http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14741
--- Comment #22 from Sebastian Pop <spop at gcc dot gnu.org> ---
Once we revert that patch, the remaining problem is that
graphite_can_represent_scev returns false on this scev:
{{(stride.12_14 + offset.13_15) + 1, +, stride.12_14}_1, +, 1}_2
the parameters are defined like this:
_12 = *n_11(D);
ubound.10_13 = (integer(kind=8)) _12;
stride.12_14 = MAX_EXPR <ubound.10_13, 0>;
offset.13_15 = ~stride.12_14;
#(Data Ref:
# bb: 5
# stmt: c__I_lsm.30_8 = *c_21(D)[_20];
# ref: *c_21(D)[_20];
# base_object: *c_21(D);
# Access function 0: {{(stride.12_14 + offset.13_15) + 1, +, stride.12_14}_1,
+, 1}_2
#)
#(Data Ref:
# bb: 6
# stmt: _29 = *a_28(D)[_27];
# ref: *a_28(D)[_27];
# base_object: *a_28(D);
# Access function 0: {{(stride.12_14 + offset.13_15) + 1, +, 1}_2, +,
stride.12_14}_3
#)
Why have these arrays been linearized?
