From stefan at sun.ac.za  Sat Apr 10 16:35:19 2010
From: stefan at sun.ac.za (=?ISO-8859-1?Q?St=E9fan_van_der_Walt?=)
Date: Sat, 10 Apr 2010 22:35:19 +0200
Subject: scikits.image performance fix
In-Reply-To: <40A9BF99-B474-49D3-8ECC-01F9C0C9EA81@yale.edu>
References: <40A9BF99-B474-49D3-8ECC-01F9C0C9EA81@yale.edu>
Message-ID: <q2j9457e7c81004101335i7ec1d5aby2e7adc43f9af81ee@mail.gmail.com>

Hey Zach

Can you believe it!  I've merged your changes and pushed to github.

Thanks
St?fan

On 9 April 2010 00:19, Zachary Pincus <zachary.pincus at yale.edu> wrote:
> Hi St?fan,
>
> I noticed a huge performance decrease in the performance of the graph
> algorithms in scikits.image when I installed from your repository on a
> new lab computer; I tracked it down to a fix you made in heap.pyx to
> prevent newer versions of cython from barfing on lines like:
> ? ? ii = (i-1)/2 # CalcPrevAbs
> For some reason, the change to:
> ? ? ii = (int)(i-1)/2 # CalcPrevAbs
> forces the use of the python division machinery! I think this is a
> "feature" (to ape "from __future__ import division" I think), but what
> a gotcha! I fixed it as:
> ? ? ii = (i-1)//2 # CalcPrevAbs
>
>
> Anyhow, if you pull from heap-perf on http://github.com/zachrahan/scikits.image
> ?, you'll get the fix (and an additional correctness issue elsewhere
> in mcp.)
>
> Zach