[IR] Introduce the opaque pointer type

The opaque pointer type is essentially just a normal pointer type with a
null pointee type.

This also adds support for the opaque pointer type to the bitcode
reader/writer, as well as to textual IR.

To avoid confusion with existing pointer types, we disallow creating a
pointer to an opaque pointer.

Opaque pointer types should not be widely used at this point since many
parts of LLVM still do not support them. The next steps are to add some
very simple use cases of opaque pointers to make sure they work, then
start pretending that all pointers are opaque pointers and see what
breaks.

https://lists.llvm.org/pipermail/llvm-dev/2021-May/150359.html

Reviewed By: dblaikie, dexonsmith, pcc

Differential Revision: https://reviews.llvm.org/D101704

llvm/docs/LangRef.rst

@@ -3307,11 +3307,17 @@ are target-specific.
 Note that LLVM does not permit pointers to void (``void*``) nor does it
 permit pointers to labels (``label*``). Use ``i8*`` instead.
 
+LLVM is in the process of transitioning to opaque pointers. Opaque pointers do
+not have a pointee type. Rather, instructions interacting through pointers
+specify the type of the underlying memory they are interacting with. Opaque
+pointers are still in the process of being worked on and are not complete.
+
 :Syntax:
 
 ::
 
       <type> *
+      ptr
 
 :Examples:
 
@@ -3320,7 +3326,11 @@ permit pointers to labels (``label*``). Use ``i8*`` instead.
 +-------------------------+--------------------------------------------------------------------------------------------------------------+
 | ``i32 (i32*) *``        | A :ref:`pointer <t_pointer>` to a :ref:`function <t_function>` that takes an ``i32*``, returning an ``i32``. |
 +-------------------------+--------------------------------------------------------------------------------------------------------------+
-| ``i32 addrspace(5)*``   | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space #5.                           |
+| ``i32 addrspace(5)*``   | A :ref:`pointer <t_pointer>` to an ``i32`` value that resides in address space 5.                            |
++-------------------------+--------------------------------------------------------------------------------------------------------------+
+| ``ptr``                 | An opaque pointer type to a value that resides in address space 0.                                           |
++-------------------------+--------------------------------------------------------------------------------------------------------------+
+| ``ptr addrspace(5)``    | An opaque pointer type to a value that resides in address space 5.                                           |
 +-------------------------+--------------------------------------------------------------------------------------------------------------+
 
 .. _t_vector:

llvm/docs/ReleaseNotes.rst

@@ -61,6 +61,9 @@ Changes to the LLVM IR
 * The ``inalloca`` attribute now has a mandatory type field, similar
   to ``byval`` and ``sret``.
 
+* The opaque pointer type ``ptr`` has been introduced. It is still in the
+  process of being worked on and should not be used yet.
+
 Changes to building LLVM
 ------------------------
 

llvm/include/llvm/Bitcode/LLVMBitCodes.h

@@ -168,8 +168,10 @@ enum TypeCodes {
 
   TYPE_CODE_TOKEN = 22, // TOKEN
 
-  TYPE_CODE_BFLOAT = 23, // BRAIN FLOATING POINT
-  TYPE_CODE_X86_AMX = 24 // X86 AMX
+  TYPE_CODE_BFLOAT = 23,  // BRAIN FLOATING POINT
+  TYPE_CODE_X86_AMX = 24, // X86 AMX
+
+  TYPE_CODE_OPAQUE_POINTER = 25, // OPAQUE_POINTER: [addrspace]
 };
 
 enum OperandBundleTagCode {

llvm/include/llvm/IR/DerivedTypes.h

@@ -633,6 +633,7 @@ inline ElementCount VectorType::getElementCount() const {
 /// Class to represent pointers.
 class PointerType : public Type {
   explicit PointerType(Type *ElType, unsigned AddrSpace);
+  explicit PointerType(LLVMContext &C, unsigned AddrSpace);
 
   Type *PointeeTy;
 
@@ -643,14 +644,28 @@ class PointerType : public Type {
   /// This constructs a pointer to an object of the specified type in a numbered
   /// address space.
   static PointerType *get(Type *ElementType, unsigned AddressSpace);
+  /// This constructs an opaque pointer to an object in a numbered address
+  /// space.
+  static PointerType *get(LLVMContext &C, unsigned AddressSpace);
 
   /// This constructs a pointer to an object of the specified type in the
-  /// generic address space (address space zero).
+  /// default address space (address space zero).
   static PointerType *getUnqual(Type *ElementType) {
     return PointerType::get(ElementType, 0);
   }
 
-  Type *getElementType() const { return PointeeTy; }
+  /// This constructs an opaque pointer to an object in the
+  /// default address space (address space zero).
+  static PointerType *getUnqual(LLVMContext &C) {
+    return PointerType::get(C, 0);
+  }
+
+  Type *getElementType() const {
+    assert(!isOpaque() && "Attempting to get element type of opaque pointer");
+    return PointeeTy;
+  }
+
+  bool isOpaque() const { return !PointeeTy; }
 
   /// Return true if the specified type is valid as a element type.
   static bool isValidElementType(Type *ElemTy);

llvm/lib/AsmParser/LLLexer.cpp

@@ -845,6 +845,7 @@ lltok::Kind LLLexer::LexIdentifier() {
   TYPEKEYWORD("x86_mmx",   Type::getX86_MMXTy(Context));
   TYPEKEYWORD("x86_amx",   Type::getX86_AMXTy(Context));
   TYPEKEYWORD("token",     Type::getTokenTy(Context));
+  TYPEKEYWORD("ptr", PointerType::getUnqual(Context));
 
 #undef TYPEKEYWORD
 

llvm/lib/AsmParser/LLParser.cpp

@@ -2595,6 +2595,13 @@ bool LLParser::parseType(Type *&Result, const Twine &Msg, bool AllowVoid) {
   }
   }
 
+  if (Result->isPointerTy() && cast<PointerType>(Result)->isOpaque()) {
+    unsigned AddrSpace;
+    if (parseOptionalAddrSpace(AddrSpace))
+      return true;
+    Result = PointerType::get(getContext(), AddrSpace);
+  }
+
   // parse the type suffixes.
   while (true) {
     switch (Lex.getKind()) {

llvm/lib/Bitcode/Reader/BitcodeReader.cpp

@@ -1807,6 +1807,13 @@ Error BitcodeReader::parseTypeTableBody() {
       ResultTy = PointerType::get(ResultTy, AddressSpace);
       break;
     }
+    case bitc::TYPE_CODE_OPAQUE_POINTER: { // OPAQUE_POINTER: [addrspace]
+      if (Record.size() != 1)
+        return error("Invalid record");
+      unsigned AddressSpace = Record[0];
+      ResultTy = PointerType::get(Context, AddressSpace);
+      break;
+    }
     case bitc::TYPE_CODE_FUNCTION_OLD: {
       // Deprecated, but still needed to read old bitcode files.
       // FUNCTION: [vararg, attrid, retty, paramty x N]

llvm/lib/Bitcode/Writer/BitcodeWriter.cpp

@@ -858,6 +858,12 @@ void ModuleBitcodeWriter::writeTypeTable() {
   Abbv->Add(BitCodeAbbrevOp(0));  // Addrspace = 0
   unsigned PtrAbbrev = Stream.EmitAbbrev(std::move(Abbv));
 
+  // Abbrev for TYPE_CODE_OPAQUE_POINTER.
+  Abbv = std::make_shared<BitCodeAbbrev>();
+  Abbv->Add(BitCodeAbbrevOp(bitc::TYPE_CODE_OPAQUE_POINTER));
+  Abbv->Add(BitCodeAbbrevOp(0)); // Addrspace = 0
+  unsigned OpaquePtrAbbrev = Stream.EmitAbbrev(std::move(Abbv));
+
   // Abbrev for TYPE_CODE_FUNCTION.
   Abbv = std::make_shared<BitCodeAbbrev>();
   Abbv->Add(BitCodeAbbrevOp(bitc::TYPE_CODE_FUNCTION));
@@ -928,12 +934,21 @@ void ModuleBitcodeWriter::writeTypeTable() {
       break;
     case Type::PointerTyID: {
       PointerType *PTy = cast<PointerType>(T);
-      // POINTER: [pointee type, address space]
-      Code = bitc::TYPE_CODE_POINTER;
-      TypeVals.push_back(VE.getTypeID(PTy->getElementType()));
       unsigned AddressSpace = PTy->getAddressSpace();
-      TypeVals.push_back(AddressSpace);
-      if (AddressSpace == 0) AbbrevToUse = PtrAbbrev;
+      if (PTy->isOpaque()) {
+        // OPAQUE_POINTER: [address space]
+        Code = bitc::TYPE_CODE_OPAQUE_POINTER;
+        TypeVals.push_back(AddressSpace);
+        if (AddressSpace == 0)
+          AbbrevToUse = OpaquePtrAbbrev;
+      } else {
+        // POINTER: [pointee type, address space]
+        Code = bitc::TYPE_CODE_POINTER;
+        TypeVals.push_back(VE.getTypeID(PTy->getElementType()));
+        TypeVals.push_back(AddressSpace);
+        if (AddressSpace == 0)
+          AbbrevToUse = PtrAbbrev;
+      }
       break;
     }
     case Type::FunctionTyID: {

llvm/lib/IR/AsmWriter.cpp

@@ -651,6 +651,12 @@ void TypePrinting::print(Type *Ty, raw_ostream &OS) {
   }
   case Type::PointerTyID: {
     PointerType *PTy = cast<PointerType>(Ty);
+    if (PTy->isOpaque()) {
+      OS << "ptr";
+      if (unsigned AddressSpace = PTy->getAddressSpace())
+        OS << " addrspace(" << AddressSpace << ')';
+      return;
+    }
     print(PTy->getElementType(), OS);
     if (unsigned AddressSpace = PTy->getAddressSpace())
       OS << " addrspace(" << AddressSpace << ')';

llvm/lib/IR/LLVMContextImpl.h

@@ -1445,6 +1445,8 @@ class LLVMContextImpl {
 
   DenseMap<std::pair<Type *, uint64_t>, ArrayType*> ArrayTypes;
   DenseMap<std::pair<Type *, ElementCount>, VectorType*> VectorTypes;
+  // TODO: clean up the following after we no longer support non-opaque pointer
+  // types.
   DenseMap<Type*, PointerType*> PointerTypes;  // Pointers in AddrSpace = 0
   DenseMap<std::pair<Type*, unsigned>, PointerType*> ASPointerTypes;
 

llvm/lib/IR/Type.cpp

@@ -699,21 +699,41 @@ PointerType *PointerType::get(Type *EltTy, unsigned AddressSpace) {
   return Entry;
 }
 
+PointerType *PointerType::get(LLVMContext &C, unsigned AddressSpace) {
+  LLVMContextImpl *CImpl = C.pImpl;
+
+  // Since AddressSpace #0 is the common case, we special case it.
+  PointerType *&Entry =
+      AddressSpace == 0
+          ? CImpl->PointerTypes[nullptr]
+          : CImpl->ASPointerTypes[std::make_pair(nullptr, AddressSpace)];
+
+  if (!Entry)
+    Entry = new (CImpl->Alloc) PointerType(C, AddressSpace);
+  return Entry;
+}
+
 PointerType::PointerType(Type *E, unsigned AddrSpace)
   : Type(E->getContext(), PointerTyID), PointeeTy(E) {
   ContainedTys = &PointeeTy;
   NumContainedTys = 1;
   setSubclassData(AddrSpace);
 }
 
+PointerType::PointerType(LLVMContext &C, unsigned AddrSpace)
+    : Type(C, PointerTyID), PointeeTy(nullptr) {
+  setSubclassData(AddrSpace);
+}
+
 PointerType *Type::getPointerTo(unsigned addrs) const {
   return PointerType::get(const_cast<Type*>(this), addrs);
 }
 
 bool PointerType::isValidElementType(Type *ElemTy) {
   return !ElemTy->isVoidTy() && !ElemTy->isLabelTy() &&
          !ElemTy->isMetadataTy() && !ElemTy->isTokenTy() &&
-         !ElemTy->isX86_AMXTy();
+         !ElemTy->isX86_AMXTy() &&
+         !(ElemTy->isPointerTy() && cast<PointerType>(ElemTy)->isOpaque());
 }
 
 bool PointerType::isLoadableOrStorableType(Type *ElemTy) {

llvm/test/Assembler/invalid-opaque-ptr.ll

@@ -0,0 +1,7 @@
+; RUN: not llvm-as < %s -disable-output 2>&1 | FileCheck %s
+
+; CHECK: pointer to this type is invalid
+define void @f(ptr %a) {
+    %b = bitcast ptr %a to ptr*
+    ret void
+}

llvm/test/Assembler/opaque-ptr.ll

@@ -0,0 +1,26 @@
+; RUN: llvm-as < %s | llvm-dis | llvm-as | llvm-dis | FileCheck %s
+; RUN: verify-uselistorder %s
+
+; CHECK: define ptr @f(ptr %a) {
+; CHECK:     %b = bitcast ptr %a to ptr
+; CHECK:     ret ptr %b
+define ptr @f(ptr %a) {
+    %b = bitcast ptr %a to ptr
+    ret ptr %b
+}
+
+; CHECK: define ptr @g(ptr addrspace(2) %a) {
+; CHECK:     %b = addrspacecast ptr addrspace(2) %a to ptr
+; CHECK:     ret ptr %b
+define ptr @g(ptr addrspace(2) %a) {
+    %b = addrspacecast ptr addrspace(2) %a to ptr addrspace(0)
+    ret ptr addrspace(0) %b
+}
+
+; CHECK: define ptr addrspace(2) @g2(ptr %a) {
+; CHECK:     %b = addrspacecast ptr %a to ptr addrspace(2)
+; CHECK:     ret ptr addrspace(2) %b
+define ptr addrspace(2) @g2(ptr addrspace(0) %a) {
+    %b = addrspacecast ptr addrspace(0) %a to ptr addrspace(2)
+    ret ptr addrspace(2) %b
+}