http://bugzilla.opensuse.org/show_bug.cgi?id=1128794
http://bugzilla.opensuse.org/show_bug.cgi?id=1128794#c5
Richard Biener changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |IN_PROGRESS
Assignee|matz@suse.com |rguenther@suse.com
--- Comment #5 from Richard Biener ---
Thanks - I still cannot reproduce the segmentation fault but valgrind reports
the error:
==5976== Invalid read of size 8
==5976== at 0x10F7104: fma_node::get_parity()
(cortex-a57-fma-steering.c:517)
==5976== by 0x10F7396: fma_node::rename(fma_forest*)
(cortex-a57-fma-steering.c:598)
==5976== by 0x10F7553: rename_fma_node(fma_forest*, fma_node*)
(cortex-a57-fma-steering.c:652)
==5976== by 0x10F7C5D: func_fma_steering::dfs(void (*)(fma_forest*), void
(*)(fma_forest*, fma_root_node*), void (*)(fma_forest*, fma_node*), bool)
(cortex-a57-fma-steering.c:880)
==5976== by 0x10F8138: func_fma_steering::rename_fma_trees()
(cortex-a57-fma-steering.c:1004)
==5976== by 0x10F8265: func_fma_steering::execute_fma_steering()
(cortex-a57-fma-steering.c:1034)
==5976== by 0x10F84F9: pass_fma_steering::execute(function*)
(cortex-a57-fma-steering.c:1069)
==5976== by 0xC7FE9A: execute_one_pass(opt_pass*) (passes.c:2465)
...
==5976== Address 0x76c0ce8 is 24 bytes inside a block of size 48 free'd
==5976== at 0x4C2B84C: operator delete(void*) (in
/usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
==5976== by 0x10F7D23: func_fma_steering::dfs(void (*)(fma_forest*), void
(*)(fma_forest*, fma_root_node*), void (*)(fma_forest*, fma_node*), bool)
(cortex-a57-fma-steering.c:894)
==5976== by 0x10F8138: func_fma_steering::rename_fma_trees()
(cortex-a57-fma-steering.c:1004)
==5976== by 0x10F8265: func_fma_steering::execute_fma_steering()
(cortex-a57-fma-steering.c:1034)
==5976== by 0x10F84F9: pass_fma_steering::execute(function*)
(cortex-a57-fma-steering.c:1069)
==5976== by 0xC7FE9A: execute_one_pass(opt_pass*) (passes.c:2465)
==5976== by 0xC801EA: execute_pass_list_1(opt_pass*) (passes.c:2554)
This looks like https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84272
Backporting the patch fixes the valgrind complaints.
I've pushed the patch to devel:gcc/gcc7 to get a build&test cycle for aarch64
and will backport upstream then.
--
You are receiving this mail because:
You are on the CC list for the bug.