C Threaded Programming - Incrementing a shared variable

views:

546

answers:

C Threaded Programming - Incrementing a shared variable

Hey guys...so I'm trying to brush up on my C threads and a question I've found is this:

Given a global variable int x = 0; implement the function void useless(int n) which creates n threads which in a loop increase x by 1, each thread terminates when x reaches 100.

I just don't have a handle on threads and need a solid example to base my foundation. This must use pthread system calls as much as possible.

+2 A:

You need a mutex to protect the variable. Each thread will lock the mutex, increment the variable and release the mutex. Each thread that doesn't do this is a rogue thread.

Jonathan Leffler 2009-10-07 18:47:27

I think that the InterlockedIncrement API doesn't use a mutext; instead it just uses a `LOCK:` opcode prefix.

ChrisW 2009-10-07 18:52:20

I don't think InterlockedIncrement alone will cut it here, since you have to protect the compare against 100 as well as the increment.

Dan Olson 2009-10-07 18:56:40

Is there an InterlockedIncrement API in POSIX pthreads?

Pete Kirkham 2009-10-07 19:41:10

@ChrisW: can you provide a context for the InterlockedIncrement API? @Pete: InterlockedIncrement is not a standard part of C, nor is it a part of POSIX. The POSIX thread primitives start 'pthread_'. You can write one easily enough - but it is not standard.

Jonathan Leffler 2009-10-07 19:48:20

@ChrisW: also, what is the 'LOCK:' prefix? That is a label in C - are you confusing C with C# perchance?

Jonathan Leffler 2009-10-07 19:49:03

@Pete: InterlockedIncrement and simplings are part of Win32. Even if there is not a standard posix API for it in phreads most platforms have something similar. On Linux you have the atomic_add and siblings you can use, on Solaris it is atomic_add_32 and similar (man atomic_ops) and I bet you'll find them on most modern OS's (at least I have). If you want to be relatively platform independent, creating a simple .h file with defines mapping in the right implementation depending on the current platform is a piece of a cake as it is normally just a handful of calls and they work the same all over.

Fredrik 2009-10-07 19:58:59

The LOCK prefix is applied to an x86 instruction to say the the processor bus is locked for the duration, making it into an atomic operation.

Pete Kirkham 2009-10-07 19:59:07

I know there's an InterlockedIncrement in win32. The OP wanted to use POSIX pthreads. Apache Portable Runtime has a portable interlocked increment, but I think the unix version used mutexes rather than mapping to all the different flavours last time I looked.

Pete Kirkham 2009-10-07 20:03:50

@Pete: the OP didn't say it had to be POSIX. He said "This must use pthread system calls as much as possible". As there are no atomic POSIX defined operations (that I know of), POSIX falls outside the "as much as possible" requirement. Using locks alone is not always a complete solution, you probably want the variable to be volatile as well.

Fredrik 2009-10-08 21:07:57

What you need is a critical section. Under windows, this would be EnterCriticalSection, but in the pthread environment, the equivalent is pthread_mutex_lock. See here for some pointers.

Bob Moore 2009-10-07 19:07:20

The markup also smartly allows you to use `backticks` to represent `__code__` which disables all other markup. Don't dis the SO markup unless you actually understand it.

Chris Lutz 2009-10-07 19:37:01

I would think that InterlockedIncrement is sufficient, if it is ok for each thread to exit if X >= 100.

I would never use a critical section unless I really have to as this can lead to a high level of contention. Whereas InterlockedIncrement has no contention at all, at least not that might affect over all performance.

Steve 2009-10-07 19:17:40

+2 A:

Pete Kirkham 2009-10-07 21:45:30

ansaurus

tags:

views:

answers:

C Threaded Programming - Incrementing a shared variable

related questions