[OpenMP] Enable the IDE support for the device runtime

[tianshilei1992]

Summary

Currently the device runtime is built as a custom target, which will not be included
in the compile commands. Those language servers using compile commands cannot
handle device runtime correctly.

In this patch, when CMAKE_EXPORT_COMPILE_COMMANDS is turned on, dummy
targets that will be excluded from all will be added. Those targets will not be
built or installed if we just simply do make or make install, but their
compilation will be included in the compile commands.

[jhuber6]

Does this work properly given the OpenMP integration? I've found that my language server doesn't know how to work around the variants.

[tianshilei1992]

In D148870#4285771, @jhuber6 wrote:

Does this work properly given the OpenMP integration? I've found that my language server doesn't know how to work around the variants.

I'm using clangd with runtime build for indexing, and it works on my side.

[jhuber6]

We should check if CMAKE_EXPORT_COMPILE_COMMANDS is on as well.

[tianshilei1992]

In D148870#4285776, @jhuber6 wrote:

We should check if CMAKE_EXPORT_COMPILE_COMMANDS is on as well.

I'm not sure that is necessary, though this is mostly only for compile commands. I'd like it similar to LLVM_ENABLE_IDE.

[jhuber6]

In D148870#4285793, @tianshilei1992 wrote:

In D148870#4285776, @jhuber6 wrote:

We should check if CMAKE_EXPORT_COMPILE_COMMANDS is on as well.

I'm not sure that is necessary, though this is mostly only for compile commands. I'd like it similar to LLVM_ENABLE_IDE.

I'm not sure familiar with that option, but I guess that makes sense. I still think this is something that should probably be on by default if we have compile commands on though.

[tianshilei1992]

In D148870#4287138, @jhuber6 wrote:

In D148870#4285793, @tianshilei1992 wrote:

In D148870#4285776, @jhuber6 wrote:

We should check if CMAKE_EXPORT_COMPILE_COMMANDS is on as well.

I'm not sure that is necessary, though this is mostly only for compile commands. I'd like it similar to LLVM_ENABLE_IDE.

I'm not sure familiar with that option, but I guess that makes sense. I still think this is something that should probably be on by default if we have compile commands on though.

I can add that, say if we have that CMake option turned on, we enable this, or LIBOMPTARGET_DEVICE_RTL_IDE_SUPPORT is on.

[tianshilei1992]

enable it when CMAKE_EXPORT_COMPILE_COMMANDS is on as well

[tianshilei1992]

remove custom variable

openmp/libomptarget/DeviceRTL/CMakeLists.txt

function(compileDeviceRTLLibrary target_cpu target_name target_triple)
   if("${CLANG_TOOL}" STREQUAL "$<TARGET_FILE:clang>")
     add_custom_command(OUTPUT ${output_name}
       DEPENDS clang
       APPEND)
   endif()
 
   set_property(DIRECTORY APPEND PROPERTY ADDITIONAL_MAKE_CLEAN_FILES ${output_name})
   set_property(TARGET omptarget.devicertl.all_objs APPEND PROPERTY IMPORTED_OBJECTS ${output_name})
+
+  if (CMAKE_EXPORT_COMPILE_COMMANDS)
+    set(ide_target_name omptarget-ide-${target_name}-${target_cpu})
+    add_library(${ide_target_name} STATIC EXCLUDE_FROM_ALL ${src_files})
+    target_compile_options(${ide_target_name} PRIVATE
+      -fopenmp --offload-arch=${target_cpu} -fopenmp-cuda-mode
+      -mllvm -openmp-opt-disable
+      -foffload-lto -fvisibility=hidden --offload-device-only
+      -nocudalib -nogpulib -nostdinc -Wno-unknown-cuda-version
+    )
+    target_compile_definitions(${ide_target_name} PRIVATE SHARED_SCRATCHPAD_SIZE=512)
+    target_include_directories(${ide_target_name} PRIVATE
+      ${include_directory}
+      ${devicertl_base_directory}/../include
+      ${LIBOMPTARGET_LLVM_INCLUDE_DIRS}
+    )
+    install(TARGETS ${ide_target_name} EXCLUDE_FROM_ALL)
+  endif()
 endfunction()
 
 # Generate a Bitcode library for all the gpu architectures the user requested.
 add_custom_target(omptarget.devicertl.nvptx)
 add_custom_target(omptarget.devicertl.amdgpu)
 foreach(gpu_arch ${LIBOMPTARGET_DEVICE_ARCHITECTURES})
   if("${gpu_arch}" IN_LIST all_amdgpu_architectures)
     compileDeviceRTLLibrary(${gpu_arch} amdgpu amdgcn-amd-amdhsa)