IV81773: XLC OPTIMIZES AWAY LOOP WITH VOLATILE VARIABLE

XL C/C++ for AIX Fix Pack 6 (February 2016 PTF) for 13.1
XL C for AIX Fix Pack 6 (February 2016 Update) for 13.1
XL C for AIX Fix Pack 7 (May 2016 Update) for 13.1
XL C/C++ for AIX Fix Pack 7 (May 2016 PTF) for 13.1
XL C for AIX Fix Pack 8 (August 2016 Update) for 13.1
XL C/C++ for AIX Fix Pack 8 (August 2016 PTF) for 13.1
XL C for AIX Fix Pack 9 (June 2017 Update) for 13.1
XL C/C++ for AIX Fix Pack 9 (June 2017 PTF) for 13.1
XL C for AIX Fix Pack 10 (February 2018 Update) for 13.1
XL C/C++ for AIX Fix Pack 10 (February 2018 PTF) for 13.1
XL C/C++ for AIX Fix Pack 11 (November 2018 PTF) for 13.1
XL C for AIX Fix Pack 11 (November 2018) for 13.1

APAR status

• Closed as program error.

Error description

• When optimization is enabled, xlc optimizes away a loop with
  volatile variable.
  
  OPT code of LOCK() [ the inner while loop is optimized away,
  although it contains volatile variable p_o4trc_header->who_pid]
  
  ie
  
  From:
  void LOCK(int my_pid)
  {
      int n=0, z=0, y=-1;
      do {
          z=0;
          while(p_o4trc_header->who_pid != 0 && z<1000000) z+=1;
          y++;
      } while (
  _check_lock((int*)&p_o4trc_header->who_pid,n,my_pid) );
      printf("inner loop count %d; outer loop count %d\n", z,y);
      return;
  }
  
  To:
  The Machne code behaves as a C code written like:
  
  void LOCK(int my_pid)
  {
      int n=0, z=0, y=-1;
      do {
          if( p_o4trc_header->who_pid != 0 ) z = 1000000 else
  z=0;
          y++;
      } while (
  _check_lock((int*)&p_o4trc_header->who_pid,n,my_pid) );
      printf("inner loop count %d; outer loop count %d\n", z,y);
      return;
  }
  
  
  
  0x100000880 (LOCK(int))      fbe1fff8         std   r31,-8(r1)
  $r31 = &p_o4trc_header;
  0x100000884 (LOCK(int)+0x4)  fbc1fff0         std   r30,-16(r1)
  0x100000888 (LOCK(int)+0x8)  3be0ffff          li   r31,-1
  0x10000088c (LOCK(int)+0xc)  3c80000f         lis   r4,0xf
  0x100000890 (LOCK(int)+0x10) 7c0802a6        mflr   r0
  0x100000894 (LOCK(int)+0x14) fba1ffe8         std   r29,-24(r1)
  0x100000898 (LOCK(int)+0x18) fb81ffe0         std   r28,-32(r1)
  0x10000089c (LOCK(int)+0x1c) fb61ffd8         std   r27,-40(r1)
  0x1000008a0 (LOCK(int)+0x20) 3bc2ff40        addi   r30,-192(r2)
  0x1000008a4 (LOCK(int)+0x24) 7c7d07b4       extsw   r29,r3
  $r29 = my_pid;
  0x1000008a8 (LOCK(int)+0x28) 3b644240        addi
  r27,0x4240(r4) $r27 = 1000000;
  0x1000008ac (LOCK(int)+0x2c) f8010010         std   r0,0x10(r1)
  0x1000008b0 (LOCK(int)+0x30) f821ff61        stdu   r1,-160(r1)
  Store stack pointer
  0x1000008b4 (LOCK(int)+0x34) 60000000         ori   r0,r0,0x0
  0x1000008b8 (LOCK(int)+0x38) 60000000         ori   r0,r0,0x0
  0x1000008bc (LOCK(int)+0x3c) 60210000         ori   r1,r1,0x0
  0x1000008c0 (LOCK(int)+0x40) e87e0000   |-->   ld   r3,0x0(r30)
  $r3 = p_o4trc_header;
  0x1000008c4 (LOCK(int)+0x44) 3bff0001   |    addi   r31,0x1(r31)
  y = y +1;
  0x1000008c8 (LOCK(int)+0x48) 38800000   |      li   r4,0x0
  $r4 = 0;
  0x1000008cc (LOCK(int)+0x4c) 63a50000   |     ori   r5,r29,0x0
  $r5 = mypid;
  0x1000008d0 (LOCK(int)+0x50) e8030002   |     lwa   r0,0x0(r3)
  $r0 = p_o4trc_header->who_pid;
  0x1000008d4 (LOCK(int)+0x54) 7cc000d0   |     neg   r6,r0
  ($r0 == 0) $r6 = 0 : $r6 = -$r0;
  0x1000008d8 (LOCK(int)+0x58) 7c003378   |      or   r0,r0,r6
  ($r0 == 0) $r0 = 0 : $r0 = negativ number;
  0x1000008dc (LOCK(int)+0x5c) 7c06fe76   |   sradi   r6,r0,0x3f
  ($r0 < 0) $r6 = -1 : $r6 = 0;
  0x1000008e0 (LOCK(int)+0x60) 7cdcd838   |     and   r28,r6,r27
  if (p_o4trc_header->who_pid == 0) z = $r28 = 0; else z = $r28 =
  1000000;
  0x1000008e4 (LOCK(int)+0x64) 48003423   |     bla
  0x0000000000003420 _check_lock($r3=&p_o4trc_header->who_pid,
  $r4=0, $r5 = my_pid)
  0x1000008e8 (LOCK(int)+0x68) 2c030000   |    cmpi
  cr0,0x0,r3,0x0
  0x1000008ec (LOCK(int)+0x6c) 4082ffd4   |<--  bne   0x1000008c0
  (LOCK(int)+0x40
  0x1000008f0 (LOCK(int)+0x70) 7fe507b4       extsw   r5,r31
  $r5 = y;
  0x1000008f4 (LOCK(int)+0x74) e86200f0          ld   r3,0xf0(r2)
  0x1000008f8 (LOCK(int)+0x78) 63840000         ori   r4,r28,0x0
  $r4 = z;
  0x1000008fc (LOCK(int)+0x7c) 48000305          bl   0x100000c00
  (printf)
  0x100000900 (LOCK(int)+0x80) e8410028          ld   r2,0x28(r1)
  0x100000904 (LOCK(int)+0x84) e98100b0          ld   r12,0xb0(r1)
  0x100000908 (LOCK(int)+0x88) 382100a0        addi   r1,0xa0(r1)
  0x10000090c (LOCK(int)+0x8c) 7d8803a6        mtlr   r12
  0x100000910 (LOCK(int)+0x90) eb61ffd8          ld   r27,-40(r1)
  0x100000914 (LOCK(int)+0x94) eb81ffe0          ld   r28,-32(r1)
  0x100000918 (LOCK(int)+0x98) eba1ffe8          ld   r29,-24(r1)
  0x10000091c (LOCK(int)+0x9c) ebc1fff0          ld   r30,-16(r1)
  0x100000920 (LOCK(int)+0xa0) ebe1fff8          ld   r31,-8(r1)
  0x100000924 (LOCK(int)+0xa4) 4e800020         blr
  
  
  =====Testcase:
  $ cat test.c
  #include <pthread.h>
  #include <sys/atomic_op.h>
  extern "C" void exit ( int s);
  
  
  #include <stdio.h>
  #include <sys/stat.h>
  #include <sys/types.h>
  #include <sys/shm.h>
  #include <unistd.h>
  
  
  #pragma pack(1)
  typedef struct _o4trc_header_t
  {
      /* offset 0x00:
       * this is lock */
       pid_t  who_pid;
      /* this is data */
       pid_t  data_pid;
  
  } o4trc_header_t;
  #pragma pack(pop)
  
  static volatile o4trc_header_t* p_o4trc_header = 0;
  
  static void o4trc_shm ( void )
  {
      int key, id = -1;
      bool created = false;
  
  
      key = 44444; // some key
      /* first, try to get existing shared memory.... */
      id = shmget ( key, 1024, S_IRUSR | S_IWUSR | S_IRGRP |
  S_IWGRP | S_IROTH | S_IWOTH );
  
      if ( id == -1 )
      {
          /* no shared memory with this key exists, so we create
  it... */
          errno = 0;
          id = shmget ( key, 1024, S_IRUSR | S_IWUSR | S_IRGRP |
  S_IWGRP | S_IROTH | S_IWOTH | IPC_CREAT );
  
          if ( id == -1 )
              exit(-1);
          else created = true;
      }
      /* the shared memory already existed */
      else created = false;
  
      /* attach to shared memory */
      errno = 0;
      p_o4trc_header = (o4trc_header_t*) shmat ( id, NULL, 0 );
      if ( ! p_o4trc_header )
          exit(-1);
  
      /* init if needed */
      if ( created )
      {
          /* initialize with default values */
          p_o4trc_header->data_pid = 0;
          _clear_lock((int*)&p_o4trc_header->who_pid, 0);
      }
  
      return;
  
  } /* o4trc_shm */
  
  void LOCK(int my_pid)
  {
      int n=0, z=0, y=-1;
      do {
          z=0;
          while(p_o4trc_header->who_pid != 0 && z<1000000) z+=1;
          y++;
      } while (
  _check_lock((int*)&p_o4trc_header->who_pid,n,my_pid) );
      printf("inner loop count %d; outer loop count %d\n", z,y);
      return;
  }
  void RELEASE(void)
  {
      _clear_lock((int*)&p_o4trc_header->who_pid, 0);
      return;
  }
  
  // erst ohne Parameter aufrufen, dann 1 Sekunde sp&#228;ter mit einem
  dummy parameter aufrufen
  int main (int argc, char** argv)
  {
      pid_t my_pid = getpid (), lock_pid, check_pid;
      int n,x, y, z=0;
  
      // get shared memory
      if (!p_o4trc_header) o4trc_shm();
      if (!p_o4trc_header) return 1;
  
      LOCK(my_pid);
  
      RELEASE();
  
      return 0;
  }
  $
  
  ===== COMPILE COMMAND:
  xlC_r -+ -g -D_AIX51 -D_LARGE_FILES -U__unix -q64
  -qtbtable=full -qhalt=e -qalias=noansi -qutf -qcpluscmt  -qfdpr
  -qstrict -qlibansi -qinlglue -qmaxmem=-1 -DNDEBUG -qarch=pwr4
  -qtune=balanced -qspill=2560 -qsaveopt -O
  -qipa=relink:level=0:partition=large -r -qdebug=ndf_bc
  -qlonglong -qalign=natural -qldbl128 -qlanglvl=extc89  -c -o
  test.o test.c
  

Local fix

• N/A
  

Problem summary

• USERS AFFECTED:
  Users who compile with IPA may be affected by this issue.
  
  PROBLEM DESCRIPTION:
  The IPA optimization was incorrectly optimizing away volatile
  variables due to incorrect information in an internal hash
  function.
  

Problem conclusion

• The problem has been resolved.
  

Temporary fix

Comments

APAR Information

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

  IV78575

• 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

• RD10 PSY