README.md

Pycallgrind

Utilities for context based profiling with callgrind. This is backed by CFFI based wrappers for callgrind's client requests.

Usage

The main api entry point is pycallgrind.callgrind. This can be used as a contextmanager. Callgrind instrumentation will be activated only within this block, and will be dumped to file at the end of the block.

>>> from pycallgrind import callgrind
>>> with callgrind(tag="foo"):
...     some_function()
...     another_function()

It can also be used as a decorator. Instrumentation is only active during the function call.

>>> @callgrind(tag="foo")
... def foo(a, b, c):
...     some_function()
...     another_function()

If the decorated function is a numba jitted function, the result will also be a numba jitted function. This works in nopython mode as well.

>>> @callgrind(tag="foo")
... @nb.njit
... def bar(a, b):
...     return a + b

The lower level client request functions are also included. These are trivial wrappers around the callgrind macros of the same name.

Note: Since instrumentation will be turned on during code execution, you probably want to include the --instr-atstart=no flag:

valgrind --tool=callgrind --instr-atstart=no python myfile.py