The SPEC CPU2000 tests bzip2, gap, and gcc cause mainline GCC to
segfault when built on powerpc64-linux with:
-m32 -O2 -ftree-vectorize -maltivec -mabi=altivec \
--param ggc-min-expand=0 --param ggc-min-heapsize=0
Every single CPU2000 test fails to build with those options and -m64
instead of -m32. gap, facerec, and apsi sometimes fail with -m64
without the ggc params defined, as mentioned in comment 4 of PR 21155.
This minimized test case demonstrates the bug:
extern void bar (unsigned char *, int, int);
unsigned char l[6][4];
int c[6][4];
void
foo (int n, int a)
{
int t, i;
int x;
for (t = 0; t < n; t++)
{
for (i = 0; i < a; i++)
x = l[t][i];
bar (&l[t][0], x, a);
}
}
gdb shows the segfault at gcc/tree-chrec.c:1048 ct=0xa5a5a5a5.
The backtrace is too messy for cut-and-paste but shows
chrec_convert
interpret_rhs_modify_expr
analyze_scalar_evolution_1
analyze_scalar_evolution
simple_iv
number_of_iterations_exit
number_of_iterations_in_loop
canonicalize_loop_induction_variables
tree_unroll_loops_completely
tree_complete_unroll
execute_pass_list
...
SPEC CPU2000 with the ggc params set to zero all build with "-O2" and
with "-O2 -ftree-loop-linear".
The failure shows up with this patch from radkver:
http://gcc.gnu.org/ml/gcc-cvs/2005-05/msg00302.html
Just to make things even more interesting, add "-fprofile-generate"
and the test case starts failing in the same place on 20050428 with
this patch from bonzini, which suggests it's a latent bug unrelated
to these patches:
http://gcc.gnu.org/ml/gcc-cvs/2005-04/msg01463.html
--
Summary: poisoned ggc memory used for -ftree-vectorize
Product: gcc
Version: 4.1.0
Status: UNCONFIRMED
Severity: normal
Priority: P2
Component: tree-optimization
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: janis at gcc dot gnu dot org
CC: dorit at il dot ibm dot com,gcc-bugs at gcc dot gnu dot
org
GCC build triplet: powerpc64-linux
GCC host triplet: powerpc64-linux
GCC target triplet: powerpc64-linux
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=21639
