[llvm] [openmp] Revert "[OFFLOAD][OPENMP] 6.0 compatible interop interface (#143491)" (PR #161279)

[Alex Duran via llvm-commits]

adurang wrote:

> Thank you @adurang. I have tested your patch and it gets rid of the linker error. However, I have also used the reproducer in [ROCm/ROCm#2897](https://github.com/ROCm/ROCm/issues/2897) to see if I get the print outs (even when they report the wrong things for AMD) but those are not working. That test runs into a segmentation fault. When I revert this patch in question, the reproducer is "working" again, i.e., it executes and prints the identifiers.
> 
> Is that something that can be easily fixed? If not I feel like we should revert for now and take some time to look how to move this forward.

I think so. Correcting a minor mistake in my previous patch, the interop test is passing with the level zero plugin (the only one I can test unfortunately):

```
-- Testing: 1 tests, 1 workers --
PASS: libomptarget :: spirv64-intel :: offloading/interop.c (1 of 1)

```

For the other plugins, it would be a matter of implementing the migrating the old interfaces. I think I can do that to the same level   that was implemented before (i.e., moving code around) but someone might need to fill the parts for the new pieces.

Note that even if that test was saying passing before it was kind of lie because the implementation of "use" and "destroy" were doing nothing. I think the tests will probably pass anyway because we're not really checking for any real result but that test wasn't really not a measure of interop working or not working.

Shall I try to fix the other plugins?

https://github.com/llvm/llvm-project/pull/161279