Page MenuHomec4science

test_all_sandia
No OneTemporary

File Metadata

Created
Thu, Aug 15, 01:53

test_all_sandia

#!/bin/bash -e
#
# Global config
#
set -o pipefail
# Determine current machine
MACHINE=""
HOSTNAME=$(hostname)
if [[ "$HOSTNAME" =~ (white|ride).* ]]; then
MACHINE=white
elif [[ "$HOSTNAME" =~ .*bowman.* ]]; then
MACHINE=bowman
elif [[ "$HOSTNAME" =~ node.* ]]; then # Warning: very generic name
MACHINE=shepard
elif [ ! -z "$SEMS_MODULEFILES_ROOT" ]; then
MACHINE=sems
else
echo "Unrecognized machine" >&2
exit 1
fi
GCC_BUILD_LIST="OpenMP,Pthread,Serial,OpenMP_Serial,Pthread_Serial"
IBM_BUILD_LIST="OpenMP,Serial,OpenMP_Serial"
INTEL_BUILD_LIST="OpenMP,Pthread,Serial,OpenMP_Serial,Pthread_Serial"
CLANG_BUILD_LIST="Pthread,Serial,Pthread_Serial"
CUDA_BUILD_LIST="Cuda_OpenMP,Cuda_Pthread,Cuda_Serial"
GCC_WARNING_FLAGS="-Wall,-Wshadow,-pedantic,-Werror,-Wsign-compare,-Wtype-limits,-Wignored-qualifiers,-Wempty-body,-Wclobbered,-Wuninitialized"
IBM_WARNING_FLAGS="-Wall,-Wshadow,-pedantic,-Werror,-Wsign-compare,-Wtype-limits,-Wuninitialized"
CLANG_WARNING_FLAGS="-Wall,-Wshadow,-pedantic,-Werror,-Wsign-compare,-Wtype-limits,-Wuninitialized"
INTEL_WARNING_FLAGS="-Wall,-Wshadow,-pedantic,-Werror,-Wsign-compare,-Wtype-limits,-Wuninitialized"
CUDA_WARNING_FLAGS=""
# Default. Machine specific can override
DEBUG=False
ARGS=""
CUSTOM_BUILD_LIST=""
DRYRUN=False
BUILD_ONLY=False
declare -i NUM_JOBS_TO_RUN_IN_PARALLEL=3
TEST_SCRIPT=False
SKIP_HWLOC=False
ARCH_FLAG=""
#
# Machine specific config
#
if [ "$MACHINE" = "sems" ]; then
source /projects/modulefiles/utils/sems-modules-init.sh
source /projects/modulefiles/utils/kokkos-modules-init.sh
BASE_MODULE_LIST="<COMPILER_NAME>/<COMPILER_VERSION>/base,hwloc/1.10.1/<COMPILER_NAME>/<COMPILER_VERSION>/base"
CUDA_MODULE_LIST="<COMPILER_NAME>/<COMPILER_VERSION>,gcc/4.7.2/base"
# Format: (compiler module-list build-list exe-name warning-flag)
COMPILERS=("gcc/4.7.2 $BASE_MODULE_LIST $GCC_BUILD_LIST g++ $GCC_WARNING_FLAGS"
"gcc/4.8.4 $BASE_MODULE_LIST $GCC_BUILD_LIST g++ $GCC_WARNING_FLAGS"
"gcc/4.9.2 $BASE_MODULE_LIST $GCC_BUILD_LIST g++ $GCC_WARNING_FLAGS"
"gcc/5.1.0 $BASE_MODULE_LIST $GCC_BUILD_LIST g++ $GCC_WARNING_FLAGS"
"intel/14.0.4 $BASE_MODULE_LIST $INTEL_BUILD_LIST icpc $INTEL_WARNING_FLAGS"
"intel/15.0.2 $BASE_MODULE_LIST $INTEL_BUILD_LIST icpc $INTEL_WARNING_FLAGS"
"intel/16.0.1 $BASE_MODULE_LIST $INTEL_BUILD_LIST icpc $INTEL_WARNING_FLAGS"
"clang/3.5.2 $BASE_MODULE_LIST $CLANG_BUILD_LIST clang++ $CLANG_WARNING_FLAGS"
"clang/3.6.1 $BASE_MODULE_LIST $CLANG_BUILD_LIST clang++ $CLANG_WARNING_FLAGS"
"cuda/6.5.14 $CUDA_MODULE_LIST $CUDA_BUILD_LIST $KOKKOS_PATH/config/nvcc_wrapper $CUDA_WARNING_FLAGS"
"cuda/7.0.28 $CUDA_MODULE_LIST $CUDA_BUILD_LIST $KOKKOS_PATH/config/nvcc_wrapper $CUDA_WARNING_FLAGS"
"cuda/7.5.18 $CUDA_MODULE_LIST $CUDA_BUILD_LIST $KOKKOS_PATH/config/nvcc_wrapper $CUDA_WARNING_FLAGS"
)
elif [ "$MACHINE" = "white" ]; then
source /etc/profile.d/modules.sh
SKIP_HWLOC=True
export SLURM_TASKS_PER_NODE=32
BASE_MODULE_LIST="<COMPILER_NAME>/<COMPILER_VERSION>"
IBM_MODULE_LIST="<COMPILER_NAME>/xl/<COMPILER_VERSION>"
CUDA_MODULE_LIST="<COMPILER_NAME>/<COMPILER_VERSION>,gcc/4.9.2"
# Don't do pthread on white
GCC_BUILD_LIST="OpenMP,Serial,OpenMP_Serial"
# Format: (compiler module-list build-list exe-name warning-flag)
COMPILERS=("gcc/4.9.2 $BASE_MODULE_LIST $IBM_BUILD_LIST g++ $GCC_WARNING_FLAGS"
"gcc/5.3.0 $BASE_MODULE_LIST $IBM_BUILD_LIST g++ $GCC_WARNING_FLAGS"
"ibm/13.1.3 $IBM_MODULE_LIST $IBM_BUILD_LIST xlC $IBM_WARNING_FLAGS"
)
ARCH_FLAG="--arch=Power8"
NUM_JOBS_TO_RUN_IN_PARALLEL=8
elif [ "$MACHINE" = "bowman" ]; then
source /etc/profile.d/modules.sh
SKIP_HWLOC=True
export SLURM_TASKS_PER_NODE=32
BASE_MODULE_LIST="<COMPILER_NAME>/compilers/<COMPILER_VERSION>"
OLD_INTEL_BUILD_LIST="Pthread,Serial,Pthread_Serial"
# Format: (compiler module-list build-list exe-name warning-flag)
COMPILERS=("intel/16.2.181 $BASE_MODULE_LIST $OLD_INTEL_BUILD_LIST icpc $INTEL_WARNING_FLAGS"
"intel/17.0.064 $BASE_MODULE_LIST $INTEL_BUILD_LIST icpc $INTEL_WARNING_FLAGS"
)
ARCH_FLAG="--arch=KNL"
NUM_JOBS_TO_RUN_IN_PARALLEL=8
elif [ "$MACHINE" = "shepard" ]; then
source /etc/profile.d/modules.sh
SKIP_HWLOC=True
export SLURM_TASKS_PER_NODE=32
BASE_MODULE_LIST="<COMPILER_NAME>/compilers/<COMPILER_VERSION>"
OLD_INTEL_BUILD_LIST="Pthread,Serial,Pthread_Serial"
# Format: (compiler module-list build-list exe-name warning-flag)
COMPILERS=("intel/16.2.181 $BASE_MODULE_LIST $OLD_INTEL_BUILD_LIST icpc $INTEL_WARNING_FLAGS"
"intel/17.0.064 $BASE_MODULE_LIST $INTEL_BUILD_LIST icpc $INTEL_WARNING_FLAGS"
)
ARCH_FLAG="--arch=HSW"
NUM_JOBS_TO_RUN_IN_PARALLEL=8
else
echo "Unhandled machine $MACHINE" >&2
exit 1
fi
export OMP_NUM_THREADS=4
