Pow very slow on x86_64

[philippe marguinaud <pmarguinaud at yahoo dot com>]

Hi all,

I have some trouble with pow on x86_64. Computing x^r with x close to 1. and r = 0.25 takes ages. The problem does not occur when compiling with -m32 and on x86. Here comes a piece of code:

#include <math.h>
#include <stdio.h>
#include <stdlib.h>

#define N 5000


int main (int argc, char * argv[])
{
  double r;
  double px[N], py[N];
  const double x = 0.9999999999999953371E+00;
  int i;

  r = atof (argv[1]);

  for (i = 0; i < N; i++)
    px[i] = x;

  printf ("x = %25.19e\nr = %25.19e\n", x, r );

  for (i = 0; i < N; i++)
    py[i] = pow (px[i], r);

  printf ("y = %25.19e\n", py[0]);


  return 0;
}

Compiled with:
cc -m64 -O3 -g main.c -o main.x  -L /home/philou/tmp/glibc-2.5/build/math -lm

Takes:

time ./main.x 0.25
x = 9.9999999999999533706e-01
r = 2.5000000000000000000e-01
y = 9.9999999999999877875e-01

real    0m11.803s
user    0m11.799s
sys     0m0.002s


While when compiled with -m32:

x = 9.9999999999999533706e-01
r = 2.5000000000000000000e-01
y = 9.9999999999999888978e-01

real    0m0.004s
user    0m0.003s
sys     0m0.001s


Computing x^0.25 with sqrt(sqrt(x)) is also very fast on x86_64.

Something must be wrong with pow. I am using libm-2.5.so and gcc 4.4.0.

What people think ?

Regards,

Philippe