[OpenMP] Fix for --opaque-offload-linker. Default remains --no-opaque-offload-linker. (llvm#1348)

Fixed lit test.

Author: gregrodgers

clang/include/clang/Driver/Options.td

@@ -3822,10 +3822,17 @@ defm openmp_optimistic_collapse : BoolFOption<"openmp-optimistic-collapse",
   NegFlag<SetFalse>, BothFlags<[NoArgumentUnused, HelpHidden], []>>;
 def static_openmp: Flag<["-"], "static-openmp">,
   HelpText<"Use the static host OpenMP runtime while linking.">;
-def opaque_offload_linker : Flag<["--"], "opaque-offload-linker">, Visibility<[ClangOption, CC1Option]>, Group<f_Group>,
-  HelpText<"Build/link omp offload binary, construct opaque cmd list instead of single clang-linker-wrapper cmd.">;
-def no_opaque_offload_linker : Flag<["--"], "no-opaque-offload-linker">, Visibility<[ClangOption, CC1Option]>, Group<f_Group>,
-  HelpText<"Build/link omp offload binary, using single clang-linker-wrapper cmd.">;
+def opaque_offload_linker
+    : Flag<["--"], "opaque-offload-linker">,
+      Visibility<[ClangOption, FlangOption]>,
+      Group<f_Group>,
+      HelpText<"Build/link omp offload binary, construct opaque cmd list "
+               "instead of single clang-linker-wrapper cmd.">;
+def no_opaque_offload_linker : Flag<["--"], "no-opaque-offload-linker">,
+                               Visibility<[ClangOption, FlangOption]>,
+                               Group<f_Group>,
+                               HelpText<"Build/link omp offload binary, using "
+                                        "single clang-linker-wrapper cmd.">;
 def fopenmp_new_driver : Flag<["-"], "fopenmp-new-driver">, Flags<[HelpHidden]>,
   HelpText<"Use the new driver for OpenMP offloading.">;
 def fno_openmp_new_driver : Flag<["-"], "fno-openmp-new-driver">,

clang/lib/Driver/ToolChains/AMDGPUOpenMP.cpp

@@ -269,10 +269,11 @@ const char *amdgpu::dlr::getLinkCommandArgs(
           << "libomptarget-amdgpu.bc";
     }
 
-    // Add the generic set of libraries, OpenMP subset only
-    BCLibs.append(amdgpu::dlr::getCommonDeviceLibNames(
-        C.getArgs(), TC.getSanitizerArgs(C.getArgs()), C.getDriver(),
-        GPUArch.str(), /* isOpenMP=*/true, RocmInstallation));
+    if (!Args.hasArg(options::OPT_no_offloadlib))
+      // Add the generic set of libraries, OpenMP subset only
+      BCLibs.append(amdgpu::dlr::getCommonDeviceLibNames(
+          C.getArgs(), TC.getSanitizerArgs(C.getArgs()), C.getDriver(),
+          GPUArch.str(), /* isOpenMP=*/true, RocmInstallation));
   }
 
   llvm::for_each(BCLibs, [&](auto BCLib) {

clang/lib/Driver/ToolChains/Clang.cpp

@@ -9370,6 +9370,24 @@ void LinkerWrapper::ConstructJob(Compilation &C, const JobAction &JA,
                                  const InputInfoList &Inputs,
                                  const ArgList &Args,
                                  const char *LinkingOutput) const {
+  bool isAMDGPU = false;
+  auto offloadTC = C.getOffloadToolChains(Action::OFK_OpenMP);
+  const auto OpenMPTCs = llvm::make_range(offloadTC.first, offloadTC.second);
+  const ToolChain *OTC;
+  for (auto &I : OpenMPTCs) {
+    OTC = I.second;
+    if (OTC->getTriple().isAMDGPU()) {
+      isAMDGPU = true;
+      break;
+    }
+  }
+  if (isAMDGPU && Args.hasFlag(options::OPT_opaque_offload_linker,
+                               options::OPT_no_opaque_offload_linker, false)) {
+    ConstructOpaqueJob(C, JA, Output, Inputs, Args, OTC->getTriple(),
+                       LinkingOutput);
+    return;
+  }
+
   using namespace options;
 
   // A list of permitted options that will be forwarded to the embedded device

clang/test/Driver/clang-offload-wrapper.c

@@ -27,33 +27,37 @@
 // ELF-WARNING: is not an ELF image, so notes cannot be added to it.
 // CHECK-IR: target triple = "x86_64-pc-linux-gnu"
 
-// CHECK-IR-DAG: [[ENTTY:%.+]] = type { ptr, ptr, i{{32|64}}, i32, i32 }
+// CHECK-IR-DAG: [[ENTTY:%.+]] = type { i64, i16, i16, i32, ptr, ptr, i64, i64, ptr }
 // CHECK-IR-DAG: [[IMAGETY:%.+]] = type { ptr, ptr, ptr, ptr }
 // CHECK-IR-DAG: [[DESCTY:%.+]] = type { i32, ptr, ptr, ptr }
-
-// CHECK-IR: [[ENTBEGIN:@.+]] = external hidden constant [[ENTTY]]
-// CHECK-IR: [[ENTEND:@.+]] = external hidden constant [[ENTTY]]
-
-// CHECK-IR: [[DUMMY:@.+]] = weak hidden constant [0 x [[ENTTY]]] zeroinitializer, section "omp_offloading_entries"
+//
+// CHECK-IR: [[ENTBEGIN:@.+]] = external hidden constant [0 x [[ENTTY]]]
+// CHECK-IR: [[ENTEND:@.+]] = external hidden constant [0 x [[ENTTY]]]
+// CHECK-IR: [[DUMMY:@.+]] = internal constant [0 x [[ENTTY]]] zeroinitializer, section "llvm_offload_entries", align 8
+// CHECK-IR: @llvm.compiler.used = appending global [1 x ptr] [ptr [[DUMMY]]], section "llvm.metadata"
 
 // CHECK-IR: [[BIN:@.+]] = internal unnamed_addr constant [[[SIZE:[0-9]+]] x i8] c"\10\FF\10\AD{{.*}}"
-
-// CHECK-IR: [[IMAGES:@.+]] = internal unnamed_addr constant [1 x [[IMAGETY]]] [[[IMAGETY]] { ptr [[BIN]], ptr getelementptr ([[[SIZE]] x i8], ptr [[BIN]], i64 0, i64 160), ptr [[ENTBEGIN]], ptr [[ENTEND]] }]
-
+  ffloading.device_image = internal unnamed_addr constant [[[SIZE]] x i8] c"\10\FF\10\AD\01\00\00\0
+// CHECK-IR: [[IMAGES:@.+]] = internal unnamed_addr constant [1 x %__tgt_device_image] [%__tgt_device_image { ptr getelementptr ([[[SIZE]] x i8], ptr [[BIN]], i64 0, i64 136), ptr getelementptr ([[[SIZE]] x i8], ptr [[BIN]], i64 0, i64 159), ptr [[ENTBEGIN]], ptr [[ENTEND]] }]
 // CHECK-IR: [[DESC:@.+]] = internal constant [[DESCTY]] { i32 1, ptr [[IMAGES]], ptr [[ENTBEGIN]], ptr [[ENTEND]] }
+// CHECK-IR: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 101, ptr [[REGFN:@.+]], ptr null }]
 
-// CHECK-IR: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 1, ptr [[REGFN:@.+]], ptr null }]
-
-// CHECK-IR: define internal void [[REGFN]]()
+// CHECK-IR: define internal void [[REGFN]]() section ".text.startup" {
 // CHECK-IR:   call void @__tgt_register_lib(ptr [[DESC]])
+// CHECK-IR:   %0 = call i32 @atexit(ptr @.omp_offloading.descriptor_unreg)
 // CHECK-IR:   ret void
+// CHECK-IR: }
 
 // CHECK-IR: declare void @__tgt_register_lib(ptr)
 
+// CHECK-IR: declare i32 @atexit(ptr)
+
+// CHECK-IR: define internal void [[DESC]]_unreg() section ".text.startup" {
 // CHECK-IR:   call void @__tgt_unregister_lib(ptr [[DESC]])
 // CHECK-IR:   ret void
+// CHECK-IR: }
 
-// CHECK-IR: declare void @__tgt_unregister_lib(ptr)
+// CHECK_IR: declare void @__tgt_unregister_lib(ptr)
 
 // Check that clang-offload-wrapper adds LLVMOMPOFFLOAD notes
 // into the ELF offload images: