You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@commons.apache.org by Eric Barnhill <er...@gmail.com> on 2015/11/24 21:58:14 UTC

MathArrays -> JogAmp -> OpenCL

Is anyone interested in some GPU support for MathArrays by using, say, the
JogAmp bindings and OpenCL methods. I have implemented such an architecture
for my own convolution library I am developing, and it would not be much
trouble for me to add this kind of support for commons-math some time in
the new year.

It would require the user to install the JOCL glue libraries to make use of
this and this may be out side of the commons mission. Overall I find JogAmp
very convenient and its creators support it.

Eric

Re: MathArrays -> JogAmp -> OpenCL

Posted by Eric Barnhill <er...@gmail.com>.
The Java JOCL / JogAmp bindings are located here:

https://jogamp.org/jocl/www/

A user would have to install what are called the "glue" libraries which 
enable use of OpenCL , via Ant.

If it is of interest, I'll put a minimal working example for, say, 
ebeMultiplication with GPU, up on my GitHub page some time next week, 
and link to it here, and then you can see what you think.

Eric

On 25/11/15 09:20, luc wrote:
> Le 2015-11-24 22:26, Gilles a écrit :
>> On Tue, 24 Nov 2015 20:58:14 +0000, Eric Barnhill wrote:
>>> Is anyone interested in some GPU support for MathArrays by using, 
>>> say, the
>>> JogAmp bindings and OpenCL methods. I have implemented such an 
>>> architecture
>>> for my own convolution library I am developing, and it would not be 
>>> much
>>> trouble for me to add this kind of support for commons-math some 
>>> time in
>>> the new year.
>>>
>>> It would require the user to install the JOCL glue libraries to make 
>>> use of
>>> this and this may be out side of the commons mission. Overall I find 
>>> JogAmp
>>> very convenient and its creators support it.
>
> Could you elaborate on this?
> Would this mean we add a dependency?
> What does "to make use of" mean, would this be an interface part of the
> library that could be compiled only if the glue libraries are available?
>
> Luc
>
>>
>> There were some recent and not so recent discussions on parallel 
>> processing
>> as a way to enhance the performance of CM.
>>
>> Actual experiments are certainly welcome to figure out the best 
>> way(s) to go.
>>
>> And this would probably be a nice addition to the userguide.
>> Could you give pointers to the utilities and a slightly more detailed
>> description
>> of the intended usage?
>>
>> Thanks,
>> Gilles
>>
>>
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
>> For additional commands, e-mail: dev-help@commons.apache.org
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
> For additional commands, e-mail: dev-help@commons.apache.org
>


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org


Re: MathArrays -> JogAmp -> OpenCL

Posted by luc <lu...@spaceroots.org>.
Le 2015-11-24 22:26, Gilles a écrit :
> On Tue, 24 Nov 2015 20:58:14 +0000, Eric Barnhill wrote:
>> Is anyone interested in some GPU support for MathArrays by using, say, 
>> the
>> JogAmp bindings and OpenCL methods. I have implemented such an 
>> architecture
>> for my own convolution library I am developing, and it would not be 
>> much
>> trouble for me to add this kind of support for commons-math some time 
>> in
>> the new year.
>> 
>> It would require the user to install the JOCL glue libraries to make 
>> use of
>> this and this may be out side of the commons mission. Overall I find 
>> JogAmp
>> very convenient and its creators support it.

Could you elaborate on this?
Would this mean we add a dependency?
What does "to make use of" mean, would this be an interface part of the
library that could be compiled only if the glue libraries are available?

Luc

> 
> There were some recent and not so recent discussions on parallel 
> processing
> as a way to enhance the performance of CM.
> 
> Actual experiments are certainly welcome to figure out the best way(s) 
> to go.
> 
> And this would probably be a nice addition to the userguide.
> Could you give pointers to the utilities and a slightly more detailed
> description
> of the intended usage?
> 
> Thanks,
> Gilles
> 
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
> For additional commands, e-mail: dev-help@commons.apache.org

---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org


Re: MathArrays -> JogAmp -> OpenCL

Posted by Gilles <gi...@harfang.homelinux.org>.
On Tue, 24 Nov 2015 20:58:14 +0000, Eric Barnhill wrote:
> Is anyone interested in some GPU support for MathArrays by using, 
> say, the
> JogAmp bindings and OpenCL methods. I have implemented such an 
> architecture
> for my own convolution library I am developing, and it would not be 
> much
> trouble for me to add this kind of support for commons-math some time 
> in
> the new year.
>
> It would require the user to install the JOCL glue libraries to make 
> use of
> this and this may be out side of the commons mission. Overall I find 
> JogAmp
> very convenient and its creators support it.

There were some recent and not so recent discussions on parallel 
processing
as a way to enhance the performance of CM.

Actual experiments are certainly welcome to figure out the best way(s) 
to go.

And this would probably be a nice addition to the userguide.
Could you give pointers to the utilities and a slightly more detailed 
description
of the intended usage?

Thanks,
Gilles


---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@commons.apache.org
For additional commands, e-mail: dev-help@commons.apache.org