diff --git a/clang/test/Driver/linker-wrapper.c b/clang/test/Driver/linker-wrapper.c
--- a/clang/test/Driver/linker-wrapper.c
+++ b/clang/test/Driver/linker-wrapper.c
@@ -130,3 +130,12 @@
 // RUN:   -o a.out 2>&1 | FileCheck %s --check-prefix=MISSING-LIBRARY
 
 // MISSING-LIBRARY: error: unable to find library -ldummy
+
+// RUN: clang-offload-packager -o %t.out \
+// RUN:   --image=file=%t.amdgpu.bc,kind=openmp,triple=amdgcn-amd-amdhsa,arch=gfx908 \
+// RUN:   --image=file=%t.amdgpu.bc,kind=openmp,triple=amdgcn-amd-amdhsa,arch=gfx908
+// RUN: %clang -cc1 %s -triple x86_64-unknown-linux-gnu -emit-obj -o %t.o -fembed-offload-object=%t.out
+// RUN: clang-linker-wrapper --host-triple=x86_64-unknown-linux-gnu --dry-run --clang-backend \
+// RUN:   --linker-path=/usr/bin/ld -- %t.o -o a.out 2>&1 | FileCheck %s --check-prefix=CLANG-BACKEND
+
+// CLANG-BACKEND: clang{{.*}} -o {{.*}}.img --target=amdgcn-amd-amdhsa -mcpu=gfx908 -O2 -Wl,--no-undefined {{.*}}.bc
diff --git a/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp b/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
--- a/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
+++ b/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
@@ -427,6 +427,17 @@
   for (StringRef Arg : Args.getAllArgValues(OPT_linker_arg_EQ))
     CmdArgs.push_back(Args.MakeArgString("-Wl," + Arg));
 
+  for (StringRef Arg : Args.getAllArgValues(OPT_builtin_bitcode_EQ)) {
+    if (llvm::Triple(Arg.split('=').first) == Triple)
+      CmdArgs.append({"-Xclang", "-mlink-builtin-bitcode", "-Xclang",
+                      Args.MakeArgString(Arg.split('=').second)});
+  }
+
+  // The OpenMPOpt pass can introduce new calls and is expensive, we do not want
+  // this when running CodeGen through clang.
+  if (Args.hasArg(OPT_clang_backend) || Args.hasArg(OPT_builtin_bitcode_EQ))
+    CmdArgs.append({"-mllvm", "-openmp-opt-disable"});
+
   if (Error Err = executeCommands(*ClangPath, CmdArgs))
     return std::move(Err);
 
@@ -629,7 +640,7 @@
   llvm::erase_if(InputFiles, [](OffloadFile &F) { return !F.getBinary(); });
 
   // LTO Module hook to output bitcode without running the backend.
-  SmallVector<StringRef, 4> BitcodeOutput;
+  SmallVector<StringRef> BitcodeOutput;
   auto OutputBitcode = [&](size_t, const Module &M) {
     auto TempFileOrErr = createOutputFile(sys::path::filename(ExecutableName) +
                                               "-jit-" + Triple.getTriple(),
@@ -648,7 +659,9 @@
 
   // We assume visibility of the whole program if every input file was bitcode.
   auto Features = getTargetFeatures(BitcodeInputFiles);
-  auto LTOBackend = Args.hasArg(OPT_embed_bitcode)
+  auto LTOBackend = Args.hasArg(OPT_embed_bitcode) ||
+                            Args.hasArg(OPT_builtin_bitcode_EQ) ||
+                            Args.hasArg(OPT_clang_backend)
                         ? createLTO(Args, Features, OutputBitcode)
                         : createLTO(Args, Features);
 
@@ -757,8 +770,12 @@
     return Error::success();
   }
 
-  // Append the new inputs to the device linker input.
-  for (StringRef File : Files)
+  // Append the new inputs to the device linker input. If the user requested an
+  // internalizing link we need to pass the bitcode to clang.
+  for (StringRef File :
+       Args.hasArg(OPT_clang_backend) || Args.hasArg(OPT_builtin_bitcode_EQ)
+           ? BitcodeOutput
+           : Files)
     OutputFiles.push_back(File);
 
   return Error::success();
diff --git a/clang/tools/clang-linker-wrapper/LinkerWrapperOpts.td b/clang/tools/clang-linker-wrapper/LinkerWrapperOpts.td
--- a/clang/tools/clang-linker-wrapper/LinkerWrapperOpts.td
+++ b/clang/tools/clang-linker-wrapper/LinkerWrapperOpts.td
@@ -25,9 +25,16 @@
 def bitcode_library_EQ : Joined<["--"], "bitcode-library=">,
   Flags<[WrapperOnlyOption]>, MetaVarName<"<kind>-<triple>-<arch>=<path>">,
   HelpText<"Extra bitcode library to link">;
+def builtin_bitcode_EQ : Joined<["--"], "builtin-bitcode=">,
+  Flags<[WrapperOnlyOption]>, MetaVarName<"<triple>=<path>">,
+  HelpText<"Perform a special internalizing link on the bitcode file. "
+           "This is necessary for some vendor libraries to be linked correctly">;
 def device_linker_args_EQ : Joined<["--"], "device-linker=">,
   Flags<[WrapperOnlyOption]>, MetaVarName<"<value> or <triple>=<value>">,
   HelpText<"Arguments to pass to the device linker invocation">;
+def clang_backend : Flag<["--"], "clang-backend">,
+  Flags<[WrapperOnlyOption]>,
+  HelpText<"Run the backend using clang rather than the LTO backend">;
 def dry_run : Flag<["--"], "dry-run">,
   Flags<[WrapperOnlyOption]>,
   HelpText<"Print program arguments without running">;