BorlandTalk.com Borland discussion newsgroups   Archives   FAQ   Search   Memberlist   Usergroups   Register   Profile   Log in to check your private messages   Log in  How to improve performance of C++ complex template? Goto page 1, 2  Next          BorlandTalk.com Forum Index -> C++ Builder (Language C++) View previous topic :: View next topic   Author Message Radek Guest Posted: Sat Dec 17, 2005 7:56 am    Post subject: How to improve performance of C++ complex template? Hi all, I wrote the simulation of technique used in UMTS and I am heavily using STL complex template and I have noticed strange thing. When I am using complex<float> instead of complex<double> it is almost 60% slower. Are there any compiler options that my improve thatI? don't know why it is like that and whether there is a way to improve performance of complex template in general or not. I need to transmit 10 000 000 bits in my simulation which takes ages and I am a little bit confused with speed of my simulation. A little more about my simulation: I generate bits in a function and bits are stored in INT array. Then there is channel coding, spreading and some more operations done on bits. Every single array that stores bits s INT. Then there is QPSK modulation which converts two bits into one complex symbol. All complex symbols are stored in dynamically declared complex<double> arrays. Before complex symbols are transmitted through the channel complex<double> is put into array of structures, because channel is written in such a way that it takes pointer to array of structures as an argument. After transmission throug the channel there are some more functions called that operates on array of structures and then again it is converted to complex<double> to make use of complex template which is quite simple and straight forward to use. And then again complex<double> is converted into bits (each complex symbol into two bits). I tried to change all int arrays into short without expected result (speed boost). I tried the same with char (the same result). As for my compiler settings I have default settings with one change: Data Aligment is Quad Word not default Double Word. Any help, hints, remarks will be much appreciated. Best regards, Radek Back to top Rob Guest Posted: Sat Dec 17, 2005 10:28 am    Post subject: Re: How to improve performance of C++ complex template? Radek wrote: Quote: Hi all, I wrote the simulation of technique used in UMTS and I am heavily using STL complex template and I have noticed strange thing. When I am using complex<float> instead of complex<double> it is almost 60% slower. Are there any compiler options that my improve thatI? don't know why it is like that and whether there is a way to improve performance of complex template in general or not. I need to transmit 10 000 000 bits in my simulation which takes ages and I am a little bit confused with speed of my simulation. I doubt it would be affected by compiler options. One weakness of C++ (which it inherits from C) is that a lot of operations on a float involve an implicit conversion of the float to a double. Several math functions also act on double argument(s) and return a double, meaning a need to convert back the other way. Depending on how it's done, those conversions can introduce a performance hit. Quote: A little more about my simulation: I generate bits in a function and bits are stored in INT array. Then there is channel coding, spreading and some more operations done on bits. Every single array that stores bits s INT. Then there is QPSK modulation which converts two bits into one complex symbol. All complex symbols are stored in dynamically declared complex<double arrays. Before complex symbols are transmitted through the channel complex written in such a way that it takes pointer to array of structures as an argument. After transmission throug the channel there are some more functions called that operates on array of structures and then again it is converted to complex<double> to make use of complex template which is quite simple and straight forward to use. And then again complex<double> is converted into bits (each complex symbol into two bits). It is possible that your conversion from complex<double> to and from bits is introducing a performance hit. Going to complex<float> will not give you much benefit if it is necessary to convert the real and imaginary parts of the complex value to double values. Quote: I tried to change all int arrays into short without expected result (speed boost). I tried the same with char (the same result). As for my compiler settings I have default settings with one change: Data Aligment is Quad Word not default Double Word. AS for the double vs float, there is no guarantee that operations on a short will be implicitly more (or less) efficient than those on a long. And an int might be either a short or a long, depending on your compiler (although with BCB, int and long are the same). I would be surprised if changing data alignment to quad word would give a performance advantage. The basic issue is that performance (however you define it) depends as much on the operating system and hardware as it does on compiler settings and your code. The only way to be sure is to test ..... Back to top Zara Guest Posted: Sat Dec 17, 2005 12:18 pm    