PRB: Overflow Doesn't Occur Using Pentium Pro or Pentium II

ID: Q168610

The information in this article applies to:

Microsoft Visual Basic Learning, Professional, and Enterprise Editions for Windows, version 6.0
Microsoft Visual Basic Control Creation, Learning, Professional, and Enterprise Editions for Windows, version 5.0
Microsoft Visual Basic Standard, Professional, and Enterprise Editions, 16-bit and 32-bit, for Windows, version 4.0
Microsoft Visual Basic for Applications, included with:
- Microsoft Office 97 for Windows
- Microsoft Office for Windows 95, versions 7.0, 7.0a, 7.0b

SYMPTOMS

When converting a floating point value to an integer value on Intel Pentium Pro and Pentium II processors, a negative value much less than the smallest negative integer does not cause a "Run-Time Error '6': Overflow," as would be expected. This problem is actually caused by an erratum in the Pentium Pro and Pentium II processors. The following code illustrates this problem:


   Debug.Print CInt(-2.59615E+33)
   'Where -2.59615E+33 = -2,596,150,000,000,000,000,000,000,000,000,000

Normally, this code will generate "Run-Time Error '6': Overflow." However, on Pentium Pro and Pentium II processors, this code does not generate an overflow.

For conversion to type Integer, the failure to report an overflow occurs only when the value is in the range -1.84 x 10^19 to -2.60 x 10^33; that is, between:

-18,400,000,000,000,000,000

-and-

-2,600,000,000,000,000,000,000,000,000,000,000

Within the range above, fewer than one out of every 65,000 values is affected.

For conversion to type Long, the failure to report an overflow occurs only when the value is in the range -1.84 x 10^19 to -3.96 x 10^28; that is, between:

-18,400,000,000,000,000,000

-and-

-39,600,000,000,000,000,000,000,000,000.

Within the range above, fewer than one out of every 4,000,000,000 values is affected.

For this problem to be relevant to the Visual Basic developer, the application must meet the following conditions:

Large negative numbers in the ranges given above could possibly be used by the application.

The application may attempt to convert those numbers to an Integer or a Long.

The application uses an On Error statement to catch the run-time error that would be caused by an overflow during the conversion.

CAUSE

Intel Corporation has identified an erratum in the Pentium Pro and Pentium II processors relating to the conversion of floating point values to signed integer values when using the Floating Point Integer Store (FIST) instruction available with these processors. The FIST instruction converts floating point numbers to 16-, 32-, or 64-bit signed integers. Because the range of a floating point number is larger than any of these formats, some floating point numbers cannot be converted to integers. When attempting to convert a floating point number that is too large to an integer, the processor should signal an "Invalid Operation" exception. Visual Basic uses this signal to report "Run-Time Error '6': Overflow." The erratum in the Pentium Pro and Pentium II processors causes them to not signal the "Invalid Operation" exception for certain numbers in the ranges given above.

RESOLUTION

An application may be affected by this problem if all three of the conditions listed under the SYMPTOMS section above are met. In that case, the application must be modified to explicitly test for large negative values and raise the exception itself. This must be done at any point that a floating point number, Single or Double, is converted to Integer or Long, including the following situations:

use of CInt or CLng.

implicit conversion by assigning to an Integer or Long variable.

implicit conversion by passing a value to an Integer or Long procedure parameter.

For conversion to either Integer or Long, the following line of code should be added immediately before the conversion:


   If FloatingPointValue < -1E+18 Then Error 6

For example:


   Dim IntegerValue As Integer
   Dim FloatingPointValue As Single
 
   FloatingPointValue = -2.59615E+33
   If FloatingPointValue < -1E+18 Then Error 6
   IntegerValue = FloatingPointValue

This test and conversion may be encapsulated into a pair of functions, as follows:


Public Function FloatToInt(Expression as Variant) As Integer
      If Expression < -1E+18 Then Error 6
      FloatToInt = Expression
   End Function

   Public Function FloatToLong(Expression as Variant) As Long
      If Expression < -1E+18 Then Error 6
      FloatToLong = Expression
   End Function

Using the FloatToInt function above, the previous example becomes:


   Dim IntegerValue As Integer
   Dim FloatingPointValue As Single

   FloatingPointValue = -2.59615E+33
   IntegerValue = FloatToInt(FloatingPointValue)

STATUS

Microsoft has confirmed that this erratum in the processor affects the Microsoft products listed at the beginning of this article. Visual Basic version 6.0 is not affected by this erratum.

MORE INFORMATION

For more details on this erratum in the Pentium Pro processor, refer to:

http://developer.intel.com/design/pro/update/pprosu.htm

For more details on this erratum in the Pentium II processor, refer to:

http://developer.intel.com/design/PentiumII/update/ppiisu.htm

Intel's documentation on this erratum will note additional ranges of values in which the "Invalid Operation" exception is not set. Visual Basic is not affected by these additional value ranges and it will report the overflow error correctly for them.

REFERENCES

For additional information on a similar problem that affected earlier versions of Visual Basic, see the following article in the Microsoft Knowledge Base:

Q126455 How to Avoid Rounding & Overflow Probs on Pentium Processors

Additional query words: kbVBp400BUG kbVBp500BUG kbVBp600 kbVBp kbdsd kbDSupport KBvba

Keywords : kbGrpVB
Version : WINDOWS:4.0,5.0,6.0,7.0,7.0a,7.0b,97
Platform : WINDOWS
Issue type : kbprb