How to use summit - 8 common examples

cwd = os.getcwd() + '/'
bin_dir = cwd + 'Bin/'
bin_name = executable_name(compiler, architecture)

log_dir  = cwd
perf_database_file = cwd + perf_database_file
day = time.strftime('%d')
month = time.strftime('%m')
year = time.strftime('%Y')

# Initialize tests
# ----------------
if args.mode == 'run':
    start_date = datetime.datetime.now().strftime("%Y_%m_%d_%H_%M_%S")
    # Set default options for compilation and execution
    config_command = get_config_command(compiler, architecture)
    # Create main result directory if does not exist
    if not os.path.exists(res_dir_base):
        os.mkdir(res_dir_base)

    # Recompile if requested
    # ----------------------
    if recompile == True:
        if pull_3_repos == True:
            git_repo = git.cmd.Git( picsar_dir )
            git_repo.pull()
            git_repo = git.cmd.Git( amrex_dir  )
            git_repo.pull()
            git_repo = git.cmd.Git( warpx_dir  )
            git_repo.pull()

        # Copy WarpX/GNUmakefile to current directory and recompile

# ------------------
source_dir_base = os.environ['AUTOMATED_PERF_TESTS']
warpx_dir = source_dir_base + '/warpx/'
picsar_dir = source_dir_base + '/picsar/'
amrex_dir = source_dir_base + '/amrex/'
res_dir_base = os.environ['SCRATCH'] + '/performance_warpx/'
perf_logs_repo = source_dir_base + 'perf_logs/'

# Define dictionaries
# -------------------
compiler_name = {'intel': 'intel', 'gnu': 'gcc', 'pgi':'pgi'}
module_Cname = {'cpu': 'haswell', 'knl': 'knl,quad,cache', 'gpu':''}
csv_file = {'cori':'cori_knl.csv', 'summit':'summit.csv'}
cwd = os.getcwd() + '/'
bin_dir = cwd + 'Bin/'
bin_name = executable_name(compiler, architecture)

log_dir  = cwd
perf_database_file = cwd + perf_database_file
day = time.strftime('%d')
month = time.strftime('%m')
year = time.strftime('%Y')

# Initialize tests
# ----------------
if args.mode == 'run':
    start_date = datetime.datetime.now().strftime("%Y_%m_%d_%H_%M_%S")
    # Set default options for compilation and execution
    config_command = get_config_command(compiler, architecture)
    # Create main result directory if does not exist
    if not os.path.exists(res_dir_base):
        os.mkdir(res_dir_base)

# One batch job submitted per n_node. Several
# tests run within the same batch job.
# --------------------------------------------
if args.mode == 'run':
    if os.path.exists( 'log_jobids_tmp.txt' ):
        os.remove( 'log_jobids_tmp.txt' )
    # loop on n_node. One batch script per n_node
    for n_node in n_node_list:
        res_dir = res_dir_base
        res_dir += '_'.join([run_name, compiler, architecture, str(n_node)]) + '/'
        runtime_param_list = []
        # Deep copy as we change the attribute n_cell of
        # each instance of class test_element
        test_list_n_node = copy.deepcopy(test_list)
        job_time_min = time_min(len(test_list))
        batch_string = get_batch_string(test_list_n_node, job_time_min, module_Cname[architecture], n_node)
        # Loop on tests
        for count, current_run in enumerate(test_list_n_node):
            current_run.scale_n_cell(n_node)
            runtime_param_string  = ' amr.n_cell=' + ' '.join(str(i) for i in current_run.n_cell)
            runtime_param_string += ' amr.max_grid_size=' + str(current_run.max_grid_size)
            runtime_param_string += ' amr.blocking_factor=' + str(current_run.blocking_factor)
            runtime_param_string += ' max_step=' + str( current_run.n_step )
            # runtime_param_list.append( runtime_param_string )
            run_string = get_run_string(current_run, architecture, n_node, count, bin_name, runtime_param_string)
            batch_string += run_string
            batch_string += 'rm -rf plotfiles lab_frame_data diags\n'

