128 lines
3.5 KiB
Python
Executable File
128 lines
3.5 KiB
Python
Executable File
import os
|
|
import subprocess
|
|
from datetime import datetime
|
|
|
|
################ HELPER FUNCTIONS ################
|
|
|
|
|
|
def load_template(template_path: str):
|
|
output_template = ""
|
|
with open(template_path, "r") as handle:
|
|
output_template = handle.read()
|
|
return output_template
|
|
|
|
|
|
def write_batch(batch_fpath: str, batch_content: str):
|
|
with open(batch_fpath, "w") as handle:
|
|
_ = handle.write(batch_content)
|
|
|
|
|
|
################### SETUP DIRS ###################
|
|
output_dir = os.getcwd()+"/output/"
|
|
err_dir = os.getcwd()+"/error/"
|
|
batch_files_dir = os.getcwd()+"/batchs/"
|
|
data_dir = os.getcwd()+"/data/"
|
|
|
|
if os.path.isdir(output_dir) == False:
|
|
os.mkdir(output_dir)
|
|
if os.path.isdir(err_dir) == False:
|
|
os.mkdir(err_dir)
|
|
if os.path.isdir(data_dir) == False:
|
|
os.mkdir(data_dir)
|
|
if os.path.isdir(batch_files_dir) == False:
|
|
os.mkdir(batch_files_dir)
|
|
|
|
################ GLOBAL DEFAULTS #################
|
|
mpi1_bin = "/home/hpc/ihpc/ihpc136h/workspace/prototyping/bin/IMB-MPI1"
|
|
default_parameter = {
|
|
"time_stamp": datetime.now().strftime("%y_%m_%d_%H-%M-%S"),
|
|
"job_name": "",
|
|
"output_dir": os.getcwd()+"/output/",
|
|
"err_dir": os.getcwd()+"/error/",
|
|
"data_dir": os.getcwd()+"/data/",
|
|
"n_procs": 18,
|
|
"off_cache_flag": "",
|
|
"bin": mpi1_bin,
|
|
"n_nodes": 1
|
|
}
|
|
|
|
collectives = [
|
|
"Reduce",
|
|
"Reduce_scatter",
|
|
"Allreduce",
|
|
"Allgather",
|
|
"Allgatherv",
|
|
"Scatter",
|
|
"Scatterv",
|
|
"Gather",
|
|
"Gatherv",
|
|
"Alltoall",
|
|
"Bcast",
|
|
]
|
|
|
|
log = ""
|
|
|
|
############### SINGLE-NODE LAUNCH ###############
|
|
procnt = [
|
|
18,
|
|
36,
|
|
54,
|
|
72
|
|
]
|
|
|
|
off_cache_flags = [
|
|
"-off_cache -1",
|
|
"-off_cache 50",
|
|
""
|
|
]
|
|
|
|
single_node_parameter = dict(default_parameter)
|
|
single_node_template = load_template("templates/singlenode.template")
|
|
|
|
for flag in off_cache_flags:
|
|
single_node_parameter["off_cache_flag"] = flag
|
|
for n_procs in procnt:
|
|
single_node_parameter["n_procs"] = n_procs
|
|
for collective in collectives:
|
|
single_node_parameter["job_name"] = collective
|
|
write_batch(batch_files_dir+collective+".sh",
|
|
single_node_template.format(**single_node_parameter))
|
|
result = subprocess.run(["sbatch", batch_files_dir+collective+".sh"],
|
|
capture_output=True, text=True)
|
|
log += f"#{collective} {n_procs}" + "\n"
|
|
log += "\tSTDOUT:" + result.stdout + "\n"
|
|
log += "\tSTDERR:" + result.stderr + "\n"
|
|
############## MULTIPLE-NODE LAUNCH ##############
|
|
off_cache_flags = [
|
|
"-off_cache -1",
|
|
"-off_cache 50",
|
|
""
|
|
]
|
|
|
|
ndcnt = [
|
|
2,
|
|
3,
|
|
4
|
|
]
|
|
|
|
proc_per_node = 72
|
|
multiple_node_parameter = dict(default_parameter)
|
|
multiple_node_template = load_template("templates/multinode.template")
|
|
|
|
for flag in off_cache_flags:
|
|
multiple_node_parameter["off_cache_flag"] = flag
|
|
for n_nodes in ndcnt:
|
|
n_procs = n_nodes*proc_per_node
|
|
multiple_node_parameter["n_procs"] = int(n_procs)
|
|
multiple_node_parameter["n_nodes"] = n_nodes
|
|
for collective in collectives:
|
|
multiple_node_parameter["job_name"] = collective
|
|
write_batch(batch_files_dir+collective+".sh",
|
|
multiple_node_template.format(**multiple_node_parameter))
|
|
result = subprocess.run(["sbatch", batch_files_dir+collective+".sh"],
|
|
capture_output=True, text=True)
|
|
log += f"#{collective} {n_procs}" + "\n"
|
|
log += "\tSTDOUT:" + result.stdout + "\n"
|
|
log += "\tSTDERR:" + result.stderr + "\n"
|
|
print(log)
|