[OpenMP][AMDGPU] Link bitcode ROCm device libraries per-TU

[jhuber6]

Summary

Previously, we linked in the ROCm device libraries which provide math
and other utility functions late. This is not stricly correct as this
library contains several flags that are only set per-TU, such as fast
math or denormalization. This patch changes this to pass the bitcode
libraries per-TU using the same method we use for the CUDA libraries.
This has the advantage that we correctly propagate attributes making
this implementation more correct. Additionally, many annoying unused
functions were not being fully removed during LTO. This lead to
erroneous warning messages and remarks on unused functions.

I am not sure if not finding these libraries should be a hard error. let
me know if it should be demoted to a warning saying that some device
utilities will not work without them.

[JonChesterfield]

We can do this but should expect an increase in code size from having multiple internalised copies of the same function. There may be an incidental benefit if we can specialise some functions to the call site without additional cloning. Address of the same functions from different TUs will be inequal, which is wrong, but probably doesn't matter in practice.

It does have the major advantage that mlink-builtin-bitcode patches up the invalid IR on the fly, which is likely easier than changing the device libs or making IR mcpu-agnostic.

[JonChesterfield]

clang/lib/Driver/ToolChains/AMDGPUOpenMP.h

58

Why hip device libs? There's a common set, plus a hip.bc plus a opencl.bc. I'd expect us to want only the common set. Hip.bc shouldn't have non-hip stuff in it.

[jhuber6]

clang/lib/Driver/ToolChains/AMDGPUOpenMP.h

58

Existing virtual function, just re-used it.

[jhuber6]

In D133726#3785040, @JonChesterfield wrote:

We can do this but should expect an increase in code size from having multiple internalised copies of the same function. There may be an incidental benefit if we can specialise some functions to the call site without additional cloning. Address of the same functions from different TUs will be inequal, which is wrong, but probably doesn't matter in practice.

It does have the major advantage that mlink-builtin-bitcode patches up the invalid IR on the fly, which is likely easier than changing the device libs or making IR mcpu-agnostic.

It will probably decrease code size in the final executable now that this will forcefully internalize all the protected functions in ocml.bc that were sticking around because LTO couldn't remove them due to visibility.

[JonChesterfield]

clang/lib/Driver/ToolChains/AMDGPUOpenMP.h

58

Rename it rocm perhaps?

[jhuber6]

Renaming virtual function to make it more generic.

[jhuber6]

Adding a test for using -nogpulib.

[yaxunl]

clang/lib/Driver/ToolChains/AMDGPU.cpp

717–723

can we emit an error if both -march and -mcpu are specified and the values are different? This is a potential source of error.

[jhuber6]

clang/lib/Driver/ToolChains/AMDGPU.cpp

717–723

I'm not exactly sure about the semantics for this, maybe someone else can help. AFAIK -mcpu is sometimes treated as an alias of -mtune which is generally implied by -march. But having -march and -mtune state different things isn't technically disallowed. -march seems to imply that we generate code that can run on a certain architecture. Since these are in some respects equivalent it would probably be fine to just take the last one specified.

[arsenm]

Does this fix the weird behavior where you needed to use -lm to use anything in the device libraries? I don't see that being removed

[jhuber6]

In D133726#3786607, @arsenm wrote:

Does this fix the weird behavior where you needed to use -lm to use anything in the device libraries? I don't see that being removed

That was removed earlier when these files were just sent directly to the link job, you can see that removed from this patch.

[yaxunl]

clang/lib/Driver/ToolChains/AMDGPU.cpp

717–723

Some toolchains use -march to specify the target processor, some toolchains use -mcpu to specify the target processor. For AMDGPU toolchain, it uses -mcpu to specify the target processor in the beginning for OpenCL. Then we add HIP support and let HIPAMD toolchain adopt using -mcpu to specify the target processor. It is better to have a consistent way for specifying target processor for AMDGPU toolchains.

HIPAMD toolchain does this by https://github.com/llvm/llvm-project/blob/main/clang/lib/Driver/ToolChains/HIPAMD.cpp#L275 .

Probably AMDGPUOpenMP toolchain can make a similar change? Then we have a consistent way to specify target processor for AMDGPU toolchains.

[jhuber6]

clang/lib/Driver/ToolChains/AMDGPU.cpp

717–723

The rest of the OpenMP toolchain uses -march so I don't see a compelling reason to change it. If this is a stopper I'll just remove the changes to this function and check -march directly in the new getROCmDeviceLibs function for the OpenMP toolchain.

