[HeterogeneousDwarf] Support translation of DIOp-based DIExpressions

[epilk]

These expressions exist downstream on amd-staging and are necessary for
debugging on device. Fixes SWDEV-445691.

[slinder1]

Took me a moment to get up-to-speed on the SPIRV representation, so I may have missed something, but aside from a few nits this LGTM!

[slinder1]

I'm not sure why the default is to just link to the GV and not look for the GVE, but I suppose we only really care about the NewElements case anyway for working symbolic debug info, so this LGTM!

(Maybe I answered my question there, and the point is to actively avoid using faulty expressions for the non-heterogeneous case?)

[AlexVlx]

@epilk something I had not considered, but is important, is that this breaks compatibility with upstream AFAICT (for example the DIExprOps header is not available), and we would like the fork to still build with upstream. Perhaps a __has_include guard would suffice here (and then based on that predicate we can also turn off the unsupported bits).

[epilk]

@epilk something I had not considered, but is important, is that this breaks compatibility with upstream AFAICT (for example the DIExprOps header is not available), and we would like the fork to still build with upstream. Perhaps a __has_include guard would suffice here (and then based on that predicate we can also turn off the unsupported bits).

Ah, okay I didn't realize that the downstream translator would need to work with upstream llvm. Using __has_include seems like a good fix, I'll write that up...

[AlexVlx]

@epilk something I had not considered, but is important, is that this breaks compatibility with upstream AFAICT (for example the DIExprOps header is not available), and we would like the fork to still build with upstream. Perhaps a __has_include guard would suffice here (and then based on that predicate we can also turn off the unsupported bits).

Ah, okay I didn't realize that the downstream translator would need to work with upstream llvm. Using __has_include seems like a good fix, I'll write that up...

Thank you and apologies for the inconvenience. I spaced out on this when I reviewed.

[epilk]

Thank you and apologies for the inconvenience. I spaced out on this when I reviewed.

No problem, thanks for spotting the issue! #14