Shapely speedups benchmarks

Results

(gispy-lab)krusty-2:Shapely-1.2 seang$ python speedups-benchmark.py 

N=10
======
Without speedups
74.75 usec/pass

With speedups
19.80 usec/pass

(gispy-lab)krusty-2:Shapely-1.2 seang$ python speedups-benchmark.py 

N=100
======
Without speedups
518.01 usec/pass

With speedups
27.57 usec/pass

(gispy-lab)krusty-2:Shapely-1.2 seang$ python speedups-benchmark.py 

N=1000
======
Without speedups
4939.36 usec/pass

With speedups
146.21 usec/pass

speedups-benchmarks.py

# Benchmark of olt's https://github.com/olt/shapely/tree/master/shapely/speedups

import array
import timeit

from shapely.geometry import LineString
import shapely.speedups

N = 100 # num of vertices
M = 100 # num of passes

print
print "N=%s" % N
print "======"

xs = array.array('d', [i for i in range(N)])
ys = array.array('d', [i for i in range(N)])
coords = zip(xs, ys)

s = """
line = LineString(coords)
"""
t = timeit.Timer(
    stmt=s,
    setup='from __main__ import coords, LineString'
    )

# No speedups
shapely.speedups.disable()
print "Without speedups"
print "%.2f usec/pass" % (1000000 * t.timeit(number=M)/M)
print

# With speedups
shapely.speedups.enable()
print "With speedups"
print "%.2f usec/pass" % (1000000 * t.timeit(number=M)/M)
print

Without speedups, cost of creating a line is proportional to N (number coordinates). With speedups, it's a flatter function of N.