charm icon indicating copy to clipboard operation
charm copied to clipboard
verified publisher icon charmplusplus

CmiCheckAffinity causes hang with UCX build on Summit

Open minitu opened this issue 4 years ago • 1 comments

On Summit, CmiCheckAffinity in _initCharm routine (src/ck-core/init.C) causes a hang when run on two nodes. Observed with the UCX and PAMILRTS builds. Runs fine when these routines are commented out.

Charm++ with UCX machine layer built with: ./build charm++ ucx-linux-ppc64le smp openpmix -j --with-production --basedir=$HOME/pmix-3.1.5/install --basedir=$HOME/ucx-1.11.1/install

Test program (1darray/hello) run with: jsrun -n2 -a1 -c2 -K1 -r1 ./hello +ppn 1 +pemap L0 +commap L1

minitu avatar Jan 13 '22 14:01 minitu

Still hangs on Summit.

minitu avatar Jul 18 '22 05:07 minitu