[Attributor] Avoid the type check in getCalledFunction

We now consistently use `CallBase::getCalledOperand` rather than
`getCalledFunction`, as we do not want the type checked performed by the
latter. This exposed various missing checks to handle mismatches
properly, but it is good to have them explicit now.

In a follow up we might want to flag certain calls as UB, but for now,
we allow everything to cut down on unexpected differences.

diff --git a/llvm/include/llvm/Transforms/IPO/Attributor.h b/llvm/include/llvm/Transforms/IPO/Attributor.h
index babafdf..49fa9f8 100644 (file)
--- a/llvm/include/llvm/Transforms/IPO/Attributor.h
+++ b/llvm/include/llvm/Transforms/IPO/Attributor.h
@@ -708,7 +708,8 @@ struct IRPosition {
       // function.
       if (Argument *Arg = getAssociatedArgument())
         return Arg->getParent();
-      return CB->getCalledFunction();
+      return dyn_cast_if_present<Function>(
+          CB->getCalledOperand()->stripPointerCasts());
     }
     return getAnchorScope();
   }
@@ -1257,7 +1258,8 @@ struct InformationCache {
 
       for (Instruction &I : instructions(*F))
         if (auto *CB = dyn_cast<CallBase>(&I))
-          if (Function *Callee = CB->getCalledFunction())
+          if (auto *Callee =
+                  dyn_cast_if_present<Function>(CB->getCalledOperand()))
             if (Seen.insert(Callee).second)
               Worklist.push_back(Callee);
     }

diff --git a/llvm/lib/Transforms/IPO/Attributor.cpp b/llvm/lib/Transforms/IPO/Attributor.cpp
index 1c0108d..24efb37 100644 (file)
--- a/llvm/lib/Transforms/IPO/Attributor.cpp
+++ b/llvm/lib/Transforms/IPO/Attributor.cpp
@@ -980,7 +980,7 @@ Argument *IRPosition::getAssociatedArgument() const {
 
   // If no callbacks were found, or none used the underlying call site operand
   // exclusively, use the direct callee argument if available.
-  const Function *Callee = CB.getCalledFunction();
+  auto *Callee = dyn_cast_if_present<Function>(CB.getCalledOperand());
   if (Callee && Callee->arg_size() > unsigned(ArgNo))
     return Callee->getArg(ArgNo);
 
@@ -1090,7 +1090,7 @@ SubsumingPositionIterator::SubsumingPositionIterator(const IRPosition &IRP) {
     // TODO: We need to look at the operand bundles similar to the redirection
     //       in CallBase.
     if (!CB->hasOperandBundles() || CanIgnoreOperandBundles(*CB))
-      if (const Function *Callee = CB->getCalledFunction())
+      if (auto *Callee = dyn_cast_if_present<Function>(CB->getCalledOperand()))
         IRPositions.emplace_back(IRPosition::function(*Callee));
     return;
   case IRPosition::IRP_CALL_SITE_RETURNED:
@@ -1098,7 +1098,8 @@ SubsumingPositionIterator::SubsumingPositionIterator(const IRPosition &IRP) {
     // TODO: We need to look at the operand bundles similar to the redirection
     //       in CallBase.
     if (!CB->hasOperandBundles() || CanIgnoreOperandBundles(*CB)) {
-      if (const Function *Callee = CB->getCalledFunction()) {
+      if (auto *Callee =
+              dyn_cast_if_present<Function>(CB->getCalledOperand())) {
         IRPositions.emplace_back(IRPosition::returned(*Callee));
         IRPositions.emplace_back(IRPosition::function(*Callee));
         for (const Argument &Arg : Callee->args())
@@ -1118,7 +1119,7 @@ SubsumingPositionIterator::SubsumingPositionIterator(const IRPosition &IRP) {
     // TODO: We need to look at the operand bundles similar to the redirection
     //       in CallBase.
     if (!CB->hasOperandBundles() || CanIgnoreOperandBundles(*CB)) {
-      const Function *Callee = CB->getCalledFunction();
+      auto *Callee = dyn_cast_if_present<Function>(CB->getCalledOperand());
       if (Callee) {
         if (Argument *Arg = IRP.getAssociatedArgument())
           IRPositions.emplace_back(IRPosition::argument(*Arg));
@@ -1375,7 +1376,8 @@ std::optional<Value *> Attributor::translateArgumentToCallSiteContent(
   if (*V == nullptr || isa<Constant>(*V))
     return V;
   if (auto *Arg = dyn_cast<Argument>(*V))
-    if (CB.getCalledFunction() == Arg->getParent())
+    if (CB.getCalledOperand() == Arg->getParent() &&
+        CB.arg_size() > Arg->getArgNo())
       if (!Arg->hasPointeeInMemoryValueAttr())
         return getAssumedSimplified(
             IRPosition::callsite_argument(CB, Arg->getArgNo()), AA,
@@ -2314,9 +2316,9 @@ ChangeStatus Attributor::cleanupIR() {
       if (CB->isArgOperand(U)) {
         unsigned Idx = CB->getArgOperandNo(U);
         CB->removeParamAttr(Idx, Attribute::NoUndef);
-        Function *Fn = CB->getCalledFunction();
-        if (Fn && Fn->arg_size() > Idx)
-          Fn->removeParamAttr(Idx, Attribute::NoUndef);
+        auto *Callee = dyn_cast_if_present<Function>(CB->getCalledOperand());
+        if (Callee && Callee->arg_size() > Idx)
+          Callee->removeParamAttr(Idx, Attribute::NoUndef);
       }
     }
     if (isa<Constant>(NewV) && isa<BranchInst>(U->getUser())) {
@@ -2715,7 +2717,10 @@ bool Attributor::isValidFunctionSignatureRewrite(
         ACS.getInstruction()->getType() !=
             ACS.getCalledFunction()->getReturnType())
       return false;
-    if (ACS.getCalledOperand()->getType() != Fn->getType())
+    if (cast<CallBase>(ACS.getInstruction())->getCalledOperand()->getType() !=
+        Fn->getType())
+      return false;
+    if (ACS.getNumArgOperands() != Fn->arg_size())
       return false;
     // Forbid must-tail calls for now.
     return !ACS.isCallbackCall() && !ACS.getInstruction()->isMustTailCall();
@@ -2741,7 +2746,8 @@ bool Attributor::isValidFunctionSignatureRewrite(
   // Avoid callbacks for now.
   bool UsedAssumedInformation = false;
   if (!checkForAllCallSites(CallSiteCanBeChanged, *Fn, true, nullptr,
-                            UsedAssumedInformation)) {
+                            UsedAssumedInformation,
+                            /* CheckPotentiallyDead */ true)) {
     LLVM_DEBUG(dbgs() << "[Attributor] Cannot rewrite all call sites\n");
     return false;
   }
@@ -3084,7 +3090,8 @@ void InformationCache::initializeInformationCache(const Function &CF,
         AddToAssumeUsesMap(*Assume->getArgOperand(0));
       } else if (cast<CallInst>(I).isMustTailCall()) {
         FI.ContainsMustTailCall = true;
-        if (const Function *Callee = cast<CallInst>(I).getCalledFunction())
+        if (auto *Callee = dyn_cast_if_present<Function>(
+                cast<CallInst>(I).getCalledOperand()))
           getFunctionInfo(*Callee).CalledViaMustTail = true;
       }
       [[fallthrough]];
@@ -3314,7 +3321,7 @@ void Attributor::identifyDefaultAbstractAttributes(Function &F) {
     // users. The return value might be dead if there are no live users.
     getOrCreateAAFor<AAIsDead>(CBInstPos);
 
-    Function *Callee = CB.getCalledFunction();
+    Function *Callee = dyn_cast_if_present<Function>(CB.getCalledOperand());
     // TODO: Even if the callee is not known now we might be able to simplify
     //       the call/callee.
     if (!Callee)

diff --git a/llvm/lib/Transforms/IPO/AttributorAttributes.cpp b/llvm/lib/Transforms/IPO/AttributorAttributes.cpp
index 099dc1a..1b3039f 100644 (file)
--- a/llvm/lib/Transforms/IPO/AttributorAttributes.cpp
+++ b/llvm/lib/Transforms/IPO/AttributorAttributes.cpp
@@ -2095,12 +2095,10 @@ public:
     ReturnedValues.clear();
 
     Function *F = getAssociatedFunction();
-    if (!F || F->isDeclaration()) {
+    if (!F || F->isDeclaration() || F->getReturnType()->isVoidTy()) {
       indicatePessimisticFixpoint();
       return;
     }
-    assert(!F->getReturnType()->isVoidTy() &&
-           "Did not expect a void return type!");
 
     // The map from instruction opcodes to those instructions in the function.
     auto &OpcodeInstMap = A.getInfoCache().getOpcodeInstMapForFunction(*F);
@@ -3068,7 +3066,7 @@ struct AANonConvergentFunction final : AANonConvergentImpl {
     // If all function calls are known to not be convergent, we are not convergent.
     auto CalleeIsNotConvergent = [&](Instruction &Inst) {
       CallBase &CB = cast<CallBase>(Inst);
-      Function *Callee = CB.getCalledFunction();
+      auto *Callee = dyn_cast_if_present<Function>(CB.getCalledOperand());
       if (!Callee || Callee->isIntrinsic()) {
         return false;
       }
@@ -3195,7 +3193,7 @@ struct AAUndefinedBehaviorImpl : public AAUndefinedBehavior {
       // Check nonnull and noundef argument attribute violation for each
       // callsite.
       CallBase &CB = cast<CallBase>(I);
-      Function *Callee = CB.getCalledFunction();
+      auto *Callee = dyn_cast_if_present<Function>(CB.getCalledOperand());
       if (!Callee)
         return true;
       for (unsigned idx = 0; idx < CB.arg_size(); idx++) {
@@ -4855,7 +4853,7 @@ struct AAIsDeadFunction : public AAIsDead {
     // functions. It can however cause dead functions to be treated as live.
     for (const Instruction &I : BB)
       if (const auto *CB = dyn_cast<CallBase>(&I))
-        if (const Function *F = CB->getCalledFunction())
+        if (auto *F = dyn_cast_if_present<Function>(CB->getCalledOperand()))
           if (F->hasLocalLinkage())
             A.markLiveInternalFunction(*F);
     return true;
@@ -5819,8 +5817,8 @@ struct AAInstanceInfoImpl : public AAInstanceInfo {
       if (auto *CB = dyn_cast<CallBase>(UserI)) {
         // This check is not guaranteeing uniqueness but for now that we cannot
         // end up with two versions of \p U thinking it was one.
-        if (!CB->getCalledFunction() ||
-            !CB->getCalledFunction()->hasLocalLinkage())
+        auto *Callee = dyn_cast_if_present<Function>(CB->getCalledOperand());
+        if (!Callee || !Callee->hasLocalLinkage())
           return true;
         if (!CB->isArgOperand(&U))
           return false;
@@ -6577,7 +6575,7 @@ struct AAValueSimplifyArgument final : AAValueSimplifyImpl {
     bool Success;
     bool UsedAssumedInformation = false;
     if (hasCallBaseContext() &&
-        getCallBaseContext()->getCalledFunction() == Arg->getParent())
+        getCallBaseContext()->getCalledOperand() == Arg->getParent())
       Success = PredForCallSite(
           AbstractCallSite(&getCallBaseContext()->getCalledOperandUse()));
     else
@@ -7538,7 +7536,9 @@ struct AAPrivatizablePtrArgument final : public AAPrivatizablePtrImpl {
     auto CallSiteCheck = [&](AbstractCallSite ACS) {
       CallBase *CB = ACS.getInstruction();
       return TTI->areTypesABICompatible(
-          CB->getCaller(), CB->getCalledFunction(), ReplacementTypes);
+          CB->getCaller(),
+          dyn_cast_if_present<Function>(CB->getCalledOperand()),
+          ReplacementTypes);
     };
     bool UsedAssumedInformation = false;
     if (!A.checkForAllCallSites(CallSiteCheck, *this, true,
@@ -7620,17 +7620,17 @@ struct AAPrivatizablePtrArgument final : public AAPrivatizablePtrImpl {
       assert(DCArgNo >= 0 && unsigned(DCArgNo) < DC->arg_size() &&
              "Expected a direct call operand for callback call operand");
 
