[clang] Use tighter lifetime bounds for C temporary arguments

In C, consecutive statements in the same scope are under
CompoundStmt/CallExpr, while in C++ they typically fall under
CompoundStmt/ExprWithCleanup. This leads to different behavior with
respect to where pushFullExprCleanUp inserts the lifetime end markers
(e.g., at the end of scope).

For these cases, we can track and insert the lifetime end markers right
after the call completes. Allowing the stack space to be reused
immediately. This partially addresses #109204 and #43598 for improving
stack usage.

Author: ilovepi

clang/lib/CodeGen/CGCall.cpp

@@ -4973,11 +4973,16 @@ void CodeGenFunction::EmitCallArg(CallArgList &args, const Expr *E,
     RawAddress ArgSlotAlloca = Address::invalid();
     ArgSlot = CreateAggTemp(E->getType(), "agg.tmp", &ArgSlotAlloca);
 
-    // Emit a lifetime start/end for this temporary at the end of the full
-    // expression.
+    // Emit a lifetime start/end for this temporary. If the type has a
+    // destructor, then we need to keep it alive for the full expression.
     if (!CGM.getCodeGenOpts().NoLifetimeMarkersForTemporaries &&
-        EmitLifetimeStart(ArgSlotAlloca.getPointer()))
-      pushFullExprCleanup<CallLifetimeEnd>(NormalAndEHCleanup, ArgSlotAlloca);
+        EmitLifetimeStart(ArgSlotAlloca.getPointer())) {
+      if (E->getType().isDestructedType()) {
+        pushFullExprCleanup<CallLifetimeEnd>(NormalAndEHCleanup, ArgSlotAlloca);
+      } else {
+        args.addLifetimeCleanup({ArgSlotAlloca.getPointer()});
+      }
+    }
   }
 
   args.add(EmitAnyExpr(E, ArgSlot), type);
@@ -6307,6 +6312,11 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
   for (CallLifetimeEnd &LifetimeEnd : CallLifetimeEndAfterCall)
     LifetimeEnd.Emit(*this, /*Flags=*/{});
 
+  if (!CGM.getCodeGenOpts().NoLifetimeMarkersForTemporaries)
+    for (const CallArgList::EndLifetimeInfo &LT :
+         CallArgs.getLifetimeCleanups())
+      EmitLifetimeEnd(LT.Addr);
+
   if (!ReturnValue.isExternallyDestructed() &&
       RetTy.isDestructedType() == QualType::DK_nontrivial_c_struct)
     pushDestroy(QualType::DK_nontrivial_c_struct, Ret.getAggregateAddress(),

clang/lib/CodeGen/CGCall.h

@@ -299,6 +299,10 @@ class CallArgList : public SmallVector<CallArg, 8> {
     llvm::Instruction *IsActiveIP;
   };
 
+  struct EndLifetimeInfo {
+    llvm::Value *Addr;
+  };
+
   void add(RValue rvalue, QualType type) { push_back(CallArg(rvalue, type)); }
 
   void addUncopiedAggregate(LValue LV, QualType type) {
@@ -312,6 +316,9 @@ class CallArgList : public SmallVector<CallArg, 8> {
     llvm::append_range(*this, other);
     llvm::append_range(Writebacks, other.Writebacks);
     llvm::append_range(CleanupsToDeactivate, other.CleanupsToDeactivate);
+    LifetimeCleanups.insert(LifetimeCleanups.end(),
+                            other.LifetimeCleanups.begin(),
+                            other.LifetimeCleanups.end());
     assert(!(StackBase && other.StackBase) && "can't merge stackbases");
     if (!StackBase)
       StackBase = other.StackBase;
@@ -352,6 +359,14 @@ class CallArgList : public SmallVector<CallArg, 8> {
   /// memory.
   bool isUsingInAlloca() const { return StackBase; }
 
+  void addLifetimeCleanup(EndLifetimeInfo Info) {
+    LifetimeCleanups.push_back(Info);
+  }
+
+  ArrayRef<EndLifetimeInfo> getLifetimeCleanups() const {
+    return LifetimeCleanups;
+  }
+
   // Support reversing writebacks for MSVC ABI.
   void reverseWritebacks() {
     std::reverse(Writebacks.begin(), Writebacks.end());
@@ -365,6 +380,10 @@ class CallArgList : public SmallVector<CallArg, 8> {
   /// occurs.
   SmallVector<CallArgCleanup, 1> CleanupsToDeactivate;
 
+  /// Lifetime information needed to call llvm.lifetime.end for any temporary
+  /// argument allocas.
+  SmallVector<EndLifetimeInfo, 2> LifetimeCleanups;
+
   /// The stacksave call.  It dominates all of the argument evaluation.
   llvm::CallInst *StackBase = nullptr;
 };

clang/test/CodeGen/stack-usage-lifetimes.c

@@ -40,23 +40,23 @@ void t1(int c) {
 }
 
 void t2(void) {
-  // x86-precise-remark@-1 {{72 stack bytes}}
+  // x86-precise-remark@-1 {{40 stack bytes}}
   // x86-sloppy-remark@-2 {{72 stack bytes}}
-  // aarch64-precise-remark@-3 {{80 stack bytes}}
+  // aarch64-precise-remark@-3 {{48 stack bytes}}
   // aarch64-sloppy-remark@-4 {{80 stack bytes}}
-  // riscv-precise-remark@-5 {{80 stack bytes}}
+  // riscv-precise-remark@-5 {{48 stack bytes}}
   // riscv-sloppy-remark@-6 {{80 stack bytes}}
 
   useA(genA());
   useA(genA());
 }
 
 void t3(void) {
-  // x86-precise-remark@-1 {{72 stack bytes}}
+  // x86-precise-remark@-1 {{40 stack bytes}}
   // x86-sloppy-remark@-2 {{72 stack bytes}}
-  // aarch64-precise-remark@-3 {{80 stack bytes}}
+  // aarch64-precise-remark@-3 {{48 stack bytes}}
   // aarch64-sloppy-remark@-4 {{80 stack bytes}}
-  // riscv-precise-remark@-5 {{80 stack bytes}}
+  // riscv-precise-remark@-5 {{48 stack bytes}}
   // riscv-sloppy-remark@-6 {{80 stack bytes}}
 
   useB(genB());

clang/test/CodeGenCXX/stack-reuse-miscompile.cpp

@@ -38,10 +38,10 @@ const char * f(S s)
 // CHECK: call void @llvm.lifetime.start.p0(ptr [[T3]])
 // CHECK: call void @llvm.lifetime.start.p0(ptr [[AGG]])
 // CHECK: [[T5:%.*]] = call noundef ptr @_ZN1TC1E1S(ptr {{[^,]*}} [[T3]], [2 x i32] %{{.*}})
+// CHECK: call void @llvm.lifetime.end.p0(ptr [[AGG]])
 //
 // CHECK: call void @_ZNK1T6concatERKS_(ptr dead_on_unwind writable sret(%class.T) align 4 [[T1]], ptr {{[^,]*}} [[T2]], ptr noundef nonnull align 4 dereferenceable(16) [[T3]])
 // CHECK: [[T6:%.*]] = call noundef ptr @_ZNK1T3strEv(ptr {{[^,]*}} [[T1]])
-// CHECK: call void @llvm.lifetime.end.p0(ptr [[AGG]])
 //
 // CHECK: call void @llvm.lifetime.end.p0(
 // CHECK: call void @llvm.lifetime.end.p0(