Refactor synthetic profile count computation. NFC.

[eraman]

Summary

Instead of using two separate callbacks to return the entry count and the
relative block frequency, use a single callback to return callsite
count. This would allow better supporting hybrid mode in the future as
the count of callsite need not always be derived from entry count (as in
sample PGO).

[davidxl]

lib/Analysis/SyntheticCountsUtils.cpp

62

To avoid overflow, AdditionalCounts be changed to a map to Scaled64?

[eraman]

Address review comments

[eraman]

Also defined AddCountTy to be function_ref<void(NodeRef, Scaled64)>

[davidxl]

lgtm

include/llvm/Analysis/SyntheticCountsUtils.h

 template <typename CallGraphType> class SyntheticCountsUtils {
 public:
   using Scaled64 = ScaledNumber<uint64_t>;
   using CGT = GraphTraits<CallGraphType>;
   using NodeRef = typename CGT::NodeRef;
   using EdgeRef = typename CGT::EdgeRef;
   using SccTy = std::vector<NodeRef>;
 
-  using GetRelBBFreqTy = function_ref<Optional<Scaled64>(EdgeRef)>;
-  using GetCountTy = function_ref<uint64_t(NodeRef)>;
-  using AddCountTy = function_ref<void(NodeRef, uint64_t)>;
+  // Not all EdgeRef have information about the source of the edge. Hence
+  // NodeRef corresponding to the source of the EdgeRef is explicitly passed.
+  using GetProfCountTy = function_ref<Optional<Scaled64>(NodeRef, EdgeRef)>;
+  using AddCountTy = function_ref<void(NodeRef, Scaled64)>;
 
-  static void propagate(const CallGraphType &CG, GetRelBBFreqTy GetRelBBFreq,
-                        GetCountTy GetCount, AddCountTy AddCount);
+  static void propagate(const CallGraphType &CG, GetProfCountTy GetProfCount,
+                        AddCountTy AddCount);
 
 private:
-  static void propagateFromSCC(const SccTy &SCC, GetRelBBFreqTy GetRelBBFreq,
-                               GetCountTy GetCount, AddCountTy AddCount);
+  static void propagateFromSCC(const SccTy &SCC, GetProfCountTy GetProfCount,
+                               AddCountTy AddCount);
 };
 } // namespace llvm
 
 #endif

lib/Analysis/SyntheticCountsUtils.cpp

 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/ModuleSummaryIndex.h"
 
 using namespace llvm;
 
 // Given an SCC, propagate entry counts along the edge of the SCC nodes.
 template <typename CallGraphType>
 void SyntheticCountsUtils<CallGraphType>::propagateFromSCC(
-    const SccTy &SCC, GetRelBBFreqTy GetRelBBFreq, GetCountTy GetCount,
-    AddCountTy AddCount) {
+    const SccTy &SCC, GetProfCountTy GetProfCount, AddCountTy AddCount) {
 
   DenseSet<NodeRef> SCCNodes;
   SmallVector<std::pair<NodeRef, EdgeRef>, 8> SCCEdges, NonSCCEdges;
 
   for (auto &Node : SCC)
     SCCNodes.insert(Node);
 
   // Partition the edges coming out of the SCC into those whose destination is
void SyntheticCountsUtils<CallGraphType>::propagateFromSCC(
   // For nodes in the same SCC, update the counts in two steps:
   // 1. Compute the additional count for each node by propagating the counts
   // along all incoming edges to the node that originate from within the same
   // SCC and summing them up.
   // 2. Add the additional counts to the nodes in the SCC.
   // This ensures that the order of
   // traversal of nodes within the SCC doesn't affect the final result.
 
-  DenseMap<NodeRef, uint64_t> AdditionalCounts;
+  DenseMap<NodeRef, Scaled64> AdditionalCounts;
   for (auto &E : SCCEdges) {
-    auto OptRelFreq = GetRelBBFreq(E.second);
-    if (!OptRelFreq)
+    auto OptProfCount = GetProfCount(E.first, E.second);
+    if (!OptProfCount)
       continue;
-    Scaled64 RelFreq = OptRelFreq.getValue();
-    auto Caller = E.first;
     auto Callee = CGT::edge_dest(E.second);
-    RelFreq *= Scaled64(GetCount(Caller), 0);
-    uint64_t AdditionalCount = RelFreq.toInt<uint64_t>();
-    AdditionalCounts[Callee] += AdditionalCount;
+    AdditionalCounts[Callee] += OptProfCount.getValue();
   }
 
   // Update the counts for the nodes in the SCC.
   for (auto &Entry : AdditionalCounts)
     AddCount(Entry.first, Entry.second);
 
   // Now update the counts for nodes outside the SCC.
   for (auto &E : NonSCCEdges) {
-    auto OptRelFreq = GetRelBBFreq(E.second);
-    if (!OptRelFreq)
+    auto OptProfCount = GetProfCount(E.first, E.second);
+    if (!OptProfCount)
       continue;
-    Scaled64 RelFreq = OptRelFreq.getValue();
-    auto Caller = E.first;
     auto Callee = CGT::edge_dest(E.second);
-    RelFreq *= Scaled64(GetCount(Caller), 0);
-    AddCount(Callee, RelFreq.toInt<uint64_t>());
+    AddCount(Callee, OptProfCount.getValue());
   }
 }
 
 /// Propgate synthetic entry counts on a callgraph \p CG.
 ///
 /// This performs a reverse post-order traversal of the callgraph SCC. For each
 /// SCC, it first propagates the entry counts to the nodes within the SCC
 /// through call edges and updates them in one shot. Then the entry counts are
 /// propagated to nodes outside the SCC. This requires \p GraphTraits
 /// to have a specialization for \p CallGraphType.
 
 template <typename CallGraphType>
 void SyntheticCountsUtils<CallGraphType>::propagate(const CallGraphType &CG,
-                                                    GetRelBBFreqTy GetRelBBFreq,
-                                                    GetCountTy GetCount,
+                                                    GetProfCountTy GetProfCount,
                                                     AddCountTy AddCount) {
   std::vector<SccTy> SCCs;
 
   // Collect all the SCCs.
   for (auto I = scc_begin(CG); !I.isAtEnd(); ++I)
     SCCs.push_back(*I);
 
   // The callgraph-scc needs to be visited in top-down order for propagation.
   // The scc iterator returns the scc in bottom-up order, so reverse the SCCs
   // and call propagateFromSCC.
   for (auto &SCC : reverse(SCCs))
-    propagateFromSCC(SCC, GetRelBBFreq, GetCount, AddCount);
+    propagateFromSCC(SCC, GetProfCount, AddCount);
 }
 
 template class llvm::SyntheticCountsUtils<const CallGraph *>;
 template class llvm::SyntheticCountsUtils<ModuleSummaryIndex *>;

lib/LTO/SummaryBasedOptimizations.cpp

auto GetEntryCount = [](ValueInfo V) {
     if (V.getSummaryList().size()) {
       auto S = V.getSummaryList().front().get()->getBaseObject();
       auto *F = cast<FunctionSummary>(S);
       return F->entryCount();
     } else {
       return UINT64_C(0);
     }
   };
-  auto AddToEntryCount = [](ValueInfo V, uint64_t New) {
+  auto AddToEntryCount = [](ValueInfo V, Scaled64 New) {
     if (!V.getSummaryList().size())
       return;
     for (auto &GVS : V.getSummaryList()) {
       auto S = GVS.get()->getBaseObject();
       auto *F = cast<FunctionSummary>(S);
-      F->setEntryCount(SaturatingAdd(F->entryCount(), New));
+      F->setEntryCount(
+          SaturatingAdd(F->entryCount(), New.template toInt<uint64_t>()));
     }
   };
 
+  auto GetProfileCount = [&](ValueInfo V, FunctionSummary::EdgeTy &Edge) {
+    auto RelFreq = GetCallSiteRelFreq(Edge);
+    Scaled64 EC(GetEntryCount(V), 0);
+    return RelFreq * EC;
+  };
   // After initializing the counts in initializeCounts above, the counts have to
   // be propagated across the combined callgraph.
   // SyntheticCountsUtils::propagate takes care of this propagation on any
   // callgraph that specialized GraphTraits.
-  SyntheticCountsUtils<ModuleSummaryIndex *>::propagate(
-      &Index, GetCallSiteRelFreq, GetEntryCount, AddToEntryCount);
+  SyntheticCountsUtils<ModuleSummaryIndex *>::propagate(&Index, GetProfileCount,
+                                                        AddToEntryCount);
   Index.setHasSyntheticEntryCounts();
 }

lib/Transforms/IPO/SyntheticCountsPropagation.cpp

 //
 //===----------------------------------------------------------------------===//
 
 #include "llvm/Transforms/IPO/SyntheticCountsPropagation.h"
 #include "llvm/ADT/DenseSet.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/Analysis/BlockFrequencyInfo.h"
 #include "llvm/Analysis/CallGraph.h"
+#include "llvm/Analysis/ProfileSummaryInfo.h"
 #include "llvm/Analysis/SyntheticCountsUtils.h"
 #include "llvm/IR/CallSite.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/Module.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
for (Function &F : M) {
     SetCount(&F, InitialCount);
   }
 }
 
 PreservedAnalyses SyntheticCountsPropagation::run(Module &M,
                                                   ModuleAnalysisManager &MAM) {
   FunctionAnalysisManager &FAM =
       MAM.getResult<FunctionAnalysisManagerModuleProxy>(M).getManager();
-  DenseMap<Function *, uint64_t> Counts;
+  DenseMap<Function *, Scaled64> Counts;
   // Set initial entry counts.
-  initializeCounts(M, [&](Function *F, uint64_t Count) { Counts[F] = Count; });
+  initializeCounts(
+      M, [&](Function *F, uint64_t Count) { Counts[F] = Scaled64(Count, 0); });
 
-  // Compute the relative block frequency for a call edge. Use scaled numbers
-  // and not integers since the relative block frequency could be less than 1.
-  auto GetCallSiteRelFreq = [&](const CallGraphNode::CallRecord &Edge) {
+  // Edge includes information about the source. Hence ignore the first
+  // parameter.
+  auto GetCallSiteProfCount = [&](const CallGraphNode *,
+                                  const CallGraphNode::CallRecord &Edge) {
     Optional<Scaled64> Res = None;
     if (!Edge.first)
       return Res;
     assert(isa<Instruction>(Edge.first));
     CallSite CS(cast<Instruction>(Edge.first));
     Function *Caller = CS.getCaller();
     auto &BFI = FAM.getResult<BlockFrequencyAnalysis>(*Caller);
+
+    // Now compute the callsite count from relative frequency and
+    // entry count:
     BasicBlock *CSBB = CS.getInstruction()->getParent();
     Scaled64 EntryFreq(BFI.getEntryFreq(), 0);
-    Scaled64 BBFreq(BFI.getBlockFreq(CSBB).getFrequency(), 0);
-    BBFreq /= EntryFreq;
-    return Optional<Scaled64>(BBFreq);
+    Scaled64 BBCount(BFI.getBlockFreq(CSBB).getFrequency(), 0);
+    BBCount /= EntryFreq;
+    BBCount *= Counts[Caller];
+    return Optional<Scaled64>(BBCount);
   };
 
   CallGraph CG(M);
   // Propgate the entry counts on the callgraph.
   SyntheticCountsUtils<const CallGraph *>::propagate(
-      &CG, GetCallSiteRelFreq,
-      [&](const CallGraphNode *N) { return Counts[N->getFunction()]; },
-      [&](const CallGraphNode *N, uint64_t New) {
+      &CG, GetCallSiteProfCount, [&](const CallGraphNode *N, Scaled64 New) {
         auto F = N->getFunction();
         if (!F || F->isDeclaration())
           return;
+
         Counts[F] += New;
       });
 
   // Set the counts as metadata.
-  for (auto Entry : Counts)
-    Entry.first->setEntryCount(
-        ProfileCount(Entry.second, Function::PCT_Synthetic));
+  for (auto Entry : Counts) {
+    Entry.first->setEntryCount(ProfileCount(
+        Entry.second.template toInt<uint64_t>(), Function::PCT_Synthetic));
+  }
 
   return PreservedAnalyses::all();
 }