PRB: LNK2001 Link Error on Floating Division Functions (128895)



The information in this article applies to:

  • Microsoft Visual C++, 32-bit Editions 2.1

This article was previously published under Q128895

SYMPTOMS

Compiling a program containing floating point divisions and linking with run-time libraries other than those supplied with Microsoft Visual C++ version 2.1 can result in LNK2001 unresolved external errors on the following functions:

   __adj_fdiv_m32
   __adj_fdiv_m64
   __adj_fdivr_m64
   __adjust_fdiv
				


This can occur when linking to libraries such as CRTDLL.LIB or the run-time libraries that came with Microsoft Visual C++ version 2.0.

CAUSE

The compiler and libraries that ship with Microsoft Visual C++ version 2.1 provide a method of avoiding Pentium division instructions. Pentium division instructions can generate inaccurate results in some cases. The compiler inserts calls to emulation functions instead of inserting floating point processor instructions. These floating point division emulation functions are in the run-time libraries shipped with version Visual C++ version 2.1, but not with earlier versions of the libraries.

RESOLUTION

The libraries and compiler that come with Microsoft Visual C++ version 2.1 are designed to work together. If other libraries must be used, use the following compiler switch to force insertion of floating point processor instructions instead of inserting calls to the emulation functions:

   /QIfdiv-          {use coprocessor instead of calls to functions}
				

STATUS

This behavior is by design.

MORE INFORMATION

The sample code below will generate the following four errors if the LIB environment variable is pointing to the Visual C++ version 2.0 libraries:

   error LNK2001: unresolved external symbol "__adj_fdiv_m32"
   error LNK2001: unresolved external symbol "__adj_fdiv_m64"
   error LNK2001: unresolved external symbol "__adj_fdivr_m64"
   error LNK2001: unresolved external symbol "__adjust_fdiv"
				


Using the compile option "/QIfdiv-" will prevent the references to these functions.

For information about a related issue on Pentium floating point functions, please see the following article in the Microsoft Knowledge Base:

128148 Procedure Entry Point fdiv Not Located in MSVCRT20.DLL



Sample Code

/* Compile options needed: none
   Environment needed    : set LIB point to a directory with the VC++
                           2.0 libraries
*/ 

void main(void) {
   double x=1.2,z;
   float y=2.1;
   z=x/y;
   x=x/z;
   x=1/z;

}

Modification Type:MinorLast Reviewed:7/5/2005
Keywords:kbCRT kbprb KB128895 kbAudDeveloper