Initial implementation of 'fence' instruction, the new C++0x-style replacement for llvm.memory.barrier.

This is just a LangRef entry and reading/writing/memory representation; optimizer+codegen support coming soon.

llvm-svn: 136009

docs/LangRef.html

@@ -170,6 +170,7 @@
           <li><a href="#i_alloca">'<tt>alloca</tt>'   Instruction</a></li>
          <li><a href="#i_load">'<tt>load</tt>'     Instruction</a></li>
          <li><a href="#i_store">'<tt>store</tt>'    Instruction</a></li>
+         <li><a href="#i_fence">'<tt>fence</tt>'    Instruction</a></li>
          <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
         </ol>
       </li>
@@ -4551,6 +4552,63 @@ that the invoke/unwind semantics are likely to change in future versions.</p>
 
 </div>
 
+<!-- _______________________________________________________________________ -->
+<div class="doc_subsubsection"> <a name="i_fence">'<tt>fence</tt>'
+Instruction</a> </div>
+
+<div class="doc_text">
+
+<h5>Syntax:</h5>
+<pre>
+  fence [singlethread] &lt;ordering&gt;                   <i>; yields {void}</i>
+</pre>
+
+<h5>Overview:</h5>
+<p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
+between operations.</p>
+
+<h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
+href="#ordering">ordering</a> argument which defines what
+<i>synchronizes-with</i> edges they add.  They can only be given
+<code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
+<code>seq_cst</code> orderings.</p>
+
+<h5>Semantics:</h5>
+<p>A fence <var>A</var> which has (at least) <code>release</code> ordering
+semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
+<code>acquire</code> ordering semantics if and only if there exist atomic
+operations <var>X</var> and <var>Y</var>, both operating on some atomic object
+<var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
+<var>X</var> modifies <var>M</var> (either directly or through some side effect
+of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
+<var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
+<i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
+than an explicit <code>fence</code>, one (but not both) of the atomic operations
+<var>X</var> or <var>Y</var> might provide a <code>release</code> or
+<code>acquire</code> (resp.) ordering constraint and still
+<i>synchronize-with</i> the explicit <code>fence</code> and establish the
+<i>happens-before</i> edge.</p>
+
+<p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
+having both <code>acquire</code> and <code>release</code> semantics specified
+above, participates in the global program order of other <code>seq_cst</code>
+operations and/or fences.</p>
+
+<p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
+specifies that the fence only synchronizes with other fences in the same
+thread.  (This is useful for interacting with signal handlers.)</p>
+
+<p>FIXME: This instruction is a work in progress; until it is finished, use
+   llvm.memory.barrier.
+
+<h5>Example:</h5>
+<pre>
+  fence acquire                          <i>; yields {void}</i>
+  fence singlethread seq_cst             <i>; yields {void}</i>
+</pre>
+
+</div>
+
 <!-- _______________________________________________________________________ -->
 <h4>
    <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>

include/llvm/Bitcode/LLVMBitCodes.h

@@ -218,6 +218,23 @@ namespace bitc {
     PEO_EXACT = 0
   };
 
+  /// Encoded AtomicOrdering values.
+  enum AtomicOrderingCodes {
+    ORDERING_NOTATOMIC = 0,
+    ORDERING_UNORDERED = 1,
+    ORDERING_MONOTONIC = 2,
+    ORDERING_ACQUIRE = 3,
+    ORDERING_RELEASE = 4,
+    ORDERING_ACQREL = 5,
+    ORDERING_SEQCST = 6
+  };
+
+  /// Encoded SynchronizationScope values.
+  enum AtomicSynchScopeCodes {
+    SYNCHSCOPE_SINGLETHREAD = 0,
+    SYNCHSCOPE_CROSSTHREAD = 1
+  };
+
   // The function body block (FUNCTION_BLOCK_ID) describes function bodies.  It
   // can contain a constant block (CONSTANTS_BLOCK_ID).
   enum FunctionCodes {
@@ -266,7 +283,8 @@ namespace bitc {
 
     FUNC_CODE_INST_CALL        = 34, // CALL:       [attr, fnty, fnid, args...]
 
-    FUNC_CODE_DEBUG_LOC        = 35  // DEBUG_LOC:  [Line,Col,ScopeVal, IAVal]
+    FUNC_CODE_DEBUG_LOC        = 35, // DEBUG_LOC:  [Line,Col,ScopeVal, IAVal]
+    FUNC_CODE_INST_FENCE       = 36  // FENCE: [ordering, synchscope]
   };
 } // End bitc namespace
 } // End llvm namespace

include/llvm/Instruction.def

@@ -133,43 +133,44 @@ HANDLE_MEMORY_INST(26, Alloca, AllocaInst)  // Stack management
 HANDLE_MEMORY_INST(27, Load  , LoadInst  )  // Memory manipulation instrs
 HANDLE_MEMORY_INST(28, Store , StoreInst )
 HANDLE_MEMORY_INST(29, GetElementPtr, GetElementPtrInst)
-  LAST_MEMORY_INST(29)
+HANDLE_MEMORY_INST(30, Fence , FenceInst )
+  LAST_MEMORY_INST(32)
 
 // Cast operators ...
 // NOTE: The order matters here because CastInst::isEliminableCastPair 
 // NOTE: (see Instructions.cpp) encodes a table based on this ordering.
- FIRST_CAST_INST(30)
+ FIRST_CAST_INST(33)
-HANDLE_CAST_INST(30, Trunc   , TruncInst   )  // Truncate integers
+HANDLE_CAST_INST(33, Trunc   , TruncInst   )  // Truncate integers
-HANDLE_CAST_INST(31, ZExt    , ZExtInst    )  // Zero extend integers
+HANDLE_CAST_INST(34, ZExt    , ZExtInst    )  // Zero extend integers
-HANDLE_CAST_INST(32, SExt    , SExtInst    )  // Sign extend integers
+HANDLE_CAST_INST(35, SExt    , SExtInst    )  // Sign extend integers
-HANDLE_CAST_INST(33, FPToUI  , FPToUIInst  )  // floating point -> UInt
+HANDLE_CAST_INST(36, FPToUI  , FPToUIInst  )  // floating point -> UInt
-HANDLE_CAST_INST(34, FPToSI  , FPToSIInst  )  // floating point -> SInt
+HANDLE_CAST_INST(37, FPToSI  , FPToSIInst  )  // floating point -> SInt
-HANDLE_CAST_INST(35, UIToFP  , UIToFPInst  )  // UInt -> floating point
+HANDLE_CAST_INST(38, UIToFP  , UIToFPInst  )  // UInt -> floating point
-HANDLE_CAST_INST(36, SIToFP  , SIToFPInst  )  // SInt -> floating point
+HANDLE_CAST_INST(39, SIToFP  , SIToFPInst  )  // SInt -> floating point
-HANDLE_CAST_INST(37, FPTrunc , FPTruncInst )  // Truncate floating point
+HANDLE_CAST_INST(40, FPTrunc , FPTruncInst )  // Truncate floating point
-HANDLE_CAST_INST(38, FPExt   , FPExtInst   )  // Extend floating point
+HANDLE_CAST_INST(41, FPExt   , FPExtInst   )  // Extend floating point
-HANDLE_CAST_INST(39, PtrToInt, PtrToIntInst)  // Pointer -> Integer
+HANDLE_CAST_INST(42, PtrToInt, PtrToIntInst)  // Pointer -> Integer
-HANDLE_CAST_INST(40, IntToPtr, IntToPtrInst)  // Integer -> Pointer
+HANDLE_CAST_INST(43, IntToPtr, IntToPtrInst)  // Integer -> Pointer
-HANDLE_CAST_INST(41, BitCast , BitCastInst )  // Type cast
+HANDLE_CAST_INST(44, BitCast , BitCastInst )  // Type cast
-  LAST_CAST_INST(41)
+  LAST_CAST_INST(44)
 