[yaxunl]

clang/lib/Driver/ToolChains/AMDGPU.cpp

717–723

There are a bunch of places in AMDGPU and HIPAMD toolchains that check the -mcpu option. If we allow -march as an alternative way to specify the target processor for the AMDGPU toolchain, it will introduce inconsistency and probably incur more changes.

Probably leaving this code in AMDGPUOpenMP toolchain is a better choice for now.

[JonChesterfield]

march for openmp, mcpu for hip seems ok. Notably llc needs to be told this as well, using mcpu, which may be an issue for save-temps

[jhuber6]

Removing use of getGPUArch and just using -march directly for OpenMP

[yaxunl]

clang/lib/Driver/ToolChains/AMDGPU.cpp

720–722

It seems the code still there.

[jhuber6]

clang/lib/Driver/ToolChains/AMDGPU.cpp

720–722

Sorry thought I reverted this file. I'll fix it.

[jhuber6]

Removing old function update for 'mcpu'

[yaxunl]

clang/include/clang/Driver/ToolChain.h

719

HIPSPV toolchain is not implemented on ROCm platform.

Probably we need to make it even more generic. How about getDeviceLibs ?

Also, you need to update the comment.

[jhuber6]

Changing interface to getAMDGPUDeviceLibs.

[yaxunl]

clang/include/clang/Driver/ToolChain.h

719

well, HIPSPV toolchain is not for AMDGPU. My thinking is that this API is to get device libraries for the toolchain, not for a specific target or platform.

[jhuber6]

clang/include/clang/Driver/ToolChain.h

719

I guess the problem is that this option can't be totally generic since it's in the top-level Toolchain namespace. Otherwise it would get confused with CUDA's device library. What's a good common name to use for it?

[yaxunl]

clang/include/clang/Driver/ToolChain.h

719

This API return device libs for the toolchain.

For AMDGPU, HIPAMD, AMDGPUOpenMP toolchains, it returns device libs for AMDGPU.

For HIPSPV toolchain, it returns device libs for HIP for Intel GPUs.

If a CUDA toolchain decides to support this API, it will return device libs for CUDA.

If a toolchain needs to return device libs for multiple targets, we can add parameters to control that, although so far there is no such need.

Therefore, I don't see issue to call it getDeviceLibs.

[jhuber6]

Changing to getDeviceLibs. I suppose in the future we could make this work for CUDA, but for now it won't be defined for that toolchain so it's fine.

[JonChesterfield]

I don't like this but will concede it's quicker than changing device libs to contain IR that doesn't have to be patched on the fly.

clang/include/clang/Driver/ToolChain.h

virtual void AddIAMCUIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                                    llvm::opt::ArgStringList &CC1Args) const;
 
   /// On Windows, returns the MSVC compatibility version.
   virtual VersionTuple computeMSVCVersion(const Driver *D,
                                           const llvm::opt::ArgList &Args) const;
 
   /// Get paths of HIP device libraries.
   virtual llvm::SmallVector<BitCodeLibraryInfo, 12>
-  getHIPDeviceLibs(const llvm::opt::ArgList &Args) const;
+  getROCmDeviceLibs(const llvm::opt::ArgList &Args) const;
 
   /// Add the system specific linker arguments to use
   /// for the given HIP runtime library type.
   virtual void AddHIPRuntimeLibArgs(const llvm::opt::ArgList &Args,
                                     llvm::opt::ArgStringList &CmdArgs) const {}
 
   /// Return sanitizers which are available in this toolchain.
   virtual SanitizerMask getSupportedSanitizers() const;

clang/lib/Driver/ToolChain.cpp

 
 void ToolChain::AddCudaIncludeArgs(const ArgList &DriverArgs,
                                    ArgStringList &CC1Args) const {}
 
 void ToolChain::AddHIPIncludeArgs(const ArgList &DriverArgs,
                                   ArgStringList &CC1Args) const {}
 
 llvm::SmallVector<ToolChain::BitCodeLibraryInfo, 12>
-ToolChain::getHIPDeviceLibs(const ArgList &DriverArgs) const {
+ToolChain::getROCmDeviceLibs(const ArgList &DriverArgs) const {
   return {};
 }
 
 void ToolChain::AddIAMCUIncludeArgs(const ArgList &DriverArgs,
                                     ArgStringList &CC1Args) const {}
 
 static VersionTuple separateMSVCFullVersion(unsigned Version) {
   if (Version < 100)

clang/lib/Driver/ToolChains/AMDGPU.cpp

if (!DriverArgs.hasArg(options::OPT_fvisibility_EQ,
                          options::OPT_fvisibility_ms_compat)) {
     CC1Args.push_back("-fvisibility=hidden");
     CC1Args.push_back("-fapply-global-visibility-to-externs");
   }
 }
 
