IJ08104: INCORRECT CODE GEN FOR POWER8

XL C for AIX, V13.1.3.5 Fix Pack (July 2018)
XL C/C++ for AIX, V13.1.3.5 Fix Pack (July 2018)
XL C for AIX, V13.1.3.6 Fix Pack (May 2019)
XL C/C++ for AIX, V13.1.3.6 Fix Pack (May 2019)

APAR status

• Closed as program error.

Error description

• The resulting test case produces an incorrect output at
  runtime when compiled for POWER8:
  
  ===== COMPILE COMMAND:
  xlc -q32 -qdfp -O -qnoansialias -qarch=pwr8 a.c b.c -o a
  
  ===== TESTCASE:
  $ cat a.c
  #include <stdio.h>
  _Decimal128 a[5] = { 1, 2, 3, 4, 5 };
  int b[5];
  void f(_Decimal128 *a, int *b, int n);
  int main() {
       f(a, b, 5);
       printf("%d %d %d %d %d\n", b[0], b[1], b[2], b[3], b[4]);
       return 0;
  }
  $
  $ cat b.c
  void f(_Decimal128 *a, int *b, int n) {
       for (int i = 0; i < n; i++)
           b[i] = (int)a[i];
  }
  $
  
  
  ===== ACTUAL OUTPUT:
  $ ./a
  0 0 0 0 2147483647
  $
  
  
  ===== EXPECTED OUTPUT:
  $ ./a
  1 2 3 4 5
  $
  

Local fix

• N/A
  

Problem summary

• USERS AFFECTED:
  Clients who compile applications meant to run on POWER8 may be
  affected by this issue
  
  PROBLEM DESCRIPTION:
  This test case exposes an issue in one of the internal compiler
  optimizations.  A certain internal variable is sometimes a
  register copy and sometimes a transfer
  between float and vector registers. The problem is that we
  propagate a float register into a vector instruction and then
  fail to save and restore properly across a call.
  

Problem conclusion

• The internal optimization has been fixed.
  

Temporary fix

Comments

APAR Information

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

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

Fix information

• Fixed component name

  XL C FOR AIX

• Fixed component ID

  5725C7100

Applicable component levels

• RD13 PSY

     UP