On 03/11/2010 11:09 PM, Mathieu Desnoyers wrote:
It could be nice to try separating memory allocation from this lockless
queue implementation. I think they belong to different problem space,
and that we should just let the users provide whatever allocator they
use. Possibly, if you need internal copies within your data structure,
that the caller should pass pointers to its malloc/free implementation.
True. Before that, the implementation would need to be genericized, as
right now it's just a test file.
Then providing a lock-free malloc/free could come as a separate lib.
Or an RCU-based malloc/free. :-)
Regarding the articles, I'll add a note. It's basically only "Fast and
Practical Non-Blocking and Blocking Concurrent Queue Algorithms", by
Maged Michael and Michael Scott (1996), except that I don't use
double-word compare-and-swap, and that this RCU-friendly allocation
strategy is "invented" by me.
@inproceeding{Michael96podc,
author = {Maged M. Michael and Michael L. Scott},
title = {Fast and Practical Non-Blocking and Blocking Concurrent
Queue Algorithms},
booktitle = {Proc. 15th ACM Symp. on Principles of Distributed
Computing},
year = {1996},
pages = {267--275},
url = {http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.86.8494}
}
Paolo
_______________________________________________
ltt-dev mailing list
[email protected]
http://lists.casi.polymtl.ca/cgi-bin/mailman/listinfo/ltt-dev
