The ABC step fails on my system (Mac OS X 10.9.3 x64) on _some_ (huge) BLIF
files generated by ODIN II. The attached file aes_flat.v can reproduce this
behavior during resyn2, after going through ODIN (tested on the latest SVN rev).
I could not attach the BLIF file due to its size (31MB), but this is the output
:
----------
$ ./abc -c "read aes.odin.blif; resyn; resyn2"
The 4-input library started with 43906 nodes and 24772 subgraphs. Time = 0.05
sec
Warning: Constant-0 drivers added to 1 non-driven nets in network
"aes_cipher_top":
top.aes_key_expand_128+u0^BUF_NODE~1815
Segmentation fault: 11
----------
The latest revision of the ABC repository doesn't crash. The bug comes from a
truncated pointer in the CUDD package used by ABC :
----------
$ lldb ./abc
Current executable set to './abc' (x86_64).
(lldb) r
Process 43615 launched: './abc' (x86_64)
The 4-input library started with 43906 nodes and 24772 subgraphs. Time = 0.04
sec
UC Berkeley, ABC for VTR 7.0 (compiled May 27 2014 14:45:36)
abc 01> read aes.odin.blif; resyn; resyn2
Warning: Constant-0 drivers added to 1 non-driven nets in network
"aes_cipher_top":
top.aes_key_expand_128+u0^BUF_NODE~1815
Process 43615 stopped
* thread #1: tid = 0x6aeee, 0x000000010013ff1a
abc`cuddGarbageCollectZdd(unique=0x0000000104731480, clearCache=1) + 248 at
cuddTable.c:896, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS
(code=1, address=0x6bd9a02)
frame #0: 0x000000010013ff1a abc`cuddGarbageCollectZdd(unique=0x0000000104731480, clearCache=1) + 248 at cuddTable.c:896
893 for (i = 0; i < slots; i++) {
894 c = &cache[i];
895 if (c->data != NULL) {
-> 896 if (cuddClean(c->f)->ref == 0 ||
897 cuddClean(c->g)->ref == 0 ||
898 (((ptruint)c->f & 0x2) && Cudd_Regular(c->h)->ref == 0) ||
899 (c->data != DD_NON_CONSTANT &&
----------
Indeed, the SIZEOF_VOID_P constant in src/bdd/mtr/mtr.h is assigned its default
value on Darwin. The proposed attached patch cudd-bugfix-darwin.h adds a few
preprocessor directives that forces the good pointer size to be used on these
systems and fixes the issue.
The "official" ABC Makefile includes a small program arch_flags.c compiled
during the build which sets the good values for SIZEOF_VOID_P (among other data
sizes).
Regards,
-- Christophe Huriaux
Original issue reported on code.google.com by c.huri...@gmail.com on 27 May 2014 at 7:15
Original issue reported on code.google.com by
c.huri...@gmail.comon 27 May 2014 at 7:15Attachments: