[Clang] Add a flag to include GPU startup files #112025

jhuber6 · 2024-10-11T17:28:03Z

Summary:
The C library for GPUs provides the ability to target regular C/C++
programs by providing the C library and a file containing kernels that
call the main function. This is mostly used for unit tests, this patch
provides a quick way to add them without needing to know the paths. I
currently do this explicitly, but according to the libc++ contributors
we don't want to need to specify these paths manually. See the
discussion in #104515.

I just default to lib/ if the target-specific one isn't found because
the linker will handle giving a reasonable error message if it's not
found. Basically the use-case looks like this.

$ clang test.c --target=amdgcn-amd-amdhsa -mcpu=native -startfiles -stdlib
$ amdhsa-loader a.out
PASS!

llvmbot · 2024-10-11T17:28:45Z

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

@llvm/pr-subscribers-clang

Author: Joseph Huber (jhuber6)

Changes

Summary:
The C library for GPUs provides the ability to target regular C/C++
programs by providing the C library and a file containing kernels that
call the main function. This is mostly used for unit tests, this patch
provides a quick way to add them without needing to know the paths. I
currently do this explicitly, but according to the libc++ contributors
we don't want to need to specify these paths manually. See the
discussion in #104515.

I just default to lib/ if the target-specific one isn't found because
the linker will handle giving a reasonable error message if it's not
found. Basically the use-case looks like this.

$ clang test.c --target=amdgcn-amd-amdhsa -mcpu=native -gpustartfiles
$ amdhsa-loader a.out
PASS!

Full diff: https://github.com/llvm/llvm-project/pull/112025.diff

4 Files Affected:

(modified) clang/include/clang/Driver/Options.td (+3)
(modified) clang/lib/Driver/ToolChains/AMDGPU.cpp (+9)
(modified) clang/lib/Driver/ToolChains/Cuda.cpp (+9)
(added) clang/test/Driver/gpustartfiles.c (+7)

diff --git a/clang/include/clang/Driver/Options.td b/clang/include/clang/Driver/Options.td
index d306c751505e98..b7f7a7cb47f754 100644
--- a/clang/include/clang/Driver/Options.td
+++ b/clang/include/clang/Driver/Options.td
@@ -1316,6 +1316,9 @@ defm offload_via_llvm : BoolFOption<"offload-via-llvm",
   BothFlags<[], [ClangOption], " LLVM/Offload as portable offloading runtime.">>;
 }
 
+def gpustartfiles : Flag<["-"], "gpustartfiles">, Group<Link_Group>,
+  HelpText<"Link the GPU C startup utilities automatically, used for testing.">;
+
 // CUDA options
 let Group = cuda_Group in {
 def cuda_include_ptx_EQ : Joined<["--"], "cuda-include-ptx=">,
diff --git a/clang/lib/Driver/ToolChains/AMDGPU.cpp b/clang/lib/Driver/ToolChains/AMDGPU.cpp
index 2c85d21ebd738c..9a648be4ea3915 100644
--- a/clang/lib/Driver/ToolChains/AMDGPU.cpp
+++ b/clang/lib/Driver/ToolChains/AMDGPU.cpp
@@ -648,6 +648,15 @@ void amdgpu::Linker::ConstructJob(Compilation &C, const JobAction &JA,
         Args.MakeArgString("-plugin-opt=-mattr=" + llvm::join(Features, ",")));
   }
 
+  if (Args.hasArg(options::OPT_gpustartfiles)) {
+    auto IncludePath = getToolChain().getStdlibPath();
+    if (!IncludePath)
+      IncludePath = "/lib";
+    SmallString<128> P(*IncludePath);
+    llvm::sys::path::append(P, "crt1.o");
+    CmdArgs.append({"-lc", "-lm", Args.MakeArgString(P)});
+  }
+
   CmdArgs.push_back("-o");
   CmdArgs.push_back(Output.getFilename());
   C.addCommand(std::make_unique<Command>(
diff --git a/clang/lib/Driver/ToolChains/Cuda.cpp b/clang/lib/Driver/ToolChains/Cuda.cpp
index 7a70cf1c5694fd..ff96ff989db630 100644
--- a/clang/lib/Driver/ToolChains/Cuda.cpp
+++ b/clang/lib/Driver/ToolChains/Cuda.cpp
@@ -641,6 +641,15 @@ void NVPTX::Linker::ConstructJob(Compilation &C, const JobAction &JA,
   llvm::sys::path::append(DefaultLibPath, CLANG_INSTALL_LIBDIR_BASENAME);
   CmdArgs.push_back(Args.MakeArgString(Twine("-L") + DefaultLibPath));
 
