[AMDGPU] Correctly determine the toolchain linker

[jhuber6]

Summary:
The AMDGPU toolchain simply took the short name to get the link job
instead of using the common utilities that respect options like
-fuse-ld. Any linker that isn't ld.lld will fail, however we should
be able to override it.

[llvmbot]

@llvm/pr-subscribers-clang-driver
@llvm/pr-subscribers-clang

@llvm/pr-subscribers-backend-amdgpu

Author: Joseph Huber (jhuber6)

Changes

Summary:
The AMDGPU toolchain simply took the short name to get the link job
instead of using the common utilities that respect options like
-fuse-ld. Any linker that isn't ld.lld will fail, however we should
be able to override it.

---

Full diff: https://github.com/llvm/llvm-project/pull/89803.diff

2 Files Affected:

• (modified) clang/lib/Driver/ToolChains/AMDGPU.cpp (+1-2)
• (modified) clang/test/Driver/amdgpu-toolchain.c (+6)

diff --git a/clang/lib/Driver/ToolChains/AMDGPU.cpp b/clang/lib/Driver/ToolChains/AMDGPU.cpp
index 4e6362a0f40632..07965b487ea79b 100644
--- a/clang/lib/Driver/ToolChains/AMDGPU.cpp
+++ b/clang/lib/Driver/ToolChains/AMDGPU.cpp
@@ -617,8 +617,7 @@ void amdgpu::Linker::ConstructJob(Compilation &C, const JobAction &JA,
                                   const InputInfoList &Inputs,
                                   const ArgList &Args,
                                   const char *LinkingOutput) const {
-
-  std::string Linker = getToolChain().GetProgramPath(getShortName());
+  std::string Linker = getToolChain().GetLinkerPath();
   ArgStringList CmdArgs;
   CmdArgs.push_back("--no-undefined");
   CmdArgs.push_back("-shared");
diff --git a/clang/test/Driver/amdgpu-toolchain.c b/clang/test/Driver/amdgpu-toolchain.c
index 4300e7e9f66705..d21ce857f3c57a 100644
--- a/clang/test/Driver/amdgpu-toolchain.c
+++ b/clang/test/Driver/amdgpu-toolchain.c
@@ -24,3 +24,9 @@
 // RUN:   -L. -fconvergent-functions %s 2>&1 | FileCheck -check-prefix=MCPU %s
 // LTO: clang{{.*}} "-flto=full"{{.*}}"-fconvergent-functions"
 // MCPU: ld.lld{{.*}}"-L."{{.*}}"-plugin-opt=mcpu=gfx906"
+
+// We do not suppor the BFD linker, but we should be able to override the
+// default even if it will error during linking.
+// RUN: %clang -### --target=amdgcn-amd-amdhsa -mcpu=gfx906 -nogpulib \
+// RUN:   -fuse-ld=bfd %s 2>&1 | FileCheck -check-prefixes=LD %s
+// LD: ld.bfd"

[jhuber6]

Note that this doesn't affect OpenMP or HIP. The former uses the clang-linker-wrapper while the latter calls lld directly.

[madhur13490]

s/suppor/support

[madhur13490]

Thanks for fixing this after our discord chats!