[ThinLTO] Add GraphTraits for FunctionSummaries

Add GraphTraits definitions to the FunctionSummary and ModuleSummaryIndex classes. These GraphTraits will be used to construct find SCC's in ThinLTO analysis passes.

Third attempt - moved function from lambda to static function due to build failures.

llvm-svn: 325506

include/llvm/IR/ModuleSummaryIndex.h

@@ -162,6 +162,24 @@ struct ValueInfo {
   bool isDSOLocal() const;
 };
 
+inline bool operator==(const ValueInfo &A, const ValueInfo &B) {
+  assert(A.getRef() && B.getRef() &&
+         "Need ValueInfo with non-null Ref to compare GUIDs");
+  return A.getRef() == B.getRef();
+}
+
+inline bool operator!=(const ValueInfo &A, const ValueInfo &B) {
+  assert(A.getRef() && B.getRef() &&
+         "Need ValueInfo with non-null Ref to compare GUIDs");
+  return A.getGUID() != B.getGUID();
+}
+
+inline bool operator<(const ValueInfo &A, const ValueInfo &B) {
+  assert(A.getRef() && B.getRef() &&
+         "Need ValueInfo with non-null Ref to compare GUIDs");
+  return A.getGUID() < B.getGUID();
+}
+
 template <> struct DenseMapInfo<ValueInfo> {
   static inline ValueInfo getEmptyKey() {
     return ValueInfo(false, (GlobalValueSummaryMapTy::value_type *)-8);
@@ -397,6 +415,25 @@ public:
     unsigned ReturnDoesNotAlias : 1;
   };
 
+  /// Create an empty FunctionSummary (with specified call edges).
+  /// Used to represent external nodes and the dummy root node.
+  static FunctionSummary
+  makeDummyFunctionSummary(std::vector<FunctionSummary::EdgeTy> Edges) {
+    return FunctionSummary(
+        FunctionSummary::GVFlags(
+            GlobalValue::LinkageTypes::AvailableExternallyLinkage,
+            /*NotEligibleToImport=*/true, /*Live=*/true, /*IsLocal=*/false),
+        0, FunctionSummary::FFlags{}, std::vector<ValueInfo>(),
+        std::move(Edges), std::vector<GlobalValue::GUID>(),
+        std::vector<FunctionSummary::VFuncId>(),
+        std::vector<FunctionSummary::VFuncId>(),
+        std::vector<FunctionSummary::ConstVCall>(),
+        std::vector<FunctionSummary::ConstVCall>());
+  }
+
+  /// A dummy node to reference external functions that aren't in the index
+  static FunctionSummary ExternalNode;
+
 private:
   /// Number of instructions (ignoring debug instructions, e.g.) computed
   /// during the initial compile step when the summary index is first built.
@@ -516,6 +553,8 @@ public:
       TIdInfo = llvm::make_unique<TypeIdInfo>();
     TIdInfo->TypeTests.push_back(Guid);
   }
+
+  friend struct GraphTraits<ValueInfo>;
 };
 
 template <> struct DenseMapInfo<FunctionSummary::VFuncId> {
@@ -718,6 +757,69 @@ public:
   const_gvsummary_iterator end() const { return GlobalValueMap.end(); }
   size_t size() const { return GlobalValueMap.size(); }
 
+  /// Convenience function for doing a DFS on a ValueInfo. Marks the function in
+  /// the FunctionHasParent map.
+  static void discoverNodes(ValueInfo V,
+                            std::map<ValueInfo, bool> &FunctionHasParent) {
+    if (!V.getSummaryList().size())
+      return; // skip external functions that don't have summaries
+
+    // Mark discovered if we haven't yet
+    auto S = FunctionHasParent.emplace(V, false);
+
+    // Stop if we've already discovered this node
+    if (!S.second)
+      return;
+
+    FunctionSummary *F =
+        dyn_cast<FunctionSummary>(V.getSummaryList().front().get());
+    assert(F != nullptr && "Expected FunctionSummary node");
+
+    for (auto &C : F->calls()) {
+      // Insert node if necessary
+      auto S = FunctionHasParent.emplace(C.first, true);
+
+      // Skip nodes that we're sure have parents
+      if (!S.second && S.first->second)
+        continue;
+
+      if (S.second)
+        discoverNodes(C.first, FunctionHasParent);
+      else
+        S.first->second = true;
+    }
+  }
+
+  // Calculate the callgraph root
+  FunctionSummary calculateCallGraphRoot() {
+    // Functions that have a parent will be marked in FunctionHasParent pair.
+    // Once we've marked all functions, the functions in the map that are false
+    // have no parent (so they're the roots)
+    std::map<ValueInfo, bool> FunctionHasParent;
+
+    for (auto &S : *this) {
+      // Skip external functions
+      if (!S.second.SummaryList.size() ||
+          !isa<FunctionSummary>(S.second.SummaryList.front().get()))
+        continue;
+      discoverNodes(ValueInfo(IsAnalysis, &S), FunctionHasParent);
+    }
+
+    std::vector<FunctionSummary::EdgeTy> Edges;
+    // create edges to all roots in the Index
+    for (auto &P : FunctionHasParent) {
+      if (P.second)
+        continue; // skip over non-root nodes
+      Edges.push_back(std::make_pair(P.first, CalleeInfo{}));
+    }
+    if (Edges.empty()) {
+      // Failed to find root - return an empty node
+      return FunctionSummary::makeDummyFunctionSummary({});
+    }
+    auto CallGraphRoot = FunctionSummary::makeDummyFunctionSummary(Edges);
+    return CallGraphRoot;
+  }
+
   bool withGlobalValueDeadStripping() const {
     return WithGlobalValueDeadStripping;
   }
@@ -748,7 +850,7 @@ public:
     return ValueInfo(IsAnalysis, getOrInsertValuePtr(GUID));
   }
 
-  /// Return a ValueInfo for \p GUID setting value \p Name. 
+  /// Return a ValueInfo for \p GUID setting value \p Name.
   ValueInfo getOrInsertValueInfo(GlobalValue::GUID GUID, StringRef Name) {
     assert(!IsAnalysis);
     auto VP = getOrInsertValuePtr(GUID);
@@ -923,6 +1025,56 @@ public:
 
   /// Export summary to dot file for GraphViz.
   void exportToDot(raw_ostream& OS) const;
+
+  /// Print out strongly connected components for debugging.
+  void dumpSCCs(raw_ostream &OS);
+};
+
+/// GraphTraits definition to build SCC for the index
+template <> struct GraphTraits<ValueInfo> {
+  typedef ValueInfo NodeRef;
+
+  static NodeRef valueInfoFromEdge(FunctionSummary::EdgeTy &P) {
+    return P.first;
+  }
+  using ChildIteratorType =
+      mapped_iterator<std::vector<FunctionSummary::EdgeTy>::iterator,
+                      decltype(&valueInfoFromEdge)>;
+
+  static NodeRef getEntryNode(ValueInfo V) { return V; }
+
+  static ChildIteratorType child_begin(NodeRef N) {
+    if (!N.getSummaryList().size()) // handle external function
+      return ChildIteratorType(
+          FunctionSummary::ExternalNode.CallGraphEdgeList.begin(),
+          &valueInfoFromEdge);
+    FunctionSummary *F =
+        cast<FunctionSummary>(N.getSummaryList().front()->getBaseObject());
+    return ChildIteratorType(F->CallGraphEdgeList.begin(), &valueInfoFromEdge);
+  }
+
+  static ChildIteratorType child_end(NodeRef N) {
+    if (!N.getSummaryList().size()) // handle external function
+      return ChildIteratorType(
+          FunctionSummary::ExternalNode.CallGraphEdgeList.end(),
+          &valueInfoFromEdge);
+    FunctionSummary *F =
+        cast<FunctionSummary>(N.getSummaryList().front()->getBaseObject());
+    return ChildIteratorType(F->CallGraphEdgeList.end(), &valueInfoFromEdge);
+  }
+};
+
+template <>
+struct GraphTraits<ModuleSummaryIndex *> : public GraphTraits<ValueInfo> {
+  static NodeRef getEntryNode(ModuleSummaryIndex *I) {
+    std::unique_ptr<GlobalValueSummary> Root =
+        make_unique<FunctionSummary>(I->calculateCallGraphRoot());
+    GlobalValueSummaryInfo G(I->isPerformingAnalysis());
+    G.SummaryList.push_back(std::move(Root));
+    static auto P =
+        GlobalValueSummaryMapTy::value_type(GlobalValue::GUID(0), std::move(G));
+    return ValueInfo(I->isPerformingAnalysis(), &P);
+  }
 };
 
 } // end namespace llvm

lib/IR/ModuleSummaryIndex.cpp

@@ -13,10 +13,14 @@
 //===----------------------------------------------------------------------===//
 
 #include "llvm/IR/ModuleSummaryIndex.h"
+#include "llvm/ADT/SCCIterator.h"
 #include "llvm/ADT/StringMap.h"
 #include "llvm/Support/Path.h"
+#include "llvm/Support/raw_ostream.h"
 using namespace llvm;
 
+FunctionSummary FunctionSummary::ExternalNode =
+    FunctionSummary::makeDummyFunctionSummary({});
 bool ValueInfo::isDSOLocal() const {
   // Need to check all summaries are local in case of hash collisions.
   return getSummaryList().size() &&
@@ -80,6 +84,26 @@ bool ModuleSummaryIndex::isGUIDLive(GlobalValue::GUID GUID) const {
   return false;
 }
 
