Subject: Re: [boost] [parallel_sort] Proposal
From: Phil Endecott (spam_from_boost_dev_at_[hidden])
Date: 2009-02-02 16:03:25
Edouard A wrote:
> On Mon, 02 Feb 2009 17:41:58 +0000, "Phil Endecott"
> <spam_from_boost_dev_at_[hidden]> wrote:
>> So basically it's something like this:
>> thread t( sort(begin,mid) );
>> merge(begin,mid,end); // [*]
> Exactly, with some differences as I do a memory copy before the thread and
> I run two threads.
> Working on different buffers allows to prevent hidden system locks when
> you reach the middle.
> so it's more like
> buffer = new ;
> thread t1(sort(buffer,...);
> buffer2 = new ;
> thread t2(sort(buffer2, ...);
> merge(buffer, buffer2, result);
That's a lot of extra copying.
What do you mean by "hidden system locks"?
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk