In the time
module, there are two timing functions: time
and clock
. time
gives you "wall" time, if this is what you care about.
However, the python docs say that clock
should be used for benchmarking. Note that clock
behaves different in separate systems:
- on MS Windows, it uses the Win32 function QueryPerformanceCounter(), with "resolution typically better than a microsecond". It has no special meaning, it's just a number (it starts counting the first time you call
clock
in your process).
# ms windows
t0= time.clock()
do_something()
t= time.clock() - t0 # t is wall seconds elapsed (floating point)
- on *nix,
clock
reports CPU time. Now, this is different, and most probably the value you want, since your program hardly ever is the only process requesting CPU time (even if you have no other processes, the kernel uses CPU time now and then). So, this number, which typically is smaller¹ than the wall time (i.e. time.time() - t0), is more meaningful when benchmarking code:
# linux
t0= time.clock()
do_something()
t= time.clock() - t0 # t is CPU seconds elapsed (floating point)
Apart from all that, the timeit module has the Timer
class that is supposed to use what's best for benchmarking from the available functionality.
¹ unless threading gets in the way…
² Python ≥3.3: there are time.perf_counter()
and time.process_time()
. perf_counter
is being used by the timeit
module.