patch for merging graphite branch (before tuplification)

[Jack Howarth]

    The current gcc trunk with the proposed graphite patch supplemented with all
the subsequent patches to the graphite branch now builds usable binaries
for almost all of the fortran polyhedron benchmarks using the
-funroll-loops -O3 -floop-block -floop-interchange -fgraphite flags.
We have one remaining compile failure with these flags in channel.f90.
The compile line...

gfortran -funroll-loops -O3 -floop-strip-mine -fgraphite channel.f90 -o channel

...or...

gfortran -funroll-loops -O3 -floop-block -fgraphite channel.f90 -o channel

produces the compile error...

channel.f90: In function ‘sw’:
channel.f90:6: internal compiler error: vector VEC(loop_p,base) index domain error, in strip_mine_profitable_p at graphite.c:4321
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

...while the compile line...

gfortran -funroll-loops -O3 -floop-interchange -fgraphite channel.f90 -o channel

...produces the compile error of...

channel.f90: In function ‘sw’:
channel.f90:6: internal compiler error: in graphite_trans_bb_move_loop, at graphite.c:4017
Please submit a full bug report,
with preprocessed source if appropriate.
See <http://gcc.gnu.org/bugs.html> for instructions.

                              Jack