chunks = [[] for _ in range(size)]
for i, chunk in enumerate(runs):
print(f'i, chunk {i} {chunk} {runs[chunk]}')
chunks[i % size].extend(runs[chunk])
else:
data = None
chunks = None
data = comm.scatter(chunks, root=0)
print(f'Node {rank} data {len(data)}')
totals = []
for run_dir in data:
try:
CommandHelper.run_command(
['mkdir', '-p', f'{main_config.out_dir}/{run_dir}'],
main_config.show_command_output,
main_config.show_command_error)
CommandHelper.run_command([
main_config.zstd, '-D', f'{main_config.stats_dir}/zstd-dict',
'-d', '-f', f'{main_config.stats_dir}/{run_dir}/stats.txt.zst',
'-o', 'stats.txt'
], main_config.show_command_output, main_config.show_command_error,
f'{main_config.out_dir}/{run_dir}')
stats = CommandHelper.run_command_output(
['awk', '/sim_insts/ {print $2}', f'stats.txt'],
f'{main_config.out_dir}/{run_dir}').splitlines()
CommandHelper.run_command(['rm', '-rf', f'{run_dir}'],
main_config.show_command_output,
main_config.show_command_error,
f'{main_config.out_dir}')
def execute(self):
try:
if len(self.workloads) != self.num_cpus:
raise AssertionError(
f'length of workloads ({len(self.workloads)}) '
f'is not equal to num_cpus ({self.num_cpus})')
gem5_args: List[str] = [
self.main_config.gem5_executable_path,
self.main_config.gem5_se_script_path, '--cpu-freq',
self.main_config.cpu_freq, '--disable-l2', '1', '--num-cores',
str(self.num_cpus)
]
gem5_args.extend(self.workloads)
CommandHelper.run_command([
'mkdir', '-p',
f'{self.main_config.out_dir}/run_{self.rank}_{self.run_id}'
], self.main_config.show_command_output,
self.main_config.show_command_error)
CommandHelper.run_command(
gem5_args, self.main_config.show_command_output,
self.main_config.show_command_error,
f'{self.main_config.out_dir}/run_{self.rank}_{self.run_id}')
stats = CommandHelper.run_command_output([
'awk', '/sim_sec/ {print $2}', f'stats.txt'
], f'{self.main_config.out_dir}/run_{self.rank}_{self.run_id}/m5out'
).splitlines()
CommandHelper.run_command([
'rm', f'-rf',
f'{self.main_config.out_dir}/run_{self.rank}_{self.run_id}'
], self.main_config.show_command_output,
self.main_config.show_command_error)
CommandHelper.run_command([
'mkdir', '-p',
f'{self.main_config.stats_dir}/run_{self.rank}_{self.run_id}'
], self.main_config.show_command_output,
self.main_config.show_command_error)
if len(stats) == 15:
total_time_for_tasks = 0
for i in range(14):
if i % 2 != 0:
total_time_for_tasks += int(stats[i][2:]) - int(
stats[i - 1][2:])
if total_time_for_tasks > self.find_runs_higher_than_this:
print(
f'node {self.rank} found higher simulation time {total_time_for_tasks} for workload {self.workloads}'
)
JsonHelper.object_as_json_to_file(
f'{self.main_config.stats_dir}/run_{self.rank}_{self.run_id}/workloads.json',
[self.workloads, total_time_for_tasks,
len(stats)])
else:
print(f'Node {self.rank} wrong stats length {len(stats)}')
JsonHelper.object_as_json_to_file(
f'{self.main_config.stats_dir}/run_{self.rank}_{self.run_id}/workloads.json',
[self.workloads, 0, len(stats)])
except OSError as e:
print(f'OSError> {e.errno} {e.strerror} {e.filename}')
except TypeError as e:
print(f'TypeError> {e}')
except AttributeError as e:
print(f'AttributeError> {e}')
except AssertionError as e:
print(f'AssertionError> {e}')
except NameError as e:
print(f'NameError> {e}')
except:
print(f'Error> {sys.exc_info()[0]}')
print(
f'Node {self.rank} ValidateSingleSimulationTask done {self.run_id:,}'
)
def execute(self):
#print(f'Starting RunSingleSimulationTask {self.run_id} {self.workloads}')
try:
if len(self.workloads) != self.num_cpus:
raise AssertionError(
f'length of workloads ({len(self.workloads)}) '
f'is not equal to num_cpus ({self.num_cpus})')
gem5_args: List[str] = [
self.main_config.gem5_executable_path,
self.main_config.gem5_se_script_path, '--cpu-freq',
self.main_config.cpu_freq, '--disable-l2', '1', '--num-cores',
str(self.num_cpus)
]
gem5_args.extend(self.workloads)
CommandHelper.run_command([
'mkdir', '-p',
f'{self.main_config.out_dir}/run_{self.rank}_{self.run_id}'
], self.main_config.show_command_output,
self.main_config.show_command_error)
CommandHelper.run_command(
gem5_args, self.main_config.show_command_output,
self.main_config.show_command_error,
f'{self.main_config.out_dir}/run_{self.rank}_{self.run_id}')
CommandHelper.run_command([
'mkdir', '-p',
f'{self.main_config.stats_dir}/run_{self.rank}_{self.run_id}'
], self.main_config.show_command_output,
self.main_config.show_command_error)
CommandHelper.run_command(
[
self.main_config.zstd, '-15', '--rm', '-f', '-D',
f'{self.main_config.stats_dir}/zstd-dict', f'stats.txt',
'-o',
f'{self.main_config.stats_dir}/run_{self.rank}_{self.run_id}/stats.txt.zst'
], self.main_config.show_command_output,
self.main_config.show_command_error,
f'{self.main_config.out_dir}/run_{self.rank}_{self.run_id}/m5out'
)
CommandHelper.run_command([
'rm', f'-rf',
f'{self.main_config.out_dir}/run_{self.rank}_{self.run_id}'
], self.main_config.show_command_output,
self.main_config.show_command_error)
JsonHelper.object_as_json_to_file(
f'{self.main_config.stats_dir}/run_{self.rank}_{self.run_id}/workloads.json',
self.workloads)
except OSError as e:
print(f'OSError> {e.errno} {e.strerror} {e.filename}')
except TypeError as e:
print(f'TypeError> {e}')
except AttributeError as e:
print(f'AttributeError> {e}')
except AssertionError as e:
print(f'AssertionError> {e}')
except:
print(f'Error> {sys.exc_info()[0]}')
print(f'Node {self.rank} RunSingleSimulationTask done {self.run_id:,}')
