, *LIS, *Loops)); - SE.reset(new SplitEditor(*SA, *LIS, *VRM, *DomTree)); + SE.reset(new SplitEditor(*SA, *LIS, *VRM, *DomTree, *MBFI)); ExtraRegInfo.clear(); ExtraRegInfo.resize(MRI->getNumVirtRegs()); NextCascade = 1; diff --git a/lib/CodeGen/RegAllocPBQP.cpp b/lib/CodeGen/RegAllocPBQP.cpp index 15a88e224fa..ecfe6ade9b5 100644 --- a/lib/CodeGen/RegAllocPBQP.cpp +++ b/lib/CodeGen/RegAllocPBQP.cpp @@ -40,9 +40,9 @@ #include "llvm/CodeGen/LiveIntervalAnalysis.h" #include "llvm/CodeGen/LiveRangeEdit.h" #include "llvm/CodeGen/LiveStackAnalysis.h" +#include "llvm/CodeGen/MachineBlockFrequencyInfo.h" #include "llvm/CodeGen/MachineDominators.h" #include "llvm/CodeGen/MachineFunctionPass.h" -#include "llvm/CodeGen/MachineLoopInfo.h" #include "llvm/CodeGen/MachineRegisterInfo.h" #include "llvm/CodeGen/PBQP/Graph.h" #include "llvm/CodeGen/PBQP/HeuristicSolver.h" @@ -96,7 +96,6 @@ public: initializeLiveIntervalsPass(*PassRegistry::getPassRegistry()); initializeCalculateSpillWeightsPass(*PassRegistry::getPassRegistry()); initializeLiveStacksPass(*PassRegistry::getPassRegistry()); - initializeMachineLoopInfoPass(*PassRegistry::getPassRegistry()); initializeVirtRegMapPass(*PassRegistry::getPassRegistry()); } @@ -130,8 +129,8 @@ private: const TargetMachine *tm; const TargetRegisterInfo *tri; const TargetInstrInfo *tii; - const MachineLoopInfo *loopInfo; MachineRegisterInfo *mri; + const MachineBlockFrequencyInfo *mbfi; OwningPtr spiller; LiveIntervals *lis; @@ -188,7 +187,7 @@ unsigned PBQPRAProblem::getPRegForOption(unsigned vreg, unsigned option) const { } PBQPRAProblem *PBQPBuilder::build(MachineFunction *mf, const LiveIntervals *lis, - const MachineLoopInfo *loopInfo, + const MachineBlockFrequencyInfo *mbfi, const RegSet &vregs) { LiveIntervals *LIS = const_cast(lis); @@ -313,10 +312,10 @@ void PBQPBuilder::addInterferenceCosts( PBQPRAProblem *PBQPBuilderWithCoalescing::build(MachineFunction *mf, const LiveIntervals *lis, - const MachineLoopInfo *loopInfo, + const MachineBlockFrequencyInfo *mbfi, const RegSet &vregs) { - OwningPtr p(PBQPBuilder::build(mf, lis, loopInfo, vregs)); + OwningPtr p(PBQPBuilder::build(mf, lis, mbfi, vregs)); PBQP::Graph &g = p->getGraph(); const TargetMachine &tm = mf->getTarget(); @@ -350,7 +349,7 @@ PBQPRAProblem *PBQPBuilderWithCoalescing::build(MachineFunction *mf, PBQP::PBQPNum cBenefit = copyFactor * LiveIntervals::getSpillWeight(false, true, - loopInfo->getLoopDepth(mbb)); + mbfi->getBlockFreq(mbb)); if (cp.isPhys()) { if (!mf->getRegInfo().isAllocatable(dst)) { @@ -435,10 +434,10 @@ void RegAllocPBQP::getAnalysisUsage(AnalysisUsage &au) const { au.addRequired(); au.addRequired(); au.addPreserved(); + au.addRequired(); + au.addPreserved(); au.addRequired(); au.addPreserved(); - au.addRequired(); - au.addPreserved(); au.addRequired(); au.addPreserved(); MachineFunctionPass::getAnalysisUsage(au); @@ -546,7 +545,7 @@ bool RegAllocPBQP::runOnMachineFunction(MachineFunction &MF) { lis = &getAnalysis(); lss = &getAnalysis(); - loopInfo = &getAnalysis(); + mbfi = &getAnalysis(); vrm = &getAnalysis(); spiller.reset(createInlineSpiller(*this, MF, *vrm)); @@ -584,7 +583,7 @@ bool RegAllocPBQP::runOnMachineFunction(MachineFunction &MF) { DEBUG(dbgs() << " PBQP Regalloc round " << round << ":\n"); OwningPtr problem( - builder->build(mf, lis, loopInfo, vregsToAlloc)); + builder->build(mf, lis, mbfi, vregsToAlloc)); #ifndef NDEBUG if (pbqpDumpGraphs) { diff --git a/lib/CodeGen/SpillPlacement.cpp b/lib/CodeGen/SpillPlacement.cpp index c5bbba3ffcc..840f05b9ff4 100644 --- a/lib/CodeGen/SpillPlacement.cpp +++ b/lib/CodeGen/SpillPlacement.cpp @@ -31,8 +31,8 @@ #include "SpillPlacement.h" #include "llvm/ADT/BitVector.h" #include "llvm/CodeGen/EdgeBundles.h" -#include "llvm/CodeGen/LiveIntervalAnalysis.h" #include "llvm/CodeGen/MachineBasicBlock.h" +#include "llvm/CodeGen/MachineBlockFrequencyInfo.h" #include "llvm/CodeGen/MachineFunction.h" #include "llvm/CodeGen/MachineLoopInfo.h" #include "llvm/CodeGen/Passes.h" @@ -53,6 +53,7 @@ char &llvm::SpillPlacementID = SpillPlacement::ID; void SpillPlacement::getAnalysisUsage(AnalysisUsage &AU) const { AU.setPreservesAll(); + AU.addRequired(); AU.addRequiredTransitive(); AU.addRequiredTransitive(); MachineFunctionPass::getAnalysisUsage(AU); @@ -178,9 +179,10 @@ bool SpillPlacement::runOnMachineFunction(MachineFunction &mf) { // Compute total ingoing and outgoing block frequencies for all bundles. BlockFrequency.resize(mf.getNumBlockIDs()); + MachineBlockFrequencyInfo &MBFI = getAnalysis(); + float EntryFreq = BlockFrequency::getEntryFrequency(); for (MachineFunction::iterator I = mf.begin(), E = mf.end(); I != E; ++I) { - float Freq = LiveIntervals::getSpillWeight(true, false, - loops->getLoopDepth(I)); + float Freq = MBFI.getBlockFreq(I).getFrequency() / EntryFreq; unsigned Num = I->getNumber(); BlockFrequency[Num] = Freq; nodes[bundles->getBundle(Num, 1)].Scale[0] += Freq; diff --git a/lib/CodeGen/SplitKit.cpp b/lib/CodeGen/SplitKit.cpp index 0a3818e43ff..e717fac2998 100644 --- a/lib/CodeGen/SplitKit.cpp +++ b/lib/CodeGen/SplitKit.cpp @@ -325,12 +325,14 @@ void SplitAnalysis::analyze(const LiveInterval *li) { SplitEditor::SplitEditor(SplitAnalysis &sa, LiveIntervals &lis, VirtRegMap &vrm, - MachineDominatorTree &mdt) + MachineDominatorTree &mdt, + MachineBlockFrequencyInfo &mbfi) : SA(sa), LIS(lis), VRM(vrm), MRI(vrm.getMachineFunction().getRegInfo()), MDT(mdt), TII(*vrm.getMachineFunction().getTarget().getInstrInfo()), TRI(*vrm.getMachineFunction().getTarget().getRegisterInfo()), + MBFI(mbfi), Edit(0), OpenIdx(0), SpillMode(SM_Partition), @@ -1119,7 +1121,7 @@ void SplitEditor::finish(SmallVectorImpl *LRMap) { } // Calculate spill weight and allocation hints for new intervals. - Edit->calculateRegClassAndHint(VRM.getMachineFunction(), SA.Loops); + Edit->calculateRegClassAndHint(VRM.getMachineFunction(), SA.Loops, MBFI); assert(!LRMap || LRMap->size() == Edit->size()); } diff --git a/lib/CodeGen/SplitKit.h b/lib/CodeGen/SplitKit.h index 4005a3d5cbb..f029c73d122 100644 --- a/lib/CodeGen/SplitKit.h +++ b/lib/CodeGen/SplitKit.h @@ -27,6 +27,7 @@ class ConnectedVNInfoEqClasses; class LiveInterval; class LiveIntervals; class LiveRangeEdit; +class MachineBlockFrequencyInfo; class MachineInstr; class MachineLoopInfo; class MachineRegisterInfo; @@ -215,6 +216,7 @@ class SplitEditor { MachineDominatorTree &MDT; const TargetInstrInfo &TII; const TargetRegisterInfo &TRI; + const MachineBlockFrequencyInfo &MBFI; public: @@ -349,7 +351,7 @@ public: /// Create a new SplitEditor for editing the LiveInterval analyzed by SA. /// Newly created intervals will be appended to newIntervals. SplitEditor(SplitAnalysis &SA, LiveIntervals&, VirtRegMap&, - MachineDominatorTree&); + MachineDominatorTree&, MachineBlockFrequencyInfo &); /// reset - Prepare for a new split. void reset(LiveRangeEdit&, ComplementSpillMode = SM_Partition); diff --git a/lib/CodeGen/StackSlotColoring.cpp b/lib/CodeGen/StackSlotColoring.cpp index f9515610d7e..3c35b8374d8 100644 --- a/lib/CodeGen/StackSlotColoring.cpp +++ b/lib/CodeGen/StackSlotColoring.cpp @@ -19,9 +19,9 @@ #include "llvm/ADT/Statistic.h" #include "llvm/CodeGen/LiveIntervalAnalysis.h" #include "llvm/CodeGen/LiveStackAnalysis.h" +#include "llvm/CodeGen/MachineBlockFrequencyInfo.h" #include "llvm/CodeGen/MachineFrameInfo.h" #include "llvm/CodeGen/MachineInstrBuilder.h" -#include "llvm/CodeGen/MachineLoopInfo.h" #include "llvm/CodeGen/MachineMemOperand.h" #include "llvm/CodeGen/MachineRegisterInfo.h" #include "llvm/CodeGen/PseudoSourceValue.h" @@ -48,7 +48,7 @@ namespace { LiveStacks* LS; MachineFrameInfo *MFI; const TargetInstrInfo *TII; - const MachineLoopInfo *loopInfo; + const MachineBlockFrequencyInfo *MBFI; // SSIntervals - Spill slot intervals. std::vector SSIntervals; @@ -89,8 +89,8 @@ namespace { AU.addRequired(); AU.addPreserved(); AU.addRequired(); - AU.addRequired(); - AU.addPreserved(); + AU.addRequired(); + AU.addPreserved(); AU.addPreservedID(MachineDominatorsID); MachineFunctionPass::getAnalysisUsage(AU); } @@ -139,7 +139,7 @@ void StackSlotColoring::ScanForSpillSlotRefs(MachineFunction &MF) { for (MachineFunction::iterator MBBI = MF.begin(), E = MF.end(); MBBI != E; ++MBBI) { MachineBasicBlock *MBB = &*MBBI; - unsigned loopDepth = loopInfo->getLoopDepth(MBB); + BlockFrequency Freq = MBFI->getBlockFreq(MBB); for (MachineBasicBlock::iterator MII = MBB->begin(), EE = MBB->end(); MII != EE; ++MII) { MachineInstr *MI = &*MII; @@ -154,7 +154,7 @@ void StackSlotColoring::ScanForSpillSlotRefs(MachineFunction &MF) { continue; LiveInterval &li = LS->getInterval(FI); if (!MI->isDebugValue()) - li.weight += LiveIntervals::getSpillWeight(false, true, loopDepth); + li.weight += LiveIntervals::getSpillWeight(false, true, Freq); SSRefs[FI].push_back(MI); } } @@ -396,7 +396,7 @@ bool StackSlotColoring::runOnMachineFunction(MachineFunction &MF) { MFI = MF.getFrameInfo(); TII = MF.getTarget().getInstrInfo(); LS = &getAnalysis(); - loopInfo = &getAnalysis(); + MBFI = &getAnalysis(); bool Changed = false; diff --git a/test/CodeGen/ARM/lsr-unfolded-offset.ll b/test/CodeGen/ARM/lsr-unfolded-offset.ll index 9b0f3e54e88..26d4be2e06f 100644 --- a/test/CodeGen/ARM/lsr-unfolded-offset.ll +++ b/test/CodeGen/ARM/lsr-unfolded-offset.ll @@ -7,8 +7,7 @@ ; CHECK: sub sp, #{{40|32|28|24}} ; CHECK: %for.inc -; CHECK: ldr{{(.w)?}} r{{.*}}, [sp, # -; CHECK: ldr{{(.w)?}} r{{.*}}, [sp, # +; CHECK-NOT: ldr ; CHECK: add target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:32-f32:32:32-f64:32:32-v64:32:64-v128:32:128-a0:0:32-n32" diff --git a/test/CodeGen/SPARC/2011-01-19-DelaySlot.ll b/test/CodeGen/SPARC/2011-01-19-DelaySlot.ll index b39c355fbdc..4a5952736ef 100644 --- a/test/CodeGen/SPARC/2011-01-19-DelaySlot.ll +++ b/test/CodeGen/SPARC/2011-01-19-DelaySlot.ll @@ -59,7 +59,7 @@ entry: ;CHECK: !NO_APP ;CHECK-NEXT: cmp ;CHECK-NEXT: bg -;CHECK-NEXT: nop +;CHECK-NEXT: or tail call void asm sideeffect "sethi 0, %g0", ""() nounwind %0 = icmp slt i32 %a, 0 br i1 %0, label %bb, label %bb1 diff --git a/test/CodeGen/X86/2012-02-20-MachineCPBug.ll b/test/CodeGen/X86/2012-02-20-MachineCPBug.ll deleted file mode 100644 index 477b4deba82..00000000000 --- a/test/CodeGen/X86/2012-02-20-MachineCPBug.ll +++ /dev/null @@ -1,78 +0,0 @@ -; RUN: llc < %s -mtriple=i386-apple-macosx -mcpu=core2 -mattr=+sse | FileCheck %s -; PR11940: Do not optimize away movb %al, %ch - -%struct.APInt = type { i64* } - -declare noalias i8* @calloc(i32, i32) nounwind - -define void @bug(%struct.APInt* noalias nocapture sret %agg.result, %struct.APInt* nocapture %this, i32 %rotateAmt) nounwind align 2 { -entry: -; CHECK: bug: - %call = tail call i8* @calloc(i32 1, i32 32) - %call.i = tail call i8* @calloc(i32 1, i32 32) nounwind - %0 = bitcast i8* %call.i to i64* - %rem.i = and i32 %rotateAmt, 63 - %div.i = lshr i32 %rotateAmt, 6 - %cmp.i = icmp eq i32 %rem.i, 0 - br i1 %cmp.i, label %for.cond.preheader.i, label %if.end.i - -for.cond.preheader.i: ; preds = %entry - %sub.i = sub i32 4, %div.i - %cmp23.i = icmp eq i32 %div.i, 4 - br i1 %cmp23.i, label %for.body9.lr.ph.i, label %for.body.lr.ph.i - -for.body.lr.ph.i: ; preds = %for.cond.preheader.i - %pVal.i = getelementptr inbounds %struct.APInt* %this, i32 0, i32 0 - %.pre5.i = load i64** %pVal.i, align 4 - br label %for.body.i - -for.body.i: ; preds = %for.body.i, %for.body.lr.ph.i - %i.04.i = phi i32 [ 0, %for.body.lr.ph.i ], [ %inc.i, %for.body.i ] - %add.i = add i32 %i.04.i, %div.i - %arrayidx.i = getelementptr inbounds i64* %.pre5.i, i32 %add.i - %1 = load i64* %arrayidx.i, align 4 - %arrayidx3.i = getelementptr inbounds i64* %0, i32 %i.04.i - store i64 %1, i64* %arrayidx3.i, align 4 - %inc.i = add i32 %i.04.i, 1 - %cmp2.i = icmp ult