[Clang] Forward -Xarch_<arch> -Wl,foo for GPU toolchains (#126248)

Summary:
This patch handles the use of `-Xarch_<arch> -Wl,foo` to send an
argument to the linker for the embedded offloading jobs in the linker
wrapper. This makes it equivalent to `-Xoffload-linker foo`.

Author: jhuber6

clang/lib/Driver/ToolChains/Clang.cpp

@@ -9187,7 +9187,7 @@ void LinkerWrapper::ConstructJob(Compilation &C, const JobAction &JA,
       OPT_fno_lto,
       OPT_flto,
       OPT_flto_EQ};
-  const llvm::DenseSet<unsigned> LinkerOptions{OPT_mllvm};
+  const llvm::DenseSet<unsigned> LinkerOptions{OPT_mllvm, OPT_Zlinker_input};
   auto ShouldForward = [&](const llvm::DenseSet<unsigned> &Set, Arg *A) {
     return Set.contains(A->getOption().getID()) ||
            (A->getOption().getGroup().isValid() &&
@@ -9205,7 +9205,9 @@ void LinkerWrapper::ConstructJob(Compilation &C, const JobAction &JA,
       ArgStringList CompilerArgs;
       ArgStringList LinkerArgs;
       for (Arg *A : C.getArgsForToolChain(TC, /*BoundArch=*/"", Kind)) {
-        if (ShouldForward(CompilerOptions, A))
+        if (A->getOption().matches(OPT_Zlinker_input))
+          LinkerArgs.emplace_back(A->getValue());
+        else if (ShouldForward(CompilerOptions, A))
           A->render(Args, CompilerArgs);
         else if (ShouldForward(LinkerOptions, A))
           A->render(Args, LinkerArgs);

clang/test/Driver/offload-Xarch.c

@@ -14,7 +14,7 @@
 // RUN:   --target=x86_64-unknown-linux-gnu -Xopenmp-target=nvptx64-nvidia-cuda --offload-arch=sm_52,sm_60 -nogpuinc \
 // RUN:   -Xopenmp-target=amdgcn-amd-amdhsa --offload-arch=gfx90a,gfx1030 -ccc-print-bindings -### %s 2>&1 \
 // RUN: | FileCheck -check-prefix=OPENMP %s
-//
+
 // OPENMP: # "x86_64-unknown-linux-gnu" - "clang", inputs: ["[[INPUT:.+]]"], output: "[[HOST_BC:.+]]"
 // OPENMP: # "amdgcn-amd-amdhsa" - "clang", inputs: ["[[INPUT]]", "[[HOST_BC]]"], output: "[[GFX1030_BC:.+]]"
 // OPENMP: # "amdgcn-amd-amdhsa" - "clang", inputs: ["[[INPUT]]", "[[HOST_BC]]"], output: "[[GFX90A_BC:.+]]"
@@ -32,3 +32,12 @@
 // CUDA: "-cc1" "-triple" "nvptx64-nvidia-cuda" {{.*}}"-target-cpu" "sm_52" {{.*}}"-O3"
 // CUDA: "-cc1" "-triple" "nvptx64-nvidia-cuda" {{.*}}"-target-cpu" "sm_60" {{.*}}"-O0"
 // CUDA: "-cc1" "-triple" "x86_64-unknown-linux-gnu" {{.*}}"-O3"
+
+// Make sure that `-Xarch_amdgcn` forwards libraries to the device linker.
+// RUN: %clang -fopenmp=libomp --offload-arch=gfx90a -nogpulib -nogpuinc \
+// RUN:   -Xarch_amdgcn -Wl,-lfoo -### %s 2>&1 \
+// RUN: | FileCheck -check-prefix=LIBS %s
+// RUN: %clang -fopenmp=libomp --offload-arch=gfx90a -nogpulib -nogpuinc \
+// RUN:   -Xoffload-linker-amdgcn-amd-amdhsa -lfoo -### %s 2>&1 \
+// RUN: | FileCheck -check-prefix=LIBS %s
+// LIBS: "--device-linker=amdgcn-amd-amdhsa=-lfoo"