[libomptarget][amdgpu][nfc] Fix build on centos

[JonChesterfield]

Summary

[libomptarget][amdgpu][nfc] Fix build on centos

rtl.cpp replaced 224 with a #define from elf.h, but that
doesn't work on a centos 7 build machine with an old elf.h

[JonChesterfield]

Adding Tony and Ron as reviewers, since I think one of you replaced the 224 literal.

We could expose amdgcnMachineID=224 in a header which is included by a file that includes llvm's elf.h, and static_assert that 224 is indeed the right value, but I don't think the line noise is worth it.

[ronlieb]

LGTM

[t-tye]

openmp/libomptarget/plugins/amdgpu/src/rtl.cpp

641

What ELF header file is being used? Should it be using the LLVM provide one which does have the definition? Or is this file part of a component that does not want a dependency on LLVM headers even though it is present in the LLVM code base?

[JonChesterfield]

This file uses the system elf.h because it uses libelf. Sadly llvm's elf.h and the gnu elf.h conflict if both are included.

I'm vaguely inclined to drop the dependency on libelf, but as libomptarget uses it anyway it wouldn't make much difference.

openmp/libomptarget/plugins/amdgpu/src/rtl.cpp

 }
 void finiAsyncInfoPtr(__tgt_async_info *async_info_ptr) {
   assert(async_info_ptr);
   assert(async_info_ptr->Queue);
   async_info_ptr->Queue = 0;
 }
 
 bool elf_machine_id_is_amdgcn(__tgt_device_image *image) {
-  const uint16_t amdgcnMachineID = EM_AMDGPU;
+  const uint16_t amdgcnMachineID = 224; // EM_AMDGPU may not be in system elf.h
   int32_t r = elf_check_machine(image, amdgcnMachineID);
   if (!r) {
     DP("Supported machine ID not found\n");
   }
   return r;
 }
 
 uint32_t elf_e_flags(__tgt_device_image *image) {