 // Other operators...
- FIRST_OTHER_INST(42)
+ FIRST_OTHER_INST(45)
-HANDLE_OTHER_INST(42, ICmp   , ICmpInst   )  // Integer comparison instruction
+HANDLE_OTHER_INST(45, ICmp   , ICmpInst   )  // Integer comparison instruction
-HANDLE_OTHER_INST(43, FCmp   , FCmpInst   )  // Floating point comparison instr.
+HANDLE_OTHER_INST(46, FCmp   , FCmpInst   )  // Floating point comparison instr.
-HANDLE_OTHER_INST(44, PHI    , PHINode    )  // PHI node instruction
+HANDLE_OTHER_INST(47, PHI    , PHINode    )  // PHI node instruction
-HANDLE_OTHER_INST(45, Call   , CallInst   )  // Call a function
+HANDLE_OTHER_INST(48, Call   , CallInst   )  // Call a function
-HANDLE_OTHER_INST(46, Select , SelectInst )  // select instruction
+HANDLE_OTHER_INST(49, Select , SelectInst )  // select instruction
-HANDLE_OTHER_INST(47, UserOp1, Instruction)  // May be used internally in a pass
+HANDLE_OTHER_INST(50, UserOp1, Instruction)  // May be used internally in a pass
-HANDLE_OTHER_INST(48, UserOp2, Instruction)  // Internal to passes only
+HANDLE_OTHER_INST(51, UserOp2, Instruction)  // Internal to passes only
-HANDLE_OTHER_INST(49, VAArg  , VAArgInst  )  // vaarg instruction
+HANDLE_OTHER_INST(52, VAArg  , VAArgInst  )  // vaarg instruction
-HANDLE_OTHER_INST(50, ExtractElement, ExtractElementInst)// extract from vector
+HANDLE_OTHER_INST(53, ExtractElement, ExtractElementInst)// extract from vector
-HANDLE_OTHER_INST(51, InsertElement, InsertElementInst)  // insert into vector
+HANDLE_OTHER_INST(54, InsertElement, InsertElementInst)  // insert into vector
-HANDLE_OTHER_INST(52, ShuffleVector, ShuffleVectorInst)  // shuffle two vectors.
+HANDLE_OTHER_INST(55, ShuffleVector, ShuffleVectorInst)  // shuffle two vectors.
-HANDLE_OTHER_INST(53, ExtractValue, ExtractValueInst)// extract from aggregate
+HANDLE_OTHER_INST(56, ExtractValue, ExtractValueInst)// extract from aggregate
-HANDLE_OTHER_INST(54, InsertValue, InsertValueInst)  // insert into aggregate
+HANDLE_OTHER_INST(57, InsertValue, InsertValueInst)  // insert into aggregate
 
-  LAST_OTHER_INST(54)
+  LAST_OTHER_INST(57)
 
 #undef  FIRST_TERM_INST
 #undef HANDLE_TERM_INST

include/llvm/Instructions.h

@@ -22,6 +22,7 @@
 #include "llvm/CallingConv.h"
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/SmallVector.h"
+#include "llvm/Support/ErrorHandling.h"
 #include <iterator>
 
 namespace llvm {
@@ -31,6 +32,22 @@ class ConstantRange;
 class APInt;
 class LLVMContext;
 
+enum AtomicOrdering {
+  NotAtomic = 0,
+  Unordered = 1,
+  Monotonic = 2,
+  // Consume = 3,  // Not specified yet.
+  Acquire = 4,
+  Release = 5,
+  AcquireRelease = 6,
+  SequentiallyConsistent = 7
+};
+
+enum SynchronizationScope {
+  SingleThread = 0,
+  CrossThread = 1
+};
+
 //===----------------------------------------------------------------------===//
 //                                AllocaInst Class
 //===----------------------------------------------------------------------===//
@@ -268,6 +285,82 @@ struct OperandTraits<StoreInst> : public FixedNumOperandTraits<StoreInst, 2> {
 
 DEFINE_TRANSPARENT_OPERAND_ACCESSORS(StoreInst, Value)
 
