Implement function prefix data as an IR feature.

Previous discussion:
http://lists.cs.uiuc.edu/pipermail/llvmdev/2013-July/063909.html

Differential Revision: http://llvm-reviews.chandlerc.com/D1191

llvm-svn: 190773

docs/BitCodeFormat.rst

@@ -718,7 +718,7 @@ global variable. The operand fields are:
 MODULE_CODE_FUNCTION Record
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-``[FUNCTION, type, callingconv, isproto, linkage, paramattr, alignment, section, visibility, gc]``
+``[FUNCTION, type, callingconv, isproto, linkage, paramattr, alignment, section, visibility, gc, prefix]``
 
 The ``FUNCTION`` record (code 8) marks the declaration or definition of a
 function. The operand fields are:
@@ -757,6 +757,9 @@ function. The operand fields are:
 * *unnamed_addr*: If present and non-zero, indicates that the function has
   ``unnamed_addr``
 
+* *prefix*: If non-zero, the value index of the prefix data for this function,
+  plus 1.
+
 MODULE_CODE_ALIAS Record
 ^^^^^^^^^^^^^^^^^^^^^^^^
 

docs/LangRef.rst

@@ -552,16 +552,16 @@ an optional ``unnamed_addr`` attribute, a return type, an optional
 name, a (possibly empty) argument list (each with optional :ref:`parameter
 attributes <paramattrs>`), optional :ref:`function attributes <fnattrs>`,
 an optional section, an optional alignment, an optional :ref:`garbage
-collector name <gc>`, an opening curly brace, a list of basic blocks,
+collector name <gc>`, an optional :ref:`prefix <prefixdata>`, an opening
-and a closing curly brace.
+curly brace, a list of basic blocks, and a closing curly brace.
 
 LLVM function declarations consist of the "``declare``" keyword, an
 optional :ref:`linkage type <linkage>`, an optional :ref:`visibility
 style <visibility>`, an optional :ref:`calling convention <callingconv>`,
 an optional ``unnamed_addr`` attribute, a return type, an optional
 :ref:`parameter attribute <paramattrs>` for the return type, a function
-name, a possibly empty list of arguments, an optional alignment, and an
+name, a possibly empty list of arguments, an optional alignment, an optional
-optional :ref:`garbage collector name <gc>`.
+:ref:`garbage collector name <gc>` and an optional :ref:`prefix <prefixdata>`.
 
 A function definition contains a list of basic blocks, forming the CFG
 (Control Flow Graph) for the function. Each basic block may optionally
@@ -598,7 +598,7 @@ Syntax::
            [cconv] [ret attrs]
            <ResultType> @<FunctionName> ([argument list])
            [fn Attrs] [section "name"] [align N]
-           [gc] { ... }
+           [gc] [prefix Constant] { ... }
 
 .. _langref_aliases:
 
@@ -757,6 +757,50 @@ The compiler declares the supported values of *name*. Specifying a
 collector which will cause the compiler to alter its output in order to
 support the named garbage collection algorithm.
 
+.. _prefixdata:
+
+Prefix Data
+-----------
+
+Prefix data is data associated with a function which the code generator
+will emit immediately before the function body.  The purpose of this feature
+is to allow frontends to associate language-specific runtime metadata with
+specific functions and make it available through the function pointer while
+still allowing the function pointer to be called.  To access the data for a
+given function, a program may bitcast the function pointer to a pointer to
+the constant's type.  This implies that the IR symbol points to the start
+of the prefix data.
+
+To maintain the semantics of ordinary function calls, the prefix data must
+have a particular format.  Specifically, it must begin with a sequence of
+bytes which decode to a sequence of machine instructions, valid for the
+module's target, which transfer control to the point immediately succeeding
+the prefix data, without performing any other visible action.  This allows
+the inliner and other passes to reason about the semantics of the function
+definition without needing to reason about the prefix data.  Obviously this
+makes the format of the prefix data highly target dependent.
+
+A trivial example of valid prefix data for the x86 architecture is ``i8 144``,
+which encodes the ``nop`` instruction:
+
+.. code-block:: llvm
+
+    define void @f() prefix i8 144 { ... }
+
+Generally prefix data can be formed by encoding a relative branch instruction
+which skips the metadata, as in this example of valid prefix data for the
+x86_64 architecture, where the first two bytes encode ``jmp .+10``:
+
+.. code-block:: llvm
+
+    %0 = type <{ i8, i8, i8* }>
+
+    define void @f() prefix %0 <{ i8 235, i8 8, i8* @md}> { ... }
+
+A function may have prefix data but no body.  This has similar semantics
+to the ``available_externally`` linkage in that the data may be used by the
+optimizers but will not be emitted in the object file.
+
 .. _attrgrp:
 
