Julia with Xeon Phi Knights Landing

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

Julia with Xeon Phi Knights Landing

Chris Rackauckas
Hey,
  From previous posts here it was noted that Julia will be able to use a Knights Landing Xeon Phi. Any idea on what that interface might look like? I was interested in starting to use the Phi via Julia, but before I venture into writing some C/OpenMP code in order to replace the perfectly fine Julia code, I have a few questions:

- Is there any time estimate on when the support will come? It was noted in the previous post that it would be part of 0.5, is this still the goal?
-What is the planned interface for the Xeon Phi? FYI most of my code is parallelized using parallel loops.
-Anyone know why only Knights Landing will be supported through LLVM? Just curious.

[FYI, I use Julia for HPC and scientific computing applications, and have a hybrid system (GPU + Xeon Phi) for code development and allocations on a few XSEDE supercomputers. Also, I will be making sure my lab gets a Knights Landing Xeon Phi soon after it drops. I am willing to do some alpha testing. Most of what I do is embarrassingly parallel, so if something is in the works and needs testing, I can give it a shot.]
Reply | Threaded
Open this post in threaded view
|

Re: Julia with Xeon Phi Knights Landing

Viral Shah
Many of these questions can only be answered after real hardware comes out and some of us get access to it. The bigger question is the multi-threaded programming model in Julia, independent of KNL specifically, but in light of upcoming multi-core hardware.

-viral

On Monday, February 22, 2016 at 7:10:35 PM UTC+5:30, Chris Rackauckas wrote:
Hey,
  From previous posts <a href="https://groups.google.com/forum/#!topic/julia-users/NXYuIHC-bwo" target="_blank" rel="nofollow" onmousedown="this.href=&#39;https://groups.google.com/forum/#!topic/julia-users/NXYuIHC-bwo&#39;;return true;" onclick="this.href=&#39;https://groups.google.com/forum/#!topic/julia-users/NXYuIHC-bwo&#39;;return true;">here it was noted that Julia will be able to use a Knights Landing Xeon Phi. Any idea on what that interface might look like? I was interested in starting to use the Phi via Julia, but before I venture into writing some C/OpenMP code in order to replace the perfectly fine Julia code, I have a few questions:

- Is there any time estimate on when the support will come? It was noted in the previous post that it would be part of 0.5, is this still the goal?
-What is the planned interface for the Xeon Phi? FYI most of my code is parallelized using parallel loops.
-Anyone know why only Knights Landing will be supported through LLVM? Just curious.

[FYI, I use Julia for HPC and scientific computing applications, and have a hybrid system (GPU + Xeon Phi) for code development and allocations on a few XSEDE supercomputers. Also, I will be making sure my lab gets a Knights Landing Xeon Phi soon after it drops. I am willing to do some alpha testing. Most of what I do is embarrassingly parallel, so if something is in the works and needs testing, I can give it a shot.]
Reply | Threaded
Open this post in threaded view
|

Re: Julia with Xeon Phi Knights Landing

Oleg Mikulchenko
Knights Landing is the self sufficient bootable CPU, binary compatible with Xeon. This is why LLVM will be easy supported. 

There is one (or more?) package, addressing parallel processing in Julia now:  ParallelAccellerator -  see, e.g.  http://www.slideshare.net/ChristianPeel/ehsan-parallel-acceleratordec2015. The generic multi-threaded programming model  in the core language is a different story, I guess. 
 
It will be very interesting to see Julia running on KNL. 
Reply | Threaded
Open this post in threaded view
|

Re: Julia with Xeon Phi Knights Landing

Alex Mellnik
Now that the developer access program for Knights Landing has started (https://software.intel.com/en-us/articles/developer-access-program-for-intel-xeon-phi-processor-codenamed-knights-landing) is anyone planning on picking one up, and if so, will you be tracking your progress on using Julia with it anywhere?  I would be interested to hear about it if so.

Thanks -A

(Disclaimer: I am an Intel employee, but don't work on anything remotely related to Knights Landing.)

On Thursday, March 3, 2016 at 6:45:41 AM UTC-8, Oleg Mikulchenko wrote:
Knights Landing is the self sufficient bootable CPU, binary compatible with Xeon. This is why LLVM will be easy supported. 

There is one (or more?) package, addressing parallel processing in Julia now:  ParallelAccellerator -  see, e.g.  <a href="http://www.slideshare.net/ChristianPeel/ehsan-parallel-acceleratordec2015" target="_blank" rel="nofollow" onmousedown="this.href=&#39;http://www.google.com/url?q\75http%3A%2F%2Fwww.slideshare.net%2FChristianPeel%2Fehsan-parallel-acceleratordec2015\46sa\75D\46sntz\0751\46usg\75AFQjCNEiYNcL4mkGk9J7d5zSctmmb0bKHg&#39;;return true;" onclick="this.href=&#39;http://www.google.com/url?q\75http%3A%2F%2Fwww.slideshare.net%2FChristianPeel%2Fehsan-parallel-acceleratordec2015\46sa\75D\46sntz\0751\46usg\75AFQjCNEiYNcL4mkGk9J7d5zSctmmb0bKHg&#39;;return true;">http://www.slideshare.net/ChristianPeel/ehsan-parallel-acceleratordec2015. The generic multi-threaded programming model  in the core language is a different story, I guess. 
 
It will be very interesting to see Julia running on KNL.