path: root/run_mrt.sh

#!/bin/bash

# Marian regression test script. Invocation examples:
#  ./run_mrt.sh
#  ./run_mrt.sh tests/training/basics
#  ./run_mrt.sh tests/training/basics/test_valid_script.sh
#  ./run_mrt.sh previous.log
# where previous.log contains a list of test files in separate lines.

# Environment variables:
#  - MARIAN - path to Marian build directory
#  - CUDA_VISIBLE_DEVICES - CUDA's variable specifying GPU device IDs
#  - NUM_DEVICES - maximum number of GPU devices to be used

SHELL=/bin/bash

export LC_ALL=C.UTF-8

function log {
    echo [$(date "+%m/%d/%Y %T")] $@
}

function logn {
    echo -n [$(date "+%m/%d/%Y %T")] $@
}


export MRT_ROOT="$( cd "$( dirname "${BASH_SOURCE[0]}" )" && pwd )"
export MRT_TOOLS=$MRT_ROOT/tools
export MRT_MARIAN="$( realpath ${MARIAN:-../build} )"
export MRT_MODELS=$MRT_ROOT/models
export MRT_DATA=$MRT_ROOT/data

# Try adding build/ to MARIAN for backward compatibility
if [[ ! -e $MRT_MARIAN/marian-decoder ]]; then
    MRT_MARIAN="$MRT_MARIAN/build"
fi

log "Using Marian: $MRT_MARIAN"

# Check if required tools are present in marian directory
for cmd in marian marian-decoder marian-scorer marian-server marian-vocab; do
    if [ ! -e $MRT_MARIAN/$cmd ]; then
        echo "Error: '$cmd' is not installed in '$MRT_MARIAN', you need to compile the toolkit first"
        exit 1
    fi
done

# Get CMake settings
cd $MRT_MARIAN
cmake -L 2> /dev/null > $MRT_ROOT/cmake.log
cd $MRT_ROOT

# Check Marian compilation settings
export MRT_MARIAN_VERSION=$($MRT_MARIAN/marian --version 2>&1)
export MRT_MARIAN_USE_MKL=$(cat $MRT_ROOT/cmake.log | grep -P "MKL_ROOT" | grep -vP "MKL_ROOT.*NOTFOUND|USE_CUDNN:BOOL=(OFF|off|0)")
export MRT_MARIAN_USE_CUDNN=$(cat $MRT_ROOT/cmake.log | grep -P "USE_CUDNN:BOOL=(ON|on|1)")

log "Using version: $MRT_MARIAN_VERSION"
log "Using MKL: $MRT_MARIAN_USE_MKL"
log "Using CUDNN: $MRT_MARIAN_USE_CUDNN"

# Number of available devices
cuda_num_devices=$(($(echo $CUDA_VISIBLE_DEVICES | grep -c ',')+1))
export MRT_NUM_DEVICES=${NUM_DEVICES:-$cuda_num_devices}

log "Using CUDA visible devices: $CUDA_VISIBLE_DEVICES"
log "Using number of devices: $MRT_NUM_DEVICES"

# Exit codes
export EXIT_CODE_SUCCESS=0
export EXIT_CODE_SKIP=100

function format_time {
    dt=$(echo "$2 - $1" | bc 2>/dev/null)
    dh=$(echo "$dt/3600" | bc 2>/dev/null)
    dt2=$(echo "$dt-3600*$dh" | bc 2>/dev/null)
    dm=$(echo "$dt2/60" | bc 2>/dev/null)
    ds=$(echo "$dt2-60*$dm" | bc 2>/dev/null)
    LANG=C printf "%02d:%02d:%02.3fs" $dh $dm $ds
}


# Default directory with all regression tests
prefix=tests
if [ $# -ge 1 ]; then
    if [[ "$1" = *.log ]]; then
        # Extract test names from .log file
        prefix=$(cat $1 | grep '/test_.*\.sh' | grep -v '/_' | sed 's/^ *- *//' | tr '\n' ' ' | sed 's/ *$//')
    else
        prefix="$@"
    fi
fi

success=true
count_passed=0
count_skipped=0
count_failed=0
count_all=0

declare -a tests_skipped
declare -a tests_failed

test_dirs=$(find $prefix -type d | grep -v "/_")

if grep -q "/test_.*\.sh\$" <<< "$prefix"; then
    test_single_files=$(printf '%s\n' $prefix | sed 's!*/!!')
    test_dirs=$(dirname $prefix | sort | uniq)
fi

time_start=$(date +%s.%N)

# Traverse test directories
cd $MRT_ROOT
for test_dir in $test_dirs
do
    log "Checking directory: $test_dir"
    nosetup=false

    # Run setup script if exists
    if [ -e $test_dir/setup.sh ]; then
        log "Running setup script"

        cd $test_dir
        $SHELL setup.sh &> setup.log
        if [ $? -ne 0 ]; then
            log "Warning: setup script returns a non-success exit code"
            success=false
            nosetup=true
        else
            rm setup.log
        fi
        cd $MRT_ROOT
    fi

    # Run tests
    for test_path in $(ls -A $test_dir/test_*.sh 2>/dev/null)
    do
        test_file=$(basename $test_path)
        test_name="${test_file%.*}"

        # In non-traverse mode skip tests if not requested
        if [[ -n "$test_single_files" && $test_single_files != *"$test_file"* ]]; then
            continue
        fi
        test_time_start=$(date +%s.%N)
        ((++count_all))

        # Tests are executed from their directory
        cd $test_dir

        # Skip tests if setup failed
        logn "Running $test_path ... "
        if [ "$nosetup" = true ]; then
            ((++count_skipped))
            tests_skipped+=($test_path)
            echo " skipped"
            cd $MRT_ROOT
            continue;
        fi

        # Run test
        # Note: all output gets written to stderr (very very few cases write to stdout)
        $SHELL -x $test_file 2> $test_file.log 1>&2
        exit_code=$?

        # Check exit code
        if [ $exit_code -eq $EXIT_CODE_SUCCESS ]; then
            ((++count_passed))
            echo " OK"
        elif [ $exit_code -eq $EXIT_CODE_SKIP ]; then
            ((++count_skipped))
            tests_skipped+=($test_path)
            echo " skipped"
        else
            ((++count_failed))
            tests_failed+=($test_path)
            echo " failed"
            success=false
        fi

        # Report time
        test_time_end=$(date +%s.%N)
        test_time=$(format_time $test_time_start $test_time_end)
        log "Test took $test_time"

        cd $MRT_ROOT
    done
    cd $MRT_ROOT

    # Run teardown script if exists
    if [ -e $test_dir/teardown.sh ]; then
        log "Running teardown script"

        cd $test_dir
        $SHELL teardown.sh &> teardown.log
        if [ $? -ne 0 ]; then
            log "Warning: teardown script returns a non-success exit code"
            success=false
        else
            rm teardown.log
        fi
        cd $MRT_ROOT
    fi
done

time_end=$(date +%s.%N)
time_total=$(format_time $time_start $time_end)

prev_log=previous.log
rm -f $prev_log

# Print skipped and failed tests
if [ -n "$tests_skipped" ] || [ -n "$tests_failed" ]; then
    echo "---------------------"
fi
[[ -z "$tests_skipped" ]] || echo "Skipped:" | tee -a $prev_log
for test_name in "${tests_skipped[@]}"; do
    echo "  - $test_name" | tee -a $prev_log
done
[[ -z "$tests_failed" ]] || echo "Failed:" | tee -a $prev_log
for test_name in "${tests_failed[@]}"; do
    echo "  - $test_name" | tee -a $prev_log
done
[[ -z "$tests_failed" ]] || echo "Logs:"
for test_name in "${tests_failed[@]}"; do
    echo "  - $(realpath $test_name | sed 's/\.sh/.sh.log/')"
done

# Print summary
echo "---------------------" | tee -a $prev_log
echo "Ran $count_all tests in $time_total, $count_passed passed, $count_skipped skipped, $count_failed failed" | tee -a $prev_log

# Return exit code
$success && [ $count_all -gt 0 ]