[llvm] [Offload] Implement the remaining initial Offload API (PR #122106)

[Callum Fare via llvm-commits]

================
@@ -1327,6 +1327,34 @@ class CUDAGlobalHandlerTy final : public GenericGlobalHandlerTy {
     DeviceGlobal.setPtr(reinterpret_cast<void *>(CUPtr));
     return Plugin::success();
   }
+
+  Error getGlobalMetadataFromImage(GenericDeviceTy &Device,
----------------
callumfare wrote:

The existing GenericGlobalHandlerTy implementation is enough for it to at least build on AMDGPU, I don't think we need to overload it.

I was a bit unclear in explaining the justification for this change in the call, it's more of a difference between cubin and fatbin that results in the generic ELF lookup failing. I'll write up some detailed documentation when I split this out into a separate PR.

https://github.com/llvm/llvm-project/pull/122106