[Attributor] Propagate Cold attribute inter procedurally

[kuter]

Summary

This patch adds AAHotCold, which propagates the cold attribute.
The Attribute marks a function as cold if all of it's callers are cold.

Other ways of propagating the hot/cold attributes are planned.

[kuter]

small fix.

[jdoerfert]

llvm/include/llvm/Transforms/IPO/Attributor.h

4636

We need more than a boolean state, no? Maybe we should use a 3-bit (or 4-bit) state here instead of an optional in the Impl.

llvm/lib/Transforms/IPO/AttributorAttributes.cpp

9605

let's implement this one

[kuter]

Fix the depgraph.ll

[kuter]

llvm/lib/Transforms/IPO/AttributorAttributes.cpp

9605

I implemented this in D107705

[kuter]

Use a BitState instead of hacky Option<bool>
I will rebase the rest of the patch set.

[jdoerfert]

Lg, one nit.

llvm/test/Transforms/Attributor/ArgumentPromotion/byval.ll

162

see below.

llvm/test/Transforms/Attributor/IPConstantProp/2009-09-24-byval-ptr.ll

300

run the script on this file until the change disappears, non-determinism.

llvm/include/llvm/Transforms/IPO/Attributor.h

struct AAPointerInfo : public AbstractAttribute {
   static bool classof(const AbstractAttribute *AA) {
     return (AA->getIdAddr() == &ID);
   }
 
   /// Unique ID (due to the unique address)
   static const char ID;
 };
 
+struct HotColdState : public BitIntegerState<uint8_t, 3, 0> {
+  enum { INVALID = 0, HOT = 1 << 0, COLD = 1 << 1, UNKNOWN = HOT | COLD };
+  static_assert(UNKNOWN == getBestState(), "Unexpected best state");
+
+  // Is this state assumed to be hot.
+  bool isAssumedHot() const { return Assumed == HOT; }
+
+  // Is this state assumed to be cold.
+  bool isAssumedCold() const { return Assumed == COLD; }
+};
+
+/// This attribute propagates hot/cold annotations.
+struct AAHotCold : public StateWrapper<HotColdState, AbstractAttribute> {
+  using Base = StateWrapper<HotColdState, AbstractAttribute>;
+  AAHotCold(const IRPosition &IRP, Attributor &A) : Base(IRP) {}
+
+  static AAHotCold &createForPosition(const IRPosition &IRP, Attributor &A);
+
+  /// See AbstractAttribute::getName()
+  const std::string getName() const override { return "AAHotCold"; }
+
+  /// See AbstractAttribute::getIdAddr()
+  const char *getIdAddr() const override { return &ID; }
+
+  /// This function should return true if the type of the \p AA is AAHotCold.
+  static bool classof(const AbstractAttribute *AA) {
+    return (AA->getIdAddr() == &ID);
+  }
+
+  /// Unique ID (due to the unique address)
+  static const char ID;
+};
+
 raw_ostream &operator<<(raw_ostream &, const AAPointerInfo::Access &);
 
 /// Run options, used by the pass manager.
 enum AttributorRunOption {
   NONE = 0,
   MODULE = 1 << 0,
   CGSCC = 1 << 1,
   ALL = MODULE | CGSCC
 };
 
 } // end namespace llvm
 
 #endif // LLVM_TRANSFORMS_IPO_ATTRIBUTOR_H

llvm/lib/Transforms/IPO/Attributor.cpp

void Attributor::identifyDefaultAbstractAttributes(Function &F) {
 
   // Every function can be "readnone/argmemonly/inaccessiblememonly/...".
   getOrCreateAAFor<AAMemoryLocation>(FPos);
 
   // Every function might be applicable for Heap-To-Stack conversion.
   if (EnableHeapToStack)
     getOrCreateAAFor<AAHeapToStack>(FPos);
 
+  // Every function can be "hot/cold"
+  getOrCreateAAFor<AAHotCold>(FPos);
+
   // Return attributes are only appropriate if the return type is non void.
   Type *ReturnType = F.getReturnType();
   if (!ReturnType->isVoidTy()) {
     // Argument attribute "returned" --- Create only one per function even
     // though it is an argument attribute.
     getOrCreateAAFor<AAReturnedValues>(FPos);
 
     IRPosition RetPos = IRPosition::returned(F);

llvm/lib/Transforms/IPO/AttributorAttributes.cpp

private:
   /// Set of functions that are unreachable, but might become reachable.
   SmallPtrSet<Function *, 8> UnreachableQueries;
 
   /// If we can reach a function with a call to a unknown function we assume
   /// that we can reach any function.
   bool CanReachUnknownCallee = false;
 };
 
+struct AAHotColdFunction : public AAHotCold {
+  AAHotColdFunction(const IRPosition &IRP, Attributor &A) : AAHotCold(IRP, A) {}
+
+  /// See AbstractAttribute::initialize(...).
+  void initialize(Attributor &A) override {
+    if (getIRPosition().hasAttr({Attribute::Cold})) {
+      addKnownBits(HotColdState::COLD);
+      indicatePessimisticFixpoint();
+      return;
+    }
+
+    if (getIRPosition().hasAttr({Attribute::Hot})) {
+      addKnownBits(HotColdState::HOT);
+      indicatePessimisticFixpoint();
+      return;
+    }
+  }
+
+  /// See AbstractAttribute::updateImpl(...).
+  ChangeStatus updateImpl(Attributor &A) override {
+    // We will use this to determine change.
+    HotColdState OldState = getState();
+
+    // If all of the callers of a function are cold, we can assume that the
+    // function is cold too. But we can't assume it is hot.
+    // TODO: Propagate accross calls inside the same block.
+    auto CallSiteCheck = [&](AbstractCallSite ACS) {
+      Function *Caller = ACS.getInstruction()->getFunction();
+      const AAHotCold &AA = A.getAAFor<AAHotCold>(
+          *this, IRPosition::function(*Caller), DepClassTy::REQUIRED);
+      *this &= AA.getState();
+      return true;
+    };
+
+    // If we don't know all call sites it is not possible for us to know
+    // if this function is hot or cold.
+    bool AllCallSitesKnown;
+    if (!A.checkForAllCallSites(CallSiteCheck, *this, true, AllCallSitesKnown))
+      indicatePessimisticFixpoint();
+
+    return OldState == getState() ? ChangeStatus::UNCHANGED
+                                  : ChangeStatus::CHANGED;
+  }
+
+  ChangeStatus manifest(Attributor &A) override {
+    if (getAssumed() == HotColdState::INVALID ||
+        getAssumed() == HotColdState::UNKNOWN)
+      return ChangeStatus::UNCHANGED;
+
+    LLVMContext &Ctx = getAssociatedFunction()->getContext();
+
+    const auto Attrs = {
+        Attribute::get(Ctx, isAssumedHot() ? Attribute::Hot : Attribute::Cold)};
+    return IRAttributeManifest::manifestAttrs(A, this->getIRPosition(), Attrs);
+  }
+
+  const std::string getAsStr() const override {
+    std::string State;
+
+    switch (getAssumed()) {
+    case HotColdState::UNKNOWN:
+      State = "Unknown";
+      break;
+    case HotColdState::INVALID:
+      State = "Invalid";
+      break;
+    case HotColdState::HOT:
+      State = "Hot";
+      break;
+    case HotColdState::COLD:
+      State = "Cold";
+      break;
+    }
+
+    return "HotCold[" + State + "]";
+  }
+
+  void trackStatistics() const override {}
+};
+
 } // namespace
 
 AACallGraphNode *AACallEdgeIterator::operator*() const {
   return static_cast<AACallGraphNode *>(const_cast<AACallEdges *>(
       &A.getOrCreateAAFor<AACallEdges>(IRPosition::function(**I))));
 }
 
 void AttributorCallGraph::print() { llvm::WriteGraph(outs(), this); }

 const char AAMemoryBehavior::ID = 0;
 const char AAMemoryLocation::ID = 0;
 const char AAValueConstantRange::ID = 0;
 const char AAPotentialValues::ID = 0;
 const char AANoUndef::ID = 0;
 const char AACallEdges::ID = 0;
 const char AAFunctionReachability::ID = 0;
 const char AAPointerInfo::ID = 0;
+const char AAHotCold::ID = 0;
 
 // Macro magic to create the static generator function for attributes that
 // follow the naming scheme.
 
 #define SWITCH_PK_INV(CLASS, PK, POS_NAME)                                     \
   case IRPosition::PK:                                                         \
     llvm_unreachable("Cannot create " #CLASS " for a " POS_NAME " position!");
 

 CREATE_ALL_ABSTRACT_ATTRIBUTE_FOR_POSITION(AAIsDead)
 CREATE_ALL_ABSTRACT_ATTRIBUTE_FOR_POSITION(AANoFree)
 
 CREATE_FUNCTION_ONLY_ABSTRACT_ATTRIBUTE_FOR_POSITION(AAHeapToStack)
 CREATE_FUNCTION_ONLY_ABSTRACT_ATTRIBUTE_FOR_POSITION(AAReachability)
 CREATE_FUNCTION_ONLY_ABSTRACT_ATTRIBUTE_FOR_POSITION(AAUndefinedBehavior)
 CREATE_FUNCTION_ONLY_ABSTRACT_ATTRIBUTE_FOR_POSITION(AACallEdges)
 CREATE_FUNCTION_ONLY_ABSTRACT_ATTRIBUTE_FOR_POSITION(AAFunctionReachability)
+CREATE_FUNCTION_ONLY_ABSTRACT_ATTRIBUTE_FOR_POSITION(AAHotCold)
 
 CREATE_NON_RET_ABSTRACT_ATTRIBUTE_FOR_POSITION(AAMemoryBehavior)
 
 #undef CREATE_FUNCTION_ONLY_ABSTRACT_ATTRIBUTE_FOR_POSITION
 #undef CREATE_FUNCTION_ABSTRACT_ATTRIBUTE_FOR_POSITION
 #undef CREATE_NON_RET_ABSTRACT_ATTRIBUTE_FOR_POSITION
 #undef CREATE_VALUE_ABSTRACT_ATTRIBUTE_FOR_POSITION
 #undef CREATE_ALL_ABSTRACT_ATTRIBUTE_FOR_POSITION
 #undef SWITCH_PK_CREATE
 #undef SWITCH_PK_INV