Post subject: Re: How to improve performance of C++ complex template? On Sat, 17 Dec 2005 08:56:20 +0100, "Radek" <radekkedzior (AT) poczta (DOT) onet.pl> wrote: Quote: Hi all, I wrote the simulation of technique used in UMTS and I am heavily using STL complex template and I have noticed strange thing. When I am using complex<float> instead of complex<double> it is almost 60% slower. Are there any compiler options that my improve thatI? don't know why it is like that and whether there is a way to improve performance of complex template in general or not. I need to transmit 10 000 000 bits in my simulation which takes ages and I am a little bit confused with speed of my simulation. Try to make all calculations using doubles, and convert to float only for transmission. The type double is the natural type for C/C++ and a s such it is faster than float. It should be the type supported by the FP coprocessor, when there is one available. Quote: A little more about my simulation: I generate bits in a function and bits are stored in INT array. Then there is channel coding, spreading and some more operations done on bits. Every single array that stores bits s INT. Then there is QPSK modulation which converts two bits into one complex symbol. All complex symbols are stored in dynamically declared complex<double> arrays. Before complex symbols are transmitted through the channel complex<double> is put into array of structures, because channel is written in such a way that it takes pointer to array of structures as an argument. After transmission throug the channel there are some more functions called that operates on array of structures and then again it is converted to complex<double> to make use of complex template which is quite simple and straight forward to use. And then again complex<double> is converted into bits (each complex symbol into two bits). I tried to change all int arrays into short without expected result (speed boost). I tried the same with char (the same result). As for my compiler settings I have default settings with one change: Data Aligment is Quad Word not default Double Word. Again, fast calculations are guaranted for int type. All the others must be internally converetd to/from int, so the may behave even more slowly. Quote: Any help, hints, remarks will be much appreciated. Best regards, Radek I think your botteleneck lies in all those structure type conversions, that may need intensive meory copying and type transformation. My 2 cents: use some profiling sytm to identify what is really slowing your system, instead of this trial-and error method you seem to be using Regards, Zara Back to top Alan Bellingham Guest Posted: Sat Dec 17, 2005 2:25 pm    Post subject: Re: How to improve performance of C++ complex template? "Rob" <nospam (AT) nonexistant (DOT) com> wrote: Quote: I doubt it would be affected by compiler options. One weakness of C++ (which it inherits from C) is that a lot of operations on a float involve an implicit conversion of the float to a double. Several math functions also act on double argument(s) and return a double, meaning a need to convert back the other way. Depending on how it's done, those conversions can introduce a performance hit. The common advice I've encountered is to use double, and only to use float if you have space problems. Alan Bellingham -- Team Thai Kingdom <url:http://www.borland.com/newsgroups/> Borland newsgroup descriptions <url:http://www.borland.com/newsgroups/netiquette.html> netiquette Back to top Radek Guest Posted: Sat Dec 17, 2005 3:44 pm    Post subject: Re: How to improve performance of C++ complex template? Quote: I would be surprised if changing data alignment to quad word would give a performance advantage. The basic issue is that performance (however you define it) depends as much on the operating system and hardware as it does on compiler settings and your code. The only way to be sure is to test ..... Thanks for reply. I have already tried different compiler settings and so on but as I have writtem without any results. I know that it highly depends on hardware and OS but I hoped there is other way to optimize the code except modyfing assembler code that I don't feel confident about. Thanks again. Radek Back to top Radek Guest Posted: Sat Dec 17, 2005 3:45 pm    Post subject: Re: How to improve performance of C++ complex template? Quote: The common advice I've encountered is to use double, and only to use float if you have space problems. Ok. I have also noticed that but at first I was really astonished that double is faster than float. Anyway I will try that out. Thanks. Radek Back to top Bob Gonder Guest Posted: Sat Dec 17, 2005 3:59 pm    Post subject: Re: How to improve performance of C++ complex template? Radek wrote: Quote: Thanks for reply. I have already tried different compiler settings and so on but as I have writtem without any results. I know that it highly depends on hardware and OS but I hoped there is other way to optimize the code except modyfing assembler code that I don't feel confident about. Just a reminder that 80% of all speed improvements come from using better algorythms. If you could eliminate the array of complex, and go directly brom array of bits to array of structs (possably using a complex, but not an array of them).. For all of these arrays, do you preallocate the entire array, or expand them? I don't know...is complex costly to construct/destruct? Might be faster to hold onto it rather than be dynamic. 