+      Function *DCCallee =
+          dyn_cast_if_present<Function>(DC->getCalledOperand());
       LLVM_DEBUG({
         dbgs() << "[AAPrivatizablePtr] Argument " << *Arg
                << " check if be privatized in the context of its parent ("
                << Arg->getParent()->getName()
                << ")\n[AAPrivatizablePtr] because it is an argument in a "
                   "direct call of ("
-               << DCArgNo << "@" << DC->getCalledFunction()->getName()
-               << ").\n";
+               << DCArgNo << "@" << DCCallee->getName() << ").\n";
       });
 
-      Function *DCCallee = DC->getCalledFunction();
       if (unsigned(DCArgNo) < DCCallee->arg_size()) {
         const auto *DCArgPrivAA = A.getAAFor<AAPrivatizablePtr>(
             *this, IRPosition::argument(*DCCallee->getArg(DCArgNo)),
@@ -7650,7 +7650,7 @@ struct AAPrivatizablePtrArgument final : public AAPrivatizablePtrImpl {
                << Arg->getParent()->getName()
                << ")\n[AAPrivatizablePtr] because it is an argument in a "
                   "direct call of ("
-               << ACS.getInstruction()->getCalledFunction()->getName()
+               << ACS.getInstruction()->getCalledOperand()->getName()
                << ").\n[AAPrivatizablePtr] for which the argument "
                   "privatization is not compatible.\n";
       });
@@ -11481,14 +11481,15 @@ struct AAPotentialValuesFloating : AAPotentialValuesImpl {
       if (V->getType()->isPointerTy()) {
         NewV = AA::getWithType(*V->stripPointerCasts(), *V->getType());
       } else {
-        auto *CB = dyn_cast<CallBase>(V);
-        if (CB && CB->getCalledFunction()) {
-          for (Argument &Arg : CB->getCalledFunction()->args())
-            if (Arg.hasReturnedAttr()) {
-              NewV = CB->getArgOperand(Arg.getArgNo());
-              break;
-            }
-        }
+        if (auto *CB = dyn_cast<CallBase>(V))
+          if (auto *Callee =
+                  dyn_cast_if_present<Function>(CB->getCalledOperand())) {
+            for (Argument &Arg : Callee->args())
+              if (Arg.hasReturnedAttr()) {
+                NewV = CB->getArgOperand(Arg.getArgNo());
+                break;
+              }
+          }
       }
       if (NewV && NewV != V) {
         Worklist.push_back({{*NewV, CtxI}, S});
@@ -11706,7 +11707,7 @@ struct AAPotentialValuesCallSiteReturned : AAPotentialValuesImpl {
           SmallVector<AA::ValueAndContext> ArgValues;
           IRPosition IRP = IRPosition::value(*V);
           if (auto *Arg = dyn_cast<Argument>(V))
-            if (Arg->getParent() == CB->getCalledFunction())
+            if (Arg->getParent() == CB->getCalledOperand())
               IRP = IRPosition::callsite_argument(*CB, Arg->getArgNo());
           if (recurseForValue(A, IRP, AA::AnyScope))
             continue;

diff --git a/llvm/test/Transforms/Attributor/IPConstantProp/arg-count-mismatch.ll b/llvm/test/Transforms/Attributor/IPConstantProp/arg-count-mismatch.ll
index 1333793..9913827 100644 (file)
--- a/llvm/test/Transforms/Attributor/IPConstantProp/arg-count-mismatch.ll
+++ b/llvm/test/Transforms/Attributor/IPConstantProp/arg-count-mismatch.ll
@@ -34,16 +34,15 @@
 ; FIXME we should recognize this as UB and make it an unreachable.
 
 define dso_local i16 @foo(i16 %a) {
-; TUNIT: Function Attrs: norecurse
+; TUNIT: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none)
 ; TUNIT-LABEL: define {{[^@]+}}@foo
 ; TUNIT-SAME: (i16 [[A:%.*]]) #[[ATTR0:[0-9]+]] {
-; TUNIT-NEXT:    [[CALL:%.*]] = call i16 bitcast (i16 (i16, i16)* @bar to i16 (i16)*)(i16 [[A]])
-; TUNIT-NEXT:    ret i16 [[CALL]]
+; TUNIT-NEXT:    ret i16 0
 ;
-; CGSCC: Function Attrs: nosync nounwind
+; CGSCC: Function Attrs: mustprogress nofree nosync nounwind willreturn
 ; CGSCC-LABEL: define {{[^@]+}}@foo
 ; CGSCC-SAME: (i16 [[A:%.*]]) #[[ATTR0:[0-9]+]] {
-; CGSCC-NEXT:    [[CALL:%.*]] = call i16 bitcast (i16 (i16, i16)* @bar to i16 (i16)*)(i16 [[A]])
+; CGSCC-NEXT:    [[CALL:%.*]] = call i16 bitcast (i16 (i16, i16)* @bar to i16 (i16)*)(i16 [[A]]) #[[ATTR2:[0-9]+]]
 ; CGSCC-NEXT:    ret i16 [[CALL]]
 ;
   %call = call i16 bitcast (i16 (i16, i16) * @bar to i16 (i16) *)(i16 %a)
@@ -51,25 +50,25 @@ define dso_local i16 @foo(i16 %a) {
 }
 
 define internal i16 @bar(i16 %p1, i16 %p2) {
-; CHECK: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none)
-; CHECK-LABEL: define {{[^@]+}}@bar
-; CHECK-SAME: (i16 [[P1:%.*]], i16 [[P2:%.*]]) #[[ATTR1:[0-9]+]] {
-; CHECK-NEXT:    ret i16 0
+; CGSCC: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none)
+; CGSCC-LABEL: define {{[^@]+}}@bar
+; CGSCC-SAME: (i16 [[P1:%.*]], i16 [[P2:%.*]]) #[[ATTR1:[0-9]+]] {
+; CGSCC-NEXT:    ret i16 0
 ;
   ret i16 0
 }
 
 define dso_local i16 @foo2(i16 %a) {
-; TUNIT: Function Attrs: norecurse
+; TUNIT: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none)
 ; TUNIT-LABEL: define {{[^@]+}}@foo2
 ; TUNIT-SAME: (i16 [[A:%.*]]) #[[ATTR0]] {
-; TUNIT-NEXT:    [[CALL:%.*]] = call i16 bitcast (i16 (i16, i16)* @bar2 to i16 (i16)*)(i16 [[A]])
+; TUNIT-NEXT:    [[CALL:%.*]] = call i16 bitcast (i16 (i16, i16)* @bar2 to i16 (i16)*)(i16 [[A]]) #[[ATTR1:[0-9]+]]
 ; TUNIT-NEXT:    ret i16 [[CALL]]
 ;
-; CGSCC: Function Attrs: nosync nounwind
+; CGSCC: Function Attrs: mustprogress nofree nosync nounwind willreturn
 ; CGSCC-LABEL: define {{[^@]+}}@foo2
 ; CGSCC-SAME: (i16 [[A:%.*]]) #[[ATTR0]] {
-; CGSCC-NEXT:    [[CALL:%.*]] = call i16 bitcast (i16 (i16, i16)* @bar2 to i16 (i16)*)(i16 [[A]])
+; CGSCC-NEXT:    [[CALL:%.*]] = call i16 bitcast (i16 (i16, i16)* @bar2 to i16 (i16)*)(i16 [[A]]) #[[ATTR2]]
 ; CGSCC-NEXT:    ret i16 [[CALL]]
 ;
   %call = call i16 bitcast (i16 (i16, i16) * @bar2 to i16 (i16) *)(i16 %a)
@@ -77,11 +76,17 @@ define dso_local i16 @foo2(i16 %a) {
 }
 
 define internal i16 @bar2(i16 %p1, i16 %p2) {
-; CHECK: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none)
-; CHECK-LABEL: define {{[^@]+}}@bar2
-; CHECK-SAME: (i16 [[P1:%.*]], i16 [[P2:%.*]]) #[[ATTR1]] {
-; CHECK-NEXT:    [[A:%.*]] = add i16 [[P1]], [[P2]]
-; CHECK-NEXT:    ret i16 [[A]]
+; TUNIT: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none)
+; TUNIT-LABEL: define {{[^@]+}}@bar2
+; TUNIT-SAME: (i16 [[P1:%.*]], i16 [[P2:%.*]]) #[[ATTR0]] {
+; TUNIT-NEXT:    [[A:%.*]] = add i16 [[P1]], [[P2]]
+; TUNIT-NEXT:    ret i16 [[A]]
+;
+; CGSCC: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none)
+; CGSCC-LABEL: define {{[^@]+}}@bar2
+; CGSCC-SAME: (i16 [[P1:%.*]], i16 [[P2:%.*]]) #[[ATTR1]] {
+; CGSCC-NEXT:    [[A:%.*]] = add i16 [[P1]], [[P2]]
+; CGSCC-NEXT:    ret i16 [[A]]
 ;
   %a = add i16 %p1, %p2
   ret i16 %a
@@ -94,18 +99,16 @@ define internal i16 @bar2(i16 %p1, i16 %p2) {
 ; been provided),
 
 define dso_local i16 @vararg_tests(i16 %a) {
-; TUNIT: Function Attrs: norecurse
+; TUNIT: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none)
 ; TUNIT-LABEL: define {{[^@]+}}@vararg_tests
 ; TUNIT-SAME: (i16 [[A:%.*]]) #[[ATTR0]] {
-; TUNIT-NEXT:    [[CALL2:%.*]] = call i16 bitcast (i16 (i16, i16, ...)* @vararg_no_prop to i16 (i16)*)(i16 noundef 7)
-; TUNIT-NEXT:    [[ADD:%.*]] = add i16 7, [[CALL2]]
-; TUNIT-NEXT:    ret i16 [[ADD]]
+; TUNIT-NEXT:    ret i16 14
 ;
-; CGSCC: Function Attrs: nosync nounwind
+; CGSCC: Function Attrs: mustprogress nofree nosync nounwind willreturn
 ; CGSCC-LABEL: define {{[^@]+}}@vararg_tests
 ; CGSCC-SAME: (i16 [[A:%.*]]) #[[ATTR0]] {
 ; CGSCC-NEXT:    [[CALL1:%.*]] = call i16 (i16, ...) @vararg_prop(i16 noundef 7, i16 noundef 8, i16 [[A]])
-; CGSCC-NEXT:    [[CALL2:%.*]] = call i16 bitcast (i16 (i16, i16, ...)* @vararg_no_prop to i16 (i16)*)(i16 7)
+; CGSCC-NEXT:    [[CALL2:%.*]] = call i16 bitcast (i16 (i16, i16, ...)* @vararg_no_prop to i16 (i16)*)(i16 7) #[[ATTR2]]
 ; CGSCC-NEXT:    [[ADD:%.*]] = add i16 [[CALL1]], [[CALL2]]
 ; CGSCC-NEXT:    ret i16 [[ADD]]
 ;
@@ -125,18 +128,21 @@ define internal i16 @vararg_prop(i16 %p1, ...) {
 }
 
 define internal i16 @vararg_no_prop(i16 %p1, i16 %p2, ...) {
-; CHECK: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none)
-; CHECK-LABEL: define {{[^@]+}}@vararg_no_prop
-; CHECK-SAME: (i16 [[P1:%.*]], i16 [[P2:%.*]], ...) #[[ATTR1]] {
-; CHECK-NEXT:    ret i16 7
+; CGSCC: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none)
+; CGSCC-LABEL: define {{[^@]+}}@vararg_no_prop
+; CGSCC-SAME: (i16 [[P1:%.*]], i16 [[P2:%.*]], ...) #[[ATTR1]] {
+; CGSCC-NEXT:    ret i16 7
 ;
   ret i16 %p1
 }
 
 ;.
-; TUNIT: attributes #[[ATTR0]] = { norecurse }
-; TUNIT: attributes #[[ATTR1]] = { mustprogress nofree norecurse nosync nounwind willreturn memory(none) }
+; TUNIT: attributes #[[ATTR0]] = { mustprogress nofree norecurse nosync nounwind willreturn memory(none) }
+; TUNIT: attributes #[[ATTR1]] = { nofree nosync nounwind willreturn memory(none) }
 ;.
-; CGSCC: attributes #[[ATTR0]] = { nosync nounwind }
+; CGSCC: attributes #[[ATTR0]] = { mustprogress nofree nosync nounwind willreturn }
 ; CGSCC: attributes #[[ATTR1]] = { mustprogress nofree norecurse nosync nounwind willreturn memory(none) }
+; CGSCC: attributes #[[ATTR2]] = { nofree nounwind willreturn }
 ;.
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; CHECK: {{.*}}

diff --git a/llvm/test/Transforms/Attributor/IPConstantProp/arg-type-mismatch.ll b/llvm/test/Transforms/Attributor/IPConstantProp/arg-type-mismatch.ll
index 57df274..5babc2f 100644 (file)
--- a/llvm/test/Transforms/Attributor/IPConstantProp/arg-type-mismatch.ll
+++ b/llvm/test/Transforms/Attributor/IPConstantProp/arg-type-mismatch.ll
@@ -6,16 +6,15 @@
 ; argument type between the caller and callee.
 
 define dso_local i16 @foo(i16 %a) {
-; TUNIT: Function Attrs: norecurse
+; TUNIT: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none)
 ; TUNIT-LABEL: define {{[^@]+}}@foo
 ; TUNIT-SAME: (i16 [[A:%.*]]) #[[ATTR0:[0-9]+]] {
-; TUNIT-NEXT:    [[CALL:%.*]] = call i16 @bar(i16 [[A]], i32 7)
-; TUNIT-NEXT:    ret i16 [[CALL]]
+; TUNIT-NEXT:    ret i16 7
 ;
-; CGSCC: Function Attrs: nosync nounwind memory(none)
+; CGSCC: Function Attrs: mustprogress nofree nosync nounwind willreturn memory(none)
 ; CGSCC-LABEL: define {{[^@]+}}@foo
 ; CGSCC-SAME: (i16 [[A:%.*]]) #[[ATTR0:[0-9]+]] {
-; CGSCC-NEXT:    [[CALL:%.*]] = call i16 @bar(i16 [[A]], i32 7)
+; CGSCC-NEXT:    [[CALL:%.*]] = call i16 @bar(i16 [[A]], i32 noundef 7)
 ; CGSCC-NEXT:    ret i16 [[CALL]]
 ;
   %call = call i16 @bar(i16 %a, i32 7)
@@ -23,10 +22,6 @@ define dso_local i16 @foo(i16 %a) {
 }
 
 define internal i16 @bar(i16 %p1, i16 %p2) {
-; TUNIT-LABEL: define {{[^@]+}}@bar
-; TUNIT-SAME: (i16 [[P1:%.*]], i16 [[P2:%.*]]) {
-; TUNIT-NEXT:    ret i16 [[P2]]
-;
 ; CGSCC: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none)
 ; CGSCC-LABEL: define {{[^@]+}}@bar
 ; CGSCC-SAME: (i16 [[P1:%.*]], i16 returned [[P2:%.*]]) #[[ATTR1:[0-9]+]] {
@@ -37,9 +32,9 @@ define internal i16 @bar(i16 %p1, i16 %p2) {
 
 
 ;.
-; TUNIT: attributes #[[ATTR0]] = { norecurse }
+; TUNIT: attributes #[[ATTR0]] = { mustprogress nofree norecurse nosync nounwind willreturn memory(none) }
 ;.
-; CGSCC: attributes #[[ATTR0]] = { nosync nounwind memory(none) }
+; CGSCC: attributes #[[ATTR0]] = { mustprogress nofree nosync nounwind willreturn memory(none) }
 ; CGSCC: attributes #[[ATTR1]] = { mustprogress nofree norecurse nosync nounwind willreturn memory(none) }
 ;.
 ;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:

diff --git a/llvm/test/Transforms/Attributor/liveness.ll b/llvm/test/Transforms/Attributor/liveness.ll
index 9e1ea70..086569d 100644 (file)
--- a/llvm/test/Transforms/Attributor/liveness.ll
+++ b/llvm/test/Transforms/Attributor/liveness.ll
@@ -2427,33 +2427,60 @@ indirectgoto:                                     ; preds = %lab0, %entry
 @e = global %struct.a* null
 
