Re: [patch, fortran, RFC] First steps towards inlining matmul

["Dominique d'Humières" <dominiq@lps.ens.fr>]

> Le 10 avr. 2015 à 16:15, Dominique d'Humières <dominiq@lps.ens.fr> a écrit :
> 
> 
>> 4. is required, I need to look at some more corner cases where bugs may still be lurking.
> 

Compiling the polyhedron test rnflow.f90 after the patch gives an ICE

f951: internal compiler error: gfc_array_dimen_size(): Bad dimension

Reduced test extracted from the subroutine evlrnf:

      subroutine evlrnf (ptrs0t, nclsm, prnf0t) 
      real, dimension (1:nclsm,1:nclsm), intent (in) :: ptrs0t
      integer, intent (in)            :: nclsm ! nombre de classes
      real, dimension (1:nclsm,1:nclsm), intent (out):: prnf0t
      real, allocatable, dimension (:,:) :: utrsft ! probas up
      real, allocatable, dimension (:)   :: vwrkft
      real, allocatable, dimension (:)   :: vwrk2t
      ncls = nclsm
      allocate (utrsft (1:ncls,1:ncls))
      allocate (vwrkft (1:ncls))
      allocate (vwrk2t (1:ncls))
         vwrk2t = matmul (utrsft, vwrkft)   ! U Fk
      deallocate (utrsft)
      deallocate (vwrk2t)
      deallocate (vwrkft)
      end subroutine evlrnf

Dominique