FIX: Library Version of atan2(0,0) Does Not Return 0

The information in this article applies to:

• The C Run-Time (CRT), included with Microsoft Visual C++, 32-bit Edition, versions 1.0, 2.0, 2.1, and 4.0

SYMPTOMS

The on-line help and documentation for Visual C++ specify that calling the atan2 function should return 0.0 if both parameters are 0. The documentation is correct, but the library version of atan2() does not return 0.0, rather, it returns a "not a number" value.

RESOLUTION

The true intrinsic version of atan2() works as documented. There are two ways to instruct the compiler to use the true intrinsic version of the atan2() function:

• Compile with the /Oi and /Og switches, or any switch that implies them both (such as /Ox or /O2). Using /Oi and /Og together instructs the compiler to replace function calls with their true intrinsic version, if one exists. Therefore, when using these switches together, the library version of atan2() is never called, and the problem does not occur.

  NOTE: /Oi and /Og are implied by the /O2 switch, which is the current default for release mode builds.

  -or-

• Use the #pragma intrinsic directive at the global level in the source file that requires the use of atan2() as in this example:

  #pragma intrinsic(atan2)

• The #pragma stays in effect for the entire source file. Use the #pragma function directive to instruct the compiler to use the library version of atan2(). Because the #pragma intrinsic is equivalent to using /Oi on the command line, the source file must be compiled with /Og, as above.

STATUS

Microsoft has confirmed this to be a bug in the Microsoft products listed at the beginning of this article. This bug was corrected in Microsoft Visual C++ 32-bit Edition, version 4.1.

MORE INFORMATION

A true intrinsic function is a function that gets expanded inline without the overhead of a function call. These can be faster than full function calls. In the case of atan2(), both arguments are pushed onto the floating point stack, and the arctangent of y/x is calculated via the fpatan instruction.

The "not a number" value is -1.#IND00.

Sample Code

/* Compile options needed to show the problem: none
   Compile options for workaround /Oi /Og
*/

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

void main(void) {
     double testreturn;

     testreturn = atan2(0.0,0.0);
     printf("testreturn = %lf",testreturn);