valentin petrov
valentin petrov
Hi @gkatev, i just tried to reproduce that on the same v5.0.x rev and couldn't. Could you plz share the build cmd line of ompi and full cmd line (by...
@angainor few comments: 1. First 3 cases look p2p related: @yosefe any ideas what can be causing ib errors ? 2. case (4): failure in allreduce with hcoll. It was...
> if UCC_CL_BASIC_TLS is not inthe environment, then is it the same as UCS_TLS @bureddy i don't understand. could you plz elaborate?
> > > if UCC_CL_BASIC_TLS is not inthe environment, then is it the same as UCC_TLS > > > > > > @bureddy i don't understand. could you plz elaborate?...
bot:retest
bot:retest
you are right. probably need to add a TL iface to query max size and in runtime disqualify TLs if user team size exceeds max supported. While UCC_RANK_MAX is internal...
ok to test
decided for team_destroy to be non-blocking with constrain: - team_destroy & team_create should not overlap - 1 team is being destroyed at a time
Hi @zasdfgbnm actually existing autotool code does check for the presence of that function at compile time. Here: https://github.com/openucx/ucc/blob/e96a6de3def951748a8c1bd9f3d074f73c594f1f/config/m4/cuda.m4#L79. So i guess it was available during compile time and in...