[libc] Can build amdgpu libc even if rocm is missing

[JonChesterfield]

Summary

Clang defaults to failing to build if it can't find rocm device libs

[jhuber6]

This should be present in _get_common_compile_options and always passed in. Which scenario is it not present?

[JonChesterfield]

• more

[jhuber6]

LG after addressing comments.

Never caught this because if you can build the tests you presumably had HSA. But we definitely don't want vendor libraries for these builds, thanks.

libc/cmake/modules/LLVMLibCObjectRules.cmake

154

Redundant.

libc/cmake/modules/LLVMLibCTestRules.cmake

578

Add here as well for NVPTX.

[JonChesterfield]

• drop first attempt

[JonChesterfield]

• nvptx as well

[JonChesterfield]

This shows up if you've built / installed HSA but rocm device libs is refusing to build with trunk and you've chosen against trying to patch it up locally.

[JonChesterfield]

• more

libc/cmake/modules/LLVMLibCObjectRules.cmake

foreach(add_gpu_obj_src ${ADD_GPU_OBJ_SRCS})
     foreach(gpu_arch ${LIBC_GPU_ARCHITECTURES})
       get_filename_component(src_name ${add_gpu_obj_src} NAME)
       set(gpu_target_name ${fq_target_name}.${src_name}.${gpu_arch})
       set(compile_options ${ADD_GPU_OBJ_COMPILE_OPTIONS})
       # Derive the triple from the specified architecture.
       if("${gpu_arch}" IN_LIST all_amdgpu_architectures)
         set(gpu_target_triple "amdgcn-amd-amdhsa")
         list(APPEND compile_options "-mcpu=${gpu_arch}")
+        list(APPEND compile_options "-nogpulib")
       elseif("${gpu_arch}" IN_LIST all_nvptx_architectures)
         set(gpu_target_triple "nvptx64-nvidia-cuda")
         get_nvptx_compile_options(nvptx_options ${gpu_arch})
         list(APPEND compile_options "${nvptx_options}")
       else()
         message(FATAL_ERROR "Unknown GPU architecture '${gpu_arch}'")
       endif()
       list(APPEND compile_options "--target=${gpu_target_triple}")

libc/cmake/modules/LLVMLibCTestRules.cmake

function(add_integration_test test_name)
   add_dependencies(${INTEGRATION_TEST_SUITE} ${fq_target_name})
 endfunction(add_integration_test)
 
 set(LIBC_HERMETIC_TEST_COMPILE_OPTIONS ${LIBC_COMPILE_OPTIONS_DEFAULT}
     -fpie -ffreestanding -fno-exceptions -fno-rtti)
 # The GPU build requires overriding the default CMake triple and architecture.
 if(LIBC_GPU_TARGET_ARCHITECTURE_IS_AMDGPU)
   list(APPEND LIBC_HERMETIC_TEST_COMPILE_OPTIONS
-       -mcpu=${LIBC_GPU_TARGET_ARCHITECTURE} -flto --target=${LIBC_GPU_TARGET_TRIPLE})
+       -nogpulib -mcpu=${LIBC_GPU_TARGET_ARCHITECTURE} -flto --target=${LIBC_GPU_TARGET_TRIPLE})
 elseif(LIBC_GPU_TARGET_ARCHITECTURE_IS_NVPTX)
   get_nvptx_compile_options(nvptx_options ${LIBC_GPU_TARGET_ARCHITECTURE})
   list(APPEND LIBC_HERMETIC_TEST_COMPILE_OPTIONS
        ${nvptx_options} -fno-use-cxa-atexit --target=${LIBC_GPU_TARGET_TRIPLE})
 endif()
 
 # Rule to add a hermetic test. A hermetic test is one whose executable is fully
 # statically linked and consists of pieces drawn only from LLVM's libc. Nothing,
 # including the startup objects, come from the system libc.
 #
 # Usage:
 #   add_libc_hermetic_test(