[PATCH] D12781: PGO IR-level instrumentation infrastructure
Xinliang David Li via llvm-commits
llvm-commits at lists.llvm.org
Tue Dec 1 14:44:23 PST 2015
>
> > +; GEN: call void @llvm.instrprof.increment(i8* getelementptr inbounds
> ([9 x i8], [9 x i8]* @__llvm_profile_name_test_br_1, i32 0, i32 0), i64
> 25571299074, i32 2, i32 0)
> > +
> > + // Set the hot/cold inline hints based on the count values.
> > + void applyFunctionAttributes(uint64_t EntryCount, uint64_t MaxCount) {
> > + if (ProgramMaxCount == 0)
> > + return;
> > + // Threshold of the hot functions.
> > + const BranchProbability HotFunctionThreshold(1, 100);
> > + // Threshold of the cold functions.
> > + const BranchProbability ColdFunctionThreshold(2, 10000);
> > + if (EntryCount >= HotFunctionThreshold.scale(ProgramMaxCount))
> > + F.addFnAttr(llvm::Attribute::InlineHint);
> > + else if (MaxCount <= ColdFunctionThreshold.scale(ProgramMaxCount))
> > + F.addFnAttr(llvm::Attribute::Cold);
>
> This is more or less a workaround for limitations of the inliner and
> should eventually be removed in favour of making the inliner smarter,
> correct?
>
>
yes. Easwaran is working on making the logic migrated to the inliner. Once
the inliner change is ready, this code together with similar/duplicated
code in FE and SampleFDO will be removed.
David
> > + }
> > +
> > +public:
> > + PGOUseFunc(Function &Func, Module *Modu,
> > + BranchProbabilityInfo *BPI_ = nullptr,
> > + BlockFrequencyInfo *BFI_ = nullptr)
> > + : F(Func), M(Modu), FuncInfo(Func, false, BPI_, BFI_) {}
> > +
> > + // Read counts for the instrumented BB from profile.
> > + bool readCounters(IndexedInstrProfReader *PGOReader);
> > +
> > + // Populate the counts for all BBs.
> > + void populateCounters();
> > +
> > + // Set the branch weights based on the count values.
> > + void setBranchWeights();
> > +};
> > +
> > +// Visit all the edges and assign the count value for the instrumented
> > +// edges and the BB.
> > +void PGOUseFunc::setInstrumentedCounts(
> > + const std::vector<uint64_t> &CountFromProfile) {
> > +
> > + // Use a worklist as we will update the vector during the iteration.
> > + std::vector<PGOUseEdge *> WorkList;
> > + for (auto &Ei : FuncInfo.MST.AllEdges)
> > + WorkList.push_back(Ei.get());
> > +
> > + uint32_t j = 0;
> > + for (auto &Ei : WorkList) {
>
> Same as before: j->J, Ei->E.
>
> > + BasicBlock *InstrBB = FuncInfo.getInstrBB(Ei);
> > + if (!InstrBB)
> > + continue;
> > + uint64_t CountValue = CountFromProfile[j++];
> > + if (!Ei->Removed) {
> > + getBBInfo(InstrBB).setBBInfoCount(CountValue);
> > + Ei->setEdgeCount(CountValue);
> > + continue;
> > + }
> > +
> > + // Need to add two new edges.
> > + BasicBlock *SrcBB = const_cast<BasicBlock *>(Ei->SrcBB);
> > + BasicBlock *DestBB = const_cast<BasicBlock *>(Ei->DestBB);
> > + // Add new edge of SrcBB->InstrBB.
> > + PGOUseEdge &NewEdge = FuncInfo.MST.addEdge(SrcBB, InstrBB, 0);
> > + NewEdge.setEdgeCount(CountValue);
> > + // Add new edge of InstrBB->DestBB.
> > + PGOUseEdge &NewEdge1 = FuncInfo.MST.addEdge(InstrBB, DestBB, 0);
>
> Why not just reuse the NewEdge variable here?
>
> > + NewEdge1.setEdgeCount(CountValue);
> > + NewEdge1.InMST = true;
> > + getBBInfo(InstrBB).setBBInfoCount(CountValue);
> > + }
> > +}
> > +
> > +// Set the count value for the unknown edge. There should be one and
> only one
> > +// unknown edge in Edges vector.
> > +void PGOUseFunc::setEdgeCount(DirectEdges &Edges, uint64_t Value) {
> > + for (auto &Ei : Edges) {
> > + if (Ei->CountValid)
> > + continue;
> > + Ei->setEdgeCount(Value);
> > +
> > + getBBInfo(Ei->SrcBB).UnknownCountOutEdge--;
> > + getBBInfo(Ei->DestBB).UnknownCountInEdge--;
> > + return;
> > + }
> > + llvm_unreachable("Cannot find the unknown count edge");
> > +}
> > +
> > +// Read the profile from ProfileFileName and assign the value to the
> > +// instrumented BB and the edges. This function also updates
> ProgramMaxCount.
> > +// Return true if the profile are successfully read, and false on
> errors.
> > +bool PGOUseFunc::readCounters(IndexedInstrProfReader *PGOReader) {
> > + auto &Ctx = M->getContext();
> > + ErrorOr<InstrProfRecord> Result =
> > + PGOReader->getInstrProfRecord(FuncInfo.FuncName,
> FuncInfo.FunctionHash);
> > + if (std::error_code EC = Result.getError()) {
> > + if (EC == instrprof_error::unknown_function)
> > + NumOfPGOMissing++;
> > + else if (EC == instrprof_error::hash_mismatch ||
> > + EC == llvm::instrprof_error::malformed)
> > + NumOfPGOMismatch++;
> > +
> > + std::string Msg = EC.message() + std::string(" ") +
> F.getName().str();
> > + Ctx.diagnose(
> > + DiagnosticInfoPGOProfile(M->getName().data(), Msg, DS_Warning));
> > + return false;
> > + }
> > + std::vector<uint64_t> &CountFromProfile = Result.get().Counts;
> > +
> > + NumOfPGOFunc++;
> > + DEBUG(dbgs() << CountFromProfile.size() << " counts\n");
> > + uint64_t ValueSum = 0;
> > + for (unsigned i = 0, e = CountFromProfile.size(); i < e; i++) {
> > + DEBUG(dbgs() << " " << i << ": " << CountFromProfile[i] << "\n");
> > + ValueSum += CountFromProfile[i];
> > + }
> > +
> > + DEBUG(dbgs() << "SUM = " << ValueSum << "\n");
> > +
> > + getBBInfo(nullptr).UnknownCountOutEdge = 2;
> > + getBBInfo(nullptr).UnknownCountInEdge = 2;
> > +
> > + setInstrumentedCounts(CountFromProfile);
> > + ProgramMaxCount = PGOReader->getMaximumFunctionCount();
> > + return true;
> > +}
> > +
> > +// Populate the counters from instrumented BBs to all BBs.
> > +// In the end of this operation, all BBs should have a valid count
> value.
> > +void PGOUseFunc::populateCounters() {
> > + // First set up Count variable for all BBs.
> > + for (auto &Ei : FuncInfo.MST.AllEdges) {
> > + if (Ei->Removed)
> > + continue;
> > +
> > + const BasicBlock *SrcBB = Ei->SrcBB;
> > + const BasicBlock *DestBB = Ei->DestBB;
> > + UseBBInfo &SrcInfo = getBBInfo(SrcBB);
> > + UseBBInfo &DestInfo = getBBInfo(DestBB);
> > + SrcInfo.OutEdges.push_back(Ei.get());
> > + DestInfo.InEdges.push_back(Ei.get());
> > + SrcInfo.UnknownCountOutEdge++;
> > + DestInfo.UnknownCountInEdge++;
> > +
> > + if (!Ei->CountValid)
> > + continue;
> > + DestInfo.UnknownCountInEdge--;
> > + SrcInfo.UnknownCountOutEdge--;
> > + }
> > +
> > + bool Changes = true;
> > + unsigned NumPasses = 0;
> > + while (Changes) {
> > + NumPasses++;
> > + Changes = false;
> > +
> > + // For efficient traversal, it's better to start from the end as
> most
> > + // of the instrumented edges are at the end.
> > + for (auto &BB : reverse(F)) {
> > + UseBBInfo &Count = getBBInfo(&BB);
> > + if (!Count.CountValid) {
> > + if (Count.UnknownCountOutEdge == 0) {
> > + Count.CountValue = sumEdgeCount(Count.OutEdges);
> > + Count.CountValid = true;
> > + Changes = true;
> > + } else if (Count.UnknownCountInEdge == 0) {
> > + Count.CountValue = sumEdgeCount(Count.InEdges);
> > + Count.CountValid = true;
> > + Changes = true;
> > + }
> > + }
> > + if (Count.CountValid) {
> > + if (Count.UnknownCountOutEdge == 1) {
> > + uint64_t Total = Count.CountValue -
> sumEdgeCount(Count.OutEdges);
> > + setEdgeCount(Count.OutEdges, Total);
> > + Changes = true;
> > + }
> > + if (Count.UnknownCountInEdge == 1) {
> > + uint64_t Total = Count.CountValue -
> sumEdgeCount(Count.InEdges);
> > + setEdgeCount(Count.InEdges, Total);
> > + Changes = true;
> > + }
> > + }
> > + }
> > + }
> > +
> > + DEBUG(dbgs() << "Populate counts in " << NumPasses << " passes.\n");
> > + // Assert every BB has a valid counter.
> > + uint64_t FuncEntryCount = getBBInfo(&*F.begin()).CountValue;
> > + uint64_t FuncMaxCount = FuncEntryCount;
> > + for (auto &BB : F) {
> > + assert(getBBInfo(&BB).CountValid && "BB count is not valid");
> > + uint64_t Count = getBBInfo(&BB).CountValue;
> > + if (Count > FuncMaxCount)
> > + FuncMaxCount = Count;
> > + }
> > + applyFunctionAttributes(FuncEntryCount, FuncMaxCount);
> > +
> > + DEBUG(FuncInfo.dumpInfo("after reading profile."));
> > +}
> > +
> > +// Assign the scaled count values to the BB with multiple out edges.
> > +void PGOUseFunc::setBranchWeights() {
> > + // Generate MD_prof metadata for every branch instruction.
> > + DEBUG(dbgs() << "\nSetting branch weights.\n");
> > + MDBuilder MDB(M->getContext());
> > + for (auto &BB : F) {
> > + TerminatorInst *TI = BB.getTerminator();
> > + if (TI->getNumSuccessors() < 2)
> > + continue;
> > + if (!isa<BranchInst>(TI) && !isa<SwitchInst>(TI))
> > + continue;
> > + if (getBBInfo(&BB).CountValue == 0)
> > + continue;
> > +
> > + // We have a non-zero Branch BB.
> > + const UseBBInfo &BBCountInfo = getBBInfo(&BB);
> > + unsigned Size = BBCountInfo.OutEdges.size();
> > + SmallVector<unsigned, 2> EdgeCounts(Size, 0);
> > + uint64_t MaxCount = 0;
> > + for (unsigned s = 0; s < Size; s++) {
> > + const PGOUseEdge *E = BBCountInfo.OutEdges[s];
> > + const BasicBlock *SrcBB = E->SrcBB;
> > + const BasicBlock *DestBB = E->DestBB;
> > + if (DestBB == 0)
> > + continue;
> > + unsigned SuccNum = GetSuccessorNumber(SrcBB, DestBB);
> > + uint64_t EdgeCount = E->CountValue;
> > + if (EdgeCount > MaxCount)
> > + MaxCount = EdgeCount;
> > + EdgeCounts[SuccNum] = EdgeCount;
> > + }
> > + assert(MaxCount > 0 && "Bad max count");
> > + uint64_t Scale = calculateCountScale(MaxCount);
> > + SmallVector<unsigned, 4> Weights;
> > + for (const auto &ECI : EdgeCounts)
> > + Weights.push_back(scaleBranchCount(ECI, Scale));
> > +
> > + TI->setMetadata(llvm::LLVMContext::MD_prof,
> > + MDB.createBranchWeights(Weights));
> > + DEBUG(dbgs() << "Weight is: "; for (const auto &W
> > + : Weights) dbgs()
> > + << W << " ";
> > + dbgs() << "\n";);
>
> This is formatted really strangely. It's possible that clang-format
> doesn't know how to deal with this code-in-DEBUG() stuff and you just
> need to format this manually (or file a bug on clang-format ;)
>
> > + }
> > +}
> > +} // end anonymous namespace
> > +
> > +bool PGOInstrumentationGen::runOnModule(Module &M) {
> > + for (auto &F : M) {
> > + if (F.isDeclaration())
> > + continue;
> > + BranchProbabilityInfo *BPI =
> > + &(getAnalysis<BranchProbabilityInfoWrapperPass>(F).getBPI());
> > + BlockFrequencyInfo *BFI =
> > + &(getAnalysis<BlockFrequencyInfoWrapperPass>(F).getBFI());
> > + instrumentOneFunc(F, &M, BPI, BFI);
> > + }
> > + return true;
> > +}
> > +
> > +static void setPGOCountOnFunc(PGOUseFunc &Func,
> > + IndexedInstrProfReader *PGOReader) {
> > + if (Func.readCounters(PGOReader)) {
> > + Func.populateCounters();
> > + Func.setBranchWeights();
> > + }
> > +}
> > +
> > +bool PGOInstrumentationUse::runOnModule(Module &M) {
> > + DEBUG(dbgs() << "Read in profile counters: ");
> > + auto &Ctx = M.getContext();
> > + // Read the counter array from file.
> > + auto ReaderOrErr = IndexedInstrProfReader::create(ProfileFileName);
> > + if (std::error_code EC = ReaderOrErr.getError()) {
> > + Ctx.diagnose(
> > + DiagnosticInfoPGOProfile(ProfileFileName.data(), EC.message()));
> > + return false;
> > + }
> > +
> > + PGOReader = std::move(ReaderOrErr.get());
> > + if (!PGOReader) {
> > + Ctx.diagnose(DiagnosticInfoPGOProfile(ProfileFileName.data(),
> > + "Cannot get PGOReader"));
> > + return false;
> > + }
> > +
> > + for (auto &F : M) {
> > + if (F.isDeclaration())
> > + continue;
> > + BranchProbabilityInfo *BPI =
> > + &(getAnalysis<BranchProbabilityInfoWrapperPass>(F).getBPI());
> > + BlockFrequencyInfo *BFI =
> > + &(getAnalysis<BlockFrequencyInfoWrapperPass>(F).getBFI());
> > + PGOUseFunc Func(F, &M, BPI, BFI);
> > + setPGOCountOnFunc(Func, PGOReader.get());
> > + }
> > + return true;
> > +}
> > Index: lib/Transforms/Instrumentation/LLVMBuild.txt
> > ===================================================================
> > --- lib/Transforms/Instrumentation/LLVMBuild.txt
> > +++ lib/Transforms/Instrumentation/LLVMBuild.txt
> > @@ -19,4 +19,4 @@
> > type = Library
> > name = Instrumentation
> > parent = Transforms
> > -required_libraries = Analysis Core MC Support TransformUtils
> > +required_libraries = Analysis Core MC Support TransformUtils ProfileData
> > Index: lib/Transforms/Instrumentation/Instrumentation.cpp
> > ===================================================================
> > --- lib/Transforms/Instrumentation/Instrumentation.cpp
> > +++ lib/Transforms/Instrumentation/Instrumentation.cpp
> > @@ -60,6 +60,8 @@
> > initializeAddressSanitizerModulePass(Registry);
> > initializeBoundsCheckingPass(Registry);
> > initializeGCOVProfilerPass(Registry);
> > + initializePGOInstrumentationGenPass(Registry);
> > + initializePGOInstrumentationUsePass(Registry);
> > initializeInstrProfilingPass(Registry);
> > initializeMemorySanitizerPass(Registry);
> > initializeThreadSanitizerPass(Registry);
> > Index: lib/Transforms/Instrumentation/CMakeLists.txt
> > ===================================================================
> > --- lib/Transforms/Instrumentation/CMakeLists.txt
> > +++ lib/Transforms/Instrumentation/CMakeLists.txt
> > @@ -6,6 +6,7 @@
> > MemorySanitizer.cpp
> > Instrumentation.cpp
> > InstrProfiling.cpp
> > + PGOInstrumentation.cpp
> > SafeStack.cpp
> > SanitizerCoverage.cpp
> > ThreadSanitizer.cpp
> > Index: lib/Transforms/Instrumentation/CFGMST.h
> > ===================================================================
> > --- /dev/null
> > +++ lib/Transforms/Instrumentation/CFGMST.h
> > @@ -0,0 +1,211 @@
> > +//===-- CFGMST.h - Minimum Spanning Tree for CFG -------*- C++ -*-===//
> > +//
> > +// The LLVM Compiler Infrastructure
> > +//
> > +// This file is distributed under the University of Illinois Open Source
> > +// License. See LICENSE.TXT for details.
> > +//
> >
> +//===----------------------------------------------------------------------===//
> > +//
> > +// This file implements Union-find algorithm to compute Minimum
> Spanning Tree
> > +// for a given CFG.
> > +//
> >
> +//===----------------------------------------------------------------------===//
> > +
> > +#include "llvm/ADT/DenseMap.h"
> > +#include "llvm/ADT/STLExtras.h"
> > +#include "llvm/Analysis/BlockFrequencyInfo.h"
> > +#include "llvm/Analysis/BranchProbabilityInfo.h"
> > +#include "llvm/Analysis/CFG.h"
> > +#include "llvm/Support/BranchProbability.h"
> > +#include "llvm/Support/Debug.h"
> > +#include "llvm/Support/raw_ostream.h"
> > +#include "llvm/Transforms/Utils/BasicBlockUtils.h"
> > +#include <string>
> > +#include <utility>
> > +#include <vector>
> > +
> > +namespace llvm {
> > +
> > +#define DEBUG_TYPE "cfgmst"
> > +
> > +template <class Edge, class BBInfo> class CFGMST {
> > +public:
> > + Function &F;
> > +
> > + // Store all the edges in CFG. It may contain some stale edges
> > + // when Removed is set.
>
> Might as well use doc comments here.
>
> > + std::vector<std::unique_ptr<Edge>> AllEdges;
> > +
> > + // This map records the auxiliary information for each BB.
> > + DenseMap<const BasicBlock *, std::unique_ptr<BBInfo>> BBInfos;
> > +
> > + // Find the root group of the G and compress the path from G to the
> root.
> > + BBInfo *findAndCompressGroup(BBInfo *G) {
> > + if (G->Group != G)
> > + G->Group = findAndCompressGroup(static_cast<BBInfo *>(G->Group));
> > + return static_cast<BBInfo *>(G->Group);
> > + }
> > +
> > + // Union BB1 and BB2 into the same group and return true.
> > + // Returns false if BB1 and BB2 are already in the same group.
> > + bool unionGroups(const BasicBlock *BB1, const BasicBlock *BB2) {
> > + BBInfo *BB1G = findAndCompressGroup(&getBBInfo(BB1));
> > + BBInfo *BB2G = findAndCompressGroup(&getBBInfo(BB2));
> > +
> > + if (BB1G == BB2G)
> > + return false;
> > +
> > + // Make the smaller rank tree a direct child or the root of high
> rank tree.
> > + if (BB1G->Rank < BB2G->Rank)
> > + BB1G->Group = BB2G;
> > + else {
> > + BB2G->Group = BB1G;
> > + // If the ranks are the same, increment root of one tree by one.
> > + if (BB1G->Rank == BB2G->Rank)
> > + BB1G->Rank++;
> > + }
> > + return true;
> > + }
> > +
> > + // Give BB, return the auxiliary information.
> > + BBInfo &getBBInfo(const BasicBlock *BB) const {
> > + auto It = BBInfos.find(BB);
> > + assert(It->second.get() != nullptr);
> > + return *It->second.get();
> > + }
> > +
> > + // Traverse the CFG using a stack. Find all the edges and assign the
> weight.
> > + // Edges with large weight will be put into MST first so they are
> less likely
> > + // to be instrumented.
> > + void buildEdges() {
> > + DEBUG(dbgs() << "Build Edge on " << F.getName() << "\n");
> > +
> > + const BasicBlock *BB = &(F.getEntryBlock());
> > + // Add a fake edge to the entry.
> > + addEdge(nullptr, BB, BFI->getEntryFreq());
> > +
> > + // Special handling for single BB functions.
> > + if (succ_empty(BB)) {
> > + addEdge(BB, nullptr, BFI->getEntryFreq());
> > + return;
> > + }
> > +
> > + static const uint32_t CriticalEdgeMultiplier = 1000;
> > +
> > + for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
> {
> > + TerminatorInst *TI = BB->getTerminator();
> > + uint64_t BBWeight = BFI->getBlockFreq(&*BB).getFrequency();
> > + uint64_t Weight;
> > + if (int successors = TI->getNumSuccessors()) {
> > + for (int i = 0; i != successors; ++i) {
> > + BasicBlock *TargetBB = TI->getSuccessor(i);
> > + bool Critical = isCriticalEdge(TI, i);
> > + uint64_t scaleFactor = BBWeight;
> > + if (Critical) {
> > + if (scaleFactor < UINT64_MAX / CriticalEdgeMultiplier)
> > + scaleFactor *= CriticalEdgeMultiplier;
> > + else
> > + scaleFactor = UINT64_MAX;
> > + }
> > + Weight = BPI->getEdgeProbability(&*BB,
> TargetBB).scale(scaleFactor);
> > + addEdge(&*BB, TargetBB, Weight).IsCritical = Critical;
> > + DEBUG(dbgs() << " Edge: from " << BB->getName() << " to "
> > + << TargetBB->getName() << " w=" << Weight <<
> "\n");
> > + }
> > + } else {
> > + addEdge(&*BB, nullptr, BBWeight);
> > + DEBUG(dbgs() << " Edge: from " << BB->getName() << " to exit"
> > + << " w = " << BBWeight << "\n");
> > + }
> > + }
> > + }
> > +
> > + // Sort CFG edges based on its weight.
> > + void sortEdgesByWeight() {
> > + std::stable_sort(
> > + AllEdges.begin(), AllEdges.end(),
> > + [](const std::unique_ptr<Edge> &Edge1, const
> std::unique_ptr<Edge> &Edge2) {
> > + return Edge1->Weight > Edge2->Weight;
> > + });
> > + }
> > +
> > + // Traverse all the edges and compute the Minimum Weight Spanning Tree
> > + // using union-find algorithm.
> > + void computeMinimumSpanningTree() {
> > + // First, put all the critical edge with landing-pad as the Dest to
> MST.
> > + // This works around the insufficient support of critical edges
> split
> > + // when destination BB is a landing pad.
> > + for (auto &Ei : AllEdges) {
> > + if (Ei->Removed)
> > + continue;
> > + if (Ei->IsCritical) {
> > + if (Ei->DestBB && Ei->DestBB->isLandingPad()) {
> > + if (unionGroups(Ei->SrcBB, Ei->DestBB))
> > + Ei->InMST = true;
> > + }
> > + }
> > + }
> > +
> > + for (auto &Ei : AllEdges) {
> > + if (Ei->Removed)
> > + continue;
> > + if (unionGroups(Ei->SrcBB, Ei->DestBB))
> > + Ei->InMST = true;
> > + }
> > + }
> > +
> > + // Dump the Debug information about the instrumentation.
> > + void dumpEdges(raw_ostream &OS, const StringRef Message =
> StringRef()) const {
> > + if (!Message.empty())
> > + OS << Message << "\n";
> > + OS << " Number of Basic Blocks: " << BBInfos.size() << "\n";
> > + for (auto &BI : BBInfos) {
> > + const BasicBlock *BB = BI.first;
> > + OS << " BB: " << (BB == nullptr ? "FakeNode" : BB->getName()) <<
> " "
> > + << BI.second->infoString() << "\n";
> > + }
> > +
> > + OS << " Number of Edges: " << AllEdges.size()
> > + << " (*: Instrument, C: CriticalEdge, -: Removed)\n";
> > + uint32_t Count = 0;
> > + for (auto &EI : AllEdges) {
> > + OS << " Edge " << Count++ << ": " << getBBInfo(EI->SrcBB).Index
> << "-->"
> > + << getBBInfo(EI->DestBB).Index << EI->infoString() << "\n";
> > + };
> > + }
> > +
> > + // Add an edge to AllEdges with weight W.
> > + Edge &addEdge(const BasicBlock *Src, const BasicBlock *Dest, uint64_t
> W) {
> > + uint32_t Index = BBInfos.size();
> > + auto Iter = BBInfos.end();
> > + bool Inserted;
> > + std::tie(Iter, Inserted) = BBInfos.insert(std::make_pair(Src,
> nullptr));
> > + if (Inserted) {
> > + // Newly inserted, update the real info.
> > + Iter->second = std::move(llvm::make_unique<BBInfo>(Index));
> > + Index++;
> > + }
> > + std::tie(Iter, Inserted) = BBInfos.insert(std::make_pair(Dest,
> nullptr));
> > + if (Inserted)
> > + // Newly inserted, update the real info.
> > + Iter->second = std::move(llvm::make_unique<BBInfo>(Index));
> > + AllEdges.emplace_back(new Edge(Src, Dest, W));
> > + return *AllEdges.back();
> > + }
> > +
> > + BranchProbabilityInfo *BPI;
> > + BlockFrequencyInfo *BFI;
> > +
> > +public:
> > + CFGMST(Function &Func, BranchProbabilityInfo *BPI_ = nullptr,
> > + BlockFrequencyInfo *BFI_ = nullptr)
> > + : F(Func), BPI(BPI_), BFI(BFI_) {
> > + buildEdges();
> > + sortEdgesByWeight();
> > + computeMinimumSpanningTree();
> > + }
> > +};
> > +
> > +#undef DEBUG_TYPE // "cfgmst"
> > +} // end namespace llvm
> > Index: lib/IR/DiagnosticInfo.cpp
> > ===================================================================
> > --- lib/IR/DiagnosticInfo.cpp
> > +++ lib/IR/DiagnosticInfo.cpp
> > @@ -132,6 +132,12 @@
> > DP << getMsg();
> > }
> >
> > +void DiagnosticInfoPGOProfile::print(DiagnosticPrinter &DP) const {
> > + if (getFileName())
> > + DP << getFileName() << ": ";
> > + DP << getMsg();
> > +}
> > +
> > bool DiagnosticInfoOptimizationBase::isLocationAvailable() const {
> > return getDebugLoc();
> > }
> > Index: include/llvm/Transforms/Instrumentation.h
> > ===================================================================
> > --- include/llvm/Transforms/Instrumentation.h
> > +++ include/llvm/Transforms/Instrumentation.h
> > @@ -79,6 +79,11 @@
> > ModulePass *createGCOVProfilerPass(const GCOVOptions &Options =
> > GCOVOptions::getDefault());
> >
> > +// PGO Instrumention
> > +ModulePass *createPGOInstrumentationGenPass();
> > +ModulePass *
> > +createPGOInstrumentationUsePass(StringRef Filename = StringRef(""));
> > +
> > /// Options for the frontend instrumentation based profiling pass.
> > struct InstrProfOptions {
> > InstrProfOptions() : NoRedZone(false) {}
> > @@ -149,6 +154,24 @@
> > /// protect against stack-based overflow vulnerabilities.
> > FunctionPass *createSafeStackPass(const TargetMachine *TM = nullptr);
> >
> > +/// \brief Calculate what to divide by to scale counts.
> > +///
> > +/// Given the maximum count, calculate a divisor that will scale all the
> > +/// weights to strictly less than UINT32_MAX.
> > +static inline uint64_t calculateCountScale(uint64_t MaxCount) {
> > + return MaxCount < UINT32_MAX ? 1 : MaxCount / UINT32_MAX + 1;
> > +}
> > +
> > +/// \brief Scale an individual branch count.
> > +///
> > +/// Scale a 64-bit weight down to 32-bits using \c Scale.
> > +///
> > +static inline uint32_t scaleBranchCount(uint64_t Count, uint64_t Scale)
> {
> > + uint64_t Scaled = Count / Scale;
> > + assert(Scaled <= UINT32_MAX && "overflow 32-bits");
> > + return Scaled;
> > +}
> > +
> > } // End llvm namespace
> >
> > #endif
> > Index: include/llvm/LinkAllPasses.h
> > ===================================================================
> > --- include/llvm/LinkAllPasses.h
> > +++ include/llvm/LinkAllPasses.h
> > @@ -85,6 +85,8 @@
> > (void) llvm::createDomOnlyViewerPass();
> > (void) llvm::createDomViewerPass();
> > (void) llvm::createGCOVProfilerPass();
> > + (void) llvm::createPGOInstrumentationGenPass();
> > + (void) llvm::createPGOInstrumentationUsePass();
> > (void) llvm::createInstrProfilingPass();
> > (void) llvm::createFunctionInliningPass();
> > (void) llvm::createAlwaysInlinerPass();
> > Index: include/llvm/InitializePasses.h
> > ===================================================================
> > --- include/llvm/InitializePasses.h
> > +++ include/llvm/InitializePasses.h
> > @@ -117,6 +117,8 @@
> > void initializeExpandPostRAPass(PassRegistry&);
> > void initializeAAResultsWrapperPassPass(PassRegistry &);
> > void initializeGCOVProfilerPass(PassRegistry&);
> > +void initializePGOInstrumentationGenPass(PassRegistry&);
> > +void initializePGOInstrumentationUsePass(PassRegistry&);
> > void initializeInstrProfilingPass(PassRegistry&);
> > void initializeAddressSanitizerPass(PassRegistry&);
> > void initializeAddressSanitizerModulePass(PassRegistry&);
> > Index: include/llvm/IR/DiagnosticInfo.h
> > ===================================================================
> > --- include/llvm/IR/DiagnosticInfo.h
> > +++ include/llvm/IR/DiagnosticInfo.h
> > @@ -60,6 +60,7 @@
> > DK_OptimizationRemarkAnalysisAliasing,
> > DK_OptimizationFailure,
> > DK_MIRParser,
> > + DK_PGOProfile,
> > DK_FirstPluginKind
> > };
> >
> > @@ -250,6 +251,31 @@
> > const Twine &Msg;
> > };
> >
> > +/// Diagnostic information for the PGO profiler.
> > +class DiagnosticInfoPGOProfile : public DiagnosticInfo {
> > +public:
> > + DiagnosticInfoPGOProfile(const char *FileName, const Twine &Msg,
> > + DiagnosticSeverity Severity = DS_Error)
> > + : DiagnosticInfo(DK_PGOProfile, Severity), FileName(FileName),
> Msg(Msg) {}
> > +
> > + /// \see DiagnosticInfo::print.
> > + void print(DiagnosticPrinter &DP) const override;
> > +
> > + static bool classof(const DiagnosticInfo *DI) {
> > + return DI->getKind() == DK_PGOProfile;
> > + }
> > +
> > + const char *getFileName() const { return FileName; }
> > + const Twine &getMsg() const { return Msg; }
> > +
> > +private:
> > + /// Name of the input file associated with this diagnostic.
> > + const char *FileName;
> > +
> > + /// Message to report.
> > + const Twine &Msg;
> > +};
> > +
> > /// Common features for diagnostics dealing with optimization remarks.
> > class DiagnosticInfoOptimizationBase : public DiagnosticInfo {
> > public:
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20151201/e3fb75d8/attachment-0001.html>
More information about the llvm-commits
mailing list