Emit !callback metadata and introduce the callback attribute

  With commit r351627, LLVM gained the ability to apply (existing) IPO
  optimizations on indirections through callbacks, or transitive calls.
  The general idea is that we use an abstraction to hide the middle man
  and represent the callback call in the context of the initial caller.
  It is described in more detail in the commit message of the LLVM patch
  r351627, the llvm::AbstractCallSite class description, and the
  language reference section on callback-metadata.

  This commit enables clang to emit !callback metadata that is
  understood by LLVM. It does so in three different cases:
    1) For known broker functions declarations that are directly
       generated, e.g., __kmpc_fork_call for the OpenMP pragma parallel.
    2) For known broker functions that are identified by their name and
       source location through the builtin detection, e.g.,
       pthread_create from the POSIX thread API.
    3) For user annotated functions that carry the "callback(callee, ...)"
       attribute. The attribute has to include the name, or index, of
       the callback callee and how the passed arguments can be
       identified (as many as the callback callee has). See the callback
       attribute documentation for detailed information.

Differential Revision: https://reviews.llvm.org/D55483

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@351629 91177308-0d34-0410-b5e6-96231b3b80d8

7 files changed, 253 insertions, 2 deletions

diff --git a/lib/AST/ASTContext.cpp b/lib/AST/ASTContext.cpp
index caf6a504cd..a34b9d78e5 100644
--- a/lib/AST/ASTContext.cpp
+++ b/lib/AST/ASTContext.cpp
@@ -9518,6 +9518,10 @@ QualType ASTContext::GetBuiltinType(unsigned Id,
                                     GetBuiltinTypeError &Error,
                                     unsigned *IntegerConstantArgs) const {
   const char *TypeStr = BuiltinInfo.getTypeString(Id);
+  if (TypeStr[0] == '\0') {
+    Error = GE_Missing_type;
+    return {};
+  }
 
   SmallVector<QualType, 8> ArgTypes;
 
diff --git a/lib/Basic/Builtins.cpp b/lib/Basic/Builtins.cpp
index 7e7f67ca87..53e7df95f8 100644
--- a/lib/Basic/Builtins.cpp
+++ b/lib/Basic/Builtins.cpp
@@ -156,6 +156,33 @@ bool Builtin::Context::isScanfLike(unsigned ID, unsigned &FormatIdx,
   return isLike(ID, FormatIdx, HasVAListArg, "sS");
 }
 
+bool Builtin::Context::performsCallback(unsigned ID,
+                                        SmallVectorImpl<int> &Encoding) const {
+  const char *CalleePos = ::strchr(getRecord(ID).Attributes, 'C');
+  if (!CalleePos)
+    return false;
+
+  ++CalleePos;
+  assert(*CalleePos == '<' &&
+         "Callback callee specifier must be followed by a '<'");
+  ++CalleePos;
+
+  char *EndPos;
+  int CalleeIdx = ::strtol(CalleePos, &EndPos, 10);
+  assert(CalleeIdx >= 0 && "Callee index is supposed to be positive!");
+  Encoding.push_back(CalleeIdx);
+
+  while (*EndPos == ',') {
+    const char *PayloadPos = EndPos + 1;
+
+    int PayloadIdx = ::strtol(PayloadPos, &EndPos, 10);
+    Encoding.push_back(PayloadIdx);
+  }
+
+  assert(*EndPos == '>' && "Callback callee specifier must end with a '>'");
+  return true;
+}
+
 bool Builtin::Context::canBeRedeclared(unsigned ID) const {
   return ID == Builtin::NotBuiltin ||
          ID == Builtin::BI__va_start ||
diff --git a/lib/CodeGen/CGOpenMPRuntime.cpp b/lib/CodeGen/CGOpenMPRuntime.cpp
index 20eb0b29f4..df377843e9 100644
--- a/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -1677,6 +1677,22 @@ CGOpenMPRuntime::createRuntimeFunction(unsigned Function) {
     auto *FnTy =
         llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ true);
     RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_fork_call");
+    if (auto *F = dyn_cast<llvm::Function>(RTLFn)) {
+      if (!F->hasMetadata(llvm::LLVMContext::MD_callback)) {
+        llvm::LLVMContext &Ctx = F->getContext();
+        llvm::MDBuilder MDB(Ctx);
+        // Annotate the callback behavior of the __kmpc_fork_call:
+        //  - The callback callee is argument number 2 (microtask).
+        //  - The first two arguments of the callback callee are unknown (-1).
+        //  - All variadic arguments to the __kmpc_fork_call are passed to the
+        //    callback callee.
+        F->addMetadata(
+            llvm::LLVMContext::MD_callback,
+            *llvm::MDNode::get(Ctx, {MDB.createCallbackEncoding(
+                                        2, {-1, -1},
+                                        /* VarArgsArePassed */ true)}));
+      }
+    }
     break;
   }
   case OMPRTL__kmpc_global_thread_num: {
@@ -2084,6 +2100,22 @@ CGOpenMPRuntime::createRuntimeFunction(unsigned Function) {
     auto *FnTy =
         llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ true);
     RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_fork_teams");
+    if (auto *F = dyn_cast<llvm::Function>(RTLFn)) {
+      if (!F->hasMetadata(llvm::LLVMContext::MD_callback)) {
+        llvm::LLVMContext &Ctx = F->getContext();
+        llvm::MDBuilder MDB(Ctx);
+        // Annotate the callback behavior of the __kmpc_fork_teams:
+        //  - The callback callee is argument number 2 (microtask).
+        //  - The first two arguments of the callback callee are unknown (-1).
+        //  - All variadic arguments to the __kmpc_fork_teams are passed to the
+        //    callback callee.
+        F->addMetadata(
+            llvm::LLVMContext::MD_callback,
+            *llvm::MDNode::get(Ctx, {MDB.createCallbackEncoding(
+                                        2, {-1, -1},
+                                        /* VarArgsArePassed */ true)}));
+      }
+    }
     break;
   }
   case OMPRTL__kmpc_taskloop: {
diff --git a/lib/CodeGen/CodeGenModule.cpp b/lib/CodeGen/CodeGenModule.cpp
index 244738042c..f5e4d4867c 100644
--- a/lib/CodeGen/CodeGenModule.cpp
+++ b/lib/CodeGen/CodeGenModule.cpp
@@ -1603,6 +1603,23 @@ void CodeGenModule::SetFunctionAttributes(GlobalDecl GD, llvm::Function *F,
 
   if (getLangOpts().OpenMP && FD->hasAttr<OMPDeclareSimdDeclAttr>())
     getOpenMPRuntime().emitDeclareSimdFunction(FD, F);
+
+  if (const auto *CB = FD->getAttr<CallbackAttr>()) {
+    // Annotate the callback behavior as metadata:
+    //  - The callback callee (as argument number).
+    //  - The callback payloads (as argument numbers).
+    llvm::LLVMContext &Ctx = F->getContext();
+    llvm::MDBuilder MDB(Ctx);
+
+    // The payload indices are all but the first one in the encoding. The first
+    // identifies the callback callee.
+    int CalleeIdx = *CB->encoding_begin();
+    ArrayRef<int> PayloadIndices(CB->encoding_begin() + 1, CB->encoding_end());
+    F->addMetadata(llvm::LLVMContext::MD_callback,
+                   *llvm::MDNode::get(Ctx, {MDB.createCallbackEncoding(
+                                               CalleeIdx, PayloadIndices,
+                                               /* VarArgsArePassed */ false)}));
+  }
 }
 
 void CodeGenModule::addUsedGlobal(llvm::GlobalValue *GV) {
diff --git a/lib/Parse/ParseDecl.cpp b/lib/Parse/ParseDecl.cpp
index 298a2bad56..be43194278 100644
--- a/lib/Parse/ParseDecl.cpp
+++ b/lib/Parse/ParseDecl.cpp
@@ -223,6 +223,15 @@ static bool attributeHasVariadicIdentifierArg(const IdentifierInfo &II) {
 #undef CLANG_ATTR_VARIADIC_IDENTIFIER_ARG_LIST
 }
 
