Very Small Single and Double-Precision Values Are Rounded to Zero on Alpha Computers

• Microsoft Windows NT Server version 4.0
• Microsoft Windows NT Workstation version 4.0
• Microsoft Windows NT Server, Enterprise Edition version 4.0
• Microsoft Windows NT Server version 4.0, Terminal Server Edition

SYMPTOMS

Very small single-precision values (values less than 1.175e-38) and very small double-precision values (values less than 2.225e-308) are always rounded towards zero (chopped,) even when the rounding mode is set to round to nearest or round towards plus or minus infinity.

You can test for the presence of this problem by checking to see if the double-precision value 1.4e-44 is converted to the single-precision value of 1.4e-44, but the double-precision value 1.4e-45 is converted to single- precision value of zero.

CAUSE

This problem is caused by inaccurate rounding in the software emulation of these very small values. This problem should not be confused with normal Alpha hardware performance, which rounds ALL very small values to zero. For information about these very small values, please search the RISC Programmer's Guide (available in the online documentation in Microsoft Visual C++ for Alpha) for topics on denormals.

RESOLUTION

To resolve this problem, obtain the latest service pack for Windows NT 4.0 or Windows NT Server 4.0, Terminal Server Edition. For additional information, please see the following article in the Microsoft Knowledge Base:

Q152734 How to Obtain the Latest Windows NT 4.0 Service Pack

STATUS

Microsoft has confirmed this to be a problem in Windows NT 4.0 and Windows NT Server 4.0, Terminal Server Edition. This problem was first corrected in Windows NT 4.0 Service Pack 4.0 and Windows NT Server 4.0, Terminal Server Edition Service Pack 4.

Keywords : NT4SP4Fix kbbug4.00 kbfix4.00.sp4 NTSrvWkst
Version : WinNT:4.0
Platform : winnt
Issue type : kbbug