llvm/test/Transforms/Attributor/ArgumentPromotion/byval.ll

 ; IS__TUNIT_NPM: Function Attrs: nofree nosync nounwind readnone willreturn
 ; IS__TUNIT_NPM-LABEL: define {{[^@]+}}@main
 ; IS__TUNIT_NPM-SAME: () #[[ATTR0]] {
 ; IS__TUNIT_NPM-NEXT:  entry:
 ; IS__TUNIT_NPM-NEXT:    [[S:%.*]] = alloca [[STRUCT_SS:%.*]], align 4
 ; IS__TUNIT_NPM-NEXT:    [[TMP1:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 0
 ; IS__TUNIT_NPM-NEXT:    store i32 1, i32* [[TMP1]], align 8
 ; IS__TUNIT_NPM-NEXT:    [[TMP4:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 1
-; IS__TUNIT_NPM-NEXT:    [[S_CAST:%.*]] = bitcast %struct.ss* [[S]] to i32*
-; IS__TUNIT_NPM-NEXT:    [[TMP0:%.*]] = load i32, i32* [[S_CAST]], align 8
-; IS__TUNIT_NPM-NEXT:    [[S_0_1:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 1
-; IS__TUNIT_NPM-NEXT:    [[TMP1:%.*]] = load i64, i64* [[S_0_1]], align 8
-; IS__TUNIT_NPM-NEXT:    [[C0:%.*]] = call i32 @f(i32 [[TMP0]], i64 [[TMP1]]) #[[ATTR0]]
 ; IS__TUNIT_NPM-NEXT:    [[S_CAST1:%.*]] = bitcast %struct.ss* [[S]] to i32*
-; IS__TUNIT_NPM-NEXT:    [[TMP2:%.*]] = load i32, i32* [[S_CAST1]], align 32
+; IS__TUNIT_NPM-NEXT:    [[TMP0:%.*]] = load i32, i32* [[S_CAST1]], align 8
 ; IS__TUNIT_NPM-NEXT:    [[S_0_12:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 1
-; IS__TUNIT_NPM-NEXT:    [[TMP3:%.*]] = load i64, i64* [[S_0_12]], align 32
+; IS__TUNIT_NPM-NEXT:    [[TMP1:%.*]] = load i64, i64* [[S_0_12]], align 8
+; IS__TUNIT_NPM-NEXT:    [[C0:%.*]] = call i32 @f(i32 [[TMP0]], i64 [[TMP1]]) #[[ATTR0]]
+; IS__TUNIT_NPM-NEXT:    [[S_CAST:%.*]] = bitcast %struct.ss* [[S]] to i32*
+; IS__TUNIT_NPM-NEXT:    [[TMP2:%.*]] = load i32, i32* [[S_CAST]], align 32
+; IS__TUNIT_NPM-NEXT:    [[S_0_1:%.*]] = getelementptr [[STRUCT_SS]], %struct.ss* [[S]], i32 0, i32 1
+; IS__TUNIT_NPM-NEXT:    [[TMP3:%.*]] = load i64, i64* [[S_0_1]], align 32
 ; IS__TUNIT_NPM-NEXT:    [[C1:%.*]] = call i32 @g(i32 [[TMP2]], i64 [[TMP3]]) #[[ATTR0]]
 ; IS__TUNIT_NPM-NEXT:    [[A:%.*]] = add i32 [[C0]], [[C1]]
 ; IS__TUNIT_NPM-NEXT:    ret i32 [[A]]
 ;
 ; IS__CGSCC_OPM: Function Attrs: nofree norecurse nosync nounwind readnone willreturn
 ; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@main
 ; IS__CGSCC_OPM-SAME: () #[[ATTR0]] {
 ; IS__CGSCC_OPM-NEXT:  entry:
 ; IS__CGSCC_OPM-NEXT:    [[S:%.*]] = alloca [[STRUCT_SS:%.*]], align 4

llvm/test/Transforms/Attributor/IPConstantProp/2009-09-24-byval-ptr.ll

 ; IS__TUNIT_OPM-NEXT:    call void @vfu1(%struct.MYstr* nocapture nofree noundef nonnull readonly byval([[STRUCT_MYSTR:%.*]]) align 8 dereferenceable(8) @mystr) #[[ATTR0]]
 ; IS__TUNIT_OPM-NEXT:    [[RESULT:%.*]] = call i32 @vfu2(%struct.MYstr* nocapture nofree noundef nonnull readonly byval([[STRUCT_MYSTR]]) align 8 dereferenceable(8) @mystr) #[[ATTR2:[0-9]+]]
 ; IS__TUNIT_OPM-NEXT:    ret i32 [[RESULT]]
 ;
 ; IS__TUNIT_NPM: Function Attrs: nounwind
 ; IS__TUNIT_NPM-LABEL: define {{[^@]+}}@unions
 ; IS__TUNIT_NPM-SAME: () #[[ATTR0]] {
 ; IS__TUNIT_NPM-NEXT:  entry:
-; IS__TUNIT_NPM-NEXT:    [[MYSTR_CAST:%.*]] = bitcast %struct.MYstr* @mystr to i8*
-; IS__TUNIT_NPM-NEXT:    [[TMP0:%.*]] = load i8, i8* [[MYSTR_CAST]], align 8
-; IS__TUNIT_NPM-NEXT:    [[MYSTR_0_1:%.*]] = getelementptr [[STRUCT_MYSTR:%.*]], %struct.MYstr* @mystr, i32 0, i32 1
-; IS__TUNIT_NPM-NEXT:    [[TMP1:%.*]] = load i32, i32* [[MYSTR_0_1]], align 8
-; IS__TUNIT_NPM-NEXT:    call void @vfu1(i8 [[TMP0]], i32 [[TMP1]]) #[[ATTR0]]
 ; IS__TUNIT_NPM-NEXT:    [[MYSTR_CAST1:%.*]] = bitcast %struct.MYstr* @mystr to i8*
-; IS__TUNIT_NPM-NEXT:    [[TMP2:%.*]] = load i8, i8* [[MYSTR_CAST1]], align 8
-; IS__TUNIT_NPM-NEXT:    [[MYSTR_0_12:%.*]] = getelementptr [[STRUCT_MYSTR]], %struct.MYstr* @mystr, i32 0, i32 1
-; IS__TUNIT_NPM-NEXT:    [[TMP3:%.*]] = load i32, i32* [[MYSTR_0_12]], align 8
+; IS__TUNIT_NPM-NEXT:    [[TMP0:%.*]] = load i8, i8* [[MYSTR_CAST1]], align 8
+; IS__TUNIT_NPM-NEXT:    [[MYSTR_0_12:%.*]] = getelementptr [[STRUCT_MYSTR:%.*]], %struct.MYstr* @mystr, i32 0, i32 1
+; IS__TUNIT_NPM-NEXT:    [[TMP1:%.*]] = load i32, i32* [[MYSTR_0_12]], align 8
+; IS__TUNIT_NPM-NEXT:    call void @vfu1(i8 [[TMP0]], i32 [[TMP1]]) #[[ATTR0]]
+; IS__TUNIT_NPM-NEXT:    [[MYSTR_CAST:%.*]] = bitcast %struct.MYstr* @mystr to i8*
+; IS__TUNIT_NPM-NEXT:    [[TMP2:%.*]] = load i8, i8* [[MYSTR_CAST]], align 8
+; IS__TUNIT_NPM-NEXT:    [[MYSTR_0_1:%.*]] = getelementptr [[STRUCT_MYSTR]], %struct.MYstr* @mystr, i32 0, i32 1
+; IS__TUNIT_NPM-NEXT:    [[TMP3:%.*]] = load i32, i32* [[MYSTR_0_1]], align 8
 ; IS__TUNIT_NPM-NEXT:    [[RESULT:%.*]] = call i32 @vfu2(i8 [[TMP2]], i32 [[TMP3]]) #[[ATTR2:[0-9]+]]
 ; IS__TUNIT_NPM-NEXT:    ret i32 [[RESULT]]
 ;
 ; IS__CGSCC_OPM: Function Attrs: nounwind
 ; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@unions
 ; IS__CGSCC_OPM-SAME: () #[[ATTR0]] {
 ; IS__CGSCC_OPM-NEXT:  entry:
 ; IS__CGSCC_OPM-NEXT:    call void @vfu1(%struct.MYstr* noalias nocapture nofree noundef nonnull readonly byval([[STRUCT_MYSTR:%.*]]) align 8 dereferenceable(8) @mystr) #[[ATTR0]]