declare -i NUM_RESULTS_TO_KEEP=7
RESULT_ROOT_PREFIX=TestAll
SCRIPT_KOKKOS_ROOT=$( cd "$( dirname "$0" )" && cd .. && pwd )
#
# Handle arguments
#
while [[ $# > 0 ]]
do
key="$1"
case $key in
--kokkos-path*)
KOKKOS_PATH="${key#*=}"
;;
--build-list*)
CUSTOM_BUILD_LIST="${key#*=}"
;;
--debug*)
DEBUG=True
;;
--build-only*)
BUILD_ONLY=True
;;
--test-script*)
TEST_SCRIPT=True
;;
--skip-hwloc*)
SKIP_HWLOC=True
;;
--num*)
NUM_JOBS_TO_RUN_IN_PARALLEL="${key#*=}"
;;
--dry-run*)
DRYRUN=True
;;
--help)
echo "test_all_sandia <ARGS> <OPTIONS>:"
echo "--kokkos-path=/Path/To/Kokkos: Path to the Kokkos root directory"
echo " Defaults to root repo containing this script"
echo "--debug: Run tests in debug. Defaults to False"
echo "--test-script: Test this script, not Kokkos"
echo "--skip-hwloc: Do not do hwloc tests"
echo "--num=N: Number of jobs to run in parallel "
echo "--dry-run: Just print what would be executed"
echo "--build-only: Just do builds, don't run anything"
echo "--build-list=BUILD,BUILD,BUILD..."
echo " Provide a comma-separated list of builds instead of running all builds"
echo " Valid items:"
echo " OpenMP, Pthread, Serial, OpenMP_Serial, Pthread_Serial"
echo " Cuda_OpenMP, Cuda_Pthread, Cuda_Serial"
echo ""
echo "ARGS: list of expressions matching compilers to test"
echo " supported compilers sems"
for COMPILER_DATA in "${COMPILERS[@]}"; do
ARR=($COMPILER_DATA)
COMPILER=${ARR[0]}
echo " $COMPILER"
done
echo ""
echo "Examples:"
echo " Run all tests"
echo " % test_all_sandia"
echo ""
echo " Run all gcc tests"
echo " % test_all_sandia gcc"
echo ""
echo " Run all gcc/4.7.2 and all intel tests"
echo " % test_all_sandia gcc/4.7.2 intel"
echo ""
echo " Run all tests in debug"
echo " % test_all_sandia --debug"
echo ""
echo " Run gcc/4.7.2 and only do OpenMP and OpenMP_Serial builds"
echo " % test_all_sandia gcc/4.7.2 --build-list=OpenMP,OpenMP_Serial"
echo ""
echo "If you want to kill the tests, do:"
echo " hit ctrl-z"
echo " % kill -9 %1"
echo
exit 0
;;
*)
# args, just append
ARGS="$ARGS $1"
;;
esac
shift
done
# set kokkos path
if [ -z "$KOKKOS_PATH" ]; then
KOKKOS_PATH=$SCRIPT_KOKKOS_ROOT
else
# Ensure KOKKOS_PATH is abs path
KOKKOS_PATH=$( cd $KOKKOS_PATH && pwd )
fi
# set build type
if [ "$DEBUG" = "True" ]; then
BUILD_TYPE=debug
else
BUILD_TYPE=release
fi
# If no args provided, do all compilers
if [ -z "$ARGS" ]; then
ARGS='?'
