Profiling

The LiteBIRD Simulation Framework implements a few tools to measure the performance of the code.

Basic profiling

The class TimeProfiler is a context manager that measures the amount of time spent running the code within a with block.

Here is an example:

from litebird_sim import TimeProfiler
from time import sleep

with TimeProfiler(name="my_code") as perf:
    sleep(1.0)

print(f"The elapsed time is {perf.elapsed_time_s():.0f} s")

The elapsed time is 1 s

Custom parameters

The constructor for TimeProfiler accepts keyword arguments; they are saved in the parameters field of the object:

from litebird_sim import TimeProfiler
from time import sleep

with TimeProfiler(name="my_code", my_param="hello") as perf:
    sleep(0.1)

print("my_param =", perf.parameters["my_param"])

my_param = hello

Export to Speedscope

You can collect several TimeProfiler objects in a list and save it in a JSON file that can be imported into Speedscope, an online viewer for performance profiles. This can be done using the function profile_list_to_speedscope(), which outputs a dictionary that can be saved to a JSON file.

Here is an example:

import json
from litebird_sim import TimeProfiler, profile_list_to_speedscope
from time import sleep

perf_list = []  # type: list[TimeProfiler]

# First code block to profile
with TimeProfiler(name="function_A") as perf:
    sleep(1.0)
perf_list.append(perf)

# Second code block to profile
with TimeProfiler(name="function_B") as perf:
    sleep(2.0)
perf_list.append(perf)

# Save the profile measurements into "profile.json"
with open("profile.json", "wt") as out_f:
    json.dump(profile_list_to_speedscope(perf_list), out_f)

# Now go to https://www.speedscope.app/ and open "profile.json"

When the file profile.json is opened in https://www.speedscope.app/, you will see a time graph with two horizontal bars representing the two with blocks:

When using the Simulation class, a number of functions are automatically profiled and JSON files are saved automatically in the output directory. See Sect. Profiling a simulation for more information.

API reference

class litebird_sim.profiler.TimeProfiler(name: str = '', **kwargs)

Bases: object

A context manager to profile the time spent by the code

This class implements a context manager that uses the performance counter provided by the system to compute the time spent within a with block.

elapsed_time_s()

valid()

litebird_sim.profiler.profile_list_to_speedscope(profile_list: list[litebird_sim.profiler.TimeProfiler]) → dict[Any, Any]

Convert a list of TimeProfiler objects into a Speedscope file

This function takes a list of TimeProfiler objects and creates a dictionary that conforms to the specifications of the Speedscope webapp. The dictionary should be saved to a JSON file and then opened within Speedscope.