commit ab851939fc6baf883a1760f18f48adc93ba2b01d
author Yaxun Liu <Yaxun.Liu@amd.com> Tue Mar 05 16:07:56 2019 +0000
committer Yaxun Liu <Yaxun.Liu@amd.com> Tue Mar 05 16:07:56 2019 +0000
tree c98f37b4e684f283a24c1e3d8e5e991bd4ec3108
parent 130322e7cc58b7236fd2181126adedc4d90d2322

[HIP] Do not unbundle object files for -fno-gpu-rdc

When -fno-gpu-rdc is set, device code is compiled, linked, and assembled into fat binary
and embedded as string in object files. The object files are normal object files which
can be linked by host linker. In the linking stage, the object files should not be unbundled
when -fno-gpu-rdc is set since they are normal object files, not bundles. The object files
only need to be unbundled when -fgpu-rdc is set.

Currently clang always unbundles object files, disregarding -fgpu-rdc option.

This patch fixes that.

Differential Revision: https://reviews.llvm.org/D58917

llvm-svn: 355410

clang/lib/Driver/Driver.cpp

diff --git a/clang/lib/Driver/Driver.cpp b/clang/lib/Driver/Driver.cpp
index eb03e6e..de6a67b 100644
--- a/clang/lib/Driver/Driver.cpp
+++ b/clang/lib/Driver/Driver.cpp
@@ -2293,6 +2293,9 @@
 
     /// Flag that is set to true if this builder acted on the current input.
     bool IsActive = false;
+
+    /// Flag for -fgpu-rdc.
+    bool Relocatable = false;
   public:
     CudaActionBuilderBase(Compilation &C, DerivedArgList &Args,
                           const Driver::InputList &Inputs,
@@ -2338,6 +2341,12 @@
 
       // If this is an unbundling action use it as is for each CUDA toolchain.
       if (auto *UA = dyn_cast<OffloadUnbundlingJobAction>(HostAction)) {
+
+        // If -fgpu-rdc is disabled, should not unbundle since there is no
+        // device code to link.
+        if (!Relocatable)
+          return ABRT_Inactive;
+
         CudaDeviceActions.clear();
         auto *IA = cast<InputAction>(UA->getInputs().back());
         std::string FileName = IA->getInputArg().getAsString(Args);
@@ -2409,6 +2418,9 @@
           !C.hasOffloadToolChain<Action::OFK_HIP>())
         return false;
 
+      Relocatable = Args.hasFlag(options::OPT_fgpu_rdc,
+          options::OPT_fno_gpu_rdc, /*Default=*/false);
+
       const ToolChain *HostTC = C.getSingleOffloadToolChain<Action::OFK_Host>();
       assert(HostTC && "No toolchain for host compilation.");
       if (HostTC->getTriple().isNVPTX() ||
@@ -2594,13 +2606,11 @@
   class HIPActionBuilder final : public CudaActionBuilderBase {
     /// The linker inputs obtained for each device arch.
     SmallVector<ActionList, 8> DeviceLinkerInputs;
-    bool Relocatable;
 
   public:
     HIPActionBuilder(Compilation &C, DerivedArgList &Args,
                      const Driver::InputList &Inputs)
-        : CudaActionBuilderBase(C, Args, Inputs, Action::OFK_HIP),
-          Relocatable(false) {}
+        : CudaActionBuilderBase(C, Args, Inputs, Action::OFK_HIP) {}
 
     bool canUseBundlerUnbundler() const override { return true; }
 
@@ -2705,13 +2715,6 @@
         ++I;
       }
     }
-
-    bool initialize() override {
-      Relocatable = Args.hasFlag(options::OPT_fgpu_rdc,
-          options::OPT_fno_gpu_rdc, /*Default=*/false);
-
-      return CudaActionBuilderBase::initialize();
-    }
   };
 
   /// OpenMP action builder. The host bitcode is passed to the device frontend