22699554347

Committed 05 Mar 2026 02:33AM UTC coverage: 19.922% (-50.1%) from 70.041%

Build Type

push

github

Committed by Rui Hu

Commit Message

Merge github/main, resolve two version conflicts by keeping gitlab local

Coverage Stats

6 of 2484 new or added lines in 35 files covered. (0.24%)

871 of 4372 relevant lines covered (19.92%)

0.2 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

0.0

/src/pipa/parser/perf_script.py

NEW 1	import pandas as pd	×
NEW 2	import re	×
NEW 3	import multiprocessing	×
NEW 4	import os	×
NEW 5	from pipa.common.logger import logger	×
NEW 6	from pipa.common.hardware.cpu import NUM_CORES_PHYSICAL	×
NEW 7	from pandarallel import pandarallel	×
8
9
NEW 10	class PerfScriptData:	×
NEW 11	def __init__(self, parsed_script_path: str, threads_num=NUM_CORES_PHYSICAL):	×
12	"""
13	Initialize the PerfScriptParser object.
14
15	Args:
16	parsed_script_path (str): The path to the parsed script file.
17	threads_num (int, optional): The number of threads to use for parsing. Defaults to min(12, NUM_CORES_PHYSICAL).
18
19	Raises:
20	FileNotFoundError: If the parsed script file does not exist.
21	"""
NEW 22	if not os.path.exists(parsed_script_path):	×
NEW 23	logger.error(f"File not found: {parsed_script_path}")	×
NEW 24	raise FileNotFoundError()	×
NEW 25	self._perf_script_data = parse_perf_script_file(parsed_script_path, threads_num)	×
26
NEW 27	if len(self._perf_script_data) == 0:	×
NEW 28	logger.warning("No data found in the perf script file.")	×
29
NEW 30	if len(self._perf_script_data) >= 10**6:	×
NEW 31	logger.info(f"Using pandarallel to speed up, threads_num is {threads_num}.")	×
NEW 32	pandarallel.initialize(nb_workers=threads_num)	×
33
NEW 34	self._df_wider = None	×
35
NEW 36	def get_raw_data(self):	×
37	"""
38	Returns the raw data of the perf script.
39
40	Returns:
41	The raw data of the perf script.
42	"""
NEW 43	return self._perf_script_data	×
44
NEW 45	def get_wider_data(self):	×
46	"""
47	Returns a wider version of the performance script data, where the cycles and instructions
48	are merged into a single DataFrame. The DataFrame includes columns for command, pid, cpu,
49	time, cycles, instructions, addr, symbol, dso_short_name, and CPI (cycles per instruction).
50
51	If the wider data has already been computed, it is returned from the cache. Otherwise, the
52	wider data is computed by merging the cycles and instructions DataFrames and calculating
53	the CPI.
54
55	Returns:
56	pandas.DataFrame: The wider version of the performance script data.
57	"""
NEW 58	if self._df_wider is not None:	×
NEW 59	return self._df_wider	×
60
NEW 61	df = self._perf_script_data	×
NEW 62	df_cycles = df.query("event == 'cycles'")	×
NEW 63	df_insns = df.query("event == 'instructions'")	×
NEW 64	df_wider = (	×
65	df_cycles.merge(
66	df_insns,
67	on=[
68	"time",
69	"cpu",
70	"command",
71	"pid",
72	"addr",
73	"symbol",
74	"dso_short_name",
75	],
76	suffixes=("_cycles", "_insns"),
77	)
78	.drop(columns=["event_cycles", "event_insns"])
79	.rename(
80	{
81	"value_cycles": "cycles",
82	"value_insns": "instructions",
83	},
84	axis=1,
85	)
86	)
87
NEW 88	df_wider = df_wider[	×
89	[
90	"command",
91	"pid",
92	"cpu",
93	"time",
94	"cycles",
95	"instructions",
96	"addr",
97	"symbol",
98	"dso_short_name",
99	]
100	]
NEW 101	df_wider["CPI"] = df_wider["cycles"] / df_wider["instructions"]	×
NEW 102	self._df_wider = df_wider	×
NEW 103	return df_wider	×
104
NEW 105	def get_tidy_data(self, thread_list: list = None):	×
106	"""
107	Returns a tidy version of the data by pivoting the wider data and renaming the columns.
108
109	Args:
110	thread_list (list): A list of hardware thread names to include in the tidy data. If None, all
111	threads are included.
112
113	Returns:
114	pandas.DataFrame: A tidy version of the data.
115	"""
NEW 116	df_wider = self.get_wider_data()	×
117
NEW 118	if thread_list is not None:	×
NEW 119	thread_list = [int(thread) for thread in thread_list]	×
NEW 120	df_wider = df_wider[df_wider["cpu"].isin(thread_list)]	×
NEW 121	if len(thread_list) == 1:	×
NEW 122	return df_wider	×
NEW 123	df_t = df_wider.pivot_table(	×
124	index=["time"], columns="cpu", aggfunc="first"
125	).reset_index()
NEW 126	df_t.columns = [f"{col[0]}_{col[1]}" for col in df_t.columns]	×
NEW 127	df_t.rename(columns={"time_": "time"}, inplace=True)	×
NEW 128	return df_t	×
129
130
NEW 131	def parse_one_line(line):	×
132	"""
133	Parses a single line of input and returns a list of parsed values.
134
135	Args:
136	line (str): The input line to be parsed.
137
138	Returns:
139	list: A list containing the parsed values from the input line.
140	The list contains the following elements in order:
141	- command (str): The command name.
142	- pid (int): The process ID.
143	- tid (int): The thread ID.
144	- cpu (int): The CPU number.
145	- time (str): The time value.
146	- value (int): The event value.
147	- event (str): The event name.
148	- addr (str): The address value.
149	- symbol (str): The symbol name.
150	- caller (str): The caller value.
151	"""
NEW 152	try:	×
NEW 153	try:	×
NEW 154	pattern = r"(\S+\|\:-\d+)\s+(\d+\|-\d+)\s+\[(\d+)]\s+(\d+\.\d+):\s+(\d+)\s+(\S+):\s+(\S+)\s+(.*?)\s+\((\S+)\)"	×
155
NEW 156	command, pid, cpu, time, value, event, addr, symbol, caller = re.match(	×
157	pattern, line.strip()
158	).groups()
NEW 159	except Exception as e:	×
NEW 160	pattern = r"(\d+\|-\d+)\s+\[(\d+)]\s+(\d+\.\d+):\s+(\d+)\s+(\S+):\s+(\S+)\s+(.*?)\s+\((\S+)\)"	×
NEW 161	(	×
162	pid,
163	cpu,
164	time,
165	value,
166	event,
167	addr,
168	symbol,
169	caller,
170	) = re.match(pattern, line[15:].strip()).groups()
171
NEW 172	command = line[:15].strip()	×
173
NEW 174	except Exception as e:	×
NEW 175	logger.warning("parse failed for line: " + line + "\n with error: " + str(e))	×
NEW 176	return None	×
177
NEW 178	return [	×
179	command,
180	int(pid),
181	int(cpu),
182	time,
183	int(value),
184	event,
185	addr,
186	symbol,
187	caller,
188	]
189
190
NEW 191	def parse_perf_script_file(parsed_script_path: str, processes_num=NUM_CORES_PHYSICAL):	×
192	"""
193	Parses a perf script file and returns the data as a pandas DataFrame.
194
195	Args:
196	parsed_script_path (str): The path to the perf script file.
197
198	Returns:
199	pandas.DataFrame: The parsed data as a DataFrame.
200	"""
201
202	# Open the perf script file in read mode
NEW 203	with open(parsed_script_path, "r") as file:	×
204	# Read all lines from the file and remove leading/trailing whitespaces
NEW 205	content = [l.strip() for l in file.readlines()]	×
206
207	# Ensure that the content is not None
NEW 208	if content is None:	×
NEW 209	logger.info("content is None")	×
NEW 210	return None	×
211
212	# Define the separator used to split the header and table sections
NEW 213	SEPARATOR = "# ========"	×
214
215	# Check if the first line is the separator, and remove it if present
NEW 216	if content[0] == SEPARATOR:	×
NEW 217	content.pop(0)	×
218
NEW 219	if SEPARATOR in content:	×
220	# Find the index of the separator in the content list
NEW 221	sep_idx = content.index(SEPARATOR)	×
222
223	# Split the content into header and table sections
NEW 224	header, table = content[:sep_idx], content[sep_idx + 2 :]	×
225	else:
NEW 226	table = content	×
227
NEW 228	with multiprocessing.Pool(processes=processes_num) as pool:	×
NEW 229	data = pool.map(parse_one_line, table)	×
230
NEW 231	data = [d for d in data if d is not None]	×
232
233	# Parse each line in the table section and create a DataFrame
NEW 234	return pd.DataFrame(	×
235	data,
236	columns=[
237	"command",
238	"pid",
239	"cpu",
240	"time",
241	"value",
242	"event",
243	"addr",
244	"symbol",
245	"dso_short_name",
246	],
247	)
248
249
NEW 250	if __name__ == "__main__":	×
NEW 251	parse_perf_script_file(input("input perf script text data path:"))	×

ZJU-SPAIL / pipa / 22699554347

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous