Skip to content

Profiling with Google PerfTools

Jump to bottom
Ben Toogood edited this page May 28, 2013 · 1 revision

Introduction

When developing stuff it's often beneficial to easily profile your running code to find performance bottlenecks before optimising the code.

The Google Performance Tools provide a library called libprofiler which generates profiles rather easily. This page describes a simple configuration file to enable the use of the google profiler from menu items within gaffer.

Details

Simply place the following in a file called ~/gaffer/startup/gui/profiler.py :

import os
import sys
import ctypes
import tempfile
import uuid

__currentProfile = None

def start( fileName=None ) :

        global __currentProfile

        if __currentProfile is not None :
                raise RuntimeError( "Profiling already in progress" )

        if fileName is None :
                fileName = os.path.join( tempfile.gettempdir(), "gafferProfile" + str( uuid.uuid4() ) + ".prof" )

        lib = ctypes.CDLL( "libprofiler.so" )
        lib.ProfilerStart( fileName )

        __currentProfile = fileName

def stop( view=False ) :

        global __currentProfile

        if __currentProfile is None :
                raise RuntimeError( "Profiling not in progress" )

        lib = ctypes.CDLL( "libprofiler.so" )
        lib.ProfilerStop()

        if view :
                pdf = os.path.splitext( __currentProfile )[0] + ".pdf"
                os.system( "pprof --pdf '%s' '%s' > '%s'" % ( sys.executable, __currentProfile, pdf )  )
                os.system( "evince '%s'&" % pdf )

        __currentProfile = None

def running() :

        return __currentProfile is not None

import GafferUI
GafferUI.ScriptWindow.menuDefinition().append( "/Tools/Profiler/Start", { "command" : start, "active" : lambda : not running() } )
GafferUI.ScriptWindow.menuDefinition().append( "/Tools/Profiler/Stop", { "command" : IECore.curry( stop, view=True ), "active" : running } )
Clone this wiki locally