On 09/13/2016 11:39 PM, Eric Botcazou wrote:
Found that the first revision causing this problem (of course with DJGPP
related patches applied) is 239113. Native bootstrap with same DJGPP
related patches applied on top of revision 239112 succeeds without
problems.
[andris@ap gcc]$ git bisect good
4bfa373ed2bfd022628609751ff34b8e51e0c2b9 is the first bad commit
commit 4bfa373ed2bfd022628609751ff34b8e51e0c2b9
Author: kugan <kugan@138bc75d-0d04-0410-961f-82ee72b054a4>
Date: Thu Aug 4 04:20:01 2016 +0000
gcc/ChangeLog:
2016-08-04 Kugan Vivekanandarajah<kug...@linaro.org>
* tree-vrp.c (set_value_range): Use vrp_equiv_obstack with
BITMAP_ALLOC.
(add_equivalence): Likewise.
(get_value_range): Allocate value range with vrp_value_range_pool.
(vrp_initialize): Initialize vrp_equiv_obstack for equiv
allocation. (vrp_finalize): Relase vrp_equiv_obstack and
vrp_value_range_pool.
Ugh, this looks like a memory corruption issue.
Thanks for good hint.
Problem solved (fixed stack size for gnat1.exe was insufficient after mentioned change). Increasing
it causes problem to go away. I'll recheck with full native DJGPP bootstrap. Currently only quick
test done (by editing stack size of 3rd stage gnat1.exe).
Andris
