LI76787: SIGNAL 11 IN IPA64 AT -O3

June 2012 Update for XL Fortran for Linux, V13.1
October 2012 Update for XL Fortran for Linux, V13.1
XL Fortran for Linux Fix Pack 8 (June 2013 Update) for 13.1

APAR status

• Closed as program error.

Error description

• IPA64 abends when compiling the following test case at -O3.
  
  ===== Compile Command:
  xlf90_r -q64 -qfixed=72 -O3 -qmaxmem=-1 -qdbg -qfullpath
  -qwarn64 -qstrict -qfloat=rrm:norsqrt:nofold -qreport
  -qflttrap=ov:zero:inv:imp:enable -qsigtrap -qsuppress=1501-245
  -qalias=intptr -qreport=smplist -qextname -qintsize=8
  -qrealsize=8 -qarch=pwr7 -qtune=pwr7 -qsimd=auto -qnocheck
  -qspillsize=32648 -qnoescape -qxflag=ngenstub
  -bmaxstack:0xF0000000 -qnoundef -qsmp=omp -qnosave
  -qtbtable=full -qxref -qattr -bnoquiet -bdatapsize:64K
  -bstackpsize:64K -btextpsize:64K -qsource -c mieint.F
  
  
  ===== Test case:
  $ cat mieint.F
  
        Subroutine MieInt(Dx, SCm, Inp, Dqv, Dqxt, Dqsc, Dbsc, Dg,
  Xs1,
       1                  Xs2, DPh, Error)
  
        Integer * 2  Imaxx
        Parameter (Imaxx = 12000)
        Real * 4     RIMax          ! largest real part of
  refractive inde
        Parameter (RIMax = 2.5)
        Real * 4     IRIMax         ! largest imaginary part of
  refractive
        Parameter (IRIMax = -2)
        Integer * 2  Itermax
        Parameter (Itermax = 12000 * 2.5)
                                  ! must be large enough to cope
  with the
                                  ! largest possible nmx = x *
  abs(scm) +
                                  ! or nmx =  Dx +
  4.05*Dx**(1./3.) + 2.0
        Integer * 2  Imaxnp
        Parameter (Imaxnp = 10000) ! Change this as required
  !     INPUT
        Real * 8     Dx
        Complex * 16  SCm
        Integer * 4  Inp
        Real * 8     Dqv(Inp)
  !     OUTPUT
        Complex * 16  Xs1(InP)
        Complex * 16  Xs2(InP)
        Real * 8     Dqxt
        Real * 8     Dqsc
        Real * 8     Dg
        Real * 8     Dbsc
        Real * 8     DPh(InP)
        Integer * 4  Error
  !     LOCAL
        Integer * 2  I
        Integer * 2  NStop
        Integer * 2  NmX
        Integer * 4  N    ! N*N > 32767 ie N > 181
        Integer * 4  Inp2
        Real * 8     Chi,Chi0,Chi1
        Real * 8     APsi,APsi0,APsi1
        Real * 8     Pi0(Imaxnp)
        Real * 8     Pi1(Imaxnp)
        Real * 8     Taun(Imaxnp)
        Real * 8     Psi,Psi0,Psi1
        Complex * 8  Ir
        Complex * 16 Cm
        Complex * 16 A,ANM1,APB
        Complex * 16 B,BNM1,AMB
        Complex * 16 D(Itermax)
        Complex * 16 Sp(Imaxnp)
        Complex * 16 Sm(Imaxnp)
        Complex * 16 Xi,Xi0,Xi1
        Complex * 16 Y
  !     ACCELERATOR VARIABLES
        Integer * 2  Tnp1
        Integer * 2  Tnm1
        Real * 8     Dn
        Real * 8     Rnx
        Real * 8     S(Imaxnp)
        Real * 8     T(Imaxnp)
        Real * 8     Turbo
        Real * 8     A2
        Complex * 16 A1
  
        If ((Dx.Gt.Imaxx) .Or. (InP.Gt.ImaxNP)) Then
          Error = 1
          Return
        EndIf
        Cm = SCm
        Ir = 1 / Cm
        Y =  Dx * Cm
        If (Dx.Lt.0.02) Then
           NStop = 2
        Else
           If (Dx.Le.8.0) Then
              NStop = Dx + 4.00*Dx**(1./3.) + 2.0
           Else
              If (Dx.Lt. 4200.0) Then
                 NStop = Dx + 4.05*Dx**(1./3.) + 2.0
              Else
                 NStop = Dx + 4.00*Dx**(1./3.) + 2.0
              End If
           End If
        End If
        NmX = Max(Real(NStop),Real(Abs(Y))) + 15.
        If (Nmx .gt. Itermax) then
            Error = 1
            Return
        End If
        Inp2 = Inp+1
        D(NmX) = Dcmplx(0,0)
        Do N = Nmx-1,1,-1
           A1 = (N+1) / Y
           D(N) = A1 - 1/(A1+D(N+1))
        End Do
        Do I =1,Inp2
           Sm(I) = Dcmplx(0,0)
           Sp(I) = Dcmplx(0,0)
           Pi0(I) = 0
           Pi1(I) = 1
        End Do
        Psi0 = Cos(Dx)
        Psi1 = Sin(Dx)
        Chi0 =-Sin(Dx)
        Chi1 = Cos(Dx)
        APsi0 = Psi0
        APsi1 = Psi1
        Xi0 = Dcmplx(APsi0,Chi0)
        Xi1 = Dcmplx(APsi1,Chi1)
        Dg = 0
        Dqsc = 0
        Dqxt = 0
        Tnp1 = 1
        Do N = 1,Nstop
           DN = N
           Tnp1 = Tnp1 + 2
           Tnm1 = Tnp1 - 2
           A2 = Tnp1 / (DN*(DN+1D0))
           Turbo = (DN+1D0) / DN
           Rnx = DN/Dx
           Psi = Dble(Tnm1)*Psi1/Dx - Psi0
           APsi = Psi
           Chi = Tnm1*Chi1/Dx       - Chi0
           Xi = Dcmplx(APsi,Chi)
           A = ((D(N)*Ir+Rnx)*APsi-APsi1) / ((D(N)*Ir+Rnx)*  Xi-
  Xi1)
           B = ((D(N)*Cm+Rnx)*APsi-APsi1) / ((D(N)*Cm+Rnx)*  Xi-
  Xi1)
           Dqxt = Tnp1 *      Dble(A + B)          + Dqxt
           Dqsc = Tnp1 * (A*Conjg(A) + B*Conjg(B)) + Dqsc
           If (N.Gt.1) then
              Dg = Dg + (dN*dN - 1) * Dble(ANM1*Conjg(A) + BNM1 *
       1           Conjg(B)) / dN + TNM1 * Dble(ANM1*Conjg(BNM1))
       2         / (dN*dN - dN)
           End If
           Anm1 = A
           Bnm1 = B
           APB = A2 * (A + B)
           AMB = A2 * (A - B)
           Do I = 1,Inp2
              If (I.GT.Inp) Then
                 S(I) = -Pi1(I)
              Else
                 S(I) = Dqv(I) * Pi1(I)
              End If
              T(I) = S(I) - Pi0(I)
              Taun(I) = N*T(I) - Pi0(I)
              Sp(I) = APB * (Pi1(I) + Taun(I)) + Sp(I)
              Sm(I) = AMB * (Pi1(I) - Taun(I)) + Sm(I)
              Pi0(I) = Pi1(I)
              Pi1(I) = S(I) + T(I)*Turbo
           End Do
           Psi0 = Psi1
           Psi1 = Psi
           Apsi1 = Psi1
           Chi0 = Chi1
           Chi1 = Chi
           Xi1 = Dcmplx(APsi1,Chi1)
        End Do
        If (Dg .GT.0) Dg = 2 * Dg / Dqsc
        Dqsc =  2 * Dqsc / Dx**2
        Dqxt =  2 * Dqxt / Dx**2
        Do I = 1,Inp
           Xs1(I) = (Sp(I)+Sm(I)) / 2
           Xs2(I) = (Sp(I)-Sm(I)) / 2
           Dph(I) = 2 * Dble(Xs1(I)*Conjg(Xs1(I)) +
  Xs2(I)*Conjg(Xs2(I)))
       1          / (Dx**2 * Dqsc)
        End Do
        Dbsc = 4 * Abs(( (Sp(Inp2)+Sm(Inp2))/2 )**2) / Dx**2
        Error = 0
        Return
        End subroutine MieInt
  $
  
  ===== ACTUAL OUTPUT:
  $xlf90_r -q64 -qfixed=72 -O3 -qmaxmem=-1 -qdbg -qfullpath
  -qwarn64 -qstrict -qfloat=rrm:norsqrt:nofold -qreport
  -qflttrap=ov:zero:inv:imp:enable -qsigtrap -qsuppress=1501-245
  -qalias=intptr -qreport=smplist -qextname -qintsize=8
  -qrealsize=8 -qarch=pwr7 -qtune=pwr7 -qsimd=auto -qnocheck
  -qspillsize=32648 -qnoescape -qxflag=ngenstub
  -bmaxstack:0xF0000000 -qnoundef -qsmp=omp -qnosave
  -qtbtable=full -qxref -qattr -bnoquiet -bdatapsize:64K
  -bstackpsize:64K -btextpsize:64K -qsource -c mieint.F
  ** mieint   === End of Compilation 1 ===
  The assert subroutine failed: lit_len <= 8 || (lit_len == 16 &&
  dt == DT_COMPLEX), file literal.cpp, line 2738
  Calling signal handler...
  1586-494 (U) INTERNAL COMPILER ERROR: Signal 6.
  /gsa/tlbgsa/projects/x/xlcmpbld/run/xlf/131/aix/solution/111020/
  usr/bin/.orig/xlf90_r: 1501-230 (S) Internal compiler error;
  please contact your Service Representative. For more information
  visit:
  http://www.ibm.com/support/docview.wss?uid=swg21110810
  1501-511  Compilation failed for file mieint.F.
  $
  
  
  ===== EXPECTED OUTPUT:
  Should compile clean
  

Local fix

• Use -O2 instead of -O3.
  

Problem summary

• USERS AFFECTED:
  XL Fortran users using O3 and up may potentially be affected by
  this issue.
  
  PROBLEM DESCRIPTION:
  A specific internal representation of literals was not designed
  to be supported in certain scenarios.
  

Problem conclusion

• The specific internal representation of literals is now
  supported via this fix.
  

Temporary fix

Comments

APAR Information

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

  IV13177

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

Fix information

• Fixed component name

  XL FORTRAN LINU

• Fixed component ID

  5724X1600

Applicable component levels

• RD10 PSN IV13177

     UP06/09/13