Skip to Content.
Sympa Menu

charm - Re: [charm] GPU Support

charm AT lists.cs.illinois.edu

Subject: Charm++ parallel programming system

List archive

Re: [charm] GPU Support


Chronological Thread 
  • From: Lukasz Wesolowski <wesolwsk AT illinois.edu>
  • To: Rafael Keller Tesser <rktesser AT inf.ufrgs.br>
  • Cc: "charm AT cs.uiuc.edu" <charm AT cs.uiuc.edu>
  • Subject: Re: [charm] GPU Support
  • Date: Thu, 12 Sep 2013 15:19:09 -0500
  • List-archive: <http://lists.cs.uiuc.edu/pipermail/charm/>
  • List-id: CHARM parallel programming system <charm.cs.uiuc.edu>

Hi Rafael,

We have a low-level task library (GPU Manager, previously called
Hybrid API) for encapsulating and offloading work to CUDA-based GPUs.
We also have the higher-level Accelerated Entry Method framework,
built on top of GPU Manager, that generates CUDA kernels from
appropriately tagged entry methods and controls for grain size by
batching or splitting the work from instances of these methods
belonging to different chares. These tools are available in the charm
repository. GPU Manager is in active use while the AEM framework may
be a little rusty.

In addition, there is G-Charm, also built on top of GPU Manager, which
is an externally developed library for agglomerating and scheduling
GPU work and optimizing for data reuse.

Lukasz

On Thu, Sep 12, 2013 at 2:18 PM, Rafael Keller Tesser
<rktesser AT inf.ufrgs.br>
wrote:
> Hello!
>
> I'd like to know what's the current state of GPU support in Charm++. I've
> read some publications on the subject but the tools (or libraries) that have
> been used in these works don't seem to be available to the general public.
>
> --
> Best regards,
> Rafael Keller Tesser
> GPPD-UFRGS
>
> _______________________________________________
> charm mailing list
> charm AT cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/charm
>




Archive powered by MHonArc 2.6.16.

Top of Page