sorry, I have overseen it somehow... Probably too late already... That's true, that array argument to a function is a pointer. Sorry.

I have tested the code with gcc under MacOS X, the function with arguments as 3 doubles was NOT necesserely faster as the boost::array param. It might be not worse implementing the macro based approach...

Here is my test app compiled with O3 optimization flags. (Below are the timing on a Dual Core 2.4 processor machine with 4 gb ram)

//============================================================================
// Name        : CppTest.cpp
// Author      :
// Version     :
// Copyright   : Your copyright notice
// Description : Hello World in C++, Ansi-style
//============================================================================

#include <iostream>

#include <numeric>

#include <boost/array.hpp>
#include <boost/progress.hpp>

using namespace std;

typedef boost::array<double, 3> array_type;

array_type::value_type sum_arr_copy(array_type a)
{
    return (a[0]+=a[1])+=a[2];
}

array_type::value_type sum_carr(array_type const& a)
{
    return a[0]+a[1]+a[2];
}

array_type::value_type sum_carr_temp_result(array_type const& a)
{
    double result=a[0];
    return (result+=a[1])+=a[2];
}

array_type::value_type sum_carr_accumulate(array_type const& a)
{
    return std::accumulate(a.begin(), a.end(), 0);
}

array_type::value_type sum_arr_copy_accumulate(array_type const& a)
{
    return std::accumulate(a.begin(), a.end(), 0);
}

double sum_doubles_copy(double d1, double d2, double d3)
{
    return d1+d2+d3;
}

double sum_doubles_copy_optimized(double d1, double d2, double d3)
{
    return (d1+=d2)+=d3;
}

double sum_doubles_copy_temp_result(double d1, double d2, double d3)
{
    double result=d1;
    return (result+=d2)+=d3;
}

double sum_doubles_ref_temp_result(double const& d1, double const& d2, double const d3)
{
    double result=d1;
    return (result+=d2)+=d3;
}

/// sorry for macro code. I was too lazy to copy paste it...
#define ARR_P() a
#define DBL_P() d1,d2,d3

#define DO_TEST(function, param_seq)            \
{                                                \
    cout << "--------------------------\n"        \
         << #function"\n";                        \
    double result=0;                            \
    {                                            \
        boost::progress_timer t;                \
        for(unsigned long i=0; i<times; ++i)    \
            result+=function(param_seq());        \
    }                                            \
    cout << "result: " << result << "\n";        \
}

int main() {

    const unsigned long times=~0;
    array_type a = {1.0, 2.3, 3.33};
    double d1 = 1.0, d2 = 2.3, d3 = 3.33;

    DO_TEST(sum_arr_copy, ARR_P);
    DO_TEST(sum_carr, ARR_P);
    DO_TEST(sum_arr_copy_accumulate, ARR_P);
    DO_TEST(sum_carr_temp_result, ARR_P);
    DO_TEST(sum_carr_accumulate, ARR_P);
    DO_TEST(sum_doubles_copy, DBL_P);
    DO_TEST(sum_doubles_copy_optimized, DBL_P);
    DO_TEST(sum_doubles_copy_temp_result, DBL_P);
    DO_TEST(sum_doubles_ref_temp_result, DBL_P);

    cout << "tests done" << endl;

    return 0;
}

Please take a look, that accumulate produced wrong result.

Resulting timings after running each test 4294967295 times:

--------------------------
sum_arr_copy
65.19 s

result: 2.84756e+10
--------------------------
sum_carr
14.99 s

result: 2.84756e+10
--------------------------
sum_arr_copy_accumulate
59.97 s

result: 2.57698e+10
--------------------------
sum_carr_temp_result
14.97 s

result: 2.84756e+10
--------------------------
sum_carr_accumulate
59.84 s

result: 2.57698e+10
--------------------------
sum_doubles_copy
15.00 s

result: 2.84756e+10
--------------------------
sum_doubles_copy_optimized
14.98 s

result: 2.84756e+10
--------------------------
sum_doubles_copy_temp_result
14.92 s

result: 2.84756e+10
--------------------------
sum_doubles_ref_temp_result
14.93 s

result: 2.84756e+10
tests done

I hope that helps and will not require you to develop some solution, which might not be worth the effort.

Good Luck!
Ovanes

On Jan 28, 2008 10:39 PM, Steven Watanabe <watanabesj@gmail.com> wrote:

AMDG

Hicham Mouline wrote:
> After some though, here is more precisely what I'd like to have...
> I apologize that it is quite different from the initial problem:
>
> template<int n> class Tree {
> static double sum(); //
> };
>
>
> If the user instantiates tree<2>, he should get:
>
> template<> class Tree<2> {
> static double sum(double d1, double d2);
> };
>
> template<> class Tree<3> {
> static double sum(double d1, double d2, double d3);
> };
> etc etc...
>
>
> so that in user code, for e.g.:
>
> double d= Tree<4>::sum(d1, d2, d3, d4);
>
> should compile.
>
>
> Is it possible for me to just define the template Tree for the n-case
> without the 2- and 3- specializations?
>

Ah. You still need the preprocessor, but you can rearrange the
definitions slightly.
(untested)

template<int N>
struct TreeSumImpl;

#define TREE_SUM_DEF(z, n, data)\
template<>\
struct TreeSumImpl<n> {\
static double sum(BOOST_PP_ENUM_PARAMS_Z(z, n, double arg)) { ... }\
};

BOOST_PP_REPEAT(20, TREE_SUM_DEF, ~)

template<int N>
struct Tree : TreeSumImpl<N> {
// other code
};

In Christ,
Steven Watanabe

_______________________________________________
Boost-users mailing list
Boost-users@lists.boost.org
http://lists.boost.org/mailman/listinfo.cgi/boost-users