[HIP] Use option -nogpulib to disable linking device lib

[tra]

The functionality looks generic enough. Should it be just flink_builtin_bitcode?

[yaxunl]

change the option name

[ashi1]

LGTM thank you

include/clang/Driver/Options.td

606 ↗

(On Diff #222746)

Since this is a more generic approach, we won't need to specify HIP ?

[yaxunl]

include/clang/Driver/Options.td

606 ↗

(On Diff #222746)

this patch only implemented this option for HIP. If it is used for other languages, this help text should be updated.

[tra]

include/clang/Driver/Options.td

606 ↗

(On Diff #222746)

Hmm. Cuda currently uses -nocudalib for essentially the same purpose (Sort of like -nostdlib, but for CUDA). Perhaps we should consolidate all these into -nogpulib and alias -nocudalib to it.

[hliao]

include/clang/Driver/Options.td

606 ↗

(On Diff #222746)

how about other relevant options, such as replacing cuda-device-only with gpu-device-only or hip-device-only to avoid confusing with CUDA.

[yaxunl]

use -nogpuarch

[tra]

include/clang/Driver/Options.td

606 ↗

(On Diff #222746)

We seem to be doing exactly that, only incrementally. In general it does make sense to consolidate the flags where the functionality is common.
cuda-device-only, cuda_host_only, cuda_compile_host_device and cuda-gpu-arch could be generalized to gpu-<something>.

This should be a separate patch, though.

[yaxunl]

include/clang/Driver/Options.td

606 ↗

(On Diff #222746)

agree

[ashi1]

LGTM - Thanks for generalizing

cfe/trunk/include/clang/Driver/Options.td

def no_canonical_prefixes : Flag<["-"], "no-canonical-prefixes">, Flags<[HelpHidden, CoreOption]>,
   HelpText<"Use relative instead of canonical paths">;
 def no_cpp_precomp : Flag<["-"], "no-cpp-precomp">, Group<clang_ignored_f_Group>;
 def no_integrated_cpp : Flag<["-", "--"], "no-integrated-cpp">, Flags<[DriverOption]>;
 def no_pedantic : Flag<["-", "--"], "no-pedantic">, Group<pedantic_Group>;
 def no__dead__strip__inits__and__terms : Flag<["-"], "no_dead_strip_inits_and_terms">;
 def nobuiltininc : Flag<["-"], "nobuiltininc">, Flags<[CC1Option, CoreOption]>,
   HelpText<"Disable builtin #include directories">;
 def nocudainc : Flag<["-"], "nocudainc">;
-def nocudalib : Flag<["-"], "nocudalib">;
+def nogpulib : Flag<["-"], "nogpulib">,
+  HelpText<"Do not link device library for CUDA/HIP device compilation">;
+def : Flag<["-"], "nocudalib">, Alias<nogpulib>;
 def nodefaultlibs : Flag<["-"], "nodefaultlibs">;
 def nofixprebinding : Flag<["-"], "nofixprebinding">;
 def nolibc : Flag<["-"], "nolibc">;
 def nomultidefs : Flag<["-"], "nomultidefs">;
 def nopie : Flag<["-"], "nopie">;
 def no_pie : Flag<["-"], "no-pie">, Alias<nopie>;
 def noprebind : Flag<["-"], "noprebind">;
 def noprofilelib : Flag<["-"], "noprofilelib">;

cfe/trunk/lib/Driver/ToolChains/Cuda.cpp

for (const char *Ver : Versions)
       Candidates.emplace_back(D.SysRoot + "/usr/local/cuda-" + Ver);
 
     if (Distro(D.getVFS()).IsDebian() || Distro(D.getVFS()).IsUbuntu())
       // Special case for Debian to have nvidia-cuda-toolkit work
       // out of the box. More info on http://bugs.debian.org/882505
       Candidates.emplace_back(D.SysRoot + "/usr/lib/cuda");
   }
 
-  bool NoCudaLib = Args.hasArg(options::OPT_nocudalib);
+  bool NoCudaLib = Args.hasArg(options::OPT_nogpulib);
 
   for (const auto &Candidate : Candidates) {
     InstallPath = Candidate.Path;
     if (InstallPath.empty() || !D.getVFS().exists(InstallPath))
       continue;
 
     BinPath = InstallPath + "/bin";
     IncludePath = InstallPath + "/include";
if (DriverArgs.hasFlag(options::OPT_fcuda_approx_transcendentals,
                            options::OPT_fno_cuda_approx_transcendentals, false))
       CC1Args.push_back("-fcuda-approx-transcendentals");
 
     if (DriverArgs.hasFlag(options::OPT_fgpu_rdc, options::OPT_fno_gpu_rdc,
                            false))
       CC1Args.push_back("-fgpu-rdc");
   }
 
-  if (DriverArgs.hasArg(options::OPT_nocudalib))
+  if (DriverArgs.hasArg(options::OPT_nogpulib))
     return;
 
   std::string LibDeviceFile = CudaInstallation.getLibDeviceFile(GpuArch);
 
   if (LibDeviceFile.empty()) {
     if (DeviceOffloadingKind == Action::OFK_OpenMP &&
         DriverArgs.hasArg(options::OPT_S))
       return;

cfe/trunk/lib/Driver/ToolChains/HIP.cpp

void HIPToolChain::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");
   }
+
+  if (DriverArgs.hasArg(options::OPT_nogpulib))
+    return;
   ArgStringList LibraryPaths;
 
   // Find in --hip-device-lib-path and HIP_LIBRARY_PATH.
   for (auto Path :
        DriverArgs.getAllArgValues(options::OPT_hip_device_lib_path_EQ))
     LibraryPaths.push_back(DriverArgs.MakeArgString(Path));
 
   addDirectoryList(DriverArgs, LibraryPaths, "-L", "HIP_DEVICE_LIB_PATH");

cfe/trunk/test/Driver/hip-no-device-libs.hip

+// REQUIRES: clang-driver
+// REQUIRES: x86-registered-target
+// REQUIRES: amdgpu-registered-target
+
+// RUN: %clang -### -nogpulib -target x86_64-linux-gnu \
+// RUN:   -x hip --cuda-gpu-arch=gfx900 \
+// RUN:   %S/Inputs/hip_multiple_inputs/b.hip \
+// RUN: 2>&1 | FileCheck %s
+
+// CHECK-NOT: "-mlink-builtin-bitcode"
+