hn wes benchmark followup

inner.py

import numpy as np
cimport numpy as np
ctypedef np.float64_t float64_t
def inner(np.ndarray[float64_t] x, np.ndarray[float64_t] y): 
    cdef Py_ssize_t i, n = len(x)
    cdef float64_t result = 0 
    for i in range(n):
        result += x[i] * y[i]
    return result

setup.py

from distutils.core import setup
from distutils.extension import Extension
from Cython.Distutils import build_ext
import numpy

setup(
    cmdclass = {'build_ext': build_ext},
    ext_modules = [Extension("inner", ["inner.pyx"],include_dirs=[numpy.get_include()])]
)

On my i7 920, I got these timings:

Julia

julia> x = randn(1,10000000)
julia> y = randn(10000000,1)
julia> @time for i=1:50 result = x * y; end
elapsed time: 0.9744820594787598 seconds
julia> 0.9744820594787598/50*1000
19.489641189575195

Numpy

In [6]: x = randn(10000000)
In [7]: y = randn(10000000)
In [8]: timeit (x*y).sum()
1 loops, best of 3: 41.1 ms per loop

cython

In [6]: from inner import inner
In [7]: x = randn(10000000)
In [8]: y = randn(10000000)
In [9]: timeit inner(x,y)
100 loops, best of 3: 13.8 ms per loop

here's the timing of inner() in Julia:

function inner(x, y)
           result = 0.0
           for i=1:length(x)
               result += x[i] * y[i]
           end
           result
       end

tic()
timing = @time for i=1:50 result = inner(x, y); end
timing = toc();
println("inner took ", (timing / 50) * 1000, " ms")

inner took 37.441840171813965 ms

See Wes's blog post for details on this stuff.

Aaaand finally array operations in Julia (via sum(x.*y)), which is really the better comparison to the numpy example, took 91.3 ms.