Fix: Distinguish CFI Metadata Checks in MergeFunctions Pass

llvm/lib/Transforms/IPO/MergeFunctions.cpp

@@ -375,9 +375,32 @@ bool MergeFunctions::doFunctionalCheck(std::vector<WeakTrackingVH> &Worklist) {
 }
 #endif
 
+/// Check whether \p F has an intrinsic which references
+/// distinct metadata as an operand. The most common
+/// instance of this would be CFI checks for function-local types.
+static bool hasDistinctMetadataIntrinsic(const Function &F) {
+  for (const BasicBlock &BB : F) {
+    for (const Instruction &I : BB.instructionsWithoutDebug()) {
+      if (!isa<IntrinsicInst>(&I))
+        continue;
+
+      for (Value *Op : I.operands()) {
+        auto *MDL = dyn_cast<MetadataAsValue>(Op);
+        if (!MDL)
+          continue;
+        if (MDNode *N = dyn_cast<MDNode>(MDL->getMetadata()))
+          if (N->isDistinct())
+            return true;
+      }
+    }
+  }
+  return false;
+}
+
 /// Check whether \p F is eligible for function merging.
 static bool isEligibleForMerging(Function &F) {
-  return !F.isDeclaration() && !F.hasAvailableExternallyLinkage();
+  return !F.isDeclaration() && !F.hasAvailableExternallyLinkage() &&
+         !hasDistinctMetadataIntrinsic(F);
 }
 
 bool MergeFunctions::runOnModule(Module &M) {

llvm/test/Transforms/MergeFunc/cfi-function-merging.ll

@@ -0,0 +1,53 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
+;; Check the cases involving internal CFI instrumented functions where we do not expect functions to be merged.
+; RUN: opt -S -passes=mergefunc < %s | FileCheck %s
+
+; Function Attrs: nocallback nofree nosync nounwind readnone speculatable willreturn
+declare i1 @llvm.type.test(ptr, metadata) #6
+
+define internal void @A__on_zero_sharedEv(ptr noundef nonnull align 8 dereferenceable(32) %this) {
+; CHECK-LABEL: define internal void @A__on_zero_sharedEv
+; CHECK-SAME: (ptr noundef nonnull align 8 dereferenceable(32) [[THIS:%.*]]) {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
+; CHECK-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
+; CHECK-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
+; CHECK-NEXT:    [[VTABLE:%.*]] = load ptr, ptr [[THIS1]], align 8
+; CHECK-NEXT:    [[TMP0:%.*]] = call i1 @llvm.type.test(ptr [[VTABLE]], metadata [[META0:![0-9]+]]), !nosanitize !1
+; CHECK-NEXT:    ret void
+;
+entry:
+  %this.addr = alloca ptr, align 8
+  store ptr %this, ptr %this.addr, align 8
+  %this1 = load ptr, ptr %this.addr, align 8
+  %vtable = load ptr, ptr %this1, align 8
+  %0 = call i1 @llvm.type.test(ptr %vtable, metadata !11), !nosanitize !47
+  ret void
+}
+
+; Function Attrs: mustprogress noinline nounwind optnone uwtable
+define internal void @B__on_zero_sharedEv(ptr noundef nonnull align 8 dereferenceable(32) %this) {
+; CHECK-LABEL: define internal void @B__on_zero_sharedEv
+; CHECK-SAME: (ptr noundef nonnull align 8 dereferenceable(32) [[THIS:%.*]]) {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
+; CHECK-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
+; CHECK-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
+; CHECK-NEXT:    [[VTABLE:%.*]] = load ptr, ptr [[THIS1]], align 8
+; CHECK-NEXT:    [[TMP0:%.*]] = call i1 @llvm.type.test(ptr [[VTABLE]], metadata [[META2:![0-9]+]]), !nosanitize !1
+; CHECK-NEXT:    ret void
+;
+entry:
+  %this.addr = alloca ptr, align 8
+  store ptr %this, ptr %this.addr, align 8
+  %this1 = load ptr, ptr %this.addr, align 8
+  %vtable = load ptr, ptr %this1, align 8
+  %0 = call i1 @llvm.type.test(ptr %vtable, metadata !22), !nosanitize !47
+  ret void
+}
+
+!10 = !{i64 16, !11}
+!11 = distinct !{}
+!21 = !{i64 16, !22}
+!22 = distinct !{}
+!47 = !{}

llvm/test/Transforms/MergeFunc/merge-fp-intrinsics.ll

@@ -0,0 +1,28 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
+;; Make sure internal constrained FP intrinsics still merge properly
+; RUN: opt -passes=mergefunc -S < %s | FileCheck %s
+
+declare float @llvm.experimental.constrained.fadd.f32(float, float, metadata, metadata)
+
+define float @func1(float %a, float %b) {
+; CHECK-LABEL: define float @func1
+; CHECK-SAME: (float [[A:%.*]], float [[B:%.*]]) {
+; CHECK-NEXT:    [[RESULT:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[A]], float [[B]], metadata !"round.dynamic", metadata !"fpexcept.strict")
+; CHECK-NEXT:    [[RESULT_2:%.*]] = call float @llvm.experimental.constrained.fadd.f32(float [[A]], float [[B]], metadata !"round.dynamic", metadata !"fpexcept.strict")
+; CHECK-NEXT:    ret float [[RESULT]]
+;
+  %result = call float @llvm.experimental.constrained.fadd.f32(float %a, float %b, metadata !"round.dynamic", metadata !"fpexcept.strict")
+  %result_2 = call float @llvm.experimental.constrained.fadd.f32(float %a, float %b, metadata !"round.dynamic", metadata !"fpexcept.strict")
+  ret float %result
+}
+
+define float @func2(float %a, float %b) {
+; CHECK-LABEL: define float @func2
+; CHECK-SAME: (float [[TMP0:%.*]], float [[TMP1:%.*]]) {
+; CHECK-NEXT:    [[TMP3:%.*]] = tail call float @func1(float [[TMP0]], float [[TMP1]])
+; CHECK-NEXT:    ret float [[TMP3]]
+;
+  %result = call float @llvm.experimental.constrained.fadd.f32(float %a, float %b, metadata !"round.dynamic", metadata !"fpexcept.strict")
+  %result_2 = call float @llvm.experimental.constrained.fadd.f32(float %a, float %b, metadata !"round.dynamic", metadata !"fpexcept.strict")
+  ret float %result
+}