[pypy-issue] [issue980] Loop with small number of iterations much faster after unrolling

[lesshaste]

New submission from lesshaste <drraph at gmail.com>:

In the following examples, the function hamdistX is called many times. If it is
implemented as a simple loop then it is 3 times slower than if the loop is
unrolled. In the case attached, the hamdistX loop only iterates 5 times.  The
slow version is

def hamdist2(pattern, index, unknown):
    """Count the # of differences between equal length strings str1 and str2"""
    diffs = 0
    for i in xrange(l):
        if pattern[i+index] != unknown[i]:
           diffs += 1
    return diffs  

and the fast version is


def hamdist3(pattern, index, unknown):
    """Count the # of differences between equal length strings str1 and str2"""
    diffs = 0
    i = 0
    if pattern[i+index] != unknown[i]:
        diffs += 1
    i += 1
    if pattern[i+index] != unknown[i]:
        diffs += 1
    i += 1
    if pattern[i+index] != unknown[i]:
        diffs += 1
    i += 1
    if pattern[i+index] != unknown[i]:
        diffs += 1
    i += 1
    if pattern[i+index] != unknown[i]:
        diffs += 1
    return diffs

----------
messages: 3635
nosy: lesshaste, pypy-issue
priority: performance bug
status: unread
title: Loop with small number of iterations much faster after unrolling

________________________________________
PyPy bug tracker <tracker at bugs.pypy.org>
<https://bugs.pypy.org/issue980>
________________________________________