[Bug tree-optimization/51000] [4.7 Regression] ICE: in vect_get_store_cost, at tree-vect-stmts.c:923 on powerpc-apple-darwin9

[irar at il dot ibm.com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51000

Ira Rosen <irar at il dot ibm.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011-11-07
                 CC|                            |irar at il dot ibm.com,
                   |                            |jakub at redhat dot com
     Ever Confirmed|0                           |1

--- Comment #1 from Ira Rosen <irar at il dot ibm.com> 2011-11-07 06:58:10 UTC ---
Confirmed on powerpc64-suse-linux.
I think it was caused by

r180424 | jakub | 2011-10-25 10:02:08 +0200 (Tue, 25 Oct 2011) | 14 lines

        PR tree-optimization/50596
        * tree-vect-stmts.c (vect_mark_relevant): Only use
        FOR_EACH_IMM_USE_FAST if lhs is SSA_NAME.
        (vectorizable_store): If is_pattern_stmt_p look through
        VIEW_CONVERT_EXPR on lhs.
        * tree-vect-patterns.c (check_bool_pattern, adjust_bool_pattern):
        Use unsigned type instead of signed.
        (vect_recog_bool_pattern): Optimize also stores into bool memory in
        addition to casts from bool to integral types.
        (vect_mark_pattern_stmts): If pattern_stmt already has vinfo
        created, don't create it again.

        * gcc.dg/vect/vect-cond-10.c: New test.

Looking at gfortran.dg/alloc_comp_assign_2.f90, the original data-refs were in
an interleaving group, so only the first store had to be aligned. The pattern
data-refs are not marked as in interleaving group, so there are 3 unsupported
unaligned stores now.