 define i32 @main() {
-; CHECK-LABEL: define {{[^@]+}}@main() {
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[F:%.*]] = alloca i32, align 4
-; CHECK-NEXT:    br label [[FOR_COND_0:%.*]]
-; CHECK:       for.cond.0:
-; CHECK-NEXT:    [[G_0:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[FOR_BODY_0:%.*]] ]
-; CHECK-NEXT:    [[CMP_0:%.*]] = icmp ult i32 [[G_0]], 100
-; CHECK-NEXT:    br i1 [[CMP_0]], label [[FOR_BODY_0]], label [[FOR_END_0:%.*]]
-; CHECK:       for.body.0:
-; CHECK-NEXT:    [[INC]] = add nuw nsw i32 [[G_0]], 1
-; CHECK-NEXT:    br label [[FOR_COND_0]]
-; CHECK:       for.end.0:
-; CHECK-NEXT:    [[CALL:%.*]] = call i8* @malloc(i64 noundef 8)
-; CHECK-NEXT:    store i8* [[CALL]], i8** bitcast (%struct.a** @e to i8**), align 8
-; CHECK-NEXT:    [[B:%.*]] = bitcast i8* [[CALL]] to %struct.a**
-; CHECK-NEXT:    store %struct.a* null, %struct.a** [[B]], align 8
-; CHECK-NEXT:    br label [[FOR_COND_1:%.*]]
-; CHECK:       for.cond.1:
-; CHECK-NEXT:    [[G_1:%.*]] = phi i32 [ 0, [[FOR_END_0]] ], [ [[INC6:%.*]], [[FOR_BODY_1:%.*]] ]
-; CHECK-NEXT:    [[CMP_1:%.*]] = icmp ult i32 [[G_1]], 100
-; CHECK-NEXT:    br i1 [[CMP_1]], label [[FOR_BODY_1]], label [[FOR_END_1:%.*]]
-; CHECK:       for.body.1:
-; CHECK-NEXT:    [[CALL4:%.*]] = call i32 (i32*, ...) bitcast (i32 (i32)* @h to i32 (i32*, ...)*)(i32* nonnull [[F]])
-; CHECK-NEXT:    [[INC6]] = add nuw nsw i32 [[G_1]], 1
-; CHECK-NEXT:    br label [[FOR_COND_1]]
-; CHECK:       for.end.1:
-; CHECK-NEXT:    ret i32 0
+; TUNIT-LABEL: define {{[^@]+}}@main() {
+; TUNIT-NEXT:  entry:
+; TUNIT-NEXT:    [[F:%.*]] = alloca i32, align 4
+; TUNIT-NEXT:    br label [[FOR_COND_0:%.*]]
+; TUNIT:       for.cond.0:
+; TUNIT-NEXT:    [[G_0:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[FOR_BODY_0:%.*]] ]
+; TUNIT-NEXT:    [[CMP_0:%.*]] = icmp ult i32 [[G_0]], 100
+; TUNIT-NEXT:    br i1 [[CMP_0]], label [[FOR_BODY_0]], label [[FOR_END_0:%.*]]
+; TUNIT:       for.body.0:
+; TUNIT-NEXT:    [[INC]] = add nuw nsw i32 [[G_0]], 1
+; TUNIT-NEXT:    br label [[FOR_COND_0]]
+; TUNIT:       for.end.0:
+; TUNIT-NEXT:    [[CALL:%.*]] = call i8* @malloc(i64 noundef 8)
+; TUNIT-NEXT:    store i8* [[CALL]], i8** bitcast (%struct.a** @e to i8**), align 8
+; TUNIT-NEXT:    [[B:%.*]] = bitcast i8* [[CALL]] to %struct.a**
+; TUNIT-NEXT:    store %struct.a* null, %struct.a** [[B]], align 8
+; TUNIT-NEXT:    br label [[FOR_COND_1:%.*]]
+; TUNIT:       for.cond.1:
+; TUNIT-NEXT:    [[G_1:%.*]] = phi i32 [ 0, [[FOR_END_0]] ], [ [[INC6:%.*]], [[FOR_BODY_1:%.*]] ]
+; TUNIT-NEXT:    [[CMP_1:%.*]] = icmp ult i32 [[G_1]], 100
+; TUNIT-NEXT:    br i1 [[CMP_1]], label [[FOR_BODY_1]], label [[FOR_END_1:%.*]]
+; TUNIT:       for.body.1:
+; TUNIT-NEXT:    [[INC6]] = add nuw nsw i32 [[G_1]], 1
+; TUNIT-NEXT:    br label [[FOR_COND_1]]
+; TUNIT:       for.end.1:
+; TUNIT-NEXT:    ret i32 0
+;
+; CGSCC-LABEL: define {{[^@]+}}@main() {
+; CGSCC-NEXT:  entry:
+; CGSCC-NEXT:    [[F:%.*]] = alloca i32, align 4
+; CGSCC-NEXT:    br label [[FOR_COND_0:%.*]]
+; CGSCC:       for.cond.0:
+; CGSCC-NEXT:    [[G_0:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[FOR_BODY_0:%.*]] ]
+; CGSCC-NEXT:    [[CMP_0:%.*]] = icmp ult i32 [[G_0]], 100
+; CGSCC-NEXT:    br i1 [[CMP_0]], label [[FOR_BODY_0]], label [[FOR_END_0:%.*]]
+; CGSCC:       for.body.0:
+; CGSCC-NEXT:    [[INC]] = add nuw nsw i32 [[G_0]], 1
+; CGSCC-NEXT:    br label [[FOR_COND_0]]
+; CGSCC:       for.end.0:
+; CGSCC-NEXT:    [[CALL:%.*]] = call i8* @malloc(i64 noundef 8)
+; CGSCC-NEXT:    store i8* [[CALL]], i8** bitcast (%struct.a** @e to i8**), align 8
+; CGSCC-NEXT:    [[B:%.*]] = bitcast i8* [[CALL]] to %struct.a**
+; CGSCC-NEXT:    store %struct.a* null, %struct.a** [[B]], align 8
+; CGSCC-NEXT:    br label [[FOR_COND_1:%.*]]
+; CGSCC:       for.cond.1:
+; CGSCC-NEXT:    [[G_1:%.*]] = phi i32 [ 0, [[FOR_END_0]] ], [ [[INC6:%.*]], [[FOR_BODY_1:%.*]] ]
+; CGSCC-NEXT:    [[CMP_1:%.*]] = icmp ult i32 [[G_1]], 100
+; CGSCC-NEXT:    br i1 [[CMP_1]], label [[FOR_BODY_1]], label [[FOR_END_1:%.*]]
+; CGSCC:       for.body.1:
+; CGSCC-NEXT:    [[CALL4:%.*]] = call i32 (i32*, ...) bitcast (i32 (i32)* @h to i32 (i32*, ...)*)(i32* nonnull [[F]]) #[[ATTR2]]
+; CGSCC-NEXT:    [[INC6]] = add nuw nsw i32 [[G_1]], 1
+; CGSCC-NEXT:    br label [[FOR_COND_1]]
+; CGSCC:       for.end.1:
+; CGSCC-NEXT:    ret i32 0
 ;
 entry:
   %f = alloca i32

diff --git a/llvm/test/Transforms/Attributor/misc_crash.ll b/llvm/test/Transforms/Attributor/misc_crash.ll
index ec9d2fb..f2755e3 100644 (file)
--- a/llvm/test/Transforms/Attributor/misc_crash.ll
+++ b/llvm/test/Transforms/Attributor/misc_crash.ll
@@ -48,10 +48,10 @@ define internal void @func2a(i32* %0) {
 }
 
