IJ24659: OPTIMIZATION MOVING ABS() CALL CAUSING INCORRECT VALUE WITH INT_MIN

APAR status

• Closed as program error.

Error description

• A code sequence that looks like:
  
             int num_base=10 ;
              ...
             int sub = value / num_base;
             int mod = value % num_base;
             sub = abs(sub);
  
  The input value for 'value' is INT_MIN, i.e. -2147483648.  The
  division "value / num_base" should always make this larger, i.e.
  closer to 0, i.e. representable by a 32-bit int (as long as
  num_base doesn't equal 1).  When -qhot=level=0 optimization is
  used, the code gets folded into:
  
             sub = abs(value) / 10;
  
  When abs(-2147483648) is calculated, the resulting positive
  integer value 2147483648 is not representable as a 32-bit signed
  integer, and so the behaviour (per the C standard) is undefined.
  
  In this case, the 'value' value stays as -2147483648, causing
  program logic issues.
  

Local fix

• Do not use -qhot
  

Problem summary

• PROBLEM DESCRIPTION:
  IPA optimization was simplifying abs(i1/i2) to abs(i1)/abs(i2),
  but this is incorrect when i1 is of value INT_MIN.
  
  USERS AFFECTED:
  Users compiling code at -O3
  

Problem conclusion

• The fix implemented now disables the optimization
  simplification of abs(i1/i2).
  

Temporary fix

Comments

APAR Information

• APAR is sysrouted FROM one or more of the following:

  LI81260

• APAR is sysrouted TO one or more of the following:

Fix information

• Fixed component name

  XL C/C++ FOR AI

• Fixed component ID

  5725C7200

Applicable component levels