10 million bits, how many complexes? Probably a lot of constructing/destructing going on. Back to top Radek Guest Posted: Sat Dec 17, 2005 6:54 pm    Post subject: Re: How to improve performance of C++ complex template? Quote: Just a reminder that 80% of all speed improvements come from using better algorythms. I know that but I think I did my best in this subject. Quote: If you could eliminate the array of complex, and go directly brom array of bits to array of structs (possably using a complex, but not an array of them).. It is possible I think. I used complex template because it is simple to use and I don't have to overload multiplying operator. Do you think it would be faster to use directly array of structures rather than complex ?? Quote: For all of these arrays, do you preallocate the entire array, or expand them? I am using new, and then delete because all those arrays are dynamically allocated. Quote: I don't know...is complex costly to construct/destruct? Might be faster to hold onto it rather than be dynamic. 10 million bits, how many complexes? Probably a lot of constructing/destructing going on. Many complexes When I transmit 10M bits I must use a lot of memory to serve it in complex. In worst it is 10M * 3 * 8 = 240 M complexes in one array... Br, Radek Back to top Leroy Casterline Guest Posted: Sat Dec 17, 2005 9:20 pm    Post subject: Re: How to improve performance of C++ complex template? On Sat, 17 Dec 2005 08:56:20 +0100, "Radek" <radekkedzior (AT) poczta (DOT) onet.pl> wrote: Quote: Any help, hints, remarks will be much appreciated. Beyond what's been said already, have you tried profiling the application? AutomatedQA (http://www.automatedqa.com/products/aqtime/) has a great profiler with, as I recall, a 30 day trial period. Back to top Radek Guest Posted: Sat Dec 17, 2005 10:51 pm    Post subject: Re: How to improve performance of C++ complex template? Quote: Beyond what's been said already, have you tried profiling the application? AutomatedQA (http://www.automatedqa.com/products/aqtime/) has a great profiler with, as I recall, a 30 day trial period. Thanks for a link. I will also try that out :-) Radek Back to top Bob Gonder Guest Posted: Sun Dec 18, 2005 3:54 am    Post subject: Re: How to improve performance of C++ complex template? Radek wrote: Quote: Just a reminder that 80% of all speed improvements come from using better algorythms. I know that but I think I did my best in this subject. Eliminating "240 M complexes in one array"...comes under the heading of "better algorythm" Quote: It is possible I think. I used complex template because it is simple to use and I don't have to overload multiplying operator. Do you think it would be faster to use directly array of structures rather than complex ?? Well, it eliminates a step, and new complex<double>[240million] can't be very fast. And, it probably causes slow page swaps, using that much memory (about 2GB?) just for intermediary values. Quote: For all of these arrays, do you preallocate the entire array, or expand them? I am using new, and then delete because all those arrays are dynamically allocated. Yes, that was understood from "dynamic", but the question was... do you allocate a few for the array, and add new complexes to the array (expand it), or do you new[] the entire array in one shot? But, it sounds like you aren't using the array template, but instead using the C-type array. Back to top Radek Guest Posted: Sun Dec 18, 2005 8:56 am    Post subject: Re: How to improve performance of C++ complex template? Quote: Eliminating "240 M complexes in one array"...comes under the heading of "better algorythm" I tried splitting this array into smaller ones but it appeared to be slower. Quote: Yes, that was understood from "dynamic", but the question was... do you allocate a few for the array, and add new complexes to the array (expand it), or do you new[] the entire array in one shot? But, it sounds like you aren't using the array template, but instead using the C-type array. Actually I don't know what's an array template if you could explain me in a few words... I "new[]" the array in one shot. I don't expand them. Radek Back to top Bob Gonder Guest Posted: Sun Dec 18, 2005 7:02 pm    Post subject: Re: How to improve performance of C++ complex template? Radek wrote: Quote: Actually I don't know what's an array template if you could explain me in a few words... Well. this _is_ the C++ group, so one never knows if you are going full-blown C++ or partial. I was thinking of the STL container classes (which I don't use, I'm a C guy). Perhaps I meant vector, though I remember reading recently array types being mentioned. There is also the BCB template DynamicArray Perhaps closer to your needs might be valarray. The BCB5.hlp has several pages on it. Also MSDN: http://msdn.microsoft.com/library/default.asp?url=/library/en-us/vcstdlib/html/vclrf_valarray_header.asp?frame=true http://msdn.microsoft.com/library/default.asp?url=/library/en-us/vclang98/html/VALARRAY_valarray.asp?frame=true Back to top Radek Guest Posted: Sun Dec 18, 2005 7:24 pm    Post subject: Re: How to improve