Accéder au contenu.
Menu Sympa

starpu-devel - Re: [Starpu-devel] StarPU code hangs when using a lot of GPU memory

Objet : Developers list for StarPU

Archives de la liste

Re: [Starpu-devel] StarPU code hangs when using a lot of GPU memory


Chronologique Discussions 
  • From: Mirko Myllykoski <mirkom@cs.umu.se>
  • To: samuel.thibault@inria.fr
  • Cc: Starpu Devel <starpu-devel@lists.gforge.inria.fr>
  • Subject: Re: [Starpu-devel] StarPU code hangs when using a lot of GPU memory
  • Date: Fri, 01 Mar 2019 18:11:57 +0100
  • Authentication-results: mail3-smtp-sop.national.inria.fr; spf=None smtp.pra=mirkom@cs.umu.se; spf=Pass smtp.mailfrom=mirkom@cs.umu.se; spf=None smtp.helo=postmaster@mail.cs.umu.se
  • Ironport-phdr: 9a23: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
  • List-archive: <http://lists.gforge.inria.fr/pipermail/starpu-devel/>
  • List-id: "Developers list. For discussion of new features, code changes, etc." <starpu-devel.lists.gforge.inria.fr>

Hi,

On 2019-02-27 23:10, Samuel Thibault wrote:

I waited until the code appeared to hang and captured a backtrace. I
then repeated the capturing after letting the code to run additional
5-10 seconds. See the attachments.

It seems StarPU has a hard time evicting data indeed. Could you run the
starpu-memusage gdb macro? That'll give us an idea of how data are
busy enough that StarPU can't evict them.

The output of the macro was not very informative:

(gdb) starpu-memusage


Node 0:
Total used: 14, 364MiB
WT: 0, 0MiB
home: 0, 0MiB
redux: 0, 0MiB
relax: 10, 307MiB
noref: 10, 307MiB
nosubdataref: 2, 28MiB
nodataref: 2, 28MiB

cached: 0, 0MiB


Node 1:
Total used: 5815, 4493MiB
WT: 0, 0MiB
home: 0, 0MiB
redux: 0, 0MiB
relax: 0, 0MiB
noref: 0, 0MiB
nosubdataref: 0, 0MiB
nodataref: 0, 0MiB

cached: 0, 0MiB


- Mirko




Archives gérées par MHonArc 2.6.19+.

Haut de le page