[libc] Install a single LLVM-IR version of the GPU library

Summary:
Recent patches have allowed us to treat these libraries as direct
builds. This makes it easier to simply build them to a single LLVM-IR
file. This matches the way these files are presented by the ROCm and
CUDA toolchains and makes it easier to work with.

Author: jhuber6

libc/cmake/modules/LLVMLibCLibraryRules.cmake

@@ -170,6 +170,46 @@ function(add_gpu_entrypoint_library target_name)
   set_target_properties(${target_name} PROPERTIES LIBRARY_OUTPUT_DIRECTORY ${LIBC_LIBRARY_DIR})
 endfunction(add_gpu_entrypoint_library)
 
+# A rule to build a library from a collection of entrypoint objects and bundle
+# it in a single LLVM-IR bitcode file.
+# Usage:
+#     add_gpu_entrypoint_library(
+#       DEPENDS <list of add_entrypoint_object targets>
+#     )
+function(add_bitcode_entrypoint_library target_name)
+  cmake_parse_arguments(
+    "ENTRYPOINT_LIBRARY"
+    "" # No optional arguments
+    "" # No single value arguments
+    "DEPENDS" # Multi-value arguments
+    ${ARGN}
+  )
+  if(NOT ENTRYPOINT_LIBRARY_DEPENDS)
+    message(FATAL_ERROR "'add_entrypoint_library' target requires a DEPENDS list "
+                        "of 'add_entrypoint_object' targets.")
+  endif()
+
+  get_fq_deps_list(fq_deps_list ${ENTRYPOINT_LIBRARY_DEPENDS})
+  get_all_object_file_deps(all_deps "${fq_deps_list}")
+
+  set(objects "")
+  foreach(dep IN LISTS all_deps)
+    set(object $<$<STREQUAL:$<TARGET_NAME_IF_EXISTS:${dep}>,${dep}>:$<TARGET_OBJECTS:${dep}>>)
+    list(APPEND objects ${object})
+  endforeach()
+
+  set(output ${CMAKE_CURRENT_BINARY_DIR}/${target_name}.bc)
+  add_custom_command(
+    OUTPUT ${output}
+    COMMAND ${LIBC_LLVM_LINK} ${objects} -o ${output}
+    DEPENDS ${all_deps}
+    COMMENT "Linking LLVM-IR bitcode for ${target_name}"
+    COMMAND_EXPAND_LISTS
+  )
+  add_custom_target(${target_name} DEPENDS ${output} ${all_deps})
+  set_target_properties(${target_name} PROPERTIES TARGET_OBJECT ${output})
+endfunction(add_bitcode_entrypoint_library)
+
 # A rule to build a library from a collection of entrypoint objects.
 # Usage:
 #     add_entrypoint_library(

libc/cmake/modules/prepare_libc_gpu_build.cmake

@@ -26,6 +26,14 @@ if(NOT LIBC_CLANG_OFFLOAD_PACKAGER)
                       "build")
 endif()
 
+# Identify llvm-link program so we can merge the output IR into a single blob.
+find_program(LIBC_LLVM_LINK
+             NAMES llvm-link NO_DEFAULT_PATH
+             PATHS ${LLVM_BINARY_DIR}/bin ${compiler_path})
+if(NOT LIBC_LLVM_LINK)
+  message(FATAL_ERROR "Cannot find 'llvm-link' for the GPU build")
+endif()
+
 # Optionally set up a job pool to limit the number of GPU tests run in parallel.
 # This is sometimes necessary as running too many tests in parallel can cause
 # the GPU or driver to run out of resources.

libc/lib/CMakeLists.txt

@@ -37,25 +37,41 @@ foreach(archive IN ZIP_LISTS
   endif()
   list(APPEND added_archive_targets ${archive_1})
 
-  # Add the offloading version of the library for offloading languages. These
-  # are installed in the standard search path separate from the other libraries.
+  # The GPU build additionally exports the libraries as both a fat binary
+  # archive and an LLVM-IR bitcode blob.
+  # FIXME: These don't properly re-run if the source gets modified. We willy
+  #        likely need source file dependencies for them.
   if(LIBC_TARGET_OS_IS_GPU)
-    set(libc_gpu_archive_target ${archive_1}gpu)
-    set(libc_gpu_archive_name ${archive_0}gpu-${LIBC_TARGET_ARCHITECTURE})
-
     add_gpu_entrypoint_library(
-      ${libc_gpu_archive_target}
+      ${archive_1}.__fatbin__
       DEPENDS
         ${${archive_2}}
     )
     set_target_properties(
-      ${libc_gpu_archive_target}
+      ${archive_1}.__fatbin__
       PROPERTIES
-        ARCHIVE_OUTPUT_NAME ${libc_gpu_archive_name}
+        ARCHIVE_OUTPUT_NAME ${archive_0}gpu-${LIBC_TARGET_ARCHITECTURE}
+        ARCHIVE_OUTPUT_DIRECTORY ${LLVM_LIBRARY_OUTPUT_INTDIR}
+    )
+
+    add_bitcode_entrypoint_library(
+      ${archive_1}.__bitcode__
+      DEPENDS
+        ${${archive_2}}
+    )
+    add_dependencies(${archive_1} ${archive_1}.__bitcode__)
+
+    install(
+      TARGETS ${added_gpu_archive_targets}
+      ARCHIVE DESTINATION lib${LLVM_LIBDIR_SUFFIX}
+      COMPONENT libc
+    )
+
+    install(FILES $<TARGET_PROPERTY:${archive_1}.__bitcode__,TARGET_OBJECT>
+            DESTINATION ${LIBC_INSTALL_LIBRARY_DIR}
+            RENAME ${archive_1}.bc
+            COMPONENT libc
     )
-    set_target_properties(${libc_gpu_archive_target} PROPERTIES 
-                          ARCHIVE_OUTPUT_DIRECTORY ${LLVM_LIBRARY_OUTPUT_INTDIR})
-    list(APPEND added_gpu_archive_targets ${libc_gpu_archive_target})
   endif()
 endforeach()
 
@@ -66,11 +82,6 @@ install(
 )
 
 if(LIBC_TARGET_OS_IS_GPU)
-  install(
-    TARGETS ${added_gpu_archive_targets}
-    ARCHIVE DESTINATION lib${LLVM_LIBDIR_SUFFIX}
-    COMPONENT libc
-  )
 endif()
 
 if(NOT LIBC_TARGET_OS_IS_BAREMETAL)