Rename options --cuda-gpu-arch and --no-cuda-gpu-arch

[yaxunl]

Summary

Options --cuda-gpu-arch and --no-cuda-gpu-arch are shared between
CUDA and HIP. It is desirable to rename them for more generic
names to avoid confusion.

One option is --gpu-arch and --no-gpu-arch. However HIP may be
ported to other devices not limited to gpu, so it may be better
to name them as --offload-arch and --no-offload-arch.

Other choices may be:

--device-arch
--offload-device-arch
--accelerator-arch
--aux-arch
--aux-cpu

The original options will be alias to the new options.

[gregrodgers]

This was discussed on llvm-dev three years ago. Here is the thread.

http://lists.llvm.org/pipermail/llvm-dev/2017-February/109930.html

The last name discussed was "-- offload-arch". I don't believe we need a list option anymore. So ignore the very old request for --offload-archs.

I am ok with the patch the way it is. In the future, we should consider renaming the CudaArch class to OffloadArch class . Also the GpuArchList is currently only initialized in CudaActionBuilder. Eventually this is will have to be done for HIPActionBuilder and OpenMPActionBuilder. Could you consider creating a function to InitializeGpuArchList ?

[tra]

+ @echristo who OK'ed the idea conditional on the actual patch. :-)

LGTM overall.

[yaxunl]

In D76987#1950366, @gregrodgers wrote:

This was discussed on llvm-dev three years ago. Here is the thread.

http://lists.llvm.org/pipermail/llvm-dev/2017-February/109930.html

The last name discussed was "-- offload-arch". I don't believe we need a list option anymore. So ignore the very old request for --offload-archs.

I am ok with the patch the way it is. In the future, we should consider renaming the CudaArch class to OffloadArch class . Also the GpuArchList is currently only initialized in CudaActionBuilder. Eventually this is will have to be done for HIPActionBuilder and OpenMPActionBuilder. Could you consider creating a function to InitializeGpuArchList ?

GpuArchList is initialized by member function initialize() of CudaActionBuilderBase, which is inherited by CudaActionBuilder and HIPActionBuilder, therefore GpuArchList is initialized in both CudaActionBuilder and HIPActionBuilder.

clang/include/clang/Driver/Options.td

HelpText<"Compile CUDA code for host only.  Has no effect on non-CUDA "
            "compilations.">;
 def cuda_compile_host_device : Flag<["--"], "cuda-compile-host-device">,
   HelpText<"Compile CUDA code for both host and device (default).  Has no "
            "effect on non-CUDA compilations.">;
 def cuda_include_ptx_EQ : Joined<["--"], "cuda-include-ptx=">, Flags<[DriverOption]>,
   HelpText<"Include PTX for the following GPU architecture (e.g. sm_35) or 'all'. May be specified more than once.">;
 def no_cuda_include_ptx_EQ : Joined<["--"], "no-cuda-include-ptx=">, Flags<[DriverOption]>,
   HelpText<"Do not include PTX for the following GPU architecture (e.g. sm_35) or 'all'. May be specified more than once.">;
+def offload_arch_EQ : Joined<["--"], "offload-arch=">, Flags<[DriverOption]>,
+  HelpText<"CUDA/HIP offloading device architecture (e.g. sm_35, gfx906).  May be specified more than once.">;
 def cuda_gpu_arch_EQ : Joined<["--"], "cuda-gpu-arch=">, Flags<[DriverOption]>,
-  HelpText<"CUDA GPU architecture (e.g. sm_35).  May be specified more than once.">;
+  Alias<offload_arch_EQ>;
 def hip_link : Flag<["--"], "hip-link">,
   HelpText<"Link clang-offload-bundler bundles for HIP">;
-def no_cuda_gpu_arch_EQ : Joined<["--"], "no-cuda-gpu-arch=">, Flags<[DriverOption]>,
-  HelpText<"Remove GPU architecture (e.g. sm_35) from the list of GPUs to compile for. "
+def no_offload_arch_EQ : Joined<["--"], "no-offload-arch=">, Flags<[DriverOption]>,
+  HelpText<"Remove CUDA/HIP offloading device architecture (e.g. sm_35, gfx906) from the list of devices to compile for. "
            "'all' resets the list to its default value.">;
+def no_cuda_gpu_arch_EQ : Joined<["--"], "no-cuda-gpu-arch=">, Flags<[DriverOption]>,
+  Alias<no_offload_arch_EQ>;
 def cuda_noopt_device_debug : Flag<["--"], "cuda-noopt-device-debug">,
   HelpText<"Enable device-side debug info generation. Disables ptxas optimizations.">;
 def no_cuda_version_check : Flag<["--"], "no-cuda-version-check">,
   HelpText<"Don't error out if the detected version of the CUDA install is "
            "too low for the requested CUDA gpu architecture.">;
 def no_cuda_noopt_device_debug : Flag<["--"], "no-cuda-noopt-device-debug">;
 def cuda_path_EQ : Joined<["--"], "cuda-path=">, Group<i_Group>,
   HelpText<"CUDA installation path">;

clang/lib/Driver/Driver.cpp

bool initialize() override {
                               options::OPT_cuda_device_only);
       EmitLLVM = Args.getLastArg(options::OPT_emit_llvm);
       EmitAsm = Args.getLastArg(options::OPT_S);
 
       // Collect all cuda_gpu_arch parameters, removing duplicates.
       std::set<CudaArch> GpuArchs;
       bool Error = false;
       for (Arg *A : Args) {
-        if (!(A->getOption().matches(options::OPT_cuda_gpu_arch_EQ) ||
-              A->getOption().matches(options::OPT_no_cuda_gpu_arch_EQ)))
+        if (!(A->getOption().matches(options::OPT_offload_arch_EQ) ||
+              A->getOption().matches(options::OPT_no_offload_arch_EQ)))
           continue;
         A->claim();
 
         const StringRef ArchStr = A->getValue();
-        if (A->getOption().matches(options::OPT_no_cuda_gpu_arch_EQ) &&
+        if (A->getOption().matches(options::OPT_no_offload_arch_EQ) &&
             ArchStr == "all") {
           GpuArchs.clear();
           continue;
         }
         CudaArch Arch = StringToCudaArch(ArchStr);
         if (Arch == CudaArch::UNKNOWN) {
           C.getDriver().Diag(clang::diag::err_drv_cuda_bad_gpu_arch) << ArchStr;
           Error = true;
-        } else if (A->getOption().matches(options::OPT_cuda_gpu_arch_EQ))
+        } else if (A->getOption().matches(options::OPT_offload_arch_EQ))
           GpuArchs.insert(Arch);
-        else if (A->getOption().matches(options::OPT_no_cuda_gpu_arch_EQ))
+        else if (A->getOption().matches(options::OPT_no_offload_arch_EQ))
           GpuArchs.erase(Arch);
         else
           llvm_unreachable("Unexpected option.");
       }
 
       // Collect list of GPUs remaining in the set.
       for (CudaArch Arch : GpuArchs)
         GpuArchList.push_back(Arch);