Compute and preserve alignment more faithfully in IR-generation.

Introduce an Address type to bundle a pointer value with an
alignment.  Introduce APIs on CGBuilderTy to work with Address
values.  Change core APIs on CGF/CGM to traffic in Address where
appropriate.  Require alignments to be non-zero.  Update a ton
of code to compute and propagate alignment information.

As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment
helper function to CGF and made use of it in a number of places in
the expression emitter.

The end result is that we should now be significantly more correct
when performing operations on objects that are locally known to
be under-aligned.  Since alignment is not reliably tracked in the
type system, there are inherent limits to this, but at least we
are no longer confused by standard operations like derived-to-base
conversions and array-to-pointer decay.  I've also fixed a large
number of bugs where we were applying the complete-object alignment
to a pointer instead of the non-virtual alignment, although most of
these were hidden by the very conservative approach we took with
member alignment.

Also, because IRGen now reliably asserts on zero alignments, we
should no longer be subject to an absurd but frustrating recurring
bug where an incomplete type would report a zero alignment and then
we'd naively do a alignmentAtOffset on it and emit code using an
alignment equal to the largest power-of-two factor of the offset.

We should also now be emitting much more aggressive alignment
attributes in the presence of over-alignment.  In particular,
field access now uses alignmentAtOffset instead of min.

Several times in this patch, I had to change the existing
code-generation pattern in order to more effectively use
the Address APIs.  For the most part, this seems to be a strict
improvement, like doing pointer arithmetic with GEPs instead of
ptrtoint.  That said, I've tried very hard to not change semantics,
but it is likely that I've failed in a few places, for which I
apologize.

ABIArgInfo now always carries the assumed alignment of indirect and
indirect byval arguments.  In order to cut down on what was already
a dauntingly large patch, I changed the code to never set align
attributes in the IR on non-byval indirect arguments.  That is,
we still generate code which assumes that indirect arguments have
the given alignment, but we don't express this information to the
backend except where it's semantically required (i.e. on byvals).
This is likely a minor regression for those targets that did provide
this information, but it'll be trivial to add it back in a later
patch.

I partially punted on applying this work to CGBuiltin.  Please
do not add more uses of the CreateDefaultAligned{Load,Store}
APIs; they will be going away eventually.

llvm-svn: 246985

1 files changed, 84 insertions, 53 deletions

diff --git a/clang/lib/CodeGen/CGCleanup.cpp b/clang/lib/CodeGen/CGCleanup.cpp
index 95d2914df93..b5a64d0a28a 100644
--- a/clang/lib/CodeGen/CGCleanup.cpp
+++ b/clang/lib/CodeGen/CGCleanup.cpp
@@ -27,7 +27,7 @@ bool DominatingValue<RValue>::saved_type::needsSaving(RValue rv) {
   if (rv.isScalar())
     return DominatingLLVMValue::needsSaving(rv.getScalarVal());
   if (rv.isAggregate())
-    return DominatingLLVMValue::needsSaving(rv.getAggregateAddr());
+    return DominatingLLVMValue::needsSaving(rv.getAggregatePointer());
   return true;
 }
 
@@ -41,9 +41,10 @@ DominatingValue<RValue>::saved_type::save(CodeGenFunction &CGF, RValue rv) {
       return saved_type(V, ScalarLiteral);
 
     // Everything else needs an alloca.
-    llvm::Value *addr = CGF.CreateTempAlloca(V->getType(), "saved-rvalue");
+    Address addr =
+      CGF.CreateDefaultAlignTempAlloca(V->getType(), "saved-rvalue");
     CGF.Builder.CreateStore(V, addr);
-    return saved_type(addr, ScalarAddress);
+    return saved_type(addr.getPointer(), ScalarAddress);
   }
 
   if (rv.isComplex()) {
@@ -51,42 +52,56 @@ DominatingValue<RValue>::saved_type::save(CodeGenFunction &CGF, RValue rv) {
     llvm::Type *ComplexTy =
       llvm::StructType::get(V.first->getType(), V.second->getType(),
                             (void*) nullptr);
-    llvm::Value *addr = CGF.CreateTempAlloca(ComplexTy, "saved-complex");
+    Address addr = CGF.CreateDefaultAlignTempAlloca(ComplexTy, "saved-complex");
     CGF.Builder.CreateStore(V.first,
-                            CGF.Builder.CreateStructGEP(ComplexTy, addr, 0));
+                            CGF.Builder.CreateStructGEP(addr, 0, CharUnits()));
+    CharUnits offset = CharUnits::fromQuantity(
+               CGF.CGM.getDataLayout().getTypeAllocSize(V.first->getType()));
     CGF.Builder.CreateStore(V.second,
-                            CGF.Builder.CreateStructGEP(ComplexTy, addr, 1));
-    return saved_type(addr, ComplexAddress);
+                            CGF.Builder.CreateStructGEP(addr, 1, offset));
+    return saved_type(addr.getPointer(), ComplexAddress);
   }
 
   assert(rv.isAggregate());
-  llvm::Value *V = rv.getAggregateAddr(); // TODO: volatile?
-  if (!DominatingLLVMValue::needsSaving(V))
-    return saved_type(V, AggregateLiteral);
-
-  llvm::Value *addr = CGF.CreateTempAlloca(V->getType(), "saved-rvalue");
-  CGF.Builder.CreateStore(V, addr);
-  return saved_type(addr, AggregateAddress);  
+  Address V = rv.getAggregateAddress(); // TODO: volatile?
+  if (!DominatingLLVMValue::needsSaving(V.getPointer()))
+    return saved_type(V.getPointer(), AggregateLiteral,
+                      V.getAlignment().getQuantity());
+
+  Address addr =
+    CGF.CreateTempAlloca(V.getType(), CGF.getPointerAlign(), "saved-rvalue");
+  CGF.Builder.CreateStore(V.getPointer(), addr);
+  return saved_type(addr.getPointer(), AggregateAddress,
+                    V.getAlignment().getQuantity());
 }
 
 /// Given a saved r-value produced by SaveRValue, perform the code
 /// necessary to restore it to usability at the current insertion
 /// point.
 RValue DominatingValue<RValue>::saved_type::restore(CodeGenFunction &CGF) {
+  auto getSavingAddress = [&](llvm::Value *value) {
+    auto alignment = cast<llvm::AllocaInst>(value)->getAlignment();
+    return Address(value, CharUnits::fromQuantity(alignment));
+  };
   switch (K) {
   case ScalarLiteral:
     return RValue::get(Value);
   case ScalarAddress:
-    return RValue::get(CGF.Builder.CreateLoad(Value));
+    return RValue::get(CGF.Builder.CreateLoad(getSavingAddress(Value)));
   case AggregateLiteral:
-    return RValue::getAggregate(Value);
-  case AggregateAddress:
-    return RValue::getAggregate(CGF.Builder.CreateLoad(Value));
+    return RValue::getAggregate(Address(Value, CharUnits::fromQuantity(Align)));
+  case AggregateAddress: {
+    auto addr = CGF.Builder.CreateLoad(getSavingAddress(Value));
+    return RValue::getAggregate(Address(addr, CharUnits::fromQuantity(Align)));
+  }
   case ComplexAddress: {
-    llvm::Value *real =
-        CGF.Builder.CreateLoad(CGF.Builder.CreateStructGEP(nullptr, Value, 0));
-    llvm::Value *imag =
-        CGF.Builder.CreateLoad(CGF.Builder.CreateStructGEP(nullptr, Value, 1));
+    Address address = getSavingAddress(Value);
+    llvm::Value *real = CGF.Builder.CreateLoad(
+                 CGF.Builder.CreateStructGEP(address, 0, CharUnits()));
+    CharUnits offset = CharUnits::fromQuantity(
+                 CGF.CGM.getDataLayout().getTypeAllocSize(real->getType()));
+    llvm::Value *imag = CGF.Builder.CreateLoad(
+                 CGF.Builder.CreateStructGEP(address, 1, offset));
     return RValue::getComplex(real, imag);
   }
   }
@@ -275,8 +290,8 @@ void EHScopeStack::popNullFixups() {
 
 void CodeGenFunction::initFullExprCleanup() {
   // Create a variable to decide whether the cleanup needs to be run.
-  llvm::AllocaInst *active
-    = CreateTempAlloca(Builder.getInt1Ty(), "cleanup.cond");
+  Address active = CreateTempAlloca(Builder.getInt1Ty(), CharUnits::One(),
+                                    "cleanup.cond");
 
   // Initialize it to false at a site that's guaranteed to be run
   // before each evaluation.
@@ -287,7 +302,7 @@ void CodeGenFunction::initFullExprCleanup() {
 
   // Set that as the active flag in the cleanup.
   EHCleanupScope &cleanup = cast<EHCleanupScope>(*EHStack.begin());
-  assert(!cleanup.getActiveFlag() && "cleanup already has active flag?");
+  assert(!cleanup.hasActiveFlag() && "cleanup already has active flag?");
   cleanup.setActiveFlag(active);
 
   if (cleanup.isNormalCleanup()) cleanup.setTestFlagInNormalCleanup();
@@ -296,6 +311,19 @@ void CodeGenFunction::initFullExprCleanup() {
 
 void EHScopeStack::Cleanup::anchor() {}
 
+static void createStoreInstBefore(llvm::Value *value, Address addr,
+                                  llvm::Instruction *beforeInst) {
+  auto store = new llvm::StoreInst(value, addr.getPointer(), beforeInst);
+  store->setAlignment(addr.getAlignment().getQuantity());
+}
+
+static llvm::LoadInst *createLoadInstBefore(Address addr, const Twine &name,
+                                            llvm::Instruction *beforeInst) {
+  auto load = new llvm::LoadInst(addr.getPointer(), name, beforeInst);
+  load->setAlignment(addr.getAlignment().getQuantity());
+  return load;
+}                                 
+
 /// All the branch fixups on the EH stack have propagated out past the
 /// outermost normal cleanup; resolve them all by adding cases to the
 /// given switch instruction.
@@ -318,9 +346,9 @@ static void ResolveAllBranchFixups(CodeGenFunction &CGF,
     // i.e. where there's an unresolved fixup inside a single cleanup
     // entry which we're currently popping.
     if (Fixup.OptimisticBranchBlock == nullptr) {
-      new llvm::StoreInst(CGF.Builder.getInt32(Fixup.DestinationIndex),
-                          CGF.getNormalCleanupDestSlot(),
-                          Fixup.InitialBranch);
+      createStoreInstBefore(CGF.Builder.getInt32(Fixup.DestinationIndex),
+                            CGF.getNormalCleanupDestSlot(),
+                            Fixup.InitialBranch);
       Fixup.InitialBranch->setSuccessor(0, CleanupEntry);
     }
 
@@ -346,8 +374,8 @@ static llvm::SwitchInst *TransitionToCleanupSwitch(CodeGenFunction &CGF,
 
   if (llvm::BranchInst *Br = dyn_cast<llvm::BranchInst>(Term)) {
     assert(Br->isUnconditional());
-    llvm::LoadInst *Load =
-      new llvm::LoadInst(CGF.getNormalCleanupDestSlot(), "cleanup.dest", Term);
+    auto Load = createLoadInstBefore(CGF.getNormalCleanupDestSlot(),
+                                     "cleanup.dest", Term);
     llvm::SwitchInst *Switch =
       llvm::SwitchInst::Create(Load, Br->getSuccessor(0), 4, Block);
     Br->eraseFromParent();
@@ -492,7 +520,7 @@ static llvm::BasicBlock *SimplifyCleanupEntry(CodeGenFunction &CGF,
 static void EmitCleanup(CodeGenFunction &CGF,
                         EHScopeStack::Cleanup *Fn,
                         EHScopeStack::Cleanup::Flags flags,
-                        llvm::Value *ActiveFlag) {
+                        Address ActiveFlag) {
   // Itanium EH cleanups occur within a terminate scope. Microsoft SEH doesn't
   // have this behavior, and the Microsoft C++ runtime will call terminate for
   // us if the cleanup throws.
@@ -505,7 +533,7 @@ static void EmitCleanup(CodeGenFunction &CGF,
   // If there's an active flag, load it and skip the cleanup if it's
   // false.
   llvm::BasicBlock *ContBB = nullptr;
-  if (ActiveFlag) {
+  if (ActiveFlag.isValid()) {
     ContBB = CGF.createBasicBlock("cleanup.done");
     llvm::BasicBlock *CleanupBB = CGF.createBasicBlock("cleanup.action");
     llvm::Value *IsActive
@@ -519,7 +547,7 @@ static void EmitCleanup(CodeGenFunction &CGF,
   assert(CGF.HaveInsertPoint() && "cleanup ended with no insertion point?");
 
   // Emit the continuation block if there was an active flag.
-  if (ActiveFlag)
+  if (ActiveFlag.isValid())
     CGF.EmitBlock(ContBB);
 
   // Leave the terminate scope.
@@ -599,10 +627,12 @@ void CodeGenFunction::PopCleanupBlock(bool FallthroughIsBranchThrough) {
 
   // Remember activation information.
   bool IsActive = Scope.isActive();
-  llvm::Value *NormalActiveFlag =
-    Scope.shouldTestFlagInNormalCleanup() ? Scope.getActiveFlag() : nullptr;
-  llvm::Value *EHActiveFlag = 
-    Scope.shouldTestFlagInEHCleanup() ? Scope.getActiveFlag() : nullptr;
+  Address NormalActiveFlag =
+    Scope.shouldTestFlagInNormalCleanup() ? Scope.getActiveFlag()
+                                          : Address::invalid();
+  Address EHActiveFlag = 
+    Scope.shouldTestFlagInEHCleanup() ? Scope.getActiveFlag()
+                                      : Address::invalid();
 
   // Check whether we need an EH cleanup.  This is only true if we've
   // generated a lazy EH cleanup block.
@@ -769,7 +799,7 @@ void CodeGenFunction::PopCleanupBlock(bool FallthroughIsBranchThrough) {
 
         // Clean up the possibly dead store to the cleanup dest slot.
         llvm::Instruction *NormalCleanupDestSlot =
-            cast<llvm::Instruction>(getNormalCleanupDestSlot());
+            cast<llvm::Instruction>(getNormalCleanupDestSlot().getPointer());
         if (NormalCleanupDestSlot->hasOneUse()) {
           NormalCleanupDestSlot->user_back()->eraseFromParent();
           NormalCleanupDestSlot->eraseFromParent();
@@ -795,7 +825,8 @@ void CodeGenFunction::PopCleanupBlock(bool FallthroughIsBranchThrough) {
         const unsigned SwitchCapacity = 10;
 
         llvm::LoadInst *Load =
-          new llvm::LoadInst(getNormalCleanupDestSlot(), "cleanup.dest");
+          createLoadInstBefore(getNormalCleanupDestSlot(), "cleanup.dest",
+                               nullptr);
         llvm::SwitchInst *Switch =
           llvm::SwitchInst::Create(Load, Default, SwitchCapacity);
 
@@ -841,9 +872,9 @@ void CodeGenFunction::PopCleanupBlock(bool FallthroughIsBranchThrough) {
         BranchFixup &Fixup = EHStack.getBranchFixup(I);
         if (!Fixup.Destination) continue;
         if (!Fixup.OptimisticBranchBlock) {
-          new llvm::StoreInst(Builder.getInt32(Fixup.DestinationIndex),
-                              getNormalCleanupDestSlot(),
-                              Fixup.InitialBranch);
+          createStoreInstBefore(Builder.getInt32(Fixup.DestinationIndex),
+                                getNormalCleanupDestSlot(),
+                                Fixup.InitialBranch);
           Fixup.InitialBranch->setSuccessor(0, NormalEntry);
         }
         Fixup.OptimisticBranchBlock = NormalExit;
@@ -908,8 +939,7 @@ void CodeGenFunction::PopCleanupBlock(bool FallthroughIsBranchThrough) {
 
     // We only actually emit the cleanup code if the cleanup is either
     // active or was used before it was deactivated.
-    if (EHActiveFlag || IsActive) {
-
+    if (EHActiveFlag.isValid() || IsActive) {
       cleanupFlags.setIsForEHCleanup();
       EmitCleanup(*this, Fn, cleanupFlags, EHActiveFlag);
     }
@@ -993,7 +1023,7 @@ void CodeGenFunction::EmitBranchThroughCleanup(JumpDest Dest) {
 
   // Store the index at the start.
   llvm::ConstantInt *Index = Builder.getInt32(Dest.getDestIndex());
-  new llvm::StoreInst(Index, getNormalCleanupDestSlot(), BI);
+  createStoreInstBefore(Index, getNormalCleanupDestSlot(), BI);
 
   // Adjust BI to point to the first cleanup block.
   {
@@ -1112,23 +1142,24 @@ static void SetupCleanupBlockActivation(CodeGenFunction &CGF,
   // If it hasn't yet been used as either, we're done.
   if (!needFlag) return;
 
-  llvm::AllocaInst *var = Scope.getActiveFlag();
-  if (!var) {
-    var = CGF.CreateTempAlloca(CGF.Builder.getInt1Ty(), "cleanup.isactive");
+  Address var = Scope.getActiveFlag();
+  if (!var.isValid()) {
+    var = CGF.CreateTempAlloca(CGF.Builder.getInt1Ty(), CharUnits::One(),
+                               "cleanup.isactive");
     Scope.setActiveFlag(var);
 
     assert(dominatingIP && "no existing variable and no dominating IP!");
 
     // Initialize to true or false depending on whether it was
     // active up to this point.
-    llvm::Value *value = CGF.Builder.getInt1(kind == ForDeactivation);
+    llvm::Constant *value = CGF.Builder.getInt1(kind == ForDeactivation);
 
     // If we're in a conditional block, ignore the dominating IP and
     // use the outermost conditional branch.
     if (CGF.isInConditionalBranch()) {
       CGF.setBeforeOutermostConditional(value, var);
     } else {
-      new llvm::StoreInst(value, var, dominatingIP);
+      createStoreInstBefore(value, var, dominatingIP);
     }
   }
 
@@ -1170,17 +1201,17 @@ void CodeGenFunction::DeactivateCleanupBlock(EHScopeStack::stable_iterator C,
   Scope.setActive(false);
 }
 
-llvm::Value *CodeGenFunction::getNormalCleanupDestSlot() {
+Address CodeGenFunction::getNormalCleanupDestSlot() {
   if (!NormalCleanupDest)
     NormalCleanupDest =
       CreateTempAlloca(Builder.getInt32Ty(), "cleanup.dest.slot");
-  return NormalCleanupDest;
+  return Address(NormalCleanupDest, CharUnits::fromQuantity(4));
 }
 
 /// Emits all the code to cause the given temporary to be cleaned up.
 void CodeGenFunction::EmitCXXTemporary(const CXXTemporary *Temporary,
                                        QualType TempType,
-                                       llvm::Value *Ptr) {
+                                       Address Ptr) {
   pushDestroy(NormalAndEHCleanup, Ptr, TempType, destroyCXXObject,
               /*useEHCleanup*/ true);
 }