On 10/2/26 1:18 PM, Mikael Morin wrote:
Le 02/10/2026 à 18:56, Jerry D a écrit :
I have attached two patches. These must be applied in order.
The first one is a cleanup I found while testing the fixes for PR127347.
It is simple and I plan to push shortly.
The second is a fixup that I renamed to PR127702 which Mikael reported.
For information, after these two go in, I have a fix for PR127684 which I also
discovered during testing. I will post that a little later.
Regression tested on x86_64.
OK for mainline?
Yes, OK.
I have a concern about the first patch wasting memory (say if only one of the
teams allocates), but that's secondary. At least there is no spurious overwrite.
I will investigate memory usage and see what can be improved.
Thanks for review.
The master branch has been updated by Jerry DeLisle <[email protected]>:
https://gcc.gnu.org/g:bd2a164d7fdf1f109a8019aa2d32a2a2ad4893c4
commit r17-4848-gbd2a164d7fdf1f109a8019aa2d32a2a2ad4893c4
Author: Jerry DeLisle <[email protected]>
Date: Sat Sep 19 12:24:05 2026 -0700
libgfortran: [PR127702] caf_shmem: do not take the size of a team from its
barrier
