IR: Split Metadata from Value

Split `Metadata` away from the `Value` class hierarchy, as part of
PR21532.  Assembly and bitcode changes are in the wings, but this is the
bulk of the change for the IR C++ API.

I have a follow-up patch prepared for `clang`.  If this breaks other
sub-projects, I apologize in advance :(.  Help me compile it on Darwin
I'll try to fix it.  FWIW, the errors should be easy to fix, so it may
be simpler to just fix it yourself.

This breaks the build for all metadata-related code that's out-of-tree.
Rest assured the transition is mechanical and the compiler should catch
almost all of the problems.

Here's a quick guide for updating your code:

  - `Metadata` is the root of a class hierarchy with three main classes:
    `MDNode`, `MDString`, and `ValueAsMetadata`.  It is distinct from
    the `Value` class hierarchy.  It is typeless -- i.e., instances do
    *not* have a `Type`.

  - `MDNode`'s operands are all `Metadata *` (instead of `Value *`).

  - `TrackingVH<MDNode>` and `WeakVH` referring to metadata can be
    replaced with `TrackingMDNodeRef` and `TrackingMDRef`, respectively.

    If you're referring solely to resolved `MDNode`s -- post graph
    construction -- just use `MDNode*`.

  - `MDNode` (and the rest of `Metadata`) have only limited support for
    `replaceAllUsesWith()`.

    As long as an `MDNode` is pointing at a forward declaration -- the
    result of `MDNode::getTemporary()` -- it maintains a side map of its
    uses and can RAUW itself.  Once the forward declarations are fully
    resolved RAUW support is dropped on the ground.  This means that
    uniquing collisions on changing operands cause nodes to become
    "distinct".  (This already happened fairly commonly, whenever an
    operand went to null.)

    If you're constructing complex (non self-reference) `MDNode` cycles,
    you need to call `MDNode::resolveCycles()` on each node (or on a
    top-level node that somehow references all of the nodes).  Also,
    don't do that.  Metadata cycles (and the RAUW machinery needed to
    construct them) are expensive.

  - An `MDNode` can only refer to a `Constant` through a bridge called
    `ConstantAsMetadata` (one of the subclasses of `ValueAsMetadata`).

    As a side effect, accessing an operand of an `MDNode` that is known
    to be, e.g., `ConstantInt`, takes three steps: first, cast from
    `Metadata` to `ConstantAsMetadata`; second, extract the `Constant`;
    third, cast down to `ConstantInt`.

    The eventual goal is to introduce `MDInt`/`MDFloat`/etc. and have
    metadata schema owners transition away from using `Constant`s when
    the type isn't important (and they don't care about referring to
    `GlobalValue`s).

    In the meantime, I've added transitional API to the `mdconst`
    namespace that matches semantics with the old code, in order to
    avoid adding the error-prone three-step equivalent to every call
    site.  If your old code was:

        MDNode *N = foo();
        bar(isa             <ConstantInt>(N->getOperand(0)));
        baz(cast            <ConstantInt>(N->getOperand(1)));
        bak(cast_or_null    <ConstantInt>(N->getOperand(2)));
        bat(dyn_cast        <ConstantInt>(N->getOperand(3)));
        bay(dyn_cast_or_null<ConstantInt>(N->getOperand(4)));

    you can trivially match its semantics with:

        MDNode *N = foo();
        bar(mdconst::hasa               <ConstantInt>(N->getOperand(0)));
        baz(mdconst::extract            <ConstantInt>(N->getOperand(1)));
        bak(mdconst::extract_or_null    <ConstantInt>(N->getOperand(2)));
        bat(mdconst::dyn_extract        <ConstantInt>(N->getOperand(3)));
        bay(mdconst::dyn_extract_or_null<ConstantInt>(N->getOperand(4)));

    and when you transition your metadata schema to `MDInt`:

        MDNode *N = foo();
        bar(isa             <MDInt>(N->getOperand(0)));
        baz(cast            <MDInt>(N->getOperand(1)));
        bak(cast_or_null    <MDInt>(N->getOperand(2)));
        bat(dyn_cast        <MDInt>(N->getOperand(3)));
        bay(dyn_cast_or_null<MDInt>(N->getOperand(4)));

  - A `CallInst` -- specifically, intrinsic instructions -- can refer to
    metadata through a bridge called `MetadataAsValue`.  This is a
    subclass of `Value` where `getType()->isMetadataTy()`.

    `MetadataAsValue` is the *only* class that can legally refer to a
    `LocalAsMetadata`, which is a bridged form of non-`Constant` values
    like `Argument` and `Instruction`.  It can also refer to any other
    `Metadata` subclass.

(I'll break all your testcases in a follow-up commit, when I propagate
this change to assembly.)

llvm-svn: 223802

1 files changed, 128 insertions, 47 deletions

diff --git a/llvm/lib/Transforms/Utils/ValueMapper.cpp b/llvm/lib/Transforms/Utils/ValueMapper.cpp
index a2f69d1f10b..d735e279d08 100644
--- a/llvm/lib/Transforms/Utils/ValueMapper.cpp
+++ b/llvm/lib/Transforms/Utils/ValueMapper.cpp
@@ -40,7 +40,7 @@ Value *llvm::MapValue(const Value *V, ValueToValueMapTy &VM, RemapFlags Flags,
 
   // Global values do not need to be seeded into the VM if they
   // are using the identity mapping.
-  if (isa<GlobalValue>(V) || isa<MDString>(V))
+  if (isa<GlobalValue>(V))
     return VM[V] = const_cast<Value*>(V);
   
   if (const InlineAsm *IA = dyn_cast<InlineAsm>(V)) {
@@ -56,57 +56,24 @@ Value *llvm::MapValue(const Value *V, ValueToValueMapTy &VM, RemapFlags Flags,
     
     return VM[V] = const_cast<Value*>(V);
   }
-  
 
-  if (const MDNode *MD = dyn_cast<MDNode>(V)) {
+  if (const auto *MDV = dyn_cast<MetadataAsValue>(V)) {
+    const Metadata *MD = MDV->getMetadata();
     // If this is a module-level metadata and we know that nothing at the module
     // level is changing, then use an identity mapping.
-    if (!MD->isFunctionLocal() && (Flags & RF_NoModuleLevelChanges))
-      return VM[V] = const_cast<Value*>(V);
-    
-    // Create a dummy node in case we have a metadata cycle.
-    MDNode *Dummy = MDNode::getTemporary(V->getContext(), None);
-    VM[V] = Dummy;
-    
-    // Check all operands to see if any need to be remapped.
-    for (unsigned i = 0, e = MD->getNumOperands(); i != e; ++i) {
-      Value *OP = MD->getOperand(i);
-      if (!OP) continue;
-      Value *Mapped_OP = MapValue(OP, VM, Flags, TypeMapper, Materializer);
-      // Use identity map if Mapped_Op is null and we can ignore missing
-      // entries.
-      if (Mapped_OP == OP ||
-          (Mapped_OP == nullptr && (Flags & RF_IgnoreMissingEntries)))
-        continue;
-
-      // Ok, at least one operand needs remapping.  
-      SmallVector<Value*, 4> Elts;
-      Elts.reserve(MD->getNumOperands());
-      for (i = 0; i != e; ++i) {
-        Value *Op = MD->getOperand(i);
-        if (!Op)
-          Elts.push_back(nullptr);
-        else {
-          Value *Mapped_Op = MapValue(Op, VM, Flags, TypeMapper, Materializer);
-          // Use identity map if Mapped_Op is null and we can ignore missing
-          // entries.
-          if (Mapped_Op == nullptr && (Flags & RF_IgnoreMissingEntries))
-            Mapped_Op = Op;
-          Elts.push_back(Mapped_Op);
-        }
-      }
-      MDNode *NewMD = MDNode::get(V->getContext(), Elts);
-      Dummy->replaceAllUsesWith(NewMD);
-      VM[V] = NewMD;
-      MDNode::deleteTemporary(Dummy);
-      return NewMD;
-    }
+    if (!isa<LocalAsMetadata>(MD) && (Flags & RF_NoModuleLevelChanges))
+      return VM[V] = const_cast<Value *>(V);
 
-    VM[V] = const_cast<Value*>(V);
-    MDNode::deleteTemporary(Dummy);
+    auto *MappedMD = MapValue(MD, VM, Flags, TypeMapper, Materializer);
+    if (MD == MappedMD || (!MappedMD && (Flags & RF_IgnoreMissingEntries)))
+      return VM[V] = const_cast<Value *>(V);
 
-    // No operands needed remapping.  Use an identity mapping.
-    return const_cast<Value*>(V);
+    // FIXME: This assert crashes during bootstrap, but I think it should be
+    // correct.  For now, just match behaviour from before the metadata/value
+    // split.
+    //
+    //    assert(MappedMD && "Referenced metadata value not in value map");
+    return VM[V] = MetadataAsValue::get(V->getContext(), MappedMD);
   }
 
   // Okay, this either must be a constant (which may or may not be mappable) or
@@ -177,6 +144,120 @@ Value *llvm::MapValue(const Value *V, ValueToValueMapTy &VM, RemapFlags Flags,
   return VM[V] = ConstantPointerNull::get(cast<PointerType>(NewTy));
 }
 
+static Metadata *map(ValueToValueMapTy &VM, const Metadata *Key,
+                     Metadata *Val) {
+  VM.MD()[Key].reset(Val);
+  return Val;
+}
+
+static Metadata *mapToSelf(ValueToValueMapTy &VM, const Metadata *MD) {
+  return map(VM, MD, const_cast<Metadata *>(MD));
+}
+
+static Metadata *MapValueImpl(const Metadata *MD, ValueToValueMapTy &VM,
+                              RemapFlags Flags,
+                              ValueMapTypeRemapper *TypeMapper,
+                              ValueMaterializer *Materializer) {
+  // If the value already exists in the map, use it.
+  if (Metadata *NewMD = VM.MD().lookup(MD).get())
+    return NewMD;
+
+  if (isa<MDString>(MD))
+    return mapToSelf(VM, MD);
+
+  if (isa<ConstantAsMetadata>(MD))
+    if ((Flags & RF_NoModuleLevelChanges))
+      return mapToSelf(VM, MD);
+
+  if (const auto *VMD = dyn_cast<ValueAsMetadata>(MD)) {
+    Value *MappedV =
+        MapValue(VMD->getValue(), VM, Flags, TypeMapper, Materializer);
+    if (VMD->getValue() == MappedV ||
+        (!MappedV && (Flags & RF_IgnoreMissingEntries)))
+      return mapToSelf(VM, MD);
+
+    // FIXME: This assert crashes during bootstrap, but I think it should be
+    // correct.  For now, just match behaviour from before the metadata/value
+    // split.
+    //
+    //    assert(MappedV && "Referenced metadata not in value map!");
+    if (MappedV)
+      return map(VM, MD, ValueAsMetadata::get(MappedV));
+    return nullptr;
+  }
+
+  const MDNode *Node = cast<MDNode>(MD);
+  assert(Node->isResolved() && "Unexpected unresolved node");
+
+  auto getMappedOp = [&](Metadata *Op) -> Metadata *{
+    if (!Op)
+      return nullptr;
+    if (Metadata *MappedOp =
+            MapValueImpl(Op, VM, Flags, TypeMapper, Materializer))
+      return MappedOp;
+    // Use identity map if MappedOp is null and we can ignore missing entries.
+    if (Flags & RF_IgnoreMissingEntries)
+      return Op;
+
+    // FIXME: This assert crashes during bootstrap, but I think it should be
+    // correct.  For now, just match behaviour from before the metadata/value
+    // split.
+    //
+    //    llvm_unreachable("Referenced metadata not in value map!");
+    return nullptr;
+  };
+
+  // If this is a module-level metadata and we know that nothing at the
+  // module level is changing, then use an identity mapping.
+  if (Flags & RF_NoModuleLevelChanges)
+    return mapToSelf(VM, MD);
+
+  // Create a dummy node in case we have a metadata cycle.
+  MDNodeFwdDecl *Dummy = MDNode::getTemporary(Node->getContext(), None);
+  map(VM, Node, Dummy);
+
+  // Check all operands to see if any need to be remapped.
+  for (unsigned I = 0, E = Node->getNumOperands(); I != E; ++I) {
+    Metadata *Op = Node->getOperand(I);
+    Metadata *MappedOp = getMappedOp(Op);
+    if (Op == MappedOp)
+      continue;
+
+    // Ok, at least one operand needs remapping.
+    SmallVector<Metadata *, 4> Elts;
+    Elts.reserve(Node->getNumOperands());
+    for (I = 0; I != E; ++I)
+      Elts.push_back(getMappedOp(Node->getOperand(I)));
+
+    MDNode *NewMD = MDNode::get(Node->getContext(), Elts);
+    Dummy->replaceAllUsesWith(NewMD);
+    MDNode::deleteTemporary(Dummy);
+    return map(VM, Node, NewMD);
+  }
+
+  // No operands needed remapping.  Use an identity mapping.
+  mapToSelf(VM, MD);
+  MDNode::deleteTemporary(Dummy);
+  return const_cast<Metadata *>(MD);
+}
+
+Metadata *llvm::MapValue(const Metadata *MD, ValueToValueMapTy &VM,
+                         RemapFlags Flags, ValueMapTypeRemapper *TypeMapper,
+                         ValueMaterializer *Materializer) {
+  Metadata *NewMD = MapValueImpl(MD, VM, Flags, TypeMapper, Materializer);
+  if (NewMD && NewMD != MD)
+    if (auto *G = dyn_cast<GenericMDNode>(NewMD))
+      G->resolveCycles();
+  return NewMD;
+}
+
+MDNode *llvm::MapValue(const MDNode *MD, ValueToValueMapTy &VM,
+                       RemapFlags Flags, ValueMapTypeRemapper *TypeMapper,
+                       ValueMaterializer *Materializer) {
+  return cast<MDNode>(MapValue(static_cast<const Metadata *>(MD), VM, Flags,
+                               TypeMapper, Materializer));
+}
+
 /// RemapInstruction - Convert the instruction operands from referencing the
 /// current values into those specified by VMap.
 ///