C++Talk.NET C++ language newsgroups   Archives   FAQ   Search   Memberlist   Usergroups   Register   Profile   Log in to check your private messages   Log in  how to prevent unwanted hoist          C++Talk.NET Forum Index -> C++ Language (Moderated) View previous topic :: View next topic   Author Message andrew_nuss@yahoo.com Guest Posted: Thu May 17, 2007 1:14 pm    Post subject: how to prevent unwanted hoist Hi, Lets say that I have an AtomicCounter whose declaration is: class AtomicCounter { volatile long val; public: AtomicCounter(); ~AtomicCounter(); long ValueOf () { return val; } void Increment () { InterlockedIncrement(&val); } }; And I use it in a function in one thread. In this thread, I only use the ValueOf() member which I'm quite afraid the compiler might think is immutable, and hence only check the value once. class MyStream { AtomicCounter* counter; // set in constructor char* buf; // alloced in constructor int buflen; int bufpos; long lastcount; ... // other members void FillBuf () { ... refill buffer lastcount = counter->Value(); } public: char NextChar () { bool dirty = false; do { if (dirty || bufpos == buflen) FillWindow(); char ch = buf[bufpos++]; if (counter->ValueOf() == lastcount) return ch; cursor == bufpos--; dirty = true; } while (true); } }; The point is that ValueOf() might be hoisted by an overly smart compiler because it appears that counter->Increment() is never called in the NextChar() function nor the FillWindow() function and so the compiler thinks it only has to fetch ValueOf() once. Nevertheless, another thread shares the same counter pointer, and could call counter- Quote: Increment() at any time. I'm wondering if the fact that ValueOf() is not a const function and the AtomicCounter data member is volatile helps any. Otherwise, what do I do? By the way, I think my AtomicCounter works fine on Windows, because InterlockedIncrement() guarantees that another CPU instantly "sees" the updated "val" member. On Unix, I have further trouble. I think that Increment() and ValueOf() both need to use a mutex to ensure that changes to "val" are seen instantly by the other CPU thread. Then ValueOf() becomes way to slow for the performance of NextChar(). I wish ValueOf() didn't need to use the mutex. Thanks, Andy -- [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Pedro Lamarão Guest Posted: Thu May 17, 2007 11:15 pm    Post subject: Re: how to prevent unwanted hoist On May 17, 10:14 am, "andrew_n...@yahoo.com" <andrew_n...@yahoo.com> wrote: Quote: The point is that ValueOf() might be hoisted by an overly smart compiler because it appears that counter->Increment() is never called in the NextChar() function nor the FillWindow() function and so the compiler thinks it only has to fetch ValueOf() once. Nevertheless, another thread shares the same counter pointer, and could call counter- Increment() at any time. I'm wondering if the fact that ValueOf() is not a const function and the AtomicCounter data member is volatile helps any. Otherwise, what do I do? Since the compiler is forbidden to assume anything about the current value of a volatile object, it can never assume anything about the result of calling ValueOf. I don't think any compiler will "cache" this result. I think you are safe. -- Pedro Lamarão [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Denise Kleingeist Guest Posted: Sun May 20, 2007 6:57 am    Post subject: Re: how to prevent unwanted hoist Hello Andrew! On May 17, 3:14 pm, "andrew_n...@yahoo.com" <andrew_n...@yahoo.com> wrote: Quote: class AtomicCounter { volatile long val; The next revision of the C++ standard will provide support for atomic operations. The exact details are still under discussion and currently I can't come up with the exact notation how this would look like. In any case, declaring your variable to be volatile won't have much of an effect with respect to multiple threads accessing it simultanously. Neither current nor future compilers will honour it in a form useful for multi-threading purposes. However, it should already prevent the compiler from hoisting access to it from the loop but since it doesn't introduce any kind for memory fences, that's about it. Good luck, Denise -- [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Le Chaud Lapin Guest Posted: Sun May 20, 2007 11:48 pm    Post subject: Re: how to prevent unwanted hoist On May 20, 1:57 am, Denise Kleingeist <denise.kleinge...@googlemail.com> wrote: Quote: Hello Andrew! On May 17, 3:14 pm, "andrew_n...@yahoo.com" <andrew_n...@yahoo.com wrote: class AtomicCounter { volatile long val; The next revision of the C++ standard will provide support for atomic operations. The exact details are still under discussion and currently I can't come up with the exact notation how this would look like. In any case, declaring your variable to be volatile won't have much of an effect with respect to multiple threads accessing it simultanously. Neither current nor future compilers will honour it in a form useful for multi-threading purposes. However, it should already prevent the compiler from hoisting access to it from the loop but since it doesn't introduce any kind for memory fences, that's about it. This is a good point. It is not clear to me what the OP's algorithm is, but the presence of an unbalanced synchronization primitive is highly suspect. InterlockedIncrement only guarantees atomicity with respect to other Interlocked* functions. A naked increment will ruin the point of using Interlocked* in the first place. This principle is not specific to Windows. In other words, there is a model/mood for doing synchronization that the compiler cannot provide, and must be conscientiously engineered by the programmer, then, after that, the elements of synchronization in the language and the library can be brought to bear upon the design. -Le Chaud Lapin- -- [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Display posts from previous: All Posts1 Day7 Days2 Weeks1 Month3 Months6 Months1 Year Oldest FirstNewest First         C++Talk.NET Forum Index -> C++ Language (Moderated) All times are GMT Page 1 of 1   Jump to: Select a forum C/C++----------------C LanguageC Language (Moderated)C++ Language (Moderated)C++ language (comp.lang.c++)C++ language, library and standards Non-English C/C++ Newsgroups----------------C++ (German)C++ (French)  You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum Powered by phpBB © 2001, 2006 phpBB Group SEO toolkit © 2004-2006 webmedic.