Merge Pull Request trilinos#7151 from bartlettroscoe/Trilinos/cdofa-1…

…05-atdm-van1-tx2

Automatically Merged using Trilinos Pull Request AutoTester
PR Title: ATDM: Add new 'van-tx2' env to support ASTRA systems
PR Author: bartlettroscoe

cmake/ctest/drivers/atdm/ctest-s-driver.sh

@@ -9,6 +9,7 @@ echo "  ==> `date`"
 echo
 
 echo "Loading env and running ctest -S comamnd to configure, build, and test ..."
+echo
 
 source ${WORKSPACE}/Trilinos/cmake/ctest/drivers/atdm/utils/setup_env.sh
 

cmake/ctest/drivers/atdm/utils/setup_env.sh

@@ -1,3 +1,5 @@
+set +x
+
 #
 # A) Load the env
 #

cmake/ctest/drivers/atdm/van1-tx2/drivers/Trilinos-atdm-van1-tx2_arm-20.0_openmpi-4.0.2_openmp_static_dbg.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+
+#export SALLOC_CTEST_TIME_LIMIT_MINUTES=1:00:00
+
+if [ "${Trilinos_TRACK}" == "" ] ; then
+  export Trilinos_TRACK=Experimental
+fi
+
+$WORKSPACE/Trilinos/cmake/ctest/drivers/atdm/van1-tx2/local-driver.sh

cmake/ctest/drivers/atdm/van1-tx2/drivers/Trilinos-atdm-van1-tx2_arm-20.0_openmpi-4.0.2_openmp_static_opt.sh

@@ -0,0 +1,9 @@
+#!/bin/bash
+
+#export SALLOC_CTEST_TIME_LIMIT_MINUTES=1:00:00
+
+if [ "${Trilinos_TRACK}" == "" ] ; then
+  export Trilinos_TRACK=Specialized
+fi
+
+$WORKSPACE/Trilinos/cmake/ctest/drivers/atdm/van1-tx2/local-driver.sh

cmake/ctest/drivers/atdm/van1-tx2/local-driver-on-allocation.sh

@@ -0,0 +1,51 @@
+#!/bin/bash
+#
+# This script is designed to run inside of an allocation which then runs the
+# build and tests on the same allocated compute node.  (See
+# van1-tx2/local-drivers.sh and atdm/READM.md for details.)
+#
+
+set +x
+
+echo
+echo "======================================================================"
+echo ""
+echo "  Running ${JOB_NAME}"
+echo "  in salloc allocation from node '$(hostname)'"
+echo ""
+echo "======================================================================"
+echo
+
+# Must use same site name for build and test results so they match up on
+# CDash!
+export CTEST_SITE=${ATDM_CONFIG_CDASH_HOSTNAME}
+
+echo
+echo "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++"
+echo "***"
+echo "*** Running start, update, configure and build on a compute node"
+echo "***"
+echo
+
+set -x
+env \
+  CTEST_DO_TEST=FALSE \
+  srun -N 1 \
+  $WORKSPACE/Trilinos/cmake/ctest/drivers/atdm/ctest-s-driver.sh
+set +x
+
+echo
+echo "++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++"
+echo "***"
+echo "*** Running tests on compute node launched from the login node"
+echo "***"
+echo
+
+set -x
+env \
+  CTEST_DO_NEW_START=FALSE \
+  CTEST_DO_UPDATES=FALSE \
+  CTEST_DO_CONFIGURE=FALSE \
+  CTEST_DO_BUILD=FALSE \
+  $WORKSPACE/Trilinos/cmake/ctest/drivers/atdm/ctest-s-driver.sh
+set +x

