[libomptarget][nfc] Add hook to easily disable building amdgcn bclib

[JonChesterfield]

Summary

[libomptarget][nfc] Add hook to easily disable building amdgcn bclib

This is useful when building LLVM with a toolchain that can't emit code
for amdgcn, e.g. because it overrides the include search path with headers
from another architecture, or the clang compiler is missing builtins.

[tianshilei1992]

LG.

[weiwang]

LGTM.

[protze.joachim]

It seems like a change in this patch introduced a dependency issue for fresh builds.

openmp/libomptarget/deviceRTLs/amdgcn/CMakeLists.txt

158

When I build openmp as LLVM runtime (-DLLVM_ENABLE_RUNTIMES=openmp) in a fresh build directory, opt is not ready when this target should get built. I usually build massively parallel(-j48), so dependency issues will pop up from time to time.
I don't know why this did not happen before, but now this target misses a dependency to the OPT_TOOL built in the outer build process.
I find these lines in my build log:

[101/206] Generating libomptarget-amdgcn-gfx701.bc
FAILED: openmp/libomptarget/deviceRTLs/amdgcn/libomptarget-amdgcn-gfx701.bc 
cd build-main/runtimes/runtimes-bins/openmp/libomptarget/deviceRTLs/amdgcn && build-main/bin/llvm-link linkout.cuda.gfx701.bc | build-main/bin/opt --always-inline -o build-main/runtimes/runtimes-bins/openmp/libomptarget/libomptarget-amdgcn-gfx701.bc
/bin/sh: build-main/bin/opt: permission denied
...
[5987/5995] Linking CXX executable bin/opt

When I revert this patch, I don't see the line about linking bin/opt at all. I have no idea, how this patch impacts the use of opt ?!?

[JonChesterfield]

That is interesting. The cmake does require opt (and llvm-link), and the only reason it does not specify a dependency on those is I do not know how to spell that in cmake.

This patch doesn't change the dependency on opt, but maybe changes the order in which cmake does things, exposing that race condition more frequently.

Does anyone know how to tell cmake not to run this until after op, llvm-link, possibly clang exist?

openmp/libomptarget/deviceRTLs/amdgcn/CMakeLists.txt

 ##===----------------------------------------------------------------------===##
 #
 # Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 # See https://llvm.org/LICENSE.txt for license information.
 # SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
 #
 ##===----------------------------------------------------------------------===##
 #
-# Build the AMDGCN Device RTL if the ROCM tools are available
+# Build the AMDGCN Device RTL bitcode library using clang -ffreestanding
 #
 ##===----------------------------------------------------------------------===##
 
+set(LIBOMPTARGET_BUILD_AMDGCN_BCLIB TRUE CACHE BOOL
+  "Can be set to false to disable building this library.")
+
+if (NOT LIBOMPTARGET_BUILD_AMDGCN_BCLIB)
+  libomptarget_say("Not building AMDGCN device RTL: Disabled by LIBOMPTARGET_BUILD_AMDGCN_BCLIB")
+  return()
+endif()
+
 # Copied from nvptx CMakeLists
 if(CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "x86_64")
   set(aux_triple x86_64-unknown-linux-gnu)
 elseif(CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "ppc64le")
   set(aux_triple powerpc64le-unknown-linux-gnu)
 elseif(CMAKE_HOST_SYSTEM_PROCESSOR MATCHES "aarch64")
   set(aux_triple aarch64-unknown-linux-gnu)
 else()

 
 foreach(mcpu ${mcpus})
   set(bc_files)
   add_cuda_bc_library(${cuda_sources})
 
   set(bc_libname lib${libname}-${mcpu}.bc)
   add_custom_command(
     OUTPUT ${bc_libname}
     COMMAND ${LINK_TOOL} ${bc_files} | ${OPT_TOOL} --always-inline -o ${OUTPUTDIR}/${bc_libname}
     DEPENDS ${bc_files})
 
   add_custom_target(lib${libname}-${mcpu} ALL DEPENDS ${bc_libname})
 
   install(FILES ${OUTPUTDIR}/${bc_libname}
      DESTINATION "${OPENMP_INSTALL_LIBDIR}"
   )
 endforeach()