Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.programming.threads > #2572

Re: Can I get code review for some shared memory RPC code?

Show all headers | View raw

> Your wait_until_different and hint_wakeup should be replaced by a
> condition variable (cond_wait, and cond_signal or cond_broadcast).
> Then the need for your pause function and spin loop go away.

Condition variables are implemented in terms of the futex system call
(on Linux only obviously).  I am reimplementing my own concurrency
primitives in terms of the lowest level OS primitives for performance
reasons and for learning experience.  If I used the condition
variables provided by GLibc it'd be highly likely that I'd still be
using spin loops, they'd just be hidden inside of GLibc.  As well,
when built with futexes enabled the code only spins for a small amount
of time and waits in a kernel wait queue otherwise.  I'm honestly kind
of offended because you seemed to have posted a very silly comment
without taking the time to learn a bit about the type of code you were
reviewing first.

> I don't know how portable the atomic_* functions are that you are
> using.  I would use a mutex protected variable, which should work
> everywhere, if portability is important.

stdatomic.h is part of the C11 standard's new functionality for
multithreading.  I'm honestly surprised that a poster on
comp.programming.threads doesn't know about stdatomic.h or can't
bother to look it up.

The futex system call is obviously unportable but it is hidden behind
a define.

_mm_pause is an Intel intrinsic and is obviously nonportable but it
should be easy to find a platform equivalent for any one that I need
to port to.

It's very amusing that you go on to recommend to use a mutex protected
variable as only the atomics part of the C11 standard is very widely
implemented, and POSIX mutexes and condition variables are not
portable to Windows.  Moreover, POSIX mutexes and condition variables
are unlikely to be portable to embedded platforms or other specialized
hardware.  atomic_* functions are probably MORE portable than using
any libraries implementations of mutex.  Besides, for performance I
need my own custom implementation in terms of the lowest level
primitives.

> I did not look at the program logic in detail, as you didn't describe
> what you intended it to do.

What's left to say? I'm implementing Shared Memory Remote Procedure
Calls between two tasks.

Back to comp.programming.threads | Previous | Next — Previous in thread | Find similar | Unroll thread

Thread

Can I get code review for some shared memory RPC code? Steven Stewart-Gallus <stevenselectronicmail@gmail.com> - 2014-08-08 19:17 -0700
  Re: Can I get code review for some shared memory RPC code? andrew@cucumber.demon.co.uk (Andrew Gabriel) - 2014-08-10 15:45 +0000
    Re: Can I get code review for some shared memory RPC code? Steven Stewart-Gallus <stevenselectronicmail@gmail.com> - 2014-08-10 20:57 -0700

csiph-web