 Attribute Groups

include/llvm/IR/Function.h

@@ -159,11 +159,11 @@ public:
   /// calling convention of this function.  The enum values for the known
   /// calling conventions are defined in CallingConv.h.
   CallingConv::ID getCallingConv() const {
-    return static_cast<CallingConv::ID>(getSubclassDataFromValue() >> 1);
+    return static_cast<CallingConv::ID>(getSubclassDataFromValue() >> 2);
   }
   void setCallingConv(CallingConv::ID CC) {
-    setValueSubclassData((getSubclassDataFromValue() & 1) |
+    setValueSubclassData((getSubclassDataFromValue() & 3) |
-                         (static_cast<unsigned>(CC) << 1));
+                         (static_cast<unsigned>(CC) << 2));
   }
 
   /// @brief Return the attribute list for this Function.
@@ -427,6 +427,13 @@ public:
   size_t arg_size() const;
   bool arg_empty() const;
 
+  bool hasPrefixData() const {
+    return getSubclassDataFromValue() & 2;
+  }
+
+  Constant *getPrefixData() const;
+  void setPrefixData(Constant *PrefixData);
+
   /// viewCFG - This function is meant for use from the debugger.  You can just
   /// say 'call F->viewCFG()' and a ghostview window should pop up from the
   /// program, displaying the CFG of the current function with the code for each

lib/AsmParser/LLLexer.cpp

@@ -540,6 +540,7 @@ lltok::Kind LLLexer::LexIdentifier() {
   KEYWORD(alignstack);
   KEYWORD(inteldialect);
   KEYWORD(gc);
+  KEYWORD(prefix);
 
   KEYWORD(ccc);
   KEYWORD(fastcc);

lib/AsmParser/LLParser.cpp

@@ -2922,7 +2922,7 @@ bool LLParser::ParseTypeAndBasicBlock(BasicBlock *&BB, LocTy &Loc,
 /// FunctionHeader
 ///   ::= OptionalLinkage OptionalVisibility OptionalCallingConv OptRetAttrs
 ///       OptUnnamedAddr Type GlobalName '(' ArgList ')' OptFuncAttrs OptSection
-///       OptionalAlign OptGC
+///       OptionalAlign OptGC OptionalPrefix
 bool LLParser::ParseFunctionHeader(Function *&Fn, bool isDefine) {
   // Parse the linkage.
   LocTy LinkageLoc = Lex.getLoc();
@@ -3001,6 +3001,7 @@ bool LLParser::ParseFunctionHeader(Function *&Fn, bool isDefine) {
   std::string GC;
   bool UnnamedAddr;
   LocTy UnnamedAddrLoc;
+  Constant *Prefix = 0;
 
   if (ParseArgumentList(ArgList, isVarArg) ||
       ParseOptionalToken(lltok::kw_unnamed_addr, UnnamedAddr,
@@ -3011,7 +3012,9 @@ bool LLParser::ParseFunctionHeader(Function *&Fn, bool isDefine) {
        ParseStringConstant(Section)) ||
       ParseOptionalAlignment(Alignment) ||
       (EatIfPresent(lltok::kw_gc) &&
-       ParseStringConstant(GC)))
+       ParseStringConstant(GC)) ||
+      (EatIfPresent(lltok::kw_prefix) &&
+       ParseGlobalTypeAndValue(Prefix)))
     return true;
 
   if (FuncAttrs.contains(Attribute::Builtin))
@@ -3109,6 +3112,7 @@ bool LLParser::ParseFunctionHeader(Function *&Fn, bool isDefine) {
   Fn->setAlignment(Alignment);
   Fn->setSection(Section);
   if (!GC.empty()) Fn->setGC(GC.c_str());
+  Fn->setPrefixData(Prefix);
   ForwardRefAttrGroups[Fn] = FwdRefAttrGrps;
 
   // Add all of the arguments we parsed to the function.

lib/AsmParser/LLToken.h

@@ -81,6 +81,7 @@ namespace lltok {
     kw_alignstack,
     kw_inteldialect,
     kw_gc,
+    kw_prefix,
     kw_c,
 
     kw_cc, kw_ccc, kw_fastcc, kw_coldcc,

lib/Bitcode/Reader/BitcodeReader.cpp

@@ -1106,9 +1106,11 @@ uint64_t BitcodeReader::decodeSignRotatedValue(uint64_t V) {
 bool BitcodeReader::ResolveGlobalAndAliasInits() {
   std::vector<std::pair<GlobalVariable*, unsigned> > GlobalInitWorklist;
   std::vector<std::pair<GlobalAlias*, unsigned> > AliasInitWorklist;
+  std::vector<std::pair<Function*, unsigned> > FunctionPrefixWorklist;
 
   GlobalInitWorklist.swap(GlobalInits);
   AliasInitWorklist.swap(AliasInits);
+  FunctionPrefixWorklist.swap(FunctionPrefixes);
 
   while (!GlobalInitWorklist.empty()) {
     unsigned ValID = GlobalInitWorklist.back().second;
@@ -1136,6 +1138,20 @@ bool BitcodeReader::ResolveGlobalAndAliasInits() {
     }
     AliasInitWorklist.pop_back();
   }
+
+  while (!FunctionPrefixWorklist.empty()) {
+    unsigned ValID = FunctionPrefixWorklist.back().second;
+    if (ValID >= ValueList.size()) {
+      FunctionPrefixes.push_back(FunctionPrefixWorklist.back());
+    } else {
+      if (Constant *C = dyn_cast<Constant>(ValueList[ValID]))
+        FunctionPrefixWorklist.back().first->setPrefixData(C);
+      else
+        return Error("Function prefix is not a constant!");
+    }
+    FunctionPrefixWorklist.pop_back();
+  }
+
   return false;
 }
 
@@ -1881,6 +1897,8 @@ bool BitcodeReader::ParseModule(bool Resume) {
       if (Record.size() > 9)
         UnnamedAddr = Record[9];
       Func->setUnnamedAddr(UnnamedAddr);
+      if (Record.size() > 10 && Record[10] != 0)
+        FunctionPrefixes.push_back(std::make_pair(Func, Record[10]-1));
       ValueList.push_back(Func);
 
       // If this is a function with a body, remember the prototype we are

lib/Bitcode/Reader/BitcodeReader.h

@@ -142,6 +142,7 @@ class BitcodeReader : public GVMaterializer {
 
   std::vector<std::pair<GlobalVariable*, unsigned> > GlobalInits;
   std::vector<std::pair<GlobalAlias*, unsigned> > AliasInits;
+  std::vector<std::pair<Function*, unsigned> > FunctionPrefixes;
 
   /// MAttributes - The set of attributes by index.  Index zero in the
   /// file is for null, and is thus not represented here.  As such all indices

lib/Bitcode/Writer/BitcodeWriter.cpp

@@ -632,7 +632,7 @@ static void WriteModuleInfo(const Module *M, const ValueEnumerator &VE,
   // Emit the function proto information.
   for (Module::const_iterator F = M->begin(), E = M->end(); F != E; ++F) {
     // FUNCTION:  [type, callingconv, isproto, linkage, paramattrs, alignment,
-    //             section, visibility, gc, unnamed_addr]
+    //             section, visibility, gc, unnamed_addr, prefix]
     Vals.push_back(VE.getTypeID(F->getType()));
     Vals.push_back(F->getCallingConv());
     Vals.push_back(F->isDeclaration());
@@ -643,6 +643,8 @@ static void WriteModuleInfo(const Module *M, const ValueEnumerator &VE,
     Vals.push_back(getEncodedVisibility(F));
     Vals.push_back(F->hasGC() ? GCMap[F->getGC()] : 0);
     Vals.push_back(F->hasUnnamedAddr());
+    Vals.push_back(F->hasPrefixData() ? (VE.getValueID(F->getPrefixData()) + 1)
+                                      : 0);
 
     unsigned AbbrevToUse = 0;
     Stream.EmitRecord(bitc::MODULE_CODE_FUNCTION, Vals, AbbrevToUse);
@@ -1863,6 +1865,8 @@ static void WriteModuleUseLists(const Module *M, ValueEnumerator &VE,
     WriteUseList(FI, VE, Stream);
     if (!FI->isDeclaration())
       WriteFunctionUseList(FI, VE, Stream);
+    if (FI->hasPrefixData())
+      WriteUseList(FI->getPrefixData(), VE, Stream);
   }
 
