"double floor(double)"

[Thomas Moor <thomas.moor@anu.edu.au>]

Dear Sirs.

I have observed that under certain circumstances a "double" beeing casted
to a "long int" and then back to "double" produces fancy results. A small
test-scenario can be found below. It runs fine on i386 architecture.

I have no specific alpha experience. I will be happy for any comments on the
following described below.

Thank you for your patience,
Thomas Moor.


******* c-source code, test.c:

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

int main(int argc,char** argv){
  long  i;
  double a;
  i=(long)(floor(.3));                     /* i should become 0*/
  a=(double)i;                                /* a should become  0.0 */
  printf("i=%d; a=%f;\n",i,a);    /* prints out funny number for a */
  exit(0);
}


******* compiled by:
$ gcc --ansi test.c -lm

******* executed:
$ ./a.out


******** observed output:
i=0; a=36028797018963968.000000;

******** version of gcc used:
$ gcc -v
Reading specs from /usr/lib/gcc-lib/alpha-linux/2.95.2/specs
gcc version 2.95.2 20000313 (Debian GNU/Linux)

******** librarays used when a.out was executed:
$ ldd a.out
 libm.so.6.1 => /lib/libm.so.6.1 (0x000002000011e000)
 libc.so.6.1 => /lib/libc.so.6.1 (0x0000020000258000)
 /lib/ld-linux.so.2 => /lib/ld-linux.so.2 (0x0000020000000000)




--

Thomas Moor

Research School of Information Sciences and Engineering
Australian National University
Canberra ACT 0200 AUSTRALIA

email: thomas.moor@anu.edu.au
phone: +61 2 6279 8680