86 lines
2.3 KiB
Python
86 lines
2.3 KiB
Python
|
# Copyright (c) OpenMMLab. All rights reserved.
|
||
|
|
import argparse
|
||
|
|
|
||
|
|
import numpy as np
|
||
|
|
from texttable import Texttable
|
||
|
|
|
||
|
|
|
||
|
|
def parse_args():
|
||
|
|
parser = argparse.ArgumentParser(
|
||
|
|
description='Analyze sdk profiler file tool.')
|
||
|
|
parser.add_argument('profile_file', help='SDK profile file path')
|
||
|
|
args = parser.parse_args()
|
||
|
|
return args
|
||
|
|
|
||
|
|
|
||
|
|
def main():
|
||
|
|
args = parse_args()
|
||
|
|
with open(args.profile_file) as f:
|
||
|
|
data = f.read()
|
||
|
|
graph, events = data.split('----\n')
|
||
|
|
graph = graph.strip().split('\n')
|
||
|
|
events = events.strip().split('\n')
|
||
|
|
|
||
|
|
name2id = {}
|
||
|
|
id2name = {}
|
||
|
|
for i, name in enumerate(graph):
|
||
|
|
name2id[name] = i
|
||
|
|
id2name[i] = name
|
||
|
|
|
||
|
|
n_active = {i: 0 for i in range(len(name2id))}
|
||
|
|
n_call = {i: 0 for i in range(len(name2id))}
|
||
|
|
t_occupy = {i: 0 for i in range(len(name2id))}
|
||
|
|
t_usage = {i: 0 for i in range(len(name2id))}
|
||
|
|
t_time = {i: [] for i in range(len(name2id))}
|
||
|
|
used_id = set()
|
||
|
|
event_start = {}
|
||
|
|
now = 0
|
||
|
|
t_min = 0
|
||
|
|
|
||
|
|
for event in events:
|
||
|
|
words = event.split()
|
||
|
|
name = words[0]
|
||
|
|
id = name2id[name]
|
||
|
|
used_id.add(id)
|
||
|
|
kind, index, ts = map(int, words[1:])
|
||
|
|
|
||
|
|
if now == 0:
|
||
|
|
t_min = ts
|
||
|
|
key = (id, index)
|
||
|
|
delta = ts - now
|
||
|
|
now = ts
|
||
|
|
|
||
|
|
for i, n_act in n_active.items():
|
||
|
|
if n_act > 0:
|
||
|
|
t_occupy[i] += delta
|
||
|
|
t_usage[i] += delta * n_act
|
||
|
|
|
||
|
|
if kind == 0:
|
||
|
|
event_start[key] = ts
|
||
|
|
n_active[id] += 1
|
||
|
|
n_call[id] += 1
|
||
|
|
else:
|
||
|
|
dt = ts - event_start[key]
|
||
|
|
t_time[id].append(dt)
|
||
|
|
event_start.pop(key)
|
||
|
|
n_active[id] -= 1
|
||
|
|
|
||
|
|
table = Texttable(max_width=0)
|
||
|
|
table.header(
|
||
|
|
['name', 'occupy', 'usage', 'n_call', 't_mean', 't_50%', 't_90%'])
|
||
|
|
|
||
|
|
for id in sorted(list(used_id)):
|
||
|
|
occupy = t_occupy[id] / (now - t_min)
|
||
|
|
usage = t_usage[id] / (now - t_min)
|
||
|
|
times = sorted(t_time[id])
|
||
|
|
t_mean = np.mean(times) / 1000
|
||
|
|
t_50 = times[int(len(times) * 0.5)] / 1000
|
||
|
|
t_90 = times[int(len(times) * 0.9)] / 1000
|
||
|
|
table.add_row(
|
||
|
|
[id2name[id], occupy, usage, n_call[id], t_mean, t_50, t_90])
|
||
|
|
print(table.draw())
|
||
|
|
|
||
|
|
|
||
|
|
if __name__ == '__main__':
|
||
|
|
main()
|