   // Write the aliases.

lib/Bitcode/Writer/ValueEnumerator.cpp

@@ -60,6 +60,11 @@ ValueEnumerator::ValueEnumerator(const Module *M) {
        I != E; ++I)
     EnumerateValue(I->getAliasee());
 
+  // Enumerate the prefix data constants.
+  for (Module::const_iterator I = M->begin(), E = M->end(); I != E; ++I)
+    if (I->hasPrefixData())
+      EnumerateValue(I->getPrefixData());
+
   // Insert constants and metadata that are named at module level into the slot
   // pool so that the module symbol table can refer to them...
   EnumerateValueSymbolTable(M->getValueSymbolTable());

lib/CodeGen/AsmPrinter/AsmPrinter.cpp

@@ -459,6 +459,10 @@ void AsmPrinter::EmitFunctionHeader() {
     OutStreamer.EmitLabel(DeadBlockSyms[i]);
   }
 
+  // Emit the prefix data.
+  if (F->hasPrefixData())
+    EmitGlobalConstant(F->getPrefixData());
+
   // Emit pre-function debug and/or EH information.
   if (DE) {
     NamedRegionTimer T(EHTimerName, DWARFGroupName, TimePassesIsEnabled);

lib/IR/AsmWriter.cpp

@@ -1647,6 +1647,10 @@ void AssemblyWriter::printFunction(const Function *F) {
     Out << " align " << F->getAlignment();
   if (F->hasGC())
     Out << " gc \"" << F->getGC() << '"';
+  if (F->hasPrefixData()) {
+    Out << " prefix ";
+    writeOperand(F->getPrefixData(), true);
+  }
   if (F->isDeclaration()) {
     Out << '\n';
   } else {

lib/IR/Function.cpp

@@ -276,6 +276,9 @@ void Function::dropAllReferences() {
   // blockaddresses, but BasicBlock's destructor takes care of those.
   while (!BasicBlocks.empty())
     BasicBlocks.begin()->eraseFromParent();
+
+  // Prefix data is stored in a side table.
+  setPrefixData(0);
 }
 
 void Function::addAttribute(unsigned i, Attribute::AttrKind attr) {
@@ -351,6 +354,10 @@ void Function::copyAttributesFrom(const GlobalValue *Src) {
     setGC(SrcF->getGC());
   else
     clearGC();
+  if (SrcF->hasPrefixData())
+    setPrefixData(SrcF->getPrefixData());
+  else
+    setPrefixData(0);
 }
 
 /// getIntrinsicID - This method returns the ID number of the specified
@@ -720,3 +727,32 @@ bool Function::callsFunctionThatReturnsTwice() const {
 
   return false;
 }
+
+Constant *Function::getPrefixData() const {
+  assert(hasPrefixData());
+  const LLVMContextImpl::PrefixDataMapTy &PDMap =
+      getContext().pImpl->PrefixDataMap;
+  assert(PDMap.find(this) != PDMap.end());
+  return cast<Constant>(PDMap.find(this)->second->getReturnValue());
+}
+
+void Function::setPrefixData(Constant *PrefixData) {
+  if (!PrefixData && !hasPrefixData())
+    return;
+
+  unsigned SCData = getSubclassDataFromValue();
+  LLVMContextImpl::PrefixDataMapTy &PDMap = getContext().pImpl->PrefixDataMap;
+  ReturnInst *&PDHolder = PDMap[this];
+  if (PrefixData) {
+    if (PDHolder)
+      PDHolder->setOperand(0, PrefixData);
+    else
+      PDHolder = ReturnInst::Create(getContext(), PrefixData);
+    SCData |= 2;
+  } else {
+    delete PDHolder;
+    PDMap.erase(this);
+    SCData &= ~2;
+  }
+  setValueSubclassData(SCData);
+}

lib/IR/LLVMContextImpl.h

@@ -355,6 +355,11 @@ public:
   typedef DenseMap<const Function*, unsigned> IntrinsicIDCacheTy;
   IntrinsicIDCacheTy IntrinsicIDCache;
 
+  /// \brief Mapping from a function to its prefix data, which is stored as the
+  /// operand of an unparented ReturnInst so that the prefix data has a Use.
+  typedef DenseMap<const Function *, ReturnInst *> PrefixDataMapTy;
+  PrefixDataMapTy PrefixDataMap;
+
   int getOrAddScopeRecordIdxEntry(MDNode *N, int ExistingIdx);
   int getOrAddScopeInlinedAtIdxEntry(MDNode *Scope, MDNode *IA,int ExistingIdx);
   

lib/IR/TypeFinder.cpp

@@ -44,6 +44,9 @@ void TypeFinder::run(const Module &M, bool onlyNamed) {
   for (Module::const_iterator FI = M.begin(), E = M.end(); FI != E; ++FI) {
     incorporateType(FI->getType());
 
+    if (FI->hasPrefixData())
+      incorporateValue(FI->getPrefixData());
+
     // First incorporate the arguments.
     for (Function::const_arg_iterator AI = FI->arg_begin(),
            AE = FI->arg_end(); AI != AE; ++AI)

lib/Linker/LinkModules.cpp

@@ -1260,6 +1260,13 @@ bool ModuleLinker::run() {
     // Skip if not linking from source.
     if (DoNotLinkFromSource.count(SF)) continue;
     
+    Function *DF = cast<Function>(ValueMap[SF]);
+    if (SF->hasPrefixData()) {
+      // Link in the prefix data.
+      DF->setPrefixData(MapValue(
+          SF->getPrefixData(), ValueMap, RF_None, &TypeMap, &ValMaterializer));
+    }
+
     // Skip if no body (function is external) or materialize.
     if (SF->isDeclaration()) {
       if (!SF->isMaterializable())
@@ -1268,7 +1275,7 @@ bool ModuleLinker::run() {
         return true;
     }
     
-    linkFunctionBody(cast<Function>(ValueMap[SF]), SF);
+    linkFunctionBody(DF, SF);
     SF->Dematerialize();
   }
 
@@ -1296,6 +1303,14 @@ bool ModuleLinker::run() {
         continue;
 
       Function *DF = cast<Function>(ValueMap[SF]);
+      if (SF->hasPrefixData()) {
+        // Link in the prefix data.
+        DF->setPrefixData(MapValue(SF->getPrefixData(),
+                                   ValueMap,
+                                   RF_None,
+                                   &TypeMap,
+                                   &ValMaterializer));
+      }
 
       // Materialize if necessary.
       if (SF->isDeclaration()) {

lib/Transforms/IPO/GlobalDCE.cpp

@@ -179,6 +179,9 @@ void GlobalDCE::GlobalIsNeeded(GlobalValue *G) {
     // any globals used will be marked as needed.
     Function *F = cast<Function>(G);
 
+    if (F->hasPrefixData())
+      MarkUsedGlobalsAsNeeded(F->getPrefixData());
+
     for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB)
       for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
         for (User::op_iterator U = I->op_begin(), E = I->op_end(); U != E; ++U)

test/CodeGen/X86/prefixdata.ll

@@ -0,0 +1,15 @@
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s
+
+@i = linkonce_odr global i32 1
+
+; CHECK: f:
+; CHECK-NEXT: .long	1
+define void @f() prefix i32 1 {
+  ret void
+}
+
+; CHECK: g:
+; CHECK-NEXT: .quad	i
+define void @g() prefix i32* @i {
+  ret void
+}

test/Feature/prefixdata.ll

@@ -0,0 +1,18 @@
+; RUN: llvm-as < %s | llvm-dis > %t1.ll
+; RUN: FileCheck %s < %t1.ll
+; RUN: llvm-as < %t1.ll | llvm-dis > %t2.ll
+; RUN: diff %t1.ll %t2.ll
+; RUN: opt -O3 -S < %t1.ll | FileCheck %s
+
+; CHECK: @i
+@i = linkonce_odr global i32 1
+
+; CHECK: f(){{.*}}prefix i32 1
+define void @f() prefix i32 1 {
+  ret void
+}
+
+; CHECK: g(){{.*}}prefix i32* @i
+define void @g() prefix i32* @i {
+  ret void
+}

test/Linker/prefixdata.ll

@@ -0,0 +1,9 @@
+; RUN: echo > %t.ll
+; RUN: llvm-link %t.ll %s -S -o - | FileCheck %s
+
+@i = linkonce_odr global i32 1
+
+; CHECK: define void @f() prefix i32* @i
+define void @f() prefix i32* @i {
+  ret void
+}