topic Hi Armando, in Software Archive

Xeon Phi development under Windows

Armando_Lazaro_Alami — Thu, 23 Jan 2014 11:46:22 GMT

I use Intel Composer for Windows development. Xeon-Phi should be a great add to our type of applications on the medical industry. But I found very little information regarding develepment for Phi under Windows.

One interesting feature of Xeon Phi is that we can use it from OpenMP and this makes the migration process easy if compared with CUDA. But as I read a few days ago working under Windows forces the use of "offload mode" and OpenMP is not a choice.

I would like to know if Intel has projects to better integrate Phi under Windows or if the only native platform for Phi will continue to be Linux.

Thanks.

In principle, the OpenMP 4

TimP — Thu, 23 Jan 2014 12:00:00 GMT

In principle, the OpenMP 4 versions of offload syntax should all work the same with Intel compilers (within this year, better than they do now), with Windows or linux host, if that is your concern.

I can't tell whether you are making a distinction between OpenMP 4 and OpenMP 3.1, or between OpenMP 4 and legacy offload, or alluding to the lack of support for OpenMPI on Windows, or wishing that Microsoft would port Windows to the coprocessor or be more supportive of Intel's efforts with MPI for Windows.

In what way is OpenMP "not a choice?" Microsoft doesn't show any signs of supporting OpenMP beyond 2.0, but I don't see that in your question.

"But as I read a few days ago

James_C_Intel2 — Thu, 23 Jan 2014 12:17:31 GMT

"But as I read a few days ago working under Windows forces the use of "offload mode" and OpenMP is not a choice."

I don't know where you read this, but to the best of my knowledge it is not true. As Tim said, using either the Intel offload pragmas or OpenMP 4.0 offload pragmas should work for a code running on Windows and offloading to a Xeon Phi coprocessor.

Thanks for your reply ! I am

Armando_Lazaro_Alami — Thu, 23 Jan 2014 13:53:03 GMT

Thanks for your reply ! I am happy to know that Windows support will be better in a near future.

If you look for information on how to code for Phi under Windows, several documents from Intel and a published book are rather too brief or skip the gory details. Look at this : Intel® Xeon Phi™ Programming Environment

Some others outside Intel wrote like this : http://www.pugetsystems.com/blog/2013/10/07/Intel-Xeon-Phi-with-Windows-510/

My question has other purpose, not only clarifications for coding: focus on applications that are very close to HPC but outside the scientific community, lot of them under Windows. Examples: CT scanners, MRI scanners, Radiotherapy Planning Systems, etc, to mention only few. On these fields a lot of legacy code exists and migration could be painful if there is no easy path, like OpenMP. Probably the reason for the lack of penetration of CUDA in some areas is the departure from traditional programming paradims.

I am not concerned with lack of support from MS-compilers, we use Intel Composer. But I do not see much effort on the side of Intel to easy the migration of code to Phi, particularly under Windows.

Conclusion: There are lot of applications running under Windows that could benefit with Xeon-Phi. The easier the migration path the deeper penetration of Phi on those fields.

Hi Armando,

BelindaLiviero — Thu, 23 Jan 2014 18:52:52 GMT

Hi Armando,

We are pleased to see your interest in Intel Xeon Phi coprocessors. The current generation is an card that fits into a host system. The coprocessor runs its own version of Linux separate from the host system. The host system may run either Linux or Windows. Applications for the coprocessor are cross compiled on the host platform (windows or Linux). Programming for the platform uses traditional threading models – we recommend OpenMP, CilkPlus threading as well as Threading Building Blocks. Many of our customer appreciate the common threading model support.

Production support for Windows host systems was effective 2H 2013. Some of the third party articles about windows hosts were written during the beta adoption period, earlier in the year. Much of the programming process is the same for windows host and Linux hosted systems. . We hope you find the documentation for windows better now than when the some of the articles you reference were posted. We work hard to provide proper support to both host environments As an example our recent articles for debugging on MIC were posted for both host platforms simultaneously (http://software.intel.com/en-us/articles/debugging-intel-xeon-phi-applications-on-linux-host and http://software.intel.com/en-us/articles/debugging-intel-xeon-phi-applications-on-windows-host). The Hardware/MPSS troubleshooting guide we have for Linux will be reworked as well.

If you have questions about programming from Windows host or suggest specific examples that would help we look forward to hearing more from you.

For reference here is a more complete list of links for windows hosted MIC programmming

Drivers: http://software.intel.com/en-us/articles/intel-manycore-platform-software-stack-mpss
Quick start guide: http://software.intel.com/en-us/articles/intel-xeon-phi-coprocessor-developers-quick-start-guide
Debugging guide: http://software.intel.com/en-us/articles/debugging-intel-xeon-phi-applications-on-windows-host
We're developing a brief illustrating “best known methods” on porting that may not be commonly known; however, most programming and porting content is already included in the compiler documentation - http://software.intel.com/en-us/articles/programming-and-compiling-for-intel-many-integrated-core-architecture

Hi Belinda. Thanks ! I will

Armando_Lazaro_Alami — Thu, 23 Jan 2014 19:30:20 GMT

Hi Belinda. Thanks ! I will check those links and let you know.