+//===----------------------------------------------------------------------===//
+//                                FenceInst Class
+//===----------------------------------------------------------------------===//
+
+/// FenceInst - an instruction for ordering other memory operations
+///
+class FenceInst : public Instruction {
+  void *operator new(size_t, unsigned);  // DO NOT IMPLEMENT
+  void Init(AtomicOrdering Ordering, SynchronizationScope SynchScope);
+protected:
+  virtual FenceInst *clone_impl() const;
+public:
+  // allocate space for exactly zero operands
+  void *operator new(size_t s) {
+    return User::operator new(s, 0);
+  }
+
+  // Ordering may only be Acquire, Release, AcquireRelease, or
+  // SequentiallyConsistent.
+  FenceInst(LLVMContext &C, AtomicOrdering Ordering,
+            SynchronizationScope SynchScope = CrossThread,
+            Instruction *InsertBefore = 0);
+  FenceInst(LLVMContext &C, AtomicOrdering Ordering,
+            SynchronizationScope SynchScope,
+            BasicBlock *InsertAtEnd);
+
+  /// Returns the ordering effect of this fence.
+  AtomicOrdering getOrdering() const {
+    return AtomicOrdering(getSubclassDataFromInstruction() >> 1);
+  }
+
+  /// Set the ordering constraint on this fence.  May only be Acquire, Release,
+  /// AcquireRelease, or SequentiallyConsistent.
+  void setOrdering(AtomicOrdering Ordering) {
+    switch (Ordering) {
+    case Acquire:
+    case Release:
+    case AcquireRelease:
+    case SequentiallyConsistent:
+      setInstructionSubclassData((getSubclassDataFromInstruction() & 1) |
+                                 (Ordering << 1));
+      return;
+    default:
+      llvm_unreachable("FenceInst ordering must be Acquire, Release,"
+                       " AcquireRelease, or SequentiallyConsistent");
+    }
+  }
+
+  SynchronizationScope getSynchScope() const {
+    return SynchronizationScope(getSubclassDataFromInstruction() & 1);
+  }
+
+  /// Specify whether this fence orders other operations with respect to all
+  /// concurrently executing threads, or only with respect to signal handlers
+  /// executing in the same thread.
+  void setSynchScope(SynchronizationScope xthread) {
+    setInstructionSubclassData((getSubclassDataFromInstruction() & ~1) |
+                               xthread);
+  }
+
+  // Methods for support type inquiry through isa, cast, and dyn_cast:
+  static inline bool classof(const FenceInst *) { return true; }
+  static inline bool classof(const Instruction *I) {
+    return I->getOpcode() == Instruction::Fence;
+  }
+  static inline bool classof(const Value *V) {
+    return isa<Instruction>(V) && classof(cast<Instruction>(V));
+  }
+private:
+  // Shadow Instruction::setInstructionSubclassData with a private forwarding
+  // method so that subclasses cannot accidentally use it.
+  void setInstructionSubclassData(unsigned short D) {
+    Instruction::setInstructionSubclassData(D);
+  }
+};
+
 //===----------------------------------------------------------------------===//
 //                             GetElementPtrInst Class
 //===----------------------------------------------------------------------===//

include/llvm/Support/IRBuilder.h

@@ -762,6 +762,10 @@ public:
   StoreInst *CreateStore(Value *Val, Value *Ptr, bool isVolatile = false) {
     return Insert(new StoreInst(Val, Ptr, isVolatile));
   }
+  FenceInst *CreateFence(AtomicOrdering Ordering,
+                         SynchronizationScope SynchScope = CrossThread) {
+    return Insert(new FenceInst(Context, Ordering, SynchScope));
+  }
   Value *CreateGEP(Value *Ptr, ArrayRef<Value *> IdxList,
                    const Twine &Name = "") {
     if (Constant *PC = dyn_cast<Constant>(Ptr)) {

include/llvm/Support/InstVisitor.h

@@ -169,6 +169,7 @@ public:
   RetTy visitAllocaInst(AllocaInst &I)              { DELEGATE(Instruction); }
   RetTy visitLoadInst(LoadInst     &I)              { DELEGATE(Instruction); }
   RetTy visitStoreInst(StoreInst   &I)              { DELEGATE(Instruction); }
+  RetTy visitFenceInst(FenceInst   &I)              { DELEGATE(Instruction); }
   RetTy visitGetElementPtrInst(GetElementPtrInst &I){ DELEGATE(Instruction); }
   RetTy visitPHINode(PHINode       &I)              { DELEGATE(Instruction); }
   RetTy visitTruncInst(TruncInst &I)                { DELEGATE(CastInst); }

lib/AsmParser/LLLexer.cpp

@@ -506,6 +506,15 @@ lltok::Kind LLLexer::LexIdentifier() {
   KEYWORD(deplibs);
   KEYWORD(datalayout);
   KEYWORD(volatile);
+  KEYWORD(atomic);
+  KEYWORD(unordered);
+  KEYWORD(monotonic);
+  KEYWORD(acquire);
+  KEYWORD(release);
+  KEYWORD(acq_rel);
+  KEYWORD(seq_cst);
+  KEYWORD(singlethread);
+
   KEYWORD(nuw);
   KEYWORD(nsw);
   KEYWORD(exact);
@@ -630,6 +639,7 @@ lltok::Kind LLLexer::LexIdentifier() {
   INSTKEYWORD(alloca,      Alloca);
   INSTKEYWORD(load,        Load);
   INSTKEYWORD(store,       Store);
+  INSTKEYWORD(fence,       Fence);
   INSTKEYWORD(getelementptr, GetElementPtr);
 
   INSTKEYWORD(extractelement, ExtractElement);

lib/AsmParser/LLParser.cpp

@@ -1145,6 +1145,32 @@ bool LLParser::ParseOptionalCommaAlign(unsigned &Alignment,
   return false;
 }
 
+/// ParseScopeAndOrdering
+///   if isAtomic: ::= 'singlethread'? AtomicOrdering
+///   else: ::=
+///
+/// This sets Scope and Ordering to the parsed values.
+bool LLParser::ParseScopeAndOrdering(bool isAtomic, SynchronizationScope &Scope,
+                                     AtomicOrdering &Ordering) {
+  if (!isAtomic)
+    return false;
+
+  Scope = CrossThread;
+  if (EatIfPresent(lltok::kw_singlethread))
+    Scope = SingleThread;
+  switch (Lex.getKind()) {
+  default: return TokError("Expected ordering on atomic instruction");
+  case lltok::kw_unordered: Ordering = Unordered; break;
+  case lltok::kw_monotonic: Ordering = Monotonic; break;
+  case lltok::kw_acquire: Ordering = Acquire; break;
+  case lltok::kw_release: Ordering = Release; break;
+  case lltok::kw_acq_rel: Ordering = AcquireRelease; break;
+  case lltok::kw_seq_cst: Ordering = SequentiallyConsistent; break;
+  }
+  Lex.Lex();
+  return false;
+}
+
 /// ParseOptionalStackAlignment
 ///   ::= /* empty */
 ///   ::= 'alignstack' '(' 4 ')'
@@ -2924,6 +2950,7 @@ int LLParser::ParseInstruction(Instruction *&Inst, BasicBlock *BB,
   case lltok::kw_alloca:         return ParseAlloc(Inst, PFS);
   case lltok::kw_load:           return ParseLoad(Inst, PFS, false);
   case lltok::kw_store:          return ParseStore(Inst, PFS, false);
+  case lltok::kw_fence:          return ParseFence(Inst, PFS);
   case lltok::kw_volatile:
     if (EatIfPresent(lltok::kw_load))
       return ParseLoad(Inst, PFS, true);
@@ -3633,6 +3660,23 @@ int LLParser::ParseStore(Instruction *&Inst, PerFunctionState &PFS,
   return AteExtraComma ? InstExtraComma : InstNormal;
 }
 
+/// ParseFence
+///   ::= 'fence' 'singlethread'? AtomicOrdering
+int LLParser::ParseFence(Instruction *&Inst, PerFunctionState &PFS) {
+  AtomicOrdering Ordering = NotAtomic;
+  SynchronizationScope Scope = CrossThread;
+  if (ParseScopeAndOrdering(true /*Always atomic*/, Scope, Ordering))
+    return true;
+
+  if (Ordering == Unordered)
+    return TokError("fence cannot be unordered");
+  if (Ordering == Monotonic)
+    return TokError("fence cannot be monotonic");
+
+  Inst = new FenceInst(Context, Ordering, Scope);
+  return InstNormal;
+}
+
 /// ParseGetElementPtr
 ///   ::= 'getelementptr' 'inbounds'? TypeAndValue (',' TypeAndValue)*
 int LLParser::ParseGetElementPtr(Instruction *&Inst, PerFunctionState &PFS) {

lib/AsmParser/LLParser.h

@@ -15,6 +15,7 @@
 #define LLVM_ASMPARSER_LLPARSER_H
 
 #include "LLLexer.h"
+#include "llvm/Instructions.h"
 #include "llvm/Module.h"
 #include "llvm/Type.h"
 #include "llvm/ADT/DenseMap.h"
@@ -178,6 +179,8 @@ namespace llvm {
     bool ParseOptionalVisibility(unsigned &Visibility);
     bool ParseOptionalCallingConv(CallingConv::ID &CC);
     bool ParseOptionalAlignment(unsigned &Alignment);
+    bool ParseScopeAndOrdering(bool isAtomic, SynchronizationScope &Scope,
+                               AtomicOrdering &Ordering);
     bool ParseOptionalStackAlignment(unsigned &Alignment);
     bool ParseOptionalCommaAlign(unsigned &Alignment, bool &AteExtraComma);
     bool ParseIndexList(SmallVectorImpl<unsigned> &Indices,bool &AteExtraComma);
@@ -360,6 +363,7 @@ namespace llvm {
     int ParseAlloc(Instruction *&I, PerFunctionState &PFS);
     int ParseLoad(Instruction *&I, PerFunctionState &PFS, bool isVolatile);
     int ParseStore(Instruction *&I, PerFunctionState &PFS, bool isVolatile);
+    int ParseFence(Instruction *&I, PerFunctionState &PFS);
     int ParseGetElementPtr(Instruction *&I, PerFunctionState &PFS);
     int ParseExtractValue(Instruction *&I, PerFunctionState &PFS);
     int ParseInsertValue(Instruction *&I, PerFunctionState &PFS);

lib/AsmParser/LLToken.h

@@ -53,6 +53,9 @@ namespace lltok {
     kw_deplibs,
     kw_datalayout,
     kw_volatile,
+    kw_atomic,
+    kw_unordered, kw_monotonic, kw_acquire, kw_release, kw_acq_rel, kw_seq_cst,
+    kw_singlethread,
     kw_nuw,
     kw_nsw,
     kw_exact,
@@ -121,7 +124,7 @@ namespace lltok {
     kw_ret, kw_br, kw_switch, kw_indirectbr, kw_invoke, kw_unwind,
     kw_unreachable,
 
-    kw_alloca, kw_load, kw_store, kw_getelementptr,
+    kw_alloca, kw_load, kw_store, kw_fence, kw_getelementptr,
 
     kw_extractelement, kw_insertelement, kw_shufflevector,
     kw_extractvalue, kw_insertvalue, kw_blockaddress,

lib/Bitcode/Reader/BitcodeReader.cpp

@@ -131,6 +131,27 @@ static int GetDecodedBinaryOpcode(unsigned Val, Type *Ty) {
   }
 }
 
+static AtomicOrdering GetDecodedOrdering(unsigned Val) {
+  switch (Val) {
+  case bitc::ORDERING_NOTATOMIC: return NotAtomic;
+  case bitc::ORDERING_UNORDERED: return Unordered;
+  case bitc::ORDERING_MONOTONIC: return Monotonic;
+  case bitc::ORDERING_ACQUIRE: return Acquire;
+  case bitc::ORDERING_RELEASE: return Release;
+  case bitc::ORDERING_ACQREL: return AcquireRelease;
+  default: // Map unknown orderings to sequentially-consistent.
+  case bitc::ORDERING_SEQCST: return SequentiallyConsistent;
+  }
+}
+
+static SynchronizationScope GetDecodedSynchScope(unsigned Val) {
+  switch (Val) {
+  case bitc::SYNCHSCOPE_SINGLETHREAD: return SingleThread;
+  default: // Map unknown scopes to cross-thread.
+  case bitc::SYNCHSCOPE_CROSSTHREAD: return CrossThread;
+  }
+}
+
 namespace llvm {
 namespace {
   /// @brief A class for maintaining the slot number definition
@@ -2534,6 +2555,18 @@ bool BitcodeReader::ParseFunctionBody(Function *F) {
       InstructionList.push_back(I);
       break;
     }
+    case bitc::FUNC_CODE_INST_FENCE: { // FENCE:[ordering, synchscope]
+      if (2 != Record.size())
+        return Error("Invalid FENCE record");
+      AtomicOrdering Ordering = GetDecodedOrdering(Record[0]);
+      if (Ordering == NotAtomic || Ordering == Unordered ||
+          Ordering == Monotonic)
+        return Error("Invalid FENCE record");
+      SynchronizationScope SynchScope = GetDecodedSynchScope(Record[1]);
+      I = new FenceInst(Context, Ordering, SynchScope);
+      InstructionList.push_back(I);
+      break;
+    }
     case bitc::FUNC_CODE_INST_CALL: {
       // CALL: [paramattrs, cc, fnty, fnid, arg0, arg1...]
       if (Record.size() < 3)

lib/Bitcode/Writer/BitcodeWriter.cpp

@@ -101,6 +101,27 @@ static unsigned GetEncodedBinaryOpcode(unsigned Opcode) {
   }
 }
 
+static unsigned GetEncodedOrdering(AtomicOrdering Ordering) {
+  switch (Ordering) {
+  default: llvm_unreachable("Unknown atomic ordering");
+  case NotAtomic: return bitc::ORDERING_NOTATOMIC;
+  case Unordered: return bitc::ORDERING_UNORDERED;
+  case Monotonic: return bitc::ORDERING_MONOTONIC;
+  case Acquire: return bitc::ORDERING_ACQUIRE;
+  case Release: return bitc::ORDERING_RELEASE;
+  case AcquireRelease: return bitc::ORDERING_ACQREL;
+  case SequentiallyConsistent: return bitc::ORDERING_SEQCST;
+  }
+}
+
+static unsigned GetEncodedSynchScope(SynchronizationScope SynchScope) {
+  switch (SynchScope) {
+  default: llvm_unreachable("Unknown synchronization scope");
+  case SingleThread: return bitc::SYNCHSCOPE_SINGLETHREAD;
+  case CrossThread: return bitc::SYNCHSCOPE_CROSSTHREAD;
+  }
+}
+
 static void WriteStringRecord(unsigned Code, StringRef Str,
                               unsigned AbbrevToUse, BitstreamWriter &Stream) {
   SmallVector<unsigned, 64> Vals;
@@ -1147,6 +1168,11 @@ static void WriteInstruction(const Instruction &I, unsigned InstID,
     Vals.push_back(Log2_32(cast<StoreInst>(I).getAlignment())+1);
     Vals.push_back(cast<StoreInst>(I).isVolatile());
     break;
+  case Instruction::Fence:
+    Code = bitc::FUNC_CODE_INST_FENCE;
+    Vals.push_back(GetEncodedOrdering(cast<FenceInst>(I).getOrdering()));
+    Vals.push_back(GetEncodedSynchScope(cast<FenceInst>(I).getSynchScope()));
+    break;
   case Instruction::Call: {
     const CallInst &CI = cast<CallInst>(I);
     PointerType *PTy = cast<PointerType>(CI.getCalledValue()->getType());

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

@@ -3211,6 +3211,10 @@ void SelectionDAGBuilder::visitStore(const StoreInst &I) {
   DAG.setRoot(StoreNode);
 }
 
+void SelectionDAGBuilder::visitFence(const FenceInst &I) {
+  llvm_unreachable("Not implemented yet");
+}
+
 /// visitTargetIntrinsic - Lower a call of a target intrinsic to an INTRINSIC
 /// node.
 void SelectionDAGBuilder::visitTargetIntrinsic(const CallInst &I,

lib/CodeGen/SelectionDAG/SelectionDAGBuilder.h

@@ -504,6 +504,7 @@ private:
   void visitAlloca(const AllocaInst &I);
   void visitLoad(const LoadInst &I);
   void visitStore(const StoreInst &I);
+  void visitFence(const FenceInst &I);
   void visitPHI(const PHINode &I);
   void visitCall(const CallInst &I);
   bool visitMemCmpCall(const CallInst &I);

lib/VMCore/AsmWriter.cpp

@@ -1098,6 +1098,7 @@ public:
 
   void writeOperand(const Value *Op, bool PrintType);
   void writeParamOperand(const Value *Operand, Attributes Attrs);
+  void writeAtomic(AtomicOrdering Ordering, SynchronizationScope SynchScope);
 
   void writeAllMDNodes();
 
@@ -1128,6 +1129,28 @@ void AssemblyWriter::writeOperand(const Value *Operand, bool PrintType) {
   WriteAsOperandInternal(Out, Operand, &TypePrinter, &Machine, TheModule);
 }
 
+void AssemblyWriter::writeAtomic(AtomicOrdering Ordering,
+                                 SynchronizationScope SynchScope) {
+  if (Ordering == NotAtomic)
+    return;
+
+  switch (SynchScope) {
+  default: Out << " <bad scope " << int(SynchScope) << ">"; break;
+  case SingleThread: Out << " singlethread"; break;
+  case CrossThread: break;
+  }
+
+  switch (Ordering) {
+  default: Out << " <bad ordering " << int(Ordering) << ">"; break;
+  case Unordered: Out << " unordered"; break;
+  case Monotonic: Out << " monotonic"; break;
+  case Acquire: Out << " acquire"; break;
+  case Release: Out << " release"; break;
+  case AcquireRelease: Out << " acq_rel"; break;
+  case SequentiallyConsistent: Out << " seq_cst"; break;
+  }
+}
+
 void AssemblyWriter::writeParamOperand(const Value *Operand,
                                        Attributes Attrs) {
   if (Operand == 0) {
@@ -1883,6 +1906,8 @@ void AssemblyWriter::printInstruction(const Instruction &I) {
     Out << ", align " << cast<LoadInst>(I).getAlignment();
   } else if (isa<StoreInst>(I) && cast<StoreInst>(I).getAlignment()) {
     Out << ", align " << cast<StoreInst>(I).getAlignment();
+  } else if (const FenceInst *FI = dyn_cast<FenceInst>(&I)) {
+    writeAtomic(FI->getOrdering(), FI->getSynchScope());
   }
 
   // Print Metadata info.

lib/VMCore/Instruction.cpp

@@ -127,6 +127,7 @@ const char *Instruction::getOpcodeName(unsigned OpCode) {
   case Alloca:        return "alloca";
   case Load:          return "load";
   case Store:         return "store";
+  case Fence:         return "fence";
   case GetElementPtr: return "getelementptr";
 
   // Convert instructions...

lib/VMCore/Instructions.cpp

@@ -995,6 +995,26 @@ void StoreInst::setAlignment(unsigned Align) {
   assert(getAlignment() == Align && "Alignment representation error!");
 }
 
+//===----------------------------------------------------------------------===//
+//                       FenceInst Implementation
+//===----------------------------------------------------------------------===//
+
+FenceInst::FenceInst(LLVMContext &C, AtomicOrdering Ordering, 
+                     SynchronizationScope SynchScope,
+                     Instruction *InsertBefore)
+  : Instruction(Type::getVoidTy(C), Fence, 0, 0, InsertBefore) {
+  setOrdering(Ordering);
+  setSynchScope(SynchScope);
+}
+
+FenceInst::FenceInst(LLVMContext &C, AtomicOrdering Ordering, 
+                     SynchronizationScope SynchScope,
+                     BasicBlock *InsertAtEnd)
+  : Instruction(Type::getVoidTy(C), Fence, 0, 0, InsertAtEnd) {
+  setOrdering(Ordering);
+  setSynchScope(SynchScope);
+}
+
 //===----------------------------------------------------------------------===//
 //                       GetElementPtrInst Implementation
 //===----------------------------------------------------------------------===//
@@ -3018,6 +3038,10 @@ StoreInst *StoreInst::clone_impl() const {
                        isVolatile(), getAlignment());
 }
 
+FenceInst *FenceInst::clone_impl() const {
+  return new FenceInst(getContext(), getOrdering(), getSynchScope());
+}
+
 TruncInst *TruncInst::clone_impl() const {
   return new TruncInst(getOperand(0), getType());
 }

lib/VMCore/Verifier.cpp

@@ -278,6 +278,7 @@ namespace {
     void visitUserOp1(Instruction &I);
     void visitUserOp2(Instruction &I) { visitUserOp1(I); }
     void visitIntrinsicFunctionCall(Intrinsic::ID ID, CallInst &CI);
+    void visitFenceInst(FenceInst &FI);
     void visitAllocaInst(AllocaInst &AI);
     void visitExtractValueInst(ExtractValueInst &EVI);
     void visitInsertValueInst(InsertValueInst &IVI);
@@ -1315,6 +1316,15 @@ void Verifier::visitAllocaInst(AllocaInst &AI) {
   visitInstruction(AI);
 }
 
+void Verifier::visitFenceInst(FenceInst &FI) {
+  const AtomicOrdering Ordering = FI.getOrdering();
+  Assert1(Ordering == Acquire || Ordering == Release ||
+          Ordering == AcquireRelease || Ordering == SequentiallyConsistent,
+          "fence instructions may only have "
+          " acquire, release, acq_rel, or seq_cst ordering.", &FI);
+  visitInstruction(FI);
+}
+
 void Verifier::visitExtractValueInst(ExtractValueInst &EVI) {
   Assert1(ExtractValueInst::getIndexedType(EVI.getAggregateOperand()->getType(),
                                            EVI.getIndices()) ==