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

Chandler Carruth via cfe-commits cfe-commits at lists.llvm.org
Sun Jan 5 13:30:35 PST 2025


chandlerc wrote:

> > @dyung -- Ok, my latest attempt to work around these MSVC issues is now pushed to this PR. It also contains a commit of specifically debugging hacks to try and extract more information from any failure here. If you could try doing another build with the latest commit ([2ec750d](https://github.com/llvm/llvm-project/pull/120534/commits/2ec750d0105e1d4d52791f85293d16d28f6ccbd8)) in this PR, would be very interested in how it works.
> 
> ```
> fatal error: error in backend: Name for __nvvm_vote_all is: ''!!!
> ```
> I don't see any instances of the `__builtin_ia32_packsswb128` from a quick search of the test log.

This is awesome news!

It seems the last change was enough and gives me a strong guess as to the bug.

It does mean MSVC is miscompiling long string literals, so when you can upgrade, I would. And we should try to exclude these MSVC versions when we can.

But I may be able to work around more of these, stay tuned for some more updates.

> 
> Here is a compressed copy of the test log file:    [log_test.2ec750d0105e1d4d52791f85293d16d28f6ccbd8.zip](https://www.dropbox.com/scl/fi/60gizuhvi85lz8jxw44t9/log_test.2ec750d0105e1d4d52791f85293d16d28f6ccbd8.zip?rlkey=esic3xnyg4i11rjzuc9i0cr2a&dl=0)



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


More information about the cfe-commits mailing list