performance of C++ complex template? Quote: Well. this _is_ the C++ group, so one never knows if you are going full-blown C++ or partial. I was thinking of the STL container classes (which I don't use, I'm a C guy). Perhaps I meant vector, though I remember reading recently array types being mentioned. There is also the BCB template DynamicArray Perhaps closer to your needs might be valarray. The BCB5.hlp has several pages on it. Also MSDN: http://msdn.microsoft.com/library/default.asp?url=/library/en-us/vcstdlib/ht ml/vclrf_valarray_header.asp?frame=true Quote: http://msdn.microsoft.com/library/default.asp?url=/library/en-us/vclang98/ht ml/VALARRAY_valarray.asp?frame=true I have let's say average knowledge on C++ and quite good on C. I read today about template arrays and syntax is not so complicated but isn't it slowet than new, delete or alloc, malloc, calloc from C syntax? My code maybe is not efficient it could be but taking conditions of my simulation into account it is not so bad When I am transmiting 1 000 000 bits in my simulation it takes about 1,5 hour running on 2 GHz Pentium M, with 1 GB RAM memory. With 10 000 000 bits it is much worse... I will read sth about valarray maybe, as you said, it will be closer to my needs. Thanks again for your help. Br, Radek Back to top Leroy Casterline Guest Posted: Sun Dec 18, 2005 8:04 pm    Post subject: Re: How to improve performance of C++ complex template? On Sun, 18 Dec 2005 20:24:06 +0100, "Radek" <radekkedzior (AT) poczta (DOT) onet.pl> wrote: Quote: I will read sth about valarray maybe, as you said, it will be closer to my needs. Have you given the profiler a try? In my experience what's causing a bottleneck isn't always obvious. It might save you time and effort to know exactly where the problem is before you spend time optimizing the wrong thing (I know - been there, done that!). Back to top Display posts from previous: All Posts1 Day7 Days2 Weeks1 Month3 Months6 Months1 Year Oldest FirstNewest First         BorlandTalk.com Forum Index -> C++ Builder (Language C++) All times are GMT Goto page 1, 2  Next Page 1 of 2   Jump to: Select a forum C++ Builder Newsgroups----------------C++ BuilderC++ Builder (ActiveX)C++ Builder (Announcements)C++ Builder (Command Line Tools)C++ Builder DatabasesC++ Builder Databases (ADO)C++ Builder Databases (Desktop)C++ Builder Databases (InterBase Express)C++ Builder Databases (Multi-Tier)C++ Builder Databases (SQL Servers)C++ Builder (Graphics)C++ Builder (IDE)C++ Builder (Internet)C++ Builder (Internet Socket)C++ Builder (Internet Web)C++ Builder (Jobs)C++ Builder (Language)C++ Builder (Language C++)C++ Builder (Compatibility with MFC)C++ Builder (Native API)C++ Builder (Non-Technical)C++ Builder (Students)C++ Builder (Third Party Tools)C++ Builder (Upgrade)C++ Builder (VCL)C++ Builder (VCL Components Usage)C++ Builder (VCL Components Development)C++ Builder (VCL Usage)C++ Builder (WinAPI)C++ Builder Installation Delphi Newsgroups----------------Delphi (General)Delphi ActiveX Controls UsageDelphi ActiveX Controls DevelopmentDelphi As400Delphi DatabasesDelphi Databases (ADO)Delphi Databases (dbExpress)Delphi Databases (Desktop)Delphi Databases (InterBase Express)Delphi Databases (Multi-Tier)Delphi Databases (SQL Servers)Delphi DeploymentDelphi GraphicsDelphi IDEDelphi InternationalizationDelphi InternetDelphi Internet Isapi-WebBrokerDelphi Internet WebSnapDelphi Internet WinsockDelphi JobsDelphi Language BASMDelphi Language ObjectPascalDelphi Model Driven Architecture (General)Delphi Model Driven Architecture (Third Party)Delphi MultimediaDelphi Native APIDelphi Non-TechnicalDelphi ObjectPascalDelphi OLE AutomationDelphi OO designDelphi OpenToolsAPIDelphi Reporting-ChartingDelphi RTLDelphi StudentsDelphi TeamSourceDelphi Thirdparty ToolsDelphi Thirdparty Tools (General)Delphi Thirdparty Tools (IntraWeb)Delphi Thirdparty Tools (RaveReports)Delphi UpgradeDelphi VCLDelphi VCL ComponentsDelphi VCL Components UsingDelphi VCL Components WritingDelphi WebServices SOAPDelphi WinAPIDelphi XMLDelphi Installationcomp.lang.pascal.delphi.components.misccomp.lang.pascal.delphi.components.usagecomp.lang.pascal.delphi.components.writingcomp.lang.pascal.delphi.databasescomp.lang.pascal.delphi.misccomp.sources.delphi JBuilder Newsgroups----------------JBuilderJBuilder AnnouncementsJBuilder Applet IssuesJBuilder As400JBuilder CompilerJBuilder CORBA/RMIJBuilder DatabasesJBuilder DB SwingJBuilder DebuggerJBuilder DeploymentJBuilder DocumentationJBuilder EnterpriseJBuilder HandheldJBuilder IDEJBuilder JavaJBuilder JavaAPIJBuilder Java LanguageJBuilder Java SwingJBuilder Java BeansJBuilder Java Beans UsageJBuilder Java Beans DevelopmentJBuilder JBCLJBuilder JDataStoreJBuilder JobsJBuilder LinuxJBuilder MacintoshJBuilder & MobileSetJBuilder MultiLingual AppsJBuilder Non-TechnicalJBuilder OpenToolsJBuilder Servlets (JSP)JBuilder StudentsJBuilder Team DevelopmentJBuilder Thirdparty ToolsJBuilder UpgradeJBuilder Installation InterBase Newsgroups----------------InterBaseInterBase Bug DiscussionInterBase GeneralInterBase IBConsoleInterBase OpenSourceInterBase ReplicationInterBase SQLInterBase ThirdpartyInterBase Installation Miscellaneous Borland Newsgroups----------------AppCenterAppServerEnteraVisiBroker  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.