Problem with shape of arrays using reallocation feature with TRANSPOSE(RESHAPE(..))

[bpichon]

Dear Fortran !

Here a possible bug with intel compilers (both Windows and Linux OS)
I don't see previous post for this problem but sorry in the converse case.

It involves :
   1)   TRANSPOSE(RESHAPE(...))             (yes, why not !)
   2)   Fortran 2003/2008 (re)allocatation feature

What do you think about this ?

Bernard
=================================================================
With ifort 14.0.0074.11 on windows 8 (via visual studio 2012)

 V(6)=[1,2,3,4,5,6]
 1 2 3 4 5 6

  D = RESHAPE(V,[3,2])
 1 2 3 4 5 6

  E = TRANSPOSE( RESHAPE(V,[3,2]) ) in one step :
            WRONG (E has not the right shape ...)
 3 2   |   2 3
 1 4 -858993460 2 5 -858993460

  F = TRANSPOSE(D)  with  D = RESHAPE(V,[3,2])  in two steps : Ok
 2 3
 1 4 2 5 3 6

  G = RESHAPE( V, Shape=[2,3] , Order=[2,1] ) in one step
                       with keyword order of RESHAPE : Ok
 2 3
 1 4 2 5 3 6

==================================================================
With ifort 14.0.0 (20130728)  on Linux (ifort with -assume realloc !!)

  V(6)=[1,2,3,4,5,6]
           1           2           3           4           5           6

  D = RESHAPE(V,[3,2])
           1           2           3           4           5           6

  E = TRANSPOSE( RESHAPE(V,[3,2]) ) in one step :
            WRONG (E has not the right shape ...)
           3           2   |             2           3
           1           4          18           2           5           0

  F = TRANSPOSE(D)  with  D = RESHAPE(V,[3,2])  in two steps : Ok
           2           3
           1           4           2           5           3           6

  G = RESHAPE( V, Shape=[2,3] , Order=[2,1] ) in one step
                       with keyworf order of RESHAPE : Ok
           2           3
           1           4           2           5           3           6

[Steven_L_Intel1]

Please reattach your test case - something went wrong.

[bpichon]

I try again as an attached file

[bpichon]

I don't know if i succeed

Here the text of the test :

program test

!ifort test.f90 -assume realloc_lhs

integer, Dimension(:,:), Allocatable :: C, D, E, F, G

integer, Dimension(6) :: V = [1,2,3,4,5,6]

Write(*,*) 'V(6)=[1,2,3,4,5,6]'

Write(*,*) V

Write(*,*)

!

D = RESHAPE(V,[3,2])

Write(*,*) " D = RESHAPE(V,[3,2]) "

Write(*,*) D

Write(*,*)

!

E = TRANSPOSE( RESHAPE(V,[3,2]) )

Write(*,*) " E = TRANSPOSE( RESHAPE(V,[3,2]) ) in one step : "

Write(*,*) " WRONG (E has not the right shape ...) "

Write(*,*) SHAPE(E), ' | ', SHAPE( TRANSPOSE( RESHAPE(V,[3,2]) ) ) ! very curious

Write(*,*) E

Write(*,*)

!

F = TRANSPOSE(D)

Write(*,*) " F = TRANSPOSE(D) with D = RESHAPE(V,[3,2]) in two steps : Ok "

Write(*,*) SHAPE(F)

Write(*,*) F

Write(*,*)

!

G = RESHAPE( V, Shape=[2,3] , Order=[2,1] )

Write(*,*) " G = RESHAPE( V, Shape=[2,3] , Order=[2,1] ) in one step "

Write(*,*) " with keyword order of RESHAPE : Ok "

Write(*,*) SHAPE(G)

Write(*,*) G

Write(*,*)

end program test

[m_sulc]

Ah, yes, with ifort version 14.0.0.080 Build 20130728, the statement
[fortran]
E = TRANSPOSE( RESHAPE(V,[3,2]) )
[/fortran]
produces incorrect results (moreover dependent on the level of optimization)...

[Steven_L_Intel1]

I can reproduce this in 14.0, but not in our next-release development stream. I have asked the developers to see if an explicit fix for this was made and if it could be added to 14.0. Issue ID is DPD200250177.

[Steven_L_Intel1]

The developers identified the fix and said it will go into Update 2, planned for January.