cmake/ctest/drivers/atdm/van1-tx2/local-driver.sh

@@ -0,0 +1,32 @@
+#!/bin/bash -l
+
+set +x
+
+if [[ "${SALLOC_CTEST_LIMIT_MINUTES}" == "" ]] ; then
+  SALLOC_CTEST_LIMIT_MINUTES=4:00:00
+  # From prior builds on 'stria', it looks like 4 hours should be plenty of
+  # time to do the build and tests as it has taken only about 2 1/2 hours to
+  # do everything.
+fi
+
+source $WORKSPACE/Trilinos/cmake/std/atdm/load-env.sh $JOB_NAME
+echo
+
+if [[ "${ATDM_CONFIG_WCID_ACCOUNT}" == "" ]] ; then
+  export ATDM_CONFIG_WCID_ACCOUNT=${ATDM_CONFIG_WCID_ACCOUNT_DEFAULT}
+fi
+
+set -x
+
+salloc -N 1 --time=${SALLOC_CTEST_LIMIT_MINUTES} -p short,batch \
+  --account=${ATDM_CONFIG_WCID_ACCOUNT} \
+  $WORKSPACE/Trilinos/cmake/ctest/drivers/atdm/van1-tx2/local-driver-on-allocation.sh
+
+set -x
+
+# NOTE: Above, we get a single compute-node allocation using 'salloc' and then
+# run the build and tests from inside of that.
+
+# NOTE: We might need to switch from salloc to the more complex sbatch
+# appraoch in the function atdm_run_script_on_compute_node.  If we see random
+# ODTE errors then that is what we will need to do.

cmake/std/atdm/ATDMDevEnvSettings.cmake

@@ -58,6 +58,7 @@ ATDM_SET_ATDM_VAR_FROM_ENV_AND_DEFAULT(FPIC OFF)
 ATDM_SET_ATDM_VAR_FROM_ENV_AND_DEFAULT(COMPLEX OFF)
 
 ATDM_SET_ATDM_VAR_FROM_ENV_AND_DEFAULT(Kokkos_ENABLE_SERIAL ON)
+#ATDM_SET_ATDM_VAR_FROM_ENV_AND_DEFAULT(HDF5_NO_SYSTEM_PATHS OFF)
 
 IF (ATDM_USE_PTHREADS)
   MESSAGE(FATAL_ERROR "Error, the Kokkos Pthreads backend is no longer supported!"
@@ -273,6 +274,7 @@ ATDM_SET_CACHE(MPI_EXEC_POST_NUMPROCS_FLAGS "$ENV{ATDM_CONFIG_MPI_POST_FLAGS}"
   CACHE STRING)
 ATDM_SET_CACHE(Trilinos_VERBOSE_CONFIGURE OFF CACHE BOOL)
 ATDM_SET_CACHE(Trilinos_ENABLE_EXPLICIT_INSTANTIATION ON CACHE BOOL)
+#ATDM_SET_CACHE(HDF5_NO_SYSTEM_PATHS ${ATDM_HDF5_NO_SYSTEM_PATHS} CACHE BOOL)
 ATDM_SET_CACHE(Trilinos_ENABLE_INSTALL_CMAKE_CONFIG_FILES ON CACHE BOOL)
 ATDM_SET_CACHE(Trilinos_ENABLE_DEVELOPMENT_MODE OFF CACHE BOOL)
 ATDM_SET_CACHE(Trilinos_ASSERT_MISSING_PACKAGES ON CACHE BOOL)
@@ -294,6 +296,7 @@ ATDM_SET_CACHE(EpetraExt_ENABLE_HDF5 OFF CACHE BOOL)
 ATDM_SET_CACHE(Panzer_ENABLE_FADTYPE "Sacado::Fad::DFad<RealType>" CACHE STRING)
 ATDM_SET_CACHE(Phalanx_KOKKOS_DEVICE_TYPE "${ATDM_NODE_TYPE}" CACHE STRING)
 ATDM_SET_CACHE(Phalanx_SHOW_DEPRECATED_WARNINGS OFF CACHE BOOL)
+ATDM_SET_CACHE(DAKOTA_ENABLE_TESTS OFF CACHE BOOL)
 ATDM_SET_CACHE(Tpetra_INST_CUDA "${ATDM_USE_CUDA}" CACHE BOOL)
 ATDM_SET_CACHE(Tpetra_INST_SERIAL "${ATDM_INST_SERIAL}" CACHE BOOL)
 ATDM_SET_CACHE(Tpetra_INST_INT_INT OFF CACHE BOOL)
@@ -398,6 +401,9 @@ ATDM_SET_CACHE(TPL_LAPACK_LIBRARIES "$ENV{ATDM_CONFIG_LAPACK_LIBS}" CACHE FILEPA
 ATDM_SET_ENABLE(TPL_ENABLE_CGNS ${ATDM_ENABLE_SPARC_SETTINGS})
 ATDM_SET_CACHE(CGNS_INCLUDE_DIRS "$ENV{CGNS_ROOT}/include" CACHE FILEPATH)
 ATDM_SET_CACHE(CGNS_LIBRARY_DIRS "$ENV{CGNS_ROOT}/lib" CACHE FILEPATH)
+IF (NOT "$ENV{ATDM_CONFIG_CGNS_LIBRARY_NAMES}" STREQUAL "")
+  ATDM_SET_CACHE(CGNS_LIBRARY_NAMES "$ENV{ATDM_CONFIG_CGNS_LIBRARY_NAMES}" CACHE FILEPATH)
+ENDIF()
 ATDM_SET_CACHE(TPL_CGNS_LIBRARIES "$ENV{ATDM_CONFIG_CGNS_LIBS}" CACHE FILEPATH)
 
