Activity

1. npryce commented on Jan 6, 2013

   

   npryce

   on Jan 6, 2013

   ContributorAuthor

   More actions

   A bit of benchmarking seems to show that raw I/O in Python3.2 is about half the speed as in Python2.7 for the kinds of writes we're doing in the GPIO library.

   The following timings were done on the 3.6.11 kernel.

   nat@raspberrypi $ python3.2 -m timeit -s 'out=open("/dev/null", "w"); s="1\n"' "out.write(s)"
   100000 loops, best of 3: 11.7 usec per loop
   nat@raspberrypi $ python2.7 -m timeit -s 'out=open("/dev/null", "w"); s="1\n"' "out.write(s)"
   100000 loops, best of 3: 5.9 usec per loop

   Oddly, opening in binary mode and writing bytes is slower than writing text:

   nat@raspberrypi $ python2.7 -m timeit -r 10 -s 'out=open("/dev/null", "wb"); bs=b"1\n"' "out.write(bs)"
   100000 loops, best of 10: 12.5 usec per loop
   nat@raspberrypi $ python3.2 -m timeit -r 10 -s 'out=open("/dev/null", "wb"); bs=b"1\n"' "out.write(bs)"
   100000 loops, best of 10: 15 usec per loop

   Converting from numeric or boolean to text is also a bit slower in Python3.2.

   nat@raspberrypi $ python2.7 -m timeit "str(1)"
   100000 loops, best of 3: 4.45 usec per loop
   nat@raspberrypi $ python3.2 -m timeit "str(1)"
   100000 loops, best of 3: 5.52 usec per loop

   nat@raspberrypi $ python3.2 -m timeit -r 10 "str(True)"
   100000 loops, best of 10: 5.34 usec per loop
   nat@raspberrypi $ python2.7 -m timeit -r 10 "str(True)"
   100000 loops, best of 10: 4.64 usec per loop