[CUDA] Allow parsing of host and device code simultaneously.
* adds -aux-triple option to specify target triple
* propagates aux target info to AST context and Preprocessor
* pulls in target specific preprocessor macros.
* pulls in target-specific builtins from aux target.
* sets appropriate host or device attribute on builtins.
Differential Revision: http://reviews.llvm.org/D12917
llvm-svn: 248299
diff --git a/clang/lib/AST/ASTContext.cpp b/clang/lib/AST/ASTContext.cpp
index 3e0d6d11..b98e3d4 100644
--- a/clang/lib/AST/ASTContext.cpp
+++ b/clang/lib/AST/ASTContext.cpp
@@ -743,10 +743,10 @@
FirstLocalImport(), LastLocalImport(), ExternCContext(nullptr),
SourceMgr(SM), LangOpts(LOpts),
SanitizerBL(new SanitizerBlacklist(LangOpts.SanitizerBlacklistFiles, SM)),
- AddrSpaceMap(nullptr), Target(nullptr), PrintingPolicy(LOpts),
- Idents(idents), Selectors(sels), BuiltinInfo(builtins),
- DeclarationNames(*this), ExternalSource(nullptr), Listener(nullptr),
- Comments(SM), CommentsLoaded(false),
+ AddrSpaceMap(nullptr), Target(nullptr), AuxTarget(nullptr),
+ PrintingPolicy(LOpts), Idents(idents), Selectors(sels),
+ BuiltinInfo(builtins), DeclarationNames(*this), ExternalSource(nullptr),
+ Listener(nullptr), Comments(SM), CommentsLoaded(false),
CommentCommandTraits(BumpAlloc, LOpts.CommentOpts), LastSDM(nullptr, 0) {
TUDecl = TranslationUnitDecl::Create(*this);
}
@@ -956,13 +956,15 @@
Types.push_back(Ty);
}
-void ASTContext::InitBuiltinTypes(const TargetInfo &Target) {
+void ASTContext::InitBuiltinTypes(const TargetInfo &Target,
+ const TargetInfo *AuxTarget) {
assert((!this->Target || this->Target == &Target) &&
"Incorrect target reinitialization");
assert(VoidTy.isNull() && "Context reinitialized?");
this->Target = &Target;
-
+ this->AuxTarget = AuxTarget;
+
ABI.reset(createCXXABI(Target));
AddrSpaceMap = getAddressSpaceMap(Target, LangOpts);
AddrSpaceMapMangling = isAddrSpaceMapManglingEnabled(Target, LangOpts);
diff --git a/clang/lib/Basic/Builtins.cpp b/clang/lib/Basic/Builtins.cpp
index 13e09b2..cb8a0b3 100644
--- a/clang/lib/Basic/Builtins.cpp
+++ b/clang/lib/Basic/Builtins.cpp
@@ -32,19 +32,27 @@
const Builtin::Info &Builtin::Context::getRecord(unsigned ID) const {
if (ID < Builtin::FirstTSBuiltin)
return BuiltinInfo[ID];
- assert(ID - Builtin::FirstTSBuiltin < NumTSRecords && "Invalid builtin ID!");
+ assert(ID - Builtin::FirstTSBuiltin < (NumTSRecords + NumAuxTSRecords) &&
+ "Invalid builtin ID!");
+ if (isAuxBuiltinID(ID))
+ return AuxTSRecords[getAuxBuiltinID(ID) - Builtin::FirstTSBuiltin];
return TSRecords[ID - Builtin::FirstTSBuiltin];
}
Builtin::Context::Context() {
// Get the target specific builtins from the target.
TSRecords = nullptr;
+ AuxTSRecords = nullptr;
NumTSRecords = 0;
+ NumAuxTSRecords = 0;
}
-void Builtin::Context::initializeTarget(const TargetInfo &Target) {
+void Builtin::Context::InitializeTarget(const TargetInfo &Target,
+ const TargetInfo *AuxTarget) {
assert(NumTSRecords == 0 && "Already initialized target?");
Target.getTargetBuiltins(TSRecords, NumTSRecords);
+ if (AuxTarget)
+ AuxTarget->getTargetBuiltins(AuxTSRecords, NumAuxTSRecords);
}
bool Builtin::Context::builtinIsSupported(const Builtin::Info &BuiltinInfo,
@@ -76,7 +84,12 @@
// Step #2: Register target-specific builtins.
for (unsigned i = 0, e = NumTSRecords; i != e; ++i)
if (builtinIsSupported(TSRecords[i], LangOpts))
- Table.get(TSRecords[i].Name).setBuiltinID(i+Builtin::FirstTSBuiltin);
+ Table.get(TSRecords[i].Name).setBuiltinID(i + Builtin::FirstTSBuiltin);
+
+ // Step #3: Register target-specific builtins for AuxTarget.
+ for (unsigned i = 0, e = NumAuxTSRecords; i != e; ++i)
+ Table.get(AuxTSRecords[i].Name)
+ .setBuiltinID(i + Builtin::FirstTSBuiltin + NumTSRecords);
}
void Builtin::Context::forgetBuiltin(unsigned ID, IdentifierTable &Table) {
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 91614a5..047cf4f 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -1868,40 +1868,54 @@
return GetUndefRValue(E->getType());
}
-Value *CodeGenFunction::EmitTargetBuiltinExpr(unsigned BuiltinID,
- const CallExpr *E) {
- switch (getTarget().getTriple().getArch()) {
+static Value *EmitTargetArchBuiltinExpr(CodeGenFunction *CGF,
+ unsigned BuiltinID, const CallExpr *E,
+ llvm::Triple::ArchType Arch) {
+ switch (Arch) {
case llvm::Triple::arm:
case llvm::Triple::armeb:
case llvm::Triple::thumb:
case llvm::Triple::thumbeb:
- return EmitARMBuiltinExpr(BuiltinID, E);
+ return CGF->EmitARMBuiltinExpr(BuiltinID, E);
case llvm::Triple::aarch64:
case llvm::Triple::aarch64_be:
- return EmitAArch64BuiltinExpr(BuiltinID, E);
+ return CGF->EmitAArch64BuiltinExpr(BuiltinID, E);
case llvm::Triple::x86:
case llvm::Triple::x86_64:
- return EmitX86BuiltinExpr(BuiltinID, E);
+ return CGF->EmitX86BuiltinExpr(BuiltinID, E);
case llvm::Triple::ppc:
case llvm::Triple::ppc64:
case llvm::Triple::ppc64le:
- return EmitPPCBuiltinExpr(BuiltinID, E);
+ return CGF->EmitPPCBuiltinExpr(BuiltinID, E);
case llvm::Triple::r600:
case llvm::Triple::amdgcn:
- return EmitAMDGPUBuiltinExpr(BuiltinID, E);
+ return CGF->EmitAMDGPUBuiltinExpr(BuiltinID, E);
case llvm::Triple::systemz:
- return EmitSystemZBuiltinExpr(BuiltinID, E);
+ return CGF->EmitSystemZBuiltinExpr(BuiltinID, E);
case llvm::Triple::nvptx:
case llvm::Triple::nvptx64:
- return EmitNVPTXBuiltinExpr(BuiltinID, E);
+ return CGF->EmitNVPTXBuiltinExpr(BuiltinID, E);
case llvm::Triple::wasm32:
case llvm::Triple::wasm64:
- return EmitWebAssemblyBuiltinExpr(BuiltinID, E);
+ return CGF->EmitWebAssemblyBuiltinExpr(BuiltinID, E);
default:
return nullptr;
}
}
+Value *CodeGenFunction::EmitTargetBuiltinExpr(unsigned BuiltinID,
+ const CallExpr *E) {
+ if (getContext().BuiltinInfo.isAuxBuiltinID(BuiltinID)) {
+ assert(getContext().getAuxTargetInfo() && "Missing aux target info");
+ return EmitTargetArchBuiltinExpr(
+ this, getContext().BuiltinInfo.getAuxBuiltinID(BuiltinID), E,
+ getContext().getAuxTargetInfo()->getTriple().getArch());
+ }
+
+ return EmitTargetArchBuiltinExpr(this, BuiltinID, E,
+ getTarget().getTriple().getArch());
+}
+
static llvm::VectorType *GetNeonType(CodeGenFunction *CGF,
NeonTypeFlags TypeFlags,
bool V1Ty=false) {
diff --git a/clang/lib/Frontend/CompilerInstance.cpp b/clang/lib/Frontend/CompilerInstance.cpp
index 3326b81..d6e5732 100644
--- a/clang/lib/Frontend/CompilerInstance.cpp
+++ b/clang/lib/Frontend/CompilerInstance.cpp
@@ -78,9 +78,8 @@
Diagnostics = Value;
}
-void CompilerInstance::setTarget(TargetInfo *Value) {
- Target = Value;
-}
+void CompilerInstance::setTarget(TargetInfo *Value) { Target = Value; }
+void CompilerInstance::setAuxTarget(TargetInfo *Value) { AuxTarget = Value; }
void CompilerInstance::setFileManager(FileManager *Value) {
FileMgr = Value;
@@ -312,7 +311,7 @@
PP = new Preprocessor(&getPreprocessorOpts(), getDiagnostics(), getLangOpts(),
getSourceManager(), *HeaderInfo, *this, PTHMgr,
/*OwnsHeaderSearch=*/true, TUKind);
- PP->Initialize(getTarget());
+ PP->Initialize(getTarget(), getAuxTarget());
// Note that this is different then passing PTHMgr to Preprocessor's ctor.
// That argument is used as the IdentifierInfoLookup argument to
@@ -396,7 +395,7 @@
auto *Context = new ASTContext(getLangOpts(), PP.getSourceManager(),
PP.getIdentifierTable(), PP.getSelectorTable(),
PP.getBuiltinInfo());
- Context->InitBuiltinTypes(getTarget());
+ Context->InitBuiltinTypes(getTarget(), getAuxTarget());
setASTContext(Context);
}
@@ -800,6 +799,13 @@
if (!hasTarget())
return false;
+ // Create TargetInfo for the other side of CUDA compilation.
+ if (getLangOpts().CUDA && !getFrontendOpts().AuxTriple.empty()) {
+ std::shared_ptr<TargetOptions> TO(new TargetOptions);
+ TO->Triple = getFrontendOpts().AuxTriple;
+ setAuxTarget(TargetInfo::CreateTargetInfo(getDiagnostics(), TO));
+ }
+
// Inform the target of the language options.
//
// FIXME: We shouldn't need to do this, the target should be immutable once
diff --git a/clang/lib/Frontend/CompilerInvocation.cpp b/clang/lib/Frontend/CompilerInvocation.cpp
index d799562..a907133 100644
--- a/clang/lib/Frontend/CompilerInvocation.cpp
+++ b/clang/lib/Frontend/CompilerInvocation.cpp
@@ -975,6 +975,9 @@
Opts.OverrideRecordLayoutsFile
= Args.getLastArgValue(OPT_foverride_record_layout_EQ);
+ Opts.AuxTriple =
+ llvm::Triple::normalize(Args.getLastArgValue(OPT_aux_triple));
+
if (const Arg *A = Args.getLastArg(OPT_arcmt_check,
OPT_arcmt_modify,
OPT_arcmt_migrate)) {
diff --git a/clang/lib/Frontend/InitPreprocessor.cpp b/clang/lib/Frontend/InitPreprocessor.cpp
index 0791494..0b445e8 100644
--- a/clang/lib/Frontend/InitPreprocessor.cpp
+++ b/clang/lib/Frontend/InitPreprocessor.cpp
@@ -918,6 +918,10 @@
// Install things like __POWERPC__, __GNUC__, etc into the macro table.
if (InitOpts.UsePredefines) {
+ if (LangOpts.CUDA && PP.getAuxTargetInfo())
+ InitializePredefinedMacros(*PP.getAuxTargetInfo(), LangOpts, FEOpts,
+ Builder);
+
InitializePredefinedMacros(PP.getTargetInfo(), LangOpts, FEOpts, Builder);
// Install definitions to make Objective-C++ ARC work well with various
diff --git a/clang/lib/Lex/Preprocessor.cpp b/clang/lib/Lex/Preprocessor.cpp
index 29a00d3..82fcde5 100644
--- a/clang/lib/Lex/Preprocessor.cpp
+++ b/clang/lib/Lex/Preprocessor.cpp
@@ -62,20 +62,19 @@
IdentifierInfoLookup *IILookup, bool OwnsHeaders,
TranslationUnitKind TUKind)
: PPOpts(PPOpts), Diags(&diags), LangOpts(opts), Target(nullptr),
- FileMgr(Headers.getFileMgr()), SourceMgr(SM),
- ScratchBuf(new ScratchBuffer(SourceMgr)),HeaderInfo(Headers),
+ AuxTarget(nullptr), FileMgr(Headers.getFileMgr()), SourceMgr(SM),
+ ScratchBuf(new ScratchBuffer(SourceMgr)), HeaderInfo(Headers),
TheModuleLoader(TheModuleLoader), ExternalSource(nullptr),
Identifiers(opts, IILookup),
PragmaHandlers(new PragmaNamespace(StringRef())),
- IncrementalProcessing(false), TUKind(TUKind),
- CodeComplete(nullptr), CodeCompletionFile(nullptr),
- CodeCompletionOffset(0), LastTokenWasAt(false),
- ModuleImportExpectsIdentifier(false), CodeCompletionReached(0),
- MainFileDir(nullptr), SkipMainFilePreamble(0, true), CurPPLexer(nullptr),
- CurDirLookup(nullptr), CurLexerKind(CLK_Lexer), CurSubmodule(nullptr),
- Callbacks(nullptr), CurSubmoduleState(&NullSubmoduleState),
- MacroArgCache(nullptr), Record(nullptr),
- MIChainHead(nullptr), DeserialMIChainHead(nullptr) {
+ IncrementalProcessing(false), TUKind(TUKind), CodeComplete(nullptr),
+ CodeCompletionFile(nullptr), CodeCompletionOffset(0),
+ LastTokenWasAt(false), ModuleImportExpectsIdentifier(false),
+ CodeCompletionReached(0), MainFileDir(nullptr),
+ SkipMainFilePreamble(0, true), CurPPLexer(nullptr), CurDirLookup(nullptr),
+ CurLexerKind(CLK_Lexer), CurSubmodule(nullptr), Callbacks(nullptr),
+ CurSubmoduleState(&NullSubmoduleState), MacroArgCache(nullptr),
+ Record(nullptr), MIChainHead(nullptr), DeserialMIChainHead(nullptr) {
OwnsHeaderSearch = OwnsHeaders;
CounterValue = 0; // __COUNTER__ starts at 0.
@@ -170,13 +169,18 @@
delete &HeaderInfo;
}
-void Preprocessor::Initialize(const TargetInfo &Target) {
+void Preprocessor::Initialize(const TargetInfo &Target,
+ const TargetInfo *AuxTarget) {
assert((!this->Target || this->Target == &Target) &&
"Invalid override of target information");
this->Target = &Target;
-
+
+ assert((!this->AuxTarget || this->AuxTarget == AuxTarget) &&
+ "Invalid override of aux target information.");
+ this->AuxTarget = AuxTarget;
+
// Initialize information about built-ins.
- BuiltinInfo.initializeTarget(Target);
+ BuiltinInfo.InitializeTarget(Target, AuxTarget);
HeaderInfo.setTarget(Target);
}
diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
index 22d2d34..9195a06 100644
--- a/clang/lib/Sema/SemaDecl.cpp
+++ b/clang/lib/Sema/SemaDecl.cpp
@@ -11293,11 +11293,11 @@
if (getLangOpts().CUDA && getLangOpts().CUDATargetOverloads &&
Context.BuiltinInfo.isTSBuiltin(BuiltinID) &&
!FD->hasAttr<CUDADeviceAttr>() && !FD->hasAttr<CUDAHostAttr>()) {
- // Target-specific builtins are assumed to be intended for use
- // in this particular CUDA compilation mode and should have
- // appropriate attribute set so we can enforce CUDA function
- // call restrictions.
- if (getLangOpts().CUDAIsDevice)
+ // Assign appropriate attribute depending on CUDA compilation
+ // mode and the target builtin belongs to. E.g. during host
+ // compilation, aux builtins are __device__, the rest are __host__.
+ if (getLangOpts().CUDAIsDevice !=
+ Context.BuiltinInfo.isAuxBuiltinID(BuiltinID))
FD->addAttr(CUDADeviceAttr::CreateImplicit(Context, FD->getLocation()));
else
FD->addAttr(CUDAHostAttr::CreateImplicit(Context, FD->getLocation()));