[clang] Patch series to reapply #118734 and substantially improve it (PR #120534)

[Chandler Carruth via cfe-commits]

chandlerc wrote:

> > > @dyung -- this PR is updated with new fixes. NVPTX hopefully works and neither of my debugging checks fires. But there may still be some other failures I need to chase down, let me know?
> > 
> > 
> > Hmm, looks like there are likely to be ARM and Hexagon failures remaining at least. An interesting question is whether there are any AArch64 failures as if so that will end all ideas I have to dodge this miscompile. I've now pushed a new update ([cf4762f](https://github.com/llvm/llvm-project/commit/cf4762f166d63fde810fd0e829e18e4a42e59e7b)) that should help isolate exactly which targets continue to hit this.
> 
> I've run your latest changes on the bot and I think there are still AArch64 failures as well as ARM failures as you predicted. I've put the test logfile up at [dropbox.com/scl/fi/lh46zur1wlm4dz16ilooq/log_test.cf4762f166d63fde810fd0e829e18e4a42e59e7b.zip?rlkey=mcggerbkvq2rfkkisp34a3tds&dl=0](https://www.dropbox.com/scl/fi/lh46zur1wlm4dz16ilooq/log_test.cf4762f166d63fde810fd0e829e18e4a42e59e7b.zip?rlkey=mcggerbkvq2rfkkisp34a3tds&dl=0)

:sigh: Yeah.

Well, that's pretty definitive. The SVE intrinsics names and types alone more than 2x the max size that this version of MSVC seems able to compile correctly. After the first 64k of string literal, it doesn't *error*, it just silently zeros the rest of the table. =[

Is there any hope of upgrading MSVC? I know you were looking at that, but not sure what progress happened there.

https://github.com/llvm/llvm-project/pull/120534