LI76635: INEFFICIENT REGISTER MOVE USING XVCPSGND

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

• When analyzing listing files that contain "xvcpsgnd"
  instructions, the compiler should be generating more efficient
  code for the that instruction.
  
  
  Compiler uses register moves xvcpsgndp. But instead, it can
  make use of the xxlor and xxpermdi instructions.
  xxlor only takes 2c (can run only on pipe 0).
  xxpermdi takes 3c (can run only on pipe 1).
  xvcpsgnd always takes 7c and can run on both pipes.
  ==> A register move with with xxlor is much faster.
  ==> 2 register moves can be implemented with xxlor + xxpermdi
  
  Moving a register pair can be required for duplicating
  _Decimal128 values.
  
  So, the following instructions:
  1435| 000044 xvcpsgnd F0021780   1     LRVS      vs0=vs2
  1435| 000048 xvcpsgnd F0231F80   1     LRVS      vs1=vs3
  
  can be replaced with:
  
  xxlor vs0,vs2,vs2
  xxpermdi vs1,vs3,vs3,0
  

Local fix

• n/a
  

Problem summary

• Users Affected:
  Users who use _Decimal128 values which internally makes use of
  xvcpsgnd can be potentially affected by this issue.
  
  Problem Description:
  The compiler inefficiently creates xvcpsgnd instructions which
  can potentially be replaced with xxlor and xxpermdi
  instructions. However, this is not always the case, as it
  depends on which architecture the final exe is being executed
  on.
  

Problem conclusion

• Replacing xvcpsgnd with xxlor and xxpermdi is beneficial on
  PWR7. As a result, the compiler will now recognize such
  circumstances under the -qtune=pwr7 option and will replace
  xvcpsgnd with xxlor and xxpermdi.
  

Temporary fix

Comments

APAR Information

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

  IV12845

• 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 IV12845

     UP06/09/13