[Mingw-users] Optimizing Compiler comparison.

[Sternbach, W. [IT] <wil...@ss...>]

Hello,

I work on some long running number crunching programs
which require hours to run.

I wanted to compare the GCC with other Compilers to see
how Optimizing the GNU C++ compiler is.
I created a small benchmark program below which displays the time
after doing each set of 1 million multiplications, divisions,
additions and subtractions.

I compared the GNU C++ compiler versus the
Microsoft MSVC Version 5 compiler and also the
Borland Turbo C++ compiler.

I used the following compiler options:

gcc -O3 cputime.c -ocputime.exe (GNU C++ compiler)
cl /O2x /G5 cputime.c (Microsoft Visual C++ Version 5 Dos prompt compiler).
tcc -G -O cputime.c         (Borland's Turbo C++ compiler).

The results are as you would expect:
Fastest: MSVC
Middle:  GNU C++
Slowest: Borland Turbo C++.

I keep a copy of these cputime.exe's around on a diskette,
and they help me evaluate PC's speed.
I learned that an Intel Pentium II 366 MHZ is the same speed
as an AMD K6 500 MHZ because they both require the same amount
of time to run the cputime.c program below.

I've attached a copy of the program which you can copy into
notepad, save, and compile.

Please let me know what you think.

The source code for the program cputime.c
which I wrote follows:

#include <stdio.h>
#include <time.h>
main()
{
    long i;
    double fp, fp_dummy, fp_accum = 0;
    time_t init_time = time(0); /* init_time = clock time now */

    for (i = 1; i < 1000001; ++i) {
       for (fp=0.1234567890;fp < 1000000;++fp) {
           fp_dummy = (((fp * fp) / 3.1415926536) + 2.718281828) -
1.23456789;
           fp_accum += fp_dummy;
       }
       printf("%ldm, time = %.0lf seconds, ignore = %lf\n",
               i, difftime(time(0), init_time), fp_accum);
    }
    return 0;
}