You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In doing portability testing of Chapel over GASNet on a BE powerpc64, Paul Hargrove found that things started breaking after we had upgraded to Qthreads 1.21. In sandialabs/qthreads#301 he fixed an issue in Qthreads that seems to have resolved the Chapel tests he's running (which I understand to be a small subset of our complete test suite), but also notes that Qthreads' make check hangs. We don't seem to run make check for Qthreads, so don't hit that specific issue, but if it's indicative of a larger problem, could be an issue for someone trying to use Qthreads on BE systems. For that reason, I wanted to note this potential issue so that in a future where someone tried a big endian chip and hit surprises, we'd have this to refer back to. Such users could presumably set CHPL_TASKS=fifo to work around the issue, and if such chips became more important to us, we could improve our default CHPL_TASKS selection to try and avoid defaulting to Qthreads on such systems.
The text was updated successfully, but these errors were encountered:
In doing portability testing of Chapel over GASNet on a BE powerpc64, Paul Hargrove found that things started breaking after we had upgraded to Qthreads 1.21. In sandialabs/qthreads#301 he fixed an issue in Qthreads that seems to have resolved the Chapel tests he's running (which I understand to be a small subset of our complete test suite), but also notes that Qthreads'
make check
hangs. We don't seem to runmake check
for Qthreads, so don't hit that specific issue, but if it's indicative of a larger problem, could be an issue for someone trying to use Qthreads on BE systems. For that reason, I wanted to note this potential issue so that in a future where someone tried a big endian chip and hit surprises, we'd have this to refer back to. Such users could presumably setCHPL_TASKS=fifo
to work around the issue, and if such chips became more important to us, we could improve our default CHPL_TASKS selection to try and avoid defaulting to Qthreads on such systems.The text was updated successfully, but these errors were encountered: