Crashing on HASWELL cpu
Issue #69
resolved
As it is close to my birthday, I bought me a new i7-4790K :)
and of cause, we are crashing :(
I have exactly the same directory mounted debian-i7-2600K is in fact the new i7-4690K (as I just put in the old hard disk and did not rename to avoid breaking my network for the moment)
and on slave 1 what is the beowulf client with an i7-920
The crash seems to have something to do with our mutex handling, I used your version for this test:
detlef@debian-i7-2600:~/tmp/oakfoam/oakfoam$ ./oakfoam
seed: 1405601249
loading opening book from 'book.dat'... error
loading opening book from '/home/detlef/tmp/oakfoam/oakfoam/book.dat'... error
no opening book loaded
genmove b
[dbg|5000] (rm:C3 r:0.48 r2:1.03) (br:A2 r:0.52)
[dbg|10000] (rm:C5 r:0.48 r2:1.05) (br:A1 r:0.67)
best move cannot change! (0.005 1.000)
oakfoam: /usr/include/boost/thread/pthread/mutex.hpp:108: boost::mutex::~mutex(): Assertion `!posix::pthread_mutex_destroy(&m)' failed.
Abgebrochen (Speicherabzug geschrieben)
detlef@debian-i7-2600:~/tmp/oakfoam/oakfoam$ ssh 192.168.1.10
detlef@slave1:~/tmp/oakfoam$ cd oakfoam/
detlef@slave1:~/tmp/oakfoam/oakfoam$ ./oakfoam
seed: 1406021142
loading opening book from 'book.dat'... error
loading opening book from '/home/detlef/tmp/oakfoam/oakfoam/book.dat'... error
no opening book loaded
genmove b
[dbg|5000] (rm:C5 r:0.50 r2:1.50) (br:A2 r:0.56)
best move cannot change! (0.058 1.053)
[genmove]: r:0.489 plts:9500 ppms:4.17 rd:0.007 r2:1.48 fs:-3.00 fsd:0.47 un:3/15 bs:0 pv:(B4,C4,C3,D3,D2,E3,B3) st:(150,8,110,12,173,83,0,36,103,59,0,28,0,0,0,234) ravepreset: -nan
= c5
Comments (4)
-
reporter -
reporter ok, I have a workaround: in Tree.cc
#define BOOST_DISABLE_ASSERTS #include "Tree.h" #include <cmath> #include <sstream>
at the beginning of the file. I think this is elision locking, which is still buggy in glibc, so should not harm
-
repo owner - changed status to resolved
Looks like you've found a good solution for the moment :)
-
reporter Very interesting :) Might be connected, but i did not check the details.....
- Log in to comment
just to show, that we are not allone I think:
https://sourceware.org/bugzilla/show_bug.cgi?id=16657