[Clang] Forward `-Xarch_<arch> -Wl,foo` for GPU toolchains #126248

jhuber6 · 2025-02-07T14:34:55Z

Summary:
This patch handles the use of -Xarch_<arch> -Wl,foo to send an
argument to the linker for the embedded offloading jobs in the linker
wrapper. This makes it equivalent to -Xoffload-linker foo.

Summary: This patch handles the use of `-Xarch_<arch> -Wl,foo` to send an argument to the linker for the embedded offloading jobs in the linker wrapper. This makes it equivalent to `-Xoffload-linker foo`.

llvmbot · 2025-02-07T14:35:32Z

@llvm/pr-subscribers-clang-driver

Author: Joseph Huber (jhuber6)

Changes

Summary:
This patch handles the use of -Xarch_<arch> -Wl,foo to send an
argument to the linker for the embedded offloading jobs in the linker
wrapper. This makes it equivalent to -Xoffload-linker foo.

Full diff: https://github.com/llvm/llvm-project/pull/126248.diff

2 Files Affected:

(modified) clang/lib/Driver/ToolChains/Clang.cpp (+5-2)
(modified) clang/test/Driver/offload-Xarch.c (+10-1)

diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
index c0891d46b0a62cd..55d2936fc5bbb53 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -9184,8 +9184,9 @@ void LinkerWrapper::ConstructJob(Compilation &C, const JobAction &JA,
       OPT_load,
       OPT_fno_lto,
       OPT_flto,
+      OPT_Zlinker_input,
       OPT_flto_EQ};
-  const llvm::DenseSet<unsigned> LinkerOptions{OPT_mllvm};
+  const llvm::DenseSet<unsigned> LinkerOptions{OPT_mllvm, OPT_Zlinker_input};
   auto ShouldForward = [&](const llvm::DenseSet<unsigned> &Set, Arg *A) {
     return Set.contains(A->getOption().getID()) ||
            (A->getOption().getGroup().isValid() &&
@@ -9203,7 +9204,9 @@ void LinkerWrapper::ConstructJob(Compilation &C, const JobAction &JA,
       ArgStringList CompilerArgs;
       ArgStringList LinkerArgs;
       for (Arg *A : C.getArgsForToolChain(TC, /*BoundArch=*/"", Kind)) {
-        if (ShouldForward(CompilerOptions, A))
+        if (A->getOption().matches(OPT_Zlinker_input))
+          LinkerArgs.emplace_back(A->getValue());
+        else if (ShouldForward(CompilerOptions, A))
           A->render(Args, CompilerArgs);
         else if (ShouldForward(LinkerOptions, A))
           A->render(Args, LinkerArgs);
diff --git a/clang/test/Driver/offload-Xarch.c b/clang/test/Driver/offload-Xarch.c
index 17db891b068340b..18c68f2acc884bc 100644
--- a/clang/test/Driver/offload-Xarch.c
+++ b/clang/test/Driver/offload-Xarch.c
@@ -14,7 +14,7 @@
 // RUN:   --target=x86_64-unknown-linux-gnu -Xopenmp-target=nvptx64-nvidia-cuda --offload-arch=sm_52,sm_60 -nogpuinc \
 // RUN:   -Xopenmp-target=amdgcn-amd-amdhsa --offload-arch=gfx90a,gfx1030 -ccc-print-bindings -### %s 2>&1 \
 // RUN: | FileCheck -check-prefix=OPENMP %s
-//
+
 // OPENMP: # "x86_64-unknown-linux-gnu" - "clang", inputs: ["[[INPUT:.+]]"], output: "[[HOST_BC:.+]]"
 // OPENMP: # "amdgcn-amd-amdhsa" - "clang", inputs: ["[[INPUT]]", "[[HOST_BC]]"], output: "[[GFX1030_BC:.+]]"
 // OPENMP: # "amdgcn-amd-amdhsa" - "clang", inputs: ["[[INPUT]]", "[[HOST_BC]]"], output: "[[GFX90A_BC:.+]]"
@@ -32,3 +32,12 @@
 // CUDA: "-cc1" "-triple" "nvptx64-nvidia-cuda" {{.*}}"-target-cpu" "sm_52" {{.*}}"-O3"
 // CUDA: "-cc1" "-triple" "nvptx64-nvidia-cuda" {{.*}}"-target-cpu" "sm_60" {{.*}}"-O0"
 // CUDA: "-cc1" "-triple" "x86_64-unknown-linux-gnu" {{.*}}"-O3"
+
+// Make sure that `-Xarch_amdgcn` forwards libraries to the device linker.
+// RUN: %clang -fopenmp=libomp --offload-arch=gfx90a -nogpulib -nogpuinc \
+// RUN:   -Xarch_amdgcn -Wl,-lfoo -### %s 2>&1 \
+// RUN: | FileCheck -check-prefix=LIBS %s
+// RUN: %clang -fopenmp=libomp --offload-arch=gfx90a -nogpulib -nogpuinc \
+// RUN:   -Xoffload-linker-amdgcn-amd-amdhsa -lfoo -### %s 2>&1 \
+// RUN: | FileCheck -check-prefix=LIBS %s
+// LIBS: "--device-linker=amdgcn-amd-amdhsa=-lfoo"

llvmbot · 2025-02-07T14:35:32Z

@llvm/pr-subscribers-clang

Author: Joseph Huber (jhuber6)

Changes

Summary:
This patch handles the use of -Xarch_<arch> -Wl,foo to send an
argument to the linker for the embedded offloading jobs in the linker
wrapper. This makes it equivalent to -Xoffload-linker foo.

Full diff: https://github.com/llvm/llvm-project/pull/126248.diff

2 Files Affected:

(modified) clang/lib/Driver/ToolChains/Clang.cpp (+5-2)
(modified) clang/test/Driver/offload-Xarch.c (+10-1)

diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
index c0891d46b0a62cd..55d2936fc5bbb53 100644
--- a/clang/lib/Driver/ToolChains/Clang.cpp
+++ b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -9184,8 +9184,9 @@ void LinkerWrapper::ConstructJob(Compilation &C, const JobAction &JA,
       OPT_load,
       OPT_fno_lto,
       OPT_flto,
+      OPT_Zlinker_input,
       OPT_flto_EQ};
-  const llvm::DenseSet<unsigned> LinkerOptions{OPT_mllvm};
+  const llvm::DenseSet<unsigned> LinkerOptions{OPT_mllvm, OPT_Zlinker_input};
   auto ShouldForward = [&](const llvm::DenseSet<unsigned> &Set, Arg *A) {
     return Set.contains(A->getOption().getID()) ||
            (A->getOption().getGroup().isValid() &&
@@ -9203,7 +9204,9 @@ void LinkerWrapper::ConstructJob(Compilation &C, const JobAction &JA,
       ArgStringList CompilerArgs;
       ArgStringList LinkerArgs;
       for (Arg *A : C.getArgsForToolChain(TC, /*BoundArch=*/"", Kind)) {
-        if (ShouldForward(CompilerOptions, A))
+        if (A->getOption().matches(OPT_Zlinker_input))
+          LinkerArgs.emplace_back(A->getValue());
+        else if (ShouldForward(CompilerOptions, A))
           A->render(Args, CompilerArgs);
         else if (ShouldForward(LinkerOptions, A))
           A->render(Args, LinkerArgs);
diff --git a/clang/test/Driver/offload-Xarch.c b/clang/test/Driver/offload-Xarch.c
index 17db891b068340b..18c68f2acc884bc 100644
--- a/clang/test/Driver/offload-Xarch.c
+++ b/clang/test/Driver/offload-Xarch.c
@@ -14,7 +14,7 @@
 // RUN:   --target=x86_64-unknown-linux-gnu -Xopenmp-target=nvptx64-nvidia-cuda --offload-arch=sm_52,sm_60 -nogpuinc \
 // RUN:   -Xopenmp-target=amdgcn-amd-amdhsa --offload-arch=gfx90a,gfx1030 -ccc-print-bindings -### %s 2>&1 \
 // RUN: | FileCheck -check-prefix=OPENMP %s
-//
+
 // OPENMP: # "x86_64-unknown-linux-gnu" - "clang", inputs: ["[[INPUT:.+]]"], output: "[[HOST_BC:.+]]"
 // OPENMP: # "amdgcn-amd-amdhsa" - "clang", inputs: ["[[INPUT]]", "[[HOST_BC]]"], output: "[[GFX1030_BC:.+]]"
 // OPENMP: # "amdgcn-amd-amdhsa" - "clang", inputs: ["[[INPUT]]", "[[HOST_BC]]"], output: "[[GFX90A_BC:.+]]"
@@ -32,3 +32,12 @@
 // CUDA: "-cc1" "-triple" "nvptx64-nvidia-cuda" {{.*}}"-target-cpu" "sm_52" {{.*}}"-O3"
 // CUDA: "-cc1" "-triple" "nvptx64-nvidia-cuda" {{.*}}"-target-cpu" "sm_60" {{.*}}"-O0"
 // CUDA: "-cc1" "-triple" "x86_64-unknown-linux-gnu" {{.*}}"-O3"
+
+// Make sure that `-Xarch_amdgcn` forwards libraries to the device linker.
+// RUN: %clang -fopenmp=libomp --offload-arch=gfx90a -nogpulib -nogpuinc \
+// RUN:   -Xarch_amdgcn -Wl,-lfoo -### %s 2>&1 \
+// RUN: | FileCheck -check-prefix=LIBS %s
+// RUN: %clang -fopenmp=libomp --offload-arch=gfx90a -nogpulib -nogpuinc \
+// RUN:   -Xoffload-linker-amdgcn-amd-amdhsa -lfoo -### %s 2>&1 \
+// RUN: | FileCheck -check-prefix=LIBS %s
+// LIBS: "--device-linker=amdgcn-amd-amdhsa=-lfoo"

clang/lib/Driver/ToolChains/Clang.cpp

dyung · 2025-02-09T06:35:11Z

@jhuber6 I've seen 2 seemingly random failures of the offload-Xarch.c test on my mac buildbots lately. Any idea why this might be happening?

https://lab.llvm.org/buildbot/#/builders/190/builds/14302
https://lab.llvm.org/buildbot/#/builders/23/builds/7414

The test generally seems to pass, but two of the bots hit the same failure in this test.

dyung · 2025-02-09T07:01:19Z

And just now another failure: https://lab.llvm.org/buildbot/#/builders/190/builds/14342

jhuber6 · 2025-02-09T13:22:19Z

And just now another failure: https://lab.llvm.org/buildbot/#/builders/190/builds/14342

Honestly, I don't know. I'd say that it's because we don't support offloading on Darwin, but the triple is supposed to be x64. I could probably just set the test to unsupported for darwin since GPU offloading isn't supported there. Hopefully a32efd8 resolved it.

dyung · 2025-02-11T19:15:37Z

And just now another failure: https://lab.llvm.org/buildbot/#/builders/190/builds/14342

Honestly, I don't know. I'd say that it's because we don't support offloading on Darwin, but the triple is supposed to be x64. I could probably just set the test to unsupported for darwin since GPU offloading isn't supported there. Hopefully a32efd8 resolved it.

I'm seeing our internal downstream builder hit the exact same error in this test, but that is building a Windows hosted compiler targeting Windows. Looking at our runs, it seems to have started appearing after your change 479ffe8 oddly enough.

jhuber6 · 2025-02-11T19:33:10Z

I'm seeing our internal downstream builder hit the exact same error in this test, but that is building a Windows hosted compiler targeting Windows. Looking at our runs, it seems to have started appearing after your change 479ffe8 oddly enough.

