[Clang] Add options in LTO mode when cross compiling for AMDGPU

[jhuber6]

Summary

The AMDGPU toolchain support directly compiling GPU images using
cross-compilation such as clang --target=amdgcn-amd-amdhsa foo.c.
However, when attempting to link bitcode this does not work because the
-mcpu options are not forwarded to the linker among others. This patch
simply adds them so that clang --target=amdgcn-amd-amdhsa foo.c -flto
works correctly.

[JonChesterfield]

clang/lib/Driver/ToolChains/AMDGPU.cpp

545

What's the test against OK_Thin for? ThinLTO is a thing but I don't know if it exists (/works) on amdgpu, is this that?

[JonChesterfield]

NVM the above, all the other call sites to addLTOOptions have that test in them so it must be fine

void tools::addLTOOptions(const ToolChain &ToolChain, const ArgList &Args,
                          ArgStringList &CmdArgs, const InputInfo &Output,
                          const InputInfo &Input, bool IsThinLTO);

seems strange that it's a bool argument and not the result of getLTOMode, but pre-existing feature.

[jhuber6]

clang/lib/Driver/ToolChains/AMDGPU.cpp

545

AMDGPU's linker is lld so it works "in theory". But in practice, separate linking of object files isn't fully supported by the ABI. So I consider it a "your mileage may vary" scenario. FWIW, it functioned properly with my experimental libc setup

$ cat main.cpp
int foo();

int main() {
  return foo();
}
$ cat foo.cpp
int foo() { return 3; }
$ clang++ crt1.o main.cpp foo.cpp --target=amdgcn-amd-amdhsa -mcpu=gfx1030 -Wl,-plugin-opt=mcpu=gfx1030 -nogpulib -o image -flto=thin
$ amdhsa_loader image; echo $?
3

[arsenm]

clang/test/Driver/amdgpu-toolchain.c

7

should add a test for thinlto?

[jhuber6]

clang/test/Driver/amdgpu-toolchain.c

7

Only thing it would change is adding -plugin-opt=thinlto. Not sure if it's worth a test since it's probably checked by other uses of addLTOOptions.

clang/lib/Driver/ToolChains/AMDGPU.cpp

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());
   ArgStringList CmdArgs;
   addLinkerCompressDebugSectionsOption(getToolChain(), Args, CmdArgs);
   AddLinkerInputs(getToolChain(), Inputs, Args, CmdArgs, JA);
+  if (C.getDriver().isUsingLTO())
+    addLTOOptions(getToolChain(), Args, CmdArgs, Output, Inputs[0],
+                  C.getDriver().getLTOMode() == LTOK_Thin);
   CmdArgs.push_back("-shared");
   CmdArgs.push_back("-o");
   CmdArgs.push_back(Output.getFilename());
   C.addCommand(std::make_unique<Command>(
       JA, *this, ResponseFileSupport::AtFileCurCP(), Args.MakeArgString(Linker),
       CmdArgs, Inputs, Output));
 }
 

clang/test/Driver/amdgpu-toolchain.c

 // RUN: %clang -### --target=amdgcn--amdhsa -x assembler -mcpu=kaveri %s 2>&1 | FileCheck -check-prefix=AS_LINK %s
 // RUN: %clang -### -g --target=amdgcn--amdhsa -mcpu=kaveri %s 2>&1 | FileCheck -check-prefix=DWARF_VER %s
 // RUN: %clang -### --target=amdgcn-amd-amdpal -x assembler -mcpu=kaveri %s 2>&1 | FileCheck -check-prefix=AS_LINK %s
 // RUN: %clang -### -g --target=amdgcn-amd-amdpal -mcpu=kaveri %s 2>&1 | FileCheck -check-prefix=DWARF_VER %s
 // RUN: %clang -### --target=amdgcn-mesa-mesa3d -x assembler -mcpu=kaveri %s 2>&1 | FileCheck -check-prefix=AS_LINK %s
 // RUN: %clang -### -g --target=amdgcn-mesa-mesa3d -mcpu=kaveri %s 2>&1 | FileCheck -check-prefix=DWARF_VER %s
 
 // AS_LINK: "-cc1as"
 // AS_LINK: ld.lld{{.*}} "-shared"
 
 // DWARF_VER: "-dwarf-version=5"
 
 // RUN: %clang -### --target=amdgcn-amd-amdhsa -mcpu=gfx906 -nogpulib \
 // RUN:   -flto %s 2>&1 | FileCheck -check-prefix=LTO %s
 // LTO: clang{{.*}} "-flto=full"
-// LTO: ld.lld{{.*}}
+// LTO: ld.lld{{.*}}-plugin-opt=mcpu=gfx906