omp declare simd support on phi

omp declare simd support on phi

  Hi 

     I am trying to compile some code using the offload model but the compiler is failing with 

        internal error: bad pointer  

     when compiling host a function with pragma omp declare simd decoration.   This does not occur when I remove all pragma offloads (i.e. compiling for the host only)

     Is OMP 4 supported on the phi? and are there any work arounds?

 Setup:    

    Red Hat Enterprise Linux Server release 6.4 (Santiago)

    icc version 15.0.0 Beta (gcc version 4.4.7 compatibility)

     MPSS 3.2.3  

    Thanks

 Jamil

11 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.

There is partial support for OpenMP 4.0; current version is 15.0.0.064.  Internal error on the current compiler is definitely a reportable bug, regardless of correctness or degree of support for your example.  Announcement said all of OpenMP 4 would be supported in the release version except for user defined reduction.  So it would be important to furnish actual examples to help establish limitations and ask for some of those to be alleviated.  Note on the other current thread about OpenMP 4 on MIC that it may be necessary currently to avoid embedded reserved words in names of functions designated by omp declare simd.

I haven't seen the C equivalent of

https://software.intel.com/en-us/articles/explicit-vector-programming-in...

showing original Intel non-standard pragmas and omp 4 equivalents, one of which may work.

https://software.intel.com/en-us/tags/43556

show a list of the videos Intel presented including the intended aspects of omp declare simd

As Tim noted your code triggered an internal compiler error that we need to address. Can you post a reproducing test case to this thread so I can report the error to Development?

With a test case we may be able to identify a work around if one exists.

If there are concerns with posting to the general forum community then you can send me a private reply with the test case attached.

Thank you

Hi Kevin

 I am in the process to creating a test that reproduces the error. It looks like it is a combination of operations as a simple reproduction of the block of code that fails in the application does not fail on its own. 

Thanks

 Jamil

Jamil - Thank you for your private messages. I received an indication of those but that feature is broken (at least for internal users) at the moment and preventing me from retrieving your message and test case. I will keep trying and submit the defect to Development once I obtain your test case.

Thank you.

 Hi Kevin

   I sent you a second message with the code inline. Let me know if you have not received it as I can send it again.

 Thanks

 Jamil

Jamil - Thank you for that. I did find that message and confirmed the error and reported it to our Developers (see internal tracking id below) for investigation/repair. It turns out the error exists in our current production Composer XE 2013 SP1 (14.0 compiler) also. I will keep you updated about a fix as I hear about it.

(Internal tracking id: DPD200358651)

(Resolution Update on 09/02/2014): This defect is fixed in the Intel® C++ Composer XE 2013 SP1 Update 4 release (Version 14.0.4.211 Build 20140805 - Linux)

 

Hi Kevin

  Do you have any indication from the Dev team when a fix is likely to be issued? I am in the process of planning a meeting with senior Intel Execs in September and I want to understand whether I will be in a position to show them any results.

 Thanks

 Jamil

Quote:

Kevin Davis (Intel) wrote:

Jamil - Thank you for that. I did find that message and confirmed the error and reported it to our Developers (see internal tracking id below) for investigation/repair. It turns out the error exists in our current production Composer XE 2013 SP1 (14.0 compiler) also. I will keep you updated about a fix as I hear about it.

(Internal tracking id: DPD200358651)

Hi Jamil - I will check on the progress and let you know shortly. Our next possible opportunity to provide a fix in an official release is in the mid-August time frame. When would you need a fix to create what you need in time for September?

Hi Kevin, mid August would be perfect but the end of the month would be the latest.

This defect is fixed in the Intel® C++ Composer XE 2013 SP1 Update 4 Release (Version 14.0.4.211 Build 20140805 - Linux) now available from our Intel Registration Center.

Leave a Comment

Please sign in to add a comment. Not a member? Join today