Maybe there's some weird implicit behavior here? Why isn't it failing on the public build bots then. I honestly don't have a good guess why.

Summary: This patch handles the use of `-Xarch_<arch> -Wl,foo` to send an argument to the linker for the embedded offloading jobs in the linker wrapper. This makes it equivalent to `-Xoffload-linker foo`.

dyung · 2025-02-18T10:40:58Z

I'm seeing our internal downstream builder hit the exact same error in this test, but that is building a Windows hosted compiler targeting Windows. Looking at our runs, it seems to have started appearing after your change 479ffe8 oddly enough.

Maybe there's some weird implicit behavior here? Why isn't it failing on the public build bots then. I honestly don't have a good guess why.

I just saw it failed on our public Windows bot:

https://lab.llvm.org/buildbot/#/builders/46/builds/12224

[Clang] Forward -Xarch_<arch> -Wl,foo for GPU toolchains

10727e9

Summary: This patch handles the use of `-Xarch_<arch> -Wl,foo` to send an argument to the linker for the embedded offloading jobs in the linker wrapper. This makes it equivalent to `-Xoffload-linker foo`.

jhuber6 requested review from Artem-B, jdoerfert, Meinersbur, shiltian and yxsamliu February 7, 2025 14:34

llvmbot added clang Clang issues not falling into any other category clang:driver 'clang' and 'clang++' user-facing binaries. Not 'clang-cl' labels Feb 7, 2025

Update Clang.cpp

cd1ad75

jhuber6 requested a review from sarnex February 7, 2025 14:37

sarnex approved these changes Feb 7, 2025

View reviewed changes

clang/lib/Driver/ToolChains/Clang.cpp Show resolved Hide resolved

jhuber6 merged commit 92eeff4 into llvm:main Feb 7, 2025
8 checks passed

jhuber6 deleted the ToolchainLinker branch February 7, 2025 17:43

Meinersbur mentioned this pull request Mar 14, 2025

[Clang][Driver] Use-after free in CreateOffloadingDeviceToolChains #131325

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[Clang] Forward `-Xarch_<arch> -Wl,foo` for GPU toolchains #126248

[Clang] Forward `-Xarch_<arch> -Wl,foo` for GPU toolchains #126248

Uh oh!

jhuber6 commented Feb 7, 2025

Uh oh!

llvmbot commented Feb 7, 2025

Uh oh!

llvmbot commented Feb 7, 2025

Uh oh!

Uh oh!

Uh oh!

dyung commented Feb 9, 2025

Uh oh!

dyung commented Feb 9, 2025

Uh oh!

jhuber6 commented Feb 9, 2025 •

edited

Loading

Uh oh!

dyung commented Feb 11, 2025

Uh oh!

jhuber6 commented Feb 11, 2025

Uh oh!

dyung commented Feb 18, 2025

Uh oh!

Uh oh!

[Clang] Forward -Xarch_<arch> -Wl,foo for GPU toolchains #126248

[Clang] Forward -Xarch_<arch> -Wl,foo for GPU toolchains #126248

Uh oh!

Conversation

jhuber6 commented Feb 7, 2025

Uh oh!

llvmbot commented Feb 7, 2025

Uh oh!

llvmbot commented Feb 7, 2025

Uh oh!

Uh oh!

Uh oh!

dyung commented Feb 9, 2025

Uh oh!

dyung commented Feb 9, 2025

Uh oh!

jhuber6 commented Feb 9, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dyung commented Feb 11, 2025

Uh oh!

jhuber6 commented Feb 11, 2025

Uh oh!

dyung commented Feb 18, 2025

Uh oh!

Uh oh!

[Clang] Forward `-Xarch_<arch> -Wl,foo` for GPU toolchains #126248

[Clang] Forward `-Xarch_<arch> -Wl,foo` for GPU toolchains #126248

jhuber6 commented Feb 9, 2025 •

edited

Loading