[SYCL][NVPTX] Obey -nocudalib when linking against libdevice (#15378)

This commit also makes a clearer distinction in code between what NVPTX
and NativeCPU are required to link against.

It refactors some nearby code which was more complicated than it needed
to be.

clang/lib/Driver/Driver.cpp

@@ -5564,7 +5564,7 @@ class OffloadingActionBuilder final {
  SYCLDeviceLibLinked = addSYCLDeviceLibs(
  TC, SYCLDeviceLibs, UseAOTLink,
  C.getDefaultToolChain().getTriple().isWindowsMSVCEnvironment(),
- IsSYCLNativeCPU, NativeCPULib);
+ IsSYCLNativeCPU, NativeCPULib, BoundArch);
  }
  JobAction *LinkSYCLLibs =
  C.MakeAction<LinkJobAction>(SYCLDeviceLibs, types::TY_LLVM_BC);
@@ -5842,7 +5842,7 @@ class OffloadingActionBuilder final {
 
  bool addSYCLDeviceLibs(const ToolChain *TC, ActionList &DeviceLinkObjects,
  bool isSpirvAOT, bool isMSVCEnv, bool isNativeCPU,
- Action *&NativeCPULib) {
+ Action *&NativeCPULib, const char *BoundArch) {
  int NumOfDeviceLibLinked = 0;
  SmallVector<SmallString<128>, 4> LibLocCandidates;
  SYCLInstallation.getSYCLDeviceLibPath(LibLocCandidates);
@@ -5852,14 +5852,10 @@ class OffloadingActionBuilder final {
  tools::SYCL::getDeviceLibraries(C, TC->getTriple(), isSpirvAOT);
 
  for (const auto &DeviceLib : DeviceLibraries) {
- bool LibLocSelected = false;
  for (const auto &LLCandidate : LibLocCandidates) {
- if (LibLocSelected)
- break;
  SmallString<128> LibName(LLCandidate);
  llvm::sys::path::append(LibName, DeviceLib);
  if (llvm::sys::fs::exists(LibName)) {
-
  // NativeCPU currently only needs libsycl-nativecpu_utils and
  // libclc, so temporarily skip other device libs in invocation.
  // Todo: remove once NativeCPU tests the other libraries.
@@ -5897,8 +5893,6 @@ class OffloadingActionBuilder final {
  C.MakeAction<InputAction>(*InputArg, types::TY_LLVM_BC);
  DeviceLinkObjects.push_back(SYCLDeviceLibsInputAction);
  }
- if (!LibLocSelected)
- LibLocSelected = !LibLocSelected;
 
  // The device link stage may remove symbols not referenced in the
  // source code. Since libsycl-nativecpu_utils contains such symbols
@@ -5909,22 +5903,26 @@ class OffloadingActionBuilder final {
  NativeCPULib = DeviceLinkObjects.back();
  DeviceLinkObjects.pop_back();
  }
+
+ break;
  }
  }
  }
 
- // For NVPTX backend we need to also link libclc and CUDA libdevice
- // at the same stage that we link all of the unbundled SYCL libdevice
- // objects together.
- if ((TC->getTriple().isNVPTX() || isNativeCPU) && NumOfDeviceLibLinked) {
+ if (!NumOfDeviceLibLinked)
+ return false;
+
+ // For NVPTX and NativeCPU we need to also link libclc at the same stage
+ // that we link all of the unbundled SYCL libdevice objects together.
+ if (TC->getTriple().isNVPTX() || isNativeCPU) {
  std::string LibSpirvFile;
  if (Args.hasArg(options::OPT_fsycl_libspirv_path_EQ)) {
  auto ProvidedPath =
  Args.getLastArgValue(options::OPT_fsycl_libspirv_path_EQ).str();
  if (llvm::sys::fs::exists(ProvidedPath))
  LibSpirvFile = ProvidedPath;
  } else {
- SmallVector<StringRef, 8> LibraryPaths;
+ SmallVector<StringRef, 2> LibraryPaths;
 
  // Expected path w/out install.
  SmallString<256> WithoutInstallPath(C.getDriver().ResourceDir);
@@ -5936,22 +5934,17 @@ class OffloadingActionBuilder final {
  llvm::sys::path::append(WithInstallPath, Twine("../../../share/clc"));
  LibraryPaths.emplace_back(WithInstallPath.c_str());
 
