------- Comment From [email protected] 2016-11-15 17:41 EDT-------
To be clear... The successful runs we have so far have been achieved by
building from Canonicals glibc sources (apt-get source libc6) with just the
following patch:
--- glibc-2.23.orig/sysdeps/unix/sysv/linux/powerpc/elision-unlock.c
+++ glibc-2.23/sysdeps/unix/sysv/linux/powerpc/elision-unlock.c
@@ -28,13 +28,12 @@ __lll_unlock_elision (int *lock, short *
__libc_tend (0);
else
{
- lll_unlock ((*lock), pshared);
-
/* Update the adapt count AFTER completing the critical section.
Doing this here prevents unneeded stalling when entering
a critical section. Saving about 8% runtime on P8. */
if (*adapt_count > 0)
(*adapt_count)--;
+ lll_unlock ((*lock), pshared);
}
return 0;
}
And swapping in just the resulting libpthread library into the test
setup. We'll continue testing with that for now. We're happy to try
something from -proposed (or whatever) if you like.
And we'll keep testing with the change to prove the problem via distinct
adapt_count value. (I _think_ I've seen the expected signature in a
crash with that code already, but want to recreate with a testcase that
will give clearer results. That testcase also runs much longer before
crashing.)
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/1641241
Title:
TensorFlow application crashes after glibc upgrade from 14.04 to 16.04
version
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/glibc/+bug/1641241/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
