[llvm][GlobalOpt] Optimize statically resolvable IFuncs

llvm/lib/Transforms/IPO/GlobalOpt.cpp

@@ -89,6 +89,8 @@ STATISTIC(NumAliasesRemoved, "Number of global aliases eliminated");
 STATISTIC(NumCXXDtorsRemoved, "Number of global C++ destructors removed");
 STATISTIC(NumInternalFunc, "Number of internal functions");
 STATISTIC(NumColdCC, "Number of functions marked coldcc");
+STATISTIC(NumIFuncsResolved, "Number of statically resolved IFuncs");
+STATISTIC(NumIFuncsDeleted, "Number of IFuncs removed");
 
 static cl::opt<bool>
     EnableColdCCStressTest("enable-coldcc-stress-test",
@@ -2404,6 +2406,60 @@ static bool OptimizeEmptyGlobalCXXDtors(Function *CXAAtExitFn) {
   return Changed;
 }
 
+static Function *hasSideeffectFreeStaticResolution(GlobalIFunc &IF) {
+  if (IF.isInterposable())
+    return nullptr;
+
+  Function *Resolver = IF.getResolverFunction();
+  if (!Resolver)
+    return nullptr;
+
+  if (Resolver->isInterposable())
+    return nullptr;
+
+  // Only handle functions that have been optimized into a single basic block.
+  auto It = Resolver->begin();
+  if (++It != Resolver->end())
+    return nullptr;
+
+  BasicBlock &BB = Resolver->getEntryBlock();
+
+  if (any_of(BB, [](Instruction &I) { return I.mayHaveSideEffects(); }))
+    return nullptr;
+
+  auto *Ret = dyn_cast<ReturnInst>(BB.getTerminator());
+  if (!Ret)
+    return nullptr;
+
+  return dyn_cast<Function>(Ret->getReturnValue());
+}
+
+/// Find IFuncs that have resolvers that always point at the same statically
+/// known callee, and replace their callers with a direct call.
+static bool OptimizeStaticIFuncs(Module &M) {
+  bool Changed = false;
+  for (GlobalIFunc &IF : M.ifuncs())
+    if (Function *Callee = hasSideeffectFreeStaticResolution(IF))
+      if (!IF.use_empty()) {
+        IF.replaceAllUsesWith(Callee);
+        NumIFuncsResolved++;
+        Changed = true;
+      }
+  return Changed;
+}
+
+static bool
+DeleteDeadIFuncs(Module &M,
+                 SmallPtrSetImpl<const Comdat *> &NotDiscardableComdats) {
+  bool Changed = false;
+  for (GlobalIFunc &IF : make_early_inc_range(M.ifuncs()))
+    if (deleteIfDead(IF, NotDiscardableComdats)) {
+      NumIFuncsDeleted++;
+      Changed = true;
+    }
+  return Changed;
+}
+
 static bool
 optimizeGlobalsInModule(Module &M, const DataLayout &DL,
                         function_ref<TargetLibraryInfo &(Function &)> GetTLI,
@@ -2464,6 +2520,12 @@ optimizeGlobalsInModule(Module &M, const DataLayout &DL,
     if (CXAAtExitFn)
       LocalChange |= OptimizeEmptyGlobalCXXDtors(CXAAtExitFn);
 
+    // Optimize IFuncs whose callee's are statically known.
+    LocalChange |= OptimizeStaticIFuncs(M);
+
+    // Remove any IFuncs that are now dead.
+    LocalChange |= DeleteDeadIFuncs(M, NotDiscardableComdats);
+
     Changed |= LocalChange;
   }
 

llvm/test/Transforms/GlobalOpt/resolve-static-ifunc.ll

@@ -0,0 +1,109 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function caller --check-globals all --version 4
+; RUN: opt --passes=globalopt -o - -S < %s | FileCheck %s --implicit-check-not=trivial\.ifunc --implicit-check-not=dead_ifunc
+
+target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64-unknown-linux-gnu"
+
+@trivial.ifunc = internal ifunc void (), ptr @trivial.resolver
+;.
+; CHECK: @unknown_condition = external local_unnamed_addr global i1
+; CHECK: @external_ifunc.ifunc = dso_local ifunc void (), ptr @external_ifunc.resolver
+; CHECK: @complex.ifunc = internal ifunc void (), ptr @complex.resolver
+; CHECK: @sideeffects.ifunc = internal ifunc void (), ptr @sideeffects.resolver
+; CHECK: @interposable_ifunc.ifunc = internal ifunc void (), ptr @interposable_ifunc.resolver
+; CHECK: @interposable_resolver.ifunc = weak ifunc void (), ptr @interposable_resolver.resolver
+;.
+define ptr @trivial.resolver() {
+  ret ptr @trivial._Msimd
+}
+define void @trivial._Msimd() {
+  ret void
+}
+define void @trivial.default() {
+  ret void
+}
+
+
+@dead_ifunc.ifunc = internal ifunc void (), ptr @trivial.resolver
+
+@external_ifunc.ifunc = dso_local ifunc void (), ptr @external_ifunc.resolver
+define ptr @external_ifunc.resolver() {
+  ret ptr @external_ifunc._Msimd
+}
+define void @external_ifunc._Msimd() {
+  ret void
+}
+define void @external_ifunc.default() {
+  ret void
+}
+
+@unknown_condition = external global i1
+@complex.ifunc = internal ifunc void (), ptr @complex.resolver
+define ptr @complex.resolver() {
+entry:
+  %v = load i1, ptr @unknown_condition
+  br i1 %v, label %fast, label %slow
+fast:
+  ret ptr @complex._Msimd
+slow:
+  ret ptr @complex._Msimd
+}
+define void @complex._Msimd() {
+  ret void
+}
+define void @complex.default() {
+  ret void
+}
+
+@sideeffects.ifunc = internal ifunc void (), ptr @sideeffects.resolver
+define ptr @sideeffects.resolver() {
+  store i1 0, ptr @unknown_condition
+  ret ptr @sideeffects.default
+}
+define void @sideeffects._Msimd() {
+  ret void
+}
+define void @sideeffects.default() {
+  ret void
+}
+
+@interposable_ifunc.ifunc = internal ifunc void (), ptr @interposable_ifunc.resolver
+define weak ptr @interposable_ifunc.resolver() {
+  ret ptr @interposable_ifunc.resolver
+}
+define void @interposable_ifunc._Msimd() {
+  ret void
+}
+define void @interposable_ifunc.default() {
+  ret void
+}
+
+@interposable_resolver.ifunc = weak ifunc void (), ptr @interposable_resolver.resolver
+define ptr @interposable_resolver.resolver() {
+  ret ptr @interposable_resolver.resolver
+}
+define void @interposable_resolver._Msimd() {
+  ret void
+}
+define void @interposable_resolver.default() {
+  ret void
+}
+
+define void @caller() {
+; CHECK-LABEL: define void @caller() local_unnamed_addr {
+; CHECK-NEXT:    call void @trivial._Msimd()
+; CHECK-NEXT:    call void @external_ifunc._Msimd()
+; CHECK-NEXT:    call void @complex.ifunc()
+; CHECK-NEXT:    call void @sideeffects.ifunc()
+; CHECK-NEXT:    call void @interposable_ifunc.ifunc()
+; CHECK-NEXT:    call void @interposable_resolver.ifunc()
+; CHECK-NEXT:    ret void
+;
+  call void @trivial.ifunc()
+  call void @external_ifunc.ifunc()
+  call void @complex.ifunc()
+  call void @sideeffects.ifunc()
+  call void @interposable_ifunc.ifunc()
+  call void @interposable_resolver.ifunc()
+  ret void
+}