 StringRef
 AMDGPUToolChain::getGPUArch(const llvm::opt::ArgList &DriverArgs) const {
+  if (DriverArgs.hasArg(options::OPT_mcpu_EQ))
     return getProcessorFromTargetID(
         getTriple(), DriverArgs.getLastArgValue(options::OPT_mcpu_EQ));
+  if (DriverArgs.hasArg(options::OPT_march_EQ))
+    return getProcessorFromTargetID(
+        getTriple(), DriverArgs.getLastArgValue(options::OPT_march_EQ));
+  return "";
 }
 
 AMDGPUToolChain::ParsedTargetIDType
 AMDGPUToolChain::getParsedTargetID(const llvm::opt::ArgList &DriverArgs) const {
   StringRef TargetID = DriverArgs.getLastArgValue(options::OPT_mcpu_EQ);
   if (TargetID.empty())
     return {None, None, None};
 

clang/lib/Driver/ToolChains/AMDGPUOpenMP.h

void AddIAMCUIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                            llvm::opt::ArgStringList &CC1Args) const override;
 
   SanitizerMask getSupportedSanitizers() const override;
 
   VersionTuple
   computeMSVCVersion(const Driver *D,
                      const llvm::opt::ArgList &Args) const override;
 
+  llvm::SmallVector<BitCodeLibraryInfo, 12>
+  getROCmDeviceLibs(const llvm::opt::ArgList &Args) const override;
+
   const ToolChain &HostTC;
 };
 
 } // end namespace toolchains
 } // end namespace driver
 } // end namespace clang
 
 #endif // LLVM_CLANG_LIB_DRIVER_TOOLCHAINS_AMDGPUOPENMP_H

clang/lib/Driver/ToolChains/AMDGPUOpenMP.cpp

void AMDGPUOpenMPToolChain::addClangTargetOptions(
 
   CC1Args.push_back("-target-cpu");
   CC1Args.push_back(DriverArgs.MakeArgStringRef(GPUArch));
   CC1Args.push_back("-fcuda-is-device");
 
   if (DriverArgs.hasArg(options::OPT_nogpulib))
     return;
 
+  for (auto BCFile : getROCmDeviceLibs(DriverArgs)) {
+    CC1Args.push_back(BCFile.ShouldInternalize ? "-mlink-builtin-bitcode"
+                                               : "-mlink-bitcode-file");
+    CC1Args.push_back(DriverArgs.MakeArgString(BCFile.Path));
+  }
+
   // Link the bitcode library late if we're using device LTO.
   if (getDriver().isUsingLTO(/* IsOffload */ true))
     return;
 
   addOpenMPDeviceRTL(getDriver(), DriverArgs, CC1Args, GPUArch, getTriple());
 }
 
 llvm::opt::DerivedArgList *AMDGPUOpenMPToolChain::TranslateArgs(
SanitizerMask AMDGPUOpenMPToolChain::getSupportedSanitizers() const {
   return HostTC.getSupportedSanitizers();
 }
 
 VersionTuple
 AMDGPUOpenMPToolChain::computeMSVCVersion(const Driver *D,
                                           const ArgList &Args) const {
   return HostTC.computeMSVCVersion(D, Args);
 }
+
+llvm::SmallVector<ToolChain::BitCodeLibraryInfo, 12>
+AMDGPUOpenMPToolChain::getROCmDeviceLibs(const llvm::opt::ArgList &Args) const {
+  if (Args.hasArg(options::OPT_nogpulib))
+    return {};
+
+  if (!RocmInstallation.hasDeviceLibrary()) {
+    getDriver().Diag(diag::err_drv_no_rocm_device_lib) << 0;
+    return {};
+  }
+
+  StringRef GpuArch = getProcessorFromTargetID(
+      getTriple(), Args.getLastArgValue(options::OPT_march_EQ));
+
+  SmallVector<BitCodeLibraryInfo, 12> BCLibs;
+  for (auto BCLib : getCommonDeviceLibNames(Args, GpuArch.str(),
+                                            /*IsOpenMP=*/true))
+    BCLibs.emplace_back(BCLib);
+
+  return BCLibs;
+}

clang/lib/Driver/ToolChains/Clang.cpp

 
 void LinkerWrapper::ConstructJob(Compilation &C, const JobAction &JA,
                                  const InputInfo &Output,
                                  const InputInfoList &Inputs,
                                  const ArgList &Args,
                                  const char *LinkingOutput) const {
   const Driver &D = getToolChain().getDriver();
   const llvm::Triple TheTriple = getToolChain().getTriple();
-  auto OpenMPTCRange = C.getOffloadToolChains<Action::OFK_OpenMP>();
   ArgStringList CmdArgs;
 
   // Pass the CUDA path to the linker wrapper tool.
   for (Action::OffloadKind Kind : {Action::OFK_Cuda, Action::OFK_OpenMP}) {
     auto TCRange = C.getOffloadToolChains(Kind);
     for (auto &I : llvm::make_range(TCRange.first, TCRange.second)) {
       const ToolChain *TC = I.second;
       if (TC->getTriple().isNVPTX()) {
         CudaInstallationDetector CudaInstallation(D, TheTriple, Args);
         if (CudaInstallation.isValid())
           CmdArgs.push_back(Args.MakeArgString(
               "--cuda-path=" + CudaInstallation.getInstallPath()));
         break;
       }
     }
   }
 
-  // Get the AMDGPU math libraries.
-  // FIXME: This method is bad, remove once AMDGPU has a proper math library
-  // (see AMDGCN::OpenMPLinker::constructLLVMLinkCommand).
-  for (auto &I : llvm::make_range(OpenMPTCRange.first, OpenMPTCRange.second)) {
-    const ToolChain *TC = I.second;
-
-    if (!TC->getTriple().isAMDGPU() || Args.hasArg(options::OPT_nogpulib))
-      continue;
-
-    const ArgList &TCArgs = C.getArgsForToolChain(TC, "", Action::OFK_OpenMP);
-    StringRef Arch = TCArgs.getLastArgValue(options::OPT_march_EQ);
-    const toolchains::ROCMToolChain RocmTC(TC->getDriver(), TC->getTriple(),
-                                           TCArgs);
-
-    SmallVector<std::string, 12> BCLibs =
-        RocmTC.getCommonDeviceLibNames(TCArgs, Arch.str());
-
-    for (StringRef LibName : BCLibs)
-      CmdArgs.push_back(Args.MakeArgString(
-          "--bitcode-library=" +
-          Action::GetOffloadKindName(Action::OFK_OpenMP) + "-" +
-          TC->getTripleString() + "-" + Arch + "=" + LibName));
-  }
-
   if (D.isUsingLTO(/* IsOffload */ true)) {
     // Pass in the optimization level to use for LTO.
     if (const Arg *A = Args.getLastArg(options::OPT_O_Group)) {
       StringRef OOpt;
       if (A->getOption().matches(options::OPT_O4) ||
           A->getOption().matches(options::OPT_Ofast))
         OOpt = "3";
       else if (A->getOption().matches(options::OPT_O)) {

clang/lib/Driver/ToolChains/HIPAMD.h

public:
   void AddClangCXXStdlibIncludeArgs(
       const llvm::opt::ArgList &Args,
       llvm::opt::ArgStringList &CC1Args) const override;
   void AddIAMCUIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                            llvm::opt::ArgStringList &CC1Args) const override;
   void AddHIPIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                          llvm::opt::ArgStringList &CC1Args) const override;
   llvm::SmallVector<BitCodeLibraryInfo, 12>
-  getHIPDeviceLibs(const llvm::opt::ArgList &Args) const override;
+  getROCmDeviceLibs(const llvm::opt::ArgList &Args) const override;
 
   SanitizerMask getSupportedSanitizers() const override;
 
   VersionTuple
   computeMSVCVersion(const Driver *D,
                      const llvm::opt::ArgList &Args) const override;
 
   unsigned GetDefaultDwarfVersion() const override { return 5; }

clang/lib/Driver/ToolChains/HIPAMD.cpp

void HIPAMDToolChain::addClangTargetOptions(
   // Default to "hidden" visibility, as object level linking will not be
   // supported for the foreseeable future.
   if (!DriverArgs.hasArg(options::OPT_fvisibility_EQ,
                          options::OPT_fvisibility_ms_compat)) {
     CC1Args.append({"-fvisibility=hidden"});
     CC1Args.push_back("-fapply-global-visibility-to-externs");
   }
 
