issue in cacoshf?

[Paul Zimmermann]

       Hi,

with the following program:

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

#define CAT1(X,Y) X ## Y
#define CAT2(X,Y) CAT1(X,Y)
#define CTYPE float complex
#define FOO1 CAT2(c,FOO)
#define FOO2 CAT2(FOO1,f)

float Xbest = -0.0f, Ybest = -0.0f;

int main (int argc, char *argv[])
{
  CTYPE y_re, y_im, z;
  printf ("Xbest=%a Ybest=%a\n", Xbest, Ybest);
  CTYPE z_in = (CTYPE) Xbest + I * (CTYPE) Ybest;
  z = FOO2 (z_in);
  printf ("libm gives (");
  printf ("%a", creal (z));
  printf (",");
  printf ("%a", cimag (z));
  printf (")\n");
  return 0;
}

I get with the GNU libc:

$ gcc -DFOO=acosh check_cacosh2.c -lm
$ ./a.out 
Xbest=-0x0p+0 Ybest=-0x0p+0
libm gives (0x0p+0,0x1.921fb6p+0)

thus we get (0,pi/2), whereas GNU MPC returns (0,-pi/2) instead.

Can someone confirm this is a bug?

Paul

PS: sorry for the cumbersome program, but with a simpler one I cannot
reproduce the issue, even with -fno-builtin.

PS2: there is the same issue with cacosh in double precision.