LI76655: LONG TYPE LOOP INDEX CAUSES SLOWDOWN AT O3

XL C/C++ for Linux Fix Pack 9 (August 2013 Update) for 11.1
June 2012 Update for XL C/C++ for Linux, V11.1
October 2012 Update for XL C/C++ for Linux, V11.1
XL C/C++ for Linux Fix Pack 8 (June 2013 Update) for 11.1

APAR status

• Closed as program error.

Error description

• Consider the following testcase:
  
  $cat t.c
  ////////////////////////////////////////
  void foo(double *x)
  {
    TYPE i;
    for (i = 0; i < 100; i++)
      *x++ += 1.1;
  }
  
  double z[1000];
  
  int main()
  {
    int i;
    memset(z,0,sizeof(z));
    for (i = 0; i < 60*1000*1000; i++)
      foo(z);
    return 0;
  }
  ////////////////////////////////////////
  
  Compile with:
  
  xlc t.c -q64 -DTYPE=long -O
  xlc t.c -q64 -DTYPE=long -O3
  
  and
  
  xlc t.c -q64 -DTYPE=int -O
  xlc t.c -q64 -DTYPE=int -O3
  
  The executable compiled with -O3 and -DTYPE=long exhibits worse
  runtime performance than the one compiled with lower level of
  optimization (-O). This regression does not occur with
  -DTYPE=int.
  

Local fix

• N/A
  

Problem summary

• Performance degradation due to over-aggressive unrolling
  

Problem conclusion

• Fixed a potential performance problem due to complete unrolling
  of high iteration count loops.
  

Temporary fix

Comments

APAR Information

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

  IV09175

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

Fix information

• Fixed component name

  XL C/C++ FOR LI

• Fixed component ID

  5724X1400

Applicable component levels

• RB10 PSN IV09175

     UP06/09/13