-  for (auto BCFile : getHIPDeviceLibs(DriverArgs)) {
+  for (auto BCFile : getROCmDeviceLibs(DriverArgs)) {
     CC1Args.push_back(BCFile.ShouldInternalize ? "-mlink-builtin-bitcode"
                                                : "-mlink-bitcode-file");
     CC1Args.push_back(DriverArgs.MakeArgString(BCFile.Path));
   }
 }
 
 llvm::opt::DerivedArgList *
 HIPAMDToolChain::TranslateArgs(const llvm::opt::DerivedArgList &Args,

 }
 
 VersionTuple HIPAMDToolChain::computeMSVCVersion(const Driver *D,
                                                  const ArgList &Args) const {
   return HostTC.computeMSVCVersion(D, Args);
 }
 
 llvm::SmallVector<ToolChain::BitCodeLibraryInfo, 12>
-HIPAMDToolChain::getHIPDeviceLibs(const llvm::opt::ArgList &DriverArgs) const {
+HIPAMDToolChain::getROCmDeviceLibs(const llvm::opt::ArgList &DriverArgs) const {
   llvm::SmallVector<BitCodeLibraryInfo, 12> BCLibs;
   if (DriverArgs.hasArg(options::OPT_nogpulib))
     return {};
   ArgStringList LibraryPaths;
 
   // Find in --hip-device-lib-path and HIP_LIBRARY_PATH.
   for (auto Path : RocmInstallation.getRocmDeviceLibPathArg())
     LibraryPaths.push_back(DriverArgs.MakeArgString(Path));

clang/lib/Driver/ToolChains/HIPSPV.h

public:
   void AddClangCXXStdlibIncludeArgs(
       const llvm::opt::ArgList &Args,
       llvm::opt::ArgStringList &CC1Args) const override;
   void AddIAMCUIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                            llvm::opt::ArgStringList &CC1Args) const override;
   void AddHIPIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                          llvm::opt::ArgStringList &CC1Args) const override;
   llvm::SmallVector<BitCodeLibraryInfo, 12>
-  getHIPDeviceLibs(const llvm::opt::ArgList &Args) const override;
+  getROCmDeviceLibs(const llvm::opt::ArgList &Args) const override;
 
   SanitizerMask getSupportedSanitizers() const override;
 
   VersionTuple
   computeMSVCVersion(const Driver *D,
                      const llvm::opt::ArgList &Args) const override;
 
   void adjustDebugInfoKind(codegenoptions::DebugInfoKind &DebugInfoKind,

clang/lib/Driver/ToolChains/HIPSPV.cpp

void HIPSPVToolChain::addClangTargetOptions(
 
   // Default to "hidden" visibility, as object level linking will not be
   // supported for the foreseeable future.
   if (!DriverArgs.hasArg(options::OPT_fvisibility_EQ,
                          options::OPT_fvisibility_ms_compat))
     CC1Args.append(
         {"-fvisibility=hidden", "-fapply-global-visibility-to-externs"});
 
-  llvm::for_each(getHIPDeviceLibs(DriverArgs),
+  llvm::for_each(getROCmDeviceLibs(DriverArgs),
                  [&](const BitCodeLibraryInfo &BCFile) {
                    CC1Args.append({"-mlink-builtin-bitcode",
                                    DriverArgs.MakeArgString(BCFile.Path)});
                  });
 }
 
 Tool *HIPSPVToolChain::buildLinker() const {
   assert(getTriple().getArch() == llvm::Triple::spirv64);
if (hipPath.empty()) {
     return;
   }
   SmallString<128> P(hipPath);
   llvm::sys::path::append(P, "include");
   CC1Args.append({"-isystem", DriverArgs.MakeArgString(P)});
 }
 
 llvm::SmallVector<ToolChain::BitCodeLibraryInfo, 12>
-HIPSPVToolChain::getHIPDeviceLibs(const llvm::opt::ArgList &DriverArgs) const {
+HIPSPVToolChain::getROCmDeviceLibs(const llvm::opt::ArgList &DriverArgs) const {
   llvm::SmallVector<ToolChain::BitCodeLibraryInfo, 12> BCLibs;
   if (DriverArgs.hasArg(options::OPT_nogpulib))
     return {};
 
   ArgStringList LibraryPaths;
   // Find device libraries in --hip-device-lib-path and HIP_DEVICE_LIB_PATH.
   auto HipDeviceLibPathArgs = DriverArgs.getAllArgValues(
       // --hip-device-lib-path is alias to this option.

clang/test/Driver/amdgpu-openmp-toolchain.c

 // CHECK-BINDINGS: "amdgcn-amd-amdhsa" - "clang", inputs: ["[[INPUT]]", "[[HOST_BC]]"], output: "[[DEVICE_BC:.+]]"
 // CHECK-BINDINGS: "x86_64-unknown-linux-gnu" - "Offload::Packager", inputs: ["[[DEVICE_BC]]"], output: "[[BINARY:.+]]"
 // CHECK-BINDINGS: "x86_64-unknown-linux-gnu" - "clang", inputs: ["[[HOST_BC]]", "[[BINARY]]"], output: "[[HOST_OBJ:.+]]"
 // CHECK-BINDINGS: "x86_64-unknown-linux-gnu" - "Offload::Linker", inputs: ["[[HOST_OBJ]]"], output: "a.out"
 
 // RUN:   %clang -### --target=x86_64-unknown-linux-gnu -emit-llvm -S -fopenmp -fopenmp-targets=amdgcn-amd-amdhsa -Xopenmp-target=amdgcn-amd-amdhsa -march=gfx803 -nogpulib %s 2>&1 | FileCheck %s --check-prefix=CHECK-EMIT-LLVM-IR
 // CHECK-EMIT-LLVM-IR: "-cc1" "-triple" "amdgcn-amd-amdhsa"{{.*}}"-emit-llvm"
 
-// RUN: %clang -### -target x86_64-pc-linux-gnu -fopenmp -fopenmp-targets=amdgcn-amd-amdhsa -Xopenmp-target=amdgcn-amd-amdhsa -march=gfx803 -lm --rocm-device-lib-path=%S/Inputs/rocm/amdgcn/bitcode %s 2>&1 | FileCheck %s --check-prefix=CHECK-LIB-DEVICE-NEW
-// CHECK-LIB-DEVICE-NEW: {{.*}}clang-linker-wrapper{{.*}}--bitcode-library=openmp-amdgcn-amd-amdhsa-gfx803={{.*}}ocml.bc"{{.*}}ockl.bc"{{.*}}oclc_daz_opt_on.bc"{{.*}}oclc_unsafe_math_off.bc"{{.*}}oclc_finite_only_off.bc"{{.*}}oclc_correctly_rounded_sqrt_on.bc"{{.*}}oclc_wavefrontsize64_on.bc"{{.*}}oclc_isa_version_803.bc"
+// RUN: %clang -### -target x86_64-pc-linux-gnu -fopenmp --offload-arch=gfx803 \
+// RUN:   --rocm-device-lib-path=%S/Inputs/rocm/amdgcn/bitcode -fopenmp-new-driver %s  2>&1 | \
+// RUN: FileCheck %s --check-prefix=CHECK-LIB-DEVICE
+// CHECK-LIB-DEVICE: "-cc1" {{.*}}ocml.bc"{{.*}}ockl.bc"{{.*}}oclc_daz_opt_on.bc"{{.*}}oclc_unsafe_math_off.bc"{{.*}}oclc_finite_only_off.bc"{{.*}}oclc_correctly_rounded_sqrt_on.bc"{{.*}}oclc_wavefrontsize64_on.bc"{{.*}}oclc_isa_version_803.bc"
+
+// RUN: %clang -### -target x86_64-pc-linux-gnu -fopenmp --offload-arch=gfx803 -nogpulib \
+// RUN:   --rocm-device-lib-path=%S/Inputs/rocm/amdgcn/bitcode -fopenmp-new-driver %s  2>&1 | \
+// RUN: FileCheck %s --check-prefix=CHECK-LIB-DEVICE-NOGPULIB
+// CHECK-LIB-DEVICE-NOGPULIB-NOT: "-cc1" {{.*}}ocml.bc"{{.*}}ockl.bc"{{.*}}oclc_daz_opt_on.bc"{{.*}}oclc_unsafe_math_off.bc"{{.*}}oclc_finite_only_off.bc"{{.*}}oclc_correctly_rounded_sqrt_on.bc"{{.*}}oclc_wavefrontsize64_on.bc"{{.*}}oclc_isa_version_803.bc"