This page list all efforts to make Hachoir faster.
Benchmark with benchmark.sh shell
To run the benchmark, use:
./benchmark.sh
Data on haypo's computer with Python 2.4.3
hachoir-grep on testcase/yellowdude.3ds:
- svn 1058: 1.05 ms
- svn 807: 1.05 sec
- svn 769: 1.09 sec
- svn 681: 1.22 sec
- svn 610: 1.20 sec
hachoir-metadata: set A (mp3, wav, png, au, mkv):
- svn 1058: 324.3 ms
- svn 807: 153.4 ms
Benchmark using program time command
Are this mesure really usefull? :-p
time python -OO script/hachoir-metadata testcase/logo-Kubuntu.png (smallest value of "real"):
- svn 807: 185 ms
- svn 769: 173 ms
- svn 681: 172 ms
- svn 610: 190 ms
- svn 508: 170 ms
- svn 479: 105 ms
- 2006-07-19: 260 ms
- version 0.3: 380 ms
Tricks
Faster Class.__init__
class Classic: def __init__(self): self._x = 0 self._y = {} self._z = 4 class Faster: _x = 0 _z = 4 def __init__(self): self._y = {} # you can't move it !
Mutable (list, dict, etc.) can't be moved outside __init__().
function calling
Call a function is slow (at least in CPython), so use less function call as possible.
You can use "inlining":
def useless(a, b): print "a=%s b=%s" % (a, b): def slow(a): b = a*2 useless(a, b) -- def fast(a): b = a*2 print "a=%s b=%s" % (a, b):
Comparaison
- "if hasattr()" is slower than "if value is not None"
- "if x != None" is slower than "if x is not None"
Links
Articles:
Tools:
