C++Talk.NET C++ language newsgroups   Archives   FAQ   Search   Memberlist   Usergroups   Register   Profile   Log in to check your private messages   Log in  Postfix is slower than postfix: how likely? Goto page 1, 2, 3  Next          C++Talk.NET Forum Index -> C++ Language (Moderated) View previous topic :: View next topic   Author Message Seungbeom Kim Guest Posted: Wed May 10, 2006 1:21 pm    Post subject: Postfix is slower than postfix: how likely? It has been said for a long time that we should prefer the prefix increment/decrement operator to the postfix because the former is never slower and sometimes it can be faster; e.g.: http://www.parashift.com/c++-faq-lite/operator-overloading.html#faq-13.15 Now I wonder: how likely are the cases where you suffer significant performance loss because you use postfix instead of prefix? Is there any benchmark that tells postfix can be measurably slower than prefix? I know I can build a class that takes arbitrarily long to copy to show that postfix can be slower, but it's not usual to have increment operators for such classes (especially in loops, where we worry about performances most). Most of the time we use the operators for built-in types and iterators, both of which are (supposed to be) cheap to copy. Moreover, contrary to my expectation, testing with g++-3.4.4 -S -O revealed no difference at all in the assembly outputs, between for (container::iterator i = c.begin(); i != c.end(); ++i) and for (container::iterator i = c.begin(); i != c.end(); i++) for any container type in the standard library. -- Seungbeom Kim [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Tomás Guest Posted: Wed May 10, 2006 10:21 pm    Post subject: Re: Postfix is slower than postfix: how likely? Seungbeom Kim posted: Quote: Now I wonder: how likely are the cases where you suffer significant performance loss because you use postfix instead of prefix? Is there any benchmark that tells postfix can be measurably slower than prefix? If we're talking about primitive, built-in types, there's no difference. If we're talking about user-defined types, then it depends entirely on the class's definition of the operators. It's commonly said that the prefix is faster than the postfix, and that's because they're commonly implemented akin to the following: class SomeClass { int i; public: SomeClass(int const a) : i(a) {} SomeClass& operator++() /* This is prefix */ { ++i; return *this; } SomeClass operator++(int) /* This is postfix */ { SomeClass temp(*this); /* Involves copy-construction of a temporary */ ++(*this); return temp; } }; Note also the the prefix returns by reference, while the postfix returns by value. This discussion has been done to death. -Tomás [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Jeffrey Schwab Guest Posted: Wed May 10, 2006 10:21 pm    Post subject: Re: Postfix is slower than postfix: how likely? Seungbeom Kim wrote: Quote: It has been said for a long time that we should prefer the prefix increment/decrement operator to the postfix because the former is never slower and sometimes it can be faster; e.g.: http://www.parashift.com/c++-faq-lite/operator-overloading.html#faq-13.15 Now I wonder: how likely are the cases where you suffer significant performance loss because you use postfix instead of prefix? Is there any benchmark that tells postfix can be measurably slower than prefix? Do whatever makes the code clearest. If a post-increment really would be that slow for a given type, the implementor probably should not provide the operation at all, or at least should give you some warning in the documentation. When in doubt, you may as well err on the side of pre-increment in case you ever want to switch to a heavier-weight iterator type. I personally find pre-inc more intuitive, anyway. [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Pete Becker Guest Posted: Wed May 10, 2006 10:21 pm    Post subject: Re: Postfix is slower than postfix: how likely? Seungbeom Kim wrote: Quote: It has been said for a long time that we should prefer the prefix increment/decrement operator to the postfix because the former is never slower and sometimes it can be faster; e.g.: http://www.parashift.com/c++-faq-lite/operator-overloading.html#faq-13.15 Yes, the new style of talking about coding guidelines is to use the weasel-word "prefer", which avoids saying anything concrete. Quote: Now I wonder: how likely are the cases where you suffer significant performance loss because you use postfix instead of prefix? Is there any benchmark that tells postfix can be measurably slower than prefix? Sure, you can write a benchmark that tests this. You might even find a measurable difference in the time it takes to run an empty loop. As you say in the following part (which I snipped), that in real-world applications it usually doesn't make any noticeable difference. This is mostly a hypothetical benefit. On the other, hand, if an application runs too slowly, it may be because there are a bunch of little things that are done sub-optimally (and you won't see that with a profiler), so it's a good habit to use hypothetically faster techniques when they don't have any downside. Since ++i and i++ in this context convey the same meaning, there's no technical reason to choose one over the other, so general performance considerations suggest using ++i. -- Pete Becker Roundhouse Consulting, Ltd. [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Seungbeom Kim Guest Posted: Thu May 11, 2006 1:21 pm    Post subject: Re: Postfix is slower than postfix: how likely? Jeffrey Schwab wrote: Quote: Seungbeom Kim wrote: It has been said for a long time that we should prefer the prefix increment/decrement operator to the postfix because the former is never slower and sometimes it can be faster; e.g.: http://www.parashift.com/c++-faq-lite/operator-overloading.html#faq-13.15 Now I wonder: how likely are the cases where you suffer significant performance loss because you use postfix instead of prefix? Is there any benchmark that tells postfix can be measurably slower than prefix? Do whatever makes the code clearest. I don't think one of '++i' or 'i++' makes the code clearer than the other; at least in those cases I am mostly concerned about (where the return values are ignored). Quote: When in doubt, you may as well err on the side of pre-increment in case you ever want to switch to a heavier-weight iterator type. I personally find pre-inc more intuitive, anyway. I'm already on that side. The question is how often I gain from it, when even something like std::set<T>::iterator shows no difference at all and I cannot think of anything more complex in common use. -- Seungbeom Kim [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Maxim Yegorushkin Guest Posted: Thu May 11, 2006 1:21 pm    Post subject: Re: Postfix is slower than postfix: how likely? Seungbeom Kim wrote: Quote: It has been said for a long time that we should prefer the prefix increment/decrement operator to the postfix because the former is never slower and sometimes it can be faster; e.g.: IMO, the main point in using languages like C++ and C is to use no more than needed and to communicate the intention of code clearly. Among other things needless use of postfix increment defeats these goals. [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top rolkA Guest Posted: Thu May 11, 2006 1:21 pm    Post subject: Re: Postfix is slower than postfix: how likely? Seungbeom Kim <musiphil (AT) bawi (DOT) org> wrote: Quote: It has been said for a long time that we should prefer the prefix increment/decrement operator to the postfix because the former is never slower and sometimes it can be faster; e.g.: http://www.parashift.com/c++-faq-lite/operator- overloading.html#f aq-13.15 Now I wonder: how likely are the cases where you suffer significant performance loss because you use postfix instead of prefix? Is there any benchmark that tells postfix can be measurably slower than prefix? I know I can build a class that takes arbitrarily long to copy to show that postfix can be slower, but it's not usual to have increment operators for such classes (especially in loops, where we worry about performances most). Most of the time we use the operators for built-in types and iterators, both of which are (supposed to be) cheap to copy. Moreover, contrary to my expectation, testing with g++-3.4.4 -S -O revealed no difference at all in the assembly outputs, between for (container::iterator i = c.begin(); i != c.end(); ++i) and for (container::iterator i = c.begin(); i != c.end(); i++) for any container type in the standard library. Hi, To my mind you miss the real point: Why using postfix form if you don't care about the previous value ? Be consistant: Prefer using the prefix form, unless you need the original value, for example: for (size_t i = v.size(); i-- > 0) // postfix when you need it but: for (size_t i = 0; i < v.size(); ++i) You're right: compilers will optimise i++ in such a loop... But what's the point in using i++ ? being consistant improve readability. -- Sam / rolkA [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Seungbeom Kim Guest Posted: Thu May 11, 2006 1:21 pm    Post subject: Re: Postfix is slower than postfix: how likely? Tomás wrote: Quote: Seungbeom Kim posted: Now I wonder: how likely are the cases where you suffer significant performance loss because you use postfix instead of prefix? Is there any benchmark that tells postfix can be measurably slower than prefix? If we're talking about primitive, built-in types, there's no difference. If we're talking about user-defined types, then it depends entirely on the class's definition of the operators. It's commonly said that the prefix is faster than the postfix, and that's because [...] This discussion has been done to death. I might not have made myself clear, but I'm afraid that's not what I asked; I know why prefix is preferred and I've been advocating it to other people for quite a long time. But people (especially those in the habit of writing postfix for no particular reason and reluctant to change) would ask me if there really was any difference and how likely it would be (so that they could be convinced to convert, or they would just stay where they were and continue to write postfix). That's why and what I asked. Hopefully I made myself clearer this time. -- Seungbeom Kim [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top James Dennett Guest Posted: Fri May 12, 2006 12:21 am    Post subject: Re: Postfix is slower than postfix: how likely? Seungbeom Kim wrote: Quote: Tomás wrote: Seungbeom Kim posted: Now I wonder: how likely are the cases where you suffer significant performance loss because you use postfix instead of prefix? Is there any benchmark that tells postfix can be measurably slower than prefix? If we're talking about primitive, built-in types, there's no difference. If we're talking about user-defined types, then it depends entirely on the class's definition of the operators. It's commonly said that the prefix is faster than the postfix, and that's because [...] This discussion has been done to death. I might not have made myself clear, but I'm afraid that's not what I asked; I know why prefix is preferred and I've been advocating it to other people for quite a long time. But people (especially those in the habit of writing postfix for no particular reason and reluctant to change) would ask me if there really was any difference and how likely it would be (so that they could be convinced to convert, or they would just stay where they were and continue to write postfix). That's why and what I asked. Hopefully I made myself clearer this time. It depends on the types involved. For inlined code, there may well be little or no difference. For types where copying is heavier, and inlining is not available, the overhead of postfix may be greater. But this is all secondary; the best reason to prefer prefix is that it is the most direct expression of intent. Postfix increment is "make a copy of X, and increment X", and usually you just mean "increment X", so why ask for more? -- James [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top James Dennett Guest Posted: Fri May 12, 2006 12:21 am    Post subject: Re: Postfix is slower than postfix: how likely? Seungbeom Kim wrote: Quote: Jeffrey Schwab wrote: Seungbeom Kim wrote: It has been said for a long time that we should prefer the prefix increment/decrement operator to the postfix because the former is never slower and sometimes it can be faster; e.g.: http://www.parashift.com/c++-faq-lite/operator-overloading.html#faq-13.15 Now I wonder: how likely are the cases where you suffer significant performance loss because you use postfix instead of prefix? Is there any benchmark that tells postfix can be measurably slower than prefix? Do whatever makes the code clearest. I don't think one of '++i' or 'i++' makes the code clearer than the other; at least in those cases I am mostly concerned about (where the return values are ignored). Nonetheless, there are good arguments with writing i++ only where we want its odd semantics -- whenever I see it in C++ code, I wonder whether (a) there's some reason why it was needed, and (b) whether the programmer who write it knows C++ well. Quote: When in doubt, you may as well err on the side of pre-increment in case you ever want to switch to a heavier-weight iterator type. I personally find pre-inc more intuitive, anyway. I'm already on that side. The question is how often I gain from it, when even something like std::set<T>::iterator shows no difference at all and I cannot think of anything more complex in common use. If you only ever increment things which are builtins or whose operator++ implementations are inline, you may well gain nothing except the (slightly subjective) improvement in code clarity. It seems to me to be a waste of time wondering if you do gain performance from writing ++x instead of x++, given that you obviously don't lose anything, and you *might* lose the other way. ++x is