[clad] Bump clad to version 2.4#22719
Conversation
Test Results 23 files 23 suites 3d 15h 27m 22s ⏱️ For more details on these failures, see this check. Results for commit be8075f. ♻️ This comment has been updated with latest results. |
|
@bellenot, @guitargeek are these failures expected? |
|
They are spurious Windows failures that are unrelated. I re-started the tests just on Windows, hopefully it works fine now! |
Clad 2.4 is now compatible with clang-12 through clang-22, having ported to LLVM 22 and shifted its support window to the last ten major releases (clang-11 has been dropped). This release strengthens both forward and reverse modes with OpenMP critical directive support, reverse-mode handling of member access on call results, and the ability to redefine builtin custom derivatives. CUDA support advances with tape offloading to VRAM and device-callable tape element destruction. A major theme is robustness: a graceful fallback when differentiating parameterless functions, a correctness fix for TBR analysis that was silently producing wrong derivatives, and an extensive round of memory-safety cleanup—scope-leak fixes, use-after-poison repairs, and explicit ownership—surfaced by new LLVM 22 AddressSanitizer CI coverage. Static-library builds now skip backend pass registration, fixing rootcling dictionary generation in ROOT. For details, please refer to the release notes.
guitargeek
left a comment
guitargeek
left a comment
There was a problem hiding this comment.
Thanks a lot for the updates!
|
Thanks for the tag. |
|
@vgvassilev , looks like this commit broke the |
|
I guess the only common denominator for @SahilPatidar's PR and this is vgvassilev/clad#1819 and I do not see how this can affect the cuda pcm... |
|
I am testing the revert of clad version (on top of latest root master) in cms-sw/cmsdist#10687 and there |
|
Thanks for investigating. Will ping Sahil. |
|
see cms-sw/cmsdist#10684 (comment) for the debug stacktrace of the failure |
4 participants
No description provided.