 define i32 @func2() {
-; CHECK: Function Attrs: norecurse
+; CHECK: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn
 ; CHECK-LABEL: define {{[^@]+}}@func2
 ; CHECK-SAME: () #[[ATTR2:[0-9]+]] {
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call i32 (i32*, ...) bitcast (void (i32*)* @func2a to i32 (i32*, ...)*)(i32* nonnull align 4 dereferenceable(4) @var2)
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call i32 (i32*, ...) bitcast (void (i32*)* @func2a to i32 (i32*, ...)*)(i32* nonnull align 4 dereferenceable(4) @var2) #[[ATTR3:[0-9]+]]
 ; CHECK-NEXT:    [[TMP2:%.*]] = load i32, i32* @var2, align 4
 ; CHECK-NEXT:    ret i32 [[TMP2]]
 ;
@@ -61,10 +61,10 @@ define i32 @func2() {
 }
 
 define i32 @func3(i1 %false) {
-; CHECK: Function Attrs: norecurse
+; CHECK: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn
 ; CHECK-LABEL: define {{[^@]+}}@func3
 ; CHECK-SAME: (i1 [[FALSE:%.*]]) #[[ATTR2]] {
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call i32 (i32*, ...) bitcast (void (i32*)* @func2a to i32 (i32*, ...)*)(i32* nonnull align 4 dereferenceable(4) @var2)
+; CHECK-NEXT:    [[TMP1:%.*]] = tail call i32 (i32*, ...) bitcast (void (i32*)* @func2a to i32 (i32*, ...)*)(i32* nonnull align 4 dereferenceable(4) @var2) #[[ATTR3]]
 ; CHECK-NEXT:    br i1 [[FALSE]], label [[USE_BB:%.*]], label [[RET_BB:%.*]]
 ; CHECK:       use_bb:
 ; CHECK-NEXT:    ret i32 [[TMP1]]
@@ -108,10 +108,10 @@ block:
 }
 
 define i16 @foo3() {
-; CHECK: Function Attrs: norecurse
+; CHECK: Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(none)
 ; CHECK-LABEL: define {{[^@]+}}@foo3
-; CHECK-SAME: () #[[ATTR2]] {
-; CHECK-NEXT:    [[CALL:%.*]] = call i16 bitcast (i16 (i16*, i16)* @bar3 to i16 ()*)()
+; CHECK-SAME: () #[[ATTR0]] {
+; CHECK-NEXT:    [[CALL:%.*]] = call i16 bitcast (i16 (i16*, i16)* @bar3 to i16 ()*)() #[[ATTR4:[0-9]+]]
 ; CHECK-NEXT:    ret i16 [[CALL]]
 ;
   %call = call i16 bitcast (i16 (i16*, i16) * @bar3 to i16 () *)()
@@ -132,5 +132,7 @@ declare void @func6(i8*)
 ;.
 ; CHECK: attributes #[[ATTR0]] = { mustprogress nofree norecurse nosync nounwind willreturn memory(none) }
 ; CHECK: attributes #[[ATTR1]] = { mustprogress nofree norecurse nosync nounwind willreturn memory(write) }
-; CHECK: attributes #[[ATTR2]] = { norecurse }
+; CHECK: attributes #[[ATTR2]] = { mustprogress nofree norecurse nosync nounwind willreturn }
+; CHECK: attributes #[[ATTR3]] = { nofree nosync nounwind willreturn memory(write) }
+; CHECK: attributes #[[ATTR4]] = { nofree nosync nounwind willreturn memory(none) }
 ;.

diff --git a/llvm/test/Transforms/Attributor/noalias.ll b/llvm/test/Transforms/Attributor/noalias.ll
index 6a72ed6..0838a29 100644 (file)
--- a/llvm/test/Transforms/Attributor/noalias.ll
+++ b/llvm/test/Transforms/Attributor/noalias.ll
@@ -570,7 +570,7 @@ define internal fastcc double @strtox(ptr %s, ptr %p, i32 %prec) unnamed_addr {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[F:%.*]] = alloca [[STRUCT__IO_FILE:%.*]], align 8
 ; CHECK-NEXT:    call void @llvm.lifetime.start.p0(i64 noundef 144, ptr nocapture nofree noundef nonnull align 8 dereferenceable(240) [[F]]) #[[ATTR11:[0-9]+]]
-; CHECK-NEXT:    [[CALL:%.*]] = call i32 @sh_fromstring(ptr nonnull align 8 dereferenceable(240) [[F]], ptr [[S]])
+; CHECK-NEXT:    [[CALL:%.*]] = call i32 @sh_fromstring(ptr noundef nonnull align 8 dereferenceable(240) [[F]], ptr [[S]])
 ; CHECK-NEXT:    call void @__shlim(ptr noundef nonnull align 8 dereferenceable(240) [[F]], i64 noundef 0)
 ; CHECK-NEXT:    [[CALL1:%.*]] = call double @__floatscan(ptr noundef nonnull align 8 dereferenceable(240) [[F]], i32 noundef 1, i32 noundef 1)
 ; CHECK-NEXT:    call void @llvm.lifetime.end.p0(i64 noundef 144, ptr nocapture nofree noundef nonnull align 8 dereferenceable(240) [[F]])

diff --git a/llvm/test/Transforms/OpenMP/custom_state_machines.ll b/llvm/test/Transforms/OpenMP/custom_state_machines.ll
index 487d813..74cfb9f 100644 (file)
--- a/llvm/test/Transforms/OpenMP/custom_state_machines.ll
+++ b/llvm/test/Transforms/OpenMP/custom_state_machines.ll
@@ -1546,13 +1546,11 @@ attributes #9 = { convergent nounwind readonly willreturn }
 ; AMDGPU:       worker_state_machine.is_active.check:
 ; AMDGPU-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
 ; AMDGPU:       worker_state_machine.parallel_region.check:
-; AMDGPU-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq ptr [[WORKER_WORK_FN_ADDR_CAST]], @__omp_outlined__19_wrapper
-; AMDGPU-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
+; AMDGPU-NEXT:    br i1 true, label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
 ; AMDGPU:       worker_state_machine.parallel_region.execute:
 ; AMDGPU-NEXT:    call void @__omp_outlined__19_wrapper(i16 0, i32 [[TMP0]])
 ; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; AMDGPU:       worker_state_machine.parallel_region.fallback.execute:
-; AMDGPU-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; AMDGPU:       worker_state_machine.parallel_region.check1:
 ; AMDGPU-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
 ; AMDGPU:       worker_state_machine.parallel_region.end:
 ; AMDGPU-NEXT:    call void @__kmpc_kernel_end_parallel()
@@ -1578,7 +1576,7 @@ attributes #9 = { convergent nounwind readonly willreturn }
 ; AMDGPU-NEXT:  entry:
 ; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 ; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
-; AMDGPU-NEXT:    [[CALL:%.*]] = call i32 @omp_get_thread_num() #[[ATTR12]]
+; AMDGPU-NEXT:    [[CALL:%.*]] = call i32 @omp_get_thread_num() #[[ATTR10]]
 ; AMDGPU-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[CALL]]) #[[ATTR10]]
 ; AMDGPU-NEXT:    ret void
 ;
@@ -2488,13 +2486,11 @@ attributes #9 = { convergent nounwind readonly willreturn }
 ; NVPTX:       worker_state_machine.is_active.check:
 ; NVPTX-NEXT:    br i1 [[WORKER_IS_ACTIVE]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK:%.*]], label [[WORKER_STATE_MACHINE_DONE_BARRIER:%.*]]
 ; NVPTX:       worker_state_machine.parallel_region.check:
-; NVPTX-NEXT:    [[WORKER_CHECK_PARALLEL_REGION:%.*]] = icmp eq ptr [[WORKER_WORK_FN_ADDR_CAST]], @__omp_outlined__19_wrapper
-; NVPTX-NEXT:    br i1 [[WORKER_CHECK_PARALLEL_REGION]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_FALLBACK_EXECUTE:%.*]]
+; NVPTX-NEXT:    br i1 true, label [[WORKER_STATE_MACHINE_PARALLEL_REGION_EXECUTE:%.*]], label [[WORKER_STATE_MACHINE_PARALLEL_REGION_CHECK1:%.*]]
 ; NVPTX:       worker_state_machine.parallel_region.execute:
 ; NVPTX-NEXT:    call void @__omp_outlined__19_wrapper(i16 0, i32 [[TMP0]])
 ; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END:%.*]]
-; NVPTX:       worker_state_machine.parallel_region.fallback.execute:
-; NVPTX-NEXT:    call void [[WORKER_WORK_FN_ADDR_CAST]](i16 0, i32 [[TMP0]])
+; NVPTX:       worker_state_machine.parallel_region.check1:
 ; NVPTX-NEXT:    br label [[WORKER_STATE_MACHINE_PARALLEL_REGION_END]]
 ; NVPTX:       worker_state_machine.parallel_region.end:
 ; NVPTX-NEXT:    call void @__kmpc_kernel_end_parallel()
@@ -2520,7 +2516,7 @@ attributes #9 = { convergent nounwind readonly willreturn }
 ; NVPTX-NEXT:  entry:
 ; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 ; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
-; NVPTX-NEXT:    [[CALL:%.*]] = call i32 @omp_get_thread_num() #[[ATTR12]]
+; NVPTX-NEXT:    [[CALL:%.*]] = call i32 @omp_get_thread_num() #[[ATTR10]]
 ; NVPTX-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[CALL]]) #[[ATTR10]]
 ; NVPTX-NEXT:    ret void
 ;
@@ -3217,7 +3213,7 @@ attributes #9 = { convergent nounwind readonly willreturn }
 ; AMDGPU-DISABLED-NEXT:  entry:
 ; AMDGPU-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 ; AMDGPU-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
-; AMDGPU-DISABLED-NEXT:    [[CALL:%.*]] = call i32 @omp_get_thread_num() #[[ATTR12]]
+; AMDGPU-DISABLED-NEXT:    [[CALL:%.*]] = call i32 @omp_get_thread_num() #[[ATTR10]]
 ; AMDGPU-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[CALL]]) #[[ATTR10]]
 ; AMDGPU-DISABLED-NEXT:    ret void
 ;
@@ -3884,7 +3880,7 @@ attributes #9 = { convergent nounwind readonly willreturn }
 ; NVPTX-DISABLED-NEXT:  entry:
 ; NVPTX-DISABLED-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 ; NVPTX-DISABLED-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
-; NVPTX-DISABLED-NEXT:    [[CALL:%.*]] = call i32 @omp_get_thread_num() #[[ATTR12]]
+; NVPTX-DISABLED-NEXT:    [[CALL:%.*]] = call i32 @omp_get_thread_num() #[[ATTR10]]
 ; NVPTX-DISABLED-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[CALL]]) #[[ATTR10]]
 ; NVPTX-DISABLED-NEXT:    ret void
 ;

diff --git a/llvm/test/Transforms/OpenMP/custom_state_machines_pre_lto.ll b/llvm/test/Transforms/OpenMP/custom_state_machines_pre_lto.ll
index 44ba276..d5f4ee1 100644 (file)
--- a/llvm/test/Transforms/OpenMP/custom_state_machines_pre_lto.ll
+++ b/llvm/test/Transforms/OpenMP/custom_state_machines_pre_lto.ll
@@ -1323,7 +1323,7 @@ attributes #9 = { convergent nounwind readonly willreturn }
 ; AMDGPU-NEXT:  entry:
 ; AMDGPU-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 ; AMDGPU-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
-; AMDGPU-NEXT:    [[CALL:%.*]] = call i32 @omp_get_thread_num() #[[ATTR12]]
+; AMDGPU-NEXT:    [[CALL:%.*]] = call i32 @omp_get_thread_num() #[[ATTR10]]
 ; AMDGPU-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[CALL]]) #[[ATTR10]]
 ; AMDGPU-NEXT:    ret void
 ;
@@ -1985,7 +1985,7 @@ attributes #9 = { convergent nounwind readonly willreturn }
 ; NVPTX-NEXT:  entry:
 ; NVPTX-NEXT:    [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
 ; NVPTX-NEXT:    [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
-; NVPTX-NEXT:    [[CALL:%.*]] = call i32 @omp_get_thread_num() #[[ATTR12]]
+; NVPTX-NEXT:    [[CALL:%.*]] = call i32 @omp_get_thread_num() #[[ATTR10]]
 ; NVPTX-NEXT:    call void @simple_state_machine_interprocedural_nested_recursive_after.internalized(i32 [[CALL]]) #[[ATTR10]]
 ; NVPTX-NEXT:    ret void
 ;