> 1. The generated C files #include <Python.h>, which includes a lot of
> the system header files --- about 16 500 lines' worth on my system.
> gcc can compile a small C source code into a shared library in tens of
> milliseconds, but parsing all these header files takes it more than a
> second on my 300MHz Pentium.
IIRC from generating Python-callable Bass
libraries, python checks a version number
passed to Py_InitModule, so you can just
hard-code (or snarf what you need from the
header files) and be no worse off (for any
given installation).
The timings I give below imply that a small
compile (invoking a shell script + gcc) can
be done in well under a second.
-Dave
:: :: ::
C is hard to beat. I though Takeuchi would
be a cakewalk after the Fibonacci results,
but until I do something with expressions
and registers (which would complicate the
gdb interface), Bass isn't significantly
faster. (and the tail calls in Bass, while
making some code easier to write, confuse
gprof immensely)
pybass.py
=========
import sys,os
if len(sys.argv) > 1:
os.system("bass -so rfmodule.b")
import rf
print rf.rf(33)
> time ./crecfib
33: 3524578
0.59user 0.01system 0:00.59elapsed 101%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (70major+9minor)pagefaults 0swaps
> time python pybass.py compile
3524578
0.55user 0.13system 0:00.67elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (2288major+851minor)pagefaults 0swaps
> time python pybass.py
3524578
0.30user 0.04system 0:00.33elapsed 100%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (182major+85minor)pagefaults 0swaps
:: :: ::
> time ./tak 36 24 12
13
363.66user 0.07system 6:08.56elapsed 98%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (73major+9minor)pagefaults 0swaps
> time ./tak.b 36 24 12
[compiling]
13
336.41user 0.19system 5:38.27elapsed 99%CPU (0avgtext+0avgdata 0maxresident)k
0inputs+0outputs (2596major+800minor)pagefaults 0swaps
