LI76663: REDUNDANT STORES AND LOADS FOR STOREBCD_URG

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

• The compiler generates redundant stores and loads for the
  storeBCD_urg() built-in which leads to a performance degradation
  at runtime.
  
  The following test case demonstrates the issue:
  
  ===== COMPILE COMMAND:
  xlC -q64 -O -qarch=pwr7 -qdfp -c load_store_bcd31.C
  
  ===== TESTCASE:
  $cat load_store_bcd31.C
  #include <builtins.h>
  
  
  extern "C" void storeBCD(_Decimal128 dfp, unsigned char *cp,
  unsigned
  long long* op)
  {
     __d128_to_signed_BCD (dfp, 1, (unsigned long long *)cp,
  (unsigned
  long long *)(cp+8));
  }
  
  extern "C" void storeBCD_urg(_Decimal128 dfp, unsigned char *cp,
  unsigned long long* op)
  {
     unsigned long long upper;
     unsigned long long lower;
     __d128_to_signed_BCD (dfp, 1, &upper, &lower);
     *((unsigned long long *)cp) = upper;
     *((unsigned long long *)(cp+8)) = lower;
  }
  
  extern "C" _Decimal128 loadBCD(unsigned char *cp)
  {
     return __signed_BCD_to_d128(*(unsigned long long *)cp,
  *(unsigned
  long long *)(cp+8));
  }
  $
  
  
  ======= ACTUAL LISTING OUTPUT:
   GPR's set/used:   su-s -u-- ---- ----  ---- ---- ---- ----
   FPR's set/used:   ssuu ---- ---- ----  ---- ---- ---- ----
   CCR's set/used:   ---- ----
    VR's set/used:   ---- ---- ---- ----  ---- ---- ---- ----
       | 000000                           PDEF     storeBCD_urg
      0|                                  PROC
  dfp,cp,op,vs2,vs3,gr5,gr6
     16| 000020 ddedpdq  FC181284   1     CVDFESB
  vs0,vs1=vs2,vs3,1
     16| 000024 stfd     D801FFD0   1     STFL
  #MX_WORK2_0(gr1,-48)=vs0
     16| 000028 stfd     D821FFE0   1     STFL
  #MX_WORK2_1(gr1,-32)=vs1
     16| 00002C ld       E801FFD0   1     L8
  gr0=#MX_WORK2_0(gr1,-48)
     16| 000030 ld       E861FFE0   1     L8
  gr3=#MX_WORK2_1(gr1,-32)
     16| 000034 std      F861FFC0   1     ST8
  lower(gr1,-64)=gr3
     16| 000038 std      F801FFC8   1     ST8
  upper(gr1,-56)=gr0
     17| 00003C std      F8050000   1     ST8       (unsigned long
  long;unsigned char)(gr5,0)=gr0
     18| 000040 std      F8650008   1     ST8       (unsigned long
  long;unsigned char)(gr5,8)=gr3
     19| 000044 bclr     4E800020   1     BA        lr
       |               Tag Table
  
  
  ======= EXPECTED LISTING OUTPUT:
   GPR's set/used:   su-s -u-- ---- ----  ---- ---- ---- ----
   FPR's set/used:   ssuu ---- ---- ----  ---- ---- ---- ----
   CCR's set/used:   ---- ----
    VR's set/used:   ---- ---- ---- ----  ---- ---- ---- ----
       | 000000                           PDEF     storeBCD_urg
      0|                                  PROC
  dfp,cp,op,vs2,vs3,gr5,gr6
     16| 000020 ddedpdq  FC181284   1     CVDFESB
  vs0,vs1=vs2,vs3,1
     16| 000024 stfd     D821FFE0   1     STFL
  lower(gr1,-32)=vs1
     16| 000028 stfd     D801FFE8   1     STFL
  upper(gr1,-24)=vs0
     17| 00002C ld       E801FFE8   1     L8
  gr0=upper(gr1,-24)
     18| 000030 ld       E861FFE0   1     L8
  gr3=lower(gr1,-32)
     17| 000034 std      F8050000   1     ST8       (unsigned long
  long;unsigned char)(gr5,0)=gr0
     18| 000038 std      F8650008   1     ST8       (unsigned long
  long;unsigned char)(gr5,8)=gr3
     19| 00003C bclr     4E800020   1     BA        lr
       |               Tag Table
  

Local fix

• n/a
  

Problem summary

• USERS AFFECTED:
  Users using -qdfp with the storeBCD_urg function and
  optimization maybe affected by this issue.
  
  PROBLEM DESCRIPTION:
  The compiler is generating redundant stores and loads for the
  storeBCD_urg function.
  

Problem conclusion

• The compiler has now been fixed to avoid generating redundant
  stores and loads for storeBCD_urg.
  

Temporary fix

Comments

APAR Information

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

  IV10080

• 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 IV10080

     UP06/09/13