 # HDF5

cmake/std/atdm/README.md

@@ -609,6 +609,7 @@ example, skip the configure, skip the build, skip running tests, etc.
 * <a href="#cee-rhel6-and-rhel7-environment">CEE RHEL6 and RHEL7 Environment</a>
 * <a href="#waterman">waterman</a>
 * <a href="#ats-2">ATS-2</a>
+* <a href="#astra-vanguard-arm-system">ASTRA (Vanguard ARM System)</a>
 
 
 ### ride/white
@@ -729,7 +730,7 @@ $ make NP=16
 $ salloc -N1 --time=0:20:00 --account=<YOUR_WCID> ctest -j4
 ```
 
-To get information on <YOUR_WCID> used above, there is a WC tool tab on
+To get information on `<YOUR_WCID>` used above, there is a WC tool tab on
 computing.sandia.gov
 
 **NOTE:** Unlike some of the other machines, one must load the environment,
@@ -1162,6 +1163,89 @@ $ ./checkin-test-atdm.sh <buildname0> <buildname1> ... \
   use caution when doing so as not to disturb the job running.
 
 
+### ASTRA (Vanguard ARM System)
+
+Once logged onto a supported Vanguard ARM system (called system 'van1-tx2')
+system like 'stria', one can build and configure on a login node.
+
+To configure, build and run the tests for the default `arm-20.0` build for
+`Kokkos` (after cloning Trilinos on the 'develop' branch), run the following
+from a login node on 'stria':
+
+```bash
+$ cd <some_build_dir>/
+
+# List available environments
+$ source $TRILINOS_DIR/cmake/std/atdm/load-env.sh help
+
+# Load hsw env and configure on the login node
+$ source $TRILINOS_DIR/cmake/std/atdm/load-env.sh arm-20.0
+$ cmake -G Ninja \
+  -DTrilinos_CONFIGURE_OPTIONS_FILE:STRING=cmake/std/atdm/ATDMDevEnv.cmake \
+  -DTrilinos_ENABLE_TESTS=ON \
+  -DTrilinos_ENABLE_Tpetra=ON \
+  $TRILINOS_DIR
+
+$ make NP=8  # This is a shared node!
+
+# Get a node allocation and run ctest
+$ salloc -N 1 --time=2:00:00  -p short,batch --account=<YOUR_WCID> ctest -j4
+```
+
+One can also get an allocation first and then configure, build on a compute
+node, and then run the test suite using:
+
+```bash
+$ salloc -N 1 --time=4:00:00 -p short,batch --account=<YOUR_WCID> bash
+# NOTE: After the above runs, hostname=stria-login<n> but now a compute node
+# has been allocated for immediately usage.
+
+$ source $TRILINOS_DIR/cmake/std/atdm/load-env.sh arm-20.0
+
+$ cmake -G Ninja \
+  -DTrilinos_CONFIGURE_OPTIONS_FILE:STRING=cmake/std/atdm/ATDMDevEnv.cmake \
+  -DTrilinos_ENABLE_TESTS=ON \
+  -DTrilinos_ENABLE_Tpetra=ON \
+  $TRILINOS_DIR
+
+$ srun -N 1 make NP=20  # We have the entire compute node to ourselves!
+
+$ ctest -j4
+```
+
+The advantage of the latter approach is that one just waits once for a node
+allocation and then one can immediately run fast parallel builds on the compute
+node (taking up the entire node).  Then one can run the test suite multiple
+times without waiting for a new allocation.
+
+One can also directly build on a compute node from the login node in one
+command:
+
+```bash
+$ srun -N 1 --time=2:00:00 -p short,batch --account=<YOUR_WCID> make NP=20
+```
+
+To use the `ctest-s-local-test-driver.sh` script, one must set one's WCID
+account using:
+
+```
+$ export ATDM_CONFIG_WCID_ACCOUNT=<YOUR_WCID>
+```
+
+If `ATDM_CONFIG_WCID_ACCOUNT` is not set, then a default account will be used.
+(But if the user is not approved for that account, then the allocation will
+fail.)
+
+**NOTES:**
+- To get information on <YOUR_WCID> used above, there is a WC tool tab on
+  computing.sandia.gov.
+- CTest runs everything using the `mpirun` command and this must be run from
+  inside a `salloc` or `sbatch` allocation and can **not** be directly
+  launched from a compute node.  For example, one cannot get an interactive
+  shell directly on a compute node using `srun ... bash' an` then run `mpirun`
+  from there.
+
+
 ## Building and installing Trilinos for ATDM Applications
 
 See the following internal SNL wiki page for instructions on building and

cmake/std/atdm/load-env.sh

@@ -56,7 +56,7 @@ fi
 source ${ATDM_CONFIG_SCRIPT_DIR}/utils/get_system_info.sh
 
 if [[ $ATDM_CONFIG_SYSTEM_NAME == "" ]] ; then
-  echo "Error, could not determine a system configuration, aborting env loading script!"
+  echo "Error, could not determine a system configuration for hostname='$realHostname', aborting env loading script!"
   return
 fi
 

cmake/std/atdm/utils/get_known_system_info.sh

@@ -31,6 +31,17 @@ fi
 source ${ATDM_CONFIG_SCRIPT_DIR}/utils/get_system_info_utils.sh
 
 realHostname=`hostname`
+if [[ "${ATDM_CONFIG_GET_KNOW_SYSTEM_INFO_REAL_HOSTNAME_OVERRIDE_FOR_UNIT_TESTING}" ]] ; then
+  echo
+  echo "***"
+  echo "*** WARNING: realHostname=$realHostname overriden to value of"
+  echo "*** ATDM_CONFIG_GET_KNOW_SYSTEM_INFO_REAL_HOSTNAME_OVERRIDE_FOR_UNIT_TESTING='${ATDM_CONFIG_GET_KNOW_SYSTEM_INFO_REAL_HOSTNAME_OVERRIDE_FOR_UNIT_TESTING}'"
+  echo "*** in <trilinos-dir>/cmake/std/atdm/utils/get_known_system_info.sh."
+  echo "*** This variable should only be set for unit testing purposes!"
+  echo "***"
+  echo
+  realHostname=${ATDM_CONFIG_GET_KNOW_SYSTEM_INFO_REAL_HOSTNAME_OVERRIDE_FOR_UNIT_TESTING}
+fi
 #echo "Hostname = '$realHostname'"
 
 #
@@ -52,6 +63,7 @@ ATDM_KNOWN_SYSTEM_NAMES_LIST=(
   mutrino   # Will be repalced by 'ats1'
   waterman
   ats2
+  van1-tx2
   cts1
   tlcc2
   sems-rhel7
@@ -124,10 +136,10 @@ fi
 # matching system type will be selected.
 #
 
-# TLCC2 systems
-if [[ $SNLSYSTEM == "tlcc2"* ]] ; then
-  systemNameTypeMatchedList+=(tlcc2)
-  systemNameTypeMatchedListHostNames[tlcc2]=$SNLCLUSTER
+# ASTRA/Van1-Tx2 systems
+if [[ $SNLSYSTEM == "astra"* ]] ; then
+  systemNameTypeMatchedList+=(van1-tx2)
+  systemNameTypeMatchedListHostNames[van1-tx2]=$SNLCLUSTER
 fi
 
 # CTS1 systems
@@ -136,13 +148,22 @@ if [[ $SNLSYSTEM == "cts1" ]] ; then
   systemNameTypeMatchedListHostNames[cts1]=$SNLCLUSTER
 fi
 
+# TLCC2 systems
+if [[ $SNLSYSTEM == "tlcc2"* ]] ; then
+  systemNameTypeMatchedList+=(tlcc2)
+  systemNameTypeMatchedListHostNames[tlcc2]=$SNLCLUSTER
+fi
+
 # SEMS RHEL6 and RHEL7 systems
 if [[ "${SEMS_PLATFORM}" == "rhel6-x86_64" ]] ; then
   systemNameTypeMatchedList+=(sems-rhel6)
   systemNameTypeMatchedListHostNames[sems-rhel6]=sems-rhel6
 elif [[ "${SEMS_PLATFORM}" == "rhel7-x86_64" ]] ; then
   systemNameTypeMatchedList+=(sems-rhel7)
   systemNameTypeMatchedListHostNames[sems-rhel7]=sems-rhel7
+elif [[ "${SNLSYSTEM}" == "astra" ]] ; then
+  echo "Don't call get-platform on 'astra' systems" > /dev/null
+  # Above logic avoids an 'ERROR: Unrecognized cluster <name>' on these systems
 elif [[ -f /projects/sems/modulefiles/utils/get-platform ]] ; then
   ATDM_SYSTEM_NAME=`source /projects/sems/modulefiles/utils/get-platform`
   if [[ $ATDM_SYSTEM_NAME == "rhel6-x86_64" ]] ; then
@@ -210,7 +231,7 @@ fi
 #
 
 if [[ $ATDM_SYSTEM_NAME != "" ]] ; then
-  echo "Hostname '$realHostname' matches known ATDM host '$ATDM_HOSTNAME' and system '$ATDM_SYSTEM_NAME'"
+  echo "Hostname '$realHostname' matches known ATDM host '$realHostname' and system '$ATDM_SYSTEM_NAME'"
   export ATDM_CONFIG_REAL_HOSTNAME=$realHostname
   export ATDM_CONFIG_CDASH_HOSTNAME=$ATDM_HOSTNAME
   export ATDM_CONFIG_SYSTEM_NAME=$ATDM_SYSTEM_NAME

cmake/std/atdm/utils/set_build_options.sh

@@ -199,6 +199,8 @@ elif [[ $ATDM_CONFIG_BUILD_NAME == *"-Volta72"* ]]; then
   export ATDM_CONFIG_KOKKOS_ARCH=Volta72
 elif [[ $ATDM_CONFIG_BUILD_NAME == *"-WSM"* ]]; then
   export ATDM_CONFIG_KOKKOS_ARCH=WSM
+elif [[ $ATDM_CONFIG_BUILD_NAME == *"-TX2"* ]]; then
+  export ATDM_CONFIG_KOKKOS_ARCH=ARMv8-TX2
 else
   export ATDM_CONFIG_KOKKOS_ARCH=DEFAULT
   if [[ $ATDM_CONFIG_VERBOSE == "1" ]] ; then

cmake/std/atdm/utils/unset_atdm_config_vars_environment.sh

@@ -41,6 +41,7 @@ unset ATDM_CONFIG_SUPERLUDIST_INCLUDE_DIRS
 unset ATDM_CONFIG_SUPERLUDIST_LIBS
 unset ATDM_CONFIG_METIS_LIBS
 unset ATDM_CONFIG_PARMETIS_LIBS
+unset ATDM_CONFIG_CGNS_LIBRARY_NAMES
 unset ATDM_CONFIG_CGNS_LIBS
 unset ATDM_CONFIG_MPI_EXEC
 unset ATDM_CONFIG_MPI_PRE_FLAGS

cmake/std/atdm/van1-tx2/all_supported_builds.sh

@@ -0,0 +1,8 @@
+# This script is sourced to return all of the supported builds
+
+export ATDM_CONFIG_CTEST_S_BUILD_NAME_PREFIX=Trilinos-atdm-
+
+export ATDM_CONFIG_ALL_SUPPORTED_BUILDS=(
+  van1-tx2_arm-20.0_openmpi-4.0.2_openmp_static_opt
+  van1-tx2_arm-20.0_openmpi-4.0.2_openmp_static_dbg
+  )

cmake/std/atdm/van1-tx2/custom_builds.sh

@@ -0,0 +1,24 @@
+#
+# Custom builds for van1-tx2 env
+#
+
+if   [[ $ATDM_CONFIG_BUILD_NAME == *"arm-20.0-openmpi-4.0.2"* ]] \
+  || [[ $ATDM_CONFIG_BUILD_NAME == *"arm-20.0_openmpi-4.0.2"* ]] \
+  || [[ $ATDM_CONFIG_BUILD_NAME == *"arm-20.0"* ]] \
+  || [[ $ATDM_CONFIG_BUILD_NAME == *"arm-20"* ]] \
+  || [[ $ATDM_CONFIG_BUILD_NAME == *"default" ]] \
+  ; then
+  export ATDM_CONFIG_COMPILER=ARM-20.0_OPENMPI-4.0.2
+
+else
+  echo
+  echo "***"
+  echo "*** ERROR: A supported compiler was not selected for 'van1-tx2' (stria) env"
+  echo "***"
+  echo "*** Supported compilers include:"
+  echo "***"
+  echo "****  arm-20.0-openmpi-4.0.2    (arm-20.0, default)"
+  echo "***"
+  return
+
+fi