Re: Performance luabind vs. SWIG

2009/11/27 Tony Kostanjsek <[email protected]>

> Did you enable any optimization switches for the build? IIRC template
> inlining doesn't kick in unless you switch on -O2 or something, and you can
> tweak additional inlining parameters.
>
> ----- Ursprüngliche Mail ----
> Von: Sebastian Wolff <[email protected]>
> An: [email protected]
> Gesendet: Freitag, den 27. November 2009, 17:39:35 Uhr
> Betreff: [luabind] Performance luabind vs. SWIG
>
> Hello list,
>
> Since I did not find any speed comparisons on bindings to Lua I decided
> to run a little test for a simple class implementation using SWIG and
> luabind.
>
> Here is the class:
>
> namespace tmmod
> {
>    class TMMatrix;
>
> //     #ifndef SWIG
>
>    class TMMatrix{
>        public:
>            TMMatrix(int i, int j) {}
>            TMMatrix operator+(const TMMatrix & rhs)const{}
>            void func1(int i){}
>            void func2(int i){}
>            void func3(int i){}
>            void func4(int i){}
>            void func5(int i){}
>            void func6(int i){}
>            void func7(int i){}
>
>    };
>
>
>    #ifndef SWIG
>    void open_lib_TM(lua_State *myLuaState);
>    #endif
>
>    void external_matrix_method(TMMatrix & mat);
>
> }
>
> and here the luabind code:
>
>    void open_lib_TM(lua_State *myLuaState)
>    {
>        luabind::module(myLuaState, "luabind")         [
>            luabind::class_<TMMatrix>("TMMatrix")
>                .def(luabind::constructor<int,int>())
>                .def("func1",&TMMatrix::func1)
>                .def("func2",&TMMatrix::func2)
>                .def("func3",&TMMatrix::func3)
>                .def("func4",&TMMatrix::func4)
>                .def("func5",&TMMatrix::func5)
>                .def("func6",&TMMatrix::func6)
>                .def("func7",&TMMatrix::func7)
>                .def(luabind::self + luabind::other<TMMatrix>())
>        ];
>
>
>    }
>
> The wrapper code for SWIG is simply "%include matrix.hpp"
>
> The functions do not anything - I only wanted to know the speed of the
> binding: So I ran through the following example:
>
> n = 1e6
>
> print("time at start: " .. TNG.Time())
>
> mat2 = luabind.TMMatrix(3,3)
> for i=1,n do
>        matrix = luabind.TMMatrix(3,3)
>        mat2 = matrix+luabind.TMMatrix(3,3)
>        mat2 = matrix+luabind.TMMatrix(3,3)
>        mat2:func1(1)
>        mat2:func3(1)
>        mat2:func7(1)
> end
>
> print("time after luabind: " .. TNG.Time())
>
> mat2 = tm.TMMatrix(3,3)
> for i=1,n do
>        matrix = tm.TMMatrix(3,3)
>        mat2 = matrix+tm.TMMatrix(3,3)
>        mat2 = matrix+tm.TMMatrix(3,3)
>        mat2:func1(1)
>        mat2:func3(1)
>        mat2:func7(1)
> end
>
> print("time after SWIG: " .. TNG.Time())
>
>
> and got:
>
> luabind: 9.4 sec
> swig:    4.78 sec
>
> This is quite astonishing since I thought that template magic may be
> able to speed up things. I did not use any special -D's, luabind was
> used as provided by my vendor (opensuse x64).
>
> I only wanted the list you know this. Maybe someone picks it up to
> finde some room for improvement in luabind.
>
> Sebastian
>
>
>
>
>
>
>
> ------------------------------------------------------------------------------
> Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
> trial. Simplify your report design, integration and deployment - and focus
> on
> what you do best, core application coding. Discover what's new with
> Crystal Reports now.  http://p.sf.net/sfu/bobj-july
> _______________________________________________
> luabind-user mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/luabind-user
>
>
> __________________________________________________
> Do You Yahoo!?
> Sie sind Spam leid? Yahoo! Mail verfügt über einen herausragenden Schutz
> gegen Massenmails.
> http://mail.yahoo.com
>
>
> ------------------------------------------------------------------------------
> Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
> trial. Simplify your report design, integration and deployment - and focus
> on
> what you do best, core application coding. Discover what's new with
> Crystal Reports now.  http://p.sf.net/sfu/bobj-july
> _______________________________________________
> luabind-user mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/luabind-user
>

I do not know the internals of either swig or luabind but I have a couple of
observations about this test.
Firstly it is a bogus profiling, anything class such as vector or matrix
should have a plain Lua implementation for speed reasons especially if you
are going to call it this many times.
Secondly I can only guess but I would assume this test calls the new
operator 5 million times. Does luabind work this way? ie when a new class is
created does it call new in C++ code?
Thirdly how did you even get the code to compile as you have a addition
operator which does not return anything yet is defined as returning an
instance.

------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day 
trial. Simplify your report design, integration and deployment - and focus on 
what you do best, core application coding. Discover what's new with
Crystal Reports now.  http://p.sf.net/sfu/bobj-july

_______________________________________________
luabind-user mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/luabind-user