        submit_job_command = get_submit_job_command()
        # Run the simulations.
        run_batch_nnode(test_list_n_node, res_dir, bin_name, config_command, batch_string, submit_job_command)
    os.chdir(cwd)

res_dir += '_'.join([run_name, compiler, architecture, str(n_node)]) + '/'
        runtime_param_list = []
        # Deep copy as we change the attribute n_cell of
        # each instance of class test_element
        test_list_n_node = copy.deepcopy(test_list)
        job_time_min = time_min(len(test_list))
        batch_string = get_batch_string(test_list_n_node, job_time_min, module_Cname[architecture], n_node)
        # Loop on tests
        for count, current_run in enumerate(test_list_n_node):
            current_run.scale_n_cell(n_node)
            runtime_param_string  = ' amr.n_cell=' + ' '.join(str(i) for i in current_run.n_cell)
            runtime_param_string += ' amr.max_grid_size=' + str(current_run.max_grid_size)
            runtime_param_string += ' amr.blocking_factor=' + str(current_run.blocking_factor)
            runtime_param_string += ' max_step=' + str( current_run.n_step )
            # runtime_param_list.append( runtime_param_string )
            run_string = get_run_string(current_run, architecture, n_node, count, bin_name, runtime_param_string)
            batch_string += run_string
            batch_string += 'rm -rf plotfiles lab_frame_data diags\n'

        submit_job_command = get_submit_job_command()
        # Run the simulations.
        run_batch_nnode(test_list_n_node, res_dir, bin_name, config_command, batch_string, submit_job_command)
    os.chdir(cwd)
    # submit batch for analysis
    if os.path.exists( 'read_error.txt' ):
        os.remove( 'read_error.txt' )
    if os.path.exists( 'read_output.txt' ):
        os.remove( 'read_output.txt' )
    process_analysis(args.automated, cwd, compiler, architecture, args.n_node_list, start_date)

# read the output file from each test and store timers in
# hdf5 file with pandas format

test_list_n_node = copy.deepcopy(test_list)
        job_time_min = time_min(len(test_list))
        batch_string = get_batch_string(test_list_n_node, job_time_min, module_Cname[architecture], n_node)
        # Loop on tests
        for count, current_run in enumerate(test_list_n_node):
            current_run.scale_n_cell(n_node)
            runtime_param_string  = ' amr.n_cell=' + ' '.join(str(i) for i in current_run.n_cell)
            runtime_param_string += ' amr.max_grid_size=' + str(current_run.max_grid_size)
            runtime_param_string += ' amr.blocking_factor=' + str(current_run.blocking_factor)
            runtime_param_string += ' max_step=' + str( current_run.n_step )
            # runtime_param_list.append( runtime_param_string )
            run_string = get_run_string(current_run, architecture, n_node, count, bin_name, runtime_param_string)
            batch_string += run_string
            batch_string += 'rm -rf plotfiles lab_frame_data diags\n'

        submit_job_command = get_submit_job_command()
        # Run the simulations.
        run_batch_nnode(test_list_n_node, res_dir, bin_name, config_command, batch_string, submit_job_command)
    os.chdir(cwd)
    # submit batch for analysis
    if os.path.exists( 'read_error.txt' ):
        os.remove( 'read_error.txt' )
    if os.path.exists( 'read_output.txt' ):
        os.remove( 'read_output.txt' )
    process_analysis(args.automated, cwd, compiler, architecture, args.n_node_list, start_date)

# read the output file from each test and store timers in
# hdf5 file with pandas format
# -------------------------------------------------------
for n_node in n_node_list:
    print(n_node)
    if browse_output_files:

rename_archive = True
    store_full_input = False
    update_perf_log_repo = True
    push_on_perf_log_repo = False
    pull_3_repos = True
    recompile = True
    if machine == 'summit':
        compiler = 'pgi'
        architecture = 'gpu'

# List of tests to perform
# ------------------------
# Each test runs n_repeat times
n_repeat = 2
# test_list is machine-specific
test_list = get_test_list(n_repeat)

# Define directories
# ------------------
source_dir_base = os.environ['AUTOMATED_PERF_TESTS']
warpx_dir = source_dir_base + '/warpx/'
picsar_dir = source_dir_base + '/picsar/'
amrex_dir = source_dir_base + '/amrex/'
res_dir_base = os.environ['SCRATCH'] + '/performance_warpx/'
perf_logs_repo = source_dir_base + 'perf_logs/'

# Define dictionaries
# -------------------
compiler_name = {'intel': 'intel', 'gnu': 'gcc', 'pgi':'pgi'}
module_Cname = {'cpu': 'haswell', 'knl': 'knl,quad,cache', 'gpu':''}
csv_file = {'cori':'cori_knl.csv', 'summit':'summit.csv'}
cwd = os.getcwd() + '/'

runtime_param_string += ' max_step=' + str( current_run.n_step )
            # runtime_param_list.append( runtime_param_string )
            run_string = get_run_string(current_run, architecture, n_node, count, bin_name, runtime_param_string)
            batch_string += run_string
            batch_string += 'rm -rf plotfiles lab_frame_data diags\n'

        submit_job_command = get_submit_job_command()
        # Run the simulations.
        run_batch_nnode(test_list_n_node, res_dir, bin_name, config_command, batch_string, submit_job_command)
    os.chdir(cwd)
    # submit batch for analysis
    if os.path.exists( 'read_error.txt' ):
        os.remove( 'read_error.txt' )
    if os.path.exists( 'read_output.txt' ):
        os.remove( 'read_output.txt' )
    process_analysis(args.automated, cwd, compiler, architecture, args.n_node_list, start_date)

# read the output file from each test and store timers in
# hdf5 file with pandas format
# -------------------------------------------------------
for n_node in n_node_list:
    print(n_node)
    if browse_output_files:
        res_dir = res_dir_base
        res_dir += '_'.join([run_name, compiler,\
                             architecture, str(n_node)]) + '/'
        for count, current_run in enumerate(test_list):
            # Read performance data from the output file
            output_filename = 'out_' + '_'.join([current_run.input_file, str(n_node), str(current_run.n_mpi_per_node), str(current_run.n_omp), str(count)]) + '.txt'
            # Read data for all test to put in hdf5 a database
            # This is an hdf5 file containing ALL the simulation
            # parameters and results. Might be too large for a repo

# Loop over the tests and run all simulations:
# One batch job submitted per n_node. Several
# tests run within the same batch job.
# --------------------------------------------
if args.mode == 'run':
    if os.path.exists( 'log_jobids_tmp.txt' ):
        os.remove( 'log_jobids_tmp.txt' )
    # loop on n_node. One batch script per n_node
    for n_node in n_node_list:
        res_dir = res_dir_base
        res_dir += '_'.join([run_name, compiler, architecture, str(n_node)]) + '/'
        runtime_param_list = []
        # Deep copy as we change the attribute n_cell of
        # each instance of class test_element
        test_list_n_node = copy.deepcopy(test_list)
        job_time_min = time_min(len(test_list))
        batch_string = get_batch_string(test_list_n_node, job_time_min, module_Cname[architecture], n_node)
        # Loop on tests
        for count, current_run in enumerate(test_list_n_node):
            current_run.scale_n_cell(n_node)
            runtime_param_string  = ' amr.n_cell=' + ' '.join(str(i) for i in current_run.n_cell)
            runtime_param_string += ' amr.max_grid_size=' + str(current_run.max_grid_size)
            runtime_param_string += ' amr.blocking_factor=' + str(current_run.blocking_factor)
            runtime_param_string += ' max_step=' + str( current_run.n_step )
            # runtime_param_list.append( runtime_param_string )
            run_string = get_run_string(current_run, architecture, n_node, count, bin_name, runtime_param_string)
            batch_string += run_string
            batch_string += 'rm -rf plotfiles lab_frame_data diags\n'

        submit_job_command = get_submit_job_command()
        # Run the simulations.
        run_batch_nnode(test_list_n_node, res_dir, bin_name, config_command, batch_string, submit_job_command)