accelerate library

accelerate library

Postby parallel-future » Wed Dec 19, 2012 2:03 am

This library looks like it might be useful for using Parallella as an accelerator. Currently it has "an experimental OpenCL backend"

http://hackage.haskell.org/package/accelerate
parallel-future
 
Posts: 6
Joined: Mon Dec 17, 2012 4:00 am
Location: Australia

Re: accelerate library

Postby aolofsson » Wed Dec 19, 2012 4:04 am

Looks like a great package! Can't wait for the release of the OpenCL programmable Parallella boards into the wild so that we can hear about developer experiences with using some of these higher level frameworks with the Epiphany.

Seems like OpenCL is becoming the defacto standard as an intermediate target for high level parallel programming frameworks. Any thoughts?
User avatar
aolofsson
 
Posts: 1005
Joined: Tue Dec 11, 2012 6:59 pm
Location: Lexington, Massachusetts,USA

Re: accelerate library

Postby adam » Wed Dec 19, 2012 4:27 am

aolofsson wrote:Looks like a great package! Can't wait for the release of the OpenCL programmable Parallella boards into the wild so that we can hear about developer experiences with using some of these higher level frameworks with the Epiphany.

Seems like OpenCL is becoming the defacto standard as an intermediate target for high level parallel programming frameworks. Any thoughts?


That is how I was thinking of using it. Given the change-compile-run turn-around time I have observed for the Epiphany SDK in demos, I am hoping that I will be able to develop a significant part of the OpenCL targeting the GPU on my development machine, and deploy to Epiphany only in later stages of development.
adam
 
Posts: 2
Joined: Mon Dec 17, 2012 3:24 am

Re: accelerate library

Postby parallel-future » Wed Dec 19, 2012 4:48 am

aolofsson wrote:Seems like OpenCL is becoming the defacto standard as an intermediate target for high level parallel programming frameworks. Any thoughts?


OpenCL is becoming the de facto standard because it's - <gasp> - open! This means it's much more portable (both in terms of software and in terms of skills). Nobody wants to be locked down to a specific vendor. But many people want vendor support, or a mature software package. Thankfully OpenCL is starting to be pretty mature.

Another thought: I think there is the potential for a wildly different sort of intermediate target, one that takes care of mobility between cores and layout (to keep communication costs down), load balancing, and so on - thus abstracting away a lot of the difficult and machine-specific parts of programming a parallel machine. This is something I am investigating.
parallel-future
 
Posts: 6
Joined: Mon Dec 17, 2012 4:00 am
Location: Australia


Return to Haskell

Who is online

Users browsing this forum: No registered users and 1 guest

cron