LI74216: INCORRECT OUTPUT WITH -QHOT

April 2009 Update for XL Fortran Advanced Edition for Blue Gene/L, V11.1
September 2010 Update for XL Fortran Advanced Edition for Blue Gene/L, V11.1

APAR status

• Closed as program error.

Error description

• The following test case produces an incorrect output with -qhot.
  
  ===== Compile Command=====
  blrts_f90 -O3 -qfixed -qarch=440d -qtune=440 -I
  /bgl/BlueLight/ppcfloor/bglsys/include -qsource -qlist
  -qreport=hotlist   -c test.f
  
  blrts_f90 test.o   -O3 -qhot
  -L/bgl/BlueLight/ppcfloor/bglsys/lib -lmpich.rts -lmsglayer.rts
  -lrts.rts -ldevices.rts   -o test
  
  
  ===== TESTCASE:
  ====Start of test.f======
        program test
        implicit REAL*8 (A-H,O-Z)
        PARAMETER(Nmax = 140)
        PARAMETER(Nsbt = 8)
        COMPLEX*16 c(NMAX,NMAX)
        COMPLEX*16 vprd(NMAX,NMAX)
        REAL*8 TDa(NMAX)
  c
        do i=1,NMAX
          TDa(i) = 1.0d0 - mod(i,2)*2.0d0
        enddo
  
        do i=1,Nmax
          do j=1,Nmax
            vprd(j,i) = DCMPLX(1.0d0,1.0d0)
          enddo
        enddo
  
        call
  test_init(c,vprd,TDa,4.0d0,2.0d0,1.0d0,0.5d0,NMAX,Nsbt)
  
        do i=1,Nsbt*2
          do j=1,Nsbt*2
            write(*,*) 'c(',j,',',i,')=',c(j,i)
          enddo
          write(*,*)
        enddo
  
        stop
        end
  
        subroutine test_reset(a,N)
        COMPLEX*16 a(N)
        integer n,i
        do i=1,n
          a(i) = 0.0D0
        enddo
        return
        end
  
  
  c@PROCESS nohot
        subroutine test_init(c,vprd,TDa,Rm0,Rm,p0,u0,NMAX,Nsbt)
        COMPLEX*16 c(NMAX,NMAX)
        COMPLEX*16 vprd(NMAX,NMAX)
        COMPLEX*16 w(NMAX,NMAX)
        REAL*8 TDa(NMAX),Rm0,Rm,p0,u0
        integer NMAX,Nsbt,i,j,k
  c
        call test_reset(c,Nmax*Nmax)
  c
        do i=1,2*Nsbt
          c(i,i) = -u0
        enddo
  c
        do i=1,Nsbt
          c(i,i+Nsbt)=(Rm0-0.5D0*Rm)*DSIGN(1.D0,TDa(i)) -
  p0*TDa(i)
        enddo
  c
        do i = 1, Nsbt
        do j = Nsbt+1, 2*Nsbt
          c(i,j) = c(i,j) + u0*vprd(i,j)
        enddo
        enddo
  c
        return
        end
  
  ========End of test.f============
  
  ===== ACTUAL OUTPUT:
  .
  .
  stdout[0]:  c( 1 , 9 )=
  (4.50000000000000000,0.500000000000000000) <--Notice the 4.5
  stdout[0]:  c( 2 , 9 )=
  (0.500000000000000000,0.500000000000000000)
  .
  .
  stdout[0]:  c( 1 , 11 )=
  (0.500000000000000000,0.500000000000000000)
  stdout[0]:  c( 2 , 11 )=
  (0.500000000000000000,0.500000000000000000)
  stdout[0]:  c( 3 , 11 )=
  (4.50000000000000000,0.500000000000000000) <--Notice the 4.5
  stdout[0]:  c( 4 , 11 )=
  (0.500000000000000000,0.500000000000000000)
  .
  .
  stdout[0]:  c( 5 , 13 )=
  (4.50000000000000000,0.500000000000000000) <--Notice the 4.5
  .
  .
  stdout[0]:  c( 6 , 15 )=
  (0.500000000000000000,0.500000000000000000)
  stdout[0]:  c( 7 , 15 )=
  (4.50000000000000000,0.500000000000000000) <--Notice the 4.5
  .
  
  
  
  ===== EXPECTED OUTPUT:
  .
  .
  stdout[0]:  c( 1 , 9 )=
  (-1.50000000000000000,0.500000000000000000) <--Should be -1.5
  stdout[0]:  c( 2 , 9 )=
  (0.500000000000000000,0.500000000000000000)
  .
  .
  stdout[0]:  c( 1 , 11 )=
  (0.500000000000000000,0.500000000000000000)
  stdout[0]:  c( 2 , 11 )=
  (0.500000000000000000,0.500000000000000000)
  stdout[0]:  c( 3 , 11 )=
  (-1.50000000000000000,0.500000000000000000) <--Should be -1.5
  stdout[0]:  c( 4 , 11 )=
  (0.500000000000000000,0.500000000000000000)
  .
  .
  stdout[0]:  c( 5 , 13 )=
  (-1.50000000000000000,0.500000000000000000) <--Should be -1.5
  .
  .
  stdout[0]:  c( 6 , 15 )=
  (0.500000000000000000,0.500000000000000000)
  stdout[0]:  c( 7 , 15 )=
  (-1.50000000000000000,0.500000000000000000) <--Should be -1.5
  

Local fix

• n/a
  

Problem summary

• Users with 440d or 450d (using the double
  hummer unit) maybe affected by this issue.
  
  There is a transformation where the compiler
  replaces pattern USEL with float compares with
  WFSEL where:
  WFSEL a, b, c represents a>=0 ? b:c
  

Problem conclusion

• The compiler has been fixed to substitute the correct
  trasnformation pattern.
  

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:

  LI74942

Fix information

• Fixed component name

  XL FORTRAN AE B

• Fixed component ID

  5799HJG00

Applicable component levels

• RB10 PSY

     UP