in no way worse (except that in code heavily influenced by C programmers it might not fit with what they're used to, or with the rest of the code -- but writing C++ to suit C programmers is not often an optimal choice). -- James [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Tomás Guest Posted: Fri May 12, 2006 12:21 am    Post subject: Re: Postfix is slower than postfix: how likely? Seungbeom Kim posted: Quote: I might not have made myself clear, but I'm afraid that's not what I asked; I know why prefix is preferred and I've been advocating it to other people for quite a long time. But people (especially those in the habit of writing postfix for no particular reason and reluctant to change) would ask me if there really was any difference and how likely it would be (so that they could be convinced to convert, or they would just stay where they were and continue to write postfix). That's why and what I asked. Hopefully I made myself clearer this time. Very simple. Get a really big, complicated class whose copy-constructor takes a long time to run. Then time a loop: for( ExpensiveClass i; i < 999; ++i) ; Versus: for( ExpensiveClass i; i < 999; i++) ; If they really need that much convincing, they won't be Grade A programmers. (Also turn off all compiler optimizations for good measure). -Tomás [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Greg Herlihy Guest Posted: Fri May 12, 2006 12:21 pm    Post subject: Re: Postfix is slower than postfix: how likely? kanze wrote: Quote: Tomás wrote: Seungbeom Kim posted: Now I wonder: how likely are the cases where you suffer significant performance loss because you use postfix instead of prefix? Is there any benchmark that tells postfix can be measurably slower than prefix? If we're talking about primitive, built-in types, there's no difference. It depends on the compiler. And the machine -- in the C world, postfix became preferred because on the earliest C compilers for the PDP-11, it was faster than prefix. For ++; the reverse was true for --, but since -- was considerably rarer... If we're talking about user-defined types, then it depends entirely on the class's definition of the operators. And the compiler, and the machine. Since the prefix operator "increments" and the postfix operator "copies, then increments" a value, it should be clear that the postfix operator will never be faster than the prefix operator and is much more likely to be slower. Quote: It's commonly said that the prefix is faster than the postfix, It's also been pointed out often enough that actual measurements don't support what is commonly said. That this is more a myth than anything else. Unless the compiler can manage to optimize away the extra copy operation required when postfix-incrementing a value, prefix-incrementation is certain to be faster. Quote: [...] This discussion has been done to death. Never the less, Seungbeom is the first beside myself to post actual facts, as opposed to speculation. To date, no one has actually shown a real life iterator where it made a measurable difference. The speed difference between prefix and postfix incrementing an iterator can readily be measured with, say, a std::vector of std::string's: #include <iostream> #include <vector> #include <string> using std::vector; using std::string; void Output(std::vector<std::string>::const_iterator iter); int main() { vector<string> v; v.resize(10000, "abc"); vector<string>::const_iterator iter = v.begin(); do { Output( ++iter); } while( iter != v.end()); } void Output(std::vector<string>::const_iterator iter) { std::cout << *iter; } The user and system time needed to run this program are: user 0m0.009s sys 0m0.007s Replacing the do loop with this while loop: while (iter != v.end()) { Output( iter++); } takes measurably more time to complete: user 0m0.016s sys 0m0.026s In other words, due to the extra copying required, the postfix-incrementing iteration is on the order of 2.5 times as slow as the same iteration performed with a prefix-incrementing iterator. A difference in time that should not be at all surprising. Greg [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top kanze Guest Posted: Fri May 12, 2006 12:21 pm    Post subject: Re: Postfix is slower than postfix: how likely? Tomás wrote: Quote: Seungbeom Kim posted: I might not have made myself clear, but I'm afraid that's not what I asked; I know why prefix is preferred and I've been advocating it to other people for quite a long time. But people (especially those in the habit of writing postfix for no particular reason and reluctant to change) would ask me if there really was any difference and how likely it would be (so that they could be convinced to convert, or they would just stay where they were and continue to write postfix). That's why and what I asked. Hopefully I made myself clearer this time. Very simple. Get a really big, complicated class whose copy-constructor takes a long time to run. Now you're being silly. Such classes can't be used as iterators. That's all there is to it. The STL copies iterators right and left -- any iterator which is expensive to copy will give unacceptable performance with the STL. Quote: Then time a loop: for( ExpensiveClass i; i < 999; ++i) ; Versus: for( ExpensiveClass i; i < 999; i++) ; If they really need that much convincing, they won't be Grade A programmers. If they pay any attention to such artificial measures, they aren't very professional. The question that Seungbeom asked was very concrete -- what is the probability of it actually making a difference in a real program? Which is the only question which would really concern a professional programmer. Quote: (Also turn off all compiler optimizations for good measure). Let's see -- we're worried about performance, so we turn off all compiler optimizations. I've never worked in an organization that worked like that. (We generally deliver without any optimizations, because we don't need the speed, and compilers are generally more reliable with optimization turned off. But when performance is an issue, compiler optimizations are generally more reliable than manual optimizations.) -- James Kanze GABI Software Conseils en informatique orientée objet/ Beratung in objektorientierter Datenverarbeitung 9 place Sémard, 78210 St.-Cyr-l'École, France, +33 (0)1 30 23 00 34 [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top kanze Guest Posted: Fri May 12, 2006 12:21 pm    Post subject: Re: Postfix is slower than postfix: how likely? James Dennett wrote: Quote: Postfix increment is "make a copy of X, and increment X", and usually you just mean "increment X", so why ask for more? Because that's what people did before you, in your existing code. Because there is a long history for preferring postfix, dating back to the PDP-11 and the early C compilers, where postfix ++ was actually faster. (Not by that much, but on a PDP-11, programs were smaller, and did less, so the difference was more often visible.) I think that the usual situation is that older people, like myself, are in the habit of preferring postfix, because that's what K&R did. And younger people prefer it because that's what their mentors used. All other things being equal, I now prefer prefix. But I don't find the difference large enough to justify bucking existing practice. -- James Kanze GABI Software Conseils en informatique orientée objet/ Beratung in objektorientierter Datenverarbeitung 9 place Sémard, 78210 St.-Cyr-l'École, France, +33 (0)1 30 23 00 34 [ See http://www.gotw.ca/resources/clcm.htm for info about ] [ comp.lang.c++.moderated. First time posters: Do this! ] Back to top Guest Posted: Sat May 13, 2006 12:21 am    Post subject: Re: Postfix is slower than postfix: how likely? Greg Herlihy wrote: Quote: The speed difference between prefix and postfix incrementing an iterator can readily be measured with, say, a std::vector of std::string's: #include <iostream #include <vector #include <string using std::vector; using std::string; void Output(std::vector<std::string>::const_iterator iter); int main() { vector<string> v; v.resize(10000, "abc"); vector<string>::const_iterator iter = v.begin(); do { Output( ++iter); } while( iter != v.end()); } void Output(std::vector<string>::const_iterator iter) { std::cout << *iter; } The user and system time needed to run this program are: user 0m0.009s sys 0m0.007s Replacing the do loop with this while loop: while (iter != v.end()) { Output( iter++); } takes measurably more time to complete: user 0m0.016s sys 0m0.026s In other words, due to the extra copying required, the postfix-incrementing iteration is on the order of 2.5 times as slow as the same iteration performed with a prefix-incrementing iterator. A difference in time that should not be at all surprising. Amazing! Copying a vector iterator (perhaps twice, since postfix will return by value) is over 2.5 times as expensive as constructing a string and sending it to cout. I think it is more likely that your timescales were such that the timings aren't meaningful, or that the undefined behavior in your prefix version meant that version exited faster. How long does it take on your system if you replace the loop with std::cout << "Hello world\n"; ??? [ 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 Goto page 1, 2, 3  Next Page 1 of 3   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.