Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

- Intel Community
- Intel oneAPI Toolkits
- Intel® oneAPI HPC Toolkit
- Can each thread on Xeon Phi be given private data areas in the offload model

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Mute
- Printer Friendly Page

Anders_S_1

Beginner

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

08-24-2015
06:48 AM

27 Views

Can each thread on Xeon Phi be given private data areas in the offload model

Hi,

I want to calculate a Jacobian matrix, which is a sum of 960 (to be simple) 3x3 matrices by distributing the calculations of these 3x3 matrices to a Xeon Phi card. The calculation of the 3x3 matrices uses a third party library whose subroutines use an interger vector not only for the storage of parameter values but also to write and read intermidiate results. It is therefore necessary for each task to have this integer vector protected from other tasks. Can this be obtained on the physical core level or even for each thread (each Xeon Phi has 60x4=240 threads.

If I got it right the MPI commands can be used on the dual 6C Xeon host to let 2x6 cores or 2x6x2 threads offload to the Xeon Phi card.

To be clear I would like to do this under Windows if possible.

Best regards

Anders S

Link Copied

1 Reply

Steve_H_Intel1

Employee

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

08-24-2015
02:00 PM

27 Views

Anders:

>I want to calculate a Jacobian matrix, which is a sum of 960 (to be simple) 3x3 matrices by distributing the calculations of these 3x3

>matrices to a Xeon Phi card. The calculation of the 3x3 matrices uses a third party library whose subroutines use an interger vector

>not only for the storage of parameter values but also to write and read intermidiate results. It is therefore necessary for each task to

>have this integer vector protected from other tasks. Can this be obtained on the physical core level or even for each thread (each

>Xeon Phi has 60x4=240 threads.

I believe that the answer is yes. One idea you might want to possibly consider is to array-expand the "integer vector" based on the number of MPI ranks, where each MPI rank is only allowed to access the portion of the "integer vector" that is associated with its rank value.

>If I got it right the MPI commands can be used on the dual 6C Xeon host to let 2x6 cores or 2x6x2 threads offload to the Xeon Phi card.

>

>

>To be clear I would like to do this under Windows if possible.

You may want to read the article titled, "Using Windows Instead of Linux as a Host for Xeon Phi Coprocessor", which can be found at the URL:

http://goparallel.sourceforge.net/using-windows-instead-linux-host-xeon-phi-coprocessor/

The URL:

https://software.intel.com/en-us/articles/debugging-intel-xeon-phi-applications-on-windows-host

references an article titled, "Debugging Intel® Xeon Phi™ Applications on Windows* Host". You might find this useful, also.

Thank you,

-Steve

For more complete information about compiler optimizations, see our Optimization Notice.