+  if (Args.hasArg(options::OPT_gpustartfiles)) {
+    auto IncludePath = getToolChain().getStdlibPath();
+    if (!IncludePath)
+      IncludePath = "/lib";
+    SmallString<128> P(*IncludePath);
+    llvm::sys::path::append(P, "crt1.o");
+    CmdArgs.append({"-lc", "-lm", Args.MakeArgString(P)});
+  }
+
   C.addCommand(std::make_unique<Command>(
       JA, *this,
       ResponseFileSupport{ResponseFileSupport::RF_Full, llvm::sys::WEM_UTF8,
diff --git a/clang/test/Driver/gpustartfiles.c b/clang/test/Driver/gpustartfiles.c
new file mode 100644
index 00000000000000..c1b7a6fa922df4
--- /dev/null
+++ b/clang/test/Driver/gpustartfiles.c
@@ -0,0 +1,7 @@
+// RUN: %clang -target nvptx64-nvidia-cuda -march=sm_61 -gpustartfiles \
+// RUN:   -nogpulib -nogpuinc -### %s 2>&1 | FileCheck -check-prefix=NVPTX %s
+// NVPTX: clang-nvlink-wrapper{{.*}}"-lc" "-lm" "{{.*}}crt1.o"
+//
+// RUN: %clang -target amdgcn-amd-amdhsa -march=gfx90a -gpustartfiles \
+// RUN:   -nogpulib -nogpuinc -### %s 2>&1 | FileCheck -check-prefix=AMDGPU %s
+// AMDGPU: ld.lld{{.*}}"-lc" "-lm" "{{.*}}crt1.o"

clang/lib/Driver/ToolChains/AMDGPU.cpp

jhuber6 · 2024-10-16T14:25:39Z

ping

Summary: The C library for GPUs provides the ability to target regular C/C++ programs by providing the C library and a file containing kernels that call the `main` function. This is mostly used for unit tests, this patch provides a quick way to add them without needing to know the paths. I currently do this explicitly, but according to the libc++ contributors we don't want to need to specify these paths manually. See the discussion in llvm#104515. I just default to `lib/` if the target-specific one isn't found because the linker will handle giving a reasonable error message if it's not found. Basically the use-case looks like this. ```console $ clang test.c --target=amdgcn-amd-amdhsa -mcpu=native -gpustartfiles $ amdhsa-loader a.out PASS! ```

ldionne · 2024-10-18T12:36:06Z

Can you explain again why the compiler isn't providing the C library and the start files implicitly by default, just like it does for non-GPU code?

jhuber6 · 2024-10-18T12:42:44Z

Can you explain again why the compiler isn't providing the C library and the start files implicitly by default, just like it does for non-GPU code?

Because the GPU is not a target that wants to provide a C library and start files by default. This is an edge case where I make the GPU look like a normal target for unit testing purposes. If you use a GPU this way, you're basically turning your 1000$+ card into a 10$ Raspberry Pi, it's not something people want by default. Kernels cannot be optimized out, and currently 0.001% of GPU applications in the wild define a main function on the device side so it would cause a linker error.

jhuber6 · 2024-10-25T16:01:49Z

ping

ldionne · 2024-10-25T18:21:15Z

Just to clarify, I'm mostly neutral on this. A Clang reviewer should chime in and evaluate the change for what it is. It does simplify the setup of the libc++ test suite for GPU, but it's not required since we have other options to set things up.

