PRB: LNK2001 Link Error on Floating Division Functions

• The Microsoft C/C++ Compiler (CL.EXE) included with: Microsoft Visual C++, 32-bit Edition, version 2.1

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

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"

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

   ARTICLE-ID: Q128148
   TITLE     : 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;