 ; IS__TUNIT_OPM-NEXT:    call void @vfu1(%struct.MYstr* nocapture nofree noundef nonnull readonly byval([[STRUCT_MYSTR:%.*]]) align 8 dereferenceable(8) @mystr) #[[ATTR0]]
 ; IS__TUNIT_OPM-NEXT:    [[RESULT:%.*]] = call i32 @vfu2_v2(%struct.MYstr* nocapture nofree noundef nonnull readonly byval([[STRUCT_MYSTR]]) align 8 dereferenceable(8) @mystr) #[[ATTR2]]
 ; IS__TUNIT_OPM-NEXT:    ret i32 [[RESULT]]
 ;
 ; IS__TUNIT_NPM: Function Attrs: nounwind
 ; IS__TUNIT_NPM-LABEL: define {{[^@]+}}@unions_v2
 ; IS__TUNIT_NPM-SAME: () #[[ATTR0]] {
 ; IS__TUNIT_NPM-NEXT:  entry:
-; IS__TUNIT_NPM-NEXT:    [[MYSTR_CAST:%.*]] = bitcast %struct.MYstr* @mystr to i8*
-; IS__TUNIT_NPM-NEXT:    [[TMP0:%.*]] = load i8, i8* [[MYSTR_CAST]], align 8
-; IS__TUNIT_NPM-NEXT:    [[MYSTR_0_1:%.*]] = getelementptr [[STRUCT_MYSTR:%.*]], %struct.MYstr* @mystr, i32 0, i32 1
-; IS__TUNIT_NPM-NEXT:    [[TMP1:%.*]] = load i32, i32* [[MYSTR_0_1]], align 8
-; IS__TUNIT_NPM-NEXT:    call void @vfu1(i8 [[TMP0]], i32 [[TMP1]]) #[[ATTR0]]
 ; IS__TUNIT_NPM-NEXT:    [[MYSTR_CAST1:%.*]] = bitcast %struct.MYstr* @mystr to i8*
-; IS__TUNIT_NPM-NEXT:    [[TMP2:%.*]] = load i8, i8* [[MYSTR_CAST1]], align 8
-; IS__TUNIT_NPM-NEXT:    [[MYSTR_0_12:%.*]] = getelementptr [[STRUCT_MYSTR]], %struct.MYstr* @mystr, i32 0, i32 1
-; IS__TUNIT_NPM-NEXT:    [[TMP3:%.*]] = load i32, i32* [[MYSTR_0_12]], align 8
+; IS__TUNIT_NPM-NEXT:    [[TMP0:%.*]] = load i8, i8* [[MYSTR_CAST1]], align 8
+; IS__TUNIT_NPM-NEXT:    [[MYSTR_0_12:%.*]] = getelementptr [[STRUCT_MYSTR:%.*]], %struct.MYstr* @mystr, i32 0, i32 1
+; IS__TUNIT_NPM-NEXT:    [[TMP1:%.*]] = load i32, i32* [[MYSTR_0_12]], align 8
+; IS__TUNIT_NPM-NEXT:    call void @vfu1(i8 [[TMP0]], i32 [[TMP1]]) #[[ATTR0]]
+; IS__TUNIT_NPM-NEXT:    [[MYSTR_CAST:%.*]] = bitcast %struct.MYstr* @mystr to i8*
+; IS__TUNIT_NPM-NEXT:    [[TMP2:%.*]] = load i8, i8* [[MYSTR_CAST]], align 8
+; IS__TUNIT_NPM-NEXT:    [[MYSTR_0_1:%.*]] = getelementptr [[STRUCT_MYSTR]], %struct.MYstr* @mystr, i32 0, i32 1
+; IS__TUNIT_NPM-NEXT:    [[TMP3:%.*]] = load i32, i32* [[MYSTR_0_1]], align 8
 ; IS__TUNIT_NPM-NEXT:    [[RESULT:%.*]] = call i32 @vfu2_v2(i8 [[TMP2]], i32 [[TMP3]]) #[[ATTR2]]
 ; IS__TUNIT_NPM-NEXT:    ret i32 [[RESULT]]
 ;
 ; IS__CGSCC_OPM: Function Attrs: nounwind
 ; IS__CGSCC_OPM-LABEL: define {{[^@]+}}@unions_v2
 ; IS__CGSCC_OPM-SAME: () #[[ATTR0]] {
 ; IS__CGSCC_OPM-NEXT:  entry:
 ; IS__CGSCC_OPM-NEXT:    call void @vfu1(%struct.MYstr* noalias nocapture nofree noundef nonnull readonly byval([[STRUCT_MYSTR:%.*]]) align 8 dereferenceable(8) @mystr) #[[ATTR0]]

llvm/test/Transforms/Attributor/depgraph.ll

 ; GRAPH-NEXT:   updates [AANoCapture] for CtxI '  %2 = load i32, i32* %0, align 4' at position {arg: [@0]} with state assumed not-captured-maybe-returned
 ; GRAPH-EMPTY:
 ; GRAPH-NEXT: [AAMemoryLocation] for CtxI '  %2 = load i32, i32* %0, align 4' at position {fn:checkAndAdvance [checkAndAdvance@-1]} with state memory:argument
 ; GRAPH-NEXT:   updates [AAMemoryLocation] for CtxI '  %6 = call i32* @checkAndAdvance(i32* %5)' at position {cs: [@-1]} with state memory:argument
 ; GRAPH-NEXT:   updates [AAMemoryLocation] for CtxI '  %6 = call i32* @checkAndAdvance(i32* %5)' at position {cs: [@-1]} with state memory:argument
 ; GRAPH-EMPTY:
 ; GRAPH-NEXT: [AAHeapToStack] for CtxI '  %2 = load i32, i32* %0, align 4' at position {fn:checkAndAdvance [checkAndAdvance@-1]} with state [H2S] Mallocs Good/Bad: 0/0
 ; GRAPH-EMPTY:
+; GRAPH-NEXT: [AAHotCold] for CtxI ' %2 = load i32, i32* %0, align 4' at position {fn:checkAndAdvance [checkAndAdvance@-1]} with state HotCold[Unknown]
+; GRAPH-EMPTY:
 ; GRAPH-NEXT: [AAValueSimplify] for CtxI '  %2 = load i32, i32* %0, align 4' at position {fn_ret:checkAndAdvance [checkAndAdvance@-1]} with state not-simple
 ; GRAPH-EMPTY:
 ; GRAPH-NEXT: [AAAlign] for CtxI '  %2 = load i32, i32* %0, align 4' at position {fn_ret:checkAndAdvance [checkAndAdvance@-1]} with state align<1-16>
 ; GRAPH-NEXT:   updates [AAAlign] for CtxI '  %6 = call i32* @checkAndAdvance(i32* %5)' at position {cs_ret: [@-1]} with state align<1-16>
 ; GRAPH-NEXT:   updates [AAAlign] for CtxI '  %6 = call i32* @checkAndAdvance(i32* %5)' at position {cs_ret: [@-1]} with state align<1-16>
 ; GRAPH-EMPTY:
 ; GRAPH-NEXT: [AANonNull] for CtxI '  %2 = load i32, i32* %0, align 4' at position {fn_ret:checkAndAdvance [checkAndAdvance@-1]} with state nonnull
 ; GRAPH-NEXT:   updates [AANonNull] for CtxI '  %6 = call i32* @checkAndAdvance(i32* %5)' at position {cs_ret: [@-1]} with state nonnull

