Preserve lifetime marker's order

clang/lib/CodeGen/CGCall.cpp

@@ -6057,6 +6057,11 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
     Builder.CreateStore(errorResult, swiftErrorArg);
   }
 
+  // Explicitly call CallLifetimeEnd::Emit just to re-use the code even though
+  // we can't use the full cleanup mechanism.
+  for (CallLifetimeEnd &LifetimeEnd : reverse(CallLifetimeEndAfterCall))
+    LifetimeEnd.Emit(*this, /*Flags=*/{});
+
   // Emit any call-associated writebacks immediately.  Arguably this
   // should happen after any return-value munging.
   if (CallArgs.hasWritebacks())
@@ -6194,11 +6199,6 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
     AllocAlignAttrEmitter.EmitAsAnAssumption(Loc, RetTy, Ret);
   }
 
-  // Explicitly call CallLifetimeEnd::Emit just to re-use the code even though
-  // we can't use the full cleanup mechanism.
-  for (CallLifetimeEnd &LifetimeEnd : CallLifetimeEndAfterCall)
-    LifetimeEnd.Emit(*this, /*Flags=*/{});
-
   if (!ReturnValue.isExternallyDestructed() &&
       RetTy.isDestructedType() == QualType::DK_nontrivial_c_struct)
     pushDestroy(QualType::DK_nontrivial_c_struct, Ret.getAggregateAddress(),

llvm/lib/Transforms/Scalar/MemCpyOptimizer.cpp

@@ -836,6 +836,7 @@ bool MemCpyOptPass::processMemSet(MemSetInst *MSI, BasicBlock::iterator &BBI) {
 /// Takes a memcpy and a call that it depends on,
 /// and checks for the possibility of a call slot optimization by having
 /// the call write its result directly into the destination of the memcpy.
+/// FIXME: it breaks the order of lifetime markers.
 bool MemCpyOptPass::performCallSlotOptzn(Instruction *cpyLoad,
                                          Instruction *cpyStore, Value *cpyDest,
                                          Value *cpySrc, TypeSize cpySize,

llvm/lib/Transforms/Scalar/SROA.cpp

@@ -2832,7 +2832,11 @@ class AllocaSliceRewriter : public InstVisitor<AllocaSliceRewriter, bool> {
     OldPtr = cast<Instruction>(OldUse->get());
 
     Instruction *OldUserI = cast<Instruction>(OldUse->getUser());
-    IRB.SetInsertPoint(OldUserI);
+    auto *OldUserII = dyn_cast<IntrinsicInst>(OldUserI);
+    if (OldUserII && OldUserII->getIntrinsicID() == Intrinsic::lifetime_end)
+      IRB.SetInsertPoint(OldUserI->getNextNode());
+    else
+      IRB.SetInsertPoint(OldUserI);
     IRB.SetCurrentDebugLocation(OldUserI->getDebugLoc());
     IRB.getInserter().SetNamePrefix(Twine(NewAI.getName()) + "." +
                                     Twine(BeginOffset) + ".");

llvm/lib/Transforms/Utils/InlineFunction.cpp

@@ -2977,6 +2977,7 @@ llvm::InlineResult llvm::InlineFunction(CallBase &CB, InlineFunctionInfo &IFI,
   if ((InsertLifetime || Caller->isPresplitCoroutine()) &&
       !IFI.StaticAllocas.empty()) {
     IRBuilder<> builder(&*FirstNewBlock, FirstNewBlock->begin());
+    SmallVector<std::pair<AllocaInst *, ConstantInt *>> AllocasWithLifetimeEnd;
     for (AllocaInst *AI : IFI.StaticAllocas) {
       // Don't mark swifterror allocas. They can't have bitcast uses.
       if (AI->isSwiftError())
@@ -3012,17 +3013,19 @@ llvm::InlineResult llvm::InlineFunction(CallBase &CB, InlineFunctionInfo &IFI,
       }
 
       builder.CreateLifetimeStart(AI, AllocaSize);
-      for (ReturnInst *RI : Returns) {
-        // Don't insert llvm.lifetime.end calls between a musttail or deoptimize
-        // call and a return.  The return kills all local allocas.
-        if (InlinedMustTailCalls &&
-            RI->getParent()->getTerminatingMustTailCall())
-          continue;
-        if (InlinedDeoptimizeCalls &&
-            RI->getParent()->getTerminatingDeoptimizeCall())
-          continue;
+      AllocasWithLifetimeEnd.emplace_back(AI, AllocaSize);
+    }
+
+    for (ReturnInst *RI : Returns) {
+      // Don't insert llvm.lifetime.end calls between a musttail or deoptimize
+      // call and a return.  The return kills all local allocas.
+      if (InlinedMustTailCalls && RI->getParent()->getTerminatingMustTailCall())
+        continue;
+      if (InlinedDeoptimizeCalls &&
+          RI->getParent()->getTerminatingDeoptimizeCall())
+        continue;
+      for (auto [AI, AllocaSize] : reverse(AllocasWithLifetimeEnd))
         IRBuilder<>(RI).CreateLifetimeEnd(AI, AllocaSize);
-      }
     }
   }