- // TODO: check if the isNVPTX() path can also use
- // TC->getTripleString() so that the conditional could be removed
- const std::string TrStr =
- isNativeCPU ? TC->getTripleString() : "nvptx64-nvidia-cuda";
-
  // Select remangled libclc variant
  StringRef LibSpirvTargetNamePref =
  TC->getAuxTriple()->isOSWindows()
  ? "remangled-l32-signed_char.libspirv-"
  : "remangled-l64-signed_char.libspirv-";
- llvm::Twine LibSpirvTargetNameTemp = LibSpirvTargetNamePref + TrStr;
- llvm::Twine LibSpirvTargetName = LibSpirvTargetNameTemp + ".bc";
 
  for (StringRef LibraryPath : LibraryPaths) {
  SmallString<128> LibSpirvTargetFile(LibraryPath);
- llvm::sys::path::append(LibSpirvTargetFile, LibSpirvTargetName);
+ llvm::sys::path::append(LibSpirvTargetFile,
+ LibSpirvTargetNamePref +
+ TC->getTripleString() + ".bc");
  if (llvm::sys::fs::exists(LibSpirvTargetFile) ||
  Args.hasArg(options::OPT__HASH_HASH_HASH)) {
  LibSpirvFile = std::string(LibSpirvTargetFile.str());
@@ -5966,30 +5959,24 @@ class OffloadingActionBuilder final {
  C.MakeAction<InputAction>(*LibClcInputArg, types::TY_LLVM_BC);
  DeviceLinkObjects.push_back(SYCLLibClcInputAction);
  }
+ }
 
- if (isNativeCPU) {
- // return here to not generate cuda actions
- return NumOfDeviceLibLinked != 0;
- }
-
+ // For NVPTX we also need to link with the CUDA libdevice
+ if (TC->getTriple().isNVPTX() && !Args.hasArg(options::OPT_nogpulib)) {
  const toolchains::CudaToolChain *CudaTC =
  static_cast<const toolchains::CudaToolChain *>(TC);
- for (const auto &LinkInputEnum : enumerate(DeviceLinkerInputs)) {
- const char *BoundArch =
- SYCLTargetInfoList[LinkInputEnum.index()].BoundArch;
- std::string LibDeviceFile =
- CudaTC->CudaInstallation.getLibDeviceFile(BoundArch);
- if (!LibDeviceFile.empty()) {
- Arg *CudaDeviceLibInputArg =
- MakeInputArg(Args, C.getDriver().getOpts(),
- Args.MakeArgString(LibDeviceFile));
- auto *SYCLDeviceLibInputAction = C.MakeAction<InputAction>(
- *CudaDeviceLibInputArg, types::TY_LLVM_BC);
- DeviceLinkObjects.push_back(SYCLDeviceLibInputAction);
- }
+ std::string LibDeviceFile =
+ CudaTC->CudaInstallation.getLibDeviceFile(BoundArch);
+ if (!LibDeviceFile.empty()) {
+ Arg *CudaDeviceLibInputArg = MakeInputArg(
+ Args, C.getDriver().getOpts(), Args.MakeArgString(LibDeviceFile));
+ auto *SYCLDeviceLibInputAction = C.MakeAction<InputAction>(
+ *CudaDeviceLibInputArg, types::TY_LLVM_BC);
+ DeviceLinkObjects.push_back(SYCLDeviceLibInputAction);
  }
  }
- return NumOfDeviceLibLinked != 0;
+
+ return true;
  }
 
  void appendLinkDependences(OffloadAction::DeviceDependences &DA) override {

clang/test/Driver/sycl-nvptx-link.cpp

@@ -28,6 +28,12 @@
 // RUN: --sysroot=%S/Inputs/SYCL --cuda-path=%S/Inputs/CUDA_90/usr/local/cuda %s 2>&1 \
 // RUN: | FileCheck %s --check-prefixes=CHECK,LIBDEVICE10
 
+// Check also that -nocudalib is obeyed
+// RUN: %clang -### -fsycl -fsycl-targets=nvptx64-nvidia-cuda -nocudalib \
+// RUN: -Xsycl-target-backend --cuda-gpu-arch=sm_35 \
+// RUN: --sysroot=%S/Inputs/SYCL --cuda-path=%S/Inputs/CUDA_90/usr/local/cuda %s 2>&1 \
+// RUN: | FileCheck %s --check-prefixes=CHECK,NOLIBDEVICE
+
 // First link command: ignored
 // CHECK: llvm-link
 
@@ -39,3 +45,4 @@
 // LIBDEVICE30-SAME: libdevice.compute_30.10.bc
 // LIBDEVICE35-SAME: libdevice.compute_35.10.bc
 // LIBDEVICE50-SAME: libdevice.compute_50.10.bc
+// NOLIBDEVICE-NOT: libdevice.{{.*}}.10.bc