llvm/test/Transforms/Attributor/hotcold.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-attributes --check-globals
+; RUN: opt -attributor -enable-new-pm=0 -attributor-manifest-internal -S < %s | FileCheck %s --check-prefixes=CHECK,NOT_CGSCC_NPM,NOT_CGSCC_OPM,NOT_TUNIT_NPM,IS__TUNIT____,IS________OPM,IS__TUNIT_OPM
+; RUN: opt -aa-pipeline=basic-aa -passes=attributor -attributor-manifest-internal -S < %s | FileCheck %s --check-prefixes=CHECK,NOT_CGSCC_OPM,NOT_CGSCC_NPM,NOT_TUNIT_OPM,IS__TUNIT____,IS________NPM,IS__TUNIT_NPM
+; RUN: opt -attributor-cgscc -enable-new-pm=0 -attributor-manifest-internal -S < %s | FileCheck %s --check-prefixes=CHECK,NOT_TUNIT_NPM,NOT_TUNIT_OPM,NOT_CGSCC_NPM,IS__CGSCC____,IS________OPM,IS__CGSCC_OPM
+; RUN: opt -aa-pipeline=basic-aa -passes=attributor-cgscc -attributor-manifest-internal -S < %s | FileCheck %s --check-prefixes=CHECK,NOT_TUNIT_NPM,NOT_TUNIT_OPM,NOT_CGSCC_OPM,IS__CGSCC____,IS________NPM,IS__CGSCC_NPM
+
+@G = global i32 0, align 4
+
+;.
+; CHECK: @[[G:[a-zA-Z0-9_$"\\.-]+]] = global i32 0, align 4
+;.
+define internal void @test() #0 {
+; IS__TUNIT____: Function Attrs: cold nofree nosync nounwind willreturn writeonly
+; IS__TUNIT____-LABEL: define {{[^@]+}}@test
+; IS__TUNIT____-SAME: () #[[ATTR0:[0-9]+]] {
+; IS__TUNIT____-NEXT:    store i32 1, i32* @G, align 4
+; IS__TUNIT____-NEXT:    ret void
+;
+; IS__CGSCC____: Function Attrs: cold nofree norecurse nosync nounwind willreturn writeonly
+; IS__CGSCC____-LABEL: define {{[^@]+}}@test
+; IS__CGSCC____-SAME: () #[[ATTR0:[0-9]+]] {
+; IS__CGSCC____-NEXT:    store i32 1, i32* @G, align 4
+; IS__CGSCC____-NEXT:    ret void
+;
+  store i32 1, i32* @G, align 4
+  ret void
+}
+
+define internal void @cold_caller1() #1 {
+; IS__TUNIT____: Function Attrs: cold nofree nosync nounwind willreturn writeonly
+; IS__TUNIT____-LABEL: define {{[^@]+}}@cold_caller1
+; IS__TUNIT____-SAME: () #[[ATTR0]] {
+; IS__TUNIT____-NEXT:    call void @test() #[[ATTR1:[0-9]+]]
+; IS__TUNIT____-NEXT:    ret void
+;
+; IS__CGSCC____: Function Attrs: cold nofree norecurse nosync nounwind willreturn writeonly
+; IS__CGSCC____-LABEL: define {{[^@]+}}@cold_caller1
+; IS__CGSCC____-SAME: () #[[ATTR0]] {
+; IS__CGSCC____-NEXT:    call void @test() #[[ATTR2:[0-9]+]]
+; IS__CGSCC____-NEXT:    ret void
+;
+  call void @test()
+  ret void
+}
+
+define internal void @cold_caller2() #1 {
+; IS__TUNIT____: Function Attrs: cold nofree nosync nounwind willreturn writeonly
+; IS__TUNIT____-LABEL: define {{[^@]+}}@cold_caller2
+; IS__TUNIT____-SAME: () #[[ATTR0]] {
+; IS__TUNIT____-NEXT:    call void @test() #[[ATTR1]]
+; IS__TUNIT____-NEXT:    ret void
+;
+; IS__CGSCC____: Function Attrs: cold nofree norecurse nosync nounwind willreturn writeonly
+; IS__CGSCC____-LABEL: define {{[^@]+}}@cold_caller2
+; IS__CGSCC____-SAME: () #[[ATTR0]] {
+; IS__CGSCC____-NEXT:    call void @test() #[[ATTR2]]
+; IS__CGSCC____-NEXT:    ret void
+;
+  call void @test()
+  ret void
+}
+
+define void @external() {
+; IS__TUNIT____: Function Attrs: nofree nosync nounwind willreturn writeonly
+; IS__TUNIT____-LABEL: define {{[^@]+}}@external
+; IS__TUNIT____-SAME: () #[[ATTR1]] {
+; IS__TUNIT____-NEXT:    call void @cold_caller2() #[[ATTR1]]
+; IS__TUNIT____-NEXT:    call void @cold_caller1() #[[ATTR1]]
+; IS__TUNIT____-NEXT:    ret void
+;
+; IS__CGSCC____: Function Attrs: nofree norecurse nosync nounwind willreturn writeonly
+; IS__CGSCC____-LABEL: define {{[^@]+}}@external
+; IS__CGSCC____-SAME: () #[[ATTR1:[0-9]+]] {
+; IS__CGSCC____-NEXT:    call void @cold_caller2() #[[ATTR2]]
+; IS__CGSCC____-NEXT:    call void @cold_caller1() #[[ATTR2]]
+; IS__CGSCC____-NEXT:    ret void
+;
+  call void @cold_caller2()
+  call void @cold_caller1()
+  ret void
+}
+
+attributes #1 = { cold }
+;.
+; IS__TUNIT____: attributes #[[ATTR0]] = { cold nofree nosync nounwind willreturn writeonly }
+; IS__TUNIT____: attributes #[[ATTR1]] = { nofree nosync nounwind willreturn writeonly }
+;.
+; IS__CGSCC____: attributes #[[ATTR0]] = { cold nofree norecurse nosync nounwind willreturn writeonly }
+; IS__CGSCC____: attributes #[[ATTR1]] = { nofree norecurse nosync nounwind willreturn writeonly }
+; IS__CGSCC____: attributes #[[ATTR2]] = { nounwind willreturn writeonly }
+;.