Re: [hwloc-users] Problem with hwloc_set_thread_cpubind() and pthread_barrier_wait() on new debianj
Αλέξανδρος Παπαδογιαννάκης, le Wed 16 Jun 2010 15:37:34 +0300, a écrit : > The program uses hwloc_set_thread_cpubind() to bind some threads to specific > cpus > but now it doesn't work. hwloc_set_thread_cpubind() returns zero but doesn't > bind > the thread. This looks to me like a glibc or kernel bug. Could you check with strace that a sched_setaffinity system call gets performed, and the returned value? > The problem is that when I call hwloc_set_thread_cpubind() the threads > are waiting on a posix barrier. That shouldn't be a problem. Samuel
Re: [hwloc-users] Problem with hwloc_set_thread_cpubind() and pthread_barrier_wait() on new debianj
hwloc_set_thread_cpubind() by itself works. I have it on a different test program and it binds the threads without a problem. The problem is when the thread is waiting on a barrier. I'll try strace and send you the results -- Alexandros Papadogiannakis > Date: Wed, 16 Jun 2010 14:47:44 +0200 > From: samuel.thiba...@inria.fr > To: hwloc-us...@open-mpi.org > Subject: Re: [hwloc-users] Problem with hwloc_set_thread_cpubind() and > pthread_barrier_wait() on new debianj > > Αλέξανδρος Παπαδογιαννάκης, le Wed 16 Jun 2010 15:37:34 +0300, a écrit : > > The program uses hwloc_set_thread_cpubind() to bind some threads to specific > > cpus > > but now it doesn't work. hwloc_set_thread_cpubind() returns zero but doesn't > > bind > > the thread. > > This looks to me like a glibc or kernel bug. Could you check with strace > that a sched_setaffinity system call gets performed, and the returned > value? > > > The problem is that when I call hwloc_set_thread_cpubind() the threads > > are waiting on a posix barrier. > > That shouldn't be a problem. > > Samuel > ___ > hwloc-users mailing list > hwloc-us...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-users _ Hotmail: Trusted email with powerful SPAM protection. https://signup.live.com/signup.aspx?id=60969
Re: [hwloc-users] Problem with hwloc_set_thread_cpubind() and pthread_barrier_wait() on new debianj
Αλέξανδρος Παπαδογιαννάκης, le Wed 16 Jun 2010 15:52:04 +0300, a écrit : > hwloc_set_thread_cpubind() by itself works. I have it on a different test > program > and it binds the threads without a problem. The problem is when the thread is > waiting on a barrier. I understand. I'm just saying that concerning hwloc, being asleep or not shouldn't change the behavior, and the bug is most probably either in glibc or the kernel. The result of sched_setaffinity should tell us. Samuel
Re: [hwloc-users] Problem with hwloc_set_thread_cpubind() and pthread_barrier_wait() on new debianj
There is no shed affinity call at all if the thread is on a barrier :/ -- Alexandros Papadogiannakis > Date: Wed, 16 Jun 2010 15:02:00 +0200 > From: samuel.thiba...@inria.fr > To: hwloc-us...@open-mpi.org > Subject: Re: [hwloc-users] Problem with hwloc_set_thread_cpubind() and > pthread_barrier_wait() on new debianj > > Αλέξανδρος Παπαδογιαννάκης, le Wed 16 Jun 2010 15:52:04 +0300, a écrit : > > hwloc_set_thread_cpubind() by itself works. I have it on a different test > > program > > and it binds the threads without a problem. The problem is when the thread > > is > > waiting on a barrier. > > I understand. I'm just saying that concerning hwloc, being asleep or not > shouldn't change the behavior, and the bug is most probably either in > glibc or the kernel. The result of sched_setaffinity should tell us. > > Samuel > ___ > hwloc-users mailing list > hwloc-us...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-users _ Your E-mail and More On-the-Go. Get Windows Live Hotmail Free. https://signup.live.com/signup.aspx?id=60969
Re: [hwloc-users] Problem with hwloc_set_thread_cpubind() and pthread_barrier_wait() on new debianj
Sorry wrong I'll send you in a minute -- Alexandros Papadogiannakis From: psxlo...@hotmail.com To: hwloc-us...@open-mpi.org Date: Wed, 16 Jun 2010 16:11:12 +0300 Subject: Re: [hwloc-users] Problem with hwloc_set_thread_cpubind() and pthread_barrier_wait() on new debianj There is no shed affinity call at all if the thread is on a barrier :/ -- Alexandros Papadogiannakis > Date: Wed, 16 Jun 2010 15:02:00 +0200 > From: samuel.thiba...@inria.fr > To: hwloc-us...@open-mpi.org > Subject: Re: [hwloc-users] Problem with hwloc_set_thread_cpubind()and > pthread_barrier_wait() on new debianj > > Αλέξανδρος Παπαδογιαννάκης, le Wed 16 Jun 2010 15:52:04 +0300, a écrit : > > hwloc_set_thread_cpubind() by itself works. I have it on a different test > > program > > and it binds the threads without a problem. The problem is when the thread > > is > > waiting on a barrier. > > I understand. I'm just saying that concerning hwloc, being asleep or not > shouldn't change the behavior, and the bug is most probably either in > glibc or the kernel. The result of sched_setaffinity should tell us. > > Samuel > ___ > hwloc-users mailing list > hwloc-us...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-users Your E-mail and More On-the-Go. Get Windows Live Hotmail Free. Sign up now. _ Hotmail: Powerful Free email with security by Microsoft. https://signup.live.com/signup.aspx?id=60969
Re: [hwloc-users] Problem with hwloc_set_thread_cpubind() and pthread_barrier_wait() on new debianj
Αλέξανδρος Παπαδογιαννάκης, le Wed 16 Jun 2010 16:11:12 +0300, a écrit : > There is no shed affinity call at all if the thread is on a barrier :/ Ok, so since hwloc_set_thread_cpubind returns 0, the issue can only be in glibc's pthread_setaffinity_np() (and I guess your are calling it on a thread != pthread_self()) Which version of glibc are you using? Samuel