FIX: Bad Codegen with /Og and Floating-Point Multiplication Class with Virtual Destructor

• The C/C++ Compiler (CL.EXE), used with:
  • Microsoft Visual C++, 32-bit Editions, version 5.0

SYMPTOMS

The Visual C++ 5.0 compiler may generate incorrect instructions when it encounters floating-point multiplication in a member of a class that has a virtual destructor.

RESOLUTION

Use one of the following four workarounds:

• Use /Op.



• Do not use /Og. Note that /O1, /O2, and /Ox are aggregate optimization switches that include /Og.



• Use the optimize pragma around the class member to disable the /Og optimization or to turn on the /Op optimization.



• Make the destructor non-virtual. This may not be an option if the given class is derived from a class that has a virtual destructor in a library that you cannot change.

STATUS

Microsoft has confirmed this to be a bug in the Microsoft products listed at the beginning of this article. This problem was corrected in Visual C++ 5.0sp3.

MORE INFORMATION

Steps to Reproduce Behavior

   // Compile options needed to reproduce the bug: /Og
   // Compile options needed to work around the bug: /Og /Op

   #include <stdio.h>
   #include <math.h>

   struct A {

       double    m_X;
       double    m_Y;

   };

   typedef double rg23[2][3];

   struct  B
   {

       rg23 m;
       B(double rotn = 0);
       void func(double *px, double *py) const;
       virtual ~B(){}

   };

   B::B(double rotn)
   {

       double cosa = cos(rotn);
       double sina = sin(rotn);

       m[0][0] =
       m[1][1] = cosa;
       m[0][1] = -sina;
       m[1][0] = sina;
       m[0][2] =
       m[1][2] = 0;

   }

   void B::func(double *px, double *py) const
   {
       double x = *px;
       double y = *py;
       *px = x * m[0][0] + y * m[0][1] + m[0][2];
       *py = x * m[1][0] + y * m[1][1] + m[1][2];
   }

   int main () {

       A a;
       a.m_X=1.0;
       a.m_Y=2.0;

       printf("Values in: x=%lf; y=%lf\n", a.m_X, a.m_Y);

       B ct;
       ct.func(&amp;a.m_X, &amp;a.m_Y);

       printf("Values out: x=%lf; y=%lf\n", a.m_X, a.m_Y);

       return 0;

   } 

Expected Output

   Values in: x=1.000000; y=2.000000
   Values out: x=1.000000; y=2.000000 

Actual Output

   Values in: x=1.000000; y=2.000000
   Values out: x=1.000000; y=0.000000 

Additional query words:

Keywords : kbcode kbVC500bug kbVC600fix
Version : winnt:5.0
Platform : winnt
Issue type : kbbug