Applications and the Math Coprocessor Under Windows

When an application for Microsoft Windows is run on a machine with a math coprocessor, the application can use inline floating-point instructions to take the fullest advantage of the hardware.

Specifying the -FPi option on the C compiler command line causes the Microsoft C optimizing compiler to produce inline 80x87 math coprocessor code for any floating-point math operation. If this code is linked with the WIN87EM.LIB library and a math coprocessor is present in the system at run time, the application will use the inline floating-point instructions for its math operations. If no coprocessor is available at run time, code in the emulator library evaluates floating-point expressions.

An application compiled for the MS-DOS environment with the -FPi option checks for the coprocessor at run time and modifies its code accordingly: if there is a coprocessor, it uses inline floating-point instructions; if there is no coprocessor, it calls software routines to emulate the coprocessor.

In the Windows environment, these run-time modifications are not performed because the Windows kernel fixes up the floating point references as it loads the application's code segments (the kernel is aware of the presence or absence of the numeric coprocessor). This means that an application for the Windows environment compiled with the -FPi option will perform direct, inline floating-point instructions without run-time coprocessor-checking. Consequently, there is no need to link in the Microsoft C Compiler inline floating- point module, which removes the run-time coprocessor-checking.