[libc] Remove special full build handling for GPU (#128572)

Summary:
Currently we default to non-fullbuild for all targets, but realistically
we should do this depending on the target OS. Some OS's like the GPU or
upcoming UEFI have no existing hosted system, so they cannot be built
with an overlay build. These are already errors so there's no reason to
complicate things and require passing it in through the runtimes build.

Author: jhuber6

libc/CMakeLists.txt

@@ -128,7 +128,13 @@ else()
   endif()
 endif()
 
-option(LLVM_LIBC_FULL_BUILD "Build and test LLVM libc as if it is the full libc" OFF)
+# Some targets can only support the full build.
+set(default_to_full_build OFF)
+if(LIBC_TARGET_OS_IS_GPU)
+  set(default_to_full_build ON)
+endif()
+
+option(LLVM_LIBC_FULL_BUILD "Build and test LLVM libc as if it is the full libc" ${default_to_full_build})
 option(LLVM_LIBC_IMPLEMENTATION_DEFINED_TEST_BEHAVIOR "Build LLVM libc tests assuming our implementation-defined behavior" ON)
 option(LLVM_LIBC_ENABLE_LINTING "Enables linting of libc source files" OFF)
 

llvm/runtimes/CMakeLists.txt

@@ -535,15 +535,13 @@ if(build_runtimes)
              "-DRUNTIMES_amdgcn-amd-amdhsa_LIBC_GPU_LOADER_EXECUTABLE=$<TARGET_FILE:amdhsa-loader>")
         list(APPEND extra_deps amdhsa-loader)
       endif()
-      list(APPEND extra_cmake_args "-DRUNTIMES_amdgcn-amd-amdhsa_LLVM_LIBC_FULL_BUILD=ON")
     endif()
     if("libc" IN_LIST RUNTIMES_nvptx64-nvidia-cuda_LLVM_ENABLE_RUNTIMES)
       if(TARGET nvptx-loader)
         list(APPEND extra_cmake_args
              "-DRUNTIMES_nvptx64-nvidia-cuda_LIBC_GPU_LOADER_EXECUTABLE=$<TARGET_FILE:nvptx-loader>")
         list(APPEND extra_deps nvptx-loader)
       endif()
-      list(APPEND extra_cmake_args "-DRUNTIMES_nvptx64-nvidia-cuda_LLVM_LIBC_FULL_BUILD=ON")
     endif()
     if(TARGET clang-offload-packager)
       list(APPEND extra_deps clang-offload-packager)