|
|
Sponsored Content
Special Forums
UNIX and Linux Applications
High Performance Computing
Memory Barriers for (Ubuntu) Linux (i686)
Post 302430449 by gorga on Thursday 17th of June 2010 03:48:05 PM
06-17-2010
Noting your help on the Programming forum too!
Are you suggesting then, that if I used such an instruction relatively frequently (say once in a loop of maybe a 100 execution statements, per core), I shouldn't notice a significant drop in throughput of the application?
You'd expect that each core accessing the XCHG variable though would have to get the value from memory though as soon as it accessed it, otherwise what use would CMPXCHG be? Not sure about this area to be honest, (but I read that these atomic operations do create a memory barrier so a core cannot execute instructions either side of said barrier out of order).
What I'm building is a thread-pool with n pthreads equal to the number of cores (so I am using pthreads). The pthreads continually execute a list of "lightweight tasks". The user can create tasks and send "messages" between them. (If you've ever used Erlang, something similar to the abstraction provided there but in my case using C).
The pthreads occasionally check the "value" of a task "state", when they reach that task in the queue, therefore if the "state" isn't "ready" they simply move on to the next task (hence the pthread has more work to do and isn't polling continuously). You see what this means, as long as a pthread "eventually" discovers a task is "ready" that's okay, even if it's not asap. It seems like a lock would be unnecessary here then, but a pthread shouldn't detect that the task state is "ready" before its other data members have been updated (hence the need for a memory barrier).
If using these atomic operations isn't going to impact throughput, then great they solve the problem, but even that seems like overkill when I only need to ensure that a handful of statements are executed in a certain order.
Quote:
I think you're overreacting... Any memory I/O monopolizes the bus*, LOCK just guarantees one instruction gets two ops in a row.
Quote:
I'm not sure LOCK XCGH even forces a real memory fetch anymore(might be simple to test, try to get back to you on that.)
Quote:
Lastly, if you're doing no mutexing, what are you doing instead -- polling? That's not going to be more efficient, untold amounts of CPU will be expended on what amounts to a while(1) loop.
The pthreads occasionally check the "value" of a task "state", when they reach that task in the queue, therefore if the "state" isn't "ready" they simply move on to the next task (hence the pthread has more work to do and isn't polling continuously). You see what this means, as long as a pthread "eventually" discovers a task is "ready" that's okay, even if it's not asap. It seems like a lock would be unnecessary here then, but a pthread shouldn't detect that the task state is "ready" before its other data members have been updated (hence the need for a memory barrier).
If using these atomic operations isn't going to impact throughput, then great they solve the problem, but even that seems like overkill when I only need to ensure that a handful of statements are executed in a certain order.
|
|
4 More Discussions You Might Find Interesting
1. UNIX for Dummies Questions & Answers
I have 512 mem on this laptop, though 'top' tells me I only have 380. However, Ubuntu is using 288 mb of memory, when I only have 3 terminals, running lynx, vim(for this file) and (of course) top. Considering it I have lynx running a 600 page txt file, which of course would eat some memory but 300?... (0 Replies)
Discussion started by: riwa
0 Replies
2. Linux
Hi, i am quite new to linux. I am interested in fedora linux distro. Fedora Project
I dont know which one to choose, either i686, x86 64 or ppc. I prefer a live cd, coz its easy to use.
And what is the difference between "Fedora Desktop Live Media" and "Fedora KDE Live Media". (3 Replies)
Discussion started by: superblacksmith
3 Replies
3. Programming
Hi guys , i need to get the total virtual memory in ubuntu but i need to write a C++ code for that, any idea on how to go about doing it? any references? or website that i can refer to ? (6 Replies)
Discussion started by: xiaojesus
6 Replies
4. Ubuntu
My PC (Esprimo, 3 yeas old) has one hard drive having 2 partitions C: (80 GB NTFS, XP) and D: (120 GB NTFS, empty) and and a 200 MB area that yet is not-partitioned.
I would like to try Ubuntu and to install Ubuntu on the not-partitioned area . The idea is to have the possibility to run... (7 Replies)
Discussion started by: C.Weidemann
7 Replies
LEARN ABOUT OPENSOLARIS
libmlib_mt
libmlib_mt(3LIB) Interface Libraries libmlib_mt(3LIB) NAME
libmlib_mt - multi-threaded mediaLib SYNOPSIS
cc [ flag... ] file... -lmlib_mt -lmlib [ library... ] #include <mlib.h> DESCRIPTION
Interfaces in this library provide functions for multimedia processing. Multi-threaded (MT) mediaLib is a software layer developed on top of mediaLib using OpenMP. When it is used with a large data set on a multi-processor system, MT mediaLib will partition data into subsets and process the subsets in parallel, thus greatly improving performance of applications that use mediaLib. INTERFACES
The shared object libmlib_mt.so.2 provides the same public interfaces as those defined in libmlib(3LIB). See Intro(3) for additional infor- mation on shared object interfaces. USAGE
There are two ways to use MT mediaLib. 1. Pre-load a multi-threaded mediaLib library during runtime by setting the LD_PRELOAD environment variable as follows before starting your application, in Bourne/Korn shell: LD_PRELOAD=libmlib_mt.so export LD_PRELOAD or in C shell: setenv LD_PRELOAD libmlib_mt.so In this way, you can take advantage of MT mediaLib without rebuilding your application. 2. Link your application with a multi-threaded mediaLib library directly as shown under SYNOPSIS. In this way, an MT mediaLib library is always used whenever your application is started. The parallelization of MT mediaLib is controlled, in part, by the PARALLEL environment variable. You can change its setting to adjust the degree of parallelization before starting your application, in Bourne/Korn shell: PARALLEL=n export PARALLEL or in C shell: setenv PARALLEL n where n is a positive integer for number of threads. Note that other factors also affect the degree of parallelization in MT mediaLib. FILES
/usr/lib/libmlib_mt.so.2 shared object /usr/lib/64/libmlib_mt.so.2 64-bit shared object ATTRIBUTES
See attributes(5) for descriptions of the following attributes: +-----------------------------+-----------------------------+ | ATTRIBUTE TYPE | ATTRIBUTE VALUE | +-----------------------------+-----------------------------+ |Availability |SUNWmlibt | +-----------------------------+-----------------------------+ |MT-Level |MT-Safe | +-----------------------------+-----------------------------+ SEE ALSO
Intro(3), libmlib(3LIB), attributes(5) mediaLib User's Manual SunOS 5.11 15 Oct 2007 libmlib_mt(3LIB)