[OpenMP] Add support for auxiliary triple specification

Summary: Device offloading requires the specification of an additional flag containing the triple of the //other// architecture the code is being compiled on if such an architecture exists. If compiling for the host, the auxiliary triple flag will contain the triple describing the device and vice versa.

Reviewers: arpith-jacob, sfantao, caomhin, carlo.bertolli, ABataev, Hahnfeld, jlebar, hfinkel, tstellar

Reviewed By: Hahnfeld

Subscribers: rengolin, cfe-commits

Differential Revision: https://reviews.llvm.org/D29339

llvm-svn: 306689

4 files changed, 25 insertions, 3 deletions

diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
index 072d62346cd..3b9805b5a92 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -129,6 +129,13 @@ forAllAssociatedToolChains(Compilation &C, const JobAction &JA,
   else if (JA.isDeviceOffloading(Action::OFK_Cuda))
     Work(*C.getSingleOffloadToolChain<Action::OFK_Host>());
 
+  if (JA.isHostOffloading(Action::OFK_OpenMP)) {
+    auto TCs = C.getOffloadToolChains<Action::OFK_OpenMP>();
+    for (auto II = TCs.first, IE = TCs.second; II != IE; ++II)
+      Work(*II->second);
+  } else if (JA.isDeviceOffloading(Action::OFK_OpenMP))
+    Work(*C.getSingleOffloadToolChain<Action::OFK_Host>());
+
   //
   // TODO: Add support for other offloading programming models here.
   //
@@ -1992,6 +1999,16 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
     CmdArgs.push_back(Args.MakeArgString(NormalizedTriple));
   }
 
+  if (IsOpenMPDevice) {
+    // We have to pass the triple of the host if compiling for an OpenMP device.
+    std::string NormalizedTriple =
+        C.getSingleOffloadToolChain<Action::OFK_Host>()
+            ->getTriple()
+            .normalize();
+    CmdArgs.push_back("-aux-triple");
+    CmdArgs.push_back(Args.MakeArgString(NormalizedTriple));
+  }
+
   if (Triple.isOSWindows() && (Triple.getArch() == llvm::Triple::arm ||
                                Triple.getArch() == llvm::Triple::thumb)) {
     unsigned Offset = Triple.getArch() == llvm::Triple::arm ? 4 : 6;
diff --git a/clang/lib/Frontend/CompilerInstance.cpp b/clang/lib/Frontend/CompilerInstance.cpp
index 3b5335478db..9be5f9cd1bd 100644
--- a/clang/lib/Frontend/CompilerInstance.cpp
+++ b/clang/lib/Frontend/CompilerInstance.cpp
@@ -936,8 +936,9 @@ bool CompilerInstance::ExecuteAction(FrontendAction &Act) {
   if (!hasTarget())
     return false;
 
-  // Create TargetInfo for the other side of CUDA compilation.
-  if (getLangOpts().CUDA && !getFrontendOpts().AuxTriple.empty()) {
+  // Create TargetInfo for the other side of CUDA and OpenMP compilation.
+  if ((getLangOpts().CUDA || getLangOpts().OpenMPIsDevice) &&
+      !getFrontendOpts().AuxTriple.empty()) {
     auto TO = std::make_shared<TargetOptions>();
     TO->Triple = getFrontendOpts().AuxTriple;
     TO->HostTriple = getTarget().getTriple().str();
diff --git a/clang/lib/Frontend/CompilerInvocation.cpp b/clang/lib/Frontend/CompilerInvocation.cpp
index 6254b0013ba..c49fda97524 100644
--- a/clang/lib/Frontend/CompilerInvocation.cpp
+++ b/clang/lib/Frontend/CompilerInvocation.cpp
@@ -2644,6 +2644,10 @@ bool CompilerInvocation::CreateFromArgs(CompilerInvocation &Res,
       Res.getTargetOpts().HostTriple = Res.getFrontendOpts().AuxTriple;
   }
 
+  // Set the triple of the host for OpenMP device compile.
+  if (LangOpts.OpenMPIsDevice)
+    Res.getTargetOpts().HostTriple = Res.getFrontendOpts().AuxTriple;
+
   // FIXME: Override value name discarding when asan or msan is used because the
   // backend passes depend on the name of the alloca in order to print out
   // names.
diff --git a/clang/lib/Frontend/InitPreprocessor.cpp b/clang/lib/Frontend/InitPreprocessor.cpp
index 08befb33c96..71420df0002 100644
--- a/clang/lib/Frontend/InitPreprocessor.cpp
+++ b/clang/lib/Frontend/InitPreprocessor.cpp
@@ -1043,7 +1043,7 @@ void clang::InitializePreprocessor(
   if (InitOpts.UsePredefines) {
     // FIXME: This will create multiple definitions for most of the predefined
     // macros. This is not the right way to handle this.
-    if (LangOpts.CUDA && PP.getAuxTargetInfo())
+    if ((LangOpts.CUDA || LangOpts.OpenMPIsDevice) && PP.getAuxTargetInfo())
       InitializePredefinedMacros(*PP.getAuxTargetInfo(), LangOpts, FEOpts,
                                  Builder);