https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114445
Tobias Burnus <burnus at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Resolution|--- |FIXED
Status|UNCONFIRMED |RESOLVED
--- Comment #7 from Tobias Burnus <burnus at gcc dot gnu.org> ---
Should be FIXED.
[Comment 6's MI300 fail seems to have occurred also in other AMD GPUs, but much
more rarely. See PR119857 for such a fail.]
NOTE: The current code (for new binaries) creates the hash bluntly on the host
– in the plugin – every time the inverse-function table is not empty - it also
leaks the memory on the host side (not freed after copying to the device).
Hence, TODO: Check how to handle multi-image load and unload, free host side
variable eventually?
→ This will be tracked in PR114690.