arsenm · 2024-10-25T22:57:47Z

clang/lib/Driver/ToolChains/AMDGPU.cpp

+  if (Args.hasArg(options::OPT_stdlib))
+    CmdArgs.append({"-lc", "-lm"});
+  if (Args.hasArg(options::OPT_startfiles)) {
+    auto IncludePath = getToolChain().getStdlibPath();


Summary: The C library for GPUs provides the ability to target regular C/C++ programs by providing the C library and a file containing kernels that call the `main` function. This is mostly used for unit tests, this patch provides a quick way to add them without needing to know the paths. I currently do this explicitly, but according to the libc++ contributors we don't want to need to specify these paths manually. See the discussion in llvm#104515. I just default to `lib/` if the target-specific one isn't found because the linker will handle giving a reasonable error message if it's not found. Basically the use-case looks like this. ```console $ clang test.c --target=amdgcn-amd-amdhsa -mcpu=native -startfiles -stdlib $ amdhsa-loader a.out PASS! ```

jhuber6 requested review from AaronBallman, arsenm, Artem-B, jdoerfert, ldionne, llvm-beanz, MaskRay, shiltian and yxsamliu October 11, 2024 17:28

llvmbot added clang Clang issues not falling into any other category backend:AMDGPU clang:driver 'clang' and 'clang++' user-facing binaries. Not 'clang-cl' labels Oct 11, 2024

jhuber6 mentioned this pull request Oct 11, 2024

[libcxx] Add testing configuration for GPU targets #104515

Merged

arsenm reviewed Oct 11, 2024

View reviewed changes

clang/lib/Driver/ToolChains/AMDGPU.cpp Outdated Show resolved Hide resolved

jhuber6 force-pushed the startfiles branch from 9b4a694 to df6ba97 Compare October 14, 2024 12:27

jhuber6 added 3 commits October 17, 2024 12:12

Add -startfiles -stdlib positive versions for GPU

6ccba02

move test

c1ec9bc

jhuber6 force-pushed the startfiles branch from df6ba97 to c1ec9bc Compare October 17, 2024 17:12

jhuber6 added 2 commits October 18, 2024 09:42

Merge branch 'main' into startfiles

aa38c43

Merge branch 'main' into startfiles

3c95de1

arsenm reviewed Oct 25, 2024

View reviewed changes

jhuber6 added 2 commits October 25, 2024 18:36

Update AMDGPU.cpp

79d1432

Update Cuda.cpp

336e319

arsenm approved these changes Oct 25, 2024

View reviewed changes

jhuber6 merged commit d4c4180 into llvm:main Oct 28, 2024
8 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[Clang] Add a flag to include GPU startup files #112025

[Clang] Add a flag to include GPU startup files #112025

Uh oh!

jhuber6 commented Oct 11, 2024 •

edited

Loading

Uh oh!

llvmbot commented Oct 11, 2024 •

edited

Loading

Uh oh!

Uh oh!

jhuber6 commented Oct 16, 2024

Uh oh!

ldionne commented Oct 18, 2024

Uh oh!

jhuber6 commented Oct 18, 2024

Uh oh!

jhuber6 commented Oct 25, 2024

Uh oh!

ldionne commented Oct 25, 2024

Uh oh!

arsenm Oct 25, 2024

Uh oh!

Uh oh!

Uh oh!

[Clang] Add a flag to include GPU startup files #112025

[Clang] Add a flag to include GPU startup files #112025

Uh oh!

Conversation

jhuber6 commented Oct 11, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Oct 11, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

jhuber6 commented Oct 16, 2024

Uh oh!

ldionne commented Oct 18, 2024

Uh oh!

jhuber6 commented Oct 18, 2024

Uh oh!

jhuber6 commented Oct 25, 2024

Uh oh!

ldionne commented Oct 25, 2024

Uh oh!

arsenm Oct 25, 2024

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

jhuber6 commented Oct 11, 2024 •

edited

Loading

llvmbot commented Oct 11, 2024 •

edited

Loading