Closed fryeguy52 closed 5 years ago
@fryeguy52 - the commands to reproduce don't seem to work:
[rppawlo@gge BUILD4]$ source /ascldap/users/cmake/std/atdm/load-env.sh Trilinos-atdm-sems-rhel6
users/
[rppawlo@gge BUILD4]$ source /ascldap/users/cmake/std/atdm/load-env.sh Trilinos-atdm-sems-rhel6
users/
[rppawlo@gge BUILD4]$ source /ascldap/users/rppawlo/Trilinos/cmake/std/atdm/load-env.sh Trilinos-atdm-sems-rhel6
Hostname 'gge.srn.sandia.gov' matches known ATDM host 'sems-rhel6' and system 'sems-rhel6'
Setting compiler and build options for buld name 'Trilinos-atdm-sems-rhel6'
***
*** ERROR: A compiler was not specified in 'Trilinos-atdm-sems-rhel6'!
***
Using SEMS RHEL6 compiler stack GNU to build DEBUG code with Kokkos node type SERIAL
[rppawlo@gge BUILD4]$
@rppawlo - sorry about that Roger I updated the instructions above. It should say
source $TRILINOS_DIR/cmake/std/atdm/load-env.sh Trilinos-atdm-sems-rhel6-gnu-opt-openmp
CC: @rppawlo, @fryeguy52
@trilinos/framework, given that this is failing in in every ATDM Trilinos build, then how did this make it through the Trilinos PR testing? What is different about the PR test builds from all of the ATDM Trilinos builds that allowed this build failure to slip through PR testing? We need to figure that out so that at least one of the Trilinos PR builds.
It looks like this build either disables kokkos or kokkos profiling define (my bet is the latter).
fix is trivial. will have a PR up in a minute.
Thanks @rppawlo
@rppawlo said:
It looks like this build either disables kokkos or kokkos profiling define (my bet is the latter).
Yup, that is it. The ATDM Trilinos configuration has Kokkos_ENABLE_Profiling=OFF
set. That came for the EMPIRE configuration for Trilinos.
Is there a good reason why we should not just remove that disable from the ATDM Trilinos configuration? Seems it would be best to change a few options as we need to change from a basic configuration of Trilinos.
We should enable Kokkos profiling by default. The only cost to Kokkos profiling is that entrance and exit of parallel regions does an if
on a function pointer to see if it's null.
All of our CUDA builds are clean today a shown here after the merge of PR #4146.
Closing as complete.
Thanks @rppawlo !
CC: @trilinos/teuchos, @jwillenbring (Trilinos Framework Product Lead), @bartlettroscoe, @fryeguy52
Next Action Status
Merge of fixing PR #4146 to 'develop' on 1/7/2019 resulted in 100% clean ATDM CUDA builds on 1/9/2019.
Description
As shown in this query the test
TeuchosComm_stacked_timer_MPI_2
is not being run in all ATDM builds due to a build failureError building packages/teuchos/comm/test/StackedTimer/CMakeFiles/TeuchosComm_stacked_timer.dir/stacked_timer.cpp.o
The new commits on the day this started failing are:
@rppawlo can you look into this?
Steps to Reproduce
One should be able to reproduce this failure on with a sems rhel6 environment as described in:
More specifically, the commands given for with a sems rhel6 environment are provided at:
The exact commands to reproduce this issue should be: