Result of _fpreset() Is Not _CW_DEFAULT

ID: Q119458

The information in this article applies to:

The C Run-Time (CRT), included with:
- Microsoft Visual C++ for Windows, versions 1.0, 1.50
- Microsoft Visual C++ 32-bit Edition, versions 1.0, 2.0, 2.1

SUMMARY

Calling _fpreset() reinitializes the floating-point math package. The header file FLOAT.H contains the following definition of _CW_DEFAULT:


    /* initial Control Word value */ 

   #ifdef  _M_IX86

   #define _CW_DEFAULT ( _RC_NEAR + _PC_64 + _EM_INVALID + _EM_ZERODIVIDE +
      _EM_OVERFLOW + _EM_UNDERFLOW + _EM_INEXACT )

   #endif

You might expect that if you call _fpreset() and then call _controlfp(0,0) to get the control word, that the control word will be _CW_DEFAULT; however, that is not the case. _CW_DEFAULT does not actually correspond to the default value of the x87 control word. The difference is that _CW_DEFAULT does not mask the denormal exception. The _fpreset() routine masks all exceptions, including denormal.

The denormal exception does not belong to the set of floating-point exceptions specified by the IEEE Standard. You cannot mask or unmask the exception using _controlfp(). The abstract control word used by the C run- time reserves the upper bits for machine-specific features. Therefore, you should not compare full control-word encodings; instead, you should compare the bits that you are interested in.

Additional query words: kbinf 1.00 1.50 2.00 2.10

Keywords : kbCRT kbVC
Version : winnt:
Platform : winnt
Issue type :