I've tried to build tbb2.1(commercially aligned release)ona Xeon server but 'text_halt.exe' hangs while "make all".
text_halt.exe linked with libtbb_debug.so passed the test run.
only therelease version hangs.
Here is the system information:
CPU:two Xeon E5345 processors
OS: linux kernel 126.96.36.199
Compiler: gcc 4.1.2 with libc 2.6
Thanks for informing about the issue. Could you please answer a couple questions to help me reproduce and investigate the issue?
- Can you provide the output of `cat /etc/issue` and`uname -rm`?
- Did you use the pre-built binaries downloaded from the site, or you built TBB from sources?
Change to the initial post: `uname -a` was an overkill, I only need the kernel version (which you already named) and whether it is 32 or 64 bits. So `uname -rm` is enough.
the linux is 188.8.131.52-21.fc7 x86_64 (FC7 Moonshine), and I built TBB from sources.
It seems that it's not relevant whether the test suite is release/debug version.
I've tried to build TBBmore times, andthe debug version of test_halt.exepassed occasionally.
I was unable to reproduce the issue on a Fedora 7 machine here; though the kernel version was 2.6.21.something, older than yours.
Could you possibly provide the binaries built on your machine (*.so.2 librariesas well as test_halt.exe, for both debug and release)?
I ran your binaries for 100 times on a FC7 machine:
for((i=0;i<100;i++)); do ./test_halt.exe 4:8 || echo $i: failed; done
butneither debug nor release version failed even once.
Of courseit does not mean there is no issue. But at the moment, the best thing I can suggest for you is to try building TBB binaries using the most recent TBB developer update. We have some post-2.1 fixes, in particular we have fixed a futex-related bug and I think it might be relevant. If that does not help, there must be something on your system that either breaks TBB or reveals a hidden bug in it.