fi
# Process args to figure out which compilers to test
COMPILERS_TO_TEST=""
for ARG in $ARGS; do
for COMPILER_DATA in "${COMPILERS[@]}"; do
ARR=($COMPILER_DATA)
COMPILER=${ARR[0]}
if [[ "$COMPILER" = $ARG* ]]; then
if [[ "$COMPILERS_TO_TEST" != *${COMPILER}* ]]; then
COMPILERS_TO_TEST="$COMPILERS_TO_TEST $COMPILER"
else
echo "Tried to add $COMPILER twice"
fi
fi
done
done
#
# Functions
#
# get_compiler_name <COMPILER>
get_compiler_name() {
echo $1 | cut -d/ -f1
}
# get_compiler_version <COMPILER>
get_compiler_version() {
echo $1 | cut -d/ -f2
}
# Do not call directly
get_compiler_data() {
local compiler=$1
local item=$2
local compiler_name=$(get_compiler_name $compiler)
local compiler_vers=$(get_compiler_version $compiler)
local compiler_data
for compiler_data in "${COMPILERS[@]}" ; do
local arr=($compiler_data)
if [ "$compiler" = "${arr[0]}" ]; then
echo "${arr[$item]}" | tr , ' ' | sed -e "s/<COMPILER_NAME>/$compiler_name/g" -e "s/<COMPILER_VERSION>/$compiler_vers/g"
return 0
fi
done
# Not found
echo "Unreconized compiler $compiler" >&2
exit 1
}
#
# For all getters, usage: <GETTER> <COMPILER>
#
get_compiler_modules() {
get_compiler_data $1 1
}
get_compiler_build_list() {
get_compiler_data $1 2
}
get_compiler_exe_name() {
get_compiler_data $1 3
}
get_compiler_warning_flags() {
get_compiler_data $1 4
}
run_cmd() {
echo "RUNNING: $*"
if [ "$DRYRUN" != "True" ]; then
eval "$* 2>&1"
fi
}
# report_and_log_test_results <SUCCESS> <DESC> <COMMENT>
report_and_log_test_result() {
# Use sane var names
local success=$1; local desc=$2; local comment=$3;
if [ "$success" = "0" ]; then
echo " PASSED $desc"
echo $comment > $PASSED_DIR/$desc
else
# For failures, comment should be the name of the phase that failed
echo " FAILED $desc" >&2
echo $comment > $FAILED_DIR/$desc
cat ${desc}.${comment}.log
fi
}
setup_env() {
local compiler=$1
local compiler_modules=$(get_compiler_modules $compiler)
module purge
local mod
for mod in $compiler_modules; do
echo "Loading module $mod"
module load $mod 2>&1
# It is ridiculously hard to check for the success of a loaded
# module. Module does not return error codes and piping to grep
# causes module to run in a subshell.
module list 2>&1 | grep "$mod" >& /dev/null || return 1
done
return 0
}
# single_build_and_test <COMPILER> <BUILD> <BUILD_TYPE>
single_build_and_test() {
# Use sane var names
local compiler=$1; local build=$2; local build_type=$3;
# set up env
mkdir -p $ROOT_DIR/$compiler/"${build}-$build_type"
cd $ROOT_DIR/$compiler/"${build}-$build_type"
local desc=$(echo "${compiler}-${build}-${build_type}" | sed 's:/:-:g')
setup_env $compiler >& ${desc}.configure.log || { report_and_log_test_result 1 ${desc} configure && return 0; }
# Set up flags
local compiler_warning_flags=$(get_compiler_warning_flags $compiler)
local compiler_exe=$(get_compiler_exe_name $compiler)
if [[ "$build_type" = hwloc* ]]; then
local extra_args=--with-hwloc=$(dirname $(dirname $(which hwloc-info)))
fi
if [[ "$build_type" = *debug* ]]; then
local extra_args="$extra_args --debug"
local cxxflags="-g $compiler_warning_flags"
else
local cxxflags="-O3 $compiler_warning_flags"
fi
if [[ "$compiler" == cuda* ]]; then
cxxflags="--keep --keep-dir=$(pwd) $cxxflags"
export TMPDIR=$(pwd)
fi
# cxxflags="-DKOKKOS_USING_EXP_VIEW=1 $cxxflags"
echo " Starting job $desc"
local comment="no_comment"
if [ "$TEST_SCRIPT" = "True" ]; then
local rand=$[ 1 + $[ RANDOM % 10 ]]
sleep $rand
