|
Boost : |
Subject: Re: [boost] [parallel_sort] Proposal
From: Edouard A. (edouard_at_[hidden])
Date: 2009-02-02 16:20:39
> That's a lot of extra copying.
[Edouard A.]
I don't do in place merge. I don't know if this would be faster.
>
> What do you mean by "hidden system locks"?
[Edouard A.]
If you have two or more CPUs trying to access the same cache line, locking
will occur. For example, let's say you are operating near the middle with
two different threads t1 & t2:
......t1..] [..t2.......
------------------------
A | A | A | B | B | B
------------------------
You have a potential (well I would say certain) false sharing because the
middle is going to be on the same cache line.
-- EA
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk