ctypes version is a little faster

[chenzhiw]

In [1]: from ctypes import CDLL, c_char_p, c_double, c_int

In [2]: import primer3

In [3]: oligotm_so = CDLL("./oligotm.so")

In [4]: oligotm = oligotm_so.oligotm

In [5]: oligotm.restype = c_double

In [6]: oligotm.argtypes = [c_char_p, c_double, c_double, c_double, c_double, c_int, c_int]

In [7]: %timeit primer3.calcTm('GTAAAACGACGGCCAGT') 3.09 µs ± 20.9 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)

In [8]: %timeit primer3.wrappers.calcTm('GTAAAACGACGGCCAGT') 3.5 ms ± 96.6 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

In [9]: %timeit oligotm('GTAAAACGACGGCCAGT'.encode(), 50, 50, 0, 0.8, 1, 1) 2 µs ± 50.3 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)

In [10]: %timeit oligotm(b'GTAAAACGACGGCCAGT', 50, 50, 0, 0.8, 1, 1) 1.93 µs ± 15.2 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)

In [11]: primer3.calcTm('GTAAAACGACGGCCAGT') Out[11]: 49.16808228911765

In [12]: primer3.wrappers.calcTm('GTAAAACGACGGCCAGT') Out[12]: 49.168082

In [13]: oligotm(b'GTAAAACGACGGCCAGT', 50, 50, 0, 0.8, 1, 1) Out[13]: 49.16808228911765

Why is that?

[benpruitt]

There is non-trivial overhead associated with accessing / unpacking locals() and the ThermoAnalysis object manipulation in the current primer3-py implementation.

[benpruitt]

In [5]: %prun [primer3.calcTm('GTAAAACGACGGCCAGT') for _ in range(100000)]
         400004 function calls in 0.368 seconds

   Ordered by: internal time

   ncalls  tottime  percall  cumtime  percall filename:lineno(function)
   100000    0.126    0.000    0.126    0.000 bindings.py:57(_setThermoArgs)
   100000    0.109    0.000    0.324    0.000 bindings.py:208(calcTm)
   100000    0.050    0.000    0.050    0.000 {method 'calcTm' of 'primer3.thermoanalysis.ThermoAnalysis' objects}
        1    0.043    0.043    0.367    0.367 <string>:1(<listcomp>)
   100000    0.040    0.000    0.040    0.000 {built-in method builtins.locals}
        1    0.001    0.001    0.368    0.368 <string>:1(<module>)
        1    0.000    0.000    0.368    0.368 {built-in method builtins.exec}
        1    0.000    0.000    0.000    0.000 {method 'disable' of '_lsprof.Profiler' objects}

_setThermoArgs accounts for 0.126 / 0.324 = 38% of the calcTm call time. So if you factor that out of your timing numbers, then the per call time of the binding is 3.09 µs * (1 - 0.38) = 1.92 µs. (this is a little hand wavy but hopefully it helps explain the discrepancy)

[chenzhiw]

Yeah, thank you. It helps me to understand it.