From: Jeff King <peff@peff.net>
Date: Sun, 20 Feb 2011 10:10:24 GMT
Subject: Re: [PATCH 2/4] bump rename limit defaults (again)
Message-ID: <20110220101024.GC988@sigill.intra.peff.net>
In-Reply-To: <AANLkTim2ETC0T8=8D5pWstZRZ95eo5V0+Pr62mNXRBH1@mail.gmail.com>

On Sat, Feb 19, 2011 at 06:54:47PM +0100, Piotr Krukowiecki wrote:

> On Sat, Feb 19, 2011 at 11:21 AM, Jeff King <peff@peff.net> wrote:
> > Let's bump the diff/merge default limits from 200/500 to
> > 400/1000. Those are 2 seconds and 10 seconds respectively on
> > my modern hardware.
> 
> Just curious - what is your modern hardware?

It's a Core i7 840QM. So it's a measly 1.8GHz, but it turboboosts up to
3.2GHz on a single-threaded process.  The multiple cores are irrelevant
for this timing, as it's very single-threaded[1].

The process used in the tens of megabytes of memory, so I think memory
size is irrelevant. It was very CPU bound in my tests.

-Peff

[1] Actually, this is one place where multi-threading the algorithm
would be very easy. It's literally an m*n double-loop, seeing which of
each "n" sources matches each "m" destination best, and the expensive
bit is comparing a source and dest. So you could parallelize score
calculation up to m*n ways, then just sort the result for each dest.