[NVPTX] Fix ctor / dtor lowering when NVPTX target is not enabled (#124116)

Summary:
We pass the `-nvptx-lower-global-ctor-dtor` option to support the `libc`
like use-case which needs global constructors sometimes. This only
affects the backend. If the NVPTX target is not enabled this option will
be unknown which prevents you from compiling generic IR for this.

Author: jhuber6

clang/lib/Driver/ToolChains/Cuda.cpp

@@ -639,6 +639,9 @@ void NVPTX::Linker::ConstructJob(Compilation &C, const JobAction &JA,
   CmdArgs.push_back(
       Args.MakeArgString("--plugin-opt=-mattr=" + llvm::join(Features, ",")));
 
+  // Enable ctor / dtor lowering for the direct / freestanding NVPTX target.
+  CmdArgs.append({"-mllvm", "--nvptx-lower-global-ctor-dtor"});
+
   // Add paths for the default clang library path.
   SmallString<256> DefaultLibPath =
       llvm::sys::path::parent_path(TC.getDriver().Dir);
@@ -783,7 +786,7 @@ void NVPTXToolChain::addClangTargetOptions(
   // If we are compiling with a standalone NVPTX toolchain we want to try to
   // mimic a standard environment as much as possible. So we enable lowering
   // ctor / dtor functions to global symbols that can be registered.
-  if (Freestanding)
+  if (Freestanding && !getDriver().isUsingLTO())
     CC1Args.append({"-mllvm", "--nvptx-lower-global-ctor-dtor"});
 }
 

clang/test/Driver/cuda-cross-compiling.c

@@ -63,8 +63,12 @@
 //
 // RUN: %clang -target nvptx64-nvidia-cuda -march=sm_70 %s -### 2>&1 \
 // RUN:   | FileCheck -check-prefix=LOWERING %s
+// RUN: %clang -target nvptx64-nvidia-cuda -march=sm_70 -flto -c %s -### 2>&1 \
+// RUN:   | FileCheck -check-prefix=LOWERING-LTO %s
 
 // LOWERING: -cc1" "-triple" "nvptx64-nvidia-cuda" {{.*}} "-mllvm" "--nvptx-lower-global-ctor-dtor"
+// LOWERING: clang-nvlink-wrapper{{.*}} "-mllvm" "--nvptx-lower-global-ctor-dtor"
+// LOWERING-LTO-NOT: "--nvptx-lower-global-ctor-dtor"
 
 //
 // Test passing arguments directly to nvlink.