if [ $rand -gt 5 ]; then
run_cmd ls fake_problem >& ${desc}.configure.log || { report_and_log_test_result 1 $desc configure && return 0; }
fi
else
run_cmd ${KOKKOS_PATH}/generate_makefile.bash --with-devices=$build $ARCH_FLAG --compiler=$(which $compiler_exe) --cxxflags=\"$cxxflags\" $extra_args &>> ${desc}.configure.log || { report_and_log_test_result 1 ${desc} configure && return 0; }
local -i build_start_time=$(date +%s)
run_cmd make build-test >& ${desc}.build.log || { report_and_log_test_result 1 ${desc} build && return 0; }
local -i build_end_time=$(date +%s)
comment="build_time=$(($build_end_time-$build_start_time))"
if [[ "$BUILD_ONLY" == False ]]; then
run_cmd make test >& ${desc}.test.log || { report_and_log_test_result 1 ${desc} test && return 0; }
local -i run_end_time=$(date +%s)
comment="$comment run_time=$(($run_end_time-$build_end_time))"
fi
fi
report_and_log_test_result 0 $desc "$comment"
return 0
}
# wait_for_jobs <NUM-JOBS>
wait_for_jobs() {
local -i max_jobs=$1
local -i num_active_jobs=$(jobs | wc -l)
while [ $num_active_jobs -ge $max_jobs ]
do
sleep 1
num_active_jobs=$(jobs | wc -l)
jobs >& /dev/null
done
}
# run_in_background <COMPILER> <BUILD> <BUILD_TYPE>
run_in_background() {
local compiler=$1
local -i num_jobs=$NUM_JOBS_TO_RUN_IN_PARALLEL
if [[ "$BUILD_ONLY" == True ]]; then
num_jobs=8
else
if [[ "$compiler" == cuda* ]]; then
num_jobs=1
fi
fi
wait_for_jobs $num_jobs
single_build_and_test $* &
}
# build_and_test_all <COMPILER>
build_and_test_all() {
# Get compiler data
local compiler=$1
if [ -z "$CUSTOM_BUILD_LIST" ]; then
local compiler_build_list=$(get_compiler_build_list $compiler)
else
local compiler_build_list=$(echo "$CUSTOM_BUILD_LIST" | tr , ' ')
fi
# do builds
local build
for build in $compiler_build_list
do
run_in_background $compiler $build $BUILD_TYPE
# If not cuda, do a hwloc test too
if [[ "$compiler" != cuda* && "$SKIP_HWLOC" == False ]]; then
run_in_background $compiler $build "hwloc-$BUILD_TYPE"
fi
done
return 0
}
get_test_root_dir() {
local existing_results=$(find . -maxdepth 1 -name "$RESULT_ROOT_PREFIX*" | sort)
local -i num_existing_results=$(echo $existing_results | tr ' ' '\n' | wc -l)
local -i num_to_delete=${num_existing_results}-${NUM_RESULTS_TO_KEEP}
if [ $num_to_delete -gt 0 ]; then
/bin/rm -rf $(echo $existing_results | tr ' ' '\n' | head -n $num_to_delete)
fi
echo $(pwd)/${RESULT_ROOT_PREFIX}_$(date +"%Y-%m-%d_%H.%M.%S")
}
wait_summarize_and_exit() {
wait_for_jobs 1
echo "#######################################################"
echo "PASSED TESTS"
echo "#######################################################"
local passed_test
for passed_test in $(\ls -1 $PASSED_DIR | sort)
do
echo $passed_test $(cat $PASSED_DIR/$passed_test)
done
echo "#######################################################"
echo "FAILED TESTS"
echo "#######################################################"
local failed_test
local -i rv=0
for failed_test in $(\ls -1 $FAILED_DIR | sort)
do
echo $failed_test "("$(cat $FAILED_DIR/$failed_test)" failed)"
rv=$rv+1
done
exit $rv
}
#
# Main
#
ROOT_DIR=$(get_test_root_dir)
mkdir -p $ROOT_DIR
cd $ROOT_DIR
PASSED_DIR=$ROOT_DIR/results/passed
FAILED_DIR=$ROOT_DIR/results/failed
mkdir -p $PASSED_DIR
mkdir -p $FAILED_DIR
echo "Going to test compilers: " $COMPILERS_TO_TEST
for COMPILER in $COMPILERS_TO_TEST; do
echo "Testing compiler $COMPILER"
build_and_test_all $COMPILER
done
wait_summarize_and_exit

Event Timeline