+// TODO: write a graphviz dumper for SCCs (see ModuleSummaryIndex::exportToDot)
+// then delete this function and update its tests
+LLVM_DUMP_METHOD
+void ModuleSummaryIndex::dumpSCCs(raw_ostream &O) {
+  for (scc_iterator<ModuleSummaryIndex *> I =
+           scc_begin<ModuleSummaryIndex *>(this);
+       !I.isAtEnd(); ++I) {
+    O << "SCC (" << utostr(I->size()) << " node" << (I->size() == 1 ? "" : "s")
+      << ") {\n";
+    for (const ValueInfo V : *I) {
+      FunctionSummary *F = nullptr;
+      if (V.getSummaryList().size())
+        F = cast<FunctionSummary>(V.getSummaryList().front().get());
+      O << " " << (F == nullptr ? "External" : "") << " " << utostr(V.getGUID())
+        << (I.hasLoop() ? " (has loop)" : "") << "\n";
+    }
+    O << "}\n";
+  }
+}
+
 namespace {
 struct Attributes {
   void add(const Twine &Name, const Twine &Value,

lib/LTO/LTO.cpp

@@ -50,6 +50,10 @@ using namespace object;
 
 #define DEBUG_TYPE "lto"
 
+static cl::opt<bool>
+    DumpThinCGSCCs("dump-thin-cg-sccs", cl::init(false), cl::Hidden,
+                   cl::desc("Dump the SCCs in the ThinLTO index's callgraph"));
+
 // The values are (type identifier, summary) pairs.
 typedef DenseMap<
     GlobalValue::GUID,
@@ -1141,6 +1145,9 @@ Error LTO::runThinLTO(AddStreamFn AddStream, NativeObjectCache Cache) {
       ThinLTO.ModuleMap.size());
   StringMap<std::map<GlobalValue::GUID, GlobalValue::LinkageTypes>> ResolvedODR;
 
+  if (DumpThinCGSCCs)
+    ThinLTO.CombinedIndex.dumpSCCs(outs());
+
   if (Conf.OptLevel > 0)
     ComputeCrossModuleImport(ThinLTO.CombinedIndex, ModuleToDefinedGVSummaries,
                              ImportLists, ExportLists);

lib/Transforms/IPO/FunctionImport.cpp

@@ -18,8 +18,8 @@
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/Statistic.h"
 #include "llvm/ADT/StringMap.h"
-#include "llvm/ADT/StringSet.h"
 #include "llvm/ADT/StringRef.h"
+#include "llvm/ADT/StringSet.h"
 #include "llvm/Bitcode/BitcodeReader.h"
 #include "llvm/IR/AutoUpgrade.h"
 #include "llvm/IR/Constants.h"

test/ThinLTO/X86/module_summary_graph_traits.ll

@@ -0,0 +1,58 @@
+; RUN: opt -module-summary %s -o %t1.bc
+; RUN: llvm-lto2 run -print-summary-global-ids -dump-thin-cg-sccs %t1.bc -o %t.index.bc \
+; RUN:     -r %t1.bc,external,px -r %t1.bc,l2,pl -r %t1.bc,l1,pl \
+; RUN:     -r %t1.bc,simple,pl -r %t1.bc,root,pl 2>&1 | FileCheck %s
+
+; CHECK: 5224464028922159466{{.*}} is external
+; CHECK: 765152853862302398{{.*}} is l2
+; CHECK: 17000277804057984823{{.*}} is l1
+; CHECK: 15440740835768581517{{.*}} is simple
+; CHECK: 5800840261926955363{{.*}} is root
+
+; CHECK: SCC (2 nodes) {
+; CHECK-NEXT: {{^}} 17000277804057984823 (has loop)
+; CHECK-NEXT: {{^}} 765152853862302398 (has loop)
+; CHECK-NEXT: }
+
+; CHECK: SCC (1 node) {
+; CHECK-NEXT: {{^}} 15440740835768581517{{$}}
+; CHECK-NEXT: }
+
+; CHECK: SCC (1 node) {
+; CHECK-NEXT: External 5224464028922159466{{$}}
+; CHECK-NEXT: }
+
+; CHECK: SCC (1 node) {
+; CHECK-NEXT: {{^}} 5800840261926955363{{$}}
+; CHECK-NEXT: }
+
+; Dummy call graph root that points at all roots of the callgraph.
+; CHECK: SCC (1 node) {
+; CHECK-NEXT: {{^}} 0{{$}}
+; CHECK-NEXT: }
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+declare void @external()
+
+define void @l2() {
+  call void @l1()
+  ret void
+}
+
+define void @l1() {
+  call void @l2()
+  ret void
+}
+
+define i32 @simple() {
+  ret i32 23
+}
+
+define void @root() {
+  call void @l1()
+  call i32 @simple()
+  call void @external()
+  ret void
+}