On Thu, Mar 31, 2011 at 5:25 PM, Eric Botcazou <ebotca...@adacore.com> wrote:
> Hi,
>
> the ACATS test cc3601a started to fail recently at -O2 after the various
> tweaks
> to the optimizer. This is an assertion failure in the gimplifier invoked from
> PRE because of a dangling PLACEHOLDER_EXPR in operand #2 of an ARRAY_REF.
>
> CCP2 turns the ARRAY_REF:
>
> D.2774_394 = MEM[(boolean[(size_type) <PLACEHOLDER_EXPR struct
> opt16__ab___XUP>.P_BOUNDS->LB0:<PLACEHOLDER_EXPR struct
> opt16__ab___XUP>.P_BOUNDS->UB0 >= <PLACEHOLDER_EXPR struct
> opt16__ab___XUP>.P_BOUNDS->LB0 ? (size_type) <PLACEHOLDER_EXPR struct
> opt16__ab___XUP>.P_BOUNDS->UB0 : (size_type) <PLACEHOLDER_EXPR struct
> opt16__ab___XUP>.P_BOUNDS->LB0 + -1] *)&ab1][D.2776_392]{lb: D.2758_393 sz:
> 1};
>
> into:
>
> D.2774_394 = MEM[(boolean[(size_type) <PLACEHOLDER_EXPR struct
> opt16__ab___XUP>.P_BOUNDS->LB0:<PLACEHOLDER_EXPR struct
> opt16__ab___XUP>.P_BOUNDS->UB0 >= <PLACEHOLDER_EXPR struct
> opt16__ab___XUP>.P_BOUNDS->LB0 ? (size_type) <PLACEHOLDER_EXPR struct
> opt16__ab___XUP>.P_BOUNDS->UB0 : (size_type) <PLACEHOLDER_EXPR struct
> opt16__ab___XUP>.P_BOUNDS->LB0 + -1] *)&ab1][D.2776_392]{lb: 0 sz: 1};
>
> i.e. it computes that operand #2 is 0. Later PRE re-creates the reference "by
> pieces" and explicitly drops the 0:
>
> if (genop2)
> {
> /* Drop zero minimum index. */
> if (tree_int_cst_equal (genop2, integer_zero_node))
> genop2 = NULL_TREE;
>
> so the gimplifier re-populates it, bringing back the PLACEHOLDER_EXPR.
>
> Fixed by not dropping the 0 in PRE. Tested on i586-suse-linux, OK for the
> mainline?
Ok.
Thanks,
Richard.
>
> 2011-03-31 Eric Botcazou <ebotca...@adacore.com>
>
> * tree-ssa-pre.c (create_component_ref_by_pieces_1) <ARRAY_REF>:
> Drop a zero minimum index only if it is redundant.
>
>
> 2011-03-31 Eric Botcazou <ebotca...@adacore.com>
>
> * gnat.dg/opt16.adb: New test.
>
>
> --
> Eric Botcazou
>