+/// Determine whether the given attribute treats kw_this as an identifier.
+static bool attributeTreatsKeywordThisAsIdentifier(const IdentifierInfo &II) {
+#define CLANG_ATTR_THIS_ISA_IDENTIFIER_ARG_LIST
+  return llvm::StringSwitch<bool>(normalizeAttrName(II.getName()))
+#include "clang/Parse/AttrParserStringSwitches.inc"
+           .Default(false);
+#undef CLANG_ATTR_THIS_ISA_IDENTIFIER_ARG_LIST
+}
+
 /// Determine whether the given attribute parses a type argument.
 static bool attributeIsTypeArgAttr(const IdentifierInfo &II) {
 #define CLANG_ATTR_TYPE_ARG_LIST
@@ -287,6 +296,12 @@ unsigned Parser::ParseAttributeArgsCommon(
   // Ignore the left paren location for now.
   ConsumeParen();
 
+  bool ChangeKWThisToIdent = attributeTreatsKeywordThisAsIdentifier(*AttrName);
+
+  // Interpret "kw_this" as an identifier if the attributed requests it.
+  if (ChangeKWThisToIdent && Tok.is(tok::kw_this))
+    Tok.setKind(tok::identifier);
+
   ArgsVector ArgExprs;
   if (Tok.is(tok::identifier)) {
     // If this attribute wants an 'identifier' argument, make it so.
@@ -314,6 +329,10 @@ unsigned Parser::ParseAttributeArgsCommon(
 
     // Parse the non-empty comma-separated list of expressions.
     do {
+      // Interpret "kw_this" as an identifier if the attributed requests it.
+      if (ChangeKWThisToIdent && Tok.is(tok::kw_this))
+        Tok.setKind(tok::identifier);
+
       ExprResult ArgExpr;
       if (Tok.is(tok::identifier) &&
           attributeHasVariadicIdentifierArg(*AttrName)) {
diff --git a/lib/Sema/SemaDecl.cpp b/lib/Sema/SemaDecl.cpp
index a5a4833a5c..f327e94468 100644
--- a/lib/Sema/SemaDecl.cpp
+++ b/lib/Sema/SemaDecl.cpp
@@ -1927,10 +1927,13 @@ static void LookupPredefedObjCSuperType(Sema &ThisSema, Scope *S,
       Context.setObjCSuperType(Context.getTagDeclType(TD));
 }
 
-static StringRef getHeaderName(ASTContext::GetBuiltinTypeError Error) {
+static StringRef getHeaderName(Builtin::Context &BuiltinInfo, unsigned ID,
+                               ASTContext::GetBuiltinTypeError Error) {
   switch (Error) {
   case ASTContext::GE_None:
     return "";
+  case ASTContext::GE_Missing_type:
+    return BuiltinInfo.getHeaderName(ID);
   case ASTContext::GE_Missing_stdio:
     return "stdio.h";
   case ASTContext::GE_Missing_setjmp:
@@ -1955,7 +1958,8 @@ NamedDecl *Sema::LazilyCreateBuiltin(IdentifierInfo *II, unsigned ID,
   if (Error) {
     if (ForRedeclaration)
       Diag(Loc, diag::warn_implicit_decl_requires_sysheader)
-          << getHeaderName(Error) << Context.BuiltinInfo.getName(ID);
+          << getHeaderName(Context.BuiltinInfo, ID, Error)
+          << Context.BuiltinInfo.getName(ID);
     return nullptr;
   }
 
@@ -13580,6 +13584,13 @@ void Sema::AddKnownFunctionAttributes(FunctionDecl *FD) {
                                               FD->getLocation()));
     }
 
+    // Handle automatically recognized callbacks.
+    SmallVector<int, 4> Encoding;
+    if (!FD->hasAttr<CallbackAttr>() &&
+        Context.BuiltinInfo.performsCallback(BuiltinID, Encoding))
+      FD->addAttr(CallbackAttr::CreateImplicit(
+          Context, Encoding.data(), Encoding.size(), FD->getLocation()));
+
     // Mark const if we don't care about errno and that is the only thing
     // preventing the function from being const. This allows IRgen to use LLVM
     // intrinsics for such functions.
diff --git a/lib/Sema/SemaDeclAttr.cpp b/lib/Sema/SemaDeclAttr.cpp
index 7cfdb9e286..84ae677d4d 100644
--- a/lib/Sema/SemaDeclAttr.cpp
+++ b/lib/Sema/SemaDeclAttr.cpp
@@ -3480,6 +3480,144 @@ static void handleFormatAttr(Sema &S, Decl *D, const ParsedAttr &AL) {
     D->addAttr(NewAttr);
 }
 
+/// Handle __attribute__((callback(CalleeIdx, PayloadIdx0, ...))) attributes.
+static void handleCallbackAttr(Sema &S, Decl *D, const ParsedAttr &AL) {
+  // The index that identifies the callback callee is mandatory.
+  if (AL.getNumArgs() == 0) {
+    S.Diag(AL.getLoc(), diag::err_callback_attribute_no_callee)
+        << AL.getRange();
+    return;
+  }
+
+  bool HasImplicitThisParam = isInstanceMethod(D);
+  int32_t NumArgs = getFunctionOrMethodNumParams(D);
+
+  FunctionDecl *FD = D->getAsFunction();
+  assert(FD && "Expected a function declaration!");
+
+  llvm::StringMap<int> NameIdxMapping;
+  NameIdxMapping["__"] = -1;
+
+  NameIdxMapping["this"] = 0;
+
+  int Idx = 1;
+  for (const ParmVarDecl *PVD : FD->parameters())
+    NameIdxMapping[PVD->getName()] = Idx++;
+
+  auto UnknownName = NameIdxMapping.end();
+
+  SmallVector<int, 8> EncodingIndices;
+  for (unsigned I = 0, E = AL.getNumArgs(); I < E; ++I) {
+    SourceRange SR;
+    int32_t ArgIdx;
+
+    if (AL.isArgIdent(I)) {
+      IdentifierLoc *IdLoc = AL.getArgAsIdent(I);
+      auto It = NameIdxMapping.find(IdLoc->Ident->getName());
+      if (It == UnknownName) {
+        S.Diag(AL.getLoc(), diag::err_callback_attribute_argument_unknown)
+            << IdLoc->Ident << IdLoc->Loc;
+        return;
+      }
+
+      SR = SourceRange(IdLoc->Loc);
+      ArgIdx = It->second;
+    } else if (AL.isArgExpr(I)) {
+      Expr *IdxExpr = AL.getArgAsExpr(I);
+
+      // If the expression is not parseable as an int32_t we have a problem.
+      if (!checkUInt32Argument(S, AL, IdxExpr, (uint32_t &)ArgIdx, I + 1,
+                               false)) {
+        S.Diag(AL.getLoc(), diag::err_attribute_argument_out_of_bounds)
+            << AL << (I + 1) << IdxExpr->getSourceRange();
+        return;
+      }
+
+      // Check oob, excluding the special values, 0 and -1.
+      if (ArgIdx < -1 || ArgIdx > NumArgs) {
+        S.Diag(AL.getLoc(), diag::err_attribute_argument_out_of_bounds)
+            << AL << (I + 1) << IdxExpr->getSourceRange();
+        return;
+      }
+
+      SR = IdxExpr->getSourceRange();
+    } else {
+      llvm_unreachable("Unexpected ParsedAttr argument type!");
+    }
+
+    if (ArgIdx == 0 && !HasImplicitThisParam) {
+      S.Diag(AL.getLoc(), diag::err_callback_implicit_this_not_available)
+          << (I + 1) << SR;
+      return;
+    }
+
+    // Adjust for the case we do not have an implicit "this" parameter. In this
+    // case we decrease all positive values by 1 to get LLVM argument indices.
+    if (!HasImplicitThisParam && ArgIdx > 0)
+      ArgIdx -= 1;
+
+    EncodingIndices.push_back(ArgIdx);
+  }
+
+  int CalleeIdx = EncodingIndices.front();
+  // Check if the callee index is proper, thus not "this" and not "unknown".
+  if (CalleeIdx < HasImplicitThisParam) {
+    S.Diag(AL.getLoc(), diag::err_callback_attribute_invalid_callee)
+        << AL.getRange();
+    return;
+  }
+
+  // Get the callee type, note the index adjustment as the AST doesn't contain
+  // the this type (which the callee cannot reference anyway!).
+  const Type *CalleeType =
+      getFunctionOrMethodParamType(D, CalleeIdx - HasImplicitThisParam)
+          .getTypePtr();
+  if (!CalleeType || !CalleeType->isFunctionPointerType()) {
+    S.Diag(AL.getLoc(), diag::err_callback_callee_no_function_type)
+        << AL.getRange();
+    return;
+  }
+
+  const Type *CalleeFnType =
+      CalleeType->getPointeeType()->getUnqualifiedDesugaredType();
+
+  // TODO: Check the type of the callee arguments.
+
+  const auto *CalleeFnProtoType = dyn_cast<FunctionProtoType>(CalleeFnType);
+  if (!CalleeFnProtoType) {
+    S.Diag(AL.getLoc(), diag::err_callback_callee_no_function_type)
+        << AL.getRange();
+    return;
+  }
+
+  if (CalleeFnProtoType->getNumParams() > EncodingIndices.size() - 1) {
+    S.Diag(AL.getLoc(), diag::err_attribute_wrong_number_arguments)
+        << AL << (unsigned)(EncodingIndices.size() - 1);
+    return;
+  }
+
+  if (CalleeFnProtoType->getNumParams() < EncodingIndices.size() - 1) {
+    S.Diag(AL.getLoc(), diag::err_attribute_wrong_number_arguments)
+        << AL << (unsigned)(EncodingIndices.size() - 1);
+    return;
+  }
+
+  if (CalleeFnProtoType->isVariadic()) {
+    S.Diag(AL.getLoc(), diag::err_callback_callee_is_variadic) << AL.getRange();
+    return;
+  }
+
+  // Do not allow multiple callback attributes.
+  if (D->hasAttr<CallbackAttr>()) {
+    S.Diag(AL.getLoc(), diag::err_callback_attribute_multiple) << AL.getRange();
+    return;
+  }
+
+  D->addAttr(::new (S.Context) CallbackAttr(
+      AL.getRange(), S.Context, EncodingIndices.data(), EncodingIndices.size(),
+      AL.getAttributeSpellingListIndex()));
+}
+
 static void handleTransparentUnionAttr(Sema &S, Decl *D, const ParsedAttr &AL) {
   // Try to find the underlying union declaration.
   RecordDecl *RD = nullptr;
@@ -6451,6 +6589,9 @@ static void ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D,
   case ParsedAttr::AT_FormatArg:
     handleFormatArgAttr(S, D, AL);
     break;
+  case ParsedAttr::AT_Callback:
+    handleCallbackAttr(S, D, AL);
+    break;
   case ParsedAttr::AT_CUDAGlobal:
     handleGlobalAttr(S, D, AL);
     break;