[Bug tree-optimization/14741] graphite with loop blocking and interchanging doesn't optimize a matrix multiplication loop

[spop at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14741

--- Comment #24 from Sebastian Pop <spop at gcc dot gnu.org> ---
Looking at t.f90.003t.original (the first dump file of -fdump-tree-all-all)
I see that the array c is already in linear form:

(*cD.1876)[((integer(kind=8)D.9) jD.1897 * stride.12D.1892 + offset.13D.1893) +
(integer(kind=8)D.9) iD.1896] = (*cD.1876)[((integer(kind=8)D.9) jD.1897 *
stride.12D.1892 + offset.13D.1893) + (integer(kind=8)D.9) iD.1896] +
(*aD.1874)[((integer(kind=8)D.9) kD.1898 * stride.2D.1880 + offset.3D.1881) +
(integer(kind=8)D.9) iD.1896] * (*bD.1875)[((integer(kind=8)D.9) jD.1897 *
stride.7D.1886 + offset.8D.1887) + (integer(kind=8)D.9) kD.1898];

Can the Fortran front-end build multi-dimensional arrays instead of this linear
form?