systemml-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Matthias Boehm <mboe...@googlemail.com>
Subject Re: Enabling CLA by default in SystemML 1.0
Date Fri, 01 Sep 2017 00:59:34 GMT
no, not yet. Before enabling it via 'auto', I wanted to completely
integrate with the new code generator as well. This is mostly done except
for some performance issues of row-wise codegen operations that require
another pass.

Regards,
Matthias

On Thu, Aug 31, 2017 at 5:52 PM, Frederick R Reiss <frreiss@us.ibm.com>
wrote:

> Did this happen?
>
> Fred
>
> [image: Inactive hide details for "Glenn Weidner" ---05/28/2017 09:38:54
> PM---+1 Glenn]"Glenn Weidner" ---05/28/2017 09:38:54 PM---+1 Glenn
>
> From: "Glenn Weidner" <gweidner@us.ibm.com>
> To: dev@systemml.apache.org
> Date: 05/28/2017 09:38 PM
> Subject: Re: Enabling CLA by default in SystemML 1.0
> ------------------------------
>
>
>
> +1
>
> Glenn
>
> "Niketan Pansare" ---05/28/2017 02:54:07 PM---+1 for the change. This will
> give us enough time to test CLA with different setting/algorithms/data
>
> From: "Niketan Pansare" <npansar@us.ibm.com>
> To: dev@systemml.apache.org
> Date: 05/28/2017 02:54 PM
> Subject: Re: Enabling CLA by default in SystemML 1.0
> ------------------------------
>
>
>
>
> +1 for the change. This will give us enough time to test CLA with different
> setting/algorithms/data characteristics before the release.
>
> > On May 28, 2017, at 2:42 PM, Matthias Boehm <mboehm7@googlemail.com>
> wrote:
> >
> > Hi all,
> >
> > just a heads-up: I intend to enable compressed linear algebra (CLA) by
> > default in our SystemML 1.0 release. So far, CLA can be configured via
> > 'compressed.linalg' as true or false, and it's disabled by default. For
> the
> > SystemML 1.0 release, I'd like to extend this configuration to take
> either
> > true, false, or auto, where auto will be the default. This 'auto' option
> > indicates to compress out-of-core matrices (i.e., matrices that do not
> fit
> > in aggregate memory), iff we support all operations over these matrices
> in
> > the given program as compressed operations to avoid implicit
> decompression.
> >
> > If you have any concerns, please let me know. Otherwise, I would make
> that
> > change in the next couple of weeks in order to give sufficient time for
> > testing in the wild.
> >
> > Regards,
> > Matthias
>
>
>
>
>
>

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message