i32 %inc.i, %sub.i - br i1 %cmp2.i, label %for.body.i, label %if.end.i - -if.end.i: ; preds = %for.body.i, %entry - %cmp81.i = icmp eq i32 %div.i, 3 - br i1 %cmp81.i, label %_ZNK5APInt4lshrEj.exit, label %for.body9.lr.ph.i - -for.body9.lr.ph.i: ; preds = %if.end.i, %for.cond.preheader.i - %sub58.i = sub i32 3, %div.i - %pVal11.i = getelementptr inbounds %struct.APInt* %this, i32 0, i32 0 - %sh_prom.i = zext i32 %rem.i to i64 - %sub17.i = sub i32 64, %rem.i - %sh_prom18.i = zext i32 %sub17.i to i64 - %.pre.i = load i64** %pVal11.i, align 4 - br label %for.body9.i - -for.body9.i: ; preds = %for.body9.i, %for.body9.lr.ph.i -; CHECK: %for.body9.i -; CHECK: movb -; CHECK: shrdl - %i6.02.i = phi i32 [ 0, %for.body9.lr.ph.i ], [ %inc21.i, %for.body9.i ] - %add10.i = add i32 %i6.02.i, %div.i - %arrayidx12.i = getelementptr inbounds i64* %.pre.i, i32 %add10.i - %2 = load i64* %arrayidx12.i, align 4 - %shr.i = lshr i64 %2, %sh_prom.i - %add14.i = add i32 %add10.i, 1 - %arrayidx16.i = getelementptr inbounds i64* %.pre.i, i32 %add14.i - %3 = load i64* %arrayidx16.i, align 4 - %shl.i = shl i64 %3, %sh_prom18.i - %or.i = or i64 %shl.i, %shr.i - %arrayidx19.i = getelementptr inbounds i64* %0, i32 %i6.02.i - store i64 %or.i, i64* %arrayidx19.i, align 4 - %inc21.i = add i32 %i6.02.i, 1 - %cmp8.i = icmp ult i32 %inc21.i, %sub58.i - br i1 %cmp8.i, label %for.body9.i, label %_ZNK5APInt4lshrEj.exit - -_ZNK5APInt4lshrEj.exit: ; preds = %for.body9.i, %if.end.i - %call.i1 = tail call i8* @calloc(i32 1, i32 32) nounwind - %4 = getelementptr inbounds %struct.APInt* %agg.result, i32 0, i32 0 - store i64* %0, i64** %4, align 4 - ret void -} diff --git a/test/CodeGen/X86/atom-bypass-slow-division-64.ll b/test/CodeGen/X86/atom-bypass-slow-division-64.ll index a3bbea3c996..26b9a1e60f1 100644 --- a/test/CodeGen/X86/atom-bypass-slow-division-64.ll +++ b/test/CodeGen/X86/atom-bypass-slow-division-64.ll @@ -4,8 +4,9 @@ define i64 @Test_get_quotient(i64 %a, i64 %b) nounwind { ; CHECK: Test_get_quotient: -; CHECK: orq %rsi, %rcx -; CHECK-NEXT: testq $-65536, %rcx +; CHECK: movq %rdi, %rax +; CHECK: orq %rsi, %rax +; CHECK-NEXT: testq $-65536, %rax ; CHECK-NEXT: je ; CHECK: idivq ; CHECK: ret @@ -17,8 +18,9 @@ define i64 @Test_get_quotient(i64 %a, i64 %b) nounwind { define i64 @Test_get_remainder(i64 %a, i64 %b) nounwind { ; CHECK: Test_get_remainder: -; CHECK: orq %rsi, %rcx -; CHECK-NEXT: testq $-65536, %rcx +; CHECK: movq %rdi, %rax +; CHECK: orq %rsi, %rax +; CHECK-NEXT: testq $-65536, %rax ; CHECK-NEXT: je ; CHECK: idivq ; CHECK: ret @@ -30,8 +32,9 @@ define i64 @Test_get_remainder(i64 %a, i64 %b) nounwind { define i64 @Test_get_quotient_and_remainder(i64 %a, i64 %b) nounwind { ; CHECK: Test_get_quotient_and_remainder: -; CHECK: orq %rsi, %rcx -; CHECK-NEXT: testq $-65536, %rcx +; CHECK: movq %rdi, %rax +; CHECK: orq %rsi, %rax +; CHECK-NEXT: testq $-65536, %rax ; CHECK-NEXT: je ; CHECK: idivq ; CHECK: divw